doi string | pmcid string | plain_text string | technical_text string | full_text string | journal string | topics list | keywords list | llm_lingua2_text string | llm_lingua2_rate float64 | llm_lingua2_preproc_time float64 |
|---|---|---|---|---|---|---|---|---|---|---|
10.1038/s41467-020-16904-3 | PMC7308400 | Single cell RNA-seq is a powerful method to assign cell identity, but the purity of cell clusters arising from this data is not clear. Here the authors present an entropy-based statistic called ROGUE to quantify the purity of cell clusters, and identify subtypes within clusters. | Single-cell RNA sequencing (scRNA-seq) is a versatile tool for discovering and annotating cell types and states, but the determination and annotation of cell subtypes is often subjective and arbitrary. Often, it is not even clear whether a given cluster is uniform. Here we present an entropy-based statistic, ROGUE, to accurately quantify the purity of identified cell clusters. We demonstrate that our ROGUE metric is broadly applicable, and enables accurate, sensitive and robust assessment of cluster purity on a wide range of simulated and real datasets. Applying this metric to fibroblast, B cell and brain data, we identify additional subtypes and demonstrate the application of ROGUE-guided analyses to detect precise signals in specific subpopulations. ROGUE can be applied to all tested scRNA-seq datasets, and has important implications for evaluating the quality of putative clusters, discovering pure cell subtypes and constructing comprehensive, detailed and standardized single cell atlas. | IntroductionTissues are complex milieus comprising various cell types and states with specialized roles1. Characterizing the property and function of each pure cell type is a long-standing challenge in biological and medical disciplines. The recent advances in scRNA-seq have transformative potential to discover and annotate cell types, providing insights into organ composition2, tumor microenvironment3, cell lineage4, and fundamental cell properties5. However, the identification of cell clusters is often determined by manually checking specific signature genes, which are arbitrary and inherently imprecise. In addition, different methods and even parameters used for normalization, feature selection, batch correction, and clustering can also confound the final identified clusters6, thus motivating the need to accurately assess the purity or quality of identified clusters (Fig. 1a).Fig. 1The expression entropy model.a Identifying pure cell subtypes in unsupervised single-cell data analysis. b The S–E plot of the Tabula Muris (droplet) dataset. Each point represents one gene. The relationship between S and E was fitted with LOESS regression for each gene. c The S–E plot of a T-cell dataset27 obtained by Smart-seq2 protocol. d Accuracy in identifying differentially expressed genes on data simulated from both NB (left) and ZINB (right) distribution, with subpopulation containing 50% of the cells. The center line indicates the median AUC value. The lower and upper hinges represent the 25th and 75th percentiles, respectively, and whiskers denote 1.5 times the interquartile range. Discriminating power of genes selected by S–E model, HVG, Gini, M3Drop, SCTransform, Fano factor, and RaceID3 (“Methods”) estimated by RF with 50 times cross-validation on both droplet-based dataset (e) and full-length-based dataset (f) listed in Supplementary Table 1. The classification accuracy was measured as the percentage of query cells that were assigned the correct label. The center line indicates the median classification accuracy. The lower and upper hinges represent the 25th and 75th percentiles, respectively, and whiskers denote 1.5 times the interquartile range. g Reproducibility of features of brain replicates (Supplementary Table 3). h ARI for the dataset comprising five cell lines6 when different feature selection methods were used.A pure cluster here is defined as a population where all cells have identical function and state without variable genes. The importance of purity assessment is particularly relevant for analyses that aim to discover novel pure subtypes and further detect the true biological signals. For example, signature genes specific to a pure subpopulation maybe mistakenly considered as the common signals of a mixture due to less guided clustering and annotation. The purity evaluation could therefore eliminate such misleading conclusions, potentially aiding our understanding of cellular function, state, and behavior. While pioneering approaches such as silhouette7, DendroSplit8, and distance ratio9 have been devoted to determining the optimal number of identified clusters by calculating the ratio of within-cluster to intercluster dissimilarity, they are not comparable among datasets and have poor interpretability of cluster purity. For example, an average silhouette value of 0.7 indicates a fairly strong consistency for a given cluster, but it is still unknown whether this cluster is a pure population or a mixture of similar subpopulations especially when frequent dropout events occur.The challenges presented by purity evaluation can be broadly addressed by investigating the number of infiltrating nonself cells and variable genes, which are suited to the intended areas of unsupervised variable gene detection. Given its importance, diverse methods10 have been proposed for the quantification and selection of highly variable genes. In particular, scran11 aims to identify variable genes by comparing variance to a local regression trend. However, the over-dispersion, coupled with the high frequency of dropout events, would often result in the selection of many lowly expressed genes12. Alternatively, although Gini coefficient13 could be used to quantify the variation in gene expression, it is specially designed for rare cell-type identification. New probabilistic approaches for variable gene selection using dropout rates have also been recently adapted12, with the advantage of supporting both pseudotime analysis and discrete clustering, but their usage of dropout metric hinders the capturing of the global distribution of gene expression. Although highly informative genes can also be determined by inspecting their weights during multiple iterations of dimensionality reduction14, such ad hoc approaches are computationally intensive, requiring potentially orders of magnitude more time than methods like HVG and M3Drop.Here, we present an entropy-based model to measure the randomness of gene expression in single cells, and demonstrate that this model is scalable across different datasets, capable of identifying variable genes with high sensitivity and precision. Based on this model, we propose the Ratio of Global Unshifted Entropy (ROGUE) statistic to quantify the purity or homogeneity of a given single-cell population while accounting for other technical factors. We demonstrate that the ROGUE metric enables accurate and unbiased assessment of cluster purity, and thus provides an effective measure to evaluate the quality of both published and newly generated cell clusters. Applying ROGUE to B cell, fibroblast, and brain data, we identified additional pure subtypes and demonstrate the application of ROGUE-guided analysis in detecting the precise biological signals. Our approach is broadly applicable for any scRNA-seq datasets, and is implemented in an open-source R package ROGUE (https://github.com/PaulingLiu/ROGUE), which is freely available.ResultsOverview of ROGUEAs scRNA-seq data can be approximated by negative binomial (NB) or zero-inflated NB (ZINB) distribution15,16, we considered the use of the statistic, S (expression entropy—differential entropy of expression distribution, as defined in “Methods”), to capture the degree of disorder or randomness of gene expression. Notably, we observed a strong relationship between S and the mean expression level (E) of genes, thus forming the basis for our expression entropy model (S–E model, Fig. 1b, c). Moreover, S is linearly related to E for the Tabula Muris dataset2 as expected (Fig. 1b), which is characteristic of current droplet experiments, hence demonstrating the NB nature of UMI-based datasets (“Methods”). For a heterogeneous cell population, certain genes would exhibit expression deviation in fractions of cells, leading to constrained randomness of its expression distribution and hence the reduction of S. Accordingly, informative genes can be obtained in an unsupervised fashion by selecting genes with maximal S-reduction (ds) against the null expectation (“Methods”).To provide a direct purity assessment of putative cell clusters or fluorescence-activated cell sorting (FACS)-sorted populations, here we take advantage of the wide applicability of S–E model to scRNA-seq data and introduce the quantitative measure, ROGUE (“Methods”). Intuitively, a cell population with no significant ds for all genes will receive a ROGUE value of 1, indicating it is a completely pure subtype or state. In contrast, a population with maximum summarization of significant ds will yield a purity score of ~0.S–E model accurately identifies informative genesTo illustrate the performance of our model, we benchmarked S–E against other competing feature selection methods (HVG11, Gini13, M3Drop12, SCTransform17, Fano factor18, and RaceID319) on data simulated from both NB and ZINB distribution (“Methods”). For a fair comparison, we generated a total of 1600 evaluation datasets with subpopulations containing 50, 20, 10, or 1% of the cells, and used AUC as a standard to test the performance of each method. Notably, S–E model consistently achieved the highest average AUC and significantly outperformed other gene selection methods in all tested cases with varied subpopulation proportions or gene abundance levels (Fig. 1d and Supplementary Figs. 1 and 2). Although SCTransform is specially designed for UMI-based scRNA-seq data, it exhibited notable performance on ZINB-distributed datasets (Fig. 1d). As a tool to identify genes specific to rare cell types, Gini showed increased performance when there were subpopulations accounting for <20% of the cells. In contrast, HVG performed better in the presence of cell subpopulations with a larger proportion (Supplementary Figs. 1 and 2).To validate our unsupervised feature selection method in real datasets, we performed cross-validation experiments using random forest classifier (RF)20. We randomly sampled 70% cells from the original dataset as reference, and classified the remaining 30% cells, with clusters defined by the original authors (“Methods”). Intuitively, gene sets that enable higher classification accuracy are more biologically meaningful21. Using 14 previously published datasets derived from both droplet-based and full-length protocols (Supplementary Table 1), we demonstrated that our method consistently identified genes with greater ability of classification when different number (30–5000) of genes were selected (Fig. 1e, f and Supplementary Figs. 3 and 4). Specially, our S–E model showed notable superiority when fewer genes (30–100) were used, demonstrating its sensitivity. Taken together, these results suggest that genes identified by our model are more informative and biologically discriminating.Since datasets derived from the same biological system are expected to have reproducible informative genes12, we tested how our expression entropy model behaves using technical replicates from different tissues (Supplementary Table 2). Notably, genes identified by our S–E model were more reproducible when top 500–2000 genes were used (Fig. 1g and Supplementary Fig. 5a–c). In addition, we also considered four pancreatic datasets (Supplementary Table 3) derived from different technologies and labs. These real datasets are more complex than technical replicates as they included systemic nuisance factors such as batch effects. Despite substantial systematic differences, our model consistently achieved high reproducibility scores (Supplementary Fig. 5d).A major task of feature selection is to identify genes that are most relevant for biological heterogeneity, which can be applied to downstream clustering. We therefore evaluate the performance of S–E model in the context of unsupervised clustering with RaceID319, SC322, and Seurat23. Here we considered five publicly available scRNA-seq datasets with high-confidence cell labels6,9,24,25 (“Methods”). These datasets include cells from different lines, FACS-purified populations, or well-characterized types (Supplementary Fig. 6 and “Methods”), and thus can be considered gold standards. To quantify the similarity between the clusters obtained by different clustering methods and the reference cell labels, we calculated the adjusted Rand index (ARI)26, which is restricted to the interval [0, 1]. For the number of features, we considered the top 100, 500, 1000, or 2000 genes. Our results illustrated that S–E model provides the best performance in terms of ARI in these scenarios (Fig. 1h and Supplementary Fig. 7).As some methods were optimized to detect rare cell types, we tested if the genes selected by our S–E model are effective in uncovering such rare subpopulations. To this end, we first simulated a scRNA-seq dataset (“Methods”), which contains three rare clusters (of 10, 30, and 20 cells, respectively) and two common clusters (of 1000 cells each), and clustered these cells with GiniClust218, RaceID3, as well as S–E model-based Seurat (“Methods”). Of note, all these three methods properly recapitulated the five cell clusters (Supplementary Fig. 8), indicating that S–E model-based Seurat is indeed effective for the recovery of both common and rare cell clusters. In addition to simulated data, we wondered how S–E model behaves in detecting real rare cell types. Since no gold standard is available for such cases, we considered four cell lines (A549, H2228, H838, and HCC827) of Tian et al.6, and generated three common cell types (A549, H2228, and H838; of 500 cells for each) and one rare cell type (HCC827; of 20 or 10 cells, respectively) by down-sampling. Similar to the analysis of simulated data, all the three methods effectively identified both common and rare cell clusters when there were 20 cells of the rare cell type (Supplementary Fig. 9a–c). For the dataset with the rare cell type accounting for lower frequency (10 cells, 0.6% of total cells here), RaceID3 and GiniClust2 exhibited their superiority in uncovering the rare cell type as opposed to S–E model-based Seurat (Supplementary Fig. 9d–f). Thus, although S–E model is effective in uncovering rare subpopulations to a certain extent, methods specifically developed for this purpose, such as GiniClust2 and RaceID3, maybe more appropriate.Evaluation of robustness of ROGUETo test how sensitive ROGUE is to the choice of informative genes, here we considered two scRNA-seq datasets: a T-cell dataset sequenced with Smart-seq25 and a droplet-based dataset2 (Tabula Muris). The results illustrated that the heterogeneity score (1-ROGUE) reached saturation when genes with significant ds were selected (Supplementary Fig. 10), thus we used significant ds to calculate ROGUE in the following analyses. We investigated the performance of ROGUE on 1860 cell populations simulated from both NB and ZINB distribution (2000 cells × 20,000 genes each), with 0.1–50% genes varied in a second cell type (“Methods”). A cell population with both fewer infiltrating nonself cells and varied genes would yield a high purity score, while a population with converse situation is expected to yield a low-purity score. It is evident that the ROGUE index decreased monotonically with the heterogeneity of cell populations (Fig. 2a, b and Supplementary Figs. 11 and 12). ROGUE performed well even when cell populations contained few varied genes (<1%) and infiltrating cells (<1%), indicating ROGUE index provides a sensitive and unbiased measure in response to the degree of cell population purity. The usage of different values of the reference factor K (“Methods”) yielded vary similar results (Supplementary Fig. 13), suggesting that ROGUE is robust to the choice of parameter K within a reasonable range.Fig. 2ROGUE use and performance.a The ROGUE index (reference factor K = 45) decreases monotonically with increasing varied genes in each simulated mixture consisting of two cell types (1:1). The center line indicates the median ROGUE value of n = 50 repeated simulations. The lower and upper hinges represent the 25th and 75th percentiles respectively, and whiskers denote 1.5 times the interquartile range. b The ROGUE values (reference factor K = 45) for the simulated mixtures with cell-type sizes ranging from 1:100 to 1:1. In each mixture, the number of varied genes was 1% of the total gene number (n = 20,000). The center line indicates the median ROUGE value of n = 50 repeated simulations. The lower and upper hinges represent the 25th and 75th percentiles respectively, and whiskers denote 1.5 times the interquartile range. c Pearson correlations of S between the randomly down-sampled datasets (n = 50 runs for each) and the entire datasets (2000 cells) simulated from both NB and ZINB distribution. The center line indicates the median correlation value. The lower and upper hinges represent the 25th and 75th percentiles respectively, and whiskers denote 1.5 times the interquartile range. d Sequencing depth distribution (total UMI counts across cells) for two simulated replicates. The replicate 2 has a sequencing depth ten times that of replicate 1. e The S–E plot of the mixture of replicates 1 and 2 is shown in d. f ROGUE values of n = 100 mixtures versus the silhouette values for every two replicates within individual mixtures. A high silhouette value indicates a substantial difference in sequencing depth between two replicates. g, h The S–E plots and corresponding ROGUE values of 10 cell populations from the PBMC dataset24. i Purity assessment of six human T-cell populations. j Purity evaluation of lung-cancer infiltrating DCs, with each point representing a patient. The center line indicates the median ROUGE value. The lower and upper hinges represent the 25th and 75th percentiles, respectively, and whiskers denote 1.5 times the interquartile range.To address the potential concern that the number of cells may represent an intrinsic challenge for S and ROGUE calculation, particularly if only a small number of cells are collected from given samples, we performed down sampling analysis to test how S was impacted by cell numbers. By calculating the Pearson correlations of S between the randomly down-sampled datasets and the entire datasets, we found the similarity values of >0.99 and demonstrated that our S and ROGUE calculation would not be affected by variation in cell number (Fig. 2c).Sequencing depth can vary significantly across cells, with variation potentially spanning orders of magnitude2, and hence contributes to a substantial technical confounder in scRNA-seq data. We sought to investigate whether ROUGE index can accurately assess the purity of single-cell population while accounting for this technical effect. As test cases, we simulated increasing molecular counts (sequencing depth) in a second mock replicate, with the fold change of gene expression means ranging from 2 to 100 (Fig. 2d and “Methods”). Despite the substantial technical effect, the mixture of each two simulated replicates is expected to be a pure cell population. Here we used silhouette to measure the degree of replicate-to-replicate differences. The results revealed ROGUE values of ~0.99–1 for each population consisting of two replicates, with silhouette values ranging from 0.25 to 0.75 (Fig. 2e, f and Supplementary Fig. 14a). Thus, ROGUE not only offers a robust and sensitive way to estimate the purity of single-cell population, but also accounts for the variation in sequencing depth.ROGUE accurately assesses the purity of cell populationsTo illustrate the applicability of ROGUE index to real data, we first considered an External RNA Controls Consortium (ERCC) dataset24, which is a highly controlled experiment dedicated to understanding the technical variability. All 1015 droplets of this dataset received the same ratio of ERCC synthetic spike-in RNAs, hence no varied RNAs should be detected in principle. We referred to this dataset as an ideal case of pure cell population and found only one RNA with significant ds. Accordingly, this ERCC dataset achieved a ROGUE value of ~1 as expected, thus confirming its purity. Further, we investigated the fresh peripheral blood mononuclear cells (PBMCs) enriched from a single healthy donor24. The authors provided multiple cell types purified by FACS, and thus representing a suitable resource for purity assessment. These cell types in Fig. 2g, including CD4/CD8 naïve T cells and CD4 memory T cells, have been shown to be highly homogeneous populations27, and were detected high ROGUE values (0.94–1) as expected. In contrast, both CD14 monocytes and CD34+ cells are mixtures of diverse subtypes24 and received relatively low ROGUE values (~0.8; Fig. 2h), thus confirming their heterogeneity.In addition to highly controlled datasets, it is also instructive to investigate how ROGUE index performs on pure subtypes identified by unsupervised clustering. Here we first considered six well-defined T-cell subtypes from human colorectal cancer5, which were generated via the Smart-seq2 protocol. All these pure subtypes achieved high ROGUE values of >0.9 (Fig. 2i), versus 0.78 for complete data (Supplementary Fig. 14b). We next examined four dendritic cell (DC) subsets collected from human lung cancers28 and sequenced with inDrop platform. Specially, tumor-infiltrating DC2 cells have been proven to be highly heterogeneous populations29,30 and deviated substantially from the other homogeneous cell types including DC1, LAMP3+ DC, and pDC (Fig. 2j). Taken together, these results illustrate that our ROGUE represents an effective and direct quantification of cell population purity without being affected by technical characteristics.ROGUE-guided analysis enhances cell-type identificationWe next evaluated the potential for ROGUE to guide clustering analysis with silhouette, which investigates whether a certain clustering has maximized intercluster dissimilarity and minimized within-cluster dissimilarity. As a test case, we simulated a scRNA-seq dataset consisting of three cell types A, B, and C (see “Methods” for details), where cell types A and B were similar subtypes with 1% varied genes. We clustered this dataset into 2, 3, 4, and 5 subpopulations respectively by adjusting the resolution parameter in Seurat23 (Fig. 3a), then evaluated the results by inspecting corresponding silhouette and average ROGUE values. Proper clustering of this dataset should result in three subpopulations, one for each cell type. However, silhouette received the maximum value when cell-type A co-clustered with B (Fig. 3b), i.e., when only two clusters were identified, suggesting that such measure is poorly interpretable for cluster purity as opposed to ROGUE, which reached saturation when there were three clusters (Fig. 3c). Repeating the simulation with varied differences in cell-type A, B, and C yielded equivalent performance for these two methods (Supplementary Fig. 15a–f). Such performance was also observed when different values of the reference factor K were used (Supplementary Fig. 16). Since ROGUE can provide direct purity quantification of a single cluster and is independent of methods used for normalization, dimensionality reduction, and clustering, it could also be applied to guide the splitting (re-clustering) or merging of specific clusters in unsupervised clustering analyses.Fig. 3ROGUE enhances single-cell clustering and cell-type identification.a t-SNE plots of a simulated dataset containing three cell types. Corresponding silhouette values (b) and average ROGUE values (c) when there were 2, 3, 4, and 5 putative clusters, respectively. d UMAP plots of lung-cancer-associated fibroblasts, color-coded by clusters in original paper (left; Supplementary Fig. 17a) and re-clustered labels (right). e ROGUE values of different clusters before (left) and after (right) re-clustering. Each point represents a patient. The center line indicates the median ROUGE value. The lower and upper hinges represent the 25th and 75th percentiles respectively, and whiskers denote 1.5 times the interquartile range. f UMAP plot of expression levels of MYH11 and MEF2C. g Differences in hallmark pathway activities scored using GSVA.To test how ROGUE could help the clustering of real datasets, we examined a previously reported dataset of cancer-associated fibroblasts (CAFs)31 from lung tumors. CAFs have been reported to represent a highly heterogeneous population and may play a tumor-supportive role in the tumor microenvironment32. We found that the seven identified fibroblast clusters received low ROGUE values (Fig. 3d, e and Supplementary Fig. 17a). We therefore performed re-clustering analysis with the goal of exploring the extent of heterogeneity and identified a total of 11 clusters with a higher average ROGUE value (Fig. 3d, e). In addition to the two classical subtypes of CAFs (myofibroblastic CAFs and inflammatory CAFs), we also found the presence of antigen-presenting CAFs (apCAFs) that was characterized by the high expression of CD74 and MHC class-II genes (Supplementary Fig. 17b). The apCAFs were firstly discovered as a fibroblast subtype in mouse pancreatic ductal adenocarcinoma (PDAC), but barely detectable in human PDAC without forming a separate cluster33. The considerable existence of apCAFs in lung cancer thus may indicate potential differences between different cancer types.Furthermore, we noted that the myCAFs (AF_C02_COL4A1, ROGUE = 0.81) identified by original authors could be further segregated into three distinct subpopulations, including BF_C01_RGS5 (ROGUE = 0.84), BF_C02_ACTA2 (ROGUE = 0.87), and BF_C03_GPX3 (ROGUE = 0.94). Interestingly, the signature genes of AF_C02_COL4A1 described by original authors were actually specific to one of these three subpopulations, including MEF2C in BF_C01_RGS5 and MYH11 in BF_C02_ACTA2 (Fig. 3f). Pathway analysis also revealed that the NOTCH signaling was activated in BF_C01_RGS5 (Fig. 3g) rather than a common signal of AF_C02_COL4A131. Despite the considerable increase of overall ROGUE index, BF_C00_AOL10A1, BF_C04_COL1A2, and BF_C05_PLA2G2A still received relatively low ROGUE values, thus deserving further investigation. Overall, ROGUE-guided analysis not only discovered novel cell subtypes, but also enabled the detection of the true signals in specific pure subpopulations.ROGUE-guided analysis identifies pure B cell subtypesB cells are key components in tumor microenvironment but have unclear functions in antitumor humoral response34. Here we investigated previously reported liver- and lung-tumor-infiltrating B cells31,35 and found that they received relatively low ROGUE values (Fig. 4a). Thus, we applied further clustering analysis coupled with ROGUE to these B cells in an attempt to discover pure subtypes. A total of seven clusters were identified, each with its specific marker genes (Fig. 4b–d). Cells from the first B-cell subset, B_C0_JUNB, specifically expressed signature genes including JUNB and FOS, thus representing activated B cells36. The second subset, B_C1_TXNIP, showed high expression of glycolysis pathway genes (Supplementary Table 4), indicating its metabolic differences. ACTB, a gene involved in antigen presenting, was highly expressed in the third subset (B_C2_ACTB). Pathway activity analysis also revealed a strong antigen processing and presentation signal in this subset (Supplementary Table 4). The fourth cluster, B_C3_FCER2, characterized by high expression of HVCN1 and genes involved in B-cell receptor signaling pathway (Supplementary Table 4), was largely composed of pre-activated B cells37. The fifth cluster, B_C4_MX1, predominantly composed of interferon-induced B cells38, expressed high levels of MX1, IFI6, and IFI44L. The sixth cluster, B_C5_CD3D, expressed key markers of both T- and B-cell lineages (Fig. 4d), thus maybe the dual expressers (DEs)-like lymphocytes39 or doublets. The remaining B cells, falling into the seventh cluster, B_C6_LRMP, exhibited high expression of LRMP and RGS13, indicative of the identity of germinal center B cells40.Fig. 4ROGUE-guided analysis in the identification of pure B-cell subtypes.a The S–E plots and ROGUE values of liver- and lung-tumor-infiltrating B cells, respectively. UMAP plots of 4291 B cells, color-coded by their associated clusters (b) and tissues (c). d Gene expression heatmap of seven B-cell clusters. Rows denote marker genes and columns denote different clusters. e ROGUE values of seven identified B-cell subtypes. Each point represents a patient. The center line indicates the median ROUGE value. The lower and upper hinges represent the 25th and 75th percentiles, respectively, and whiskers denote 1.5 times the interquartile range. f Tissue preference of each B-cell subtype in liver cancer estimated by RO/E27, the ratio of observed to expected cell numbers calculated by the chi-square test. g The average fractions of B_C02_ACTB and B_C04_MX1 in each patient across tissues, where error bars representing ±s.e.m. *p < 0.05, **p < 0.005, Student’s t test. The Kaplan–Meier curves of TCGA LUAD (h) and LIHC (i) patients grouped by the 13 markers (Supplementary Table 5) of B_C02_ACTB.Both DEs/doublets-like and germinal center B cells exhibited low ROGUE values (Fig. 4e), but the limited cells did not permit further clustering. For germinal center B cells, we readily detected the high expression of proliferating marker genes, including MKI67 and STMN1 (Supplementary Fig. 18), in a fraction of these cells, thus explaining the heterogeneity to some extent. In contrast to these two clusters, we found ROGUE values of >0.92 for each of the remaining five clusters (Fig. 4e), demonstrating that they were all highly homogeneous B-cell subtypes. By calculating the ratio of observed to expected cell numbers with the chi-square test (RO/E), we noted that both B_C02_ACTB and B_C04_MX1 contained mainly cells from tumor, with RO/E values >1 (Fig. 4f). Similar analyses stratified by patient further confirmed this trend (Fig. 4g). Based on the independent TCGA lung adenocarcinoma (LUAD) cohort dataset, patients with higher expression of the marker genes of B_C02_ACTB (normalized by MS4A1; Supplementary Table 5) showed significantly worse overall survival (Fig. 4h). Such survival difference was also observed in TCGA liver hepatocellular carcinoma (LIHC) cohort dataset (Fig. 4i). Thus, the clinical implication deserves further study to investigate what specific roles B_C02_ACTB cells play in tumor microenvironment. In summary, identifying pure subtypes with ROGUE-guided analysis could enable a deeper biological understanding of cell state and behavior.Application to brain data and batch effect evaluationIn addition to cancer data, we also demonstrated the application of ROGUE in analyzing the brain transcriptome dataset2, which harbors a high degree of heterogeneity for those encapsulated cell classes. This dataset identified seven distinct cell types, of which oligodendrocyte and neuron cell types had low ROGUE values of <0.8, versus ~0.9–1 for the remaining five cell classes (Fig. 5a). We therefore applied further clustering guided by ROGUE to oligodendrocyte which is of enough cells (n = 3401), and identified ten refined cell subtypes, each with its specific marker genes (Fig. 5b, c). Except cluster 6, we found ROGUE values of ~0.9–1 for all the other nine clusters, suggesting their purity (Fig. 5d). To investigate potential functions of these subtypes, we compared pathway activities and found considerable phenotypic diversity. For example, cluster 5 showed a strong signal of axon guidance signaling (Fig. 5e), while neurotrophin signaling pathway was highly activated in cluster 1 (Fig. 5f). This example further illustrates how ROGUE plays a key role in uncovering pure subpopulations.Fig. 5The application of ROGUE in brain data and batch effect evaluation.a ROGUE values of seven distinct brain cell types as defined by the original publication2, with each point representing a sample. The center line indicates the median ROUGE value. The lower and upper hinges represent the 25th and 75th percentiles respectively, and whiskers denote 1.5 times the interquartile range. b UMAP plot of the ten identified clusters of oligodendrocytes (n = 3401), color-coded by their associated clusters. c Expression heatmap of cell-type-specific genes of the ten oligodendrocyte clusters. d ROGUE values of oligodendrocyte clusters. Each point represents a sample. The center line indicates the median ROUGE value. The lower and upper hinges represent the 25th and 75th percentiles, respectively, and whiskers denote 1.5 times the interquartile range. e, f Enriched pathways for cluster 5 (e) and cluster 1 (f), respectively. g ROGUE values were shown for batch 1 (the control group), batch 2 (the stimulation group), and aggregated cell population (batch 1 and batch 2) for each cell type. For fair comparison, we equalized the number of cells in each group by down-sampling. The center line indicates the median ROGUE value. The lower and upper hinges represent the 25th and 75th percentiles, respectively, and whiskers denote 1.5 times the interquartile range. *p < 0.05, **p < 0.005, Student’s t test. h ROGUE values for individual-specific cell populations and aggregated populations (all individuals). All cells used here were from the control group. Subsampling was performed to equalize the number of cells in each group. The center line indicates the median ROUGE value. The lower and upper hinges represent the 25th and 75th percentiles, respectively, and whiskers denote 1.5 times the interquartile range. *p < 0.05, **p < 0.005, Student’s t test.To investigate if ROGUE is effective in evaluating the impact of batch effect, we studied a dataset of human PBMCs containing multiple distinct cell types38. Cells of this dataset were previously split into two groups—the interferon-beta (IFN-β)-stimulated group and the culture-matched control group, thus could be considered as two batches. Then we applied ROGUE to assess the purity of each cell type (as defined by the original authors) in individual bathes as well as the aggregated cell population (batch 1 and batch 2), and found that ROGUE detected considerable purity reduction in the aggregated group (Fig. 5g).As cells of this dataset were collected from eight unrelated individuals, we also tested how ROGUE behaves in estimating the variability (i.e., batch effect) among patients. Here we only used cells from the control group so that the evaluation would not be influenced by IFN-β perturbation. As expected, the aggregated cell populations of all individuals received significantly lower ROGUE values as opposed to patient-specific populations for each cell type (Fig. 5h). Thus, ROGUE offers a reasonable method for estimating the impact of batch effect.DiscussionPurity assessment of identified cell clusters is paramount to the interpretation of scRNA-seq data. This assessment is especially pertinent as increasingly rare and subtle cell subtypes are being uncovered. To address this computational challenge, we present the S–E model and demonstrate that this model is capable of identifying variable genes with high sensitivity and precision, and thus could be applied to both clustering and potentially pseudotime analyses. By taking advantage of the wide applicability of S–E model, we develop the statistic ROGUE to quantify the purity of single-cell populations. Through a wide range of tests, we demonstrate that our entropy-based measure, ROGUE, is broadly applicable for datasets from different platforms, protocols and operators, and able to successfully quantify the purity of singl-cell populations regardless of uncontrollable cell-to-cell variation.When using ROGUE to assess the purity of four DC subtypes from human lung tumors, we found that DC2 was a heterogeneous population, which is consistent with previous findings30. Such heterogeneous populations like DC2 may have different properties and specialized roles in the cancer microenvironment, and could be assessed in a similar fashion with ROGUE. Accordingly, future studies could focus on these cell populations and hence may deepen our understanding of cellular origins of cancer. In addition, ROGUE addresses an important need in unsupervised single-cell data analyses, i.e., to effectively assess the quality of published or newly generated clusters. Often, unsupervised clustering may lead to under- or over-clustering of cells due to the lack of universal stands for clustering quality. By quantifying cluster purity with ROGUE before and after clustering or re-clustering, we were able to detect low-purity clusters and perform further analysis to discover pure subtypes. Improving the purity and credibility of the ever-increasing number of cell types is a mounting challenge with explosive efforts toward single-cell sequencing, and ROGUE could become a potential standard for judging the quality of cell clusters.Our ROGUE-guided analysis on fibroblasts identified a novel subpopulation in lung cancer, apCAFs, which highly expressed CD74 as well as MHC class-II genes and had a strong antigen-presenting signal. These cells have been speculated to deactivate CD4 T cells and decrease the CD8+ to Treg ratio in mouse PDAC33, but have unclear role in the lung-cancer microenvironment, hence requiring further investigation. Moreover, when applying ROUGE to B-cell analysis, we found an interesting pure cluster B_C02_ACTB that displayed high expression of genes involved in antigen processing and presentation. Cells from this cluster were preferentially enriched in tumors and were associated with poor prognostic outcomes in both lung and liver cancer. We therefore hypothesize that these cells may contribute to immune suppression in the cancer microenvironment and hence curtail antitumor immunity, although further studies are required to define the roles of these cells. Such approaches for discovering novel or additional pure subtypes can also be extended to other published or newly generated scRNA-seq datasets.When determining the purity of cell clusters, we recommend a ROGUE value of 0.9 as a suitable threshold, at which the number of infiltrating cells and varied genes is well constrained. But for low-quality data or continuous data, the threshold could be determined by considering the global ROGUE values. Although ROGUE can be very efficient and effective, we anticipate that additional extensions could enable enhanced performance, for example, assessing the purity of integrated cell populations from different protocols and platforms. Overall, our ROGUE metric provides a robust and direct measure for cluster purity in the presence of substantial technical confounders. We expect the ROGUE metric to be broadly applicable to any scRNA-seq datasets, and anticipate that our strategy will improve the rigor and quality of unsupervised single-cell data analysis.MethodsExpression entropy modelFor droplet datasets, the observed UMI count can be modeled as a NB random variable, which also arises as a Poisson–Gamma mixture411\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{l}X_{ij}\sim {\mathrm{Poisson}} \,( {s_j\lambda _{ij}})\\ \, \,\,\,\,\,\lambda _{ij}\sim {\mathrm{Gamma}}\,( {\alpha _{ij},\beta _{ij}})\end{array},$$\end{document}Xij~Poisson(sjλij)λij~Gamma(αij,βij),where λij represents the true expression value that underlies the observed UMI count Xij of gene i in cell j, and sj denotes the size normalization factor in cell j. The αij and βij are shape parameter and rate parameter respectively. Given the assumption that the shape parameter α is a constant across cells and genes, and that the rate parameter β is a constant of gene i across cells41,42, αij and βij can be expressed as α and βi, respectively. Then the distributions can be recognized as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda _i\sim {\mathrm{Gamma}}\left( {\alpha ,\beta _i} \right)$$\end{document}λi~Gammaα,βi and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_{ij}\sim {\mathrm{Poisson}}\,( {s_j\lambda _i})$$\end{document}Xij~Poisson(sjλi). We denote2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_{ij}^\prime = \frac{{X_{ij}}}{{s_j}},$$\end{document}Xij′=Xijsj,as the normalized expression of gene i in cell j, and use \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Bbb E}\left( {X_i^\prime } \right)$$\end{document}EXi′ (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_i^\prime$$\end{document}Xi′ is the normalized expression assigned to gene i and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Bbb E}\left( {X_i^\prime } \right)$$\end{document}EXi′ is the expectation across cells) as the moment estimation of λi. For the Gamma distribution, the rate parameter could therefore be calculated based on the maximum likelihood estimation3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta _i = \frac{\alpha }{{\lambda _i}} = \frac{\alpha }{{{\Bbb E}\left( {X_i^\prime } \right)}}.$$\end{document}βi=αλi=αEXi′.To capture the degree of disorder or randomness of gene expression, here we considered the use of differential entropy defined as434\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H\left( X \right) = - \int_{ - \infty }^{ + \infty } {p\left( x \right) \cdot {\mathrm{ln}}\,p\left( x \right)dx},$$\end{document}HX=−∫−∞+∞px⋅lnpxdx,where X is a continuous random variable and p(x) is the probability density function. Differential entropy is an extension of Shannon entropy, which is used to measure the average surprisal of a continuous probability distribution, and has shown notable performance in our supervised gene selection method E-test44. Specially, for the gamma distributed random variable λi, its differential entropy can be computed as5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_i = \alpha - {\mathrm{ln}}\beta _i + {\mathrm{ln}}\Gamma \left( \alpha \right) + \left( {1 - \alpha } \right) \cdot \varphi \left( \alpha \right) = {\mathrm{ln}}\frac{\alpha }{{\beta _i}} + a = {\mathrm{ln}}{\Bbb E}\left( {X_i^\prime } \right) + a,$$\end{document}Si=α−lnβi+lnΓα+1−α⋅φα=lnαβi+a=lnEXi′+a,where φ is the digamma function, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a = \alpha - {\mathrm{ln}}\alpha + {\mathrm{ln}}\Gamma \left( \alpha \right) + \left( {1 - \alpha } \right) \cdot {\upvarphi}\left( \alpha \right)$$\end{document}a=α−lnα+lnΓα+1−α⋅φα is a constant. Although other pioneering methods such as Scnorm45, scran46, and BASiCS47 can be used to calculate size factors, we considered the library size normalization of each cell defined as the total UMI counts divided by the mean total UMI counts across cells41. Accordingly, the expectation of library size factor across cells is equal to 1. Given Eq. (2) and that the gene expression and library size are two independent random variables42, for a given gene i, we have6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Bbb E}\left( {X_i} \right) = {\Bbb E}\left( {X_i^\prime \times s} \right) = {\Bbb E}\left( {X_i^\prime } \right) \times {\Bbb E}\left( s \right) = {\Bbb E}\left( {X_i^\prime } \right) \times 1 = {\Bbb E}\left( {X_i^\prime } \right),$$\end{document}EXi=EXi′×s=EXi′×Es=EXi′×1=EXi′,where Xi is the observed expression assigned to gene i and s is the library size assigned to cells. Thus, for each cell type, the differential entropy of λi could be computed as7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_i = {\mathrm{ln}}{\Bbb E}\left( {X_i} \right) + a.$$\end{document}Si=lnEXi+a.We formulate the null hypothesis that there is only one Poisson–Gamma component for each gene in a given population (H0) and thus the corresponding differential entropy can be calculated with Eq. (7). Then we assume that each cell represents its own cluster and use Xij as a moment estimation of the mean expression of such cluster. In this way, we define the entropy reduction of gene i across n cells as8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ds_i = {\mathrm{differential}}\,{\mathrm{entropy}}\,{\mathrm{under}}\,H_0 - {\mathrm{average}}\,{\mathrm{actual}}\,{\mathrm{differential}}\,{\mathrm{entropy}}\\ = {\mathrm{ln}}{\Bbb E}\left( {X_i} \right) - \frac{{\mathop {\sum }\nolimits_{j = 1}^n \,( {{\mathrm{ln}}X_{ij}} )}}{n},$$\end{document}dsi=differentialentropyunderH0−averageactualdifferentialentropy=lnEXi−∑j=1n(lnXij)n,which captures the degree of disorder or randomness of gene expression44. Given that genes under H0 (non-variable genes) account for the major proportion, we fit the relationship between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{ln}}\,{\Bbb E}\left( {X_i} \right)$$\end{document}lnEXi and average actual differential entropy, and calculate corresponding residual as dsi to improve the performance (Fig. 1b, c). The significance of ds is estimated based on a normal distribution approximation and is adjusted using Benjamini–Hochberg method. We also extended such procedure to full-length datasets and found that our approach consistently outperformed other gene selection methods (Fig. 1f and Supplementary Fig. 4).Data simulationWe simulated droplet datasets with NB distribution. Mean gene abundance levels E were sampled from the log-normal distribution\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ln\left( E \right)\sim {N}\left( {\mu ,\sigma ^2} \right),$$\end{document}lnE~Nμ,σ2,with parameters μ = 0 and σ = 2. The number of transcripts for each gene were drawn from\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{ij} \sim {\mathrm{NB}}\left( {E_i,r} \right).$$\end{document}Nij~NBEi,r.For each simulated dataset, the dispersion parameter r (r = α)48 was set to a fixed value, ranging from 5 to 20 (Supplementary Fig. 1). In addition, we simulated full-transcript datasets with ZINB distribution. The dropout rates for each gene was modeled with the sigmoid function49\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_i \sim {\mathrm{sigm}}\left( { - \left( {\gamma _0 + \gamma _1E_i} \right)} \right),$$\end{document}Pi~sigm−γ0+γ1Ei,with parameters γ0 = −1.5 and γ1 = 1/median(E). Each simulated scRNA-seq dataset contained 20,000 genes and 2000 cells (Supplementary Fig. 2).Differentially expressed genes were added in a fraction of cells (1–50%, Supplementary Fig. 1 and 2), with fold changes sampled from the log-normal distribution (μ = 0 and σ = 2). Genes with a >1.5-fold decrease or increase in mean expression were considered as ground truth DE genes.Feature selection methodsThe HVG method11 identifies variable genes by comparing the coefficient of variation squared to a local regression trend, and was implemented with the BrenneckeGetVariableGenes function in the M3Drop12 package. In the Gini index model proposed in GiniClust13, a gene is considered as informative if its Gini is higher than expected from the maximum observed expression. We copied the source code of original GiniClust (GiniClust_Preprocess.R, GiniClust_Filtering.R and GiniClust_Fitting.R) (https://github.com/lanjiangboston/GiniClust/tree/master/Rfunction), and defined the Gini_fun function in our scripts to select genes. M3Drop uses dropout rates for variable gene selection and was implemented with the M3DropFeatureSelection function in the M3Drop package. The SCTransform method17 selects genes with Pearson residuals from the regularized negative binominal regression and was implemented with the SCTransform function in Seurat package. In addition, we implemented the Fano factor method as used in the script GiniClust2_Fano_clustering.R from GiniClust218. The feature selection step in RaceID319, which selects genes with a second-order polynomial fit between the expression variance and log-transformed mean, was implemented according to the fitBackVar function in RaceID3.Datasets used for clustering-based evaluationTo evaluate the performance of different feature selection methods in the context of unsupervised clustering, here we considered five publicly available scRNA-seq datasets. The first dataset6 consists of five cell lines (A549, H1975, H2228, H838, and HCC827) and was sequenced with 10X Genomics protocol, with a total of 3918 cells. The second dataset was generated by the same study6. This dataset comprises three cell lines (H1975, H2228, and HCC827) and was sequenced with CEL-seq2 protocol. The third dataset24 was created by processing multiple FACS-purified cell populations and was sequenced with 10X Genomics protocol. Considering that some populations such as CD8+ cytotoxic T cells were relatively heterogenous24, here we only used CD19 B cells, CD4 naïve T cells, CD56 NK cells, and CD14 monocytes, which were readily distinguishable (Supplementary Fig. 6a). The fourth dataset contains cells from human pancreatic islet and was generated by Smart-seq protocol25. These pancreatic cell types including alpha, beta, delta, and gamma cells are well-characterized and have been shown to be distinct23,44, thus were used for benchmarking (Supplementary Fig. 6b). The remaining dataset comprises multiple immune cell types9, with cells sequenced by Smart-seq2 protocol. Although the cell labels in original publication were assigned using unsupervised clustering, cross-validation experiments revealed that the major cell types (macrophages, DCs, lymphocytes, and exhausted CD8 T cells) were readily distinguishable (Supplementary Fig. 6c). We therefore also consider this dataset for benchmarking.Cross-validation experiments and gene reproducibilityTo illustrate the performance of S–E model in real datasets (Supplementary Table 1), we performed cross-validation experiments using the procedure as implemented in scmap: (i) we randomly selected 70% of the cells as the reference set, (ii) we then identified informative genes (based on the reference set) with different feature selection methods respectively, (iii) we further trained the RF classifier50 using the reference set with only informative genes selected by different methods (cell labels were defined with unsupervised clustering by the original authors), (iv) the remaining 30% cells were considered as query set, and corresponding cell types were predicted with the trained classifier, (v) the classification accuracy was then quantified with the accuracy score50, which is the similarity between the predicted cell types and the original cell types of the query set, (vi) finally, we repeated this entire procedure for n = 50 times for each dataset.We calculated the reproducibility by intersecting the corresponding sets of variable genes as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Reproducibility}} = \frac{{{\mathrm{Geneset}}_{{\mathrm{replicate}} - 1}^{m - n} \cap {\mathrm{Geneset}}_{{\mathrm{replicate}} - 2}^{m - n}}}{n},$$\end{document}Reproducibility=Genesetreplicate−1m−n∩Genesetreplicate−2m−nn,where m denotes the adapted gene selection method and n is the number of top-ranked variable genes.Rare cell-type simulationWe simulated the synthetic scRNA-seq data following the same approach in GiniClust2 (https://github.com/dtsoucas/GiniClust2/blob/master/Rfunction/Generate_Simulated_Data.R), specifying two large 1000 cell clusters, and three rare clusters of 10, 20, and 30 cells, respectively. To test the performance of our method, we applied our S–E model to the raw count data to select informative genes and performed follow-up clustering with standard clustering procedure in Seurat. The R scripts of RaceID3 and GiniClust2 were accessed through https://github.com/dgrun/RaceID3_StemID2_package and https://github.com/dtsoucas/GiniClust2, respectively.ROGUE calculationBy taking advantage of the wide applicability of S–E model to scRNA-seq data, we introduce the statistic ROGUE to measure the purity of a cell population as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{ROGUE}} = 1 - \frac{{\mathop {\sum }\nolimits_{{\mathrm{sig}}} ds}}{{\mathop {\sum }\nolimits_{{\mathrm{sig}}} ds + K}},$$\end{document}ROGUE=1−∑sigds∑sigds+K,where the parameter K is used for two purposes: (i) constrain the ROGUE value between 0 and 1, (ii) serve as a reference factor to provide the purity evaluation. Consider a reference dataset with maximum summarization of significant ds. We set the value of K to one-half of the maximum. In this way, ROGUE will receive a value of 0.5 when summarized significant ds is equivalent to one-half of the maximum. A cell population with no significant ds for all genes will receive a ROGUE value of 1, while a population with large summarization of significant ds is supposed to yield a small purity score. We reasoned that Tabula Muris can be considered as such a plausible reference dataset because it comprises cells from 20 organs, which represents a highly heterogeneous population and was sequenced with both 10X Genomics and Smart-seq2 protocols2. As the technical variation associated with PCR, which is present in full-length-based but not droplet-based technology, will affect the value of ds, we calculated the summarization of significant ds of Tabula Muris for both 10X Genomics and Smart-seq2 datasets (Supplementary Fig. 19). Accordingly, we set the default value of K to one-half of the summarization, i.e., 45 for droplet-based data and 500 for full-length-based data, receptively. The K value can also be determined in a similar way by specifying a different reference dataset in particular scRNA-seq data analyses. Users should be careful when using the default K value on datasets of different species, and we recommend the user to determine the K value by specifying a highly heterogeneous dataset of that species with the DetermineK function in ROGUE package.Silhouette coefficientTo assess the differences of simulated replicates and the separation of different cell clusters, we calculated the silhouette width7, which is the ratio of within-cluster to intercluster dissimilarity. Let a(i) denote the average dissimilarity of cell i to all other cells of its cluster A, and let b(i) denote the average dissimilarity of cell i to all data points assigned to the neighboring cluster, whose dissimilarity with cluster A is minimal. The silhouette width for a given cell i is defined as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s\left( i \right) = \frac{{b\left( i \right) - a(i)}}{{{\mathrm{max}}\left( {a\left( i \right),b\left( i \right)} \right)}}.$$\end{document}si=bi−a(i)maxai,bi.A high s(i) value suggests that the cell i is well assigned to its own cluster but poorly assigned to neighboring clusters.Sequencing depth simulationSequencing depth can vary significantly across cells and thus contributes to a substantial technical confounder in scRNA-seq data analysis. To illustrate that ROGUE is robust to sequencing depth, we generated simulated populations, each consisting of two replicates with only differences in sequencing depth (Fig. 4d and Supplementary Fig. 7a). In each simulation, we varied the sequencing depth of the two replicates as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _{{\mathrm{replicate}} - 2,i} = \mu _{{\mathrm{replicate}} - 1,i} \cdot \delta ,\,i \in \left\{ {1, \ldots ,n} \right\},$$\end{document}μreplicate−2,i=μreplicate−1,i⋅δ,i∈1,…,n,where n is the number of genes, μ is the mean expression level, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta \in \left\{ {2,\,5,\,10,\,20,\,50,\,70,\,100} \right\}$$\end{document}δ∈2,5,10,20,50,70,100.Generation of simulated cell typesTo demonstrate the potential for ROGUE to guide single-cell clustering, we used NB model as aforementioned to simulate different scRNA-seq datasets, each consisting of three cell types A, B, and C (1000 cells × 10,000 genes each), where A and B were similar subtypes. For the three scenarios shown in Fig. 3a and Supplementary Fig. 15a, d, we introduced 500, 1000, and 800 varied genes between cell-type A and cell-type B/C, respectively, with fold changes drawn from the log-normal distribution (μ = 0 and σ = 2). In addition, we simulated 100, 100, and 120 highly variable genes between cell-type B and C respectively, with fold changes sampled from a log-normal distribution with μ = 0 and σ = 1. The results were visualized using t-distributed stochastic neighbor embedding (t-SNE) implemented in R package Rtsne.Analysis of the fibroblast and B-cell datasetsTo demonstrate the application of ROGUE-guided analysis in identifying pure subpopulations and detecting precise biological signals, we performed re-clustering analysis of the fibroblast and B-cell datasets31,35. We filtered out low-quality cells with either <600 expressed genes, over 25,000 or below 600 UMIs. After filtration, a total of 4291 B cells and 1465 fibroblasts were remained. We further applied our S–E model to the raw count data to select informative genes. Although other pioneering methods could be used to calculate size factors39,40,45, we normalized the gene expression matrices using regularized NB regression in Seurat23. The top 3000 genes with maximal ds were used for PCA analysis. To remove batch effects between donors, we performed batch correction using BBKNN51 with the first 50 PCs. Using the leiden clustering approach implemented in scanpy52, each cell cluster was identified by its principle components. This yielded 11 fibroblast subtypes and 7 B-cell subtypes as shown in Figs. 3d and 4b, which were visualized in 2D projection of UMAP53 with default parameters. Accordingly, the purity score of each cluster was calculated with the rogue function in our R package. The calculation of ROGUE is based on raw count data and is independent of methods used for normalization, dimensionality reduction, and clustering.Pathway and TCGA data analysisTo characterize and detect the pathway signals in specific fibroblast subtypes, we performed pathway analyses using hallmark pathways from the molecular signature database54 with GSVA55. The TCGA LUAD and LIHC data were used to investigate the prognostic effect of 13 signature genes (Supplementary Table 5) derived from B_C2_ACTB. To eliminate the effects of different B-cell proportions, we normalized the mean abundance level of these 13 marker genes by the expression of MS4A1 gene, and performed subsequent statistical analyses using GEPIA256 with default parameters.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary | nature communications | [
"Article"
] | [
"RNA sequencing",
"Computational models",
"Software"
] | complex cell types specialized roles1 Characterizing function cell type challenge advances scRNA-seq discover annotate cell types insights organ tumor microenvironment3 cell cell clusters checking signature genes arbitrary imprecise methods parameters normalization feature selection batch correction clustering confound clusters6 assess purity clusters (Fig. 1a).Fig. expression entropy model Identifying pure cell subtypes single-cell data analysis S–E plot Tabula Muris dataset Each point represents one gene relationship E fitted with LOESS regression S–E plot T-cell dataset27 Smart-seq2 protocol Accuracy identifying differentially expressed genes data NB ZINB distribution subpopulation 50% cells center line median AUC value lower upper hinges represent 25th 75th percentiles whiskers 1.5 times interquartile range Discriminating power of genes S–E model HVG Gini M3Drop SCTransform Fano factor RaceID3 estimated by RF with 50 times cross-validation Supplementary Table 1. classification accuracy measured as percentage query cells assigned correct label center line median classification accuracylower upper hinges represent 25th 75th percentiles whiskers 1.5 interquartile range Reproducibility features brain replicates (Supplementary Table 3) ARI dataset five cell lines6 different feature selection methods used pure cluster defined population cells identical function state without variable genes purity assessment relevant for analyses novel pure subtypes detect true biological signals signature genes pure subpopulation mistakenly common signals mixture guided clustering annotation purity evaluation could eliminate misleading conclusions understanding cellular function state behavior approaches silhouette7 DendroSplit8 distance ratio9 optimal number clusters not comparable poor interpretability cluster purity average silhouette value 0.7 indicates strong consistency cluster unknown whether cluster pure or mixture subpopulations frequent dropout events challenges purity evaluation addressed investigating infiltrating nonself cells variable genes unsupervised variable gene detection diverse proposed for quantification selection highly variable genes scran11 identify variable genes comparing variance local regression trend over-dispersion high frequency dropout events selection lowly expressed genes12 Gini coefficient13 quantify variation gene expression designed for rare cell-typeprobabilistic approaches for variable gene selection using dropout rates adapted12 pseudotime analysis discrete clustering dropout metric hinders global distribution gene expression informative genes determined by inspecting weights during dimensionality approaches computationally intensive more time than HVG M3Drop present entropy-based model randomness gene expression in single cells scalable across datasets identifying variable genes with high precision propose Ratio of Global Unshifted Entropy (ROGUE) statistic to quantify purity single-cell population technical factors ROGUE metric enables unbiased assessment cluster purity effective measure quality of published generated cell clusters Applying ROGUE to B cell fibroblast brain data identified pure subtypes application ROGUE-guided analysis detecting precise biological signals approach applicable for scRNA-seq datasets implemented in open-source R package ROGUE freely available scRNA-seq data approximated by negative binomial considered use statistic S (expression to capture disorder randomness gene expression observed strong relationship between S mean expression level (E of genes basis for expression entropy modelS linearly related to E Tabula Muris dataset2 (Fig. characteristic of current droplet experiments nature UMI-based datasets heterogeneous cell population genes exhibit expression deviation constrained randomness expression distribution reduction S informative genes obtained selecting genes with maximal S-reduction (ds) against null expectation purity assessment cell populations S–E model scRNA-seq data quantitative measure ROGUE cell population with no significant ds genes ROGUE value 1 pure subtype population with maximum summarization significant ds purity score ~0.S–E model informative benchmarked S–E against competing feature selection methods (HVG11 Gini13 M3Drop12 SCTransform17 Fano factor18 RaceID319) data NB ZINB distribution generated 1600 evaluation datasets subpopulations 50 20 10 1% cells used AUC standard S–E model achieved highest average AUC outperformed other gene selection methods cases varied subpopulation proportions gene abundance levels (Fig. 1d Supplementary Figs. 1 and 2)SCTransform designed for UMI-based scRNA-seq data exhibited performance on ZINB-distributed datasets (Fig. 1d). genes rare cell types Gini increased performance subpopulations <20% cells HVG performed better subpopulations larger proportion Figs. 1 2) unsupervised feature selection method performed cross-validation experiments using random forest classifier (RF randomly sampled 70% cells original dataset classified remaining 30% defined authors gene sets higher classification accuracy biologically meaningful21 14 datasets method identified genes greater classification when (30–5000) genes selected (Fig. 1e f 3 4) S–E model showed superiority when fewer genes (30–100) used results suggest genes identified model more informative biologically discriminating datasets same system reproducible tested expression entropy model using technical replicates from different tissues Table 2) genes identified S–E model more reproducible when top 500–2000 genes used (Fig. 1g considered four pancreatic datasets datasets more complex included systemic nuisance factorsdifferences model achieved high reproducibility scores Fig. task feature selection genes relevant for biological heterogeneity downstream clustering performance S–E model unsupervised clustering with RaceID319 SC322 Seurat23 considered five scRNA-seq datasets with high-confidence cell labels6,9,24,25 datasets include cells from different lines FACS-purified populations well-characterized types Fig. 6 gold standards similarity between clusters calculated adjusted Rand index (ARI)26 restricted to interval [0, 1] considered top 100 500 1000 2000 genes results S–E model best performance scenarios (Fig. 1h Fig. 7) methods rare cell types tested if genes S–E model effective uncovering rare subpopulations simulated scRNA-seq dataset three rare clusters two common clusters 1000 clustered with GiniClust218 RaceID3 S–E model-based Seurat methods recapitulated five cell clusters Fig. 8) S–E model-based Seurat effective for recovery of common rare cell clusters wondered S–E model detecting real rare cell typesno gold standard considered four cell lines (A549 H2228 H838 HCC827) Tian et al.6 generated three common cell types (A549 H2228 H838 500 cells one rare type (HCC827 20 or 10 cells by down-sampling three methods identified common rare cell clusters 20 cells rare type Fig. 9a–c). dataset rare cell type lower frequency (10 cells 0.6% total cells RaceID3 GiniClust2 rare cell type S–E model Seurat 9d–f). S–E model effective rare subpopulations methods GiniClust2 RaceID3 appropriate.Evaluation robustness ROGUETo ROGUE genes considered two scRNA-seq datasets T-cell dataset Smart-seq25 droplet-based dataset2 heterogeneity score (1-ROGUE) saturation when genes significant ds selected used significant ds calculate ROGUE investigated performance ROGUE on 1860 cell populations NB ZINB distribution 0.1–50% genes varied second cell type population fewer nonself cells varied genes high purity score converse low-purity score ROGUE index decreased with heterogeneity cell populationsFigs. 11 ROGUE performed populations varied genes (<1%) infiltrating cells<1%) index sensitive unbiased measure cell population purity different values reference factor K similar results Fig. ROGUE robust parameter K.Fig. 2ROGUE use performance ROGUE index K = 45 decreases increasing varied genes mixture cell types (1:1) center line median ROGUE value n = 50 simulations hinges 25th 75th percentiles whiskers 1.5 times interquartile range ROGUE values 45 mixtures cell-type sizes 1:100 to 1:1 varied genes 1% total gene number (n = center line median ROUGE value n = 50 simulations 25th 75th percentiles whiskers 1.5 times interquartile range Pearson correlations randomly down-sampled datasets = 50 entire datasets (2000 cells) center line median correlation value hinges 25th 75th percentiles whiskers 1.5 times interquartile range Sequencing depth distribution two replicates replicate 2 sequencing depth ten times replicate 1. S–E plot mixture replicates 1 2 shown d.ROGUE values n = 100 mixtures versus silhouette values two replicates high silhouette value indicates difference sequencing depth replicates S–E plots ROGUE values 10 cell populations PBMC dataset24. Purity assessment six human T-cell populations Purity evaluation lung-cancer infiltrating DCs point patient center line median ROUGE value lower upper hinges represent 25th 75th percentiles whiskers 1.5 times interquartile range number cells challenge S ROGUE calculation down sampling analysis S cell numbers Pearson correlations S down-sampled datasets datasets similarity values >0.99 S ROGUE calculation affected variation cell number (Fig. 2c).Sequencing depth across cells orders contributes technical confounder scRNA-seq data ROUGE index assess purity single-cell population technical effect simulated increasing molecular counts (sequencing second mock replicate fold change gene expression means 2 to 100 (Fig. 2d two replicates expected pure cell population used silhouette measure replicate-to-replicate differences results revealed ROGUE values ~0.99–1 population two replicates silhouette values 0.25 to 0.75 (Figf Supplementary Fig. 14a). ROGUE purity single-cell population accounts variation sequencing.ROGUE assesses purity cell considered External RNA Controls Consortium (ERCC) dataset24 controlled experiment technical variability 1015 droplets received same ratio ERCC synthetic spike-in RNAs no varied RNAs ideal pure cell population found one RNA with significant ds. dataset achieved ROGUE value ~1 confirming purity investigated fresh peripheral blood mononuclear cells) from single healthy provided multiple cell types purified by FACS suitable resource purity assessment cell types Fig. 2g CD4/CD8 naïve T CD4 memory T cells homogeneous detected high ROGUE values (0.94–1) CD14 monocytes CD34+ cells diverse received low ROGUE values (~0.8 confirming heterogeneity ROGUE index on pure subtypes unsupervised clustering considered six T-cell subtypes from human colorectal cancer5 generated Smart-seq2 protocol pure subtypes achieved high ROGUE values >0.9 (Fig. versus 0.78 complete dataexamined four dendritic cell (DC) subsets from human lung sequenced with inDrop tumor-infiltrating DC2 cells heterogeneous deviated from cell types DC1 LAMP3+ DC pDC (Fig. results illustrate ROGUE effective cell population purity without affected technical characteristics.ROGUE-guided analysis enhances cell-type evaluated potential ROGUE guide clustering analysis with silhouette intercluster dissimilarity within-cluster dissimilarity simulated scRNA-seq dataset three cell types A B C similar 1% varied genes clustered dataset into 2 3 4 5 subpopulations resolution parameter Seurat23 evaluated results silhouette ROGUE values Proper clustering should three subpopulations one each cell type silhouette maximum value when cell-type A co-clustered with B two clusters poorly interpretable for cluster purity ROGUE saturation when three clusters Repeating simulation with varied differences cell-type A B C yielded equivalent performance performance observed when different values reference factor K usedROGUE purity quantification cluster independent normalization reduction clustering splitting merging clusters unsupervised clustering analyses.Fig. 3ROGUE enhances single-cell clustering cell-type identification t-SNE plots simulated dataset three cell types silhouette values average ROGUE values 2 3 4 5 clusters UMAP plots lung-cancer-associated fibroblasts color-coded clusters original paper Supplementary Fig. re-clustered labels ROGUE values clusters before after re-clustering point represents patient center line median ROUGE value lower upper hinges represent 25th 75th percentiles whiskers 1.5 times interquartile range UMAP plot expression levels MYH11 MEF2C Differences hallmark pathway activities scored GSVA ROGUE clustering examined dataset cancer-associated fibroblasts lung tumors heterogeneous population tumor-supportive seven fibroblast clusters low ROGUE values (Fig. 3d Fig. performed re-clustering analysis heterogeneity identified 11 clusters higher average ROGUE value(myofibroblastic inflammatory antigen-presenting (apCAFs high expression CD74 MHC class-II genes Fig. 17b). apCAFs discovered fibroblast subtype mouse pancreatic ductal adenocarcinoma barely detectable in human PDAC apCAFs lung cancer differences cancer types myCAFs (AF_C02_COL4A1 ROGUE = 0.81) segregated into three subpopulations BF_C01_RGS5 (ROGUE = 0.84), BF_C02_ACTA2 0.87) BF_C03_GPX3 = 0.94) signature genes AF_C02_COL4A1 specific to subpopulations MEF2C BF_C01_RGS5 MYH11 BF_C02_ACTA2 NOTCH signaling activated in BF_C01_RGS5 AF_C02_COL4A131 increase ROGUE index BF_C00_AOL10A1 BF_C04_COL1A2 BF_C05_PLA2G2A low ROGUE values further investigation ROGUE-guided analysis discovered novel cell subtypes enabled detection true signals in subpopulationsROGUE-guided analysis B cell subtypesB key tumor microenvironment unclear functions antitumor investigated liver- lung-tumor-infiltrating B low ROGUE values applied clustering analysis ROGUE cells discover pure subtypes seven clusters identified specific marker genes (Fig. 4b–d). first B-cell subset B_C0_JUNB expressed signature genes JUNB FOS activated B second subset B_C1_TXNIP high expression glycolysis pathway genes metabolic differences ACTB gene antigen presenting expressed third subset (B_C2_ACTB). analysis strong antigen processing presentation signal fourth cluster B_C3_FCER2 high expression HVCN1 genes B-cell receptor signaling pathway pre-activated B cells37 fifth cluster B_C4_MX1 interferon-induced B expressed high levels MX1 IFI6 IFI44L sixth cluster B_C5_CD3D expressed markers T- B-cell lineages remaining cells seventh cluster B_C6_LRMP high expression LRMP RGS13 germinal center B cells40-guided analysis B-cell subtypes S–E plots ROGUE values liver lung-tumor-infiltrating B cells UMAP plots 4291 B cells color-coded clusters tissues Gene expression heatmap seven B-cell clusters Rows genes columns clusters ROGUE values seven B-cell subtypes point represents patient center line median ROUGE value lower upper hinges 25th 75th percentiles whiskers 1.5 interquartile range Tissue preference B-cell subtype cancer RO/E27 ratio observed expected cell numbers chi-square test average fractions B_C02_ACTB B_C04_MX1 patient error bars < 0.05 < 0.005 Kaplan–Meier curves TCGA LUAD LIHC patients grouped 13 markers B_C02_ACTB DEs/doublets-like germinal center B cells low ROGUE values limited cells permit clustering germinal center B cells high expression proliferating marker genes MKI67 STMN1 explaining heterogeneity ROGUE values >0.92 remaining five clusters homogeneous B-cell subtypesratio observed expected cell numbers chi-square test B_C02_ACTB B_C04_MX1 contained cells tumor RO/E values >1 (Fig. analyses patient confirmed trend (Fig. TCGA lung adenocarcinoma patients higher expression genes B_C02_ACTB showed worse survival (Fig. survival difference observed TCGA hepatocellular carcinoma) cohort dataset. clinical implication deserves further study roles B_C02_ACTB cells tumor microenvironment identifying pure subtypes ROGUE-guided analysis deeper understanding cell state behavior.Application brain data batch effect application ROGUE brain transcriptome dataset2 high heterogeneity identified seven cell types oligodendrocyte neuron cell types low ROGUE values <0.8 versus ~0.9–1 remaining five (Fig. applied clustering ROGUE oligodendrocyte identified ten refined cell subtypes specific marker genes (Fig. 5b Except cluster 6 found ROGUE values ~0.9–1 other nine clusters suggesting purity (Fig compared pathway activities found diversity cluster 5 showed strong axon guidance signalingneurotrophin activated cluster 1 (Fig. illustrates ROGUE uncovering subpopulations. application ROGUE brain data batch effect evaluation ROGUE values seven brain cell types each point sample center line median ROUGE value lower upper hinges represent 25th 75th percentiles whiskers 1.5 times interquartile range UMAP plot ten clusters oligodendrocytes (n = 3401) color-coded heatmap cell-type-specific genes ten oligodendrocyte clusters ROGUE values clusters Each point represents sample center line median ROUGE value lower upper hinges 25th 75th percentiles whiskers 1.5 times interquartile range Enriched pathways cluster 5 1 ROGUE values batch 1 2 cell population equalized cells down-sampling center line median ROGUE value lower upper hinges 25th 75th percentiles whiskers denote 1.5 times interquartile range *p < 0.05 **p < 0.005 test ROGUE values individual-specific cell populations populations cells control group Subsampling equalize cells center line median ROUGE valuelower upper hinges represent 25th 75th percentiles whiskers denote 1.5 times interquartile range. *p < 0.05 **p < 0.005 Student’s t test ROGUE impact batch effect studied dataset human PBMCs multiple cell Cells split into two interferon-beta (IFN-β)-stimulated group culture-matched control group two batches applied ROGUE purity each cell type individual bathes cell population (batch 1 batch 2) ROGUE detected purity reduction in group (Fig. cells collected from eight unrelated individuals tested ROGUE variability batch effect among patients used cells control group IFN-β perturbation aggregated cell populations received lower ROGUE values patient-specific populations (Fig. 5h). ROGUE offers reasonable method for estimating impact batch effect.DiscussionPurity assessment of cell clusters paramount interpretation scRNA-seq data pertinent rare subtle cell subtypes uncovered S–E model variable genes high sensitivity precision applied to clustering pseudotime analyses statistic ROGUE quantify purity of single-cell populationsentropy-based measure ROGUE applicable for datasets different platforms protocols operators purity of-cell populations cell-to-cell variation ROGUE purity four DC subtypes human lung tumors DC2 heterogeneous consistent with previous heterogeneous populations different properties roles in cancer microenvironment assessed with ROGUE future studies could focus deepen understanding cellular origins cancer ROGUE addresses need unsupervised single-cell data analyses quality of clusters unsupervised clustering under- over-clustering universal clustering quality quantifying cluster purity with ROGUE low-purity clusters analysis pure subtypes Improving purity credibility of cell types challenge single-cell sequencing ROGUE could potential standard for quality cell clusters ROGUE analysis on fibroblasts identified novel subpopulation in lung cancer apCAFs expressed CD74 MHC class-II genes strong antigen-presenting signal cells deactivate CD4 T cells decrease CD8+ to Treg ratio in mouse PDAC33 unclear role in lung-cancer microenvironment further investigation ROUGE to B-cell analysis found pure cluster B_C02_ACTB high expression of genes antigen processing presentationCells from cluster enriched in tumors associated with poor outcomes in lung liver cancer hypothesize cells may contribute to immune suppression cancer curtail antitumor immunity further studies required approaches for discovering pure subtypes to other scRNA-seq datasets determining purity cell recommend ROGUE value 0.9 threshold infiltrating cells varied genes constrained for low-quality threshold determined global ROGUE values ROGUE efficient effective anticipate additional extensions enhanced performance assessing purity integrated cell populations protocols ROGUE metric robust measure for cluster purity technical confounders expect applicable to scRNA-seq datasets strategy improve rigor quality of unsupervised single-cell data analysisMethodsExpression entropy droplet datasets observed UMI count modeled NB random variable Poisson–Gamma mixture411[12pt{minimal}\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}{array}}X_{ij}\mathrm{Poisson}}\lambda _{ij}}{ij{Gamma}}\alpha _{ij\beta _{ij}}{array}{document}Xij~Poisson(sjλij)λij~Gamma(αij λij expression value observed UMI count Xij gene i cell j sj size normalization factor cell j αij βij shape parameter rate parameter shape parameter α constant across cells genes rate parameter β constant gene i αij βij expressed as α and βidistributions recognized as[12pt]{minimal{amsmath{wasysym\oddsidemargin}{-69pt}{document}$$\lambda _i\sim\mathrm{Gamma}}\left\alpha \beta _i\end{document}λi~Gammaα,βi[12pt]{minimal}{amsmath{wasysym{upgreek}{\oddsidemargin}{-69pt}{document}$$X_{ij}\sim\mathrm{Poisson}}\lambda _i\end{document}Xij~Poisson(sjλi).[12pt]{minimal}{amsmath}{wasysym{amsfonts{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$X_{ij} =\frac{{X_{ij}}}{{s_j}}\end{document}Xij′=Xijsj normalized expression of gene i in cell j use[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin{-69pt}{document}\Bbb E}\left( {X_i^\prime }\right\end{document}EXi′[12pt]{minimal{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin{-69pt}$X_i^\prime\end{document}Xi′ normalized expression gene i[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}\setlength\oddsidemargin{-69pt}{document}{\Bbb E}\left( {X_i^\prime }\right\end{document}EXi′ expectation across cells moment estimation of λi.Gamma distribution rate parameter calculated maximum likelihood estimation3\documentclass[12pt]{minimal{amsmath\oddsidemargin-69pt}{document}\beta _i = \frac{\alpha\lambda\frac{\alpha\Bbb E}\left( {X_i^\prime \right\end{document}βi=αλi=αEXi′ disorder randomness gene expression differential entropy\documentclass[12pt]{minimal}{amsmath\oddsidemargin-69pt}{document}$H\left( X \right) = - \int{ - \infty + \infty }\left( x \right)\mathrm{ln}}\left( x \right)dx\end{document}HX=−∫−∞+∞px⋅lnpxdx X continuous random variable p(x) probability density function Differential entropy extension Shannon entropy average surprisal continuous probability distribution performance supervised gene selection method E-test44.gamma distributed random variable λi differential entropy computed[12pt]\usepackage{amsmath-69pt$S_i =\alpha -\mathrm{ln}}\beta _i +\mathrm{ln}}\Gamma\left\alpha \right) + {1 - \alpha \right) =\mathrm{ln}}\alpha\beta _i}} + a =\mathrm{ln}}\Bbb\left(_i\prime \right) + a{document}Si=α−lnβi+lnΓα+1−α⋅φα=lnαβi+a=lnEXi′+a φ digamma function[12pt]{minimal\usepackage{amsmath\oddsidemargin{-69pt}}$$a = \alpha -\mathrm{ln}}\alpha +\mathrm{ln}}\Gamma \left( \alpha \right) +\left( {1 - \alpha \right)\upvarphi\left(alpha \right)$\end{document}a=α−lnα+lnΓα+1−α⋅φα constant methods Scnorm45 scran46 BASiCS47 calculate size factors considered library size normalization cell total UMI counts divided by mean total UMI counts across cells41 expectation library size factor across cells equal to 1. Eq. (2) gene expression library size independent random variables42 gene i\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}\Bbb E}\left( {X_i \right) = s\right 1\end{document}EXi=EXi′×s=EXi′×Es=EXi′×1=EXi′ Xi observed expression gene i s library size cells.each cell type differential entropy λi computed[12pt]{minimal{amsmath{wasysym{upgreek}\oddsidemargin-69pt}{document}$S_i =\mathrm{ln}} E}\left {X_i}\right) + a\end{document}Si=lnEXi+a formulate null hypothesis one Poisson–Gamma component each gene population (H0) differential entropy calculated Eq. (7) each cell represents cluster use Xij moment estimation mean expression clusterdefine entropy reduction gene i across n cells\documentclass[12pt{minimal{amsmath{upgreek\oddsidemargin-69pt}{document}_i =\mathrm{differential}}{entropy}}{average}}{actual}}{differential}}{entropy}}\\{ln}}\Bbb E\left {X_i}\right) -\frac{{\mathop\sum\nolimits_{j = 1}^n\mathrm{ln}}X_{ij}}\end{document}dsi=differentialentropyunderH0−averageactualdifferentialentropy=lnEXi−∑j=1n(lnXij)n captures disorder randomness gene expression44genes under H0 (non-variable major proportion relationship between[12pt{minimal{amsmath{wasysym\oddsidemargin-69pt}{document\mathrm{ln}}_i{document}lnEXi average differential entropy calculate residual as dsi improve performance (Fig. 1b, c). significance of ds estimated normal distribution adjusted Benjamini–Hochberg method extended procedure to full-length datasets approach outperformed gene selection methods (Fig. 1f Supplementary Fig. 4) simulated droplet datasets with NB distributiongene abundance levels E sampled from log-normal distribution[12pt]{minimal{amsmath{wasysym\oddsidemargin-69pt}}$$ln\left E\right\sim {N} ^2{document}lnE~Nμ,σ2 parameters μ = 0 σ = 2. transcripts each gene drawn[12pt]{minimal}{amsmath{wasysym\oddsidemargin}{-69pt}}$$N_{ij}\mathrm{NB}}\left {E_i,r} \end{document}Nij~NBEi,r simulated dataset dispersion parameter r = α)48 fixed value 5 to 20 (Supplementary Fig. 1) simulated full-transcript datasets with ZINB distributiondropout rates gene modeled sigmoid function49\documentclass[12pt{minimal{amsmath{wasysym\oddsidemargin-69pt\mathrm{sigm}}\left\gamma _0 _1E_i~sigm−γ0+γ1Ei parameters γ0 = −1.5 γ1 = 1/median(E). simulated scRNA-seq dataset 20,000 genes 2000 cells.Differentially expressed genes added fraction cells (1–50% changes sampled log-normal distribution (μ = 0 σ = 2) Genes >1.5-fold decrease increase mean expression ground truth DE genes.Feature selection HVG method11 identifies variable genes coefficient variation squared local regression trend implemented BrenneckeGetVariableGenes function M3Drop12 package Gini index model GiniClust13 gene informative if Gini higher than expected maximum observed expression copied source code original GiniClust defined Gini_fun function scripts select genesM3Drop uses dropout rates variable gene selection implemented M3DropFeatureSelection function SCTransform selects genes with Pearson residuals from negative binominal regression implemented SCTransform function Seurat package implemented Fano factor method GiniClust2_Fano_clustering.R feature selection step RaceID319 selects genes with second-order polynomial fit between expression variance log-transformed mean implemented fitBackVar function RaceID3.Datasets clustering feature selection methods unsupervised clustering considered five scRNA-seq datasets first five cell lines (A549 sequenced with 10X Genomics protocol 3918 cells second dataset three cell lines (H1975 H2228 HCC827) sequenced-seq2 protocol third multiple FACS-purified cell populations sequenced with 10X Genomics protocol used CD19 B cells CD4 naïve T cells CD56 NK cells CD14 monocytes distinguishable fourth dataset cells human pancreatic islet generated Smart-seq pancreatic cell types alpha beta delta gamma cells well-characterized used for benchmarkingremaining dataset multiple immune cell types9 cells sequenced by Smart-seq2 protocol cell labels assigned unsupervised clustering cross-validation experiments revealed major cell types (macrophages DCs lymphocytes exhausted CD8 T cells distinguishable Fig. 6c). consider dataset for benchmarking.Cross-validation experiments gene reproducibilityTo performance S–E model datasets performed cross-validation experiments procedure scmap randomly selected 70% cells reference set identified informative genes different selection methods trained RF classifier50 genes labels defined unsupervised clustering remaining 30% cells query set cell types predicted with trained classifier classification accuracy quantified with accuracy score50 similarity between predicted cell types original cell types repeated procedure = 50 times each datasetcalculated reproducibility intersecting variable genes[12pt{amsmath{wasysym{upgreek\oddsidemargin-69pt\mathrm{Reproducibility}} ={Geneset}} - 1}{m - n} - 2 - n}Reproducibility=Genesetreplicate−1m−n∩Genesetreplicate−2m−nn m adapted gene selection method n top-ranked variable genes.Rare cell-type simulated synthetic scRNA-seq data GiniClust2 specifying two large 1000 cell clusters three rare clusters 10, 20 30 cells test performance applied S–E model raw count data genes performed follow-up clustering standard procedure Seurat R scripts RaceID3 GiniClust2 accessed https://github/dgrun/RaceID3_StemID2_package/dtsoucas/GiniClust2ROGUE calculationBy S–E model scRNA-seq data introduce statistic ROGUE purity cell population[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{mathrsfs{upgreek}\oddsidemargin{-69pt}{document\mathrm{ROGUE}} = 1 - ds + K}}}ROGUE=1−∑sigds∑sigds+K parameter K constrain ROGUE value between 0 and 1 reference factor purity evaluation reference dataset with maximum summarization significant ds set value K to one-half of maximum ROGUE value 0.5 when ds equivalent to one-half maximum cell population with no significant ds for all genes ROGUE value 1 population large summarization significant ds small purity score Tabula Muris plausible reference dataset cells from 20 organs heterogeneous population sequenced with 10X Genomics and Smart-seq2 protocols2. technical variation PCR value ds calculated summarization significant ds Tabula Muris for 10X Genomics and Smart-seq2 datasets Fig.set default value K to one-half summarization 45 for droplet-based 500 full-length-based data K value determined specifying different reference dataset scRNA-seq data analyses careful using default K value datasets different species recommend determine K value specifying heterogeneous dataset with DetermineK function ROGUE package.Silhouette coefficientTo assess differences simulated replicates separation cell clusters calculated silhouette width7 ratio within-cluster to intercluster dissimilarity a(i) average dissimilarity cell i to other cells cluster A b(i) average dissimilarity cell i to all data points neighboring cluster dissimilarity minimal silhouette width for cell i defined as\documentclass[12pt]{minimal}\usepackage{amsmath\left\right =\right - a(i}si=bi−a(i)maxai,bi high s(i) value suggests cell i well assigned to own cluster poorly assigned to neighboring clustersSequencing depth across cells technical confounder scRNA-seq data analysis ROGUE robust generated simulated populations two replicates differences sequencing depth (Fig. 4d Supplementary Fig.7a). each simulation varied sequencing depth two replicates[12pt]{minimal}{amsmath{wasysym-69pt}{document}$\mu\mathrm{replicate}} - 2,i} ={replicate}} - 1,i} \delta\left {1 ,n}\right\end{document}μreplicate−2,i=μreplicate−1,i⋅δ,i∈1,...,n n number genes μ mean expression level[12pt]{minimal}{amsmath{upgreek\oddsidemargin{-69pt}{document}$\delta\left {2,,5,10,20,50,70,,100}\right\end{document}δ∈2,5,10,20,50,70,100.Generation simulated cell potential ROGUE guide single-cell clustering used NB model simulate scRNA-seq datasets three cell types A B C (1000 cells × 10,000 genes A B similar subtypesscenarios Fig. 3a 15a introduced 500 1000 800 varied genes cell A B/C changes log-normal distribution (μ 0 σ = 2) simulated 100 120 variable genes between B C fold changes log-normal distribution μ = 0 σ = 1. results visualized t-distributed stochastic neighbor embedding-SNE) R package fibroblast B-cell ROGUE-guided analysis subpopulations biological signals re-clustering analysis fibroblast-cell filtered low-quality cells <600 expressed genes over 25,000 below 600 UMIs 4291 B cells 1465 fibroblasts remained applied S–E model raw count data informative genes normalized gene expression matrices NB regression Seurat23 top 3000 genes maximal ds used PCA analysis batch effects batch correction BBKNN51 first 50 PCs clustering approach cell cluster identified principle components yielded 11 fibroblast subtypes 7 B-cell subtypes Figs. 3d 4b visualized 2D projection UMAP53 default parameters purity score cluster calculated rogue function R package calculation based raw count data independent normalization reduction clusteringPathway TCGA data pathway signals fibroblast subtypes performed pathway analyses hallmark pathways molecular signature database54 GSVA55 TCGA LUAD LIHC data prognostic effect 13 signature genes Table 5) B_C2_ACTB normalized abundance 13 genes expression MS4A1 gene performed statistical analyses GEPIA256 default parameters.Reporting Nature Research Reporting Summary.Supplementary information Peer Review File Reporting Summary | 50.4 | 1.137954 |
10.1038/s41467-020-20466-9 | PMC7794343 | Dual oxidases (DUOXs), assembled from the catalytic DUOX and the auxiliary DUOXA subunits, produce hydrogen peroxide by transferring electrons from intracellular NADPH to extracellular oxygen in a calcium-activated manner. Here authors report the cryo-EM structures of human DUOX1-DUOXA1 complex in both high-calcium and low-calcium states. | Dual oxidases (DUOXs) produce hydrogen peroxide by transferring electrons from intracellular NADPH to extracellular oxygen. They are involved in many crucial biological processes and human diseases, especially in thyroid diseases. DUOXs are protein complexes co-assembled from the catalytic DUOX subunits and the auxiliary DUOXA subunits and their activities are regulated by intracellular calcium concentrations. Here, we report the cryo-EM structures of human DUOX1-DUOXA1 complex in both high-calcium and low-calcium states. These structures reveal the DUOX1 complex is a symmetric 2:2 hetero-tetramer stabilized by extensive inter-subunit interactions. Substrate NADPH and cofactor FAD are sandwiched between transmembrane domain and the cytosolic dehydrogenase domain of DUOX. In the presence of calcium ions, intracellular EF-hand modules might enhance the catalytic activity of DUOX by stabilizing the dehydrogenase domain in a conformation that allows electron transfer. | IntroductionReactive oxygen species (ROS) are oxygen-containing chemical species that are highly reactive, such as hydrogen peroxide and superoxide anion1. They participate in many physiological processes and are implicated in several pathological conditions1. ROS can be generated by a class of dedicated enzymes called NADPH oxidase (NOX) in a highly regulated manner. These enzymes are multi-pass transmembrane proteins that catalyze the reduction of extracellular or luminal oxygen by intracellular NADPH to generate superoxide anion or hydrogen peroxide. NOX proteins are involved in many biological processes, including host defense, differentiation, development, cell growth and survival, cytoskeletal reorganization, and modification of the extracellular matrix2.Comprising the human NOX protein family are NOX1–5 and DUOX1–2 (ref. 2). NOX2 protein catalyzes the production of superoxide anion during phagocytosis in neutrophils and is essential for host defense3. DUOX1–2 proteins are highly expressed in thyroid tissue and they catalyze the production of hydrogen peroxide, which is important for the biosynthesis of thyroid hormones4. The function of DUOX protein requires physical interactions with an auxiliary protein called dual oxidase maturation factor (DUOXA)5. DUOXA promotes the maturation and proper plasma membrane localization of DUOX5. DUOX protein is encoded by two homologous genes in human, namely DUOX1 and DUOX2. Similarly, DUOXA protein is encoded by DUOXA1 and DUOXA2. Loss-of-function mutations of DUOX2 or DUOXA2 in human cause congenital hypothyroidism6. Because of the important role of DUOX in thyroid tissue, they are also named thyroid oxidase4.NOX family proteins share a common catalytic core, formed by a heme-coordinating transmembrane domain (TMD) and a cytosolic dehydrogenase (DH) domain7. The DH domain binds intracellular substrate NADPH and cofactor FAD, and shares sequence homology to the ferredoxin-NADP + reductase (FNR), which is composed of two subdomains8. In addition to the shared TMD-DH catalytic core of NOX, the functional DUOX protein has an additional large N-terminal extracellular peroxidase homology domain (PHD) and a long intracellular loop 0 containing two EF-hand domains, and it requires an auxiliary DUOXA protein for proper function. The activity of DUOX is regulated by intracellular calcium concentration4. Prior to our studies, the structures of NOX family members are only available in the form of isolated domains, including the DH domain (PDB ID: 5O0X)9 and TMD (PDB ID: 5O0T)9 of NOX5 from the algea Cylindrospermum stagnale (csNOX5) and a subdomain of human NOX2 DH domain (PDB ID: 3A1F). Despite the functional importance of DUOX and other NOX family members, their structures in the context of full-length functional protein complex are still unknown. Several open questions for DUOX remain elusive: How is the DH domain engaged with TMD to perform the catalytic redox reaction? How does DUOXA protein interact and co-assemble with DUOX? How is the activity of DUOX regulated by intracellular calcium? To answer these fundamental questions, we sought to characterize DUOX–DUOXA protein complex both structurally and functionally. Here, we present the cryo-EM structures of human DUOX1–DUOXA1 (hDUOX1–hDUOXA1) complex in both high-calcium and low-calcium states, providing insights into the structure and mechanism of calcium activation for DUOX.ResultsStructure determinationTo express the hDUOX1 protein, we constructed the matured hDUOX1 protein (20–1551) in frame with N-terminal GFP tag guided by a rat FSHβ signal peptide for efficient secretion10. The molecular weight of GFP-tagged hDUOX1 is 219 kDa. To monitor the formation of DUOX1–DUOXA1 complex, we fused the hDUOXA1 protein with a C-terminal MBP-mScarlet tag to increase its molecular weight to 106 kDa. Fluorescence size-exclusion chromatography (FSEC) showed the co-expression of hDUOXA1 effectively shifted the peak of hDUOX1 toward higher molecular weight, suggesting the formation of a stable hDUOX1–hDUOXA1 hetero-complex (Fig. S1a, b). The peak positions indicated hDUOX1 migrated as a monomer, while hDUOX1–hDUOXA1 complex migrated as a heterotetramer (Fig. S1b). Moreover, we found the co-expression of DUOX1 and DUOXA1 resulted in cell membranes that showed robust calcium-activated, NADPH-dependent hydrogen peroxide production detected by the Amplex Red assay11 (Fig. 1a–c). In the low-calcium condition, DUOX1–DUOXA1 complex showed low basal activity (Fig. 1b, c). Addition of calcium not only reduced the Km, but also increased the Kcat of DUOX1 complex, leading to the overall enhancement of enzymatic activity (Fig. 1b,c).Fig. 1Structure of human DUOX1–DUOXA1 complex in the high-calcium state.a Schematic of the DUOX enzymatic assay. In the presence of H2O2 (produced by DUOX), horseradish peroxidase (HRP) converts nonfluorescent Amplex Red to fluorescent resorufin, which is measurable and proportional to H2O2. b Calcium-dependent activation of hDUOX1–hDUOXA1 complex. Data are shown as means ± standard deviations, n = 3 biologically independent samples. Source data are provided as a Source data file. c Steady state enzyme activity of hDUOX1–hDUOXA1 complex as the function of NADPH concentration in the presence or absence of calcium. Data were fit to the Michaelis–Menten equation to obtain the Km and Kcat value. Data are shown as means ± standard deviations, n = 3 biologically independent samples. Source data are provided as a Source data file. d Side view of the cryo-EM map of hDUOX1–hDUOXA1 complex in the high-calcium state. The approximate boundaries of phospholipid bilayer are indicated as gray thick lines. One protomer of hDUOX1 and hDUOXA1 complex is colored as blue and green, the other one is colored as yellow and red, respectively. e A 90° rotated top view compared to d. f A 180° rotated bottom view compared to e. g Top view of the cross-section of the transmembrane layer at the position indicated as a dashed line in d. The large cavity in the transmembrane layer is indicated by dashed oval. For clarity, the cryo-EM map was low-pass filtered to 6 Å. h Topology of hDUOX1 and hDUOXA1 subunits. Transmembrane helices are shown as cylinders, unmodeled disordered regions are shown as dashed lines. The phospholipid bilayer is shown as gray layers. PHD peroxidase homology domain of hDUOX1, PHLD pleckstrin homology-like domain of hDUOX1, EF EF-hand calcium-binding module of hDUOX1, DH dehydrogenase domain of hDUOX1, CLD claudin-like domain of hDUOXA1. i Structure of one protomer of hDUOX1 and hDUOXA1 complex in cartoon representation. The colors of each individual domain are the same as in g. The approximate boundaries of phospholipid bilayer are indicated as gray thick lines. Sugar moieties, hemes, FAD, and NADPH are shown as black, yellow, pink, and green sticks, respectively.We solubilized and reconstituted hDUOX1–hDUOXA1 complex into peptidisc using NSPr12. Purified peptidisc sample showed a homogenous peak on SEC (Fig. S1c) and two major protein bands on SDS–PAGE, both of which could be trimmed upon PNGase F treatment (Fig. S1d), suggesting both of hDUOX1 and hDUOXA1 were modified by N-linked glycosylation. UV–vis spectrum showed the peptidisc sample has characteristic Soret band with peak at 415 nm (Fig. S1e), indicating proper Fe (III) heme incorporation. Moreover, the highly purified peptidisc sample recapitulated the calcium-activated NADPH-dependent hydrogen peroxide production observed on membrane (Fig. S1f), confirming that the calcium-dependent activation is a built-in mechanism of hDUOX1–hDUOXA1 protein complex. However, we found the maximum activity of purified peptidisc sample was lower than the activity measured using crude cell membrane (Fig. 1b and Fig. S1f), suggesting either membrane bilayer or endogenous lipids might play a role on DUOX activity. We prepared cryo-EM grids using the peptidisc sample, either in the presence of 2.5 mM ethylene glycol tetraacetic acid (EGTA; low calcium) or 0.5 mM free calcium (high calcium). Both samples contained 0.1 mM FAD as cofactor and 0.5 mM NADPH as substrate.Single particle cryo-EM analysis showed the purified protein was homogeneous and showed twofold symmetry (Figs. S2–S4). The overall resolution of cryo-EM maps in the low-calcium and high-calcium states reached 2.7 and 2.6 Å, respectively (Table S1). The extracellular domains and TMD showed better local resolution than the cytosolic domains, suggesting the higher mobility of the cytosolic domains (Figs. S2g and S4g). To further improve the map quality of cytosolic domains, we exploited symmetry expansion13 and multibody refinement14 by dividing one protomer into the large body (the extracellular domain and TMD) and the small body (the cytosolic domains; Figs. S2c and S4c). The final resolutions of cytosolic domain reached 3.4 and 3.2 Å for the low-calcium and high-calcium states, respectively (Figs. S2–4 and Table S1). The high map quality and available homology structures allowed us to build the order regions of the complex, which encompassed 88% of DUOX1 and 79% of DUOXA1 (Figs. S5–8 and Table S1). In the following text, we will focus on the high-calcium state structure unless noted otherwise, because of its higher resolution.The architecture of hDUOX1–hDUOXA1 protein complexhDUOX1 subunits and hDUOXA1 subunits co-assemble into a 2:2 heterotetrameric protein complex with molecular weight ~457 kDa. The complex encompasses 140 Å × 105 Å × 160 Å 3D space and has an overall twofold rotational symmetry (Fig. 1d–f). Vertically, the complex can be divided into three layers: the extracellular layer, the transmembrane layer, and the cytosolic layer (Fig. 1d). In the extracellular layer, the two large N-terminal PHD domains of hDUOX1 pack against each other diagonally and are buttressed by the extracellular domain of DUOXA1 from beneath (Fig. 1d–f). The transmembrane layer is formed by 24 transmembrane helices and harbors the heme-binding sites that provide the electron transfer pathway across the membrane (Fig. 1g). At the center of the transmembrane layer, there is a large cavity without discernable protein densities. The interior surface of this cavity is highly hydrophobic (Fig. S3i) and there are several lipid molecules bound on this surface (Fig. S3i), suggesting this cavity is probably filled by phospholipids on the cell membrane. The cytosolic layer is comprised of the catalytic DH domain and regulatory domains for intracellular calcium sensing (Fig. 1f, i).Structure of the catalytic hDUOX1 subunithDUOX1 is the catalytic subunit of the complex (Fig. 2). On the extracellular side of hDUOX1 resides the large N-terminal PHD domain which shares sequence homology with several peroxidases, such as peroxidase A from Dictyostelium discoideum (DdPoxA, PDB ID: 6ERC)15 (Fig. S9a). Functional peroxidases utilize histidine-coordinated heme as the cofactor for catalysis. However, key residues for heme binding, such as the heme ligand histidine, are missing in the PHD of hDUOX1. Indeed, we did not observe any heme density in the structure of hDUOX1 PHD, suggesting PHD is probably not enzymatic functional in term of peroxidase activity. Close inspection of the map reveals two putative cation densities in PHD. One cation (cation binding site 1, CBS1) is coordinated by the side chains of D397 and T332, and the main chain carbonyl groups of V399, T332, and R395 (Fig. S9b). The second cation (CBS2) is coordinated by the side chain of D109, D174, S176, and T170, and the carbonyl groups of T170 and W172 (Fig. S9b). We observed strong densities in these two sites in both low-calcium and high-calcium conditions (Fig. S9b), suggesting the bound cations might be sodium ions which were present in large quantities in our protein sample or calcium ions that bind very tightly. Both CBS1 and CBS2 are evolutionary conserved in DUOX (Fig. S5) and DdPoxA15 (Fig. S9c), indicating their functional importance. Interestingly, we found both CBS1 mutant (D397A + T332A) and CBS2 mutant (D109A + D174A) of DUOX1 failed to co-assemble with DUOXA1 (Fig. S9d). Because CBS1 and CBS2 are away from the subunit interfaces in the DUOX1–DUOXA1 complex, we speculate these mutants probably affect the folding of PHD domain, suggesting the role of CBS1 and CBS2 in protein stability.Fig. 2Structure of hDUOX1 subunit.a Side view of hDUOX1 subunit in the high-calcium state, highlighting the key interfaces (boxed by dashed lines). Each domain is colored as in Fig. 1h. The surface of hDUOX1 is shown in transparency. b The binding site of outer heme in the TMD. Heme is shown as sticks and colored in yellow. Unrelated helices in TMD are omitted for clarity. The putative oxygen-reducing center is indicated by arrow. c The binding site of inner heme in the TMD. d The interface between PHD and TMD boxed in a. Disulfide bond between C118–C1165 is shown as golden sticks. e The interface between PHLD and TMD boxed in a, the hydrogen bonds are indicated with dashed lines. f The interface between PHLD and DH domain. g The interface between EF module and DH domain. h The FAD-binding site located at the interface between TMD and DH domain. Ligands and interacting residues are shown as sticks. i The NADPH-binding site located at the interface between TMD and DH domain.The PHD packs on top of the TMD of DUOX1 through multiple noncovalent interactions (Fig. 2a). Moreover, a disulfide bond between C118 on PHD and C1165 on loop C of TMD further staples the bottom of PHD onto the top of TMD (Fig. 2a, d). In the TMD, hDUOX1 has an extra bent M0 helix at the periphery of M3 and M4, together with the canonical six TM helices of NOX protein family. M1–M6 of hDUOX1 form two heme-binding sites within the TMD. H1144 on M3 and H1238 on M5 coordinate the outer heme (Fig. 2b). H1130 on M3 and H1225 on M5 coordinate the inner heme (Fig. 2c). These four histidines are absolutely conserved in NOX family proteins (Fig. S6). We observed a spherical density surrounded by the invariant R1087 on M2, H1148 on M3, and outer heme-coordinating residue H1144 (Fig. 2b and Fig. S3c). Previous studies showed mutations of the csNOX5 residues corresponding to R1087 and H1148 of hDUOX1 affected the reoxidation of dithionite-reduced TMD by oxygen and this site was proposed to be the oxygen substrate binding site, namely oxygen-reducing center (Fig. S10a, b)9. Our structure observations in hDUOX1 support the hypothesis.Preceding the M1 helix of DUOX1 TMD, an amphipathic preM1 helix floats on the inner leaflet of plasma membrane (Fig. 1h). This helix was previously observed in csNOX5 (ref. 9) and is probably a shared feature of NOX family proteins. Between M0 and preM1 is a long cytosolic fragment loop 0. Cryo-EM maps reveal that the N-terminal of loop 0 is a domain rich of β sheets (Figs. S3g and S10c). Structural search using DALI server16 identified the β sheets-rich domain is a crypto pleckstrin homology-like domain (PHLD) that shares little sequence homology, but high structural similarity to the PH domain proteins (Fig. S10c)17.Following the PHLD, two EF-hand type calcium-binding domains (EF1 and EF2) form a compact helical module that is connected to the PHLD through αC (Fig. S6). Residues predicted to be responsible for calcium binding in EF1 and EF2 are evolutionary conserved in DUOX family proteins (Fig. S6). Although we did not observe the strong densities for small calcium ions due to poor local resolution (Figs. S2–3), the structure of EF-hand module closely resembles the small subunit of calcium-dependent protein phosphatase calcineurin in the calcium-bound state (PDB ID: 4IL1)18 (Fig. S10d), suggesting both EF1 and EF2 are loaded with calcium in the high-calcium state. Based on the homology structure (4IL1), side chains of D828, D830, N832, and E839 and the main chain carbonyl group of Y834 chelate one calcium ion in EF1 (Fig. S10e) and side chains of D864, D866, N868, E875, and the main chain carbonyl group of L870 chelate another calcium ion in EF2 (Fig. S10f). It is reported that mutations of any of these calcium-binding sites abolished calcium activation19 and E879K mutation in hDUOX2 (E875 in hDUOX1) leads to congenital hypothyroidism20, emphasizing their importance in calcium activation.The C-terminal catalytic DH domain is connected to M6 of DUOX1 TMD via a short linker (Fig. 1h, i). The DH domain of hDUOX1 has a canonical DH fold and its structure is similar to csNOX5 (ref. 9; Fig.1i and Fig. S7). We observed strong densities for both FAD cofactor and NADPH substrate, and their binding sites were contributed from not only DH domain, but also TMD (Fig. 2h, i), as described later.Inter-domain interactions in the high-calcium stateIn the high-calcium state, individual domains of DUOX1 in the cytosolic layer are stabilized by multiple inter-domain interactions. The PHLD interacts with adjacent TMD and DH domains (Fig. 2e). The main chain carbonyl group of K653 on PHLD makes hydrogen bond with R1215 on loop D of TMD (Fig. 2e). Side chain of R674 of PHLD interacts with the main chain carbonyl group of E1348 and I1349 on α1 of the DH domain (Fig. 2f). The EF1–EF2 module in the high-calcium state shapes a crevice that embraces α4 and post α4 loop of the DH domain (Fig. 2g and Fig. S10g). The interactions between the EF module and DH are mainly hydrophobic and involve F768, F772, F807, F819, F840, and F847 of the EF module, L1463, M1467, I1470, F1475, V1478, and F1484 of the DH domain (Fig. 2g). In addition, K814 of the EF module makes electrostatic interaction with E1281 on β2 of DH (Fig. 2g). The interactions between the EF module and the DH domain of hDUOX1 mimic the interactions between calcineurin subunit B and A in the calcium-bound state (PDB ID: 4IL1)18 (Fig. S10g, h).The linker between the EF module and preM1 helices binds in a groove on the surface of the DH domain (Fig. 1i). DH docks onto the bottom of TMD via polar interaction between R1270 on M6 and D1367 on β7, and between R1113 on loop B of TMD and N1550 of DH (Fig. 2h, i). It is reported that R1111Q mutation in hDUOX2 (R1113 in hDUOX1) was identified in congenital hypothyroidism patients20, highlighting the importance of this inter-domain interaction. Moreover, both the FAD cofactor and NADPH substrate bind at the interface between DH and TMD. R1214 and R1131 in TMD form electrostatic interaction with phosphate of FAD. D1128 makes hydrogen bonding with ribose of FAD (Fig. 2h). E1039 and N1040 in TMD make hydrogen bonding with adenosine ring of NADPH, and R1036 make cation-π interaction with both adenosine ring and electrostatic interaction with phosphate group of NADPH (Fig. 2i). Notably, R1495, R1424, and R1036 all participate in electrostatic interactions with the phosphate group of NADPH ribose, providing structural mechanism to distinguish NADPH from NADH (Fig. 2i). Through structural comparison, we found the NADPH-binding site in the csNOX5 structure was blocked by the artificially engineered C-terminal insertion, which was introduced into previous crystallization construct9 (Fig. S10i). Moreover, the adenosine group of FAD has a 180° flip compared with structure of the isolated DH of csNOX5 (Fig. S10j). This is probably because D1128 on TMD stabilizes the ribose of FAD in such a conformation to make the connecting adenosine group of FAD in close proximity with inner heme for electron transfer (Fig. 2h). Taken together, the binding of FAD and NADPH at the interface between DH and TMD of hDUOX1 would stabilize the docking of DH onto the bottom of TMD.The putative electron transfer pathwayThe measured edge-to-edge distances between NADPH and FAD, between FAD and inner heme, and between inner heme and outer heme are 8.2, 3.9, and 6.7 Å, respectively (Fig. 3a). Although, it is possible that there are additional protein residues on DUOX1 that rely electrons from NADPH to FAD, such as W378 between two hemes in csNOX5 (ref. 9), the distance between NADPH and FAD is larger than that in the canonical FNR protein, such as 3.2 Å in spinach FNR (sFNR, PDB ID: 1QFZ)21. Through structural comparison, we found the DH domain of DUOX1 shows a relaxed conformation, in which two subdomains are loosely packed, while both the DH of csNOX5 and sFNR show a tense conformation and their two subdomains are tightly packed against each other to bring FAD and NADPH into close proximity for electron transfer (Fig. 3b–e). Therefore, the electron transfer efficiency in the current structure of DUOX1 is not optimal. Because the DUOX1 complex on cell membrane exhibited higher activity (Fig. 1b and Fig. S1f), it is possible that lipids on cell membrane or the bilayer environment could somehow affect the structure of DUOX1 to enhance its electron transfer efficiency.Fig. 3Electron transfer pathway in hDUOX1 subunit in the high-calcium state.a The edge-to-edge distances between NADPH and FAD, FAD and inner heme, and two hemes are shown beside dashes. The ligands are shown as sticks, each domain of hDUOX1 are shown in surface, and colored the same as Fig. 1h. Only one hDUOX subunit is shown for clarity. The putative oxygen-reducing center is boxed by dashed lines. b–e The DH domain of hDUOX1 in a relaxed conformation (b), DH domain of csNOX5 (c), and sFNR (d) in a tense conformation. The ligands are shown as sticks, two subdomains (FAD-binding domain, FBD, and NADPH-binding domain, NBD) of DH are shown as cartoon with surface. Distances between Cα atoms of the Arg (Lys in sFNR) of FBD and the Cys of NBD (shown as spheres) are labeled. e Structural comparison of the hDUOX1 DH domain (cyan) and csNOX5 (purple). FBD is used for structural alignment. f–i The close-up view of the putative oxygen-reducing center. Four predicted tunnels for oxygen substrate entrance and product exit are shown as surface in yellow, green, magenta, and orange, respectively. Residues surrounding the tunnels are shown as sticks. j Calculated radii of tunnels shown in f–i. The putative oxygen-reducing center is used as the starting point for calculation.At the terminus of electron transfer chain near the extracellular side, the initial product of oxygen-reducing reaction is superoxide anion. We probed the possible pathways for oxygen entrance and for superoxide anion exit with CAVER22, using the oxygen-reducing center as the starting point. We located four possible tunnels: tunnel A is formed by M1, M2, M5, and M6 and is capped by loop E on top (Fig. 3f); tunnel B is surrounded by M2, loop A, loop C, and loop E (Fig. 3g); tunnel C is embraced by M3, M4, and loop C (Fig. 3h); and tunnel D is enclosed by M3, M4, loop C, and M0 (Fig. 3i). The bottleneck radii of these tunnels are ~1 Å (Fig. 3j), which may allow the permeation of small oxygen substrate under dynamic motion of DUOX1 protein. Further analysis showed tunnels B–D are all surrounded by hydrophobic residues (Fig. 3g–i), which are unfavorable for superoxide anion permeation. In contrast, tunnel A is gated by hydrophilic R1087 on M2, R1062 on M1, and R1248 and Q1245 on loop E (Fig. 3f). We speculate the highly positively charged constriction of tunnel A would strongly attract the negatively charged superoxide anions, and this might be essential for the dismutation reaction between two superoxide anions to generate uncharged hydrogen peroxide for diffusion. Therefore, manipulations that may alter the constrictions of tunnels A–D would affect superoxide anion intermediate leakage. Indeed, it is reported that mutations on DUOX1 loop A or on DUOXA1 N-terminus peptide (NTP) which interacts with and stabilizes loop A would change the ratio of superoxide anion and hydrogen peroxide produced, probably by affecting the leakage of superoxide anions through these tunnels23–25.Structure of hDUOXA1 and mechanism of complex assemblyDUOXA protein is an essential auxiliary subunit for DUOX enzyme5 (Fig. S11a) and it has an extracellular N-terminus that is important for hydrogen peroxide generation24,25. We observed the NTP of hDUOXA1 extends and packs onto the PHD–TMD junction of the distal hDUOX1 subunit (Fig. 4a–e). Side chains of F8, F10, and Y11 of NTP insert into the hydrophobic groove formed by loop C, loop A, and PHD of hDUOX1 (Fig. 4c). In addition, K15 of DUOXA1 NTP makes electrostatic interactions with D1077 of DUOX1 (Fig. 4d). This agrees with previous data showing DUOXA1 NTP interacts with DUOX1 loop A (ref. 23). hDUOXA1 has five transmembrane helices. Lower part of TM1 interacts with preM1 and M1 of hDUOX1 (Fig. S11a). The remaining four helices and associated extracellular loops share structural similarity with claudin superfamily members, such as claudin-9 (PDB ID: 6OV2)26 (Fig. S11). The extracellular loop between TM2 and TM3 folds into a compact claudin-like domain (CLD) composed of four β strands and two α helices (Figs. S8 and S11). CLD forms extensive interactions with both distal and proximal DUOX1 subunits (Fig. 4a, b), emphasizing its important role in the complex assembly. This agrees with previous studies showing that splicing variants at TM2–TM3 loop have distinct behavior in supporting the activity of DUOX1 (ref. 24). Moreover, we found an ordered N-linked glycosylation decoration on N109 of hDUOXA1 and its branched sugar moieties make extensive polar interactions with both DUOXA1 and DUOX1 subunits (Fig. 4a, b). The PHD of two DUOX1 subunits also interact with each other (Fig. 4a, b). Close to the dyad axis, R50 and R507 on one PHD make polar interactions with E41 and F313 on the opposite PHD (Fig. 4e). We further analyzed the effects of interface mutations on the tetramer assembly, and found mutations of R50E, R507E, and R507A all severely affect tetrameric peak formation on FSEC (Fig. 4f). These structural information and biochemical data revealed the detailed inter-subunit interactions that dictate the heterotetramer assembly.Fig. 4Mechanism of hDUOX1–hDUOXA1 tetramer assembly.a The side view of hDUOX1–hDUOXA1 protein complex shown in surface representation and colored the same as in Fig. 1d. b The open-book view of the inter-subunit interfaces. Residues of hDUOX1 subunits that interact with hDUOXA1 subunit are colored in green. Residues of hDUOXA1 subunit that interact with hDUOX1 subunits are colored in yellow and blue. c The close-up view of the interactions between NTP of hDUOXA1 and hDUOX1 boxed in a. d The close-up view of additional interactions between NTP of hDUOXA1 and hDUOX1 boxed in c. e The top view of interactions between PHD of two opposing hDUOX1 subunits. f Representative FSEC traces of hDOUX1 R50E, R507E, and R507A mutants are compared to that of wild-type (WT) hDOUX1. The peak position of the hDOUX1 peak is denoted by the hollow circles. Asterisks denote the peak position of hDUOX1–hDUOXA1 protein complex.Conformational change of DUOX1 complex upon calcium activationThe consensus map in the low-calcium state showed the cytosolic layer had poor local resolution, which was improved by multibody refinement14 (Fig. S4). Further molecular flexibility analysis14 showed the cytosolic domains (small body) in the low-calcium state were sampling a broad range of orientations relative to the TMD, evidenced by the plateau-shaped distribution on the histogram of the major eigenvector (Fig. S4f). This is in great contrast to the normal distribution in the high-calcium state (Fig. S2f), suggesting the cytosolic layer in the low-calcium state is more flexible. We compared the structures in the low-calcium state and high-calcium state, and found structural changes in the extracellular layer and transmembrane layer are small (Fig. 5a). However, there are large conformational changes of the regulatory PHLD and EF-hand module in the cytosolic layer (Fig. 5a–c and Movie S1). In the absence of calcium, the EF module switches from an extended shape into a more contracted shape (Fig. 5d, e), which reconfigures the interface between the EF module and α4 of the DH domain, resulting in a loosely packed structure (Fig. 5f). In the low-calcium state, EF2 moves away from the DH domain. The Cα atom of A894 on αJ of EF2 has 40 Å displacement (Fig. 5b). PHLD rotates away from the TMD and DH domains, and αA of PHLD has 17.2° outward rotation (Fig. 5c). As a result, several inter-domain interactions observed in the high-calcium state were disrupted and therefore the docking of DH domain onto TMD is weakened by these structural changes, leading to a higher mobility of DH domain (Fig. S4g). We propose the increased mobility of DH domain negatively correlates with the electron transfer efficiency and thus the catalytic activity of DUOX. In addition, because TMD also contributes to FAD and NADPH binding, the increased mobility of the DH domain would result in the reduced affinity of NADPH as well. This is in agreement with the markedly reduced Kcat and moderately increased Km in the low-calcium state, as we observed (Fig. 1c).Fig. 5Conformational change of hDUOX1 complex during calcium activation.a Structural comparison of hDUOX–hDUOXA1 complex between the high-calcium state (colored) and the low-calcium states (gray). Protein is shown as cartoon. Regions with large conformational changes are boxed by dashed lines. b Close-up view of the conformational changes of EF-hand module. Cα atom of A894 on αJ helix is used as marker to measure the movement of EF2. c Close-up view of the conformational change of PHLD. The angle between αA helices in the high-calcium and low-calcium states was measured. d, e Conformational differences of EF-hand module between the high-calcium state and the low-calcium state. f Reconfiguration of the interface between EF-hand module and α4 helix of DH domain. Arrows denote movements from high-calcium state into the low-calcium state.During the preparation of this manuscript, another group reported the structures of mouse DUOX1–DUOXA1 complex27. Interestingly, they found mouse DUOX1 complex exists in both heterodimeric and heterotetrameric form, and they proposed the activity of DUOX1 complex is regulated by dimer–tetramer assembly27. This is in contrast to our observation that majority of hDUOX1 complex is in a homogenous tetrameric form (Fig. S1c). Whether this difference is due to different protein preparation procedure or different species (mouse vs human) remain elusive. Moreover, the intracellular PHLD and EF domains were not resolved in mouse DUOX1 complex structure because of insufficient map quality27. The overall structures of resolved parts between mouse and hDUOX1 complex are similar with root mean square deviation of 1.521 and 0.908 Å for DUOX1 and DUOXA1 subunit, respectively (Fig. S11c). However, detailed structural comparison revealed several differences especially in the atomic models of FAD (Fig. S11d) and NADPH (Fig. S11e), probably due to the poor local map quality of mouse DUOX1 complex (EMD-21964).DiscussionIn this study, we provided the structures of hDUOX1–hDUOXA1 as a peptidisc-stabilized heterotetrameric protein complex in both high-calcium and low-calcium states. The structure of hDUOX1 complex in the high-calcium state reveals multiple inter-domain interactions that orientate DH and TMD for electron transfer, and thus redox reaction. Removal of calcium ions results in the reconfiguration of cytosolic inter-domain interactions which in turn mobilizes the DH domain and lowers the electron transfer efficiency (Fig. 6). These structures provide mechanistic insights into the structure and mechanism of DUOX and other NOX enzymes.Fig. 6Activation mechanism of DUOX1 complex by calcium.Two DUOX1 and one DUOXA1 subunit are shown as cartoon, and colored the same as Fig. 1d. Calcium ions are presented as green spheres. Electron transfer pathways are indicted with gray arrows.MethodsCell cultureHEK293F suspension cells (Thermo Fisher Scientific) were cultured in Freestyle 293 medium (Thermo Fisher Scientific) supplemented with 1% FBS at 37 °C with 6% CO2 and 70% humidity. Sf9 insect cells (Thermo Fisher Scientific) were cultured in SIM SF (Sino Biological) at 27 °C. The cell lines were routinely checked to be negative for mycoplasma contamination.Protein expression and purificationWe constructed a modified BacMam vector28,29 with N-terminal GFP tag guided by rat FSHβ signal peptide10 and cloned hDUOX1 cDNA into this vector (please see Supplementary Table 2 for primer list). The hDUOX1 cDNA, we obtained from Prof. Han, has the same protein sequence as NP_059130.2 except the L1178F mutation, a SNP previously observed in AAI14939.1. The cDNAs of hDUOXA1 were cloned into a non-tagged BacMam vector or a modified BacMam vector with C-terminal MBP-mScarlet tag (please see Supplementary Table 2 for primer list)28,29. The hDUOX1 mutants were generated by Quick Change methods (please see Supplementary Table 2 for primer list). The two expression cassette were further merged into one bicistronic vector by the LINK sequence on the modified vector28,30. The baculoviruses were produced using the Bac-to-Bac system and amplified in Sf9 cells. For protein expression, HEK293F cells cultured in Freestyle 293 medium at density of 2.8 × 106 ml−1 were infected with 15% volume of P2 virus. A total of 10 mM sodium butyrate was added to the culture 12 h post infection and transferred to a 30 °C incubator for another 36 h before harvesting. Cells were collected by centrifugation at 3999 × g (JLA 8.1000, Beckman) for 10 min, and washed with 20 mM Tris (pH 8.0 at 4 °C), 150 mM NaCl, 2 mM EGTA, 2 μg ml−1 aprotinin, 2 μg ml−1 pepstatin, 2 μg ml−1 leupeptin, flash-frozen, and storage at −80 °C.For each batch of protein purification, cell pellet corresponding to 0.5 liter culture was thawed and extracted with 20 ml buffer A (20 mM Tris pH 8.0 at 4 °C, 150 mM NaCl, 5 μg ml−1 aprotinin, 5 μg ml−1 pepstatin, 5 μg ml−1 leupeptin, 20% (v/v) glycerol, and 2 mM EGTA) containing 1 mM phenylmethanesulfonyl fluoride and 1% (w/v) digitonin (Biosynth) at 4 °C for 50 min. A total of 1 mg ml−1 iodoacetamide (Sigma—I1149) was added during the detergent extraction procedure to reduce nonspecific cysteine crosslinking. The supernatant was ultracentrifuged at 135,300 × g (TLA100.3, Beckman) for 50 min. The solubilized proteins were loaded onto 5 ml Streptactin Beads 4FF (Smart-Lifesciences) column and washed with 20 ml buffer A + 0.1% digitonin. The column was washed with 100 ml buffer A + 0.1% digitonin plus 10 mM MgCl2 and 1 mM adenosine triphosphate (ATP) to remove contamination of heat shock proteins. Then the column was washed with 40 ml buffer A + 0.1% digitonin again to remove residual MgCl2 and ATP. The target protein was assembled into the peptidisc on the Streptactin Beads through washing with 4 ml 1 mg ml−1 NSPr in 20 mM Tris pH 8.0 (ref. 12). Then the column was washed with 100 ml buffer A to remove free NSPr. The assembled peptidiscs were eluted with 40 ml buffer A + 5 mM D-desthiobiotin (IBA). Eluted protein was concentrated using 100-kDa cutoff concentrator (Millipore) and further purified by Superose 6 increase (GE Healthcare) running in HBS (20 mM Hepes pH 7.5, 150 mM NaCl) + 0.5 mM EGTA. Fraction 19 corresponding to DUOX1 + DUOXA1 peptidisc complex was concentrated to A280/415 = 4.4/1.4 with estimated concentration of 10.7 μM DUOX1 subunits (ε415 = 0.131 μM−1 cm−1).Enzymatic assayThe membrane fractions of DUOX1 for enzymatic assay were prepared as previously reported with minor modification31. Briefly, cells were washed with buffer A. After centrifuging at 3999 × g for 10 min at 4 °C, the cell pellets were broken using a needle for 12 times in 1 ml of 20 mM Tris pH 8.0 at 4 °C containing 0.1 mM dithiothreitol, 10 mM EGTA (pH 8.0), and the mixture of protease inhibitors. The pellet was removed by centrifuging at 3615 × g for 15 min, and the supernatant was collected and then centrifuged at 228,600 × g (TLA100.3, Beckman) for 1 h. The membrane pellet was resuspended in HBS containing 1 mM EGTA. Meanwhile, 10 μl membrane was solubilized by 100 μl TBS + 1% digitonin with the mixture of protease inhibitors for 1 h at 4 °C for FSEC. The protein concentrations in the membrane were estimated by comparing their GFP fluorescence signal to that of a purified GFP-tagged DUOX1 complex.The H2O2-generating activity of DUOX1 complex was determined using the amplex red assay32. The concentrations of H2O2 solution were determined by measuring UV–vis absorbance at 240 nm with spectrophotometer (Pultton) and calculated using molar extinction coefficient of 43.6 M−1 cm−1. The concentration of H2O2 solution was further validated by reacting with Amplex red to generate resorufin which has ε571 = 69,000 M−1 cm−1 (ref. 32). Then the H2O2 solution with known concentration was used to calibrate the resorufin fluorescence curve (excitation, 530 nm; emission, 590 nm) measured using a Microplate Reader (BioTek Synergy HT) at 37 °C.The H2O2-generating reaction of the membrane fraction containing DUOX1 complex was performed at 37 °C in 0.15 ml of HBS with 1 mM EGTA, 10 μM FAD, 100 μM NADPH, 50 μM amplex red, 0.067 mg ml−1 horseradish peroxidase, and 0.0576 mg ml−1 SOD. Ca2+ concentrations were determined using fluorescent indicators fura-2 or fluo3-FF. The Kcat and Km values of the membrane fraction containing DUOX1 complex were determined at 37 °C with different concentrations of NADPH in the presence or absence of 1.4 mM CaCl2. The H2O2-generating reaction of the purified DUOX1 complex in peptidisc was performed at 27 °C in 0.15 ml of HBS + 1 mM EGTA, 10 μM FAD, 100 μM NADPH, 50 μM amplex red, 0.067 mg ml−1 horseradish peroxidase, and 0.0576 mg ml−1 SOD in the presence or absence of 1.1 mM CaCl2. Progress of the reactions was monitored continuously by following the increase of the resorufin fluorescence, and the initial reaction rates were obtained by fitting the curve with linear equation. The activity of DUOX1 complex was determined by subtracting the background of the corresponding buffer without enzyme. The data were processed with Microsoft Excel-2013, SigmaPlot-12.0, and GraphPad Prism 6.Cryo-EM sample preparation and data acquisitionThe peptidisc sample was supplemented with 2.5 mM EGTA (low calcium) or 0.5 mM free calcium (high calcium) for cryo-EM analysis, respectively. Both samples contain 100 μM FAD as the cofactor and 500 μM NADPH as the substrate. To overcome the preferred orientation problem, 0.5 mM non-solubilizing detergent fluorinated octyl-maltoside was added to the sample before cyro-EM sample preparation. Aliquots of 1.5 μL protein sample were placed on graphene oxide-coated grids, as previously reported33. Grids were blotted for 3 s at 100% humidity and flash-frozen in liquid ethane cooled by liquid nitrogen using Vitrobot Mark I (FEI). Grids were then transferred to a Titan Krios (FEI) electron microscope that was equipped with a Gatan GIF Quantum energy filter and operated at 300 kV accelerating voltage. Image stacks were recorded on a Gatan K2 Summit direct detector in super-resolution counting mode using Serial EM at a nominal magnification of 130,000× (calibrated pixel size of 1.045 Å pixel−1), with a defocus ranging from −1.5 to −2.0 μm. Each stack of 32 frames was exposed for 7.12 s, with a total dose ~50 e− Å−2 and a dose rate of 8 e− pixel−1 s−1 on detector.Image processingThe image processing workflow is illustrated in Figs. S2 and S4. A total of 7076 super-resolution movie stacks of the high-calcium state sample and 2076 stacks of the low-calcium state sample were collected using Serial EM, and motion-corrected, dose weighted, and twofold binned to a pixel size of 1.045 Å using MotionCor2 (ref. 34). Contrast transfer function (CTF) parameters were estimated with Gctf35. Micrographs with ice or ethane contamination, and empty carbon were removed manually. Autopicking were performed using Gautomatch (kindly provided by Kai Zhang). All subsequent classification and reconstruction was performed in Relion 3.1 (ref. 36) unless otherwise stated. Reference-free 2D classification was performed to remove contaminants. Initial model was generated using cryoSPARC37. Particles were subjected to multi-reference 3D classification38,39 and random-phase 3D classification38,39. Phase-randomized models were generated from the model obtained from previous refinement using randomize software (from the lab of Nikolaus Grigorieff). Further CTF refinement was then performed with Relion 3.1 using C2 symmetry. The particles were then re-extracted, re-centered, and re-boxed from 256 pixels to 320 pixels for consensus refinement in Relion 3.1 (ref. 36) and cryoSPARC37. To improve the density of cytosolic layer, particles were symmetry expended13 for multibody refinement14. One soft mask (the large body) that covers the extracellular domain together with TMD of one protomer was generated from the consensus map, using UCSF Chimera and Relion 3.1 (ref. 36). The other soft mask (the small body) covers the cytosolic domains of the same protomer. 3D multibody refinements14 were performed using the two soft masks and the parameters determined from previous consensus refinement. The motions of the bodies were analyzed by relion_flex_analyse in Relion 3.1 (ref. 36). The two half-maps of each body generated by 3D multibody refinement were subjected to post-processing in Relion 3.1 (ref. 36). The masked and sharpened maps of each body were aligned to the consensus map using UCSF Chimera40, zoned to isolated nonoverlapping regions and summed using Relion 3.1 (ref. 36) to generate the composite maps for visualization and model building. All of the resolution estimations were based on a Fourier shell correlation of 0.143 cutoff after correction of the masking effect. B-factor used for map sharpening was automatically determined by the post-processing procedure in Relion 3.1 (ref. 36). The local resolution was estimated with Relion 3.1 (ref. 36).Model buildingThe composite maps derived from multibody refinement were used for model building. The structures of PHD, TMD, EF1–2, and DH domains of hDUOX1 were generated using phyre2 server41 based on PDB ID: 6ERC, 5O0T, 4IL1, and 5O0X, and manually docked into the cryo-EM maps using Chimera40. Initial models of PHLD were generated by Rosetta Web Server using ab initio mode42, manually selected according to the distances calculated by RaptorX Contact Prediction server43, and validated by the fitting between model and cryo-EM densities, especially the location of bulky aromatic residues. The partial model of hDUOXA1 were generated using EM builder44. The initial models were iteratively built using Coot45 and refined using Phenix in real space46. Figures were prepared using UCSF chimera40, Chimera X47, and Pymol.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationDescription of Additional Supplementary FilesSupplementary Movie 1Reporting Summary | nature communications | [
"Article"
] | [
"Enzyme mechanisms",
"Oxidoreductases",
"Cryoelectron microscopy"
] | oxygen species (ROS) are oxygen reactive hydrogen peroxide superoxide participate in processes pathological ROS generated by NADPH oxidase (NOX) multi-pass transmembrane proteins catalyze reduction oxygen superoxide anion hydrogen peroxide NOX proteins in biological processes host defense differentiation development cell growth survival cytoskeletal reorganization modification extracellular human NOX protein family NOX1–5 DUOX1–2 NOX2 protein catalyzes superoxide anion essential for host DUOX1–2 proteins expressed in thyroid tissue catalyze production hydrogen peroxide important for thyroid DUOX protein interactions with dual oxidase maturation factor (DUOXA promotes maturation plasma membrane localization DUOX encoded by genes DUOX1 DUOX2. encoded by DUOXA1 DUOXA2. mutations of DUOX2 DUOXA2 cause congenital hypothyroidism6 named thyroid oxidase4 family proteins share catalytic core heme-coordinating transmembrane domain cytosolic dehydrogenaseDH domain binds NADPH FAD shares sequence homology ferredoxin-NADP + reductase two TMD-DH catalytic core functional DUOX protein large N-terminal extracellular peroxidase homology domain) long intracellular loop 0 two EF-hand domains requires auxiliary DUOXA protein for function activity DUOX regulated by intracellular calcium structures NOX family members isolated domains DH domain 5O0X TMD 5O0T)9 NOX5 from Cylindrospermum stagnale subdomain human NOX2 DH domain ID 3A1F). functional importance DUOX NOX family members structures full-length functional protein complex unknown open questions DUOX DH domain with TMD DUOXA protein with DUOX activity regulated by intracellular calcium? DUOX–DUOXA protein complex structurally functionally cryo-EM structures human DUOX1–DUOXA1 complex in high-calcium low-calcium states structure calcium activation DUOX constructed matured protein (20–1551) N-terminal GFP tag guided rat FSHβ signal peptide for molecular weight GFP-tagged hDUOX1 219 kDa.DUOX1–DUOXA1 complex fused hDUOXA1 protein C-terminal MBP-mScarlet tag molecular weight 106 kDa chromatography co-expression hDUOXA1 shifted peak hDUOX1 higher molecular weight stable hDUOX1–hDUOXA1 hetero-complex (Fig. S1a peak positions hDUOX1 migrated monomer heterotetramer co-expression DUOX1 DUOXA1 cell membranes robust calcium-activated-dependent hydrogen peroxide production (Fig. low-calcium DUOX1–DUOXA1 complex low basal activity Addition calcium reduced Km increased Kcat DUOX1 enzymatic activity 1Structure DUOX1–DUOXA1 complex high-calcium state Schematic DUOX enzymatic assay H2O2 horseradish peroxidase converts nonfluorescent Amplex Red to fluorescent resorufin H2O2. Calcium-dependent activation hDUOX1–hDUOXA1 complex Data means ± standard deviations n = 3 samples Source data Steady state enzyme activity NADPH concentration presence absence calcium Data Michaelis–Menten equation Km Kcat valueData shown as means ± standard deviations = 3 independent samples Source data file Side view cryo-EM map hDUOX1–hDUOXA1 complex high-calcium state boundaries phospholipid bilayer gray lines One protomer hDUOX1 colored blue green other yellow red 90° rotated top view 180° rotated bottom view cross-section transmembrane layer dashed line large cavity dashed cryo-EM map low-pass filtered to 6 Å Topology of hDUOX1 hDUOXA1 subunits Transmembrane helices cylinders disordered regions dashed lines phospholipid bilayer gray layers PHD peroxidase PHLD pleckstrin EF-binding DH dehydrogenase CLD Structure of protomer hDUOX1 cartoon colors same boundaries phospholipid bilayer gray lines Sugar moieties hemes FAD NADPH as black yellow pink green sticks solubilized reconstituted hDUOX1–hDUOXA1 complex into peptidisc using NSPr12 Purified peptidisc sample homogenous peak on SEC (Figtwo protein bands SDS–PAGE trimmed PNGase F treatment (Fig. hDUOX1 hDUOXA1 modified N-linked glycosylation UV–vis spectrum peptidisc sample Soret band peak 415 nm (Fig. Fe (III) heme incorporation purified peptidisc sample recapitulated calcium-activated-dependent hydrogen peroxide production. calcium-dependent activation hDUOX1–hDUOXA1 protein complex maximum activity purified peptidisc lower crude cell membrane. 1b membrane bilayer endogenous lipids DUOX activity prepared cryo-EM grids peptidisc sample 2.5 mM ethylene glycol tetraacetic acid 0.5 mM free calcium samples 0.1 mM FAD cofactor 0.5 mM NADPH substrate particle cryo-EM analysis purified protein homogeneous twofold symmetry (Figs. S2–S4) resolution cryo-EM maps low-calcium high-calcium states 2.7 2.6 Å S1) extracellular domains TMD better resolution cytosolic domains higher mobility (Figs. S2g quality exploited symmetry multibody protomer large smallS2c S4c). final resolutions cytosolic domain reached 3.4 3.2 Å low high-calcium states (Figs. S2–4 Table S1) high map quality homology structures order regions complex 88% DUOX1 79% DUOXA1 (Figs. S5–8 Table S1) focus high-calcium state structure higher resolution hDUOX1–hDUOXA1 protein co-assemble 2:2 heterotetrameric protein complex molecular weight ~457 kDa complex encompasses 140 Å × 105 Å 160 Å 3D space twofold rotational symmetry (Fig. 1d–f). divided three layers extracellular transmembrane layer cytosolic layer extracellular layer two N-terminal PHD domains hDUOX1 pack buttressed DUOXA1 transmembrane layer 24 transmembrane helices heme-binding sites electron transfer pathway (Fig. center large cavity protein densities interior surface hydrophobic (Fig. S3i lipid molecules bound filled phospholipids cytosolic layer catalytic DH domain regulatory domains intracellular calcium sensingcatalytic hDUOX1 catalytic subunit complex (Fig. 2) extracellular side large N-terminal PHD domain shares sequence homology peroxidases peroxidase A Dictyostelium discoideum (Fig. S9a). Functional peroxidases utilize histidine-coordinated heme catalysis key residues heme binding ligand histidine missing PHD hDUOX1 heme density hDUOX1 PHD not enzymatic functional peroxidase activity reveals two cation densities PHD One cation CBS1) coordinated side chains D397 T332 chain carbonyl groups V399 T332 R395 (Fig. second cation (CBS2) coordinated side chain D109 D174 S176 T170 carbonyl groups T170 W172 observed strong densities sites low-calcium high-calcium conditions cations sodium ions calcium ions CBS1 CBS2 evolutionary conserved in DUOX (Fig. S5) DdPoxA15 functional importance CBS1 mutant CBS2 mutant (D109A D174A) DUOX1 failed co-assemble with DUOXA1CBS1 CBS2 subunit interfaces DUOX1–DUOXA1 complex mutants affect folding PHD domain role protein stability.Fig. 2Structure hDUOX1 subunit Side view high-calcium state key interfaces domain colored Fig. 1h surface transparency site outer heme TMD sticks colored yellow Unrelated helices omitted oxygen-reducing center arrow site inner heme interface between PHD TMD Disulfide bond between C118–C1165 golden sticks interface PHLD TMD hydrogen bonds dashed lines interface PHLD DH domain EF module DH domain FAD-binding site interface TMD DH domain Ligands interacting residues sticks NADPH-binding site interface TMD DH domain PHD packs top TMD DUOX1 noncovalent interactions (Fig. disulfide bond between C118 PHD C1165 loop C staples bottom PHD top TMD (Fig. hDUOX1 extra bent M0 helix periphery M3 M4 six TM helices NOX protein family M1–M6 form two heme-binding sites TMD H1144 on M3 H1238 on M5 coordinate outer hemeH1130 M3 H1225 M5 coordinate inner heme. histidines conserved NOX proteins. S6) observed spherical density R1087 M2 H1148 M3 heme residue H1144 (Fig. 2b mutations csNOX5 residues R1087 H1148 reoxidation dithionite-reduced TMD oxygen substrate site oxygen-reducing center. S10a structure observations hDUOX1 support hypothesis amphipathic preM1 helix inner leaflet plasma membrane. observed csNOX5 shared feature NOX proteins Between M0 preM1 cytosolic fragment loop 0 Cryo-EM maps N-terminal loop domain rich β sheets (Figs. S3g Structural search DALI identified β sheets-rich domain pleckstrin homology-like domain high structural similarity PH domain proteins. S10c EF-hand calcium-binding domains (EF1 EF2) compact helical module connected PHLD αC (Fig. S6) Residues calcium binding EF1 EF2 conserved DUOX proteins S6) strong densities small calcium ions poor local resolutionstructure EF-hand module resembles calcium-dependent protein phosphatase calcineurin state ID 4IL1)18 EF1 EF2 loaded calcium high-calcium state homology structure side chains D828 D830 N832 E839 carbonyl group Y834 chelate one calcium ion EF1 chains D864 D866 N868 E875 carbonyl L870 chelate another EF2 mutations calcium-binding sites abolished calcium E879K mutation hDUOX2 (E875 hDUOX1) leads congenital hypothyroidism20 activation C-terminal catalytic DH domain connected to M6 DUOX1 TMD short linker (Fig. 1h DH domain hDUOX1 canonical DH fold structure similar csNOX5 strong densities FAD cofactor NADPH substrate sites contributed from DH domain TMD 2h-domain interactions high-calcium domains DUOX1 stabilized inter-domain interactions PHLD interacts with TMD DH domains main chain carbonyl group K653 PHLD bond with R1215 loop D TMDR674 PHLD interacts E1348 I1349 α1 DH domain (Fig. EF1–EF2 module high-calcium α4 α4 loop DH domain (Fig. 2g interactions EF module DH hydrophobic involve F768 F772 F807 F819 F840 F847 L1463 M1467 I1470 F1475 V1478 F1484 DH K814 EF E1281 β2 DH interactions EF module DH domain hDUOX1 mimic calcineurin subunit B A calcium-bound state (Fig. S10g linker EF module preM1 helices binds groove surface DH domain (Fig. DH docks bottom TMD interaction R1270 M6 D1367 β7 R1113 B TMD N1550 DH (Fig. 2h R1111Q mutation hDUOX2 (R1113 hDUOX1) identified congenital hypothyroidism inter-domain interaction FAD cofactor NADPH substrate bind interface DH TMD R1214 R1131 TMD electrostatic interaction FAD D1128 hydrogen ribose FADE1039 N1040 TMD hydrogen bonding adenosine ring NADPH R1036 cation-π interaction adenosine electrostatic interaction phosphate group NADPH (Fig. R1495 R1424 R1036 participate electrostatic interactions phosphate group NADPH ribose distinguish NADPH from NADH NADPH-binding site csNOX5 blocked by engineered C-terminal previous crystallization (Fig. adenosine group FAD 180° flip DH csNOX5 D1128 TMD stabilizes ribose FAD adenosine group inner heme electron transfer binding FAD NADPH interface DH TMD hDUOX1 stabilize docking DH bottom TMD putative electron transfer edge-to-edge distances between NADPH FAD FAD inner heme outer heme 8.2 3.9 6.7 Å (Fig. 3a). additional protein residues on DUOX1 rely electrons NADPH to FAD W378 hemes csNOX5 distance between NADPH FAD larger than canonical FNR protein 3.2 Å spinach FNRcomparison DH domain DUOX1 relaxed conformation two subdomains loosely packed DH csNOX5 sFNR tense conformation tightly packed FAD NADPH electron transfer (Fig. 3b–e). electron transfer efficiency not optimal DUOX1 complex cell membrane higher activity (Fig. 1b lipids bilayer environment affect structure electron transfer efficiency. 3Electron transfer pathway hDUOX1 subunit high-calcium state edge-to-edge distances between NADPH FAD FAD inner heme two hemes shown ligands sticks domain hDUOX1 surface colored same Fig. 1h one hDUOX subunit clarity putative oxygen-reducing center boxed dashed lines DH domain hDUOX1 relaxed conformation DH csNOX5 sFNR tense conformation ligands sticks two subdomains (FAD NADPH-binding cartoon Distances between Cα atoms Arg FBD Cys NBD labeled Structural comparison hDUOX1 DH domain csNOX5 FBD for structural alignment close-up view putative oxygen-reducing center Four tunnels for oxygen substrate entrance product exit yellow green magenta orangeResidues tunnels as sticks Calculated radii tunnels f–i oxygen-reducing center starting point terminus electron transfer chain extracellular initial product oxygen-reducing reaction superoxide anion probed pathways oxygen entrance superoxide anion exit with CAVER22 center starting point located four tunnels A formed by M1 M2 M5 M6 capped by loop E B surrounded by M2 A C E C embraced by M3 M4 C D enclosed by M3 M4 C M0 bottleneck radii ~1 Å permeation small oxygen substrate DUOX1 protein tunnels B–D surrounded by hydrophobic residues unfavorable for superoxide anion permeation tunnel A gated by hydrophilic R1087 R1062 R1248 Q1245 loop E positively charged constriction tunnel A negatively charged superoxide anions dismutation reaction uncharged hydrogen peroxide constrictions tunnels A–D affect superoxide anion leakagemutations DUOX1 loop A N-terminus peptide) ratio superoxide hydrogen peroxide leakage hDUOXA1 complex essential auxiliary subunit DUOX enzyme5. S11a extracellular N-terminus important hydrogen peroxide NTP extends packs PHD–TMD junction distal hDUOX1 subunit (Fig. 4a–e). Side chains F8 F10 Y11 NTP insert into hydrophobic groove loop C A PHD hDUOX1 K15 DUOXA1 NTP interactions with D1077 DUOX1 (Fig. data NTP interacts DUOX1 loop A hDUOXA1 five transmembrane helices TM1 interacts with preM1 M1 hDUOX1. remaining four helices extracellular loops structural similarity with claudin superfamily members extracellular loop between TM2 TM3 into compact claudin-like domain (CLD) four β strands two α helices (Figs. S8 S11) CLD interactions distal proximal DUOX1 subunits. 4a role complex assemblyagrees studies splicing variants TM2–TM3 loop activity DUOX1 ordered N-linked glycosylation decoration N109 hDUOXA1 branched sugar moieties polar interactions with DUOXA1 DUOX1 subunits (Fig. 4a PHD two DUOX1 subunits interact (Fig R50 R507 PHD polar interactions with E41 F313 opposite PHD (Fig. 4e). analyzed effects interface mutations tetramer assembly mutations R50E R507E R507A affect tetrameric peak formation FSEC (Fig. 4f). structural biochemical data inter-subunit interactions heterotetramer assembly.Fig. 4Mechanism hDUOX1–hDUOXA1 tetramer assembly side view hDUOX1–hDUOXA1 protein complex Fig 1d open-book view inter-subunit interfaces Residues hDUOX1 subunits hDUOXA1 colored green yellow blue close-up view interactions NTP hDUOXA1 hDUOX1 interactions top view interactions between PHD two hDUOX1 subunitsFSEC traces hDOUX1 R50E R507E R507A mutants compared to wild-type hDOUX1 peak position hDOUX1 denoted hollow circles Asterisks denote peak position hDUOX1–hDUOXA1 protein complex.Conformational change DUOX1 complex upon calcium low-calcium state cytosolic layer poor local resolution improved by multibody refinement14 (Fig S4) molecular flexibility cytosolic domains broad orientations TMD plateau-shaped distribution histogram major eigenvector contrast normal distribution high-calcium state cytosolic layer more flexible compared structures structural changes extracellular transmembrane layer small (Fig. large conformational changes regulatory PHLD EF-hand module in cytosolic layer calcium EF module switches extended contracted reconfigures interface α4 DH domain loosely packed structure low-calcium state EF2 moves away from DH domain Cα atom A894 on αJ EF2 has 40 Å displacement PHLD rotates away from TMD DH domains αA 17.2° outward rotationinter-domain interactions high-calcium state disrupted docking DH domain TMD weakened higher mobility DH (Fig. S4g). increased mobility DH correlates with electron transfer efficiency catalytic activity DUOX TMD contributes to FAD NADPH binding increased mobility reduced affinity NADPH with reduced Kcat increased Km low-calcium state (Fig. 1c).Fig. 5Conformational change hDUOX1 complex during calcium activation Structural comparison hDUOX–hDUOXA1 complex high-calcium low-calcium states Protein cartoon large conformational changes dashed lines Close-up conformational changes EF-hand module Cα atom A894 on αJ helix movement EF2. conformational change PHLD angle between αA helices high low-calcium states measured Conformational differences EF-hand module high-calcium low-calcium state Reconfiguration interface between EF-hand module α4 helix DH domain Arrows movements group reported structures mouse DUOX1–DUOXA1 complex27 heterodimeric heterotetrameric form activity regulated by dimer–tetramer assembly27contrast majority hDUOX1 complex homogenous tetrameric form (Fig. difference protein preparation procedure species elusive intracellular PHLD EF domains not resolved mouse DUOX1 complex insufficient map structures parts mouse hDUOX1 complex similar mean deviation 1.521 0.908 Å DUOX1 DUOXA1 subunit (Fig. comparison differences atomic models FAD NADPH due poor map quality DUOX1 structures hDUOX1–hDUOXA1 peptidisc-stabilized heterotetrameric protein complex high-calcium low-calcium states structure high-calcium state reveals inter-domain interactions DH TMD electron transfer redox reaction Removal calcium ions cytosolic-domain interactions mobilizes DH domain lowers electron transfer efficiency (Fig. 6) structures insights structure mechanism DUOX NOX enzymes. 6Activation mechanism DUOX1 complex by calcium DUOX1 DUOXA1 subunit Calcium ions green spheres Electron transfer pathways gray arrows cultureHEK293F suspension cells cultured Freestyle 293 medium 1% FBS 37 °C 6% CO2 70% humiditySf9 insect cells Fisher Scientific cultured SIM SF 27 °C cell lines checked negative mycoplasma contamination expression constructed modified BacMam N-terminal GFP tag rat FSHβ cloned hDUOX1 cDNA Table 2 Prof. Han same protein sequence_059130.2 L1178F mutation AAI14939.1. cDNAs hDUOXA1 cloned non BacMam C-terminal MBP-mScarlet tag hDUOX1 mutants generated Quick Change methods merged bicistronic vector LINK sequence baculoviruses produced Bac-to-Bac system amplified Sf9 cells HEK293F cells cultured Freestyle 293 medium × 106 ml−1 infected 15% P2 virus 10 mM sodium butyrate added culture 12 h post infection transferred 30 °C incubator 36 h before harvesting Cells collected centrifugation 3999 × g 10 min washed 20 mM Tris (pH 8.0 4 150 mM NaCl 2 mM EGTA 2 aprotinin pepstatin leupeptin flash-frozen storage −80 °Cprotein purification cell pellet 0.5 liter culture extracted 20 ml buffer pH 8.0 4 °C 150 mM NaCl 5 μg aprotinin pepstatin leupeptin 20% glycerol 2 mM EGTA 1 mM fluoride 1% digitonin 4 °C 50 min 1 mg iodoacetamide added cysteine crosslinking supernatant ultracentrifuged 135,300 × g 50 min solubilized proteins loaded 5 ml Streptactin Beads column washed 20 ml buffer 0.1% digitonin 100 ml buffer 0.1% digitonin 10 mM MgCl2 1 mM adenosine triphosphate 40 ml buffer 0.1% digitonin MgCl2 protein peptidisc 4 ml 1 mg NSPr 20 mM Tris pH 8.0 washed 100 ml buffer NSPr peptidiscs eluted 40 ml buffer A 5 mM D-desthiobiotin Eluted protein concentrated 100-kDa concentrator purified Superose 6 HBS (20 pH 7.5 150 mM NaCl 0.5 mM EGTAFraction 19 DUOX1 A280/415 4.4/1.4 10.7 μM DUOX1 subunits (ε415 0.131 μM−1 cm−1) membrane DUOX1 cells washed buffer A centrifuging 3999 10 min cell pellets broken 12 times 1 ml 20 mM Tris pH 8.0 0.1 mM dithiothreitol 10 mM EGTA protease inhibitors pellet removed 3615 g 15 min supernatant collected centrifuged 228,600 × g 1 h membrane pellet resuspended HBS 1 mM EGTA 10 μl membrane solubilized 100 μl TBS 1% digitonin protease inhibitors 1 h 4 °C protein concentrations estimated GFP fluorescence purified GFP-tagged DUOX1 complex H2O2-generating activity DUOX1 determined amplex red concentrations H2O2 UV–vis absorbance 240 nm molar extinction coefficient 43.6 M−1 cm−1 validated Amplex red resorufin ε571 69,000 M−1 cm−1H2O2 solution resorufin fluorescence curve 530 emission 590 Microplate Reader 37 °C H2O2-generating reaction DUOX1 complex 37 °C 0.15 ml HBS 1 mM EGTA 10 μM FAD 100 μM NADPH 50 μM amplex red 0.067 horseradish peroxidase 0.0576 mg SOD Ca2+ concentrations determined Kcat Km values determined 37 °C concentrations NADPH 1.4 mM CaCl2. H2O2-generating reaction purified DUOX1 27 °C 0.15 ml HBS 1 mM EGTA 10 μM FAD 100 μM NADPH 50 μM amplex red 0.067 horseradish peroxidase 0.0576 mg SOD 1.1 mM CaCl2. monitored resorufin fluorescence initial reaction rates curve activity DUOX1 complex subtracting background buffer enzyme data processed Excel-2013 SigmaPlot-12.0 GraphPad Prism 6.Cryo-EM sample supplemented 2.5 mM EGTA 0.5 mM free calcium samples 100 μM FAD 500 μM NADPH substrateorientation 0.5 mM fluorinated octyl-maltoside added before cyro-EM 1.5 μL protein sample on oxide-coated grids blotted 3 s 100% humidity flash-frozen ethane cooled nitrogen Vitrobot transferred to Titan Krios electron microscope Gatan GIF Quantum energy filter 300 kV voltage Image stacks recorded Gatan K2 Summit detector super-resolution Serial EM magnification 130,000× pixel size 1.045 Å defocus −1.5 to −2.0 μm stack 32 frames exposed 7.12 s total dose ~50 e− Å−2 rate 8 e− pixel−1 s−1 Figs. S2 S4 7076 super-resolution stacks 2076 low-calcium collected Serial EM motion-corrected dose weighted twofold binned pixel size 1.045 Å MotionCor2 Contrast transfer parameters estimated Gctf35 Micrographs ice ethane contamination empty carbon removed Autopicking Gautomatch classification reconstruction Relion 3.1 36) Reference-free 2D classification contaminants Initial model generated cryoSPARC37 Particles subjected multi-reference 3D random-phase 3DPhase-randomized models generated from previous refinement CTF refinement performed Relion 3.1 symmetry particles re-extracted-centered re-boxed from 256 to 320 pixels for consensus refinement Relion 3.1 cryoSPARC37 density cytosolic layer particles symmetry for multibody soft mask large extracellular domain TMD protomer generated from consensus map UCSF Chimera Relion 3.1 other covers cytosolic domains protomer 3D multibody performed using two soft masks parameters previous consensus refinement motions analyzed by relion_flex_analyse Relion 3.1 two half-maps post-processing Relion 3.1 masked sharpened maps aligned to consensus map UCSF zoned regions summed Relion 3.1 composite maps for visualization model building resolution estimations on Fourier correlation 0.143 cutoff after correction masking effect B-factor map sharpening determined post-processing Relion 3.1 local resolution estimated with Relion 3.1 composite maps multibody refinement used for model buildingstructures PHD TMD EF1–2 DH domains hDUOX1 generated PDB ID 6ERC 5O0T 4IL1 5O0X docked cryo-EM maps Chimera40 models PHLD generated Rosetta Web Server ab initio selected distances RaptorX Contact Prediction validated fitting model cryo-EM densities bulky aromatic residues partial model hDUOXA1 generated EM builder44 initial models built Coot45 refined Phenix Figures prepared UCSF chimera40 Chimera X47 Pymol Nature Research Reporting Summary.Supplementary Additional | 50.3 | 0.740802 |
10.1038/s41467-020-17590-x | PMC7387547 | In this study the authors show that monotonous basaltic volcanoes can host a range of melts in their sub-volcanic systems, extending to rhyolitic compositions. The study implies that volcanoes which have produced monotonous basaltic lavas on long timescales could transition to more explosive, silica-rich eruptions in the future. | Many volcanoes erupt compositionally homogeneous magmas over timescales ranging from decades to millennia. This monotonous activity is thought to reflect a high degree of chemical homogeneity in their magmatic systems, leading to predictable eruptive behaviour. We combine petrological analyses of erupted crystals with new thermodynamic models to characterise the diversity of melts in magmatic systems beneath monotonous shield volcanoes in the Galápagos Archipelago (Wolf and Fernandina). In contrast with the uniform basaltic magmas erupted at the surface over long timescales, we find that the sub-volcanic systems contain extreme heterogeneity, with melts extending to rhyolitic compositions. Evolved melts are in low abundance and large volumes of basalt flushing through the crust from depth overprint their chemical signatures. This process will only maintain monotonous activity while the volume of melt entering the crust is high, raising the possibility of transitions to more silicic activity given a decrease in the crustal melt flux. | IntroductionVolcanoes are underlain by complex and dynamic magmatic systems that often span tens of kilometres of crust from the Moho to the near-surface1,2. As melts ascend through the lithosphere they undergo diverse processes, including crystallisation, volatile exsolution, assimilation of the surrounding country rock, mixing between different magma batches, and interaction with mush crystallised from previous magmas3–5. These processes modify magma compositions, creating significant diversity in the chemistry of igneous rocks observed at the Earth’s surface6. Despite this multitude of sub-surface processes, many magmatic systems exhibit remarkably monotonous volcanic behaviour, erupting chemically homogeneous liquids over long timescales (several decades to millennia)7–14. The causes of monotonous volcanism are not currently well understood, but constraining the architecture and dynamics of monotonous systems is essential for determining their longevity and identifying the potential for future changes in eruptive behaviour that may result in more hazardous activity15.Current models for producing chemically monotonous eruption sequences typically involve: (1) uniformity by reactive filtration8, whereby distinct magma batches interact with surrounding gabbroic material to form chemically homogeneous products or (2) uniformity by processing, whereby successive batches of ascending magma evolve under the same P-T conditions16. The first model typically requires hot primitive melts to cool and react as they ascend through a super-solidus mush column, buffering their temperature and composition1,8,17,18. In many cases, however, the products of monotonous eruption series contain multiple macrocryst populations that record complex growth histories. Geophysical and geobarometric constraints on the crustal structure beneath monotonous volcanoes are also inconsistent with single large storage regions9,19–22. The second model often includes repeated recharge of upper crustal magma bodies by chemically consistent melts ascending from depth, maintaining the crustal system within a narrow temperature and compositional range7,10,11,23. In this case, monotonous activity necessitates an exact thermal balance between the heat supplied by ascending magma and the heat lost by advection and eruption.The western Galápagos Archipelago is an ideal location for studying compositionally monotonous volcanism because it hosts several volcanoes that have erupted near-homogeneous basaltic magmas for several millennia8,17,24,25. These include Fernandina volcano (on the island of the same name) and Wolf, Darwin, Ecuador, and Sierra Negra volcanoes on Isabela Island (Fig. 1a)8,26. All of the basalts emplaced during monotonous Galápagos eruptions have undergone extensive olivine, clinopyroxene and plagioclase crystallisation8,26,27. However, historic lavas at each individual centre have pre-eruptive storage temperatures within a range of only ~22–30 °C17,28. Monotonous volcanoes on Fernandina and Isabela are located near the centre of the inferred Galápagos plume29. The prevailing hypothesis is that the uniformity in their erupted products is related to the flux of mantle-derived magma entering the crust; the volcanoes receive enough thermal input to sustain thick, thermochemically steady-state gabbroic mush zones in the mid-crust, which interact with ascending magmas, buffering their temperature and composition8. More compositionally diverse Galápagos volcanoes are thought to lack such large mush zones: Cerro Azul is also proximal to the inferred plume centre but erupts basalts with a greater range of MgO concentrations, likely because there has been insufficient magma input to the crust for a thermally stable magmatic system to develop30; Alcedo is downstream from the plume centre and has produced dacitic and rhyolitic eruptions, which may reflect the mush reaching a phase where melts entering the crust are able to cool and fractionate31. Harpp and Geist26 extend this geographic trend to volcanoes in the eastern archipelago, suggesting that their more primitive and compositionally variable eruptions reflect the absence of sustained magma plumbing systems and thus limited crustal processing.Fig. 1Locations of Wolf and Fernandina volcanoes and the sampling sites in this study.a Regional map of the Galápagos Archipelago showing the different volcanic centres on Isabela Island. Boxes show the locations of maps (b) and (c). b Detailed map of Wolf volcano showing the locations of the 2015 circumferential fissure and intra-caldera vent. The 2015 lava flow extent is in red, after Bernard et al.71. The sampling locations of the lavas and tephra analysed in this study are shown as green circles and blue diamonds, respectively. c Detailed map of Fernandina showing the extent of the pyroclastic flow (dashed line) and isopachs of the total tephra (in cm; dotted lines) produced during the 1968 eruption, after Howard et al.48. The sampling locations of the nodules analysed in this study are shown as dark red squares. Contours are 200 m.Recent Galápagos eruptions have afforded geophysical constraints on magma storage depths19,21,32,33, making them good targets for understanding the processes responsible for homogenising erupted liquids. In this study, we use detailed microanalyses of mineral phases to constrain magmatic processes at Wolf and Fernandina volcanoes (Fig. 1), which have erupted monotonous basaltic melts for several millennia (Fig. 2)8,17,24,25. We focus on integrating petrographic and geochemical observations of pre-eruptive compositional heterogeneity with thermodynamic models that reveal the range of magmas present in the sub-volcanic systems. This allows us to identify the sub-volcanic process that regulates the diversity of erupted liquids. By comparing Wolf and Fernandina with other Galápagos volcanoes, we identify controls on the transition between monotonous and more compositionally varied volcanism.Fig. 2Erupted liquid compositions from the Galápagos Archipelago.TiO2 vs. Mg#liq of erupted liquids at a Wolf volcano and b Fernandina. The points show: whole-rock, tephra glass, submarine glass and melt inclusion literature data from all volcanoes in the western Galápagos Archipelago (excluding intrusive rocks and plagioclase-ultraphyric lavas); historic erupted liquids (i.e. whole-rocks, tephra glasses and submarine glasses) from Wolf and Fernandina; and whole-rock and tephra glass data from the 2015 Wolf and 1968 Fernandina eruptions (see legend). References for Wolf and Fernandina historic erupted liquids and literature data from all western Galápagos volcanoes are in Supplementary Note 2. Glass data for the 2015 Wolf eruption are from Stock et al.19 and whole-rock data are from this study. Glass and whole-rock data for the 1968 Fernandina eruption are from Allan and Simkin28. Characteristic 2σ analytical uncertainties for our whole-rock analyses are less than the size of a data point. The black lines show liquid lines of descent calculated using Rhyolite-MELTS at 50 MPa (solid line), 300 MPa (dashed line) and 500 MPa (dotted line). The kernel density estimates above each panel show the Mg#liq distribution of the historic erupted liquids from Wolf and Fernandina, the red points show their median Mg#liq, and the red bars show their Mg#liq interquartile range. The grey dotted bars above each panel show the Mg#liq interquartile range of historic lavas from Santiago26 (centred around the Wolf and Fernandina medians) for comparison.Results and discussionSamples and petrographyThe samples used in this study include basaltic lava and reticulitic tephra from the 2015 Wolf eruption and gabbroic nodules from the 1968 Fernandina eruption (see Supplementary Note 1 for eruption chronologies). They were selected because they have existing petrological constraints that provide information on their storage depths within their respective sub-volcanic systems. The Wolf lavas are from the circumferential fissure phase of the eruption and contain ≳50 µm, euhedral–subhedral plagioclase (~5 vol.%), clinopyroxene (~2 vol.%) and olivine (<1 vol.%) macrocrysts within a microcrystalline groundmass. The macrocrysts occur in three distinct textural associations: (1) isolated phenocrysts surrounded by groundmass; (2) glomerocrystic aggregates containing plagioclase + clinopyroxene ± olivine (Fig. 3a, b); (3) plagioclase aggregates attached by synneusis that share common rims. The Wolf tephra was produced during an initial explosion and contains the same macrocryst assemblage as the lava samples, except that anhedral quartz is also present in very low abundance (≪1 vol.%; Fig. 3c) and clinopyroxene crystals occasionally contain ilmenite inclusions (Fig. 3c–e). The Fernandina nodules were exhumed during a hydromagmatic paroxysm and have a phaneritic texture, indicating slow cooling, with a gabbroic mineral assemblage comprising euhedral plagioclase (60–70 vol.%), subhedral clinopyroxene (≤20 vol.%) and subhedral–anhedral olivine (2–15 vol.%). Plagioclase occurs both as independent grains and as inclusions within earlier-crystallising phases (usually clinopyroxene). Miarolitic cavities (<2 mm diameter) are also present, with clinopyroxene lining many of the void walls. Hydrothermally altered samples exhibit secondary pyrite, iddingsite replacement of olivine and epidotisation of other primary magmatic minerals (Fig. 3f–g).Fig. 3Petrographic evidence for heterogeneous liquids and multi-level magma storage.a Quantitative Evaluation of Minerals by SCANing electron microscopy (QEMSCAN) phase map and b Ca map of a glomerocryst in a 2015 Wolf lava sample. The values in (b) show the An# of plagioclase zones. c−e False colour backscattered electron (BSE) images of quartz and clinopyroxene crystals in tephra samples from the 2015 Wolf eruption (d and e are adapted from Stock et al.19). The values in (d) and (e) show the Mg#cpx of clinopyroxene zones. f, g False colour BSE images of nodule samples from the 1968 Fernandina eruption, showing the phaneritic texture, a miarolitic cavity and evidence of hydrothermal alteration. An# and Mg#cpx values in black and grey show averages from crystal cores and rims, respectively. Ol – olivine, Cpx – clinopyroxene, Pl – plagioclase, Qtz – quartz, Gl – glass, Ilm – ilmenite, Py – pyrite, MC – miarolitic cavity.Homogeneous erupted liquidsWe compiled published whole-rock and matrix glass analyses from historic Wolf and Fernandina eruptions to evaluate the degree of variability in erupted liquid compositions (Fig. 2; see Supplementary Note 2 for data sources). The Mg#liq (Mg#liq = atomic Mg/[Mg + Fe2+]·100; see Methods) interquartile ranges of subaerial lavas from Wolf and Fernandina are only 4.85 (median = 52.4) and 2.86 (median = 53.1), respectively. Some lavas from both volcanoes contain accumulated plagioclase, which slightly decreases their whole-rock Mg#liq, and a small number of lavas from Fernandina have accumulated olivine, which significantly increases their bulk Mg#liq17,28,34. Additionally, a few spatially restricted submarine lavas erupted on the southwest flank of Fernandina have more evolved (low MgO, high K2O; evolved series) compositions34; including submarine samples increases the Fernandina Mg#liq interquartile range to 5.56 (median = 52.8). Nevertheless, the Mg#liq interquartile ranges of historic eruptions at Wolf and Fernandina are extremely limited and reflect a remarkable degree of homogeneity (cf. Mg#liq interquartile ranges of 12.7 at Santa Cruz and 21.4 at Santiago in the eastern Galápagos Archipelago8,26; Fig. 2), especially given that the lavas represent thousands of years of eruptive activity17,25.We analysed 18 whole-rock samples from the 2015 Wolf eruption (Fig. 2; ref. 19; Supplementary Dataset 1). These do not show any significant compositional variability outside of uncertainty, despite sampling lava flows that erupted at different times through the circumferential fissure phase of the eruption. The bulk samples have an average Mg#liq of 53.0 ± 0.23 (1σ of all whole-rock analyses), which is higher than most previous measurements of the 2015 tephra glass (Mg#liq = 45.4 ± 1.22 [1σ of all glass analyses]; Fig. 2a)19. Although there are fewer data for the 1968 Fernandina eruption, the scoria glass also has a lower Mg#liq than the bulk lava (Fig. 2b)28. These more evolved glass compositions likely reflect minor pre-eruptive crystallisation of the bulk magmas. Whole-rock analyses from the 2015 Wolf eruption have similar Al2O3 contents to the tephra glass and most historic erupted liquids, indicating that there was no significant plagioclase accumulation (Supplementary Fig. 1). In contrast, bulk lavas from the 1968 Fernandina eruption do have elevated Al2O3 concentrations, demonstrating that they have accumulated plagioclase (Supplementary Fig. 2). Both the matrix glass and whole-rock Mg#liq of material erupted in 2015 and 1968 fall within the interquartile ranges of liquids historically erupted from Wolf and Fernandina (Fig. 2).Heterogeneous mineral compositionsWe collected ~800 plagioclase and ~90 olivine analyses from the 2015 Wolf eruption (Fig. 4; Supplementary Datasets 2, 3), which we have integrated with ~500 clinopyroxene analyses from the same samples reported by Stock et al.19. We also collected ~160 plagioclase, ~100 clinopyroxene and ~80 olivine analyses from 1968 Fernandina nodules (Fig. 4; Supplementary Datasets 4–6) and compare these with mineral analyses from historic Fernandina lavas28,35. In contrast with the monotony of erupted liquid compositions, these mineral analyses show striking compositional diversity.Fig. 4Major element compositions of minerals from Wolf and Fernandina.a, b Plagioclase An#, c, d clinopyroxene Mg#cpx and e, f olivine Mg#ol in lava and tephra samples from the 2015 Wolf eruption, nodule samples from the 1968 Fernandina eruption and lava samples from historic Fernandina eruptions. Crystals are classified according to their textural association (see legend). Clinopyroxene compositions from the 2015 Wolf eruption are from Stock et al.19 and crystal compositions from historic Fernandina lavas are from Allan and Simkin28 and Koleszar et al.35. Characteristic 2σ analytical uncertainties for our mineral analyses are either shown or are less than the size of a data point. The kernel density estimates above each panel show the distributions of the crystal compositions. The vertical grey lines show the compositions of crystals calculated to be in equilibrium with the 2015 Wolf tephra glass (solid lines—average composition; dashed lines—1σ compositional range)19 and 1968 Fernandina scoria glass28. Equilibrium calculations were performed at 1160 °C for Wolf and 1130 °C for Fernandina (the approximate pre-eruptive crystallisation temperatures)19, 28, using the models of Namur et al.67 for plagioclase, Putirka68 for clinopyroxene and Herzberg and O’Hara69 for olivine.In the Wolf samples, plagioclase An# (An# = atomic Ca/[Na + Ca + K]⋅100) ranges from 48.8 to 86.8. Tephra crystals and lava glomerocrysts extend to the lowest values (Fig. 4a). Regardless of textural association, plagioclase An# kernel density estimates (KDEs) are asymmetric, with the highest peaks at high anorthite contents (An# = 78–82) and long tails extending to lower An#. The Mg#cpx (Mg#cpx and Mg#ol = atomic Mg/[Mg + Fe*]⋅100, where Fe* is Fe2+ + Fe3+; see “Methods”) of phenocrystic and glomerocrystic clinopyroxene ranges from 71.8 to 84.6. The largest Mg#cpx KDE peak for the clinopyroxene phenocrysts is at 74.5, with a subsidiary peak at 82.3, whereas the largest peak for the glomerocrysts is at 82.5, with a short tail to lower Mg#cpx. The Mg#cpx KDE for clinopyroxene crystals from tephra samples is highly asymmetric, with a peak at 81.5 and crystal compositions extending down to Mg#cpx = 33.0 (Fig. 4c). Olivine crystals in the Wolf samples typically have high Mg#ol in the range 75.7–82.8; only two phenocryst analyses have lower Mg#ol (~72; Fig. 4e).In the Fernandina nodules, independent plagioclase grains (i.e. crystals that are not inclusions) have highly variable An# (33.7–83.0), with plagioclase inclusions and crystals bounding miarolitic cavities at intermediate compositions (54.8–76.5 and 57.4–69.8, respectively). Regardless of their textural association, plagioclase An# KDEs are non-Gaussian, with their largest peaks at similar An# (67.6–68.8) and tails towards albitic compositions. Plagioclase crystals in historic Fernandina lavas have similarly diverse An# (43.5–89.7), with the largest KDE peak at 63.5. The Mg#cpx of clinopyroxene crystals in the 1968 Fernandina nodules and historic lavas typically range ~64–82, with the largest KDE peaks at similar values (81.0 and 79.9, respectively; Fig. 4d). Olivine crystals in the nodule samples are stoichiometric but have low Mg#ol (51.7–63.8) with the largest KDE peak at 54.7. Olivine crystals in historic Fernandina lavas have been identified with very low Mg#ol (>37.8) but most analyses are in the range 73.9–86.7, with the largest KDE peak at 86.1 (Fig. 4f)28,35.Plagioclase minor element concentrations in our samples vary systematically with An#. K2O concentrations range from below detection limit (~0.02 wt%) to 0.38 wt% in the Wolf samples and from 0.12 to 1.22 wt% in the Fernandina nodules. K2O correlates negatively with An# in both eruptions, but the gradient is significantly steeper in the Fernandina samples. Plagioclase crystals from historic Fernandina lavas typically have lower K2O concentrations than in the 1968 nodules but overlap with the Wolf lavas (Fig. 5a). Plagioclase TiO2 concentrations are 0.04–0.22 wt% and 0.09–0.30 wt% in the Wolf samples and Fernandina nodules, respectively. In both cases, TiO2 correlates negatively with An# at high An# but positively with An# at low An#. In Wolf samples, the inflection occurs at An# ≈63 and TiO2 ≈0.19 wt%, whereas in Fernandina samples it is at lower An# (~57) and higher TiO2 (~0.25 wt%; Fig. 5b). FeO concentrations are typically highest at intermediate An#, with the lowest concentrations in the most albitic and anorthitic plagioclase analyses, although there is significant scatter (Fig. 5c). Plagioclase MgO concentrations in the Wolf samples define two populations, which each correlate negatively with An#: the first has high MgO concentrations (0.08–0.28 wt%) and includes crystals in all textural associations from lava and tephra samples; the second has lower MgO concentrations (0.03–0.10 wt%) and includes only a sub-set of crystals from the tephra samples. Plagioclase crystals in the Fernandina nodules typically have lower MgO contents than the Wolf lavas; only a few analyses extend up to 0.19 wt%. In contrast, most plagioclase analyses from historic Fernandina lavas have higher MgO contents that correlate with the high-MgO Wolf population (Fig. 5d).Fig. 5Plagioclase minor element compositions from Wolf and Fernandina.Minor elements vs. An# of plagioclase crystals in lava and tephra samples from the 2015 Wolf eruption (blue squares), nodule samples from the 1968 Fernandina eruption (red diamonds) and lava samples from historic Fernandina eruptions (open diamonds). Crystal compositions from historic Fernandina lavas are from Allan and Simkin28. Characteristic 2σ analytical uncertainties for our plagioclase analyses are shown or are less than the size of a data point. The lines in (d) show the MgO contents of plagioclase crystals calculated to be in equilibrium with liquids containing 1, 2, 4, 6 and 8 wt% MgO (at appropriate temperatures from Wolf Rhyolite-MELTS models at 300 MPa; black dashed lines) and with the average 2015 Wolf tephra glass at 1160 °C (the approximate pre-eruptive crystallisation temperature; solid grey line)19, using the MgO partitioning model of Nielsen et al.43. Insets show the general theoretical trends of crystal compositions during growth (i.e. fractional crystallisation; solid lines) and diffusive re-equilibration (dashed lines), after Humphreys39.In samples from the 2015 Wolf eruption, plagioclase crystals are typically either unzoned or show concentric oscillatory zoning. Some plagioclase glomerocrysts in lava samples have highly resorbed cores with very low An# and high K2O and MgO concentrations, overgrown by higher An# mantles (Figs. 3b, 6a). Additionally, a few crystals in the tephra samples have high-An# cores and normally zoned mantles that extend to low An# and high K2O; these crystals can have anorthite contents covering almost the full range identified in our Wolf samples (Fig. 6b). Some crystals in both lava and tephra samples have thin, normally zoned rims that show a small decrease in An#, accompanied by increases in K2O, MgO, FeO and TiO2 (Fig. 6a); other crystals have elevated FeO at their rims even at constant An#. Zoning in clinopyroxene crystals from the 2015 Wolf eruption was described by Stock et al.19, who identified rare patchy zoned crystals with low Mg#cpx and reverse zoned crystals with highly resorbed low-Mg#cpx cores in tephra samples. We have subsequently identified ilmenite inclusions within these crystals (Fig. 3d). However, most clinopyroxene crystals from the 2015 eruption are either unzoned or oscillatory zoned, occasionally with slightly higher Mg#cpx cores and/or sector zoning (Fig. 3e). Olivine is typically unzoned, but some crystals show normal zoning with decreasing Mg#ol towards their rims. Most plagioclase crystals in nodule samples from the 1968 Fernandina eruption show minor oscillatory zoning in An#, K2O, FeO and/or MgO, but a minority contain more significant normal and reverse zones. Many plagioclase grains have a normal zone at their rims, extending to low An# and high K2O (Fig. 6c) and FeO; these zones are typically thick (<100 µm) in isolated crystals. Clinopyroxene crystals are similarly unzoned or show slight oscillatory zoning and many have a thick (<250 µm) normally zoned rim, extending to low Mg#cpx. Olivine crystals in the Fernandina nodules are generally unzoned but, in contrast with other phases in our samples, some have reverse zoned rims, characterised by a small increase in Mg#ol.Fig. 6Selected plagioclase zoning profiles.Core-to-rim zoning profiles across a a glomerocrystic plagioclase crystal from a 2015 Wolf lava sample, b a plagioclase crystal from a 2015 Wolf tephra sample, and c a plagioclase crystal from a 1968 Fernandina nodule sample. MgO is typically below detection limit in (c). Characteristic 2σ analytical uncertainties are shown or are less than the size of a data point.Evaluating crystal−liquid equilibriaTo investigate the degree of compositional heterogeneity in sub-volcanic melts, we calculated the compositions of minerals that would have been in equilibrium with their carrier liquids (represented by tephra glasses from each eruption) and compared these with the compositional distributions of our mineral analyses (Fig. 4; see “Methods”). For Wolf, the average equilibrium plagioclase An# is 64.1 ± 1.7 (1σ of the range calculated from all available glass analyses), clinopyroxene Mg#cpx is 75.5 ± 0.9 and olivine Mg#ol is 75.2 ± 0.8. For Fernandina, the average equilibrium plagioclase An# is 63.3 (1σ cannot be determined as only one glass analysis is available), clinopyroxene Mg#cpx is 73.6 and olivine Mg#ol is 73.1. Most of our mineral analyses from the 2015 Wolf and 1968 Fernandina samples have higher An# and Mg# than predicted to be in equilibrium with their carrier liquids (Fig. 4). This is typical of ocean island volcanoes globally, where magmas evolve through fractional crystallisation, precipitating minerals with progressively lower An# and Mg# until the ascending carrier liquid entrains earlier-formed crystals and brings them to the surface36. However, many of the mineral compositions in our samples extend to much lower An# and Mg# values than predicted to be in equilibrium with their carrier liquids. These highly evolved compositions are found in the interiors of plagioclase and clinopyroxene crystals from Wolf tephra, plagioclase glomerocrysts from Wolf lavas, and all mineral phases analysed in the Fernandina nodules. Evolved compositions have also been measured in rare olivine, plagioclase and clinopyroxene crystals from historic Fernandina lavas28. Typically, they form long tails in non-Gaussian An# and Mg# KDEs, suggesting that crystals grew or equilibrated with a range of variably evolved melts (Fig. 4). Hence, the crystal cargos of the 2015 and 1968 eruptions record highly heterogeneous liquids in the Wolf and Fernandina sub-volcanic systems, despite the volcanoes erupting monotonous melts at the surface over long timescales.Characterising equilibrium melt compositionsDiffusive re-equilibration of An# (i.e. CaAl–NaSi interdiffusion) and highly charged cations (e.g. Ti4+) is very slow in plagioclase, causing minerals to retain their original compositions over millennia, even at magmatic temperatures (insets Fig. 5)37,38. Hence, we interpret plagioclase TiO2 variations in our Wolf and Fernandina samples as reflecting compositional changes in their host liquids at the time of crystallisation. By comparing models of plagioclase TiO2 concentrations during fractional crystallisation with our analyses from Galápagos, we can constrain the composition of their equilibrium liquids and quantify the extent of heterogeneity in sub-volcanic melts39–41. Our modelling approach calculates plagioclase major element and residual liquid compositions during isobaric cooling and fractional crystallisation using Rhyolite-MELTS42, and computes equilibrium plagioclase TiO2 concentrations at each temperature step using the temperature- and anorthite-dependent partitioning model of Nielsen et al.43 (DTi = 0.03–0.10; Supplementary Dataset 7; see “Methods”). Comparing natural plagioclase analyses to the Rhyolite-MELTS outputs provides the major element composition, and hence an estimate of the physical characteristics (density and viscosity), of their equilibrium liquids.The trajectory of plagioclase TiO2 vs. An# predicted by our models matches the compositional trend defined by crystals from Wolf and Fernandina (Figs. 7, 8). We ran the models over a range of pressures, guided by independent petrological and geophysical constraints on magma storage depths at Wolf and Fernandina19,21,32,34. The model outputs are relatively insensitive to pressure but taken at face value the best fits to the natural data are at 300 MPa for Wolf and 500 MPa for Fernandina, in agreement with previous estimates of the main pressures of magma storage19,34. Simulated crystal compositions underpredict the TiO2 content of our very lowest An# crystals from Fernandina, potentially due to a paucity of experimental data for these compositions43. Both Wolf and Fernandina models accurately predict the inflection where TiO2 transitions from correlating negatively with An# to correlating positively. In our models, these inflections occur when ilmenite comes onto the liquidus, suggesting that crystals with An# ≲57–63 grew from ilmenite-saturated melts, despite the absence of ilmenite as a phenocryst phase in any of the western Galápagos basaltic shield volcanoes. Ilmenite inclusions are, however, found within the most evolved (lowest Mg#cpx) clinopyroxene cores from the 2015 Wolf eruption (Fig. 3d), validating the model predictions.Fig. 7TiO2 in plagioclase model for the 2015 Wolf eruption.a Plagioclase TiO2 vs. An# in lava and tephra samples from the 2015 Wolf eruption. Crystals are classified according to their textural association (see legend). Characteristic 2σ analytical uncertainty is shown for TiO2 and is less than the size of a data point for An#. The grey lines show the compositions of crystals calculated to be in equilibrium with the 2015 Wolf tephra glass (solid lines—average composition; dashed lines—1σ compositional range)19, using the models of Namur et al.67 for An# and Nielsen et al.43 for TiO2 at 1160 °C (the approximate pre-eruptive crystallisation temperature)19. The black lines show the trajectory of plagioclase compositional evolution calculated using Rhyolite-MELTS and the TiO2 partitioning model of Nielsen et al.43 at 50 MPa (solid line), 300 MPa (dashed line) and 500 MPa (dotted line). Plagioclase comes onto the liquidus at 1214, 1229 and 1244 °C in 50, 300 and 500 MPa models, respectively. The plots above show b the physical and c the compositional evolution of the liquid predicted by Rhyolite-MELTS at 300 MPa (the approximate crystallisation pressure of the main magma storage zone at Wolf)19.Fig. 8TiO2 in plagioclase model for the 1968 Fernandina eruption.a Plagioclase TiO2 vs. An# in nodule samples from the 1968 Fernandina eruption and lava samples from historic Fernandina eruptions. Crystals are classified according to their textural association (see legend). Crystal compositions from historic Fernandina lavas are from Allan and Simkin28. Characteristic 2σ analytical uncertainty for our plagioclase analyses is shown for TiO2 and is less than the size of a data point for An#. The grey lines show the compositions of crystals calculated to be in equilibrium with the 1968 Fernandina scoria glass28, using the models of Namur et al.67 for An# and Nielsen et al.43 for TiO2 at 1130 °C (the approximate pre-eruptive crystallisation temperature)28. The black lines show the trajectory of plagioclase compositional evolution calculated using Rhyolite-MELTS and the TiO2 partitioning model of Nielsen et al.43 at 50 MPa (solid line), 300 MPa (dashed line) and 500 MPa (dotted line). Plagioclase comes onto the liquidus at 1191, 1207 and 1222 °C in 50, 300 and 500 MPa models, respectively. The plots above show b the physical and c the compositional evolution of the liquid predicted by Rhyolite-MELTS at 500 MPa (the approximate crystallisation pressure at Fernandina)34.Our models suggest that the most evolved natural plagioclase crystals (i.e. lowest An#) from Wolf and Fernandina grew from plagioclase- and clinopyroxene-saturated basaltic trachy-andesitic melts with 2.46 wt% MgO (Mg#liq = 37.5; at 300 MPa) and trachy-andesitic melts with 1.08 wt% MgO (Mg#liq = 25.1; at 500 MPa), respectively (Figs. 7, 8). These melt compositions are substantially more evolved than any known eruptive rocks from Wolf or Fernandina (Fig. 2). In fact, the presence of resorbed quartz in tephra from the 2015 eruption suggests that our plagioclase data may not capture the full compositional range of sub-volcanic liquids because quartz does not saturate in our models until melts reach dacitic-to-rhyolitic compositions at low temperatures (<830 °C). Quartz is not a phenocryst or groundmass phase in the rhyodacites of Alcedo or Rabida volcanoes (or any other Galápagos lava that has been inspected)8,31, but it is present in Rabida crustal xenoliths44. Hence, our data suggest that Galápagos volcanoes that have erupted monotonous basaltic lava over millennial timescales contain highly evolved, silicic melts within their sub-volcanic systems that have not previously been identified from material erupted at the surface.Diffusive re-equilibrationIron and Mg diffuse faster than Ti in plagioclase, and initial crystal compositions for these elements are hence more likely to be overprinted by diffusive re-equilibration38,39. In our samples, Fe shows significant scatter (Fig. 5) and is often enriched at crystal rims where An# and other elements remain constant. We believe that these apparently high-Fe rims likely reflect secondary fluorescence induced during electron probe microanalysis (EPMA) and are not of geological significance45.Magnesium is negligibly affected by secondary fluorescence45 and if the Mg content of Wolf and Fernandina plagioclase were controlled by changes in melt chemistry during fractional crystallisation we would anticipate a positive correlation between An# and MgO (inset Fig. 5c). Our data, however, define three populations, each with nearly constant MgO or a negative correlation between An# and MgO (Fig. 5). This relationship is consistent with diffusive re-equilibration where MgO concentrations reflect lower solid−liquid Mg partition coefficients in more anorthitic plagioclase39,43. To determine the compositions of the liquids with which these plagioclases equilibrated, we modelled crystal compositions in equilibrium with liquids of varying MgO concentration, using temperatures from Rhyolite-MELTS and the MgO partitioning model of Nielsen et al.43 (DMg = 0.02–0.05). We find that most crystals from the 1968 Fernandina nodules are consistent with having re-equilibrated with melts containing 1–2 wt% MgO, similar to the most evolved magma predicted from our Ti modelling (Figs. 5d, 8). Although crystals from the 2015 Wolf eruption do not preserve an MgO fractional crystallisation trend, they are also not in equilibrium with any melt at fixed MgO (Fig. 5d). We interpret this as recording disequilibrium, whereby the higher MgO population have partially re-equilibrated with melts analogous to the carrier liquid (>4 wt% MgO) and the tephra crystals with lower MgO contents are partially equilibrated with more evolved melts (~2–3 wt% MgO; Fig. 5d). Some crystals contain zones with slightly elevated MgO close to their rims, potentially due to intermittent growth from more primitive melts or boundary layer effects during rapid crystallisation (e.g. Fig. 6a, c)46.Olivine crystals in our Fernandina nodule samples have ubiquitously low Mg#ol, consistent with growth or re-equilibration with evolved liquids. They are typically unzoned, but a minority of crystals have reverse rims (in contrast with normal plagioclase and clinopyroxene rim zones). As Fe–Mg interdiffusion in olivine is geologically fast47, this could reflect either growth or diffusive re-equilibration with more primitive liquids on very short pre-eruptive timescales. Although a sub-set of plagioclase crystals in Wolf tephra samples have similarly low MgO contents, no low-Mg#ol olivine has been identified in these samples.Storage depths and origin of Galápagos evolved liquidsWe find evidence for evolved liquids in different types of sample (lava, tephra, nodules) and crystal associations (e.g. tephra crystals, glomerocrysts) from Wolf and Fernandina volcanoes. In our Wolf lava samples, low-An# plagioclase crystals that grew from basaltic trachy-andesitic melts are in the same glomerocrystic aggregates as pyroxenes that crystallised at ~300 MPa (from clinopyroxene-melt barometry, ±140 MPa standard error of estimate)19; this is consistent with glomerocrysts being sourced from a magma storage region at >6.1–8.8 km 19 and provides strong evidence for evolved liquids in the lower crust. However, albitic plagioclase, low-Mg#cpx pyroxenes and resorbed quartz crystals are also present in tephra from the 2015 eruption. Stock et al.19 interpret the tephra as deriving from an upper crustal storage region (identified geophysically at ~1 km depth), based on it being the first material to have erupted and having a crystal cargo distinct from that of later lavas. Similarly, the Fernandina nodules contain evolved olivine and feldspar crystals, as well as miarolitic cavities and evidence of alteration by an active hydrothermal system, which demonstrate that they formed in an upper crustal storage region, at a similar level to a shallow geophysical source at ~1 km depth21,48. Hence, our data qualitatively indicate that evolved liquids occur at a variety of depths beneath Galápagos volcanoes.Rhyolite-MELTS models show that the viscosity of Galápagos melts increases significantly after ilmenite saturation, accompanied by a reduction in density (Figs. 7b, 8b). Magma buoyancy alone is often insufficient to drive eruptions49, and the low H2O content of Galápagos primary melts50 would delay crystallisation-induced volatile saturation (i.e. second boiling) and the generation of volatile overpressure until very low melt fractions (e.g. refs. 51,52). Hence, we suggest that once ilmenite-saturated magmas stall and stagnate in the crust beneath Galápagos volcanoes, they become highly viscous (due to their high melt viscosities and crystallinities) and are then unlikely to ascend further without external influence (e.g. mafic recharge)53 or until they have undergone substantial crystallisation.Basalt flushing beneath monotonous volcanoesAlthough crystal compositions provide unequivocal evidence of highly variable melts in Galápagos sub-volcanic systems, such heterogeneity is not reflected in the geochemistry of monotonous lavas erupted at the surface. To determine the fate of these magmas, we studied the petrographic contexts of evolved mineral zones. Plagioclase and clinopyroxene crystals in our samples have diverse textures, with crystals from Wolf lavas and tephra including both normal and reverse zoning (Figs. 3d, e, 6a, b), and crystals from Fernandina nodules containing oscillatory normal and reverse zones, often with the lowest An# and Mg#cpx at their rims. Although the scarcity of crystals with evolved zones inhibits robust characterisation of populations, these textures indicate open systems, with intermittent interactions between primitive and evolved melts54. Glomerocrysts in the 2015 Wolf eruption derive from disaggregated sub-volcanic mush19 and some glomerocrystic plagioclase grains contain evolved (low An#) cores enclosed by fully concentric primitive (high An#) mantles, consistent with mafic recharge before the crystals were incorporated into a cumulate pile (i.e. while they were surrounded by melt; Fig. 3b). Mush accumulation likely occurred over long timescales and this textural evidence, along with diffusive re-equilibration of plagioclase MgO contents, suggests that some mixing events recorded in zoned crystals significantly pre-date eruption. Furthermore, as clinopyroxene-melt barometry indicates that glomerocrysts are derived from the lower crust19, at least some of this mixing likely occurred at depth.Given the petrographic evidence for open-system behaviour, we constructed K2O/TiO2 vs. Mg#liq mixing models to determine the proportions of primitive and evolved melts in the monotonous lavas erupted at Wolf and Fernandina (Fig. 9). This ratio is useful because it increases dramatically over a short crystallisation interval after ilmenite saturation and is unaffected by plagioclase accumulation. The evolved end-member in our models is taken as the melt calculated to be in equilibrium with our lowest An# plagioclase crystals from Wolf and Fernandina (from Rhyolite-MELTS) and the primitive end-members are the highest and lowest Mg#liq erupted liquid or melt inclusion from each volcano (some higher Mg#liq whole-rock analyses from Fernandina are excluded as they have accumulated olivine and are not true liquids; Supplementary Table 1). Figure 9 shows that all the liquids (whole-rocks and glasses) erupted at Wolf have low K2O/TiO2 ratios, approximately along the liquid line of descent, and contain <10% of the evolved magma. Most liquids erupted at Fernandina also have low K2O/TiO2 ratios and contain <10% of the evolved end-member. However, Geist et al.34 identified a small number of submarine lavas on the southwest flank of the volcano that have more evolved compositions and are derived from the lower crustal storage region. These evolved series lavas have a restricted range of Mg#liq (~40–50), suggesting that their spatially related vents were fed by compositionally similar primitive liquids ascending from depth, but have elevated K2O/TiO2 ratios and may contain up to ~50% of the evolved magma. Our estimates of the amount of evolved material in erupted magmas are probably maxima, as quartz crystals in our tephra samples suggest that magmas fractionated beyond the extent recorded by our plagioclase analyses; real evolved end-member melts may have significantly higher K2O/TiO2 ratios than those in our models.Fig. 9Mixing models between evolved and primitive end-member liquids.K2O/TiO2 vs. Mg#liq of erupted liquids at a Wolf volcano and b Fernandina. The K2O/TiO2 ratio of evolving liquids increases significantly when ilmenite (ilm) comes onto the liquidus. The points show: whole-rock, tephra glass, submarine glass and melt inclusion literature data from all volcanoes in the western Galápagos Archipelago (excluding intrusive rocks and plagioclase-ultraphyric lavas); historic erupted liquids (i.e. whole-rocks, tephra glasses and submarine glasses) from Wolf and Fernandina; and whole-rock and tephra glass data from the 2015 Wolf and 1968 Fernandina eruptions (see legend). References for Wolf and Fernandina historic erupted liquids and literature data from all western Galápagos volcanoes are in Supplementary Note 2. Glass data for the 2015 Wolf eruption are from Stock et al.19 and whole-rock data are from this study. Glass and whole-rock data for the 1968 Fernandina eruption are from Allan and Simkin28. Characteristic 2σ analytical uncertainties for our whole-rock analyses are less than the size of a data point. The black lines show liquid lines of descent predicted by Rhyolite-MELTS at 50 MPa (solid line), 300 MPa (dashed line) and 500 MPa (dotted line). The solid grey lines show mixing models between evolved and primitive liquids, contoured by increasing proportions of the evolved end-member (dashed grey lines). For both volcanoes, the evolved (high K2O/TiO2) end-members are the liquids calculated to be in equilibrium with our lowest An# plagioclase crystals using Rhyolite-MELTS (at 300 MPa for Wolf and 500 MPa for Fernandina; Supplementary Table 1). The primitive (low K2O/TiO2) Wolf end-members are the highest (whole-rock W9562) and lowest (glass D4A) Mg#liq liquids measured in historic eruptions by Geist et al.17. The primitive Fernandina end-members are the highest (melt inclusion D25C-2-34) and lowest (glass D30-A) Mg#liq normal series liquids (i.e. excluding whole-rock samples that include accumulated olivine) measured in historic eruptions by Koleszar et al.35 and Geist et al.34, respectively (Supplementary Table 1).Our petrological data and models show that evolved melts exist in Galápagos sub-volcanic systems and periodically mix with more primitive melts. However, they typically comprise <10% of the magma erupted at the surface. We thus suggest that their absence from the erupted record is due to mass-balance, whereby large volumes of basaltic magma ascending through the crust from a primitive lower crustal storage region interact with much smaller quantities of more evolved, heterogeneous magma at higher levels (Fig. 10). The liquids mix, but because of the disparity in their relative proportions, the basaltic component dominates the mass-balance and its composition remains almost unaltered; only evolved crystals that were not fully resorbed during the mixing event preserve a record of the earlier heterogeneity. Hence, monotonous activity does not reflect simplicity or chemical homogeneity in magmatic systems. Instead, just as CO2 flushing through the crust from depth can impact the composition of magmatic volatiles at higher levels55, we suggest that monotonous volcanism reflects large volumes of basalt flushing through the crust from depth reacting with and overwhelming smaller volumes of more evolved, heterogeneous material at shallower levels (Fig. 10).Fig. 10Cartoon summarising the architecture of the Wolf and Fernandina plumbing systems.The grey bars (left) show relative estimates of magma storage depths at Wolf and Fernandina from Interferometric Synthetic Aperture Radar (InSAR) inversions and petrological barometry19,21,32,33. Not to scale. a The volcanoes are underlain by large lower crustal magma storage regions and smaller shallow storage regions, which are recharged by new magma ascending from depth. Both regions contain compositionally heterogeneous melts. We cannot discount mid-crustal magma storage but there is no petrological or geophysical evidence from recent eruptions. At Fernandina, evolved series submarine eruptions are periodically fed by melts ascending directly from the deeper storage region. b Large volumes of basaltic melt periodically ascend from depth, flushing through the systems and mixing with liquids stored at shallower levels. Some sub-volcanic mush is entrained into the ascending liquids. As the volume of ascending basalt is much greater than the volume of evolved material at shallower levels, its composition remains almost unaltered; only crystals that were not fully resorbed during mixing preserve evidence of the earlier heterogeneity.A large body of theory considers basaltic injections into silicic magma reservoirs, largely in the context of driving large-volume explosive eruptions (e.g. refs. 56,57). In contrast, the hybridised monotonous basalts of Galápagos erupt in a Hawaiian or Strombolian style, despite originating from mixing between basaltic and rhyolitic magmas. We note that the 2018 lower-rift zone eruption of Kīlauea volcano (Hawai’i) also involved injection of basalt into a silicic crustal reservoir, yet the eruption was overwhelmingly effusive58. The critical factor in controlling mixing dynamics must therefore be the proportion of mafic and silicic magmas59: in both the Galápagos and Hawai’i cases, the volumes of basalt were substantially greater than those of the resident silicic magmas.Basalt flushing will only be able to maintain monotonous eruptions while the melt flux from the lower crust is high. When the melt flux is lower or a thermally stable lower crustal storage region is yet to develop, there may be insufficient volumes of primitive melt moving through the crust to fully overprint heterogeneity at higher levels. Model results indicate that a magma flux >1 ⋅ 10−4 km3 year−1 is required to stabilise a super-solidus crustal mush zone60,61. This less than the long-term eruptive fluxes at Wolf (0.4–1 ⋅ 10−3 km3 year−1)17 and Fernandina (4.4 ⋅ 10−3 km3 year−1)25, but not in the eastern Galápagos Archipelago where eruptive fluxes are several orders of magnitude lower26. Hence, our findings support a model8 whereby the diversity of erupted products at Galápagos volcanoes is dictated by their position relative to the centre of the hotspot (and thus the melt flux). More broadly, our findings have implications for volcano monitoring, suggesting that even volcanoes that have reliably erupted basaltic lavas for millennia can contain evolved liquids in their sub-volcanic systems. Although basalt flushing can maintain monotonous eruptions over long timescales, external influences (e.g. changes in the regional stress field) or a decrease in the crustal melt flux (e.g. resulting in further fractionation and an increased likelihood of second boiling) might allow these melts to ascend, generating explosive silicic eruptions.MethodsSample selection and preparationSamples from the 2015 Wolf eruption include the lava and reticulitic tephra analysed by Stock et al.19, plus additional lava samples from the east flank of the volcano collected during fieldwork in June 2017 (Fig. 1b). All the material was unaltered and lava samples were collected from dense flow interiors where possible. Samples from the 1968 Fernandina eruption are gabbroic nodules collected on the floor and rim of the caldera during fieldwork in July 1970 (Fig. 1c) and are described by Howard et al.48. Most of the nodules are fresh, but some show extensive hydrothermal modification48. The Wolf lava samples and Fernandina nodule samples were prepared as polished thin sections. Crystals in the Wolf tephra samples were separated from the 40–500 µm size fraction by heavy liquid and magnetic separation and mounted in epoxy along with the quenched glass.Analytical methodsWhole-rock samples from the 2015 Wolf eruption were analysed by X-ray fluorescence spectrometry (XRF) for major and trace elements. Analyses were performed using a Philips PW 2404 instrument at the University of Edinburgh (UK) following the method of Fitton et al.62 with modifications by Fitton and Godard63. Analytical precision, encompassing errors associated with sample preparation and heterogeneity, was estimated by preparing and analysing three replicates of the same sample. Relative 1σ precision is better than ±1% for major elements (>1 wt%) and better than ±2% for minor and trace elements (<1 wt%), except Th (±8%) and Pb (±20%).Mineral compositions were measured by EPMA using Cameca SX100 (for Wolf samples) and Cameca SXFive (for Fernandina samples) instruments in the Departments of Earth Sciences at the University of Cambridge (UK) and Syracuse University (USA), respectively. To ensure consistency across instruments and multiple analytical sessions, measurements were internally calibrated using appropriate Smithsonian Microbeam Standards64. Relative 1σ precision was monitored by repeat analysis of mineral standards on the Cambridge instrument and is assumed to be similar on the Syracuse instrument. This is better than ±2% for major elements and ±15% for minor elements, except MnO in clinopyroxene (±17%). Typical 1σ relative errors for each element analysed by XRF and EPMA are provided in Supplementary Datasets 1–6 and full details of the analytical methods are provided in Supplementary Note 3.We assume a melt Fe2+/Fe* (Fe* = Fe2+ + Fe3+) value of 0.85 when calculating Mg#liq. This equates to an oxygen fugacity (fO2) near the quartz-fayalite-magnetite (QFM) buffer, as measured in Galápagos lavas by Peterson et al.65. We do not assume an Fe speciation for minerals, instead calculating the number of Fe2+ and Fe3+ ions per formula unit by stoichiometry and using the sum of these (Fe*) in our calculations of Mg#cpx and Mg#ol. We evaluate the distribution of our major element datasets using KDEs, with bandwidths calculated after Sheather and Jones66.Modelling crystal–liquid equilibriaEquilibrium plagioclase, clinopyroxene and olivine compositions were calculated using the models of Namur et al.67, Putirka68, and Herzberg and O’Hara69, respectively, given pre-eruptive storage temperatures of 1160 °C for Wolf and 1130 °C for Fernandina19,28. The model of Namur et al.67 is calibrated for anhydrous melt compositions; H2O in the melt would increase the equilibrium An# but this effect is assumed to be negligible as Galápagos magmas have consistently low H2O contents (typically < 1 wt%)35,50.Modelling TiO2 in plagioclasePlagioclase major element and residual liquid compositions were calculated during isobaric cooling and fractional crystallisation using Rhyolite-MELTS42. We use primitive whole-rock (W9562)17 and melt inclusion (D25C-2-34)35 compositions as our Wolf and Fernandina model starting liquids, respectively. Equilibrium plagioclase TiO2 concentrations were then calculated at each temperature step using the partitioning model of Nielsen et al.43 (Supplementary Dataset 7). By comparing modelled plagioclase An# and TiO2 concentrations with our natural mineral analyses, we can characterise their equilibrium melt compositions. This method of characterising mineral-melt relationships is preferable to that of Scruggs and Putirka70 where the eruptive record does not encompass the full range of melt compositions in a sub-volcanic system.Models were run at 50, 300 and 500 MPa, an fO2 at the QFM buffer and 0.15 wt% initial H2O, which approximate the range of Galápagos magma storage conditions identified in previous studies19,27,28,50,65. The liquid lines of descent predicted by Rhyolite-MELTS fractional crystallisation models are generally a good fit to previously analysed whole-rock and glass data from Wolf and Fernandina (Supplementary Figs. 1, 2) and simulated clinopyroxene compositions also match natural crystal compositions (Supplementary Fig. 3). This validates our choice of intrinsic variables and suggests that assimilation of compositionally distinct wall rock material has a negligible impact on the compositional evolution of Galápagos magmas, in agreement with previous studies27.Supplementary information
Supplementary Information
Peer Review File
Descriptions of Additional Supplementary Files
Supplementary Data 1
Supplementary Data 2
Supplementary Data 3
Supplementary Data 4
Supplementary Data 5
Supplementary Data 6
Supplementary Data 7 | nature communications | [
"Article"
] | [
"Geochemistry",
"Geology",
"Petrology",
"Volcanology"
] | by complex magmatic systems kilometres crust from Moho to near-surface1,2 melts ascend undergo processes crystallisation volatile exsolution assimilation mixing between magma batches interaction with mush processes modify magma compositions diversity in chemistry igneous rocks magmatic systems exhibit monotonous volcanic behaviour erupting chemically homogeneous liquids over long timescales to millennia causes monotonous volcanism understood constraining dynamics systems essential for longevity potential future changes eruptive behaviour hazardous models monotonous eruption sequences involve uniformity by reactive filtration8 products uniformity by processing under same first model requires hot melts to cool react through super-solidus mush column buffering temperature products monotonous eruption contain multiple macrocryst populations complex growth histories Geophysical geobarometric constraints crustal structure volcanoes inconsistent with large storage regions9 second model includes repeated recharge of upper crustal magma by consistent melts ascending from maintaining crustal system narrow temperature compositional range7monotonous activity necessitates thermal balance between heat ascending magma lost advection eruption western Galápagos Archipelago ideal monotonous volcanism hosts volcanoes erupted near-homogeneous basaltic magmas include Fernandina volcano Wolf Darwin Ecuador Sierra Negra volcanoes Isabela Island basalts during monotonous Galápagos eruptions olivine clinopyroxene plagioclase crystallisation8 historic lavas have pre-eruptive storage temperatures ~22–30 °C17 Monotonous volcanoes on Fernandina Isabela near centre inferred Galápagos plume29 uniformity in erupted products related to flux mantle-derived magma entering crust volcanoes receive thermal input thick steady-state gabbroic mush zones mid-crust interact with ascending magmas temperature diverse Galápagos volcanoes lack large mush zones Cerro Azul proximal erupts basalts greater MgO concentrations insufficient magma input stable system Alcedo downstream produced dacitic rhyolitic eruptions mush meltsHarpp extend geographic trend to volcanoes eastern archipelago primitive variable eruptions reflect absence magma plumbing systems limited crustal processing.Fig. 1Locations Wolf Fernandina volcanoes sampling sites Regional map Galápagos Archipelago volcanic centres Isabela Island maps (b) (c). map Wolf volcano 2015 circumferential fissure intra-caldera vent 2015 lava flow extent red Bernard sampling locations lavas tephra shown green circles blue diamonds map Fernandina pyroclastic flow total tephra 1968 eruption Howard sampling locations nodules dark red Contours 200 m Galápagos eruptions constraints magma storage good targets understanding processes homogenising erupted liquids study microanalyses mineral phases constrain magmatic processes Wolf Fernandina volcanoes erupted monotonous basaltic melts millennia focus petrographic geochemical observations pre-eruptive compositional heterogeneity with thermodynamic models range magmas sub-volcanic systems sub-volcanic process diversity erupted liquidscomparing Wolf Fernandina Galápagos volcanoes transition monotonous varied volcanism. 2Erupted liquid compositions Galápagos Archipelago.TiO2 Mg#liq liquids Wolf volcano Fernandina points show whole-rock tephra glass submarine glass volcanoes western Galápagos Archipelago historic erupted liquids Wolf Fernandina 2015 Wolf 1968 Fernandina eruptions References Wolf Fernandina historic erupted liquids data Supplementary Note 2. Glass data 2015 Wolf eruption Stock et al.19 whole-rock study 1968 Fernandina eruption Allan Simkin28 2σ analytical uncertainties whole-rock analyses less than size data point black lines show liquid lines descent calculated Rhyolite-MELTS 50 MPa 300 500 MPa kernel density estimates show Mg#liq distribution erupted liquids Wolf Fernandina red points median Mg#liq bars Mg#liq interquartile range grey bars show Mg#liq interquartile range historic lavas Santiago26Results discussionSamples basaltic lava reticulitic tephra 2015 Wolf eruption gabbroic nodules 1968 Fernandina eruption Supplementary Note 1 selected petrological constraints storage depths sub Wolf lavas circumferential fissure phase contain 50 μm euhedral–subhedral plagioclase~5 clinopyroxene~2 olivine<1 macrocrysts microcrystalline groundmass macrocrysts associations isolated phenocrysts groundmass glomerocrystic aggregates plagioclase clinopyroxene olivine plagioclase aggregates synneusis Wolf tephra initial explosion same macrocryst assemblage anhedral quartz low (≪1 vol% clinopyroxene crystals ilmenite inclusions Fernandina nodules exhumed hydromagmatic paroxysm phaneritic texture slow cooling gabbroic mineral euhedral plagioclase (60–70 subhedral clinopyroxene (≤20 olivine (2–15 vol.%) Plagioclase grains inclusions-crystallising phases Miarolitic cavities<2 mm diameter present clinopyroxene void wallsHydrothermally altered samples exhibit secondary pyrite replacement olivine epidotisation magmatic minerals (Fig. evidence heterogeneous liquids multi magma storage Quantitative Evaluation Minerals map Ca map glomerocryst 2015 Wolf lava sample values show An# plagioclase zones False images quartz clinopyroxene crystals samples 2015 Wolf eruption Stock values show Mg#cpx clinopyroxene zones images nodule samples 1968 Fernandina eruption phaneritic texture miarolitic cavity hydrothermal alteration An# Mg#cpx values averages crystal cores rims Ol olivine Cpx clinopyroxene Pl plagioclase Qtz quartz Gl glass Ilm ilmenite Py pyrite MC miarolitic cavity.Homogeneous erupted whole-rock matrix glass analyses Wolf Fernandina eruptions variability erupted liquid compositions (Fig 2 Note 2 Mg#liq interquartile ranges subaerial lavas Wolf Fernandina 4.85 2.86lavas from volcanoes contain accumulated plagioclase decreases whole-rock Mg#liq small lavas Fernandina accumulated olivine increases bulk Mg#liq17 submarine lavas southwest flank Fernandina have evolved (low MgO high K2O Fernandina Mg#liq interquartile range to 5.56 (median = 52.8). Mg#liq interquartile ranges eruptions at Wolf Fernandina limited reflect homogeneity ranges 12.7 at Santa Cruz 21.4 at Santiago eastern Galápagos lavas represent thousands of years eruptive analysed 18 whole-rock samples from 2015 Wolf eruption significant compositional variability bulk samples average Mg#liq 53.0 0.23 higher than 2015 tephra glass 45.4 fewer data 1968 Fernandina eruption scoria glass lower Mg#liq than bulk lava evolved glass compositions reflect minor pre-eruptive crystallisation of bulk magmas Whole-rock analyses from 2015 Wolf eruption similar Al2O3 contents to tephra glass historic erupted liquids no significant plagioclase accumulationlavas 1968 Fernandina eruption elevated Al2O3 concentrations accumulated plagioclase matrix glass whole-rock Mg#liq 2015 1968 interquartile ranges liquids Wolf Fernandina mineral collected ~800 plagioclase ~90 olivine analyses 2015 Wolf eruption integrated with ~500 clinopyroxene analyses Stock et al.19 collected ~160 plagioclase ~100 clinopyroxene ~80 olivine analyses 1968 Fernandina nodules with analyses historic Fernandina mineral analyses show compositional diversity compositions minerals Wolf Fernandina Plagioclase clinopyroxene olivine Mg#ol in lava tephra samples 2015 Wolf nodule 1968 Fernandina lava Fernandina Crystals classified textural association Clinopyroxene compositions 2015 Wolf eruption Stock et al.19 crystal compositions historic Fernandina Allan Simkin28 Koleszar et al.35 2σ uncertainties analyses or less than size data point kernel density estimates show distributions crystal compositionsvertical grey lines show crystals 2015 Wolf tephra glass 1968 Fernandina scoria Equilibrium calculations 1160 °C Wolf 1130 °C Fernandina pre-eruptive crystallisation Namur et al.67 plagioclase Putirka68 clinopyroxene Herzberg O’Hara69 olivine Wolf samples plagioclase An# K ranges 48.8 to 86.8. Tephra crystals lava glomerocrysts extend lowest values. plagioclase An# kernel density estimates asymmetric highest peaks high anorthite contents 78–82) long tails lower An# Mg#cpx Fe* glomerocrystic clinopyroxene ranges 71.8 to 84.6. largest Mg#cpx peak clinopyroxene phenocrysts 74.5 subsidiary peak 82.3 glomerocrysts 82.5 short tail lower#cpx Mg#cpx KDE clinopyroxene crystals tephra asymmetric peak 81.5 compositions Mg = 33.0 (Fig. Olivine crystals Wolf samples high Mg#ol 75.7–82.8 two phenocryst analyses lower Mg#olFernandina nodules plagioclase grains variable An# inclusions miarolitic cavities intermediate compositions (54.8–76.5 57.4–69.8 plagioclase An# KDEs non-Gaussian largest peaks (67.6–68.8) tails albitic compositions Plagioclase crystals lavas diverse An# (43.5–89.7) largest KDE peak 63.5. Mg#cpx clinopyroxene crystals range ~64–82 largest KDE peaks (81.0 79.9 Olivine crystals stoichiometric low Mg#ol (51.7–63.8) largest KDE peak 54.7. low Mg#ol (>37.8) 73.9–86.7 largest KDE peak 86.1 minor element concentrations vary with An# K2O concentrations to 0.38 Wolf samples 0.12 to 1.22 Fernandina K2O correlates negatively with An# gradient steeper in Fernandina samples crystals lower K2O concentrations 1968 overlap with Wolf lavas Plagioclase TiO2 concentrations 0.04–0.22 0.09–0.30 wt% in Wolf samples Fernandina nodulesTiO2 correlates negatively high positively low Wolf samples inflection An# ≈63 TiO2 ≈0.19 wt%, Fernandina lower An# (~57) higher TiO2 (~0.25 wt% FeO concentrations highest intermediate An# lowest albitic anorthitic plagioclase analyses significant (Fig. 5c). Plagioclase MgO concentrations Wolf samples define two populations negatively An# first high MgO concentrations (0.08–0.28 wt%) second lower MgO concentrations (0.03–0.10 wt%) crystals crystals Fernandina nodules lower MgO few up to 0.19 wt% historic Fernandina lavas higher MgO high-MgO Wolf population (Fig. 5d).Fig 5Plagioclase minor element compositions Wolf Fernandina plagioclase crystals lava 2015 Wolf eruption 1968 Fernandina Crystal compositions historic Fernandina lavas Allan Simkin28 2σ analytical uncertainties plagioclase analyses less than size data pointlines show MgO contents plagioclase crystals liquids 1 2 4 6 8 wt% MgO Wolf Rhyolite-MELTS models 300 MPa average 2015 Wolf tephra glass 1160 °C pre-eruptive crystallisation MgO partitioning model Nielsen al show trends crystal compositions growth crystallisation diffusive re-equilibration Humphreys39 2015 Wolf eruption plagioclase crystals unzoned concentric oscillatory zoning glomerocrysts lava resorbed cores low An# high K2O MgO concentrations overgrown higher An# mantles (Figs. 3b crystals tephra high-An# cores zoned mantles low An# high K2O anorthite contents full range samples thin zoned rims decrease An# increases K2O MgO FeO TiO2 other elevated FeO rims constant An# Zoning clinopyroxene crystals 2015 Wolf eruption described Stock et al patchy zoned crystals low Mg#cpx resorbed low-Mg#cpx cores tephra identified ilmenite inclusionsclinopyroxene crystals 2015 eruption unzoned or oscillatory zoned occasionally higher Mg#cpx cores sector zoning (Fig. Olivine typically unzoned some normal zoning decreasing Mg#ol rims plagioclase crystals 1968 Fernandina eruption show minor oscillatory zoning in An# K2O FeO MgO minority contain normal reverse zones plagioclase grains normal zone at rims low An# high K2O FeO zones thick (<100 μm) Clinopyroxene crystals unzoned or slight oscillatory zoning many thick<250 μm) normally zoned rim to low Mg#cpx Olivine crystals Fernandina nodules unzoned some reverse zoned rims small increase Mg#ol.Fig. plagioclase zoning profiles.Core-rim lava MgO below detection limit (c). 2σ analytical uncertainties less than size data point.Evaluating crystal−liquid equilibriaTo compositional heterogeneity sub-volcanic melts calculated compositions minerals carrier liquids compared with distributions mineral analyses (FigWolf average equilibrium plagioclase An# 64.1 ± 1.7 (1σ glass clinopyroxene Mg#cpx 75.5 ± 0.9 olivine Mg#ol 75.2 ± 0.8 Fernandina average equilibrium plagioclase An# 63.3 one glass analysis clinopyroxene Mg#cpx 73.6 olivine Mg#ol 73.1. mineral analyses 2015 Wolf 1968 Fernandina samples higher An# Mg# carrier liquids. typical ocean island volcanoes magmas evolve fractional crystallisation minerals lower An# Mg# ascending carrier liquid entrains crystals mineral compositions lower An# Mg# values evolved compositions plagioclase clinopyroxene crystals Wolf tephra glomerocrysts Wolf lavas Fernandina nodules Evolved compositions measured rare olivine plagioclase clinopyroxene crystals Fernandina form long tails non-Gaussian An# Mg# KDEs crystals variably evolved melts 2015 1968 eruptions record heterogeneous liquids Wolf Fernandina sub-volcanic systems monotonous melts equilibrium melt compositionsDiffusive re-equilibration An#CaAl–NaSi charged cations Ti4+) slow plagioclase minerals retain original compositions millennia magmatic temperatures plagioclase TiO2 variations Wolf Fernandina samples changes liquids crystallisation plagioclase TiO2 concentrations analyses Galápagos constrain composition equilibrium liquids quantify heterogeneity sub modelling calculates plagioclase element residual liquid compositions isobaric cooling fractional crystallisation Rhyolite computes equilibrium plagioclase TiO2 concentrations each temperature step temperature anorthite-dependent partitioning model Nielsen et al = 0.03–0.10 natural plagioclase analyses Rhyolite-MELTS outputs provides major element composition estimate physical characteristics equilibrium liquids trajectory plagioclase TiO2 vs. An# matches compositional trend crystals Wolf Fernandina. 7 ran models pressures petrological geophysical constraints magma storage depths Wolf model outputs insensitive pressure best fits natural data 300 MPa Wolf 500 MPa Fernandina pressures magmacrystal compositions underpredict TiO2 content lowest An# crystals Fernandina experimental data Wolf Fernandina models predict TiO2 transitions An# to inflections occur ilmenite liquidus crystals An# 57–63 grew from ilmenite-saturated melts despite absence ilmenite western Galápagos volcanoes Ilmenite inclusions evolved (lowest Mg#cpx clinopyroxene cores 2015 Wolf eruption (Fig. validating model predictions. 7TiO2 plagioclase model 2015 Wolf eruption TiO2 vs An# lava tephra samples Crystals classified textural association 2σ analytical uncertainty for TiO2 less than data point An# grey lines show compositions crystals equilibrium with 2015 Wolf tephra glass models Namur et al.67 Nielsen et al.43 TiO2 1160 °C black lines show trajectory plagioclase compositional evolution Rhyolite-MELTS model Nielsen et.43 at 50 300 500 Plagioclase liquidus at 1214 1229 1244 °C in 50 300 500 MPa modelsplots show physical compositional evolution liquid Rhyolite-MELTS 300 MPa crystallisation pressure magma Wolf. 8TiO2 plagioclase model 1968 Fernandina eruption TiO2 vs An# nodule samples 1968 Crystals classified textural association compositions Fernandina Allan 2σ analytical uncertainty TiO2 less data point An# grey lines show compositions crystals 1968 Fernandina scoria models Namur et al.67 An# Nielsen et al.43 TiO2 1130 °C pre-eruptive crystallisation black lines show plagioclase compositional evolution Rhyolite-MELTS TiO2 model Nielsen et al.43 at 50 MPa 300 MPa 500 MPa Plagioclase liquidus at 1191 1207 1222 °C in 50 300 500 MPa models plots show physical compositional evolution liquid predicted Rhyolite-MELTS at 500 MPa Fernandina evolved natural plagioclase crystalsAn# Wolf Fernandina grew from plagioclase clinopyroxene-saturated basaltic trachy-andesitic melts 2.46 wt% MgO 37.5 300 MPa 1.08 wt% MgO 25.1 500 (Figs. 7 compositions more evolved than eruptive rocks Wolf Fernandina (Fig. 2) resorbed quartz in tephra 2015 eruption plagioclase data capture full range sub-volcanic liquids quartz saturate until dacitic-to-rhyolitic compositions low temperatures (<830 Quartz not phenocryst groundmass phase in rhyodacites Alcedo Rabida volcanoes present in Rabida crustal xenoliths44 data suggest Galápagos volcanoes contain evolved silicic melts sub.Diffusive re-equilibrationIron Mg diffuse faster than Ti in plagioclase crystal compositions likely overprinted by re Fe shows scatter (Fig. 5) enriched at crystal rims where An# elements constant high-Fe rims reflect secondary fluorescence not geologicalMagnesium negligibly affected by secondary if Mg content Wolf Fernandina plagioclase controlled melt chemistry crystallisation positive correlation between An# MgO data three populations constant MgO or negative correlation. 5) consistent with diffusive re-equilibration MgO lower solid−liquid Mg partition coefficients in anorthitic plagioclase39 modelled crystal compositions varying MgO concentration Rhyolite-MELTS MgO partitioning model Nielsen et al 0.02–0.05) crystals 1968 Fernandina nodules with melts 1–2 wt% MgO magma. 5d crystals 2015 Wolf eruption preserve MgO fractional crystallisation trend not in equilibrium with melt fixed MgO (Fig. disequilibrium higher MgO partially re-equilibrated with melts (>4 wt% MgO crystals lower MgO partially equilibrated with evolved melts (~2–3 wt% MgO Some crystals contain elevated MgO rims due to intermittent growth effects rapid crystallisationOlivine crystals Fernandina nodule samples low Mg#ol consistent growth-equilibration evolved liquids typically unzoned minority reverse rims plagioclase clinopyroxene Fe–Mg interdiffusion olivine reflect growth or re-equilibration primitive liquids pre-eruptive plagioclase crystals Wolf tephra samples low MgO no low-Mg#ol olivine origin Galápagos evolved evidence evolved liquids sample associations Wolf Fernandina volcanoes Wolf lava samples low-An# plagioclase crystals basaltic trachy-andesitic melts same glomerocrystic pyroxenes crystallised ~300 MPa consistent with glomerocrysts magma storage region >6.1–8.8 km evidence evolved liquids lower crust albitic plagioclase low-Mg#cpx pyroxenes resorbed quartz crystals present in tephra from 2015 eruption Stock tephra from upper crustal storage region ~1 km first material erupted crystal cargo distinct from later lavasFernandina nodules contain olivine feldspar crystals miarolitic cavities alteration active hydrothermal system formed upper crustal storage region similar to shallow source ~1 km data indicate evolved liquids beneath Galápagos volcanoes-MELTS models show viscosity Galápagos melts increases after ilmenite saturation reduction density. 7b Magma buoyancy insufficient low H2O content delay crystallisation volatile saturation volatile overpressure until low melt fractions ilmenite-saturated magmas stagnate Galápagos volcanoes become highly viscous high viscosities unlikely to ascend without external influence until substantial crystallisation.Basalt flushing beneath monotonous crystal compositions variable melts in Galápagos sub-volcanic systems heterogeneity not reflected in geochemistry of monotonous lavas surface studied petrographic contexts of evolved mineral zones Plagioclase clinopyroxene crystals diverse textures Wolf lavas tephra normal reverse zoning3d 6a crystals Fernandina nodules normal reverse zones lowest An# Mg#cpx rims scarcity evolved zones inhibits characterisation textures indicate open systems intermittent interactions primitive evolved Glomerocrysts 2015 Wolf eruption from sub-volcanic plagioclase grains evolved (low An# cores concentric primitive (high An# mantles recharge before crystals pile Mush accumulation long timescales textural evidence re-equilibration plagioclase MgO contents mixing events crystals pre-date eruption clinopyroxene-melt barometry glomerocrysts lower mixing occurred at petrographic evidence open-system behaviour constructed K2O/TiO2 vs. Mg#liq mixing models proportions primitive evolved melts Wolf Fernandina (Fig. 9) ratio increases crystallisation after saturation unaffected by plagioclase accumulationevolved end-member equilibrium with lowest An# plagioclase crystals from Wolf Fernandina primitive end-members highest lowest Mg#liq erupted liquid from each volcano higher Mg#liq analyses from Fernandina excluded not true liquids Figure 9 liquids at Wolf low K2O/TiO2 ratios <10% evolved magma liquids at Fernandina low K2O/TiO2 ratios <10% evolved end-member Geist identified submarine lavas southwest flank volcano more evolved compositions from lower crustal storage region restricted range of Mg#liq (~40–50), fed by primitive liquids elevated K2O/TiO2 ratios contain ~50% evolved magma estimates evolved material in erupted magmas quartz crystals magmas fractionated beyond evolved end-member melts may have higher K2O/TiO2 ratios.Fig. models between evolved primitive end-member liquids.K2O/TiO2 vs Mg#liq of liquids at Wolf volcano Fernandina K2O/TiO2 ratio increases when ilmenite (ilm)points show whole-rock tephra submarine glass melt literature data volcanoes western Galápagos Archipelago intrusive rocks-ultraphyric historic erupted liquids Wolf Fernandina data 2015 Wolf 1968 Fernandina eruptions References Wolf Fernandina liquids literature data Supplementary Note 2. Glass data 2015 Wolf eruption Stock et al.19 whole-rock data study 1968 Fernandina eruption Allan Simkin28 2σ analytical uncertainties whole-rock analyses less than size data point black lines show liquid lines descent Rhyolite-MELTS 50 MPa 300 MPa 500 MPa grey lines show mixing models evolved primitive liquids proportions evolved end-member evolved (high K2O/TiO2) end-members with lowest plagioclase crystals Rhyolite-MELTS 300 MPa Wolf 500 MPa Fernandina primitive (low K2O/TiO2) Wolf end-members highest lowest Mg#liq liquids measured eruptions primitive Fernandina end-members highest lowest Mg#liq liquidswhole-rock samples accumulated olivine measured eruptions by Koleszar Geist et.34 Table 1) petrological data show evolved melts in Galápagos sub-volcanic systems mix with primitive melts <10% magma erupted surface absence due to mass-balance large volumes basaltic magma interact with smaller evolved magma higher levels (Fig. liquids mix disparity basaltic component dominates composition remains unaltered evolved crystals not resorbed preserve heterogeneity monotonous activity reflect simplicity chemical homogeneity magmatic systems CO2 flushing composition higher monotonous volcanism reflects large volumes basalt evolved heterogeneous material shallower levels (Fig.. architecture Wolf Fernandina plumbing systems grey bars show estimates magma storage depths at Wolf Fernandina from Interferometric Synthetic Aperture Radar) inversions petrological barometry19 volcanoes by large lower crustal magma storage regions smaller shallow storage regions recharged by new magma from regions contain compositionally heterogeneous meltsmid-crustal magma storage no petrological geophysical evidence from recent eruptions Fernandina submarine eruptions fed by melts from deeper storage Large volumes basaltic melt ascend from mixing with liquids shallower levels sub-volcanic mush entrained into ascending liquids volume ascending basalt greater than evolved material shallower levels composition unaltered crystals resorbed preserve heterogeneity theory considers basaltic injections into silicic magma large explosive eruptions hybridised monotonous basalts of Galápagos erupt Hawaiian Strombolian style from mixing basaltic rhyolitic magmas 2018 eruption Kīlauea volcano (Hawai’i involved injection basalt eruption effusive58 critical mixing dynamics proportion of mafic silicic Galápagos Hawai’i volumes basalt greater than resident silicic magmas flushing monotonous eruptions melt flux crust high melt flux lower insufficient volumes primitive melt to overprint heterogeneity higher levels magma flux >1 ⋅ 10−4 km3 year−1 required stabilise super-solidus crustal mushless long-term eruptive fluxes at Wolf (0.4–1 ⋅ 10−3 km3 Fernandina (4.4 ⋅ 10−3 km3 not eastern Galápagos Archipelago eruptive fluxes findings support diversity erupted products Galápagos volcanoes dictated by position hotspot findings volcano monitoring volcanoes erupted basaltic contain evolved liquids sub systems basalt flushing monotonous eruptions external influences changes regional stress field decrease crustal melt flux second boiling allow melts ascend generating explosive silicic eruptions selection preparationSamples 2015 Wolf eruption lava reticulitic tephra additional lava samples east flank volcano 2017 material unaltered lava dense flow interiors Samples 1968 Fernandina eruption gabbroic nodules floor rim caldera described Howard et al nodules fresh some show extensive hydrothermal Wolf lava Fernandina nodule samples prepared as polished thin sections Crystals Wolf tephra separated 40–500 μm fraction by liquid magnetic separation mounted in epoxy quenched glassmethodsWhole-rock samples 2015 Wolf eruption analysed X-ray spectrometry for major trace elements Philips PW 2404 instrument University of Edinburgh Fitton et al modifications Fitton Analytical precision estimated three replicates same sample Relative 1σ precision better than ±1% for major elements>1 ±2% minor trace elements<1 except Th (±8%) Pb (±20%).Mineral compositions measured by EPMA Cameca SX100 SXFive instruments Departments Earth Sciences University of Cambridge Syracuse University measurements internally calibrated Smithsonian Microbeam Standards64 Relative 1σ precision monitored Cambridge similar Syracuse instrument better than ±2% for major ±15% minor elements except MnO in clinopyroxene (±17%) Typical 1σ errors in Supplementary Datasets 1–6 Supplementary Note melt Fe2+/Fe* value 0.85 calculating Mg#liq oxygen fugacity near quartz-fayalite-magnetite (QFM) buffer measured in Galápagos lavas by Peterson et alFe speciation minerals Fe2+ Fe3+ ions per formula unit sum calculations Mg#cpx Mg#ol evaluate distribution major element datasets KDEs bandwidths calculated after Sheather Jones66 crystal–liquid plagioclase clinopyroxene olivine compositions calculated models Namur et al.67 Putirka68 Herzberg O’Hara69 pre-eruptive storage temperatures 1160 °C Wolf 1130 °C Namur.67 anhydrous melt H2O equilibrium An# negligible Galápagos magmas low H2O contents < 1 TiO2 in major element residual liquid compositions calculated isobaric cooling fractional crystallisation Rhyolite-MELTS42 primitive whole-rock melt inclusion compositions Wolf Fernandina model starting liquids Equilibrium plagioclase TiO2 concentrations calculated each temperature step model Nielsen et al.43 modelled plagioclase An# TiO2 concentrations natural mineral analyses characterise equilibrium melt compositions preferable Scruggs Putirka70 meltModels run at 50 300 500 MPa fO2 QFM buffer 0.15 wt% initial H2O approximate Galápagos magma storage conditions studies19,27 liquid lines descent Rhyolite-MELTS crystallisation models good fit whole-rock glass data Wolf Fernandina Figs 1 2) simulated clinopyroxene compositions match natural crystal compositions 3) validates intrinsic variables suggests assimilation distinct wall rock material negligible impact evolution Galápagos magmas agreement studies27.Supplementary information Review Files 1 7 | 49.9 | 0.838315 |
10.1038/s41467-020-17400-4 | PMC7371698 | Assessments of future virtual water trading are still lacking. Here the authors estimated the global virtual water trade throughout the century and found that virtual green water exports and virtual blue water exports at least triple to more than 3200 bcm and 170 bcm, respectively, by the end of the century. | Water stressed regions rely heavily on the import of water-intensive goods to offset insufficient food production driven by socioeconomic and environmental factors. The water embedded in these traded commodities, virtual water, has received increasing interest in the scientific community. However, comprehensive future projections of virtual water trading remain absent. Here we show, for the first time, changes over the 21st century in the amount of various water types required to meet international agricultural demands. Accounting for evolution in socioeconomic and climatic conditions, we estimate future interregional virtual water trading and find trading of renewable water sources may triple by 2100 while nonrenewable groundwater trading may at least double. Basins in North America, and the La Plata and Nile Rivers are found to contribute extensively to virtual water exports, while much of Africa, India, and the Middle East relies heavily on virtual water imports by the end of the century. | IntroductionVirtual water trade (VWT) is the amount of water, either green (soil moisture) or blue (renewable and nonrenewable), that is consumed in the production of agricultural goods that are then traded in the international market1. This trading acts to alleviate stresses in several water stressed regions2–4. Potential water savings associated with, and analyses of past virtual water trading have received increased attention in the research community5–13 as future water stresses may leave some regions unable to meet their agricultural demands through domestic production alone5,8. These stresses can be driven by future socioeconomic conditions which are expected to cause a relative increase in global VWT14,15. While much of the water traded globally is green water or renewable blue water, recent studies have found increasing extraction of nonrenewable groundwater from deep aquifers to grow crops16–20 that are then traded internationally21–24. Continued depletion and over-exploitation of nonrenewable groundwater has significant negative effects both regionally and globally, including, but not limited to, land subsidence, eventual sea-level rise, and water quality degradation17,25,26. However, the spatial and temporal characteristics of future all sources of VWT are generally unknown. Further, although many analyses of VWT have attempted to reconstruct the historical evolution of the virtual water trade network2,4,8,27 and have concluded that VWT doubled between 1986 and 20076,8, assessments of future virtual water trading remain absent from the current literature.Here we have found and quantified increases in global VWT throughout the century, using the Global Change Analysis Model (GCAM), a market equilibrium model that links socioeconomics, climate, water, energy, and land systems (Methods), and a business-as-usual scenario combination of Shared Socioeconomic Pathway 2 (SSP2) and Representative Concentration Pathway 6.0 (RCP 6.0)28. Accounting for future human and climate influences, VWT is shown to increase throughout the century for all water types. Exports of virtual water are found to originate from select basins around the world while showing a dependence on socioeconomic changes throughout the century, particularly population dynamics in China.ResultsFuture water exports are concentrated in particular regionsIn total, virtual green water exports and virtual blue water exports more than triple from 905 billion m3 and 56 billion m3 in 2010 to more than 3200 billion m3 and 170 billion m3, respectively, by the end of the century in response to increases in population and the resultant demand increases (Fig. 1). Uncertainties in projected values of virtual water exports increase because of differences in climate impacts, as simulated by general circulation models (GCM) for the RCP6.0 scenario (Methods). Virtual nonrenewable groundwater exports at least double by the end of the century. Comparisons to previous global estimates (Table 1) show differences in virtual water trading intensities as a result of different trade data products and resultant aggregation of crop products. This study includes direct agricultural products, prior to processing, whereas other studies may include aggregate crop textiles and by-products, thus increasing trading values.Fig. 1Annual water flows of green, blue, and groundwater embedded in agricultural trade for SSP2-RCP6.0.Range of virtual green and blue water exports and the amount of nonrenewable groundwater depletion embedded in agricultural trade for all SSP2-RCP6.0 scenarios (n = 6 total scenarios), including effects of GCM uncertainty. Virtual green exports are shown on the primary y-axis (left) while virtual blue and nonrenewable groundwater exports are represented on the secondary y-axis (right). Solid lines represent the average for each water flow in SSP2-RCP6.0, while ribbons depict the full range of GCMs for the RCP6.0 scenario. Previous estimates of virtual water trade are shown as points and expanded upon in Table 1. Current study values are based on FAO country-level trade in 2010, all future estimates are between 32 GCAM regions.Table 1Global physical water flows with comparisons to previous historical studies7,21.Water flowsAnnual flows (billion mc per year)Source1996–2005201020502100Virtual green exports1352Hoekstra and Mekonnen71239This studya9052745–3040b3222–3708bThis study – SSP2-RCP6.0cVirtual blue exports255Hoekstra and Mekonnen7101This studya56122–145b179–208bThis study – SSP2-RCP6.0cVirtual nonrenewable groundwater exports25Dalin et al.21 d17This studya413.5–23.5b7.5–11.5bThis study – SSP2-RCP6.0caCalculated using trade between each country from 2010 FAO country-level crop export data.bRange across the five GCM suite of SSP2-RCP6.0 model runs.cCalculated using trade between each of the 32 regions in GCAM. Does not include intraregional trade.dCalculated using groundwater depletion rather than groundwater consumption.A large proportion of the virtual green water trade in 2010 is associated with oil crops (e.g. soybeans), a result consistent with previous studies29 (Supplementary Fig. 4). Increases in corn, wheat, and oil crops and lead to significant virtual green export increases by 2100 (Fig. 2a). These three crop commodities represent the largest proportion of current VWT and the highest green to blue water ratio required for production30. Africa, Europe, and India represent the largest importers of virtual green water.Fig. 2Virtual water trade fluxes by water type, region, and crop in 2100.a Average global virtual green water trade (billion m3), b virtual blue water trade (billion m3), and c virtual nonrenewable groundwater trade (billion m3) by crop and aggregate GCAM region in 2100 for all SSP2-RCP6.0 GCM scenarios (n = 5). Positive values represent exports while negative values imply virtual water imports.Virtual blue water trading shows significant differences arising in China, Pakistan, India, and the Middle East as the availability of water for irrigation decreases and populations change throughout the century31 (Fig. 2b; Supplementary Figs. 4–7). In 2100, globally, China represents a large source of virtual water exports due to the trading of wheat and rice products (Fig. 2b). Interestingly, China shifts from importer currently6,8,11 to exporter in the future12, because of a reduced growth rate after 2030 that ultimately causes population to decline31. Reduced domestic demands allow the use of all excess production to meet international agricultural demands. Regions in Africa experience nearly the opposite effect as population rapidly increases throughout the century, resulting in increasing demand that is unmet by domestic production (Supplementary Fig. 3). The United States represents another main source of future virtual blue exports through corn, fibers, and oil crops, with a corresponding import of only miscellaneous crops (MiscCrops, e.g. fruits, vegetables, nuts), as part of the southwestern United States shifts production away from MiscCrops toward the end of the century. Finally, we have found an intensification of VWT in the early part of the century as population growth continues and exports originate from water-intensive regions of the Middle East, Pakistan, and India, while toward the end of the century, exports come from water-rich areas that require smaller amounts of water to grow (Supplementary Fig. 2). This shifting of global food production accounts for demand changes, water scarcity changes, and groundwater depletion that together result in an inability to meet demands from domestic production alone, consistent with previous studies16,32,33.Our results show a fivefold increase in virtual nonrenewable groundwater trade by mid-century, with an end-of-century value doubling that of 2010. The export of nonrenewable groundwater to meet international demands is concentrated in several main regions: The United States, Mexico, western South America, and northern Africa. On a temporal scale, water scarce regions export nonrenewable groundwater early in the century but cease to do so after mid-century as demand changes and groundwater depletion worsens (Supplementary Fig. 2).Only a few basins address global agricultural deficitsAnalysis of GCAM results at the 235 water-basin scale (Methods) permits for the identification of specific locations where virtual water exports originate. Comparing the 2050–2100 values of virtual green and blue exports (Fig. 3a–d) reveals an intensification of exports in most Chinese river basins. Blue and green exports are also concentrated in the Missouri River basin, the La Plata basin in South America, and the Murray-Darling basin in Australia. Each of these basins contains significant agricultural production and their water supplies will be used heavily to meet future demands.Fig. 3Basin-level virtual water exports in 2050 and 2100 for all sectors.Virtual green water exports (billion m3) a and b, and blue water exports (billion m3), c, d in 2050 and 2100 respectively for the average of five GCM runs for SSP2-RCP6.0. Virtual nonrenewable groundwater exports (billion m3) in 2050 and 2100 for the same averaged GCM runs for SSP2-RCP.6.0 is shown in e and f. All values are considering the exports of agricultural crops only with additional, potentially necessary virtual water imports not considered.Tracing the virtual nonrenewable groundwater exports over time shows an important evolution as basins in Saudi Arabia and the Indus River basin export substantial volumes in 2050, but do not contribute to the global nonrenewable groundwater exports in 2100. This is because extraction in the first half of the century from the large underground aquifers in Saudi Arabia and India34,35 causes additional pumping to become too expensive to sustain in these regions.The High Plains, Central Valley, and Mississippi Embayment aquifers are the three most over-exploited aquifers in the United States currently, particularly for irrigated agriculture20,36, and our results show this trend continues through 2100 in terms of virtual nonrenewable groundwater exports. Exports are not shown from the Missouri River basin, on top of the deepest portion of the High Plains aquifer since groundwater recharge is greater than nonrenewable extraction in this basin37; therefore, the groundwater withdrawn in the Missouri River basin is classified as renewable and is thus captured in the virtual blue exports. The Nile, La Plata, and Murray-Darling basins use nonrenewable groundwater to produce rice, fibers, and corn that is demanded outside of their regional boundaries (Supplementary Fig. 2C). These basins currently use significant amounts of groundwater for irrigated agriculture and have extensive, yet declining groundwater reserves38–40.DiscussionUsing GCAM to project the evolution of the global trade market, based upon changes in socioeconomic and climate conditions, provides a first assessment of changes in virtual water trade towards the end of the century, and addresses one of the major gaps identified in virtual water analyses13. In this analysis, we have built upon previous advances in the reconstruction of the historical global virtual water trade network6,7,21,27 by linking potential future socioeconomic and climatic changes to alterations in the production of agricultural goods, with the resulting price fluctuations in the global trade market causing a potential restructuring of global agricultural trading. Further, the work provides a first assessment of the quantities of nonrenewable groundwater extraction from aquifers around the world required to meet the international crop demand.This study focused only on water used to grow agricultural crops, and while nearly 90% of blue water consumption is used for agricultural purposes41, energy and industrial goods are also extensively traded in the global market; therefore, it is important to understand the international trade in these different sectors and how it may change into the future. Projecting such trade changes into the future is not trivial, but obtaining estimates based upon different climate and socioeconomic conditions can produce a wide range of potential trade development pathways. Further, although the main focus of this study has been on one socioeconomic scenario (SSP2) and one climate scenario (RCP6.0), it is important that future studies consider changes in the VWT network that result from different socioeconomic and climate conditions, and that are likely to produce a larger range of potential outcomes. We provide an initial estimate of the uncertainty surrounding mid and end-of-century VWT values across a suite of SSP-RCP scenarios (Supplementary Figs. 8 and 9) and encourage further investigation to understand the drivers of change for various socioeconomic and climate mitigation scenarios.MethodsOverall descriptionThis analysis uses GCAM to quantify the amount of water embedded in the global trading of agricultural goods. This water, called virtual water1, is calculated based on how much water is consumed by the individual exported crop in the region where it was grown. In order to account for an evolving market and changing production conditions, we use a defined future socioeconomic scenario, SSP228,42, matched with the RCP6.0 climate forcing scenario43. We introduce climate derived impacts from five general circulation models (GCM) to allow for changing water supply, crop yields, hydropower availability, and building energy demands (i.e., cooling and heating). We analyze the amount of green and blue water consumption that is embedded in global trade and differentiate between renewable surface and groundwater recharge, as well as nonrenewable groundwater to provide global estimates, regional contributions, and basin-level usage. Below we describe the GCAM model, scenario components, virtual water calculations in GCAM, and assumptions for the downscaling of exports and estimations of virtual water imports.GCAMGCAM is a market equilibrium model that links energy, water, land, economy, and climate systems44–46. GCAM adjusts prices of goods and services within each model time step to equilibrate the supply and demand of goods and services at each time step, and thus simultaneously clears markets across sectors. This study accounts for a limited supply of water by employing cost resource curves across all 235 basins that follow a logit formulation to determine the share of each water source (renewable, nonrenewable, and desalinated water) needed to meet the water demands within all basins47,48. As depletion of various water sources increases, the extraction price increases, which leads to compounding price increases in the goods and services that require higher-priced water sources.Agricultural production in GCAM is computed endogenously by accounting for historical crop growth representations from MIRCA 2000 data in combination with yield estimates and a breakdown of irrigated and rainfed production. Water consumption coefficients, both biophysical and blue water, are exogenous inputs in GCAM by country and crop type7. These are aggregated to the GCAM region scale for 12 crop types in GCAM, with two additional biomass crop-type water coefficients49.Agricultural trade within GCAM is modeled following a Heckscher–Ohlin method in which commodities are traded in a single global market where each region will see the same global price for that commodity. This allows each region to determine how much it will supply or demand of each commodity at that price. Using this method results in no preference for any region to demand certain commodities from another particular region.Key scenario componentsThe SSP2 scenario, often referred to as a reference or business-as-usual scenario, represents a world with steady population growth through the middle of the century, at which time the global population begins to equilibrate toward a 2100 value of 9 billion people. Economic growth continues at present-day values, and thus fuel and energy preferences remain very similar to what they are today. For these reasons, this scenario represents one with medium challenges to both climate mitigation and adaptation42. Combining these socioeconomic features with future climate changes, we implement a future RCP6.0 trajectory that results in end of century climatic forcing of 6.0 W/m2. Quantitative assumptions for the SSP2 scenario are documented in separate studies28,50.GCM derived climate impactsThis study includes climatic impacts on water supply, agricultural productivity, hydropower availability, and building energy demands that are derived from five different general circulation models (GCM). We calculate each of these impacts by using downscaled and bias-corrected climate data from the Inter-Sectoral Impact Model Intercomparison Project (ISI-MIP)51. The global hydrologic model, Xanthos52–54, calculates climate derived changes to renewable water supply at the GCAM 235-basin scale using necessary GCM outputs. Climate derived impacts to crop yield changes55, hydropower availability56, and building energy demands57 are calculated from the same set of ISI-MIP models and the climate varying impacts are added to the SSP2 scenario. All datasets have been made publicly available for future use58,59.Calculation of all virtual water components in GCAMVirtual water calculations in GCAM require several assumptions to account for its representation of trade as occurring across 32 regions, demands as regional, and production as basin level; further, the origins of imported goods are not traceable once exports are placed in the global market. In order to calculate the different components of virtual water trade, we must first calculate the regional and basin level trade. Although basin level imports are not calculated, all exports are trackable to the basin level, using the proportion of production as a proxy.The second term in Eq. 1 takes the proportion of production, P, of any crop, c, and growth type, g, within a basin, b, to the total production of that crop in that region. This proportion is then multiplied by the regional demand, D, of that crop. This is due to GCAM modeling crop demands at the regional level. In order to scale this at the basin level we assume that the demand of a basin is proportional to the production from that basin. While demands are only modeled at the regional level, this is a good first-order approximation for estimating demands at a finer scale in GCAM. Growth types are classified as either rainfed, RFD, or irrigated, IRR and are determined endogenously within GCAM based upon the profitability of each crop type after the calibration period.Once the proportion of regional demands is determined, it can then be subtracted from the basin level production, P, to determine the net surplus or deficit of a crop in a basin, T. Positive values of T, represent exports, E, whereas negative values represent the need for imports, I.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_{b,c,g}(t) = P_{b,c,g} - \left[ {D_C \ast \left( {\frac{{P_{b,c,g}}}{{\mathop {\sum }\nolimits_{i = 1}^b P_{b,c,g}}}} \right)} \right]$$\end{document}Tb,c,g(t)=Pb,c,g−DC*Pb,c,g∑i=1bPb,c,gVirtual green water exports, VGE, are calculated by considering the green water consumption, GWC, the basin level rainfed crop production, PRFD, and the rainfed exports, ERFD. Virtual green imports, VGI, must consider the amount of virtual green water that is in the global market summed over all regions, r, and basins, b, VGE, the ratio of imports, I, in a region, r, and total global imports of each crop. Finally, the total virtual green water trade (VGT) is calculated at the regional level as the combination of the exports and imports of virtual green water.2A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{VGE}}_{b,c}(t) = \frac{{{\mathrm{GWC}}_{b,c}}}{{\left( {\frac{{P_{b,c,{\mathrm{RFD}}}}}{{\mathop {\sum }\nolimits_{i = 1}^b P_{b,c,{\mathrm{RFD}}}}}} \right)}} \, * \, E_{b,c,{\mathrm{RFD}}}$$\end{document}VGEb,c(t)=GWCb,cPb,c,RFD∑i=1bPb,c,RFD*Eb,c,RFD2B\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{VGI}}_{r,c}(t) = \left( {\mathop {\sum }\limits_{i = 1}^{b,r} {\mathrm{VGE}}_{b,c}} \right) \, * \, \frac{{I_{r,c,{\mathrm{RFD}}}}}{{\mathop {\sum }\nolimits_{i = 1}^r I_{r,c,{\mathrm{RFD}}}}}$$\end{document}VGIr,c(t)=∑i=1b,rVGEb,c*Ir,c,RFD∑i=1rIr,c,RFD2C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{VGT}}_{r,c}(t) = \left( {\mathop {\sum }\limits_{i = 1}^{b_r} {\mathrm{VGE}}_{b,c}} \right) + {\mathrm{VGI}}_{r,c}(t)$$\end{document}VGTr,c(t)=∑i=1brVGEb,c+VGIr,c(t)Virtual blue water analysis follows the same process as for green water, with the slight adjustment of accounting for irrigated production and trade, as well as the blue water consumption, BWC. Here virtual blue water exports (VBE), virtual blue water imports (VBI), and virtual blue water trade (VBT) require the production of irrigated agriculture.3A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{VBE}}_{b,c}(t) = \frac{{{\mathrm{BWC}}_{b,c}}}{{\left( {\frac{{P_{b,c,{\mathrm{IRR}}}}}{{\mathop {\sum }\nolimits_{i = 1}^b P_{b,c,{\mathrm{IRR}}}}}} \right)}} \, * \, E_{b,c,{\mathrm{IRR}}}$$\end{document}VBEb,c(t)=BWCb,cPb,c,IRR∑i=1bPb,c,IRR*Eb,c,IRR3B\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{VBI}}_{r,c}(t) = \left( {\mathop {\sum }\limits_{i = 1}^{b,r} {\mathrm{VBE}}_{b,c}} \right) \, * \, \frac{{I_{r,c,{\mathrm{IRR}}}}}{{\mathop {\sum }\nolimits_{i = 1}^r I_{r,c,{\mathrm{IRR}}}}}$$\end{document}VBIr,c(t)=∑i=1b,rVBEb,c*Ir,c,IRR∑i=1rIr,c,IRR3C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{VBT}}_{r,c}\left( t \right) = \left( {\mathop {\sum }\limits_{i = 1}^{b_r} {\mathrm{VBE}}_{b,c}} \right) + {\mathrm{VBI}}_{r,c}\left( t \right) = 0$$\end{document}VBTr,ct=∑i=1brVBEb,c+VBIr,ct=0Finally, the calculation of virtual groundwater exports (VGWE) considers the ratio of groundwater depletion in a basin, GWD, to the total blue water withdrawals in the basin, BWW. Multiplying this proportion by the virtual blue water exports yields the amount of the blue water exports that is from nonrenewable groundwater sources.4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{VGWE}}_{b,c}(t) = {\mathrm{VBE}}_{b,c}(t) \, * \, \frac{{{\mathrm{GWD}}_b}}{{{\mathrm{BWW}}_b}}$$\end{document}VGWEb,c(t)=VBEb,c(t)*GWDbBWWbTotal virtual groundwater trade (VGWT) and virtual groundwater imports (VGWI) are calculated in the same manner as Eq. 4, by considering the blue water imports and total trade as the first term on the right-hand side of the equation.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary | nature communications | [
"Article"
] | [
"Sustainability",
"Hydrology"
] | IntroductionVirtual water trade (VWT) is water green or blue consumed in production agricultural goods traded in international market1 trading stresses in water stressed regions2–4 Potential water savings analyses past virtual water trading attention future water stresses may leave regions meet agricultural demands domestic production stresses driven by future socioeconomic conditions expected cause increase in global VWT14 water traded globally is green or renewable blue studies found increasing extraction of nonrenewable groundwater from traded internationally21–24. depletion over-exploitation of nonrenewable groundwater negative effects land subsidence sea-level rise water quality degradation17 spatial temporal characteristics of future sources of VWT unknown analyses evolution VWT doubled between 1986 and 20076,8 assessments of future virtual water trading absent from found quantified increases in global VWT using Global Change Analysis Model market equilibrium model socioeconomics climate water energy scenario Shared Socioeconomic Pathway 2 Pathway 6.0 VWT for all water types Exports of virtual water originate from select basins on socioeconomic changes particularly population dynamics in ChinaResultsFuture water exports in virtual green blue water triple from 905 56 2010 to 3200 170 billion by end century population demand increases Uncertainties in water exports increase climate impacts RCP6.0 scenario nonrenewable groundwater exports double by end century Comparisons estimates show differences in virtual water trading intensities different trade data products crop products study includes direct agricultural products other studies crop textiles by-products increasing trading values. 1Annual water flows green blue groundwater agricultural trade for SSP2-RCP6.0 green blue water exports nonrenewable groundwater depletion trade SSP2-RCP6.0 scenarios GCM uncertainty green exports primary y-axis blue nonrenewable groundwater exports secondary lines average each water flow in SSP2-RCP6.0 ribbons full range GCMs RCP6.0 scenario Previous estimates virtual water trade in Table 1. study values based on FAO country-level trade 2010, future estimates between 32 GCAM regions 1Global water flows comparisons toWater flowsAnnual (billion mc per year)Source1996–2005201020502100Virtual green exports1352Hoekstra SSP2-RCP6.0cVirtual blue exports255Hoekstra SSP2-RCP6 nonrenewable groundwater exports25Dalin et al.21 SSP2-RCP6 trade 2010 FAO crop export data five GCM SSP2-RCP6.0 model runs 32 regions GCAM intraregional trade groundwater depletion consumption virtual green water trade 2010 oil crops Increases corn wheat oil crops virtual green export increases by 2100 largest proportion VWT highest green to blue water ratio Africa Europe India largest importers green water.Fig. 2Virtual water trade fluxes by water type region crop 2100 Average global virtual green water trade (billion blue water nonrenewable groundwater trade by crop GCAM region 2100 SSP2-RCP6.0 GCM scenarios 5) Positive values exports negative values imports blue water trading differences China Pakistan India Middle East availability water irrigation decreases populations changeFigs. 4–7) 2100 China large source virtual water exports wheat rice products (Fig shifts from importer to exporter reduced growth rate after 2030 population Reduced domestic demands allow excess production international agricultural demands Africa opposite effect population increases increasing demand unmet by domestic production Fig 3) United States main source future virtual blue exports through corn fibers oil crops import miscellaneous crops southwestern United States shifts production from MiscCrops end century intensification of VWT early century population growth exports from water-intensive Middle East Pakistan India end exports from water-rich areas smaller water 2) shifting global food production for demand changes water scarcity groundwater depletion inability meet demands from domestic production results fivefold increase in virtual nonrenewable groundwater trade by mid-century end-of-century value doubling 2010. export nonrenewable groundwater in United States Mexico western South America northern Africa water scarce regions export nonrenewable groundwater early cease after mid-century demand changes groundwater depletion worsens Fig 2)few basins address global agricultural GCAM results 235 water-basin scale specific locations virtual water exports Comparing 2050–2100 virtual green blue exports (Fig. 3a–d intensification Chinese river basins Blue green exports Missouri River basin La Plata basin America Murray-Darling basin Australia basins significant agricultural production water supplies used future demands. 3Basin-level virtual water exports 2050 2100 sectors green water exports (billion m3) a b blue water (billion m3) c d in 2050 2100 five GCM runs SSP2-RCP6.0 Virtual nonrenewable groundwater exports (billion m3) 2050 2100 e f values exports agricultural crops additional water imports not virtual nonrenewable groundwater exports evolution basins Saudi Arabia Indus River basin export volumes 2050 contribute global exports in 2100 extraction aquifers Saudi Arabia additional pumping expensive High Plains Central Valley Mississippi Embayment aquifers most over-exploited United States irrigated trend continues through 2100Exports not shown from Missouri River basin High Plains aquifer groundwater recharge greater than nonrenewable extraction groundwater withdrawn renewable captured in virtual blue exports Nile La Plata Murray-Darling basins use nonrenewable groundwater rice fibers corn demanded outside regional boundaries Fig. 2C). basins use significant groundwater for irrigated have declining groundwater reserves38–40 GCAM to project evolution global trade market socioeconomic climate conditions first assessment of changes virtual water trade end century addresses gaps in virtual water analyses13 built trade linking future socioeconomic climatic changes to alterations production agricultural goods price fluctuations potential restructuring global agricultural trading provides first assessment of nonrenewable groundwater extraction international crop demand study focused on water agricultural crops 90% blue water consumption for agricultural energy industrial goods traded global important to understand international trade in sectors future Projecting trade changes not trivial estimates based climate socioeconomic conditions produce potential trade development pathways main focus study on one socioeconomic scenario (SSP2) climate scenario (RCP6important future studies consider changes VWT network different socioeconomic climate conditions potential outcomes provide initial estimate uncertainty mid end-of-century VWT values SSP-RCP scenarios Figs. 8 9) encourage investigation understand drivers change socioeconomic climate mitigation scenarios analysis uses GCAM water global trading agricultural goods virtual water1 calculated consumed exported crop account evolving market changing production conditions use future socioeconomic scenario SSP228,42 RCP6.0 climate forcing scenario43 introduce climate impacts from five general circulation models) changing water supply crop yields hydropower availability building energy demands analyze green blue water consumption global trade differentiate renewable surface groundwater recharge nonrenewable groundwater global estimates regional contributions basin-level usage describe GCAM model scenario components virtual water calculations assumptions downscaling exports estimations virtual water imports market equilibrium model energy water land economy climate adjusts prices goods services equilibrate supply demand clears markets across sectorsstudy accounts limited supply water cost resource curves across 235 basins share each water source needed water demands depletion water sources increases extraction price increases price increases in goods services higher-priced water sources.Agricultural production in GCAM computed historical crop growth MIRCA 2000 data yield estimates breakdown irrigated rainfed production Water consumption coefficients biophysical blue water exogenous inputs by country crop type7 aggregated to GCAM region scale for 12 crop types two additional biomass crop-type water coefficients49.Agricultural trade GCAM modeled Heckscher–Ohlin method commodities traded in single global market region same global price region supply demand commodity no preference for commodities scenario SSP2 scenario represents steady population growth middle century global population toward 2100 value 9 billion people Economic growth continues at present-day values fuel energy preferences similar medium challenges to climate mitigation adaptation42 future climate changes future RCP6.0 trajectory end of century climatic forcing of 6.0 W/m2. Quantitative assumptions for SSP2 scenario documented in separate studies28GCM climate study includes impacts on water supply agricultural productivity hydropower availability building energy demands from five general circulation models downscaled bias-corrected climate data from Inter-Sectoral Impact Model Intercomparison Project global hydrologic model Xanthos52–54 calculates climate changes renewable water supply at GCAM 235-basin scale GCM outputs Climate impacts to crop yield hydropower availability56 building energy calculated from-MIP models impacts added to SSP2 scenario datasets publicly available for future virtual water components in calculations require assumptions trade 32 regions demands regional production basin level origins of imported goods not traceable exports global market calculate trade calculate regional basin level trade basin level imports not exports trackable basin level proportion of production proxy second term in Eq. 1 proportion of production of crop growth type basin to total production multiplied by regional demand D due to GCAM modeling crop demands regional level basin level demand proportional to production basin approximation for estimating demands finer scale in GCAMGrowth types classified as rainfed or irrigated determined within GCAM based profitability each crop type after calibration period proportion of regional demands determined subtracted from basin level production P net surplus or deficit of crop basin Positive values T represent exports negative values represent need for imports\documentclass[12pt{minimal}\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}_{b,c,g} = P_{b,c,g_{b,c,g = 1 P_,c{document}Tb,c,g=Pb,c,g−DC*Pb,gVirtual green water exports calculated green water consumption basin level rainfed crop production rainfed exports green imports consider green water in global market regions ratio of imports in region total global imports of each crop total virtual green water trade) calculated at regional level combination of exports and imports virtual green water[12pt]{minimal{amsmath-69pt}}\mathrm{VGE}}_{b,c}(t) =\frac\mathrm{GWC}}_{b,c\left\frac{{P_{b,c\mathrm{RFD\nolimits_{i = 1}^b P_{b,c\mathrm{RFD}}}}}}{b,c\mathrm{RFD}}}\end{document}VGEb,c(t)=GWCb,cPb,c,RFD∑i=1bPb,c,RFD*Eb,c,RFD2B[12pt]{minimal}{amsmath}}{upgreek}{}{-69pt}}{\mathrm{VGI}}_{r,c}(t) =\left(\limits_{i = 1}^{b,r} {\mathrm{VGE}}_{b,c}}\right\frac{{I_{r,c,{\mathrm{RFD}}}}}{{\mathop {\sum\nolimits_{i = 1}^r{r\mathrm{RFD\end{document}VGIr,c(t)=∑i=1b,rVGEb,RFD∑i=1rIr,RFD2C\documentclass[12pt]{minimal\usepackage{amsmath}{upgreek}\setlength\oddsidemargin-69pt}{document}\mathrm{VGT}}_{r,c}(t) =\left\sum\limits{i = 1}{b_r}\mathrm{VGE}}{b,c}}\right) + {\mathrm{VGI}}_{r,c}(t\end{document}VGTr,c(t)=∑i=1brVGEb,c+VGIr(t)Virtual blue water analysis process green water accounting irrigated production trade blue consumption exports imports trade require irrigated\documentclass[12pt]{minimal{amsmath-69pt}}\mathrm{VBE}}_{b,c}(t) =\frac\mathrm{BWC}}_{b,c\left\frac{{P_{b,c\mathrm{IRR\nolimits_{i = 1}^b P_{b,c\mathrm{IRR}}}}}}{b,c\mathrm{IRR}}}\end{document}VBEb,c(t)=BWCb,cPb,c,IRR∑i=1bPb,c,IRR*Eb,c,IRR3B[12pt]{minimal}{amsmath}}}{upgreek}{}{-69pt}}{\mathrm{VBI}}_{r,c}(t) =\left(\limits_{i = 1}^{b,r}\mathrm{VBE}}_{b,c}}\right\frac{{I_{r,c,{\mathrm{IRR}}}}}{{\mathop {\sum\nolimits_{i = 1}^r{r\mathrm{IRR\end{document}VBIr,c=∑i=1b,rVBEb,IRR∑i=1rIr,IRR3C\documentclass[12pt]{minimal}\usepackage{amsmath}{upgreek}\oddsidemargin-69pt}{document}\mathrm{VBT}}_{r,c}\left\right) =\mathop\sum\limits{i = 1}{b_r}\mathrm{VBE}}\right +\mathrm{VBI}}{r,c}\left =\end{document}VBTr,ct=∑i=1brVBEb,c+VBIr calculation virtual groundwater exports ratio groundwater depletion to blue water withdrawals Multiplying by blue water exports exports from nonrenewable groundwater sources[12pt{minimal\usepackage{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}{document}\mathrm{VGWE}}{b,c(t =\mathrm{VBE}}_{b,c}(t)\mathrm{GWD}}_b}}\mathrm{BWW}}_b}}{document}VGWEb,c(t)=VBEb,c(t)*GWDbBWWbTotal virtual groundwater trade imports calculated Eq. 4 considering blue water imports total trade first term right-hand side equation.Reporting research design Nature Research Reporting Summary.Supplementary information Peer Review File Reporting Summary | 49.5 | 0.424733 |
10.1038/s41467-021-21363-5 | PMC7910301 | Here, advanced scanning transmission electron microscopy techniques are used to image the atomic structure at the interface between 2D MoS2 and 3D Au nanoislands, revealing a moiré superlattice and illustrating the potential for (opto-)electronic moiré engineering at the 2D/3D interface. | The atomic structure at the interface between two-dimensional (2D) and three-dimensional (3D) materials influences properties such as contact resistance, photo-response, and high-frequency electrical performance. Moiré engineering is yet to be utilized for tailoring this 2D/3D interface, despite its success in enabling correlated physics at 2D/2D interfaces. Using epitaxially aligned MoS2/Au{111} as a model system, we demonstrate the use of advanced scanning transmission electron microscopy (STEM) combined with a geometric convolution technique in imaging the crystallographic 32 Å moiré pattern at the 2D/3D interface. This moiré period is often hidden in conventional electron microscopy, where the Au structure is seen in projection. We show, via ab initio electronic structure calculations, that charge density is modulated according to the moiré period, illustrating the potential for (opto-)electronic moiré engineering at the 2D/3D interface. Our work presents a general pathway to directly image periodic modulation at interfaces using this combination of emerging microscopy techniques. | IntroductionFollowing the success of moiré engineering in modulating (opto-)electronic properties of graphene/hexagonal boron nitride (hBN) heterostructures1,2 and twisted bilayer graphene3–6, studies have extended the moiré toolbox to include systems such as double bilayer graphene7, trilayer graphene8, and van der Waals (vdW) heterostructures composed of transition metal dichalcogenides9,10 and hBN-graphene-hBN stacks11,12. Recently, moiré engineering has been extended beyond vdW heterostructures, to 3D/3D oxides13. Moiré engineering is yet to be utilized for tailoring the quasi-vdW interface between a 2D material and 3D metal. Engineering such 2D/3D interfaces is key to device applications where 2D materials make contact, through a well-controlled junction, to a 3D material such as a metal or semiconductor14–16. In contrast to 2D/2D heterostructures, moiré engineering at the 2D/3D interface requires consideration of the stacking of atomic planes in the out-of-plane direction. 3D stacking introduces an additional tuning parameter in 2D/3D systems for modulating moiré properties that is not available in 2D/2D heterostructures17.The ability to image moiré superlattices directly is required to map electronic property modulation onto atomically-resolved structure18. Various techniques have been used to observe moiré superlattices. These include reciprocal space imaging via low energy electron diffraction19,20 and convergent beam electron diffraction (CBED)21; spatially resolved property measurement via scanning tunneling microscopy (STM)22,23, atomic force microscopy (AFM) modalities2,24, near-field optical microscopy13, and infrared nano-imaging5; and imaging of transmitted intensity via high-resolution and dark field (scanning) transmission electron microscopy, (S)TEM25,26. Of these techniques, STM and (S)TEM are the only two that exhibit real-space atomic resolution. STM is widely used to characterize moiré patterns in 2D materials on bulk substrates, such as graphene on Ru27, Ir28, and Cu29. However, STM measurements are challenging for deeply buried interfaces and for the suspended layers that are gaining traction in 2D device physics30,31. (S)TEM, on the other hand, provides detailed information for suspended moiré systems fabricated from solely 2D materials25,26. Interpretation is more challenging for 2D/3D interfaces due to the necessity of considering the 3D structure of layers away from the interface32,33. This has resulted in discrepancies in periodicity measurement between imaging techniques22,33. The MoS2/Au{111} system highlights these challenges, with different values reported for the periodicities of superlattices measured via STM and (S)TEM, 32 Å and 18 Å respectively22,33.To reconcile such discrepancies and map moiré structure-property relations at the 2D/3D interface, we combine an analytic convolution technique and a range of STEM imaging techniques, integrated differential phase contrast (iDPC) and four-dimensional (4D) STEM, to decouple the spectrum of higher order moiré patterns. We investigate MoS2 /Au{111} as a model 2D/3D system, relevant to TMDC (opto-) electronics14, and also examine hBN/Au{111}, relevant in plasmonics34,35. iDPC STEM measures the phase of the sample transmission function, enabling direct interpretation as the projected electrostatic potential in thin samples36–38. 4D STEM is a rapidly developing technique in which a pixelated array detector is used to collect a CBED pattern at each probe position in the STEM image. The resulting 4D dataset can be filtered post-acquisition to produce reconstructions such as bright field, annular bright field, annular dark field (ADF), ptychography, and iDPC39. 4D STEM has been applied to analysis of materials including Cu40, ZrO241, LiNiO242, DyScO343, graphene44, MoS245 and WS246, with 2D materials particularly well-suited due to their small thickness47. We show that iDPC and 4D STEM are able to decouple higher order moiré periods to form real space images of the moiré pattern at the 2D/3D interface of MoS2/Au{111}, revealing the crystallographic 32 Å period. We explain the difference compared to conventional (S)TEM in terms of projection effects of the ABC stacking of the 3D metal. We then use ab initio electronic structure calculations to corroborate that MoS2/Au{111} charge density modulation is concentrated at the interface and follows the 32 Å moiré periodicity. Together these findings demonstrate the utility of direct imaging via iDPC and 4D STEM for understanding the structure and electronic properties of 2D/3D heterostructures.ResultsMicroscopy of the MoS2/Au{111} systemAn example of the MoS2/Au{111} interface is shown in Fig. 1. In contrast to the mechanical transfer processes employed for fabricating vdW heterostructures, the 2D/3D systems studied here were formed by direct epitaxial growth48 in ultra-high vacuum conditions (Methods). The resulting samples consist of flat, faceted Au{111} nanoislands with an average edge length of 25 nm and height of 8 nm (Fig. 1a, Supplementary Fig. 1) that are epitaxially aligned on suspended MoS2{0001} (Fig. 1b), with uniform moiré periodicities across micrometre-scale areas. Selected-area electron diffraction (SAED) confirms 0° rotation between Au and MoS2 with a standard deviation of 0.2° (Supplementary Fig. 2). In Fig. 1b and other SAEDs, we observe spots indexed as 1/3{422} Au reflections. These are classically forbidden for the FCC structure but their presence is consistent with Au nanoisland literature49 (Supplementary Note 1). High resolution (HR) TEM shows that the islands are single crystalline, with no evidence of misfit dislocations and grain boundaries (Fig. 1c). The discontinuity in the moiré pattern visible at some boundaries arises from island coalescence. Here, both rigid body displacements and twin boundaries arise from stacking faults between coalesced islands (Fig. 1d, blue arrows).Fig. 1HRTEM and STEM demonstrating epitaxial MoS2/Au{111} moiré.a Schematic of epitaxially aligned Au deposited on suspended MoS2 supported on a SiNx TEM grid. Orange atoms represent Au, yellow S, and purple Mo. The SiNx membrane is shown in dark blue. b Reciprocal space model and experimental selected area electron diffraction pattern of the Au {111} zone aligned on MoS2 {0001}, with weak intensity 1/3{422}Au spots (see text) aligned with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{10\overline 10\}_{\rm{MoS}_{2}}$$\end{document}{101¯0}MoS2 and higher intensity {220}Au spots aligned with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{2\overline {11}0\}_{\rm{MoS}_{2}}$$\end{document}{211¯0}MoS2. Orange dots represent frequencies from Au crystal planes, while purple represent frequencies from MoS2 crystal planes. c HRTEM image of Au nanoisland on MoS2, showing apparent 18 Å-period moiré pattern. Scale bar, 40 Å. d High angle annular dark field (HAADF) STEM image. Scale bar, 80 Å. Coalescence boundaries are marked by blue arrows.The uniform moiré periodicity and sinusoidal intensity modulation show that the Au and MoS2 lattices are undistorted in the plane of the interface, even near island edges. This is different from the case of twisted vdW structures, which frequently display reconstructions25,26. The absence of distortion can likely be attributed to weak quasi-vdW bonding at the MoS2/Au{111} interface14. Motion of islands at room temperature is consistent in suggesting loose binding of the Au nanoislands to the underlying substrate (Supplementary Movie 1). During their motion, the islands exhibit rotation up to 0.3°, visually amplified in the angle of the moiré pattern (Supplementary Fig. 2). These MoS2/Au{111} interface characteristics are consistent for a range of Au thicknesses and uniform across samples (Supplementary Fig. 3).Moiré site inequivalenceAt first glance, the period of the MoS2/Au{111} moiré superlattice in Fig. 1c, d is 18 Å. While this is in agreement with previous HRTEM studies33, it is a consequence of the projective nature of conventional (S)TEM imaging. To illustrate this, one can consider a thought-experiment in which the out-of-plane coordinate of the 3D Au{111} structure is ignored; this results in a “projected” hexagonal Au lattice with atomic spacing of 1.66 Å, which indeed yields a moiré pattern of 18 Å with the MoS2 substrate. A more accurate view of electron scattering through the Au crystal requires us to include the full face-centred cubic (FCC) Au structure, as shown in Fig. 2a, b. Consider a location where an Au atom from the A layer (orange) is directly above a pair of S atoms, as in the centre of Fig. 2c-top. This site repeats every 32 Å, shown by the orange squares in Fig. 2b. Sites that appear similar (red and blue squares in Fig. 2b) instead have Au atoms from the B or C layers above the S atoms (Fig. 2c-middle, 2c-bottom). The inequivalence of the three sites can be further illustrated via radial distribution functions (RDFs), which show the quantitative difference in atomic locations (Fig. 2d). Although HRTEM (Fig. 1c) and STEM (Figs. 1d, 2e, g) do not distinguish the three sites, we find that iDPC STEM imaging (Fig. 2f), sensitive to the projected electrostatic potential50,51, shows small changes in contrast that are statistically significant (Fig. 2h, Methods) and are confirmed by multislice simulations (Supplementary Fig. 4). iDPC can therefore detect the true 32 Å moiré cell at the MoS2/Au{111} interface. However, although this modulation is qualitatively and statistically observable, the translation and rotation of the quasi-vdW islands, as in Supplementary Movie 1, preclude a quantitative analysis.Fig. 2Atomic models, RDFs, HAADF, and iDPC characterization of 32 Å moiré structure.a Atomic model [100] zone axis for the 32 Å moiré with the Au atoms indicated (orange, red, blue) to highlight relative stacking of A, B, and C sites. b Plan view atomic model for the 32 Å moiré. Boxed areas represent three inequivalent sites in the 32 Å moiré. c Close up plan-view image of each of the sites highlighted in (b). r is the projected distance from the central aligned sites, and the dotted circles show two representative r values. d Corresponding RDFs of the three inequivalent sites in the 32 Å moiré. e HAADF and f iDPC STEM images showing the (apparent) 18 Å and 32 Å moiré cells, respectively. Scale bars 200 Å. g, h Relative intensity distributions and statistical variation of the three inequivalent sites in the island immediately above in the corresponding images. The equivalent disk radius for each spot was calculated and partitioned to inequivalent sites (red, green, blue). The histograms were smoothed using a gaussian kernel of radius 0.5 Å for visual clarity.Although visible, the iDPC signal from the 32 Å moiré is weak. To consider the full set of spatial frequencies of the Au{111} FCC crystal and obtain a clear real-space image of the 32 Å moiré, we turn to a reciprocal space convolution theorem to predict the entire spectrum of possible moirés in the MoS2/Au{111} system (Methods). The geometric interpretation of the convolution theorem indicates that periodicities arise from the pairwise vectors connecting all spatial frequencies of the MoS2 and Au lattices52 (Fig. 3a). In Supplementary Fig. 5 and Supplementary Table 1, we calculate these periodicities and intensities as a function of rotation angle between the two crystals. The four largest periodicities are shown in Fig. 3b. Additional higher order moirés are also predicted which often exhibit smaller periodicities and weaker intensities (Supplementary Fig. 5). At zero rotation, we indeed recover the 32 Å moiré period, alongside the apparent 18 Å moiré (Fig. 3b). Note that 32 Å moiré periodicity is obscured by the 230% higher intensity reflections of the 18 Å period convolution. We confirm this assignment of moiré periods by showing the experimental fast fourier transform (FFT) of Fig. 1c (Methods, Fig. 3c). The moiré superlattice periods emerge as two sets of satellite peaks around the central beam spot. The simulated diffraction pattern in Fig. 3d is in quantitative agreement with the FFT of the acquired image (Fig. 3c), predicting all the higher order moiré periodicities at the 2D/3D interface.Fig. 3Geometric convolution technique to predict moiré spectrum.a Schematic representation of satellite spot generation. Spatial frequencies due to a single lattice shown in the top left panel (orange) are overlaid on those arising from a second lattice on the top right (cyan). The convolution of these two sets of spatial frequencies (i, ii) can be understood as the pairwise vectors connecting spatial frequencies of the two lattices (iii, iv, v - bottom, left). These convolutions generate moiré frequencies (iii, iv, v) shown as black dots in the bottom right panel. b Calculated moiré period vs rotation angle for the four largest moiré supercells in the MoS2/Au{111} system, illustrated for small (±1%) Au lattice strain. Dot dashed lines represent 0% strain, while the two solid lines on either side represent ±1 strain as a bound. Black dashed lines represent the experimentally observed moiré periods from the FFT, two of which (18 Å and 32 Å) are predicted at 0° relative rotation angle. The moirés are color coded according to the reflections they arise from, with blue arising from the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{2\overline {11}0\}_{\rm{MoS}_{2}}$$\end{document}{211¯0}MoS2: {220}Au, green \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{10\bar 10\}_{\rm{MoS}_{2}}$$\end{document}{101¯0}MoS2:1/3{422}Au, orange \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{6\overline {33}0\}_{\rm{MoS}_{2}}$$\end{document}{633¯0}MoS2: {642}Au, and grey \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{20\overline 20\}_{\rm{MoS}_{2}}$$\end{document}{202¯0}MoS2: {220}Au reflections, respectively. The inset shows the variation of moiré angle with relative rotation angle near 0°. c FFT of atomic resolution HRTEM image of the MoS2/Au{111} image in Fig. 1c showing 1/3{422} reflection and two visible moiré periodicities around the central spot. Illustrative orange dots represent frequencies from Au crystal planes, while purple represent frequencies from MoS2 crystal planes. Scale bar, 0.5 Å−1. d Simulated FFT for Au/MoS2 generated via the geometric convolution technique with each spot colored to show its origin (orange: Au, purple: MoS2, blue: 32 Å crystallographic moiré, green: apparent 18 Å moiré). Area of spots is proportional to absolute intensity, but with inner moiré spots magnified 2x for clarity. Scale bar, 0.5 Å−1.To extract a real space image of the weak 32 Å moiré, we employ the technique of 4D STEM (Fig. 4a)39. Subsequently we select, with a virtual ADF detector, an annular area of each diffraction pattern to reconstruct an image from the average pattern (Fig. 4b) using certain diffraction spots only. Using an annulus that includes the Au{220} spots and the MoS2 {2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline {11}$$\end{document}11¯0} spots (Fig. 4c, Methods), we observe the high intensity 18 Å moiré pattern (Fig. 4e). The moiré shows uniform periodicity and sinusoidal intensity modulation across the islands. The symmetry is reduced to periodic line patterns in some areas due to sample tilt, but 18 Å periodicity appears across all islands. If instead we generate a second virtual ADF image using the weaker 1/3{422} Au and {10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline 1$$\end{document}1¯0} MoS2 reflections (Fig. 4d), we observe a hexagonal pattern of spots with 32 Å moiré periodicity, consistent with our predictions from geometric convolution and the true crystallographic moiré accounting for 3D structure (Fig. 4f).Fig. 44D STEM imaging of 18 Å and 32 Å moiré periodicities.a Schematic of 4D STEM technique showing rastered beam (red) on MoS2/Au{111} with corresponding CBED pattern at each point. The green and blue scattered beams are centred on the spots of the annuli shown in (c and d). b CBED pattern formed by averaging patterns collected over the entire scan area, c 4D STEM annulus used to isolate 18 Å moiré periodicity (angular range 31–43 mrad, green) and d 4D STEM annulus used to isolate 32 Å periodicity (angular range 11–24 mrad, blue). e, f Virtual ADF STEM images revealing 18 (green) and 32 Å (blue) period moirés, respectively. Scale bar, 200 Å. Insets show unit cells.Charge density modulationTo explore the impact of the moiré periodicity on ground state charge density of our 2D/3D structure, we next turn to ab initio electronic structure calculations (Methods). Figure 5a shows a calculated isosurface of ground-state charge density difference for MoS2/Au{111} in side-view. The charge density difference is concentrated at the interface, specifically on the upper S layer of atoms, with some penetration to the underlying Mo layer. On the Au side, the charge density difference is concentrated on the first atomic plane, with negligible charge density found in the second Au{111} layer. The charge density modulation due to the 2D/3D interface indeed has a periodicity of 32 Å (Fig. 5b). To quantify the effect of moiré modulation on band structure and density of states, the supercell electronic states can be unfolded onto a single MoS2 unit cell (Fig. 5c). Accounting for the 32 Å moiré, band structure calculations are in agreement with prior angle resolved photoemission spectroscopy and scanning tunnelling spectroscopy measurements of the MoS2/Au{111} system53.Fig. 5Electronic structure calculations at the 2D/3D interface.a Charge density difference viewed down a [100] cross section of the 32 Å commensurate moiré (11Au x 10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_{\rm{MoS}_{2}}$$\end{document}MoS2 superstructure). Orange atoms represent Au, yellow S, and purple Mo. Purple denotes negative and green positive charge density isosurface contours. b Calculated charge density difference at the MoS2/Au{111} interface, as viewed down the [001] axis, showing electronic modulation following the 32 Å moiré periodicity. Black line indicates the 32 Å crystallographic moiré unit cell. c Unfolded band structure for MoS2/Au{111} system (left) and corresponding density of states (right). Color corresponds to the band’s spectral weight.Application of the method to the hBN/Au{111} interfaceTo explore the generality of the geometric convolution technique, we also apply it to the hBN/Au{111} structure (Supplementary Fig. 6). Here, the Au lattice is rotated by 10° with respect to the hBN, leading to a more complex situation than the symmetric 0° epitaxy of MoS2/Au{111}. This 10° rotation leads to a strong moiré periodicity of 11 Å. Prediction via geometric convolution technique is necessary in these rotated systems to uncover higher order moirés. For this interface, the convolution technique predicts and explains an additional 19 Å periodicity observed experimentally and re-creates the experimental diffraction pattern (Supplementary Fig. 6). The predictions of the convolution technique rely solely on inputs of crystal structure, lattice parameters, and rotation. This analysis, as well as previous literature, illustrate the wide applicability of the technique in moiré analysis52.DiscussionFor the MoS2/Au{111} interface, the combination of several different imaging techniques with electronic calculations provides a clearer picture of the moiré structure than is possible with any single measurement. In HRTEM the 18 Å moiré is the strongest visible, leading to the possibility of erroneously predicting that electronic properties should be modulated with this period. Our calculations reveal that electronic modulation instead follows the true crystallographic 32 Å periodicity. This periodicity is hidden in conventional TEM due to projection effects. Instead, for this interface, 4D STEM imaging, combined with a geometric convolution analysis of the full moiré spectrum, allows a direct real-space observational link between atomic structure and moiré-induced electronic modulation at this 2D/3D interface. The combination of analysis techniques also explains the discrepancy between moiré patterns observed by TEM and STM at this 2D/3D interface. These results highlight electronic modulation at the 2D/3D interface, and showcase the growing opportunities for advanced STEM techniques for direct imaging of moiré structures at the atomic scale.We envision that the coupled application of 4D STEM and the geometric convolution theorem, presented here for analysis of the 2D/3D interface, could also be extended to the direct imaging of higher order moirés in systems with multiple interfaces and could expand opportunities across the field of moiré engineering. Potential applications lie in multiple overlaid moiré superlattices, which have been found to coexist in vdW heterostructures such as hBN-graphene-hBN stacks11,12, or in so-called “moiré of moirés” structures arising from relaxation of twisted trilayer graphene and WSe254. Although the effects of these coexisting moirés have been reported, they have not yet been directly imaged. This is because the overall moiré observed in HRTEM and STEM is a convolved projection of all the moirés in the system. Using 4D STEM and geometric convolution, moiré characterization could in theory be performed at each interface in the structure by highlighting the relevant diffraction spots. Virtual ADF images could then be used to decouple and directly image each separate moiré. Moreover, 4D STEM could enable simultaneous mapping of crystal orientation, strain, sample thickness, polarization, electric fields, and 3D ptychographic reconstructions of relevant moiré structures39. To date, most 2D/3D moiré investigations (including this study) have focussed on epitaxially grown interfaces exhibiting a single orientation. However, future practical development of 2D/3D moiré engineering will require complete control of the structure and orientation of 2D and 3D materials. Emerging fabrication methods using direct transfer of a 3D metal, such as Au, onto 2D materials55, or nanomechanical rotation of a 3D nanocrystal using AFM or STM cantilevers56, suggest that such control of interfacial orientation is increasingly feasible, extending opportunities of 2D/3D moiré engineering.MethodsSuspended MoS2 sample fabricationCustom TEM chips were fabricated that include a SiNx membrane supported on Si, with 9 holes each 4 µm in diameter. We employed a wedging transfer process to suspend MoS2 on these TEM grids57. Thermally grown 90 nm SiO2/Si wafers were pre-treated with oxygen plasma and MoS2 was mechanically exfoliated onto them using the conventional Scotch tape method. Flakes of suitable thickness were identified by their contrast in optical microscopy. A solution of 25 g cellulose acetate butyrate (CAB) in 100 ml ethyl acetate was spin coated onto the sample and baked at 80 °C for 6 min. MoS2 flakes were transferred to the TEM grids using a wedging transfer technique57. Here, a scalpel is used to cut the CAB around the desired flake. A drop of deionised water can then be intercalated between the CAB and SiO2/Si surface and the entire flake transferred to the TEM grid with the CAB polymer handle using a tweezers. The transferred flakes were baked at 140 °C for 5–10 min to improve adhesion. After dissolving the CAB in acetone for 15 min, the flakes were dipped in isopropanol and dried using a critical point dryer.Ultra-high vacuum (UHV) epitaxial depositionTo create epitaxial nanoislands, UHV deposition is used. This reduces impurities trapped at the metal-2D interface58. The main source of interfacial impurities is polymer residues, which create heterogeneous nucleation sites. Therefore, polymer residue remaining on the 2D material nucleates non-epitaxially aligned nanoislands (Supplementary Fig. 1). The combination of CAB polymer and heat treatment is effective in removing carbon and polymeric contamination31; material transferred using other polymers such as PMMA cannot be cleaned as effectively. MoS2/SiNx substrates were loaded into a UHV sample preparation chamber and cleaned of residual polymer by heating resistively in UHV to ~550 °C for several hours. Au deposition was carried out in the same multichamber UHV system (base pressure 2 × 10−9 Torr), and was deposited in a homebuilt K-cell, using sheet metal placed in a BN crucible, at a rate of 0.5 Å/min. The deposited thickness was calibrated by measuring the evaporation rate with a quartz crystal microbalance immediately before and after deposition. AFM analyses of island thickness were performed in a Veeco Metrology Nanoscope V in tapping mode. There is no intentional heating during deposition, but thermocouple measurements show that the sample temperature rises to 50–60 °C.TEM imaging and data analysisA field-emission TEM (JEOL 2010F) was used for selected area electron diffraction and bright-field imaging, operated at 200 kV. HRTEM imaging was performed with a Hitachi HF-3300V with CEOS BCOR imaging aberration corrector, operated at 60 kV. Figure 1c was obtained from a drift corrected mean of 25 images, where each image was an 8 second exposure, so the total exposure on to the camera was 200 seconds. The electron flux was 500 e−/Å2/sec so the final image exhibited a total ~100,000 e−/Å2. Drift tracking over the images gave an average drift of < 7 pm/sec, although most images exhibited less drift. FFTs and line-scans were obtained using Fiji ImageJ software. FFTs of the real-space image are used for observing moiré peaks instead of SAED patterns since moiré peaks were not readily observed at the energies (80–300 keV) used in TEM59. The FFT in Fig. 3c was produced by multiplying the source image (Fig. 1c) by a Hanning window prior to taking its FFT to minimise streaking incurred due to the hard edge of the image.Multislice image simulationsSTEM image simulations in Supplementary Fig. 4 were performed using an orthorhombic supercell consisting of 3 Au layers on an MoS2 monolayer (7956 atoms), sliced along the [001] direction. A repeating unit from the supercell was cropped and simulated using custom Python-based STEM image simulation software. Simulation parameters similar to experiments were used, with an accelerating voltage of 60 kV, convergence angle of 24.7 mrad, and collection angles of 25–153 mrad (ADF) and 6–24 (iDPC). Simulated ADF and iDPC images were convolved with a gaussian kernel having FWHM of 80 pm, approximately accounting for the finite effective source size.4D STEM imaging and data analysis4D STEM imaging was performed with a probe-corrected Thermo Fisher Scientific Themis Z G3 60–300 kV S/TEM operated at 60 kV with a beam current of 50–60 pA in the microprobe mode and a semi-convergence angle of 5.42 mrad, using an Electron Microscopy Pixel Array Detector. The equivalent probe size used in Fig. 4 was ~1 nm and the pixel size was 0.813 nm. Virtual ADF STEM images were generated from the 4D STEM dataset using virtual detectors in the ‘4D STEM Explorer’ program60. The HAADF and iDPC images in Fig. 2 were acquired at 200 kV, 25 mrad convergence angle, and a current of 30 pA. The quantification in Fig. 2g,h was performed as follows: input images were convolved with the Laplacian of a gaussian kernel with radius 3.75 Å prior to peak detection; peaks were segmented using a watershed transform and an equivalent disk radius for each spot was calculated and partitioned to inequivalent sites (red, green, blue); a Student t-test was used to test the null hypothesis that the different sample means were equal, at the 0.001 significance level.Geometric convolution techniqueThe geometric convolution code was implemented in the computational package Wolfram Mathematica 12.0 and builds on a model previously described for hexagonal lattices52. Frequencies arising from the superposition of the two lattice functions were obtained by the convolution theorem, F{t x b} = F{t} ⊗ F{b}, where t and b are the top and bottom lattice functions respectively, F{\,} denotes the Fourier transform, and ⊗ denotes the convolution operation. All possible spatial frequencies arising from observed spots in the SAED/FFT were initially obtained and we make no assumptions in the simulation other than the bulk structure of Au and its {111} orientation. The full set of spatial frequencies of the FCC crystal along the [111] zone axis were used to calculate the moiré periods for all possibilities within the experimentally observed FFT as a function of the relative rotation, while allowing for small (±1%) Au lattice strain. The angles can also be calculated (Supplementary Fig. 2). We then evaluate the most likely candidates to explain the experimentally measured moiré periods and angles (Supplementary Fig. 5).Electronic structure calculationsThe ground-state charge density difference (Δρ) between the Au/MoS2 heterostructure \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\rho _{{\mathrm{Au}}/{\mathrm{MoS}}_2})$$\end{document}(ρAu/MoS2), and pristine Au (ρAu) and MoS2
\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\rho _{{\mathrm{MoS}}_2})$$\end{document}(ρMoS2) is given by1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{{\Delta}}}\rho = \rho _{{\mathrm{Au}}/{\mathrm{MoS}}_2} - \rho _{{\mathrm{Au}}} - \rho _{{\mathrm{MoS}}_2}$$\end{document}Δρ=ρAu/MoS2−ρAu−ρMoS2Density functional theory calculations were carried out using the projector augmented wave method implemented in the Vienna ab initio simulation package, VASP61,62. We account for the vdW dispersion interactions using the generalized gradient optB86b-vdW functional63. We use a cut-off energy of 400 eV on an equivalent Monkhorst-Pack k-points grid of 40x40x1 MoS2 unit cell (and similar density supercell). Bandstructure unfolding was performed using the BandUP code64.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Movie 1 | nature communications | [
"Article"
] | [
"Surfaces, interfaces and thin films",
"Electronic properties and materials",
"Two-dimensional materials",
"Imaging techniques"
] | success moiré engineering properties/hexagonal boron nitride heterostructures1,2 twisted bilayer studies extended moiré toolbox double bilayer graphene7 trilayer graphene8 van der Waals (vdW) heterostructures transition metal dichalcogenides9 hBN-graphene-hBN moiré engineering extended beyond 3D/3D oxides13 engineering quasi-vdW interface between 2D material 3D metal Engineering 2D/3D interfaces key applications 2D materials contact 3D material metal moiré engineering requires stacking atomic planes out-of-plane direction 3D stacking introduces additional tuning parameter moiré properties not image moiré superlattices map electronic property modulation atomically-resolved techniques superlattices include reciprocal space imaging low energy electron convergent beam diffraction spatially resolved property measurement scanning tunneling microscopy atomic force microscopy near-field optical microscopy13 infrared nano-imaging5 imaging transmitted intensity high-resolution field transmission electron microscopy STM (S)TEM exhibit real-space atomic resolutionSTM moiré patterns 2D materials graphene Ru27 Ir28 Cu29 STM measurements buried interfaces suspended layers 2D device physics30 (S)TEM provides information suspended moiré systems 2D Interpretation challenging 2D/3D interfaces 3D structure layers discrepancies periodicity measurement imaging MoS2/Au{111} system highlights challenges different values periodicities superlattices STM (S)TEM 18 Å reconcile discrepancies map moiré structure-property relations 2D/3D combine convolution technique STEM imaging techniques differential phase contrast four-dimensional (4D) STEM decouple higher order moiré patterns investigate MoS2 /Au{111} model 2D/3D system TMDC (opto- examine hBN/Au{111} plasmonics34 iDPC STEM measures sample transmission function interpretation projected electrostatic potential thin 4D STEM pixelated array detector CBED pattern each probe position STEM resulting 4D dataset filtered post-acquisition reconstructions bright field annular dark field ptychography iDPC394D STEM applied materials Cu40 ZrO241 LiNiO242 DyScO343 MoS245 WS246 2D materials-suited small thickness47 iDPC 4D STEM decouple higher moiré periods space images moiré pattern 2D/3D interface MoS2/Au{111} crystallographic 32 Å period explain difference projection effects ABC stacking 3D metal electronic structure calculations MoS2/Au{111} charge density modulation concentrated interface follows 32 Å moiré periodicity findings demonstrate utility direct imaging iDPC 4D STEM structure electronic properties 2D/3D heterostructures MoS2/Au{111} example interface Fig. 1. 2D/3D systems formed direct epitaxial ultra-high vacuum conditions samples flat faceted Au{111} nanoislands average edge length 25 nm height 8 nm epitaxially aligned on suspended MoS2{0001} (Fig uniform moiré periodicities diffraction) confirms 0° rotation between Au MoS2 standard deviation 0.2° spots 1/3{422} Au reflections forbidden for FCC structure consistent with Au nanoisland literature49TEM islands single crystalline no dislocations grain boundaries. discontinuity moiré pattern coalescence rigid body displacements twin boundaries stacking faults islands. 1d. STEM epitaxial MoS2/Au{111} moiré epitaxially aligned Au suspended MoS2 SiNx TEM grid Orange atoms Au yellow S purple Mo SiNx membrane dark blueReciprocal space model electron diffraction pattern Au {111} zone aligned MoS2 {0001} weak intensity 1/3{422}Au spots aligned\documentclass[12pt]{minimal}{amsmath-69pt\overline{MoS{101 ̄0}MoS2 higher intensity {220}Au spots aligned[12pt]{minimal}-69pt\overline {11}0{211 ̄0}MoS2. Orange dots frequencies Au crystal planes purple MoS2 crystal planes HRTEM image Au nanoisland MoS2 18 Å-period moiré pattern Scale bar 40 Å. High angle annular dark field (HAADF) STEM image Scale bar 80 Å. Coalescence boundaries marked blue arrows uniform moiré periodicity sinusoidal intensity modulation Au MoS2 lattices undistorted plane interface island edgesdifferent from twisted vdW structures display reconstructions25 distortion to weak quasi-vdW bonding at MoS2/Au{111} Motion islands at room temperature loose binding Au nanoislands to substrate islands exhibit rotation up to 0.3° amplified in moiré pattern MoS2/Au{111} interface characteristics consistent for Au thicknesses uniform across samples.Moiré site MoS2/Au{111} moiré superlattice in Fig. 1c d 18 Å with HRTEM consequence of projective nature conventional)TEM imaging out-of-plane coordinate 3D Au{111} structure ignored in “projected” hexagonal Au lattice atomic spacing 1.66 Å yields moiré pattern 18 Å with MoS2 substrate accurate view electron scattering Au full face-centred cubic Au structure in Fig. 2a, b Au atom from A layer above S atoms site repeats every 32 Å squares Fig 2b Sites similar have Au atoms from B or C layers above S atoms2c-middle-bottom). inequivalence three sites illustrated via radial distribution functions difference in atomic locations (Fig. 2d). HRTEM distinguish sites iDPC STEM imaging shows small changes in contrast statistically significant 2h confirmed by multislice simulations iDPC detect 32 Å moiré cell at MoS2/Au{111} interface modulation observable translation rotation of quasi-vdW islands preclude quantitative analysis.Fig. 2Atomic models RDFs HAADF iDPC 32 Å moiré structure Atomic model zone axis 32 Å moiré Au atoms stacking A B C sites Plan view atomic model Å Boxed areas represent three inequivalent sites Close up plan-view image sites r projected distance from central aligned sites dotted circles show representative r values RDFs of three inequivalent sites HAADF iDPC STEM images showing 18 Å 32 Å moiré cells Scale bars 200 Å Relative intensity distributions statistical variation of three inequivalent sitesequivalent disk radius spot calculated partitioned inequivalent sites green histograms smoothed gaussian kernel radius 0.5 Å clarity iDPC signal 32 Å moiré weak spatial frequencies Au{111} FCC crystal image 32 Å moiré reciprocal space convolution theorem predict moirés MoS2/Au{111} system periodicities from pairwise vectors connecting frequencies MoS2 Au Supplementary Fig 5 calculate periodicities intensities function rotation angle crystals four largest periodicities Fig. 3b Additional higher order moirés predicted smaller periodicities weaker intensities zero rotation recover 32 Å moiré period 18 Å moiré 32 Å moiré periodicity obscured by 230% higher intensity reflections 18 Å period convolution fast fourier transform) Fig. 1c moiré superlattice periods emerge satellite peaks around central beam spot simulated diffraction pattern Fig. 3d FFT higher order moiré periodicities 2D/3D interface. 3Geometric convolution technique predict moiré spectrum satellite spot generationfrequencies single lattice top left panel overlaid second lattice right convolution frequencies (i ii pairwise vectors connecting (iii iv v convolutions generate moiré frequencies (iii iv v black dots bottom right panel moiré period rotation angle four largest supercells MoS2/Au{111 system small (±1%) lattice strain Dot dashed lines 0% strain two solid lines ±1 strain Black dashed lines observed moiré periods two (18 Å 32 Å) predicted 0° rotation anglemoirés color coded reflections blue[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek-69pt}{document}{2\overline {11}0\{MoS}{document{211}MoS2: {220}Au, green[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}-69pt}{document}{10\bar 10\}{MoS}{2}}{document{101 ̄0}MoS2:1/3{422}Au, orange[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}{6\overline {33}0\{MoS}{2}}{document{633 ̄0}MoS2: {642}Au, grey[12pt]{minimal}{amsmath}{wasysym{amsfonts-69pt}Au reflections variation moiré angle rotation angle 0° atomic resolution HRTEM image MoS2/Au{111} image Fig. 1c 1/3{422} reflection two moiré periodicities central spot orange dots Au crystal planes purple MoS2 planes Scale bar 0.5 Å−1 Simulated FFT Au/MoS2 geometric convolution spot colored origin (orange Au purple MoS2 blue 32 Å moiré green 18 Å Area spots proportional intensity inner moiré spots magnified 2x clarity Scale bar 0.5 Å−1 extract image weak 32 Å moiré 4D (Fig. 4a)39 select virtual ADF detector annular area diffraction pattern reconstruct image average pattern (Fig. 4b) diffraction spotsannulus Au{220} spots MoS2[12pt{amsmath{wasysym-69pt spots (Fig. 4c high intensity 18 Å moiré pattern (Fig. 4e). moiré shows uniform periodicity sinusoidal intensity modulation across islands symmetry reduced to periodic line patterns sample tilt 18 Å periodicity across all islands second virtual ADF image weaker 1/3{422} Au[12pt{minimal{amsmath-69pt MoS2 reflections (Fig. 4d), hexagonal pattern of spots 32 Å moiré periodicity consistent predictions geometric convolution true crystallographic moiré 3D structure (Fig. 4f).Fig. 44D STEM imaging of 18 Å 32 Å moiré periodicities Schematic 4D STEM technique rastered beam on MoS2/Au{111} CBED pattern each pointgreen blue beams on annuli (c d). CBED pattern patterns scan area 4D STEM annulus 18 Å moiré periodicity 31–43 mrad 32 Å 11–24 mrad Virtual ADF STEM images 18 32 Å moirés Scale bar 200 Å Insets show unit cells.Charge density impact moiré periodicity ground state charge density 2D/3D structure electronic structure calculations Figure 5a ground-state charge density difference MoS2/Au{111} side-view difference concentrated interface upper S layer penetration Mo layer Au side first atomic plane negligible second Au{111} layer charge density modulation 2D/3D interface periodicity 32 Å (Fig. 5b). moiré modulation band structure supercell electronic states unfolded single MoS2 unit cell (Fig. 32 Å moiré band structure calculations photoemission spectroscopy scanning tunnelling spectroscopy measurements MoS2/Au{111}. 5Electronic structure calculations 2D/3D interfaceCharge density difference [100] cross section 32 Å moiré (11Au x 10[12pt{amsmath{wasysym{upgreek\oddsidemargin-69pt}}MoS2 superstructure). Orange atoms represent Au yellow S purple Mo Purple negative green positive charge density isosurface contours Calculated charge density difference MoS2/Au{111} interface [001] axis electronic modulation 32 Å moiré periodicity Black line 32 Å crystallographic moiré unit cell Unfolded band structure MoS2/Au{111} system density states Color band’s spectral weight.Application method hBN/Au{111} geometric convolution technique hBN/Au{111} structure Fig. 6) Au lattice rotated 10° hBN complex situation 0° epitaxy 10° rotation strong moiré periodicity 11 Å Prediction geometric convolution technique necessary uncover higher order moirés predicts additional 19 Å periodicity re-creates experimental diffraction pattern predictions crystal structure lattice parameters rotationanalysis previous literature illustrate applicability technique in moiré MoS2/Au{111} interface combination imaging techniques electronic calculations provides clearer picture moiré structure single measurement In HRTEM 18 Å moiré strongest electronic properties calculations reveal electronic modulation follows crystallographic 32 Å periodicity hidden in TEM projection effects 4D STEM imaging geometric convolution analysis allows direct link between atomic structure moiré electronic modulation 2D/3D explains discrepancy between moiré patterns TEM STM results highlight electronic modulation showcase opportunities for advanced STEM techniques direct imaging moiré structures atomic scale application 4D STEM geometric convolution theorem to imaging higher order moirés in systems multiple interfaces expand opportunities moiré engineering Potential applications in overlaid moiré superlattices in vdW heterostructures hBN-graphene-hBN “moiré of moirés” structures from relaxation twisted trilayer graphene WSe254 effects coexisting moirés reported not directly imaged moiré in HRTEM STEM convolved projection of moirés Using 4D STEM geometric convolution moiré characterization could performed at each interface highlighting diffraction.ADF images decouple image moiré 4D STEM mapping crystal orientation strain sample thickness polarization electric fields 3D ptychographic reconstructions moiré structures39 2D/3D moiré investigations epitaxially grown interfaces single orientation future control structure orientation 2D 3D materials fabrication methods transfer 3D metal 2D nanomechanical rotation 3D nanocrystal control interfacial orientation feasible opportunities 2D/3D moiré engineering.MethodsSuspended MoS2 sample fabricationCustom TEM chips SiNx membrane Si 9 holes each 4 μm wedging transfer process suspend MoS2 Thermally grown 90 nm SiO2/Si wafers pre-treated oxygen plasma MoS2 exfoliated Flakes identified by contrast optical microscopy 25 g cellulose acetate butyrate) in 100 ml ethyl acetate coated baked at 80 °C 6 min MoS2 flakes transferred to TEM grids wedging transfer scalpel CAB flake deionised water intercalated between CAB SiO2/Si surface flake transferred to TEM grid flakes baked at 140 °C 5–10 min adhesiondissolving CAB acetone 15 min flakes dipped isopropanol dried dryer-high vacuum (UHV) epitaxial UHV reduces impurities metal-2D source polymer residues heterogeneous nucleation sites polymer residue 2D nucleates non aligned nanoislands CAB polymer heat treatment carbon polymeric material polymers PMMA MoS2/SiNx substrates loaded UHV sample preparation chamber cleaned residual polymer UHV ~550 °C hours deposition multichamber UHV system pressure 2 × 10−9 deposited homebuilt K-cell sheet metal BN crucible 0.5 Å/min deposited thickness calibrated evaporation rate quartz crystal microbalance deposition analyses thickness Veeco Metrology Nanoscope V no intentional heating deposition thermocouple measurements sample temperature rises 50–60 °C imaging data field-emission TEM 2010F) electron diffraction bright-field imaging 200 kV HRTEM imaging Hitachi HF-3300V CEOS BCOR imaging aberration corrector 60 kV Figure 1c drift corrected 25 images 8 second exposure total exposure 200 secondselectron flux 500 e−/Å2/sec final image ~100,000 e−/Å2. Drift average < 7 pm/sec most less drift FFTs line-scans Fiji ImageJ software FFTs moiré peaks SAED (80–300 keV TEM59 FFT Fig. 3c multiplying source image 1c Hanning window streaking.Multislice image simulationsSTEM simulations Fig. 4 orthorhombic supercell 3 Au layers MoS2 monolayer (7956 sliced [001] direction repeating unit cropped simulated Python STEM image simulation software parameters accelerating voltage 60 kV convergence angle 24.7 mrad collection angles 25–153 mrad (ADF) 6–24 (iDPC). ADF iDPC images convolved gaussian kernel FWHM 80 pm finite effective source size.4D STEM imaging data probe-corrected Thermo Fisher Scientific Themis Z G3 60–300 kV S/TEM beam current 50–60 pA semi-convergence angle 5.42 mrad Electron Microscopy Pixel Array Detector probe size 4 ~1 nm pixel size 0.813 nm Virtual ADF STEM images generated STEM Explorer’HAADF iDPC images Fig. 2 acquired at 200 kV 25 mrad convergence angle current 30 pA quantification Fig. 2g input images convolved with Laplacian gaussian kernel radius 3.75 Å peak detection peaks segmented watershed transform equivalent disk radius calculated partitioned to inequivalent sites Student t-test null hypothesis sample means 0.001 significance level.Geometric convolution Wolfram Mathematica 12.0 model hexagonal lattices52 Frequencies superposition lattice functions obtained convolution theorem F{t x b} = F{t} F{b} t b top lattice functions Fourier transform convolution operation spatial frequencies from spots SAED/FFT obtained no assumptions bulk structure Au {111} orientation full spatial frequencies FCC crystal [111] zone axis calculate moiré periods possibilities small (±1%) Au lattice strain angles calculated Fig. 2) evaluate likely candidates explain moiré periods angles Fig. 5)Electronic structure ground-state charge density difference (Δρ Au/MoS2 heterostructure[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin{-69pt}\rho\mathrm{Au}}{MoS}}_2}{document}(ρAu/MoS2) pristine Au (ρAu) MoS2[12pt]{minimal}{amsmath{wasysym}}\oddsidemargin{-69pt}\rho\mathrm{MoS}}_2}{document}(ρMoS2)[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}\mathrm{{\Delta}}}\rho = \rho _{{\mathrm{Au}}\mathrm{MoS}}_2} -\rho\mathrm{Au}}} -{MoS}}_2\end{document}=ρAu theory calculations projector augmented wave method Vienna initio simulation package VASP61,62 vdW dispersion interactions generalized gradient optB86b-vdW cut-off energy 400 eV Monkhorst-Pack k-points 40x40x1 MoS2 unit cell density Bandstructure unfolding BandUP code64.Supplementary Review Additional Movie | 49 | 0.584603 |
10.1038/s41467-020-17515-8 | PMC7393140 | Here, the authors apply live-cell and in situ fluorescence imaging at the single-molecule level to examine lambda DNA replication in single cells, finding that individual phage DNAs sequester host factors to their own vicinity and confine their replicated DNAs into separate compartments, suggesting that phage decision-making transcripts are spatially organized in separate compartments to allow distinct subcellular decisions to develop. | Spatial organization of biological processes allows for variability in molecular outcomes and coordinated development. Here, we investigate how organization underpins phage lambda development and decision-making by characterizing viral components and processes in subcellular space. We use live-cell and in situ fluorescence imaging at the single-molecule level to examine lambda DNA replication, transcription, virion assembly, and resource recruitment in single-cell infections, uniting key processes of the infection cycle into a coherent model of phage development encompassing space and time. We find that different viral DNAs establish separate subcellular compartments within cells, which sustains heterogeneous viral development in single cells. These individual phage compartments are physically separated by the E. coli nucleoid. Our results provide mechanistic details describing how separate viruses develop heterogeneously to resemble single-cell phenotypes. | IntroductionOrganization is a fundamental part of life. For complex organisms, the spatial development of body parts is controlled for proper function1. In the cells comprising these organisms, separate organelles are organized by membranes2. Bacterial cells utilize proteins to localize processes in lieu of intracellular membranes3. Furthermore, the physical inhomogeneity of cytoplasmic DNAs, RNAs, and proteins favor the segregation of components for different processes4–6. Viruses are even simpler than cellular life, but have also been reported to organize their development within cytoplasmic inclusions or proteinaceous compartments7,8.Bacteriophages, bacterial viruses, are among the simplest biological systems and serve as models for advanced cellular processes. High-resolution fluorescence microscopy and mathematical modeling have been used to examine the phage lambda lysis–lysogeny paradigm for cellular decision-making to uncover surprising phenomena9–11. By labeling single virus particles, we characterized that separate votes for decisions by co-infecting lambda phages determined cell fates12. We next provided distinct voices for separate phages to expound upon our voting model by incorporating different lytic and lysogenic reporters into different phages, finding that phage DNAs compete and cooperate as subcellular individuals13. It is curious how identical viral DNA molecules can commit to divergent trajectories while inhabiting a single cellular environment. While stochasticity is understood to definitely influence development, where noisy elements of gene expression might partially explain differential subcellular behaviors14, it is also evident that properly targeted, high-resolution studies can elicit molecular mechanisms beyond the intrinsic stochasticity of cellular biochemistry15–17. We hypothesize that subcellular organization allows different phages to develop as individuals in a single cell and expect to detect the underlying subcellular heterogeneity by specifically investigating the spatial distribution of viral/host biomolecules. Here, we show that cascading events during viral transcription, DNA replication, and gene expression combine to establish an organized subcellular unit of phage development. This organization permits multiple individual viruses in the same cell to develop via different pathways in separate areas of that cell.ResultsLive-cell fluorescent reporters of phage developmentTo work toward a unified model of individual lambda development in cellular space, we used live-cell time-lapse microscopy targeting the initial infecting phage DNAs, host cell’s replication resources, replicated phage DNAs, and phage decisions (Fig. 1a). To visualize the initial infecting phage DNA, we modified cells to be dam− and carry a seqA-mKO2 (mKO2 signal defined as yellow) translational fusion18. This system has been validated previously to correctly label single DNA molecules from infecting phages11,13,18. Upon ejection of methylated phage DNA into the host, SeqA-mKO2 binds exclusively to the ejected phage DNA and any DNA copies retaining the methylated parent strands with single-DNA labeling sensitivity, but not subsequent replicated DNA copies (SeqA system, Fig. 1b)11,18. Since the SeqA system cannot target all replicated phage DNAs, we recombineered an array of tetO sequences into the phage genome. With the host cell harboring a TetR-mCherry plasmid (mCherry signal defined as red), all phage genomes are bound and labeled at tetO sites by TetR (Tet system, Fig. 1d) (Supplementary Discussion). The Tet-labeling scheme lacks single-DNA sensitivity under our experimental conditions, so we used both Tet and SeqA systems to target phage DNA. As lambda depends on host factors for viral DNA replication, we translationally fused the Escherichia coli helicase19, DnaB, with mTurquoise2 (mTurquoise2 signal defined as blue) by replacing the native dnaB gene with dnaB-mTurquoise2 on the E. coli chromosome (Fig. 1c; Supplementary Fig. 2f). DnaB is essential for phage/E. coli DNA replication and directly interacts with lambda P (analog of E. coli DnaC)20. The DnaB construct does not appear to impose major detriment on E. coli or phage growth (Supplementary Fig. 2g–j). Finally, we reported lambda lysis–lysogeny decision-making using previously developed systems13. Briefly, we modified phages with a D-mNeongreen (mNeongreen signal defined as green) translational fusion, reporting the lytic pathway because progeny phages are assembled with green gpD, and a cI-mKO2 transcriptional fusion, reporting the lysogenic pathway because the cI operon(s) are expressed during lysogeny (Fig. 1a). Accordingly, we developed a data analysis framework for these reporters to detail the spatial organization of subcellular events during infection (Supplementary Fig. 1). Notably, all presented images of individual cells unambiguously represent single cells, because early expression of the Kil protein by lambda inhibits cell division during infection21.Fig. 1Phage DNAs organize developmental processes into subcellular locations during infection.a Combination of phage processes results in decision-making. Lytic decisions reported by a D-mNeongreen translational fusion and lysogenic decisions reported by a cI-mKO2 transcriptional fusion. b SeqA system detects single molecules of phage DNA. Methylated phages infect dam− cells. Phage DNA is bound by SeqA-mKO2 proteins. Only phage DNA retaining methylation is labeled. c DnaB is an essential DNA replication resource. DnaB-mTurquoise2 fusion protein reports localization of DnaB. d Tet system detects replicated phage DNAs. Phage DNA bearing tetO arrays is labeled by TetR-mCherry binding. e Representative infected cell with reporters described in a–d undergoes lytic development. Representative cells in e and g chosen from three independent infection experiments. * indicates contrast is adjusted for each time point for clarity. DnaB and replicated DNA fixed contrast images are shown in Supplementary Fig. 2b–e. All scale bars in this figure are 2 μm. f Kymograph of the cell in e. Explanations for data analysis in Supplementary Fig. 1 and Supplementary Discussion. Fluorescence is normalized to the population maximum. g Representative infected, lytic cell with two subcellular areas of development. h Kymograph of the cell in g. Fluorescence is normalized to the population maximum. i DnaB heat maps for lytic cells at their DnaB appearance time point are arranged by the position of DnaB. Cell to the left of i describes how location is represented for i–l. Fluorescence of each cell is normalized to its own peak brightness for i–l. n = 91 cells for i–l. j SeqA heat maps for lytic cells are arranged in the same order and time points as in i to compare SeqA and DnaB. k TetR heat maps for lytic cells at their TetR cluster appearance time point are arranged by the position of TetR. l DnaB heat maps for lytic cells are arranged in the same order and time points as in k to compare DnaB and TetR. Source data are provided as a Source Data file.Organization of resources and replication by individual phages in single cellsWe expect that compartmentalization of biomolecules reflects the establishment and perpetuation of individualistic development by lambda. Uninfected cells (LZ1557) displayed diffuse blue, yellow, and red fluorescence, indicating that DnaB, SeqA, and TetR do not compartmentalize without phage infection (Supplementary Fig. 2a; Supplementary Movie 1). Functional DnaB is required for growing cells19; thus, DnaB localization under cell growth conditions represents a basal DnaB state. Importantly, active DnaB-mTurquoise2 does not localize as a focus under phage-free, normal growth conditions. When infecting these cells with our phages (λLZ1576), the fluorescence patterns change significantly from the phage-free state. We primarily focus on analyzing lytic cells because extensive DNA replication is required for successful lytic propagation and note that lysogenic development markedly differs from lytic behaviors (Supplementary Fig. 3b, c; Supplementary Movie 2; Supplementary Discussion). SeqA foci appear within the cells, typically early during infection (Supplementary Fig. 3a), indicating that phage DNA entered the cell (Fig. 1e). As phage DNA replication demands host resources, we tracked DnaB. DnaB foci typically formed after SeqA foci and were commonly colocalized with SeqA foci over time (Fig. 1e, f, i, j; Supplementary Fig. 4a, b). This indicates that phage DNA directly alters the natural behavior of DnaB by collecting essential resources to its own location. DnaB foci represent multiple DnaB complexes, specifically aggregated to single phage DNAs for extensive phage DNA replication. In lytic cells, blue foci remained over time (median: 90 min), suggesting that lambda persistently recruits resources for itself (Supplementary Fig. 3b).Resource recruitment precedes phage DNA replication. Red fluorescence is initially spread throughout the cell, representing free TetR as background signal, not phage DNA signal (Supplementary Fig. 2c, e, at 40 min). TetR signal later rearranged into small clusters over time, signifying the production of additional phage DNAs nearby the specific locations of the single DNAs and resources (Fig. 1e–h; Supplementary Movie 3). These subcellular local maxima, or clusters, of TetR signal represent phage DNA. DnaB was pre-localized near the eventual red clusters and remained within the clusters as they expanded within the cell, suggesting that replicating phage DNAs predictably arise at the location where previous DNAs gathered resources, and then maintain this sequestration (Fig. 1k, l; Supplementary Figs. 3d and 4c, d). In lytic cells, gpD (green) signal increased throughout the cell over time (Fig. 1e–h). This signal is initially diffused throughout the cell, but later, green foci, corresponding to phage capsids, preferentially formed in the red clusters (Supplementary Fig. 5a). This suggests that the locations of phage DNA determine where progeny phages are assembled, consistent with the characterized mechanisms of phage DNA packaging22. Altogether, the data indicate that a single phage DNA organizes its own subcellular phage factory, persistently maintaining its clones and resources proximal to itself. We designated this phage-derived, subcellular compartment as a “phactory” for future reference.As the single phage DNA had the capacity to assemble its own compartment, we predicted that multiple, individual phage DNAs could form separate phactories. We identified cells with single phage DNAs in different locations (Fig. 1g, h; Supplementary Movie 4), where each DNA collected its own stockpile of DnaB. Separate red clusters appeared and grew at each DnaB location, and finally, green foci grew into clusters nearby the separated DNA clusters. Intriguingly, the phage-related biomolecules remain separated in space as different viral microenvironments, where the levels and identities of the phage DNA, RNA, and protein in the microenvironments may change and be exchanged while still maintaining spatial separation. These data suggest that each subcellular phage DNA can be its own entity and can organize its own phactory within single cells. In conditions that specifically stimulate spatially segregated phage DNAs to develop, lysogenic induction (Fig. 2a; Supplementary Discussion), multiple phactories separately form and progress in single cells (Fig. 2b–d; Supplementary Figs. 5b and 6). Phactories have unequal DNA and lytic reporter levels over time, suggesting that heterogeneous development is sustained within individual phactories (Fig. 2e; Supplementary Figs. 7 and 8).Fig. 2Organization of multiple heterogeneous subcellular areas during phage induction.a Schematic for the spatial organization of phage development after induction. Lambda prophage is integrated into the bacterial chromosome, and different copies of the chromosome locate to different areas of the cell during growth. Induction of lysogens forces phages to develop in different areas of the cell. The prophage bears the gpD-mNeongreen lytic reporter and carries a tetO array. The cell harbors a TetR-mCherry plasmid and a DnaB-mTurquoise2 reporter. b Overlay images of lysogens after induction. At 0 min the cells were not yet induced (*indicates that the contrast is adjusted for each time point shown for clarity). All scale bars in this figure are 2 μm. c, d Intracellular areas of phage DNA and DnaB form. Histograms of the number of DnaB (c) and replicated DNA clusters (d) are shown for each time point. e Phage DNA replication varies intracellularly. For cells with more than one DNA cluster, the standard deviation of the size of the clusters is represented in boxplots for each time point, as a measure of intracellular phage DNA variability. The median is indicated by the dot at the center of the box, the box bounds the interquartile range of the data, the whiskers span the range of the data excluding the outliers, and the outliers are indicated as individual points. The approximate limit of our resolution is around 250 nm. Source data are provided as a Source Data file.Host nucleoids maintain separation of individually developing phagesOur results indicate that phage DNAs in different phactories remain separated as the basis of their individuality, but do not indicate obvious barriers that segregate phage DNAs, obscuring the detailed mechanisms of their heterogeneous development. We observed that the growth of phage DNA clusters decreased as the cell filled with phage DNA, as if their size might be physically limited by something in the cell (Supplementary Fig. 9a, c; Supplementary Discussion). We hypothesized that replicating phage DNAs are physically separated from E. coli DNA, a presumed barrier shaping the phactories. We tested this hypothesis by labeling E. coli attB with a lacO array and LacI-EYFP construct, adapted from previous work11. We infected these cells (LZ1643) with phages (λLZ1629) (Fig. 3a). Phage DNAs and capsids behave similarly to the above infections with λLZ1576 and LZ1557 (Supplementary Figs. 5c, 9b, d and 10). The terminus of lytic cells is cell lysis, so we tracked the location of attB in the final time points to determine how preceding lytic development affects E. coli DNA, finding that attB locations were biased near the poles (Fig. 3h). Conversely, the terminus of non-lytic development is cell division, and attB was localized between the mid-cell and quarter-cell in non-lytic cells, suggesting differential development (Fig. 3h; Supplementary Fig. 11a; Supplementary Movie 5). We subcategorized the lytic cells based on the interaction of phage DNA with attB (push, spread, and squeeze, Fig. 3b-g; Supplementary Fig. 11b-d; Supplementary Movies 6–8). In the largest class (push, 60%, 108 out of 179 lytic cells), we found that attB was pushed towards one side of the cell, away from a single expanding phage DNA cluster (Fig. 3b, c; Supplementary Movie 6). Within this population of cells, as phage DNA cluster sizes increased over time, attB moved closer to cell poles (Fig. 3i, j). Furthermore, after phage DNA clusters expand near attB, they do not move past it, suggesting that phage and bacterial DNA do not mix together (Supplementary Fig. 11e, f; Supplementary Discussion). These data agree with our previous observations regarding the polar movement of attB in lytic cells11, provide a clear phage-active mechanism, where the spatial expansion of phage DNA due to replication explains the results, and corroborate our hypothesis that E. coli DNA helps determine phactory localization and segregation. Because the two phactories do not merge around the bacterial marker (Fig. 3f), it suggests that bacterial DNA is a physical barrier which allows subcellular viruses to maintain different identities.Fig. 3Bacterial DNA physically separates different intracellular phactories.a Detection of phage and bacterial DNA locations. Phage carries the Tet reporter as in Fig. 1d and the D-mTurquoise2 lytic reporter. Cell carries a lacO array at its attB locus and bears a plasmid expressing both TetR-mCherry and LacI-EYFP to label phage and bacterial DNA. b, d, f Representative lytic cells with different bacterial DNA interactions. Phage DNA will push (b, 108/179 cells), spread (d, 56/179 cells), or squeeze (f, 15/179 cells) bacterial DNA as phage DNA expands. Representative cells chosen from four independent infection experiments. All scale bars in this figure are 2 μm. c, e, g Kymographs corresponding to the cells in b, d, f. Fluorescence is normalized to the population maximum. h Spatial distribution of bacterial DNA in lytic and non-lytic cells differs. In non-lytic cells, the locations of attB for three time points prior to cell division (orange), and in lytic cells, the location of the attB marker for three time points prior to cell lysis (blue) represent the location preference of bacterial loci in different developmental paths. The cell below shows how locations are represented. n = 653 data points for lytic and 382 data points for non-lytic categories. i, j Expanse of phage DNA pushes bacterial DNA. The sizes of phage DNA in each lytic cell that pushes bacterial DNA are shown as violin plots (i). The maximum (nearest to mid-cell) location of attB of the cells in i are shown as violin plots (j). Cell on the right shows how locations are represented for i, j. Cells were oriented such that the TetR clusters were all aligned toward one direction. In the violin plots, the solid line represents the median, and the dashed lines mark the bounds of the interquartile range of the data. Source data are provided as a Source Data file.We next used alternate techniques to examine phage infection, utilizing bacteria/phages without genetically engineered reporters to minimize the likelihood that the genetic modifications influenced biological behaviors. We performed single-molecule fluorescence in situ hybridization (FISH) to characterize the spatial aspect of phage transcription and to examine phage DNA replication in fixed cells, to better characterize how different subcellular viruses develop23–25.To compare FISH with our live-cell techniques, we labeled phage λLZ613 DNA during infection using lambda DNA-specific probes. At early time points, phage DNA existed as small foci/clusters in cells (Fig. 4a). The localization of FISH foci resembles that of SeqA foci in live cells (Fig. 4p, q). At later time points, we found that DNA signal increased with time, and appeared as larger clusters, similar to our Tet reporter system (Fig. 4b). The amount of DNA per cell and within clusters varied (Fig. 4c, d; Supplementary Fig. 12a). To study bacterial DNA, we treated the cells with DAPI in FISH experiments. While DAPI stains DNA non-specifically, the size of E. coli DNA (4.6 Mbp) far outstrips that of lambda DNA (48.5 kbp); thus, DAPI will primarily stain bacterial DNA at early time points before extensive phage DNA replication, allowing phage DNA FISH signal and DAPI to be mutually exclusive (Supplementary Discussion). We observed that the spatial distribution of the DAPI signal in E. coli leaves distinct nucleoid-free zones (Fig. 4h, p; Supplementary Fig. 13b), and phage DNAs preferred these areas in early time points (Fig. 4g, h, j, p; Supplementary Fig. 13a, b), supporting earlier observations of negative bacterial and phage DNA spatial correlation (Fig. 3). Phage DNA and DAPI signals overlap in later time points, after presumed phage DNA replication (Fig. 4j; Supplementary Figs. 12b–e and 13a, b, g). Furthermore, we generated probes against E. coli attB, and found that attB generally localizes with DAPI (Supplementary Fig. 13d–f), but avoids phage DNA locations (Fig. 4g, i, k; Supplementary Fig. 13a, c, h), congruent with our live-cell data (Supplementary Discussion). These results suggest that the organization of heterogeneous, separated phage DNA units exists and is reported faithfully by both live-cell and fixed-cell methods.Fig. 4Phage transcription and DNA replication are collectively organized within nucleoid-free regions.a, b DNA FISH labels phage and bacterial DNA. Representative images of DNA FISH experiments targeting phage DNA and attB at 10 min (a) and 50 min (b) post-infection with cells stained with DAPI. Representative cells chosen from six independent infection experiments. All scale bars in this figure are 2 μm. c, d Violin plots of whole-cell phage DNA signal (c) and sizes of phage DNA clusters (d). In the violin plots, the solid line represents the median, and the dashed lines mark the bounds of the interquartile range of the data. e, f RNA FISH labels phage pR transcript at 6 min (e) and 40 min (f) post-infection. Cells stained with DAPI. Representative cells chosen from six independent infection experiments. g–i Phage DNA (g), DAPI (h), and E. coli attB (i) heat maps arranged by the location of peak brightness of phage DNA at 10 min post-infection. Cell below (g) shows how locations are represented for g–n. Fluorescence of each cell is normalized to its own peak brightness for g–i and l, m. n = 907 cells for g–i. j, k Phage DNA prefers nucleoid-free locations and avoids attB. j Difference maps (see Supplementary Discussion), calculated as g, h, show contrast in locations of phage DNA and DAPI. Negative values are set to 0. k Difference maps as in j, except showing differences between phage DNA and attB. n for j and k = n of c for each time point. l–n Phage mRNA prefers nucleoid-free locations. pR RNA FISH (l) and DAPI (m) heat maps arranged by the location of peak brightness of pR at 15 min post-infection. n Difference map of l, m, similar to j–k. n = 2035 for l–n. o–q Phage DNA and mRNA colocalize away from E. coli DNA. o Histograms of peak location of pR and DAPI from RNA FISH in l, m. p Histograms of peak location of phage DNA and DAPI from DNA FISH (g, h). q Histogram of peak location of SeqA signal from experiments in Fig. 1 (lytic cells, combined first three time points). Cell below (q) shows how locations are represented for o–q. Source data are provided as a Source Data file.Phages maintain localized transcriptionGiven the organization of segregated phage DNAs, our next step was to investigate the organization of phage mRNAs because gene expression is a key component of phage development and decision-making. We studied phage transcription with RNA FISH24, where we first targeted pR, an early transcriptional unit comprising genes for decision-making and phage DNA replication26. At early time points, pR transcripts existed as small clusters (Fig. 4e). We found that pR localizes to areas with lower DAPI signal (Fig. 4l–o; Supplementary Fig. 14a–d), near poles and/or mid-cells, regions well-characterized to typically be nucleoid-free27,28, suggesting that phage mRNAs, like phage DNAs, reside away from nucleoids. Remarkably, even at later time points, pR retains subcellular localization as opposed to diffusing evenly throughout the cell (Supplementary Fig. 14a). Notably, DAPI signals increase in pR locations at later time points following presumed phage DNA replication. From the convergence of our experimental data, we concluded that the locations of pR transcripts in FISH also represent the locations of phage DNA. These results suggest that intracellular phactories possess their own gene expression profiles in single cells. This surprising degree of mRNA spatial organization for lambda may be promoted by the physical attachment of mRNAs to phage DNAs during transcript elongation29, and whole-cell diffusion may be discouraged by the relatively short lifetimes of mRNAs in E. coli30, combined with similarly localized ribosomes translating the transcripts27,31. Notably, phage mRNA localization remains even after transcription is finished and further transcription is blocked with rifampicin treatment32, demonstrating that phage RNAs are continuously localized with phage DNAs in the phactory (Supplementary Fig. 15).Separate decisions made in different locations of the same cellDownstream of phage DNA replication and early gene expression is decision-making, and we predicted that the organization of upstream processes would impact the execution of specific cell-fate expression programs26. The pRʹ transcript encodes the lysis and morphogenesis proteins during lytic development; the pRE/pRM (referred to as pRE) transcripts encode CI to establish and maintain lysogeny. Therefore, we targeted the pRʹ and pRE transcripts with FISH to characterize subcellular decision-making in subcellular space (Fig. 5a, b).Fig. 5Individual phage decision-making occurs in separated subcellular locations.a Colored arrows represent the transcripts targeted by FISH. The arrow direction corresponds to the transcription direction. The colored line segments on the black line mark the approximate locations of the FISH probes relative to the transcripts. b Decision-making transcripts localize to subcellular locations. Representative infected cells at 15 min post-infection with cells DAPI-stained. Representative cells chosen from four independent infection experiments. All scale bars in this figure are 2 μm. c Average pRE FISH signal is plotted against pR signal (Pearson’s ρ in plot, p value <0.001). d–g Phages make different decisions in separate subcellular locations. pR transcripts, separated by DAPI clusters, occupy separate areas. Of 2035 cells, 645 have pRE foci. Of 2035 cells, 439 have pRʹ foci. In all, 504/2035 cells have pRE without pRʹ foci. Of 2035, 298 cells have pRʹ without pRE foci. d–e Representative cells showing lytic decisions in a subset of different locations (75/2035 cells). f–g Representative cells showing conflicting lytic/lysogenic decisions in different locations (48/2035 cells). Representative cells chosen from four independent infection experiments. In total, 141/2035 cells have both pRE and pRʹ in the same cell at any locations. h Model of lambda decision-making in subcellular space. Phage DNAs occupy different subcellular areas, separated by bacterial DNA. Phage DNAs undergo gene expression in their locations, consisting of transcription and translation. It is unknown where key proteins localize after detaching from mRNA. Phage DNAs sequester essential DNA replication resources to their own locations. Phage DNA replication transpires where individual DNAs are, and continued transcription remains localized with phage DNA. Individual units remain separated by bacterial DNA and can differ in composition. Expanding phage DNAs physically push bacterial DNA inside the cell. Individual decisions may be enacted by segregated phage DNAs. Source data are provided as a Source Data file.We performed a set of FISH experiments at 15 min after infection because decision-making is expected to occur around this timeframe24. In support of our model, pR levels vary at different locations of the same cell, more so than pR varies between different cells (Supplementary Fig. 14g). These data reiterate that pR expression is localized and suggest that the variation of localized pR mRNA levels may lead to different localized decisions. The lysogenic decision will eventually repress transcription from pR due to the action of CI26. Accordingly, we found that cellular pRE levels negatively correlate with cellular pR levels (Fig. 5c). Furthermore, focusing on the spatial aspect of transcription, we determined that the cellular coordinates of the brightest pRE signals are offset from the brightest pR signals (Supplementary Fig. 14e, h). These data corroborate earlier hypotheses that pRE/pR head-on transcription might commit the lambda genetic circuit towards lysogeny when pRE dominates33. The data suggest that lysogenic decisions may be locally initiated by individual DNAs, where pRE activation halts pR transcription in a single phage genome, and that these localized actions precede cell-wide repression by CI. As for the pRʹ transcripts, our FISH data show that the brightest pRʹ and pR signals coincide within the cell. Altogether, the data suggest that decisions are enacted by separate phage DNAs in specific subcellular areas (Supplementary Fig. 14f, i).DiscussionOur model for lambda decision-making involves voting for decisions by co-infecting phages, signifying that individual lambda DNAs are capable of heterogeneous development in single cells12,13. Given our conclusions up until now, we would expect that these voting behaviors occur in separate subcellular areas. Accordingly, we found that in our FISH experiments, pRʹ transcripts can exist in a subset of intracellular pR clusters (Fig. 5d, e). We also found that pRE and pRʹ transcripts can coexist in different locations in single cells, suggesting that one intracellular phactory might vote differently from a neighboring phactory (Fig. 5f, g). The organization underpinning these behaviors persists even after new transcription initiation is blocked (Supplementary Fig. 15f–k, m). These data indicate that multiple intracellular lambda DNAs can execute divergent developmental pathways, where the spatial organization of phage/bacterial biomolecules and processes supports their individuality (Fig. 5h).We described how multiple aspects of lambda development coalesce via spatial organization, allowing multiple viruses to develop separately in a cell. We characterized how the single phage DNA collects DnaB, essential for DNA replication, to its own location, begetting its clones and their corresponding development to its own coordinates. We showed that phage DNAs organize multiple transcripts locally at early time points, suggesting that different gene expression patterns are enacted by separate sets of viral DNA in separate cellular spaces, altogether resulting in individual, subcellular compartments of phage development, or phactories. The individual phactory is formed both due to the phage’s own actions, including DnaB recruitment and phage mRNA organization from active and finished transcripts, and also due to the physical separation of the viral biomolecules by the bacterial chromosome. It is unclear to what degree phage proteins are separated during the infection cycle, which is another important element contributing to divergent development in single cells (Fig. 5h).It is intriguing that even simpler viruses, such as lambda, have evolved their own means of organization, which allows relatively complex behaviors to be achieved. When compared to particularly complex phages, such as Pseudomonas phage 201phi2-1, with a genome over 300 kbp encoding over 400 putative genes, lambda is simplistic34. 201phi2-1 houses its own cytoskeletal genes that position and build a protein-enclosed compartment encapsulating its DNA replication and transcription processes separately from its translation, evoking comparison to a nucleus8. Simpler phages, such as phage phi29, may have increased reliance or exploitation of host-specific proteins and processes. Efficient phi29 DNA replication relies on interactions with the host’s MreB cytoskeleton to position phage DNA and replication machinery35. Phi29 DNA replication also depends on the organization of a replication resource, phage DNA polymerase, via the phage-driven actions of its terminal protein36. Phage lambda has a similar strategy to organize replication, as it both utilizes its own proteins to actively reorganize its essential resource, DnaB, and also depends on its host’s architecture, the location of the bacterial nucleoid. The relative simplicity of lambda can also be compared to some high-copy number plasmids lacking partitioning systems, which tend to localize to non-nucleoid locations, perhaps passively37. There are major differences between lambda and plasmid replication. Our data demonstrate that lambda retains its DNA into a single phactory, which pushes the nucleoid as it grows in space. This would obviously be detrimental for plasmids, because a single plasmid cluster in one cellular position would become easily lost during cell division, and if plasmids interfered with bacterial DNA localization, it would disrupt cell viability. Additionally, lambda actively commandeers DnaB, which clearly diverges from plasmid behavior, since this would kill the cell.The application of high-resolution methods to phage lambda has the potential to reveal distinct and novel mechanisms buried within the long-standing lambda paradigm and could be equally applicable in other phage systems for comparison38. Further investigations into the biophysical interplay of viral and bacterial biomolecules, especially those which leverage continually advancing high-resolution techniques capable of fine spatial discrimination, promise new insights into the varied mechanisms behind viral development.MethodsStrains, plasmids, and primersBacterial strains, phages, and plasmids used in this study are listed in Supplementary Table 1.Primers and homology regions used in this study are listed in Supplementary Table 3.Single-cell infection imaging assayHost cells, LZ1557, were grown from a −80 °C frozen permanent stock in 1 ml of M9 + 0.4% maltose (M9M), supplemented with antibiotics, 100 μg/ml ampicillin (Amp100) + 50 μg/ml kanamycin (Kan50) + 10 μg/ml chloramphenicol (Cm10) in a shaker at 37 °C, 265 r.p.m., overnight for ~24 h. This overnight culture was then diluted 1:1000 into 5 ml of M9M + antibiotics as above and grown under the same conditions for ~16–18 h, to OD600 ~0.3–0.4. Then, 1 ml of this culture was pelleted in a tabletop centrifuge, 3000 × g for 4 min at room temperature. During this time, 20 μl of purified reporter phage (λLZ1576) at ~3–4 × 1010 pfu/ml is pipetted into a microcentrifuge at room temperature. After the cells were centrifuged, the supernatant was pipetted away, and the pellet was resuspended in 200 μl of room temperature M9M. Twenty microliters of this suspension was then mixed with the 20 μl of phage solution via gentle pipetting, resulting in an average phage input (API, the phage:bacterium ratio) of ~4, and then 80 μl of room temperature M9M was added to the mixture and mixed via gentle pipetting. This new mixture was then moved to a pre-warmed 35 °C water bath for 4 min to allow for phage adsorption and DNA ejection. During this time, a small (1–1.5 cm2) section from a room temperature M9M agarose pad (all pads in all experiments were 1.5% agarose), freshly made, was set onto a small No. 1 coverslip (18 × 18 mm). Following the incubation of the mixture, 1 μl of the mixture was deposited onto the M9M pad. After the mixture visibly dried, ~1 min, a larger No. 1 coverslip (24 × 50 mm) was overlaid onto the M9M pad, sandwiching it, and the sample was moved to the microscope for time-lapse imaging at 30 °C.For the experiment using bacterial strain LZ1643 and phage λLZ1629, the M9M is supplemented with Amp100 + Kan50 for the bacterial growth. A colony from a plate was grown overnight for ~16–18 h. This overnight culture was then diluted 1:100 into fresh media and grown until OD600 ~0.3–0.4, about 3–4 h. The cells were then pelleted and resuspended as described above. In these experiments, after the phage and cell suspensions were mixed, the mixture was set on ice for 30 min to pre-adsorb the phages before moving the mixture to the 35 °C water bath for 5 min. The following steps up to imaging are identical to the above description.Induction imaging assayA lysogen colony, LZ1596, from a plate was grown overnight in 1 ml of LB + 10 mM MgSO4 (LBM) supplemented with Amp100 + Cm10, ~16–18 h in a 30 °C shaker, 225 r.p.m. The overnight culture was then diluted 1:100 into 5 ml of fresh LBM with the above antibiotics and grown under the same conditions until OD600 ~0.3–0.4. For the 0 min time point, 1 μl of the culture was taken from the flask and deposited onto a phosphate-buffered saline (PBS) agarose pad, similarly as described above with the M9M pad, and then imaged. The lysogen culture was then induced by moving the flask to a 42 °C (our phages bear the cI857 temperature-sensitive allele39), 225 r.p.m. shaking water bath for 15 min. The 5-min time point occurs after 5 min at 42 °C, and the sample was processed and imaged at that time, as the 0 min sample was. The 15-min time point occurs at the end of the 42 °C incubation. While the 15-min time point sample was processed, the induction culture was moved to a 37 °C, 225 r.p.m. shaking water bath, and imaging of the 15-min time point occurred during this 37 °C incubation. The remaining time points were taken as the culture was shaking at 37 °C.Bacterial growth and lysogen induction assaysTo generate bacterial growth curves, bacterial strains were plated on a standard LB agar plate supplemented with appropriate antibiotics. A single colony was used to inoculate a 1 ml overnight culture in LB or M9 + 0.4% maltose (M9M) in a 37 °C, 265 r.p.m. shaker. The overnight culture was then diluted 1:100 into 30 ml of LB or M9 + 0.4% maltose (M9M) supplemented with appropriate antibiotics in a flask and grown in a 37 °C, 265 r.p.m. shaking water bath. OD600 was measured using a spectrophotometer.For lysogen induction assays, a lysogen colony from an LB plate was grown overnight in 1 ml of LB + 10 mM MgSO4 (LBM) supplemented with appropriate antibiotics in a 30 °C, 180 r.p.m. shaking water bath. An overnight culture was then diluted 1:100 into 25 ml of fresh LBM with appropriate antibiotics in a flask and grown under the same conditions until an OD600 of ~0.3, which serves as the 0-min time point. The lysogen culture was then transferred to a 42 °C, 180 r.p.m. shaking water bath for 15 min for thermal induction. The OD600 of the culture was measured at the end of the 42 °C incubation, serving as the 15-min time point. The induction culture was moved to a 37 °C, 180 r.p.m. shaking water bath. The OD600 of the culture was measured every 5 min until lysis with an OD600 of ~0.05 using a spectrophotometer. Subsequently, 5 ml of the induction culture was taken from the flask and mixed with chloroform to a final concentration of 2% (vol/vol). The culture was agitated using a nutator for 15 min at room temperature and then centrifuged at 3000 × g for 10 min to obtain the phage lysate. The concentration of the phage lysate was determined via a standard phage titration assay.DNA FISHFor DNA FISH, probes for lambda DNA were produced by PCR amplifying ~3 kbp of the lambda genome (f-lambda-dnafish and r-lambda-dnafish primer pair), using a phage lysate as the template, and treating the purified PCR product with a PromoFluor500-dUTP nick translation kit (PromoCell) to generate DNA-PromoFluor500 fragments ranging from 100 to 500 bp. To generate probes for the E. coli attB region, an ~3 kbp region of the E. coli genome, including the attB region, was amplified with PCR (f-attb-dnafish and r-attb-dnafish primer pair) and treated with a PromoFluor640-dUTP nick translation kit (PromoCell). Equal amounts of these probes were mixed together to form a probe mixture.To perform DNA FISH on infection samples, cells (MG1655) were first grown from a colony overnight, ~16–18 h, in LBMM (LB + 0.2% maltose + 10 mM MgSO4). The overnight was then diluted 1:1000 into 50 ml of fresh LBMM and grown at 37 °C, 265 r.p.m., until OD600 ~0.3–0.4, about 3–3.5 h. The culture was then pelleted via tabletop centrifuge (2000 × g, 4 °C, 15 min), the supernatant was discarded, and the pellet was resuspended in LBM at 1/10th the original volume to concentrate the cells. Four milliliters of the cells were placed on ice, and ~40 μl of λLZ613 phage, at ~1 × 1011 pfu/ml, was added to the cells and gently mixed. 2 × 500 μl aliquots of cells were also separated as a control without phages. After leaving the infection mixture for 30 min on ice, the tube was moved to a 35 °C water bath for 5 min for phage DNA ejection. At this point, 500 μl of the infection mixture was aliquoted into a culture tube with 4.5 ml of LB + 0.2% glucose + 10 mM MgSO4 (LBGM) for each time point, all tubes were then moved to a 30 °C shaker at 265 r.p.m. At each given time, a tube was taken and fixed by pouring the mixture into a 15 ml centrifuge tube with 550 μl of 37% formaldehyde. This tube was left to shake on a nutator for 30 min, and then centrifuged at 4000 × g for 3 min to pellet the cells. The control sample was fixed after the 35 °C incubation.Details of fixation, permeabilization, and hybridization are detailed in other studies25. Briefly, the fixed cells were washed with 1 ml of ice-cold 1× PBS three times and resuspended in 1 ml of GTE solution (50 mM glucose, 20 mM Tris-HCl [pH 7.5], 10 mM EDTA). For the control sample, three separate 500 μl aliquots of the cell suspension were then mixed with 10 μl of 0.01 μg/μl lysozyme solution and incubated at room temperature for 2, 4, and 6 min followed by three washes with GTE, pelleting the cells via centrifugation at 10,000 × g for 30 s. The cells were then resuspended in ~150 μl of GTE. For each control sample, 1 μl of the cells was deposited onto a PBS agarose pad and imaged. The lysozyme treatment time yielding ~90–95% intact cells (~1–5% lysed cells) represents the optimal treatment time for the samples. The actual time point samples, from the initial GTE wash, were then processed as the control was, using the optimal lysozyme time. For each time point, 10 μl of cells were deposited onto poly-l-lysine-coated large coverslips (24 × 50 mm), then covered with a smaller, normal, coverslip (22 × 22 mm). The coverslips were then immersed in 1× PBS and the smaller coverslip was removed, leaving only the sample coverslip. The cells were then dehydrated by immersing the coverslip in increasing concentrations of ethanol (70, 90, then 100%). Samples were then ready for hybridization.For each sample, approximately 160 μg of the probe mixture was combined with 10 μl of hybridization solution (50% formamide, 10% dextran sulfate, 50 mM NaPO4/pH 7, 2× SSC). The dsDNA probes were denatured at 75 °C in a thermocycler, then placed on ice. Ten microliters of the denatured probe mixture were then deposited onto the center of the sample on the coverslip and overlaid with a small coverslip (22 × 22 mm). The small coverslip was then sealed with nail polish, forming a sample chamber. The chambers were incubated at 80 °C for 5 min to denature the cellular DNA, and then placed on Kimwipes over ice for 5 min. The chambers were then incubated in a 37 °C incubator overnight to complete hybridization.The next day, the chambers were immersed in 2× SSC until the smaller coverslip dislodged. The remaining coverslips were soaked in wash solution (2× SSC, 50% formamide) for 20 min at 37 °C twice. The coverslips were then washed with a series of increasing SSC concentration washes (1, 2, then 4×), each for 5 min at room temperature. A DAPI solution was then made by mixing 1 μl of 10 mg/ml DAPI to 1 ml of 4× SSC. For each sample, 500 μl of the DAPI solution was added over the sample, covering it, and incubated for 5 min at room temperature. After drying the coverslip, 10 μl of 2× SSC was added over the sample and overlaid with a small coverslip (22 × 22 mm). The samples were then imaged.RNA FISHDifferent probes were synthesized to target different phage transcripts (Biosearch Technologies). Probes targeting pR and pRE were designed following previous studies23,24, labeled with Cy5 and TAMRA, respectively. Probes targeting pR′ followed the same design principles as pR and pRE, and were labeled with AlexaFluor488 (pR′ probes listed in Supplementary Table 2).To perform RNA FISH, we follow the same infection protocols as described above for DNA FISH. At given time points, the cells were fixed in formaldehyde and pelleted. In one set of experiments, samples were taken between 6 and 40 min, after infection by phages at ~2 × 1010 pfu/ml. In another set of experiments, samples were taken at 15 min, after infection by phages at ~1, 2, 3, and 4 × 1011 pfu/ml. The processing of the samples is detailed in our previous study24. Briefly, after fixation, the cells were washed three times with 1× PBS. Subsequently, the cells were permeabilized by resuspension in 70% ethanol for 1 h at room temperature and centrifuged to collect the cells. The pellet was then resuspended in wash solution (40% formamide, 2× SSC) and incubated for 5 min at room temperature, and pelleted again, ready for hybridization.The cells were then resuspended in 25 μl hybridization solution (40% formamide, 2× SSC, 1 mg/ml E. coli tRNA, 2 mM ribonucleoside-vanadyl complex, and 0.2 mg/ml BSA) with each set of probes reaching a final concentration of 1 μM. The samples were then incubated in a 30 °C water bath overnight. The next day, the cells were washed three times using wash solution by incubating the cell pellet for 30 min in a 30 °C water bath. After the final wash, the cells were resuspended in wash solution + 10 μg/ml DAPI and incubated for 10 min at room temperature. This suspension was then pelleted and resuspended in 2× SSC. The sample was then ready for imaging.For the infections with rifampicin, rifampicin was added at a final concentration of 50 μg/ml to a 50 ml infection mixture in a flask at 15 min after the 35 °C step. Instead of pre-aliquoting separate infection tubes, 5 ml of the infection mixture was withdrawn at each given time point after addition of rifampicin for fixation and further processing as described.E. coli nucleoid imaging after rifampicin treatmentBacterial cells (MG1655) were inoculated from a colony into 1 ml of LB and grown at 37 °C, 265 r.p.m. for overnight. The overnight culture was then diluted 1:1000 into 5 ml of LB and grown under the same conditions until an OD600 of ~0.3. To study the effect of rifampicin on E. coli nucleoid morphology, rifampicin was added to the culture aliquots at final concentrations of 50, 100, and 300 μg/ml, and treated the cells for 15 min and 30 min at 37 °C, 265 r.p.m. One milliliter of the cell cultures was then fixed in 3.7% formaldehyde for 30 min at room temperature, followed by a washing step with 1 ml of PBS. The cell pellet was resuspended in 100 μl of PBS to reach an optimal cell density for microscopy imaging. For nucleoid imaging, 10 μl of cells were mixed with 10 μl of 20 μg/ml DAPI (to reach the final concentration of 10 μg/ml DAPI) for 10 min at room temperature. One microliter of the DAPI-stained cells were spotted onto a PBS agarose pad and imaged.SeqA cell lysogenization cultureAn overnight culture of cells bearing the SeqA reporter (LZ1557) were diluted 1:1000 and grown in 10 ml of LBMM + Kan50 + Amp100 + Cm10 to OD600 ~0.4. The culture was centrifuged, and the pellet was resuspended in 1 ml of LBM to concentrate the cells by 10-fold. Two hundred and fifty microliters of cells were mixed with phage (λLZ1576) to reach an API of ~4. The infection mixture was placed on ice for 30 min, and then moved to a 35 °C water bath for 5 min. The mixture was then diluted into 5 ml of fresh LBM and incubated in a 30 °C shaking water bath at 265 r.p.m. Samples were withdrawn at given time points for imaging.Microscopy imagingEach set of experiments (live-cell with λLZ1576/LZ1557, live-cell with λLZ1629/LZ1643, lysogen induction, DNA FISH, and RNA FISH) had its own set of imaging parameters according to the specific strains and fluorophores employed. All imaging was performed on a Nikon Eclipse Ti inverted epifluorescence microscope using a 100× objective (Plan Fluo, NA 1.40, oil immersion) with a 2.5× TV relay lens, using a mercury lamp as the light source (X-Cite 200DC, Excelitas Technologies), within a cage incubator (InVivo Scientific) at 30 °C, and acquired using a cooled EMCCD (electron multiplying charge-coupled device) camera (iXon3 897; Andor, Belfast, United Kingdom). The software images each stage through each filter sequentially for each time point before moving to the next stage. For the induction and fixed-cell experiments, stages with abundant cells were chosen for imaging. The stages were imaged under phase-contrast and specific filter cubes. The fluorescent filters used in the study were as follows (X, Y [excitation bandwidth] excitation filter/dichroic beamsplitter wavelength/X, Y [emission bandwidth] emission filter/company, product #): DAPI (350 nm, 50ex/400 nm/460 nm, 50em/Nikon, 96310), blue (436 nm, 20ex/455 nm/480 nm, 40em/Nikon, 96361), custom green (490 nm, 20ex/505 nm/525 nm, 30em/Chroma, custom 49308), yellow (500 nm, 20ex/515 nm/535 nm, 30em/Nikon, 96363), orange (539 nm, 21ex/556 nm/576 nm, 31em/Chroma, 49309), Cy3 (545 nm, 30ex/570 nm/610 nm, 75em/Nikon, 96323), red (560 nm, 40 nm/585 nm/630 nm, 75 nm/Nikon, 96365), far red (592 nm, 21ex/610 nm/630 nm, 30em/Chroma, 49310), and Cy5 (615 nm, 70ex/660 nm/700 nm, 75em/Nikon, 96366).For imaging each experiment, samples were exposed to the named filter cube in this order (with this exposure time, against this target).Infection movies using λLZ1576/LZ1557: phase-contrast (100 ms), blue (1 s, DnaB), orange (100 ms, single phage DNA), far red (200 ms, replicated phage DNA), and green (40 ms, capsid). This cycle of imaging occurred automatically once every 10 min for at least 3 h in each movie.Infection movies using λLZ1629/LZ1643: phase-contrast (100 ms), blue (100 ms, capsid), yellow (100 ms, attB), and red (200 ms, replicated phage DNA). This cycle of imaging occurred automatically once every 5 min for at least 2 h in each movie.Lysogen induction: phase-contrast (100 ms), blue (1 s, DnaB), far red (200 ms, replicated phage DNA), and green (40 ms, capsid). This was not a movie, so the imaging cycle was repeated for each stage/time point in the dataset.DNA FISH: phase-contrast (100 ms), yellow (200 ms, phage DNA), Cy5 (200 ms, attB), and DAPI (30 ms, DAPI). This imaging cycle was repeated for each stage in the dataset.RNA FISH: phase-contrast (100 ms), Cy5 (200 ms, pR), Cy3 (200 ms, pRE), yellow (200 ms, pRʹ), and DAPI (30 ms, DAPI). This imaging cycle was repeated for each stage in the dataset.E. coli nucleoid imaging after rifampicin treatment: phase-contrast (100 ms) and DAPI (100 ms).Data analysisMicroscopy images were analyzed using the cell recognition program Schnitzcells (gift of Michael Elowitz, California Institute of Technology), the spot recognition program MicrobeTracker (gift of Christine Jacob-Wagner, Yale University), and homemade scripts in Matlab (Supplementary Discussion).Strain constructionTo construct the phages with tetO arrays, phages bearing fluorescent reporters (λLZ1269, λLZ1369, λLZ1527, details for constructing these types phages were reported in previous studies) served as the parents13. Then a tetO-recombination plasmid was constructed to replace the bor::KanR region of the fluorescent phages with a bor::CmR 24×tetO array construct, using homology regions (upstreambor and downstreambor) The tetO array was derived from another study40, and inserted adjacent to a CmR cassette. Phages were titered onto host cells bearing the tetO-recombination plasmid (pBR322 24×tetO bor::CmR) and a pLate*D plasmid, then lysogenized the resulting plate lysate into MG1655, selecting for Cm resistance and Kan sensitivity in single-integration lysogens. The genomic construct was then verified by PCR.To construct the triple reporter strain, bearing SeqA (single phage DNA), TetR (replicated phage DNA), and DnaB reporters, strain LZ1383 (MG1655 seqA-mKO2 CmR-FRT) served as the parent. Plasmid PCP20 was transformed into this strain to recombine out the CmR cassette flanked by FRT sites41, imparting Cm-sensitivity to the cell (LZ1535). Then, MG1655 dnaB-mTurquoise2-CmR-FRT (LZ1510) was generated by first constructing the plasmid, pdnaB-mTurquoise2-CmR-H (H is the downstream homologous region of dnaB). Next, PCR was performed to create a linear dsDNA of the dnaB-mTurquoise2-CmR-H region, which was used for red-recombination42, to generate LZ1510. Then P1 transduction43 was performed to move the dnaB-mTurquiose2 reporter from LZ1510 (donor) to LZ1535 (recipient), making a strain with both the SeqA and DnaB reporters (LZ1552). Next, the Δdam-KanR marker was transduced from LZ1386 to LZ1552 to complete the SeqA reporter (LZ1555). This new strain was transformed with pACYC177 pFtsKi tetR-mCherry to complete the triple reporter strain, LZ1557.To construct the attB reporter strain, MG1655 served as the parent, and a 96× lacO array was inserted via red-recombination. This was done by first inserting upstream (E. coli genome region amplified with the f-up-attb and f-up-attb primer pair) and downstream (E. coli genome region amplified with the f-down-attb and f-down-attb primer pair) homology regions flanking the 96× lacO KanR region of a plasmid40, and then digesting this plasmid (pattB 96× lacO KanR) within the homology regions to produce linear dsDNA for red-recombination. This strain was then transformed with pACYC177 pFtsKi tetR-mCherry lacI-eyfp.To construct the lysogen with DnaB, TetR, and gpD reporters, LZ1510 (dnaB-mTurquoise2 CmR-FRT) served as the parent. The CmR cassette was removed using PCP20 to generate LZ1511 (dnaB-mTurquoise2) and then transformed pACYC177 pFtsKi tetR-mCherry into the strain. This host was then lysogenized with λLZ1575 (cI857
D-mNeongreen bor::CmR 24×tetO), producing LZ1596.Phage purificationThe purification protocol was adapted from other sources12,44,45. Briefly, a single colony of desired lysogens was grown with appropriate antibiotics at 30 °C overnight. The overnight culture was then diluted into 500 ml and induced. The phages were then precipitated using 10% PEG8000 + 1 M NaCl. The resulting phage pellet was soaked in a total of 8 ml of cold SM buffer and incubated at 4 °C overnight, ~16 h. An organic extraction was performed by mixing the SM suspension gently with an equal volume of chloroform and centrifuging at 3000 × g for 15 min at 4 °C. The supernatant was removed to exclude the PEG pellet, and the extraction step was done two more times, to finally yield a clear supernatant containing the phage. A step gradient was made for each desired phage using 1.5 ml each of 1.3, 1.5, and 1.7 g/ml CsCl + SM buffer solutions, and the phage (~8 ml) was layered on top in a 13.2 ml ultraclear tube (Beckman Coulter), then ultracentrifuged in a Beckman SW41Ti rotor at 24,000 r.p.m. for 6–8 h at 4 °C. The phages migrated to a band and were then extracted from the side wall of the tube. This phage extraction was then loaded into a 5 ml ultraclear tube (Beckman Coulter) and then filled with a 1.5 g/ml CsCl + SM buffer and ultracentrifuged in a Beckman SW50 rotor at 35,000 r.p.m. for 24 h at 4 °C, and then was extracted in the same manner. This new phage extraction was then dialyzed 1:1000 against SM buffer over ~24 h. This phage solution was then extracted and stored away from light and at 4 °C to be used in the experiments.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this Article.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Movie 1
Supplementary Movie 2
Supplementary Movie 3
Supplementary Movie 4
Supplementary Movie 5
Supplementary Movie 6
Supplementary Movie 7
Supplementary Movie 8
Reporting Summary | nature communications | [
"Article"
] | [
"Bacteriophages",
"Cellular noise",
"Single-cell imaging"
] | IntroductionOrganization fundamental life complex organisms spatial development controlled for function1 cells organelles organized by membranes2. Bacterial cells utilize proteins localize processes intracellular membranes3 inhomogeneity of cytoplasmic DNAs proteins segregation Viruses simpler life organize development within cytoplasmic inclusions.Bacteriophages simplest biological systems models for advanced cellular processes High-resolution fluorescence microscopy mathematical modeling phage lambda lysis–lysogeny paradigm cellular decision-making virus particles separate votes co-infecting lambda phages cell fates12 distinct voices for separate phages lytic lysogenic reporters phages phage DNAs compete cooperate subcellular identical viral DNA molecules divergent trajectories single cellular environment stochasticity development differential subcellular behaviors14 targeted high-resolution studies elicit molecular mechanisms beyond stochasticity hypothesize subcellular organization allows phages develop single cell detect subcellular heterogeneity investigating spatial distribution of viral/host biomolecules cascading events during viral transcription DNA replication gene expression establish organized subcellular unit of phage developmentorganization permits viruses cell develop different pathways separate areas-cell fluorescent reporters phage model lambda development used live-cell time-lapse microscopy targeting initial infecting phage DNAs host replication resources replicated DNAs decisions (Fig. phage DNA modified cells dam− carry seqA-mKO2) translational system validated label single DNA molecules infecting ejection methylated phage DNA SeqA-mKO2 binds ejected phage DNA DNA copies not replicated DNA copies SeqA system target all replicated phage DNAs recombineered tetO sequences phage genome host cell TetR-mCherry plasmid all phage genomes bound labeled tetO sites TetR Tet-labeling scheme lacks single-DNA sensitivity used Tet SeqA systems target phage DNA depends host factors viral DNA replication translationally fused Escherichia coli helicase19 DnaB with mTurquoise2 native dnaB gene dnaB-mTurquoise2 E. coli chromosome (Fig. 1c DnaB essential for phage/E.coli DNA interacts with lambda P E. coli DnaC DnaB construct E. coli phage growth Fig. reported lambda lysis–lysogeny decision-making modified phages with D-mNeongreen translational fusion cI-mKO2 transcriptional fusion lysogenic pathway (Fig. developed data analysis framework spatial organization subcellular events infection Fig. 1) images represent single cells early expression Kil protein inhibits cell division. 1Phage DNAs organize developmental processes phage processes decision-making Lytic decisions D-mNeongreen translational fusion lysogenic decisions cI-mKO2 transcriptional fusion SeqA system detects phage DNA Methylated phages infect cells Phage DNA bound by SeqA-mKO2 proteins DNA methylation labeled DnaB essential DNA replication resource DnaB-mTurquoise2 fusion protein reports localization DnaB Tet system detects replicated phage DNAs DNA labeled by TetR-mCherry binding infected cell undergoes lytic development cells e g chosen from three experiments contrast adjustedDnaB replicated DNA contrast images Supplementary Fig. 2b–e scale bars 2 μm Kymograph cell e Explanations data analysis Supplementary Fig 1 Discussion Fluorescence normalized population maximum infected lytic cell two subcellular areas development Kymograph cell g Fluorescence normalized population maximum DnaB heat maps cells arranged by position DnaB Cell left location i–l Fluorescence normalized peak brightness i–l n = 91 cells i–l SeqA heat maps arranged compare TetR heat maps arranged position TetR DnaB heat maps compare DnaB TetR Source data file.Organization resources replication phages compartmentalization biomolecules reflects individualistic development lambda Uninfected cells (LZ1557) diffuse blue yellow red fluorescence DnaB SeqA TetR compartmentalize without phage infection (Supplementary Fig. 2a; Supplementary Movie 1) Functional DnaB required growing cells19 DnaB localization growth basal DnaB state active DnaB-mTurquoise2 localize under phage-free normal growth conditionscells with phages (λLZ1576) fluorescence patterns change from phage-free lytic cells extensive DNA replication for lytic propagation lysogenic development differs from lytic behaviors Fig. 3b c SeqA foci appear early during infection phage DNA entered cell phage DNA replication demands resources DnaB foci formed after SeqA colocalized with over time (Fig. 1e f i j 4a phage DNA alters behavior DnaB resources DnaB foci represent multiple complexes to phage DNAs for replication lytic cells blue foci remained over time 90 lambda recruits resources 3b).Resource recruitment precedes phage DNA replication Red fluorescence spread free TetR background signal not phage DNA Fig 2c e 40 min). TetR signal rearranged into clusters production additional phage DNAs (Fig. 1e–h 3) subcellular TetR signal represent phage DNA DnaB pre-localized near red clusters remained replicating phage DNAs arise at previous gathered maintain sequestration1k l Figs. 3d 4c lytic cells gpD (green) signal increased (Fig. 1e–h). signal diffused green foci phage capsids formed in red clusters Fig. 5a). locations phage DNA determine progeny phages phage DNA data single phage DNA organizes subcellular phage factory maintaining clones resources proximal designated compartment “phactory” single phage DNA compartment predicted multiple DNAs form separate phactories identified cells with single phage DNAs different locations (Fig. 1g h 4) each DNA collected stockpile DnaB red clusters grew each DnaB location green foci grew clusters nearby DNA clusters phage-related biomolecules remain separated viral microenvironments levels identities phage DNA RNA protein change suggest each phage DNA organize phactory within single cells conditions segregated phage DNAs develop multiple phactories form progress in single cells (Fig. 2b–d Figs. 5b 6) Phactories unequal DNA lytic reporter levels over time heterogeneous development within phactories (FigSupplementary Figs. 7 2Organization subcellular areas during phage induction Schematic spatial organization phage development after induction Lambda prophage integrated into bacterial chromosome copies locate areas cell during growth Induction lysogens forces phages develop areas prophage bears gpD-mNeongreen lytic reporter carries tetO array cell harbors TetR-mCherry plasmid DnaB-mTurquoise2 reporter Overlay images lysogens after induction 0 min cells not induced contrast adjusted scale bars 2 μm Intracellular areas phage DNA DnaB form Histograms number DnaB replicated DNA clusters shown each point Phage DNA replication varies intracellularly more one DNA cluster standard deviation size clusters represented in boxplots variability median indicated by dot center box interquartile range whiskers span range outliers outliers points limit resolution 250 nm Source data provided file nucleoids maintain separation developing results indicate phage DNAs phactories separated obvious barriers obscuring mechanisms heterogeneous developmentobserved growth phage DNA clusters decreased cell filled DNA limited Fig. 9a c hypothesized replicating phage DNAs separated from E. coli DNA phactories tested labeling E. coli attB with lacO array LacI-EYFP construct infected cells (LZ1643) with phages (λLZ1629) (Fig. 3a). Phage DNAs capsids λLZ1576 LZ1557 Figs. 5c 9b d 10). terminus lytic cells cell lysis tracked location attB attB locations biased near poles (Fig. terminus non-lytic development cell division attB localized between mid-cell quarter-cell differential development. 3h 11a subcategorized lytic cells interaction phage DNA with attB Fig. 3b-g 11b-d largest class 60% 108 out of 179 lytic attB pushed towards one side cell away from expanding phage DNA cluster (Fig. 3b c phage DNA cluster sizes increased attB moved closer to cell polesphage DNA clusters expand near attB move past bacterial DNA mix (Supplementary Fig. 11e f data agree observations polar movement attB in lytic phage-active mechanism spatial expansion phage DNA replication explains results corroborate hypothesis E. coli DNA phactory localization segregation two phactories merge around bacterial marker (Fig. bacterial DNA barrier viruses different identities. 3Bacterial DNA separates intracellular phactories Detection phage bacterial DNA locations Phage carries Tet reporter D-mTurquoise2 lytic reporter Cell carries lacO array at attB locus bears plasmid expressing TetR-mCherry LacI-EYFP label phage bacterial DNA f lytic cells bacterial DNA interactions Phage DNA push spread bacterial DNA as expands cells from four infection experiments scale bars 2 μm Kymographs cells Fluorescence normalized to population maximum Spatial distribution bacterial DNA in lytic non-lytic cells differs locations attB prior cell division lytic lysis represent location preference bacterial loci developmental paths shows locations653 data points lytic 382 non-lytic categories phage DNA pushes bacterial DNA sizes phage DNA lytic cell shown violin plots maximum mid-cell location attB cells i violin plots (j). Cell right shows locations i, j Cells oriented TetR clusters aligned one direction violin plots solid line median dashed lines interquartile range Source data provided Source Data file used alternate techniques phage infection bacteria/phages without genetically engineered reporters performed single-molecule fluorescence in situ hybridization (FISH) phage transcription examine phage DNA replication fixed cells FISH live-cell techniques labeled phage λLZ613 DNA infection lambda DNA-specific probes early phage DNA small foci/clusters cells (Fig. localization FISH foci resembles SeqA foci live cells later DNA signal increased larger clusters similar reporter system amount DNA per cell clusters varied (Fig. 4c d bacterial DNA treated cells DAPI FISH experiments DAPI stains DNA size E.DNA (4.6 Mbp outstrips lambda DNA (48.5 DAPI bacterial DNA early before phage DNA replication phage DNA FISH signal DAPI mutually exclusive spatial distribution DAPI signal E. leaves nucleoid-free zones. 4h phage preferred areas early 4g supporting phage DNA correlation Phage DNA DAPI signals overlap later after replication (Fig. 4j 12b–e 13a generated probes against E. coli attB localizes with DAPI avoids phage DNA locations 4g congruent with live-cell data results suggest heterogeneous separated phage DNA units exists reported by live-cell fixed-cell methods.Fig. 4Phage transcription DNA replication organized within nucleoid-free regions DNA FISH labels phage bacterial DNA images experiments targeting phage DNA attB 10 min 50 min post-infection cells stained DAPI cells from six infection experiments scale bars 2 μm.d Violin plots phage DNA signal clusters solid line median dashed lines interquartile range RNA FISH labels phage pR transcript 6 min 40 min post-infection Cells stained DAPI six experiments g–i Phage DNA DAPI E. attB heat maps peak brightness phage DNA 10 min post-infection Cell locations g–n Fluorescence normalized peak brightness g–i l m n = 907 cells g–i j k Phage DNA prefers nucleoid-free locations avoids attB Difference maps contrast locations phage DNA DAPI Negative values 0. Difference maps phage DNA attB n k l–n Phage mRNA prefers nucleoid-free locations RNA FISH DAPI (m heat maps peak brightness pR 15 min post-infection Difference map l m j–k n = 2035 l–n o–q Phage DNA mRNA colocalize E. coli DNA Histograms peak location pR DAPI RNA FISH phage DNA DAPI DNA FISH peak location SeqA signal Fig(lytic cells first three time Cell below (q locations o–q Source data file.Phages maintain localized phage DNAs next step investigate phage mRNAs gene expression key phage development decision-making studied phage transcription with FISH24 targeted pR early transcriptional unit decision-making phage DNA replication26 pR transcripts small clusters (Fig. 4e). pR localizes lower DAPI signal (Fig. 4l–o near poles mid-cells regions nucleoid-free27 phage mRNAs reside away from nucleoids pR retains subcellular localization DAPI signals increase pR locations phage DNA replication locations pR transcripts in FISH represent locations phage DNA intracellular phactories possess own gene expression profiles cells mRNA spatial organization promoted attachment mRNAs to phage DNAs during transcript whole-cell diffusion discouraged by short lifetimes mRNAs in E. coli30 localized ribosomes translating transcripts27phage mRNA localization remains after transcription rifampicin phage RNAs localized with DNAs phactory decisions locations phage DNA replication gene expression decision-making cell-fate expression pRʹ transcript encodes lysis morphogenesis proteins lytic development pRE/pRM transcripts encode CI lysogeny targeted pRʹ pRE transcripts with FISH subcellular decision-making (Fig. 5a phage decision-making separated subcellular locations Colored represent transcripts targeted FISH direction transcription direction locations FISH probes-making transcripts localize subcellular locations infected cells 15 min post-infection-stained four experiments scale bars 2 μm Average pRE FISH signal plotted against pR signal value <0.001) Phages decisions in separate subcellular locations pR transcripts separated by DAPI clusters occupy separate areas 2035 cells 645 have pRE foci 439 pRʹ foci 504/2035 cells have pRE without pRʹ foci 298 pRʹ without pRE focid–e cells lytic decisions different locations (75/2035 cells). f–g conflicting lytic/lysogenic decisions locations (48/2035 cells). from four infection experiments 141/2035 cells have pRE and pRʹ same cell Model lambda decision-making subcellular space Phage DNAs occupy subcellular areas separated by bacterial DNA undergo gene expression transcription translation unknown key proteins localize after detaching mRNA sequester DNA replication resources locations replication transpires transcription localized units separated bacterial DNA differ composition Expanding phage DNAs push bacterial DNA inside cell decisions enacted by segregated phage DNAs Source data provided file performed FISH experiments 15 min after infection decision-making pR levels vary locations cell cells pR expression localized variation pR mRNA levels lead localized decisions lysogenic decision transcription from pR CI26 cellular pRE levels negatively correlate with pR levels (Fig. 5c). brightest pRE signals offset from brightest pR signalsdata corroborate hypotheses pRE/pR transcription lambda genetic circuit towards lysogeny pRE suggest lysogenic decisions initiated by DNAs pRE activation halts pR transcription single phage genome actions precede cell-wide repression CI FISH data show brightest pRʹ signals coincide within cell data suggest decisions enacted by separate phage DNAs subcellular areas Fig. 14f model lambda decision-making involves voting co-infecting phages lambda DNAs heterogeneous development in single voting behaviors occur separate subcellular areas FISH experiments pRʹ transcripts exist in intracellular pR clusters (Fig. 5d pRE pRʹ transcripts coexist in different locations cells phactory vote differently 5f organization behaviors persists after transcription initiation blocked Fig. 15f–k data indicate multiple intracellular lambda DNAs execute divergent developmental pathways spatial organization supports individuality (Fig. lambda development spatial organization viruses develop separately cell single phage DNA collects DnaB DNA replication clones developmentphage DNAs organize transcripts early different gene expression patterns enacted by separate viral DNA cellular spaces resulting in individual subcellular compartments phage development phactories individual phactory formed due to phage’s actions DnaB recruitment mRNA organization transcripts physical separation viral biomolecules by bacterial chromosome unclear phage proteins separated during infection cycle divergent development in single cells (Fig. simpler viruses lambda evolved organization complex behaviors compared to complex phages Pseudomonas phage 201phi2-1 lambda simplistic34 201phi2-1 houses cytoskeletal genes protein-enclosed compartment DNA replication transcription processes from translation comparison to Simpler phages phage phi29 host-specific proteins processes phi29 DNA replication relies on interactions with host’s MreB cytoskeleton DNA DNA replication depends on organization replication resource phage DNA polymerase terminal Phage lambda similar strategy replication utilizes own proteins reorganize DnaB depends on host’s architecture bacterial nucleoidsimplicity lambda compared to high-copy plasmids lacking partitioning non-nucleoid locations differences between lambda plasmid replication lambda retains DNA into single phactory pushes nucleoid detrimental for plasmids cluster lost during cell division bacterial DNA localization viability lambda commandeers DnaB diverges from plasmid behavior cell high-resolution methods to phage lambda reveal mechanisms applicable in other phage systems Further investigations into interplay viral bacterial biomolecules high-resolution techniques promise insights viral development.MethodsStrains plasmids primersBacterial strains phages in Supplementary Table 1.Primers homology regions Table 3.Single-cell infection imaging assayHost cells LZ1557 grown from −80 °C frozen stock in 1 ml M9 + 0.4% maltose antibiotics 100 μg/ml ampicillin 50 μg/ml kanamycin 10 μg/ml chloramphenicol 37 °C 265 r.p.m. overnight ~24 h culture diluted 1:1000 into 5 ml M9M + antibiotics grown same conditions ~16–18 h to OD600 ~0.3–0.4.1 ml culture pelleted tabletop centrifuge 4 min room temperature 20 μl purified phage (λLZ1576) ~3–4 × 1010 pfu/ml pipetted microcentrifuge temperature cells supernatant pipetted pellet resuspended 200 μl room temperature M9M Twenty microliters suspension mixed with 20 μl phage average phage input ~4 80 μl room temperature M9M added mixture moved pre-warmed 35 °C water bath 4 min phage adsorption DNA ejection small (1–1.5 cm2) section room temperature M9M agarose pad set No. 1 coverslip (18 × 18 1 μl deposited M9M pad larger No. 1 coverslip (24 × 50 mm) overlaid M9M pad sample moved microscope time-lapse imaging 30 °C bacterial strain LZ1643 phage λLZ1629 M9M supplemented Amp100 + Kan50 bacterial growth colony grown overnight ~16–18 h diluted 1:100 fresh media grown until OD600 ~0.3–0.4 3–4 h cells pelleted resuspendedexperiments phage cell suspensions mixed set on ice 30 min-adsorb 35 °C water bath 5 min steps imaging identical.Induction imaging lysogen colony LZ1596 grown overnight in 1 ml LB + 10 mM MgSO4) Amp100 + Cm10 ~16–18 h 30 °C shaker 225 r.m. culture diluted 1:100 into 5 ml fresh LBM antibiotics grown until OD600 ~0.3–0.4 1 μl culture deposited-buffered saline) agarose pad imaged lysogen culture induced flask 42 °C 225 r.p.m. shaking water bath 15 min 5-min time point after 5 min 42 °C sample processed imaged 15-min time point end 42 °C incubation culture moved to 37 °C 225 r.m. shaking water bath imaging 15-min 37 incubation remaining time points culture shaking 37 °C.Bacterial growth lysogen induction bacterial strains plated on standard LB agar plate antibiotics single colony 1 ml overnight culture in LB M9 + 0.4% maltose) 37 °C, 265 r.p.m. shakerovernight culture diluted 1:100 30 ml LB M9 + 0.4% maltose) antibiotics grown 37 °C 265 r.p.m. shaking water bath OD600 measured spectrophotometer lysogen induction colony LB plate grown overnight 1 ml LB + 10 mM MgSO4 antibiotics 30 °C 180 r. water bath culture diluted 1:100 into 25 ml fresh LBM antibiotics grown until OD600 ~0.3 0-min time point culture transferred 42 °C 180 r.p.m. shaking water bath 15 min thermal induction OD600 measured end incubation 15-min point moved 37 °C 180 r.p.m. water bath OD600 measured every 5 min until OD600 ~0.05 5 ml induction culture mixed chloroform 2% agitated 15 min centrifuged 3000 × g 10 min phage lysate concentration determined standard phage titration assayDNA probes lambda DNA PCR amplifying ~3 kbp lambda genome phage lysate PromoFluor500-dUTP kit DNA fragments 100 to 500 bp E. coli attB region ~3 kbp region amplified PCR treated PromoFluor640-dUTP kit probes mixed mixture DNA FISH samples cells (MG1655) grown from colony overnight LBMM 0.2% maltose 10 mM MgSO4) diluted 1:1000 into 50 ml fresh LBMM grown at 37 °C, 265 r.p.m. until OD600 ~0.3–0.4 3–3.5 h culture pelleted via tabletop centrifuge (2000 × g 4 °C 15 supernatant discarded pellet resuspended in LBM 1/10th volume concentrate cells Four milliliters cells on ice ~40 μl λLZ613 phage ~1 × 1011 pfu/ml added mixed 2 × 500 μl aliquots separated 30 min tube moved to 35 °C water bath 5 min for phage DNA ejection500 μl infection mixture aliquoted culture tube 4.5 ml LB 0.2% glucose 10 mM MgSO4 tubes moved 30 °C shaker 265 r.p.m tube fixed 15 ml centrifuge tube 550 μl 37% formaldehyde 30 min centrifuged 4000 × g 3 min pellet cells control sample fixed after 35 °C incubation fixation permeabilization hybridization fixed cells washed 1 ml-cold PBS resuspended 1 ml GTE solution (50 mM glucose 20 mM Tris-HCl [pH 7.5 10 mM three 500 μl aliquots mixed 10 μl 0.01 μg/μl lysozyme solution incubated room temperature 2 4 6 min three washes GTE 10,000 × g 30 s resuspended ~150 μl GTE 1 μl deposited PBS agarose pad imaged lysozyme treatment time ~90–95% intact cells~1–5% lysed cells optimal samples processed optimal lysozyme time10 μl cells deposited poly-l-lysine coverslips (24 × 50 covered (22 × 22 immersed 1× PBS removed cells dehydrated ethanol (70 90 Samples ready for hybridization 160 μg probe mixture combined 10 μl hybridization solution (50% formamide 10% dextran sulfate 50 mM NaPO4/pH 7 2× dsDNA probes denatured 75 °C thermocycler placed Ten microliters denatured probe mixture deposited center sample overlaid small coverslip (22 × 22 sealed nail polish sample chamber chambers incubated 80 °C 5 min Kimwipes 5 min incubated 37 °C incubator overnight hybridization next day chambers immersed 2× SSC until smaller coverslip dislodged remaining coverslips soaked wash solution (2× SSC 50% formamide 20 min 37 °C washed SSC washes (1 2 5 min DAPI solution 1 μl 10 mg/ml DAPI to 1 ml 4× SSC sample 500 μl DAPI solution added incubated 5 mindrying coverslip 10 μl 2× SSC added sample overlaid small coverslip (22 × 22 samples imaged.RNA FISHDifferent probes synthesized phage transcripts (Biosearch Technologies). Probes pR pRE designed studies23 labeled Cy5 TAMRA pR′ labeled AlexaFluor488 (pR′ Supplementary Table 2) RNA FISH infection protocols DNA FISH cells fixed formaldehyde pelleted samples taken 6 40 min infection ~2 × 1010 pfu/ml samples 15 min ~1 2 3 4 × 1011 pfu/ml processing detailed previous fixation cells washed three times 1× PBS permeabilized 70% ethanol 1 h centrifuged resuspended wash solution (40% formamide 2× SSC incubated 5 min pelleted ready hybridization cells resuspended 25 μl hybridization solution (40% formamide 2× SSC 1 mg/ml E. tRNA 2 mM ribonucleoside-vanadyl complex 0.2 mg/ml BSA final concentration 1 μM samples incubated 30 °C water bath overnight next day cells washed three times wash solution 30 minwash cells resuspended in wash solution + 10 μg/ml DAPI incubated 10 min room temperature pelleted resuspended 2× SSC sample ready for imaging infections rifampicin added 50 μg/ml 50 ml infection mixture 15 min after 35 °C 5 ml withdrawn after rifampicin fixation processing. nucleoid imaging after rifampicin treatmentBacterial cells (MG1655) inoculated 1 ml LB grown 37 °C 265 r.m. overnight culture diluted 1:1000 into 5 ml LB grown until OD600 ~0.3 rifampicin E. coli nucleoid morphology added culture aliquots 50 100 300 μg/ml treated cells 15 30 min 37 °C 265 r.p fixed 3.7% formaldehyde 30 min 1 ml PBS cell pellet resuspended in 100 μl PBS density microscopy imaging nucleoid imaging 10 μl cells mixed with 10 20 μg/ml DAPI 10 min-stained cells spotted PBS agarose pad imaged.SeqA cell lysogenization overnight culture cells (LZ1557) diluted 1:1000 grown in 10 ml LBMM + + to OD600 ~0.4.culture centrifuged pellet resuspended in 1 ml LBM cells 10-fold Two hundred fifty microliters cells mixed with phage (λLZ1576) API ~4. mixture on ice 30 min moved 35 °C water bath 5 min diluted into 5 ml fresh LBM incubated in 30 °C water bath 265 r.m. Samples withdrawn for imaging.Microscopy experiments (live-cell λLZ1576/LZ1557 λLZ1629/LZ1643 lysogen induction DNA FISH RNA FISH imaging parameters strains fluorophores imaging on Nikon Eclipse Ti inverted epifluorescence microscope 100× objective 2.5× TV relay lens mercury lamp cage incubator (InVivo) 30 °C cooled EMCCD camera software images each stage induction fixed-cell experiments stages with abundant cells chosen imaged under phase-contrast specific filter cubesfluorescent filters study DAPI (350 nm 50ex/400 nm/460 nm/Nikon blue (436 nm 20ex/455 nm/480 nm 40em custom green (490 nm 20ex/505 nm/525 nm 49308) yellow (500 nm 20ex/515 nm/535 nm 30em orange (539 nm 21ex/556 nm/576 nm 31em 49309) Cy3 (545 nm 30ex/570 nm/610 nm 75em/Nikon 96323) red (560 nm 40 nm/585 nm/630 nm 75 nm/Nikon far red (592 nm 21ex/610 nm/630 nm 30em/Chroma Cy5 (615 nm 70ex/660 nm/700 nm 75em/Nikon samples exposed filter cube movies λLZ1576/LZ1557 phase-contrast (100 blue orange far red (200 green (40 ms every 10 min 3 h λLZ1629/LZ1643 phase-contrast (100 blue yellow red (200 ms every 5 min 2 hLysogen induction phase-contrast (100 blue (1 s DnaB), far red (200 ms phage green (40 ms capsid). not movie imaging cycle repeated each stage.DNA FISH phase-contrast (100 yellow (200 ms phage Cy5 (200 ms DAPI (30 ms.RNA FISH phase-contrast (100 Cy5 (200 ms Cy3 (200 ms yellow (200 ms DAPI (30 ms.E. nucleoid imaging after rifampicin treatment phase-contrast (100 ms DAPI (100 ms).Data analysisMicroscopy images analyzed cell recognition Schnitzcells spot recognition MicrobeTracker homemade scripts Matlab phages tetO arrays fluorescent reporters (λLZ1269 λLZ1369 λLZ1527 tetO-recombination plasmid replace bor::KanR region bor::CmR 24×tetO array homology regions downstreambor tetO array derived inserted adjacent CmR cassettePhages titered host cells tetO-recombination plasmid pLate*D plasmid lysogenized lysate MG1655 Cm resistance Kan sensitivity-integration lysogens genomic verified PCR triple reporter strain SeqA TetR DnaB strain LZ1383 (MG1655 seqA-mKO2 CmR-FRT parent Plasmid PCP20 transformed strain CmR cassette FRT Cm-sensitivity cell (LZ1535) MG1655 dnaB-mTurquoise2-CmR-FRT (LZ1510) generated plasmid pdnaB-mTurquoise2-CmR-H PCR linear dsDNA dnaB-mTurquoise2-CmR-H red LZ1510 P1 dnaB-mTurquiose2 reporter LZ1510 to LZ1535 strain SeqA DnaB reporters (LZ1552) Δdam-KanR marker transduced LZ1386 to LZ1552 SeqA reporter strain transformed pACYC177 pFtsKi tetR-mCherry triple reporter strain LZ1557 reporter strain MG1655 parent 96× lacO array inserted red-recombination upstreamcoli f-up-attb primer pair 96× lacO KanR digesting plasmid linear dsDNA red-recombination strain transformed pACYC177 pFtsKi tetR-mCherry lacI-eyfp lysogen DnaB TetR LZ1510 (dnaB-mTurquoise2 CmR-FRT parent CmR cassette removed PCP20 LZ1511 (dnaB-mTurquoise2) transformed pACYC177 pFtsKi tetR-mCherry strain lysogenized λLZ1575 (cI857 D-mNeongreen bor::CmR 24×tetO), producing LZ1596.Phage adapted single colony lysogens grown antibiotics 30 °C overnight culture diluted 500 ml induced phages precipitated 10% PEG8000 + 1 M NaCl phage pellet soaked 8 ml cold SM buffer incubated 4 °C overnight ~16 h organic extraction mixing SM suspension chloroform centrifuging 3000 × g 15 min 4 °C supernatant removed exclude PEG pellet extraction two times clear supernatant phagestep gradient phage 1.5 ml 1.3 1.5 1.7 g/ml CsCl + SM buffer solutions phage~8 ml layered 13.2 ml ultraclear tube ultracentrifuged Beckman SW41Ti rotor 24,000 r.p.m. 6–8 h 4 °C phages migrated band extracted side wall tube phage extraction loaded 5 ml ultraclear tube filled 1.5 g/ml CsCl + SM buffer ultracentrifuged Beckman SW50 rotor 35,000 r.p. 24 h 4 °C extracted new phage extraction dialyzed 1:1000 against SM buffer ~24 h phage solution extracted stored 4 °C experiments Nature Research Reporting Summary.Supplementary information Peer Review File Additional Supplementary Files 1 2 7 8 | 48.8 | 0.917512 |
10.1038/s41467-020-14616-2 | PMC7005897 | Humans are normally not aware that their eyes are always in motion, even when attempting to maintain steady gaze on a point. Here the authors show that these small eye movements are finely controlled and contribute more than two lines in a standard eye-chart test of visual acuity. | High visual acuity is essential for many tasks, from recognizing distant friends to driving a car. While much is known about how the eye’s optics and anatomy contribute to spatial resolution, possible influences from eye movements are rarely considered. Yet humans incessantly move their eyes, and it has long been suggested that oculomotor activity enhances fine pattern vision. Here we examine the role of eye movements in the most common assessment of visual acuity, the Snellen eye chart. By precisely localizing gaze and actively controlling retinal stimulation, we show that fixational behavior improves acuity by more than 0.15 logMAR, at least 2 lines of the Snellen chart. This improvement is achieved by adapting both microsaccades and ocular drifts to precisely position the image on the retina and adjust its motion. These findings show that humans finely tune their fixational eye movements so that they greatly contribute to normal visual acuity. | IntroductionHumans critically rely on high visual acuity. Although fine spatial resolution is restricted to the foveola, a tiny region of the retina that covers less than 0.1% of the visual field, its loss has devastating consequences in everyday life, as experienced by subjects affected by deficits in foveal vision1.Individuals with normal or corrected-to-normal vision typically achieve a resolution of 0 logMAR—the 20/20 line of a Snellen chart—or better, which corresponds to the capability of resolving lines as thin as 1 min of arc, an astounding 1/60th of a degree (Fig. 1a). Since this level of resolution roughly matches the filtering of the optics2 and the spacing of receptors within the foveola3,4, it is often assumed that visual acuity is primarily determined by spatial factors, i.e., the spatial rules determining how an image can be discretized with minimal loss of information by a lattice of receptors5–7. However, the eyes are never stationary, and neurons in the visual system are strongly selective not just for spatial patterns, but also for temporally changing stimuli. Thus, unlike a stationary camera, it is doubtful that human acuity relies on purely spatial mechanisms.Fig. 1Fixational eye movements and Snellen acuity.a Several lines from a standard eye chart. The 20/20 line corresponds to a minimum angle of resolution (MAR) of 1 arcmin (logMAR = 0). Fixational eye movements (green arrows) cause the image to move on the retina (blue arrow). b Example of eye movements during examination of the 20/20 line. An oculomotor trace is shown superimposed onto the stimulus (top) and over time (bottom). Green and pink colors mark the periods of drifts and microsaccades, respectively. The black triangles mark the time at which the subject reported each optotype in the array. Source data are provided as a source data file.Humans always move their eyes during the acquisition of visual information, even when attempting to maintain steady gaze on a single point (Fig. 1b). Rapid gaze shifts known as saccades typically occur 2–3 times per second, bringing a new portion of the visual scene into the foveola. In between these movements, the so-called periods of “fixation”, the eyes wander incessantly, following seemingly random trajectories (ocular/eye drift) occasionally interrupted by miniature replicas of saccades (microsaccades)8–11. These fixational eye movements continually modulate the luminance flow impinging onto the retina and downstream neuronal activity12–17. Given the slow temporal integration of retinal neurons18,19, it has long been questioned how this motion does not impair spatial resolution, resulting in a percept similar to a blurred photograph acquired by a shaky camera20,21.Although less known, the opposite argument has also been made. It has long been argued that eye movements could be beneficial, rather than detrimental, to visual acuity22–25. With stimuli far from the limits of spatial resolution, both eye drifts26,27 and microsaccades28 have been found to facilitate pattern vision. However, experimental evidence on acuity has been contradictory. While no study has examined the consequences of eye movements in the standard Snellen test, conflicting results have been reported in other tasks. Pioneering experiments reported no benefits from eye movements on the minimum width of bars and vernier offsets that can be detected29,30. In contrast, retinal image motion seems advantageous when high-acuity stimuli are directly flickered on the retina, effectively bypassing the optics of the eye31. The reasons for these discrepancies, whether technical limitations in the earlier studies or the differences in stimulus delivery, are presently unclear.The recent findings with direct retinal stimulation31 are particularly interesting because the known characteristics of fixational eye movements do not appear suited to enhance acuity. Within the sensitivity range of parvocellular (P) ganglion cells18,19—the neurons primarily responsible for high-acuity vision32,33—the power delivered by drift modulations peaks at much lower spatial frequencies than those needed for visual acuity (e.g., Fig. 2d), effectively yielding noisy signals in this range. Furthermore, very fine control of microsaccades also seems necessary to properly position the stimulus on the foveola28. Precisely directed microsaccades down to amplitudes of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${20}^{\prime}$$\end{document}20′ have been previously observed28,34. However, the separation between adjacent optotypes in a 20/20 line of the Snellen eye chart is only \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10}^{\prime}$$\end{document}10′ (Fig. 1b), and it is unclear whether microsaccades this small can be precisely directed.Fig. 2Drift characteristics.Comparison of oculomotor data collected in the Snellen test and during sustained fixation. Data represent averages and SEM across N = 7 subjects. a, b Mean distributions of a drift speed and b curvature in the two conditions. Dashed lines indicate the means of the distributions (*p = 0.047, **p = 0.031, two-tailed Wilcoxon signed-rank test). c Average constants of drift diffusion, D, in the two conditions (*p = 0.035, two-tailed paired t-test). d Average power in the luminance modulations resulting from ocular drift. Power at different temporal frequencies was weighted by the temporal sensitivity of P cells and integrated to estimate the driving input for these neurons. Changes in eye drift shift modulations to higher spatial frequencies (horizontal arrow) amplifying power in the range of the optotypes (dashed line and vertical arrow). Dotted lines with triangles mark the peaks of the distributions (*p = 0.009, two-tailed paired t-test, **p = 0.022, two-tailed paired t-test). e Changes in the critical spatial frequency, kc, the frequency that delivers the largest luminance modulations (* as in d). Shaded regions and error bars represent SEM. Triangles in c and e represent data from individual subjects. Source data are provided as a source data file.Theoretical considerations26,35 indicate that changes in drift characteristics, specifically a slower and more curved drift, would have the desired consequence of shifting power to a higher range of spatial frequencies. Do humans adjust their eye drifts to reach their acuity limits? Do they precisely direct the smallest microsaccades? And if so, how much do eye movements contribute to standard assessments of visual acuity? Until recently, investigation of these questions was prevented by the technical challenges inherent in precisely measuring eye movements, estimating how they affect the spatiotemporal input, and controlling the luminance flow on the retina. These challenges can now be overcome by means of recently developed methods for gaze-contingent control36.Building upon these recent advances, here we investigate the functions of eye movements in the most common test of visual acuity, the Snellen eye chart. We show that humans actively tune both major components of fixational eye movements, ocular drift and microsaccades, to benefit from the spatial and temporal properties of retinal processing. Acuity is impaired when eye movements can no longer exert their normal consequences on the luminance flow entering the eyes. These results suggest that fine control of eye movements plays a critical role in achieving the limits of visual resolution: high acuity is not a purely visual accomplishment but the outcome of a visuomotor process that requires active control.ResultsTuning fixational eye movementsTo investigate the importance of fixational eye movements in visual acuity, we first examined whether human observers tune them to the task. To this end, we compared the characteristics of the eye movements recorded during inspection of the 20/20 line of a Snellen chart (Fig. 1, Supplementary Movie 1) with those recorded during the initial period of each trial before the appearance of the optotypes, when observers were simply asked to maintain steady gaze on a fixation marker (a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${2}^{\prime}$$\end{document}2′ dot). Sustained fixation is a standard condition for studying fixational control, and the eye movements it elicits have been extensively characterized8–11,37.Both ocular drift and microsaccades, the two main components of fixational eye movements, differed in important ways in the two periods. In contrast to the widespread assumption that eye drift is caused by limits in oculomotor control38,39, striking differences in the characteristics of this movement occurred. First, drift was slower in the Snellen test, with an average speed reduction of approximately 15% relative to fixation (Fig. 2a). Second, drift was also more curved in the high-acuity task than during fixation (Fig. 2b). These differences were not just present in the average data across subjects; they were also clearly visible in the distributions exhibited by every individual observer. For each subject, the distribution of drift speed was significantly narrower in the Snellen test compared to fixation, whereas the distribution of drift curvature was broader (p < 0.001, two-tailed two-sample Kolmogorov–Smirnov test).Although, taken individually, these differences in speed and curvature may appear small, their joint action on the motion of the stimulus on the retina are profound. These two effects compound in maintaining the line of sight closer to its post-saccadic location, and their consequences can be summarized by a single parameter, the diffusion constant, D, in a Brownian motion model of ocular drift. This model captures many characteristics of ocular drift35,40,41, and indeed in both the Snellen task and sustained fixation, the variance of drift displacement increased approximately linearly with time—a signature of Brownian motion (r2 = 0.98 and 0.84 for Snellen and fixation, respectively). However, because of both the reduction in speed and the increment in curvature, the estimated D measured in the Snellen test was much smaller than that obtained during the fixation period (Fig. 2c).In fact, the average diffusion constant observed across our subjects during execution of the Snellen test was also substantially smaller than the corresponding values measured in two separate groups of subjects that either maintained fixation for the entire duration of a trial, or freely observed natural scenes. In the sustained fixation group (N = 29), the mean diffusion constant ± SEM across observers was 17.5 ± 2.2 arcmin2 s−1, a 50% increase (p = 0.019, ANOVA with post-hoc Tukey–Kramer test). In the free-viewing group (N = 17), D = 26.2 ± 2.6 arcmin2 s−1, a two-fold increase (p < 0.001). Thus, drift displaced the image on the retina more slowly and by a smaller amount during execution of the Snellen test, so that retinal receptors received input from narrower regions of the visual field in this task.The observed changes in ocular drift have important repercussions on the visual flow impinging onto the retina. Previous studies have shown that ocular drift reformats spatial patterns into highly structured luminance modulations on the retina26,35. Specifically, at every non-zero temporal frequency the power of the resulting input signal depends non-monotonically on the spatial frequency (k) of the stimulus: it increases with k up to a critical frequency, kc, and then decreases for k > kc (Fig. 2d). Critically, kc depends on the amount of retinal image motion: it shifts toward higher spatial frequencies as the diffusion constant of motion decreases.Because of the observed differences in drift motion, the luminance modulations delivered by drift to retinal receptors also differed in the two periods (Fig. 2d, e). Within the temporal range of sensitivity of parvocellular (P) ganglion cells18,19, drift modulations in the Snellen task possessed considerably more power at high spatial frequencies. Whereas at fixation the average critical frequency kc lies around 15 ± 1 cycles per degree (cpd; gray curve in Fig. 2d), the smaller D measured in the Snellen test shifts kc up to approximately 23 ± 1 cpd (blue curve in Fig. 2d).This effect creates a high frequency range (above ~17 cpd) in which ocular drift delivers more power in the Snellen task. The optotypes in the 20/20 line are well within this amplification range, as they contain primarily energy at 30 cpd, so that the power of the input luminance flow on the retina increased by ~50% within the range of P cell sensitivity (vertical arrow in Fig. 2d). This signal was significantly stronger than the input that would have resulted from the drifts measured in other tasks. Relative to the separate group of subjects who maintained strict fixation for the entire duration of a trial, the input power in the Snellen test increased by 22% (p = 0.002, ANOVA with post-hoc Tukey–Kramer comparison). A 48% increment occurred relative to the eye drift recorded during free viewing of natural scenes (p < 10−7).Thus, by varying the amount of eye drift, the subjects in our experiments effectively amplified luminance modulations in the range of spatial frequencies relevant to the task. This signal enhancement is immediately obvious in reconstructions of the drift-induced temporal modulations impinging onto the retina (Supplementary Movie 2): a narrower and slower drift, like the one measured in the Snellen task, significantly sharpens the important edges of the optotypes in the 20/20 line of a Snellen chart.These results were highly robust relative to the specific methods of data collection and analysis. Similar results were obtained when the Snellen oculomotor data were compared to those measured in a separate control experiment in which subjects were specifically instructed to maintain very accurate fixation for the entire duration of a trial (Supplementary Fig. 1), rather than in the period preceding the Snellen task. Additionally, the diffusion constants measured in the Snellen test were smaller than those measured in the same observers when performing a non-acuity task with the same 20/20 line (judging a ±4∘ tilt in the overall line; Supplementary Fig. 2).Furthermore, conclusions were not influenced by the different durations of drift segments, which—because of the difference in microsaccade rates—were longer in the Snellen task. Indeed, very similar results were obtained by selectively focusing only on the first (or the last) 300 ms of each drift segment to make the durations of the periods of analysis identical in the two conditions. Similarly, differences in the amplitudes of the preceding saccades—on average smaller in the Snellen task (Fig. 3a)—were also inconsequential: results remained virtually identical when data analysis was restricted to drift segments preceded by microsaccades with comparable amplitudes in the two conditions. Thus, the changes in drift characteristics and their luminance modulations were robust effects, which did not depend on the specific design of our experiments.Fig. 3Microsaccade characteristics.a Average distribution of microsaccade amplitudes. Oculomotor data collected in the Snellen test (blue) are compared to those measured during sustained fixation (gray). Dashed lines indicate the means of the distributions. b 90th percentiles of the amplitude distributions in the two conditions (*p = 0.0009; two-tailed paired t-test). c Average distributions of microsaccade directions. d Distributions of gaze probability at different times during the course of the trial. The squares mark the optotypes positions. Note the left-to-right gaze shift mediated by microsaccades. e Average distance between microsaccade landing and the nearest optotype. Data from the Snellen test are compared to those obtained when microsaccades in Snellen were randomly replaced by those executed during fixation (Fixation) or when they were randomly permutated (Shuffled; *p = 0.00008, **p = 0.0009; two-tailed paired t-test). Shaded regions and error bars represent SEM. Triangles in b and e represent data from individual subjects. Source data are provided as a source data file.Microsaccades were also tuned to the task. In keeping with the lower microsaccade rates generally observed in high-acuity tasks34,42,43, microsaccades were less frequent during the Snellen test than during sustained fixation (mean rate ± SEM across observers in Snellen: 1.2 ± 0.1 microsaccades s−1 vs. Fixation: 2.5 ± 0.3 microsaccades s−1; p = 0.016, two-tailed Wilcoxon signed-rank test). Interestingly, the frequency of occurrence was not the only dimension in which microsaccades differed between the two tasks.Microsaccades in the Snellen test were also much smaller and more directionally selective than when the same subjects maintained fixation on a dot. Their amplitudes were approximately half the value measured during fixation (Fig. 3a), with the 90th percentile of the distribution decreasing from approximately \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${40}^{\prime}$$\end{document}40′ to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${20}^{\prime}$$\end{document}20′ between the two conditions (Fig. 3b). Notably, in the Snellen test, the distribution peaked at just 10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${}^{\prime}$$\end{document}′, an amplitude that matches the center-to-center spacing between neighboring optotypes in the 20/20 line. Furthermore, microsaccades in the Snellen test exhibited a strong bias for shifting gaze to the right, an effect reflected in their narrower angular variance relative to sustained fixation (Snellen: 0.51 ± 0.05 vs. Fixation: 0.64 ± 0.05; p = 0.047, two-tailed Wilcoxon signed-rank test; Fig. 3c). Similar effects were also observed when comparing the average microsaccade characteristics recorded in the Snellen test to those measured in the two separate groups of subjects that either maintained fixation for the entire duration of the trial or freely observed natural scenes. In both cases, microsaccades were larger than in the Snellen test (90th percentile amplitude in fixation: 33.6 ± 1.6 arcmin; p = 0.0013; free-viewing: 52.7 ± 1.4 arcmin; p < 10−9; ANOVA with post-hoc Tukey–Kramer tests), and they were also less likely to shift gaze to the right (percentages of rightwards microsaccades in fixation: 19.6%, p = 0.00003; free-viewing: 15.7%, p < 10−7).Together, these changes in amplitude and direction shifted the average position of gaze progressively rightwards during the course of a each trial, bringing the retinal projection of each optotype close to the preferred retinal locus of fixation (PRL), the narrow region on the retina at the very center of gaze (Fig. 3d, Supplementary Movie 3). Microsaccades were highly efficient in positioning stimuli on the retina. On average, each optotype fell within \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${7}^{\prime}$$\end{document}7′ from the center of gaze (Fig. 3e). Furthermore, when the microsaccades executed in the Snellen test were randomly replaced by those acquired during fixation, significant increments occurred in both (a) the average distance of each microsaccade landing to the closest optotype (Fig. 3e) and (b) the average distance of each optotype to the nearest microsaccade landing (a ~30% increase; p = 0.0043, two-tailed paired t-test). As shown in Fig. 3e, the distance by which microsaccades brought the line of sight close to an optotype also increased considerably when microsaccades were substituted by other randomly selected microsaccades occurring in the Snellen task.Microsaccades with amplitudes close to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1{0}^{\prime}$$\end{document}10′ played an important role in these effects. Both distances mentioned above remained virtually unchanged when only microsaccades in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${7.5}^{\prime}$$\end{document}7.5′–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${12.5}^{\prime}$$\end{document}12.5′ amplitude range were included in the analyses. When these microsaccades were randomly replaced with similar amplitude microsaccades executed during fixation, the distance from each microsaccade landing to the nearest optotype increased by 27% (p = 0.0006; two-tailed paired t-test). Also the distance between each optotype and nearest saccade landing increased significantly (by 12%; p = 0.02).Like for ocular drift, these effects were very robust. Significant differences in both the microsaccade amplitude and direction distributions measured in the two tasks were present in the data from every individual observer (amplitude: p < 10−9, two-tailed Mann–Whitney U-test; direction: p ≤ 0.001, two-tailed two-sample Kuiper’s test). Furthermore, the microsaccades measured in the Snellen task were significantly smaller than those measured in the same subjects when they were asked—in a separate control experiment—to maintain strict fixation for the entire duration of the trial, rather than in the period preceding the Snellen test (Supplementary Fig. 1). Microsaccades were also smaller than those measured in the same observers when performing a non-acuity task with the 20/20 line (judging a ±4∘ line tilt; Supplementary Fig. 2).These results reveal an oculomotor strategy for positioning the PRL close to each individual target. This strategy is remarkable both because of the small size of the eye movements involved and the high degree of control that it entails. Precisely directed microsaccades have been previously reported in the literature28,34. However, the targeted movements observed by these previous studies were considerably larger than those measured here, approximately the double in amplitude. Thus, even the smallest microsaccades appear to be tailored to the needs of the task.Oculomotor contributions to visual acuityHaving established that fixational eye movements, both microsaccades and ocular drift, are tuned to the Snellen test, we quantified their contributions to visual acuity. To this end, we counteracted their normal consequences on the visual flow by maintaining the stimulus immobile on the retina, a process known as retinal stabilization (Fig. 4a). This was achieved by means of a custom system for gaze-contingent control44, which enabled real-time updating of the stimulus on the display according to the observer’s eye movements. This system has been extensively tested and has been shown to yield high quality of retinal stabilization45.Fig. 4Oculomotor contributions to acuity.a Retinal stabilization. Stimuli moved on the display under real-time computer control (red arrow) to counteract the motion of the stimulus on the retina caused by eye movements (blue arrow). b Performance as a function of optotype size under retinal stabilization. The red line is the average psychometric function of N = 7 observers. Contrast was individually adjusted to yield ~5% correct identification during normal unstabilized viewing of 0 logMAR optotypes (black circle). Retinal stabilization greatly impaired performance (red arrow). Increasing the optotype size reestablished threshold level (blue arrow). The shaded region represents SEM (*p = 0.003, two-tailed paired t-test; **p = 0.002, two-tailed paired t-test). c Performance by optotype position in the 20/20 line. Filled circles represent averages across subjects for optotypes at different eccentricities on the display. Performance under retinal stabilization was significantly lower at all positions (p < 0.03; one-tailed paired t-test). Squares and triangles represent individual data under normal and stabilized viewing, respectively. Error bars represent SEM. d Acuity loss in the central 0–15′ and in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1{5}^{\prime}$$\end{document}15′–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3{0}^{\prime}$$\end{document}30′ range of foveal eccentricity. Black circles and error bar represent medians and SEM across subjects (*p = 0.0008, two-tailed paired t-test). Triangles are individual subject data. Source data are provided as a source data file.We compared performance in the discrimination of 0 logMAR optotypes (the 20/20 line of the eye chart) between two conditions: normal viewing and retinal stabilization. In the former condition, when eye movements normally moved the stimulus on the retina, the contrast of the stimulus was individually adjusted so that the mean percentage of correct identification was close to a threshold of 75% (dotted line in Fig. 4b). Performance was drastically impaired under retinal stabilization (red arrow in Fig. 4b). On average across subjects, the proportion of correct responses fell by approximately 25% when eye movements could no longer exert their visual consequences on the retina (p = 0.003; two-tailed paired t-test).Increasing the size of the optotypes improved performance under retinal stabilization and enabled reestablishing the same threshold level obtained in the normal, unstabilized condition (blue arrow in Fig. 4b). On average, this happened by enlarging each optotype by approximately 0.15 ± 0.02 logMAR, a loss in the minimum angle of resolution of about 40% (p = 0.002; two-tailed paired t-test). These effects were highly consistent across subjects and statistically significant in each individual observer (p < 0.01; one-sample two-tailed permutation test). They correspond to a visual loss of approximately two lines of the Snellen chart: from the 20/20 line in the presence of normal retinal image motion to the 20/30 line in its absence.The negative consequences of retinal stabilization could not be counteracted by increasing contrast. Unlike the immediate benefit resulting from enlarging the optotypes, performance under retinal stabilization improved little when contrast was increased while maintaining fixed-size optotypes. Performance remained far below threshold even at maximum contrast (p = 0.003; two-tailed paired t-test; Supplementary Fig. 3a). This marginal influence of the contrast of a stabilized stimulus is consistent with dynamic theories of spatial vision22–26,35,46 and deviates sharply from the strong facilitation that contrast exerts during normal, unstabilized viewing (Supplementary Fig. 3b).The adverse consequences of retinal stabilization on acuity were evident across the entire foveola, but varied with eccentricity (Fig. 4c). Around the very center of gaze (the two locations at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pm \!{5}^{\prime}$$\end{document}±5′ eccentricity), retinal stabilization resulted in an impairment in performance of approximately 16% (p < 0.03; one-tailed paired t-test). Given that these small eccentricities do not normally trigger microsaccades28,47, this loss was presumably caused by the absence of drift luminance modulations. At eccentricities larger than \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${5}^{\prime}$$\end{document}5′, retinal stabilization was even more disruptive (p < 0.02). This greater impairment was likely caused by the impossibility of using microsaccades to recenter the preferred retinal locus onto the more eccentric optotypes, as it occurred in the normal condition. In keeping with this idea, performance was also impaired when image motion was normally allowed on the retina, but subjects were asked to suppress their microsaccade scanning strategy by maintaining fixation at the center of the array (Supplementary Fig. 4). Asymmetries were also visible between the nasal and temporal retina, with the majority of subjects performing better at the largest nasal eccentricity, relative, for example, to the adjacent one (the optotype at ~\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${5}^{\prime}$$\end{document}5′ eccentricity), an effect statistically significant in two observers (p < 0.05; Z-test).Because of the dependence on eccentricity, the amount by which the stimulus had to be enlarged to recover threshold performance also varied across the foveola. The median loss in acuity increased from 0.14 ± 0.02 logMAR close to the preferred retinal locus (eccentricity\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$<1{5}^{\prime}$$\end{document}<15′) to 0.19 ± 0.01 logMAR in the larger range of tested eccentricities (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1{5}^{\prime}$$\end{document}15′–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3{0}^{\prime}$$\end{document}30′; Fig. 4d). These results show how rapidly acuity drops with eccentricity across the foveola and highlight the importance of both drift and microsaccades in ensuring high acuity during normal examination of fine details.DiscussionLittle attention is usually paid to eye movements when measuring visual acuity. Yet, the human eyes are always in motion, even when the retinal projection of the attended stimulus already falls within the high-acuity foveola, as during examination of the finest rows in an eye chart. In this study, we coupled high-resolution eye-tracking with real-time control of retinal stimulation to investigate the roles of eye movements in a standard acuity test. Our results show that humans finely tune their eye movements to enact an oculomotor strategy that takes advantage of both the spatial and temporal selectivities of retinal neurons. Visual acuity is impaired when this oculomotor strategy is prevented from exerting its normal consequences on the visual input.This study provides important contributions on two fronts: the control of eye movements, and their visual functions. In terms of oculomotor control, our results show that humans are capable of adapting their fixational eye movements to an unexpected degree, a behavior that paradoxically leads to a more stable retinal image during an acuity test than when observers are actually asked not to move at all, as when fixating on a point. The most suprising element of this oculomotor strategy is the tuning of ocular drift. The incessant jitter of the eye is widely believed to be an involuntary, random motion, presumably resulting from physiological limits in oculomotor precision38,39. Contrary to this idea, it has long been suggested that eye drift may actually represent a form of controlled motion48,49, a proposal consistent with findings of control at very low speeds in other types of eye movements, such as pursuit50 and the vestibulo-ocular reflex51. Our study shows, for the first time, that humans tune their eye drift in a way that is consistent with active theories of vision26,46. The changes in drift speed and curvature measured during examination of the finest lines of a Snellen chart are functionally important, as they increase power within the frequency range relevant to the task (Fig. 2d). This effect can be directly observed in reconstructions of the spatiotemporal flow impinging onto the retina (see Supplementary Movie 2).Microsaccades also exhibited a remarkable degree of control. They redirected the line of sight from one optotype to the next, even though the optotypes were only \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${10}^{\prime}$$\end{document}10′ apart. Control of microsaccades has been previously reported in the literature28,34,40. In particular, using methods for accurate gaze localization similar to the ones employed here, two previous studies have observed targeted microsaccades28,34. Critically, however, microsaccades in these previous studies were considerably larger, almost the double of those measured here. Microsaccades with amplitudes around \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1{0}^{\prime}$$\end{document}10′ were rare34 or virtually absent28, and determination of whether such small gaze shifts were also targeted was not possible.For example, in a simulated needle-threading task, microsaccades shifted the line of sight back and forth between the tip of the thread and the eye of the needle34; these movements became smaller as the thread approached the needle, but their average amplitude at the end of the task was still close to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${20}^{\prime}$$\end{document}20′, which matched the microsaccade amplitude measured from the same observers during sustained fixation. In the Snellen test, the peak amplitude of microsaccades was only 10′ and matched the spacing between adjacent optotypes. These movements were precisely directed, as revealed by their landing positions and by their random permutations or substitutions. Thus, during normal examination of stimuli at the limits of visual resolution, precise control extends to very small microsaccade amplitudes.In terms of visual functions, by showing that eye movements play a fundamental role in the outcome of a standard acuity test, this study provides support to the so-called dynamic theories of visual acuity22–26, the long-standing idea that oculomotor activity is instrumental for acuity. Reduced performance in the absence of retinal image motion has been recently reported with stimuli directly projected on the retina31. In this previous study, isolated optotypes at the limits of resolution were flickered at 30 Hz—a low frequency that affects contrast sensitivity52—while counteracting for the effects of the eye optics. However, earlier studies with more natural stimulation reached the opposite conclusion of no effects of eye movements on acuity29,30, and the reasons for this discrepancy have remained unclear.Our data show that, contrary to these older reports, the impairments observed with direct retinal stimulation extend to more natural conditions, with stimuli displayed at high refresh rates and normally viewed through the eye optics. Critically, our results go beyond the previous literature in several important ways, including (a) the finding of oculomotor tuning in both drifts and microsaccades; (b) the observations that these motor behaviors increase power in the range of neuronal sensitivity and enable precise re-centering of the stimulus on the retina; (c) the estimation of the acuity loss resulting from lack of visuomotor consequences, i.e., how much optoptypes need to be enlarged to maintain performance; and (d) the quantification of the consequences of eye movements in the Snellen eye chart, the most common test of visual acuity, where stimuli are not isolated and their layout plays an important role.Regarding the mechanisms by which eye movements enhance acuity, several possibilities remain open. One possibility is via spatial mechanisms similar to the super-resolution algorithms developed in computer vision53,54. These algorithms enable estimation of higher resolution images than those afforded by the sensor in the camera. A similar approach, in which the motion of the image enables overcoming sampling limitations imposed by the receptor array in the retina, was favored by Ratnam et al. (2017)31 as an explanation for their findings. An alternative explanation, in principle not mutually exclusive with the previous one, relies on the characteristics of the spatiotemporal flow impinging onto the retina. With larger stimuli—stimuli far from the limits of acuity—beneficial influences from fixational eye movements have been previously reported: both the temporal luminance modulations resulting from eye drifts26,35 and the positioning of the stimulus on the retina operated by microsaccades28 enhance foveal vision. Without the oculomotor adaptation observed in our experiments, these effects would not extend to visual acuity, as the characteristics of both microsaccades and drifts typically measured in non-acuity tasks are too coarse to enhance features at this scale: drift modulations would peak at too low spatial frequencies, and re-centering of each optotype on the preferred retinal locus would be difficult without control of the smallest microsaccades. However, the observed oculomotor tuning extends these benefits to to the limits of spatial resolution, suggesting that the strive for acuity is the primary factor driving these behaviors. While other factors unrelated to acuity could also contribute to the production of these visuomotor strategies, acuity is impaired in their absence, as during retinal stabilization or when examining a Snellen line in the absence of microsaccades.Our findings suggest a possible link between deficits in visual acuity and fixational eye movements. Oculomotor activity is rarely monitored during assessment of visual acuity, and poor outcomes in the Snellen test are commonly attributed to defects in the optical, structural, and/or physiological properties of the eye, not eye movements. Yet, abnormal fixational eye movements and impairments in fine spatial vision co-occur in multiple disorders. For example, poor fixational control accompanies the visual impairments present in conditions such as amblyopia55 and dyslexia56, and reduced visual acuity co-exists with the motor abnormalities present in conditions such as nystagmus57,58 and Parkinson’s disease59,60. That is, all these conditions exhibit both abnormal fixational eye movements and impaired acuity. Furthermore, theoretical considerations suggest that chronic exposure to the retinal input resulting from poor fixational control affects the maturation of the receptive fields of cortical neurons during development61–63. These considerations together with the findings of our study point at the need to examine in greater depth the consequences of abnormal eye movements for visual acuity.The acuity impairment measured in our experiments likely underestimates the real contribution of eye movements. One reason for this has to do with the way the Snellen chart itself is structured. While this eye chart represents the most widespread method for measuring visual acuity, it does not control for the possible effects of crowding, the negative consequences on visibility exerted by nearby stimuli64. In the Snellen chart, the number of optotypes does not remain constant across rows, but increases as the optotypes become smaller. This implies that the increased difficulty of the task with finer optotypes may stem not just from the required higher acuity, but also from more severe crowding. In our experiments, this effect may have partly compensated for the impairment caused by stabilization when the optotypes were progressively enlarged.Furthermore, one has to keep in mind that perfect retinal stabilization is not experimentally achievable, and theoretical considerations suggest that stabilization errors may also contribute to underestimating the real impact of eye movements. Our apparatus provides state-of-the-art quality of retinal stabilization, leaving a residual motion on the retina of approximately 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${}^{\prime}$$\end{document}′45. The resulting luminance modulations are not just smaller in amplitude, they also emphasize higher spatial frequencies. This happens because reducing the scale of retinal image motion is functionally equivalent to enlarging the stimulus, which translates into a compression of the spatial frequency axis in the frequency domain. As a consequence, while the strength of the input flow is reduced under retinal stabilization, it may still provide useful temporal power in the spatial frequency range of the smallest optotypes. Changes in the shape of the contrast sensitivity function under retinal stabilization are consistent with the idea that the visual system is sensitive to this residual motion65.It is worth pointing out that our results appear to have little to do with image fading, the gradual disappearance of stimuli observed under prolonged retinal stabilization. Image fading is typically observed with low-contrast and low spatial frequency stimuli displayed far from the fovea66. In agreement with previous studies31,45, none of our participants reported fading with the sharp, high-contrast foveal stimuli of our experiments. In fact, under retinal stabilization, recovery of threshold performance could only be achieved by increasing the size of the optotypes, not their contrast (Supplementary Fig. 3). This behavior deviates from the strong beneficial influence exerted by contrast in the presence of the physiological motion of the retinal image. Such changes support the notion that the visual system uses luminance modulations from eye movements to encode spatial information22–26,46.In sum, our results show that humans fine-tune their eye movements in tasks at the limits of spatial resolution. The resulting motion of the image on the retina plays a critical role in the outcome of the most common assessment of visual acuity. These results suggest that low performance in acuity measurements may result from suboptimal eye movements and stress the importance to carefully examine fixational eye movements in subjects with impaired acuity.MethodsSubjectsA total of 13 emmetropic subjects participated in the main experiments of this study (6 females and 7 males; average age: 23; age range: 20–35): seven subjects took part in the experiments of Figs. 2–4 and Supplementary Figs. 3 and 4; six other subjects participated in the control experiments of Supplementary Figs. 1 and 2. The oculomotor data collected in these experiments were compared to those collected from other 46 subjects (21 females and 25 males; average age: 22), who either maintained fixation or freely examined pictures of natural scenes. Subjects were naive about the purpose of the study and were compensated for their participation. To qualify, subjects had to possess at least 20/20 acuity in the right eye, which was assessed by correct identification of at least 75% of the optotypes in the 20/20 line during a standard execution of the Snellen test. Experiments followed the ethical procedures approved by the Charles River Campus Institutional Review Board at Boston University and the Research Subjects Review Board at the University of Rochester. Informed consent was obtained from all subjects.Stimuli and apparatusStimuli consisted of horizontal arrays of black tumbling-E optotypes displayed over a white uniform background (14 cd m−2). They were displayed at the center of a calibrated fast-phosphor CRT (Iiyama HM204DT; 1024 × 768 pixel resolution, 150 Hz refresh rate) placed in front of the observer in a dimly illuminated room. In every trial, a single array was presented with all the optotypes of equal size and contrast. Each optotype had equal probability to be oriented along four possible directions (legs up, down, left, or right). The size of the optotypes ranged from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${5}^{\prime}$$\end{document}5′ to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${16}^{\prime}$$\end{document}16′, with adjacent optotypes always separated by a space equal to the optotype width. Stimuli were viewed monocularly with the right eye, while the left eye was patched. A dental imprint bite-bar and head-rest minimized head movements and maintained the observer at a fixed distance from the display.Vertical and horizontal eye position data were measured by means of a Dual Purkinje Image (DPI) eye-tracker (Fourward Technology). Analog oculomotor signals were low-pass filtered at a cutoff frequency of 500 Hz and sampled at 1 kHz. Stimuli were rendered by means of EyeRIS, a hardware/software system for gaze-contingent display control that enables precise synchronization between eye movement data and the refresh of the image on the monitor44.Experimental proceduresData were collected in multiple experimental sessions, each lasting approximately 1 h. Every session started with preliminary steps aimed at ensuring optimal eye-tracking and gaze-contingent control. Blocks of trials then followed, each lasting 10–15 min. Breaks in between blocks allowed the subject to rest.To achieve localization of the line of sight, subjects underwent a two-stage calibration procedure. In the first phase, they sequentially looked at each of the 9 points of a standard 3 × 3 grid. This yielded a first estimate of the parameters of a bilinear transformation that mapped DPI voltages into visual angles. Parameters were then refined in a second gaze-contingent phase, in which subjects manually fine-tuned the estimated position of gaze, displayed in real-time on the monitor for each of the grid points. This approach enables accurate determination of the intersection between the line of sight and the display—i.e., the point in the stimulus that projects onto the center of the preferred retinal locus of fixation (PRL). This method has been shown to increase the accuracy of gaze localization by approximately one order of magnitude relative to standard eye-tracking calibrations36. Note that this procedure estimates the distance of a stimulus from the PRL in visual field coordinates; it does not allow determination of where the PRL is located on the observer’s retina. To counteract possible misalignment caused by drifts in the apparatus and/or minute head movements, the gaze-contingent procedure was repeated for the central fixation point before every trial.In a forced-choice procedure, subjects sequentially reported the orientations of all the optotypes in the array. They were instructed to proceed from left to right, as in a standard Snellen test, using four keys on a joypad. Each trial started with the subject maintaining strict fixation on a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${2}^{\prime}$$\end{document}2′ dot at the center of a uniform field for 1 s. The stimulus then appeared with the contrast of the array gradually increasing over the course of 1 s and then remaining at a fixed value until the subject had completed the task. To help maintain track of progress within a trial, a brief sound marked the registration of each response. The trial ended, and the stimulus disappeared, after reporting the orientation of the last optotype. At 0 logMAR (the 20/20 line), the average trial duration was 12 s.Blocks of trials alternated between two conditions. In the normal condition, stimuli remained at a fixed location of the display and moved normally on the retina because of eye movements. In the stabilized condition, the entire array of optotypes moved with the eye, under EyeRIS control, to counteract the consequences of eye movements and minimize retinal image motion. In the normal condition, subjects were always presented with the 20/20 line, a row of six optotypes, each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${5}^{\prime}$$\end{document}5′ in width (0 logMAR). The contrast of the optotypes varied adaptively across trials, following the Parametric Estimation by Sequential Testing (PEST) procedure67 to determine the contrast value yielding 75% correct discrimination for each individual observer.In the stabilized condition, we examined the effect of varying the optotype size. The contrast of the array remained fixed at the individual threshold value established in the normal condition, while the optotypes were systematically enlarged to determine the angle of resolution needed to reestablish threshold performance (method of constant stimuli: 11 optotypes within \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${5}^{\prime}$$\end{document}5′–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${16}^{\prime}$$\end{document}16′). As in a standard Snellen chart, the number of optotypes in the array decreased as they became larger, with the entire array spanning no more than 1∘.Four control experiments examined possible influences from various factors. Two experiments focused on performance in the Snellen test with 0 logMAR optotypes. The experiment of Supplementary Fig. 3 examined the consequence of varying contrast under retinal stabilization. Procedures were identical to those of the normal condition in Fig. 4, except that stimuli were now stabilized on the retina. The experiment of Supplementary Fig. 4 examined the effect of actively suppressing the microsaccade sequence. Subjects were asked to maintain fixation for the entire duration of the trial on a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${2}^{\prime}$$\end{document}2′ dot, which was presented at the center of the array. This request overruled the normal microsaccade behavior. Optotypes were displayed at the contrast threshold level determined in the normal condition.Two further experiments focused on the characteristics of eye movements. In the experiment of Supplementary Fig. 1, the oculomotor data collected in the Snellen test were compared to those acquired when observers maintained fixation for the entire duration of the trial (1.5 s), rather than in the initial period preceding the Snellen task. Subjects were instructed to fixate as accurately as possible on the fixation dot at the center of the display, which was presented at maximum contrast over a uniform gray background. In the control of Supplementary Fig. 2, subjects judged whether the entire 0 logMAR Snellen line was tilted by ±4∘ relative to the horizontal axis. The orientation of the line randomly alternated between trials. The optotypes were always presented at the contrast threshold level determined in the normal condition.Data collected in the Snellen task were also compared to those acquired in separate experiments in which subjects either maintained fixation or freely observed natural scenes. The procedures of these experiments were similar to those described in previous publications35,37. In the fixation condition, subject were asked to fixate as accurately as possible on a marker at the center of the display (a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${4}^{\prime}$$\end{document}4′ dot) for at least 1.5 s. The fixation marker was displayed at maximum contrast over a uniform background, and no other task preceded or followed fixation. In the free-viewing condition, subjects were instructed to memorize grayscale pictures of natural scenes, which were were displayed sequentially, each for 10 s. Each pixel subtended \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${1}^{\prime}$$\end{document}1′, an angle similar to that covered when the image was originally acquired.Data analysisAll effects are reported as group statistics. Individual statistics are reported for the most important results to point out that these are very robust effects, clearly visible in the data from each individual observer. Contrast thresholds yielding 75% correct identification were obtained by fitting a cumulative normal function to the data via a maximum likelihood procedure68 (Supplementary Fig. 3a). A similar approach was followed to estimate visual acuity thresholds using the cumulative Weibull function (Fig. 4b). This function, used by previous studies4,69, better fits acuity data than a cumulative normal. For each subject, we tested whether the change in the minimum angle of resolution was significant by measuring acuity thresholds via parametric bootstrap on the responses to individual optotypes (N = 1000).Oculomotor data from different periods in a trial were examined separately. The first 200 ms of each trial were discarded to ensure that subjects had acquired fixation. Fixation data in Figs. 2 and 3 refer to the remaining 800 ms period of each trial, when the fixation marker was displayed. Snellen data refer to the later period in which subjects reported the optotype orientations. This period started with the visual fixation preceding the first response and ended with the reporting of the last optotype.Oculomotor traces were segmented into complementary periods of saccades and drifts based on velocity. Events in which eye speed exceeded 3∘ s−1 were classified as saccades, with onset and offsets marked as the times at which the speed reached 2∘ s−1. Consecutive events separated by less than 15 ms were automatically merged to exclude post-saccadic overshoots. Remaining trace segments were classified as eye drifts. All events were classified automatically and verified visually. All trials with suboptimal eye-tracking or in which the subject looked away from the stimulus (trials with saccades larger than 2∘) were discarded from data analysis. Blinks were detected automatically by the DPI eye-tracker as the sudden loss and recovery of both Purkinje images. They were removed from analysis together with their surrounding segments.Saccade amplitudes and directions (Fig. 3a–c) were determined based on the difference between eye positions at saccade onset and offset. The distributions of saccade amplitudes and directions in the Snellen task and during fixation were compared by means of the Kolomogorov–Smirnov and Kuiper’s tests, respectively. The latter, being circularly invariant, is better suited for comparing angular directions. To take into account the width of the distribution, we compare microsaccade amplitudes by using their 90th percentile. The distribution of gaze position over time reported in Fig. 3d and Supplementary Movie 3 represent the average across subjects. Since no time limits were posed on the completion of the task, some trials took longer than others. To discount this variability, each trial was normalized by its duration and then subdivided into consecutive intervals (6 bins in Fig. 3d; 20 bins in Supplementary Movie 3). The time label on the y-axis of Fig. 3d indicates the average time of all the data points contained in the corresponding bin.In Fig. 3e, data points represent the distance from the landing position of a microsaccade to the nearest optotype, averaged across all microsaccades. These measurements are compared to those obtained in two conditions: when microsaccades in the Snellen task were substituted by microsaccades randomly selected from (a) the pool recorded during fixation and (b) the pool of microsaccades recorded in the Snellen task. In both cases, the new microsaccade was positioned so to possess the same starting position as the original one. Similar analyses were conducted to also estimate the distance between the center of each optotype and the nearest saccade landing position, averaged across optotypes. To evaluate the effectiveness of 10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${}^{\prime}$$\end{document}′ saccades, these analyses were repeated considering only saccades in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${7.5}^{\prime}$$\end{document}7.5′–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${12.5}^{\prime}$$\end{document}12.5′ amplitude range.To attenuate the impact of measurement noise during the low-velocity intersaccadic periods, drift segments were filtered by means of a low-pass third-order Savitzky-Golay filter with cutoff frequency at approximately ~30 Hz. Drift periods within 50 ms from saccades were discarded from data analysis to eliminate possible saccadic influences. Results in Fig. 2 are averages across all drift segments, independent of their durations. Virtually identical results were obtained on the initial, or the final, 300 ms of each drift segment or when only drifts following microsaccades smaller than \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${30}^{\prime}$$\end{document}30′ were considered.Spectral analysis of retinal inputWe estimated the power spectrum of the luminance modulations delivered by ocular drift on the retina. To this end, we used a Brownian motion model of ocular drift, a model that allows analytical formulation of the gain, Q, by which eye movements redistribute the power of the stimulus35:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Q({\bf{k}},\omega ;D)=\frac{2D{{\bf{k}}}^{2}}{{D}^{2}{{\bf{k}}}^{4}+{\omega }^{2}},$$\end{document}Q(k,ω;D)=2Dk2D2k4+ω2,where k = (kx, ky) represents spatial frequency, ω temporal frequency, and D the diffusion constant of motion.We first fitted the model for each observer, by estimating the equivalent diffusion constant of eye drift (Fig. 2c). This was accomplished by linear regression of the variance of the eye displacement as a function of time: σ2(t) ∝ 4Dt. We then measured the average power made available by eye motion at all spatial frequencies. We specifically examined the spatial frequency at which the distribution peaked (Fig. 2e) and how the change in diffusion constant in the Snellen task affected power at 30 cpd, the main frequency of a 0 logMAR optotype (Fig. 2d).To quantify the efficacy of the visual flow in driving neural responses, spectral distributions were weighted by the temporal frequency sensitivity of parvocellular ganglion cells. This was modeled by a series of filters, as previously proposed70:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H(\omega )=A\exp \left(-{\rm{i}}\omega d\right)\left(1-\frac{{H}_{{\rm{S}}}}{1+{\rm{i}}\omega {\tau }_{{\rm{S}}}}\right){\left(\frac{1}{1+{\rm{i}}\omega {\tau }_{\rm{L}}}\right)}^{{N}_{{\rm{L}}}}.$$\end{document}H(ω)=Aexp−iωd1−HS1+iωτS11+iωτLNL.Parameters were adjusted based on neurophysiological data19: A = 12.63, d = 0.0022, HS = 0.62, NL = 46.15, τS = 0.0259, τL = 0.0012. Since drift changes the spectral distribution of the input signal to the retina before any neural filter, our results are extremely robust with respect to the specific values of these parameters.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary
Description of Additional Supplementary Files
Supplementary Movie 1
Supplementary Movie 2
Supplementary Movie 3 | nature communications | [
"Article"
] | [
"Neural encoding",
"Oculomotor system",
"Sensorimotor processing",
"Visual system"
] | rely on high visual acuity fine spatial resolution restricted to foveola tiny region retina less than 0.1% visual field loss has consequences experienced by subjects deficits foveal normal or vision achieve resolution 0 20/20 line Snellen chart—or better resolving lines 1 min of arc 1/60th of degree (Fig. 1a). resolution matches filtering optics2 spacing receptors foveola3,4 assumed visual acuity determined by spatial factors eyes stationary neurons selective for spatial patterns temporally changing stimuli doubtful human acuity relies on purely spatial mechanisms.Fig. 1Fixational eye movements Snellen acuity standard eye chart 20/20 line minimum angle resolution 1 arcmin (logMAR = 0). eye movements cause image to move on retina Example of eye movements during 20/20 line oculomotor trace shown superimposed onto stimulus over time Green and pink colors mark periods of drifts microsaccades black triangles mark time subject reported each optotype Source data provided file.Humans move eyes during acquisition visual information steady gazeRapid gaze shifts saccades occur 2–3 times per second new visual scene into foveola between movements eyes wander random trajectories interrupted by miniature replicas saccades fixational eye movements modulate luminance flow retina downstream neuronal activity12–17 slow integration of retinal neurons18 questioned motion spatial resolution to blurred photograph opposite argument eye movements beneficial to visual acuity22–25 eye drifts26 microsaccades28 facilitate pattern vision experimental evidence on acuity contradictory no study examined consequences eye movements in standard Snellen test conflicting results reported in other tasks experiments reported no benefits from eye movements on minimum of bars vernier offsets retinal image motion advantageous when high-acuity stimuli directly flickered on retina bypassing optics reasons for discrepancies technical limitations differences in stimulus delivery unclear recent findings with direct retinal stimulation31 interesting fixational eye movements enhance acuitysensitivity range of parvocellular (P) ganglion cells18 responsible for high-acuity vision32 power by drift modulations peaks at lower frequencies than for visual acuity Fig. 2d), yielding noisy signals fine control of microsaccades necessary to position stimulus on foveola28 directed microsaccades amplitudes of previously observed28,34 separation between adjacent optotypes in 20/20 line of Snellen eye chart is only (Fig. 1b), unclear whether microsaccades this small can be precisely directed.Fig. 2Drift characteristics.Comparison of oculomotor data in Snellen test and sustained fixation Data represent averages and SEM across N = 7 subjects. Mean distributions of drift speed and curvature in two conditions.Dashed lines indicate distributions (*p = 0.047, **p = 0.031, two Wilcoxon test). Average constants drift diffusion two conditions (*p = 0.035 two-tailed t-test). Average power luminance modulations from ocular drift Power temporal frequencies weighted by P cells driving input Changes eye drift shift modulations to higher frequencies amplifying power optotypes Dotted lines triangles mark peaks distributions (*p = 0.009 **p = 0.022 Changes critical spatial frequency kc delivers largest luminance modulations Shaded regions error bars represent SEM. Triangles c e represent data subjects Source data.Theoretical indicate changes drift characteristics slower curved drift shifting power higher spatial frequencies humans adjust eye drifts acuity limits direct smallest microsaccades eye movements contribute to assessments visual acuity? investigation prevented technical challenges measuring eye movements estimating input controlling luminance flow retina challenges overcome gaze-contingent investigate functions eye movements in visual acuity Snellen eye charthumans tune fixational eye movements ocular drift microsaccades benefit from spatial temporal properties retinal processing Acuity impaired when eye movements exert normal consequences on luminance flow eyes results suggest fine control of eye movements critical in visual resolution high acuity not purely visual accomplishment outcome visuomotor process active control.ResultsTuning fixational eye movementsTo importance movements in visual acuity examined human observers tune them to task compared characteristics eye movements recorded during inspection 20/20 line Snellen chart (Fig. 1 with initial period each trial before optotypes steady gaze on fixation marker Sustained fixation standard condition for fixational control eye movements extensively characterized8–11 ocular drift microsaccades main components fixational eye movements differed in two periods contrast eye drift caused by oculomotor striking differences in characteristics occurred drift slower in Snellen test average speed reduction approximately 15% relative to fixation (Fig.drift more curved in high-acuity task than fixation (Fig. 2b). differences in average data visible in distributions observer drift speed narrower in Snellen test fixation drift curvature broader (p < 0.001 differences in speed curvature joint action on motion stimulus retina profound effects compound line of sight closer to post-saccadic location consequences summarized by diffusion constant D in Brownian motion model of ocular drift Snellen task sustained fixation variance of drift displacement increased linearly with (r2 = 0.98 and 0.84 for Snellen reduction in speed in curvature estimated D in Snellen test smaller than fixation (Fig. average diffusion constant Snellen test smaller than values in groups fixation or freely observed natural scenes sustained fixation group (N = 29), mean diffusion constant ± 17.5 ± 2.2 arcmin2 s−1 50% increase (p = 0.019 free-viewing group (N = 17), D = 26.2 ± 2.6 arcmin2 s−1 two-fold increase (p < 0.001)drift displaced image retina slowly smaller during Snellen test retinal receptors received input from narrower regions visual field changes in ocular drift visual flow retina reformats spatial patterns into structured luminance modulations non-zero temporal frequency power input signal depends on spatial frequency (k) stimulus increases with up to critical frequency decreases for k > kc (Fig. kc depends on retinal image motion shifts toward higher frequencies as diffusion constant motion decreases differences drift motion luminance modulations retinal receptors differed (Fig. drift modulations in Snellen more power at high spatial frequencies at fixation average critical frequency kc 15 ± 1 cycles per degree smaller D Snellen shifts kc to 23 ± 1 cpd creates high frequency range (above ~17 cpd) ocular drift delivers more power Snellen optotypes in 20/20 line within amplification range contain energy at 30 cpd power input luminance flow retina increased ~50% cell sensitivity signal stronger than input from drifts other tasksseparate group subjects strict fixation entire duration trial input power Snellen test increased 22% (p = 0.002 ANOVA post-hoc Tukey–Kramer comparison). 48% eye drift during free viewing natural scenes (p < 10−7) varying eye drift subjects amplified luminance modulations spatial frequencies relevant task signal enhancement obvious in reconstructions drift-induced temporal modulations retina narrower slower drift Snellen task sharpens edges optotypes in 20/20 line Snellen chart results robust specific methods data collection analysis Similar results Snellen oculomotor data compared to separate control experiment maintain accurate fixation entire duration trial diffusion constants measured in Snellen test smaller than same observers non-acuity task same 20/20 line ±4∘ tilt line conclusions not influenced by different durations drift segments difference microsaccade longer in Snellen task similar results by focusing on first 300 ms each drift segment durations identical differences in amplitudes of preceding saccades—on smaller in Snellen task (Figinconsequential results identical restricted to drift segments microsaccades comparable amplitudes conditions changes in drift characteristics luminance modulations robust effects depend on design.Fig. 3Microsaccade characteristics Average distribution microsaccade amplitudes data Snellen test compared to sustained fixation Dashed lines indicate distributions 90th percentiles amplitude distributions conditions (*p = 0.0009 Average distributions microsaccade directions Distributions gaze probability different times trial mark optotypes positions left-to-right gaze shift mediated by microsaccades Average distance between microsaccade landing nearest optotype Data Snellen test compared to microsaccades replaced fixation permutated Shaded regions error bars represent SEM Triangles b e represent data individual subjects Source data.Microsaccades tuned to task lower rates high-acuity less frequent during Snellen test sustained fixation (mean rate ± 1.2 ± 0.1 microsaccades s−1 vsFixation: 2.5 ± 0.3 microsaccades s−1; p = 0.016 two-tailed Wilcoxon signed-rank test). frequency occurrence not only microsaccades differed between tasks.Microsaccades in Snellen test smaller more directionally selective subjects fixation on dot amplitudes half value during fixation (Fig. 3a), 90th percentile distribution decreasing\documentclass[12pt]{minimal}{amsmath{wasysym\oddsidemargin-69pt}{document}{40}\prime\end{document}40′ to[12pt{minimal{amsmath\oddsidemargin-69pt}{document{20}\end{document}20′ between two conditions (Fig.Snellen test distribution peaked at 10[12pt{minimal{amsmath\oddsidemargin-69pt} amplitude center-center spacing between optotypes in 20/20 line microsaccades Snellen shifting gaze right narrower angular variance fixation (Snellen: 0.51 ± 0.05 vs Fixation: 0.64 ± 0.05; p = 0.047 Fig. 3c). Similar effects average microsaccade characteristics Snellen test to groups fixation natural scenes microsaccades larger than Snellen test (90th percentile amplitude fixation: 33.6 ± 1.6 arcmin; p = 0.0013 free-viewing 52.7 ± 1.4 arcmin p < 10−9 less likely to shift gaze right (percentages rightwards microsaccades in fixation: 19.6% p = 0.00003 free-viewing 15.7% p < 10−7)changes in amplitude direction shifted average position gaze rightwards trial retinal projection each optotype close to preferred retinal locus of fixation region at center gaze (Fig. 3d, Supplementary Movie 3) Microsaccades efficient in positioning stimuli retina average each optotype fell within-69pt from center gaze (Fig. 3e). microsaccades in Snellen test randomly replaced by acquired during fixation increments occurred in average distance each microsaccade to closest optotype average distance optotype to nearest microsaccade landing ~30% increase; p = 0.0043 Fig. 3e distance microsaccades brought line of sight close to optotype increased when substituted by other randomly selected microsaccades SnellenMicrosaccades amplitudes close to[12pt]{minimal{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin{-69pt}{document}$$1{0}{document}10′ effects distances unchanged microsaccades in[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin}{-69pt{document}$${7.5}^\prime{document}7.5′–[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}{document}$${12.5}\prime{document}12.5′ amplitude range included in analyses microsaccades replaced with similar amplitude microsaccades fixation distance each microsaccade to nearest optotype increased by 27% (p = 0.0006 t distance between each optotype nearest saccade landing increased 12%; p = 0.02).ocular drift effects robust differences in microsaccade amplitude direction distributions in two tasks present data observer (amplitude p < 10−9 Mann–Whitney U-test direction p ≤ 0.001 Kuiper’s test). microsaccades in Snellen task smaller than same subjects strict fixation trial Fig. 1) Microsaccades smaller non-acuity task with 20/20 line ±4∘ line tilt 2) results reveal oculomotor strategy for positioning PRL close to target strategy remarkable small size eye movements high degree control Precisely directed microsaccades previously reported targeted movements previous studies larger double in amplitude smallest microsaccades tailored to needs task.Oculomotor contributions to visual fixational eye movements microsaccades ocular drift tuned to Snellen test quantified contributions to visual acuity counteracted consequences visual flow by maintaining stimulus immobile on retina retinal stabilization (Fig. 4a). achieved custom system for gaze-contingent control44 real-time updating stimulus display according to eye movements system extensively tested high quality retinal stabilization454Oculomotor contributions to acuity Retinal stabilization Stimuli moved computer control Performance optotype size stabilization red line average psychometric function of N = 7 observers Contrast adjusted ~5% identification normal viewing 0 optotypes stabilization impaired performance optotype size reestablished threshold level shaded region represents SEM (*p = 0.003 = 0.002 Performance by optotype position 20/20 line Filled circles averages eccentricities Performance under stabilization lower at all positions (p < 0.03 Squares triangles represent data normal stabilized viewing Error bars represent SEMAcuity loss central 0–15′\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{upgreek\oddsidemargin{-69pt}{document}$$1{5}^\prime{document}15′–[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek-69pt}$$3{0}^{\prime{document}30′ range foveal eccentricity Black circles error bar medians SEM subjects (*p = 0.0008 two-tailed paired t-test). Triangles individual subject data Source data source data file compared performance discrimination 0 logMAR optotypes 20/20 line eye chart normal viewing retinal stabilization former condition contrast adjusted mean percentage correct identification close 75% (dotted line Fig. 4b). Performance impaired retinal stabilization (red arrow Fig.average correct responses fell 25% when eye movements visual retina (p = 0.003 two-tailed t-test).Increasing size optotypes improved performance stabilization threshold level normal condition (blue arrow Fig. 4b). enlarging each optotype 0.15 ± 0.02 logMAR loss minimum angle resolution 40% (p = 0.002 effects consistent across statistically significant each observer (p < 0.01 correspond visual loss two lines Snellen chart 20/20 line normal retinal motion to 20/30 negative consequences retinal stabilization counteracted increasing contrast performance improved little when contrast increased fixed-size optotypes Performance remained below threshold at maximum contrast (p = 0.003 t-test Fig. 3a). marginal influence contrast consistent with dynamic theories spatial deviates from facilitation normal unstabilized viewing Fig. adverse consequences retinal stabilization acuity evident foveola varied with eccentricity (Fig.4c). center of gaze two locations at\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek-69pt}±5′ eccentricity), retinal stabilization impairment performance 16% (p < 0.03; one-tailed paired t-test). small eccentricities trigger microsaccades28 loss caused by of drift luminance modulations At eccentricities larger than[12pt]{minimal}-69pt retinal stabilization more disruptive (p < 0.02) impairment likely caused by impossibility of using microsaccades to recenter preferred retinal locus onto eccentric optotypes performance impaired when image motion allowed on retina subjects suppress microsaccade scanning strategy by maintaining fixation at center array (Supplementary Fig. 4).Asymmetries visible between nasal temporal retina majority subjects performing better at largest nasal eccentricity relative adjacent optotype[12pt{minimal{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}\prime}5′ eccentricity), effect statistically significant in two observers (p < 0.05; Z on eccentricity stimulus enlarged recover threshold performance varied across foveolamedian loss in acuity increased from 0.14 ± 0.02 logMAR close to preferred retinal locus\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts to 0.19 ± 0.01 logMAR in larger range of tested eccentricities\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek-69pt}15′–[12pt]{minimal}{amsmath{wasysym{amsfonts{amsbsy{mathrsfs{upgreek-69pt{document}30′; Fig. 4d). results show rapidly acuity drops with eccentricity across foveola highlight importance of drift and microsaccades in ensuring high acuity during normal examination of fine details.DiscussionLittle attention paid to eye movements when measuring visual acuityhuman eyes in motion even when retinal projection within high-acuity foveola during examination finest rows eye chart study coupled high-resolution eye-tracking with real-time control of retinal stimulation roles eye movements in standard acuity test results show humans tune eye movements oculomotor strategy spatial temporal selectivities retinal neurons Visual acuity impaired when oculomotor strategy prevented visual study provides contributions on control of eye movements visual functions results show humans fixational eye movements leads to more stable retinal image during acuity test than suprising strategy tuning of ocular drift jitter eye believed involuntary random motion from physiological limits in oculomotor precision38 eye drift may represent controlled motion48 consistent with control at low speeds in eye movements study shows humans tune eye drift consistent with active theories of vision26 changes in drift speed curvature measured during examination finest lines Snellen chart functionally important increase power within frequency range relevant to task (Fig. 2d). effect observed in reconstructions of spatiotemporal flow retina (see Supplementary Movie 2)Microsaccades exhibited control redirected sight from optotype to optotypes\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}}10′ apart. Control of microsaccades previously reported in literature28,34,40 two previous studies observed targeted microsaccades28,34 microsaccades previous studies larger almost double of measured here Microsaccades with amplitudes around\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts{mathrsfs{upgreek-69pt}}10′ rare34 or virtually absent28 determination of small gaze shifts targeted not possible.in simulated needle-threading task microsaccades shifted line sight between tip thread eye movements smaller thread approached needle average amplitude end close to\documentclass[12pt{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}}20′ matched microsaccade amplitude observers during sustained fixation In Snellen test peak amplitude microsaccades 10′ matched spacing between adjacent optotypes movements precisely directed by landing positions random permutations substitutions normal examination stimuli at limits visual resolution precise control extends to small microsaccade amplitudes eye movements fundamental in outcome standard acuity test study dynamic theories of visual acuity22–26 oculomotor activity instrumental for acuity Reduced performance retinal image motion reported with stimuli projected on retina31 previous study isolated optotypes at limits resolution flickered at 30 contrast counteracting effects eye optics earlier studies opposite no effects eye movements on acuity29 reasons discrepancy uncleardata show contrary older reports impairments with direct retinal stimulation extend to natural conditions stimuli displayed at high refresh rates viewed through eye optics results go beyond previous literature oculomotor tuning in drifts microsaccades motor behaviors increase neuronal enable precise re-centering stimulus on retina estimation acuity loss from lack visuomotor consequences optoptypes quantification consequences eye movements in Snellen eye chart common test visual acuity stimuli isolated layout important mechanisms eye movements enhance acuity possibilities open spatial mechanisms similar super-resolution algorithms in computer algorithms enable higher resolution images similar approach motion image overcoming sampling limitations receptor array retina favored by Ratnam et al. explanation alternative explanation on characteristics spatiotemporal flow retina larger far from limits acuity—beneficial influences from fixational eye movements reported temporal luminance modulations from eye drifts26 positioning stimulus on retina by microsaccades28 enhance foveal visionoculomotor adaptation effects extend to visual acuity characteristics microsaccades drifts coarse drift modulations peak at low frequencies re-centering optotype difficult without control smallest microsaccades oculomotor tuning extends benefits to limits spatial resolution strive for acuity primary driving behaviors factors unrelated acuity contribute strategies acuity impaired in during retinal stabilization Snellen line findings suggest link between deficits visual acuity fixational eye movements Oculomotor activity rarely monitored during acuity poor outcomes Snellen test attributed to defects optical structural physiological properties not movements abnormal fixational eye movements impairments spatial vision co-occur in multiple disorders poor fixational control impairments amblyopia55 dyslexia56 reduced visual acuity-exists with motor abnormalities nystagmus57 Parkinson’s conditions exhibit abnormal fixational eye movements impaired acuity chronic exposure retinal input poor fixational control affects maturation receptive fields cortical neurons need to examine consequences of abnormal eye movements for visual acuity acuity impairment underestimates contribution eye movementsSnellen chart widespread method for measuring visual acuity control possible effects crowding negative consequences on visibility nearby stimuli64 In Snellen chart number of optotypes constant across increases as smaller implies increased difficulty with finer optotypes may stem from higher acuity severe crowding In experiments effect may have partly compensated for impairment stabilization when optotypes enlarged perfect retinal stabilization not experimentally achievable suggest stabilization errors may to underestimating real impact of eye movements Our apparatus provides state-of-the-art quality retinal stabilization residual motion on retina of approximately 1 resulting luminance modulations smaller in amplitude emphasize higher spatial frequencies reducing scale of retinal image motion equivalent to enlarging stimulus translates into compression of spatial frequency axis strength input flow reduced under retinal stabilization may provide useful temporal power in spatial frequency range of smallest optotypesChanges in contrast under retinal stabilization consistent with visual system sensitive to residual motion65 results little with image fading disappearance stimuli under retinal stabilization fading observed with low-contrast low spatial frequency stimuli far from fovea66 participants reported fading with sharp high-contrast foveal stimuli under retinal stabilization recovery of threshold performance by increasing size optotypes not contrast behavior deviates from influence contrast physiological motion retinal image changes support visual system uses luminance modulations to encode spatial results show humans fine-tune eye movements in at limits spatial resolution motion of image on retina critical in visual acuity results suggest low performance acuity may from suboptimal eye movements importance to examine fixational eye movements in impaired acuity 13 subjects participated in experiments (6 females 7 males average age 23 20–35) seven in experiments Figs. 2–4 3 4 six in control experiments 1 oculomotor data compared to 46 subjects (21 females 25 males average age maintained fixation or examined pictures natural Subjects compensated for participationqualify subjects 20/20 acuity right eye assessed identification 75% optotypes 20/20 line Snellen test Experiments followed ethical procedures Charles River Campus Review Board Boston University Research Subjects Review Board University Rochester Informed consent obtained subjects horizontal arrays black tumbling-E optotypes white background (14 cd m−2) center fast-phosphor CRT (Iiyama HM204DT 1024 × resolution 150 Hz refresh rate observer dimly illuminated room single array optotypes equal size contrastoptotype equal probability oriented along four directions (legs up down left right). size optotypes ranged from[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek to}16′ adjacent optotypes separated by space equal to optotype width Stimuli viewed monocularly with right eye left eye patched. dental imprint bite-bar head-rest minimized head movements maintained observer at fixed distance from display.Vertical horizontal eye position data measured Dual Purkinje Image (DPI) eye-tracker Analog oculomotor signals low-pass filtered at 500 Hz sampled at 1 kHz Stimuli rendered by EyeRIS, system for gaze-contingent display control synchronization between eye movement data refresh image.Experimental proceduresData collected in multiple experimental sessions each lasting approximately 1 h.session started preliminary steps optimal eye-tracking gaze-contingent control Blocks trials followed lasting 10–15 min Breaks allowed rest localization line of sight subjects underwent two-stage calibration procedure first phase looked at 9 points standard 3 × 3 grid yielded first estimate parameters bilinear transformation DPI voltages visual angles Parameters refined second gaze-contingent phase fine-tuned estimated position gaze displayed real-time approach enables accurate determination intersection between line of sight point stimulus center preferred retinal locus of fixation accuracy gaze localization one order magnitude standard eye-tracking procedure estimates distance stimulus from PRL visual field coordinates determination PRL retina misalignment drifts head movements gaze-contingent procedure repeated central fixation point before trial forced-choice procedure subjects reported orientations optotypes array instructed proceed left to right standard Snellen test using four keys joypadtrial started subject fixation on\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts{amssymb}{amsbsy{mathrsfs{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}{2\prime\end{document}2′ dot center uniform field for 1 s stimulus appeared contrast increasing 1 s fixed value until subject completed task brief sound marked registration response trial ended stimulus disappeared after last optotype 0 logMAR 20/20 average trial duration 12 s trials alternated two conditions normal condition stimuli fixed location moved normally retina eye movements stabilized condition optotypes moved with eye EyeRIS control counteract movements minimize retinal image motionnormal condition subjects presented 20/20 line six optotypes\documentclass[12pt{minimal}\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}{5}\prime{document}5′ (0 contrast optotypes varied adaptively across trials Parametric Estimation Sequential Testing (PEST) procedure67 contrast value 75% correct discrimination observer stabilized condition examined effect varying optotype sizecontrast array fixed at threshold value normal condition optotypes enlarged to determine angle resolution reestablish threshold performance (method constant stimuli: 11 optotypes\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek-69pt{document{5}5′–-69pt{16}}16′). standard Snellen chart number of optotypes in array decreased as became larger entire array spanning no more than 1∘.Four control experiments examined influences from factors Two on performance in Snellen test with 0 logMAR optotypes experiment Supplementary Fig. 3 examined varying contrast under retinal stabilization Procedures identical to normal condition Fig. 4 stimuli stabilized on retina experiment Supplementary Fig. 4 examined effect of suppressing microsaccade sequenceSubjects asked maintain fixation duration trial on\documentclass[12pt{minimal}{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document}{2\prime\end{document}2′ dot presented center array request overruled normal microsaccade behavior Optotypes displayed contrast threshold level normal condition.Two experiments focused eye movements experiment Supplementary Fig. 1 oculomotor data Snellen test compared maintained fixation entire duration trial (1.5 Subjects instructed fixate accurately on fixation dot center display presented maximum contrast uniform gray background control Supplementary Fig. 2 subjects judged 0 logMAR Snellen line tilted ±4∘ horizontal axis line randomly alternated trials optotypes presented at contrast threshold level normal condition.Data Snellen task compared separate experiments maintained fixation or observed natural scenes procedures similar previous publications35,37.fixation condition subject asked fixate accurately on marker center display\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{upgreek-69pt for 1.5 s fixation marker displayed at maximum contrast uniform background no other task preceded followed fixation free-viewing condition subjects instructed memorize grayscale pictures of natural scenes displayed sequentially each for 10 s Each pixel subtended[12pt{minimal-69pt angle similar to covered image originally acquired.Data analysisAll effects reported as group statistics Individual statistics reported for important results robust effects visible data individual observer Contrast thresholds 75% correct identification obtained by fitting cumulative normal function to data maximum likelihood procedure68 (Supplementary Fig. 3a). similar approach visual acuity thresholds using cumulative Weibull function (Fig. 4b). function better fits acuity data than cumulative normalsubject tested change minimum angle resolution measuring acuity thresholds parametric bootstrap responses optotypes (N = 1000).Oculomotor data periods examined separately first 200 ms discarded fixation Fixation data Figs. 2 3 remaining 800 ms period marker displayed Snellen data later period reported optotype orientations started fixation first ended last optotype.Oculomotor traces segmented into saccades drifts based velocity eye speed exceeded 3∘ s−1 classified as saccades onset offsets marked speed reached 2∘ s−1 Consecutive events separated less than 15 ms merged exclude post-saccadic overshoots Remaining segments classified as eye drifts events classified verified visually trials suboptimal eye-tracking subject looked away from stimulus saccades larger than 2∘ discarded Blinks detected sudden loss recovery images removed from analysis.Saccade amplitudes directions (Fig. 3a–c determined difference eye positions at saccade onset offset distributions Snellen task fixation compared Kolomogorov–Smirnov Kuiper’s tests comparing angular directions microsaccade amplitudes 90th percentile distribution gaze position over time Fig.3d Supplementary Movie 3 average subjects no time limits some trials longer variability each trial normalized by duration subdivided into intervals (6 bins Fig. 3d 20 bins Supplementary Movie 3) time label y-axis Fig. 3d indicates average time data points bin Fig. 3e data points distance from landing position microsaccade to nearest optotype averaged across microsaccades measurements compared to microsaccades Snellen task substituted by randomly selected fixation Snellen new microsaccade same starting position original Similar analyses estimate distance between center optotype nearest saccade landing position averaged across optotypesevaluate effectiveness of\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{mathrsfs{upgreek-69pt}{document} saccades analyses repeated considering saccades[12pt{amsmath{wasysym{amsfonts{mathrsfs{upgreek{-69pt}{document}{7.5}\prime{document[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}\oddsidemargin{-69pt}{12.5}{\prime{document}12.5′ amplitude range attenuate measurement noise low-velocity intersaccadic periods drift segments filtered low-pass third-order Savitzky-Golay filter cutoff frequency ~30 Hz Drift periods within 50 ms from saccades discarded eliminate saccadic influences Results in Fig. 2 averages across all drift segments independent durationsidentical results obtained initial final 300 ms each drift segment drifts following microsaccades smaller\documentclass[12pt]{minimal}{amsmath{wasysym\oddsidemargin-69pt}{document}30′ considered.Spectral analysis retinal estimated power spectrum luminance modulations ocular drift retina used Brownian motion model ocular drift formulation gain Q eye movements redistribute power stimulus35:1\documentclass[12pt]{minimal}{amsmath-69pt}{document}(\bf{k}},\omega ;D)={2D{{\bf{k^{2}}{{D^^{4}+\omega^{2}}{document}Q(k,ω;D)=2Dk2D2k4+ω2 k = (kx ky) spatial frequency ω temporal frequency D diffusion constant motion fitted model each observer estimating equivalent diffusion constant eye drift (Fig.accomplished by linear regression eye displacement function time: σ2(t) 4Dt measured average power eye motion at spatial frequencies examined spatial frequency distribution peaked (Fig. 2e) change diffusion constant Snellen task affected power at 30 cpd main frequency 0 logMAR optotype (Fig. efficacy visual flow neural responses spectral distributions weighted by temporal frequency parvocellular ganglion cells modeled by filters:2\documentclass[12pt{minimal\usepackage{amsmath\oddsidemargin-69pt}{document}\omega )=A\exp{document}H(ω)=Aexp−iωd1−HS1+iωτS11+iωτLNL.Parameters adjusted based neurophysiological data19 A = 12.63, d = 0.0022 HS = 0.62 NL = 46.15 τS = 0.0259 τL = 0.0012.drift changes spectral distribution input signal retina before neural filter results robust specific values parameters.Reporting research design Nature Research Reporting Summary.Supplementary information Peer Review File Reporting Summary Additional Supplementary Files Movie 1 2 3 | 46.6 | 1.105571 |
10.1038/s41467-020-17401-3 | PMC7374574 | Orexin signaling is provided by diffusely distributed fibers and involved in different brain circuits that orchestrate sleep and wakefulness states. Here, the authors show that a proportion of orexin neurons project to the sublaterodorsal tegmental nucleus and exhibit rapid eye movement (REM) sleep-related actions. | The relationship between orexin/hypocretin and rapid eye movement (REM) sleep remains elusive. Here, we find that a proportion of orexin neurons project to the sublaterodorsal tegmental nucleus (SLD) and exhibit REM sleep-related activation. In SLD, orexin directly excites orexin receptor-positive neurons (occupying ~3/4 of total-population) and increases gap junction conductance among neurons. Their interaction spreads the orexin-elicited partial-excitation to activate SLD network globally. Besides, the activated SLD network exhibits increased probability of synchronized firings. This synchronized excitation promotes the correspondence between SLD and its downstream target to enhance SLD output. Using optogenetics and fiber-photometry, we consequently find that orexin-enhanced SLD output prolongs REM sleep episodes through consolidating brain state activation/muscle tone inhibition. After chemogenetic silencing of SLD orexin signaling, a ~17% reduction of REM sleep amounts and disruptions of REM sleep muscle atonia are observed. These findings reveal a stabilization role of orexin in REM sleep. | IntroductionStable vigilance states, which fundamentally serve vital brain functions, depend on the regulation of diverse neurochemical signals1. Among them, the hypothalamic neuropeptide orexin/hypocretin is indispensably involved in this process2–5. Orexin signaling is provided by diffusely distributed fibers, and separately assigned to different brain circuitries that orchestrate sleep/wakefulness states6,7. Intriguingly, in addition to the prominent deficits in maintaining wakefulness state after orexin deficiency7, impaired quality of rapid eye movement (REM) sleep has been observed in increasing clinical observations8–10. The sublaterodorsal tegmental nucleus (SLD), which is regarded as both necessary and sufficient for the generation and maintenance of REM sleep11–13, has been reported to integrate a variety of neurochemical signals in sleep/wakefulness cycles14. Orexin fibers are also detected in the SLD15. Moreover, a previous report has found that microinjection of orexin into the SLD produces a REM sleep-like pattern of muscle tone decrease in decerebrated rats16. Therefore, the REM sleep relevance and the exact function of orexin signaling in the SLD, especially the underlying mechanisms need to be clarified.Orexin signaling is generally excitatory in nearly all target brain regions, including the SLD17. The SLD mainly contains glutamatergic neurons that constitute elementary units of its REM sleep-related output12,18. During REM sleep, the firing activities of the SLD glutamatergic neurons are carriers of output information for brain state activation, such as fast electroencephalogram (EEG) activities in the corticohippocampal region, and muscle atonia11,19,20. In general, the elevated firing activities in individual SLD glutamatergic neurons have been found to contribute to the SLD output19,21,22. Intriguingly, in addition to the readily observed membrane potential level of the SLD neurons17,23, several forms of sub-threshold fluctuations, such as gamma band and spikelet activities, are also involved in shaping their firings24,25. Furthermore, correct orchestration of the elevated firings are also thought to be required for the output of a given neuronal network, including the SLD26–28. It has been found that the pontine wave, which is thought to mirror a coordinated firing pattern of the SLD17,29, is highly correlated with the strength of the hippocampal theta activities during REM sleep27. While loss of orexin signaling has been reported to correlate with impaired brain states reflected by abnormal fast-EEG activities and muscle tone regulation in either wakefulness or REM sleep9,10,30, the specific influences of orexin on the neuronal membrane and network dynamics in the SLD, and their contributions to the functional brain state during REM sleep have not been investigated.Here, the REM sleep relevance of SLD orexin signaling is first uncovered through retrograde tracing and c-Fos screening of the SLD-projecting orexin (OXSLD) neurons. These orexin neurons are non-overlapping from those projecting to the wake-promoting locus coeruleus (LC). A global excitation of orexin in the SLD is next identified. Through in vitro/vivo electrophysiological assays, we identified an interaction between orexin’s effects on neuronal membrane dynamics and electrical connections, which supports global and synchronized excitation in the SLD network. The consequent enhancement of the SLD output and its contribution to consolidating brain state activation and muscle tone inhibition are revealed through optogenetics. Combining fiber photometry, we further find that all these substrates underlying SLD orexin signaling are employed during REM sleep to prolong individual episodes. A reduced REM sleep amount caused by disrupted REM sleep performances, including insufficient levels of muscle atonia and EEG theta oscillation power, is finally observed after chemogenetic inhibition of SLD orexin signaling. Thus, an essential role for SLD orexin signaling in REM sleep stabilization is presented.ResultsOXSLD neurons show REM sleep-related activationWe first identified the OXSLD neurons and examined whether their activities were related to REM sleep. Cholera toxin subunit B (CTB)-488 and CTB-555 were injected into the SLD and its neighboring LC, respectively, to label the origins of orexin signaling (Fig. 1a, b). CTB-488 and CTB-555 labeled orexin-A+ neurons accounted for 7.1 ± 0.8% and 15.2 ± 2.1%, respectively, of the total orexin-A+ neurons (n = 6 rats). Both of them were sporadically distributed across the lateral hypothalamus (LH) (Supplementary Fig. 1a–d), indicating that the injected region (Supplementary Fig. 2a) of the SLD was innervated by orexin neurons. Intriguingly, the OXSLD neurons were largely non-overlapping with the LC-projecting orexin (OXLC) neurons (overlap: 1.9 ± 0.3%; n = 6 rats) (Fig. 1d). The orexin receptors (OXRs) and bead-like orexin-A+ varicosities were also detected in the rat SLD (Fig. 1e), suggesting actions of orexin through volume transmission in this area. In the SLD, the OXRs were abundantly expressed in vesicular glutamate transporter-1 (VGLUT1)-positive glutamatergic neurons (Fig. 1f), and few OXRs/glutamate decarboxylase-67 (GAD-67)-positive GABAergic somas were detected (Fig. 1g). The abundance of glutamatergic, but not GABAergic, neurons in this SLD region was also observed by using glutamate and GABA antibodies (Supplementary Fig. 2b, c).Fig. 1LH OXSLD neurons show REM sleep-related activation.a, b Schematic drawing (a) and representative CTB-555 (red)/CTB-488 (green) injections (b) for retrograde tracing. LH lateral hypothalamus, scp superior cerebellar peduncle. Mo5 trigeminal motor nucleus, 4V 4th ventricle, DMTg dorsomedial tegmental area, PnC caudal pontine reticular nucleus. Scale bar: 500 μm. c Representative SLD-projecting (OXSLD, arrow heads, CTB-488+/orexin-A+, cyan) and LC-projecting (OXLC, arrows, CTB-555+/orexin-A+, purple) orexin-A+ (blue) neurons in the rat LH. Scale bar: 50 μm. Examples (square boxes) were magnified on the right. Scale bars: 10 μm. d Percentage of orexin (OX) neurons projecting to the SLD, LC, and both (OXSLD/LC), respectively (n = 6 rats; analysis of 3902 orexin neurons). e Expression of OXRs (green) and orexin-A+ varicosities (red) in the rat SLD. Scale bars: left, 50 μm; right, 10 μm. f, g Expression of OXRs (red) with VGLUT1 (green, f) and GAD 67 (green, g) in the rat SLD. All scale bars: 20 μm. h Representative sleep/wakefulness architecture following the 72-h RSD. The rat was sacrificed for c-Fos immunostaining (IHC) after 2.5 h from the first REM sleep episode (gray arrow). Control rats were set with normal sleep/wakefulness cycle and underwent same processes. i Percentage of time spent in REM sleep (R), wakefulness (W), and NREM sleep (NR) during the 2.5 h preceding sacrifice in the control (black, n = 4) and RSD (red, n = 7) rats. j Representative c-Fos (red) expression in orexin-A+ (blue) and OXSLD (cyan) neurons (RSD group). C-Fos-positive OXSLD neurons (arrows) from the medial (n1) and lateral (n2) part of the LH were presented below. All scale bars, 20 μm. k Percentage of c-Fos expression in the OXentirety (blue, two-sided unpaired t-test; t9 = 0.817, P = 0.435) and OXSLD neurons (cyan, two-sided unpaired t-test; t9 = 4.839, P = 9.221 × 10−4) in the control (n = 4, analysis of 2767 OXentirety and 178 OXSLD neurons) and RSD (n = 7, analysis of 6012 OXentirety and 370 OXSLD neurons) rats, respectively. Data represent mean ± SEM. **P < 0.01. Source data are provided as a Source Data file.We then assessed the REM sleep relevance of SLD orexin signaling. Rats with CTB-488 injection in the SLD were subjected to a 72-h REM sleep deprivation (RSD), and c-Fos screening was performed after the following REM sleep rebound (Fig. 1h). The REM sleep amount of these rats (n = 7) increased to 39.0 ± 2.8% in this period, compared to that of the control rats (9.1 ± 0.8%; n = 4) with normal sleep/wakefulness cycles (Fig. 1i). When analyzed as an entirety, the orexin neurons (OXentirety) exhibited no difference in c-Fos expression between the control and RSD groups (control: 7.1 ± 1.3%, n = 4 rats; RSD: 6.0 ± 0.6%; n = 7 rats; P = 0.435) (Fig. 1j, k). Nevertheless, c-Fos expression was dramatically increased within the OXSLD neurons (control: 6.8 ± 2.8%, n = 4 rats; RSD: 28.2 ± 2.9%, n = 7 rats; P = 9.221 × 10−4) (Fig. 1j, k). These data revealed that the OXSLD neurons exhibited increased activity in the presence of enriched REM sleep, implying the REM sleep relevance of SLD orexin signaling. In contrast, a proportion of the LH melanin-concentrating hormone (MCH) neurons also innervated the SLD, and they exhibited similar c-Fos expression level to that of the total MCH population after the REM sleep rebound (Supplementary Fig. 3a–d). Moreover, the abundant expression of OXRs in the SLD REM-on (c-Fos+) neurons was also observed after the REM sleep rebound (Supplementary Fig. 4a).Orexin globally excites the electrically coupled SLD networkWhole-cell patch-clamp recordings were employed to examine the electrophysiological effects of orexin in this glutamatergic neuron-enriched SLD region in rats (Supplementary Fig. 5a). The membrane potential of recorded neurons was adjusted to silence their spontaneous firing, and a cocktail of blockers (picrotoxin/MK-801/DNQX) was applied to block chemical synaptic transmission. In this condition, orexin-A (100 nM) strongly depolarized all tested SLD neurons (baseline: −63.3 ± 1.3 mV, orexin-A: −55.4 ± 1.4 mV; n = 32; P = 7.940 × 10−7) (Fig. 2a, c), revealing a global excitation effect. In addition, before orexin-A application, carbenoxolone (CBX)-sensitive spikelet activities were detected in 21 of 32 (65.6%) tested neurons (Fig. 2b, d), suggesting that the SLD glutamatergic neurons were electrically connected. This connection was further confirmed by the expression of the neuronal gap junction (GJ) protein Connexin-36 (Cx-36)31 in the SLD glutamatergic neurons (Supplementary Fig. 6a) and the spread of biocytin among them in patch-clamp recordings (Supplementary Fig. 6b–d). We also noticed that orexin-A (100 nM) elicited spikelet activities in six neurons without spikelet activities at baseline, increasing the percentage of SLD neurons exhibiting spikelet activities to 84.4% (27 of 32) (Fig. 2d).Fig. 2Orexin elicits depolarization and increases spikelets in the SLD neurons.a Orexin-A (100 nM) depolarizes a rat SLD neuron in DNQX/MK-801 (MK)/picrotoxin (PTX). Expanded traces showing that orexin-A also elicited consecutive spikelets. b The spikelets emerged after holding down the membrane potential in a rat SLD neuron and were blocked by CBX (100 μM). The spikelet had similar rise kinetics to action potential (box labeled). c Membrane potential of the SLD neurons before and after orexin-A applications (n = 32, two-sided Wilcoxon signed-rank test; z = 4.937, P = 7.940 × 10−7). d Orexin-A increased the percentage of the SLD neurons exhibiting spikelets (n = 32). e–g Representative traces (e) and statistics showing that 30–300 nM orexin-A concentration-dependently elicited inward current (f) and noise increase of the current (g) in rat SLD neurons (n = 8). h–j Representative traces (h) and statistics showing that the amplitude (i, two-sided paired t-test; t6 = 0.670, P = 0.528) and the noise increase (j, two-sided paired t-test; t6 = 0.192, P = 0.854) of the orexin-elicited inward current remained unaffected by DNQX/MK/PTX in rat SLD neurons (n = 7). k–m Representative traces (k) and statistics showing that the amplitude (l; one-way repeated-measures ANOVA, F(2, 10) = 22.137, P = 2.124 × 10−4; post hoc LSD comparison test; control vs. TCS OX2 29, P = 0.0122; control vs. TCS OX2 29/SB, P = 2.218 × 10−3; TCS OX2 29 vs. TCS OX2 29/SB, P = 0.0204) and the noise increase (m; one-way repeated-measures ANOVA, F(2, 10) = 8.721, P = 6.425 × 10−3; post hoc LSD comparison test; control vs. TCS OX2 29, P = 0.0422; control vs. TCS OX2 29/SB, P = 0.0249; TCS OX2 29 vs. TCS OX2 29/SB, P = 0.0387) of the orexin-elicited inward current were partially blocked by TCS OX2 29, and totally abolished by TCS OX2 29/SB 334867 (SB) in rat SLD neurons (n = 6). Data represent mean ± SEM. *P < 0.05; **P < 0.01. Source data are provided as a Source Data file.The current basis of the orexin-elicited global excitation was further analyzed. At −70 mV holding voltage, orexin-A (30–300 nM) elicited inward currents in all 58 tested SLD neurons in normal artificial cerebrospinal fluid (ACSF). The orexin-elicited inward current was dose-dependent (Fig. 2e–g), and was independent of chemical synaptic transmission (control: −25.3 ± 7.9 pA, blockers: −24.6 ± 7.6 pA; n = 7; P = 0.528) (Fig. 2h, i). Furthermore, the whole-cell current were usually noisy with rhythmic fluctuations (Fig. 2h), reflecting the GJ currents underlying spikelet activities32. With the onset of the orexin-elicited inward currents, the current noise apparently increased, and the increase persisted after blocking chemical synaptic transmission (control: 59.9 ± 26.2%, blockers: 57.4 ± 13.6%; n = 7; P = 0.854) (Fig. 2j). We further found that the orexin-elicited noisy inward currents were mediated by the activation of orexin-1 and orexin-2 receptors (Fig. 2k–m).The global excitation of orexin in the SLD depends on GJsSurprisingly, after blockage of GJ by 100 μM CBX, the orexin-elicited inward current was dramatically decreased by 42.9% (control: 26.6 ± 6.2 pA, CBX: 15.2 ± 4.3 pA; n = 15; P = 3.143 × 10−3) (Fig. 3a–c). Furthermore, in 4 of 15 (26.7%) neurons, the inward current was completely abolished by CBX (Fig. 3b, c). These data indicate an interaction between the orexin-elicited excitation and SLD GJ activities. We further employed another GJ blocker, mefloquine (MEF), for confirmation. In addition, biocytin-based post hoc immunostaining was employed to determine the expression of OXRs in each electrophysiologically tested neuron. Pretreatment (2 h) with MEF (10 µM) also reduced the amplitude of the orexin-elicited inward current by 36.2% (control: 24.3 ± 2.1 pA, n = 58; MEF: 15.5 ± 3.4 pA, n = 25; P = 1.018 × 10−3) (Fig. 3d, e). In addition, 6 of 25 (24%) neurons showed no response to orexin-A after MEF pretreatment (Fig. 3d, e). Post hoc immunostaining further revealed that the orexin-A non-responsive neurons did not express OXRs (Fig. 3d). After blocking GJ activities, ~1/4 of the SLD neurons were found to be orexin-A non-responsive, suggesting that SLD orexin signaling directly targets ~3/4 of the SLD neurons. We then determined the substrates of the orexin-elicited effects that may spread through GJs to generate the global excitation. After addition of TTX to block spike generation, the spikelet currents were all abolished (Fig. 3f, g). Nevertheless, orexin-A still elicited inward currents in all tested neurons and the amplitude remained unaffected (control: 20.7 ± 2.6 pA, blockers: 22.6 ± 2.4 pA; n = 8; P = 0.0589) (Fig. 3h). Together, these results indicate that the orexin-elicited sustained depolarization by this current on the SLD OXRs+ neurons spreads through GJs to form the orexin-elicited global excitatory effects. In addition, the depolarization may eventually evoke firings in the SLD neurons and thus contribute to the increase in spikelet activities. We also found that the orexin-elicited excitation originates from the activation of the non-selective cationic conductance (NSCC) (Supplementary Fig. 7a–g).Fig. 3The excitation of orexin in ~3/4 of SLD neurons spreads via GJs.a, b Representative traces showing that the orexin-A (100 nM) elicited inward currents were partially blocked by CBX (100 μM) in a rat SLD neuron (a) and completely abolished by CBX in another (b). Note that the spikelet currents meditated by GJs were all blocked by CBX. c The orexin-elicited inward current was reduced in the presence of CBX in rat SLD neurons (n = 15, two-sided Wilcoxon signed-rank test; z = 2.953, P = 3.143 × 10−3). Pie charts showing that the orexin-elicited inward current was completely abolished in 4 of 15 (26.7%) tested neurons. d Biocytin (green)-based post hoc immunostaining showed that an orexin-A non-responsive neuron from the rat SLD (left) in the presence of mefloquine (MEF) did not express OXRs (red), and another orexin-A-responsive neuron (right) from the rat SLD in the presence of MEF expressed OXRs. Scale bars, 20 μm. e The orexin-elicited inward current was also reduced in the presence of MEF, compared to the normal ACSF condition in rat SLD neurons (control: n = 58, MEF: n = 25; two-sided Mann–Whitney rank-sum test; z = 3.286, P = 1.018 × 10−3). Pie charts showing that the orexin-elicited inward current was completely abolished in 6 of 25 tested neurons. f Representative traces showing that orexin-A (100 nM) elicited inward current in a rat SLD neuron before and after the application of TTX (1 μM) plus DNQX/MK/PTX. Note that spikelet currents were observed after orexin-A application in normal ACSF. In the presence of TTX, rhythmic spikelet currents cannot be elicited by orexin-A. g, h The orexin-elicited noise increase of the inward current was attenuated in the presence of TTX (g, two-sided paired t-test; t = 3.430, P = 0.0110), but the amplitude of the orexin-elicited inward current remained unaffected in rat SLD neurons (h, two-sided paired t-test; t = 2.253, P = 0.0589) (n = 8). Data represent mean ± SEM. *P < 0.05; **P < 0.01. Source data are provided as a Source Data file.Orexin increases the GJ conductance in the SLD networkAlthough the orexin-elicited excitatory effects on individual SLD neurons contribute to the depolarization and the increased spikelet activities, the orexin’s effect on GJs may also be involved. We thus recorded pairs of electrically connected SLD neurons in rats. In electrically connected SLD neurons, current injections in either side caused voltage responses in both neurons (Fig. 4a). After orexin-A (100 nM) application, the coupling coefficient was bilaterally increased (n = 7 pairs; controlcell1→cell2: 0.046 ± 0.015, orexin-Acell1→cell2: 0.059 ± 0.017, P = 0.0180; controlcell2→cell1: 0.042 ± 0.016, orexin-Acell2→cell1: 0.054 ± 0.019, P = 0.0280) (Fig. 4b). While the coupling coefficient describes functional coupling relationship, it does not identify the electrophysiological mechanisms, e.g. the underlying GJ conductance or input resistance of the connected cells33. Therefore, we further calculated the GJ conductance in these pairs. Orexin-A (100 nM) bilaterally increased the conductance (n = 7 pairs; controlcell1→cell2: 0.35 ± 0.09 ns, orexin-Acell1→cell2: 0.46 ± 0.13 ns, P = 0.0180; controlcell2→cell1: 0.31 ± 0.08 ns, orexin-Acell2→cell1: 0.43 ± 0.12 ns, P = 0.0180), while the input resistances were not changed (n = 7 pairs; controlcell1: 143.0 ± 43.7 MΩ, orexin-Acell1: 139.8 ± 38.9 MΩ, P = 0.735; controlcell2: 147.3 ± 44.8 MΩ, orexin-Acell2: 153.5 ± 45.8 MΩ, P = 0.0910) (Fig. 4c, d). Thus, in addition to providing the original power that drives the depolarization and the increase in spikelet activities, orexin also promotes the spread of these effects in the SLD network by increasing the GJ conductance.Fig. 4Orexin increases the GJ conductance in coupled SLD neurons.a (Left) Double patch-clamp recordings in a pair of electrically connected SLD neurons in rat. (Right) Representative traces from this electrically connected pair of neurons showing that similar amplitude of voltage change in n1 (Vinjection) induced larger voltage response (Vresponse) in n2 in the presence of orexin-A (100 nM), and vice versa. b Group data showing that orexin-A bilaterally increased the coupling coefficient from cell 1 to cell 2 (left, two-sided Wilcoxon signed-rank test; z = 2.366, P = 0.0180) and cell 2 to cell 1 (right, two-sided Wilcoxon signed-rank test; z = 2.197, P = 0.0280) (n = 7 pairs of neurons). c Group data showing that the input resistances of cell 1 (left, two-sided Wilcoxon signed-rank test; z = 0.338, P = 0.735) and cell 2 (right, two-sided Wilcoxon signed-rank test; z = 1.690, P = 0.0910) were not affected by orexin-A (n = 7 pairs of neurons). d Group data showing that orexin-A bilaterally increased the GJ conductance from cell 1 to cell 2 (left, two-sided Wilcoxon signed-rank test; z = 2.366, P = 0.0180) and cell 2 to cell 1 (right, two-sided Wilcoxon signed-rank test; z = 2.366, P = 0.0180) (n = 7 pairs of neurons). Data represent mean ± SEM. *P < 0.05. Source data are provided as a Source Data file.Orexin modulates synchronized excitation in the SLD in vivoWe next examined the modulations of orexin on the network dynamics of the SLD in vivo through multiple channel recordings in urethane-anesthetized rats (Fig. 5a, b). Microinjections of orexin-A (30 µM, 0.3 μl) in the SLD increased the firing frequency in 67 of 97 (69.1%) SLD units (baseline: 8.0 ± 0.9 Hz, orexin-A: 14.7 ± 1.6 Hz; P = 1.120 × 10−12) (Fig. 5c, d). In addition, the GJ activities were reflected in vivo by coincidental spiking within a sharp time window of ±1 ms between unit pairs. In this condition, 18 of 215 (8.4%) SLD unit pairs showed significant coincident spiking activities. After orexin-A microinjection, this percentage increased to 28 of 215 (13%). Moreover, the coincident probability from the 18 SLD unit pairs with significant interactions in both baseline and orexin-A conditions increased from 1.5 ± 0.2% to 2.4 ± 0.3% (P = 1.076 × 10−4) (Fig. 5e, f). These data suggest that the orexin-elicited excitation and increase in GJ conductance drive synchronized excitation in the SLD network in vivo.Fig. 5Orexin increases synchronized spiking activities in the SLD in vivo.a, b Schematic drawing (a) and representative locations (b) of electrode arrays (pre-covered by DiI, red)/drug canulas (red rectangle) for the SLD multiple channel recordings in urethane-anesthetized rats in vivo. Scale bar: 500 μm. c Orexin-A (30 µM, 0.3 μl) microinjection increased the firing frequency of a SLD unit in vivo. Overlaid spike waveforms and raster plot of spikes in 10-s epochs before and after orexin-A application were expanded. d Orexin-A significantly increased the firing frequency of the SLD units (n = 67/97 units, two-sided Wilcoxon signed-rank test; z = 7.115, P = 1.120 × 10−12). e Two example cross-correlograms with significant coincidental spiking within ±1 ms before and after orexin-A microinjections from the rat in c. Note that GJ-mediated interactions occurred either within ±1 ms (pair 1) or ±5 ms (pair 2). We focused on the coincident activities within ±1 ms. A significant interaction was counted if a pairwise 1% threshold (dotted gray lines) was crossed. The mean of the chance coincident (solid blue line) was computed from 30 shuffled cross-correlograms. f Orexin-A significantly increased the coincident spiking probability of the SLD unit pairs with significant interactions in both baseline and orexin-A conditions (n = 18/215 pairs, two-sided Wilcoxon signed-rank test; z = 3.881, P = 1.076 × 10−4). Only unit pairs between the adjacent recording sites were analyzed. g Orexin-A (30 μM, 0.3 μl) microinjection still increased the firing frequency of SLD neurons after the pre-injection of CBX (100 mM) (n = 39/72 units, two-sided Wilcoxon signed-rank test; z = 5.442, P = 5.255 × 10−8). h Two example cross-correlograms with significant coincidental spiking within ±1 ms before and after orexin-A microinjection in the presence of CBX. i Orexin-A failed to increase the coincident spiking probability of the SLD unit pairs with significant interactions in both baseline and orexin-A conditions in the presence of CBX (n = 13/182 pairs, two-sided Wilcoxon signed-rank test; z = 0.622, P = 0.542). Data represent mean ± SEM. **P < 0.01. Source data are provided as a Source Data file.To further assess the contribution of GJ activities in the orexin-elicited synchronized excitation, we tried to use CBX to interfere with the SLD GJs. Although CBX microinjection alone (100 mM, 0.3 μl) only partially reduced the coincident spiking probability of the SLD unit pairs in vivo (Supplementary Fig. 8a–c), we found that CBX pre-injection was able to abolish the orexin-elicited elevation of synchronized spiking and orexin-A still increased the spiking rate in 39 of 72 (54.2%) SLD units (CBX: 5.2 ± 0.9 Hz, CBX/orexin-A: 8.9 ± 1.5 Hz; P = 5.255 × 10−8) (Fig. 5g–i). In this circumstances, orexin-A microinjection after CBX did not change the number of SLD unit pairs with significant interactions (CBX: 7.7%, 14 of 182 pairs; CBX/orexin-A: 7.7%, 14 of 182 pairs) and the coincident probability from 13 pairs with significant interactions in both CBX and CBX/orexin-A conditions (CBX: 2.5 ± 0.3%; CBX/orexin-A: 2.4 ± 0.3%; P = 0.542). These data further suggest that the elevated GJ activities are necessarily involved in the orexin-elicited synchronized excitation of the SLD network.To evaluate changes of the SLD output induced by drug injections, we also simultaneously recorded the hippocampal local field potential (LFP) activities, which largely depend on the SLD output during REM sleep11. In urethane-anesthesia, the hippocampal activities were dominated by slow oscillations at ~1 Hz (Supplementary Fig. 9a). We found that microinjections of orexin-A into the SLD increased the power of slow oscillations (0.3–2.5 Hz) and the phase-locking strength between the SLD spiking activities and these oscillations (Supplementary Fig. 9b–d), suggesting that orexin facilitates SLD output and the enhanced SLD-hippocampus correspondence underlies. Moreover, these effects were all blunted by pre-injection of CBX into the SLD (Supplementary Fig. 9e–l). Collectively, these data suggest that SLD orexin signaling promotes the correspondence between the SLD and its downstream target to enhance the SLD output, and the orexin-elicited synchronized excitation may contribute to this process.Activation of SLD orexin signaling prolongs REM sleep episodeOptogenetics was then employed to examine the contributions of the orexin-enhanced SLD output in brain state regulation. AAV-Ef1α-DIO-ChR2-mCherry was bilaterally injected into the LH of orexin-Cre mice (orexinChR2 mice), and optical fibers were implanted in the bilateral SLD (Fig. 6a, b). Through patch-clamp recordings, we found that in orexinChR2 mice, optogenetic activation of the SLD orexin terminals excited SLD neurons through the release of orexin34 (Supplementary Fig. 10a–f). In addition, the functional expression of light-sensitive opsins in the LH orexin neurons after virus infections was also confirmed by immunostaining and patch-clamp methods (Supplementary Fig. 11a–e).Fig. 6Optogenetic activation of SLD orexin signaling prolongs REM sleep bouts.a Schematic drawing for optogenetic activation of the SLD orexin terminals. b Representative viral infections of ChR2-mCherry+ (red) in the LH orexin-A+ (blue) neurons (left, scale bar: 50 μm) and optical-fiber location above the SLD (right, scale bars: 500 and 50 μm in magnification). c Representative trials of raw EEG/EMG recordings during 20-Hz optical activation (20 pulses every 3 s) of SLD orexin signaling during NREM sleep. Purple arrows indicated wakefulness. d Heatmap of EEG (left)/EMG (right) recordings from all tested trials in the mouse in (c). EEG/EMG recordings before wakefulness (purple line) were normalized and averaged in the bottom. Blue shadow represented SEM. e, f Effects of optical activation during NREM sleep on the total EEG power (e, 5-Hz: t5 = 2.442, P = 0.0585; 10 Hz: t5 = 3.939, P = 0.0110; 20 Hz: t5 = 4.633, P = 5.667 × 10−3) and EMG amplitude (f, 5 Hz: t5 = 1.610, P = 0.168; 10 Hz: t5 = 2.180, P = 0.0811; 20 Hz: t5 = 5.056, P = 3.912 × 10−3) (n = 6 orexinChR2 mice). OrexinmCherry mice was set as control (n = 5 mice; EEG (20 Hz): t4 = 0.840, P = 0.448; EMG (20 Hz): t4 = 0.506, P = 0.640). All analyzed by two-sided paired t-test. g Effects of optical activation during NREM sleep on the EEG theta/total power ratio (n = 6 orexinChR2 mice, two-sided Wilcoxon signed-rank test; 5 Hz: z = 0.105, P = 0.917; 10 Hz: z = 1.572, P = 0.116; 20 Hz: z = 2.201, P = 0.0277). h Procedures of closed-loop optogenetics during REM sleep. i, j Effects of optical activation (20 Hz, 20 pulses every 3 s) during REM sleep on the EEG theta power (i, t8 = 3.381, P = 9.624 × 10−3), EMGREM/NREM ratio (i, t8 = 2.126, P = 0.0662), and episode duration (j, t8 = 2.368, P = 0.0454) (n = 9 orexinChR2 mice). All analyzed by two-sided paired t-test. Data represent mean ± SEM. *P < 0.05; **P < 0.01. Source data are provided as a Source Data file.During NREM sleep, 20-Hz light stimulation (473 nm, 20 pulses every 3 s) of the SLD orexin terminals caused an immediate (latency: 2.4 ± 0.3 s) decrease in the total EEG power (baseline: 963.2 ± 196.6 μV2, light: 574.8 ± 186.8 μV2; P = 5.667 × 10−3), followed by a late-onset (latency: 9.4 ± 1.7 s) decrease in EMG amplitude (baseline: 6.6 ± 1.7 μV, light: 5.6 ± 1.6 μV; P = 3.912 × 10−3) unless disrupted by transitions to wakefulness before the light-off (n = 6 orexinChR2 mice) (Fig. 6c–f). The EEG theta component (theta/total power) increased before the disruption of wakefulness (baseline: 0.19 ± 0.02, light: 0.23 ± 0.02; n = 6 orexinChR2 mice; P = 0.0277) (Fig. 6g), suggesting an activation of brain state in this period. In the control group expressing only mCherry (orexinmCherry mice), light (20 Hz) stimulation did not affect the total EEG power (baseline: 836.9 ± 126.2 μV2, light: 870.2 ± 142.9 μV2; P = 0.448) or EMG amplitude (baseline: 5.6 ± 1.5 μV, light: 5.5 ± 1.5 μV; P = 0.640) (n = 5 orexinmCherry mice) (Fig. 6e, f). The activation of brain state followed by a decrease of EMG amplitude also occurred during physiological NREM to REM sleep transitions. We thus analyzed the EEG power spectrogram during the light stimulation (20-Hz) and found that consecutive theta oscillations characterizing REM sleep were not induced (Supplementary Fig. 12a). Moreover, the latency to REM sleep was not changed by light (Supplementary Fig. 13a). Together, these data demonstrate that the activation of SLD orexin signaling does not induce REM sleep transitions, but indeed causes the activation of brain state with increased EEG theta component and muscle tone decrease. In NREM sleep, this activation may cause a short awakening effect. Although the latency from NREM sleep to wakefulness was also not significantly affected, the NREM sleep to wakefulness transition probability slightly increased between ~20 and ~30 s after light delivery (Supplementary Fig. 13b–e). Increasing the stimulation intensity (20 Hz for 20 s) still only caused the brain state activation during NREM sleep, but failed to induce REM sleep transitions. In this condition, the transition probability from NREM sleep to wakefulness was high (Supplementary Fig. 14a–d).We next examined whether the brain state activation and muscle tone decrease induced by SLD orexin signaling contribute to REM sleep regulation. Optogenetic activation of the SLD orexin terminals was applied during individual REM sleep episodes with 50% probability, and the REM sleep performances and length were compared between the light-on and light-off trials (Fig. 6h). In REM sleep, light stimulation (473 nm, 20 Hz, 20 pulses every 3 s) further consolidated the EEG theta oscillations (theta power; light-off: 295.3 ± 72.6 μV2, light-on: 321.1 ± 77.0 μV2; n = 9 orexinChR2 mice); P = 9.624 × 10−3) (Fig. 6i). In addition, to eliminate biased EMG modifications induced by posture changes of different sleep epochs35, the EMGREM/NREM ratio was used to examine changes of muscle atonia. In this condition, a decreasing trend in the EMGREM/NREM ratio was also observed (EMGREM/NREM ratio; light-off: 0.94 ± 0.01, light-on: 0.91 ± 0.02; n = 9 orexinChR2 mice; P = 0.0662) (Fig. 6i). Consequently, optical activation significantly prolonged the REM sleep episode duration by 12.1% (light-off: 65.4 ± 3.1 s, light-on: 73.3 ± 2.3 s; n = 9 orexinChR2 mice; P = 0.0454) (Fig. 6j). All these effects were not observed in the mice expressing only mCherry (Supplementary Fig. 15a–c). These data suggest that SLD orexin signaling prolonged REM sleep episodes through consolidating the brain state activation and muscle tone decrease, and may thereby, contribute to REM sleep stabilization.Silencing of SLD orexin signaling destabilizes REM sleepThe contributions of SLD orexin signaling in the maintenance of REM sleep episodes were then examined. To observe the activities of SLD orexin signaling during REM sleep, AAV-CAG-FLEX-jGCaMP7b was injected into the LH of the orexin-Cre mice, followed by implantation of optical fibers above the SLD (Fig. 7a, b). Intriguingly, jGCaMP fluorescence changes (ΔF/F) of the SLD orexin terminals were observed during the REM sleep episodes (0.26 ± 0.03%), which was significantly higher than that (0.02 ± 0.01%) of the preceding NREM sleep episodes and lower than that (0.89 ± 0.20%) of the following wakefulness episodes (n = 7 mice; one-way repeated-measures ANOVA; F(2, 12) = 16.758; P = 3.358 × 10−4; post hoc LSD comparison test; REM vs. NREM: P = 2.481 × 10−4; REM vs. wakefulness: P = 0.0127; NREM vs. wakefulness: P = 4.563 × 10−3) (Fig. 7c, d), suggesting that the activity of SLD orexin signaling increased during natural REM sleep episodes. In addition, the jGCaMP signals elevated throughout the entire period, including both phasic and tonic REM sleep (Supplementary Fig. 16a). We thus performed optogenetic inhibition of SLD orexin signaling during REM sleep. AAV-CAG-FLEX-ArchT-GFP was bilaterally injected into the LH of orexin-Cre mice (orexinArchT mice) followed by implantation of optical fibers bilaterally in the SLD (Fig. 7e). The effectiveness of optogenetic inhibition was validated by immunostaining and patch-clamp methods (Supplementary Fig. 11a–d). Optogenetic inhibition consistently decreased the theta oscillation power (light-off: 405.0 ± 61.0 μV2, light-on: 375.1 ± 58.3 μV2; P = 0.0325), and the REM sleep episode duration (light-off: 76.2 ± 4.4 s; light-on: 62.3 ± 4.1 s; P = 0.0363) (n = 9 orexinArchT mice) (Fig. 7f, g). Changes in the EMGREM/NREM ratio (light-off: 0.91 ± 0.02, light-on: 0.90 ± 0.01; n = 9 orexinArchT mice; P = 0.320; Fig. 7f) were not detected by this inhibition. These data thus suggest that SLD orexin signaling is involved in the maintenance of REM sleep episodes.Fig. 7Optogenetic inhibition of SLD orexin signaling shortens REM sleep bouts.a Schematic drawing of fiber photometry recordings in the SLD orexin terminals. b Representative viral infections of jGCaMP7b (green) in the LH orexin-A+ neurons (blue) (top, scale bar: 25 μm), and optical fiber location above the SLD orexin terminals (bottom, scale bar: 100 μm). c Representative jGCaMP fluorescence traces of the SLD orexin terminals, spectrogram of EEG recordings, and raw EMG recorded simultaneously in a episode of REM sleep (R), the preceding NREM sleep (NR), and the following wakefulness (W). d Averaged changes of jGCaMP fluorescence in REM sleep, the preceding NREM sleep, and the following wakefulness (n = 7 mice; one-way repeated-measures ANOVA; F(2, 12) = 16.758, P = 3.358 × 10−4; post hoc LSD comparison test; REM vs. NREM: P = 2.481 × 10−4; wake vs. NREM: P = 4.563 × 10−3; REM vs. wake: P = 0.0127). e Schematic drawing for optogenetic inhibition of the SLD orexin terminals (left) and procedures for closed-loop optogenetic inhibition during REM sleep (right). The EEG/EMG signals were visually inspected on-line. Lasers were manually turned on with 50% probability after the detection of REM sleep (~10 s), and turned off when the REM sleep episode ended. f Effects of optical inhibition of the SLD orexin terminals during REM sleep on the EEG theta power (left, two-sided paired t-test; t8 = 2.582; P = 0.0325) and EMGREM/NREM ratio (right, two-sided paired t-test; t8 = 1.060; P = 0.320) (n = 9 orexinArchT mice). g Effects of optical inhibition of the SLD orexin terminals during REM sleep on the episode duration of REM sleep (n = 9 orexinArchT mice, two-sided paired t-test; t8 = 2.512; P = 0.0363). Data represent mean ± SEM. *P < 0.05; **P < 0.01. Source data are provided as a Source Data file.To silence SLD orexin signaling on an hourly time scale, AAV-retro-hSyn-DIO-hM4D (Gi)-mCherry (mixed with AAV-hSyn-EGFP for histological analysis) was bilaterally injected into the SLD of the orexin-Cre mice (orexinSLD-hM4D mice) (Fig. 8a–c). The effectiveness of chemogenetic manipulation was confirmed by immunostaining and patch-clamp methods (Supplementary Fig. 17a–d). Clozapine N-oxide (CNO, 3 mg/kg) or vehicle (0.9% NaCl) was injected intraperitoneally (i.p.) in the orexinSLD-hM4D mice at ZT 4 (12:00 a.m.), and the brain states were monitored for the next 4 h. Compared to vehicle, CNO caused a 17.0% reduction in the total REM sleep amount (vehicle: 8.8 ± 1.2%, CNO: 7.3 ± 1.2%; P = 6.170 × 10−3), and the amount of wakefulness (vehicle: 37.3 ± 2.7%, CNO: 39.7 ± 2.7%; P = 0.217) or NREM sleep (vehicle: 53.9 ± 2.7%, CNO: 53.1 ± 2.9%; P = 0.615) was not affected (n = 8 orexinSLD-hM4D mice) (Fig. 8d, e). The reduced REM sleep amount was due to a decrease in the episode duration (vehicle: 51.3 ± 5.1 s, CNO: 40.0 ± 2.7 s; P = 0.0269), whereas the episode number was not altered (vehicle: 25.3 ± 3.5, CNO: 25.3 ± 3.3; P = 1.000) (n = 8 orexinSLD-hM4D mice) (Fig. 8f). We also assessed the regulation of SLD orexin signaling in REM sleep through pharmacological manipulation in rats (Supplementary Fig. 18a, b). Similarly, through microinjection of orexin-A or TCS 1102 in the SLD, we found that SLD orexin signaling increased the REM sleep amount via prolonging the episode duration (Supplementary Fig. 18a, b). These data demonstrate a routinely involved role of SLD orexin signaling in the stabilization of REM sleep.Fig. 8Chemogenetic silencing of SLD orexin signaling destabilizes REM sleep.a Schematic drawing for chemogenetic silencing of SLD orexin signaling. b Representative virus injection sites in the SLD (scale bar: 500 μm). mlf medial longitudinal fasciculus. c Representative expression of hM4D (Gi)-mCherry (red) in the orexin-A+ (blue) neurons. Scale bar: 20 μm. d Hypnogram showing the sleep/wakefulness states in the following 4 h after vehicle and CNO injections in a tested orexinSLD-hM4D mouse. e The amount of REM sleep (two-sided paired t-test; t7 = 3.865, P = 6.170 × 10−3), wakefulness (two-sided paired t-test; t7 = 1.355, P = 0.217), and NREM sleep (two-sided paired t-test; t7 = 0.527, P = 0.615) during the 4-h recording period after CNO and vehicle injections (n = 8 orexinSLD-hM4D mice). f The episode duration (left, two-sided paired t-test; t7 = 2.789, P = 0.0269) and number (right, two-sided paired t-test; t7 = 0, P = 1.000) of REM sleep during the 4-h recording period after CNO and vehicle injections (n = 8 orexinSLD-hM4D mice). g Representative raw EEG/EMG activities in REM sleep, the preceding NREM sleep, and the following wakefulness after vehicle and CNO injections in a tested orexinSLD-hM4D mouse. Note that a prominent failure of muscle atonia in REM sleep occurred after CNO injection. h EMGREM/NREM activity map after vehicle (top) and CNO (bottom) injections from all REM sleep episodes and the preceding NREM sleep in a orexinSLD-hM4D mice. These episodes are sorted vertically from the highest (purple) to the lowest (blue) mean EMGREM/NREM ratio. The x-axis represents the episode duration normalized between 0 and 2π. Note that CNO injection disrupted the muscle tone decrease from NREM to REM sleep, and failure of muscle atonia occurred in REM sleep. i, j Changes of EMGREM/NREM ratio (i, two-sided paired t-test; t7 = 3.619, P = 8.518 × 10−3) and EEG theta power (j, two-sided paired t-test; t7 = 2.387, P = 0.0484) after CNO and vehicle injections (n = 8 orexinSLD-hM4D mice). Data represent mean ± SEM. *P < 0.05; **P < 0.01. Source data are provided as a Source Data file.Intriguingly, through EEG/EMG analysis in REM sleep episodes of the entire 4-h recordings, a general failure of muscle atonia was obviously observed after CNO injections in all tested orexinSLD-hM4D mice. The physiological muscle tone decrease from NREM to REM sleep (EMGREM/NREM ratio) was abolished (vehicle: 93.4 ± 1.6%; CNO: 99.0 ± 1.5%; n = 8 orexinSLD-hM4D mice; P = 8.518 × 10−3) (Fig. 8g–i). In this case, an abnormal behavior phenotype characterized by severe disruption of muscle atonia was observed after constructing the EMG activity map (Fig. 8h). These observations may explain that more than half of narcoleptic patients suffer from failure of REM sleep paralysis36. Moreover, the theta oscillation power (vehicle: 409.3 ± 52.3 μV2; CNO: 372.9 ± 55.6 μV2; n = 8 orexinSLD-hM4D mice; P = 0.0484) (Fig. 8j) in REM sleep were also decreased after silencing of SLD orexin signaling, and the power distribution remained unchanged (Supplementary Fig. 19a, b). All these effects were not observed in the control mice expressing only mCherry (Supplementary Fig. 20a, b). Together, the disruption of these core features in REM sleep behavior after the loss of SLD orexin signaling may eventually destabilize individual REM sleep episodes, and thus reduce REM sleep amount, further suggesting the essential role of SLD orexin signaling in REM sleep stabilization.DiscussionThe present study demonstrated a direct role of SLD orexin signaling in REM sleep stabilization. The REM sleep relevance was firstly identified in a sub-cluster of orexin neurons, which constituted a specific efferent node of orexin signaling to the glutamatergic neuron-enriched SLD region. We next observed the electrophysiological modulations of orexin on the SLD neurons. Intriguingly, in addition to the reported excitatory effects in the SLD neurons14,17, orexin also actively increased the GJ conductance among the SLD network. These parallel actions interacted to restrain the SLD neuronal activities towards a synchronized excitation pattern. Moreover, the SLD output was consequently enhanced to induce brain state activation and muscle tone decrease in vivo. During REM sleep, this mechanism contributed to consolidating the EEG theta-band activities and muscle atonia. In this way, the stabilization of REM sleep was achieved under physiological conditions.Orexin neurons innervate multiple neuronal circuitries controlling sleep/wakefulness states7. In general, many wakefulness-promoting structures receive abundant orexin innervations6. It thus readily leads to assumptions of secondary effects on sleep by orexin signaling1. However, accumulating evidence has shown orexin innervations in sleep-promoting regions7,15,37. The orexin innervations have been observed in the SLD by previous reports and the present study6,15. We further found that the OXRs were abundantly expressed on the SLD glutamatergic neurons or REM-on neurons. These findings rationally suggest an alternatively direct effect of SLD orexin signaling in REM sleep regulation. Moreover, the observed characteristics of the OXSLD neurons were also consistent with direct REM sleep regulation. These neurons were sporadically distributed and distinct from the OXLC neurons, even though the LC was adjacent to the SLD. These findings support notions that orexin neurons can be classified into different groups based on collateral downstream projection patterns38. Actually, it has long been suspected that orexin neurons are divided into distinct groups, according to their roles in seemingly disparate behaviors, such as arousal, reward seeking, and motor control39,40. A group of REM sleep-related orexin neurons may also exist, as micro-dialysis has reported that orexin release increased during REM sleep41. Intriguingly, with projection-specific tagging, REM sleep-related activation emerged in the relatively minor (~7%) OXSLD class. Furthermore, by using fiber photometry on regional SLD orexin terminals, their increased activities during REM sleep was also observed in the present study. Another interesting finding of the c-Fos screening was that the OXSLD neurons were activated against the generally silent background of the OXentirety during REM sleep. This phenomenon may also account for the conflicting results from previous reports37. Based on extracellular recordings from indiscriminate orexin neurons, the firing activities of them in REM sleep have been described as intermediate, low, or even silent42–44.The elevated activities of the OXSLD neurons in REM sleep indicate their essential roles on the SLD neurons. Intriguingly, following the activation of orexin neurons, complex modulations have been reported in different sleep/wakefulness-related structures, such as the suprachiasmatic nuclei (SCN) and the tuberomammillary nucleus (TMN)45–48. Importantly, we found that the SLD GJs provided specific conditions for orexin to elicit a more dedicated modulation pattern, rather than just excitation17. CBX-sensitive spikelet activities and functional Cx-36 expression were both abundantly observed in the SLD neurons, suggesting possibly extensive electrical connections in the SLD network. With the aid of GJs, the orexin-elicited direct and partial excitation could spread from the OXRs+ SLD neurons (occupying ~3/4 of the total population) to affect the large-scale SLD network. The substrates exchanged in this process are still unclear, as GJs are reciprocal pathways for either ionic current or small molecules49. Alternatively, we showed that the electrophysiological influences generated by exchange may largely affect the SLD neuronal activity pattern. In detail, the propagation of orexin-elicited direct excitation, including the sub-threshold depolarization and depolarization-evoked spikes, caused two forms of consequences. On the one hand, the spread of sustained depolarization via GJs alone may increase the effect range50. The OXRs− SLD neurons can thus be recruited in the excitation of orexin. This phenomenon was reflected by the reduced number of orexin-excited neurons after blocking GJs by CBX/MEF and the unchanged excitation amplitude after blocking spike generation by TTX. On the other hand, the activation-evoked spikes would form another discrete spread source, increasing the spike to spikelet transmission32. We indeed observed increased spikelet activities after orexin applications in the SLD. Intriguingly, spikelet activities are thought to be the basis for synchronized firings32,49,50. Therefore, orexin elicited widespread activation in the electrically connected SLD network, and the activation-evoked firings may tend to be synchronized.The significance of orexin-elicited synchronized excitation on the SLD output is also demonstrated. The increased SLD firings promote several types of REM sleep physiology, such as ponto-geniculo-occipital (PGO) waves, hippocampal theta activities and muscle atonia1,11. The global excitation of orexin may thus enhance the SLD output to regulate REM sleep. In fact, the regulation of firing frequency in the SLD neurons appears to be a common way to influence REM sleep19,21,22. Intriguingly, it has been assumed that the synchrony of SLD firings constitutes another essential force for enhancing the output26–28. This phenomenon is directly evidenced here, as the orexin-elevated synchronized SLD firings were more effectively phase-locked to the hippocampal oscillations (0.3–2.5 Hz) in anesthesia, and consequently, increased its power. Furthermore, disruption of the orexin-induced synchronized excitation in the SLD by CBX blunted all these effects. Therefore, both the orexin-elicited excitation and firing synchrony are essentially involved in the SLD output.Considering that the SLD output with synchronized information may be fundamental for recruiting coherence activities in REM sleep-related brain regions28, the orexin-elicited synchronized excitation may contribute to all processes involved in REM sleep, such as the core features, e.g., corticohippocampal activation and muscle atonia. Actually, for the corticohippocampal theta oscillations, several reports have indeed provided the anatomical basis for the orexin-enhanced SLD output to influence them. The neural pathways may include ascending projections from the SLD to the intralaminar thalamus12, the precoeruleus region11, or the hippocampus51. In addition, muscle atonia can also be influenced through descending projections from the SLD to the gigantocellular reticular nucleus or the spinal cord19,35. Consistently, we observed brain state activation with an increased theta component and muscle tone decrease after optogenetic activation of SLD orexin signaling in free-moving animals. Moreover, insufficiencies of EEG theta oscillation and muscle atonia were observed after chemogenetic inhibition of SLD orexin signaling. Notably, alterations in EEG theta oscillations and failure of muscle atonia are frequently reported during REM sleep in narcoleptic patients8–10. Given the tight relationship between EEG theta oscillation/muscle atonia and REM sleep stability1, the orexin-enhanced SLD output seems to be essentially involved in REM sleep stabilization. We consequently found that the REM episode duration was increased by optogenetic activation and decreased by optogenetic/chemogenetic inhibition of SLD orexin signaling. Thus, a decrease in REM sleep amount was observed after prolonged pharmacological/chemogenetic silencing of this signaling. These findings provide evidence that the orexin-enhanced SLD output and the consequent brain state activation/muscle tone decrease contribute to REM sleep stabilization under physiological conditions.In summary, the maintenance of REM sleep, another active brain state in addition to wakefulness, also requires excitatory orexin signaling. In fact, it has been reported that multiple projection sites of orexin signaling help to fulfill its diverse physiological roles6,34,40. In this condition, the present study rationally suggests that the orexin projections to the SLD stabilizes REM sleep, expanding previously reported findings that its projections to wakefulness-promoting nuclei mediate arousal52–56. Our findings provide additional insights to understand that both REM sleep symptoms and wakefulness-maintaining deficits exist after loss of central orexin signaling9,10,36,. Intriguingly, based on the demands of internal and external states, the orexin neuronal entirety is thought to operate as an integrator in the homeostatic regulation of diverse physiological functions10,40,57. For integral sleep/wakefulness cycling behavior, loss of orexin signaling indeed causes severely disrupted arrangements of vigilance states, while the total time of the individual state remains unchanged58. Therefore, the fine division of the orexin neuronal entirety in wakefulness and REM sleep stabilization may work together to orchestrate this basic brain function.MethodsAnimalsSprague-Dawley rats (Laboratory Animal Center, Third Military Medical University) were used for experiments in Figs. 1–5 and Supplementary Figs. 1–9 and 18. To selectively manipulate SLD orexin signaling in behavioral experiments, orexin-Cre mice (gift from L. de Lecea, Stanford University, USA)59 and Ai27D mice (012567, Jackson Laboratory, USA) were used for experiments in Figs. 6–8 and Supplementary Figs. 10–17 and 19, 20. All animal care and experimental procedures were approved by the Guide for the Care and Use of Laboratory Animals of the Third Military Medical University. Animals were housed in 12-h light/dark cycle, with lights-on at 8:00 a.m. (ZT 0) and lights-off at 8:00 p.m. (ZT 12). The environment temperature was kept constant at 22 ± 1 °C and the relative humidity was kept between 40% and 60%. Food and water were available ad libitum.CTB retrograde tracingMale rats (250–300 g) were anesthetized with sodium pentobarbital (75 mg/kg, i.p.) and then fixed on a stereotaxic apparatus (RWD Life Science, China). For this and the following experiments in rats, we focused on the SLD region from AP: −9.20 to −9.80 mm according to the rat brain atlas60, as previous studies reported that this region was enriched of glutamatergic neurons that were responsible for REM sleep generation18. To target this region, CTB-488 (69 nl, 5 µg/µl) was injected into the SLD unilaterally (bregma: AP: −9.50 mm, ML: −1.30 mm, DV: −8.20 mm) (Supplementary Fig. 2a). Besides, CTB-555 (69 nl, 5 µg/µl) was injected into the LC (AP: −9.60 mm, ML: −1.40 mm, DV: −7.20 mm), ipsilaterally to the SLD CTB-488 injection. CTB tracers were injected using Nanoject II (Drummond Scientific, USA) via a silicate-glass micro-pipette (tip diameter: ~20 μm). Multiple 2.3 nl injections were made at 10-s intervals and the pipette was left in place for additional 10 min after the injections. At least 2 weeks after injections, rats were sacrificed and coronal brain slices containing the LH and the SLD/LC were collected for following immunohistological processes.REM sleep deprivationRats with CTB-488 injection in the unilateral SLD were subjected to RSD through employing the well-accepted inverted flowerpot technique. Briefly, rats were placed on a small-round platform (diameter: 6.5 cm) surrounded by water (1 cm under the platform) at 10:00 a.m. to prevent REM sleep, but not NREM sleep18. Rats were available to food and water ad libitum and kept in the 12-h light/dark cycles. After 72-h RSD, rats were removed from the platform at 10:00 a.m. and were placed to their original cages to allow REM sleep recovery. EEG and EMG recordings were performed to monitor the sleep/wakefulness states. Another group of rats that remained in their home cages throughout the experiment were set as control. All rats were anesthetized and sacrificed after 2.5 h from the first REM sleep episode in recording sessions. The coronal brain sections containing LH and SLD were collected for further c-Fos immunostaining and histological processes, respectively.ImmunohistochemistryRats or mice were deeply anesthetized with pentobarbital sodium and transcardially perfused with saline, followed by 4% paraformaldehyde. Brains were fixed in 4% paraformaldehyde and kept in 30% sucrose/PBS at 4 °C. Coronal slices of 10–20 μm were made for immunohistochemical staining, using a vibratome (CM 3050S, Leica, Germany). Slices for immunostaining were sequentially incubated in the primary and secondary antibodies following the instructions. The primary antibodies used in the present study were as follows, goat anti-orexin-A (1:500, sc-8070, Santa Cruz, USA), rabbit anti-c-Fos (1:1000, ABE-457, Millipore, USA), mouse anti-c-Fos (1:1000, AB208942, Abcam, USA), rabbit anti-orexin 1/2R (1:500, bs-1095R, Bioss, China), mouse anti-Cx-36 (1:1000, sc-398063, Santa Cruz, USA), mouse-anti-NeuN (1:1000, MAB377, Millipore, USA), guinea-pig-anti-VGLUT1 (1:1000, AB 5905, Millipore, USA), mouse anti-GAD 67 (1:500, MAB5406, Millipore, USA), rabbit anti-glutamate (1:500, G6642, Sigma, USA), rabbit anti-GABA (1:1000, A2052, Sigma, USA), and rabbit anti-MCH (1:2000, H-070-47, Phoenix Pharmaceuticals, USA). Images were acquired with a LSM 800 (Carl Zeiss, Germany) and analyzed by Zen software 2012 (Carl Zeiss, Germany). Histological processes were also performed to locate the sites of CTB/virus injections and drug canula/electrode/optical fiber implantations. Data were excluded if the locations were not correct.Cell countingTo quantify the SLD-projecting or LC-projecting orexin neurons, six coronal sections of each CTB-injected rat containing the LH with a distance of 200 µm (between AP: −2.30 and −3.80 mm) were collected. After orexin-A immunostaining, we counted the number of total orexin neurons, CTB-488-labeled orexin neurons (SLD-projecting), CTB-555-labeled orexin neurons (LC-projecting), and orexin neurons labeled by both CTB tracers (SLD/LC-projecting). The ratio of each group was then reported. To examine the REM sleep-related activities of OXSLD neurons, c-Fos immunostaining was further applied in rats with CTB-488 injection after recovery from RSD and the control rats. The counting method was similar.EEG/EMG recordings and drug injection in free-moving ratsRats which underwent EEG/EMG recording were chronically implanted with EEG/EMG electrodes. Two EEG electrodes were implanted at the frontal and entorhinal cortex region. Two EMG electrodes were placed between the neck musculature. Two drug canulas were implanted bilaterally above the SLD (bregma: AP: −9.50 mm, ML: −1.70 mm, DV: −7.80 mm) for drug injections in the pharmacological tests. All electrodes were soldered to a micro-pin connector, and then affixed to the skull with dental cement. Before tests, rats were allowed to recover within their home cages for at least 7 days and acclimated in the recording cage for 2 days. The EEG/EMG signals were recorded and stored for further analysis using omniplex neural data acquisition system (PlexControl 1.10, Plexon, USA). The EEG/EMG signals were digitized at 1000 Hz and band-pass filtered (EEG: 1–30 Hz, EMG: 20–100 Hz). EEG/EMG signals were on-line monitored and off-line analyzed by two investigators who are blind to treatments based on spectral signatures of EEG–EMG waveforms in 5-s epochs. Wakefulness was defined as desynchronized, low-amplitude EEG rhythms and elevated EMG activity with phasic bursts. NREM sleep was defined as synchronized, high amplitude and low frequency (1–4 Hz, delta) EEG activity and lower EMG activity compared with wakefulness. REM sleep was defined as containing consecutive theta (6–9 Hz) oscillations with further decreased EMG activity compared with the preceding NREM sleep. In pharmacological tests, drugs were injected into SLD by a micro-syringe (filled with drugs) connected to a micro-drive (KD310, KD Scientific, USA). Drugs were delivered in a random pattern with 2 days interval at a speed of 0.06 μl/min between ZT4 (11:00 a.m.) and ZT5 (12:00 a.m.). The percentage of time spent in each state in the following hour after drug injection was reported, as drug effects returned to baseline level after the first post-injection hour.Whole-cell patch-clamp recordingCoronal brainstem slices containing the SLD (300–400 μm) were prepared with a vibroslicer (VT 1200S, Leica, Germany) from rats aged 9–14 days. During recordings, the slices were continuously superfused with 95% O2 and 5% CO2 oxygenated ACSF (composition in mM: 125 NaCl, 2.5 KCl, 1.25 NaH2PO4, 1.3 MgSO4, 26 NaHCO3, 2 CaCl2 and 20 d-glucose) at 2 ml/min in room temperature. Whole-cell recordings were performed on SLD neurons with borosilicate glass pipettes (3–5 MΩ) filled with an internal solution (composition in mM: 130 K-methylsulfate, 5 KCl, 2 MgCl2, 10 HEPES, 0.1 EGTA, 2 Na2-ATP, 0.2 Na2-GTP, adjusted to pH 7.25 with 1 M KOH). Several previous reports61–63 and our immunostaing observations (Supplementary Fig. 2b, c) found that this SLD region mainly contained glutamatergic neurons and low density of small (5–15 μm) GABAergic neurons. In order to investigate the effects of orexin on SLD glutamatergic neurons, SLD neurons had a soma diameter larger than 15 μm (membrane capacitance > 80 pF) were recorded. We also used biocytin (0.5%) to label a group of recorded neurons. Post hoc immunostaining revealed that all pre-tested neurons were co-labeled by biocytin and VGLUT1 in this condition (Supplementary Fig. 5a). Recordings were performed in current/voltage clamp mode by Clampex 10.3 (Molecular Devices, USA) using a Multiclamp-700B amplifier (Molecular Devices, USA). Data were analyzed by Clampfit 10.3 (Molecular Devices, USA). In all experiments, neurons were excluded from the study if the series resistance exceeded 20 MΩ or changed by 20%.Double patch-clamp recordings were applied in pairs of electrically connected SLD neurons. Recorded neurons were first adjusted to around −60 mV by injecting constant current in current-clamp. During orexin application, the orexin-elicited depolarization in membrane potential at the steady state was adjusted to the baseline by constant current injection. Coupling coefficient and input resistance of both cells were determined by applying a series of hyperpolarizing current steps in either cell before and during the steady state of the orexin’s effects. Voltage changes from each step were averaged from at least three sweeps. Input resistance was calculated from the slope of the relationship between injected current and the corresponding voltage changes. Coupling coefficient was calculated from the slope of the relationship between ΔVCoupled cell/ΔVInjected cell. All linear fits has a R-square value higher than 0.97. To examine whether orexin directly modulates GJs, we calculated the GJ conductance by the following equation64:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{\mathrm{c}},{\mathrm{cell}}1} = {R}_{{\mathrm{in}},{\mathrm{cell}}1} \times {\mathrm{cc}}_{12}/\left[ {\left( {{R}_{{\mathrm{in}},{\mathrm{cell}}1} \times {R}_{{\mathrm{in}},{\mathrm{cell}}2}} \right) - \left( {{R}_{{\mathrm{in}},{\mathrm{cell}}1} \times {\mathrm{cc}}_{12}} \right)^2} \right]$$\end{document}Rc,cell1=Rin,cell1×cc12/Rin,cell1×Rin,cell2−Rin,cell1×cc122where Rc,cell1 represents the GJ resistance from cell1 to cell2, Rin,cell1 represents the input resistance of cell1, Rin,cell2 represents the input resistance of cell2, cc12 represents the coupling coefficient from cell1 to cell2. GJ conductance from cell1 to cell2 is the inverse of Rc,cell1.Multiple channel recordingRats were anesthetized with urethane (1.5 g/kg, i.p.), and supplemental urethane doses of 0.3 g/kg were administered as needed. The skull surface was then exposed. The LFP was obtained from the dorsal hippocampal CA1, using a steel teflon-coated electrode (76.2 μm diameter, 777000, A-M System, USA). Multiple channel recordings were performed by a single-shank silicon probe with eight channels (site diameter: 20 µm, 300–500 kΩ, site interval: 200 µm; Plexon, USA) in the SLD (AP: −9.50 mm, ML: −1.30 mm, DV: −7.60 to −8.60 mm). The probes were pre-covered by DiI for histological analysis, and data were excluded if the recording channels exceeded the SLD. Besides, a micro-syringe (filled with drugs) connecting to a micro-drive (KD310, KD Scientific, USA) was implanted above the SLD (AP = −9.50 mm, ML = −1.50 mm, DV = −7.60 mm) for drug injection. A small screw was fixed above cerebellum as reference electrode. Electrodes were connected to a head-stage with a preamplifier and an omniplex neural data acquisition system (PlexControl 1.10, Plexon, USA) for data collection. Offline Sorter 3.3 (Plexon, USA), Neuroexplorer 4.1 (Nex Technologies, USA), and MATLAB 2014a (MathWorks, USA) were used for data analysis. Wide-band and field potential signals were digitally sampled at 40 and 1 kHz, respectively. After stable recordings in the hippocampus and SLD for at least 15 min, drugs were delivered at a speed of 0.06 μl/min. The baseline of the SLD neuronal activities and hippocampal LFP was calculated from a 600-s epoch before drug applications. The drug responses were evaluated from an equal-epoch when maximal effects were observed. After experiments, the brains were extracted for histological analysis.Spike sortingTo exact spikes in the SLD, the recorded wide-band signal from each channel within the SLD was high-pass filtered at 250 Hz. A threshold that was −4 times the standard deviation of the channel noise was set to detect spikes. A 1 ms refractory period was used to avoid detection of a subsequent spike during this window. Detected spike waveforms were then stored from −0.4 to 1.0 ms around the threshold crossing. Through using off-line sorter 3.3 (Plexon, USA), principal component (PC) analysis of these detected waveforms was performed and the first three linearly uncorrelated PCs were extracted. Then a clustering algorithm with standard expectation-maximation measures operating on the first three PCs was used to distinguish different single units. After the automated sorting, we manually checked the clustering through the three-dimensional plot of PCs. We verified spike times with auto-correlograms to assess the refractory period violations and used cross-correlograms to eliminate duplicates. To assess the response of single units to drugs, the histograms of firing frequency before and after drug application were generated in 1-s bins. A drug response was considered to be present when the change in the firing frequency was larger than twice the SD of the baseline65.Coincidental spiking analysisTo assess the involvement of GJ activities, we examined coincident spiking within the time scale of ±1 ms by cross-correlogram analysis between spike times of single units (bin size: 1 ms). The mean of the chance coincident was computed from 30 shuffled cross-correlograms. The probability of coincidental spiking was calculated by subtracting the probability of the mean chance from that of the raw cross-correlogram. This method controlled for chance effects in firing rate between pairs32. An interaction was then counted if a pairwise 1% threshold was crossed after the subtraction. Note that only pairs of single units between the adjacent recording sites were analyzed, as GJ-mediated interactions occurred between spatially confined recordings66. And because prevented by the 1 ms refractory period in spike-detecting process, pairs within the same recording sites were not included. The changes in the probability of coincident spiking within ±1 ms from these data were still sufficient to reflect the drug effects on GJs.LFP and phase-locking analysisTo extract hippocampal LFP, the field potential signals were filtered between 0.3 and 30 Hz. To determine the changes of LFP activities before and after drug applications, the power spectrogram (10 s sliding windows, sequentially shifted by 5 s increments) of the hippocampal LFP were generated using multi-taper methods (five Slepian taper functions, time bandwidth product of 3) with the Chronux data analysis toolbox for MATLAB (http://chronux.org/)67. The phase-locking analyses were next performed between SLD single-unit spikes and hippocampal oscillations (0.3–2.5 Hz). Briefly, the hippocampal LFP was band-pass filtered (0.3–2.5 Hz) with a zero-phase filter, and then the instantaneous 0.3–2.5 Hz phase was extracted with a Hilbert transform. Every spike was assigned to its corresponding phase. Rayleigh’s test for circular uniformity was applied to test the significance of phase-locking (P < 0.01). The locking strength was defined as the modulus of the average vector of all spike events corresponding to the 0.3–2.5 Hz phase.Optogenetics and chemogeneticsOrexin-Cre mice (male, 8–12 weeks old) were anesthetized with sodium pentobarbital (50 mg/kg, i.p.) and then fixed on a stereotaxic apparatus (RWD Life Science, China). For optogenetics, AAV-Ef1α-DIO-ChR2(H134R)-mCherry, AAV-CAG-FLEX-ArchT-GFP, and AAV-Ef1α-DIO-mCherry (150 nl, OBiO, China) were bilaterally injected into the LH (bregma, AP: −1.55 mm, ML: ±1.05 mm, DV: −5.25 mm) in different groups of orexin-Cre mice. Optical fibers (200 μm in diameter, NA of 0.37, Newdoon, China) were bilaterally implanted above the mice SLD (AP: −5.10 mm, ML: ±1.20 mm, DV: −3.75 mm) for light delivery68. For chemogenetics, AAV-retro-hSyn-DIO-hM4D (Gi)-mCherry or AAV-retro-hSyn-DIO-mCherry mixed with AAV-hSyn-EGFP-WPRE-pA (1:1, 23 nl, BrainVTA, China) were bilaterally injected into the SLD (bregma, AP: −5.10 mm, ML: ± 0.95 mm, DV: −4.25 mm). The infected region by AAV-hSyn-EGFP-WPRE-pA was used to evaluate the injection site. The viruses were stereotaxically injected using Nanoject II (Drummond Scientific, USA) via a silicate-glass micro-pipette (tip diameter: ~20 μm). Multiple 23 nl (13 nl/s) injections were made at 30-s intervals. After injections, the pipette was left in place for additional 10 min and then slowly retracted, to avoid potential damage to brain tissue.We crossed orexin-Cre and Ai27D mice to generate the Orexin-Cre;Ai27D offspring (10–14 days) for the validation of activating orexin signaling in SLD by light stimulation in slice physiology. Ai27D mice express the ChR2(H134R)/tdTomato fusion protein in a Cre-dependent manner69. The myelination in SLD prevented the identification of SLD neurons in adult mice in the brain slice17. But at this age (10–14 days), we found that ChR2 was already abundantly expressed in orexin neurons (Supplementary Fig. 6a). The patch-clamp recording procedure was similar to that of rats. After stale recordings of the baseline, an optical fiber (diameter: 200 μm) was used for light delivery, with the fiber tip positioned above the brain slices. The laser power was gradually increased to induce maximal responses of recorded neurons, and the final power used is estimated to be 5–20 mW. Light was delivered in pulse trains (473 nm, 5 ms, 20-Hz) for 1 s.After at least 3 weeks of virus injections, mice were implanted with EEG/EMG electrodes. The surgery, recording, and analysis procedures were similar to that of rats. To generate the power spectrum for the EEG signals (5 s sliding windows, sequentially shifted by 0.5 s increments), the multi-taper methods (five Slepian taper functions, time bandwidth product of 3) were used. Hypnograms were built in chemogenetics to calculate the total amount, episode duration, and number of each brain state.Light lasers (473 or 589 nm, Viashow, China) were controlled by a waveform generator (Master-8, AMPI, Israel). The laser intensity was calibrated to 10–15 mW at the tip by an optical power meter (PM100D, Thorlabs). After the EEG/EMG signals indicating NREM sleep appeared for 30 s, pulses (5 ms) of 473 nm light at 1, 5, and 20-Hz were delivered in the SLD for 1 in every 3 s, and the changes of EMG/EEG signals were recorded in orexinChR2 mice. At least 10 trials were conducted for each mice. The inter-trial interval was randomly chosen from a uniform distribution from 15 to 25 min. Optical activation during NREM sleep induced fast decrease in EEG power and a late-onset decrease in EMG amplitude, which can be disrupted by behavioral bouts of wakefulness before the light-off. We thus averaged EEG/EMG data from the trials till the light-off or disrupted by wakefulness. Control experiments were conducted in orexin-Cre mice that received AAV-Ef1α-DIO-mCherry injection in the LH.To test the role of SLD orexin signaling in REM sleep, we applied a closed-loop stimulation protocol70 in orexinArchT mice. The EEG signals were visually inspected on-line by an experienced experimenter. After the EEG/EMG signals indicating REM sleep appeared, the laser (473 nm: 20-Hz, 20 pulses every 3 s; 589 nm: a single pulse) was turned on with 50% probability and turned off only when the REM episode ended. An ~10 s delay existed for manual detection to ensure that the mice were in REM sleep. This allowed comparisons of the REM episode durations with and without laser stimulation within the same recording session. The experiments were conducted for 20 trials in each mice. Data were excluded if the duration of the optical stimulation was <10 s due to transitions out of REM sleep. The same criterion was also applied to the control group without laser stimulation. All optogenetics were conducted between ZT 2 and ZT 10.After recovery from surgery, orexinSLD-hM4D mice were allowed to habituate to the recording environment for at least 5 days. On the test day, the EEG/EMG recordings started at ZT 3 (11:00 a.m.). After 1-h stable recordings, CNO (3 mg/kg) or vehicle (0.9% NaCl) were injected intraperitoneally (i.p.) at ZT 4 (12:00 a.m.). CNO and vehicle injections were performed on 2 consecutive days in a random arrangement. After drug injections, brain states were monitored for the next 4 h through EEG/EMG recordings.To objectively evaluate effects on EEG/EMG performances during REM sleep, EEG/EMG signals during REM sleep (>20 s) and in 10-s epoch of the preceding NREM sleep were computed to extract EMG/EEG values for each REM sleep episodes. Note that the first 10 s of REM sleep was not included to make sure that signals from REM sleep were sampled from a stable state. The EEG theta power during REM sleep was reported to reflect the quality of EEG theta oscillation. Besides, the EMGREM/NREM ratio was reported to reflect the changes in EMG tone. This may eliminate the posture changes of the animal between NREM sleep and REM sleep that can induce biased EMG modifications35. After analyzing all epodes, the hot-map of EMG activities in each mouse were constructed.Fiber photometryWe injected AAV-CAG-FLEX-jGCaMP7b (150 nl, BrainVTA, China) bilaterally into the LH (bregma, AP: −1.55 mm, ML: ±1.05 mm, DV: −5.25 mm) of orexin-Cre mice, to selectively express jGCaMP7b in the orexin neurons. Six weeks after virus injections, mice were implanted with EEG–EMG electrodes and ceramic ferrules above the SLD (AP: −5.10 mm, ML: ±1.20 mm, DV: −3.80 mm). The surgery procedure was similar as above. To observe the REM sleep-related activity of the SLD orexin terminals, a fiber photometry system (Inperstudio Alpha 8.2, Inper, China) was used for recording jGCaMP signals. Data were analyzed by MATLAB 2014a (MathWorks, USA). During recording, an optical fiber (200 μm in diameter, NA of 0.37, Newdoon, China) was inserted into the ferrule. A 488 nm and a 405 nm laser beam were used for jGCaMP7b excitation and isosbestic wavelength, respectively. The power of 488-nm imaging light was set at 30-40 μW, and the 405-nm light power was adjusted to approximately match the jGCaMP fluorescence signals. The emitted signals were captured at 30 Hz with alternating pulses of 488 and 405-nm light, resulting in frame rates of 15 Hz for jGCaMP and the control signals. To synchronize fiber photometry and EEG/EMG recordings, a BNC cable carrying TTL pulses from the Inper system was connected to a digital input channel of the EEG/EMG recording system (Plexon, USA). The sampled signals were low-pass filtered at 2 Hz with a zero-phase filter from the two excitation wavelengths, 488 and 405 nm. The filtered 405 nm signal was aligned to the 488 nm signal through using a least-squares linear fit. ∆F/F was then calculated according to: (488 nm signal−fitted 405 nm signal)/(fitted 405 nm signal). We recorded signals from 10 REM sleep episodes (>20 s) from each tested animal. The preceding NREM sleep and the following wakefulness of the 10 REM sleep episodes were analyzed for comparison. The ∆F/F of 10 s from the steady state of each brain state were then averaged.Statistics and reproducibilityAll data were plotted and reported as mean ± SEM. Statistical analyses were performed in SPSS Statistics 22.0 (IBM, USA). Shapiro–Wilk test was first used to test the normality on each dataset. If the dataset passed the normality test, parametric tests (two-sided paired or unparied t-test) were used. Otherwise, non-parametric tests (Mann–Whitney rank-sum test or Wilcoxon signed-rank test) were used. One-way repeated-measures ANOVA followed by post hoc LSD comparison tests was used for tests in three or more groups. A threshold of P < 0.05 was accepted as statistically different. Significance levels of data are denoted as *P < 0.05 and **P < 0.01. P > 0.05 was considered non-significant and was denoted as n.s. Statistical methods used were all reported in the figure legends.Experiments were repeated independently in 6 rats for Fig. 1b, c, three times from 3 rats for Fig. 1e–g, in 7 rats for Fig. 1j, in 14 rats for Fig. 5b, in 9 orexinChR2 mice for Fig. 6b (left panel), in 27 orexin-cre mice for Fig. 6b (right panel), in 7 orexinSLD-jGCaMP7b mice for Figs. 7b, 8, and in orexinSLD-hM4D mice for Fig. 8b, c. In the Supplementary figures, experiments were repeated independently in 14 rats for Fig. 2a, three times from 3 rats for Fig. 2b, c, in 3 rats for Fig. 3b, in 16 SLD neurons for Fig. 5a, five times from 5 rats for Fig. 6a, three times from 3 orexin-Cre;Ai27D offsprings for Fig. 10b, in 9 orexinArchT-GFP mice for Fig. 11b, in 8 orexinSLD-hM4D mice for Fig 17b, and in 6 rats for Fig. 18a.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplemental InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Hypocretin",
"Hypocretin",
"Orexin",
"Orexin",
"REM sleep"
] | vigilance states serve brain functions depend on neurochemical signals1 hypothalamic neuropeptide orexin/hypocretin involved Orexin signaling provided by distributed fibers assigned to brain circuitries sleep/wakefulness states6,7 deficits wakefulness after orexin deficiency7 impaired quality rapid eye movement (REM) sleep observed sublaterodorsal tegmental nucleus necessary for REM neurochemical signals in sleep/wakefulness cycles14 Orexin fibers detected in SLD15 microinjection of orexin into SLD produces REM sleep-like muscle tone decrease in rats16 REM sleep relevance function of orexin signaling in SLD need.Orexin signaling excitatory in all brain regions including SLD17 SLD contains glutamatergic neurons elementary units REM sleep-related output12 firing activities SLD glutamatergic neurons carriers of output information for brain state activation muscle elevated firing activities in SLD glutamatergic neurons contribute to SLD output19membrane potential SLD neurons17 sub-threshold fluctuations gamma band spikelet activities firings24 correct orchestration elevated firings for output neuronal network SLD26–28 pontine wave coordinated firing pattern SLD17 with hippocampal theta activities during REM sleep27 loss of orexin signaling with impaired brain states abnormal fast-EEG activities muscle tone regulation wakefulness REM influences orexin on neuronal membrane network dynamics SLD brain state REM sleep investigated REM sleep relevance of SLD orexin signaling uncovered through retrograde tracing c-Fos screening SLD-projecting orexin neurons non-overlapping projecting to wake-promoting locus coeruleus global excitation of orexin in SLD identified interaction between orexin’s neuronal membrane dynamics electrical connections global synchronized excitation SLD network enhancement SLD output brain state activation muscle tone inhibition revealed through optogenetics fiber photometry substrates SLD orexin signaling employed during REM sleep prolong episodesreduced REM sleep insufficient muscle atonia EEG theta oscillation observed after inhibition SLD orexin signaling essential role for SLD orexin signaling in REM sleep stabilization neurons show REM identified REM sleep Cholera toxin subunit B-488 CTB-555 injected into SLD LC orexin signaling CTB-488 CTB-555 orexin-A+ neurons 7.1 ± 0.8% 15.2 ± 2.1% total orexin-A+ neurons 6 sporadically distributed across lateral hypothalamus innervated by orexin neurons OXSLD neurons non-overlapping with LC-projecting orexin) neurons (overlap 1.9 ± 0.3% orexin receptors (OXRs) bead-like orexin-A+ varicosities detected in rat SLD orexin transmission SLD OXRs expressed in vesicular glutamate transporter-1-positive glutamatergic neurons few OXRs/glutamate decarboxylase-67-positive GABAergic somas detectedglutamatergic not GABAergic neurons SLD region observed glutamate GABA antibodies Fig 2b OXSLD neurons REM sleep activation Schematic drawing CTB-555-488 injections retrograde tracing LH lateral hypothalamus superior cerebellar peduncle Mo5 trigeminal motor nucleus 4V 4th ventricle dorsomedial tegmental area caudal pontine reticular nucleus Scale bar 500 μm SLD-projecting CTB-488+/orexin-A LC-projecting-555 neurons Scale bar 50 μm 10 μm Percentage orexin neurons SLD LC 6 rats 3902 orexin Expression orexin-A+ varicosities SLD Scale bars 50 μm 10 μm Expression OXRs VGLUT1 GAD 67 SLD scale bars 20 μm sleep/wakefulness architecture 72-h RSD rat sacrificed c-Fos immunostaining after 2.5 h first REM sleep episode Control rats normal sleep/wakefulness cycle processesPercentage time REM sleep wakefulness NREM sleep 2.5 h preceding sacrifice control 4) RSD 7) rats c-Fos expression orexin-A+ OXSLD (cyan neurons (RSD C-Fos-positive OXSLD neurons medial lateral (n2) LH presented scale bars 20 μm Percentage c-Fos expression OXentirety (blue 0.817 P = 0.435) OXSLD neurons (cyan 4.839 9.221 × 10−4) control = 4 RSD 7 370 rats Data mean ± SEM. **P < 0.01. Source data assessed REM sleep relevance SLD orexin signaling Rats CTB-488 injection subjected 72-h REM sleep deprivation c-Fos screening REM sleep rebound REM sleep amount 7) increased 39.0 ± 2.8% compared control rats (9.1 ± 0.8% 4) normal sleep cycles orexin neurons (OXentirety no difference c-Fos expression control RSD groups (control 7.1 ± 1.3% RSD 6.0 ± 0.7 rats P 0.435) (Fig. 1j c-Fos expression increased OXSLD neurons (control 6.8 ± 2.8% 4 RSD 28.2 ± 2.9% 7 rats P = 9.221 × 10−4) OXSLD neurons increased activity REM sleep relevance SLD orexin signaling LH melanin-concentrating hormone neurons SLD similar c-Fos expression MCH population after REM sleep rebound Fig. 3a–d). abundant expression OXRs SLD REM neurons after REM rebound 4a).Orexin excites SLD patch-clamp recordings membrane potential adjusted silence firing blockers applied transmission orexin-A (100 nM) depolarized SLD neurons (baseline −63.3 ± 1.3 mV orexin-A −55.4 ± 1.4 mV n 32 P = 7.940 × 10−7) (Fig. 2a global excitation effect before orexin-A application carbenoxolone (CBX)-sensitive spikelet activities 21 of 32 (65.6%) neuronsSLD glutamatergic neurons electrically connected confirmed by neuronal gap junction protein Connexin-36 spread biocytin in patch-clamp recordings orexin-A (100 nM) elicited spikelet activities six neurons SLD neurons to 84.4% (27 of 32) (Fig. 2d).Fig 2Orexin elicits depolarization increases spikelets SLD neurons Orexin-A (100 depolarizes rat SLD neuron DNQX/MK-801/picrotoxin elicited consecutive spikelets spikelets emerged after membrane potential blocked by CBX (100 similar rise kinetics action potential Membrane potential SLD neurons before after orexin-A applications (n = 32 z = 4.937 P = × 10−7) Orexin-A increased percentage SLD neurons exhibiting spikelets (n = 32). e–g 30–300 nM orexin-A elicited inward current noise increase in rat SLD neurons (n = 8). h–j amplitudeP = 0.528) noise increase 0.192 = 0.854) orexin-elicited current unaffected by DNQX/MK/PTX in rat SLD neurons (n = 7) traces amplitude = 22.137 P = 2.124 × 10−4 P = 0.0122 = 2.218 × 10−3 = 0.0204 noise increase 8.721 P = 6.425 × 10−3 P = 0.0422 = 0.0249 P = 0.0387) current partially blocked by TCS OX2 29, abolished by TCS OX2 29/SB in rat SLD neurons (n = 6) Data mean ± SEM. *P < 0.05; **P < 0.01. Source data orexin-elicited global excitation analyzed −70 mV orexin-A (30–300 nM) elicited currents in 58 SLD neurons current dose-dependent independent of chemical synaptic transmission (control −25.3 ± 7.9 pA blockers −24.6 ± 7.6 pA n = 7 P = 0.528)-cell current noisy fluctuations. GJ currents orexin-elicited currents noise increased blocking transmission (control 59.9 ± 26.2% blockers 57.4 ± 13.6% 7 P 0.854) orexin currents mediated orexin-1 receptors. excitation orexin depends blockage GJ 100 μM CBX orexin-elicited current decreased 42.9% (control 26.6 ± 6.2 pA CBX 15.2 ± 4.3 pA P = 3.143 × 10−3) (Fig. 4 of 15 (26.7% neurons inward current abolished CBX. 3b interaction orexin-elicited excitation SLD GJ activities GJ blocker mefloquine confirmation biocytin-based immunostaining Pretreatment MEF (10 μM reduced orexin-elicited inward current 36.2% (control 24.3 ± 2.1 pA MEF 15.5 ± 3.4 pA 25 P 1.018 × 10−3). 3d 6 of 25 (24%) neurons no response orexin-A after MEF pretreatmentimmunostaining revealed orexin-A non-responsive neurons express OXRs (Fig. blocking GJ ~1/4 SLD neurons orexin-A non-responsive SLD orexin signaling targets ~3/4 SLD neurons determined substrates orexin-elicited effects through GJs global excitation TTX spikelet currents abolished (Fig. 3f orexin-A elicited currents neurons amplitude unaffected (control 20.7 ± 2.6 pA blockers 22.6 ± 2.4 pA n = 8 P = 0.0589) (Fig. 3h). results indicate orexin-elicited depolarization spreads through GJs global excitatory effects depolarization evoke firings spikelet activities orexin-elicited excitation originates from non-selective cationic conductance (NSCC) Fig. 7a–g).Fig. excitation orexin ~3/4 SLD neurons spreads via GJs orexin-A currents partially blocked by CBX abolished by CBX spikelet currents blocked by CBXorexin-elicited inward current reduced CBX rat SLD neurons (n = 15 Wilcoxon test z = 2.953 P = 3.143 × 10−3) abolished 4 of 15 (26.7%) neurons Biocytin immunostaining orexin-A non-responsive neuron mefloquine express OXRs another-A-responsive neuron MEF expressed OXRs orexin-elicited inward current reduced MEF neurons (control n = 58 MEF n = 25 Mann–Whitney rank-sum test z = 3.286 P = 1.018 × 10−3) current abolished 6 of 25 neurons orexin-A (100 nM) elicited current SLD neuron before after TTX (1 μM) plus DNQX/MK/PTX spikelet currents observed after orexin-A application normal ACSF TTX spikelet currents elicited orexin-A orexin-elicited noise increase attenuated TTX-test = 3.430 P = 0.0110), unaffected SLD neurons t = 2.253 P = 0.0589) (n = Data mean ± SEM.< 0.05 **P < 0.01. Source data.Orexin increases GJ conductance SLD orexin effects depolarization activities effect recorded connected SLD neurons rats current injections voltage responses. orexin-A (100 coupling coefficient increased 7 pairs controlcell1→cell2 0.046 ± 0.015 0.059 ± 0.017 0.042 ± 0.016 0.054 ± 0.019 P = 0.0280. coupling coefficient identify electrophysiological mechanisms GJ conductance input resistance calculated GJ conductance Orexin-A (100 increased conductance 7 pairs controlcell1→cell2 0.35 ± 0.09 ns 0 ± 0.13 ns 0.31 ± 0.08 0.43 ± 0.12 ns input resistances changed 7 pairs controlcell1: 143.0 ± 43.7 MΩ orexin-Acell1: 139.8 ± 38.9 MΩ P = 0.735 controlcell2 147.3 ± 44.8 orexin-Acell2 153.5 ± 45.8 MΩ P = 0.0910)4c, d). providing original power depolarization increase spikelet activities orexin promotes spread effects in SLD network increasing GJ conductance.Fig. 4Orexin increases GJ conductance in coupled SLD neurons (Left Double patch-clamp recordings in electrically connected SLD neurons in rat. (Right traces similar amplitude voltage change in n1 (Vinjection induced larger voltage response in n2 presence orexin-A (100 vice versa Group data orexin-A increased coupling coefficient from cell 1 to cell 2 (n = 7 pairs neurons). Group data input resistances of cell 1 and cell 2 0.0910) not affected by orexin-A (n = 7 pairs neurons). Group data orexin-A increased GJ conductance from cell 1 to cell 2 7 Data represent mean ± SEM. *P < 0.05. Source data provided as Source Data file.Orexin modulates synchronized excitation in SLD in examined modulations orexin on network dynamics SLD in vivo through multiple channel recordings in urethane-anesthetized rats (FigMicroinjections orexin-A (30 0.3 μl increased firing frequency 67 of 97 (69.1%) units (baseline 8.0 ± 0.9 Hz orexin-A 14.7 ± 1.6 Hz P = 1.120 × 10−12). GJ activities reflected coincidental spiking ±1 ms pairs 18 of 215 (8.4%) pairs spiking After orexin-A microinjection increased 28 of 215 (13%) coincident probability increased 1.5 ± 0.2% to 2.4 0.3% (P = 1.076 × 10−4) (Fig. orexin-elicited excitation GJ conductance drive synchronized excitation 5Orexin increases spiking Schematic drawing electrode arrays canulas SLD recordings urethane-anesthetized rats Orexin-A (30 0.3 μl microinjection increased firing frequency spike waveforms plot before after expanded Orexin-A increased firing frequency SLD units (n = 67/97 units z = 7.115 P = 1.120 × 10−12) cross-correlograms coincidental spiking ±1 ms before after orexin-A microinjectionsGJ interactions within ±1 ms or ±5 ms 2) focused on activities within ±1 ms significant interaction counted if 1% threshold crossed mean coincident computed from 30 cross-correlograms Orexin-A increased coincident spiking probability SLD pairs = 18/215 pairs = 3.881 P = 1.076 × 10−4) pairs between sites analyzed Orexin-A (30 μM, 0.3 μl microinjection increased firing frequency SLD neurons after pre-injection CBX (100 mM 39/72 units = 5.442 P = 5.255 × 10−8) Two cross-correlograms coincidental spiking within ±1 ms before after orexin-A microinjection CBX Orexin-A coincident spiking probability SLD pairs (n = 13/182 pairs z = 0.622 P = 0.542). Data mean ± SEM < 0.01. Source data CBX interfere SLD GJs CBX microinjection (100 mM 0.3 μl partially reduced coincident spiking probability SLDCBX pre-injection orexin spiking orexin-A increased spiking rate in 39 of 72 (54.2%) SLD units (CBX 5.2 ± 0.9 Hz/orexin-A 8.9 ± 1.5 Hz P = 5.255 × 10−8) (Fig. orexin-A microinjection after CBX change SLD pairs interactions (CBX 7.7% 14 of 182/orexin-A 7.7% 14 coincident probability 13 pairs CBX/orexin-A (CBX 2.5 ± 0.3%/orexin-A 2.4 ± 0.3% P = 0.542). suggest elevated GJ activities involved orexin-elicited excitation SLD network recorded hippocampal activities REM urethane-anesthesia hippocampal activities slow oscillations ~1 Hz microinjections orexin-A increased slow oscillations (0.3–2.5 Hz phase-locking strength SLD spiking orexin facilitates SLD output enhanced SLD-hippocampus correspondence effects blunted by pre-injection CBXdata suggest SLD orexin signaling promotes correspondence SLD target output orexin-elicited synchronized excitation SLD orexin signaling prolongs REM sleep episodeOptogenetics employed orexin-enhanced SLD output brain state regulation AAV-Ef1α-DIO-ChR2-mCherry injected LH orexin mice optical fibers implanted SLD (Fig. 6a mice optogenetic activation SLD orexin terminals excited SLD neurons release Fig functional expression light-sensitive opsins LH orexin neurons after virus infections confirmed immunostaining patch-clamp methods Fig 11a–e).Fig. 6Optogenetic activation SLD orexin signaling prolongs sleep Schematic drawing optogenetic activation SLD orexin terminals viral infections ChR2-mCherry+ LH orexin-A+ neurons optical-fiber location above trials EEG/EMG recordings during 20-Hz optical activation 3 SLD orexin signaling NREM sleep Purple wakefulness Heatmap EEG/EMG recordings trials recordings before wakefulness normalized averaged Blue shadow SEMoptical activation NREM EEG power 5-Hz t5 2.442 0.0585 10 Hz 3.939 0.0110 20 Hz 4.633 5.667 × 10−3) EMG amplitude 5 Hz t5 1.610 P 0.168 10 Hz 2.180 0.0811 20 Hz 5.056 6 orexinChR2 5 EEG t4 P 0.448 EMG (20 t4 0.506 P analyzed two paired t-test optical activation EEG theta power ratio 6 orexinChR2 mice 5 Hz 0.105 P 0.917 10 Hz 1.572 P 0.116 20 Hz 2.201 P 0.0277) closed-loop optogenetics REM sleep optical activation (20 Hz pulses 3 s EEG theta power t8 3.381 P 9.624 × 10−3) EMGREM/NREM ratio t8 2.126 P episode duration 2.368 P 0.0454 9 orexinChR2 analyzed two-sided paired t-test mean ± SEM < 0.05 < 0.01. dataNREM sleep 20-Hz light stimulation 20 pulses 3 s SLD orexin 2.4 ± 0.3 s EEG power 963.2 ± 196.6 μV2 light 574.8 ± 186.8 μV2 P 5.667 × 10−3) late 9.4 ± 1.7 s decrease EMG amplitude 6.6 ± 1.7 μV 5.6 ± 1.6 μV × 10−3) 6 mice EEG theta component increased before wakefulness (baseline 0.19 ± 0.02 light 0.23 ± 0.02 6 P 0.0277) activation brain state control group (20 Hz stimulation affect EEG power (baseline 836.9 ± 126.2 μV2 light 870.2 ± 142.9 μV2 P = 0.448) EMG amplitude 5.6 ± 1.5 5.5 ± 1.5 0.640) 5 activation brain decrease EMG amplitude NREM REM sleep transitions EEG power light stimulation theta oscillations REM induced latency REM sleep changed lightdata demonstrate SLD orexin signaling induce REM sleep transitions causes activation brain state increased EEG theta muscle tone decrease NREM sleep short awakening latency wakefulness affected transition probability increased s after light delivery Fig 13b–e). stimulation intensity (20 Hz for 20 s caused brain activation induce REM transitions transition probability from to wakefulness high Fig. examined brain state activation muscle tone decrease SLD orexin signaling REM sleep regulation Optogenetic activation SLD orexin terminals applied during REM sleep episodes 50% probability REM sleep performances length compared light-on light-off trials (Fig. 6h). light stimulation (473 nm, 20 Hz 20 pulses every 3 s consolidated EEG theta oscillations 295.3 72.6 321.1 77.0 P = 9.624 × 10−3) (Fig. EMGREM/NREM ratio used muscle atonia decreasing trend EMGREM/NREM ratio observed 0.94 ± 0.01 0.91 ± 0.02 P = 0.0662optical activation prolonged REM sleep duration 12.1% (light-off 65.4 ± 3.1 s light-on 73.3 ± 2.3 s 9 orexinChR2 mice P = 0.0454 (Fig. effects not observed mCherry suggest SLD orexin signaling prolonged REM sleep brain state activation muscle tone REM sleep stabilization SLD orexin signaling destabilizes REM contributions examined AAV-CAG-FLEX-jGCaMP7b injected LH orexin mice optical fibers above SLD (Fig. 7a jGCaMP fluorescence changes SLD orexin terminals observed REM sleep episodes (0.26 ± higher than (0.02 ± 0.01%) NREM sleep episodes lower (0.89 ± 0.20% wakefulness episodes (n = 7 mice F(2 = 16.758 P = 3.358 × 10−4 REM NREM = 2.481 × 10−4 (Fig. 7c activity SLD orexin signaling increased during REM sleep episodes jGCaMP signals elevated phasic tonic REM sleepperformed optogenetic inhibition SLD orexin signaling REM sleep AAV-CAG-FLEX-ArchT-GFP injected LH orexin mice optical fibers SLD inhibition validated immunostaining patch-clamp methods inhibition decreased theta oscillation power 405.0 61.0 μV2 375.1 58.3 μV2 P = 0.0325) REM sleep episode duration 76.2 4.4 s 62.3 4.1 s P = 0.0363 9 mice Changes EMGREM/NREM ratio 0.91 ± 0.02 0.90 ± 0.01 P 0.320 not detected suggest SLD orexin signaling REM sleep episodes inhibition shortens REM sleep bouts fiber photometry recordings SLD orexin terminalsviral infections jGCaMP7b LH orexin-A+ neurons 25 optical fiber location above SLD orexin terminals 100 μm). jGCaMP fluorescence traces SLD orexin terminals spectrogram EEG recordings raw EMG REM sleep NREM sleep wakefulness Averaged changes jGCaMP fluorescence REM sleep NREM sleep wakefulness (n = 7 mice ANOVA F(2, 12) = 16.758 P = 3.358 × 10−4 LSD comparison test REM vs. NREM P = 2.481 × 10−4 wake vs NREM P = 4.563 × 10−3 P = 0.0127) Schematic drawing optogenetic inhibition SLD orexin terminals closed-loop optogenetic inhibition REM sleep EEG/EMG signals inspected on-line Lasers turned on 50% after detection REM sleep turned off Effects optical inhibition SLD orexin terminals EEG theta power EMGREM/NREM ratio 9 mice). Effects inhibition episode duration 9 mice P = 0.0363). Data mean ± SEM.< 0.05 **P < 0.01. Source data silence SLD orexin signaling AAV-retro-hSyn-DIO-hM4D-mCherry AAV-hSyn-EGFP injected SLD orexin. effectiveness chemogenetic manipulation confirmed immunostaining patch-clamp methods Clozapine N-oxide 3 mg/kg vehicle (0.9% NaCl) injected intraperitoneally orexinSLD-hM4D mice ZT 4 (12:00 a brain states monitored 4 h CNO 17.0% reduction REM sleep 8.8 ± 1.2% 7.3 ± 1.2% 6.170 × wakefulness 37.3 ± 2.7% ± 2.7% NREM sleep 53.9 ± 2.7% ± 2.9% affected 8 orexinSLD-hM4D mice 8d reduced REM sleep episode duration 51.3 ± 5.1 s 40.0 ± 2.7 s P = 0.0269) episode number not altered 25.3 ± 3.5 ± 3.3 8 mice assessed regulation SLD orexin signaling REM sleep manipulation rats18a b). microinjection orexin-A TCS 1102 SLD SLD orexin signaling increased REM sleep episode duration Fig. 18a, b). data demonstrate role SLD orexin signaling REM sleep.Fig. 8Chemogenetic silencing SLD orexin signaling destabilizes REM sleep Schematic drawing chemogenetic silencing virus injection sites SLD 500 μm). expression hM4D (Gi)-mCherry) orexin-A+ neurons 20 μm Hypnogram sleep/wakefulness states 4 h after vehicle CNO injections orexinSLD-hM4D mouse REM sleep = 3.865 P = 6.170 × 10−3) wakefulness 1.355 P = 0.217) NREM sleep 0.527 P = 0.615) during 4-h after CNO injections (n = 8 episode duration 2.789 P = 0.0269) REM sleep 4-h injections EEG/EMG activities REM sleep NREM sleep wakefulness after vehicle CNO injections orexinSLD-hM4D mouse failure of muscle atonia in REM sleep after CNO injectionEMGREM/NREM activity map after vehicle CNO injections REM NREM-hM4D mice episodes sorted highest lowest EMGREM/NREM ratio x-axis episode duration 0 2π CNO injection disrupted muscle tone NREM to REM failure muscle atonia REM Changes EMGREM/NREM ratio = 3.619 P = × 10−3) EEG theta power 2.387 P 0.0484 after CNO vehicle injections 8 Data mean ± SEM *P < 0.05; **P < 0.01. Source data EEG failure of muscle atonia observed after CNO injections mice muscle tone decrease NREM to REM ratio abolished (vehicle: 93.4 ± 1.6%; CNO 99.0 ± 1.5% 8 P = 8.518 × 10−3) (Fig. abnormal behavior severe disruption muscle atonia after EMG activity map half narcoleptic patients suffer failure REM sleep theta oscillation power (vehicle: 409.3 ± 52.3 μV2 CNO 372.± 55.6 μV2 8 orexinSLD-hM4D mice P = 0.0484 (Fig. 8j REM sleep decreased after silencing SLD orexin signaling power distribution unchanged Fig. 19a b). effects not observed in control mice mCherry Fig. 20a disruption REM sleep after loss SLD orexin signaling may destabilize sleep episodes reduce REM sleep amount role SLD orexin signaling in REM sleep stabilization study demonstrated SLD orexin signaling in REM sleep stabilization REM sleep relevance identified in sub-cluster orexin neurons glutamatergic neuron-enriched SLD region observed electrophysiological modulations orexin on SLD neurons orexin increased GJ conductance SLD synchronized excitation SLD output enhanced brain state activation muscle tone decrease EEG theta-band activities muscle atonia stabilization REM sleep achieved.Orexin neurons innervate circuitries controlling sleep/wakefulness wakefulness-promoting structures orexin secondary effects sleep orexin evidence shown orexin innervations in sleep-promoting regions7orexin innervations observed in SLD previous reports present study6 found OXRs expressed on SLD glutamatergic neurons findings suggest direct effect SLD orexin signaling in REM sleep regulation characteristics OXSLD neurons consistent with REM sleep regulation sporadically distributed distinct from OXLC neurons adjacent to SLD findings support orexin neurons different groups projection suspected orexin neurons divided into distinct groups roles arousal reward seeking motor control39 REM sleep-related orexin neurons exist orexin release increased during REM-specific tagging REM sleep-related activation in minor (~7%) OXSLD class fiber photometry SLD orexin terminals increased activities during REM sleep observed present study OXSLD neurons activated against silent background OXentirety during REM sleep conflicting results previous extracellular recordings orexin neurons firing activities REM sleep intermediate low elevated activities OXSLD neurons REM sleep indicate essential roles on SLD neuronsactivation orexin neurons modulations reported in sleep/wakefulness structures suprachiasmatic nuclei tuberomammillary nucleus SLD GJs provided conditions for orexin modulation CBX-sensitive spikelet activities functional Cx-36 expression observed in SLD neurons extensive electrical connections SLD network orexin-elicited excitation could spread from OXRs+ SLD neurons ~3/4 population large-scale SLD network substrates unclear GJs reciprocal pathways for ionic current small molecules49 electrophysiological influences affect SLD neuronal activity propagation orexin-elicited excitation sub-threshold depolarization spikes consequences spread sustained depolarization via GJs may increase effect OXRs− SLD neurons recruited in excitation orexin reduced number orexin-excited neurons after blocking GJs by CBX unchanged excitation amplitude after blocking spike generation by TTX activation-evoked spikes spread source spike spikelet increased spikelet activities after orexin applications SLD spikelet activities basis for synchronized firings32orexin elicited activation SLD network firings synchronized significance of orexin-elicited synchronized excitation on SLD output demonstrated increased SLD firings promote REM sleep physiology ponto-geniculo-occipital waves hippocampal theta activities muscle atonia1 excitation orexin SLD output REM sleep regulation firing frequency SLD REM sleep19 synchrony of SLD firings enhancing orexin-elevated synchronized SLD firings phase-locked to hippocampal oscillations in anesthesia increased power disruption orexin excitation by CBX blunted effects orexin-elicited excitation firing synchrony involved in SLD output SLD output synchronized fundamental for recruiting coherence activities REM sleep brain orexin-elicited synchronized excitation to processes REM sleep corticohippocampal activation muscle atonia corticohippocampal theta oscillations reports anatomical basis for orexin-enhanced SLD output neural pathways include projections from SLD to intralaminar thalamus12 precoeruleus hippocampus51muscle atonia influenced through projections SLD to gigantocellular reticular nucleus spinal cord19 observed brain state activation increased theta muscle tone decrease after activation SLD orexin signaling in free-moving animals insufficiencies EEG theta oscillation muscle atonia observed after chemogenetic inhibition SLD orexin signaling alterations EEG failure muscle atonia during REM sleep in narcoleptic EEG oscillation atonia REM sleep orexin-enhanced SLD output involved in REM sleep stabilization REM episode duration increased by optogenetic activation decreased by inhibition SLD orexin signaling decrease in REM sleep amount after/chemogenetic silencing orexin-enhanced SLD output brain state activation/muscle tone decrease contribute to REM sleep stabilization maintenance REM sleep requires orexin signaling multiple projection sites orexin signaling study suggests orexin projections to SLD stabilizes REM sleep findings REM sleep symptoms wakefulness-maintaining deficits after loss central orexin signaling9orexin neuronal regulation physiological functions10 sleep/wakefulness loss of orexin signaling causes vigilance states total time unchanged58 orexin neuronal in wakefulness REM sleep stabilization brain function.MethodsAnimalsSprague-Dawley rats (Laboratory Animal Center Third Military Medical University) used for experiments Figs. 1–5. 1–9 18. SLD orexin signaling orexin-Cre mice Ai27D mice Jackson Laboratory used for experiments Figs. 6–8 Figs. 10–17 animal care procedures approved by Guide for Care Use Laboratory Animals Third Military Medical University housed in 12-h light/dark cycle lights-on 8:00 a.m.-off 8:00 p. environment temperature constant 22 ± 1 °C relative humidity between 40% and 60% Food water available ad libitum tracingMale rats (250–300 g) anesthetized with sodium pentobarbital (75 mg/kg fixed on stereotaxic apparatus (RWD Life Science, focused on SLD region from AP: −9.20 to −9.80 mm enriched glutamatergic neurons for REM sleepCTB-488 (69 nl 5 μg/μl injected SLD AP −9.50 mm ML −1.30 mm DV −8.20 mm Fig. CTB-555 (69 nl 5 μg/μl injected LC (AP −9.60 mm ML −1.40 DV −7.20 ipsilaterally SLD CTB-488 injection CTB tracers injected Nanoject II (Drummond Scientific silicate-glass micro-pipette ~20 2.3 nl injections 10-s intervals pipette left 10 min 2 weeks after injections rats sacrificed coronal brain slices LH SLD/LC collected immunohistological processes.REM sleep deprivationRats CTB-488 injection SLD subjected RSD inverted technique small-round platform 6.5 cm water (1 cm prevent REM sleep NREM food water 12-h light/dark cycles 72-h RSD removed original cages REM sleep recovery EEG EMG recordings sleep home cages control anesthetized sacrificed after 2.5 h first REM sleep episode coronal brain sections LH SLD collected immunostaining histological processesImmunohistochemistryRats mice anesthetized pentobarbital sodium perfused saline 4% paraformaldehyde Brains fixed 4% paraformaldehyde kept 30% sucrose/PBS 4 °C Coronal slices 10–20 μm made for immunohistochemical staining vibratome (CM 3050S Leica Slices incubated primary secondary antibodies primary antibodies goat anti-orexin-A sc-8070 rabbit anti-Fos Millipore mouse anti-c-Fos AB208942 rabbit anti-orexin 1/2R mouse anti-Cx-36-anti-NeuN MAB377 guinea-pig-anti-VGLUT1 mouse anti-GAD 67 MAB5406 rabbit anti-glutamate anti-GABA anti-MCH H-070-47 Phoenix Pharmaceuticals Images acquired LSM 800 (Carl Zeiss analyzed Zen software 2012 Zeiss Histological processes performed sites CTB/virus injections drug canula/electrode/optical fiber implantations Data excluded if locations not correctCell SLD orexin neurons six coronal sections CTB-injected rat LH 200 μm (between AP −2.30 −3.80 mm collected After orexin-A immunostaining counted total orexin CTB-488 CTB-555 ratio reported REM sleep activities OXSLD neurons c-Fos immunostaining applied rats CTB-488 injection RSD control rats counting method similar.EEG/EMG recordings drug injection free-moving ratsRats EEG/EMG recording chronically implanted electrodes Two frontal entorhinal cortex neck musculature Two drug canulas implanted bilaterally above SLD AP −9.50 mm ML −1.70 mm DV −7.80 mm drug injections electrodes soldered micro-pin connector affixed skull dental cement rats recover home cages 7 days acclimated recording cage 2 daysEEG/EMG signals recorded stored for analysis omniplex digitized at 1000 Hz band-pass filtered 1–30 Hz 20–100 monitored off analyzed by two investigators blind treatments Wakefulness desynchronized low-amplitude EEG elevated EMG NREM sleep synchronized high amplitude low frequency (1–4 Hz EEG lower EMG REM sleep consecutive theta (6–9 Hz) oscillations decreased EMG activity tests drugs injected into SLD micro-syringe micro-drive Scientific random 2 days interval 0.06 μl/min between ZT4 ZT5 (12:00 time spent state after drug injection effects returned to baseline after first post-injection hour.Whole-cell patch-clamp recordingCoronal brainstem slices SLD (300–400 μm prepared vibroslicer from rats 9–14 days superfused with 95% O2 5% CO2 oxygenated ACSF 125 NaCl 2.5 KCl 1.25 NaH2PO4 1.3 MgSO4 26 NaHCO3 2 CaCl2 20 d-glucose at 2 ml/min room temperaturerecordings SLD neurons borosilicate glass pipettes (3–5 MΩ internal solution 130 K-methylsulfate 5 KCl 2 MgCl2 10 HEPES 0.1 EGTA 2 Na2-ATP 0.2 Na2-GTP adjusted pH 7.25 1 M KOH). observations SLD region glutamatergic small (5–15 μm GABAergic neurons effects orexin on glutamatergic neurons soma diameter larger than 15 μm (membrane capacitance > 80 pF recorded used biocytin (0.5%) label neurons immunostaining neurons co-labeled by biocytin VGLUT1 Recordings current/voltage clamp mode Clampex 10.3 Multiclamp-700B amplifier analyzed by Clampfit 10.3 neurons excluded if series resistance exceeded 20 MΩ or changed 20% patch-clamp recordings connected SLD neurons neurons adjusted to −60 mV constant current orexin application depolarization membrane potential adjusted to baseline Coupling coefficient input resistance determined hyperpolarizing current stepsVoltage changes averaged three sweeps Input resistance calculated injected current voltage changes Coupling coefficient calculated ΔVCoupled/ΔVInjected cell linear fits R-square value higher 0.97.orexin modulates GJs calculated GJ conductance equation64\documentclass[12pt{minimal}\usepackage{amsmath}{wasysym{upgreek\oddsidemargin-69pt}{document}{R}\mathrm{c}}{cell}}1} =\mathrm{cc}}_{12}\mathrm{cell}}1}{cell}}2}}\mathrm{cc}}{12}}^2}{document}Rc,cell1=Rin,cell1×cc12/Rin,cell1×Rin,cell2−Rin Rc,cell1 GJ resistance cell1 to cell2 Rin,cell1 input resistance cell1 Rin,cell2 resistance cell2 coupling coefficient cell1 to cell2. GJ conductance cell1 to cell2 inverse Rc,cell1 channel recordingRats anesthetized urethane (1.5 g/kg supplemental urethane doses 0.3 g/kg administered needed skull surface exposedLFP dorsal hippocampal CA1 steel teflon-coated electrode (76.2 μm diameter 777000 A-M System channel recordings single-shank silicon probe eight channels diameter 20 μm 300–500 kΩ interval 200 μm Plexon SLD (AP −9.50 ML −1.30 mm DV −7.60 to −8.60 probes pre-covered by DiI histological analysis data excluded channels exceeded SLD micro-syringe drugs micro-drive implanted above SLD drug injection small screw above cerebellum reference electrode Electrodes connected head-stage preamplifier omniplex neural data system data collection Offline Sorter 3.3 Neuroexplorer 4.1 MATLAB 2014a data analysis Wide-band field potential signals digitally sampled at 40 1 kHz recordings 15 drugs delivered 0.06 μl/min baseline SLD neuronal activities hippocampal LFP calculated 600-s epoch before drug drug responses evaluated equal-epoch effects brains extracted for histological analysis wide-band signal high-pass filtered at 250 Hzthreshold −4 times standard deviation channel noise set detect spikes 1 ms refractory period avoid subsequent spike Detected spike waveforms stored −0.4 to 1.0 ms around threshold crossing off-line sorter 3.3 (Plexon,) analysis performed first three uncorrelated PCs extracted clustering algorithm standard expectation-maximation measures distinguish units manually checked clustering verified spike times with auto-correlograms used cross-correlograms eliminate duplicates response to drugs histograms firing frequency before after drug application generated in 1-s bins drug response when change firing frequency larger than twice SD baseline65.Coincidental spiking examined coincident spiking within time scale ms cross-correlogram analysis between spike times units mean computed from 30 shuffled cross-correlograms probability spiking calculated subtracting mean chance from raw cross-correlogram method controlled chance effects firing rate between interaction counted if pairwise 1% threshold crossed after subtraction only pairs between adjacent recording sites analyzed 1 ms refractory period pairs within same recording sites not includedchanges spiking within ±1 ms drug effects on GJs.LFP phase-locking hippocampal LFP field signals filtered between 0.3 30 Hz LFP before after drug applications power spectrogram (10 hippocampal LFP generated multi-taper methods Chronux data phase-locking analyses performed between SLD-unit spikes hippocampal oscillations (0.3–2.5 hippocampal LFP band-pass filtered (0.3–2.5 Hz zero-phase filter instantaneous 0.3–2.5 Hz phase extracted Hilbert transform spike assigned corresponding phase Rayleigh’s test circular uniformity phase-locking (P < locking strength defined average vector spike events 0.3–2.5 Hz phase.Optogenetics chemogeneticsOrexin-Cre mice 8–12 weeks anesthetized with sodium pentobarbital (50 mg/kg fixed stereotaxic apparatus (RWD Life Science optogenetics AAV-Ef1α-DIO-ChR2(H134R)-mCherry-CAG-FLEX-ArchT-GFP-Ef1α-DIO injected into LH in groups orexin-Cre miceOptical fibers (200 μm NA 0.37 Newdoon China implanted SLD −5.10 mm ML ±1.20 mm DV −3.75 mm light chemogenetics AAV-retro-hSyn-DIO-hM4D AAV-hSyn-EGFP-WPRE-pA (1:1 23 nl BrainVTA China injected SLD AP −5.10 mm ML 0.95 mm DV −4.25 infected region injection site viruses injected Nanoject II (Drummond Scientific USA silicate-glass micro-pipette ~20 23 nl (13 nl/s injections 30-s intervals pipette left 10 min retracted damage brain crossed orexin-Cre Ai27D mice offspring (10–14 days activating orexin signaling SLD light stimulation Ai27D mice express ChR2(H134R)/tdTomato fusion protein Cre-dependent myelination SLD prevented identification SLD neurons adult mice brain ChR2 expressed in orexin neurons patch-clamp recording procedure rats optical fiber 200 μm light delivery tip above brain sliceslaser power increased maximal responses final power 5–20 mW Light delivered pulse trains (473 nm, 5 ms 20-Hz 1 s 3 weeks virus injections mice implanted EEG/EMG electrodes surgery recording analysis similar rats power spectrum EEG (5 s windows 0.5 s multi methods Hypnograms chemogenetics amount duration brain state lasers (473 or 589 nm Viashow China controlled by waveform generator (Master-8, AMPI intensity calibrated to 10–15 mW optical power meter (PM100D EEG/EMG signals NREM sleep 30 s pulses (5 ms 473 nm light 1 5 20-Hz delivered 1 every 3 s changes EMG/EEG recorded in orexinChR2 mice 10 trials inter-trial interval randomly chosen 15 to 25 min Optical activation NREM sleep EEG power EMG amplitude averaged EEG/EMG data trials-off wakefulness Control experiments in orexin-Cre mice AAV-Ef1α-DIO-mCherry injection SLD orexin signaling REM sleep closed-loop stimulation in orexinArchT miceEEG signals inspected on-line by experimenter After REM sleep laser (473 nm 20 pulses every 3 s; 589 nm single pulse turned on 50% probability off when REM episode ~10 s delay for manual detection allowed comparisons REM durations with without laser stimulation experiments for 20 trials Data excluded if stimulation <10 s criterion to control group without laser stimulation optogenetics conducted between ZT 2 and ZT 10.After recovery surgery orexinSLD-hM4D mice habituate to recording environment for 5 days EEG/EMG recordings started at ZT 3 (11:00 a.m After 1-h recordings CNO (3 mg/kg) or vehicle (0.9% NaCl) injected intraperitoneally at ZT 4 (12:00 a CNO vehicle injections 2 consecutive days After brain states monitored 4 h through EEG/EMG recordings effects during REM sleep signals NREM sleep computed values first 10 s of REM sleep not included EEG theta power during REM oscillation EMGREM/NREM ratio changes EMG toneeliminate posture changes NREM REM sleep biased EMG analyzing epodes hot-map EMG activities mouse constructed injected AAV-CAG-FLEX-jGCaMP7b (150 BrainVTA China LH −1.55 ±1.05 −5.25 orexin-Cre mice express jGCaMP7b neurons Six weeks after injections mice implanted EEG–EMG electrodes ferrules above SLD −5.10.20 −3.80 surgery procedure REM sleep activity SLD orexin terminals fiber photometry system (Inperstudio Alpha 8.2 recording jGCaMP signals Data analyzed MATLAB 2014a optical fiber (200 μm inserted ferrule 488 nm 405 nm laser beam jGCaMP7b excitation wavelength 488-nm 30-40 μW 405-nm match jGCaMP fluorescence signals emitted signals captured 30 Hz alternating 488 405-nm light frame rates 15 Hz jGCaMP control signalsfiber photometry EEG/EMG recordings BNC cable TTL Inper connected digital input channel EEG/EMG (Plexon, sampled signals low-pass filtered 2 Hz zero-phase filter 405 nm filtered 405 nm signal aligned 488 nm least-squares linear fit ∆F/F calculated (488 405 nm)/ recorded signals 10 REM sleep episodes each animal preceding NREM sleep following wakefulness analyzed ∆F/F 10 s steady state averaged.Statistics data plotted reported mean ± SEM Statistical analyses SPSS Statistics 22.0 (IBM Shapiro–Wilk test normality passed parametric tests non-parametric tests (Mann–Whitney rank-sum test Wilcoxon signed-rank test One-way repeated-measures ANOVA post hoc LSD comparison tests three groups threshold P < 0.05 accepted statistically different Significance levels denoted *P < 0.05 **P < 0.01. P > 0.05 non-significant denoted n.s. Statistical methods reported figure legends.Experiments repeated 6 rats Fig. 1b c 3 1e–g 7 1j 149 orexinChR2 Fig 6b 27 orexin-cre 6b 7 orexinSLD-jGCaMP7b Figs 7b 8 orexinSLD-hM4D Fig 8b c Supplementary figures experiments repeated 14 rats Fig. 2a three 3 Fig 2b 3 3b 16 SLD neurons Fig 5a five 5 rats Fig 6a 3 orexin-Cre;Ai27D Fig 10b 9 orexinArchT-GFP Fig. 11b 8 orexinSLD-hM4D 17b 6 rats Fig. 18a Nature Research Reporting Summary Review Summary | 47.3 | 1.381432 |
10.1038/s41467-020-19306-7 | PMC7595090 | A conclusive account on how the brain translates audiovisual evidence into a rapid decision is still lacking. Here, using a neurally-informed modelling approach, the authors show that sounds amplify visual evidence later in the decision process, in line with higher-order multisensory effects. | Despite recent progress in understanding multisensory decision-making, a conclusive mechanistic account of how the brain translates the relevant evidence into a decision is lacking. Specifically, it remains unclear whether perceptual improvements during rapid multisensory decisions are best explained by sensory (i.e., ‘Early’) processing benefits or post-sensory (i.e., ‘Late’) changes in decision dynamics. Here, we employ a well-established visual object categorisation task in which early sensory and post-sensory decision evidence can be dissociated using multivariate pattern analysis of the electroencephalogram (EEG). We capitalize on these distinct neural components to identify when and how complementary auditory information influences the encoding of decision-relevant visual evidence in a multisensory context. We show that it is primarily the post-sensory, rather than the early sensory, EEG component amplitudes that are being amplified during rapid audiovisual decision-making. Using a neurally informed drift diffusion model we demonstrate that a multisensory behavioral improvement in accuracy arises from an enhanced quality of the relevant decision evidence, as captured by the post-sensory EEG component, consistent with the emergence of multisensory evidence in higher-order brain areas. | IntroductionIn everyday life, we often encounter situations that demand rapid decisions based on ambiguous sensory information. Consolidating the available evidence requires processing information presented in more than one sensory modality and exploiting this for multisensory decision-making1–4. For example, the decision to cross a street on a foggy morning will be based on a combination of visual evidence about hazy objects in your field of view and muffled sounds from various sources.The presence of complimentary audiovisual (AV) information can improve our ability to make perceptual decisions, when compared to visual (V) information alone5–8. While recent studies have provided a detailed picture of the emergence of different types of uni- and multisensory representations in the brain4,9–11, these studies have not provided a conclusive mechanistic account of how the brain encodes and ultimately translates the relevant sensory evidence into a decision2. Specifically, it remains unclear whether the perceptual improvements of multisensory decision-making are best explained by a benefit in the early encoding of sensory information, changes in the efficiency of post-sensory processes, such as the accumulation of evidence, or changes in the required amount of accumulated evidence before committing to a choice.These questions can be addressed within the general framework of sequential sampling models, such as the drift diffusion model (DDM), which posit that decisions are formed by a stochastic accumulation of evidence over time12–16. The DDM decomposes behavioral data into internal processes that reflect the rate of evidence accumulation (drift rate), the amount of evidence required to make a decision (starting point and decision boundaries corresponding to the different decision alternatives), and latencies induced by early stimulus encoding and response production (nondecision time; nDT). Importantly, different signatures of brain activity were shown to reflect distinct aspects of this mechanistic model, and therefore, single-trial measurements of the relevant brain activity can be used to constrain these models based on the underlying neural processes17–22.To date, few studies have exploited such neural markers of dissociable representations associated with sensory and decision evidence to arbitrate between different accounts of how multisensory evidence influences decisions in the human brain23. While some studies have performed careful comparisons between diffusion models and behavioral data24–27, they did not constrain these models against neural activity. Other studies, in contrast, tried to dissociate pre- and post-perceptual mechanisms by traditional activation mapping, but without a clear mechanistic model reflecting the decision process to support the interpretation of brain activity28–31. Furthermore, many studies focusing on visual judgements have considered only very simplistic stimulus features, such as contrast, salience, random-dot motion, or orientation7,32–35, which may be encoded locally at the level of early sensory processing, and hence may not generalize to complex real-life conditions. As a result, the neural mechanisms governing the influence of information from one modality on the decision-making process of another modality remain unknown.In this work, we employ a well-established visual object categorization task, in which early sensory evidence and post-sensory decision evidence can be properly dissociated based on electroencephalography (EEG) recordings. Specifically, using a face-vs-car categorization task, we have previously profiled two temporally distinct neural components that discriminate between the two stimulus categories: an early component, appearing ~170–200 ms poststimulus onset, and a late component, seen after 300–400 ms following the stimulus presentation36–41. In this previous work, we found that the late component was a better predictor of behavior than the early component, as it predicted changes in the rate of evidence accumulation in a DDM and shifted later in time with longer deliberation times36,42–44. Taken together, these findings established that the early component encodes the initial sensory evidence, while the late component encodes post-sensory decision evidence.Here, we capitalized on these distinct validated neural representations of visual information to identify the stage at which complimentary auditory information influences the encoding of decision-relevant visual evidence in a multisensory context. Based on recent results9–11,40, we hypothesized that using AV information to discriminate complex object categories—rather than more primitive visual features—would lead primarily to enhancements in the Late, as opposed to the Early, component, consistent with a post-sensory account. Importantly, by combining single-trial modelling and EEG data, we exploited the trial-by-trial variability in the strength of the Early and Late neural components in a neurally informed DDM to derive mechanistic insights into the specific role of these representations in decision-making with AV information. In short, we demonstrate in this work that multisensory behavioral improvements in accuracy arise from enhancements in the quality of post-sensory, rather than early sensory, decision evidence, consistent with the emergence of multisensory information in higher-order brain networks.ResultsBehavioral performanceWe collected behavioral and EEG data from 40 participants during a speeded face-vs-car categorization task (Fig. 1). Participants were required to identify a noisy image as being either a face or a car, presented in a randomly interleaved fashion either alone (visual trials; V trials) or simultaneously with distorted speech or car sounds (audiovisual trials; AV trials) for 50 ms. The amount of visual evidence (image phase coherence) varied consistently across participants over four levels, whereas the quality of the auditory evidence (distortion level) was set at a participant-specific level throughout the task. This level was determined by calculating the amount of distortion required for correct discrimination of 68–72% of trials during an auditory-only training session on the previous day (see “Methods”).Fig. 1Experimental paradigm.Schematic representation of the task design illustrating the order of presented events on the testing day. Participants had to categorize noisy representations of faces and cars. A brief stimulus, which was either an image (V) or a congruent image and sound (AV), was presented for 50 ms and followed by a delay period of up to 1500 ms during which participants were required to indicate their decision with a button press. Their response was followed by an intertrial interval (blank gray screen), jittered between 1000 and 1500 ms in duration, before the next stimulus was presented.We used generalized linear mixed-effects models (GLMMs) and post hoc likelihood-ratio (χ2) model comparisons to evaluate decision accuracy and response times (RTs; using a binomial logit and a gamma model, respectively), both as a function of modality (V/AV) and the levels of visual evidence (see “Methods”). We find that participants perform more accurately during AV than during V trials (χ2 = 30.02, df = 1, p < 0.001; Fig. 2a–c), as well as with increases in the amount of visual evidence (χ2 = 204.51, df = 3, p < 0.001; Fig. 2a, b). Our data shows no significant interactions between modality and the level of visual evidence (χ2 = 0.60, df = 1, p = 0.4376; χ2 = 0.01, df = 1, p = 0.9142; χ2 = 0.69, df = 1, p = 0.4047, respectively; sorted by increasing coherence level), and very strong evidence for an alternative model without interactions given our data from a Bayesian mixed model analysis (BF10 = 5030.05 ± 0.62%; see “Methods”). RTs increase somewhat with AV evidence (χ2 = 18.78, df = 1, p < 0.001) and decrease with the amount of visual evidence (χ2 = 48.71, df = 3, p = 0.0011; Fig. 2d, e). The RT model shows no significant interactions between modality and the level of visual evidence (χ2 = 1.43, df = 1, p = 0.2322; χ2 = 1.53, df = 1, p = 0.2156; χ2 = 0.004, df = 1, p = 0.9522), and very strong evidence for an alternative model without interactions given our data (BF10 = 1848.15 ± 0.53%).Fig. 2Behavioral performance.a, d Group averages of a decision accuracy (mean) and d response time (RT; median) across the four levels of visual evidence (phase coherence) and as a function of the visual (V; turquoise) and audiovisual (AV; red) trials. Shaded error bars indicate standard errors of the mean (SEM) and median across participants (n = 40), respectively. b, e Individual participant behavioral performance changes (AV–V trials) for b decision accuracy and e RT across the four levels of visual evidence (phase coherence). Group averages were computed across n = 40 independent participants. Solid black lines indicate group averages. c Robust bend correlation between individual participant decision accuracy, computed across all four levels of visual evidence (one value per participant), during V and AV trials. Bending (i.e., down-weighting) was performed on 20% of the data points in each direction. Dashed gray line represents equal performance during V and AV trials. Indicated correlation statistics were obtained from a robust bend correlation. f Standardized RTs of single trials. RTs were standardized on the participant level. Dots of the raincloud plot denote single-trial RTs108. Source data for this figure are provided as a Source data file.To ensure that our choice in the amount of participant-specific auditory evidence could not independently explain the overall improvements in accuracy during AV trials, we quantified the extent to which participants provided with higher levels of auditory evidence benefited more in AV trials. We find that the amount of auditory evidence explains only a minimal fraction of the variance in accuracy across participants (R2 = 0.01). In addition, we find that participants who perform well in V trials also perform well in AV trials (rbend(38) = 0.64, p < 0.0001), and the majority of participants (90%) show improved decision accuracy with additional auditory evidence (Fig. 2c).The nature of our experimental design (i.e., learning to associate short sounds with specific visual categories) may have encouraged some participants to adopt a strategy in which—in some trials—they only used the complementary auditory information when the visual evidence alone did not allow them to categorize the stimulus (rather than a consistent combination of both pieces of evidence). It follows that in this subset of trials, RTs would increase leading to a bimodality in the RT distribution, which could have been concealed by group differences. To rule this out, we first standardized each participant’s RTs (by z-scoring) and then tested the resulting distributions for bimodality, using a mixture of one or two exponentially modified Gaussian distributions (see “Methods”). We find that one exponential Gaussian fits our RT data best (BIC = −1064 vs −981 for V and BIC = −912 vs −688 for AV; Fig. 2f).Taken together, these results suggest that the combined influence of audiovisual information indeed contributes to an increased likelihood of making a correct decision (overall improvement M = 4.14%, standard deviation (SD) = 3.91%), but at the cost of response speed (overall slowing across visual coherence levels M = 33.1 ms, SD = 35.02 ms). The latter is likely due to additional time required for encoding the auditory stimulus (see “Discussion”).Time course of the impact of sounds on visual representationsNext, we analyzed the EEG data to identify temporally distinct components that discriminate between face and car stimulus categories. We performed this analysis separately for V and AV trials to characterize the extent to which the visual representations encoded in these components were affected by the additional auditory evidence. Specifically, for each participant separately, we performed a single-trial multivariate discriminant analysis45,46 to estimate linear spatial weights (i.e., spatial filters) that maximally discriminated face-vs-car trials within short predefined temporal windows, locked either to the onset of the stimulus or the response (see “Methods”).Applying the resulting spatial filters to single-trial data produces a measure of the discriminating component amplitudes (henceforth y), which can be used as an index of the quality of the visual evidence in each trial36,42,47,48. In other words, more extreme amplitudes, positive or negative, indicate more face or car evidence respectively, while values closer to zero indicate less evidence. To quantify the discriminator’s performance over time and identify the relevant components, we used the area under a receiver operating characteristic (ROC) curve (henceforth Az value) with a leave-one-trial-out cross-validation approach to control for overfitting.The discriminator’s performance as a function of stimulus-locked time reveals a broad window over which face-vs-car decoding was statistically reliable for both V and AV conditions (i.e., 180–600 ms poststimulus; Fig. 3a). To identify the number of relevant components in this range, we applied temporal clustering on the resulting scalp topographies as in previous work (see “Methods”)41. This procedure reveals the presence of two temporally distinct scalp representations with a transition point at 380 ms poststimulus for both the V and AV conditions. These spatial representations are consistent with our previously reported Early and Late components, with centrofrontal and bilateral occipitotemporal activations for the Early and a prominent centroparietal activation cluster for the Late component (Fig. 3a, top)36–38,42,43.Fig. 3Stimulus-locked face-vs-car discrimination analysis.a Mean discriminator performance (Az) during face-vs-car discrimination of stimulus-locked EEG data after a leave-one-trial-out cross-validation procedure, as a function of the visual (V; turquoise) and audiovisual (AV; red) conditions. Dashed black line represents the group average permutation threshold at p < 0.05. Shaded error bars indicate bootstrapped standard errors (SEM) across participants. Shaded gray vertical bars indicate Early and Late EEG component windows determined using temporal clustering of scalp topographies. The number of clusters and their extent was obtained via a k-means clustering algorithm that used a Euclidean distance metric and optimized k (for details, see “Methods”). These bars do not indicate statistical significance. Scalp topographies at representative time windows corresponding to the Early and Late EEG components, encoding sensory and post-sensory visual evidence, respectively. Colourbar represents normalized amplitude (µV). b Bootstrapped difference in discriminator performance (AV–V; thick black line) with 95% confidence intervals (2.5–97.5%; thin black lines). Horizontal thick black line above the x-axis in b illustrates significant temporal windows resulting from this one-sided permutation test (i.e., those in which the lower confidence interval is greater than zero (p < 0.025) with an added data-driven minimum requirement of three contiguous windows to correct for multiple comparisons; see “Methods”). c Fraction of participants showing discriminator performance (Az) in the same direction as the group-level mean. d Late EEG component amplitudes reflecting the relative separation across all face and car trials (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y$$\end{document}yfaces–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y$$\end{document}ycars) at the point of maximum Az separation between V and AV trials (see a, b). e Late EEG component amplitudes (y) as in d separated by decision accuracy. Reproducibility was ensured using an out of sample leave-one-trial-out cross-validation procedure per participant, thereby each participant becomes its own replication unit (for details, see “Methods”). Source data for this figure are provided as a Source data file.We then extracted participant-specific component latencies—for each condition separately—by identifying the time points leading to peak Az performance within each of the two windows identified by the clustering procedure. We allowed a 40 ms gap centered on the transition point to avoid potential multiplexing effects (i.e., we considered stimulus-locked windows 180–360 ms and 400–600 ms, for the Early and Late components, respectively). The mean peak times for the Early component for the V and AV conditions are 293 ms (SDV = 53.84 ms, SDAV = 57.52 ms). The mean peak times for the Late component are 500.25 ms (SDV = 40.92 ms) and 508.25 ms (SDAV = 40.76 ms) for the V and AV conditions, respectively. Our data shows no statistically significant latency differences across V and AV conditions (Early: two-sided paired t test, t(39) = 0.00, p = 1; Late: two-sided paired t test, t(39) = −1.09, p = 0.281).Moreover, the seemingly separate peaks in the discriminator performance (Fig. 3a) within the earlier temporal window (180–360 ms poststimulus) are likely due to interindividual differences in the onset of the Early component (i.e., differences in early sensory encoding). We tested this formally by demonstrating that the distributions of the Early component peak latencies are best approximated by a mixture of two—rather than one—Gaussians (BIC = 433 vs 438 for V and BIC = 440 vs 444 for AV) with means ~240 and ~330 ms, respectively, which coincide with the two peaks in the aforementioned window.Our main goal in this work is to determine when the decision-relevant category information is enhanced in the AV condition. We, therefore, sought to identify temporal windows during which the discriminator performance differs systematically between V and AV trials, and test the extent to which they overlap with the Early and/or Late components. Specifically, we used a temporal cluster-based permutation analysis, whereby for each temporal sample we created a bootstrap distribution of group-level Az difference scores (AV–V) and compared the bootstrapped median difference score against the lower bound of the estimated confidence interval of the distribution (supporting a significance level of p < 0.025)49,50. To form contiguous temporal clusters and avoid transient effects due to false positives, we required a data-driven minimum temporal cluster size of at least three significant samples (see “Methods”).This analysis shows only a single temporal cluster overlapping with the Late component (490–540 ms) over which the discriminator performance for AV trials is significantly improved compared to V trials (Fig. 3a, b). During this time, up to 78% of participants exhibited increases in the discriminator’s performance for AV trials, compared to only 60% of participants during the Early component (Fig. 3c). These findings indicate that the addition of auditory information in our task enhances primarily the quality of visual evidence (as reflected in our discriminator component amplitudes y) during post-sensory decision-related processing (Fig. 3d). This enhancement of the quality of decision evidence during AV trials is comparable across both correct and incorrect trials—with the quality of evidence being overall higher during correct compared to incorrect trials (Fig. 3e; BF10 = 10.27 ± 0.9%). Further, our data provide evidence against an interaction between AV benefit and decision accuracy (BF10 = 2.06 ± 1.3%).To rule out that these post-sensory enhancements are not driven by the speed (and hence the efficiency) with which participants encode the early sensory evidence, we performed two complementary analyses. Specifically, we correlated our Early component peak times with (1) the differential Late component amplitude effects (AV–V) and (2) the peak times of the Late component. In both analyses, we observe that the latency of the Early component has no significant leverage on the neural correlates of the Late component (rbend = −0.28, p = 0.071 and rbend = 0.22, p = 0.1752, respectively).In previous work, we showed that the Late component activity starts out as being stimulus-locked, but persists and becomes more robust near the response36,41,42, consistent with the notion that decision evidence reverberates and accumulates continuously until one commits to a choice. We therefore repeated the single-trial multivariate discrimination analysis on response-locked data. Importantly, this analysis also helps to rule out potential motor confounds associated with differences in RTs across V and AV trials by abolishing potential temporal lags near the time of the response.As with the stimulus-locked analysis, we compared the face-vs-car discriminator performance between V and AV trials. This analysis reveals a temporal cluster leading up to the eventual choice (−110 to −60 ms pre-response) during which discriminator performance is significantly enhanced for AV compared to V trials (Fig. 4a, b), with consistent effects (>70%) appearing across participants (Fig. 4c). Inspection of the resulting scalp maps during this period indicates that the spatial topographies, featuring a prominent centroparietal cluster, are consistent with the Late component seen in the stimulus-locked analysis (rbend(38) = 0.88 for V and 0.86 for AV; compare scalp topographies for LateS and LateR in Figs. 3a and 4a), in line with previous work19,41,44. These findings further highlight that it is primarily late, decision-related visual evidence that is being amplified during audiovisual object categorization (Fig. 4d). Similar to the stimulus-locked Late component, we find very strong evidence that this amplification in AV trials arises independently of the accuracy of the decision, while overall neural evidence is higher for correct trials (Fig. 4e; BF10 = 506.3 ± 0.73%). There is no interaction between this amplification of neural evidence and the accuracy of a decision (Fig. 4e; BF10 = 3.05 ± 0.86%).Fig. 4Response-locked face-vs-car discrimination analysis.a Mean discriminator performance (Az) during face-vs-car discrimination of response-locked EEG data after a leave-one-trial-out cross-validation procedure, as a function of the visual (V; turquoise) and audiovisual (AV; red) conditions. Dashed black line represents the group average permutation threshold at p < 0.05. Shaded error bars indicate bootstrapped standard errors (SEM) across participants. Shaded gray vertical bar indicates Late component window used in the selection of component amplitudes (y) shown in d. This window resulted from a temporal cluster-based bootstrap analysis performed on median Az difference scores (for details, see description of b and “Methods” section). Scalp topographies at representative time windows corresponding to the Late EEG component (indicated by dashed lines) encoding persistent post-sensory visual evidence up until the eventual commitment to choose. Colourbar represents normalized amplitude (µV). b Bootstrapped difference in discriminator performance (AV–V; thick black line) with 95% confidence intervals (2.5–97.5%; thin black lines). Gray shaded area in a and horizontal thick black line above the x-axis in b illustrate significant temporal windows resulting from this one-sided permutation test (i.e., those in which the lower confidence interval is greater than zero (p < 0.025) with an added data-driven minimum requirement of three contiguous windows to correct for multiple comparisons; see “Methods”). This procedure corrects for multiple comparisons. c Fraction of participants showing discriminator performance (Az) in the same direction as the group-level mean. d Late EEG component amplitudes reflecting the relative separation across face and car trials (yfaces–ycars) at the point of maximum Az separation between V and AV trials (see a, b). e Late EEG component amplitudes (y) as in d separated by decision accuracy. Reproducibility was ensured using an out of sample leave-one-trial-out cross-validation procedure per participant, whereby each participant becomes its own replication unit (for details, see “Methods”). Source data for this figure are provided as a Source data file.Neurally informed modelling explains multisensory effectsHaving characterized whether the added influence of auditory information enhances early sensory or late post-sensory visual representations, we then asked whether the identified single-trial neural responses are directly linked to improvements in behavior between V and AV trials. To this end, we employed a neurally informed variant of the traditional hierarchical drift diffusion model (HDDM; see “Methods”), a well-known psychological model for characterizing rapid decision-making14,51,52 to offer a mechanistic account of how the human brain translates the relevant evidence into a decision. In doing so, we directly constrained the model based on additional neural evidence, hence closing this persistent gap in the literature26,28,30.In brief, the traditional HDDM decomposes task performance (i.e., choice and RT) into internal components of processing representing the rate of evidence integration (drift rate, δ), the amount of evidence required to make a choice (decision boundary separation, α), the duration of other processes, such as stimulus encoding and response production (nDT), and a potential bias or prior information favouring one of the two choices (starting point, β). Ultimately, by comparing the obtained values of all these HDDM parameters across the V and AV trials, we could associate any behavioral differences resulting from the addition of auditory information (improved performance and longer RTs as in Fig. 2) to the constituent internal processes reflected by each of these parameters.Importantly, we deployed a neurally informed HDDM (nHDDM), whereby we incorporated single-trial EEG component amplitudes (y-values) into the parameter estimation (Fig. 5a). Specifically, we extracted single-trial discriminator amplitudes from participant-specific temporal windows (i.e., peak Az difference across AV–V) corresponding to both the Early and the Late stimulus-locked EEG components (see “Methods”). Since these values represent the amount of face or car evidence available for the decision (i.e., indexing the quality of the visual evidence as we demonstrated in previous work38,42,43), we used them to construct regressors for the drift rate parameter in the model (γEarly, γLate), based on the idea that evidence accumulation is faster when the neural evidence for one of the choices is higher. We therefore estimated these regression coefficients (γEarly, γLate) to directly assess the relationship between trial-to-trial variations in EEG component amplitudes and drift rate.Fig. 5Neurally informed cognitive modelling.a Graphical representation showing hierarchical estimation of nHDDM parameters. Round nodes represent continuous random variables and double-bordered nodes represent deterministic variables, defined in terms of other variables. Shaded nodes represent recorded or computed signals, i.e., single-trial behavioral data (choice, RT) and EEG component amplitudes (y’s). Parameters are modeled as random variables with inferred means μ and variances σ2. Plates denote that multiple random variables share the same parents and children. The outer plate is over sensory conditions (V, AV) and the inner plate is over participants (n). b Histogram and nHDDM model fits for RT distributions of car (left) and face (right) choices in the visual (V; top—in turquoise) and audiovisual (AV; bottom—in red) conditions. c, d Regression coefficients (γ) of the c Early and d Late EEG component amplitudes (y’s) in V (turquoise) and AV (red) conditions, as predictors of the drift rate (δ) of the nHDDM shown in a. Coefficients derived from nHDDM, including n = 40 independent participants and 28,540 trials. e Starting point values (β) estimated by the nHDDM for V (turquoise) and AV (red) conditions. f Boundary separation values (α) estimated by the nHDDM for V (turquoise) and AV (red) conditions. g Nondecision times (nDT) estimated by the nHDDM for V (turquoise) and AV (red) conditions. Dots indicate single-participant values and gray lines connect the population means in c–g. Source data for this figure are provided as a Source data file.We further hypothesized that the reliability of sensory information (as reflected by the visual coherence levels) would affect the rate of information integration. Thus, as per common practice14,53, we modeled a linear relationship between drift rate and coherence levels. To investigate whether this relationship is modulated by the Early and/or Late EEG component amplitudes, we tested three models where coherence scaled (a) yEarly, (b) yLate, or (c) both components. We find the best fit for the model where coherence scales yLate (deviance information criterions (DICs) = 767, 517, and 661, respectively), indicating that the modulation of the Late component with the reliability of available evidence predicts the rate of evidence accumulation. In other words, the best fitting model suggests that the effect of task difficulty on behavioral performance is captured by post-sensory mechanisms. Critically, this result dissociates the roles of the Early and Late EEG components in the decision-making process and is consistent with the role of the Late component in indexing the quality of the evidence entering the decision process (as has been shown in past work38,42,43), which, unlike early sensory encoding, is more closely associated with the accuracy of perceptual choices.When applying this model to the behavioral data, we obtain a good fit, accounting for most of the variance in the choice and RT data (average R2 = 0.94; Fig. 5b). Consistent with the functional role of the Early and Late EEG components in conveying sensory and post-sensory evidence, respectively, the within-participant single-trial discriminator amplitudes of both components are predictive of drift rate in both sensory conditions (Fig. 5c, d; γEarly and γLate significantly larger than zero for both V and AV, t(39) = 17.67, t(39) = 15.55 for γEarly(V), γEarly(AV), respectively, and t(39) = 11.92, t(39) = 16.02 for γLate(V), γLate(AV), respectively, all p values < 0.001; all two-sided paired t tests). Our results also show that the drift rates of correct trials are on average higher than those of incorrect trials (δcorrect = 0.27 ± 0.09, δincorrect = 0.14 ± 0.11 for face choices and δcorrect = −0.65 ± 0.11, δincorrect = −0.34 ± 0.10 for car choices—with the convention of positive signs for faces and negative signs for cars). This finding provides strong support that accuracy effects are effectively directly captured by our nHDDM.Crucially, the contribution of the Late but not the Early component (i.e., γLate, but not γEarly) is significantly higher in AV compared to V trials (Fig. 5d; two-sided paired t tests: t(39) = −0.6891, p = 0.4984 for γEarly, t(39) = −2.66, p = 0.011 for γLate). This is consistent with the increased discrimination power of the Late component in AV trials, and suggests that this component underpins the behavioral facilitation of evidence accumulation via post-sensory amplification of the available decision evidence (via the added auditory information) entering the decision process.We subsequently investigated the effect of the additional auditory information on the three other parameters of the nHDDM. Our data shows no reliable difference in starting point and boundary separation between the two sensory conditions (βV = 0.5475 ± 0.005, βAV = 0.5495 ± 0.005; two-sided paired t test: t(39) = −0.4964, p = 0.6224; Fig. 5e and αV = 1.13 ± 0.03, αAV = 0.12 ± 0.03; two-sided paired t test: t(39) = 0.8191, p = 0.4177; Fig. 5f), and significantly longer nDTs during AV trials (370 ± 9 ms for V vs 408 ± 10 ms for AV; two-sided paired t test: t(39) = 2.81, p = 0.0063; Fig. 5g). The latter result is likely related to longer stimulus encoding in AV trials, which may result from the extra time required to process the auditory stimulus (see “Discussion”). Notably, the average difference in RTs (33 ms) is comparable with the average nondecision difference between the two conditions (38 ms), which provides further evidence for the early sensory origins of the longer RTs in AV trials.Neurally informed modelling of choice biasesNext, we explored additional analyses that had no direct impact on the multisensory effects reported above, but nonetheless captured relevant idiosyncratic strategies in choice behavior. Specifically, we observe a starting point bias closer to face choices (expressed as a proportion of the boundary separation, two-sided paired t tests: t(39) = 9.06 for V and t(39) = 10.63 for AV, both p values < 0.001) and a higher drift rate for car choices, in both the V and AV conditions (δcar-V = −0.92 ± 0.14, δface-V = 0.15 ± 0.13, δcar-AV = −0.96 ± 0.14, δface-AV = 0.07 ± 0.12—positive (negative) signs indicate face (car) choices). To understand these results, we examined potential differences in the behavioral results between face and car choices. In particular, participants choose cars more often than faces (60 ± 8% of the V trials and 52 ± 8% of the AV trials were car choices), and are more accurate (84% vs 74% on average) in trials with a face choice. This accuracy effect is comparable across V and AV trials (two-sided paired t test: t(39) = −0.98, p = 0.3353; 83.2 ± 1.34% for V and 86.59 ± 1.14% for AV face choices, 72.84 ± 0.98% for V and 76.9 ± 0.91% for AV car choices).These results suggest that the bias in the starting point is likely driving the accuracy difference between face and car choices (i.e., more (fewer) errors when the starting point is further from (closer to) the correct boundary), while the higher number of car responses could be explained by a higher drift rate during car choices. The latter is also consistent with the RT distributions for face and car choices (Fig. 5b), where car RTs appear to have slightly longer tails but modes similar to face RTs. Importantly, these two parameter differences are consistent across the two conditions (V and AV), and thus they have no impact on the behavioral multisensory effects and their underlying neural mechanisms.Finally, given the timing of the Early component, we also considered whether it relates to the duration of sensory processing mechanisms captured in the nHDDM by nDTs. Thus, we also tested a model with yEarly as a regressor for nDT (rather than drift rate). However, this model demonstrates no relationship between yEarly and nDT (regression coefficients are not significantly different from 0 for both V and AV). Moreover, this model yields a poorer fit of the data compared to the model of choice (DIC = 1277 vs 517 for the chosen nHDDM). This finding is also consistent with the notion that increases in the nDTs observed in AV trials are likely driven by increases in the early encoding time of the added auditory information.Neurally informed model outperforms behaviorally constrained modelGiven that most previous studies in multisensory decision-making have fit DDMs only to behavioral data, it is worth asking whether the inclusion of EEG-derived regressors actually improves model performance and/or shapes the conclusion derived from the model. We formally compared the neurally inspired HDDM to a standard HDDM without neurally informed constraints. The traditional model yields a poorer trade-off between goodness-of-fit and complexity, as assessed by the DIC for model selection54, compared to its neurally informed counterpart (DICHDDM = 758 vs DICnHDDM = 517). In addition, the conclusions that would have been derived from such a poorer model contradict those reported above. For example, the conventional HDDM yields larger boundary separations for AV trials (two-sided paired t test: t(39) = −3.52, p = 0.0011), the nDTs estimated by this model are ~100–120 ms longer for both sensory conditions compared to the nHDDM (490 ± 10 ms for V and 509 ± 11 ms for AV), and the difference in average nDTs across conditions (19 ms) does not track the mean RT difference as closely as the nDTs estimated by the nHDDM. Hence, this poorer performing model constrained only on the behavioral data could lead to the misleading conclusion that the auditory information also affects the response caution (or individual speed-accuracy trade-off strategies via boundary adjustments). This supports the importance of constraining behavioral models with neural data and suggests that integrating neural information in these models can potentially enable a more accurate characterization of the behavioral effects, as well as a mechanistic interpretation of their neural correlates.DiscussionIn this work, we used multivariate single-trial EEG analysis and behavioral modelling to investigate the enhancement of visual perceptual decisions by complementary auditory information. We showed that significant improvements in behavioral performance in AV trials were accompanied primarily by enhancements in a late EEG component indexing decision-related processes36–38,42,43. In contrast, an earlier EEG component reflecting sensory (visual) evidence remained unaffected by the addition of auditory evidence. Using neurally informed cognitive modelling, we showed that these multisensory behavioral and neural benefits could be explained primarily by improvements in the rate of evidence accumulation in the decision process itself.The processing of multisensory information requires the coordination of multiple mechanisms serving bottom-up, top-down, and predictive coding processing3. The neural implementation of these mechanisms involves a distributed neural network, including primary sensory, parietal, and frontal brain areas that interact with each other to form and shape multisensory perception10,11.In this study, we were particularly interested in when multisensory information is combined to improve perceptual judgements. In the field of multisensory decision-making, there are two prominent theories that emphasize either the role of early or late integration of multisensory information, respectively2. The early integration hypothesis55–57 posits that sensory evidence is combined at the stage of early sensory encoding. This hypothesis is supported by evidence for direct pathways between early visual and auditory regions, or cross-modal influences on neural responses in early visual cortices55,58–62 and studies demonstrating benefits for the perception of simplistic visual stimuli, such as contrast7,63, motion direction5,32, and simple shape discrimination29 from acoustic information. However, the use of such simple stimuli may have specifically engaged only early sensory regions, hence providing a biased interpretation that does not generalize to more complex objects.In contrast, the late integration hypothesis, postulates that evidence from each sensory modality is instead processed separately during early sensory encoding, and is combined into a single source of evidence downstream, during the process of decision formation itself2. Support for this hypothesis comes from both animal and human experiments demonstrating that multisensory information is accumulated right up to the point of a decision, while processing of unisensory information occurs prior to the formation of a multisensory decision8,64. Similarly, recent neuroimaging work has provided new insights that flexible behavior can be accounted for by causal inference models65, with multisensory representations converging on higher-level parietal and prefrontal regions (e.g., inferior parietal sulcus, superior frontal gyrus) previously linked to the process of evidence accumulation9–11,66,67.Our findings appear to be at odds with the early integration hypothesis, since we found no evidence that the addition of auditory information had any impact on the encoding of early visual evidence, which remained comparable between V and AV trials. Instead, we offered support for post-sensory enhancements of decision evidence with the addition of auditory information that is most consistent with the late integration hypothesis. Importantly, these later visual representations are likely to reside in higher-order visual areas involved in object recognition and categorization (e.g., lateral occipital cortex), as we have shown previously38, consistent with the emergence of multisensory evidence only after early sensory encoding9. Specifically, the timing of these representations (starting after early sensory encoding and lasting until the commitment to choice) suggests that they unfold concurrently with the decision and provide the input to the process of evidence accumulation in prefrontal and parietal cortex66–70.A potential confounding factor for the late multisensory effects observed in our data could be differences in attention between V and AV conditions. If such unspecific effects were indeed at play, they would have likely impacted both early and late processing stages in a similar manner. Moreover, recent work suggests that the influences of multisensory information and attention operate independently across cortical columns71, and that attentional resources are largely shared across sensory modalities72. Hence, arguing against a competition between sensory modalities for attentional resources. Experimentally, we fully randomized all trials to ensure participants were equally likely to encounter (and expect) V or AV stimuli during each trial, thereby minimizing differences in attention between sensory conditions.Another potential point of departure from previous studies is that we observed increased RTs during multisensory trials. This finding likely stems from the auditory information being context-dependent and complementary to the visual information, rather than redundant as in previous work73. In other words, the sounds in our task are treated as supplementary evidence, instead of simply providing confirmation of the visual evidence, that require the deployment of additional processing resources, consistent with the observed increases in nDTs for multisensory trials in our modelling results.An alternative interpretation of these increases in nDT during multisensory trials provoked by a lack of latency differences in our Early (sensory) component across V and AV trials could be differences in motor preparation. This interpretation is highly unlikely because participants indicated their decision, using the same motor effectors and button press in both V and AV trials. Furthermore, our hypothesized increases in sensory encoding time due to the additional processing of auditory evidence during AV trials would not have been reflected in the latency of our Early component, which remained unaffected by the presence of the additional auditory evidence.Correspondingly, reaction time differences could arise due to the particular choice of sensory modalities and/or interindividual choice strategies employed by participants. For example, a recent study26 using time-varying multisensory information (visual and vestibular) reported faster, but slightly less accurate choices for multisensory compared to unisensory decisions. The authors modeled these results with a variant of the DDM model that incorporates the effects of time-varying information and sensory cue reliability and reported consistent drift rate improvements in the multisensory condition across participants. In other words, despite differences in the behavioral outcomes, their findings are in line with the increase in drift rate in AV trials we observed in the present study; that is, both studies suggest that multisensory information leads to faster accumulation of sensory evidence.Crucially in this work, we were able to characterize the neural underpinnings of the behavioral benefits obtained from the addition of auditory information. This contribution was made possible by the joint cognitive modelling of behavioral and neural data that linked the neural correlates of sensory and decision evidence with the internal processes involved in decision-making. Our neurally informed DDM indicated that the improvement in behavioral performance is derived mainly from enhanced post-sensory representations that modulate the rate of evidence accumulation. This result ran contrary to the behavioral-only version of a standard DDM, which attributed the longer RTs in AV trials to additional changes (increases) in the decision boundary and to a lesser extent to early encoding of the auditory stimulus.We suggest that the reason for this discrepancy is a less accurate account of the trial-by-trial variability in the decision dynamics (also indicated by the poorer fit of the single-trial data) than its neurally informed counterpart. In other words, the inclusion of the two well-characterized EEG components provided a more accurate account of the contributions of early sensory and post-sensory decision evidence to the formation of decision dynamics. Thus, this approach enabled the disambiguation of the internal processing stages that yielded such a behavioral benefit. Additional support for this claim is provided by the fact that the behavioral model yielded longer stimulus encoding times, whose difference across conditions did not track the difference in measured RTs equally well.Our findings suggest that constraining models of perceptual decision-making with neural data can provide key mechanistic insights, which may remain unobserved using behavioral modelling alone. This argument is in line with recent research, suggesting that the high complexity of decision-making models may yield neurally incompatible outcomes74–76. However, when informed by neural measurements, these models cannot only yield more reliable parameter estimates, but also shed light on the neural mechanisms underpinning behavioral effects43,51,77–81.It is worth noting that several previous studies have used DDMs to study multisensory decision-making. Some of these considered models in which the combination of multisensory information was explicitly hardwired, for example, to converge during sensory accumulation25,82,83. By doing so, these models can describe certain aspects of human behavior, but they cannot evaluate competing hypotheses about the locus of convergence. Other multisensory studies have combined behavioral modelling using DDMs and EEG, but did not use the neural data to constrain the behavioral model. Using such an approach, we have previously argued that the encoding of visual random-dot motion in early sensory regions is affected by acoustic motion5, speaking in favor of a sensory-level integration effect. However, this sensory-level effect was not validated using an EEG-inspired DDM model, as performed here.One explanation for these diverging findings is that the use of simpler stimuli, such as random-dot motion, may have biased the earlier study to a sensory-level effect, whereas multisensory information about more complex objects is instead combined at a post-sensory stage. This interpretation is supported by neuroimaging studies that have reported audiovisual interactions for complex stimuli mostly at longer poststimulus latencies or in high-level brain regions84–87.Another potentially important difference that might explain these divergent findings is the particular construction of the multisensory context across tasks. Many audiovisual integration studies use tasks in which there is a direct mapping between the source of the evidence across the two modalities, for instance, seeing a person’s mouth while producing speech (i.e., lip reading) to compensate for noisy acoustic information in a bar. In the present task, as in many real-world scenarios, however, this direct audiovisual mapping is not immediately available. In our earlier example, the decision to cross the street on a foggy morning will be based on hazy objects in your visual field together with street sounds that cannot immediately be matched to individual objects. In other words, the decision to step off the curb will be based on a broader audiovisual context and a higher-level conceptualization of the evidence, such as the presence of car-like objects and sounds signaling a busy street. This is a subtle but critical distinction in deciphering the mechanisms underlying audiovisual integration and reconciling discrepancies across different experimental designs.MethodsParticipantsWe estimated a minimum sample size of 35 participants by an a priori power analysis for a fixed linear multiple regression model with two predictors, a medium effect size of 0.5, an alpha of 0.05, and a power of 0.95. We therefore tested 40 participants (male = 18, female = 22; mean age = 23.85, SD = 5.47) on a speeded face-vs-car categorization task. All participants were right-handed with normal or corrected-to-normal vision and no self-reported history of neurological disorders. This study was approved by the ethics committee of the College of Science and Engineering at the University of Glasgow (CSE 300150102). All participants provided written informed consent prior to participation.StimuliWe used a set of 30 grayscale images—15 of faces and 15 of cars (image size 670 × 670 pixels, 8-bits per pixel)—adapted from our previous experiments36–38,42. The original face images were selected from the face database of the Max Planck Institute of Biological Cybernetics88 and car images were sourced from the Internet. Upon retrieval of the images, the background was removed and the image placed on uniform gray background.All images were equated for spatial frequency, contrast, and luminance, and had identical magnitude spectra (average magnitude spectrum of all images in the database). We manipulated the phase spectra of the images using the weighted mean phase technique89, whereby we changed the amount of visual evidence in the stimuli as characterized by their percentage phase coherence. To manipulate task difficulty, we used four levels of sensory visual evidence (27.5, 30, 32.5, and 35% phase coherence). These levels were based on our previous studies36–38,42, as they are known to result in performance spanning psychophysical threshold. Both image categories (i.e., faces and cars) contained an equal number of frontal and side views (up to ±45 degrees). We displayed all pictures on light gray background (RGB [128, 128, 128]), using the PsychoPy software90 (version 1.83.04) for a duration of 50 ms.Auditory sounds (15 car- and 15 face-related) were used in addition to the visually presented images in a random half of trials. Sounds were either human speech or car/street-related sounds obtained from online sources. No copyright restrictions were in place and modifications of the sound files were allowed. These were sampled at a rate of 22.05 kHz and stored as .wav files. In MATLAB (version 2015b, The MathWorks, 2015, Natick, Massachusetts), we added a 10 ms cosine on/off ramp to reduce the effects of sudden sound onsets and normalized all sounds by their SD. Subsequently, we reduced the intensity of these normalized sounds by lowering their amplitude by 80%. Sounds were embedded in Gaussian white noise, and the relative amplitude of the sounds and noise was manipulated to create 17 different levels of relative noise-to-signal ratios (ranging from 12.5 to 200% of noise relative to the lowered amplitude signal in increments of 12.5%). The resulting noisy speech- and car-related sounds were presented binaurally for 50 ms through Sennheiser stereo headphones HD 215.The stimulus display was controlled by a Dell 64 bit-based machine (16 GB RAM) with an NVIDIA Quadro K620 (Santa Clara, CA) graphics card running Windows Professional 7 or Linux-x86_64 and PsychoPy presentation software90. All images were presented on an Asus ROG Swift PG278Q monitor (resolution, 2560 × 1440 pixels; refresh rate set to 120 Hz). Participants were seated 75 cm from the stimulus display, and each image subtended ~11 × 11 degrees of visual angle.Behavioral taskWe employed an adapted audiovisual version of the widely used visual face-vs-car image categorization task36–38,42. This task required participants to decide whether they saw a face or a car embedded in the stimulus. Participants were asked to indicate their decision via button press on a standard keyboard as soon as they had formed a decision. The response deadline was set at 1.5 seconds. During half of the trials, participants were also given an additional auditory cue in the form of a brief noisy sound that was congruent with the picture’s content. Audiovisual face trials were accompanied by a human speech sound, whereas audiovisual car trials were accompanied by a car-related sound, such as squeaking tires or a slammed door. All stimuli were presented for 50 ms in the center of the screen and on AV trials to both ears. Participants were explicitly instructed to pay equal attention to and base their decision on information presented in both modalities in all trials. During AV trials, pictures and sounds were presented simultaneously. More specifically, we used four levels of visual noise, but only one participant-specific auditory difficulty level, obtained at perithreshold performance during an initial auditory training task (see below). Thereby, we accounted for interindividual differences in auditory perception, independently of visual image difficulty.This experimental paradigm required participants to attend a training and a testing session on two consecutive days at the same time of the day. On the first day (i.e., the training day), participants were asked to perform three separate simple categorization tasks to familiarize themselves with the task: (1) a visual image discrimination task (face-vs-car), (2) an auditory sound discrimination task (face/speech vs car/street sounds) and (3) an audiovisual discrimination task (face-vs-car). During the training session, participants also received visual feedback following each response (on all three tasks). Feedback was presented centrally for each of the possible three outcomes: ‘Correct’ written in green, ‘Incorrect’ written in red, and ‘Too slow’ written in blue (when participants exceeded the response deadline). Stimuli presentation duration for all stimuli and tasks was set to 50 ms for comparability between the training and testing days.During the visual training task, we used the same images and all four levels of visual evidence as on the second day (i.e., the testing day). During the auditory training task, we presented sounds to participants using eight different levels of relative noise-to-signal ratios (12.5%, 37.5%, 62.5%, 93.75%, 125%, 150%, 175%, and 200% of added noise). We estimated participant-specific noise levels supporting individual perithreshold performance (i.e., ~70% decision accuracy), including levels that might have fallen in between the eight noise-to-signal ratios used in this training task (from the larger set of 17; M = 140%, SD = 45%). We used these individual levels for the audiovisual training task and the main experiment. During the audiovisual training task, we used all images at the four levels of visual evidence together with the participant-specific perithreshold noise level determined above. This audiovisual training task mimicked the main task presented on the second (testing) day, with the addition that participants received feedback on their choices.Overall, on the training day, we presented 480 trials for each of the visual and auditory discrimination training tasks split into four blocks of 120 trials with a 60-second rest period between blocks. We presented 240 trials split into two blocks during the audiovisual training task. Taken together, all three training tasks lasted approximately 55 minutes on the first (training) day.On the second day, we collected behavioral and EEG data using randomly interleaved visual (unisensory) and audiovisual (multisensory) trials in a combined task (Fig. 1). Stimuli presentation employed the same task timings as outlined above on both days. Crucially, we did not provide any feedback to participants during testing. Using only one auditory noise level per participant on the testing day allowed us to evaluate the effects of auditory benefit at different levels of visual evidence. We presented 720 trials—divided equally between all stimulus categories (i.e., face/car, V/AV, and four levels of visual evidence)—in short blocks of 60 trials with 60-second breaks between blocks. The entire task on the testing day lasted approximately 45 minutes. EEG data were collected only during the testing day.Behavioral analysisOur main behavioral analysis quantified participants’ behavioral performance (i.e., decision accuracy and RTs) in the data collected during the testing day, using two separate GLMMs. GLMMs are superior to traditional repeated measures ANOVA analysis as their random effects structure better accounts for inter-participant variability, and allows for mixing of categorical and continuous variables91. Both models included all main effects and interactions of our two predictor variables, modality (V and AV) and visual evidence (27.5, 30, 32.5, and 35%), along with by-participant random slopes and random intercepts for the modality main effect. These random effects structure was justified by our design and adopted for reasons of parsimony. We employed post hoc likelihood-ratio (χ2) model comparisons to quantify the predictive power and significance of all main effects and interactions included in both GLMMs. These likelihood-ratio (χ2) model comparisons compared the full model (i.e., a model including all main effects, interactions, and random effects) to a reduced model, excluding the predictor or the set of predictors in question. Only results and statistics of the post hoc model comparisons are reported in the main results section. We performed these GLMM analyses using the lme4 package92 in RStudio93, specifying a binomial logit model in the family argument of the glmer function for decision accuracy, a binary dependent variable, and a gamma model for RT, a continuous dependent variable while selecting the bobyca optimizer. The predictor modality was entered in mean-centered form (deviation coding), whereas the predictor visual evidence (four levels) was entered using mean-centered backward difference coding. By using mean-centered coding schemes, we accounted for small imbalances in trial numbers between a predictor’s levels. Random correlations were excluded for both GLMMs.To quantify evidence for and against specific nonsignificant interaction effects in our two GLMMs, we complemented these models with model comparisons of two Bayesian linear mixed models (using the lmBF function and default priors of the BayesFactor package94 in RStudio93). We report a Bayes Factor indicating the available evidence for the alternative model (i.e., a reduced model omitting the interaction in question) given the data and a proportional error estimate for the Bayes Factor resulting from 500.000 Markov chain Monte Carlo (MCMC) iterations. All models in this study used single-trial data as input and are based on the following mean amount of trials per condition across participants: Vcar = 178.53, Vface = 178.78, AVcar = 178.33, and AVface = 177.88. The respective mean absolute deviation was Vcar = 1.74, Vface = 1.6, AVcar = 2.18, and AVface = 2.81 trials. Note that 180 trials per condition were originally presented to all participants.To quantify whether single-trial RT distributions are bimodal, we standardized (z-scored) RTs on the participant level before fitting a mixture of exponentially modified Gaussian (expGaussian) distributions (using maximum likelihood estimation) to the resulting RT distribution. Further, to formally rule out that our choice of participant-specific levels of auditory evidence could exclusively explain individual improvements in decision accuracy in AV trials, we correlated these measures across participants, using a robust bend correlation analysis95. Specifically, we evaluated whether the individual levels of auditory noise correlated with the difference in accuracy between V and AV trials (i.e., accuracyAV–accuracyV) across participants. As part of this correlation analysis, we computed the mean accuracy across all trials of each level of visual evidence and modality for each participant separately. In addition, to demonstrate that participants who performed well in V trials also performed well in AV trials, we complemented the above analysis by correlating decision accuracy (one value per participant calculated across visual coherence levels) between V and AV trials, using robust bend correlation analysis95.EEG data acquisition and preprocessingWe acquired continuous EEG data in a sound-attenuated and electrostatically shielded room from a 64-channel EEG amplifier system (BrainAmps MR-Plus, Brain Products GmbH, Germany) with Ag/AgCl scalp electrodes placed according to the international 10–20 system on an EasyCap (Brain Products GmbH, Germany). A chin electrode acted as ground and all channels were referenced to the left mastoid during recording. We adjusted the input impedance of all channels to <20 kΩ. The data were sampled at a rate of 1000 Hz and underwent online (hardware) filtering by a 0.0016–250 Hz analog band-pass filter. We used PsychoPy90 and Brain Vision Recorder (BVR; version 1.10, Brain Products GmbH, Germany) to record trial-specific information, including experimental event codes and button responses simultaneously with the EEG data. These data were collected and stored for offline analysis in MATLAB. Offline data preprocessing included applying a software-based fourth-order butterworth band-pass filter with cutoff frequencies between 0.5 and 40 Hz. To avoid phase-related distortions, we applied these filters noncausally (using MATLAB filtfilt). Finally, the EEG data were re-referenced to the average of all channels.We removed eye movement artifacts, such as blinks and saccades, using data from an eye movement calibration task completed by participants before the main task on the testing day. During this calibration task, participants were instructed to blink repeatedly upon the appearance of a black fixation cross on light gray background in the center of the screen before making several lateral and horizontal saccades according to the location of the fixation cross on the screen. Using principal component analysis, we identified linear EEG sensor weights associated with eye movement artifacts, which were then projected onto the broadband data from the main task and subtracted out45. We excluded all trials from all subsequent analyses where participants exceeded the RT limit of 1.5 s, indicated a response within <300 ms after onset of the stimulus or the EEG signal exceeded a maximum amplitude of 150 μV during the trial (0.8%, 0.06%, and 0.03% of all trials across participants, respectively).EEG data analysisWe employed a linear multivariate single-trial discriminant analysis of stimulus- and response-locked EEG data45,46 to identify early sensory and late decision-related EEG components discriminating between face and car trials as in previous work (e.g., refs. 36–38,42). We performed this analysis separately for V and AV trials to independently identify the sensor signals discriminating the relevant visual evidence in each sensory modality condition, and allow direct comparisons between them in terms of overall discrimination performance. All single trials were included in all discriminant analyses.Specifically, we identified a projection of the multichannel EEG signal, xi, where i = (1…N trials), within short time windows (i.e., a sliding window approach) that maximally discriminated between face and car trials (i.e., V discrimination: face-vs-car; AV discrimination: face/speech vs car/street sounds). All time windows had a width of 60 ms and onset intervals every 10 ms. These windows were centered on and shifted from −100 to 1000 ms relative to stimulus onset on stimulus-locked data and from −600 to 500 ms relative to the response button press on response-locked data. Specifically, a 64-channel spatial weighting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w(\tau )$$\end{document}w(τ) was learned by means of logistic regression45 that achieved maximal discrimination within each time window, arriving at the one-dimensional projection yi(τ), for each trial i and a given window τ:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y(\tau ) = w(\tau )^Tx\left( \tau \right) = \mathop {\sum }\limits_{i = 1}^D w_i(\tau )x_i(\tau ).$$\end{document}y(τ)=w(τ)Txτ=∑i=1Dwi(τ)xi(τ).Here, T refers to the transpose operator and D refers to the number of EEG sensors. In separating the two stimulus categories, the discriminator was designed to map component amplitudes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_i(\tau )$$\end{document}yi(τ) for face and car trials, to positive and negative values, respectively. These values are a weighted reflection of all available neural evidence with respect to the specific decision task (face-vs-car) that we asked participants to perform. By performing separate analyses for each modality condition, any unspecific effects present across trials—such as memory recollection or attention—would not be contributing to the estimation of the relevant classification weights separating face from car trials, and would effectively be subtracted out96.To quantify the performance of our discriminator for each time window, we used the area under a ROC curve97, referred to as an Az value, combined with a leave-one-trial-out cross-validation procedure to control for overfitting36–38,42. Specifically, for every iteration, we used N-1 trials to estimate a spatial filter w, which was then applied to the left out trial to obtain out-of-sample discriminant component amplitudes (y) and compute the Az value. Moreover, we determined significance thresholds for the discriminator performance (rather than assuming an Az of 0.5 as chance performance) using a bootstrap analysis, whereby face and car labels were randomized and submitted to a separate leave-one-trial-out test. This randomization procedure was repeated 1000 times, producing a probability distribution for Az, which we used as reference to estimate the Az value leading to a significance level of p < 0.05 (participant average Azsig = 0.57). Note that this EEG analysis pipeline was performed on individual participants such that each participant became their own replication unit98.Finally, the linearity of our model allowed us to compute scalp projections of our discriminating components resulting from Eq. (1) by estimating a forward model as:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a\left( \tau \right) = \frac{{x(\tau )y(\tau )}}{{y\left( \tau \right)^Ty(\tau )}},$$\end{document}aτ=x(τ)y(τ)yτTy(τ),where the EEG data (x) and discriminating components (y) are now in a matrix and vector notation, respectively, for convenience. Such forward models can be displayed as scalp topographies and interpreted, as the coupling between the observed EEG and the discriminating component amplitudes (i.e., vector α reflects the electrical coupling of the discriminating component y that explains most of the activity in x). These forward models were computed separately for V and AV face-vs-car discriminant analyses.Optimizing number of distinct spatiotemporal componentsDuring periods of sustained significant discriminating activity, we used the forward model estimates resulting from Eq. (2) above to identify temporal transitions between different components based on differences in scalp distribution, which are typically suggestive of changes in the underlying cortical sources. Specifically, we used a k-means clustering algorithm using a Euclidean distance metric on the intensities of vector a(τ) for the entire time range of interest and optimized k (i.e., the number of different time windows with similar scalp topographies) using silhouette values99, as implemented in MATLAB’s evalclusters function. Our results remained robust regardless of the choice of criterion (e.g., Silhouette, CalinskiHarabasz, etc.), the distance metric used for clustering, and the conditions it was applied to (i.e., V or AV trials). We used the resulting temporal components in all relevant EEG analyses.Temporal cluster-based bootstrap analysisTo quantify if and when the discriminator performance differed between V and AV trials, we used a percentile bootstrap technique for comparing the group-level Az difference between two dependent samples49. Specifically, on a sample-by-sample basis, we created a distribution of shuffled Az difference scores (i.e., AV–V) across participants (drawing with replacement). We repeated this shuffling procedure 1000 times for each sample, whereby we created a random bootstrap distribution of median Az difference scores from every iteration. We computed the median of this bootstrap distribution for a given sample along with the 95% confidence interval (2.5–97.5%) of the resulting distribution of median difference scores. To test whether our bootstrapped median difference was significantly different from zero for each sample, we compared it against the lower bound of the estimated confidence interval (i.e., at the 2.5% threshold; p < 0.025).To form contiguous temporal clusters and avoid transient effects due to false positives, we required a minimum temporal cluster size of at least three significant samples. This threshold was determined by means of the 95th percentile of a data-driven null distribution of maximum cluster sizes. Specifically, while in the analysis above the relationship between adjacent samples was preserved, here, we first applied a permutation procedure (i.e., shuffling temporal samples without replacement) to abolish the relationship across temporal samples, while keeping the relative difference between V and AV Az values unchanged, for each sample and participant. We generated the null distribution of maximum cluster sizes by computing and storing the maximum number of adjacent significant samples of the largest cluster for each of the 1000 iterations. Similar to the analysis on our original data, we performed this analysis on the discriminator performance (Az) of both stimulus- and response-locked data (Figs. 3b and 4b, respectively), which yielded an average of at least three significant samples. This procedure corrects for multiple comparisons and is comparable to the temporal cluster-based nonparametric permutation test reported in ref. 100.This entire procedure determined the extent of the temporal window used for the selection of the single-trial EEG component amplitudes (y-values), which we subsequently included in the neurally informed drift diffusion modelling analysis (see section below). Since our sample-based procedure was performed directly on discriminator accuracy (Az), these times effectively represent the centers of the original discrimination windows, which consider data from a wider window (60 ms). To capture the full extent of these windows, we extended the selection window by 30 ms on either side of the significant clusters determined by our temporal cluster-based bootstrap analysis.To ensure that neural effects were also reliably traceable in individual participants without group-level averages masking variability, we also computed the proportion of participants who demonstrated a participant-level effect in line with the general group-level effect per sample (that is, higher AV Az value for a given sample, see Figs. 3c and 4c). We performed these statistical analyses building on MATLAB code obtained from the Figshare and GitHub repositories associated with refs. 49,50.To quantify evidence for and against the effects of sensory modality and decision accuracy on the subject-specific component amplitudes (y) based on trial accuracy (Figs. 3e and 4e), we computed two additional Bayesian linear mixed models analyses (using the generalTestBF function and default priors of the BayesFactor package94 in RStudio93). Here, splitting trials into correct and incorrect responses, we report a Bayes Factor indicating the available evidence for the alternative model (i.e., a larger model including the predictor in question compared to a reduced model omitting the predictor in question) given the data. Note, when examining an interaction between sensory modality and decision accuracy the alternative model is the one omitting the interaction term.Lastly, we performed a robust bend correlation analysis95 to test the topographical consistency between the late stimulus-locked and response-locked components. We computed the average scalp map (i.e., forward models) across participants at the point of peak discrimination for the two components (500 ms poststimulus and 100 ms prestimulus, respectively) and assessed their similarity by computing their correlation. We also used two similar bend correlation analyses to test the extent to which the individual onset times in the Early component predicted (1) the difference in the stimulus-locked Late component discriminator amplitudes across the two modalities (AV vs V), and (2) the peak time of the stimulus-locked Late component.Hierarchical drift diffusion modelling of behavioral dataWe fit the participants’ performance (i.e., face or car choice and RT) with an HDDM101. Similar to the traditional DDM, the HDDM assumes a stochastic accumulation of sensory evidence over time toward one of two decision boundaries representing the two choices (face or car). The model returns estimates of internal components of processing, such as the rate of evidence accumulation (drift rate), the distance between decision boundaries controlling the amount of evidence required for a decision (decision boundary), a possible bias toward one of the two choices (starting point) and the duration of nDT processes, which include stimulus encoding and response production.The HDDM uses MCMC sampling to iteratively adjust the above parameters to maximize the summed log-likelihood of the predicted mean RT and accuracy. The DDM parameters were estimated in a hierarchical Bayesian framework, in which prior distributions of the model parameters were updated on the basis of the likelihood of the data given the model, to yield posterior distributions52,101,102. The use of Bayesian analysis, and specifically the HDDM, has several benefits relative to traditional DDM analysis. First and foremost, this framework supports the use of other variables as regressors of the model parameters to assess relations of the parameters with other physiological or behavioral data51,76,78–80,103. This property of the HDDM allowed us to establish the link between the EEG components and the aspects of the decision-making process they are implicated in. Second, posterior distributions directly convey the uncertainty associated with parameter estimates102,104. Third, the Bayesian hierarchical framework has been shown to be especially effective when the number of observations is low105. Fourth, within this hierarchical framework, all observers in a dataset are assumed to be drawn from a group, which yields more stable parameter estimates for individual participants52.To implement the hierarchical DDM, we used the Wiener module101 in JAGS106, via the Matjags interface in MATLAB to estimate posterior distributions. For each trial, the likelihood of accuracy and RT was assessed by providing the Wiener first-passage time distribution with the three model parameters (boundary separation, nDT, and drift rate). Parameters were drawn from group-level Gaussian distributions. The means and SDs of these group-level distributions had non-informative normally or uniformly distributed priors. Specifically, all SD priors were uniformly distributed U(0.01, 2). The mean priors of nDT, boundary separation, and starting point were also uniformly distributed: nDT ~ U(0.01, 1), α ~ U(0.01, 3), β ~ U(0.1, 0.9). The priors of all the regression coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _i$$\end{document}γi means were Gaussians N(0, 3). For each model, we ran five separate Markov chains with 5500 samples each; the first 500 were discarded (as “burn-in”) and the rest were subsampled (“thinned”) by a factor of 50 following the conventional approach to MCMC sampling, whereby initial samples are likely to be unreliable due to the selection of a random starting point, and neighboring samples are likely to be highly correlated101. The remaining samples constituted the probability distributions of each estimated parameter from which individual parameter estimates were computed.To ensure convergence of the chains, we computed the Gelman–Rubin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat R$$\end{document}R^ statistic (which compares within-chain and between-chain variance), and verified that all group-level parameters had an \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat R$$\end{document}R^ close to 1 and always lower than 1.03. For comparison between models, we used the DIC, a measure widely used for fit assessment and comparison of hierarchical models54. DIC selects the model that achieves the best trade-off between goodness-of-fit and model complexity. Lower DIC values favor models with the highest likelihood and least degrees of freedom.We first estimated a nHDDM that used our EEG discrimination analysis to inform the fitting of the behavioral data. In this model, we input the single-trial RTs and (face or car) choices of all 40 participants, and hypothesized that the evidence accumulation rate during each trial would be dependent on the amount of neural evidence about face or car choice in that trial. Therefore, as part of the model fitting within the HDDM framework, we used the single-trial EEG measures of the face-vs-car discrimination analysis as regressors of the drift rate (δ) as follows:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta = \gamma _0 + \gamma _1 * y_{{\mathrm{Early}}}^{\mathrm{s}} + \gamma _2 * y_{{\mathrm{Late}}}^{\mathrm{s}} * C,$$\end{document}δ=γ0+γ1*yEarlys+γ2*yLates*C,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{{\mathrm{Early}}}^{\mathrm{s}}$$\end{document}yEarlys and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{{\mathrm{Late}}}^{\mathrm{s}}$$\end{document}yLates are the single-trial discriminator amplitudes of participant-specific stimulus-locked Early EEG components (individual peak Az across V and AV in the time range 180–360 ms poststimulus) and Late EEG components (individual peak Az difference between AV and V in the time range established in Fig. 3b; 490–540 ms (expanded further by 30 ms on either side to account for the resulting Az values being obtained with 60 ms training windows centered on these times)), respectively. The coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _i$$\end{document}γi weight the slope of the drift rate by the values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{{\mathrm{Early}}}^{\mathrm{s}}$$\end{document}yEarlys and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_{{\mathrm{Late}}}^{\mathrm{s}}$$\end{document}yLates of that specific trial, with an intercept \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _0$$\end{document}γ0. Here, we estimated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _i$$\end{document}γi’s for each participant and sensory condition. C is the phase coherence level of the image presented in each trial. This value represents the quality of visual evidence available in each trial and has been shown to be proportional to the amplitude of the Late component38,42,43. Hence, by using these regression coefficients, we were able to test the influences of each of the two identified components on the drift rate in both sensory conditions78. Posterior probability densities of each regression coefficient were estimated using the sampling procedure described above. Significantly positive (negative) effects were determined when >99.9% of the posterior density was higher (lower) than 0. To test the significance of differences between the two sensory conditions (V vs AV), we performed a “hierarchical” t test comparing the population-level distributions of the parameters under consideration. This statistical testing has been shown to reduce biases induced by ignoring the hierarchical structure of the model (and testing at the participant level) and to actually yield conservative effect sizes107.For comparison, we also estimated a HDDM without including any neural correlates. We fit the HDDM to RT distributions for face and car choices conditioned on the sensory condition (V or AV) for each trial. Overall drift rate, boundary separation, starting point, and nDT were estimated for each individual participant and were dependent on the sensory condition. As per common practice, we assumed that evidence strength affected the drift rate; thus, we modeled a linear relationship between drift rate and coherence level81.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Neuroscience",
"Psychology"
] | life situations rapid decisions ambiguous sensory information Consolidating evidence requires processing information more sensory modality multisensory decision-making1–4 decision to cross street foggy morning based on visual evidence hazy objects muffled sounds audiovisual (AV information perceptual decisions compared to visual information alone5–8 recent studies uni- multisensory representations in brain4,9–11 conclusive mechanistic account brain encodes translates sensory evidence into decision2. unclear perceptual improvements of multisensory decision-making explained by benefit early encoding sensory information efficiency post-sensory processes required amount accumulated evidence before choice questions addressed within sequential sampling models drift diffusion model (DDM), decisions formed by stochastic accumulation evidence over time12–16 DDM decomposes behavioral data into processes rate evidence accumulation (drift amount evidence required decision boundaries latencies early stimulus encoding response production different signatures brain activity reflect distinct aspects mechanistic model single-trial measurements brain activity can constrain models neural processes17–22 few studies exploited neural markers of dissociable representations sensory decision evidence arbitrate between multisensory evidence decisions human brain23studies between diffusion models behavioral data24–27 constrain models against neural activity Other studies tried pre- post-perceptual mechanisms by activation mapping without clear mechanistic model decision brain studies on visual judgements considered simplistic stimulus features contrast salience random motion encoded locally early sensory processing generalize to complex real-life conditions neural mechanisms governing influence information one modality on decision-making unknown work visual object categorization task early sensory post-sensory decision evidence dissociated based electroencephalography (EEG recordings face-vs-car categorization task profiled two distinct neural components between stimulus categories early ~170–200 ms poststimulus onset late component after 300–400 ms late component better predictor behavior predicted changes in evidence accumulation longer deliberation times36 early component encodes initial sensory evidence late component encodes post-sensory decision evidence capitalized on neural representations visual to identify auditory information influences encoding decision-relevant visual evidence in multisensory contextrecent results9–11 hypothesized using AV information discriminate complex object primitive visual enhancements Late Early component consistent post-sensory account combining single-trial modelling EEG data exploited variability Early Late neural components DDM derive insights decision-making with AV information multisensory behavioral improvements accuracy arise from enhancements post-sensory decision evidence consistent with emergence multisensory information higher-order brain networks.ResultsBehavioral collected behavioral EEG data from 40 participants speeded face-vs-car categorization task (Fig. 1) Participants identify noisy image face or car randomly alone or with distorted speech car sounds for 50 ms visual evidence coherence varied across participants over four levels auditory evidence (distortion level set participant-specific level level determined calculating distortion required for correct discrimination of 68–72% trials auditory-only training session previous. 1Experimental paradigm task design order events Participants categorize noisy representations faces carsstimulus image or sound presented 50 ms delay 1500 ms decision button press response followed intertrial interval gray 1000 1500 ms before next stimulus used linear mixed-effects models post hoc likelihood-ratio model comparisons decision accuracy response times binomial modality visual evidence participants perform accurately during AV V trials (χ2 = 30.02 df = 1 p < 0.001 visual evidence (χ2 = 204.51 df = 3 p < 0.001 no interactions between modality visual evidence (χ2 = 0.60 df = 1 p = 0.4376 = 0.01 0.69 strong evidence for alternative model without interactions Bayesian mixed model analysis = 5030.05 ± 0.62% RTs increase with AV evidence (χ2 = 18.78 df = 1 p < 0.001) decrease visual evidence (χ2 = 48.71 df = 3 p = 0.0011 RT model shows no interactions between modality visual evidence (χ2 = 1.43 df = 1 p = 0.2322; χ2 = 1.53 df = 1 = 0.2156 0.004strong evidence alternative model without interactions data (BF10 = 1848.15 ± 0.53%).Fig. 2Behavioral performance Group averages decision accuracy response time four levels visual evidence visual audiovisual trials Shaded error bars standard errors mean median participants (n = Individual participant behavioral performance changes (AV–V trials decision accuracy RT four levels Group averages computed n = 40 participants Solid black lines indicate group averages Robust bend correlation individual decision accuracy levels visual evidence V AV trials Bending-weighting 20% data points each Dashed gray line equal performance V AV trials correlation statistics robust bend correlation Standardized RTs single trials participant level raincloud plot single-trial Source data choice participant-specific auditory evidence improvements accuracy AV trials quantified participants higher levels auditory evidence benefited AV trials auditory evidence explains minimal variance accuracy (R2 = 0.01). participants well V trials perform well AV trials (rbend(38) = 0.64 p < 0.0001) majority (90% show improved decision accuracy with additional auditory evidence (Fig. experimental designshort sounds visual categories encouraged participants strategy complementary auditory information when visual evidence stimulus RTs bimodality in RT distribution concealed by group differences standardized RTs tested distributions for bimodality one or two exponentially modified Gaussian distributions one exponential Gaussian fits RT data best (BIC = −1064 vs −981 for V BIC = −912 vs −688 for AV Fig. results suggest combined influence audiovisual information contributes to increased likelihood correct decision (overall improvement M = 4.14% standard deviation (SD) = cost response speed slowing across visual coherence levels M = 33.1 ms SD = 35.02 likely due to additional time encoding auditory stimulus impact of sounds on visual analyzed EEG data distinct components between face car stimulus categories performed analysis separately for V AV trials visual representations additional auditory evidence performed single-trial multivariate discriminant estimate linear spatial weightsspatial filters discriminated face-vs-car trials temporal windows locked to onset stimulus or response spatial filters to single-trial data discriminating component amplitudes index quality visual evidence extreme amplitudes indicate more face car evidence closer zero indicate less evidence performance relevant components used area under receiver operating characteristic (ROC) curve Az leave-one-trial-out cross-validation approach control overfitting performance stimulus-locked time reveals broad window face-vs-car decoding reliable for V AV conditions 180–600 ms poststimulus Fig. identify relevant components applied temporal clustering on scalp topographies two temporally distinct scalp representations transition point at 380 ms poststimulus V AV conditions representations consistent with Early Late components centrofrontal bilateral occipitotemporal activations Early centroparietal activation cluster Late.. 3Stimulus-locked face-vs-car discrimination analysis discriminator performance (Az during-car discrimination stimulus-locked EEG data after leave-one-trial-out cross-validation function visual audiovisual conditionsblack line group average permutation threshold p < 0.05. Shaded error bars standard errors gray bars indicate Early Late EEG windows temporal clustering scalp topographies clusters extent obtained k-means clustering algorithm Euclidean distance k statistical significance Scalp topographies time windows Early Late EEG sensory post-sensory visual evidence Colourbar normalized amplitude difference discriminator performance 95% confidence intervals (2.5–97.5% Horizontal thick black line x-axis illustrates temporal windows one-sided permutation test lower confidence interval greater zero (p < 0.025) requirement three contiguous windows comparisons Fraction participants discriminator performance same direction group-level meanLate EEG amplitudes reflecting separation across face car trials\documentclass[12pt{minimal{amsmath{wasysym at maximum Az separation between V and AV trials Late EEG amplitudes separated by decision accuracy Reproducibility ensured using out of sample leave-one-trial-out cross-validation procedure per participant each participant replication unit “Methods”). Source data provided Source data file extracted participant-specific latencies—for identifying time points leading to peak Az performance within two windows clustering allowed 40 ms gap on transition point to avoid multiplexing effects considered stimulus-locked windows 180–360 ms and 400–600 ms for Early and Late components mean peak times for Early component V and AV conditions are 293 ms (SDV = 53.84 ms, SDAV = 57.52 ms). mean peak times for Late component are 500.25 ms (SDV = 40.92 ms 508.25 ms 40.76 ms V AV conditions data shows no significant latency differences across V AV (Early t(39) = 0.00 p = 1; Late t(39) = −1.09 p = separate peaks in discriminator performance (Fig. 3a) earlier temporal window (180–360 ms poststimulus due to interindividual differences Early component sensory Early component peak latencies approximated by (BIC = 433 vs 438 for V BIC 440 vs 444 for AV means ~240 ~330 ms coincide with two peaks goal determine decision-relevant category information enhanced in AV condition identify temporal windows discriminator performance differs between V AV trials test with Early Late components used temporal cluster-based permutation analysis bootstrap distribution group-level difference scores compared median against lower bound estimated confidence interval p < required minimum temporal cluster size three samples analysis shows single temporal cluster overlapping with Late component (490–540 ms) discriminator performance for AV trials improved compared to V (Fig78% participants increases performance AV trials compared 60% Early component (Fig. 3c). addition auditory information enhances quality visual evidence discriminator component amplitudes post-sensory decision processing enhancement quality decision evidence AV trials comparable correct incorrect quality higher during correct incorrect trials (Fig. 3e; BF10 = 10.27 ± 0.9%). data evidence against interaction between AV benefit decision accuracy (BF10 = 2.06 ± 1.3%) post-sensory enhancements not driven by speed early sensory evidence performed analyses correlated Early component peak times with differential Late component amplitude effects peak times Late latency Early component no significant leverage on neural correlates Late component (rbend = −0.28, p = 0.071 rbend = 0.22, p = 0.1752 Late component activity stimulus-locked persists robust near decision evidence reverberates until choice repeated single-trial multivariate discrimination analysis on response-locked data out potential motor confounds RTs V AV trials temporal lags near response compared face-vs-car discriminator performance between V AV trialsanalysis reveals temporal cluster choice (−110 to −60 ms pre-response discriminator performance enhanced for AV trials (Fig. 4a consistent effects (>70% across participants (Fig. scalp maps spatial topographies centroparietal cluster consistent with Late component stimulus-locked analysis (rbend(38) = 0.88 for V 0.86 AV LateS Figs. 3a late decision-related visual evidence amplified during audiovisual object categorization (Fig. amplification in AV trials independently accuracy neural evidence higher for correct trials (Fig. 4e = 506.3 ± 0.73%). no interaction between accuracy decision = 3.05 ± 0.86%).Fig. 4Response-locked face-vs-car discrimination analysis discriminator performance during discrimination after leave-one-trial-out cross-validation visual audiovisual conditions Dashed black line group average permutation threshold p < 0.05. Shaded error bars indicate standard errors gray vertical bar indicates Late component window selection component amplitudeswindow temporal cluster bootstrap analysis median Az difference scores Scalp topographies time windows Late EEG component dashed lines post-sensory visual evidence commitment choose Colourbar normalized amplitude (μV). Bootstrapped difference discriminator performance (AV–V 95% confidence intervals (2.5–97.5% Gray shaded area horizontal black line x-axis temporal windows one-sided permutation test lower confidence interval greater than zero (p < 0.025) three contiguous windows correct multiple comparisons procedure corrects multiple comparisons Fraction participants discriminator performance (Az same direction group-level mean Late EEG amplitudes separation face car trials maximum Az separation between V AV trials Late EEG amplitudes separated by decision accuracy Reproducibility ensured out of sample leave-one-trial-out cross-validation procedure Source data file.Neurally informed modelling explains multisensory auditory information early post-sensory visual representations asked single-trial neural responses linked to improvements behavior between V AV trialsemployed neurally informed variant traditional hierarchical drift diffusion model (HDDM rapid decision-making14 account human brain translates evidence into decision constrained model additional neural evidence closing gap in literature26 traditional HDDM decomposes task performance choice RT) into internal components processing representing rate evidence integration (drift rate amount evidence required choice (decision boundary separation duration of processes potential bias information favouring one choices (starting point β). comparing values HDDM parameters across V and AV trials associate behavioral differences from addition auditory information (improved performance longer RTs to internal processes reflected deployed neurally informed HDDM incorporated single-trial EEG component amplitudes into parameter estimation (Fig. extracted single-trial discriminator amplitudes from participant-specific temporal windows AV–V Early and Late stimulus-locked EEG components values represent amount evidence for decision quality visual evidence used construct regressors for drift rate parameter model (γEarly, γLate), evidence accumulation faster when neural evidence for one higherestimated regression coefficients (γEarly γLate relationship trial-to-trial variations EEG amplitudes drift rate.Fig. 5Neurally informed cognitive modelling Graphical hierarchical estimation nHDDM parameters Round nodes represent continuous random variables double-bordered deterministic variables Shaded nodes represent recorded signals single-trial behavioral data EEG amplitudes Parameters modeled as random variables inferred means μ variances σ2. Plates variables share same parents children outer plate over sensory conditions inner plate over participants nHDDM model fits RT distributions car face choices visual audiovisual conditions Regression coefficients (γ Early Late EEG amplitudes in V AV conditions predictors drift rate (δ) nHDDM Coefficients derived from nHDDM n = 40 participants 28,540 trials Starting point values (β estimated conditions Boundary separation values (α Nondecision times (nDT) estimated Dots indicate single-participant values gray lines connect population means c–g Source datahypothesized reliability sensory information visual coherence information integration modeled linear relationship between drift rate coherence levels by Early Late EEG amplitudes tested three models coherence scaled yEarly yLate both components best fit model coherence scales yLate (deviance information criterions (DICs) = 517 661 modulation Late component reliability evidence predicts evidence accumulation model suggests effect task difficulty on behavioral performance captured by post-sensory mechanisms result dissociates roles Early Late EEG in decision-making consistent with Late component indexing quality evidence decision process associated with accuracy of perceptual choices applying model to behavioral data good fit accounting for most variance in choice RT data (average R2 = 0.94; Fig. 5b). Early Late EEG evidence within-participant single-trial discriminator amplitudes both predictive of drift rate in both sensory conditions (Fig.γEarly γLate larger than zero V AV t(39) = 17.67 15.55 11.92 16.02 for γLate p values < 0.001 two-sided paired t drift rates correct trials higher than incorrect (δcorrect = 0.27 ± 0.09 δincorrect = 0.14 ± 0.11 face −0.65 ± 0.11 −0.34 ± 0.10 car positive faces negative accuracy effects captured by nHDDM contribution Late not Early component higher in AV V trials (Fig. 5d t(39) = −0.6891 p = 0.4984 for γEarly t(39) = −2.66, p = 0.011 for γLate). consistent with increased discrimination power Late component AV facilitation evidence accumulation investigated effect additional auditory information on parameters nHDDM no difference in starting point boundary separation between sensory conditions (βV = 0.5475 ± 0.005 βAV = 0.5495 ± 0.005 = −0.4964 p = 0.6224αV = 1.13 ± 0.03 αAV 0.12 ± 0.03 t test t(39) = 0.8191 p = 0.4177 Fig. longer nDTs AV trials (370 ± 9 ms V vs 408 ± 10 ms AV t(39) = 2.81 p = 0.0063 related longer stimulus encoding AV extra time auditory stimulus average difference RTs (33 ms comparable nondecision difference (38 early sensory origins longer RTs AV informed modelling choice explored analyses effects captured strategies choice behavior starting point bias closer face choices t(39) = 9.06 V t(39) 10.63 AV p values < 0.001) higher drift rate car choices (δcar-V = −0.92 ± 0.14 δface-V = 0.15 ± 0.13-AV −0.96 ± 0.14-AV = 0.07 ± examined differences results face car choices participants choose cars often than (60 ± 8% V 52 ± 8% AV more accurate (84% vs 74% average face choiceaccuracy effect comparable across V AV trials (two t test t(39) = −0.98 p = 0.3353; 83.2 ± 1.34% V 86.59 ± 1.14% AV face choices 72.84 ± 0.98% V 76.9 ± 0.91% AV car results suggest bias starting point accuracy difference between face car choices more errors higher car responses higher drift rate car choices consistent with RT distributions face car choices (Fig. car RTs longer tails modes similar face parameter differences consistent across conditions no impact on behavioral multisensory effects neural mechanisms timing Early component duration sensory processing mechanisms nHDDM tested model with yEarly regressor for nDT no relationship between yEarly nDT coefficients not different 0 V model yields poorer fit data model choice (DIC = 1277 vs 517 nHDDM). consistent increases nDTs AV trials driven by increases early encoding time auditory informationNeurally informed model outperforms behaviorally constrained previous studies fit DDMs to behavioral data inclusion EEG-derived regressors improves model performance conclusion compared neurally inspired HDDM to standard HDDM without traditional model poorer trade-off between goodness-of-fit complexity assessed DIC neurally informed (DICHDDM = 758 vs DICnHDDM = 517) conclusions contradict conventional HDDM yields larger boundary separations for AV trials −3.52 p = 0.0011) nDTs estimated ~100–120 ms longer for conditions nHDDM (490 ± 10 ms V 509 ± 11 ms for difference in average nDTs across conditions (19 ms track mean RT difference closely poorer model could lead misleading conclusion auditory information affects response caution supports importance constraining behavioral models with neural data integrating neural information accurate characterization behavioral effects interpretation used multivariate single-trial EEG analysis behavioral modelling enhancement visual perceptual decisions by auditory informationimprovements in behavioral performance in AV trials late EEG component decision processes36–38 earlier EEG component unaffected by addition auditory evidence neurally informed cognitive modelling multisensory behavioral neural benefits explained by improvements evidence accumulation decision process processing multisensory information requires coordination mechanisms bottom-up top-down predictive coding neural implementation involves distributed neural network primary sensory parietal frontal brain areas multisensory perception10 study interested multisensory information combined perceptual judgements multisensory decision-making two theories emphasize early or late integration multisensory information early integration posits sensory evidence combined early sensory encoding supported by direct pathways between early visual auditory regions cross-modal influences neural responses early visual studies benefits for perception simplistic visual stimuli motion shape from acoustic information simple stimuli may early sensory regions biased interpretation complex objectslate integration hypothesis postulates evidence each sensory modality processed separately during early encoding combined into single source downstream during decision formation Support animal human experiments multisensory information accumulated to decision processing unisensory prior recent neuroimaging work flexible behavior accounted by causal inference multisensory representations on higher-level parietal prefrontal regions to evidence findings odds with early integration hypothesis no evidence addition auditory information encoding early visual evidence comparable between V AV trials support for post-sensory enhancements decision evidence with addition auditory information consistent with late integration hypothesis later visual representations likely reside in higher-order visual areas object recognition categorization lateral occipital consistent with emergence multisensory evidence after early sensory encoding9 timing representations after early encoding until commitment to choice unfold concurrently with decision input to evidence accumulation in prefrontal parietal cortex66–70 potential confounding factor for late multisensory effects differences in attention between V and AV conditions effects impacted early late processing stagesrecent work suggests influences multisensory information attention operate independently across cortical columns71 attentional resources shared across sensory modalities72 arguing against competition between sensory modalities for attentional resources randomized trials V or AV stimuli minimizing differences attention between sensory conditions potential departure observed increased RTs during multisensory trials stems from auditory information context-dependent complementary to visual information sounds as supplementary evidence additional processing resources consistent with increases in nDTs for multisensory trials alternative interpretation increases could differences in motor preparation unlikely participants decision same motor effectors button press in hypothesized increases in sensory encoding time due additional processing auditory evidence during AV trials reflected in latency Early component unaffected by additional auditory evidence reaction time differences could arise due to choice sensory modalities interindividual choice strategies recent using time-varying multisensory information reported faster less accurate choices for unisensory decisions authors modeled results with DDM model time-varying information sensory cue reliability reported consistent drift rate improvements in multisensory condition acrossdespite differences in behavioral outcomes findings line with increase drift rate in AV trials both studies suggest multisensory information leads to faster accumulation sensory evidence neural underpinnings behavioral benefits from addition auditory information joint cognitive modelling behavioral neural correlates sensory decision evidence with processes decision-making neurally informed DDM improvement behavioral performance from enhanced post-sensory representations evidence accumulation contrary to version standard DDM attributed longer RTs in AV trials to additional changes decision boundary early encoding auditory stimulus discrepancy less accurate account of trial-by-trial variability in decision dynamics single-trial data inclusion of EEG components provided accurate account contributions early sensory post-sensory decision evidence approach enabled disambiguation of internal processing stages behavioral benefit behavioral model yielded longer stimulus encoding times track difference measured RTs findings suggest constraining models perceptual decision-making with neural data provide mechanistic insights unobserved using behavioral modelling line with recent research high complexity decision-making models may yield neurally incompatibleinformed by neural measurements models yield reliable parameter estimates light on neural mechanisms behavioral effects43,51,77–81 previous studies used DDMs multisensory decision-making models multisensory information hardwired to converge during sensory accumulation25,82,83 models describe human behavior evaluate hypotheses about locus convergence Other studies combined behavioral modelling DDMs and EEG use neural data constrain behavioral model argued encoding of visual random-dot motion affected by acoustic motion5 sensory-level integration effect not validated using EEG-inspired DDM model explanation for diverging findings simpler stimuli biased earlier study to sensory-level effect multisensory information about complex objects combined at post-sensory stage interpretation supported by neuroimaging studies reported audiovisual interactions for complex stimuli at longer poststimulus latencies high-level brain regions84–87 difference divergent findings construction of multisensory context across tasks audiovisual integration studies use direct mapping between source evidence for noisy acoustic information In present task direct audiovisual mapping not availabledecision cross street foggy morning based on hazy objects street sounds decision step off curb based broader audiovisual context higher-level conceptualization evidence car-like objects sounds signaling busy street subtle critical distinction deciphering mechanisms audiovisual integration reconciling discrepancies experimental designs estimated minimum sample size 35 participants a priori power analysis fixed linear multiple regression model two predictors medium effect size 0.5 alpha 0.05 power 0.95 tested 40 participants (male 18 female 22 mean age 23.85 SD 5.47) speeded face-vs-car categorization task participants right-handed normal vision no neurological disorders study approved ethics committee College of Science and Engineering University of Glasgow participants provided written informed consent used 30 grayscale faces 15 cars size 670 × pixels 8-bits per pixel from previous original face images Max Planck Institute of Biological Cybernetics88 car images Internet background removed uniform gray background images equated for spatial frequency contrast luminance identical magnitude spectra manipulated phase spectra weighted mean phase changed visual evidence stimuli percentage phase coherencetask difficulty used four levels sensory visual evidence (27.5 30 32.5 35% phase levels based previous studies36–38,42 performance spanning psychophysical threshold image categories faces cars equal frontal side views ±45 displayed pictures light gray background (RGB PsychoPy software90 1.83.04) duration 50 ms.Auditory sounds (15 car 15 face-related used images random trials Sounds human speech car/street-related sounds online No copyright restrictions modifications sound files allowed sampled 22.05 kHz stored .wav files MATLAB 2015b added 10 ms cosine on/off ramp sound onsets normalized sounds reduced intensity amplitude 80% Sounds embedded Gaussian white noise relative amplitude manipulated 17 levels noise-to-signal ratios 12.5 to 200% noise noisy speech car-related sounds presented binaurally 50 ms Sennheiser stereo headphones HD 215 stimulus display controlled Dell 64 bit machine (16 GB RAM NVIDIA Quadro K620 card Windows Professional 7 Linux-x86_64 PsychoPy presentation software90images presented on Asus ROG Swift PG278Q monitor (resolution 2560 × 1440 pixels refresh rate 120 Participants seated 75 cm from stimulus display each image subtended ~11 × 11 degrees visual angle employed adapted audiovisual visual face-vs-car image categorization task36–38 required participants decide face or car in stimulus decision via button press keyboard response deadline 1.5 seconds additional auditory cue brief noisy sound congruent with content face trials accompanied by human speech sound car trials car-related sound stimuli presented for 50 ms in center screen AV trials both ears instructed pay equal attention base decision on information both modalities pictures sounds presented simultaneously used four levels of visual noise one participant-specific auditory difficulty level obtained performance training accounted for interindividual differences in auditory perception experimental paradigm required participants attend training testing session two consecutive days same first daytraining participants asked perform three categorization tasks visual image discrimination (face-vs auditory sound discrimination (3) audiovisual discrimination (face-vs-car). received visual feedback each response Feedback presented centrally for three outcomes ‘Correct’ green ‘Incorrect’ red ‘Too slow’ blue response Stimuli presentation duration set 50 ms for comparability testing visual training task used same images four levels visual evidence second day auditory training task presented sounds eight levels noise-to-signal ratios (12.5%, 37.5% 62.5% 93.75% 125% 150%, 175% 200% added noise). estimated participant-specific noise levels performance ~70% decision accuracy), including levels between eight noise-to-signal ratios used levels for audiovisual training task main experiment audiovisual training task used all images four levels visual evidence participant-specific perithreshold noise level mimicked main task second day participants received feedback on choices training day presented 480 trials for visual auditory discrimination training tasks split four blocks of 120 trials 60-second rest period between blocks 240 trials two blocks audiovisual training task all three training tasks lasted 55 minutes first daysecond day collected behavioral EEG data using randomly interleaved visual audiovisual trials combined task (Fig. 1) same task timings both feedback during testing one auditory noise level per participant effects benefit different levels visual evidence presented 720 trials—divided between stimulus categories face/car V/AV four levels visual evidence blocks 60 trials 60-second breaks task lasted 45 minutes EEG data collected testing day.Behavioral analysis quantified behavioral performance decision accuracy RTs using two GLMMs GLMMs superior ANOVA analysis random effects structure accounts inter-participant variability allows mixing categorical continuous variables91 models included main effects interactions variables modality (V AV visual evidence (27.5 30 32.5 by-participant random slopes random intercepts for modality main effect random effects structure justified design adopted for parsimony post hoc likelihood-ratio (χ2) model comparisons quantify predictive power significance main effects interactions GLMMs compared full model reduced model excluding results statistics post hoc model comparisons reported in main results sectionperformed GLMM analyses lme4 package92 RStudio93 binomial logit model glmer function decision accuracy binary dependent variable gamma model RT continuous dependent variable bobyca optimizer predictor modality entered mean-centered form visual evidence levels mean-centered backward difference coding accounted small imbalances trial numbers levels Random correlations excluded GLMMs interaction effects complemented models comparisons Bayesian linear mixed models lmBF function default BayesFactor package94 RStudio93) Bayes Factor evidence alternative model reduced model omitting interaction proportional error estimate 500.000 Markov chain Monte Carlo) iterations models single-trial data mean trials per condition Vcar = 178.53 Vface 178.78 AVcar 178.33 AVface 177.88 mean absolute deviation Vcar = 1.74 Vface 1.6 AVcar 2.18 AVface 2.81 trials 180 trials per condition presented participants quantify single-trial RT distributions bimodal standardized (z-scored RTs participant level fitting exponentially modified Gaussian distributions RT distributionrule out participant-specific levels auditory evidence improvements decision accuracy in AV trials correlated measures across participants bend correlation evaluated levels auditory noise with difference accuracy between V AV trials computed mean accuracy across trials each visual evidence modality participants well in V trials AV trials decision accuracy between V AV trials.EEG data acquisition preprocessingWe acquired continuous EEG data sound-attenuated electrostatically room from 64-channel EEG amplifier system with Ag/AgCl scalp electrodes system on EasyCap chin electrode acted as ground channels referenced to left mastoid adjusted input impedance to <20 kΩ data sampled at 1000 Hz filtering by 0.0016–250 Hz analog band-pass filter used PsychoPy90 Brain Vision Recorder record trial-specific information experimental event codes button responses with EEG data data collected stored for offline analysis in MATLAB Offline data preprocessing included software-based fourth-order butterworth band-pass filter cutoff frequencies between 0.5 and 40 Hzavoid phase-related distortions applied filters noncausally MATLAB filtfilt). EEG data re-referenced to average channels removed eye movement artifacts blinks saccades data from eye movement calibration task before main task participants instructed blink black fixation cross on light gray background screen before lateral horizontal saccades according location fixation identified linear EEG sensor weights eye movement artifacts projected broadband data main task subtracted excluded trials where exceeded RT limit 1.5 s response within <300 ms after onset stimulus EEG signal exceeded maximum amplitude 150 μV (0.8%, 0.06% 0.03% of trials data linear multivariate single-trial discriminant analysis of stimulus- response-locked EEG identify early sensory late decision-related EEG components discriminating between face car trials performed analysis separately for V AV trials identify sensor signals visual comparisons discrimination performance single trials included in discriminant analyses identified projection multichannel EEG signal xi = (1...N short time windows discriminated between face car trials time windows 60 ms onset intervals every 10 mswindows centered shifted −100 to 1000 ms stimulus onset-locked data −600 to 500 ms response button press-locked data 64-channel spatial weighting\documentclass[12pt]{minimal{amsmath-69pt}w(τ) learned logistic regression45 maximal discrimination each time window one-dimensional projection yi(τ), each trial given window τ:1[12pt]{minimal{amsmath\oddsidemargin-69pt}$$y(\tau ) = w(\tau )^Tx\left\tau \right) = \mathop {\sum\limits_{i = 1}^D w_i(\tau )x_i(\tau ).\end{document}y(τ)=w(τ)Txτ=∑i=1Dwi(τ)xi T refers transpose operator D number of EEG sensorsseparating stimulus categories discriminator component amplitudes\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek\oddsidemargin-69pt}{document}$$y_i(}yi(τ) face car trials positive negative values values weighted reflection neural evidence specific decision task (face-vs-car) separate analyses each modality condition unspecific effects memory recollection estimation classification weights face car trials subtracted out96 quantify performance discriminator time window used area under ROC curve97 Az value leave-one-trial-out cross-validation procedure control overfitting36–38 used N-1 trials estimate spatial filter w applied left out trial obtain out-of-sample discriminant component amplitudes compute Az value determined significance thresholds for discriminator performance using bootstrap analysis face car labels randomized separate leave-one-trial-out test randomization repeated 1000 times probability distribution for Az estimate Az value significance level p < 0.05 (participant average Azsig = 0.57)EEG analysis pipeline performed on individual participants replication unit98 linearity model scalp projections discriminating components Eq. (1) estimating forward model\documentclass[12pt{minimal{amsmath\oddsidemargin-69pt{document\left =\frac{{x(\tau )y( )}}{{y\left^Ty{document}aτ=x(τ)y(τ)yτTy EEG data discriminating components (y) in matrix and vector notation forward models displayed as scalp topographies interpreted coupling between observed EEG discriminating component amplitudes vector α reflects electrical coupling discriminating component y activity forward models computed separately for V and AV face-vs-car discriminant analyses.Optimizing spatiotemporal discriminating activity used forward model estimates Eq. (2) identify temporal transitions between components differences scalp distribution changes cortical sources used k-means clustering algorithm Euclidean distance on intensities vector a(τ) for time range optimized ktime windows similar scalp topographies using silhouette implemented in MATLAB’s evalclusters function results robust regardless criterion Silhouette CalinskiHarabasz distance conditions V or AV used temporal components in EEG analyses.Temporal cluster-based bootstrap discriminator performance between V AV trials used percentile bootstrap technique comparing group-level Az difference between samples49 sample-by-sample created distribution shuffled Az difference scores AV–V across participants repeated 1000 times each created random bootstrap distribution median Az difference scores computed median 95% confidence interval (2.5–97.5% distribution median difference scores test median difference compared against lower estimated confidence interval 2.5% threshold; p < 0.025) form contiguous temporal clusters avoid effects false positives required minimum temporal cluster size of three significant samples threshold determined 95th percentile data-driven null distribution of maximum cluster sizes relationship between adjacent samples preserved applied permutation procedure shuffling temporal samples without replacement abolish relationship relative difference between V AV Az values unchanged eachgenerated null distribution maximum cluster sizes storing maximum samples largest cluster 1000 iterations Similar analysis original data performed discriminator performance (Az stimulus- response-locked data (Figs. 3b 4b yielded average three significant samples procedure corrects multiple comparisons comparable temporal cluster-based nonparametric permutation test ref. 100 procedure determined temporal window selection single-trial EEG component amplitudes included neurally informed drift diffusion modelling analysis sample-based procedure performed discriminator accuracy times represent centers original discrimination windows wider window (60 extended selection window 30 ms significant clusters analysis neural effects traceable participants variability computed proportion participants participant-level effect general group-level effect per sample higher AV Az value sample Figs. 3c 4c). performed statistical analyses MATLAB code Figshare GitHub repositories refs. 49,50 quantify evidence effects sensory modality decision accuracy subject-specific component amplitudes (y trial accuracy (Figs. 3e 4e), computed two additional Bayesian linear mixed models analyses generalTestBF function BayesFactor RStudio93) splitting trials correct incorrect responses Bayes Factor evidence alternative modellarger model including predictor compared to reduced model omitting predictor examining interaction between sensory modality decision accuracy alternative model omitting interaction term performed bend correlation topographical consistency between late stimulus-locked response-locked components computed average scalp map forward models across participants peak discrimination components (500 ms poststimulus 100 ms prestimulus assessed similarity correlation used two similar bend correlation analyses test individual onset times Early component difference stimulus-locked Late component discriminator amplitudes peak time stimulus-locked Late component.Hierarchical drift diffusion modelling behavioral fit participants’ performance face or car choice RT with HDDM101 HDDM assumes stochastic accumulation sensory evidence toward two decision boundaries model returns estimates of internal components processing rate of evidence accumulation distance between decision boundaries decision bias toward choices duration of nDT processes HDDM uses MCMC sampling adjust parameters maximize summed log-likelihood of predicted mean RT accuracy DDM parameters estimated in hierarchical Bayesian framework prior distributions updated likelihood data posteriorBayesian analysis HDDM benefits traditional DDM analysis supports variables as regressors model relations with physiological behavioral data51,78–80 HDDM link between EEG components decision-making process posterior distributions convey uncertainty parameter estimates102 Bayesian hierarchical framework effective when observations low105 all observers dataset assumed from group yields stable parameter estimates for hierarchical DDM used Wiener module101 JAGS106 Matjags interface in MATLAB to estimate posterior distributions each trial likelihood of accuracy RT assessed Wiener first-passage time distribution with three model parameters (boundary separation nDT drift Parameters drawn from group-level Gaussian distributions means SDs had non-informative uniformly distributed priors all SD priors uniformly distributed(0.01mean priors of nDT boundary separation starting point uniformly distributed: nDT ~ U(0.01, 1), α ~ U(0.01, 3) β ~ U(0.1, 0.9). priors of regression coefficients{minimal{amsmath-69pt Gaussians N(0, 3) ran five Markov chains with 5500 samples each first 500 discarded rest subsampled by factor of 50 conventional approach MCMC sampling initial samples unreliable random starting point neighboring samples highly correlated101 remaining samples probability distributions of each estimated parameter estimates computedensure convergence chains computed Gelman–Rubin\documentclass[12pt{minimal}\usepackage{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek\oddsidemargin-69pt} statistic compares within-chain between-chain variance), verified group-level parameters[12pt{minimal}{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}^ close to 1 lower than 1.03. comparison models used DIC measure fit assessment comparison hierarchical DIC selects model best trade-off between goodness-of-fit model complexity Lower DIC values favor models highest likelihood least degrees freedom estimated nHDDM EEG discrimination analysis inform fitting behavioral data single-trial RTs (face or car choices of 40 participants hypothesized evidence accumulation rate each trial dependent on neural evidence face or car choicemodel fitting HDDM framework used single-trial EEG measures face-vs-car discrimination analysis regressors drift rate (δ)\documentclass[12pt]{minimal}{amsmath}{wasysym}}{upgreek}\oddsidemargin-69pt}{document}\delta = \gamma _0 + _1 *\mathrm{Early{s}} +\gamma _2 *\mathrm{Late}}}{s}}\end{document}δ=γ0+γ1*yEarlys+γ2*yLates*C\documentclass[12pt]{minimal}{amsmath}{wasysym}{upgreek}\oddsidemargin}{-69pt}\begin{document}$$y\mathrm{Early}}}\mathrm{s}}\end{document}yEarlys[12pt]{minimal}{amsmath}{wasysym}}}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}\mathrm{Lasingle-trial discriminator amplitudes participant-specific stimulus-locked Early EEG peak Az across V AV 180–360 ms poststimulus Late EEG peak Az difference between AV V Fig. 3b 490–540 ms (expanded 30 ms side Az values 60 ms training windowscoefficients\documentclass[12pt]{minimal}{amsmath{wasysym}{amsfonts}{amssymb{amsbsy{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}\gamma\end{document weight slope drift rate values[12pt]{minimal}{amsmath{wasysym{amsfonts{amsbsy{mathrsfs{upgreek}{\oddsidemargin}{-69pt}{document\mathrm{Early}}}\mathrm{s}}\end{document[12pt]{minimal}{amsmath}{wasysym{amsfonts}{amssymb{amsbsy{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}\mathrm{Late}}}\mathrm{s}}\end{document}yLates specific trial intercept[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin-69pt}\begin{document$\gamma\end}γ0 estimated\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{mathrsfs{upgreek\setlength\oddsidemargin-69pt}$\gamma\end{document}γi’s each participant sensory condition C phase coherence level image each trial represents quality visual evidence proportional to amplitude Late component38,42,43 regression coefficients influences components on drift rate in both sensory conditions78 Posterior probability densities each regression coefficient estimated sampling procedure positive effects determined when >99.9% posterior density higher than 0 test differences sensory conditions performed “hierarchical” t test population-level distributions parameters testing biases hierarchical structure model conservative effect sizes107 estimated HDDM without neural correlates fit HDDM to RT distributions for face car choices conditioned sensory condition (V or AV) each trial drift rate separation starting point nDT estimated for each participant dependent on sensory conditionassumed evidence strength drift rate modeled linear relationship drift rate coherence information Nature Research Reporting Summary article.Supplementary informationPeer Review FileReporting Summary | 47 | 1.331961 |
10.1038/s41467-021-20907-z | PMC7846738 | It is known that invasive lung adenocarcinomas evolve from pre-cancerous dysplastic lesions. In this study, the authors show that evolution of pre-cancerous lesions is accompanied by DNA methylation alterations, and that global hypomethylation correlates with immune infiltration, mutational burden and copy number alterations. | The evolution of DNA methylome and methylation intra-tumor heterogeneity (ITH) during early carcinogenesis of lung adenocarcinoma has not been systematically studied. We perform reduced representation bisulfite sequencing of invasive lung adenocarcinoma and its precursors, atypical adenomatous hyperplasia, adenocarcinoma in situ and minimally invasive adenocarcinoma. We observe gradual increase of methylation aberrations and significantly higher level of methylation ITH in later-stage lesions. The phylogenetic patterns inferred from methylation aberrations resemble those based on somatic mutations suggesting parallel methylation and genetic evolution. De-convolution reveal higher ratio of T regulatory cells (Tregs) versus CD8 + T cells in later-stage diseases, implying progressive immunosuppression with neoplastic progression. Furthermore, increased global hypomethylation is associated with higher mutation burden, copy number variation burden and AI burden as well as higher Treg/CD8 ratio, highlighting the potential impact of methylation on chromosomal instability, mutagenesis and tumor immune microenvironment during early carcinogenesis of lung adenocarcinomas. | IntroductionLung cancer remains the leading cause of cancer-related death worldwide, yet it is curable if treated early. Many cancers, including lung cancers, are preceded by precancers. Treating precancers to prevent invasive lung cancer is theoretically an attractive approach to reduce lung cancer-associated morbidity and mortality. However, developing strategies for lung cancer prevention has been challenging owing to our limited understanding of neoplastic progression from precancers to invasive lung cancers1. Lung adenocarcinoma (ADC) is the most common histologic subtype, accounting for >50% of all lung cancers2. It has been proposed that invasive lung ADC evolves from atypical adenomatous hyperplasia (AAH), the only recognized precancer for lung ADC, which could progress to ADC in situ (AIS), a preinvasive lung cancer, then to minimally invasive ADC (MIA), and eventually frankly invasive ADC3. AAH, AIS, MIA, and some ADC often present as pulmonary nodules with a unique radiologic feature termed ground-glass opacity (hazy nodular opacity with the preservation of underlying bronchial and vascular margins) on computed tomography scans. The biology and clinical course of these lesions are not well defined and their management is controversial. Surgical resection is not the standard of care for treating these lung nodules and the diagnostic yield of biopsy is often low, particularly for ground-glass opacity-dominant nodules4. Therefore, these lung nodules are often referred to as indeterminate pulmonary nodules (IPNs). However, as surgical resection is not often offered, obtaining adequate tissue for comprehensive profiling of these IPNs is difficult, hindering our understanding of the biology underlying these lesions.We initiated an international collaboration for the collection and characterization of these lung precancers, preinvasive, and early invasive ADC presenting as IPNs. We recently reported the genomic landscape, including the genomic intra-tumor heterogeneity (ITH), and revealed evidence of progressive evolution from AAH to AIS, MIA, and ADC5. In addition to mutations, epigenetic alterations such as DNA methylation can also impact neoplastic transformation and fitness. Recent genome-wide methylation profiling studies have revealed that certain alterations, such as the silencing of tumor suppressor genes (TSGs) and the activation of genes in stem-like cellular programs6,7, may contribute to carcinogenesis. Our previous study has demonstrated that complex methylation of ITH was associated with larger tumor size and increased risk of postsurgical recurrence in patients with invasive lung ADC8. Methylation aberrations have been reported in AAH lesions and tumor-adjacent lung tissues, suggesting that methylation changes may be early molecular events9,10. However, these pioneer studies only analyzed small numbers of genes implicated in lung carcinogenesis. The dynamic changes in methylation at the genome level and the evolutionary trajectories of methylation ITH during the initiation and progression of lung precancers have not been studied systematically.In this study, using a unique cohort of resected IPNs of different histologic stages, we delineate the evolution of the methylation landscape and reveal increased methylation ITH in lung ADC than its precursors of early stages, as well as global hypomethylation correlates with immune infiltration, mutational burden, and copy number alterations.ResultsDNA methylation aberrations increase with the progression of precancersWe performed reduced representation bisulfite sequencing (RRBS) of 62 resected IPNs (14 AAH, 15 AIS, 22 MIA, and 11 invasive ADC) and their paired normal lung tissues from 39 patients (Supplementary Data 1). There was no significant difference regarding age (p = 0.6288, Kruskal–Wallis H test), sex (p = 0.6482, χ2 test), or smoking status (p = 0.5696, χ2 test) between different histologic groups (Supplementary Data 2). To minimize the impact of “contamination” from nonmalignant cells, each IPN specimen was reviewed by two lung cancer pathologists to confirm the diagnosis, mark the areas of diseases, and estimate the purity. Only specimens with a minimum of 40% of premalignant or malignant cells were included and manual macrodissection was applied to obtain premalignant or malignant cells for DNA extraction. Furthermore, we estimated the tumor purity of these specimens in silico based on whole-exome sequencing (WES) data by ABSOLUTE11. With the caveat that these IPNs had a high level of genomic ITH leading to underestimation of tumor purity5, we observed no significant difference in tumor purity between different stages (Supplementary Fig. 1).The mean RRBS sequencing coverage was 58.98 reads. With a minimum of ten reads shared across all samples, the RRBS profiling allowed the quantification of methylation status at 751,462 CpG sites mapped to 15,761 known genes. The principal component analysis demonstrated that the DNA methylome of AAH was more similar to that of normal lung tissue, whereas those of AIS, MIA, and ADC were clearly different from that of the normal lung (Fig. 1a). Similarly, unsupervised hierarchical clustering identified two separate clusters: one comprising normal lung and AAH, and the other including AIS, MIA, and ADC (Supplementary Fig. 2). In addition, different spatially separated specimens from the same IPNs tended to cluster together, suggesting that interlesion heterogeneity was more prevalent than intralesion heterogeneity. Similarly, the methylation profiles of different spatially separated specimens from the same IPNs were highly correlated (median coefficient r = 0.884 [range, 0.736–0.969], p < 2.2 × 10−16; Supplementary Fig. 3).Fig. 1DNA methylation aberrations increase with the progression of precancers.a Principal component analysis of the methylation profiles of the average methylation values in 39,148 5-kb tiling regions (shared in all samples) composed of autosomal, non-polymorphism CpG sites supported by at least 10× read coverage in IPNs of different stages. The solid dots of different colors represent IPNs of different stages and the star (*) represented the centers of specimens of each histologic stage. b Overall genome-wide DNA methylation correlation between IPNs of different stages and matched normal tissue from the same patients. The yellow-purple clouded dots in the smooth scatter plot represent the CpG sites covered by IPN specimens (n = 14 for AAH, n = 11 for AIS, n = 18 for MIA, n = 10 for ADC) and paired normal lung. Two-tailed Pearson’s correlation coefficient r values are shown on the top. P < 2.2 × 10−16 in all four histologic stages. The red dots represent CpG sites with hypermethylation in IPNs (within DMRs of methylation gain in IPNs and methylation ≤20% of CpG sites in the corresponding normal lung) and the green dots represent CpG sites with hypomethylation (within DMRs of methylation loss in IPNs and methylation ≥20% of CpG sites in the corresponding normal lung). c The number of CpG sites overlapping with DMRs showing hypermethylation (left) or hypomethylation (right) in IPNs of different histologic stages. The green dots represent the mean of normalized numbers of CpG sites overlapping with DMRs in each IPN. The solid blue dots represent the mean numbers of CpG sites overlapping with DMRs in IPNs of each histologic stage with a 95% confidence interval as error bars. The differences among all stages were assessed using the Kruskal–Wallis H test. Source data is provided as a source data file.There was a progressive increase in the number of CpG sites displaying hypermethylation or hypomethylation (as compared with matched normal lung tissues from the same patients) from AAH to AIS, MIA, and invasive ADC. Hypermethylation appeared to emerge as early as AAH, whereas hypomethylation was evident in only AIS, MIA, and ADC. Meanwhile, the correlation coefficients between the methylation profiles of IPNs and those of their paired normal tissues progressively decreased (r = 0.885 for AAH–normal, 0.824 for AIS–normal, 0.796 for MIA–normal, and 0.740 for ADC–normal, p < 2.2 × 10−16, for all comparisons; Fig. 1b). These decreases were marginal, reflecting the substantial heterogeneity among different patients and suggesting that overall methylation changes may be subtle during early lung carcinogenesis. Nonetheless, these marginal decreases represent methylation alterations in thousands of CpG sites. Further quantification of differentially methylated regions (DMRs) in these IPNs compared with their paired normal lung tissues revealed that later-stage IPNs had numerically more CpG sites with hypermethylation (p = 0.3635, Kruskal–Wallis H test) and significantly more CpG sites with hypomethylation (p = 0.000287, Kruskal–Wallis H test) (Fig. 1c).Evolution of methylome shows similar trend across all genomic regulatory regionsThe epigenome is composed of regions with various regulatory elements. To further depict the methylation evolution at different genomic regulatory regions, we first partitioned genomic regions into promoters, enhancers, transcribed regions (transcription), and repressed regions (heterochromatin) based on the peaks of histone marks (H3K04me1, H3K04me2, H3K04me3, H3K09ac, H3K09me3, H3K27ac, H3K27me3, H3K36me3, and H3K79me2) using ChromHMM12. We then calculated the numbers of DMRs that overlap with these regions. As shown in Fig. 2a–d, the number of DMRs was higher in later-stage IPNs in all genomic regulatory regions. Next, we selected DMRs overlapping with repetitive elements (UCSC RepeatMasker). Later-stage IPNs demonstrated increased DMRs in repetitive regions as well as nonrepetitive regions (Fig. 2e, f). Furthermore, as partially methylated domains (PMDs) have been reported to represent a major source of DNA methylation variation in a variety of cancer types13, we quantified DMRs inside PMDs versus DMRs outside PMDs. Similar to the distribution of DMRs in genomic regulatory regions, there were significantly more DMRs in later-stage IPNs both inside and outside PMDs (Fig. 2g, h). Taken together, these results suggested that the majority of methylation alterations during early carcinogenesis of lung ADC may be stochastic across all genomic regulatory regions.Fig. 2The number of DMRs in IPNs of different stages by genomic regulatory regions and PMDs.The number of DMRs are shown in promoters (a), enhancers (b), transcribed regions (c), repressed regions (heterochromatin) (d), repetitive regions (e), non-repetitive regions (f), inside PMDs (g), and outside PMDs (h). The solid blue dots represent the means of normalized numbers of DMRs in IPNs of each histologic stage with a 95% confidence interval as error bars. The differences among all stages were assessed using the Kruskal-Wallis H test. Source data is provided as a source data file.Enrichment of transcription factor (TF) binding sites at DMRsDNA methylation can impact the DNA binding of TFs to their target sequence, termed “motifs”14. We searched the motifs that were covered by DMRs in the IPNs and identified 50 motifs in AAH, 57 in AIS, 64 in MIA, and 66 in ADC that were significantly enriched over all DMRs in each histologic stage (Supplementary Data 3). Many of these motifs were aligned with known binding sites recognized by different TFs known to be involved in critical biological processes, including cell cycle progression, cell proliferation, apoptosis, tumor metastasis, angiogenesis, and immune response15–18 (Supplementary Data 4). Motifs associated with BHE40, BMAL1, CLOCK, EPAS1, MAX, MITF, MXI1, MYC, MYCN, TFE3, USF1, and USF2 were significantly enriched at DMRs across all stages, suggesting that the methylation changes at these CpG sites may be early events before the establishment of precancers. On the other hand, some motifs were significantly enriched in DMRs in later-stage IPNs, including ATF3, E2F1, E2F3, E2F4, E2F6, E2F7, EGR1, EGR2, KLF1, KLF12, KLF3, KLF4, KLF6, MYOG, PATZ1, SP1, SP2, SP3, SP4, TFDP1, ZBT14, ZIC1, ZN281, ZN335 exclusively in ADC; ASCL1, HEN1, KLF9, SNAI2 only in ADC and MIA; SNAI1 and TFEB in ADC, MIA, and AIS but not in AAH, suggesting that methylation changes at these CpG sites may be associated with later processes during carcinogenesis of these IPNs.Later-stage disease has higher methylation ITHMolecular ITH can have a profound impact on tumor biology19,20, and our previous work demonstrated that methylation ITH was associated with clinical outcome in patients with invasive ADC8. To assess the evolution of methylation ITH during the progression of lung precancers, we first calculated the “epiallele shifts,” the combinatory difference in epiallele status in all captured loci (>60 reads), for each IPN specimen compared to paired normal lung tissues from the same patients. MIA and invasive ADC showed more pronounced epiallele shifts than AAH or AIS did, indicating a higher level of methylation ITH in IPNs of later histologic stages (Fig. 3a). The corresponding abundance of eloci (loci with distinct epiallele shifts) was also higher in later-stage IPNs (p = 0.004567, Kruskal–Wallis H test) (Fig. 3b). When focusing on the 15 patients with multiple IPNs of different stages (IPNs of different stages with the identical genetic background and exposure history), the abundance of eloci was higher in later-stage IPNs than in early-stage IPNs in 13 of the 15 patients (Supplementary Fig. 4), further suggesting that later-stage IPNs have a higher level of methylation ITH. On the other hand, some patients (C23 and J43 for example) had a relative high level of methylation ITH in early-stage IPNs. Since the postoperative follow-up was rather short and the majority of patients are doing well without recurrence, the biological or clinical significance of a higher level of methylation ITH in early-stage IPNs remains to be determined. Furthermore, we calculated the frequency of 16 combinatory DNA methylation patterns at four consecutive CpG sites covered by the same RRBS reads (termed “epipolymorphism”) in each specimen. Accordingly, later-stage IPNs showed higher epiallele diversity than early-stage IPNs did (Supplementary Fig. 5). Moreover, ADCs had higher levels of epipolymorphism at eloci than MIA, AIS, or AAH did (Fig. 3c), indicating that DNA methylation status had undergone a greater extent of drifting in later-stage IPNs than in early-stage IPNs.Fig. 3Increased DNA methylation ITH in later-stage IPNs.a Cumulative distribution curves of epiallele shifts of the DNA methylome in AAH, AIS, MIA, and ADC specimens compared to normal lung, with 1st percentile of entropy as −21.46 for AAH, −28.88 for AIS, −30.75 for MIA, and −42.01 for ADC by quantile regression. b The abundance of eloci (loci with distinct epiallele shifts) among IPNs of different histologic stages. Each green dot represents the mean of normalized number of eloci in each IPN. The solid blue dots represent the mean numbers of eloci in the IPN of each histologic stage with 95% confidence interval as error bars. The differences among all stages were assessed using the Kruskal–Wallis H test. c The cumulative distribution curves of epipolymorphism for loci with significant epiallele shifts (ΔS < −60) in IPN specimens of AAH, AIS, MIA, and ADC, with 50th percentile of epipolymorphism as 0.295 for AAH, 0.362 for AIS, 0.359 for MIA, and 0.463 for ADC by quantile regression. X-axis denotes the epipolymorphism in IPN specimens at each loci. Y-axis denotes the cumulative fraction of epipolymorphism from all IPN specimens of each stage. Source data is provided as a source data file.In addition, we sought to assess whether methylation ITH differs at different genomic regulatory regions. As shown in Fig. 4a, loci at heterochromatic regions show the most pronounced epiallele shifts indicating a higher level of methylation ITH at genomic regions of repressed chromatin states. Moreover, there were more eloci inside PMDs than outside PMDs across all histologic stages (Fig. 4b–e). PMDs are known to be more epigenetically plastic21, which makes CpG sites inside PMDs more prone to develop methylation ITH. In addition, PMDs have been reported to associate with repressive chromatin domains, gene-poor regions, and low transcription21,22; therefore, a higher level of methylation ITH may be better tolerated inside PMDs.Fig. 4Comparison of methylation ITH by epiallele shifts in different genomic regulatory regions and PMDs.a Cumulative distribution curves of epiallele shifts (from IPN specimens versus its matched normal lung tissues) of consecutive loci located in promoters, enhancers, transcribed regions, and repressed regions (heterochromatin) are shown, with 1st percentile of entropy as −34.07 for all genomic regions, −30.87 for promoters, −38.12 for enhancers, −42.44 for transcribed regions, −45.48 for repressed regions, and −29.51 for repetitive regions by quantile regression. Cumulative distribution of epiallele shifts (from IPN specimens versus its matched normal lung tissues) of consecutive loci located inside PMDs (red) versus outside PMDs (blue), with 1st percentile of entropy as −29.10 inside PMDs and −20.73 outside PMDs in AAH (b), −37.60 inside PMDs and −28.05 outside PMDs in AIS (c), −44.74 inside PMDs and −29.25 outside PMDs in MIA (d), −50.82 inside PMDs and −41.25 outside PMDs in ADC (e) by quantile regression. The boundary of PMDs is derived from WGBS profiling of normal lung. Source data is provided as a source data file.We next examined the relative distance of eloci to the nearest transcription start sites (TSSs). Interestingly, the vast majority of the eloci were much closer to TSSs in invasive ADC than in AAH, AIS, or MIA (Supplementary Fig. 6). To explore the potential association between methylation ITH and histone modification, we performed locus overlap analysis (LOLA)23 of genomic regions identified as eloci to evaluate the overlap between genomic regions with methylation ITH and genomic regions targeted by diverse histone posttranslational modifications24,25. LOLA calculates the number of overlapping versus nonoverlapping regions to assess the significance of the overlap. After adjustment for false discovery rate considering more eloci in later-stage IPNs, MIA and ADC had higher incidences of eloci significantly enriched in genomic regions occupied by H3K27me3, H3K9me3, and H3K9me2 (Supplementary Data 5), histone modifications strongly associated with transcriptional repression26–28 than AAH or AIS, supporting the concept that histone modifications cooperate with DNA methylation alterations along the evolution of lung precancers29,30.Genomic and methylation evolution was primarily in parallel during early lung carcinogenesisTo dissect the evolutionary relationship between epigenome and genome in lung ADC, we constructed phylogenetic trees. To avoid overfitting, only five patients (C5, J7, J8, J9, and J43) with IPNs having a minimum of four spatially separated specimens were included. The overall structure of methylation-based phylogenetic trees was similar to that of phylogenetic trees based on mutations5 (Fig. 5a–e). Furthermore, the genomic distance based on somatic mutations was positively correlated with the methylation distance between any pair of specimens from the same IPNs (Fig. 5f), suggesting parallel evolution (either collaboratively or independently) in play during early carcinogenesis of these lung ADCs. Particularly, four out of five patients with multiregional specimens (C5, J7, J8, J9, and J43) included for phylogenetic analysis demonstrated a strong correlation between methylation distance and genomic distance, with the only exception of patient J9 (Supplementary Fig. 7). Of note, J9 was a never smoker with the lowest genomic aberration burden (the lowest total mutation burden (TMB), lowest allelic imbalance (AI) burden, and the second lowest copy number variation (CNV) burden) and the highest level of genomic ITH (the lowest proportion of trunk mutations) among these five IPNs, but similar methylation aberration burden comparable to the other four IPNs (Supplementary Data 6).Fig. 5The evolutionary relationship between genomic and methylation landscape.Phylogenetic trees based on mutations (blue) and methylation values (green) in patient C5 (a), J7 (b), J8 (c), J9 (d), and J43 (e). The length of each branch indicates the similarity of mutation or methylation profiles between any pair of two spatially separated specimens from each patient. To avoid overfitting, only patients with IPNs having a minimum of four spatially separated specimens were included for this analysis. f Correlation of genetic distance (Hamming distance based on all mutations) and methylation distance (Euclidean distance based on methylation values of all CpG sites) between different spatially separated specimens from the same IPNs assessed by two-tailed Spearman’s correlation analysis (p = 2.62 × 10−38). Each dot represents the normalized distance between each pair of specimens from the same IPNs. Source data is provided as a source data file.Interestingly, in patient C10, promoter hypermethylation of TSC2, a candidate TSG known to inhibit cell growth in the lung31, was identified in AIS specimens, whereas copy number loss was identified in AAH specimens from the same patient. Similar phenomena of different putative TSGs were observed in several other patients (Supplementary Data 7). Taken together, these data suggested convergent evolution, whereby the same genes or pathways are activated or inactivated by different mechanisms in different cancer cell clones during lung cancer development and progression.Global hypomethylation and methylation ITH were associated with increased chromosomal instability (CIN) in precancerous IPNsAs an essential chemical modification, the methylation status can directly impact the chromosomal structure and DNA mutagenesis. It has been well documented that global hypomethylation is associated with CIN and increased mutational rates in cancers32,33. To further depict the interaction between genome and epigenome during early carcinogenesis of lung ADC, we assessed the global methylation status of these IPNs using long interspersed transposable elements-1 (LINE-1), a widely used surrogate marker for global DNA methylation34. As shown in Fig. 6a, we observed a significant decrease of LINE-1 methylation in AIS, MIA, and ADC compared to normal lung tissues or AAH indicating increased global hypomethylation in IPNs of later histologic stages. Importantly, the methylation level of LINE-1 was inversely correlated with CNV burden (Fig. 6b), AI burden (Fig. 6c), and TMB (Fig. 6d), indicating that global hypomethylation is associated with a higher level of CIN. Interestingly, LINE-1 methylation status was also inversely correlated with the proportion of clonal mutations in each specimen (Fig. 6e). Furthermore, there was a significant positive correlation between the abundance of eloci and CNV burden as well as AI burden (Supplementary Fig. 8), indicating higher methylation ITH was associated with increased CIN.Fig. 6Correlation of LINE-1 methylation with genomic features in IPNs of different stages.a LINE-1 methylation level in IPNs of different stages. Each green dot represents LINE-1 methylation level in each IPN and the solid blue dots represent the means of LINE-1 methylation level in IPNs of each histologic stage with 95% confidence intervals as error bars. The differences among all stages were assessed using the Kruskal–Wallis H test. Correlation between LINE-1 methylation levels and percent of genes with copy number changes (b), number of events with allelic imbalance (AI) (c), mutation burden (log 2 transformed) (d), and proportion of clonal mutations (e), assessed by two-tailed Spearman’s correlation analysis. Each dot represents each IPN specimen. Source data is provided as a source data file.Given the complex interaction between genome and epigenome, we next categorized CpG sites located in chromosomal regions with various CNV status to assess whether CNV status impacted aforementioned methylation analyses. As shown in Supplementary Fig. 9, the evolutionary patterns of methylation ITH (represented by the number of eloci) were similar regardless of using CpG sites located in genomic regions with copy number gain, loss, or neutral. We also plotted the cumulative distribution of epiallele shifts of AAH, AIS, MIA, and ADC using only CpG sites located in copy number-neutral chromosomal regions and observed that later-stage IPNs had a higher cumulative distribution of epiallele shifts than early-stage IPNs (Supplementary Fig. 10), similar to that using all CpG sites (Fig. 3a). Furthermore, we regenerated methylation-based phylogenetic trees using only CpG sites located in copy number-neutral chromosomal regions. As shown in Supplementary Fig. 11, the patterns of these methylation-based phylogenetic trees are almost identical to those using all CpG sites. Taken together, these results suggested that in these IPNs, most methylation changes were early molecular events that have occurred before the copy number changes, and once established, these patterns were inherited without significant changes from one cell generation to the next.Global hypomethylation was associated with suppressed T cell infiltrationCancer evolution is shaped by the interaction between cancer cells and host factors, particularly the host immune response. Given T cells’ central role in antitumor immune surveillance20,35, we depicted the T cell infiltration in AIS, MIA, and invasive ADC by deconvoluting RNA-sequencing (RNA-seq) data using ImmuCellAI from an independent cohort recently published36. Our analysis demonstrated an increase of CD4+ T regulatory cells (Tregs) (p = 4.25e − 27) and decrease of CD8+ T cells (although the difference was not significant, p = 0.1374) from normal lung tissues to AIS/MIA and invasive ADC leading to significantly higher Treg/CD8 ratio in invasive ADC (p = 3.17e − 31) (Supplementary Fig. 12a–c). We next applied MethylCIBERSORT37 to delineate the T cell infiltration of IPNs in current cohort. Similarly, we observed higher infiltration of Tregs (Fig. 7a, p = 0.009758) and lower infiltration of CD8+ T cells in later-stage IPNs, although the difference did not reach statistical significance (Fig. 7b, p = 0.1472), leading to significantly higher Treg/CD8 ratio in later-stage IPNs (Fig. 7c, p = 0.00194). As increased Treg/CD8 ratio is known to associate with suppressed antitumor immune surveillance38, these results indicated a more suppressive immune microenvironment in later-stage IPNs, consistent with our previous findings39. Interestingly, Treg/CD8 ratio was inversely correlated with LINE-1 methylation level (Fig. 7d), implying potential association between global hypomethylation and immunosuppression.Fig. 7T cell infiltration in IPNs of different stages.The immune cell fraction of T regulatory cells (Tregs) (a), CD8+ T cells (b), and Treg/CD8 ratio (c) in IPNs of different stages. Each green dot represents the mean immune cell fraction in each IPN and the blue dots represent the means of immune cell fraction in IPNs of each histologic stage with 95% confidence intervals as error bars. The differences among all stages were assessed using the Kruskal–Wallis H test. d Correlation between LINE-1 methylation and Treg/CD8 ratio was assessed by Spearman’s correlation analysis. Each dot represents each specimen. Source data is provided as a source data file.DiscussionThe methylation landscape has been studied extensively in various malignancies40–42. However, methylation aberrations in precancers are poorly defined, largely due to the lack of appropriate specimens. Using small panels of genes implicated in lung carcinogenesis, previous studies have demonstrated gradual changes in DNA methylation in AAH, AIS, and ADC9,10. However, due to a small number of loci, primitive technology and the lack of genomic data, many critical questions regarding methylation evolution during early carcinogenesis were not addressed. Leveraging a unique collection of resected IPNs of different stages, we delineated the DNA methylome in lung precancers, preinvasive ADC, and invasive ADC. Our results demonstrated a trend of increase in both hypermethylation and hypomethylation in later-stage diseases. Interestingly, compared to normal lung tissues, hypermethylation appeared to emerge as early as at AAH, whereas hypomethylation only became obvious after AIS (Fig. 1b), implying that somatic hypermethylation may have preceded hypomethylation during early carcinogenesis of lung ADC.Different cells within the same tumor can exhibit different molecular and phenotypic features, a phenomenon termed ITH. ITH may foster tumor evolution by providing diverse cell populations, and the dynamics of ITH architecture may evolve with neoplastic progression8,43. Methylation ITH has been observed in various advanced malignancies and higher levels of methylation ITH have been reported to associate with inferior clinical outcomes8,44–46. However, there are only a few reports on methylation ITH in precancers. For example, methylation ITH in the Barrett esophagus, a precursor to esophageal ADC, was associated with the risk of malignant transformation46. In the current study, we demonstrated a higher level of methylation ITH in later-stage IPNs than in early stages (Fig. 3 and Supplementary Figs. 4 and 5). These results are in line with the findings in advanced malignancies, suggesting that complex methylation ITH may be associated with more aggressive tumor entities. As methylation ITH is linked to evolutionary plasticity and phenotypic diversity44,47,48, a higher level of ITH could provide a higher probability to survive and progress. Interestingly, eloci were significantly more abundant around TSSs in ADC than in AAH, AIS, or MIA. One plausible explanation is that although somatic methylation ITH may be stochastic during early lung carcinogenesis, some of the methylation aberrations (particularly those that are close to TSSs and potentially impact gene expression) may convey survival and/or growth advantages, resulting in the selection of cells with higher densities of eloci around TSSs.Parallel evolution of genome and methylome has been reported in various advanced malignancies, including lung cancers8,49. The current study demonstrated similar phylogenetic patterns and correlated genetic and methylation distances in IPNs of different stages (Fig. 5), suggesting that genetic alterations and DNA methylation changes also evolve in parallel during early carcinogenesis of most lung ADCs. Meanwhile, promoter hypermethylation and copy number loss or mutations of putative TSGs were identified in independent IPNs within the same patients. These results were reminiscent of previous findings showing that distinct mutations of the same cancer genes were present in different spatial regions of the same tumors19,50,51 or in different primary tumors from the same patients52, indicating convergent evolution. Although genetic events (e.g., copy number loss of TSGs) or methylation changes (e.g., promoter hypermethylation of TSGs) may independently or cooperatively offer proliferation or survival advantages to cancer cells, these processes may be constrained around certain genes or pathways (e.g., inactivation of TSC2 in the case of patient C10) that are essential to carcinogenesis in certain patients.In our cohort, global hypomethylation was found to associate with significantly higher TMB and CNV burden, as well as AI burden (Fig. 6b–d), consistent with previous reports that global hypomethylation is associated with CIN32 and increased rate of somatic mutations33. These data suggest that methylation aberrations have not only evolved in parallel with genomic aberrations but may have also facilitated accumulating genomic alterations that may have led to more drastic phenotypic changes in IPNs of later stages. Interestingly, a higher level of global hypomethylation was associated with a higher proportion of clonal mutations (Fig. 6e). As the progression of lung precancers into invasive lung ADC predominantly follows a clonal sweeping model with a selective outgrowth of fit subclones5, one plausible explanation for this association is that cells within the precancers with a higher level of global hypomethylation may be prone to accumulate genomic aberrations, which subsequently provide growth advantages to these cell clones to develop into major clones in invasive lung ADC.Cancer evolution results from the accumulation of molecular alterations and is constantly shaped by selection pressure such as anti-tumor immune surveillance and therapeutic interventions. These molecular aberrations such as point mutations may accumulate gradually, a model termed microevolution, or in a punctuated or catastrophic manner through processes such as chromoplexy and chromothripsis, a model termed macroevolution, both of which have been reported in advanced malignancies53,54. Largely due to the lack of appropriate study materials, our understanding of the molecular evolutionary pattern during early carcinogenesis of lung ADC is rudimentary. Our previous study on genomic profiling of lung ADC and its precursors has demonstrated progressive accumulation of somatic mutations from AAH to AIS, MIA and ADC in line with the microevolution model. Meanwhile, there was a distinct increase in CNV burden from AAH to AIS and an increase in AI burden from AIS to MIA. In the current study, we observed a progressive increase of methylation changes from AAH to AIS, MIA, and ADC. However, the overall difference in methylation aberrations appeared to be subtle between IPNs of different histologic stages (Fig. 1b) and methylation evolution from AAH to ADC was similar across different epigenetically defined genomic regulatory regions. These observations suggest that methylation aberrations may have primarily contributed to the stepwise microevolution during early carcinogenesis of these lung ADCs and most methylation changes were stochastic “passengers,” as are the majority of somatic mutations55. On the other hand, these seemingly stochastic genomic and epigenetic alterations may give rise to heterogeneous subclones in precancers with various biological features; therefore, increasing the possibility of establishing fit subclones leading to malignant transformation. These principles may also apply to later neoplastic evolution, including invasion, metastasis, and development of drug resistance, where large-scale genomic sequencing studies have only depicted the underlying molecular mechanisms in a small subset of patients56,57. Comprehensive molecular profiling incorporating genomic, epigenomics, and transcriptomic profiling are warranted in future studies to depict these critical cancer evolutionary processes.Antitumor immune surveillance plays a central role during the initiation and progression of precancers. We have previously reported that the immune microenvironment was suppressed in invasive lung cancers compared to preinvasive cancers or precancers58. In this study, we demonstrated a higher Treg/CD8 ratio in later-stage IPNs (Fig. 7c and Supplementary Fig. 12c), implying a more suppressed T cell infiltrate in later-stage diseases, in line with a concomitant study of immune profiling of the same cohort of IPNs59. These findings are consistent with the concept of immune editing, whereby the immunogenicity of cancer cells evolves under the selective pressure from antitumor immune response, resulting in the emergence of immune-resistant cancer clones in later-stage diseases. Interestingly, global hypomethylation was associated with a higher Treg/CD8 ratio (Fig. 7d). Methylation aberrations may affect antitumor immune surveillance directly by regulating the expression of immune-related genes60 and/or potential neoantigens or indirectly via modifying chromosomal vulnerability for CNV and mutations, both of which are well known to influence the tumor immune microenvironment61,62. However, these impacts are complicated as many processes can affect antitumor immune surveillance both positively and negatively. For example, a high level of global hypomethylation may lead to a high CNV burden known to associate with a cold tumor immune microenvironment61; meanwhile, global hypomethylation is also associated with increased mutation rate, which may increase tumor immunogenicity62. In the end, the selection of cancer cell clones under immune pressure is determined by the cumulative effects of these molecular aberrations and only the cells with the best combination of molecular features including methylation status, mutation and CNV burden will survive and develop into dominant clones in invasive cancers.There has been increasing enthusiasm toward moving interventions successfully applied to metastatic cancers to early-stage cancers and even precancers, a concept called interception63. Compared with invasive cancers, precancers and preinvasive cancers may exhibit less complexity in aberrant molecular landscapes, as well as better preserved immune contextures, and thus may be easier to eradicate. Accordingly, we have launched the IMPRINT-Lung Clinical Trial (NCT03634241), in which patients with high-risk IPNs (many of which may be AAH or AIS) are treated with immune checkpoint inhibitors. In the current study, we demonstrated that DNA methylation aberrations are less complex in precancers and preinvasive lung cancers than in invasive cancers. Therefore, therapeutic agents that can modulate methylation by targeting aberrant methylations and potentially reprogram the immune microenvironment may also have potential in treating precancers and preinvasive cancers to prevent invasive lung cancers.We delineated the evolution of genome-wide DNA methylation during the early carcinogenesis of lung ADC using RRBS profiling on invasive lung ADC precursors of different stages. Rather than aggregated changes from populations of cancer cells measured by methylation array, RRBS assesses DNA methylation heterogeneity of single molecules derived from individual cancer cells, which made it possible to delineate methylation ITH at the allelic level. Furthermore, the WES data from the same IPNs has provided a unique opportunity to depict the relationship between methylation and genomic evolution. Due to the scarcity of IPN materials, our study has inevitable limitations. First, the sample size was relatively small for each histologic stage. Given the substantial heterogeneity between IPNs even within the same stages, our data was not powered to address some essential questions relevant to methylation evolution, such as whether certain methylation aberrations are significantly more common in early-stage IPNs than later-stage lesions representing “dead-end” IPNs. Second, most of these IPNs were very small after pathological assessment, which prevented the acquisition of sufficient data in more patients for in-depth analyses such as multiregion profiling to dissect the interaction between parallel evolution and CIN, transcriptomic profiling to determine the biological impact of observed methylation changes. Although a substantial number of DMRs were associated with TF binding sites, which could potentially regulate the transcriptome of the IPNs, these remained to be speculation without confirmation from transcriptomic data. Third, follow-up time for all patients in this study was relatively short, so we were unable to investigate the impact of these methylation changes on recurrence or survival. Finally, the resected specimens in this study could only provide a single molecular snapshot of the evolutionary process of IPNs. Whether all AAH will evolve into AIS, MIA, and ADC; whether all ADC evolve from AAH; and whether the observed methylation changes in IPNs of different stages represent the true evolutionary dynamics or simply reflect the distinct methylation patterns of IPNs with different malignant potentials is unknown. Deciphering the temporal evolution during neoplastic progression will require specimens obtained over the course of disease progression. Clinical trials collecting longitudinal biopsy specimens, such as IMPRINT-Lung (NCT03634241), may provide such opportunities going forward.MethodsSample acquisitionA total of 53 resected pulmonary nodules and paired normal lung tissues from 39 patients treated at Nagasaki University Hospital or Zhejiang Cancer Hospital between 2014 and 2017 were used in the study. None of the patients received chemotherapy or radiotherapy before surgery. Twenty-nine lung nodules from 15 patients had multiregional specimens for spatial heterogeneity assessment (Supplementary Data 1). WES data were available for all specimens (EGAS00001004960)5. Written informed consent was obtained from all patients. The study was approved by the Institutional Review Boards of MD Anderson Cancer Center, Nagasaki University Graduate School of Biomedical Sciences, and Zhejiang Cancer Hospital.DNA methylation profiling by RRBS and data processingDNA was extracted using the QIAamp DNA FFPE Tissue Kit (Qiagen), and 200 ng–1 μg of DNA was subjected to RRBS for genome-wide DNA methylation profiling64. Briefly, TrimGalore v.0.4.3 was used to trim the Illumina adapter sequences (a minimum of 5 bp in a read was required to overlap with the adapter sequence); then Bismark v.0.18.1 integrating bowtie2 v.2.2.3 was used to align the trimmed reads to the GRCh37 assembly of the human genome. FastQC v.0.11.7 was used for quality control. The DNA methylation levels for individual CpGs were calculated using methyKit (v.1.16.0)65. Methylated reads (containing Cs) and unmethylated reads (containing Ts) at each cytosine site were counted, and the percentage of methylated reads among total reads covering the corresponding cytosine was calculated to quantify the DNA methylome for each sample at the single-base resolution. The CpG sites, DMRs, and loci of known genes, as well as genomic features, including CpG islands, were annotated using the R package “ChIPSeeker (v.1.26.0)”66 and the toolkit “genomation” (v.1.22.0)67, which is based on the “TxDb.Hsapiens.UCSC.hg19.knownGene” annotation database and the UCSC Genome Browser CpG islands table. To avoid bias, we kept CpG sites mapped to the autosome and removed CpG sites overlapping with the single-nucleotide polymorphism (SNP) positions in dbSNP137. DNA methylation was analyzed at either a single-CpG resolution or at genomic region bins, in which DNA methylation values were averaged across 5-kb regions. Promoter methylation was calculated as the averaged DNA methylation values based on GENCODE promoter regions (i.e., 1 kb upstream to 500 bp downstream of the annotated TSSs).Comparison of methylation profiles between different specimensTo assess the DNA methylome profiles of distinct IPNs of different pathological stages and examine the heterogeneity between IPN samples, we first aggregated the DNA methylation levels of 5-kb tiling regions across the genome in each sample by retaining only CpG sites with ≥10 sequencing reads. We then applied principal component analysis to identify global DNA methylation patterns between samples. To evaluate consistent clusters of these DNA methylation profiles, we performed an unsupervised hierarchical agglomerative analysis of CpG sites covered by ≥50 reads across all samples; these reads were based on single CpG methylation calls without any binning. To evaluate the correlation between overall DNA methylation in all samples from each patient, we used a pairwise approach to compare distance and similarity matrices on the basis for all CpGs with coverage of ≥10 reads.Differential DNA methylation analysisDMRs encompassing the differentially methylated CpGs between paired disease and normal tissue samples were identified using triangular kernel to smooth the number of methylated reads and the total number of reads by applying the “noise filter” function in the “DMR caller” (v.1.22.0) R package68. The differentially methylated CpG of paired samples were identified by selecting CpG sites located in DMRs; only CpG sites covered by ≥10 reads in paired samples were included.Partition of genomic regulatory regions based on chromatin states modelChromHMM v.1.21 was applied to build the chromatin states model at 200-bp resolution with default parameters12 on A549 cell line (Encode Broad) for histone marks: H3K04me1, H3K04me2, H3K04me3, H3K09ac, H3K09me3, H3K27ac, H3K27me3, H3K36me3, and H3K79me2. The genomic regions, including promoter regions, enhancer regions, transcribed regions, and repressed regions (heterochromatin), were partitioned based on resulting segments inferred as defined chromatin states, which were deduced from chromatin-state signatures using a multivariate hidden Markov model that explicitly models the combinatorial presence or absence of each histone mark.Repetitive elementsThe repetitive elements were retrieved from UCSC RepeatMasker track (https://genome.ucsc.edu/cgi-bin/hgTrackUi?g=rmsk), which was created using Arian Smit’s RepeatMasker program, through screening human genomic DNA sequences for interspersed repeats and low complexity DNA sequences.Identification of PMDsThe boundary of PMDs was inferred using whole-genome shotgun bisulfite sequencing of lung cells from a healthy donor (GSM983647) applying MethylSeekR (v.1.30.0) package69. Before PMD calling, CpG sites overlapping common SNPs (dbSNP build 137) were removed.Motif identification and prediction of TF binding sitesThe de novo methylated DNA motifs in IPNs of each stage were identified by mEpigram v.0.07, which discovers motifs by using position-specific weight matrices from the k-mers that are most enriched in the positive sequences compared with the negative sequences as “seeds” and extending the motifs in both directions70. The Tomtom tool (v.5.2.0) from the MEME suite was used to select significantly enriched methylated DNA motifs based on the database of known transcript factors (HOCOMOCO_v11)71.Estimation of DNA methylation ITHTo estimate DNA methylation ITH, we applied “methclone” (v.0.1) to identify epigenetic loci whose distributions of epigenetic allele (“epiallele”) clonality differed between paired tumor and normal samples by quantifying the degree to which the compositions of epialleles at given loci in the tumor sample were distinct from those in the normal tissue sample. An epiallele was defined by setting 60 reads in four consecutive CpG sites as the threshold to consider the epigenetic allele composition of the locus. We then calculated the differences in epiallele entropy between each IPN sample and its matched normal tissue sample. Loci with combinatorial entropy changes (ΔS) below −60 between each IPN sample and its paired normal sample were defined as epigenetic shift loci (termed “eloci”). To reduce the bias due to the different coverage for each sample, we then calculated relative epiallele shifts (i.e., the normalized number of eloci) via dividing the number of eloci by the total number of assessed loci in each sample and then multiplying that ratio by the average number of total loci across all samples72. We also assessed epigenetic polymorphism (“epipolymorphism”) to measure the epiallelic diversity in each IPN sample by calculating the frequency of each specific epiallele from multiple stochastic changes in the frequencies of many epialleles73. We calculated 16 epiallele status (0000, 0100, 0010, 0001, 1000, 1100, 0110, 0011, 0101, 1010, 1001, 0111, 1011, 1101, 1110, and 1111, where 1 represents a methylated CpG site and 0 represents an unmethylated CpG site), then epipolymorphism for each loci is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathop {\sum}\nolimits_{i = 1}^{16} {Fi^2}$$\end{document}∑i=116Fi2, where F is the fraction of each epiallele status.Locus overlap analysisWe applied LOLA23 to the genomic regions identified as eloci (ΔS < −60) in all samples of each stage to evaluate the overlap between genomic regions with methylation ITH and chromatin marks. The genomic regions of all loci with ≥ 60 reads were used as background genomic regions, and the selected genomic regions were mapped to a compendium of publicly available histone mark profiles, including CTCF, H2AZ, H3K4me1, H3K4me2, H3K4me3, H3K9ac, H3K9me3, H3K27me3, H3K27ac, H3K36me3, H3K79me2, and H4K20me1 in A549 lung ADC cell line (http://hgdownload.soe.ucsc.edu/goldenPath/hg19/encodeDCC/wgEncodeBroadHistone) and H3K27me3, H3K4me3, H3K9me2, H3K9ac, and CTCF in an immortalized human bronchial epithelial cell line (BEAS-2B; GSE56053)25. P-values in the enrichment analyses were calculated using one-sided Fisher’s exact test. Adjustment for multiple testing with q-value (false discovery rate adjusted p-value) was performed using the Benjamini–Yekutieli method.Construction of phylogenetic treesTo construct methylation-based phylogenetic trees from the RRBS data, we used promoter CpG sites (≥50 reads per CpG site selected for all samples from each patient) with the most variable methylation values (mean absolute deviation >10%) shared by all samples, including a normal tissue sample used as the tree root, to build a Euclidean distance matrix. We built the phylogenetic trees by applying a neighbor-joining algorithm from the “ape” (v.5.4.1) package to independently infer phylogenetic relationships between IPN specimens for each patient from the mutation and methylation profiles. To assess the phylogenetic similarity between the genetic and epigenetic profiles, we employed an independent but parallel distance matrix construction from the mutation and methylation profiles for each IPN, in which the genetic distance was quantified by Hamming distance based on all filtered mutations and methylation distance was quantified by Euclidean distance based on methylation values of all CpG sites (≥20 reads, MAD ≥ 20 for all samples in each IPN), and then calculated the Spearman’s correlation coefficient for all pairwise samples grouped by IPNs.Estimation of global hypomethylationTo determine global methylation levels, we chose CpG sites (covered by ≥10 aligned reads) within evolutionarily young subfamilies of LINE-1 repeat elements (L1HS and L1PA). LINE-1 family annotation was obtained from the RepeatMasker of the UCSC genome browser. We averaged the methylation values of the chosen CpG sites, to represent the global methylation level of each sample.Deconvolution of T cell profilesTo derive tumor-infiltrating T cell subtypes from transcriptomic data previously published36, processed RNA-seq dataset (normalized and log 2 transformed) comprise 197 normal lung tissues, 98 AIS/MIA, and 99 ADC samples were retrieved from EGAS00001004006. ImmuCellAI74 was applied to infer immune cell components for each sample.To derive the infiltration of T cell repertoire from RRBS data, we first obtained a reference methylation signature by retrieving the methylation calls of whole-genome bisulfite datasets from the BLUEPRINT epigenome project, including those for samples of regulatory T cells (EGAX00001343016/EGAX00001236257 in EGAD00001002492) and CD8+ T cells (EGAX00001195937/EGAX00001195943 in EGAD00001002486). Only promoter CpG sites that were covered by ≥5 reads in all samples were retained and binned with a 50-bp window by mean methylation values, and then only 50-bp binned regions that overlap with 50-bp binned regions of CpG sites covered by the RRBS methylation profiles of preneoplastic lesions in all samples were used for reference signature extraction by non-negative matrix factorization implemented in the “MethylCIBERSORT” (v.0.2.0) package. We then performed DNA methylation deconvolution (average methylation level by 50-bp bin) using aforementioned BLUEPRINT signature and the CIBERSORT webserver (https://cibersort.stanford.edu). We performed T cell deconvolution in relative mode, running 100 permutations with quantile normalization disabled75. The resulting immune-cellular fractions were used to compare samples of different pathological stages. The average values of immune cell infiltration for each sample inferred by whole-genome bisulfite analysis from two patients independently were used to infer Treg/CD8 ratio.Statistical analysisViolin plots were created using the “geom_violin” function in R statistical package ggplot2 (v.0.9.1) to represent data point density along the Y-axis, and the “stat_summary” function from ggplot2 (v.0.9.1) was used to calculate the mean as the center point. Differences in DMR numbers, eloci numbers, and immune cell infiltration between IPN specimens of different stages were assessed using the Kruskal–Wallis H test. We used quantile regression to assess the differential entropy values at the 1st percentile (low percentile of entropy was chosen to reflect the loci with biological interest) and epipolymorphism values at the 50th percentile between different groups. We used a two-sided Pearson’s correlation coefficient to compare methylation profiles between two samples and between groups of samples of different stages. We used a two-sided Spearman’s correlation coefficient to determine the extent to which distance matrices were correlated with DNA methylation profiles and somatic mutation profiles.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationDescription of Additional Supplementary FilesSupplementary Data 1-7Reporting Summary | nature communications | [
"Article"
] | [
"Cancer genomics",
"Non-small-cell lung cancer",
"Tumour heterogeneity",
"Epigenomics"
] | cancer leading cause death curable if treated early cancers preceded by precancers Treating precancers prevent invasive lung cancer morbidity mortality strategies prevention challenging limited understanding of neoplastic progression from precancers to invasive lung Lung adenocarcinoma (ADC) common subtype >50% of all lung cancers2. invasive lung ADC evolves from atypical adenomatous hyperplasia progress to ADC in situ preinvasive minimally invasive ADC invasive ADC3 AAH AIS MIA ADC present as pulmonary nodules with ground-glass opacity on tomography scans biology clinical course lesions defined management controversial Surgical resection not standard of care diagnostic yield of biopsy low for ground-glass opacity-dominant nodules4 nodules referred as indeterminate pulmonary nodules (IPNs). surgical resection not often tissue for profiling difficult understanding biology initiated international collaboration for collection characterization of lung precancers preinvasive early invasive ADC IPNsreported genomic landscape intra-tumor heterogeneity evolution from AAH to AIS MIA ADC5 mutations epigenetic alterations DNA methylation impact neoplastic transformation fitness methylation profiling studies alterations silencing tumor contribute carcinogenesis previous study complex methylation ITH larger tumor size increased risk postsurgical recurrence invasive lung Methylation aberrations reported in AAH lesions tumor-adjacent lung tissues early molecular studies analyzed small genes lung carcinogenesis dynamic changes methylation genome level evolutionary trajectories progression lung precancers not studied study unique cohort resected IPNs histologic stages evolution methylation landscape increased methylation ITH in lung ADC global hypomethylation correlates with immune infiltration mutational burden copy number alterations methylation aberrations increase progression sequencing of 62 resected IPNs (14 AAH 15 AIS 22 MIA 11 invasive ADC paired normal lung tissues from 39 patients no significant difference age (p = sex =χ2 smoking status (p = 0.5696 between histologic groups Data 2) nonmalignant cells each IPN specimen reviewed by two lung cancer pathologists estimate purity specimens with 40% premalignant or malignant cells included manual macrodissection for DNA extraction estimated tumor purity in silico-exome sequencing data by ABSOLUTE11 IPNs high genomic ITH tumor no significant difference tumor purity between stages Fig. 1) mean RRBS sequencing coverage 58.98 reads minimum ten reads samples RRBS profiling methylation status at 751,462 CpG sites 15,761 genes DNA methylome of AAH similar normal lung tissue AIS MIA ADC different lung (Fig. unsupervised hierarchical clustering identified two clusters one normal lung AAH AIS MIA ADC spatially separated specimens together interlesion heterogeneity prevalent than intralesion heterogeneity methylation profiles of specimens highly correlated (median coefficient r = 0.884.736–0.969] p < 2.2 × 10−16 Fig. 3)1DNA methylation aberrations increase progression precancers analysis methylation profiles average values 39,148 5-kb tiling regions autosomal non-polymorphism CpG sites 10× read coverage different stages solid dots different colors represent IPNs stages star (*) centers specimens each stage genome-wide DNA methylation correlation between IPNs different stages normal tissue same patients yellow-purple dots represent CpG sites covered IPN specimens (n = 14 AAH 11 AIS 18 MIA 10 ADC normal lung Two-tailed Pearson’s correlation coefficient r values top. P < 2.2 × 10−16 all four histologic stages red dots represent CpG sites hypermethylation green dots sites hypomethylation number CpG sites overlapping DMRs hypermethylation hypomethylation different histologic stages green dots represent mean normalized numbers CpG sites overlapping DMRs each blue dots represent mean numbers CpG sites overlapping DMRs each 95% confidence interval error differences stages assessed Kruskal–Wallis H test Source data provided fileincrease CpG sites normal lung tissues AAH to AIS MIA ADC Hypermethylation early AAH hypomethylation AIS MIA ADC correlation coefficients between methylation profiles IPNs paired normal tissues decreased (r = 0.885 AAH–normal AIS–normal 0.796 MIA–normal 0.740 ADC–normal p < 2.2 × 10−16 Fig 1b). decreases marginal heterogeneity patients methylation changes subtle during early lung carcinogenesis decreases represent methylation alterations in CpG sites quantification differentially methylated regions IPNs later-stage IPNs more CpG sites hypermethylation (p = 0.3635 more hypomethylation (p = 0.000287 (Fig. 1c).Evolution methylome similar trend across genomic regulatory epigenome regions regulatory elementsmethylation evolution genomic regions partitioned regions into promoters enhancers transcribed repressed based on histone marks (H3K04me1 using ChromHMM12 calculated DMRs with regions Fig. 2a–d DMRs higher in later-stage IPNs regions selected DMRs overlapping with repetitive elements Later-stage IPNs increased DMRs in repetitive nonrepetitive regions (Fig. 2e f). partially methylated domains (PMDs major source DNA methylation variation in cancer quantified DMRs inside PMDs versus outside more in later-stage IPNs inside outside PMDs (Fig. 2g h). results majority methylation alterations during early carcinogenesis lung ADC stochastic across all genomic regulatory regions.Fig. number DMRs in IPNs different stages by genomic regulatory regions PMDsDMRs shown in promoters enhancers transcribed repressed repetitive non-repetitive inside outside blue dots represent normalized numbers DMRs IPNs each histologic stage 95% confidence interval error bars differences stages assessed Kruskal-Wallis H test Source data provided file.Enrichment transcription factor (TF) binding sites DMRsDNA methylation DNA binding TFs target sequence identified 50 motifs AAH 57 AIS 64 MIA 66 ADC enriched over DMRs each histologic stage motifs aligned with sites TFs critical biological processes cell cycle progression cell proliferation apoptosis tumor metastasis angiogenesis immune Motifs BHE40 BMAL1 CLOCK EPAS1 MAX MITF MXI1 MYC MYCN TFE3 USF1 USF2 enriched at DMRs across all stages methylation changes sites early events before precancersmotifs enriched in DMRs later-stage IPNs including ATF3 E2F1 E2F3 EGR1 KLF1 MYOG PATZ1 SP1 SP2 SP3 SP4 TFDP1 ZBT14 ZIC1 ZN281 ZN335 ADC ASCL1 HEN1 KLF9 SNAI2 ADC MIA SNAI1 TFEB ADC MIA AIS not AAH methylation changes CpG sites later processes carcinogenesis.Later-stage disease higher methylation tumor methylation associated clinical outcome invasive ADC8 evolution methylation ITH lung precancers calculated “epiallele shifts loci>60 each IPN specimen normal lung tissues MIA invasive ADC showed more pronounced epiallele shifts than AAH AIS higher level methylation ITH later stages (Fig. 3a). abundance of eloci distinct epiallele shifts) higher in later-stage IPNs (p = 0.004567on 15 patients abundance eloci higher in later-stage than early-stage in 13 15 4) suggesting later-stage IPNs higher methylation ITH some patients (C23 J43 had high methylation ITH in early-stage IPNs postoperative follow-up short majority well without recurrence biological significance of higher methylation ITH early-stage IPNs calculated frequency of 16 combinatory DNA methylation patterns at four consecutive CpG sites later-stage IPNs showed higher epiallele diversity ADCs had higher epipolymorphism at eloci than MIA AIS AAH DNA methylation status in later-stage IPNs.Fig. 3Increased DNA methylation ITH in later-stage IPNs Cumulative distribution curves of epiallele shifts DNA methylome in AAH AIS MIA ADC specimens compared to normal lung 1st percentile entropy −21.46 for AAH −28.88 AIS −30.75 MIA −42.01 ADC abundance of eloci among different histologic stages green dot mean normalized number elociblue dots represent mean eloci IPN each stage 95% confidence interval error differences stages assessed Kruskal–Wallis H test cumulative distribution curves epipolymorphism epiallele shifts (ΔS < −60) IPN specimens AAH AIS MIA ADC 50th percentile epipolymorphism 0.295 AAH 0.362 AIS 0.359 MIA 0.463 ADC regression X-axis epipolymorphism Y-axis cumulative epipolymorphism specimens Source data methylation ITH genomic regulatory regions Fig. 4a loci heterochromatic regions show pronounced epiallele shifts higher methylation ITH repressed chromatin more eloci inside PMDs outside stages PMDs epigenetically CpG sites prone develop methylation ITH PMDs associate repressive chromatin domains gene-poor regions low higher methylation ITH tolerated PMDs. 4Comparison methylation ITH epiallele shifts genomic regulatory regions PMDs Cumulative distribution curves epiallele shifts IPN normal lung tissues loci promoters enhancers transcribed repressed regions 1st percentile entropy.genomic regions −30.87 promoters −38.12 enhancers −42.44 transcribed −45.48 repressed −29.51 repetitive regions quantile regression distribution epiallele shifts IPN specimens normal lung tissues loci inside outside 1st percentile entropy −29.10 inside −20.73 outside AAH −37.60 −28.05 outside AIS −44.74 inside −29.25 outside MIA −50.82 inside −41.25 outside ADC regression PMDs derived WGBS profiling normal lung Source data examined distance eloci nearest transcription start sites majority eloci closer TSSs invasive ADC AAH AIS MIA Fig. 6) potential association methylation ITH histone modification performed locus overlap analysis (LOLA genomic regions eloci overlap methylation ITH diverse histone posttranslational LOLA calculates overlapping nonoverlapping regions significanceadjustment false discovery rate more eloci later-stage IPNs MIA ADC higher incidences genomic regions H3K27me3 H3K9me3 H3K9me2 histone modifications associated with transcriptional AAH AIS histone modifications cooperate DNA methylation alterations lung precancers29.Genomic methylation evolution parallel early lung relationship epigenome genome lung ADC constructed trees five patients (C5 J7 J8 J9 J43) four spatially separated specimens included structure methylation-based phylogenetic trees similar trees (Fig. genomic distance somatic mutations correlated with methylation distance specimens same IPNs parallel evolution early carcinogenesis lung four out of five patients multiregional specimens (C5 J7 J8 J9 J43) strong correlation between methylation distance genomic distance exception patient J9 7)J9 never smoker lowest genomic aberration burden allelic imbalance lowest copy number variation highest genomic ITH lowest trunk mutations among five IPNs similar methylation aberration burden other four IPNs (Supplementary Data 6).Fig. evolutionary relationship genomic methylation landscape.Phylogenetic trees mutations methylation values in patient C5 J7 J8 J9 J43 length branch indicates similarity mutation methylation profiles between two separated specimens patients minimum four separated specimens included Correlation genetic distance methylation distance between separated specimens assessed by two-tailed Spearman’s correlation analysis (p = 2.62 × 10−38) dot represents normalized distance Source data file patient C10 promoter hypermethylation of TSC2 TSG cell growth identified in specimens copy number loss identified in AAH specimens Similar phenomena different TSGs observed in other patients (Supplementary Data 7) data suggested convergent evolution same genes activated inactivated different mechanisms cancer cell clones during lung cancer development progressionhypomethylation ITH associated with increased chromosomal instability (CIN in precancerous IPNsAs methylation chromosomal structure DNA mutagenesis global hypomethylation with CIN increased mutational rates in cancers32 early carcinogenesis lung ADC assessed global methylation using transposable elements-1 (LINE-1) surrogate marker for global DNA methylation34 decrease of LINE-1 methylation in MIA ADC to normal lung tissues increased global hypomethylation in later stages methylation level LINE-1 inversely correlated with CNV burden AI burden TMB higher CIN LINE-1 methylation inversely correlated with clonal mutations significant positive correlation between abundance of eloci CNV burden AI burden higher methylation increased CIN. 6Correlation of LINE-1 methylation with genomic features in IPNs different stages LINE-1 methylation level green dot LINE-1 methylation level blue dots LINE-1 methylation level 95% confidence intervals differences among stages assessed using Kruskal–Wallis H testCorrelation between LINE-1 methylation levels genes with copy number changes events allelic imbalance mutation burden proportion clonal mutations assessed by Spearman’s correlation analysis Each dot represents IPN specimen Source data file complex interaction between genome epigenome categorized CpG sites in chromosomal regions CNV status methylation analyses Fig 9 evolutionary patterns methylation similar CpG sites regions copy number gain loss neutral plotted cumulative distribution of epiallele shifts of AAH AIS MIA ADC using CpG sites copy number-neutral regions later-stage IPNs higher distribution than early-stage regenerated methylation-based phylogenetic trees using CpG sites copy number-neutral chromosomal regions 11 patterns almost identical to all CpG sites results most methylation changes early molecular events before copy number changes patterns inherited without changes generation hypomethylation associated with suppressed T cell infiltrationCancer evolution shaped by interaction between cancer cells host factors immune responseT cells’ role antitumor depicted cell infiltration AIS MIA invasive ADC RNA-sequencing data ImmuCellAI independent cohort analysis increase CD4+ T cells = 4.25e − 27 decrease CD8+ T cells p = 0.1374 normal lung to AIS/MIA invasive ADC higher Treg/CD8 ratio (p = 3.17e − 31 applied MethylCIBERSORT37 T cell infiltration IPNs cohort higher infiltration Tregs p = lower CD8+ T cells later-stage IPNs statistical significance p = higher Treg/CD8 ratio 7c p = 0.00194) increased Treg/CD8 ratio suppressed antitumor immune suppressive immune microenvironment later-stage IPNs Treg/CD8 ratio inversely correlated with LINE-1 methylation level association global hypomethylation immunosuppression. 7T cell infiltration IPNs stages T regulatory cells CD8+ T cells Treg/CD8 ratio stagesgreen dot mean immune cell fraction IPN blue dots immune fraction stage 95% confidence intervals error bars differences stages assessed Kruskal–Wallis H test Correlation between LINE-1 methylation Treg/CD8 ratio assessed by Spearman’s correlation analysis Each dot represents specimen Source data provided file methylation landscape studied in methylation aberrations in precancers poorly defined lack of specimens studies demonstrated changes in DNA methylation in AAH AIS ADC9 small loci primitive technology lack of genomic data critical questions methylation evolution early carcinogenesis not addressed unique collection resected IPNs delineated DNA methylome in lung precancers preinvasive ADC invasive ADC results increase hypermethylation hypomethylation later-stage diseases hypermethylation early AAH hypomethylation after AIS hypermethylation preceded hypomethylation early carcinogenesis.Different cells tumor exhibit different molecular phenotypic features ITH tumor evolution diverse evolve with neoplastic Methylation ITH observed in advanced malignancies higher levels inferior clinical outcomes8few reports on methylation ITH in precancers methylation ITH in Barrett esophagus ADC associated risk malignant current study higher level methylation ITH in later-stage IPNs early stages (Fig. 3 4 results with advanced malignancies complex methylation ITH aggressive tumor entities methylation ITH linked to evolutionary plasticity higher level ITH higher probability to survive progress eloci more abundant around TSSs in ADC than AAH AIS MIA somatic methylation ITH stochastic early lung carcinogenesis methylation aberrations convey survival growth advantages selection cells with higher densities eloci around TSSs.Parallel evolution of genome methylome reported in advanced malignancies lung current study similar patterns correlated genetic methylation distances in IPNs different stages (Fig. 5) genetic alterations DNA methylation changes evolve parallel during early carcinogenesis lung ADCs promoter hypermethylation copy number loss mutations of TSGs identified in independent IPNs same patients mutations cancer genes different regions convergent evolution genetic eventsloss TSGs or methylation changes may offer proliferation survival advantages to cancer cells constrained around genes pathways inactivation of TSC2 essential to carcinogenesis cohort global hypomethylation with higher TMB CNV AI burden (Fig. 6b–d), consistent with with CIN32 increased somatic mutations33 suggest methylation aberrations evolved with genomic aberrations facilitated genomic alterations to drastic phenotypic changes in IPNs later stages higher global hypomethylation with higher clonal mutations (Fig. progression of lung precancers into invasive lung ADC follows clonal sweeping model selective outgrowth of subclones5 explanation cells with higher global hypomethylation accumulate genomic aberrations provide growth advantages.Cancer evolution results from accumulation molecular alterations shaped by selection pressure anti-tumor immune surveillance therapeutic interventions molecular aberrations mutations may accumulate gradually or through chromoplexy chromothripsis macroevolution reported in advanced malignancies53 lack of study materials understanding of molecular evolutionary pattern during early carcinogenesis of lung ADC rudimentaryprevious study lung ADC demonstrated accumulation somatic mutations from AAH to AIS MIA ADC increase CNV burden from AAH to AIS AI burden AIS to MIA current study observed increase methylation changes from AAH to AIS MIA ADC difference methylation aberrations subtle between IPNs histologic stages (Fig. 1b methylation evolution to ADC similar across genomic regulatory regions suggest methylation aberrations contributed to stepwise microevolution early carcinogenesis lung ADCs methylation changes stochastic “passengers majority stochastic genomic epigenetic alterations may heterogeneous subclones in precancers possibility subclones malignant transformation principles apply to later neoplastic evolution invasion metastasis drug resistance-scale genomic sequencing studies small Comprehensive molecular profiling genomic transcriptomic profiling warranted future studies cancer.Antitumor immune surveillance initiation progression precancers immune microenvironment suppressed in invasive lung cancers study demonstrated higher Treg/CD8 ratio in later-stage IPNs (Fig. 7cimplying suppressed T cell infiltrate later-stage diseases with study immune profiling cohort findings consistent with immune editing immunogenicity cancer cells evolves under pressure antitumor immune response immune-resistant cancer clones later-stage global hypomethylation higher Treg/CD8 ratio (Fig. Methylation aberrations affect antitumor immune surveillance expression immune-related potential neoantigens modifying chromosomal vulnerability for CNV mutations tumor immune impacts complicated surveillance high global hypomethylation high CNV burden global hypomethylation increased mutation rate tumor immunogenicity62 selection of cancer cell clones under immune pressure determined by effects molecular aberrations cells with best combination molecular features methylation status mutation CNV burden survive develop dominant clones in invasive cancers increasing enthusiasm toward moving interventions metastatic cancers to early-stage cancers precancers concept interception63 less complexity aberrant molecular landscapes better preserved immune contextures easier to eradicatelaunched IMPRINT-Lung Clinical Trial (NCT03634241) patients with high-risk IPNs AAH or AIS treated with immune checkpoint inhibitors study demonstrated DNA methylation aberrations less complex in precancers preinvasive lung cancers therapeutic agents modulate methylation methylations reprogram immune microenvironment prevent invasive lung cancers delineated evolution of genome-wide DNA methylation during early carcinogenesis of lung ADC using RRBS profiling on invasive lung ADC precursors RRBS assesses DNA methylation heterogeneity of single molecules cancer cells methylation ITH at allelic level WES data from IPNs relationship between methylation genomic evolution scarcity of IPN materials study limitations sample size small for each histologic stage substantial heterogeneity between IPNs data not address essential questions methylation evolution aberrations common in early-stage IPNs later-stage IPNs small after pathological assessment prevented sufficient data patients for in-depth analyses multiregion profiling transcriptomic profiling impact methylation changesDMRs associated TF binding sites regulate transcriptome IPNs speculation without confirmation transcriptomic data follow-up time short investigate impact methylation changes on recurrence survival resected specimens molecular snapshot evolutionary AAH into AIS MIA ADC AAH methylation changes evolutionary dynamics methylation patterns unknown temporal evolution neoplastic progression specimens Clinical trials longitudinal biopsy specimens IMPRINT-Lung may provide opportunities 53 resected pulmonary nodules normal lung tissues from 39 patients Nagasaki University Hospital Zhejiang Cancer Hospital 2014 2017 used received chemotherapy radiotherapy before surgery Twenty-nine lung nodules from 15 patients multiregional specimens for spatial heterogeneity assessment WES data available for specimens Written informed consent obtained from patients approved by Institutional Review Boards MD Anderson Cancer Center Nagasaki University Graduate School Biomedical Sciences Zhejiang Cancer Hospital methylation profiling RRBS processingDNA extracted DNA Tissue Kit 200 ng–1 μg DNA subjected RRBS for genome-wide methylation TrimGalore v.0.4.Illumina adapter sequences 5 bp overlap Bismark v.0.18.1 bowtie2 v.2.2.3 trimmed reads GRCh37 assembly human genome FastQC v.0.11.7 quality control DNA methylation levels CpGs calculated methyKit (v.1.16 Methylated unmethylated cytosine site counted percentage calculated DNA methylome single-base CpG sites DMRs loci genes genomic features islands annotated R package “ChIPSeeker (v.1.26 toolkit “genomation” (v.1.22.0 based “TxDb.Hsapiens.UCSC.hg19.knownGene” annotation database UCSC Genome Browser CpG islands table bias kept CpG sites mapped autosome removed overlapping single polymorphism (SNP positions dbSNP137 DNA methylation analyzed single-CpG resolution genomic region bins averaged 5-kb regions Promoter methylation calculated averaged GENCODE promoter regions1 kb upstream to 500 bp downstream annotated TSSs).Comparison methylation profiles DNA methylome profiles IPNs pathological stages heterogeneity IPN samples aggregated DNA methylation levels 5-kb tiling regions retaining CpG sites ≥10 reads applied component analysis global DNA methylation patterns evaluate performed hierarchical agglomerative analysis CpG sites ≥50 reads samples based single CpG methylation calls binning correlation DNA methylation samples pairwise approach distance similarity matrices CpGs ≥10 reads.Differential DNA methylation analysisDMRs differentially methylated CpGs paired disease normal tissue samples identified triangular kernel “noise filter” function “DMR caller” differentially methylated CpG paired samples identified selecting CpG sites DMRs CpG sites ≥10 reads includedPartition genomic regulatory regions chromatin states modelChromHMM v.1.21 chromatin states model 200-bp resolution default parameters12 A549 cell line (Encode Broad) histone marks H3K04me1 genomic regions promoter enhancer transcribed repressed partitioned defined chromatin states deduced chromatin-state signatures multivariate hidden Markov model combinatorial presence histone mark.Repetitive retrieved UCSC RepeatMasker track screening human genomic DNA sequences interspersed repeats low complexity DNA sequences.Identification PMDs inferred whole-genome shotgun bisulfite sequencing lung cells healthy donor (GSM983647) MethylSeekR (v.1.30.0) PMD calling CpG sites overlapping common SNPs build 137) removed.Motif identification prediction TF binding de novo methylated DNA motifs IPNs stage identified mEpigram v.0.07 discovers motifs position-specific weight matrices k-mers positive sequences negative motifsTomtom tool (v.5.2.0) MEME suite enriched methylated DNA motifs based database transcript factors (HOCOMOCO_v11)71.Estimation DNA methylation applied “methclone” (v.0.1) identify epigenetic loci between tumor normal samples epialleles tumor normal tissue epiallele defined 60 reads in four consecutive CpG sites threshold epigenetic allele composition calculated differences in epiallele entropy between IPN sample normal tissue sample Loci with combinatorial entropy changes) below −60 defined as epigenetic shift loci “eloci”). reduce bias calculated relative epiallele shifts normalized number eloci via dividing eloci by total assessed loci multiplying ratio by average total loci across all samples72 assessed epigenetic polymorphism epiallelic diversity IPN sample calculating frequency each epiallele from multiple stochastic changescalculated 16 epiallele status (0000 0100 0010 0001 1000 1100 0110 0011 0101 1010 1001 1101 1111 1 methylated CpG site 0 unmethylated CpG epipolymorphism each loci defined\documentclass[12pt{minimal{amsmath-69pt\nolimits = 1{16^2}}∑i=116Fi2 F each epiallele status.Locus overlap applied LOLA23 genomic regions eloci (ΔS < −60) samples stage evaluate overlap methylation ITH chromatin marks loci with ≥ 60 reads background selected mapped to histone mark profiles CTCF H2AZ H3K4me1 H3K4me2 H3K4me3 H3K9ac H3K79me2 H4K20me1 A549 lung ADC cell line (http://hgdownload.soe.ucsc/goldenPath/hg19/encodeDCC/wgEncodeBroadHistone H3K27me3 H3K4me3 H3K9me2 H3K9ac CTCF immortalized human bronchial epithelial cell line (BEAS-2B GSE56053)25 P-values enrichment analyses calculated one-sided Fisher’s exact test Adjustment multiple testing q-value discovery rate adjusted p-value Benjamini–Yekutieli method.Construction phylogenetic methylation-based trees RRBS data used promoter CpG sites (≥50 reads per variable methylation values deviation >10% normal tissue sample Euclidean distance matrix built trees neighbor-joining algorithm infer phylogenetic relationships IPN specimens assess phylogenetic similarity genetic epigenetic profiles independent parallel distance matrix construction genetic distance quantified Hamming distance methylation distance Euclidean distance CpG sites (≥20 reads MAD ≥ 20 calculated Spearman’s correlation coefficient pairwise samplesEstimation global chose CpG sites ≥10 reads young subfamilies LINE-1 repeat elements (L1HS LINE-1 family annotation RepeatMasker UCSC genome browser averaged methylation values CpG sites global methylation level.Deconvolution T cell derive tumor-infiltrating T cell subtypes transcriptomic data processed RNA-seq dataset 197 normal lung tissues 98 AIS/MIA 99 ADC samples EGAS00001004006. ImmuCellAI74 applied infer immune cell components infiltration T cell repertoire RRBS data obtained reference methylation signature methylation calls whole-genome bisulfite datasets BLUEPRINT epigenome project regulatory T cells CD8+ T promoter CpG sites covered ≥5 reads retained binned 50-bp window mean methylation values 50-bp binned regions overlap CpG sites methylation profiles used reference signature extraction non-negative matrix factorization “MethylCIBERSORT” performed DNA methylation deconvolution level 50-bp bin BLUEPRINT signature CIBERSORT webserverperformed T cell deconvolution 100 permutations quantile normalization resulting immune-cellular fractions samples pathological stages average values immune cell infiltration inferred whole-genome bisulfite analysis two patients infer Treg/CD8 ratio.Statistical analysisViolin plots created “geom_violin” function R statistical package ggplot2 data point density Y-axis “stat_summary” function calculate mean center point Differences DMR numbers eloci numbers immune cell infiltration IPN specimens stages assessed Kruskal–Wallis H test used quantile regression differential entropy values 1st percentile epipolymorphism values 50th percentile groups used two-sided Pearson’s correlation coefficient compare methylation profiles samples stages two-sided Spearman’s correlation coefficient distance matrices DNA methylation profiles somatic mutation profiles information research Nature Research Reporting Summary.Supplementary Additional Data | 49.1 | 0.889898 |
10.1038/s41467-020-18240-y | PMC7462860 | Fusarium graminearum is a major fungal pathogen of cereals. Here the authors show that F. graminearum secretes an effector, Osp24, that induces degradation of the wheat TaSnRK1α kinase to promote disease while an orphan wheat protein, TaFROG1, can compete with Osp24 for binding to TaSnRK1α and protect it from degradation | Fusarium graminearum is a causal agent of Fusarium head blight (FHB) and a deoxynivalenol (DON) producer. In this study, OSP24 is identified as an important virulence factor in systematic characterization of the 50 orphan secreted protein (OSP) genes of F. graminearum. Although dispensable for growth and initial penetration, OSP24 is important for infectious growth in wheat rachis tissues. OSP24 is specifically expressed during pathogenesis and its transient expression suppresses BAX- or INF1-induced cell death. Osp24 is translocated into plant cells and two of its 8 cysteine-residues are required for its function. Wheat SNF1-related kinase TaSnRK1α is identified as an Osp24-interacting protein and shows to be important for FHB resistance in TaSnRK1α-overexpressing or silencing transgenic plants. Osp24 accelerates the degradation of TaSnRK1α by facilitating its association with the ubiquitin-26S proteasome. Interestingly, TaSnRK1α also interacts with TaFROG, an orphan wheat protein induced by DON. TaFROG competes against Osp24 for binding with the same region of TaSnRKα and protects it from degradation. Overexpression of TaFROG stabilizes TaSnRK1α and increases FHB resistance. Taken together, Osp24 functions as a cytoplasmic effector by competing against TaFROG for binding with TaSnRK1α, demonstrating the counteracting roles of orphan proteins of both host and fungal pathogens during their interactions. | IntroductionComparative genome studies have revealed that fungal pathogens have a cadre of orphan genes that are restricted to a single species or narrow clade. Although the vast majority of them are of unknown functions, these taxonomically restricted orphan genes are thought to play important roles in lineage-specific adaptations1–3. Plant pathogenic fungi may evolve novel orphan genes to facilitate their infection or enhance virulence. Fungal effectors are good examples of orphan genes that have evolved for plant infection as many of them lack homologs in closely related species. To date, effectors of various molecular mechanisms have been identified and characterized in different fungi4,5. Most of them are small secreted proteins that are cysteine-rich but lack a common structural motif. In Magnaporthe oryzae, a model fungal pathogen, whereas apoplastic effectors are usually secreted via the conventional endoplasmic reticulum to Golgi route and accumulated in the apoplast, some cytoplasmic effectors are able to be secreted through the biotrophic interfacial complex (BIC) and translocated into plant cells4,6. The targets of fungal effectors also vary significantly in their functions, including transcription factors, protein kinases, and proteins or compounds that are involved in plant defense, signaling, and metabolic pathways7–10. For example, the Tin2 and Cmu1 effectors of Ustilago maydis target the ZmTTK1 kinase for affecting cell wall lignification and chorismate for affecting the metabolic status of infected plant cell, respectively11–13.Fusarium graminearum is a causal agent of Fusarium head blight (FHB), which is one of the most important diseases of wheat and barley worldwide. It develops compound appressoria and infection cushions for plant penetration14. After penetration, invasive hyphae grow inter-cellularity and intra-cellularly in infected plant tissues and develop bulbous, irregular invasive hyphae that are morphologically distinct from epiphytic hyphae15,16. Infectious growth then spreads from the initial infection site to neighboring spikelets on the same wheat heads via the rachis, resulting in the blight of entire wheat heads. In the past decade, many genes important for different infection processes have been identified in F. graminearum17–20, including the putative FGL1 and FgNahG effector genes. FGL1 encodes a secreted lipase that can release free fatty acids to inhibit innate immunity-related callose formation during wheat head infection21. FgNahG is predicted to encode a secreted salicylate hydroxylase that can covert salicylic acid (SA) to catechol. The FgnahG deletion mutant was reduced in virulence and expression of FgNahG in Arabidopsis reduced its resistance against F. graminearum22. Recently, a non-ribosomal octapeptide, fusaoctaxin A, was identified as a virulence factor that is required by F. graminearum for cell-to-cell invasion in wheat coleoptiles23.Besides causing severe yield losses, F. graminearum is a producer of the trichothecene mycotoxin deoxynivalenol (DON)24,25. As an inhibitor of eukaryotic protein synthesis, DON is also phytotoxic and an important virulence factor. The TRI5 gene, which is essential for DON biosynthesis, is expressed in infection cushions during early stages of infection14. The tri5 deletion mutant is defective in spreading from the infected florets to other spikelets on the same head via the rachis26 and transgenic plants expressing a UDP-glucotransferase gene that glycosylates DON are increased in resistance against FHB27. Overexpression of TaFROG that encodes a Pooideae-specific orphan protein induced by DON treatment also increased resistance against F. graminearum28. In wheat, TaFROG interacts with the TaSnRK1α protein kinase28 and a NAC transcription factor29. Although germplasm with complete resistance or immunity to F. graminearum is lacking, more than 50 quantitative trait loci (QTLs) conferring various degrees of FHB resistance have been identified in wheat. To date, only Fhb1, a QTL that confers resistance to pathogen spread but not the initial infection, has been characterized at the molecular level. The candidate Fhb1 genes that have been reported include TaPFT encoding a pore-forming toxin-like protein30 and TaHRC-R or HisR encoding a histidine-rich calcium-binding protein31,32. However, the underlying mechanism of these candidate Fhb1 genes for FHB resistance remains to be characterized.The F. graminearum genome is predicted to encode hundreds of orphan genes33 but their roles in the pathogenic interaction and co-evolution of F. graminearum with its host plants remain to be characterized. Because small secretory proteins may function as effectors to interfere with plant immunity during infection34–36 and no cytoplasmic effectors have been identified in F. graminearum, in this study we systematically characterized all the 50 genes encoding orphan secretory proteins (OSPs) and identified OSP24 as an effector that is specifically expressed during plant infection and important for infectious growth in the rachis tissues of infected wheat heads. Transient expression of Osp24 in plant cells suppressed BAX-induced or INF1-induced program cell death. Furthermore, we identified the SNF1-related kinase TaSnRK1α as an Osp24-interacting protein and showed that TaSnRK1α is important for resistance against F. graminearum by RNA silencing or over-expression in transgenic plants. Osp24 may promote the degradation of TaSnRK1α through the ubiquitin-26S proteasome system37. TaFROG competed against Osp24 for binding to the same region of TaSnRK1α. Overexpression of TaFROG in wheat stabilized TaSnRK1α and increased resistance against F. graminearum. Overall, results from this study showed that the OSP Osp24 in F. graminearum functions as a cytoplasmic effector targeting TaSnRK1α for degradation but the wheat orphan protein TaFROG competes with Osp24 for binding with TaSnRK1α and prevents its degradation, indicating the evolving and active adoption of orphan proteins in the arms race between the pathogen and its host.ResultsIdentification and characterization of OSP genes in F. graminearumAmong all the predicted protein-coding genes in the F. graminearum genome, 971 of them (~7.3%) were identified as orphan genes using the bioinformatics analyses described in the “Methods” section. On average, these orphan genes encode smaller proteins than genes conserved in other species (Fig. 1a) and tend to be less transcribed or transcribed at lower levels (Fig. 1b) based on published RNA-seq data20,30. Fifty of them were predicted to encode proteins with signal peptides (SPs) and named OSP genes in this study. For a number of these OSPs, the mature peptides have fewer than 100 amino acid (aa) residues and are cysteine-rich (Fig. 1c), which is similar to the general characteristics of fungal effector proteins4,5.Fig. 1Characterization of orphan genes encoding secreted proteins in Fusarium graminearum.a Comparative analysis of the length of proteins encoded by orphan and non-orphan genes with the Wilcoxon rank-sum test (p < 2.2e−16). b Comparative analysis of the expression levels of orphan and non-orphan genes with the Wilcoxon rank-sum test (p < 2.2e−16). For each gene, the maximum gene expression level was estimated from its transcripts per kilobase million (TPM) in RNA-seq data of conidia, vegetative hyphae, perithecia, and infected wheat heads. c Length and Cys (cysteine) content (%) of mature orphan secretory proteins (OSPs). The three OSPs important for virulence are in red. d Representative images of 3-day-old PDA cultures and 8-dpf (8 days post-fertilization) mating plates of the wild-type strain PH-1 (WT) and the osp24, osp25, and osp44 deletion mutants. e Representative images of wheat heads infected with marked strains were photographed at 14 dpi. f The disease index of WT and three osp mutants. Error bar represents standard deviation (SD) from mean (marked with black dots on the bars) of three independent experiments (n = 3) with at least 10 wheat heads examined in each experiment. Different letters indicate significant differences based on ANOVA analysis followed by Duncan’s multiple range test (P = 0.05). g Mean and standard deviations of DON levels in diseased wheat spikelets inoculated with WT or three osp mutants based on data from three biological replicates (n = 3). No significant differences was observed based on ANOVA analysis followed by Duncan’s multiple range test (P = 0.05). h Distribution of the OSP genes (blue vertical bars) on the four chromosomes of F. graminearum. Red bars and arrowheads indicate the positions of OSP24, OSP25, and OSP44. i Expression levels of the indicated OSP genes based on their TPMs in RNA-seq data of conidia and infected wheat heads20 sampled at 1-dpi, 2-dpi, or 3-dpi. Error bar represents SD from three biological replicates (n = 3).To determine their functions, gene replacement mutants were generated for the 50 OSP genes by the split-marker approach in the wild-type strain PH-1 (Supplementary Table 1, Supplementary Fig. 1). All the resulting mutants were normal in vegetative growth, conidiation, and sexual reproduction (Fig. 1d; Supplementary Table 2). In wheat head infection assays, most of the mutants were normal in virulence (Supplementary Table 2). However, mutants deleted of three individual OSP genes, OSP24 (FGSG_11564), OSP25 (FGSG_11647), and OSP44 (FGSG_13464), were significantly reduced in virulence in repeated infection assays with wheat heads of cultivar Xiaoyan 22 (Fig. 1e). On average, the disease index of the osp24, osp25, and osp44 deletion mutants was reduced 51%, 37%, and 36%, respectively, in comparison with that of PH-1 (Fig. 1f). The osp24, osp25, and osp44 mutants also were reduced in virulence in infection assays with wheat cultivar Zhoumai 3638 (Supplementary Fig. 2), indicating that the role of OSP24 in plant infection is not cultivar-specific. When assayed with the inoculated spikelets sampled at 14 dpi, DON production was normal in these three mutants (Fig. 1g). The osp24 mutant that had the lowest virulence was selected to assay for TRI5 gene expression. In the inoculated spikelets, the transcription level of TRI5 also was similar between the wild type and osp24 mutant (Supplementary Fig. 3). Interestingly, the OSP24, OSP25, and OSP44 genes all are located in the sub-telomeric regions (Fig. 1h), which may facilitate their rapid evolution39. Moreover, they were all specifically expressed or significantly up-regulated in infected wheat heads (Fig. 1i; Supplementary Table 3). These results indicate that the secreted Osp24, Osp25, and Osp44 proteins are important for the full virulence of F. graminearum, possibly by functioning as effectors during plant infection.The osp24 mutant is defective in infectious growth in the rachis tissuesBecause the osp24 deletion mutant was most significantly reduced in virulence, we selected it for further characterization. In infection assays with corn silks, tomatoes and Arabidopsis floral tissues, the osp24 mutant was normal in virulence (Supplementary Fig. 4), suggesting a host-specific role of Osp24 during plant infection. To further characterize its function during wheat infection, the osp24 mutant was assayed for the formation of infection cushions and growth of invasive hyphae. When examined by scanning electron microscopy (SEM), abundant infection cushions were observed on wheat lemma inoculated with the osp24 mutant at 2 days post-inoculation (dpi). In comparison with the wild type, the osp24 mutant had no obvious defects in infection cushion formation (Fig. 2a). Deletion of OSP24 also had no obvious effects on the initial plant penetration and development of invasive hyphae in lemma tissues at 2 dpi (Supplementary Fig. 5). However, extensive hyphal growth to the wild-type level was not observed in the rachis tissues of osp24-infected wheat heads at 5 dpi (Fig. 2b). When assayed by qPCR with genomic DNA isolated from infected wheat heads at 5 dpi, fungal biomass was significantly reduced in wheat heads inoculated with the osp24 mutant in comparison with those inoculated with PH-1 (Fig. 2c). These results indicate that OSP24 plays an important role in infectious growth and spreading in the wheat rachis tissues.Fig. 2Functions of Osp24 and its signal peptide and cysteine residues.a Infection cushions formed by the wild-type strain PH-1 (WT) and osp24 deletion mutant on wheat lemma at 2 dpi were examined by SEM under ×800 amplification. Scale bar, 10 µm. b. Thick sections of infected wheat heads were examined for invasive hyphae (red arrowheads) in the rachis tissues at 5 dpi. Scale bar, 20 µm. c Relative biomass of F. graminearum in infected wheat heads at 5 dpi was determined by qPCR. Mean and standard deviation were estimated with data from three (n = 3) independent biological replicates (marked as black dots on the bar). The asterisk * indicates significant differences (P = 0.05) based on Bootstrap analysis. d The positions of eight cysteine residues in Osp24 and the predicted intra-molecular disulfide bond between C94 and C105. SP, signal peptide. e The yeast suc2 mutant YTK12 and its transformants expressing the empty vector pSUC2 or vectors with the signal peptide from Osp24 and Avr1 (positive control) were assayed for growth on CMD-W or YPRAA plates and invertase activity in TTC medium. f Representative images of wheat heads infected with PH-1, osp24 mutant, and the osp24/OSP24 and osp24/OSP24ΔSP transformants were photographed at 14 dpi. g Wheat coleoptiles were infected with PH-1 expressing the Osp24:mCherry:NLS construct and examined for mCherry signals in plant cells. Scale bar, 20 µm. h Representative images of wheat heads infected with PH-1, osp24 mutant, and transformants of osp24 expressing OSP24 mutant alleles carrying the indicated C-to-A mutations were photographed at 14 dpi. i Western blots of mixtures of equal amounts of total proteins isolated from wheat heads of cultivar Xiaoyan 22 (XY22) and recombinant Osp24-GST, Osp24C94A-GST, or Osp24C105A–GST proteins incubated for the indicated times after the addition of 10 mM ATP were detected with an anti-GST antibody. Detection with an anti-actin antibody was used as a loading control. j Different concentrations (cells/ml) of yeast transformants expressing the indicated bait and prey constructs were assayed for growth on SD-Trp-Leu-His plates and LacZ activity. k Transient expression of Osp24 suppressed programmed cell death triggered by BAX or INF1. At the indicated spots, N. benthamiana leaves were infiltrated with Agrobacterium cells expressing GFP/BAX/INF1 alone or infiltrated with BAX/INF1-expressing cells at 18 h after infiltration with GFP or Osp24 first. Representative leaves were photographed 5 days after infiltration.The SP of Osp24 is required for its secretion and functionThe SP-SUC2 construct was generated by cloning the 22-aa SP of Osp24 (Fig. 2d) into the vector pSUC240 and transformed into the yeast suc2 mutant41. The resulting SP-SUC2 transformants were able to grow on YPRAA agar and had secreted invertase activity (Fig. 2e), indicating that the SP of Osp24 is functional in yeast.To test its function in F. graminearum, we then generated the OSP24 and OSP24ΔSP constructs42,43 and transformed them into the osp24 mutant. The resulting osp24/OSP24 transformants were normal in virulence. However, the osp24/OSP24ΔSP transformants were similar to the osp24 mutant in virulence on wheat heads (Fig. 2f), indicating that the SP is essential for the complementation of osp24 mutant by ectopic expression of OSP24. Therefore, the secretion of Osp24 is important for its function during plant infection.Osp24 is a cytoplasmic effector that is translocated into plant cellsTo determine whether Osp24 is delivered into plant cells, we generated the POSP24-OSP24-mCherry-NLS construct with the NLS from SV40 T-Antigen and transformed it into PH-1. In wheat seedlings infected with transformants expressing the Osp24-mCherry-NLS construct, mCherry signals were observed in the nucleus in coleoptile cells at the inoculation sites (Fig. 2g), suggesting that Osp24 is a cytoplasmic effector that is translocated into plant cells.The C94 and C105 cysteine residues are important for the function of Osp24Similar to many other fungal effectors, Osp24 is a cysteine-rich protein with eight cysteine (C) residues (Fig. 2d). To determine their roles in Osp24 function, constructs of OSP24 with individual Cs changed to alanine (A) were generated and transformed into the osp24 mutant. Whereas mutations of six other cysteine residues had no effect on virulence, the osp24/OSP24C94A and osp24/OSP24C105A transformants, similar to the osp24 mutant, were defective in wheat head infection (Fig. 2h), indicating that C94 and C105 are essential for the function of Osp24 proteins during pathogenesis.When analyzed for the formation of intra-molecular disulfide bonds with the EDBCP tool44, the disulfide bridge between C94 and C105 was predicted with the highest probability (Probability = 0.89774). To characterize their functions, we expressed and purified Osp24-, Osp24C94A-, and Osp24C105A-GST fusion proteins. Equal amounts of these recombinant proteins were co-incubated with total proteins isolated from wheat heads inoculated with PH-1 sampled at 3 dpi and assayed for their degradation by western blot analyses. In comparison with Osp24-GST proteins that were degraded gradually over time, the degradation of Osp24C94A and Osp24C105A GST fusion proteins occurred much more rapidly (Fig. 2i), indicating that Osp24 proteins may be less stable when C94 or C105 is changed to A. Therefore, the intra-molecular disulfide bond between C94 and C105 may be important for the folding and stability of Osp24 proteins.When tested by yeast two-hybrid assays, the Osp24 protein was found to interact with itself (Fig. 2j), suggesting that Osp24 may form homodimers. To determine their roles in the formation of homodimers, we also generated the bait constructs of Osp24 carrying the C94A or C105A mutation. Similar to the Osp24–Osp24 interaction, Osp24 interacted with both Osp24C94A and Osp24C105A in yeast two-hybrid assays (Fig. 2j), indicating that the C-to-A mutation at C94 or C105 had no obvious effect on its intermolecular interaction. Therefore, C94 and C105 are likely only important for the formation of intramolecular disulfide bonds.Osp24 suppresses cell death and the expression of metabolism-related genesTo characterize its role as an effector to suppress host immunity, we first assayed the effect of Osp24 on programmed cell death (PCD) in Nicotiana benthamiana induced by BAX, a mouse proapoptotic protein45. The PCaMV 35S-Osp24ΔSP construct was generated and transformed into Agrobacterium tumefaciens. In N. benthamiana leaves infiltrated with Agrobacterium expressing Osp24ΔSP alone, no cell death was observed. Under the same conditions, PCD was observed on leaves infiltrated with cells expressing BAX. However, on leaves infiltrated with A. tumefaciens expressing Osp24ΔSP 18 h before infiltration with cells expressing BAX, cell death was not observed when Osp24ΔSP and BAX proteins were co-expressed (Fig. 2k, Supplementary Fig. 6), indicating suppression of BAX-induced cell death by Osp24. We also tested the effects of Osp24 on PCD triggered by INF1, a pathogen-associated molecular pattern (PAMP) from Phytophthora infestans46. In infiltration assays with N. benthamiana leaves, INF1-induced cell death also was suppressed by Osp24 (Fig. 2k).To identify genes suppressed or induced by Osp24 during plant infection, we conducted RNA-seq analysis with infected wheat heads that were collected at 3 dpi. A total of 478 differentially expressed genes (DEGs) with at least two-fold changes were identified in wheat infected with the wild type and osp24 mutant. Among them, 296 DEGs were up-regulated in wheat inoculated with the osp24 mutant, suggesting that their expression may be suppressed by Osp24 during F. graminearum infection. FunCat-enrichment analysis showed that up-regulated DEGs were significantly enriched for genes functionally related to metabolism, information pathway, and perception and response to stimuli (Supplementary Fig. 7). Gene Ontology (GO)-enrichment analysis revealed that many GO categories associated with metabolism were significantly enriched, including oxidation–reduction, glycolytic, glycine catabolic, malate metabolic, peroxisome fission, chlorophyll biosynthetic, fructose 2,6-bisphosphate metabolic, fructose metabolic, l-serine metabolic, photosynthesis, and ATP synthesis coupled proton transport processes (Supplementary Data 1). Interestingly, 17 (~7%) of the up-regulated DEGs in osp24-infected wheat heads encode putative NBS-LRR proteins that are known to be involved in plant immunity against microbial pathogens47. Putative NBS-LRR genes were absent in the down-regulated DEGs. It is possible that some of these NBS-LRR genes with up-regulated expression in osp24-infected wheat heads may contribute to defense responses against F. graminearum infection.In comparison with the wild type, only 54 and 83 fungal genes were up-regulated and down-regulated, respectively, in the osp24 mutant during wheat infection. Because none of them are related to known virulence factors or putative effector genes (Supplementary Data 2), their altered expression levels may not contribute significantly to the defect of the osp24 mutant in virulence.Osp24 interacts with wheat TaSnRK1αTo further characterize its function during plant infection, we generated an Osp24 bait construct and screened a yeast two-hybrid library constructed with RNA isolated from infected wheat heads collected at 3 dpi. After screening over five coverage of this library, 16 putative Osp24-interacting clones (OICs) were identified (Supplementary Table 4). Sequencing analysis showed that five clones were the SNF1-related protein kinase gene TaSnRK1α28 that is highly similar to Arabidopsis SnRK1. In Arabidopsis, SnRK1 is a central integrator of stress and energy signaling that regulates plant metabolism, growth, and immunity48,49.The wheat genome has three TaSnRK1α homoelogues (A, B, and D) that share the same AA sequence and have only 12–14 nucleotide differences in the coding regions50. Because all the five TnSnRK1α clones identified in the original yeast two-hybrid library screen were TaSnRK1α-A, we used TnSnRK1α-A for all the experiments related to TaSnRK1α unless otherwise stated. We then generated the full-length TaSnRK1α prey construct and confirmed its interaction with Osp24 in yeast two-hybrid assays (Fig. 3a). To verify their interaction in plant cells, the Osp24-nYFP and TaSnRK1α-cYFP fusion constructs were generated and co-expressed in N. benthamiana leaves by Agrobacterium infiltration. In epidermal cells of infiltrated tobacco leaves, YFP signals were observed in the nucleus but not in the cytoplasm (Fig. 3b), indicating the expression of Osp24-nYFP and TaSnRK1α-cYFP and their interaction in the nucleus of plant cells. We further verified the interaction between Osp24 and TaSnRK1α by in vitro pull-down assays. Recombinant Osp24-GST and TaSnRK1α-HIS proteins were purified from Escherichia coli and co-incubated with glutathione resins. In western blots of proteins eluted from glutathione resins, the TaSnRK1α-HIS band was detected (Fig. 3c), indicating its interaction with Osp24-GST in vitro.Fig. 3Interaction of Osp24 with TaSnRK1α and function of TaSnRK1α in wheat resistance against F. graminearum.a Yeast two-hybrid assays to detect the interaction of Osp24 (Bait) with TaSnRK1α (Prey). Different concentrations of the labeled yeast transformants were assayed for growth on SD-Trp-Leu-His plates and LacZ activity. b BiFC assays for the interaction of Osp24 with TaSnRK1α. Leaves of N. benthamiana were agroinfiltrated with a mixture of A. tumefaciens strains expressing the Osp24-nYFP and TaSnRK1α-cYFP constructs. YFP signals were observed at 2 days post‐agroinfiltration. Infiltration with Agrobacterium expressing the TaSnRK1α or Osp24 construct alone was used as the negative control. No YFP signals was observed in these negative controls. Scale bar, 20 µm. c Verification of the Osp24–TaSnRK1α interaction by GST pull down assays. Western blots of the marked protein mixtures (Input) or proteins co-purified with TaSnRK1α-HIS from these mixtures (GST IP) were detected with the anti-HIS and anti-GST antibodies. d Yeast transformants expressing the TaSnRK1α (prey) and marked full-length or truncated Osp24 (bait) were assayed for growth on SD-Trp-Leu-His plates and LacZ activities. SP signal peptide. e Representative wheat heads of cultivar KN199 and its transgenic lines expressing the TaSnRK1α overexpression (TaSnRK1α OE 5 and 6) or silencing (TaSnRK1α RNAi 2 and 12) construct were drop-inoculated with PH-1 and photographed at 7 dpi. f Mean and standard deviation (SD) of the disease index of PH-1 on the labeled wheat lines were estimated from three independent experiments (n = 3) with at least 10 infected wheat heads in each experiment. Different letters indicate significant differences based on ANOVA analysis followed by Duncan’s multiple range test (P = 0.05). g Thick sections of the rachises of wheat heads of KN199 and its TaSnRK1α OE or TaSnRK1α RNAi transgenic line inoculated with PH-1 were examined for invasive hyphae (red arrowheads) at 5 dpi. Scale bar, 50 µm. h Relative biomass of F. graminearum was determined by qPCR in infected wheat heads of KN199 and labeled transgenic lines sampled at 5 dpi. Mean and standard deviation were estimated with data from three (n = 3) independent biological replicates (indicated with black dots on the bar). *Indicates significant differences (P = 0.05) based on Bootstrap analysis.To determine the region required for its association with TaSnRK1α, three different fragments of Osp24 (Fig. 3d) were amplified and tested in yeast two-hybrid assays. Deletion of the C-terminal region (aa 99–135) of Osp24 eliminated its interaction with TaSnRK1α (Fig. 3d). However, deletion of the N-terminal (aa 1–60) or middle (aa 61–98) region had no obvious effect on the Osp24–TaSnRK1α interaction (Fig. 3d). Therefore, the C-terminal region of Osp24 is involved in its interaction with TaSnRK1α. In Agrobacterium infiltration assays, expression of Ops24 with a truncated C-terminal region failed to suppress BAX-induced or INF1-induced cell death in N. benthamiana (Supplementary Fig. 8). These results indicate that the C-terminal region of Osp24 is related to its function in PCD suppression, likely by interacting with SnRK1.TaSnRK1α contributes to resistance against F. graminearumTo determine its function in resistance against F. graminearum, we first generated a TaSnRK1α RNAi construct and transformed it into wheat cultivar KN19951, which is a cultivar amenable to efficient transformation but more tolerant to FHB than cultivar Xiaoyan 22. In transgenic lines, the expression level of TaSnRK1α was reduced approximately two-fold based on qRT-PCR assays (Supplementary Fig. 9a). In wheat head infection assays, transgenic lines with reduced TaSnRK1α expression displayed more severe FHB symptoms than cultivar KN199 when inoculated with the wild-type strain PH-1 (Fig. 3e). The disease index of PH-1 was increased in lines expressing the TaSnRK1α RNAi construct in comparison with KN199 (Fig. 3f).We also generated transgenic lines with the TaSnRK1α gene under the control of the CaMV 35S promoter52. Increased expression of TaSnRK1α (over 3-fold) was detected in five transgenic lines (Supplementary Fig. 9b). In wheat head infection assays, transgenic plants overexpressing TaSnRK1α were increased in resistance to F. graminearum (Fig. 3e). The disease index of PH-1 on TaSnRK1α overexpressing plants was decreased ~40% compared to that on KN199 (Fig. 3f).Overexpression of TaSnRK1α also significantly reduced infectious growth and only limited invasive hyphae were observed in the rachis tissues of transgenic wheat plants overexpressing TaSnRK1α (Fig. 3g). When assayed by qPCR, less fungal biomass was detected in infected wheat heads in transgenic lines overexpressing TaSnRK1α than in KN199 (Fig. 3h). In contrast, more abundant infectious hyphae were observed in the rachis tissues (Fig. 3g), and more fungal biomass was detected by qPCR (Fig. 3h) in wheat heads of transgenic plants expressing the TaSnRK1α RNAi construct than in the control KN199 plants. Nevertheless, DON production in the diseased spikelets inoculated with PH-1 was similar between KN199 and TaSnRK1α RNAi plants (Supplementary Fig. 10). These results indicate that overexpression of TaSnRK1α in wheat plants increased the resistance against F. graminearum and silencing of TaSnRK1α had the opposite effect.Osp24 accelerates TaSnRK1α degradation during infectionTo assay the effect of Osp24 on TaSnRK1α proteins in a cell-free degradation assay, equal amounts of TaSnRK1α-HIS recombinant proteins were mixed and co-incubated with total proteins isolated from wheat heads of cultivar Xiaoyan 22 inoculated with the wild-type strain PH-1 or osp24 mutant (sampled at 3 dpi). In the presence of ATP, TaSnRK1α was degraded over time (45–90 min) in both co-incubation mixtures (Fig. 4a). However, the rate of TaSnRK1 degradation was reduced in co-incubation mixtures with proteins from osp24-infected wheat heads in comparison with those with proteins from PH-1-infected samples (Fig. 4a). These results suggest that Osp24 may accelerate TaSnRK1α degradation during infection. Under the same conditions, TaSnRK1α degradation was not observed in the co-incubation mixture of TaSnRK1α-HIS and Osp24-GST without proteins from infected wheat heads (Fig. 4b). However, the degradation of TaSnRK1α-HIS was observed in its co-incubation mixture with Osp24-GST and protein extracts from non-inoculated wheat heads (Fig. 4c), indicating that Osp24-stimulated degradation of TaSnRK1α is dependent on some components of wheat head protein extracts.Fig. 4Roles of Osp24 and TaFROG in modulating TaSnRK1α stability.a Western blots of the mixtures of equal amount of TaSnRK1α-HIS and total proteins isolated from wheat heads of cultivar Xiaoyan22 inoculated with the wild type PH-1 (XY22 + WT) or osp24 mutant (XY22 + osp24) incubated for the indicated times after the addition of 10 mM ATP were detected with the anti-HIS or anti-actin (loading control) antibody. b Western blots of the mixtures of equal amount of TaSnRK1α-HIS and Osp24-GST incubated for the indicated times were detected with an anti-HIS or anti-GST antibody. Degradation of TaSnRK1α-HIS by Osp24 was not observed. c Recombinant TaSnRK1α-HIS and Osp24-GST proteins were incubated with equal amounts of total proteins isolated from wheat head (XY22) in the presence of 10 mM ATP and used for western blot analyses with the indicated antibodies. d Degradation of TaSnRK1α by total proteins isolated from wheat head inoculated with PH-1 (XY22 + PH-1) was inhibited by proteasome inhibitor MG132. e Western blots of the indicated protein mixtures (Input) or proteins co-purified with TaSnRK1α-HIS from these protein mixtures (HIS IP) were detected with the anti-20S, anti-RBX1, anti-HIS, and anti-GST antibodies. f Yeast transformants expressing the indicated Osp24 or TaFROG bait construct and prey constructs of the N-terminal (1–266 aa) or C-terminal (267–499 aa) region of TaSnRK1α (TaSnRK1α-N and TaSnRK1α-C) were assayed for growth on SD-Trp-Leu-His plates and LacZ activity. g Western blots of the indicated protein mixtures (Input) or proteins co-purified with Osp24-GST from these mixtures (GST IP) were detected with the anti-GST, anti-HIS, or anti-S-tag antibody. The amount of TaSnRK1α-HIS proteins co-immunoprecipitated with Osp24-GST was reduced by the addition of increasing concentrations of TaFROG-S-tag proteins. h Representative wheat heads of cultivar KN199 and its TaFROG overexpressing transgenic line (TaFROG OE) infected with PH-1 were photographed at 8 dpi. i Mean and standard deviation (SD) of the disease index of PH-1 on the labeled wheat lines were estimated from three independent experiments (n = 3) with at least 10 infected wheat heads in each replicate. Different letters indicate significant differences based on ANOVA analysis followed by Duncan’s multiple range test (P = 0.05). j Cell-free degradation assays with recombinant TaSnRK1α-HIS incubated with equal amounts of total proteins isolated from wheat heads of KN199 and TaFROG OE transgenic lines infected by PH-1 for the indicated time after addition of 10 mM ATP. k Western blots of the indicated protein mixtures (Input) or proteins co-purified with TaSnRK1α-HIS from these protein mixtures (HIS IP) were detected with the anti-20S, anti-RBX1, and anti-HIS antibodies. For a, c, d, e, j, and k, detection with an anti-actin antibody was used as the control.We also assayed the transcription level of TaSnRK1α during F. graminearum infection. TaSnRK1α transcription was not up-regulated in infected wheat heads in comparison with non-inoculated samples (Supplementary Fig. 11a). Furthermore, wheat plants inoculated with PH-1 or the osp24 mutant had no obvious difference in the abundance of TaSnRK1α transcripts (Supplementary Fig. 11b). These results indicate that Osp24 does not affect the transcription of TaSnRK1α but may stimulate its degradation in infected wheat heads.Osp24 facilitates the interaction of TaSnRK1α with the ubiquitin-26S proteasome systemIn Arabidopsis, SnRK1 interacts with both the SCF ubiquitin ligase complex and 26S proteasome53. Osp24 may promote the degradation of TaSnRK1α through the ubiquitin-26S proteasome pathway54 during fungal infection. To test this hypothesis, we first assayed the effect of proteasomal inhibitor MG132 on TaSnRK1α degradation. Even in the presence of Osp24, the degradation of TaSnRK1α in co-mixtures with proteins isolated from infected wheat heads was suppressed by MG132 (Fig. 4d), indicating the involvement of the 26S proteasome.We then investigated the effects of Osp24 on the interaction between TaSnRK1α and the ubiquitin-26S proteasome by in vitro pull-down assays. TaSnRK1α-HIS proteins were co-incubated with total proteins isolated from wheat heads in the presence or absence of Osp24-GST for 30 min before mixing with anti-HIS beads. Proteins bound to anti-HIS beads were then eluted and assayed for the presence of SCF ubiquitin ligase complex and 26S proteasome with the anti-RBX1 and anti-20S proteasome antibodies37, respectively. With or without the addition of Osp24-GST, both SCF ubiquitin ligase complex and 26S proteasome were detected in proteins co-immunoprecipitated with TaSnRK1α-HIS (Fig. 4e). However, the presence of Osp24 increased the amount of these proteins co-purified with TaSnRK1α-HIS (Fig. 4e). These results indicate that Osp24 may accelerate TaSnRK1α degradation by enhancing its association with the ubiquitin-26S proteasome system.The wheat orphan protein TaFROG competes with Osp24 in binding with TaSnRK1αTaSnRK1α is known to interact with a wheat orphan protein TaFROG, which is induced by DON28. In yeast two-hybrid assays, the C-terminal region of TaSnRK1α (267–499 aa) was found to interact with both Osp24 and TaFROG (Fig. 4f). In contrast, the N-terminal region of TaSnRK1α (1–266 aa) was dispensable for its interaction with Osp24 or TaFROG. The direct interaction between Osp24 and TaFROG was not detected in yeast two-hybrid assays (Supplementary Fig. 12).Because both Osp24 and TaFROG interacted with the same region of TaSnRK1α, it is possible that they may compete with each other for interacting with TaSnRK1α. To test this hypothesis, TaSnRK1α-HIS, Osp24-GST, and TaFROG-S-tag fusion proteins were purified and used in in vitro pull-down assays. When equal amounts of TaSnRK1α-HIS and Osp24-GST proteins were co-incubated and mixed with anti-HIS beads, abundant TaSnRK1α proteins were detected in proteins co-purified with Osp24-GST. With the addition of increasing concentrations of TaFROG proteins to the TaSnRK1α–Osp24 protein mixtures, the amount of TaSnRK1α proteins co-purified with Osp24-GST was gradually reduced in a concentration-dependent manner (Fig. 4g), indicating the competition between TaFROG and Osp24 in binding with TaSnRK1α.Overexpressing TaFROG enhances FHB resistance by stabilizing TaSnRK1αTo determine the role of TaFROG in regulating TaSnRK1α stability during F. graminearum infection, we generated transgenic wheat plants in which the TaFROG gene was overexpressed with the CaMV 35S promoter. When assayed by qRT-PCR, the expression level of TaFROG was increased over 100-fold in five transgenic lines (Supplementary Fig. 13). In wheat head infection assays, fewer spikelets developed FHB symptoms in the TaFROG overexpression transgenic lines than the control KN199 plants (Fig. 4h, i), indicating an increase in resistance against F. graminearum by TaFROG overexpression, which is consistent with an earlier report28. We then conducted cell-free degradation assays with equal amounts of TaSnRK1α-HIS recombinant proteins co-incubated with total proteins isolated from wheats heads of control KN199 and TaFROG overexpression plants infected with PH-1. Although TaSnRK1α degradation was observed in all the reaction mixtures, the degradation rate was significantly reduced in co-incubation mixtures with proteins isolated from transgenic lines overexpressing TaFROG in comparison with those from KN199 plants (Fig. 4j), confirming that overexpression of TaFROG increases the stability of TaSnRK1α. In the same degradation assays, the rate of TaSnRK1α degradation was similar in co-incubation mixtures with proteins isolated from wheat heads infected with PH-1 or the osp24 mutant (Supplementary Fig. 14). Therefore, overexpression of TaFROG may enable its effective protection of TaSnRK1α against Osp24 binding and degradation.TaFROG may compete with Osp24 in binding with TaSnRK1α and reduce its degradation via the ubiquitin-26S proteasome system. To test this hypothesis, TaSnRK1α-HIS proteins were mixed with total proteins isolated from wheat heads infected with PH-1 and anti-HIS beads. Western blots of proteins eluted from anti-HIS beads were then detected with the anti-RBX1 and anti-20s proteasome antibodies. The density of both SCF ubiquitin ligase complex and 26S proteasome bands was weaker in samples with proteins isolated from transgenic plants overexpressing TaFROG than those with proteins from KN199 (Fig. 4k). These results indicate that overexpression of TaFROG may reduce the interaction of TaSnRK1α with the ubiquitin-26S proteasome system in these transgenic plants, possibly by competing with Osp24.DiscussionLike many other plant pathogenic fungi, F. graminearum may use secreted proteins or effectors to suppress plant defense responses and its genome contains hundreds of orphan genes that are specifically expressed or highly up-regulated during plant infection. In this study, a total of 50 secretory proteins unique to F. graminearum were identified and functionally characterized. Among them, OSP24, OSP25, and OSP44, were found to be important for virulence. They all have typical features of fungal effectors4,5 and are in the fast-evolving telomeric regions55. The location of these three putative effector genes in the telomeric regions may allow the rapid gain and loss or translocation to supernumerary chromosomes as effector reservoirs39,56 although supernumerary chromosomes have not been reported in the few sequenced F. graminearum strains.OSP24 was selected for further characterization because the osp24 deletion mutant had the most significant reduction in virulence. Although dispensable for vegetative growth, reproduction, and initial penetration, OSP24 is important for infectious growth in the rachis tissues in infected wheat heads. Osp24 lacks any conserved domain but is a cysteine-rich protein. Two of the eight cysteine residues, C94 and C105, were found to be important for the function of Osp24 in plant infection. In Sclerotinia sclerotiorum, two cysteine residues of effector SsSSVP1 are essential for the formation of a homo-dimer and its interaction with the host target57. Although Osp24 may form homo dimers, C94 and C105 were dispensable for the Osp24–Osp24 interaction in yeast two-hybrid assays. In fact, these two cysteine residues were predicted to form an intra-molecular disulfide bond. In Stagonospora nodorum, cysteine residues of effector SnTox1 may form multiple disulfide bonds to resist degradation58. In F. graminearum, the formation of a disulfide bond between C94 and C105 may be important for the proper folding and stability of Osp24 in infected wheat tissues.Osp24 is a cytoplasmic effector that is translocated into plant cells during infection. In M. oryzae, a number of cytoplasmic effectors are accumulated in the BIC before being delivered into plant cells34. It is possible that Osp24 is translocated into wheat cells through BIC-like structures that may be formed by F. graminearum. Osp24 strongly interacted with TaSnRK1α in yeast two-hybrid assays and their interactions were confirmed by BiFC and in vitro pull-down assays. The highly conserved SnRK1 kinases function as metabolic regulators of energy homeostasis and are important for development and stress responses in plants48,59. SnRK1 kinases also are involved in regulating plant immunity and known to be targeted by viral and bacterial effector proteins60–63. Although SnRK1 is likely a conserved target for fungal pathogens, interactions between fungal secretory proteins and plant SnRK1 kinases have not been reported. In rice, the expression of OsSnRK1α was reported to be associated with disease resistance against M. oryzae, Cochliobolus miyabeanus, and Rhizoctonia solani64,65 but its exact role in disease resistance has not yet been characterized. In wheat, the kinase activity of TaSnRK1α was increased in the presence of DON produced by F. graminearum50. In this study, we showed that overexpression of TaSnRK1α in wheat increased resistance against F. graminearum but transgenic plants expressing the SnRK1 silencing construct were more susceptible. It is likely that TaSnRK1α plays an important role in regulating resistance responses to F. graminearum and is a target of effector protein Osp24 secreted by the pathogen during infection.The interaction of Osp24 with TaSnRK1α likely results in its degradation in infected wheat heads because the degradation of TaSnRK1α was accelerated in the presence of Osp24 in cell-free degradation assays. In Arabidopsis, the degradation of SnRK1 kinases is commonly associated with sumoylation and ubiquitylation66. The association of SnRK1 kinases with SCF ubiquitin ligases complex and components of the 26S proteasome also have been well documented53. In this study, we showed that treatments with MG132, an inhibitor of the 26S proteasome67, reduced Osp24-mediated TaSnRK1α degradation. Furthermore, we showed that Osp24 increased the association of TaSnRK1α with the SCF ubiquitin ligase complex and 26S proteasome, indicating that the stimulation of TaSnRK1α degradation by Osp24 likely occurs through the ubiquitination-26S proteasomal pathway67. Fungal effectors are known to target the ubiquitin-proteasome pathway in other pathosystems. Whereas effectors Pit2 of U. maydis and Avr2 of Cladosporium fulvum inhibit host proteases that are required for basal defense68,69, AvrPiz-t of M. oryzae suppresses the RING E3 ubiquitin ligase APIP6 in plant cells70. Our results suggest that Osp24 is secreted by F. graminearum as an effector interacting with TaSnRK1α to stimulate its ubiquitination and proteasomal degradation, which in turn negatively impacts resistance responses to fungal infection in wheat plants.Transient expression of Osp24 but not Osp24 truncated of its C-terminal region suppressed BAX-induced or INF1-induced cell death in N. benthamiana. Because the C-terminal region of Osp24 mediates its interaction with TaSnRK1α, it is possible that Osp24 suppresses PCD by targeting N. benthamiana SnRK1. The role of SnRK1 in PCD suppression was also reported in Pepper. Silencing of the pepper SnRK1 transcript resulted in a significant reduction of hypersensitive response (HR) that is elicited by protein AvrBs1 from Xanthomonas campestris pv. vesicatoria (Xcv)62. The YopJ effector homolog AvrBsT from Xcv targeted SnRK1 to suppress AvrBs1-induced plant immunity62.In F. graminearum, DON is not essential for the initial infection but plays a critical role in the spreading of infectious growth via the rachis in infected wheat heads. However, DON is not important for infecting Arabidopsis floral tissues71, likely because infectious growth can spread by hyphae grown on the surface of floral tissues. Interestingly, the expression of TaFROG that encodes a Pooideae-specific orphan protein was induced by DON treatment. TaFROG interacts with TaSnRK1α, and overexpression of TaFROG increased resistance against F. graminearum28. In this study, we showed that Osp24 and TaFROG interacted with the same region of TaSnRK1α. Furthermore, we showed that their binding with TaSnRK1α was competitive and the binding of TaFROG with TaSnRK1α increased its stability (Fig. 5). It is likely that TaFROG functions in defense against F. graminearum by protecting TaSnRK1α against Osp24-mediated degradation. A recent report showed that TaFROG also interacts with TaNACL-D1, a Poaceae-divergent NAC transcription factor with its NAC C‐terminal region specific to the Triticeae, to enhance FHB resistance independent of its interaction with TaSnRK1α29. It is possible that Osp24 also interacts with TaNACL-D1 to suppress defense against FHB. Osp24 and TaFROG are orphan proteins in the pathogens and hosts, respectively, and each may be subjected to co-evolution during fungal–plant interactions. To our knowledge, the active adoption of competing orphan proteins in both fungal pathogen and plant hosts has not been reported in other pathosystems. Expressing engineered TaFROG alleles with stronger interactions with TaSnRK1α or using the host-induced gene silencing (HIGS) approach to silence OSP24 may improve resistance against F. graminearum without yield penalties because they encode orphan proteins specifically expressed during infection.Fig. 5A schematic summary of the roles of two orphan proteins from wheat and Fusarium graminearum during fungal–plant interactions.The cytoplasmic effector Osp24 secreted by invasive hyphae that develop in infected wheat tissues are translocated into plant cells. It interacts with TaSnRK1α, a protein kinase activated by DON and important for resistance against F. graminearum in wheat. The association of Osp24 with TaSnRK1α stimulates its degradation by the SCF ubiquitin ligase complex and 26S proteasome in infected plant cells, resulting in increased susceptibility to Fusarium head blight (FHB). However, DON produced by the fungal pathogen during infection induces the expression of TaFROG in wheat heads. TaFROG, a Pooideae-specific orphan protein, competes with Osp24 for binding with TaSnRK1α and increases its stability. Protection by TaFROG from proteasome degradation and activation by DON enhance the functions of TaSnRK1α in regulating defense responses against F. graminearum infection.MethodsIdentification of OSPs in F. graminearumThe genome and predicted protein sequences of F. graminearum, F. verticillioides, and F. oxysporum, were downloaded from the Broad Institute website (ftp://ftp.broadinstitute.org/pub/annotation/fungi/fusarium/). To identify unique genes, protein sequences of F. graminearum were first used as the queries to search against the predicted proteomes of F. verticillioides and F. oxysporum by BLASTp. Sequences of the F. graminearum proteins also were used to search against the genome sequences of F. verticillioides and F. oxysporum by tBLASTn for possible genes that might not be predicted by automated annotation. (This study was initiated in 2010 when only the genome sequences of F. verticillioides and F. oxysporum were publicly available as the most closely related species of F. graminearum.) All protein sequences of F. graminearum without homologs in these searches (E value cutoff of 1e−5) were extracted and then used as queries to search against NCBI number database (excluding F. graminearum sequences) by BLASTp. The proteins without detectable homologs in these searches (E value cut off of 1e−5) were considered as orphan proteins of F. graminearum. Some of them are not unique to F. graminearum because they have homologs in the genome sequences of other Fusarium species that late became available in the public domain. Therefore, these sequences were described as orphan genes in F. graminearum and its close-related species. These predicted orphan proteins were further analyzed with SignalP 3.0 (http://www.cbs.dtu.dk/services/SignalP-3.0/) to identify secretory proteins.Culture conditions and fungal transformationThe wild-type F. graminearum strain PH-133 and deletion mutants of OSP genes were routinely cultured on potato dextrose agar (PDA) plates at 25 °C. PDA cultures grown at 25 °C were used for measuring growth rate or colony morphology. Conidiation was assayed with 5-day-old liquid carboxymethyl cellulose (CMC) medium. For assaying defects in sexual reproduction, aerial hyphae of 5-day-old carrot agar cultures were pressed down with sterile 0.1% Tween 20 and then incubated at 25 °C under black light. Perithecium formation was examined 7 days after induction for sexual reproduction17,72.Gene replacement constructs were generated with the split-marker approach and transformed into protoplasts of PH-117,20. For each OSP gene, at least two independent gene replacement mutants were identified. For complementation assays, the entire OSP24 gene with its native promoter was cloned into plasmid pFL2 by gap repair42,43 and transformed into the osp24 mutant. The osp24/OSP24 transformants were identified by PCR and assayed for phenotype complementation17. All the primers used in this study are described in Supplementary Data 3.Infection assays with flowering wheat headsConidia were harvested from 5-day-old CMC cultures and re-suspended to 105 spores/ml in sterile distilled water (DDW). Flowering heads of 6-week-old wheat plants of cultivar Xiaoyan 2273, Zhoumai 3638, or KN199 were inoculated with 10 μl of conidium suspensions at the fifth spikelet from the base72,74. Inoculated wheat heads were capped with a plastic bag for 48 h to keep the moisture. Infected wheat heads were examined for diseased spikelets at 7 or 14 dpi to estimate the disease index (number of diseased spikeletsper head)75. Mean and standard deviation of the disease index were estimated with data from three independent replicates with at least 10 wheat heads examined in each replicate. DON production in the inoculated spikelets sampled at 14 dpi was assayed by GCMS-QP2010 system with AOC-20i autoinjector (Shimadzu Co. Japan). To assay infection cushion formation, infected lemmas were sampled at 2 dpi, fixed with 4% (vol/vol) glutaraldehyde, and coated with gold–palladium before examination with a JEOL 6360 scanning electron microscope (Jeol Ltd., Japan)14,20,76. For assaying infectious growth, infected rachis tissues were embedded in Spurr resin after fixation with 3% glutaraldehyde and dehydration in graded series of 30–100% of ethanol before being sectioned77. Thick sections were then prepared and stained with 0.5% (wt/vol) toluidine blue before examination with an Olympus BX-53 microscope. Differences between the wild-type and mutant strains in infection cushion formation and infectious growth were determined with results from at least three independent replicates. Corn silks inoculated with culture blocks (5 mm) of F. graminearum were examined for discoloration at 7 dpi78. For infection assays with tomatoes, each fruit was injected with 10 μl of conidium suspensions after surface sterilization. Inoculated tomatoes were examined for tissue maceration after incubation at 25 °C in dark for 7 days79. For infection assays with Arabidopsis, flowers were sprayed with conidium suspensions of F. graminearum (105 spores/ml). Inoculated plants were cultured in plastic propagators to keep the moisture and examined for necrosis in floral tissues at 7 dpi80.RNA-seq analysisThe RNA-seq reads of wheat heads infected with PH-1 and osp24 mutant were mapped to the reference genome of PH-1 via hisat281. The expression level of each gene was counted with featureCounts82. DEGs were identified by edgeRun83. Functional Catalog (FunCat)84 and GO-enrichment analyses were performed with custom script (https://github.com/xulab-nwafu/funcat) and Blast2GO85, respectively.Assays for the function of the SP of Osp24The predicted SP (22 aa) of Osp24 (SP22) was cloned into the pSUC2 vector86 that carries the yeast SUC2 gene deleted of its SP sequence. The resulting SP22-SUC2 construct was transformed into the yeast suc2 mutant YTK1287 and assayed for growth on CMD-W (0.67% yeast nitrogen base without AAs, 0.075% tryptophan dropout supplement, 2% sucrose, 0.1% glucose, and 2% agar) and YPRAA medium plate (1% yeast extract, 2% peptone, 2% raffinose, and 2 μg/ml antimicyn A)88. Transformants of YTK12 carrying the empty pSUC vector or pSUC2-Avr1bSP89 were used as the negative and positive controls, respectively. The invertase enzymatic activity was detected by the reduction of 2,3,5-triphenyltetrazolium chloride (TTC) to insoluble red colored 1,3,5-triphenylformazan (TPF).Assays for the suppression of BAX-induced or INF1-induced cell death by Osp24PCaMV 35S-Osp24ΔSP was cloned into pGR106 and transformed into A. tumefaciens strain GV310190 expressing the BAX and INF1 constructs91. For infiltration assays with N. benthamiana leaves, A. tumefaciens cells were resuspended to OD600 of 0.8 in infiltration solution (10 mM MES, 10 mM MgCl2, and 150 μM acetosyringone)90. At 18 h after the initial infiltration of tobacco leaves with A. tumefaciens transformant carrying the PCaMV 35S-Osp24ΔSP construct, the same sites were infiltrated with cells carrying GFP, BAX, or INF1 constructs40. Plant cell death symptoms were examined 5 days after infiltration with cells expressing BAX or INF1. A. tumefaciens cells carrying the GFP vector40 were used as the negative control. Each infiltration experiment was repeated at least three times with a minimum of three leaves tested. The expression of BAX and Osp24 in N. benthamiana leaves were assayed by western blot analysis with the anti-BAX and anti-GFP antibodies.Assays for the localization of Osp24 during plant infectionThe Osp24-mCherry-NLS fusion construct under the control of its native promoter was generated by overlapping PCR with the NLS sequence from simian virus large T-antigen92 and transformed into the wild-type stain PH-1. Conidium suspensions (105 spores/ml) of the resulting transformants were used for infection assays with wheat seedlings15. Infectious hyphae and mCherry signals in plant tissues were examined at 2 dpi with a Nikon A1 microscope at excitation/emission wavelengths 543 nm/560–615 nm.Yeast two-hybrid assaysFor library construction, RNA was isolated from wheat heads of Xiaoyan 22 inoculated with PH-1 and sampled at 3 dpi. The yeast two-hybrid library of 6 million primary clones was constructed with vector pGADT7 (Takara Bio, Japan) by OEBiotech (Shanghai, China). For library screening, the Osp24ΔSP bait construct was generated with vector pGBKT7 (Takara Bio, Japan). Yeast colonies that grew on SD-Trp-Leu-His-Ade and had β-galactosidase activity were isolated as putative OICs. After sequencing with primer T7, wheat genes corresponding to the inserts in these clones were identified by Blast searches. To directly assay their interactions, full-length cDNAs or fragments of TaSnRK1α, Osp24, and TaFROG were amplified and cloned into pGADT7 or pGBKT7. The resulting bait and prey constructs of TaSnRK1α, Osp24, and TaFROG were transformed in pairs into yeast strain AH109. The Leu+ and Trp+ transformants were isolated and assayed for growth on SD-Trp-Leu-His medium and galactosidase activities in filter lift assays78.BiFC assays for the TaSnRK1α–Osp24 interactionThe TaSnRK1α and Osp24ΔSP fragments were cloned into the BiFC vectors pSPYNE-35S and pSPYCE-35S93, respectively. The resulting Osp24-nYFP and TaSnRK1α-cYFP constructs were transformed into A. tumefaciens strain GV3101. Leaves of N. benthamiana were then infiltrated with Agrobacterium cells expressing Osp24-nYFP and/or TaSnRK1α-cYFP as described above91. Two-days after infiltration, fluorescence signals were examined with a Nikon A1 microscope. Nuclei were stained with 4,6-diamidino-2-phenylindole (DAPI).In vitro pull-down assaysOsp24, TaSnRK1α, and TaFROG cDNA fragments were amplified and cloned into pGEX4T194 and pCold95 for purification of Osp24-GST, TaSnRK1α-His, and TaFROG-S recombinant proteins96. To assay their interactions, equal amounts of Osp24-GST and TaSnRK1α-HIS fusion proteins were incubated at 4 °C for 2 h and mixed with glutathione or Ni-NTA resins (GenScript, China) for affinity purification97,98. The presence of Osp24-GST and TaSnRK1α-HIS in proteins eluted from Ni-NTA resins was detected by western blot analysis with the anti-HIS (1:5000 dilution, CW0286, CoWin Bioscience Co., China) and anti-GST (1:5000 dilution, CW0084, CoWin Bioscience Co., China) antibodies. For assaying competition between Osp24 and TaFROG, equal amounts of Osp24-GST and TaSnRK1α-HIS fusion proteins were co-incubated with varying amounts of TaFROG-S-tag proteins (1×, 2×, and 4×) before mixing with Ni-NTA resins. To assay the interaction between TaSnRK1α and the ubiquitin-26S proteasome system, total proteins were isolated from wheat heads20,78 and co-incubated with TaSnRK1α-HIS for 30 min before mixing with Ni-NTA resins in the presence of 50 μM proteasome inhibitor MG13267. Western blots of proteins eluted from the Ni-NTA resins were detected with the anti-RBX1 (1:1000 dilution, ab133565, Abcam, UK) and anti-20s (1:1000 dilution, ab22674, Abcam, UK) antibodies for the presence of RBX1 protein and 20S proteasome37.Cell-free protein degradation assaysTotal proteins isolated from wheat heads20,78 and TaSnRK1α-HIS recombinant proteins were used for the cell-free protein degradation assays99,100. In brief, a final concentration of 10 mM ATP was added to the mixture of equal amounts of TaSnRK1α-HIS and crude protein extract isolated from infected wheat heads100. Detection with an anti-actin antibody (1:1000 dilution, BE0027, Easybio, China) is used as a loading control. To inhibit the 26S proteasome activity, a final concentration of 50 μM MG132 was added to the reaction mixtures67. The degradation reaction was stopped after incubation at 25 °C for 0, 45, and 90 min by boiling for 5 min in SDS sample buffer. Western blots of these reaction mixtures were detected for the remaining TaSnRK1α-HIS proteins with an anti-HIS antibody (1:5000 dilution, CW0286, CoWin Bioscience Co., China). Each experiment was repeated at least three times.Generation of transgenic wheat plantsFor overexpression, the full-length TaSnRK1α ORF was cloned into the pANIC-5E vector behind the 35S promoter101. For silencing, a 107-bp fragment of TaSnRK1α was amplified and cloned into vector pANIC-7E in both antisense and sense orientations by Gateway cloning101. The resulting constructs were transformed into immature embryos of wheat cultivar KN199 by particle bombardment32 at Genovo Bio (Tianjin, China). Transgenic plants resistant to BASTA were verified by PCR for carrying transforming constructs with DNA isolated from leaves of T0 plants. The expression level of TaSnRK1α was assayed by qRT-PCR with RNA isolated from wheat heads of selected T1 and T2 plants. Transgenic lines with over 2-fold increase or reduction of TaSnRK1α expression were selected for infection assays with F. graminearum as described above20. Similar approaches were used to generate TaFROG overexpression constructs and transgenic plants. To quantify fungal biomass in infected plant tissues, genomic DNA was extracted from infected wheat heads sampled at 5 dpi and used for qPCR assays with primers specific for the wheat GAPDH102 and F. graminearum CHS5 genes103. Results from three independent biological replicates were used to estimate the ratio of wheat and F. graminearum biomasses.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary informationPeer Review FileDescription of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2Supplementary Data 3Reporting Summary | nature communications | [
"Article"
] | [
"Fungi",
"Fungal pathogenesis",
"Pathogens",
"Plant immunity"
] | genome studies fungal pathogens orphan genes restricted to single species clade majority unknown functions genes in lineage-specific Plant pathogenic fungi evolve novel orphan genes infection virulence Fungal effectors orphan genes for plant infection lack homologs in related species effectors mechanisms identified in small secreted proteins cysteine-rich lack common structural motif In Magnaporthe oryzae fungal pathogen apoplastic effectors secreted endoplasmic reticulum Golgi route apoplast some cytoplasmic effectors secreted through biotrophic interfacial complex) translocated into plant cells4,6 targets fungal effectors vary transcription factors protein kinases plant defense signaling metabolic Tin2 Cmu1 effectors of Ustilago target ZmTTK1 kinase cell wall lignification metabolic status.Fusarium graminearum causal agent of Fusarium head blight wheat barley develops compound appressoria infection cushions for plant After invasive hyphae grow in infected plant tissues develop bulbous irregular invasive hyphae distinct from epiphytic hyphae15Infectious growth spreads to neighboring spikelets wheat rachis blight wheat heads genes for infection processes identified in F. FGL1 FgNahG genes FGL1 encodes lipase fatty callose formation wheat FgNahG salicylate hydroxylase salicylic acid) to catechol FgnahG deletion mutant reduced virulence expression FgNahG Arabidopsis reduced resistance against F. non-ribosomal octapeptide fusaoctaxin A identified virulence factor required F. graminearum for invasion in wheat coleoptiles23 yield losses F. graminearum producer trichothecene mycotoxin deoxynivalenol (DON inhibitor eukaryotic protein synthesis DON important virulence factor TRI5 gene essential for DON biosynthesis expressed in infection cushions early stages tri5 deletion mutant florets spikelets transgenic plants expressing UDP-glucotransferase gene DON increased resistance Overexpression TaFROG protein DON resistance against F. interacts with TaSnRK1α protein NAC transcriptiongermplasm complete resistance F. graminearum lacking 50 trait loci FHB resistance identified in wheat only Fhb1 resistance pathogen spread not initial infection characterized molecular candidate Fhb1 genes include TaPFT pore toxin TaHRC-R histidine-rich calcium-binding underlying mechanism genes FHB resistance F. graminearum genome hundreds orphan roles pathogenic interaction co-evolution small secretory proteins plant immunity no cytoplasmic effectors identified in characterized 50 genes encoding orphan secretory proteins identified OSP24 effector expressed during infection important for infectious growth rachis tissues infected wheat Transient expression Osp24 suppressed BAX-induced INF1-induced cell death identified SNF1-related kinase TaSnRK1α Osp24-interacting protein important for resistance against F. graminearum silencing transgenic plants Osp24 degradation TaSnRK1α ubiquitin-26S proteasome TaFROG competed against Osp24 TaSnRK1α Overexpression TaFROG in wheat stabilized TaSnRK1α increased resistance against F.results study showed OSP Osp24 in F. graminearum cytoplasmic effector targeting TaSnRK1α degradation wheat orphan protein TaFROG competes with Osp24 prevents degradation adoption of orphan proteins pathogen host.ResultsIdentification characterization of OSP genes in F. protein-coding genes genome 971 (~7.3%) identified as orphan genes bioinformatics analyses orphan genes encode smaller proteins less transcribed lower levels Fifty predicted to encode proteins with signal peptides named OSP genes study mature peptides have fewer than 100 amino acid) residues cysteine-rich (Fig. similar to characteristics fungal effector proteins4,5. 1Characterization of orphan genes encoding proteins in Fusarium graminearum analysis length proteins encoded by orphan genes Wilcoxon rank-sum test expression levels genes maximum gene expression level estimated from transcripts per kilobase million in RNA-seq data of conidia perithecia infected wheat headsLength content (%) mature orphan secretory proteins three OSPs important virulence red images 3-day-old PDA cultures 8-dpf mating plates wild-type strain PH-1 (WT) osp24 deletion mutants images wheat heads infected strains 14 dpi disease index WT three osp mutants Error bar standard deviation) mean three experiments = 3) 10 wheat heads significant differences ANOVA analysis Duncan’s multiple range test (P = 0.05) Mean standard deviations DON levels diseased wheat spikelets inoculated WT three osp mutants three replicates 3) No significant differences ANOVA test Distribution OSP genes four chromosomes F. graminearum Red bars positions OSP24 OSP25 OSP44 Expression levels OSP genes TPMs RNA-seq data conidia infected wheat sampled 1-dpi 2-dpi 3-dpi Error bar represents SD from three replicates 3) gene replacement mutants generated 50 OSP genes split-marker wild-type strain PH-1 mutants normal in vegetative growth conidiation sexual reproductionwheat head infection assays mutants normal virulence Table 2) mutants deleted OSP genes reduced virulence assays wheat heads cultivar Xiaoyan 22 (Fig. 1e). disease index osp24 deletion mutants reduced 51% 37% 36% PH-1 (Fig. 1f). osp24 mutants reduced virulence assays wheat cultivar Zhoumai 3638 2) role OSP24 infection not cultivar-specific inoculated spikelets 14 dpi DON production normal in three mutants. 1g). osp24 mutant lowest virulence selected for TRI5 gene expression transcription level TRI5 similar wild type osp24 mutant Fig. 3) OSP24 OSP25 OSP44 genes sub-telomeric regions rapid evolution39 expressed up-regulated in infected wheat heads (Fig. 1i Table 3) results secreted Osp24 Osp25 Osp44 proteins important for full virulence F. graminearum plant infection osp24 mutant defective in infectious growth deletion mutant reduced virulence selected for further characterizationinfection assays corn silks tomatoes Arabidopsis floral tissues osp24 mutant normal virulence host-specific role plant infection wheat infection mutant assayed for formation infection cushions growth invasive hyphae abundant infection cushions on wheat lemma inoculated osp24 mutant 2 days post-inoculation mutant no defects infection cushion formation Deletion OSP24 effects plant penetration development invasive hyphae lemma tissues 2 dpi extensive hyphal growth not observed tissues osp24-infected wheat heads at 5 dpi qPCR 5 dpi fungal biomass reduced in wheat heads inoculated osp24 mutant PH-1 OSP24 infectious growth spreading wheat rachis tissues 2Functions Osp24 signal peptide cysteine residues Infection cushions wild-type strain PH-1 osp24 deletion mutant on wheat lemma at 2 dpi examined SEM ×800 amplification infected wheat heads examined for invasive hyphae rachis tissues at 5 dpi Relative biomass F.infected wheat heads 5 dpi determined qPCR Mean standard deviation estimated three replicates asterisk * indicates differences (P = 0.05) Bootstrap analysis cysteine residues Osp24 intra disulfide bond between C94 C105 signal peptide yeast suc2 mutant YTK12 transformants pSUC2 Osp24 Avr1 assayed for growth CMD-W YPRAA plates invertase activity TTC medium images wheat heads infected PH-1 osp24 mutant/OSP24 transformants photographed 14 dpi Wheat coleoptiles infected PH-1 Osp24:mCherry:NLS examined mCherry signals plant cells 20 μm images wheat heads infected PH-1 osp24 mutant transformants photographed 14 dpi blots proteins wheat heads cultivar Xiaoyan 22 recombinant Osp24-GST-GST Osp24C105A–GST proteins 10 mM ATP detected anti-GST antibody anti-actin antibody loading control concentrations yeast transformants bait prey assayed for growth SD-Trp-Leu-His plates LacZ activityOsp24 cell death BAX INF1 N. benthamiana leaves infiltrated with Agrobacterium cells GFP/BAX/INF1 cells 18 h after GFP Osp24 leaves photographed 5 days after infiltration SP Osp24 required for secretion SP-SUC2 construct generated 22-aa SP Osp24 into pSUC240 transformed into yeast suc2 SP transformants on YPRAA agar invertase activity SP Osp24 functional in yeast in F. graminearum generated OSP24 OSP24ΔSP transformed into osp24 mutant transformants normal in virulence similar mutant in virulence on wheat heads SP essential for complementation osp24 mutant expression OSP24. secretion Osp24 important plant infection cytoplasmic effector translocated into plant generated POSP24-OSP24-mCherry-NLS construct SV40 T-Antigen transformed into PH-1 In wheat seedlings infected transformants Osp24-mCherry-NLS mCherry signals observed in nucleus coleoptile cells inoculation sitesOsp24 cytoplasmic effector translocated into plant cells C94 C105 cysteine residues important for function fungal Osp24 cysteine-rich protein eight (C residues (Fig. constructs OSP24 Cs changed to alanine (A transformed into osp24 mutant mutations six cysteine residues virulence/OSP24C94A transformants defective in wheat head infection (Fig. C94 C105 essential for function Osp24 proteins pathogenesis intra-molecular disulfide bonds disulfide bridge between C94 C105 predicted highest probability (Probability = expressed purified Osp24---GST fusion proteins proteins co-incubated with proteins wheat heads inoculated PH-1 sampled 3 dpi assayed for degradation analyses Osp24C94A Osp24C105A fusion proteins rapidly (Fig. Osp24 proteins less stable when C94 or C105 changed to A-molecular disulfide bond between C94 C105 important for folding stability Osp24 proteinstested yeast-hybrid assays Osp24 protein (Fig. form homodimers generated bait constructs Osp24 C94A or C105A mutation interacted with yeast assays (Fig. mutation C94 C105 effect intermolecular interaction C94 C105 important for formation intramolecular disulfide bonds.Osp24 suppresses cell death expression metabolism-related assayed effect on cell death) in Nicotiana benthamiana induced BAX mouse proapoptotic PCaMV 35S-Osp24ΔSP construct generated transformed into Agrobacterium tumefaciens N. benthamiana leaves infiltrated with Agrobacterium expressing Osp24ΔSP no cell death PCD observed on leaves infiltrated with cells expressing BAX. tumefaciens Osp24ΔSP h death not observed Osp24ΔSP BAX co-expressed (Fig. 2k suppression BAX-induced cell death tested effects Osp24 on PCD triggered INF1 Phytophthora infestans46 infiltration assays N. benthamiana leaves INF1-induced cell death suppressed by Osp24genes suppressed Osp24 infection conducted RNA-seq analysis infected wheat heads 478 differentially expressed genes two-fold changes identified wheat infected wild osp24 mutant DEGs up-regulated in wheat inoculated osp24 mutant expression suppressed Osp24 F. graminearum infection FunCat-enrichment analysis up-regulated DEGs enriched genes metabolism information pathway perception response stimuli Gene Ontology-enrichment analysis categories metabolism enriched oxidation–reduction glycolytic glycine catabolic malate metabolic peroxisome fission chlorophyll biosynthetic fructose 2,6-bisphosphate l-serine photosynthesis ATP synthesis transport 17 (~7%) up-regulated DEGs-infected wheat encode NBS-LRR proteins plant immunity NBS-LRR genes absent in down-regulated DEGs contribute defense responses against F. graminearum 54 83 fungal genes up-regulated down-regulated in osp24 mutant during wheat infectionrelated to virulence factors genes altered expression levels contribute to osp24 mutant interacts with wheat generated Osp24 bait construct screened yeast two-hybrid library infected wheat heads 16 Osp24-interacting clones identified Table 4) Sequencing five clones SNF1-related protein kinase gene TaSnRK1α28 similar to Arabidopsis SnRK1 SnRK1 integrator stress energy signaling plant metabolism growth wheat genome has three TaSnRK1α homoelogues AA sequence 12–14 nucleotide differences coding five TnSnRK1α clones TaSnRK1α-A used TnSnRK1α-A for experiments generated full-length TaSnRK1α prey construct confirmed interaction with Osp24 in yeast two-hybrid assays (Fig. Osp24-nYFP TaSnRK1α-cYFP fusion constructs generated co-expressed in N. benthamiana leaves Agrobacterium infiltration epidermal cells infiltrated tobacco leaves YFP signals observed in nucleus not cytoplasmOsp24-nYFP TaSnRK1α-cYFP nucleus plant cells verified interaction Osp24 TaSnRK1α in vitro assays Recombinant Osp24-GST TaSnRK1α-HIS proteins purified Escherichia co glutathione resins western blots TaSnRK1α-HIS band detected. interaction Osp24-GST Osp24 TaSnRK1α wheat resistance against F. graminearum Yeast two-hybrid assays Osp24 TaSnRK1α concentrations yeast transformants SD-Trp-Leu-His plates BiFC assays interaction Osp24 TaSnRK1α Leaves N. benthamiana agroinfiltrated A. tumefaciens strains Osp24-nYFP TaSnRK1α-cYFP YFP signals observed 2 days post‐agroinfiltration Infiltration Agrobacterium TaSnRK1α Osp24 negative control No YFP signals Verification Osp24–TaSnRK1α interaction GST assays Western blots protein co-purified TaSnRK1α-HIS detected anti-HIS anti-GST antibodiesYeast transformants TaSnRK1α Osp24 assayed for growth SD-Trp-Leu-His plates LacZ activities wheat heads cultivar KN199 transgenic lines TaSnRK1α overexpression drop-inoculated with PH-1 photographed 7 dpi Mean standard deviation disease index PH-1 wheat lines estimated from three experiments 3) 10 infected wheat heads differences ANOVA analysis Duncan’s multiple range test (P = 0.05) sections rachises wheat heads KN199 TaSnRK1α inoculated PH-1 examined for invasive hyphae) at 5 dpi biomass F. graminearum determined by qPCR in infected wheat heads KN199 transgenic lines 5 dpi Mean standard deviation estimated three biological replicates differences (P = 0.05) Bootstrap analysis association TaSnRK1α three fragments Osp24 (Fig 3d amplified tested in yeast two-hybrid assays Deletion C-terminal region Osp24 eliminated interaction with TaSnRK1αdeletion N-terminal 1–60) middle 61–98) region effect Osp24–TaSnRK1α interaction C-terminal region Osp24 involved TaSnRK1α Agrobacterium infiltration Ops24 truncated C-terminal suppress BAX cell death in N. benthamiana C-terminal Osp24 related PCD suppression SnRK1.TaSnRK1α contributes resistance against F. generated TaSnRK1α RNAi transformed into wheat cultivar KN19951 tolerant FHB Xiaoyan 22. transgenic lines expression TaSnRK1α reduced two-fold qRT-PCR assays wheat head infection transgenic lines reduced TaSnRK1α expression displayed severe FHB symptoms than KN199 inoculated strain PH-1 disease index PH-1 increased in lines TaSnRK1α generated transgenic lines TaSnRK1α gene CaMV 35S promoter52 Increased expression TaSnRK1α (over 3-fold detected in five transgenic lines infection transgenic plants overexpressing TaSnRK1α increased resistance to F. graminearumdisease index PH-1 TaSnRK1α overexpressing plants decreased ~40% compared KN199 TaSnRK1α reduced infectious growth limited invasive hyphae rachis tissues transgenic wheat plants overexpressing less fungal biomass infected wheat heads transgenic overexpressing KN199 more abundant infectious hyphae tissues more fungal biomass transgenic plants control KN199 plants DON production diseased spikelets inoculated PH-1 similar KN199 TaSnRK1α RNAi plants overexpression TaSnRK1α increased resistance against F. graminearum silencing TaSnRK1α opposite effect.Osp24 accelerates TaSnRK1α degradation effect Osp24 TaSnRK1α proteins TaSnRK1α-HIS recombinant proteins mixed co-incubated wheat heads cultivar Xiaoyan 22 inoculated wild-type strain PH-1 osp24 mutant TaSnRK1α degraded over time (45–90 min-incubation mixturesTaSnRK1 degradation reduced in co-incubation mixtures with osp24-infected wheat heads PH-1-infected samples Osp24 accelerate TaSnRK1α degradation infection TaSnRK1α degradation not observed in co-incubation mixture TaSnRK1α-HIS Osp24-GST without infected wheat degradation TaSnRK1α-HIS observed in co mixture with Osp24-GST non-inoculated wheat heads Osp24-stimulated degradation TaSnRK1α dependent on wheat head protein extracts Osp24 TaFROG modulating TaSnRK1α stability Western blots TaSnRK1α-HIS proteins wheat heads inoculated PH-1 after 10 mM ATP detected with anti-HIS anti-actin antibody Western blots TaSnRK1α-HIS Osp24-GST detected with anti-HIS or anti-GST antibody Degradation TaSnRK1α-HIS by Osp24 not observed Recombinant TaSnRK1α-HIS Osp24-GST proteins incubated with equal proteins wheat head 10 mM ATP analysesDegradation TaSnRK1α by proteins wheat head inoculated PH-1 inhibited by proteasome inhibitor MG132 Western blots protein-purified TaSnRK1α-HIS detected with anti-20S-RBX1-HIS anti-GST antibodies Yeast transformants Osp24 TaFROG N-terminal (1–266 C-terminal TaSnRK1α assayed for growth on SD-Trp-Leu-His plates LacZ activity Western blots protein-purified-GST detected with anti-GST anti-HIS anti-S-tag antibody TaSnRK1α-HIS proteins co-immunoprecipitated with Osp24-GST reduced by TaFROG-S-tag proteins wheat heads cultivar KN199 infected with PH-1 photographed at 8 dpi Mean standard deviation) disease index PH-1 on wheat lines estimated from three experiments 3) 10 infected wheat heads significant differences ANOVA analysis Duncan’s multiple range test (P = 0.05)Cell-free degradation assays TaSnRK1α-HIS proteins wheat heads KN199 TaFROG OE lines infected PH-1 after 10 mM ATP Western blots protein mixtures co-purified TaSnRK1α-HIS detected with anti-20S anti-RBX1 anti-HIS antibodies detection anti-actin antibody control assayed transcription level TaSnRK1α during F. graminearum infection transcription not up-regulated in infected wheat heads non samples wheat plants inoculated with PH-1 osp24 mutant difference abundance TaSnRK1α transcripts Osp24 affect transcription TaSnRK1α may stimulate degradation infected wheat heads facilitates interaction TaSnRK1α ubiquitin-26S proteasome Arabidopsis interacts SCF ubiquitin ligase complex 26S Osp24 degradation TaSnRK1α proteasome during fungal infection assayed effect proteasomal inhibitor MG132 TaSnRK1α degradation Osp24 degradation TaSnRK1α suppressed by MG132 involvement 26S proteasomeinvestigated effects Osp24 TaSnRK1α ubiquitin-26S proteasome in vitro assays TaSnRK1α-HIS proteins co-incubated wheat heads Osp24-GST 30 min before mixing anti-HIS beads Proteins bound eluted assayed SCF ubiquitin ligase complex 26S proteasome anti-RBX1 anti-20S proteasome Osp24-GST SCF ubiquitin ligase complex 26S proteasome detected proteins co-immunoprecipitated TaSnRK1α-HIS Osp24 increased proteins co-purified TaSnRK1α-HIS Osp24 TaSnRK1α degradation association ubiquitin-26S proteasome wheat orphan protein TaFROG competes Osp24 wheat TaFROG induced DON28 yeast-hybrid assays C-terminal region TaSnRK1α (267–499 Osp24 TaFROG N-terminal region (1–266 aa) dispensable Osp24 TaFROG direct interaction Osp24 TaFROG not detected assaysOsp24 TaFROG TaSnRK1α compete TaSnRK1α TaSnRK1α-HIS Osp24-GST TaFROG-S-tag fusion proteins purified in vitro assays equal TaSnRK1α-HIS Osp24-GST proteins co-incubated mixed anti-HIS beads abundant TaSnRK1α proteins detected proteins co-purified Osp24-GST TaFROG proteins TaSnRK1α proteins-purified Osp24-GST reduced (Fig. competition TaFROG Osp24 TaSnRK1α TaFROG enhances FHB resistance TaSnRK1α F. graminearum infection generated transgenic wheat plants TaFROG gene overexpressed CaMV 35S assayed qRT-PCR expression TaFROG increased 100-fold in five transgenic lines wheat infection fewer spikelets developed FHB symptoms TaFROG overexpression transgenic lines control KN199 plants increase resistance F. graminearum TaFROG overexpression consistent earlierconducted cell-free degradation assays TaSnRK1α-HIS proteins proteins KN199 TaFROG overexpression plants infected PH-1 TaSnRK1α degradation observed mixtures reduced in mixtures proteins transgenic lines overexpressing TaFROG KN199 plants overexpression TaFROG increases stability TaSnRK1α TaSnRK1α degradation similar mixtures proteins wheat heads infected PH-1 osp24 mutant overexpression TaFROG TaSnRK1α against Osp24 binding degradation compete Osp24 reduce degradation ubiquitin-26S proteasome system TaSnRK1α-HIS proteins mixed with proteins wheat heads infected PH-1 anti-HIS beads blots proteins anti-HIS detected with anti-RBX1 anti-20s proteasome antibodies density SCF ubiquitin ligase complex 26S proteasome bands weaker in samples transgenic plants overexpressing TaFROG KN199 overexpression TaFROG may reduce interaction TaSnRK1α ubiquitin-26S proteasome system transgenic plants F.graminearum proteins plant defense responses genome contains orphan genes during infection 50 secretory proteins F. graminearum identified characterized OSP24 OSP25 OSP44 important for virulence typical features fungal in fast-evolving telomeric regions55 genes rapid gain loss translocation to supernumerary chromosomes not reported in sequenced F. graminearum strains.OSP24 selected deletion mutant reduction virulence dispensable for vegetative growth reproduction important for infectious growth in tissues infected wheat heads Osp24 lacks conserved domain cysteine-rich protein Two cysteine residues C94 C105 important for Osp24 plant infection In Sclerotinia two cysteine residues SsSSVP1 essential for formation homo-dimer interaction host Osp24 C94 C105 dispensable for interaction in yeast two-hybrid assays cysteine residues form intra-molecular disulfide bond In Stagonospora nodorum cysteine residues SnTox1 form multiple disulfide bonds resist F.disulfide bond between C94 C105 for folding stability Osp24 in infected wheat tissues cytoplasmic effector translocated into plant cells during infection In M. oryzae effectors accumulated in BIC before Osp24 translocated into wheat cells through BIC structures F. graminearum Osp24 interacted with TaSnRK1α in yeast-hybrid assays confirmed by BiFC in vitro pull-down assays SnRK1 kinases metabolic regulators energy important for development stress responses plant immunity targeted viral bacterial SnRK1 target fungal pathogens interactions between fungal secretory proteins plant SnRK1 kinases reported rice expression OsSnRK1α with disease resistance against M. oryzae Cochliobolus miyabeanus Rhizoctonia role wheat kinase activity TaSnRK1α increased F. overexpression TaSnRK1α in wheat increased resistance against F. graminearum transgenic plants more susceptible TaSnRK1α resistance F. graminearum target effector protein Osp24 secreted infectioninteraction Osp24 with TaSnRK1α degradation in infected wheat heads In Arabidopsis degradation SnRK1 kinases associated with sumoylation association SnRK1 kinases with SCF ubiquitin ligases 26S proteasome treatments with MG132 inhibitor 26S reduced Osp24 TaSnRK1α degradation Osp24 increased association TaSnRK1α with SCF ubiquitin ligase complex 26S proteasome stimulation degradation through ubiquitination-26S proteasomal Fungal effectors target ubiquitin-proteasome pathway pathosystems effectors Pit2 U. maydis Avr2 Cladosporium fulvum inhibit proteases basal AvrPiz-t of M. oryzae suppresses E3 ubiquitin ligase APIP6 plant Osp24 secreted by F. graminearum TaSnRK1α ubiquitination proteasomal degradation impacts resistance fungal infection in wheat plants expression Osp24 cell death in N. benthamiana mediates TaSnRK1α suppresses N. benthamiana SnRK1role SnRK1 PCD suppression reported in Pepper Silencing pepper SnRK1 transcript hypersensitive response (HR) protein AvrBs1 from Xanthomonas campestris pv vesicatoria YopJ effector AvrBsT SnRK1 AvrBs1-induced plant F. graminearum DON not essential for initial infection infectious growth via rachis wheat heads not important infecting Arabidopsis floral growth by hyphae floral tissues expression TaFROG Pooideae-specific orphan protein induced by DON treatment TaFROG interacts with TaSnRK1α overexpression increased resistance against F. Osp24 TaFROG interacted same region TaSnRK1α with TaSnRK1α competitive increased stability. TaFROG F. graminearum TaSnRK1α against Osp24 degradation TaFROG interacts with TaNACL-D1 Poaceae-divergent NAC transcription factor FHB resistance Osp24 interacts with TaNACL-D1 defense FHB Osp24 TaFROG orphan proteins in pathogens hosts co-evolution during fungal–plant interactionsadoption orphan proteins in fungal plant hosts reported pathosystems Expressing engineered TaFROG alleles interactions TaSnRK1α host gene silencing OSP24 improve resistance against F. graminearum penalties proteins.Fig. 5A roles orphan proteins from wheat Fusarium graminearum during fungal–plant interactions cytoplasmic effector Osp24 secreted hyphae wheat plant cells interacts with TaSnRK1α protein important resistance against F. graminearum wheat association Osp24 TaSnRK1α stimulates degradation SCF ubiquitin ligase 26S proteasome susceptibility to Fusarium head blight DON produced induces expression TaFROG in wheat heads TaFROG competes with Osp24 TaSnRK1α increases stability Protection TaFROG from degradation activation DON enhance TaSnRK1α defense against F. graminearum OSPs in F. genome protein sequences of F verticillioides downloaded from Broad Institute website genes protein sequences F. graminearum used proteomesoxysporum BLASTp F. graminearum proteins F. verticillioides F. oxysporum tBLASTn genes study initiated 2010 sequences F. verticillioides F. oxysporum available protein sequences F. graminearum without homologs (E value cutoff 1e−5) extracted NCBI number database BLASTp proteins without homologs orphan proteins F. graminearum Some not unique homologs other Fusarium species described orphan genes F. graminearum-related species orphan proteins analyzed with SignalP 3.0 identify secretory proteins.Culture conditions fungal wild-type F. graminearum strain PH-133 deletion mutants OSP genes cultured on potato dextrose agar (PDA) plates at 25 °C cultures growth rate colony morphology Conidiation assayed with 5-day-old liquid carboxymethyl cellulose) medium defects sexual reproduction 5-day-old carrot agar cultures pressed sterile 0.1% Tween 20 incubated at 25 °C black light Perithecium formation examined 7 days after induction for sexualGene replacement constructs generated split-marker transformed protoplasts PH-117 OSP gene two mutants identified OSP24 gene cloned plasmid pFL2 transformed osp24 mutant transformants identified PCR assayed phenotype primers Supplementary Data 3.Infection assays flowering wheat headsConidia harvested 5-day CMC cultures re-suspended 105 spores/ml sterile distilled water Flowering heads 6-week-old wheat plants 2273 inoculated 10 μl conidium suspensions fifth spikelet Inoculated wheat heads capped plastic bag 48 h Infected wheat heads examined diseased spikelets 7 14 dpi disease index Mean standard deviation disease index estimated three replicates 10 wheat heads DON production inoculated spikelets 14 assayed-QP2010 system AOC-20i autoinjector infection cushion formation infected lemmas sampled 2 dpi fixed 4% glutaraldehyde coated gold–palladium examination JEOL 6360 scanning electron microscopeinfectious growth infected rachis tissues embedded Spurr resin fixation 3% glutaraldehyde dehydration 30–100% ethanol sections stained 0.5% toluidine blue examination Olympus BX-53 microscope Differences wild-type mutant strains infection cushion formation infectious growth determined three replicates Corn silks inoculated blocks F. graminearum examined discoloration 7 tomatoes fruit injected 10 μl conidium suspensions sterilization Inoculated tomatoes examined tissue maceration incubation 25 °C 7 Arabidopsis flowers sprayed conidium suspensions F. graminearum (105 spores Inoculated plants cultured plastic propagators examined necrosis 7-seq reads wheat heads infected PH-1 osp24 mutant mapped genome PH-1 expression level gene counted DEGs identified Functional Catalog GO-enrichment analyses custom script Blast2GO85 function SP predicted SP Osp24 cloned pSUC2 vector86 yeast SUC2 gene sequenceSP22-SUC2 transformed yeast mutant YTK1287 assayed CMD-W (0.67% yeast nitrogen 0.075% tryptophan 2% sucrose 0.1% glucose 2% agar YPRAA plate (1% yeast extract 2% peptone raffinose 2 μg/ml antimicyn Transformants YTK12 pSUC negative positive controls invertase activity 2,3,5-triphenyltetrazolium chloride 1,3,5-triphenylformazan BAX cell death-Osp24ΔSP cloned pGR106 transformed A. tumefaciens strain GV310190 BAX INF1 infiltration N. benthamiana leaves. tumefaciens cells resuspended OD600 infiltration solution (10 mM MES MgCl2 150 μM acetosyringone 18 h infiltration tobacco leaves. tumefaciens infiltrated cells GFP BAX INF1 Plant cell death symptoms examined 5 days after. tumefaciens cells GFP negative control repeated three times three leaves expression BAX Osp24 leaves assayed western blot analysis anti-BAX anti-GFP antibodiesOsp24 plant Osp24-mCherry-NLS fusion generated PCR NLS simian virus T-antigen92 transformed wild-type stain PH-1 Conidium suspensions (105 spores/ml infection assays wheat Infectious hyphae mCherry signals plant examined 2 dpi Nikon A1 microscope wavelengths 543/560–615 nm two-hybrid isolated wheat heads Xiaoyan 22 inoculated PH-1 sampled 3 dpi yeast two-hybrid library 6 million clones constructed pGADT7 Bio OEBiotech Osp24ΔSP bait construct generated pGBKT7 Bio Yeast colonies SD-Trp-Leu-His-Ade β-galactosidase activity isolated putative OICs sequencing T7 wheat genes identified Blast searches TaSnRK1α Osp24 TaFROG amplified cloned pGADT7 pGBKT7 bait prey constructs transformed yeast strain AH109 Leu+ Trp+ transformants isolated assayed growth SD-Trp-Leu-His galactosidase activities filterBiFC assays TaSnRK1α–Osp24 TaSnRK1α Osp24ΔSP fragments cloned BiFC vectors pSPYNE-35S Osp24-nYFP TaSnRK1α-cYFP transformed A. tumefaciens strain GV3101 Leaves N. benthamiana infiltrated Agrobacterium cells Osp24-nYFP TaSnRK1α-cYFP-days fluorescence signals examined Nikon A1 microscope Nuclei stained-diamidino-2-phenylindole vitro assaysOsp24 TaSnRK1α TaFROG cDNA fragments cloned pGEX4T194 pCold95 purification Osp24-GST TaSnRK1α-His TaFROG-S recombinant Osp24-GST TaSnRK1α-HIS proteins incubated 4 °C 2 h mixed glutathione Ni-NTA resins Osp24-GST TaSnRK1α-HIS proteins detected western blot analysis antibodiescompetition Osp24 TaFROG Osp24-GST TaSnRK1α-HIS fusion proteins co-incubated with TaFROG-S-tag proteins 4× before mixing Ni-NTA resins interaction TaSnRK1α ubiquitin-26S proteasome system proteins isolated wheat heads20,78 co-incubated with TaSnRK1α-HIS 30 min before mixing Ni-NTA resins 50 μM proteasome inhibitor MG13267 Western blots proteins detected anti-RBX1 anti-20s antibodies protein 20S proteasome37-free protein degradation proteins isolated wheat heads20,78 TaSnRK1α-HIS recombinant proteins final concentration 10 mM ATP added TaSnRK1α-HIS crude protein extract infected wheat Detection anti-actin antibody loading control inhibit 26S proteasome activity 50 μM MG132 added reaction degradation reaction stopped after incubation 25 °C 90 min boiling 5 min SDS sample buffer Western blots detected remaining TaSnRK1α-HIS proteins anti-HIS antibody experiment repeated three timestransgenic wheat overexpression TaSnRK1α ORF cloned pANIC-5E vector 35S silencing 107-bp fragment TaSnRK1α amplified cloned pANIC-7E Gateway constructs transformed embryos wheat cultivar KN199 particle Genovo Bio (Tianjin Transgenic plants resistant BASTA verified PCR transforming constructs DNA leaves T0 plants expression level TaSnRK1α assayed qRT-PCR RNA wheat heads T1 T2 plants Transgenic lines 2-fold increase reduction TaSnRK1α expression selected infection assays F. graminearum approaches TaFROG overexpression constructs transgenic plants fungal biomass genomic DNA infected wheat heads dpi qPCR assays wheat GAPDH102 F. graminearum CHS5 Results three biological replicates estimate ratio wheat F. graminearum biomasses Nature Research Reporting Summary.Supplementary Review Additional Supplementary Data | 50.8 | 1.19977 |
10.1038/s41467-021-21677-4 | PMC7921134 | Photoluminescence printing is a widely applied anticounterfeiting technique but there are still challenges in developing new generation anticounterfeiting materials providing a high security level. Here, the authors demonstrate coordination dependent photochromic luminescence in a supramolecular coordination polyelectrolyte for multiple information authentication. | While photoluminescence printing is a widely applied anticounterfeiting technique, there are still challenges in developing new generation anticounterfeiting materials with high security. Here we report the construction of a photoresponsive supramolecular coordination polyelectrolyte (SCP) through hierarchical self-assembly of lanthanide ion, bis-ligand and diarylethene unit, driven by metal-ligand coordination and ionic interaction. Owing to the conformation-dependent photochromic fluorescence resonance energy transfer between the lanthanide donor and diarylethene acceptor, the ring-closure/ring-opening isomerization of the diarylethene unit leads to a photoreversible luminescence on/off switch in the SCP. The SCP is then utilized as security ink to print various patterns, through which photoreversible multiple information patterns with visible/invisible transformations are realized by simply alternating the irradiation with UV and visible light. This work demonstrates the possibility of developing a new class of smart anticounterfeiting materials, which could be operated in a noninvasive manner with a higher level of security. | IntroductionCounterfeit goods such as currency, microelectronics, software, movie films, pharmaceutics, and clothing in the market not only cause economic loss to customer and copyright owners, but also bring potential risks to the health and lives of consumers1–3. Governments and copyright holders are forced to increase their investments in developing anticounterfeiting technologies. The global market size of anticounterfeiting technologies was 51.8 billion USD in 2017, and the global anticounterfeiting packaging market is expected to grow to 208.4 billion USD in 20234. Amongst the anticounterfeiting techniques and signal outputs, photoluminescence printing is the most widely applied one, because it offers advantages such as easy handling, high-throughput, facile design, and tunable optical properties in multiple dimensions5–7. For instance, a series of optical materials, including but not limited to upconversion nanoparticles8,9, organic dyes10–12, quantum dots13, metal-organic frameworks14,15, and perovskites16,17, are promising candidates as anticounterfeiting taggants. Lanthanide complexes are also widely applied in anticounterfeiting due to their inherent optical properties, including distinguishable spectroscopic fingerprint, large Stokes shift, and long excited lifetime18–22. For example, Eu2+/Eu3+ are used in Euro banknotes as a luminescence anticounterfeiting label23.However, there are still several challenges in developing new generations of anticounterfeiting materials with more covert and reliable features capable of providing higher security level. (1) Quit a large number of luminescent inks are suspended/dissolved in organic solvents, or contain toxic ions, thus limiting their applications in authenticating food and medicine4,24. (2) Authentic information recorded in materials with static optical outputs is often visible under ambient condition or the excitation of UV light25–27. Thus, stimulus-responsive materials that can respond to external stimuli and alter their optical outputs would be ideal to bring additional security features, making them more difficult to forge2,28–31. On the other hand, invasive stimulus approaches (e.g., thermal, chemical, and mechanical means) may not only contaminate or destroy the goods, but also be inconvenient to operate32–37. For example, it is unrealistic for untrained consumer to add acid, alkali or other chemicals to the labels by themselves. Heating approach may cause damage to goods. 3) In terms of printing technologies, inkjet printing is the most common printing form today38. In many cases, however, the widespread use of inkjet printing fluorescent nanoparticles/nanocrystals requires either complicated assembly and coating procedures to achieve sufficient loading of nanoparticles and long-term stability of the inks, or the modification of preexisting commercial inkjet printers to cope with high viscosity inks or inks containing oversized nanoparticles (such as aerosol jet printers)39–42.To address above-discussed issues, herein, we developed a photoresponsive supramolecular coordination polyelectrolyte (SCP) via the electrostatic interactions of an anionic lanthanide coordination polymer with a cationic photochrome (Fig. 1). Reversible on/off switching of the luminescence signal was realized by remotely alternating UV and visible light irradiation, allowing the fabrication of anticounterfeiting tags for multiple-time verifications. The anionic lanthanide coordination polymer was prepared by the coordination between Eu3+ and alkyl bridged bis-2,6-pyridinedicarboxylic acid ligand, followed by mixing with a cationic diarylethene derivative to form SCP in pure water. The diarylethene unit with the features of high photoisomerization yield, excellent fatigue resistance, and thermal irreversibility was chosen as a photoswitch43–47, since the photochromic fluorescence resonance energy transfer (FRET) between Eu3+ and diarylethene unit is typically governed by the conformation of diarylethene48–50. Thus, the as-prepared SCP exhibits characteristic emission of Eu3+, because the emission spectrum of Eu3+ does not overlap with the absorption spectrum of open-form diarylethene. The irradiation of SCP with UV light leads to the isomerization of open-form diarylethene to its close-form conformation, whose absorption band perfectly overlaps with the emission band of Eu3+. As a result, the luminescence is quenched due to the activation of the photochromic FRET between Eu3+ and photocyclized diarylethene. After subsequent visible light irradiation, the close-form diarylethene isomerizes back to its open-form, and the luminescence intensity is totally recovered. According to this unique property, SCP was filled into a commercially available desktop inkjet printer cartridge to print various high-resolution anticounterfeiting marks. Reversible authentic information with visible/invisible transformation was thus achieved by simple light stimuli, making it suitable for high-security anticounterfeiting applications. Hence, the ring-close and ring-open photoisomerization of the diarylethene moiety regulates the FRET process, leading to reversible luminescence on/off switch in SCP capable of multiple information authentication. In this system, both the lanthanide coordination polymer and the diarylethene derivative are water soluble, and thus, water is the only solvent used in preparing the security ink, enabling its usage in a green condition and its good compatibility with commercial printers. In addition, light irradiation offers clear triggers and spatiotemporal control over the anticounterfeiting patterns in a noninvasive manner.Fig. 1Schematic illustration.The construction of the photochromic supramolecular coordination polyelectrolyte, and the chemical structures of corresponding components.ResultsSynthesis and characterizationBis-2,6-pyridinedicarboxylic acid ligand (L) was synthesized by a two-step procedure and comprehensively characterized (Supplementary Methods and Supplementary Figs. 1–7). Luminescence titration revealed that the coordination stoichiometry between 2,6-pyridinedicarboxylic acid (DPA) and Eu3+ is 3:1 (Supplementary Fig. 8), which is inconsistent with the previous report51,52. Trimeric lanthanide coordination polymer (Eu3+-L) was prepared by mixing compound L and EuCl3 in water with a molar ratio of 1.5:1 and characterized by FTIR spectra (Supplementary Fig. 9) and 1H NMR spectra. Compared to individual L, the absorption band at 1724 cm‒1 assigned to the C = O stretching vibration of DPA underwent a red shift to 1625 cm‒1 in the FTIR spectrum of Eu3+-L, implying the successful coordination of DPA with Eu3+ ion53. In the 1H NMR spectra (Fig. 2a, b), the proton signals assigned to ligand L became highly broadening after the coordination with Eu3+, further confirming the formation of the coordination polymer. The high coordination number of Eu3+-L not only benefits to sufficient sensitization of the Eu3+ ion based on the antenna effect, but also prevents luminescent quenching caused by the infiltration of water molecule, thus endowing the lanthanide coordination polymer with characteristic emission color and brightness in both aqueous solution and the solid state under UV light excitation (Supplementary Figs. 10, 11)54. The luminescence quantum yield of Eu3+-L aqueous solution was measured to be 23.31%. The imidazolium salt modified open-form diarylethene (OF-1) was synthesized through a robust two-step procedure in a yield of 72%, along with full characterizations (Supplementary Figs. 12–21). UV–Vis (Supplementary Figs. 22–26) and 1H NMR spectra (Supplementary Figs. 27, 28) revealed that compound 1 had excellent reversible ring-open/ring-close photoisomerization behavior (Supplementary Notes 1 and 2).Fig. 21H NMR spectral studies.Partial 1H NMR spectra (DMSO:D2O = 4:1, 400 MHz, 25 °C) of a compound L, b Eu3+-L, c-e Eu3+-L-OF-1: c before and d after the irradiation by UV light (300 nm, 60 min), and e subsequent irradiation with visible light (> 450 nm, 60 min). [Eu3+] = 1.4 × 10-4 M, [L] = [OF-1] = 2.1 × 10−4 M.The lanthanide coordination polymer carries three negative net charges (6COO− + Eu3+) per coordination center, allowing it to further assemble with positively charged OF-1 based on electrostatic interaction55–59. The SCP (Eu3+-L-OF-1) was then prepared by mixing Eu3+-L and OF-1 at charge stoichiometry (Eu3+:OF-1 = 1:1.5). Zeta potential experiments were carried out to verify the existence of electrostatic interaction between Eu3+-L and OF-1 (Supplementary Fig. 29). Individual Eu3+-L displayed a negative potential of −19.53 mV, while the ζ-potential value of individual OF-1 was measured to be 20.15 mV. The Eu3+-L-OF-1 solution was almost electrically neutral (1.54 mV). These results confirmed the presence of electrostatic interaction between Eu3+-L and OF-1, enabling suitable distance between the energy donor and acceptor. Dynamic light scattering (DLS) measurements confirmed the formation of supramolecular assembly between Eu3+-L and OF-1. The DLS experiment of Eu3+-L (Fig. 3a) shows a hydrodynamic radius of 220 nm, indirectly proving the formation of large-scaled coordination polymer in solution. The hydrodynamic radius of Eu3+-L-OF-1 increases to around 500 nm, much larger than that of Eu3+-L, revealing that Eu3+-L assembles with OF-1 to form the supramolecular polymer60,61. Meanwhile, uniform spheres with an average diameter of 300 nm were observed by transmission electron microscopy (Supplementary Fig. 30), providing intuitive evidence for the formation of the self-assembly between Eu3+-L and OF-1.Fig. 3DLS size distribution and UV–Vis spectral studies.a DLS size distribution of Eu3+-L (black curve) and Eu3+-L-OF-1 (red curve) ([Eu3+] = 1.4 × 10−4 M, [L] = [OF-1] = 2.1 × 10-4 M). b UV–Vis spectral changes and corresponding photographic images of Eu3+-L-OF-1 and Eu3+-L-CF-1 with alternating 300 nm UV and >450 nm visible light irradiation in water for up to 60 s each time ([Eu3+] = 1.4 × 10-5 M, [L] = [OF-1] = 2.1 × 10−5 M).Photoresponsive propertyWe then investigated the photoresponsive property of Eu3+-L-OF-1 resulting from the isomerization of the diarylethene moiety. The UV–Vis spectra of Eu3+-L-OF-1 (Fig. 3b) showed an absorption band at 294 nm corresponding to OF-1 unit, and no absorption over 400 nm was observed. Upon the irradiation with UV light (300 nm), the absorption at 294 nm gradually decreased, two new absorption bands centered at 380 nm and 596 nm appeared. Meanwhile, the colorless aqueous solution changed to dark blue (insert of Fig. 3b). These phenomena jointly demonstrated the OF-1 transformed to its close form (CF-1) after the irradiation. All these changes levelled off in 60 s (Supplementary Fig. 31). Moreover, a well-defined isosbestic point was observed at 323 nm, indicating that ring-open isomer cleanly transformed into the photocyclized form in SCP62,63. We further measured the photocyclization yield at the photostationary state by 1H NMR spectra (Fig. 2c,d). Since the proton signals of OF-1 showed serious broadening in aqueous media (Supplementary Fig. 32), the 1H NMR spectral study was carried out in mixed deuterated solvent (DMSO-d6:D2O = 4:1). After irradiated by UV light (300 nm, 60 min), the thiophene protons (Hb) underwent an obvious upfield shift from 7.30 to 6.81 ppm, mainly due to the electronic shielding effect in the large conjugated closed ring isomers64. The methyl protons (Ha) of the diarylethene unit underwent an apparent downfield shift from 1.88 to 2.00 ppm. Meanwhile, the aromatic protons Hc and Hd showed downfield shifts from 7.52 ppm to 7.63 ppm and from 6.98 ppm to 7.04 ppm, respectively. All these shifts were thorough, and no apparent residual peaks retained in the original chemical shifts after the UV light irradiation (Supplementary Fig. 33). The molar ratio of CF-1: OF-1 was determined to be 0.94:0.06 according to the integrating resonance of protons Ha, indicating nearly quantitative (~94%) conversion from Eu3+-L-OF-1 to Eu3+-L-CF-1 upon exposure to UV light65. Interestingly, a complete recovery in both UV–Vis (Supplementary Figs. 34, 35) and 1H NMR spectra (Fig. 2e) was achieved upon subsequent irradiation of the resulting Eu3+-L-CF-1 solution with >450 nm visible light, accompanied by color change back to colorless, revealing that this photoisomerization behavior was fully reversible.The photoresponsive luminescent behavior of SCP was then investigated. In the as-prepared Eu3+-L-OF-1, no FRET was observed, because there was no spectral overlapping between the UV–Vis absorption of OF-1 and the emission spectrum of Eu3+-L. Eu3+-L-OF-1 exhibited the characteristic spectral line of lanthanide. The excitation spectrum of Eu3+-L-OF-1 showed a broad band centered at 265 nm, attributed to the absorption of the DPA moiety (Supplementary Fig. 36). The corresponding emission spectrum was composed of five sharp peaks at 580, 594, 615, 649, and 692 nm, referred to the 5D0 to 7FJ (J = 0-4) transitions of Eu3+ respectively, in which the 5D0 → 7F2 transition at 615 nm is dominant and responsible for the bright red emitting color (Supplementary Fig. 36)66. On the other hand, the luminescence emission spectrum of lanthanide coordination polymer Eu3+-L completely overlapped with the absorption spectrum of CF-1 in the range of 500-700 nm (Fig. 4a), implying that efficient FRET process may occur from Eu3+ to CF-1 in Eu3+-L-CF-1. As expected, the luminescence of Eu3+ (Fig. 4b) was quenched gradually upon irradiating SCP with UV light. The luminescence quenching followed a biexponential attenuation law, containing a fast process, followed by a slow process to the photostationary state in 60 s (inset of Fig. 4b)49. The luminescence intensity was quenched completely at the end, and the decay decreased from 1,289 to 12 μs (Supplementary Figs. 37–41), with concomitant decrease of the luminescence quantum yield from 15.84% to 0.85%. These phenomena confirmed the occurrence of the FRET process with an efficiency (E) of 98%, calculated according to the reported method67. The quenched luminescence of Eu3+-L-CF-1 could completely recover to its original level upon subsequent visible light irradiation, ascribing to the photocycloreversion reaction (Fig. 4c). In particular, the photocontrolled luminescence on/off switch of SCP presented outstanding reversibility, and no apparent deterioration in the luminescence intensity (less than 4%) was observed after 20 consecutive cycles of alternating UV and visible light irradiations (Fig. 4d). Thus, SCP exhibited excellent fatigue resistance, which is of utmost importance for multiple anticounterfeiting applications.Fig. 4Photophysical studies.a Partial emission spectrum (black curve) of Eu3+-L, and absorption spectra of OF-1 (red curve) before and (blue curve) after irradiation with 300 nm UV light for 60 s. b, c Luminescence emission spectral changes of Eu3+-L-OF-1 upon (b) UV light (300 nm) irradiation and c subsequent visible light (>450 nm) irradiation in water. Insets show corresponding emission intensity changes at 615 nm. d Luminescence emission changes of Eu3+-L-OF-1 upon consecutive alternating exposure to UV and visible light. Insets show corresponding intensity changes at 615 nm (upper) and the photographs of the SCP solution under 254 nm UV lamp (lower). [Eu3+] = 1.4 × 10−4 M, [L] = [OF-1] = 2.1 × 10−4 M.It is worth noticing that the diarylethene derivative is bistable37, which means that the spontaneous photocycloreversion reaction is extremely slow under natural conditions. The half-life (t1/2) of Eu3+-L-CF-1 at 25 °C was estimated to be 376.7 min (Supplementary Figs. 42, 43, Supplementary Table 1, and Supplementary Note 3), ranking one of the longest t1/2 values reported so far in diarylethene derivatives68,69, which confirmed that the self-switching is negligible. Only slight self-switching of Eu3+-L-CF-1 was observed upon continuous exposure to sunlight for 90 min (Supplementary Fig. 44). When the Eu3+-L-CF-1 solution was kept at an elevated temperature (60 °C) in the dark, no sign of thermal ring opening was observed from the UV–Vis spectra, supporting the good thermal stability of Eu3+-L-CF-1 (Supplementary Fig. 45).Pattern printingThe developed SCP with important features of rapid response, prominent anti-fatigue capability and thermally irreversible luminescence on/off photoswitch encouraged us to further explore its performance in smart anticounterfeiting. We directly filled the Eu3+-L-OF-1 aqueous solution in a commercial inkjet printer (canon PIXMA ip1180) cartridge with the concentration low to 2.1 × 10-4 M (according to the concentration of OF-1), and printed various high-resolution quick response (QR) codes on commercial blue polyester terephthalate (PET) films (Fig. 5a, b). The obtained QR code was invisible under daylight due to the colorless nature of Eu3+-L-OF-1 aqueous solution (Fig. 5c and Supplementary Movie 1). However, bright red luminescent pattern was observed under 254 nm UV lamp, allowing to retrieve the encoded information quickly and accurately by scanning through a smartphone (Fig. 5d and Supplementary Movie 2). It should be noted that the UV absorbance intensity of OF-1 at 254 nm is low, and thus the conversion from OF-1 to CF-1 under 254 nm UV lamp is very slow, providing enough time for recognizing the authentic information recorded in the QR code. The luminescence was quenched upon 300 nm UV light irradiation, making the QR code invisible under UV light. Although the pattern turned to blue under daylight, it can be completely masked by the blue background of the PET film. Through which, the absolutely and really invisible security pattern was achieved under both daylight (Fig. 5e and Supplementary Movie 3) and UV light (Fig. 5f and Supplementary Movie 4), which was highly sufficient for confidential information encryption.Fig. 5Pattern printing using SCP as the ink.a, b Schematic illustration of the pattern printing process. Light triggered QR code with visible/invisible transformation behavior was achieved by using supramolecular coordination polyelectrolyte (SCP) as the smart ink. c–f Digital photos of SCP-based luminescent QR code on commercial blue PET film (size: 5 × 5 cm) upon alternating UV (300 nm, 60 s) and visible light (>450 nm, 120 s) irradiation. c, d Photos under daylight. e, f Photos under 254 nm UV lamp. [Eu3+] = 1.4 × 10−4 M, [L] = [OF-1] = 2.1 × 10−4 M in the SCP ink.As discussed above, CF-1 is the photostable state under daylight, especially in solid state. Consequently, the erased pattern remained unreadable even after placing under sunlight for one month (Supplementary Movies 5 and 6). The erased pattern could be completely recovered and recognized upon further irradiating with visible light (>450 nm). Moreover, even after 20 consecutive switching cycles, the quality of the remote light triggered information pattern with visible/invisible transformation process still remained unaffected (Supplementary Movies 7 and 8). Thus, the rapid response, noninvasive regulation, excellent fatigue resistance, and thermal irreversibility of SCP-based system made it a suitable anticounterfeiting ink for multiple authentic information encryption and decryption.DiscussionIn summary, we have developed a hierarchical self-assembly approach to realize a photoresponsive supramolecular coordination polyelectrolyte capable of reversible multiple information encryption and decryption. An anionic lanthanide coordination polymer, obtained from the coordination between lanthanide ion and a bis-ligand, further assembles with a cationic diarylethene derivative based on ionic interaction to afford the SCP. Significantly, the ring-open/ring-close photoisomerization of the diarylethene moiety governs the FRET process between the lanthanide emitting center and the diarylethene component, leading to reversible luminescence on/off switch in the SCP. This SCP has been directly utilized as a security ink to realize reversible authentic information patterning with visible/invisible transformation by simply alternating the exposure to UV and visible light. The developed materials and its associated patterning technology with environmentally friendly preparation process, remote light control, rapid response, excellent fatigue resistance and thermal irreversibility have demonstrated a promising potential as a high-security anticounterfeiting ink in various fields, including authenticating food and medicine.MethodsSynthetic routes for compounds L and OF-1 are shown in Supplementary Figs. 1, 12.Synthesis of compound 5A mixture of compound 6 (474 mg, 1.98 mmol), 1,4-dibromobutane (60 μL, 0.5 mmol) and K2CO3 (248 mg, 1.8 mmol) was stirred in N,N-dimethylformamide (10 mL) under N2 at 80 °C for 48 h. Then, the reaction mixture was poured into water (100 mL). The resulting white precipitate was collected and washed three times with water. The precipitate was then dissolved in CH2Cl2 (100 mL) and washed with a solution of 5 % aqueous NaOH solution (2 × 50 mL). The organic phase was concentrated and dried under vacuum to give compound 5 as a white solid in 80% yield. 1H NMR (400 MHz, CDCl3, ppm): δ 7.79 (s, 4H), 4.48 (q, J = 7.1 Hz, 8H), 4.27 (m, 4H), 2.08 (m, 4H), 1.46 (t, J = 7.1 Hz, 12H). 13C NMR (100 MHz, CDCl3, ppm): δ 166.7, 164.7, 150.2, 114.1, 68.2, 62.4, 25.3, 14.2. HRMS [M + H]+ calcd. for C26H33N2O10+ 533.2135; found: 533.2126; Anal. Cald. for C26H32N2O10: C, 58.64; H, 6.06; N, 5.26; Found: C, 58.58; H, 6.10; N, 5.22.Synthesis of compound LA mixture of compound 5 (168 mg, 0.4 mmol), KOH (224 mg, 4 mmol), methanol (10 mL) and water (10 mL) was stirred at 60 °C for 12 h, and then acidified with HCl (3 M) to pH 4. The precipitate was collected by centrifugation, washed with H2O, and dried under vacuum to give compound L as a white solid in 60% yield. 1H NMR (400 MHz, D2O, ppm): δ 7.59 (s, 4H), 4.34 (m, 4H), 2.08 (m, 4H). 13C NMR (100 Hz, H2O, ppm): δ 172.7, 166.7, 154.9, 111.4, 68.3, 24.7. HRMS [M-H]– calcd. for C18H15N2O10– 419.0727; found: 419.0734; Anal. Cald. for C18H16N2O10: C, 51.44; H, 3.84; N, 6.66; Found: C, 51.29; H, 3.78; N, 6.60.Preparation of the coordination polymer Eu3+-LCompound L (42 mg, 0.1 mmol) and potassium hydroxide (22.4 mg, 0.4 mmol) were dissolved in water (10 mL). Then, europium chloride hexahydrate (24.5 mg, 0.067 mmol) was added with stirring for 30 min. The mixture was dried under vacuum to give Eu3+-L as a white solid.Synthesis of compound 44-Hydroxyphenylboronicacidpinacolester (220 mg, 1 mmol), 1,2-dibromoethane (940 mg, 5 mmol), and K2CO3 (690 mg, 5 mmol) were added into acetonitrile (20 mL) with stirring. The mixture was heated at 70 °C under N2 atmosphere for 24 h. After cooling down to room temperature, the reaction mixture was filtered and the residue was washed with CH2Cl2. Then, the filtrate was concentrated under reduced pressure. The residue was dissolved by CH2Cl2 (50 mL) and washed twice with saturated NaCl solution. The organic phase was concentrated. The crude product was purified by column chromatography over silica gel (eluent: petroleum ether/ethyl acetate = 20:1), and compound 4 was obtained as white powder in 80% yield. 1H NMR (400 MHz, CDCl3, ppm): δ 7.75 (d, J = 8.5 Hz, 2H), 6.90 (d, J = 8.5 Hz, 2H), 4.32 (t, J = 6.4 Hz, 2H), 3.64 (t, J = 6.3 Hz, 2H), 1.33 (s, 12H). 13C NMR (100 MHz, CDCl3, ppm): δ 160.6, 136.6, 114.0, 83.6, 67.6, 28.9, 24.9. HRMS [M + H]+ calcd. for C14H21BBrO3+ 327.0767; found: 327.0754; Anal. Cald. for C14H20BBrO3: C, 51.42; H, 6.16; Found: C, 51.38; H, 6.18.Synthesis of compound 2Compound 4 (327 mg, 1 mmol), compound 3 (210 mg, 0.4 mmol), Pd(PPh3)4 (70 mg, 0.06 mmol), and Na2CO3 (680 mg, 6.4 mmol) were added to a mixed solution of water (4 mL) and dimethoxyethane (30 mL). The mixture was refluxed under N2 at 90 °C in dark for 24 h. After cooling down to room temperature, the solvent was removed under vacuum. The residue was extracted by dichloromethane, and purified on a silica gel column using petroleum ether/ethyl acetate (20:1) as the eluent. 1H NMR (400 MHz, CDCl3, ppm): δ 7.47 (d, J = 8.6 Hz, 4H), 7.17 (s, 2H), 6.93 (d, J = 8.6 Hz, 4H), 4.32 (t, J = 6.2 Hz, 4H), 3.66 (t, J = 6.2 Hz, 4H), 1.94 (s, 6H). 13C NMR (100 MHz, CDCl3, ppm): δ 158.0, 141.9, 140.5, 127.0, 126.9, 125.8, 121.5, 115.2, 68.0, 28.9, 14.5. HRMS [M + H]+ calcd. for C31H25Br2F6O2S2+ 766.9546; found: 766.9537; Anal. Cald. for C31H24Br2F6O2S2: C, 48.58; H, 3.16; Found: C, 48.51; H, 3.19.Synthesis of compound OF-1Compound 2 (383 mg, 0.5 mmol) was dissolved in acetonitrile (10 mL), and then 1-methylimidazole (410 mg, 5 mmol) was added. The reaction mixture was stirred at 80 °C for 12 h. After cooling down to room temperature, the obtained precipitate was collected by centrifugation and washed with diethyl ether for three times to afford the desired product OF-1 in 90% yield. 1H NMR (400 MHz, DMSO-d6, ppm): δ 9.20 (s, 2H), 7.83 (s, 2H), 7.73 (s, 2H), 7.58 (d, J = 8.4 Hz, 4H), 7.39 (s, 2H), 7.02 (d, J = 8.6 Hz, 4H), 4.61 (t, J = 4.6 Hz, 4H), 4.39 (t, J = 4.7 Hz, 4H), 3.88 (s, 6H), 1.93 (s, 6H). 13C NMR (100 MHz, DMSO-d6, ppm): δ 158.1, 142.0, 140.7, 137.5, 127.2, 126.4, 125.4, 124.0, 123.3, 121.8, 115.8, 66.4, 48.8, 36.3, 14.5. HRMS [M-2Br]2+ calcd. for C39H36F6N4O2S22+ 385.1086; found: 385.1081. Anal. Cald. for C39H36Br2F6N4O2S2: C, 50.33; H, 3.90; N, 6.02; Found: C, 50.39; H, 3.98; N, 5.94.Preparation of the QR codeIn a standard procedure, commercial PET film was first printed with blue background, and the QR-pattern was then directly printed on the blue PET film. The QR code was scanned by a commercially available smartphone APP.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Movie 1Supplementary Movie 2Supplementary Movie 3Supplementary Movie 4Supplementary Movie 5Supplementary Movie 6Supplementary Movie 7Supplementary Movie 8 | nature communications | [
"Article"
] | [
"Coordination polymers",
"Self-assembly",
"Supramolecular polymers"
] | IntroductionCounterfeit goods currency microelectronics software films pharmaceutics clothing cause economic loss to copyright owners bring potential risks to health lives Governments copyright holders increase investments anticounterfeiting technologies global market size anticounterfeiting 51.8 billion USD in 2017 packaging market expected to 208.4 billion USD in 20234 photoluminescence printing applied easy handling high-throughput facile design tunable optical properties optical materials upconversion nanoparticles8,9 organic quantum dots13 metal-organic frameworks14 perovskites16 promising candidates anticounterfeiting Lanthanide complexes applied in anticounterfeiting optical properties spectroscopic fingerprint large Stokes shift long excited Eu2+/Eu3+ used in Euro banknotes as luminescence anticounterfeiting label23 challenges in developing new anticounterfeiting materials higher security level luminescent inks suspended/dissolved in organic solvents contain toxic ions applications authenticating food medicine4 Authentic information in materials static optical outputs visible under ambient UV stimulus-responsive materials optical outputs additional security features difficult to forge2 invasive stimulus approachesthermal chemical mechanical means contaminate destroy goods inconvenient operate32–37 unrealistic untrained consumer add acid alkali chemicals to labels Heating damage goods inkjet printing common fluorescent nanoparticles requires complicated assembly coating stability modification inkjet printers high viscosity inks oversized nanoparticles developed photoresponsive supramolecular coordination polyelectrolyte) electrostatic interactions anionic lanthanide coordination polymer with cationic photochrome Reversible on/off switching luminescence signal alternating UV visible light irradiation fabrication anticounterfeiting tags multiple-time verifications anionic lanthanide polymer prepared Eu3+ alkyl bridged bis-2,6-pyridinedicarboxylic acid ligand mixing with cationic diarylethene derivative SCP in pure water diarylethene unit high photoisomerization yield fatigue resistance thermal irreversibility chosen as photoswitch43–47 photochromic fluorescence resonance transfer) Eu3+ diarylethene governed by conformation diarylethene48–50-prepared SCP exhibits emission Eu3+ overlap with open-form diarylethene irradiation UV leads isomerization open-form diarylethene to close-form overlaps with Eu3+ luminescence quenched activation photochromic FRET between Eu3+ diarylethene After irradiation close-form diarylethene isomerizes to open-form luminescence intensity recovered SCP filled into desktop inkjet printer cartridge high anticounterfeiting marks Reversible information transformation achieved by light stimuli suitable for high-security anticounterfeiting applications ring-close ring-open photoisomerization diarylethene regulates FRET reversible luminescence on/off switch SCP multiple information authentication lanthanide coordination polymer diarylethene derivative water soluble water only solvent security ink green condition compatibility with commercial printers light irradiation offers triggers spatiotemporal control anticounterfeiting patterns. construction photochromic supramolecular coordination polyelectrolyte chemical structures components characterizationBis-2,6-pyridinedicarboxylic acid ligand) synthesized two-step procedure characterized 1–7)Luminescence titration between 2,6-pyridinedicarboxylic acid (DPA Eu3+ 3:1 8) inconsistent with previous Trimeric lanthanide coordination polymer (Eu3+-L) prepared mixing L EuCl3 in water molar ratio 1.5:1 characterized FTIR 1H NMR spectra absorption band at 1724 cm‒1 to 1625 cm‒1 FTIR spectrum Eu3+-L successful coordination DPA with Eu3+ 1H NMR spectra proton signals ligand L broadening after coordination with Eu3+ formation coordination polymer high coordination Eu3+-L benefits sensitization Eu3+ prevents luminescent quenching water polymer characteristic emission color brightness in aqueous solution solid state UV luminescence quantum yield of Eu3+-L aqueous solution 23.31% imidazolium salt modified open-form diarylethene (OF-1) synthesized procedure yield 72% characterizations 12–21) UV–Vis 1H NMR spectra compound 1 excellent reversible ring-open/ring-close photoisomerization behavior. 21H NMR spectral studies1H NMR spectra:D2O 4:1 400 MHz 25 °C Eu3+-L Eu3-L-OF-1 before after irradiation UV (300 nm 60 visible light> 450 nm 60 [Eu3+ 1.4 × 10-4 M [OF-1] 2.1 × 10−4 M lanthanide coordination polymer three negative charges (6COO− + Eu3+) charged OF-1 SCP (Eu3+-L-OF-1) Eu3+-L OF-1 charge stoichiometry Zeta potential experiments electrostatic interaction Eu3+-L OF-1 Eu3+-L negative potential −19.53 mV ζ-potential OF-1 20.15 mV Eu3+-L-OF-1 solution electrically neutral (1.54 mV). electrostatic interaction Eu3+-L OF-1 distance energy donor acceptor Dynamic light scattering measurements supramolecular assembly Eu3+-L OF-1 DLS experiment Eu3+-L hydrodynamic radius 220 nm large-scaled coordination polymer radius Eu3+-L-OF-1 increases 500 nm Eu3+-L assembles OF-1 supramolecularuniform 300 nm observed electron microscopy Fig. evidence self-assembly between Eu3+-L OF-1. 3DLS size distribution UV–Vis spectral studies size distribution Eu3+-L-OF-1+ 1.4 × 10−4 M-1] 2.1 × 10-4 UV–Vis spectral changes photographic images Eu3+-L-OF-1-CF-1 300 nm UV >450 nm visible light irradiation water 60 s+ 1.4 × 10-5 M-1] 2.1 × 10−5 M).Photoresponsive investigated property Eu3+-L-OF-1 isomerization diarylethene moiety UV–Vis spectra absorption band 294 nm no absorption over 400 nm irradiation UV (300 absorption 294 nm decreased new absorption bands 380 596 nm colorless aqueous solution changed dark blue OF-1 transformed form (CF-1) after irradiation changes levelled off in 60 s isosbestic point observed 323 nm ring-open isomer transformed photocyclized form SCP62,63 measured photocyclization yield state 1H NMR spectraproton signals OF-1 broadening aqueous media 1H NMR spectral study mixed deuterated solvent (DMSO-d6:D2O = 4:1) UV light (300 nm thiophene protons (Hb upfield shift 7.30 to 6.81 ppm electronic methyl protons (Ha diarylethene downfield shift 1.88 to 2.00 ppm aromatic protons Hc Hd downfield shifts 7.52 ppm to 7.63 ppm 6.98 to 7.04 ppm shifts no residual peaks UV irradiation molar ratio CF-1: OF-1 0.94:0.06 (~94%) conversion from Eu3+-L-OF-1 to Eu3+-L-CF-1 UV complete recovery in UV–Vis 1H NMR spectra irradiation Eu3+-L-CF-1 solution >450 nm light color change to photoisomerization behavior reversible photoresponsive luminescent behavior SCP investigated-OF-1 no FRET observed no spectral overlapping UV–Vis absorption OF-1 emission spectrum Eu3+-L-L-OF-1 characteristic spectral line of lanthanideexcitation spectrum Eu3+-L-OF-1 broad band at 265 nm absorption DPA moiety Fig emission spectrum five peaks at 580 615 649 692 nm 5D0 to 7FJ transitions Eu3+ 5D0 → 7F2 transition at 615 nm dominant bright red color luminescence spectrum Eu3+-L overlapped with absorption spectrum CF-1 500-700 nm (Fig. efficient FRET process from Eu3+ to CF-1 luminescence Eu3+ 4b quenched irradiating SCP UV light quenching followed biexponential attenuation law fast process slow process to photostationary state in 60 s luminescence intensity quenched end decay decreased from 1,289 to 12 μs 37–41) luminescence quantum yield from 15.84% to 0.85% FRET process efficiency (E) 98% quenched luminescence Eu3+-L-CF-1 original level visible light irradiation photocycloreversion reactionphotocontrolled luminescence on/off switch SCP outstanding reversibility no deterioration luminescence intensity than 4% after 20 cycles UV visible light irradiations (Fig. SCP excellent fatigue resistance for anticounterfeiting applications. 4Photophysical studies emission spectrum Eu3+-L absorption OF-1 before after irradiation 300 nm UV light 60 s Luminescence emission changes Eu3+-L-OF-1 UV nm visible light>450 nm irradiation 615 nm Luminescence emission changes Eu3+-L-OF-1 exposure UV visible light 615 nm SCP solution under 254 nm UV lamp [Eu3+] = 1.4 × 10−4 M [L [OF-1] = 2.1 × 10−4 M diarylethene derivative bistable37 spontaneous photocycloreversion reaction slow natural conditions half-life Eu3+-L-CF-1 at 25 °C 376.7 min longest t1/2 diarylethene self-switching negligible slight self-switching Eu3+-L-CF-1 continuous exposure sunlight 90 minEu3+-L-CF-1 solution kept at elevated temperature (60 °C) dark no thermal ring opening UV–Vis spectra good thermal stability Fig. 45).Pattern printingThe SCP rapid response anti-fatigue capability thermally irreversible luminescence on/off performance smart anticounterfeiting filled Eu3+-L-OF-1 aqueous solution in commercial inkjet printer PIXMA ip1180) cartridge concentration to 2.1 × 10-4 M printed high-resolution quick response (QR) codes on blue polyester terephthalate (PET) films (Fig. 5a QR code invisible under daylight colorless solution bright red luminescent pattern observed under 254 nm UV lamp encoded information scanning smartphone UV absorbance intensity OF-1 at 254 nm low conversion from to CF-1 under slow time recognizing authentic information QR code luminescence quenched upon 300 nm UV light irradiation QR code invisible under UV light pattern turned to blue under daylight masked by blue background PET film invisible security pattern achieved under daylight UV light for confidential information encryption.Fig. 5Pattern printing using SCP illustration processLight triggered QR code visible/invisible transformation supramolecular coordination polyelectrolyte (SCP) smart ink photos SCP luminescent QR code blue PET film 5 × 5 cm alternating UV (300 nm visible light>450 nm 120 irradiation Photos daylight 254 nm UV lamp [Eu3+] 1.4 × 10−4 M [L [OF-1] 2.1 × 10−4 M SCP ink CF-1 photostable daylight solid erased pattern unreadable sunlight one month erased pattern recovered irradiating visible light>450 20 switching cycles remote light triggered information pattern visible/invisible transformation unaffected rapid response noninvasive regulation fatigue resistance thermal irreversibility SCP system suitable anticounterfeiting ink information encryption decryption developed hierarchical self-assembly approach photoresponsive supramolecular coordination polyelectrolyte reversible multiple information encryption decryption anionic lanthanide coordination polymer cationic diarylethene derivative SCP ring-open-close photoisomerization diarylethene governs FRET process reversible luminescence on/off switch SCPSCP utilized security ink reversible information patterning alternating UV light materials patterning technology environmentally friendly preparation remote light control rapid response fatigue resistance thermal irreversibility potential high-security anticounterfeiting ink authenticating food medicine routes compounds L OF-1 Figs. 1 12.Synthesis 5A 6 (474 mg 1.98 1,4-dibromobutane (60 K2CO3 (248 mg 1.8 mmol stirred N-dimethylformamide (10 mL 80 °C 48 h poured water (100 precipitate collected washed three times dissolved CH2Cl2 (100 mL washed 5 % NaOH (2 50 organic phase concentrated dried vacuum compound 5 white solid 80% yield 1H NMR (400 MHz CDCl3 δ 7.79 4.48 4.27 2.08 1.46 13C NMR (100 MHz CDCl3 δ 166.7 164.7 150.2 114.1 68.2 62.4 25.3 14.2. C26H33N2O10 533.2135 C26H32N2O10 C 58.64 6.06 5.26H 6.10 N 5.22.Synthesis 5 (168 mg KOH (224 mg 4 methanol water stirred 60 °C 12 h acidified HCl pH 4. precipitate collected washed H2O dried vacuum white solid 60% yield 1H NMR (400 MHz D2O δ 7.59 4.34 2.08 13C NMR (100 Hz H2O δ 172.7 166.7 154.9 111.4 68.3 24.7. 419.0727.0734 C18H16N2O10 C 51.44 H 3.84 N 6.66 polymer Eu3+-LCompound L (42 mg 0.1 mmol potassium hydroxide (22.4 mg mmol dissolved water (10 europium chloride hexahydrate (24.5 mg 0.067 mmol added 30 dried white solid 44-Hydroxyphenylboronicacidpinacolester (220 mg 1 1,2-dibromoethane (940 mg 5 K2CO3 (690 mg 5 mmol acetonitrile (20 mL heated 70 °C 24 h filtered washed CH2Cl2. concentrated reduced pressureresidue dissolved CH2Cl2 (50 mL washed NaCl organic phase concentrated crude purified column chromatography silica gel petroleum ether/ethyl acetate 20:1) compound 4 white powder 80% yield 1H NMR (400 MHz CDCl3 δ 7.75 6.90 4.32 3.64 1.33 13C NMR (100 MHz CDCl3 δ 160.6 136.6 114.0 83.6 67.6 28.9 24.9. C14H21BBrO3 327.0767 327.0754 C14H20BBrO3 51.42 compound 4 (327 mg 1 3 (210 mg Pd(PPh3)4 (70 mg 0.06 Na2CO3 (680 mg 6.4 mmol added water (4 mL dimethoxyethane (30 refluxed N2 90 °C 24 h solvent removed vacuum residue extracted dichloromethane purified silica gel column petroleum ether/ethyl acetate (20:1)NMR MHz CDCl3 7.47 Hz 7.17 6.93 4.32 6.2 3.66 1.94 13C NMR (100 MHz CDCl3 δ 158.0 141.9 140.5 127.0 126.9 125.8 121.5 115.2 68.0 28.9 14.5. C31H25Br2F6O2S2 766.9546 766.9537 C31H24Br2F6O2S2 48.58 3.16 48.51 3.19 2 (383 mg 0.5 mmol dissolved acetonitrile (10 1-methylimidazole (410 mg 5 mmol added stirred 80 °C 12 h precipitate washed diethyl ether three times OF-1 90% yield 1H NMR (400 MHz DMSO-d6 δ 9.20 7.83 7.73 7.58 Hz 7.39 7.02 8.6 Hz 4.61 Hz 4.39 4.7 Hz 3.88 1.93 13C NMR (100 MHz DMSO-d6 δ 158.1, 142.0 140.7, 137.5, 127.2 126.4 125.4 124.0 123.3 121.8 115.8 66.48.8 36.3 14.5. HRMS [M-2Br]2+ C39H36F6N4O2S22+.1086.1081 C39H36Br2F6N4O2S2 C 50.33; H 3.90 N 6.02 C 50.39 H 3.98 N 5.94.Preparation QR commercial PET film printed blue background QR-pattern printed blue PET film QR code scanned smartphone APP.Supplementary Review Additional FilesSupplementary Movie 8 | 49.6 | 0.463204 |
10.1038/s41467-020-17105-8 | PMC7335163 | The sodium-leak channel NALCN controls the resting membrane potentials of neurons. Here, the authors identified two subunits of NALCN, UNC80 and UNC79. Domains in UNC80, which are mutated in individuals with intellectual disability, interact to achieve the dendritic localization of NALCN complex. | The sodium-leak channel NALCN forms a subthreshold sodium conductance that controls the resting membrane potentials of neurons. The auxiliary subunits of the channel and their functions in mammals are largely unknown. In this study, we demonstrate that two large proteins UNC80 and UNC79 are subunits of the NALCN complex. UNC80 knockout mice are neonatal lethal. The C-terminus of UNC80 contains a domain that interacts with UNC79 and overcomes a soma-retention signal to achieve dendritic localization. UNC80 lacking this domain, as found in human patients, still supports whole-cell NALCN currents but lacks dendritic localization. Our results establish the subunit composition of the NALCN complex, uncover the inter-subunit interaction domains, reveal the functional significance of regulation of dendritic membrane potential by the sodium-leak channel complex, and provide evidence supporting that genetic variations found in individuals with intellectual disability are the causes for the phenotype observed in patients. | IntroductionThe resting membrane potentials (RMPs) of mammalian neurons vary significantly among cells from different brain regions and within the same region. In some neurons such as the spiny striatal neurons, they can be as polarized as approximately −90 mV, close to the equilibrium potential of K+ (EK), whereas in other neurons such as the spontaneously firing neurons dissociated from cerebellar nuclei and hypothalamus, they can be as depolarized as approximately −40 mV1–3. Similarly, RMPs recorded from the same region, such as the suprachiasmatic nucleus (SCN), span a large range between approximately −70 to −50 mV4. In addition, some bistable neurons can have two RPMs that are different by as much as 30 mV, depending on the history of synaptic input or time of the day (for a review)5.A major mechanism to generate RMPs above EK is via Na+ conductances. The resting Na+ permeability (PNa) is very small (a few percent of PK), but regulating the permeability ratio (PNa/PK) provides a powerful way to control RMPs6. The resting K+ counductances are formed by many K+ channels including the 15 K2P leak channels and several voltage-gated K+ channels (for reviews)7,8. The resting Na+ conductance is contributed by several types of ion channels. In many CNS neurons, TTX-sensitive, voltage-gated Na+ channels (NaVs) generate persistent and resurgent currents, particularly at more depolarized RMPs (for reviews)9,10. At more hyperpolarized membrane potentials (MPs), the hyperpolarization-activated cation channels (HCNs) are active in many neurons11. In most neurons, there is also a TTX-resistant, voltage-independent true Na+ leak conductance formed by the NALCN channel1,3,12,13. Consistent with NALCN’s major contribution, RMPs recorded from neurons with NALCN disrupted are hyperpolarized compared with wild-type (WT) by ~10–20 mV, and are little sensitive to changes in extracellular Na+ concentration, as found in hippocampal, retrotrapezoid nucleus, SCN, and spinal cord neurons12,14–16. In the midbrain dopaminergic neurons, NALCN contributes to the spontaneous activities, and is inhibited by the D2 dopamine receptor and GABA-B receptor17. NALCN is also activated by neuropeptides substance P (SP) and neurotensin. Its contribution to SP’s excitatory action is dominant despite the peptide’s extensively studied modulation of several K+ channels: neurons lacking NALCN are not excited by SP, as found in several types of neurons from hippocampi, VTA, RTN, the pre-Botzinger complex and spinal cord15,16,18,19. In addition, the channel is inhibited by extracellular Ca2+ through the Ca2+-sensing receptor and G-proteins. Decreases in extracellular [Ca2+] ([Ca2+]e) activates the channel20. In NALCN knockout (KO) mice, drop in [Ca2+]e to sub-millimolar no longer excites neurons as in WT, suggesting that an increase in Na+ leak is a major mechanism by which lowering [Ca2+]e leads to neuronal excitation20,21.Mutant animals from various species have established NALCN as one of the most essential ion channels. NALCK KO mice have severe apnea and die within 24 h of birth12. In C. elegans and Drosophila, NALCN mutant animals have disrupted locomotion, abnormal circadian rhythms, and altered sensitivity to anesthetics22–26. In humans, NALCN deficiency is associated with severe neurological phenotypes including hypotonia, central apnea, inability to sit or stand, lack of speech development, absence of meaningful communication, and severe intellectual disability27–33.NALCN belongs to the superfamily of 24 transmembrane-spanning ion channel proteins that contain four homologous repeats of six transmembrane-spanning segments (4 × 6TM)21,34,35. This superfamily also includes the family of ten voltage-gated Ca2+ channels (CaVs) and the family of nine NaVs. The protein complexes of CaV and NaV families have been extensively studied38,36,37. In contrast, the subunit composition of NALCN has not been determined. In C. elegans and Drosophila, NALCN orthologs (Nca-1, Nca-2 in C. elegans, NA in Drosophila) genetically interact with several other genes including UNC-79, UNC-80, NLF-1, UNC-7, and Synaptojanin22–25,39,40. NLF-1 is localized in the endoplasmic reticulum where it facilitates trafficking NALCN to the plasma membrane39. In mouse brains, NALCN physically interacts with both UNC79 and UNC8018,20. However, it is not known whether UNC80 and UNC79 are simply two of many NALCN-interacting proteins or they are exclusively associated with NALCN and can be considered as bona fide auxiliary subunits.Despite the large size of UNC80 (3326 aa in the mouse isoform, 3258 in humans) and its high degree of conservation among vertebrates (97% identity between mouse and human, 33% between mouse and C. elegans), there is no functional domain predicted or experimentally identified. The in vivo function of UNC80 in mammals is also not established as no KO mouse has been reported. We and others have recently found human individuals with variations along the open reading frame of UNC80, including truncations at the very N-terminal (e.g., R51*) and the C-terminal ends (e.g., L2586*) of the protein (Supplementary Fig. 1). Those individuals have hypotonia, impaired speech development, severe intellectual disability, and premature death41–47. The genetic variations were largely discovered using whole-exome sequencing (WES). As for any WES-based diagnosis, the individuals also have other detected genetic variations in exons and likely also have undetected intronic variations that cannot be convincingly excluded as causes of the phenotypes. As a result, the causal relationship between the UNC80 variations/mutations and the severe diseases remains to be firmly established.In this study, we generated targeted UNC80 mutations in the mice to test the relationships. UNC80 null, like those of NALCN and UNC79, have severe apnea and die shortly after birth. The severe phenotype provides the strongest evidence that the phenotypes in the UNC80 human patients are the results of the mutations detected with WES. We also used the mutant mice to reveal UNC80 domains important for inter-subunit interaction and dendritic localization.ResultsTargeted disruption of UNC80 leads to severe apnea and neonatal lethalityTo test whether disruption in UNC80 is sufficient to lead to severe phenotypes in mammals, we used the CRISPR/Cas9 technique to generate a KO mouse line with UNC80 truncated at V47 (thereafter called UNC80 KO; Fig. 1a, b), close to R51, the position of a truncation found in several human patients [R51*, 44]. This truncation removes 3279 of 3326 residues of the protein (GenBank # NM_175510 as coordinate). As expected, an antibody raised against the C-terminal 15 residues failed to detect any UNC80 protein in whole brain lysate from mutant pups, confirming the lack of UNC80 protein expression (Fig. 1c).Fig. 1Targeted disruption of UNC80 leads to apnea and neonatal lethality.a The design of UNC80 knockout (KO) using the CRISPR technique to delete exon 3. Exon 3 sequence is in capital and the surrounding introns are in lower case. The 5′ and 3′ target sequences including the PAM motif (XGG) against which the two CRISPR sgRNAs targeted are underlined. Deleted sequences including exon 3 and part of the surrounding introns are shaded. Deletion of exon 3 (total of 157 nucleotides) leads to truncation after V47. The codon encoding R51 (CGA) corresponding to the residue mutated to a stop codon found in human patients are in red. PCR primers used for genotyping in b are in italic and boxed. b Genotyping PCR products from WT (+/+), heterozygote (+/−), and homozygous KO (−/−) pups using primers in a. c Total brain proteins from +/+ and UNC80 −/− were blotted with anti-UNC80 (upper), anti-NALCN (middle), or anti-UNC79 (lower) antibody. d Representative appearances of WT and KO P0 pup. For (b) and (c), three or more independent repeats were performed with similar results. For apnea phenotype in the KO, see Supplementary Movie 1. Source data are provided as a Source Data file.Heterozygous UNC80 KO mice were viable, fertile, and had no gross abnormality. From matings between heterozygous, pups were born with genotypes close to a Mendelian fashion (47 litters, 131+/+, 189+/−, and 85−/−). Homozygous mutant (−/−) pups appeared normal at birth (Fig. 1d). However, no −/− pups survived beyond 24 h of birth. Close inspection revealed that −/− pups had severe apnea (Supplementary Video 1). The apnea and neonatal lethality phenotypes are similar to those found in the NALCN KO12. In humans, the phenotypes found in individuals with UNC80 null mutations and those with NALCN mutations are also similar to each other27,28,44. The severe phenotypes caused by the targeted disruption in UNC80 strongly support a causal relationship between the UNC80 genetic variations and the phenotypes found in the human individuals.UNC80 is required for the NALCN-mediated TTX-resistant Na+ leak current and its regulationBecause of the similar phenotypes in UNC80 and NALCN KOs and the association between the two proteins, we compared the NALCN-dependent Na+ leak current (ΔILNa) in WT and UNC80 KO neurons. Due to the small sizes of the whole-cell currents (~10 pA), we measured ΔILNa as the change in the sizes of holding current (at −70 mV) when bath [Na+] was lowered from 140 to 14 mM, in the presence of TTX and Cs+ to block Navs and HCNs, respectively3,12 (Fig. 2a). The TTX-resistant Na+ leak current at subthreshold MPs is mediated by NALCN and is abolished in NALCN KO12,14,15,19,48.Fig. 2UNC80 is required for the TTX-resistant Na+ leak current and its regulation by extracellular Ca2+.a, b Representative TTX-resistant Na+ leak current recorded in hippocampal neurons cultured from WT (+/+) (a) and UNC80 KO (−/−) (b) pups. Each trace represents 1 s of currents recorded at −70 mV in bath solutions with varying [Na+] (140 or 14 mM) and [Ca2+] (2 or 0.1 mM). The Na+-leak current (ΔILNa) is calculated as the difference of current sizes between those recorded under 140 mM [Na+] and 14 mM [Na+], as indicated by dashed lines. c–e Similar to a, but recorded from UNC80 KO neurons transfected with UNC80 (c), NALCN (d), or UNC79 (e) cDNA. f Averaged sizes of currents in a–e. ILCa is the current activated by lowering [Ca2+] from 2 to 0.1 mM, as defined as the size difference of ΔILNa under the two [Ca2+] conditions). Data are presented as mean values ± SEM. Numbers of neurons are in parentheses. In (f), two sample t test of each group against the “+/+” group (n = 5) was performed: “−/−” (n = 8, p ≤ 0.001), “−/−; UNC80” (n = 6, p = 0.326), “−/−; NALCN” (n = 5, p ≤ 0.001), “−/−; UNC79” (n = 7, p ≤ 0.001). Asterisk “*” indicates p < 0.05. Source data are provided as a Source Data file.Compared with WT, hippocampal neurons cultured from UNC80 KOs had drastically decreased ΔILNa (WT: 12.0 ± 1.5 pA, N = 5; KO: 2.4 ± 0.4 pA, N = 8) (Fig. 2a, b, f). Transfecting an UNC80 cDNA driven by a strong CMV promoter into the KO neurons fully restored ΔILNa, suggesting that the reduction of ΔILNa in the KO neurons was not a result of nonspecific developmental defect or off-target disruption of another gene by the CRISPR technique (Fig. 2c, f).In mammalian neurons, a drop in extracellular Ca2+ concentration ([Ca2+]e), found under both physiological and pathophysiological conditions, generally excites neurons49. Lowering [Ca2+]e to sub-millimolar activates an inward current50–53. We previously found that the current results from an increase in the NALCN-dependent current (INALCN)20. Consistent with the obligate role of UNC80 in the regulation of INALCN by [Ca2+]e, lowering [Ca2+]e from 2 to 0.1 mM increased ΔILNa in the WT by approximately threefold, but did not lead to obvious increase of ΔILNa in the KO neurons (Fig. 2b, f). Again, transfecting UNC80 cDNA fully rescued the Ca2+ sensitivity (Fig. 2c, f).In total brain protein extracts, the level of NALCN protein in the KO was comparable to that in the WT (Fig. 1c). The drastically smaller ΔILNa in the KO neurons suggests that UNC80 potentiates INALCN. Increasing the level of NALCN protein via NALCN cDNA transfection in the UNC80 KO neurons also increased ΔILNa (Fig. 2d, f), suggesting that NALCN can form channels in the absence of UNC80. However, ΔILNa without UNC80 was not potentiated by lowering [Ca2+]e, further confirming that INALCN’s sensitivity to [Ca2+]e requires UNC80.In UNC80 KO brains, the UNC79 protein level was also drastically lower (Fig. 1c). We previously demonstrated that in UNC79 KO neurons, INALCN is insensitive to [Ca2+]e, presumably because of a lowered UNC80 level20. Transfecting UNC79 cDNA into the UNC79 KO neurons could rescue ΔILNa’s [Ca2+]e sensitivity20. In UNC80 KO neurons, however, UNC79 transfection could not restore ΔILNa, suggesting that the large reduction of ΔILNa in the UNC80 KO was not due to the lack of UNC79 (Fig. 2e, f). These data further support the idea that UNC79 indirectly controls NALCN function through UNC8021.Haploinsufficient reduction in UNC80 function is associated with severe intellectual disabilityIn both humans and mice, individuals heterozygous for UNC80 null mutations develop normally, are fertile and do not have obvious abnormalities such as lethality and severe intellectual disability, suggesting that a reduction in UNC80 gene dosage by 50% is tolerated by the organisms. To determine the level of UNC80 function (measured as INALCN) below which severe phenotypes are present in humans, we searched for UNC80 mutations that retain residual function at a level below 50%.One individual we came across has hypotonia, feeding difficulties, seizures, developmental delay, intellectual disability, and is nonverbal. His phenotype is similar to, although milder than, those found in the individuals with UNC80 null mutations44. Using WES with samples from the individual and his parents, we detected biallelic variations in the UNC80 gene (Fig. 3a). In one allele, he inherited variations of c.1020G>T and c.1021C>T (p.Q340_P341delinsHS) from his mother. In the other, he inherited c.3883G>C (p.E1295Q) from his father (Fig. 3a). The three residues (Q340, P341, and E1295) mutated in the individual are highly conserved among deuterostome animals, from sea urchins to fishes, and humans (Fig. 3b).Fig. 3Haploinsufficient reduction in UNC80 is associated with severe intellectual disability.a Sanger sequencing chromatograms confirming the whole-exome sequence findings of genetic variations in the proband leading to protein changes of Q340H, P341S (left), and E1295Q (right). Upper: from the proband; lower: control. Codons encoding Q340 (CAG), P341 (CCC) and E1295 (GAA) are underlined. b Protein sequence alignments showing conservation in the Q340, P341 (left) and E1295 (right) regions. Accession numbers of the sequences used are NM_032504 (human), NP_780719 (mouse), XP_015144792 (chicken), XP_009300567 (zebrafish), and XP_011676014 (sea urchin). c Western blots with proteins prepared from non-transfected HEK293T cells and those transfected with mouse UNC80 wild type, Q341H;P342S (corresponding to human Q340H;P341S) or E1296Q (corresponding to human E1295Q) mutants, blotted with anti-UNC80 antibody (upper) or anti-actin (lower) for loading control. More than three independent repeats were performed with similar results. d TTX-resistant Na+ leak current recorded from cultured UNC80 KO hippocampal neurons transfected with wild-type UNC80, E1296Q, Q341H;P342Q, or mixture of the two mutants, as indicated. Representative currents are in the left four subpanels and averaged current amplitudes are summarized in the right. Recordings were from −70 mV and were done with bath solutions with varying [Na+] (140 mM or 14 mM) and [Ca2+] (2 mM (2 Ca) or 0.1 mM (0.1 Ca)) (see Fig. 2 legend for details). Data are presented as mean values ± SEM. Numbers of neurons recorded are in parentheses. In the right bar graph, two sample t test of each group against the “−/−; WT UNC80” group (n = 17) was performed and the p values are as follows. ΔILNa (2 Ca): “−/−; E1296Q” (n = 17, p ≤ 0.001), “−/−; E1296Q + Q341H;P342S” (n = 10, p ≤ 0.001), “−/−; Q341H;P342S” (n = 9, p ≤ 0.001)). ΔILNa (0.1 Ca): “−/−; E1296Q” (p ≤ 0.001), “−/−; E1296Q + Q341H;P342S” (p ≤ 0.001), “−/−; Q341H;P342S” (p ≤ 0.001). ΔILCa (0.1 Ca–2 Ca): “−/−; E1296Q” (p ≤ 0.001), “−/−; E1296Q + Q341H;P342S” (p ≤ 0.001), “−/−; Q341H;P342S” (p ≤ 0.001). Asterisk “*” indicates p < 0.05. Source data are provided as a Source Data file.To determine whether the variations alter UNC80 function, we introduced into mouse UNC80 (97% identical to human UNC80) the corresponding mutations (Q341H;P342S for human p.Q340_P341delinsHS; E1296Q for human p.E1295Q). When transfected into HEK293T cells, the mutant cDNAs generated proteins at levels comparable to those of WT (Fig. 3c). When transfected into UNC80 KO neurons, the Q341H;P342S mutant UNC80 generated little or no ΔILNa at 2 mM or 0.1 mM [Ca2+]e. Thus, the two-amino-acid variation found in one allele of the individual largely disrupts UNC80 function (Fig. 3d).The E1296Q mutant partially restored ΔILNa when expressed in UNC80 KO neurons. Like that from the WT-transfected neurons, the E1296Q -generated current was sensitive to [Ca2+]e. However, the sizes of the currents were only <50% of those from the WT cDNA-transfected ones under both 2 and 0.1 mM [Ca2+]e (Fig. 3d).In the individual carrying the essentially non-functional Q340H;P341S allele and the partially functional E1295Q allele, the UNC80-mediated INALCN is expected to be reduced by ~75%. Consistent with this prediction, co-transfection of both of the UNC80 mutant cDNA constructs at equal amount into UNC80 KO neurons generated ΔILNa with a size of ~25% of those from WT UNC80-transfected ones (Fig. 3d). Taken together, these data suggest that while a reduction of 50% in UNC80 gene dosage is tolerated, further reduction of UNC80 function to below 25% likely leads to severe phenotypes.UNC80 and UNC79 are subunits of the NALCN complexUNC80’s essential physiological roles could be primarily through its regulation of NALCN and/or through its other functions. We used protein fractionation and protein depletion assays to test whether UNC80 is exclusively associated with the NALCN complex. Consistent with being a 24 transmembrane-spanning protein, NALCN from mouse brains co-segregated with the microsomal fraction after centrifugation at >100,000 × g (Fig. 4a). UNC80 and UNC79 do not have obvious predictable transmembrane-spanning segments. Nevertheless, both of them co-segregated with NALCN in the microsomal fraction, suggesting that the two proteins are membrane associated (Fig. 4a). We further tested whether depleting NALCN from solubilized total brain protein also depletes UNC80 and UNC79. To facilitate the depletion, we generated a knock-in (KI) mouse line in which NALCN is fused with a GFP-HA-6xHis triple tag (the NALCN-GFP-HA-His mice, Fig. 4b), against which commercially antibodies and histidine-binding resins suitable for affinity depletion are readily available. Anti-His antibody detected tagged NALCN in protein prepared from the KI brains but not in that from WT (Fig. 4b). The sizes of the Na+-leak current and its Ca2+ sensitivity recorded from neurons cultured from the KI mice (Fig. 4c) are comparable to those of the WT (Fig. 2f). In addition, the KI mice are viable, fertile, and do not have gross abnormality, suggesting that the tagged NALCN functionally replaced the native one. After NALCN was depleted with histidine-binding Ni-column and anti-GFP antibody, both UNC80 and UNC79 became undetectable, suggesting that nearly all UNC80 and UNC79 proteins are in the NALCN complex (Fig. 4d). Based on this finding of apparently exclusive physical association of UNC80 and UNC79 with NALCN, and previous findings of functional and genetic interaction among the three, we propose that UNC80 and UNC79 are auxiliary subunits of the NALCN complex.Fig. 4UNC80 and UNC79 are auxiliary subunits of the NALCN channel complex.a Protein fractionation demonstrating that UNC80 and UNC79 co-segregate with NALCN to the membrane fraction. Total proteins (T) from adult brains were centrifuged at 200,000 × g and separated into the cytosolic (supernatant, S) and microsomal membrane (pellet, P) fractions, as illustrated in the schematic diagram (upper). Each fraction was blotted with anti-NALCN, anti-UNC80, anti-UNC79, or anti-actin antibody. b A knock-in mouse line with NALCN tagged with GFP, HA, and His tags (NALCN-GFP-HA-His mice). Upper, schematic design. Lower, total brain proteins (100 μg) prepared from the triple-tagged mice and wild-type (non-tagged) mice were immunoblotted with anti-His antibody. c TTX-resistant Na+ leak currents recorded from neurons cultured from the KI pups (n = 5). Left, representative current traces. Right, averaged current amplitudes. Recordings were from −70 mV and were done with bath solutions with varying [Na+] (140 mM or 14 mM) and [Ca2+] (2 mM (2 Ca) or 0.1 mM (0.1 Ca)) (see Fig. 2 legend for details). Numbers of neurons recorded are in parentheses. d Protein depletion demonstrating that all UNC80 and UNC79 proteins are associated with NALCN. Total brain protein lysates were prepared from the NALCN-GFP-HA-His mice. NALCN was depleted by incubating with Ni column (binding to 6-His) followed by further immune depletion with anti-GFP antibody. Lysates before (lane 1) and after (lane 2: with Ni-beads, lane 3: with α-GFP agarose) depletion were blotted with anti-NALCN, anti-UNC79, or anti-UNC80 antibody. Anti-actin was used as a control. For (a, b, d), three or more independent repeats were performed with similar results. Data are presented as mean values ± SEM. Source data are provided as a Source Data file.UNC80’s N-terminal half interacts with NALCNUNC80 associates with NALCN when the two are co-transfected into HEK293T cells18. We used co-immunoprecipitation assays to define the regions that mediate UNC80’s interaction with NALCN. The N-terminal half (aa 300–1700) was sufficient for the UNC80–NALCN association. Further deletion from the N- or C-terminal side greatly diminished the association (Fig. 5a).Fig. 5UNC80’s N-terminal half interacts with NALCN.a Co-immunoprecipitation assays to locate fragments on UNC80 required for its association with NALCN. Upper panels, schematic presentation of mouse UNC80 truncation mutants used in the studies. Lower panels, HEK293T cells were co-transfected with NALCN and GFP-tagged full-length or truncated UNC80 containing residues as indicated. Cell lysates were immunoprecipitated (IP) with anti-GFP followed by immunoblotting with anti-GFP (lower) or anti-NALCN (top). GFP was used in “ctrl”. It migrated at ~20 kDa (outside the molecular ranges shown) and is not visible in the blots. More than three independent repeats were performed with similar results. b
INALCN from cells transfected with NALCN and wild-type or truncated UNC80 mutants as indicated. Recordings were done using a ramp protocol from −100 to +100 mV in 1 s (holding voltage Vh = 0 mV). To ensure that the current was not from nonspecific leak, bath cations were replaced with large non-permeant ion NMDG after each recording (see “Methods” for details). Bar graphs show averaged INALCN amplitude (at −100 mV). Two sample t test of each group against the “full-length” group (n = 11) was performed and the p values are: “Mock” (n = 5, p ≤ 0.001), “1–2387” (n = 9, p ≤ 0.001), “1–2554” (n = 5, p = 0.019), “1–2885” (n = 6, p = 0.818), “1–3000” (n = 5, p = 0.933). Asterisk “*” indicates p < 0.05. Data are presented as mean values ± SEM. Numbers of cells are in parentheses. Source data are provided as a Source Data file.We also used deletions to determine the regions functionally important for UNC80’s potentiation of INALCN. When UNC80 and NALCN are co-transfected in HEK293T cells, cellular dialysis with a peptide Src activator via pipette perfusion potentiates INALCN in an UNC80-dependent fashion18,42 (Fig. 5b). UNC80 with residues deleted from the C-terminal end up to aa 2554 was fully functional in supporting INALCN (Fig. 5b). Additional deletion after aa 2387 (Fig. 5b) disrupted the protein’s function. Those findings suggest that aa 1701–2387, though not required for NALCN binding, is essential for UNC80’s ability to potentiate INALCN.UNC80 increases NALCN surface expression in a manner independent of UNC80’s distal C-terminal regionOur finding that the C-terminal 653 residues after aa 2554 appeared to be nonessential for INALCN was surprising (Fig. 5b). We and others previously reported that several patients with three truncational mutations within this C-terminal region had phenotypes including hypotonia, seizure, lack of speech development, and severe intellectual disability42 (Supplementary Fig. 1). Two such individuals carry c.2033del on one allele, leading to a premature truncation at Q678 (p.Q678Tfs*1, a likely null), and c.6657T>A on the other allele, leading to truncation at L2586 (p.L2586*, corresponding to L2654* in mouse UNC80). The C-terminal truncation mutation is in a region apparently not required for the whole-cell INALCN.As expected, UNC80 truncated at L2654 (containing the first 2653 residues, 1–2653, Fig. 6a) potentiated INALCN when transfected into HEK293T cells. It also fully rescued ΔILNa in UNC80 KO neurons (Fig. 6b). The currents generated from the truncation mutant were significantly larger than those generated with full-length UNC80 in both HEK293T cells and in UNC80 KO neurons. Like in WT, ΔILNa generated by the truncation mutant was also potentiated by lowering [Ca2+]e (Fig. 6b).Fig. 6UNC80’s C-terminus truncated in human patients is not required for whole-cell INALCN but is essential for survival.a, b
INALCN (a done as in Fig. 5b, averaged amplitudes at −100 mV given in the bar graph) or Na+-leak current (ΔILNa) (b) from HEK293T cells (a) or cultured UNC80 KO neurons (b recorded with 2 mM Ca2+ (2 Ca) or 0.1 mM Ca2+ (0.1 Ca) in the bath) transfected with full-length or mutant UNC80 truncated at L2654 (aa1–2653). c Surface biotinylation assays. Left, representative western blots. Surface proteins in cells transfected with cDNA combinations as indicated were biotinylated, isolated using streptavidin-conjugated beads, and probed with α-NALCN. Total protein (whole-cell lysate, WCL) was probed with α-NALCN or α-GFP. In the lower panel, GFP alone in the control lane migrated at ~20 kDa (outside the molecular range shown) and is not visible in the blot. Right, quantification. Signal intensity of each band from cells co-transfected with NALCN and full-length or truncated UNC80 was normalized to that from cells co-transfected with NALCN and GFP. d Sanger sequencing of the knock-in mice (L2654*, lower) and WT (upper). In the L2654* mice, the leucine-encoding codon (CTA, L2654) is mutated to a stop codon (TAA,*). e Representative appearances of WT and homozygous L2654* mutant P0 pups. For apnea phenotypes, see Supplementary Movie 2. f ΔILNa recorded from neurons cultured from WT and L2654* pups. Data are presented as mean values ± SEM. In the bar graphs, numbers of recordings or repeats are in parentheses. Two sample t tests were performed. Asterisk “*” indicates p < 0.05. p values are as follows. a (against the “full-length” group): “Mock” (p ≤ 0.001), “1–2653” (p ≤ 0.001), b (between the group of “−/−; UNC80 full-length” (full-length UNC80-transfected) and the group of “−/−; (1–2653)” (truncation mutant-transfected)): ΔILNa (2 Ca) (p ≤ 0.001), ΔILNa (0.1 Ca) (p ≤ 0.001), ΔILCa (0.1 Ca–2 Ca) (p ≤ 0.001), c (against the group of “GFP” (GFP-transfected)): “full-length” (p = 0.025), “1–2653” (p = 0.049). Source data are provided as a Source Data file.NaV and CaV β subunits potentiate channel currents partly by increasing the surface localization of the pore-forming α subunits (for reviews)54,55. We used the surface biotinylation assay to test whether UNC80 facilitates NALCN surface localization and whether the C-terminally truncated UNC80 is also able to do so. When co-transfected with NALCN in HEK293T cells, both the full-length and the UNC80 truncated at L2654 increased the surface fraction of NALCN by approximately threefold, without affecting the total levels of NALCN protein (Fig. 6c). Thus, the C-terminally truncated UNC80 is as efficient as the full-length one in increasing NALCN surface localization.UNC80’s distal C-terminal region is required for survivalA potential explanation for the apparent inconsistence between our findings of seemingly normal in vitro function of the C-terminally truncated UNC80 and the severe phenotypes in the human individuals is that the phenotypes found in humans are not directly caused by the UNC80 mutation but are from other genetic variations yet to be discovered. To test this possibility, we generated a mouse line in which the codon (CTA) encoding L2654 was substituted with a stop codon (TAA) (Fig. 6d). Homozygous mutants (thereafter referred to as L2654*) were born close to a Mendelian ratio (9 litters, total of 76 pups, 24−/−, 22+/+, and 30+/−) (Fig. 6e). In neurons cultured from the mutant pups homozygous for L2654*, ΔILNa was significantly larger than that of WT, and like in WT, was increased by lowering [Ca2+]e (Fig. 6f). Heterozygous mice were viable, fertile and did not have obvious abnormality. However, none of homozygotes survived beyond P1. Similar to the KOs, the newborn pups had severe apnea (Supplementary Video 2). The similar lethal phenotypes in the L2654* and in the null mice suggest that the C-terminal truncational mutations cause severe symptoms in humans.A C-terminal domain overcomes soma-retention for UNC80’s dendritic localizationAnother potential explanation of the severe phenotypes in mice and humans caused by the UNC80 C-terminal truncation could be that the mutant protein, though supporting functional currents, has aberrant localizations. When transfected into cultured hippocampal neurons, C-terminally RFP-tagged UNC80 (UNC80-RFP) was found in both soma and neuronal processes (Fig. 7a, upper panels). This ubiquitous localization was not an artifact of the RFP-tagging since N-terminally GFP-tagged UNC80 (GFP-UNC80) had a similar localization pattern (Fig. 7a, lower panels). In striking contrast, UNC80 truncated at L2654, whether with RFP-tagged at the C-terminus (Fig. 7b, upper panels) or with GFP tagged at the N-terminus (Fig. 7b, lower panels), was restricted in soma and was absent in axons and dendrites, suggesting essential roles of the C-terminus in the trafficking of UNC80.Fig. 7UNC80’s C-terminal domain overcomes soma-retention for dendritic localization.a C-terminally RFP (upper) or N-terminally GFP (lower)-tagged UNC80 was co-transfected with GFP (upper) or RFP (lower) into cultured wild-type hippocampal neurons. Scale bars: 50 μm. b–d Similar to (a) but transfected with UNC80 mutants truncated at L2654 (containing residues 1–2653) (b), lacking the last six residues (containing residues 1–3320) (c) or containing residues 2234–2758 only (d). Representative pictures from similar results of >10 are presented for each condition.We used additional truncations to narrow down the region required for UNC80’s dendritic localization. The last six residues (LDESHV; 3321–3326) of UNC80 contain a class I PDZ binding motif (S/T-X-Φ), and are highly conserved among vertebrates. Like WT, UNC80 lacking the last six residues were detected throughout neuronal processes (Fig. 7c). Similar localization patterns were also observed in UNC80 lacking the last 156 aa (UNC80 (1–3171)) (Supplementary Fig. 2a). Further deleting UNC80 from the C-terminus up to 326 aa (UNC80 (1–3000), Supplementary Fig. 2b) started to compromise UNC80’s localization, as the truncated protein had limited dendritic localization in some neurons (Supplementary Fig. 2b, upper panels) but was restricted to soma in the other (Supplementary Fig. 2b, lower panels). Together, these data suggest that the distal C-terminal ~300 aa (before ~aa 3000), while not required for the generation of whole-cell INALCN, mediates UNC80’s dendritic localization.The requirement of a C-terminal domain for dendritic localization also suggests the existence of a soma-retention signal that prevents UNC80’s diffusion from soma to dendrites. Indeed, UNC80 truncated at aa 2233 (Supplementary Fig. 2c), aa 1265 (Supplementary Fig. 2d), or aa 2387 (Supplementary Fig. 2e), similar to WT, was found in both soma and dendrites, suggesting the existence of the soma-retention domain between ~aa 2300–2600. Consistent with this idea, a fragment of 525 amino acids alone in the region (aa 2234–2657) was sufficient to retain GFP in soma (Fig. 7d).UNC80’s C-terminus interacts with UNC79UNC80 interacts with both NALCN and UNC7920. We used immunoprecipitation assays in transfected HEK293T cells to test whether the disease-associated L2654* truncation affects UNC80’s association with NALCN and UNC79. As expected from its ability to potentiate INALCN, the truncated UNC80 was fully able to associate with NALCN (Fig. 8a).Fig. 8UNC80’s C-terminus contains an UNC79-interacting domain.a Association between UNC80 mutant truncated at L2654 (containing residues 1–2653) and NALCN. Cell lysates from HEK293T cells co-transfected with NALCN and GFP (as control), GFP-tagged full-length UNC80 or the truncation mutant were immunoprecipitated (IP) with anti-GFP and blotted with anti-NALCN. b Lack of association between UNC80 mutant truncated at L2654 and UNC79. Cell lysates from HEK293T cells co-transfected with UNC79 and GFP (as control), GFP-tagged full-length UNC80, or truncated UNC80 containing only residues 1–2653 or 1–1266 were immunoprecipitated (IP) with anti-GFP and blotted with anti-UNC79 (upper). Lower 3 panels: whole lysates were blotted with anti-UNC79, anti-GFP or anti-actin (for loading control). c Mapping UNC80’s UNC79-interacting domain. Cell lysates from HEK293T cells co-transfected with UNC79 and GFP-tagged full-length UNC80, or truncated UNC80 containing residues as indicated were immunoprecipitated (IP) with anti-GFP and blotted with anti-UNC79. d Disruption of UNC79-NALCN association in mouse brains lacking the C-terminal part of UNC80. Brain proteins were prepared from the NALCN-GFP-HA mice and those that also carry the homozygous L2654* mutation in the NALCN-GFP-HA-His background. Upper panel: NALCN was pulled down with anti-GFP antibody and the complex was probed with anti-UNC79. Lower 3 panels: whole lysates were blotted with anti-UNC79, anti-NALCN or anti-actin (lower 3 panels). e Summary of UNC80’s functional domains: NALCN-interacting domain (residues 301–1700), soma-retention domain (2387–2657), soma-retention relieve domain (2758-3000), and UNC79-interacting domain (2758–2947). f, g Reduced UNC80–UNC79 interaction strength associated with intellectual disability. f UNC79 was co-transfected with GFP, GFP-tagged wild-type UNC80 or GFP-tagged R2910Q UNC80 mutant in HEK293T cells. Immunoprecipitates (IP with anti-GFP, upper panel) and whole-cell lysates (lower two panels) were blotted (IB) with anti-UNC79 or anti-GFP as indicated. In the lower panel, GFP alone in the control lane migrated at ~20 kDa (outside the molecular range shown) and is not visible in the blot. g Protein levels normalized to that obtained with WT UNC80. Results were from four independent experiments. In (a–d, f), three or more independent repeats were performed with similar results. Data are presented as mean values ± SEM. Numbers of repeats are in parentheses. Source data are provided as a Source Data file.In contrast to the WT, however, UNC80 lacking the last 554 aa failed to associate with UNC79 (Fig. 8b). Similarly, R1266* (corresponding to human R1265* found in patients43) did not associate with UNC79 (Fig. 8b). Additional truncation mutants located UNC80’s UNC79-association domain within a small region of 190 aa (aa 2758–2947, Fig. 8c). We also tested the inter-subunit association in the L2654* mouse brains. Consistent with the detection of INALCN in the mutant neurons, NALCN protein was present in the mutant brains (Fig. 8d). Unlike in the UNC80 KO brains where UNC79 was undetectable (Fig. 1c), the level of UNC79 protein in the L2654* brain was similar to that of WT (Fig. 8d), suggesting that the physical interaction between UNC79 and UNC80 mediated by the fragment of aa 2758–2947 is not required for the UNC79 stability.Based on co-immunoprecipitation experiments, we previously proposed that UNC80 bridges UNC79 into the NALCN complex20,21. We tested this model in brains using the L2586* mutant where both NALCN and UNC79 are present but the UNC79–UNC80 association is eliminated. Unlike in the WT, the association of UNC79 to the NALCN complex was absent in the L2654* mutant (Fig. 8d). These data suggest that the symptoms in the patients with truncations at the C-terminus (Supplementary Fig. 1) likely result from the inability of UNC80 to recruit UNC79 into the NALCN complex, in addition to mislocalization of UNC80.Reduced UNC80–UNC79 interaction strength is associated with intellectual disabilityThe L2654* mutation leads to neonatal lethality in mice and the corresponding mutation is associated with severe phenotypes in human42. Since the truncation eliminates the UNC80–UNC79 interaction domain (aa 2758–2947) and a highly conserved region in the C-terminal end (aa 2948–3207) outside the domain required for dendritic localization, we searched for more subtle mutations limited to the UNC79–UNC80 interaction to test specifically the function of UNC80–UNC79 interaction. A recently reported individual has a homozygous variation leading to an R2842Q substitution (corresponding to R2910Q in mice)45. R2910 is in the UNC79-interacting domain (aa 2758–2947, Fig. 8c, e) and is highly conserved in vertebrates. The individual can walk independently, has basic communication skills, and her symptoms are much milder than the ones with null UNC80 mutations or with the L2586* mutation. However, she is nonverbal and has severe intellectual disability45. To test whether R2910Q compromises UNC80–UNC79 interaction, we compared the ability of WT and R2910Q-bearing UNC80 to pull-down UNC79 in a co-immunoprecipitation assay. The mutant protein expressed at levels comparable to that of the WT. However, the apparent UNC80–UNC79 affinity was reduced by ~60% in the mutant (Fig. 8f, g).DiscussionIn this study, we established the subunit composition of the NALCN complex and revealed UNC80’s functional domains important for INALCN potentiation, NALCN association, UNC79 association, soma-retention, and dendritic localization. Using KO mice and functional analysis, we also established a causal relationship between various UNC80 mutations and severe phenotypes in humans. In particular, UNC80 mutants truncated at the C-terminal end retain the protein’s somatic function, but lack dendritic localization and cause severe intellectual disability, suggesting the importance for the proper regulation of dendritic resting membrane potential through the NALCN channel.There are now biochemical, functional, and genetic evidences that overwhelmingly support the idea that the Na+-leak channel has at least three subunits: NALCN, UNC80, and UNC79. First, essentially all the UNC80 and UNC79 protein is complexed with NALCN, as depleting NALCN from brain lysates also depletes UNC80 and UNC79. Second, UNC80 is required for the normal sizes and the regulation of INALCN. Third, UNC80 and UNC79 control each other’s protein level. Finally, mutations in any of the three genes have comparable, although nonidentical, phenotypes. These include abnormal locomotion in C. elegans and Drosophila, severe apnea and neonatal lethality in mice, hypotonia and severe intellectual disability in humans22–25,27,28,42–44.Several functional domains of UNC80 are now defined: an UNC79-interacting domain (U7-ID) in the C-terminal half, a NALCN association-domain (NALCN-ID) in the N-terminal half, a soma-retention domain, and a domain required to overcome the soma retention (Fig. 8e). In addition, the residues between 1700 and 2500, though not required for NALCN-association, are essential for INALCN potentiation (Fig. 5b). While the U7-ID is only 190 aa (aa 2758–2947), the NALCN-ID is as large as ~1500 aa and further deletion abolishes the UNC80–NALCN interaction. The requirement of such a large segment for UNC80–NALCN association suggests existence of multiple contacts between the two subunits in the tertiary structure. Further studies with structural analysis are required to reveal the details of the interaction.UNC80 enhances INALCN likely through two major mechanisms. First, it increases the number of NALCN molecules on cell surface by controlling channel trafficking/insertion (Fig. 6c). This mechanism alone, however, is unlikely sufficient to explain the potentiation of INALCN by UNC80. UNC80 increases NALCN surface expression by approximately threefold when tested in HEK293T cells (Fig. 6c). In the UNC80 KO, which has a total NALCN protein level comparable to that of WT, INALCN is reduced by >> threefold, especially when recorded at 0.1 mM [Ca2+] (Fig. 2). In addition, both the full-length and C-terminally truncated UNC80 increase NALCN surface expression to comparable levels. INALCN with the truncated UNC80, however, is significantly larger than that with the full-length UNC80, both in transfected HEK293T cells and in neurons (Figs. 5, 6). The second mechanism by which UNC80 potentiates INALCN can be through the increase of channel opening via protein modification or inter-subunit interaction. UNC80 is known to mediate increases in INALCN by a Src-kinase pathway that can be activated by neuropeptides15,16,18,19,56. The resting level of INALCN presumably is influenced by the basal Src activity. The potentiation of NALCN by the auxiliary subunit UNC80 is reminiscent of that of CaV channels by the β subunits via the dual mechanism involving increase of surface expression level of the α1 subunits and the potentiation of channel opening57,58. Finally, overexpressing UNC80 can further increase INALCN above the level in WT (Fig. 2f), suggesting that UNC80 availability is a limiting factor in INALCN. Regulating the protein level or modification of UNC80 could be another way to regulate the basal excitability of neurons.The NALCN-mediated Na+ conductance is extremely small (~0.1 nS whole cell), generating only ~10 pA inward current in the hippocampal neurons at rest, almost at the detection limit of whole-cell patch clamp. This conductance is ~100- and 10-fold smaller than the peak NaV and CaV conductances, respectively, making its contribution to the total conductance negligible during action potentials. However, at subthreshold MPs of approximately −70 mV, at which NaVs and HCNs are minimally open, NALCN is likely a major contributor for Na+ permeability.Neurons lacking NALCN, though hyperpolarized toward EK by ~10–20 mV, are at least partially functional and are able to generate action potentials12,14–16. Indeed, human patients with null NALCN or UNC80 mutations have normal muscle stretch reflexes, can raise hands above midline, and are able to perceive sounds, which require the basic functions of the nervous system to generate and transmit action potentials. It is perhaps surprising that the nervous system with neurons hyperpolarized by >10 mV still maintains some of the complex functions. However, the individuals lack eye fixation, normal communication and speech synthesis, and have some of the most severe intellectual disability. NALCN perhaps contributes to the complexity of the nervous system for higher cognitive function by helping generate and regulate RMP heterogeneity.It’s known that in mammalian neurons the dendritic MPs can be significantly different from those of soma, and can fluctuate during behavior, as shown in brain slices and in freely behaving animals59–62. Voltage-gated ion channels such as KVs, CaVs, Navs, and HCNs have been discovered in dendrites63–65 for reviews). Those channels regulate dendritic excitability and information integration. When transfected into cultured hippocampal neurons, NALCN, UNC80, and UNC79 are detected in soma, axons and dendrites. Similar wide-spread localizations of the complex were also observed in Drosophila and C. elegans22–24,26. Recording INALCN from dendrites is technically very challenging because of the small sizes of the current. Given the dominant function of UNC80 in controlling INALCN, the localization of UNC80 likely regulates dendritic excitability. Like in soma, NALCN in dendrites perhaps primarily functions at MPs between the active ranges of HCNs and NaVs. Because of its small conductance, the channel regulates PNa/PK and RMPs without significantly affecting the total dendritic impedance. Direct dendritic recordings from WT and mutant brain slices will be required to reveal the relative contribution of the NALCN complex to dendritic RMPs.The dendritic localization of UNC80 is not simply by diffusion, but rather requires overcoming a somatic retention signal localized between aa 2234 and 2657 by the C-terminal segment (~aa 2600–3000). The dendrite localization domain also overlaps with the UNC79-interacting domain. In the UNC79 KO, there is a drastic reduction of UNC80 protein, suggesting that UNC79 is required for UNC80 stability20. In the UNC80 KO, the UNC79 protein level is also reduced. Intriguingly, the ability of UNC80 to stabilize UNC79 does not appear to require the UNC79-interacting domain since UNC79 level is normal in mice lacking this domain (L2654*, Fig. 8d). We cannot rule out additional interactions between UNC80 and UNC79 that are not strong enough to stand the detergent-containing immunoprecipitation assays used in the study but is responsible for the stability of each other. How UNC80 is trafficked to dendrites and whether it requires the UNC80–UNC79 interaction require further studies. In neurons with UNC80 lacking the C-terminus required for dendritic localization, we detected robust whole-cell INALCN. Mice without the C-terminus are neonatal lethal, which precludes further behavioral studies. The human patients with similar C-terminal UNC80 truncations have basic motor skills, but lack fine motor coordination, do not have speech development, and have severe intellectual disability. Those severe phenotypes support the in vivo significance of UNC80’s dendritic localization and the importance of the proper regulation of dendritic RMPs.MethodsAnimalsAnimal uses were approved by the University of Pennsylvania IACUC. Mice were housed under a 12 h light/dark cycle in rooms with ambient temperature of 19–26 °C and 30–70% humidity. KO and KI mouse lines were generated using the CRISPR/Cas9 technique66. Single-guide RNAs (sgRNAs) and Cas9 RNA were synthesized using in vitro transcription with MEGAshortscript T7 kit (for sgRNAs) or mMESSAGE mMACHINE T7 ULTRA kit (for Cas9 RNA), and purified using the MEGAclear kit (Life Technologies). The sgRNA-targeted sequences used for the generation of each line were (with PAM sequences in italic): TGAGTTCTATAATTATCTTT [TGG] and TCATCGTTGCCATTTATAAAA [GGG] (for UNC80 knockout), TATCCAGTCTCTTTAGAATG [AGG] (for UNC80 L2654* KI), and TGACCTCCTGGATATTTAGA [TGG] (for GFP-HA-His triple-tagged NALCN KI). For the NALCN GFP-HA-His KI, the DNA donor contained 1.5 kb genomic DNA (left arm), followed by ~0.8 kb sequence encoding HA and 6 × His-tagged GFP at the C-terminus (GFP-HA-His) to replace the NALCN stop codon, and 6 kb genomic DNA (right arm). For the L2654* KI, single strand DNA (sequence:CATCATGGAGATGCTCCCCATTACTGATTGGTCAGCAGAGGCTGTGAGGCCAGCTCTTATCCTCATT TAA AGAGACTGGATAGAATGTTCAACAAAATCCATAAGATGCCCACCTTGAGGTGAGAAGGC) was used as the donor. Cas9 RNA, sgRNA, and for KI, donor DNA were co-injected into embryos (done by the Transgenic and Chimeric Mouse Core of the University of Pennsylvania). Embryos used for injection were collected from C57BL6/J (JAX) (for UNC80 KO) or B6SJLF1/J (heterozygous for C57BL/6J and SJL/J, for L2654*, and NALCN KI). F0 founders were crossed to C57BL6/J to obtain germ-line transmission. Three independent UNC80 KO sublines (deleted sequences indicated in Fig. 1a), three UNC80 L2654* KI sublines (sequence substitution indicated in Fig. 6d) and one NALCN GFP-HA-His KI line were established. Mice were backcrossed to C57BL6/J for two to ten generations before being used to generate P0 pups for neuronal culture. Littermates were used as controls. Genotyping was performed using PCR, restriction digestion, and/or Sanger sequencing. Sequences of PCR primers used for genotyping the UNC80 KO mice are TGATAACAAATGGGTTGCTATGTGAAGAAG (mU8.CRSP3.P0F, forward primer) and TGGAAATAAACAGTAAATCACAGACTAGG (mU8.CRSP3.P0R, reverse primer). Sequences of PCR primers used for genotyping the NALCN GFP-HA-His KI mice are AGATGACCTCCTGGATATTTAG (NCDWT.PF, NALCN-specific forward primer), GTGAACTTCAAGATCCGCCACAACATCG (EGFP.P3F, eGFP-specific forward primer), and TGAAAAACCCATGCTTGGGTGG (NL.CRKI.gtP1R, NALCN-specific reverse primer).Cell cultureHEK293T was purchased from ATCC and was not retested in-house for mycoplasma contamination. Cells were maintained at 37 °C and 5% CO2 in DMEM (Gibco) medium supplemented with 1× penicillin–streptomycin (Invitrogen) and 10% Fetal Bovine Serum (Atlanta biologicals). Neuronal cultures were made from P0 pups. Genders of the pups were not determined. Hippocampi were dissociated and digested with papain (Worthington). Cultured neurons were plated on 12 mm poly-L-lysine coated coverslips. The starting medium composed of 80% DMEM (Lonza), 10% bovine calf serum (Hyclone), 10% Ham’s F-12 (Lonza), and 0.5× penicillin–streptomycin. Medium was changed the next day (DIV1) to Neurobasal A medium (Gibco) supplemented with 1× B-27 (Gibco), 1× penicillin–streptomycin, L-glutamate (25 µM), and 0.5 mM Glutamax-I (Gibco), and on DIV2, to Neurobasal A medium supplemented with 1× B-27, 1× penicillin–streptomycin and 0.5 mM Glutamax-I.TransfectionHEK293T cells were transfected using PolyjetTM (Signa Gen) transfection reagent. Transfected cells were replated on 12-mm poly-L-lysine coated coverslips ~48 h after transfection. Neurons between DIV 5 and 7 were transfected using Lipofectamine LTX (Invitrogen).cDNA constructsWT non-tagged NALCN (rat), UNC79 (mouse), and UNC80 (mouse) used for transfections were described in20. N-terminally GFP-tagged WT and mutant mouse UNC80 constructs were made in the vector peGFP-C1 cut with EcoRI and ApaI. C-terminally mCherry RFP-tagged UNC80 constructs were made in EcoRI and XhoI cut vector pcDNA3.1(+), with mCherry amplified from the vector pmCherry-C1. Mutations were introduced using PCR fragments and assembled with T4 ligase or with the Gibson Assembly kit (NEB), and were confirmed by restriction digestion and Sanger sequencing.ElectrophysiologyAll experiments were performed at room temperature. Recording were done 48–60 h after transfection. For HEK293 cells, pipette solution contained (in mM) 150 Cs, 120 Mes, 10 NaCl, 10 EGTA, 4 CaCl2, 0.3 Na2GTP, 2 Mg-ATP, 0.002 Src family kinase activator (Santa Cruz), and 10 HEPES (pH 7.4). Bath solutions contained (in mM) 150 NaCl, 3.5 KCl, 1 MgCl2, 20 glucose, 2 CaCl2, and 10 HEPES (pH 7.4). In the NMDG+ bath, Na+ and K+ were replaced with NMDG+. For patch clamp recording with neurons, pipette solution contained (in mM) 120 CsCl, 4 EGTA, 2 CaCl2, 2 MgCl2, 4 Mg-ATP, 0.3 Tris-GTP, 14 phosphocreatine (di-tris salt), and 10 HEPES (pH 7.4). The 140 mM Na-containing bath contained (in mM) 140 NaCl, 5 KCl, 2 (or 0.1) CaCl2, 1 MgCl2, 6 glucose, 2 CsCl, and 10 HEPES (pH 7.4). In the 14 mM Na-containing bath, tris-Cl was used to replace 126 mM NaCl. TTX (1 µM), APV (10 μM), bicuculline (20 μM), and CNQX (20 μM) were applied in the bath to block Nav and synaptic currents. Sodium-leak current (ΔILNa) was measured by subtracting the holding currents obtained with 14 mM Na-containing bath from that obtained with 140 mM Na-containing bath12. Low [Ca2+]-activated leak current (ILCa) was measured by subtracting ΔILNa recorded in high (2 mM) Ca2+ from that obtained in low (0.1 mM) Ca2+. Patch clamp recordings were amplified and filtered at 1 kHz using a MultiClamp 700B amplifier and digitized at 5 kHz with a Digidata 1400A digitizer, both controlled with Clampex 10.4 (Molecular Devices). Data were analyzed using Clampfit 10.4 (Molecular Devices), Excel (Microsoft), and Origin (Origin Laboratory).Immunoprecipitation, Cell surface biotinylation, and western blotting (WB)The anti-NALCN (used at 1 μg ml−1 for WB), anti-UNC79 (used at 1 μg ml−1), and anti-UNC80 (used at 1 μg ml−1) polyclonal antibodies used in this study have been described18,20. Other antibodies were from Sigma-Aldrich (anti-actin, #A5441, used at 1:1000 dilution for WB), Invitrogen (anti-GFP, #A11120, used at 1:1000 dilution for WB), and Rockland (anti-His, #600–401–382, used at 1:1000 dilution for WB). For HEK293T cells, cells from a 35 mm dish were lysed for 30 min with 360 µl IP buffer containing (in mM) 50 Tris-HCl (pH 7.4), 150 NaCl, 1% NP-40, and 1 EDTA supplemented with protease inhibitor cocktail (PIC, Roche). The IP buffer (RIPA) used in Fig. 8 also contained 0.5% (w/v) deoxycholate and 0.1% (w/v) SDS. Lysate was centrifuged at 20,000 × g for 30 min 4 °C. The supernatant was mixed with 1 µg antibody and incubated at 4 °C for 2 h. Samples were mixed with buffer-equilibrated protein A-agarose at 4 °C for 2 h, and washed with binding buffer 3 times (5 min each). Proteins were eluted with 1× lithium dodecyl sulfate (LDS) sample buffer (Invitrogen, #NP0007) containing 100 mM DTT.For cell surface biotinylation assay, transfected cells were washed twice with PBS and incubated with Sulfo-NHS-LC-Biotin (Thermo Fisher, #21335, 0.5 mg ml−1 in PBS) for 30 min at 4 °C. Cells were washed with 100 mM glycine in PBS (1 time) and were lysed with 400 µl IP buffer and rotated at 4 °C for 1 h. Cell lysates were centrifuged 20,000 × g for 30 min at 4 °C. Sixty microliters supernatant was saved for input. Samples were mixed with 100 µl NeutrAvidin agarose (Thermo Fisher, #29200, pre-equilibrated 50% slurry) overnight at 4 °C, and were washed 3 times (5 min each) with IP buffer. Proteins were eluted with 1× lithium dodecyl sulfate (LDS) sample buffer (Invitrogen) containing 100 mM DTT.For total brain protein prep from P0 pups used for WB in Fig. 1c, brains were homogenized in IP buffer without detergent and spun for 10 min at 5000 × g. Supernatant was collected, added with 1% NP-40 and spun for 20 min at 20,000 × g. Protein concentration was measured using the BCA assay (Pierce). Hundred micrograms of total protein was loaded onto each lane. For protein preparation from P0 pups used for immunoprecipitation in Fig. 8d, whole brains were homogenized in IP buffer, solubilized for 30 min, and spun for 30 min at 20,000 × g. One mg total protein in 500 μl was precipitated with 1 µg antibody for 2 h. Samples were mixed with 60 μl buffer-equilibrated protein A-agarose at 4 °C for 2 h, and washed with binding buffer 3 times (5 min each). Proteins were eluted with 30 μl 1× LDS sample buffer with 100 mM DTT and used for WB analysis. For membrane fractionation experiments described in Fig. 4, frozen adult brains were homogenized with dunce in buffer containing 250 mM sucrose, 5 mM Tris (pH 7.4), and 1× PIC. Homogenate was spun for 10 min at 1000 × g followed by another spin for 10 min at 3220 × g for 20 min. Supernatant (used as total protein) was spun for 2 h at 200,000 × g at 4 °C. Pellet (microsomal fraction) was resuspended in the same volume as starting volume of homogenization buffer. Equal volume of total and each fraction was loaded for WB. For affinity depletion experiments (Fig. 4), adult brains from the GFP-HA-His NALCN KI mice were homogenized in binding buffer containing 300 mM NaCl, 20 mM HEPES (pH8.0), and 1× PIC (EDTA-free). Homogenate was spun at 1000 × g for 10 min. The supernatant was mixed with CHAPS (1% final) for 1 h at 4 °C and spun at 18,000 × g for 20 min followed by mixing with pre-equilibrated nickel beads (Qiagen, 1 ml beads per 10 ml) for 1 h. The mixture was spun at 200 × g for 3 min. The supernatant was referred as Ni-depleted. For additional antibody depletion, one ml of supernatant was mixed with 20 μl agarose-conjugated anti-GFP (MBL) and mixed for 2 h at 4 °C.For WB, proteins were separated using NuPage 4–12% Bis-Tris gels (Invitrogen) with denaturing running buffer (MOPS SDS running buffer, Novex), and transferred onto polyvinylidene difluoride (PVDF) membranes for 2–3 h. Membranes were pre-blocked with 5% nonfat dry milk in PBS with 0.1% Tween-20 (PBST) and incubated with primary antibodies at 4 °C overnight or at room temperature for 2 h. Following washes (2 times, 5 min each), membranes were incubated with HRP-conjugated secondary antibody for 1 h at room temperature, followed by 5× washes (5 min each) and additional incubation with detection reagents (Super Signal West Pico ECL from Thermo Scientific, or Hi/Lo Digital-ECL WB detection kit from Kindle biosciences). Signals were detected using X-ray films or with a camera (Fujifilm corporation digital camera (X-A2)).Protein localizationHippocampal neurons on polylysine coated coverslips were transfected for 48 h using Lipofectamine LTX (Thermo Fisher). Cells were washed three times with PBS and fixed with 4% paraformaldehyde (in PBS) for 20 min at room temperature, followed by 5× washes with PBS. Samples were mounted with ProlongTM Diamond Antifade Mountant (Invitrogen). Images were taken using a Nikon Eclipse Ti inverted microscope with a 20× lens using a 543 nm laser (for RFP) or a 488 nm laser (for GFP) for excitation.Clinical studies and whole-exome sequencingThe studies were approved by Western Institutional Review Board (WIRB, Protocol #20120789, Genetic Studies of Patients and their Families with Diseases of Unknown Genetic Etiology). The patient described in Fig. 3 is a 9-year-old male with global developmental delay, epileptic encephalopathy, hypotonia, ataxia, and a muscle biopsy suggestive of mitochondrial dysfunction. He was born by cesarean section due to nuchal cord and heart rate deceleration after an uneventful pregnancy. During the first year of life, he had feeding difficulties and was diagnosed with failure to thrive and gastroesophageal reflux disease. Parents noted he was missing his milestones at 6–12 months. He smiled at 6 months, babbled at 1 year, and crawled and cruised at 3 years. Two MRIs during this time were normal and a muscle biopsy indicated mitochondrial disease. At 4 years of age a repeat muscle biopsy showed Complex I and III deficiency by Oxidative phosphorylation enzymology testing and a decreased Complex I through clear native in-gel OXPHOS enzyme activity testing. Light microscopy of skeletal muscle showed mild to moderate increase in myofiber size variation that appeared to be due to early Type II fiber atrophy and increased presence of lipids. Urine organic acid testing showed significant ketosis and his blood pyruvate levels were high. CSF testing showed a glucose level of 33 (normal 60–80). At 6 years of age he had his first seizure. An additional MRI was normal and an EEG showed bilateral central, centrotemporal and centroparietal spike-wave discharges, with a fairly dramatic activation during sleep. A continuous EEG at around 7 years showed independent epileptiform discharges over the left and right central region during wakefulness, which were occasionally continuous over the right hemisphere as well as continuous epileptiform activity over the right central region during sleep, consistent with electrical status epilepticus. At 9 years of age a physical assessment showed mild hypotonia and choreoathetoid movements. He had good strength in his muscles and was able to walk with support but used a stroller. He was friendly, alert, and interactive; nonverbal but vocalized. He used a communication device to indicate choices. Dysmorphic features included triangular facies, mild frontal bossing, simplified shape of his auricles, and camptodactyly of the 4th and 5th digits of the right and 5th digit of the left hand. Electromyography and nerve conduction velocity testing, CSF exam for neurotransmitters, echocardiogram, lysosomal enzymes, VLCFA, MePCR for PWS, MED12, 7-dehydrocholesterol, lactate, ammonia, CPK, chromosomal microarray, Fragile X, DM1 and DM2, were all negative.The participating family provided written consent and was enrolled into the Center for Rare Childhood Disorders at the Translational Genomics Research Institute (TGen). The patient was under 7 years of age at the time of enrollment and verbal assent was not required according to Western Institutional Review Board (WIRB Protocol #20120789). Whole blood was collected from the proband and parents for DNA extraction using the QiaAmp blood kit. Genomic libraries were prepared with the Illumina’s Truseq DNA Sample Preparation Kit (Illumina, Inc., San Diego, CA, USA), following the manufacturer’s protocol. A final sequencing library was prepared using the TruSeq Exome Library Prep Kit v1 and protocol from Illumina, Inc. WES was performed on the trio using the Illumina HiSeq2000 sequencing platform. Filtered reads were aligned to the Human genome (Hg19/GRC37) using the Burrows-Wheeler transform (BWA v.0.7.5)67. PCR duplicates were removed using Picard v1.9268 and base quality recalibration, indel realignment and SNP and indel discovery were performed using the Genome Analysis Toolkit (GATK v2.5–2)69. Data were filtered against dbSNP137, 1000 Genomes, an in-house exome database, and then annotated with SnpEff 3.2a against Ensembl v66 to identify novel damaging mutations. An annotated variant file containing variants in three family members was filtered to include novel, private, or rare variants according to the Exome Aggregation Consortium (ExAC) database and the Genome Aggregation Database (gnomAD). Mode of inheritance and disease association, followed by detailed analyst assessment for genotype–phenotype correlation, disease mechanism, and literature review were performed. Variants predicted to be damaging by multiple tools—the Combined Annotation Dependent Depletion (CADD), ExAC’s probability of loss-of-function intolerance score and missense z-score (pLi, z-score), Genomic Evolutionary Rate Profiling (GERP), Polymorphism Phenotyping v2 (PolyPhen2) algorithms were considered as candidate genes responsible for the child’s phenotype.The patient is compound heterozygous for variants in the UNC80 gene. On the paternal allele, he inherited the c.3883G>C, p.Glu1295Gln variant. On the maternal allele he inherited the double nucleotide polymorphism *c.1020_1021delGCinsTT, p.Gln340_P341delinsHS. Both variants were confirmed by Sanger sequencing. In the gnomAD database, which presumably only includes individuals without severe phenotype, both of the variations are also present in multiple heterozygous carriers (E1295Q variant in 30 heterozygotes and Q340_P341delinsHS in 152 heterozygotes and 1 homozygote), making a definitive genetic diagnosis of the severe phenotype found in our patient challenging. However, curators of the gnomAD database note that some individuals with severe disease may still be included in the dataset, albeit at a frequency equivalent to or lower than that seen in the general population. Therefore, the one homozygote could be an affected individual that was still included in the dataset.Quantification and statistical analysisOrigin 8.0 software was used for all electrophysiology data analyses. Protein level analysis (Fig. 6c) was done with Image J (US National Institutes of Health). Experimental sample sizes were chosen based on previous experiments to reach statistical significance and were not predetermined using statistical methods. Some of the experiments, but not all, comparing two groups were done in a blind fashion. The experimentalist was blind to the genotype during data collection and data analysis. No data were excluded for analysis. Comparisons between two groups were made using two sample t test (two sided). No adjustments were made. Numeric data were represented as mean ± SEM.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Info
Supplementary Movie_Audio_Data 1
Supplementary Movie_Audio_Data 2
Reporting Summary | nature communications | [
"Article"
] | [
"Ion channels in the nervous system",
"Physiology"
] | resting membrane potentials (RMPs mammalian neurons vary among different brain regions some spiny striatal polarized −90 mV close to potential K+ other spontaneously firing depolarized −40 RMPs same region suprachiasmatic nucleus −70 to −50 mV4 bistable neurons RPMs different by 30 mV depending on history synaptic input time day mechanism RMPs above EK via Na+ conductances resting Na+ permeability (PNa) small few percent of regulating permeability ratio (PNa/PK RMPs6 resting K+ counductances formed by K+ channels 15 K2P leak channels voltage-gated K+ channels resting Na+ conductance contributed by ion channels In CNS neurons TTX-sensitive voltage-gated Na+ channels generate persistent resurgent currents at depolarized RMPs hyperpolarized potentials hyperpolarization-activated cation channels (HCNs active most TTX-resistant voltage-independent Na+ leak conductance by NALCN channel1NALCN’s RMPs neurons disrupted hyperpolarized-type ~10–20 mV sensitive to extracellular Na+ concentration in hippocampal retrotrapezoid nucleus SCN spinal cord neurons12 midbrain dopaminergic neurons NALCN contributes activities inhibited by D2 dopamine GABA-B NALCN activated by neuropeptides P (SP) neurotensin contribution excitatory action dominant modulation K+ neurons lacking NALCN not excited by SP hippocampi VTA RTN pre-Botzinger complex spinal channel inhibited by extracellular Ca2+-sensing receptor G-proteins Decreases extracellular [Ca2+] activates NALCN knockout mice [Ca2+]e sub-millimolar excites neurons Na+ leak lowering [Ca2+]e neuronal animals NALCN NALCK KO mice apnea die 24 h In C. elegans Drosophila NALCN mutant animals disrupted locomotion altered sensitivity anesthetics22–26NALCN deficiency with severe neurological phenotypes hypotonia central apnea inability sit stand lack speech development communication severe intellectual disability27–33.NALCN superfamily 24 transmembrane-spanning ion channel proteins four repeats six segments includes ten voltage-gated Ca2+ channels) nine NaVs protein complexes CaV NaV studied38 subunit composition NALCN determined In C. elegans Drosophila NALCN orthologs interact with genes UNC-79 UNC-80 NLF-1 UNC-7 Synaptojanin22–25 NLF-1 localized endoplasmic reticulum NALCN plasma membrane39 mouse brains NALCN interacts with UNC79 UNC8018 known UNC80 UNC79 NALCN-interacting proteins or associated with NALCN auxiliary subunits large size UNC80 (3326 aa mouse isoform 3258 humans high conservation among vertebrates (97% identity 33% no functional domain predicted identified in vivo function UNC80 in mammals not established no KO mouse reported found human individuals with variations along frame UNC80 truncations at N-terminal C-terminal endsL2586*) protein Fig. 1) individuals have hypotonia impaired speech severe intellectual disability premature genetic variations discovered using whole-exome sequencing individuals have other genetic variations undetected variations causal relationship between UNC80 variations/mutations severe diseases established study generated UNC80 mutations in mice UNC80 null NALCN UNC79 have severe apnea die after birth severe phenotype evidence results mutations WES mutant mice reveal UNC80 domains inter-subunit interaction dendritic localization disruption UNC80 leads to severe apnea neonatal lethalityTo disruption UNC80 used CRISPR/Cas9 technique KO mouse line with UNC80 truncated at V47 KO R51 truncation patients truncation removes 3279 of 3326 residues protein (GenBank # NM_175510 antibody raised against C-terminal 15 residues detect UNC80 protein in brain lysate mutant pups lack of UNC80 protein expression (Fig. disruption UNC80 leads to apnea neonatal lethalitydesign UNC80 knockout CRISPR delete exon 3. Exon 3 sequence capital introns lower case 5′ 3′ target sequences PAM motif underlined Deleted sequences 3 introns shaded Deletion exon 3 157 nucleotides leads truncation after V47 codon encoding R51) red PCR primers genotyping italic boxed Genotyping PCR WT (+/+) heterozygote (+ homozygous KO (−) pups brain proteins +/+ UNC80 −/− blotted anti-UNC80-NALCN anti-UNC79 antibody appearances WT KO P0 pup three repeats similar results apnea phenotype KO Supplementary Movie 1. Source data file.Heterozygous UNC80 KO mice viable fertile no gross abnormality matings pups born genotypes close Mendelian fashion (47 litters 131+/+ 189+/− 85−/−). Homozygous mutant (−/−) pups normal birth no −/− pups survived 24 h birth severe apnea apnea neonatal lethality similar NALCN KO12.phenotypes UNC80 NALCN mutations similar,28 severe phenotypes disruption UNC80 support causal relationship between UNC80 genetic variations phenotypes.UNC80 required for NALCN-mediated TTX-resistant Na+ leak current similar phenotypes UNC80 NALCN association compared NALCN-dependent Na+ leak current (ΔILNa) in WT UNC80 KO neurons~10 measured ΔILNa as change holding current −70 when bath [Na+] lowered from 140 to 14 mM TTX Cs+ (Fig. TTX-resistant Na+ leak current at subthreshold MPs mediated by NALCN abolished in NALCN KO12 2UNC80 required for TTX-resistant Na+ leak current regulation by extracellular Ca2+ TTX-resistant Na+ leak current in hippocampal neurons WT UNC80 currents −70 mV varying [Na+] 14 [Ca2+] (2 0.1 mM). Na+-leak current (ΔILNa) calculated as difference current sizes 140 mM and 14 mM linesc–e UNC80 KO neurons transfected with UNC80 NALCN UNC79 cDNA Averaged sizes currents a–e ILCa activated lowering [Ca2+] 2 to 0.1 mM size difference ΔILNa under Data mean values ± SEM Numbers in parentheses two sample t test against “+/+” group = 5) performed “−/−” 8 p ≤ 0.001) 6 0.326) 5 UNC79” 7 0.001) “*” indicates p < 0.05. Source data file hippocampal neurons from UNC80 KOs decreased ΔILNa (WT 12.0 ± 1.5 pA 5 KO 2.4 ± 0.4 pA 8) (Fig. 2a Transfecting UNC80 cDNA into KO neurons restored ΔILNa reduction not developmental defect disruption CRISPR (Fig. 2c mammalian neurons drop in extracellular Ca2+ concentration excites neurons49 Lowering [Ca2+]e to sub-millimolar activates inward current results from increase NALCN-dependent currentUNC80 INALCN lowering from 2 to 0.1 mM increased ΔILNa WT threefold increase ΔILNa KO neurons (Fig. 2b transfecting UNC80 cDNA rescued Ca2+ (Fig. 2c NALCN protein KO comparable to WT (Fig. smaller ΔILNa KO neurons UNC80 potentiates INALCN Increasing NALCN protein UNC80 KO neurons increased ΔILNa 2d NALCN form channels absence UNC80 ΔILNa without UNC80 not potentiated lowering [Ca2+]e INALCN’s sensitivity requires UNC80 UNC80 KO brains UNC79 protein level lower (Fig. UNC79 KO neurons INALCN insensitive to [Ca2+]e lowered UNC80 Transfecting UNC79 cDNA KO could rescue [Ca2+ UNC80 KO neurons UNC79 transfection restore ΔILNa reduction ΔILNa UNC80 KO not due lack UNC79 (Fig. 2e support UNC79 controls NALCN functionHaploinsufficient reduction UNC80 function severe intellectual disabilityIn humans mice individuals heterozygous for UNC80 null mutations develop normally fertile abnormalities lethality severe intellectual disability reduction UNC80 gene dosage by 50% tolerated UNC80 function severe phenotypes searched for mutations function below 50% individual has hypotonia feeding difficulties seizures developmental delay intellectual disability nonverbal phenotype similar to milder UNC80 null detected biallelic variations in UNC80 gene. inherited variations c.1020G>T c.1021C>T from mother c.3883G>C (p.E1295Q) from father three residues (Q340 P341 E1295) mutated conserved among deuterostome animals sea urchins fishes humans 3Haploinsufficient reduction severe intellectual disability Sanger sequencing chromatograms genetic variations proband protein changes of Q340H P341S E1295Q control Codons encoding Q340 P341 E1295 underlinedProtein sequence alignments conservation Q340 P341 E1295 regions Accession numbers NM_032504 NP_780719 XP_015144792 XP_009300567 XP_011676014 (sea Western blots proteins non HEK293T cells transfected with mouse UNC80 wild type Q341H;P342S E1296Q mutants blotted anti-UNC80 antibody anti-actin for loading control three repeats similar results TTX-resistant Na+ leak current from cultured UNC80 KO hippocampal neurons transfected with wild-type UNC80 E1296Q Q341H;P342Q Representative currents averaged current amplitudes right Recordings from −70 mV bath solutions varying [Na+] [Ca2+] Fig. 2 Data mean values ± SEM Numbers neurons recorded in parentheses graph two sample t test against WT UNC80” group (n = 17) p valuesΔILNa (2/−; E1296Q” 17, p ≤ 0.001) + Q341H;P342S” 10 Q341H;P342S” 9 ΔILNa (0.1 Ca):/−; E1296Q” 0.001) + Q341H;P342S” Q341H;P342S” ΔILCa (0.1 Ca–2 Ca): “−/−; E1296Q” 0.001) + Q341H;P342S” Asterisk “*” indicates p < 0.05. Source data UNC80 function introduced mouse UNC80 mutations (Q341H;P342S E1296Q transfected HEK293T cells mutant cDNAs generated proteins comparable WT (Fig. UNC80 KO neurons Q341H;P342S mutant generated little no ΔILNa at 2 mM or 0.1 mM [Ca2+]e two-amino-acid variation allele disrupts UNC80 function E1296Q mutant restored ΔILNa UNC80 KO neurons E1296Q current sensitive to [Ca2+]esizes currents <50% WT cDNA-transfected ones under 2 0.1 mM [Ca2+]e (Fig. individual carrying non-functional Q340H;P341S allele partially functional E1295Q allele UNC80-mediated INALCN reduced ~75% co-transfection UNC80 mutant cDNA constructs UNC80 KO neurons generated ΔILNa size ~25% WT UNC80-transfected ones (Fig. suggest reduction 50% UNC80 gene dosage tolerated reduction UNC80 function below 25% leads severe phenotypes.UNC80 UNC79 subunits NALCN roles NALCN other functions used protein protein depletion assays UNC80 associated NALCN complex NALCN mouse brains co-segregated microsomal fraction after centrifugation at >100,000 × g (Fig. 4a). UNC80 UNC79 transmembrane-spanning segments co-segregated NALCN microsomal fraction proteins associated (Fig. depleting NALCN brain protein depletes UNC80 UNC79 generated knock-in mouse line NALCN fused GFP-HA-6xHis triple tagantibodies histidine-binding resins affinity depletion available Anti-His antibody detected NALCN in KI brains not WT sizes Na+-leak current Ca2+ from neurons KI mice comparable to WT KI mice viable fertile gross abnormality NALCN replaced native After NALCN depleted with histidine-column anti-GFP antibody UNC80 UNC79 undetectable all UNC80 proteins in NALCN complex (Fig. 4d). association UNC80 UNC79 with NALCN genetic interaction UNC80 UNC79 auxiliary subunits of NALCN complex 4UNC80 UNC79 auxiliary subunits NALCN complex Protein fractionation UNC80 UNC79 co-segregate with NALCN membrane fraction proteins from adult brains centrifuged at 200,000 × g separated into cytosolic microsomal membrane fractions Each fraction blotted with anti-NALCN anti-UNC80 anti-UNC79 anti-actin antibody knock-in mouse line with NALCN tagged GFP HA His tagstotal brain proteins (100 μg triple-tagged wild immunoblotted with anti-His antibody TTX-resistant Na+ leak currents from neurons KI pups (n = 5) current traces averaged current amplitudes Recordings from −70 mV bath solutions varying [Na+] (140 mM 14 mM [Ca2+] (2 Fig. 2 Numbers neurons in parentheses Protein depletion UNC80 UNC79 proteins associated with NALCN brain protein lysates from-GFP-HA-His mice NALCN depleted incubating with Ni column immune depletion with anti-GFP antibody Lysates blotted with anti-NALCN anti-UNC79 anti-UNC80 antibody Anti-actin control three repeats similar results Data mean values ± SEM Source data.UNC80’s N-terminal half interacts with associates co-transfected into HEK293T co-immunoprecipitation assays interaction N-terminal half sufficient for UNC80–NALCN association deletion from-terminal diminished association (Fig. N-terminal half interacts with NALCNCo-immunoprecipitation assays fragments UNC80 association NALCN Upper panels mouse UNC80 truncation mutants Lower panels HEK293T cells co-transfected with NALCN GFP-tagged full-length truncated UNC80 residues Cell lysates immunoprecipitated anti-GFP immunoblotting anti-GFP anti-NALCN GFP used migrated ~20 kDa not visible blots three repeats similar results INALCN cells transfected NALCN wild-type truncated UNC80 mutants Recordings ramp protocol −100 to +100 mV 1 s voltage Vh = 0 bath cations replaced with large non-permeant ion NMDG recording Bar averaged INALCN amplitude −100 mV). Two sample t test group against “full-length” group (n = 11) p values “Mock” 5 ≤ 0.001) “1–2387” 0.933) Asterisk “*” indicates p < 0.05. Data mean values ± SEM Numbers cells parentheses Source data file used deletions determine regions important UNC80’s potentiation INALCNUNC80 NALCN co-transfected in HEK293T cells cellular dialysis peptide Src activator potentiates INALCN UNC80-dependent UNC80 residues deleted C-terminal to aa 2554 functional INALCN Additional deletion after aa 2387 disrupted function aa 1701–2387 not required NALCN binding essential for UNC80’s INALCN.UNC80 increases NALCN expression independent C-terminal C-terminal 653 residues after aa 2554 nonessential for INALCN patients with three truncational mutations C-terminal hypotonia seizure lack speech development severe intellectual disability42 Two individuals carry c.2033del one allele premature truncation at Q678 c.6657T>A truncation at L2586 UNC80) C-terminal truncation mutation not required for whole-cell INALCN UNC80 truncated at L2654 potentiated INALCN transfected HEK293T cells rescued ΔILNa in UNC80 KO neuronscurrents truncation mutant larger full-length UNC80 in HEK293T cells UNC80 KO neurons ΔILNa potentiated lowering [Ca2+ (Fig. 6UNC80’s C-terminus truncated not required for whole-cell INALCN essential for survival INALCN amplitudes −100 mV Na+-leak current from HEK293T cells UNC80 KO neurons 2 mM Ca2+ or 0.1 mM Ca2+ transfected with full-length mutant UNC80 truncated at L2654 Surface biotinylation assays proteins cells transfected biotinylated isolated streptavidin beads probed with α-NALCN Total protein probed with α-NALCN or α-GFP GFP lane migrated at ~20 kDa not visible blot Signal intensity from cells co-transfected with NALCN full-length truncated UNC80 normalized to co-transfected NALCN GFP Sanger sequencing knock-in mice (L2654 WT L2654* leucine-encoding codon mutated to stop codonappearances WT L2654* mutant P0 pups apnea phenotypes Supplementary Movie 2. ΔILNa recorded neurons WT L2654* pups Data mean values ± SEM bar graphs recordings parentheses Two t tests Asterisk “*” p < 0.05. values “Mock” (p ≤ 0.001) “1–2653” (p ≤ 0.001) (between ΔILNa (2 Ca) ΔILNa (0.1 Ca ΔILCa (0.1 Ca–2 Ca “full-length” (p = 0.025) “1–2653” (p = 0.049) Source data file.NaV CaV β subunits potentiate channel currents surface localization pore-forming α subunits surface biotinylation assay UNC80 NALCN localization C-terminally truncated UNC80 co-transfected NALCN HEK293T cells full-length UNC80 truncated L2654 increased surface fraction NALCN threefold without affecting levels NALCN protein. C-terminally truncated UNC80 efficient full-length NALCN surface localizationUNC80’s distal C-terminal region for inconsistence between normal vitro function truncated UNC80 severe phenotypes human not caused by UNC80 mutation from other genetic variations generated mouse line L2654 substituted with stop codon Homozygous mutants L2654*) born close to Mendelian ratio (9 litters 76 pups 24−/− 22+/+ 30+/−) neurons from mutant pups L2654 ΔILNa larger than WT increased by lowering [Ca2+]e Heterozygous mice viable fertile abnormality homozygotes survived beyond P1 newborn pups had severe apnea similar lethal phenotypes in L2654* null mice suggest C-terminal truncational mutations cause severe symptoms in humans C-terminal domain overcomes soma-retention for UNC80’s dendritic severe phenotypes UNC80 truncation mutant protein aberrant localizations transfected into hippocampal neurons RFP-tagged UNC80) found in soma neuronal processesubiquitous localization artifact RFP-tagging GFP-tagged UNC80 similar localization pattern UNC80 truncated at L2654 restricted in soma absent in axons dendrites roles C-terminus trafficking UNC80 7UNC80’s C-terminal domain overcomes soma-retention for dendritic localization C N-tagged UNC80 co-transfected into cultured wild-type hippocampal neurons Scale bars 50 μm transfected UNC80 mutants truncated at L2654 residues 1–2653) lacking last six residues 1–3320) residues 2234–2758 only Representative pictures similar results >10 each condition additional truncations region for UNC80’s dendritic localization last six residues 3321–3326) UNC80 contain class I PDZ binding motif conserved among vertebrates UNC80 lacking last six residues detected neuronal processes (Fig. 7c). Similar localization patterns in UNC80 lacking last 156deleting UNC80 from C-terminus up to 326 aa localization truncated protein limited dendritic localization in neurons restricted to soma suggest distal C-terminal ~300 aa (before not required for-cell INALCN mediates UNC80’s dendritic localization requirement C-terminal for suggests soma-retention signal UNC80’s diffusion from to dendrites UNC80 truncated at aa 2233 1265 2387 found in soma and dendrites soma-retention domain between ~aa 2300–2600 fragment of 525 amino acids in region (aa 2234–2657) to retain GFP in soma 7d).UNC80’s C-terminus interacts with interacts with NALCN and UNC7920 immunoprecipitation assays in HEK293T cells L2654* truncation UNC80’s association with NALCN UNC79 truncated UNC80 with NALCN 8UNC80’s C-terminus contains UNC79-interacting domainAssociation UNC80 mutant L2654 residues 1–2653) NALCN lysates HEK293T cells co-transfected with NALCN GFP UNC80 immunoprecipitated anti-GFP blotted anti-NALCN Lack association UNC80 mutant L2654 UNC79 lysates co UNC79 GFP truncated UNC80 residues 1–2653 1–1266 immunoprecipitated anti-GFP blotted anti-UNC79 lysates blotted anti-UNC79 anti-GFP anti-actin UNC80’s UNC79-interacting domain lysates co-transfected UNC79 UNC80 UNC80 immunoprecipitated anti-GFP blotted anti-UNC79 Disruption UNC79-NALCN association mouse brains lacking C-terminal part UNC80 Brain proteins prepared from NALCN-GFP-HA mice L2654* mutation NALCN pulled down anti-GFP antibody complex probed anti-UNC79Lower 3 panels lysates blotted with anti-UNC79 anti-NALCN anti-actin UNC80’s functional domains NALCN-interacting domain soma-retention domain (2387–2657) relieve domain (2758 UNC79-interacting domain (2758–2947) Reduced UNC80–UNC79 interaction strength intellectual disability UNC79 co-transfected with GFP-tagged wild-type UNC80 R2910Q UNC80 mutant in HEK293T cells Immunoprecipitates anti-GFP whole-cell lysates blotted with anti-UNC79 or anti-GFP GFP control lane migrated at ~20 kDa not visible Protein levels normalized WT UNC80 Results four experiments three repeats similar results Data mean values ± SEM parentheses Source data UNC80 lacking last 554 aa failed associate with UNC79 R1266* associate with UNC79 Additional truncation mutants located UNC80’s UNC79-association domain within region 190 aa 2758–2947 tested inter-subunit association in L2654* mouse brains NALCN protein present mutant brainsUnlike UNC80 KO brains UNC79 undetectable level UNC79 protein in L2654* brain similar to WT interaction between UNC79 UNC80 aa 2758–2947 not required for UNC79 stability UNC80 bridges UNC79 into NALCN tested in L2586* mutant NALCN UNC79 present UNC79–UNC80 association eliminated Unlike WT association UNC79 to NALCN complex absent in L2654* mutant suggest symptoms in patients with truncations at C-terminus result from inability UNC80 to recruit UNC79 into NALCN complex mislocalization UNC80.Reduced UNC80–UNC79 interaction strength intellectual L2654* mutation leads to neonatal lethality in mice severe phenotypes in truncation eliminates UNC80–UNC79 interaction domain 2758–2947) region C-terminal end searched for subtle mutations UNC79–UNC80 interaction test individual homozygous variation R2842Q substitution R2910Q in mice R2910 in UNC79-interacting domain 2758–2947 conserved in vertebratesindividual communication skills symptoms milder than null UNC80 mutations L2586* mutation nonverbal severe intellectual R2910Q UNC80–UNC79 interaction compared WT R2910Q UNC80 UNC79 mutant protein expressed comparable WT UNC80–UNC79 affinity reduced ~60% mutant (Fig. 8f established subunit composition NALCN complex UNC80’s functional domains for INALCN potentiation NALCN association UNC79 association soma-retention dendritic localization causal relationship between UNC80 mutations severe phenotypes UNC80 mutants truncated-terminal retain somatic function lack dendritic localization cause severe intellectual disability regulation dendritic resting membrane potential NALCN channel biochemical functional genetic evidences support Na+-leak channel three subunits NALCN UNC80 UNC79 UNC80 UNC79 protein complexed with NALCN depleting NALCN depletes UNC80 UNC79 UNC80 required for normal sizes regulation INALCN UNC80 UNC79 control protein level mutations in genes comparable phenotypes abnormal locomotion C.elegans Drosophila apnea neonatal lethality mice hypotonia intellectual disability functional domains UNC80 defined UNC79-interacting domain (U7-ID C-terminal half NALCN association-domain N-terminal half soma-retention domain domain overcome soma retention (Fig. residues between 1700 and 2500 not NALCN-association essential for INALCN potentiation (Fig. U7-ID 190 aa NALCN-ID ~1500 aa deletion abolishes UNC80–NALCN interaction large segment for UNC80–NALCN association suggests multiple contacts subunits structure Further studies.UNC80 enhances INALCN increases NALCN molecules cell surface channel trafficking/insertion unlikely potentiation INALCN UNC80 increases NALCN surface expression threefold in HEK293T cells UNC80 KO NALCN protein level INALCN reduced threefold at 0.1 mM [Ca2+] (Fig. 2) full-length C truncated UNC80 increase NALCN surface expression comparable INALCN truncated UNC80 larger than full-length in transfected HEK293T cells neurons (Figs. 5 6)UNC80 potentiates INALCN channel opening protein modification inter-subunit interaction UNC80 increases INALCN Src-kinase pathway activated by neuropeptides15 resting level INALCN influenced by basal Src activity potentiation NALCN by UNC80 reminiscent CaV channels β subunits increase surface expression level α1 subunits potentiation channel overexpressing UNC80 INALCN above WT (Fig. UNC80 availability INALCN Regulating protein level modification UNC80 basal excitability NALCN-mediated Na+ conductance small (~0.1 nS generating ~10 pA current hippocampal neurons at rest conductance ~100- 10-fold smaller than peak NaV and CaV conductances contribution negligible during action potentials at subthreshold MPs −70 mV NaVs HCNs open NALCN for Na+ permeability.Neurons lacking NALCN hyperpolarized toward EK ~10–20 mV partially functional generate action potentials12 patients with null NALCN or UNC80 mutations have normal muscle stretch reflexes raise hands perceive sounds action potentialsnervous system with neurons hyperpolarized >10 mV maintains complex functions individuals lack eye fixation communication speech synthesis have severe intellectual disability NALCN contributes to complexity RMP heterogeneity in mammalian neurons dendritic MPs different from soma fluctuate during behavior Voltage-gated ion channels KVs CaVs Navs HCNs discovered in regulate dendritic excitability information integration transfected into hippocampal neurons NALCN UNC80 UNC79 detected in soma axons dendrites Similar localizations observed in Drosophila C. Recording INALCN from dendrites challenging small sizes current dominant function UNC80 regulates dendritic excitability NALCN in dendrites functions at MPs between active ranges HCNs NaVs small conductance regulates PNa/PK RMPs without affecting dendritic impedance dendritic recordings reveal contribution NALCN to RMPs dendritic localization of UNC80 overcoming somatic retention signal between aa 2234 2657 C-terminal segment localization domain overlaps with UNC79-interacting domainUNC79 KO reduction UNC80 protein required for UNC80 UNC80 KO UNC79 protein level reduced UNC80 stabilize UNC79 require UNC79-interacting domain UNC79 normal mice lacking domain additional interactions between UNC80 UNC79 responsible stability UNC80 UNC80–UNC79 interaction require further studies neurons UNC80 lacking C-terminus localization detected robust whole-cell INALCN Mice without C-terminus neonatal lethal behavioral studies human patients similar C-terminal UNC80 truncations basic motor skills lack fine motor coordination speech development severe intellectual disability phenotypes support significance UNC80’s dendritic localization importance regulation dendritic RMPs uses approved University of Pennsylvania IACUC Mice housed 12 h light/dark cycle ambient temperature 19–26 °C 30–70% humidity KO KI mouse lines generated CRISPR/Cas9Single-guide RNAs Cas9 RNA synthesized in vitro transcription MEGAshortscript T7 kit mMESSAGE mMACHINE T7 ULTRA kit purified MEGAclear kit sgRNA-targeted sequences TGAGTTCTATAATTATCTTT [TGG] [GGG UNC80 TATCCAGTCTCTTTAGAATG [AGG UNC80 L2654* TGACCTCCTGGATATTTAGA [TGG] GFP-HA-His triple-tagged NALCN NALCN GFP-HA-His KI DNA donor 1.5 kb genomic DNA (left ~0.8 kb sequence HA 6 × His-tagged GFP C-terminus 6 kb genomic DNA (right L2654* KI single strand DNA donor Cas9 RNA sgRNA donor DNA co-injected into embryos Transgenic and Chimeric Mouse Core University of Embryos collected from C57BL6/J UNC80 B6SJLF1/J L2654* NALCN F0 founders crossed to C57BL6/J germ-line transmissionThree UNC80 KO sublines Fig. UNC80 L2654* KI sublines Fig. 6d one NALCN GFP-HA-His KI line established Mice backcrossed to C57BL6/J two ten generations P0 pups neuronal culture Littermates controls Genotyping PCR restriction digestion Sanger sequencing PCR primers KO mice.CRSP3.P0F GFP-HA-His KI AGATGACCTCCTGGATATTTAG TGAAAAACCCATGCTTGGGTGG cultureHEK293T purchased from ATCC not retested mycoplasma contamination Cells maintained at 37 °C 5% CO2 DMEM (Gibco) medium 1× penicillin–streptomycin 10% Fetal Bovine Serum Neuronal cultures from P0 pups Genders not determined Hippocampi dissociated digested with papain Cultured neurons plated on 12 mm poly-L-lysine coated coverslipsstarting medium 80% DMEM 10% bovine calf serum 10% Ham’s F-12 0.5× penicillin–streptomycin changed Neurobasal B-27 penicillin–streptomycin L-glutamate 0.5 Glutamax-I DIV2 Neurobasal A B-27 penicillin–streptomycin 0.5 Glutamax-I cells transfected PolyjetTM reagent cells replated 12-mm poly-L-lysine coverslips ~48 h after Neurons DIV 5 7 transfected Lipofectamine LTX (Invitrogen).cDNA non-tagged NALCN UNC79 UNC80 N-terminally GFP-tagged WT mutant mouse UNC80 peGFP-C1 EcoRI-terminally mCherry RFP-tagged UNC80 EcoRI XhoI pcDNA3.1-C1 Mutations introduced PCR fragments T4 ligase Gibson Assembly kit confirmed restriction digestion Sanger sequencing experiments room temperature Recording 48–60 h after transfectionHEK293 cells pipette solution 150 Cs 120 Mes 10 NaCl EGTA 4 CaCl2 0.3 Na2GTP 2 Mg-ATP 0.002 kinase activator 10 HEPES 7.4) Bath solutions 150 NaCl 3.5 KCl 1 MgCl2 20 glucose 2 CaCl2 10 HEPES+ bath Na+ K+ replaced+ patch clamp recording pipette solution 120 CsCl 4 EGTA 2 CaCl2 2 MgCl2 4 Mg-ATP 0.3 Tris-GTP 14 10 HEPES 140 mM Na bath 140 NaCl 5 KCl 2 CaCl2 1 MgCl2 6 glucose 2 CsCl 10 HEPES 14 mM Na bath tris-Cl 126 mM NaCl TTX (1 APV (10 bicuculline (20 CNQX block Nav synaptic currents Sodium-leak current measured currents 14 Low [Ca2+-activated leak currentclamp recordings amplified filtered 1 kHz MultiClamp 700B amplifier digitized 5 kHz Digidata 1400A digitizer controlled Clampex 10.4 Data analyzed Clampfit 10.4 Excel Origin Cell surface biotinylation western blotting anti-NALCN 1 μg anti-UNC79-UNC80 polyclonal antibodies antibodies Sigma-Aldrich-actin Invitrogen-GFP Rockland (anti-His HEK293T cells 35 mm dish lysed 30 min 360 μl IP buffer 50 Tris-HCl 150 NaCl 1% NP-40 1 EDTA protease inhibitor cocktail IP buffer 0.5% deoxycholate 0.1% SDS Lysate centrifuged × g 30 min 4 °C supernatant mixed 1 μg antibody incubated 4 °C 2 h Samples mixed buffer-equilibrated protein A-agarose 4 °C 2 h washed buffer 3 times (5 min Proteins eluted 1× lithium dodecyl sulfate sample buffer (Invitrogen #NP0007) 100 mM DTTcell surface biotinylation transfected cells washed incubated Sulfo-NHS-LC-Biotin 30 min 4 °C washed 100 mM glycine lysed 400 μl IP buffer rotated 4 °C 1 h lysates centrifuged 20,000 30 min 4 Sixty microliters supernatant saved Samples mixed 100 μl NeutrAvidin agarose overnight 4 °C washed 3 times (5 min buffer Proteins eluted 1× lithium dodecyl sulfate buffer 100 mM DTT brain protein prep pups homogenized buffer spun 10 min 5000 Supernatant 1% NP-40 spun 20 min Protein concentration measured BCA assay micrograms protein lane protein preparation brains homogenized buffer solubilized 30 min spun One mg protein 500 μl precipitated 1 μg antibody 2 h Samples mixed 60 μl buffer-equilibrated protein A-agarose 4 °C 2 h washed buffer 3 times (5 Proteins eluted 30 μl LDS buffer 100 mM DTT WB analysis membrane fractionation experimentsfrozen adult brains homogenized dunce buffer 250 mM sucrose 5 mM Tris 7.4) 1× PIC Homogenate spun 10 min 1000 g 3220 20 Supernatant spun 2 h 200,000 g 4 °C Pellet resuspended buffer Equal loaded WB affinity depletion experiments adult brains homogenized buffer 300 mM NaCl 20 mM HEPES.0) 1× Homogenate spun 1000 g 10 min supernatant mixed CHAPS 1 h 4 °C 18,000 g 20 min pre-equilibrated nickel beads 1 h spun 200 × g 3 min supernatant Ni-depleted antibody depletion one ml supernatant mixed 20 μl agarose-conjugated anti-GFP) 2 h 4 °C proteins separated NuPage 4–12% Bis-Tris gels buffer transferred polyvinylidene difluoride membranes 2–3 h Membranes pre-blocked 5% nonfat dry milk 0.1% Tween-20 incubated antibodies 4 °C overnight room temperature 2 hwashes 5 min membranes incubated HRP secondary antibody 1 h washes (5 min each incubation detection reagents Signal West Pico ECL Thermo Scientific Hi/Lo Digital-ECL WB detection kit Kindle Signals detected X-ray films camera (Fujifilm camera localizationHippocampal neurons polylysine transfected 48 h Lipofectamine LTX Cells washed three PBS fixed 4% paraformaldehyde 20 min 5× washes Samples ProlongTM Diamond Antifade Mountant Images Nikon Eclipse Ti microscope 20× lens 543 nm laser 488 nm laser studies whole-exome approved Western Institutional Review Board Unknown patient 3 9-year male developmental delay epileptic encephalopathy hypotonia ataxia muscle biopsy mitochondrial dysfunction born cesarean section nuchal cord heart rate deceleration feeding difficulties diagnosed failure thrive gastroesophageal reflux disease milestones 6–12 months smiled 6 months babbled 1 year crawled cruised 3 years MRIs normal muscle biopsy indicated mitochondrial disease4 years muscle biopsy showed Complex I III deficiency decreased Complex I OXPHOS enzyme activity microscopy skeletal muscle myofiber size early Type II fiber atrophy increased lipids Urine ketosis blood pyruvate levels high CSF testing glucose 33 (normal 60–80) 6 years first seizure MRI normal EEG bilateral central centrotemporal centroparietal spike-wave discharges activation sleep EEG 7 years epileptiform discharges left right central region wakefulness right right sleep 9 years physical assessment mild hypotonia choreoathetoid movements good strength muscles support used stroller friendly alert interactive nonverbal vocalized used communication device Dysmorphic features triangular facies mild frontal bossing simplified shape auricles camptodactyly 4th 5th digits Electromyography nerve conduction velocity testing CSF exam neurotransmitters echocardiogram lysosomal enzymes VLCFA MePCR MED12 7-dehydrocholesterol lactate ammonia CPK chromosomal microarray Fragile X DM1 DM2 negativefamily consent enrolled Center for Rare Childhood Disorders Translational Genomics Research Institute patient under 7 assent not required Review Board Protocol blood collected parents for DNA extraction QiaAmp blood kit Genomic libraries prepared Illumina’s Truseq DNA Sample Preparation Kit final sequencing library TruSeq Exome Library Prep Kit v1 Illumina WES performed Illumina HiSeq2000 sequencing platform Filtered reads aligned Human genome/GRC37) Burrows-Wheeler transform PCR duplicates removed Picard v1.9268 base quality recalibration indel realignment SNP discovery Genome Analysis Toolkit Data filtered against dbSNP137 1000 Genomes exome database annotated SnpEff 3.2a Ensembl v66 novel damaging mutations annotated variant file three family members filtered novel rare variants Exome Aggregation Consortium inheritance disease association analyst assessment genotype–phenotype correlation disease mechanism literature reviewVariants predicted damaging by Combined Annotation Dependent Depletion ExAC’s loss-of-function intolerance score z-score Genomic Evolutionary Rate Profiling Polymorphism Phenotyping v2 algorithms genes child’s phenotype patient compound heterozygous for variants UNC80 gene paternal allele inherited c.3883G>C, p.Glu1295Gln variant maternal allele double nucleotide polymorphism *c.1020_1021delGCinsTT p.Gln340_P341delinsHS variants confirmed by Sanger sequencing gnomAD database variations present in multiple heterozygous carriers (E1295Q variant 30 Q340_P341delinsHS 152 heterozygotes 1 definitive genetic severe phenotype challenging individuals with severe disease included dataset frequency equivalent lower population one homozygote could affected statistical analysisOrigin 8.0 software used electrophysiology analyses Protein level analysis with Image J National Institutes of Experimental sample sizes chosen based previous experiments not predetermined experiments groups blind experimentalist blind to genotype during data collection analysis No data excludedComparisons between two groups two sample t test No adjustments data represented as mean ± SEM.Reporting Nature Research Reporting Summary.Supplementary information Movie_Audio_Data 1 2 Reporting Summary | 48.7 | 1.138242 |
10.1038/s41467-020-16588-9 | PMC7265461 | Immunopeptidomics allows identifying the cellular repertoire of MHC-bound peptides, but quantifying them remains challenging. Here, the authors present a method to efficiently generate internal peptide MHC standards and calibration curves, facilitating relative and absolute quantitative immunopeptidomics. | Peptides bound to class I major histocompatibility complexes (MHC) play a critical role in immune cell recognition and can trigger an antitumor immune response in cancer. Surface MHC levels can be modulated by anticancer agents, altering immunity. However, understanding the peptide repertoire’s response to treatment remains challenging and is limited by quantitative mass spectrometry-based strategies lacking normalization controls. We describe an experimental platform that leverages recombinant heavy isotope-coded peptide MHCs (hipMHCs) and multiplex isotope tagging to quantify peptide repertoire alterations using low sample input. HipMHCs improve quantitative accuracy of peptide repertoire changes by normalizing for variation across analyses and enable absolute quantification using internal calibrants to determine copies per cell of MHC antigens, which can inform immunotherapy design. Applying this platform in melanoma cell lines to profile the immunopeptidome response to CDK4/6 inhibition and interferon-γ — known modulators of antigen presentation — uncovers treatment-specific alterations, connecting the intracellular response to extracellular immune presentation. | IntroductionCells present signals on the extracellular surface that serve as targets for immune cell recognition. These signals, peptides presented by class I major histocompatibility complexes (MHCs), are typically derived from intracellular source proteins, and may therefore provide an external representation of the internal cell state1. As a reflection of this, the peptide MHC (pMHC) repertoire, or “immunopeptidome”, of cancer cells may contain tumor-associated or mutation-containing antigens that serve as tumor-specific markers to activate T cells and initiate an antitumor immune response. This interaction can be strengthened with checkpoint blockade (CB) immunotherapies; however, low response rates and toxicity remain barriers to their broad clinical success2,3. A growing body of evidence suggests that combining CB with other treatments, such as small-molecule inhibitors, cytotoxic agents, and radiotherapy, could potentiate the response to CB, in part, by augmenting tumor immunogenicity through increased surface pMHC expression4–7. While clinical trials in this space have shown promise8,9, the optimal combination of agents, as well as the order and timing of administration, are only beginning to be understood. In order to improve combinatorial strategies, a quantitative, molecular understanding of how different perturbations shift the immunopeptidome is required. Furthermore, achieving absolute quantification of presented antigens is necessary to inform immunotherapy drug design, as targeted strategies have varying thresholds of antigen expression required for an optimal antitumor response.Traditional data-dependent acquisition (DDA) methods to profile pMHC repertoires using mass spectrometry (MS) are well documented10–12, but quantitative methods have critical limitations. Specifically, most common relative quantification pMHC methods lack a normalization strategy to account for variations in sample input and processing13–18. Peptide losses during processing vary across peptide sequences, concentrations, and samples, underscoring the need for normalization19,20. Absolute quantification of pMHCs to date is most commonly performed by comparing endogenous levels of pMHCs to exogenous peptide standards, again failing to account for sample losses21–24. Losses can be accounted for with internal pMHC standards, but require laborious refolding of pMHCs for every target of interest19,25. Nevertheless, this approach relies on single point calibration, ignoring the effects of ion suppression, thereby inaccurately estimating absolute pMHC levels in quantitative analyses.To combat these challenges in quantitative immunopeptidomic profiling, we present a platform that utilizes ultraviolet (UV)-mediated peptide exchange of recombinant MHC monomers to generate on demand heavy-isotope-labeled pMHCs for relative and absolute quantification of pMHC repertoires using low sample input. We demonstrate that the addition of heavy-isotope pMHCs (hipMHCs) spiked into sample lysates for normalization improves quantitative accuracy between samples for both label-free (LF) and multiplexed (tandem mass tags (TMTs) labeled) analyses and provides an estimate of ion suppression through regression against a titrated internal calibrant. Furthermore, we utilize hipMHC multipoint-embedded standard curves coupled with isobaric mass tags to accurately quantify the absolute number of copies per cell of target antigens within a single analysis. We apply this platform to profile immunopeptidomic changes in melanoma cell lines, comparing treatment with palbociclib (a small-molecule CDK4/6 inhibitor) and interferon-γ (IFN-γ), both known modulators of antigen presentation7,26. Peptides derived from proteins implicated in the biological response to palbociclib and IFN-γ are selectively enriched in the pMHC repertoire following treatment, connecting the intracellular response to extracellular immune presentation. Furthermore, peptides derived from the metabolic response to palbociclib, along with known tumor-associated antigens (TAAs), display significantly increased presentation with palbociclib treatment. We propose this platform can be broadly applied to profile immunopeptidomic changes in a high-throughput, low-input format across sample types and treatments to inform combination therapy strategies and can be used to identify and quantify treatment-modulated antigen targets for targeted immunotherapy.ResultsPlatform for relative and absolute pMHC quantitationWe set out to develop a platform to provide accurate relative and absolute quantification of pMHCs across multiple samples while controlling for losses associated with sample processing and enrichment. Accurate quantitative analysis is best performed with internal standards and multipoint internal calibration curves. To generate internal standards, heavy-isotope-labeled MHC peptides of interest were synthesized and loaded onto biotinylated MHC monomers through UV-mediated peptide exchange27 (Fig. 1a). To control for loading efficiency of synthetic peptides into recombinant MHC proteins, the concentration of stable hipMHC complexes was determined by an enzyme-linked immunosorbent assay (ELISA). Stable hipMHC complexes were then used in two ways: selected hipMHC complexes were spiked at the same concentration into the whole-cell lysate from each sample to provide a normalization correction for relative quantification across samples, while other hipMHC complexes were titrated at different concentrations into each sample to verify correction parameters, estimate dynamic range suppression for quantification, and/or create an internal standard curve for absolute quantification of a specific peptide. After adding hipMHCs, endogenous and exogenous pMHCs were isolated by immunoprecipitation (IP), acid elution, and molecular weight size-exclusion filtration.Fig. 1Platform for quantitative immunopeptidomics using hipMHCs.a HipMHCs were generated through UV-mediated peptide exchange of HLA*A2:01 monomers with a heavy leucine HLA-A*02:01 binding peptide. Stable hipMHCs concentrations were measured with an ELISA, and hipMHC complexes were added to lysate samples prior to immunoprecipitation (IP), at the same concentration for quantification correction (blue/teal) or titrated in to create an internal standard curve (red). Heavy and light pMHCs were isolated with IP, acid elution, and molecular weight cut-off (MWCO) filters. b Peptides were analyzed by LC-MS/MS three ways. Relative quantification label-free analyses were quantified by integrating the area under the curve (AUC) of the chromatographic elution across samples, and quantification was normalized by applying correction factors determined by hipMHC AUC intensity ratios between samples. Samples for multiplexed analysis were TMT-labeled and relative quantification was implemented using reporter ion intensities. Normalization was performed using hipMHC reporter ion intensity ratios across TMT channels. For absolute quantification, TMT-labeled samples containing a hipMHC internal standard curve were used to calculate the endogenous copies per cell of the pMHC of interest.Peptide mixtures were next analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) in three different ways (Fig. 1b). For LF analyses, samples were analyzed individually, and peptides were quantified by integrating the area under the curve (AUC) for the chromatographic elution of precursor masses for each peptide-spectrum match (PSM). Relative AUC intensities of quantification correction hipMHCs were used to normalize AUC intensities of endogenous peptides across analyses. To analyze multiple samples simultaneously, we labeled samples with TMT and relative TMT ion intensity ratios of hipMHCs were used for normalization to correct the relative quantification in multiplexed samples. TMT-labeled titrated hipMHCs were also used for absolute quantification of endogenous peptides. Apex TMT intensities of hipMHCs generated a peptide-specific multipoint calibration curve to calculate the average number of copies per cell. As a control, heavy-isotope-coded synthetic peptides not complexed to MHCs were spiked into the whole-cell lysate prior to IP. These peptides were not detected in the subsequent LC-MS/MS analysis, demonstrating that only peptides in stable complexes were isolated in our workflow and that excess free peptides did not displace endogenously presented peptides.HipMHC standards improve quantitative accuracyTo demonstrate the improved quantitative accuracy obtained with hipMHCs, we used five LF and six TMT-labeled technical replicates of 1 × 107 MDA-MB-231 breast cancer cells to measure variance between replicates before and after hipMHC correction (Fig. 2a, Supplementary Data 1). In both LF and TMT-labeled workflows, we spiked 30 fmol of two quantification correction hipMHCs into each sample, as adding additional correction hipMHCs gave minimal improvement in quantitative accuracy. We also added 30–300 fmol of a titrated hipMHC across samples (Fig. 2b). A total of 2369 unique pMHCs were identified in total across five LF analyses, 1352 of which were quantifiable via AUC integration (Fig. 2c). Of these quantifiable peptides, only 589 were quantified in all five analyses, highlighting the poor run-to-run overlap of LF analyses, even with replicate samples (Supplementary Fig. 1a). By comparison, 1754 unique peptides were quantifiable with TMT-labeled analyses. The extra sample handling steps associated with TMT labeling can result in losses, so to achieve high coverage of the immunopeptidome, labeled samples were divided into six separate analyses, thereby increasing the number of unique identifications (Supplementary Fig. 1b). In both LF and TMT analyses, peptides matched expected length distributions of 8–11 amino acids (Supplementary Fig. 1c), and 82% of LF and 92% of TMT 9-mers were predicted to be binders with <500 nM predicted affinity28 (Fig. 2d). The peptides were similarly apportioned across alleles between LF and TMT analyses, and the allele motifs aligned with those previously reported29 (Supplementary Fig. 1d, e). Further reducing the input material to 5 × 106 cells still resulted in 86% of the number of unique peptides identified with 1 × 107 cells in a single LF analysis, establishing the sensitivity of this method for low-input pMHC analyses (Supplementary Fig. 1f).Fig. 2HipMHCs improve quantitation in LF and TMT-labeled samples.a Experimental design. Five LF (orange) and six TMT (blue) technical replicates of 1 × 107 cells + hipMHCs were used to compare LF and TMT quantification. b Peptide sequence and amount of hipMHC added into each sample. L7 denotes heavy-isotope-labeled leucine (+7). ALNEQIARL7 and SLEEPIGHL7 were used as quantification correction hipMHCs, and SVVESVKFL7 was titrated in across samples. For LF analysis, sample #6 was omitted. c Two thousand three hundred and sixty-nine unique LF peptides were identified across five analyses (black), 1352 of which were quantifiable (gray) via AUC quantification, and 589 quantifiable peptides which were identified in all five analyses (orange). One thousand seven hundred and fifty-four unique peptides were quantified with TMT-labeled analyses combining six TMT fractions (blue). d Ninety-two percent of TMT, 82% of LF, and 5.6% of random peptide 9-mers derived from the proteome (gray) are predicted to bind to an HLA allele in MDA-MB-231 cells with an affinity < 500 nM. e Linear fit of titrated hipMHC peptide for LF (left) and TMT (right). Raw r2 = 0.48 (LF) and r2 = 0.91 (TMT), hipMHC-adjusted (adj) r2 = 0.99 (LF) and r2 = 0.96 (TMT). f Distribution of the log2 fold change (FC) of each peptide’s quantification (x) over the mean (μ) peptide quantification across samples for raw (left) and hipMHC adjusted (right). g Gaussian fit of the frequency distribution of log2(FC) of (x) over (μ) for raw and hipMHC-adjusted LF and TMT samples. In all, 99.7% of variance between peptide quantitation (3× SD) is captured within a 1.65 (raw) and 1.52 (adj) FC from the mean for LF samples, and 1.30 (raw) and 1.23 (adj.) for TMT samples. h Median coefficient of variation (CV) for LF (24.27% raw, 20.99% adj) and TMT (14.00% raw, 7.48% adj). Error bars represent the interquartile range. Source data are provided in the Source data file.To normalize LF and TMT-labeled data sets, we applied correction parameters calculated from the quantification correction hipMHCs. The titrated peptide, SVVESVKFL7, displayed an improved linear fit after correction, with an even more pronounced effect in the LF samples (Fig. 2e). We observed dynamic range suppression for this peptide in TMT-labeled (4.7×) and LF (2.7×) data sets, demonstrating in both cases that quantitative differences are likely larger than what is measured.In both analyses, hipMHC quantification correction reduced variation, for example, peptides from TMT-labeled sample 5 have lower intensities than the other samples, which was corrected by hipMHC normalization (Fig. 2f). The standard deviation from the mean for replicate samples decreased with hipMHC correction in LF and TMT-labeled samples (Fig. 2g), although TMT labeling showed lower variation between replicates (Fig. 2h), allowing for higher confidence in small shifts within the immunopeptidome. We investigated whether peptides with lower abundance had higher quantitative variation across samples, but found no correlation in LF or TMT-labeled analyses (Supplementary Fig. 1g).Absolute quantification of endogenous pMHCsTo demonstrate the ability of hipMHCs to quantify pMHC copies per cell, we selected two peptides identified in TMT-labeled MDA-MB-231 cells for absolute quantification: KLDVGNAEV derived from B cell receptor-associated protein 31 (BCAP31) and KQVSDLISV from DEAD-box RNA helicase 5 (DDX5). BCAP31 regulates the transport of membrane proteins from the endoplasmic reticulum to the Golgi, a central component of antigen processing, and is a known TAA peptide30. DDX5 is important in gene expression regulation and has been implicated in proliferation, metastasis, and tumorigenesis in cancer31. These peptides were detected at differing levels with the highly abundant BCAP31 peptide falling in the 98th percentile and DDX5 falling in the 33rd percentile of abundance (Fig. 3a). Both peptides were synthesized with heavy-isotope-labeled leucine (+7), and hipMHC normalization standards were added to three replicates of 1 × 107 MDA-MB-231 cells along with titrated amounts of BCAP31 and DDX5 hipMHC (Fig. 3b). We then labeled samples with TMT and performed LC-MS/MS analysis using an inclusion list, so only targeted peptides of interest were selected for fragmentation (Supplementary Data 2).Fig. 3Absolute quantification of pMHCs with hipMHC standards and isobaric labeling.a Peptide intensities for TMT-labeled MDA-MB-231 cells from Fig. 2 were determined by AUC quantification. Percentile of abundance represents a peptide’s rank relative to the most abundant peptide. b Experimental design. Normalization standards along with 5, 15, and 50 fmol of BCAP31 and 0.3, 1, and 3 fmol of DDX5 hipMHCs were added to three biological replicates of 1 × 107 MDA-MB-231 cells, and peptides from MHC complexes were isolated, labeled, and analyzed via LC-MS/MS. c Chromatographic elution profiles for the three TMT reporter ion intensities of the hipMHC standard curve (colored), along with the average (n = 3) TMT reporter ion intensity trace of the endogenous peptide (black). Each MS2 scan is represented as a single point, and elution profiles are fitted with a gaussian distribution (line). d Adjusted (hipMHC normalized) apex intensity versus fmol of hipMHC added creates a standard curve from which the endogenous concentration of antigen is calculated. For both linear fits of BCAP31 and DDX5, r2 > 0.999. The endogenous peptide is presented as the mean value ± SD for n = 3 biological replicates. Source data are provided in the Source data file.Chromatographic traces of the three TMT reporter ions for heavy BCAP31 and DDX5 peptides displayed increasing ion intensities with increasing amount of hipMHC added (Fig. 3c). In order to quantify peptide expression, the apex intensities of reporter ions were adjusted based on the normalization hipMHCs, and a linear fit was used to determine pMHC concentration present in the sample. Cells had an average of 1740 copies per cell of the BCAP31 peptide, and 81 copies per cell of the DDX5 peptide (Fig. 3d). Concentrations of the DDX5 hipMHC as low as 100 attomole were detected (six copies per cell) showcasing the broad range of pMHC expression levels quantifiable by our method (Supplementary Fig. 2). Furthermore, BCAP31 and DDX5 had a dynamic range suppression of 1.9× and 2.5×, respectively, illustrating that the ion suppression is not uniform across peptides and that peptide-specific internal standards may be required for absolute quantification of each pMHC of interest.CDK4/6 inhibition alters the pMHC repertoire in melanomaCyclin-dependent kinases 4 and 6 (CDK4/6) control cell cycle progression by phosphorylating Rb1, thereby releasing the E2F family of transcription factors that drive progression through the G1 checkpoint32. CDK4/6 is often dysregulated and overactive in cancer, leading to uncontrolled proliferation33. As such, CDK4/6 inhibitors have emerged as a potentially powerful class of anticancer agents, active against a spectrum of tumor types including melanoma34. In recent years, CDK4/6 inhibitors have also been shown to enhance tumor immunogenicity by increasing surface MHC class I expression and boosting T cell activation and infiltration7,35. These data highlight CDK4/6 inhibitors as an attractive candidate to combine with CB or other immunotherapies to augment immunotherapy response rates in melanoma. However, to date, the effect of CDK4/6 inhibition on the MHC class I peptide repertoire has not been characterized. We therefore applied our platform to quantify how pMHC repertoires in melanoma change in vitro upon treatment with the CDK4/6 inhibitor, palbociclib, to better understand how CDK4/6 inhibitors could be leveraged in combination therapy regimes to improve patient outcomes.We selected four melanoma cell lines for analysis: SKMEL5 and SKMEL28 (BRAF mutant), and SKMEL2 and IPC298 (NRAS mutant). Based on sensitivity analyses for each cell line (Fig. 4a), we selected two doses of palbociclib for further study: a low dose of 1 μM, below the half-maximal inhibitory concentrations (IC50) of all four cell lines, and a high dose of 10 μM, near the IC50. Three biological replicates of 1 × 107 cells of each cell line were then treated with dimethyl sulfoxide (DMSO), and low- or high-dose palbociclib for 72 h (Fig. 4b). Low-dose treatment increased surface class I MHC presentation, as measured by flow cytometry, by 1.5–2× across cell lines, whereas high-dose treatment had a milder effect (Fig. 4c, Supplementary Fig. 3a).Fig. 4Palbociclib alters the immunopeptidome in melanoma.a Viability at 72 h after drug treatment; data are represented as a fraction (%) of the DMSO control. Calculated IC50s: SKMEL5 = 12.74 μM, SKMEL28 = 14.62 μM, SKMEL2 = 16.98 μM, and IPC298 = 10.62 μM. Data are presented as mean values ± SD for n = 3 experimental replicates for all cell lines, except SKMEL5 (n = 4). b Experimental setup of TMT-labeled immunopeptidomics experiments in melanoma cell lines. c Flow cytometry measurements of surface HLA expression in SKMEL5 cells. Data are represented as % of maximum signal, and the distributions are representative of three independent experiments. d Histogram distribution of log2 fold change (FC) of (palbociclib/DMSO) for unique pMHCs, where FC is calculated from the mean intensity of n = 3 biological replicates per condition. Data are represented as a % of total unique peptides identified. e Volcano plot displaying log2(FC) of 1 μM treated SKMEL5 cells versus significance (mean-adjusted p value, unpaired two-sided t test). Colored points (p < 0.05, log2(FC) > 1.56) correspond to processes in f. f Log2(FC) of significantly enriched peptides from GO term enrichment processes labeled with source protein name. False discovery rate (FDR)-adj. p value < 0.05. g Four-way Venn diagram of the number of source proteins of peptides significantly enriched (top value) or significantly increasing (bottom value) with 1 μM palbociclib. h Log2(FC) of pIRS2 peptide following 1 μM (gray) or 10 μM (black) palbociclib, *p < 0.05, unpaired two-sided t test. i Source protein name, peptide sequence, and log2(FC) of TAAs in SKMEL5 (left) and IPC298 (right) cells. Exact significance values and other source data are reported in the Source Data File.To characterize the pMHC repertoire alterations induced by palbociclib, multiplexed relative quantitation was performed comparing low- and high-dose palbociclib to DMSO for each cell line, and data were normalized using hipMHC standards (Supplementary Fig. 3b, Supplementary Data 3). As with our previous analysis, identified peptides matched expected length distributions, and a majority were predicted to be MHC class I binders (Supplementary Fig. 3c, d). Immunopeptidomic analysis for each cell line and treatment showed a similar trend to the flow cytometry data: low-dose palbociclib shifted mean pMHC expression higher than DMSO treatment in all cell lines, and a high-dose palbociclib showed a small increase in mean expression for SKMEL5 compared to DMSO and no significant change for the other cell lines (Fig. 4d, Supplementary Fig. 3e). We measured a wider distribution of changes in peptide presentation following low-dose treatment, with several peptides increasing eight- to ten-fold, even before considering the effect of dynamic range suppression (Supplementary Fig. 3b).To gain insight into the biology underlying palbociclib-modulated pMHC alterations, we analyzed our data in two ways. First, we determined which peptides and source proteins were significantly increased with palbociclib treatment over DMSO. Because many peptides were significantly increased with low-dose treatment, we also identified the peptides and source proteins that were significantly enriched in presentation with treatment relative to the mean fold change of all peptides, highlighting peptides preferentially modulated by palbociclib.Using these data, we performed GO term enrichment on the 127 peptides significantly enriched in low-dose-treated SKMEL5 cells (Fig. 4e), and identified enriched biological processes of interest, including ribosomal biogenesis, glucose metabolic process, and antigen processing, a reflection of the expected biological response to palbociclib7,36,37 (Fig. 4f, Supplementary Fig. 3f). We performed the same analysis with the raw, non-normalized values, and found only 66 of these peptides were significantly enriched without hipMHC quantification correction, altering the Gene Oncology (GO) term pathway analysis results (Supplementary Fig. 3g). While peptides mapping to ribosomal biogenesis were still significantly enriched, the other two biological processes were not, underscoring the importance of using hipMHCs for quantification correction to accurately interpret alterations in pMHC repertoires.To determine if the measured pMHC alterations to SKMEL5 cells were common across cell lines, we compared the source proteins of peptides that were significantly enriched with low-dose treatment compared to DMSO across all four cell lines. Surprisingly, a majority (72–88%) of enriched source proteins were unique to each cell line, and we discovered only three proteins in common: vimentin, putative β-actin-like protein 3, and SIL1 nucleotide exchange factor (Fig. 4g, Supplementary Fig. 3h). Even when comparing source proteins of all peptides significantly increasing to any extent, just 17 proteins in common were identified, further illustrating the uniqueness of the proteins altered by palbociclib in each immunopeptidomic landscape (Supplementary Fig. 3i). We investigated whether the commonality of these 17 proteins could be explained by having high abundance in the peptide mixtures, but in SKMEL5 cells they were scattered throughout the distribution of AUC intensities (Supplementary Fig. 3j).While the list of shared pMHCs and source proteins in common is limited, of interest is the serine-phosphorylated IRS2 (pIRS2) peptide, RVA[pS]PTSGVK. This post-translationally modified sequence has previously been shown to be restricted to malignant cells, with only the phosphorylated form demonstrating immunogenic potential38,39. Even though there are no alleles in common across the four cell lines40 (Supplementary Fig. 3k), we observed the pIRS2 peptide increasing across all cell lines with low-dose treatment (Fig. 4h). Furthermore, RVA[pS]PTSGVK has high expression among pMHCs (Supplementary Fig. 3j), and can be isolated without phospho-enrichment41. As a result, this peptide may be uniquely positioned as a broadly targetable antigen whose expression can be modulated by CDK4/6 inhibition. As a general effect of palbociclib treatment, TAAs derived from proteins like MLANA (MART1), PMEL (gp100), and TYR, among others, also increased in presentation following treatment (Fig. 4i). While these antigens and their source proteins are not universally conserved across our cell lines, the effect of increased TAA presentation following 1 μM palbociclib treatment could be applied to increase antigen presentation prior to immunotherapies targeting these well-documented antigens.Response to palbociclib is reflected in the immunopeptidomeTo further assess whether quantitative differences in the immunopeptidome after palbociclib treatment are reflective of the cell signaling response to a perturbation, we performed a nonparametric test to identify positively and negatively enriched pathways. Gene names for source proteins were rank ordered according to fold change with treatment and searched against the MSigDB Hallmarks gene set database using Gene Set Enrichment Analysis (GSEA)42–44. This analysis did not reveal any significantly enriched pathways for the low-dose treatment, but high-dose palbociclib showed significant enrichment among downregulated pMHCs of E2F targets, G2M checkpoint, DNA repair, mitotic spindle, and MTORC1 signaling pathways in one or more cell lines (Fig. 5a). These findings reflect the known biological effects of CDK4/6 inhibition. For instance, inhibiting CDK4/6 decreases expression of E2F targets, and peptides derived from E2F targets like Ki-67, a proliferation marker, were depleted in all four cell lines (Fig. 5b). E2F also controls genes involved in DNA damage repair, and consistently, γH2AX levels, a marker of DNA double-strand breaks, increased at 72 h with palbociclib treatment in a dose-dependent manner45 (Supplementary Fig. 4a). Although similar biological processes are enriched across the four cell lines, source proteins for significantly depleted E2F peptides showed little overlap between the cell lines (Fig. 5b), again emphasizing the individuality of the source proteins contributing to each cell line’s detected pMHC repertoire.Fig. 5Pathway analysis of palbociclib-altered immunopeptidome.a Normalized enrichment score (NES) of significantly enriched pathways with 10 μM palbociclib, where +/− NES scores reflect enrichment directionality. For all, q < 0.25, and *p < 0.05, **p < 0.01, and ***p < 0.001. b, c String network of protein–protein interactions of all source proteins from E2F peptides (b) significantly decreasing with 10 μM palbociclib, and OxPhos peptides (c) significantly increasing with 1 μM palbociclib for all cell lines, except SKMEL28, where peptides from 10 μM are depicted. Node color corresponds to cell line. d Quantification (n = 9) of MitoTraker green intensity normalized to cell number following 72 h palbociclib treatment. Data are represented as a box and whiskers plot, with whiskers displaying minimum and maximum signal. Significance was determined using Dunnett’s multiple comparisons test for each condition versus DMSO. *p < 0.05 and ****p < 0.0001. e Correlation between log2 fold change (FC) of (palbociclib/DMSO) for RNA expression (y-axis) and pMHC presentation (x-axis) of SKMEL5 cells treated for 72 h with 1 μM palbociclib, r2 = 0.04. FC is calculated from the mean intensity of n = 3 biological replicates per condition. f Significantly enriched pathways using RNA-seq data (p < 0.05, q < 0.25). Annotated pathways reflect pathways also identified in the immunopeptidome analysis (blue), and those that match with previous reported data (red)7. g Log2(FC) for SKMEL5 OxPhos peptides significantly increasing (p < 0.05, blue) with 1 μM palbociclib, and matched log2(FC) of RNA expression (black). Significant differences in RNA expression (palbociclib versus DMSO) are indicated. **p < 0.01 and ****p < 0.0001 (Wald test, Benjamini–Hochberg (BH) adjusted). Exact significance values and other Source data are reported in the Source Data File.Only one pathway, oxidative phosphorylation (OxPhos), was significantly upregulated in SKMEL28 cells. However, all cell lines presented peptides derived from the OxPhos pathway that increased significantly with palbociclib treatment, although this effect was more prominent with low-dose treatment in SKMEL5, IPC298, and SKMEL2 cells, in contrast to the results of SKMEL28 cells (Fig. 5c). OxPhos has been shown to increase with CDK4/6 inhibition due to increased ATP levels and mitochondrial mass, elevating metabolic activity. Comparably, all samples showed elevated mitochondrial levels following treatment, suggesting that enriched pMHC presentation of OxPhos-derived peptides reflects a change in the metabolic cell state (Fig. 5d).Because alterations to the pMHC repertoire align with previously characterized biological responses to CDK4/6 inhibition, we tested whether changes in RNA expression could predict the quantitative immunopeptidome changes (Supplementary Data 4). No bulk correlation (r2 = 0.04) was observed between pMHC expression and RNA expression (Fig. 5e). This was unsurprising, as many mechanisms beyond gene expression regulate pMHC presentation, including protein synthesis, degradation, post-transitional modifications, processing, and more. Despite this poor correlation, significantly enriched gene sets in the immunopeptidome were also present in our RNA-sequencing (RNA-seq) analysis (Fig. 5f). While E2F pMHCs significantly depleted in SKMEL5 cells correlated with significantly decreased gene expression of the same source proteins (Supplementary Fig. 3b), only five of the 15 positively enriched OxPhos peptides displayed significantly higher gene expression with palbociclib treatment, with three decreasing in expression, and seven remaining unchanged (Fig. 5g). Collectively, these data suggest that while changes in gene expression and pMHC repertoires map to the same biological pathways, individual gene expression changes are not necessarily predictive of alterations in the immunopeptidome.IFN-γ-induced pMHC alterations are distinct from palbociclibPrevious work has demonstrated that CDK4/6 inhibition stimulates IFN signaling, augmenting antigen presentation levels15. We also observed upregulation of IFN-γ response genes with low-dose palbociclib treatment, as well as increased expression of genes relating to antigen presentation (Figs. 5f, 6a). Consequently, we tested whether direct IFN-γ stimulation would shift the repertoire similarly to CDK4/6 inhibition. Cells were stimulated with DMSO or 10 ng mL−1 IFN-γ for 72 h and the resulting pMHC repertoires were quantified using our multiplexed hipMHC platform (Supplementary Data 5). IFN-γ increased surface pMHC levels >2× for each cell line (Fig. 6b), a trend that was reflected in the immunopeptidome, as nearly every identified pMHC increased in presentation with stimulation (Fig. 6c, Supplementary Fig. 5a).Fig. 6Quantifying the pMHC repertoire response to IFN-γ stimulation.a RNA-seq (black) and pMHC (blue) log2 fold change (FC) of 1 μM palbociclib/DMSO, calculated from the mean intensity of n = 3 biological replicates per condition, for antigen-processing genes. b Surface HLA expression via flow cytometry of cells treated with 72 h IFN-γ shown as log2(FC) (IFN-γ/DMSO). Errors bars represent ± SD, biological replicates are n = 8, 11, 9, and 9 for SKMEL5, SKMEL28, SKMEL2, and IPC298, respectively. c Immunopeptidome log2(FC), dotted lines display quartiles, and mean fold changes (solid line) are 2.42, 2.50, 2.08, and 3.04 for SKMEL5, SKMEL28, SKMEL2, and IPC298 cells, respectively. d Significantly enriched pathways in SKMEL5 cells with 72 h IFN-γ, q < 0.25, *p < 0.05, **p < 0.01. e Enrichment plot of IFN-γ response enrichment in SKMEL5 cells displays running enrichment score (green, right y-axis), and the log2(FC) (left y-axis) versus rank (x-axis) for each peptide (gray). Open circles show significantly enriched IFN-γ peptides. f Volcano plot of IFN-γ-induced changes in SKMEL5 cells. Peptides are presented as the log2(FC) versus mean-adjusted p value (unpaired two-sided t test). Red points represent peptides significantly enriched (p < 0.05, fold change > 2.42). g Venn diagram of significantly enriched source proteins (black) or peptides (gray) between IFN-γ and 1 μM palbociclib-treated SKMEL5 cells. h Protein–protein interaction network of significantly enriched source proteins in common, annotated by enriched gene ontology cellular components (CCs) and/or biological professes (BPs). Significance values are false discovery rate (FDR) adjusted. Exact significance values and other Source data are reported in the Source Data File.To determine the similarity of response to palbociclib treatment, we again performed GSEA against the hallmark gene sets. The most significantly upregulated pathway in SKMEL5 cells with IFN-γ stimulation was the “IFN-γ response,” including peptides derived from proteins involved in antigen processing like STAT1 and HLA-A, in line with previous findings46 (Fig. 6d, e). In fact, IFN-γ response was the top enriched pathway in every cell line, reiterating that the cellular response to stimulus is reflected in quantitative differences in pMHC presentation, and that IFN-γ-related peptides are preferentially upregulated by IFN-γ stimulation (Supplementary Fig. 5b). Other pathways such as G2M checkpoint and mitotic spindle were positively enriched in IFN-γ stimulated cells, in contrast to the results of palbociclib treatment.Although the cell lines showed differential pMHC pathway enrichment upon CDK4/6 inhibition with palbociclib and IFN-γ stimulation, we tested whether any pMHCs or source proteins were commonly enriched in response to these perturbations. In SKMEL5 cells, we identified just 20 peptides and 31 source proteins significantly enriched in both conditions (Fig. 6f, g), which primarily map to the cytoplasm and contain multiple ribosomal and translation initiation proteins frequently overrepresented in immunopeptidomic data sets (i.e., DRiPs)47 (Fig. 6h). These data demonstrate that while CDK4/6 inhibition may induce an IFN-γ response, stimulating cells with IFN-γ does not recapitulate the distinct peptide repertoire alterations observed with palbociclib treatment. Instead, IFN-γ stimulation alters the repertoire by augmenting the presentation of IFN-γ-related peptides.DiscussionThe addition of hipMHCs as internal standards improves relative quantitative accuracy for both LF and multiplexed, labeled analyses, although multiplexed labeling with TMT showed superior accuracy and peptide binding specificity and yielded a higher number of quantifiable unique peptides using equivalent sample input. These internal hipMHC standards, which travel through the entire pMHC workflow, also account for variation across samples and provide an estimate for dynamic range suppression, which varies across peptides. We demonstrate that hipMHC correction alters the biological interpretation of quantitative pMHC repertoire changes, even in a relatively simple, in vitro system. Utilizing hipMHCs will be increasingly beneficial in accounting for variation in sample losses across heterogeneous in vivo samples, and in large studies to compare and correct quantitation across many multiplexed analyses or clinical sites. While we use TMT 6-plex in our analyses, this method is compatible with other isobaric labeling strategies, including iTRAQ (isobaric tags for relative and absolute quantitation), TMT 11-plex, and TMTpro, to analyze up to 16 samples simultaneously. For rapid profiling of immunopeptidome changes, we elected to use minimal sample input, making this protocol easily translatable for in vivo-derived tissue (e.g., clinical and animal) samples. While further reductions in sample input mirroring the amount obtained with a 14-gauge needle biopsy48 resulted in a notable decrease in the number of unique peptides identified (Supplementary Fig. 1f), we believe advancements in the speed and sensitivity of mass spectrometers, as well as in sample preparation techniques to reduce sample losses will enable pMHC profiling at even lower sample inputs in the future. Alternatively, using this same general platform of hipMHCs and isobaric multiplex labeling, the sample amount could be increased and coupled with fractionation for deeper sequencing of the pMHC repertoire, including neoantigen identification49.In addition to improved relative quantification, we also demonstrated the utility of hipMHCs for pMHC absolute quantification by generating an embedded multipoint standard curve. Using targeted MS to detect attomole levels of antigen from just 1 × 107 cells and regressing this signal against the titrated hipMHC standard, we were able to extract accurate absolute quantification in terms of copies per cell for two pMHC’s with ~20-fold difference in abundance. While absolute quantification is limited to just two peptides in this study, applying advanced targeted MS methods could enable the quantitation of hundreds of peptides in a single analysis50. The ability to readily determine the absolute quantification of detectable antigens of interest without the need for a pMHC-specific antibody will aid in targeted immunotherapy design. For instance, peptides of lower abundances may be better suited for engineered TCR-based therapies, as TCRs have been shown to be incredibly sensitive with as few as one pMHC complex being capable of initiating detectable T cell activation51. Alternatively, antibody-based therapies targeting specific pMHCs, for example, bi-specific T cell engagers or antibody–drug conjugates, may benefit from higher antigen expression levels, although results vary across antigen targets and antibody affinities52. Moreover, absolute quantification of pMHC expression can help to untangle the biological relationships among antigen processing, epitope abundances, immunogenicity, and off-target toxicity (e.g., tumor versus non-tumor abundance).It is worth noting that one existing restriction to using hipMHCs is the commercial availability of UV-mediated MHC monomers and ELISA control reagents, which are limited to a handful of common human class I alleles. While matched allele hipMHCs are not required for normalization correction if MHC molecules are isolated using a pan-specific antibody, they are necessary for accurate absolute quantification with embedded standard curves. An analogous technology, disulfide-stabilized HLA molecules, could be used in place of UV-mediated exchange53. These HLA–B2M complexes show increased stability and higher exchange efficiency of lower-affinity peptides, potentially eliminating the need for an ELISA to quantify exchange efficiency and simplifying MHC refolding to expand this protocol to other alleles and species.We applied our quantitative multiplexed hipMHC normalization to determine the pMHC repertoire response to CDK4/6 inhibition with palbociclib treatment in melanoma. These results indicate that extracellular changes in pMHC abundance are reflective of the intracellular response to CDK4/6 inhibition. Moreover, palbociclib treatment increased the presentation of TAAs and peptides derived from metabolic processes. Recently, high tumor antigen and metabolic protein expression levels have been shown to be predictive of checkpoint inhibitor response in melanoma, suggesting that palbociclib could be used in conjunction with CB- or TIL-based therapies to increase tumor immunogenicity54. As an alternate therapeutic strategy, peptide antigens whose surface expression was selectively increased by palbociclib could be utilized for targeted immunotherapy, either alone or in combination.Indeed, the landscape of clinical trials exploring combination treatment regimens coupling checkpoint blockade with other therapies is rapidly expanding55–57. Quantifying the molecular consequences of these combination regimes with our platform could provide insight into these trials and enable the informed design of new therapeutic combinations, potentially with targeted immunotherapies. Taken together, our relative and absolute quantitative immunopeptidomic data demonstrate the utility of quantitative immunopeptidomics in evaluating the pMHC repertoire response to therapy. The multiplexed nature of this platform allows for analyses of many samples in a short timescale, an important feature in the context of clinical trials. Further analyses of pMHC repertoire changes will be useful in understanding the order and timing of therapies to achieve optimal success and may enable predictions as to how to tune the immunopeptidome to be most applicable to immunotherapy targeting.MethodsHuman cell linesSKMEL5, SKMEL28, and MDA-MB-231 cell lines were obtained from ATCC (ATCC HTB-70, ATCC HTB-72, and HTB-26, respectively) and maintained in DMEM (Dulbecco’s modified Eagle’s medium) (Corning). IPC298 and SKMEL2 cells were provided by Array Biopharma and maintained in RPMI-1640 (Gibco) and minimum essential medium-α (Gibco), respectively. All media were supplemented with 10% fetal bovine serum (FBS) (Gibco) and 1% penicillin/streptomycin (Gibco). Cells were routinely tested for mycoplasma contamination, and maintained in 37 °C, 5% CO2.Phenotypic assaysIC50 of palbociclib (Selleckchem, PD-0332991) were determined for each cell line using CellTiter-Glo luminescent cell viability assay (Promega). Cells were seeded at density of 10,000 (SKMEL2, SKMEL28, IPC298) or 5,000 (SKMEL5) cells per well in a 96-well plate and allowed to adhere overnight. Cells were then treated with palbociclib or DMSO as a vehicle control in fresh medium for 72 h and assayed. Data were acquired using a Tecan plate reader Infinite 200 with Tecan icontrol version 1.7.1.12. IC50 values were calculated using a four-parameter logistic curve in Prism 8.4.1.Mitochondrial content was measured using a fluorescent mitochondrial stain. Cells were seeded at a density of 20,000 cells per well in a 24-well plate and allowed to adhere overnight. Cells were then treated with 1 μM or 10 μM palbociclib or DMSO vehicle control in fresh medium for 72 h. Cells were assayed by incubating 200 nM of MitoTracker Green FM (Thermo Fisher) and a 1:1000 dilution of NuclearID Red DNA stain (Enzo Life Biosciences) for 15 min in a serum-free medium at 37 °C. After staining and medium exchange, cells were imaged and analyzed using the Incucyte Live Cell Analysis System (IncuCyte Zoom version 6.2.9200.0, Essen BioScience). The integrated intensity of MitoTracker dye was calculated for each image (n = 3 experimental replicates, n = 3 images per sample) and divided by the number of cells (counted using nuclear counterstain) to determine the mitochondrial intensity per cell. A one-way analysis of variance followed by Dunnett’s multiple comparisons statistical test was performed in Prism to compare the significance of treated cells versus vehicle DMSO control. Significance values represent multiplicity-adjusted p values.Flow cytometryFor analysis of cells by flow cytometry, cells were lifted with 0.05% Trypsin-EDTA and 106 cells/mL were spun at 300 × g for 3 min, washed with ice-cold phosphate-buffered saline (PBS) supplemented with 1% FBS and 0.1% sodium azide (flow buffer), and incubated with fluorophore-conjugated antibody at 0.5 μg mL−1 in flow buffer for 30 min on ice. After incubation, cells were washed again, and resuspended in flow buffer plus 5 μL of propidium iodide staining solution (10 μg mL−1, Invitrogen) per sample. Analyses were performed on an LSRII (BD Biosciences) and data were analyzed using FlowJo (version 10.6.2). All antibodies were purchased from BioLegend: Alexa Fluor 488 HLA-A, -B, -C, clone W6/32 (cat. # 311413), Alexa Fluor 488 anti-H2A.X Phospho (Ser139), clone 2F3 (cat. # 613406). The gating strategy used for all experiments is provided in Supplementary Fig. 6.UV-mediated peptide exchange for hipMHCsUV-mediated peptide exchange was performed using recombinant, biotinylated Flex-T HLA-A*02:01 monomers (BioLegend), using a modified version of the commercial protocol. Briefly, 4 μL of 500 μM peptide stock, 2 μL of Flex-T monomer, and 32 μL of 1× PBS were combined in a 96-well U-bottom plate. On ice, plates were illuminated with UV light (365 nm) for 30 min, followed by a 30-min incubation at 37 °C protected from light. Concentration of stable complexes following peptide exchange was quantified using the Flex-T HLA class I ELISA assay (BioLegend) as per the manufacturer’s instructions for HLA-A*02:01. ELISA results were acquired using a Tecan plate reader Infinite 200 with Tecan icontrol version 1.7.1.12.pMHC isolationCultured cells were seeded in 10 cm plates, allowed to adhere overnight, and treated for 72 h with palbociclib, 10 ng mL−1 human recombinant IFN-γ (ProSpec Bio), or DMSO vehicle control. At the time of harvest, cells were washed with 1× PBS, and lifted using 0.05% Trypsin-EDTA (Gibco). Cells were pelleted at 500 × g for 5 min, washed twice more in 1× PBS, and pelleted again. Cells were resuspended in 1 mL lysis buffer [20 nM Tris-HCl pH 8.0, 150 mM NaCl, 0.2 mM PMSO, 1% CHAPS, and 1× HALT Protease/Phosphatase Inhibitor Cocktail (Thermo Fisher)], followed by brief sonication (3 × 10 s microtip sonicator pulses) to disrupt cell membranes. The lysate was cleared by centrifugation at 5000 × g for 5 min and quantified using Bicinchoninic Acid Protein Assay Kit (Pierce).pMHCs were isolated from 1 × 107 cells per condition with IP and size-exclusion filtration, as previously described58 Briefly, for each condition 0.5 mg of pan-specific anti-human MHC class I (HLA-A, HLA-B, HLA-C) antibody [clone W6/32, Bio X Cell (cat. # BE0079)] was bound to 20 μL FastFlow Protein A Sepharose bead slurry (GE Healthcare) for 3 h rotating at 4 °C. Beads were washed 2× with IP buffer (20 nM Tris-HCl pH 8.0, 150 mM NaCl) prior to lysate and hipMHC addition, and incubated rotating overnight at 4 °C to isolate pMHCs. Beads were washed with 1× TBS and water, and pMHCs were eluted in 10% formic acid for 20 min at room temperature (RT). Peptides were isolated from antibody and MHC molecules using a passivated 10 K molecule weight cut-off filters (PALL Life Science), lyophilized, and stored at −80 °C prior to analysis.pMHC labeling with TMTs and SP3 cleanupFor labeled analyses, 100 μg of pre-aliquoted TMT 6-plex (TMT) was resuspended in 30 μL anhydrous acetonitrile (MeCN), and lyophilized peptides were resuspended in 100 μL 150 mM triethylammonium bicarbonate and 50% ethanol. Both were gently vortexed, centrifuged at 13,400 × g for 1 min, and combined. TMT/peptide mixtures were incubated on a shaker for 1 h at RT, followed by 15 min of vacuum centrifugation. After combining labeled samples, we washed tubes 2× with 25% MeCN in 0.1% acetic acid (AcOH) and added it to the labeled mixture, which was subsequently centrifuged to dryness.Sample cleanup was performed using single-pot solid-phase-enhanced sample preparation (SP3) as previously described59. Briefly, a 1:1 mix of hydrophobic/hydrophilic Sera-mag carboxylate-modified speed beads (GE Healthcare) was prepared at a final bead concentration of 10 μg μL−1. Labeled samples were resuspended in 30 μL of 100 mM ammonium bicarbonate (pH 7–8) and added to 500 μg of bead mix with 1 mL MeCN. Peptides were allowed to bind for 10 min at RT, washed 2× with MeCN, and eluted with 2% DMSO for 1 min of sonication in a bath sonicator. TMT-labeled peptides were transferred to a fresh microcentrifuge tube and centrifuged to dryness.Synthetic peptide standardsHeavy leucine-containing peptides were synthesized at the MIT Biopolymers and Proteomics Lab using standard Fmoc chemistry using an Intavis model MultiPep peptide synthesizer with HATU activation and 5 μmol chemistry cycles. Starting resin used was Fmoc-Amide Resin (Applied Biosystems). Cleavage from resin and simultaneous amino acid side chain deprotection was accomplished using: trifluoroacetic acid (81.5% v/v); phenol (5% v/v); water (5% v/v); thioanisole (5% v/v); 1,2-ethanedithiol (2.5% v/v); 1% triisopropylsilane for 1.5 h. Standard Fmoc amino acids were procured from NovaBiochem and Fmoc-Leu (13C6, 15N) was obtained from Cambridge Isotope Laboratories.Peptides were quality controlled by MSy and reverse phase chromatography using a Bruker MiroFlex MALDI-TOF and Agilent model 1100 HPLC system with a Vydac C18 column (300 Å, 5 μm, 2.1 × 150 mm2) at 300 μL/min monitoring at 210 and 280 nm with a trifluoroacetic acid/H2O/MeCN mobile phase survey gradient. All peptides contain C-terminal amidation, with the exception of the BCAP31 and DDX5 peptides used for absolute quantification. For amidated peptides, we observe C-terminal amidation and C-terminal carboxyl groups on peptides synthesized with an amide group. Therefore, both are considered in downstream analyses.RNA-sequencingRNA was isolated from 10 cm plates of SKMEL5 cells with three biological replicates per condition. Prior to harvest, cells were washed with ice-cold 1× PBS over ice and lysed in TRIzol reagent (Thermo Fisher). Total RNA was isolated from each sample using Direct-zol RNA MiniPrep kit (Zymo Research) according to the manufacturer’s instructions.RNAs were confirmed for quality using the Agilent Fragment Analyzer and 300 ng of material was polyA selected using NEBNext Poly(A) mRNA Magnetic Isolation Module (E7490) modified to include two rounds of polyA binding and 10 min incubations. cDNA was generated using the NEB Ultra II Directional Kit (E7760) following the manufacturer’s protocol using 12 cycles of PCR and a 0.9X SPRI clean. The resulting libraries were quality assessed using the Fragment Analyzer and quantified by quantitative PCR prior to be sequenced on the Illumina HiSeq2000. The 40 nt single-end reads with an average depth of five million reads per sample were sequenced for all conditions.RNA-seq reads were aligned to the human transcriptome prepared with the hg38 primary assembly and the Ensembl version 95 annotation using STAR version 2.5.3a60. Gene expression was summarized with RSEM version 1.3.0 and SAMtools version 1.3 (refs61,62). Differential expression analysis was performed with DESeq2 version 1.24.0 running under R version 3.6.0 with normal log fold change shrinkage63. Significance values (adjusted p value) are determined using the Wald test, and are multiple hypothesis corrected using Benjamini–Hochberg method. The resulting data were parsed and assembled using Tibco Spotfire Analyst version 7.11.1.MS data acquisitionFor MS analysis, peptides were resuspended in 0.1% AcOH and loaded on a precolumn packed in-house [100 μm ID × 10 cm packed with 10 μm C18 beads (YMC gel, ODS-A, 12 nm, S-10 μm, AA12S11)]. The precolumn was then washed with 0.1% AcOH and connected in series to an analytical capillary column with an integrated electrospray tip (~1 μm orifice) with 5 μM C18 beads, prepared in-house [(50 μm ID × 12 cm with 5 μm C18 beads (YMC gel, ODS-AQ, 12 nm, S-5 μm, AQ12S05)].Peptides were eluted using a 130-min gradient with 10–45% buffer B (70% MeCN, 0.2 M AcOH) from 5 to 100 min and 45–55% buffer B from 100 to 120 min at a flow rate of 0.2 mL/min for a flow split of ~10,000:1. Peptides were analyzed using a Thermo Fisher Q Exactive HF-X Hybrid Quadrupole-Orbitrap mass spectrometer, and data were acquired using Thermo Fisher Scientific Xcalibur version 2.9.0.2923. Standard MS parameters were as follows: spray voltage, 2.5 kV; no sheath or auxiliary gas flow; heated capillary temperature, 250 °C.The HF-X was operated in DDA mode for LF and TMT analyses. LF: Full-scan MS spectra (mass/charge ratio (m/z), 350–2000; resolution, 60,000) were detected in the Orbitrap analyzer after accumulation of ions at 3e6 target value with a maximum injection time (IT) of 50 ms. For every full scan, the top 20 most intense ions were isolated (isolation width of 0.4 m/z) and fragmented (collision energy: 28%) by higher energy collisional dissociation with a maximum injection time of 300 ms, automatic gain control target 1e5, and 60,000 resolution. Charge states <2 and >4 were excluded, and dynamic exclusion was set to 30 s. TMT: Full-scan MS spectra (m/z, 400–2000; resolution, 120,000) were detected in the Orbitrap analyzer after accumulation of ions at 3e6 target value with a maximum IT of 50 ms. For every full scan, the 20 most intense ions were isolated (isolation width of 0.4 m/z) and fragmented (collision energy: 29%) by higher energy collisional dissociation with a maximum injection time of 350 ms, AGC target 1e5, and 30,000 resolution. Charge states <2 and >4 were excluded, and dynamic exclusion was set to 60 s. To ensure fragmentation of normalization standards, one fraction may be analyzed using targeted selected ion monitoring used in tandem with DDA with an inclusion list of hipMHC standards. For absolute quantification, the HF-X was operated in DDA mode with inclusion list enabled. Parameters mirror those of the TMT DDA method, with several exceptions. Full-scan mass spectra m/z range: 300–1200, maximum MS2 injection time 200 ms, only charge states of 2 and 3 were considered. Inclusion list masses and charge states listed in Supplementary Data 6.MS search space and filteringAll mass spectra were analyzed with Proteome Discoverer (PD, version 2.2) and searched using Mascot (version 2.4) against the human SwissProt database. No enzyme was used, and variable modifications included oxidized methionine for all analyses and phosphorylated serine, threonine, and tyrosine for cell treatment analyses. Treatment analyses were also searched against a previously published catalog of over 40,000 predicted antigenic mutations in cancer cell lines64. Heavy leucine-containing peptides were searched for separately with heavy leucine (+7), C-terminal amidation, and methionine oxidation as dynamic modifications against a custom database of the synthetic peptide standards. All analyses were filtered with the following criteria: search engine rank = 1, isolation interference ≤ 30%, and length between 8 and 15 amino acids. LF analyses were filtered with ion score ≥ 20, and labeled samples were filtered with ion score ≥15 and percolator q value ≤ 0.05. AUC quantitation was performed using the minora feature detector in PD with match between runs enabled and filtered for ion score ≥20. For targeted, absolute quantification analyses, total ion count values for each scan and peak intensities were extracted using Skyline (version 19.1.0.193)65.MS data analysis with hipMHC correctionFor LF analyses, correction parameters were determined by calculating the ratio of AUC intensities in each sample against a reference sample and taking the mean across hipMHCs. For TMT-labeled samples, ratios against a reference channel (usually TMT126) were calculated and the median of all ratios for correction hipMHCs was used to determine the final correction parameters. Only PSMs of heavy leucine-coded peptides with an average reporter ion intensity within 10-fold of the interquartile range of endogenous PSM reporter ion intensities were used for correction, as we observed drift in the correction factors when PSM TMT intensities were well beyond endogenous levels. For absolute quantification analyses, correction factors were generated as described for TMT analyses, and used to normalize maximum peak intensity values for DDX5 and BCAP31. Notably, with mean fold changes >2× between samples (e.g., IFN-γ stimulation), in our hands hipMHCs are no longer able to correct between conditions despite narrow isolation window (0.4 m/z). This inaccuracy may be due to co-isolation, as the calculated correction factors reflect median fold changes of endogenous peptides. In this case, we generated correction factors for each treatment condition separately.Correction factors were applied to AUC values in LF analyses for all peptides that were quantifiable across samples. For labeled samples, ion intensities of PSMs for each unique peptide across analyses of the same sample were summed, after which normalization factors were applied. To evaluate differences between conditions, the log2-transformed ratio of arithmetic mean intensity for drug- and DMSO-treated samples (n = 3) was calculated. To determine if peptides were significantly increasing, an unpaired, two-sided t test was performed, and peptides with p ≤ 0.05 were considered significantly increasing. To evaluate which peptides were significantly enriched above the mean, treated samples were mean centered by dividing the ion intensity of each peptide by the mean fold change across all peptides, after which a Student’s two-tailed t test was performed on adjusted values. Peptides with a mean-adjusted p value ≤ 0.05 were considered significantly enriched. Mean centering was not performed on samples where the mean log2 fold change was between −0.07 and 0.07. Data analyses were performed using Matlab version R2019b, and Microsoft Excel version 16.34.pMHC binding affinityBinding affinity of pMHCs was estimated using NetMHCpan-4.0 against each cell line’s allelic profile28,40 (Supplementary Fig. 3k). Only 9-mers were evaluated, and the minimum predicted affinity (nM) of each peptide was used to assign peptides to their best predicted allele. The threshold for binding was set to 500 nM. Binding motifs for the alleles were generated using 9-mers with predicted affinity <500 nM, and visualized using WebLogo 2.8.2 (ref.66). To estimate the proportion of peptides predicted to be binders by chance, 10 sets of 2000 random 9-mers were created by selecting with equal probability any amino acid more than 8 amino acid from a protein C terminus as a start site from human proteins in SwissProt version 2019_2, and binding affinity prediction was performed against the alleles of MDA-MB-231 cells. Data presented in Fig. 2d are a representative example.Enrichment analysesFor pMHC pathway enrichment analyses, gene names from peptide source proteins were extracted and rank ordered according to the average log2 fold change over DMSO-treated cells. In cases where more than one peptide mapped to the same source protein, the maximum/minimum was chosen, depending on the directionality of enrichment analysis. For RNA-seq data, gene sets were rank ordered according to the mean log2 fold change value with only protein encoding genes considered. We utilized GSEA 4.0.3 pre-ranked tool against the Molecular Signatures Database hallmarks gene sets with 1000 permutations, weighted enrichment statistic (p = 1), and a minimum gene size of 8 for pMHC analyses and 15 for RNA-seq42–44. Results were filtered for false discovery rate q value ≤0.25, and nominal p value ≤ 0.05.Significantly enriched peptides (mean-adjusted p value ≤0.05) were analyzed using STRING v.11 for GO term enrichment against biological processes and cellular components data sets67,68. Enriched categories were filtered according to false discovery rate q value ≤0.05.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Description of Additional Supplementary Files
Supplementary Data 1
Supplementary Data 2
Supplementary Data 3
Supplementary Data 4
Supplementary Data 5
Supplementary Data 6
Reporting Summary | nature communications | [
"Article"
] | [
"Peptides",
"Mass spectrometry",
"Proteomics",
"Melanoma",
"MHC class I"
] | present signals targets for immune cell recognition signals peptides class I major histocompatibility complexes derived from intracellular proteins external representation internal cell peptide MHC repertoire of cancer cells contain tumor-associated antigens tumor-specific markers activate T cells antitumor immune response interaction strengthened with checkpoint blockade (CB) immunotherapies low response rates barriers to clinical evidence suggests combining CB with treatments small-molecule inhibitors cytotoxic agents radiotherapy could potentiate response CB tumor immunogenicity surface pMHC clinical trials optimal combination of agents order timing administration beginning strategies quantitative understanding of immunopeptidome required absolute quantification of antigens necessary immunotherapy drug design strategies varying thresholds antigen expression for optimal antitumor response.Traditional data-dependent methods profile pMHC repertoires quantitative methods have limitations quantification pMHC methods lack normalization strategy for variations sample input Peptide losses processing vary across peptide sequences concentrations samples need for normalization19quantification pMHCs comparing endogenous levels to exogenous peptide standards sample Losses accounted with internal pMHC standards require refolding target approach relies on single point calibration ion suppression pMHC levels challenges platform ultraviolet-mediated peptide exchange MHC monomers heavy-isotope-labeled pMHCs for quantification pMHC low sample input addition of heavy-isotope pMHCs sample lysates improves quantitative accuracy label-free analyses provides estimate ion suppression through regression internal calibrant utilize hipMHC multipoint-embedded standard curves isobaric mass tags quantify number copies per cell of target antigens platform profile immunopeptidomic changes in melanoma cell lines comparing treatment with palbociclib CDK4/6 inhibitor interferon-γ modulators antigen Peptides from response palbociclib IFN-γ enriched in pMHC repertoire following treatment connecting intracellular response to extracellular immune presentation peptides from metabolic response to palbociclib tumor display increased presentation with palbociclib treatmentplatform profile immunopeptidomic changes high low-input format sample types treatments combination therapy strategies identify quantify treatment-modulated antigen targets targeted immunotherapy relative absolute pMHC platform accurate quantification pMHCs samples controlling losses processing enrichment analysis internal standards multipoint calibration curves heavy-isotope-labeled MHC peptides synthesized loaded biotinylated MHC monomers UV-mediated peptide loading MHC proteins concentration stable hipMHC complexes determined enzyme-linked immunosorbent assay used spiked same concentration whole-cell lysate normalization correction other titrated different concentrations verify correction estimate dynamic range suppression create internal standard curve quantification peptide adding hipMHCs endogenous exogenous pMHCs isolated immunoprecipitation acid elution molecular weight size-exclusion filtration. 1Platform quantitative immunopeptidomics hipMHCs generated UV-mediated peptide exchange HLA*A2:01 monomers heavy leucine HLA-A*02:01 binding peptidehipMHCs concentrations measured ELISA complexes added to lysate samples immunoprecipitation for quantification correction or titrated internal standard curve Heavy light pMHCs isolated with IP acid elution molecular filters Peptides analyzed LC-MS/MS chromatographic elution correction factors hipMHC AUC intensity ratios Samples multiplexed analysis TMT-labeled relative quantification reporter ion intensities Normalization hipMHC reporter ion intensity ratios across TMT channels absolute quantification TMT-labeled samples hipMHC internal standard curve calculate endogenous copies per cell pMHC.Peptide mixtures analyzed by liquid chromatography spectrometry three ways (Fig. analyses samples analyzed individually peptides quantified integrating area under curve) chromatographic elution precursor peptide-spectrum match Relative AUC intensities hipMHCs normalize AUC intensities peptides labeled samples TMT ion intensity ratios hipMHCs normalization TMT-labeled titrated hipMHCs for absolute quantification peptidesTMT intensities hipMHCs generated peptide-specific multipoint calibration curve average copies per cell control heavy-isotope-coded synthetic peptides not MHCs spiked into whole-cell lysate prior IP peptides not detected LC-MS/MS analysis peptides stable complexes isolated excess free peptides displace peptides.HipMHC standards improve quantitative accuracy used five LF six TMT-labeled replicates 1 × 107 MDA-MB-231 breast cancer cells variance before after hipMHC correction spiked 30 fmol two correction hipMHCs each sample added 30–300 fmol titrated hipMHC across samples 2369 unique pMHCs identified across five LF analyses 1352 quantifiable via AUC integration 589 quantified in all five analyses poor run-to-run overlap LF analyses 1754 unique peptides quantifiable with TMT-labeled analyses extra sample handling steps TMT losses labeled samples divided into six analyses increasing unique LF TMT analyses peptides matched expected length distributions of 8–11 amino acids82% LF 92% TMT 9-mers <500 nM (Fig. peptides apportioned alleles LF TMT analyses allele motifs aligned reducing input material 5 × 106 cells 86% unique peptides identified × 107 cells LF analysis low-input pMHC analyses Fig 2HipMHCs improve quantitation LF TMT samples Experimental design Five LF six TMT replicates 1 107 cells hipMHCs LF TMT quantification Peptide sequence hipMHC heavy-isotope-labeled leucine ALNEQIARL7 SLEEPIGHL7 quantification correction hipMHCs SVVESVKFL7 titrated LF sample #6 omitted Two thousand three hundred sixty-nine LF peptides identified five analyses 1352 quantifiable 589 quantifiable five analyses One thousand seven hundred fifty-four unique peptides quantified TMT-labeled analyses six Ninety-two percent TMT 82% LF 5.6% random peptide 9-mers bind HLA allele MDA-MB-231 cells affinity < 500 nMLinear fit titrated hipMHC peptide LF TMT Raw r2 0.48 (LF 0.91 hipMHC-adjusted 0.99 (LF 0.96 Distribution log2 change mean peptide quantification raw hipMHC Gaussian fit frequency distribution log2(FC) raw hipMHC-adjusted LF TMT samples 99.7% variance peptide quantitation 1.65 (raw 1.52 (adj) FC mean LF 1.30 1.23 TMT Median coefficient variation LF (24.27% raw 20.99% adj TMT (14.00% raw 7.48% adj). Error bars interquartile range Source data LF TMT data applied correction parameters quantification hipMHCs titrated peptide SVVESVKFL7 improved linear fit after correction pronounced effect LF samples dynamic range suppression peptide TMT-labeled (4.7× LF (2.7×) data sets quantitative differences larger measured hipMHC quantification correction reduced variation peptides TMT sample 5 lower intensities corrected hipMHC normalizationstandard deviation samples decreased hipMHC correction LF TMT-labeled samples TMT labeling lower variation replicates higher confidence shifts immunopeptidome peptides lower abundance higher variation no correlation in LF TMT analyses quantification endogenous selected two peptides TMT-labeled MDA-MB-231 cells quantification KLDVGNAEV B cell receptor protein 31 KQVSDLISV DEAD-box RNA helicase 5 BCAP31 regulates transport membrane proteins Golgi antigen processing TAA DDX5 important gene expression implicated proliferation metastasis tumorigenesis peptides detected differing levels abundant BCAP31 98th percentile DDX5 33rd percentile peptides synthesized with heavy-isotope-labeled leucine hipMHC normalization standards added to three replicates 107 MDA-MB-231 cells BCAP31 DDX5 hipMHC labeled samples with TMT performed LC-MS/MS analysis list targeted peptides selected for fragmentation 3Absolute quantification of pMHCs with hipMHC standards isobaric labelingPeptide intensities TMT-labeled MDA-MB-231 cells Fig. 2 determined by AUC quantification Percentile abundance abundant peptide Experimental design Normalization standards 5 15 50 fmol BCAP31 0.3 1 3 fmol DDX5 hipMHCs added to three replicates 107 MDA-MB-231 cells peptides isolated labeled analyzed via LC-MS/MS Chromatographic elution profiles three TMT ion intensities average 3) TMT ion intensity endogenous peptide MS2 scan single point elution profiles gaussian distribution Adjusted apex intensity fmol hipMHC added standard curve endogenous concentration antigen fits BCAP31 DDX5 r2 > 0.999 endogenous peptide mean value ± SD for n = 3 replicates Source data traces TMT BCAP31 DDX5 peptides increasing intensities hipMHC added (Fig. apex intensities adjusted normalization linear fit pMHC concentration average 1740 copies per cell BCAP31 81 copies DDX5 peptideConcentrations DDX5 hipMHC low 100 attomole detected (six copies per cell broad pMHC expression levels method Fig 2) BCAP31 DDX5 dynamic range suppression 1.9× 2.5× ion suppression not uniform across peptides peptide-specific standards required quantification pMHC.CDK4/6 inhibition alters pMHC repertoire kinases 4 6 control cell cycle progression Rb1 releasing E2F transcription factors progression G1 CDK4/6 dysregulated overactive in cancer uncontrolled inhibitors powerful anticancer agents active tumor types including inhibitors enhance tumor immunogenicity surface MHC class I expression T cell activation data highlight CDK4/6 inhibitors combine immunotherapies response melanoma effect CDK4/6 inhibition MHC class I peptide repertoire not characterized applied platform quantify pMHC repertoires melanoma vitro treatment CDK4/6 inhibitor palbociclib understand combination therapy improve outcomes selected four melanoma cell lines analysis SKMEL5 SKMEL28 (BRAF SKMEL2 IPC298 (NRAS analysesselected two doses palbociclib low 1 μM below cell high 10 μM near IC50 Three replicates 107 cells treated dimethyl sulfoxide low- high-dose palbociclib 72 h Low-dose increased class I MHC presentation 1.5–2× high-dose milder effect 4Palbociclib alters immunopeptidome melanoma Viability 72 h treatment data DMSO control IC50s SKMEL5 12.74 μM SKMEL28 14.62 μM SKMEL2 16.98 μM IPC298 10.62 μM mean values ± n = 3 replicates except SKMEL5 setup TMT-labeled immunopeptidomics experiments melanoma cell lines Flow cytometry measurements surface HLA expression SKMEL5 cells Data % maximum signal distributions three experiments distribution log2 change) (palbociclib/DMSO unique pMHCs mean intensity n = 3 replicates Data % unique peptides Volcano plot log2(FC) 1 μM treated SKMEL5 cells Colored points (p < 0.05 log2(FC) > 1.56) correspond processesLog2(FC) enriched peptides GO enrichment processes source protein name False discovery rate p < 0.05. Four-way Venn diagram source proteins peptides enriched 1 μM palbociclib Log2(FC) pIRS2 peptide 1 μM 10 μM palbociclib *p < 0.05 unpaired two-sided t test Source protein name peptide sequence log2(FC) TAAs SKMEL5 IPC298 cells significance values source data Source Data File pMHC repertoire alterations palbociclib multiplexed relative quantitation low- high-dose palbociclib DMSO data normalized hipMHC standards 3b peptides matched length distributions majority MHC class I binders Immunopeptidomic analysis trend low-dose palbociclib shifted pMHC expression higher DMSO high palbociclib small increase expression SKMEL5 no change other cell lines 4d distribution changes peptide presentation low-dose treatment several peptides increasing eight- to ten-fold dynamic range suppression biology palbociclib-modulated pMHC alterations analyzed data twodetermined peptides source proteins increased with palbociclib treatment over DMSO peptides increased low-dose identified peptides enriched highlighting peptides modulated by palbociclib performed GO enrichment on 127 peptides enriched in low-dose-treated SKMEL5 cells (Fig. identified enriched biological processes ribosomal biogenesis glucose metabolic process antigen processing response palbociclib7 (Fig. 4f analysis raw non-normalized values 66 peptides enriched without hipMHC correction altering Gene Oncology) term pathway analysis results peptides mapping ribosomal biogenesis enriched other two biological processes not importance using hipMHCs for quantification correction pMHC alterations compared source proteins peptides enriched low-dose treatment DMSO across four cell lines majority (72–88%) enriched source proteins unique to each cell line three proteins common vimentin β-actin-like protein 3 SIL1 nucleotide exchange factor (Fig. 4g 17 proteins common identified uniqueness proteins altered by palbociclib each immunopeptidomic landscapeinvestigated commonality 17 proteins high abundance peptide mixtures SKMEL5 cells scattered AUC intensities Fig shared pMHCs source proteins limited interest serine-phosphorylated IRS2 peptide RVA[pS]PTSGVK modified sequence restricted malignant cells phosphorylated form immunogenic no alleles four cell observed pIRS2 peptide increasing cell lines low-dose treatment (Fig. RVA[pS]PTSGVK high expression among pMHCs isolated without-enrichment41 peptide broadly targetable antigen expression modulated CDK4/6 inhibition palbociclib treatment TAAs proteins MLANA (MART1) PMEL TYR increased presentation (Fig. 4i). antigens source proteins not universally conserved cell lines increased TAA presentation 1 μM palbociclib treatment increase antigen presentation immunotherapies antigens.Response palbociclib reflected in immunopeptidomeTo quantitative differences immunopeptidome after palbociclib treatment cell signaling response performed nonparametric test positively negatively enriched pathwaysGene source proteins change treatment searched MSigDB Hallmarks gene database enriched pathways low-dose treatment high-dose palbociclib enrichment pMHCs E2F targets G2M checkpoint DNA repair mitotic spindle MTORC1 signaling pathways cell lines. reflect CDK4/6 inhibition inhibiting CDK4/6 decreases expression E2F targets peptides Ki-67 proliferation depleted four cell lines E2F controls genes DNA damage repair γH2AX levels DNA double-strand breaks increased 72 h palbociclib treatment dose processes cell proteins depleted E2F peptides little overlap cell lines individuality source proteins pMHC repertoire. 5Pathway analysis palbociclib-altered immunopeptidome score enriched pathways 10 μM palbociclib enrichment directionality q < 0.25 *p < 0.05 **p < 0.01 ***p < 0.001interactions E2F peptides 10 μM palbociclib OxPhos peptides increasing 1 μM palbociclib except SKMEL28 peptides 10 μM Node color cell line MitoTraker green intensity 72 h palbociclib treatment box whiskers plot minimum maximum signal Dunnett’s comparisons test DMSO *p < 0.05 < 0.0001 Correlation log2 fold change (palbociclib/DMSO RNA expression pMHC presentation SKMEL5 cells treated 72 h 1 μM palbociclib r2 = 0.04. FC mean intensity 3 replicates per condition enriched pathways RNA-seq data (p < 0.05 q < 0.25) pathways reflect immunopeptidome analysis Log2 SKMEL5 OxPhos peptides < 0.05 1 μM palbociclib RNA expression differences RNA expression (palbociclib versus DMSO **p < 0.01 ****p < 0.0001 data Source Data File one pathway oxidative phosphorylation upregulated SKMEL28 cellscell lines presented peptides OxPhos increased with palbociclib treatment prominent low-dose treatment in SKMEL5 IPC298 SKMEL2 cells SKMEL28 OxPhos CDK4/6 inhibition increased ATP mitochondrial mass elevating metabolic activity all samples showed elevated mitochondrial levels following treatment enriched pMHC presentation reflects change metabolic cell state alterations pMHC repertoire responses CDK4/6 inhibition tested changes expression immunopeptidome changes No correlation (r2 = 0.04) observed between pMHC expression RNA expression mechanisms beyond gene expression regulate pMHC presentation enriched gene sets immunopeptidome present in RNA-sequencing analysis E2F pMHCs depleted in SKMEL5 cells correlated decreased gene expression proteins five of 15 enriched OxPhos peptides higher gene expression with palbociclib treatment three decreasing seven unchanged data suggest changes gene expression pMHC pathways not predictive alterations immunopeptidomeIFN-γ pMHC alterations CDK4/6 inhibition stimulates IFN signaling antigen presentation observed upregulation IFN-γ genes low-dose palbociclib treatment increased expression genes antigen presentation tested IFN-γ stimulation repertoire CDK4/6 inhibition Cells stimulated DMSO 10 ng mL−1 IFN-γ 72 h pMHC repertoires quantified multiplexed hipMHC platform IFN-γ increased pMHC levels >2× cell line immunopeptidome pMHC increased stimulation pMHC repertoire response IFN-γ stimulation RNA-seq pMHC log2 change 1 μM palbociclib/DMSO 3 replicates per antigen-processing genes Surface HLA expression flow cytometry cells treated 72 h IFN-γ Errors biological replicates n = 8 11 9 9 SKMEL5 SKMEL28 SKMEL2 IPC298 Immunopeptidome log2 mean changes 2.42 2.50 2.08 3.04 for SKMEL5 IPC298 cellsenriched pathways SKMEL5 cells 72 h IFN-γ q < 0.25 < 0.05 < 0.01. Enrichment plot IFN-γ enrichment score log2(FC) rank peptide Open circles show enriched IFN-γ peptides Volcano plot IFN-γ changes Peptides log2(FC) versus mean-adjusted p value Red points represent peptides enriched (p < 0.05 change > 2.42) diagram enriched source proteins peptides between IFN-γ 1 μM palbociclib-treated SKMEL5 cells interaction network enriched source proteins enriched gene ontology Significance values adjusted data Source Data File similarity response palbociclib treatment performed GSEA against hallmark gene sets upregulated pathway IFN-γ stimulation “IFN-γ response peptides proteins antigen processing STAT1 HLA-A (Fig IFN-γ response top enriched pathway cell line response differences pMHC presentation IFN-γ-related peptides upregulated by IFN-γ stimulationpathways G2M checkpoint mitotic spindle enriched in IFN-γ stimulated cells contrast palbociclib treatment cell lines showed differential pMHC pathway enrichment CDK4/6 inhibition palbociclib IFN-γ stimulation tested pMHCs source proteins enriched SKMEL5 cells identified 20 peptides 31 source proteins enriched conditions (Fig. 6f cytoplasm ribosomal translation initiation proteins overrepresented immunopeptidomic data sets CDK4/6 inhibition IFN-γ response stimulating cells IFN-γ recapitulate peptide repertoire alterations palbociclib IFN-γ stimulation alters repertoire presentation IFN-γ-related peptides addition hipMHCs improves quantitative accuracy LF multiplexed labeled analyses multiplexed labeling TMT superior accuracy peptide specificity higher unique peptides hipMHC standards account for variation across samples estimate dynamic range suppression hipMHC correction alters interpretation pMHC repertoire changes vitro system Utilizing hipMHCs beneficial accounting variation sample losses vivo samples quantitation analysesuse TMT 6-plex compatible with isobaric labeling strategies iTRAQ TMT 11-plex TMTpro 16 samples simultaneously rapid profiling immunopeptidome changes minimal sample input translatable for vivo tissue samples reductions in sample input 14-gauge needle unique peptides identified advancements speed mass spectrometers sample preparation techniques reduce losses enable pMHC profiling lower sample inputs hipMHCs labeling sample amount increased coupled with fractionation for deeper sequencing pMHC repertoire neoantigen identification49 improved relative quantification utility hipMHCs for pMHC absolute quantification generating embedded multipoint standard curve targeted MS detect levels antigen from 1 × 107 cells titrated hipMHC standard accurate absolute quantification copies per cell for two pMHC’s ~20-fold difference in abundance absolute quantification limited to two peptides advanced targeted MS methods enable quantitation of hundreds of peptides single determine absolute quantification antigens without pMHC-specific antibody targeted immunotherapy designpeptides lower abundances for TCR therapies TCRs sensitive pMHC complex T cell antibody-based therapies targeting specific pMHCs benefit higher antigen expression levels results vary across antigen targets quantification pMHC expression relationships among antigen processing epitope abundances immunogenicity off-target restriction hipMHCs commercial availability UV-mediated MHC monomers ELISA control reagents limited to common human class I alleles matched allele hipMHCs not required for normalization correction necessary for accurate quantification curves disulfide-stabilized HLA molecules place UV-mediated complexes show increased stability higher exchange efficiency lower-affinity peptides need ELISA simplifying MHC refolding other alleles species applied quantitative multiplexed hipMHC normalization pMHC response to CDK4/6 inhibition palbociclib treatment in melanoma results extracellular changes pMHC abundance reflective intracellular response CDK4/6 inhibition palbociclib treatment increased presentation of TAAs peptides metabolic processeshigh tumor antigen metabolic protein expression of checkpoint inhibitor response in melanoma palbociclib with CB- or TIL-based therapies increase tumor immunogenicity54 peptide antigens expression increased by palbociclib for targeted immunotherapy clinical trials combination treatment regimens coupling checkpoint blockade therapies expanding55–57 Quantifying molecular consequences regimes with platform insight design new therapeutic combinations targeted immunotherapies quantitative immunopeptidomic data demonstrate utility in evaluating pMHC repertoire response therapy multiplexed platform allows analyses many samples short timescale Further analyses of pMHC repertoire changes order timing therapies predictions immunopeptidome immunotherapy targeting.MethodsHuman cell linesSKMEL5 SKMEL28 MDA-MB-231 cell lines obtained from ATCC maintained in DMEM IPC298 SKMEL2 cells provided by Array Biopharma maintained in RPMI-1640 (Gibco) minimum essential medium-αmedia supplemented 10% fetal bovine serum 1% penicillin/streptomycin Cells tested for mycoplasma contamination maintained 37 °C 5% CO2.Phenotypic assaysIC50 palbociclib determined cell line CellTiter-Glo cell viability assay Cells seeded 10,000 or cells per 96-well plate adhere overnight treated palbociclib or DMSO fresh 72 h assayed Data acquired Tecan plate reader Infinite 200 icontrol version 1.7.1.12. IC50 values calculated four-parameter logistic curve Prism 8.4.1.Mitochondrial content measured fluorescent mitochondrial stain Cells seeded cells per well 24-well plate adhere overnight treated 1 μM or 10 μM palbociclib DMSO 72 h assayed 200 nM MitoTracker Green FM 1:1000 NuclearID Red DNA stain 15 min serum-free medium 37 °C After cells imaged analyzed Incucyte Live Cell Analysis System integrated intensity MitoTracker dye calculated each image divided by number cells mitochondrial intensityone-way analysis variance comparisons test Prism treated cells vehicle DMSO control values multiplicity-adjusted p values lifted 0.05% Trypsin-EDTA 106 cells/mL spun 300 × g 3 min washed ice-cold phosphate-buffered saline 1% FBS 0.1% sodium azide incubated fluorophore-conjugated antibody 0.5 μg mL−1 flow buffer 30 min washed resuspended flow buffer 5 μL propidium iodide staining solution Analyses LSRII (BD Biosciences FlowJo antibodies purchased BioLegend Alexa Fluor 488 HLA-A -B -C clone Fluor 488 anti-H2A.X Phospho 2F3 gating strategy Supplementary Fig. 6.UV-mediated peptide exchange recombinant biotinylated Flex-T HLA-A*02:01 monomers 4 μL 500 μM peptide stock 2 μL Flex-T monomer 32 μL PBS 96-well U-bottom plate illuminated UV light (365 nm 30 min 30-min incubation 37 °Cstable complexes peptide exchange quantified Flex-T HLA I ELISA assay HLA-A*02:01 ELISA results Tecan plate reader Infinite 200 icontrol 1.7.1.12.pMHC cells seeded 10 cm plates overnight treated 72 h palbociclib 10 ng mL−1 recombinant IFN-γ DMSO control cells washed 1× PBS lifted 0.05% Trypsin-EDTA pelleted 500 × g 5 min resuspended 1 mL lysis buffer nM Tris-HCl pH 8.0 150 mM NaCl 0.2 mM PMSO 1% CHAPS HALT Protease/Phosphatase Inhibitor Cocktail sonication lysate cleared centrifugation 5000 × g 5 min quantified Bicinchoninic Acid Protein Assay Kit isolated 107 cells condition IP size-exclusion filtration 0.5 mg pan-specific anti-human MHC class I-A-B-C antibody/32 20 μL FastFlow Protein Sepharose bead slurry 3 h 4 °C Beads washed 2× IP buffer-HCl 150 NaCl incubated overnight 4 °C isolateBeads washed TBS water pMHCs eluted 10% formic acid 20 min Peptides isolated antibody MHC molecules 10 filters lyophilized stored −80 °C labeling TMTs SP3 100 μg TMT 6-plex) resuspended 30 μL anhydrous acetonitrile lyophilized peptides 100 μL 150 mM triethylammonium bicarbonate 50% ethanol vortexed centrifuged 13,400 1 min combined TMT/peptide mixtures incubated shaker 1 h 15 min vacuum centrifugation washed tubes 25% MeCN 0.1% acetic acid added mixture centrifuged dryness cleanup single-pot solid preparation 1:1 mix hydrophobic/hydrophilic Sera-mag carboxylate-modified speed beads 10 μg samples resuspended 30 μL 100 mM ammonium bicarbonate added 500 μg bead mix 1 mL MeCN Peptides bind 10 min washed MeCN eluted 2% DMSO 1 min TMT-labeled peptides microcentrifuge centrifuged drynessSynthetic peptide leucine peptides synthesized MIT Biopolymers Proteomics Lab Fmoc chemistry Intavis MultiPep peptide synthesizer HATU activation 5 μmol chemistry cycles resin Fmoc-Amide Resin Cleavage amino acid deprotection trifluoroacetic acid water thioanisole 1,2-ethanedithiol (2.5% 1% triisopropylsilane 1.5 h Fmoc amino acids NovaBiochem Fmoc-Leu Cambridge Isotope Laboratories quality controlled MSy reverse phase chromatography Bruker MiroFlex MALDI-TOF Agilent 1100 HPLC system Vydac C18 column 300 μL/min 210 280 nm trifluoroacetic acid/H2O/MeCN phase survey gradient peptides C-terminal amidation BCAP31 DDX5 peptides amidated peptides C-terminal amidation carboxyl groups analyses isolated 10 SKMEL5 cells three replicates cells washed ice-cold lysed TRIzol reagent Total RNA isolated Direct-zol RNA MiniPrep kit (Zymo ResearchRNAs confirmed Agilent Fragment Analyzer 300 ng material selected NEBNext Poly(A) mRNA Magnetic Isolation Module (E7490) two rounds polyA binding 10 min incubations cDNA generated NEB Ultra II Directional Kit (E7760) 12 cycles PCR 0.9X SPRI clean libraries quality assessed Fragment Analyzer quantified quantitative PCR Illumina HiSeq2000 40 nt single-end reads average five million reads sample sequenced conditions.RNA-seq reads aligned human transcriptome hg38 primary assembly Ensembl 95 annotation STAR 2.5 Gene expression summarized RSEM 1.3.0 SAMtools 1.3 Differential expression analysis DESeq2 1.24.0 R version 3.6.0 normal log fold change Significance values determined Wald test hypothesis corrected Benjamini–Hochberg method data parsed assembled Tibco Spotfire Analyst 7.11.1.MS data peptides resuspended 0.1% AcOH loaded precolumn μm ID 10 cm 10 μm C18 beads precolumn washed 0.1% AcOH connected analytical capillary column 5 μM C18 beadsPeptides eluted 130-min gradient 10–45% buffer B MeCN 0.2 M AcOH 5 to 100 min 45–55% buffer B 100 to 120 min flow rate 0.2 mL/min split ~10,000:1 analyzed Thermo Fisher Exactive HF-X Hybrid Quadrupole-Orbitrap mass spectrometer data Thermo Fisher Scientific Xcalibur 2.9.0.2923 parameters spray voltage 2.5 kV no gas flow heated capillary temperature 250 °C HF-X DDA mode LF TMT analyses LF Full-scan MS spectra 350–2000 resolution 60,000 detected 3e6 50 20 intense ions fragmented 28%) energy dissociation injection time 300 ms control target 1e5 60,000 resolution Charge states <2 >4 excluded dynamic exclusion 30 s TMT Full-scan MS spectra/z 400–2000 resolution 120,000 detected 3e6 50 ms 20 intense ions fragmented 29%) injection 350 ms AGC target 1e5 30,000 resolution Charge states <2 >4 excluded exclusion 60 sfragmentation normalization fraction analyzed ion monitoring DDA hipMHC standards HF-X operated DDA mode inclusion list enabled Parameters mirror TMT DDA method exceptions Full-scan mass spectra m/z range 300–1200 MS2 injection time 200 ms charge states 2 3 considered Inclusion list charge states Supplementary Data 6.MS search mass spectra analyzed Proteome Discoverer searched Mascot 2.4 SwissProt database No enzyme used modifications oxidized methionine phosphorylated serine threonine tyrosine treatment searched 40,000 antigenic mutations cancer cell Heavy leucine-containing peptides searched-terminal amidation methionine oxidation synthetic peptide standards analyses filtered search engine rank 1 interference ≤ 30% length 8 15 amino acids LF analyses filtered ion score ≥ 20 labeled samples ion score ≥15 percolator q value ≤ 0.05. AUC quantitation minora feature detector filtered ion score ≥20 quantification total ion count values peak intensities extracted Skyline data hipMHC correction parameters determined AUC intensities reference sample mean hipMHCsTMT-labeled samples ratios reference channel TMT126 calculated median ratios correction hipMHCs correction parameters PSMs heavy leucine-coded peptides ion intensity within 10-fold endogenous PSM used correction drift intensities beyond endogenous levels quantification analyses correction factors generated peak intensity DDX5 BCAP31 mean changes >2× samples IFN-γ hipMHCs correct conditions narrow isolation window (0.4 m inaccuracy co-isolation correction factors reflect median fold changes endogenous peptides generated correction factors each treatment condition separately factors applied AUC values LF analyses peptides labeled samples ion intensities PSMs peptide summed normalization factors applied differences log2-transformed ratio arithmetic mean intensity drug- DMSO-treated samples (n = 3) calculated peptides increasing unpaired two-sided t test performed peptides p ≤ 0.05 considered significantly increasing peptides enriched samples centered dividing ion intensity mean fold change Student’s two-tailed t test adjusted values Peptides mean-adjusted p value ≤ 0.05 considered significantly enrichedcentering not performed on samples log2 fold change between −0.07 0.07. analyses Matlab R2019b Microsoft Excel 16.34.pMHC binding estimated NetMHCpan-4.0 cell allelic 9-mers evaluated minimum predicted affinity peptide best allele threshold binding 500 nM Binding motifs generated using 9-mers <500 nM visualized WebLogo 2.8.2 10 sets 2000 random 9-mers created protein C terminus SwissProt 2019_2 binding affinity prediction against alleles MDA-MB-231 cells Fig. 2d representative.Enrichment analysesFor pMHC analyses gene names peptide source proteins extracted rank ordered average log2 fold change over DMSO-treated cells peptide source protein maximum/minimum chosen RNA-seq data gene sets rank ordered mean log2 fold change value protein encoding genes utilized GSEA 4.0.3 pre-ranked tool against Molecular Signatures Database gene sets with 1000 permutations weighted enrichment statistic minimum gene size 8 for pMHC analyses 15 for RNA-seq42–44Results filtered false discovery rate ≤0.25 nominal p ≤ enriched peptides-adjusted p value ≤0.05) analyzed STRING v.11 for GO term enrichment against biological processes cellular components data Enriched categories filtered false discovery rate q ≤0.05.Reporting Nature Research Reporting Summary.Supplementary information Supplementary Files 1 Summary | 49.2 | 1.001984 |
10.1038/s41467-020-15757-0 | PMC7181735 | Applying first-principles molecular dynamic simulations and thermodynamic modelling, the authors suggest a vertical oxygen fugacity gradient in magma oceans of Earth, Mars, and the Moon. Consequently, the study proposes larger planets like Earth to have stronger oxidized upper mantles than smaller bodies such as Mars or the Moon. | Magma oceans were once ubiquitous in the early solar system, setting up the initial conditions for different evolutionary paths of planetary bodies. In particular, the redox conditions of magma oceans may have profound influence on the redox state of subsequently formed mantles and the overlying atmospheres. The relevant redox buffering reactions, however, remain poorly constrained. Using first-principles simulations combined with thermodynamic modeling, we show that magma oceans of Earth, Mars, and the Moon are likely characterized with a vertical gradient in oxygen fugacity with deeper magma oceans invoking more oxidizing surface conditions. This redox zonation may be the major cause for the Earth’s upper mantle being more oxidized than Mars’ and the Moon’s. These contrasting redox profiles also suggest that Earth’s early atmosphere was dominated by CO2 and H2O, in contrast to those enriched in H2O and H2 for Mars, and H2 and CO for the Moon. | IntroductionThe redox condition of planetary bodies influences their chemical differentiation and governs the composition of overlying atmospheres1–5. For instance, to understand how bio-essential volatiles such as carbon and hydrogen were initially incorporated near Earth’s surface requires knowledge about the redox state during early stages of Earth’s history. A number of studies have shown that the uppermost mantle of present-day Earth is considerably oxidized (IW + 3.5, that is, 3.5 log units above the iron-wüstite buffer)6. Petrological evidence also suggests that such oxidized conditions formed early, 4.3–4.4 Ga ago7. Unlike Earth, the present-day Martian and Lunar mantles are considered to be much more reduced (~IW − 1)8–11.These contrasting oxidization states may have been set up during the early phase of planetary formation when magma oceans (MOs) could have existed3,12. Such an MO involved mechanism has yet to be fully established because relevant redox controlling reactions are still poorly constrained in realistic magma ocean scenarios. Several studies have inferred the oxygen fugacity (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{{\mathrm{O}}_{2}}$$\end{document}fO2) profile of silicate melts at high pressures using the experimental data at zero or relatively low pressures and is applicable only for shallow magma oceans13,14. The oxygen fugacity is a function of pressure, temperature, and composition, thus likely varying greatly within MOs that could have extended very deep, even covering the entire mantle regime.Here, we study the redox controlling reactions in magma oceans by simulating silicate melts containing ferrous and ferric iron with first-principles molecular dynamics (FPMD) and perform thermodynamic modeling at pressures that cover the entire Earth’s mantle and temperatures up to 5000 K. The results suggest that ferric iron becomes increasingly energetically favorable with pressure mainly due to its small partial molar volume in silicate melts under large compression. Consequently, the magma oceans of Earth, Mars, and the Moon, if compositionally homogeneous due to vigorous mixing, would be characterized with a vertical gradient in oxygen fugacity. Specifically, a deeper magma ocean existing in the early Earth would have more oxidizing surface conditions compared with those of smaller bodies like Mars and the Moon. The contrasting surface conditions between these planetary bodies suggest that the early atmosphere in equilibrium with Earth’s surface may have been dominated by CO2 and H2O, in contrast to those enriched in H2O and H2 for Mars and H2 and CO for the Moon.Results and discussionEquations of state of silicate meltsAt the base of a MO where metallic melts may pond before sinking into the core15, the oxygen fugacity is governed by the equilibrium between the metallic and silicate melts, and can be directly calculated given the compositions of these melts are known. Away from the base where metallicmeltis absent due to its rapid sinking velocity16, the MO redox state is controlled by the following redox buffering reaction3,17:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{FeO}}\left( {{\mathrm{melt}}} \right)\;+\frac{1}{4}{{\mathrm{O}}_{2}}={{\mathrm{FeO}}_{1.5}}\left( {{\mathrm{melt}}} \right)$$\end{document}FeOmelt+14O2=FeO1.5meltThe thermodynamic behavior of the above reaction informs how oxygen fugacity varies with temperature and pressure. Taking the oxygen fugacity at the MO base as the boundary condition, one may, in principle, obtain the oxygen fugacity throughout the MO if the thermodynamic properties of the reactants and products in Eq. 1 are known. One key-parameter is the difference in molar volumes between FeO1.5 and FeO in the melts, ΔV. Its value has been directly measured only at 1 bar18 and also inferred from experiments performed up to 23 GPa and ~2500 K12,13,18–23. However, these conditions are still far from what are expected in MOs of Earth and Mars. Our goal is to calculate ΔV as a function of pressure, temperature, and composition so that we can constrain oxygen fugacity in the redox buffering Eq. (1) under directly applicable conditions. Moreover, we evaluate the MO redox states of Earth, Mars, and the Moon in order to understand their oxidation conditions of the present-day mantle and the chemistry of earliest atmosphere.We first present the results from FPMD simulations of iron-bearing MgSiO3 liquids with iron in different valence states at 2000–4000 K and up to 140 GPa (Methods). The calculated pressure–volume–temperature (P–V–T) relationships can be described with the following equation:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(V,T) \;=\; P\left( {V,T_0} \right)\; +\; B_{TH}(T\; -\; T_0)$$\end{document}P(V,T)=PV,T0+BTH(T−T0)Here \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(V,T_0)$$\end{document}P(V,T0) represents the reference isotherm at T0 = 3000 K using a fourth-order Birch-Murnaghan equation of state. The second term contains a thermal pressure coefficient, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B_{{\mathrm{TH}}}\left( V \right) = \left[ {a \;-\; b\left( {\frac{V}{{V_0}}} \right)\; +\; c\left( {\frac{V}{{V_0}}} \right)^2} \right]/1000$$\end{document}BTHV=a−bVV0+cVV02/1000 where a, b, and c are constants for a given melt composition. The bulk moduli of the Fe2+ -bearing melts are systematically larger than those of the Fe3+ bearing melts (Supplementary Table 1). This means that the Fe3+ bearing melts are more compressible at the conditions investigated (Fig. 1), consistent with previous low-pressure studies18,24.Fig. 1Molar volume difference (ΔV) between FeO1.5 and FeO in silicate melts.The calculated ΔV is shown as a function of pressure at different temperatures: a 12.5 mol% iron for Earth- and Moon-like magma ocean and b 25 mol% iron for Mars-like magma ocean. Insets show the corresponding pressure–volume relationships for melts containing 12.5 and 25 mol% iron as Fe2+ (solid symbols and curves) and Fe3+ (open symbols and dashed curves). Volumes are plotted along isotherms only to pressures where the simulated systems were in a liquid state. The 1σ standard deviation of ΔV is ~0.2–0.5 cm3 mol−1.Using the pressure–volume results of simulated silicate melts for the same molar content of Fe3+ and Fe2+, we calculate the difference in molar volume (ΔV) between FeO1.5 and FeO in the melts as a function of pressure (Fig. 1). Our calculated value of ΔV at zero pressure agrees well with existing experimental data18,24 (Supplementary Fig. 1). As pressure increases, ΔV decreases rapidly initially in the low-pressure regime. Thereafter, ΔV increases slightly and then decreases gradually at higher pressures. The predicted non-monotonicpressure trend weakens at higher temperatures. For silicate melts of different iron contents (i.e., 12.5 and 25 mol%), ΔV takes slightly different values, showing a weak positive trend with iron content. This is consistent with the observed weak dependency of ΔV on the melt composition18. Our results thus show that ΔV remains positive at all pressures up to 140 GPa irrespective of temperature and composition. This finding contradicts previous inferences that ΔV would keep on decreasing and eventually become negative within the pressure range of Earth’s mantle3,12.Previous models on ΔV either adopt a bulk modulus derivative of 4 or use an equation of state fit to experimental data within a limited pressure range12–14,23. We compare model values with our calculated results for silicate melts of 12.5 mol% iron, as these models are designed for Earth’s relevant composition (Supplementary Fig. 1). At low pressures (<10 GPa), our results are in good agreement with the recent model by ref. 12, both showing a sharp decrease of ΔV at low pressures, whereas at higher pressures, our results are in better line with other earlier models13,14,23, all showing that ΔV gradually levels out. These ΔV differences arise mainly due to the different pressure dependencies of the incompressibility (Kʼ) of FeO1.5 and FeO in silicate melts adopted by the previous studies. Our 4th order Birch-Murnaghan fit yields a lightly larger Kʼ for FeO1.5 (4.6) than that of FeO (3.3). Previous studies other than ref. 12 assume Kʼ of FeO1.5 and FeO to be 4, thus exhibiting similar pressure dependency of ΔV to our study. The contrasting behavior of ΔV from ref. 12 is caused by drastically different Kʼ values, 1.3 and 8, respectively, for FeO1.5 and FeO. These extreme values of Kʼ are not consistent with other experimental studies on silicate melts for which Kʼ is 3–825,26 and on FeO liquid for which Kʼ is 3–427,28. The reason for this inconsistency is, however, unclear. Our analysis of the coordination environment of iron in the silicate melts shows that the mean Fe–O coordination increases rapidly initially with pressure and more gradually at pressures beyond 40 GPa (Supplementary Fig. 2). This pressure trend is similar to that of ΔV. This implies an inherent correlation between the local iron-oxygen bonding environment in the silicate melt and ΔV.We stress that our first-principles results make no assumption on the value of Kʼ of FeO1.5 and FeO, so they are directly applicable over the entire mantle regime of Earth. To explore this implication further, we evaluate ΔV along two representative magma ocean thermal profiles referred to as “cold” and “hot” hereafter (Supplementary Fig. 3). The calculated ΔV varies considerably but remains positive over wide ranges of pressure and temperature of magma ocean relevance (Supplementary Fig. 4), thus indicating a positive contribution of pressure to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{{\mathrm{O}}_2}$$\end{document}fO2.Redox profilesOur calculated ΔV profiles along the magma ocean thermal profiles are used to assess the redox state of magma oceans of relevance to Earth, Mars, and the Moon. We assume that the MOs are fully convective and well-mixed, resulting in a homogeneous Fe3+ to the total Fe ratio (Fe3+/ΣFe)3. The thermodynamic relationship for the reaction (1) is3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$- \frac{{\Delta G_r^0\left( {P_0,T} \right)\; + \;\mathop {\smallint }\nolimits_{P_0}^P \Delta V\left( {P,T} \right)dP}}{{RT}} = \;{\mathrm{ln}}\frac{{X_{{\mathrm{FeO}}_{1.5}}^{{\mathrm{melt}}}}}{{X_{{\mathrm{FeO}}}^{{\mathrm{melt}}}}} \\ +\, {\mathrm{ln}}\frac{{\gamma _{{\mathrm{FeO}}_{1.5}}^{{\mathrm{melt}}}}}{{\gamma _{{\mathrm{FeO}}}^{{\mathrm{melt}}}}}\; - \;\frac{1}{4}{\mathrm{ln}} \, \,f_{{\mathrm{O}}_2},$$\end{document}−ΔGr0P0,T+ ∫P0PΔVP,TdPRT=lnXFeO1.5meltXFeOmelt+lnγFeO1.5meltγFeOmelt−14lnfO2,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta G_r^0\left( {P_0,T} \right)$$\end{document}ΔGr0P0,T is the free energy of the reaction (Eq.(1)) at reference pressure P0(1 bar) and temperature T, X and γ are the molar fractions and activity coefficients of the Fe-oxide component, respectively, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{{\mathrm{O}}_2}$$\end{document}fO2 is the oxygen fugacity, and R is the gas constant. The above equation has been widely used in many literatures3,12,13,23 and it suggests that the variation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{{\mathrm{O}}_2}$$\end{document}fO2 with pressure explicitly hinges on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV only. However, one should note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V(P,T)$$\end{document}ΔV(P,T) not only depends on pressure and temperature but also implicitly on many extensive properties, including the configuration entropy, and excess enthalpy. We first evaluate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta G_r^0\left( {P_0,T} \right)$$\end{document}ΔGr0P0,T for FeO1.5, FeO, and O2 as a function of temperature (Supplementary Fig. 5 and Supplementary Note 1). We then estimate the activity ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{ln}}\frac{{\gamma _{{\mathrm{FeO}}_{1.5}}^{{\mathrm{melt}}}}}{{\gamma _{{\mathrm{FeO}}}^{{\mathrm{melt}}}}}$$\end{document}lnγFeO1.5meltγFeOmelt by relating it to the interaction parameters between all the components following ref. 19. Moreover, the experimental results on ferric iron content (Fe3+/ΣFe) at various conditions (listed in Supplementary Table 2) are fit to the Eq. (3) to resolve the interaction parameters (Supplementary Table 3, Supplementary Fig. 6, Supplementary Note 2). We explore four different methods to fit the interaction parameters, but all models yield very similar redox profiles for MOs (Supplementary Fig. 7). We choose the one with smallest reduced chi-square as the best model and our predicted ferric iron contents (shown in Supplementary Fig. 8) are broadly consistent with the observations by both 1-bar experiments18–22 and the recent high-pressure experiments12,13,23 (Supplementary Note 3).The redox gradients in MOs of Earth, Mars, and the Moon are calculated using Eq. (3) along a cold thermal profile where 2100 K is assumed to be temperature at the surface (Fig. 2). Similar results are obtained for a hot geotherm with the surface temperature set at 2500 K (Supplementary Fig. 9).The uncertainties of all the parameters in Eq. (3) are propagated to calculate the oxygen fugacity using LMFIT package29. We use ΔV of 12.5 mol% Fe in silicate melts as a representative value for Earth30 and the Moon31 and that of 25 mol% Fe for Mars32. This assumption is justified for given mantle compositions of these three planetary bodies (Supplementary Table 4) because of relatively small effects of iron content on ΔV (Fig. 1). We quantify the redox states in terms of oxygen fugacity relative to IW buffers, that is, ΔIW = log\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{{\mathrm{O}}_{2}}$$\end{document}fO2 − IW, where the reference IW is taken from ref. 33. Since the temperatures considered are higher than the temperature at which this IW buffer is calibrated, we extrapolate this buffer equation to high temperatures12,13. We also assume that the bases of MOs are at depths of 55 GPa34, 14 GPa35, and 5 GPa36, and the corresponding redox states (ΔIW) are −2, −1.5, and −2 for Earth, Mars, and the Moon, respectively13. These redox values are representatives for terrestrial bodies when the molten iron ponds are assumed to be in local equilibrium with the overlaying MOs13. The pressures considered here are based on the single stage model and the complete equilibrium between the silicate melt and iron melt. More general consideration of magma ocean depths is discussed below.Fig. 2Redox profiles of magma oceans (MOs) for Earth, Mars, and the Moon.Redox state defined by ΔIW = log\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{{\mathrm{O}}_{2}}$$\end{document}fO2 – IW is shown as a function of pressure along a cold thermal profile.The MO bases are taken to be at depths corresponding to 55, 14, and 5 GPa with redox states (ΔIW) of −2, −1.5, and −2 for Earth, Mars, and the Moon, respectively. The 1σ standard deviation of the oxygen fugacity is ~0.5 log unit for this study (thick solid curves). The previous model results from ref. 12 (dashed-dotted curves) and ref. 23 (dotted curves) are also shown within their applicable ranges. For the model of ref. 13 (dashed curves), we follow the model to extrapolate to 15 and 25 GPa to predict the oxygen fugacities of the Martian and Earth’s MOs, respectively. The redox states of the present (upper) mantle of Earth, Mars, and the Moon are displayed in horizontal bars55. See Supplementary Fig. 9 for MO redox profiles along a hot thermal profile and Supplementary Fig. 3 for the thermal profiles.Along both thermal profiles considered, the absolute oxygen fugacities of the MOs of Earth, Mars, and the Moon all increase with depth, though more gradually at greater depths (Supplementary Fig. 9). This is expected because ΔV decreases with increasing pressure and always remains positive over the conditions we investigate. However, the relative oxygen fugacity (ΔIW) first increases slightly with pressure by ~0.3 log unit in the uppermost mantle and then gradually decreases with pressure throughout the rest of mantle (Fig. 2). Our results show that the upper mantle should have been relatively more oxidized. Therefore, an oxidized upper mantle is a natural consequence of a MO because the pressure- and temperature-dependent ΔV and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta G_r^0\left( {P_0,T} \right)$$\end{document}ΔGr0P0,T of the Eq. (3) make ferric iron increasingly stable at greater depths even at relatively reduced conditions (this raises Fe3+/ΣFe of silicate melts in equilibrium with metal alloy). Additionally, our derived redox profiles of Earth, Mars, and the Moon are nearly parallel owing to the similar ΔV values. They show that Earth is ~2 log units more oxidized than Mars which, in turn, is ~2 log units more oxidized than the Moon at the same depth. This order of relative redox states of the MOs of the early Earth, Mars, and the Moon coincides with that of their present-day mantles, implying a possible inheritance of present-day oxidation states of these planets from their early MOs.The comparison between the predicted redox profile of the MO with that of the present-day mantle for each planet informs us how the MO stage influences the subsequent redox evolution of each planet throughout its history. The oxidation state of the uppermost mantle of the present-day Earth is near IW + 3.537 and has remained constant within ∼1.0 log unit since at least the early Archean6,7. Our predicted redox state of the uppermost MO of Earth is at the lower bound of present-day values. Likewise, the ferric iron content corresponding to this redox profile is 1.0–3.5%, overlapping with the lower end of the present-day ferric iron abundances of the upper mantle37. The predicted redox state and ferric iron content suggest that Earth’s oxidizing uppermost mantle is a natural outcome of the thermodynamic equilibrium across the deep MO during the MO stage. Secondary contributions may arise from other mechanisms, including disproportionation of Fe2+ in the lower mantle by crystallization of bridgmanite38,39, and/or late accretion of oxidized materials40,41. Compared to the silicate Earth, the Martian uppermost mantle is less oxidized with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f_{{\mathrm{O}}_2}$$\end{document}fO2 ~IW11,42, which is consistent with our predicted redox state of the shallow Martian MO. This similarity may suggest negligible effects of subsequent tectonic processes and other oxidizing mechanisms mentioned above on the redox state of the Martian mantle43. Lunar basalts record oxygen fugacity ranging from IW to IW‒28–10 and our predicted redox state falls into the lower end of the observed values. Our predicted redox profiles differ considerably from those based on previous models (Fig. 2). Previous models have generally predicted relatively more reduced MOs of the Moon and Mars and either very reducing13 or very oxidizing MO of Earth12. It is important to note that the previously used data are limited with respect to pressure and temperature, for example, up to 3 GPa and 1673 K23, 7 GPa and 2023 K13, and 23 GPa and 2300 K12 (Supplementary Table 3).We also investigate the effects of varying depth of the MOs on the redox states of the surface and equilibrium ferric iron content (Fig. 3 and Supplementary Fig. 10). The redox states of the MO bases (ΔIW) are assumed to be fixed at −2, −1.5, −2, respectively, for Earth, Mars, and the Moon. A deeper MO generally shifts upwards its oxygen fugacity profile at shallower depths (Fig. 3a). The redox states and ferric iron contents of the Lunar and Martian MOs are marginally affected due to their small sizes. In contrast, the Earth’s MO may have reached 25–90 GPa based on moderately siderophile elements abundances, assuming models for single or multi-stage core formation with partial or complete equilibrium between impactor and proto Earth34,44. The oxygen fugacity of Earth’s surface would decrease by ~1.5 log units if the base of MO moved upwards from 55 GPa to 25 GPa. Concurrently, Fe3+/ΣFe would also drop by a factor of two. An even deeper magma ocean may induce the spin transition of iron in the silicate melts. However, the effect of the spin transition on the oxygen fugacity is shown to be insignificant within the MO thermal profiles considered here (Methods and Supplementary Fig. 11). Note our assumption that the ferric iron distribution is homogeneous within the MO due to vigorous convection. However, this ferric iron content profile likely evolves during the solidification of the MO. The evolution is controlled by how the MO crystallizes and the partitioning of iron species between the melt and crystal, which are still poorly constrained. Nevertheless, our study suggests that the whole mantles of Earth and Mars could have been as enriched in ferric iron as the present-day upper mantle since the MO stage.Fig. 3Redox state and ferric iron contents at the surfaces of magma oceans (MOs) of varying depths.The calculated relative redox state a and Fe3+/ΣFe ratio b of MOs of Earth (blue), Mars (red) and the Moon (green) as a function of pressuresat the base of MOs considering a cold thermal profile (Supplementary Note 2). Calculations are performed at the plausible pressure ranges of the bases of the MOs suggested by previous studies34–36,44. The redox state/ferric iron content of the present (upper) mantle of Earth (blue), Mars (red) and the Moon (green) are presented as horizontal bars55. The ferric iron contents of the mantle of Mars and the Moon are poorly constrained (not shown) because the available samples suffer from alterations and post-formation oxidations and cannot reflect the ferric iron contents of the source mantle56,57. The 1σ standard deviation is ~0.5 log unit for of the oxygen fugacity and ~0.03–0.06 for the ferric iron content.Chemistry of early atmospheresThe redox states of the MOs may have dictated the chemical speciation of the early atmospheres. For simplicity, we consider a case where the early atmosphere is at chemical equilibrium with the underlying MO3 and use the approach of ref. 45 to calculate the speciation of volatiles. Based on our results shown in Fig. 3, the redox state at the MO surface corresponds to ~IW + 2 for Earth, ~IW − 0.3 for Mars, and ~IW − 2 for the Moon. Assuming a simple C–O–H atmosphere with a mass H/C ratio of 0.5 at 1 bar and 1800 K, we show that the Earth’s early atmosphere would be enriched in H2O (~70 mol%) and CO2 (~15 mol%) but depleted in CO and H2. The early Martian atmosphere would consist of H2O and H2 in almost equal amounts (each ~40 mol%), ~15 mol% CO, and ~5 mol% CO2. In contrast, the early lunar atmosphere would be enriched in H2 (>70 mol%) and CO (~20 mol%) and relatively depleted in H2O (10 mol%)3,45 (Fig. 4). These early atmospheres further evolve as the planets cool down. The speciation and mass of the atmosphere would likely change over time due to the thermodynamic re-equilibrium, hydrodynamic loss, as well as subsequent degassing and ingassing/regassing. Nevertheless, these distinct early atmospheric compositions may have fundamentally influenced the subsequent evolution of these terrestrial planets, including climate, magma ocean solidification, and the evolution of surficial conditions3,46.Fig. 4Inferred compositions of early atmospheres.The redox states of magma oceans and the dominant chemical speciation of the overlying atmospheres are shown for Earth, Mars and the Moon. Refer to the text for estimated fraction of each species. The thicknesses and oxidation states of atmospheres are not scaled.The vertical gradient in the MO redox state predicted here may also apply to other rocky planets where MOs were once formed. For example, Earth’s sister planet Venus is of similar size and has similar iron content. The redox state of the post-MO upper mantle of Venus, to first order, may be similar to that of Earth and could be tested by future Venusian missions. In addition, super-Earths close to their host stars may have MOs extended to various depths and their atmospheres can potentially be detected in the near future with space telescope missions47.MethodsComputational detailsFirst-principles molecular dynamics (FPMD) simulations were carried out using the VASP software48 in the NVT-canonical ensemble with temperature controlled by a Nosé thermostat49. The projector augmented wave potentials50,51 were employed together with the generalized gradient approximation (GGA) to the exchange-correlation potential52. The plane-wave basis cutoff was set at 400 eV and Brillouin zone sampling was performed at the Gamma point. Pulay stress corrections were applied to the calculated pressures. We simulated Mg14Fe2Si16O48(ferrous) and Mg14Fe2Si16O49(ferric) melts for 12.5 mol% iron and Mg12Fe4Si16O48(ferrous) and Mg12Fe4Si16O50(ferric) melts for 25 mol% iron. Iron was set to be in high-spin state in all simulations. At each volume, the system was initially melted and thermalized at 6000 K, and then subsequently quenched down to desired lower temperatures of 4000, 3000, 2500, and 2000 K. Simulations were run for 10–30 picoseconds with time step of 1 femtosecond. Further details can be found in ref. 53.Calculation of volume difference (ΔV)The molar volume difference between FeO1.5 and FeO defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{\Delta }}V = V_{{\mathrm{FeO}}_{1.5}} - V_{{\mathrm{FeO}}}$$\end{document}ΔV=VFeO1.5−VFeO is calculated as the volume difference between the ferric and ferrous iron-bearing silicate melts. Take the silicate melt with 12.5 mol% iron as an example. The volumes of Mg14Fe2Si16O48 and Mg14Fe2Si16O49 melts referred to as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{\mathrm{Mg}}_{14}{\mathrm{Fe}}_2{\mathrm{Si}}_{16}{\mathrm{O}}_{48}}$$\end{document}VMg14Fe2Si16O48 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{\mathrm{Mg}}_{14}{\mathrm{Fe}}_2{\mathrm{Si}}_{16}{\mathrm{O}}_{49}}$$\end{document}VMg14Fe2Si16O49, respectively, are calculated at the same pressure and temperature conditions using the resolved equation of state parameters (Supplementary Table 1). These volumes can be related to the partial volumes of components by:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{\mathrm{Mg}}_{14}{\mathrm{Fe}}_2{\mathrm{Si}}_{16}{\mathrm{O}}_{48}} = 14V_{{\mathrm{MgSiO}}_3} + 2V_{{\mathrm{FeO}}} + 2V_{{\mathrm{SiO}}_2} + V^{{\mathrm{E}},{\mathrm{reduced}}}$$\end{document}VMg14Fe2Si16O48=14VMgSiO3+2VFeO+2VSiO2+VE,reducedand5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{\mathrm{Mg}}_{14}{\mathrm{Fe}}_2{\mathrm{Si}}_{16}{\mathrm{O}}_{49}} = 14V_{{\mathrm{MgSiO}}_3} + 2V_{{\mathrm{FeO}}_{1.5}} + 2V_{{\mathrm{SiO}}_2} + V^{{\mathrm{E}},{\mathrm{oxidized}}},$$\end{document}VMg14Fe2Si16O49=14VMgSiO3+2VFeO1.5+2VSiO2+VE,oxidized,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{{\mathrm{E}},{\mathrm{reduced}}}$$\end{document}VE,reduced and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{{\mathrm{E}},{\mathrm{oxidized}}}$$\end{document}VE,oxidized are the excess volumes for reduced and oxidized systems, respectively, and are sensitive to the amount of iron. Several previous low-pressure experiments show that the excess terms are small for silicate melts if Na2O, Al2O3, and TiO2 components are absent18,24. In this case, we can approximate ΔV by6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{\Delta }}V = V_{{\mathrm{FeO}}_{1.5}} - V_{{\mathrm{FeO}}} \approx \left( {V_{{\mathrm{Mg}}_{14}{\mathrm{Fe}}_2{\mathrm{Si}}_{16}{\mathrm{O}}_{49}} - V_{{\mathrm{Mg}}_{14}{\mathrm{Fe}}_2{\mathrm{Si}}_{16}{\mathrm{O}}_{48}}} \right)/2$$\end{document}ΔV=VFeO1.5−VFeO≈VMg14Fe2Si16O49−VMg14Fe2Si16O48/2Similarly, for 25 mol% iron content, we use7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{\Delta }}V = V_{{\mathrm{FeO}}_{1.5}} - V_{{\mathrm{FeO}}} \approx \left( {V_{{\mathrm{Mg}}_{12}{\mathrm{Fe}}_4{\mathrm{Si}}_{16}{\mathrm{O}}_{50}} - V_{{\mathrm{Mg}}_{12}{\mathrm{Fe}}_4{\mathrm{Si}}_{16}{\mathrm{O}}_{48}}} \right)/4.$$\end{document}ΔV=VFeO1.5−VFeO≈VMg12Fe4Si16O50−VMg12Fe4Si16O48/4.By using the above equation to calculate ΔV, we assume that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{{\mathrm{E}},{\mathrm{oxidized}}}$$\end{document}VE,oxidized and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{{\mathrm{E}},{\mathrm{reduced}}}$$\end{document}VE,reduced take small similar values so \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{{\mathrm{E}},{\mathrm{oxidized}}} - V^{{\mathrm{E}},{\mathrm{reduced}}} \approx 0$$\end{document}VE,oxidized−VE,reduced≈0. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{{\mathrm{E}},{\mathrm{oxidized}}} - V^{{\mathrm{E}},{\mathrm{reduced}}}$$\end{document}VE,oxidized−VE,reduced is a large non-zero value, one would expect that the ΔV differs significantly between the two compositions considered (12.5 and 25 mol% iron in silicate melts). However, our calculated results show that the ΔV values of 12.5 and 25 mol% iron contents differ slightly from each other and the difference diminishes especially at high pressures, which justifies our assumptions.It should be noted that the small excess volume is not conflicted with the large Margules interaction parameters resolved for silicate melts. The excess volume is thermodynamically defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^{\mathrm{E}} = \left( {\frac{{\partial G_{{\mathrm{mix}}}}}{{\partial {\mathrm{P}}}}} \right)_{\mathrm{T}} = \left( {\frac{{\partial H_{{\mathrm{mix}}}}}{{\partial {\mathrm{P}}}}} \right)_{\mathrm{T}}$$\end{document}VE=∂Gmix∂PT=∂Hmix∂PT, where Gmix and Hmix are the Gibbs free energy and enthalpy of mixing, respectively; P is pressure; and T is temperature. Hmix is a function of interaction parameter (W) and composition54. For a binary system with endmember components A and B, Hmix = WXAXB, where XA and XB are the molar fractions of A and B, respectively. Therefore, a small VE requires that the pressure derivative of the interaction parameter to be small but does not necessarily mean that the value of W is small. Indeed, both in this study and many other studies13,23, W is assumed to be pressure independent, which is in line with the assumption that VE is small.Effects of spin transition of iron on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV and oxygen fugacityBoth ferric and ferrous irons undergo electronic spin transitions at high pressure as predicted by a recent FPMD study53. The high- to low-spin transition of Fe3+ and Fe2+ occurs gradually over pressure intervals centered around 90 and 110 GPa, respectively, at 3000 K. These transition pressures are higher than the maximum pressures of the magma oceans considered in this study (Fig. 2). As both Fe3+ and Fe2+ will be mostly in high-spin (HS) state at relevant magma ocean pressures, we evaluate the volume difference between FeO1.5 and FeO as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V\; = \;V_{{\mathrm{HS}}}^{{\mathrm{FeO}}_{1.5}}\; -\; V_{{\mathrm{HS}}}^{{\mathrm{FeO}}}$$\end{document}ΔV=VHSFeO1.5−VHSFeO. However, all Fe3+ and Fe2+ will not undergo the HS-LS transition at a given condition. This means that the spin transition-induced changes in volume also contribute to our ΔV evaluation. We assess the spin effects on ΔV using the spin phase diagrams from Karki et al.53. Considering exact HS and LS distributions for both ferrous and ferric irons, we can evaluate the volume difference between FeO1.5 and FeO as8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V_{{\mathrm{exact}}} \;= \;(V_{{\mathrm{HS}}}^{{\mathrm{FeO}}_{1.5}} \;-\; V_{{\mathrm{HS}}}^{{\mathrm{FeO}}})\; -\; n_{{\mathrm{LS}}}^{{\mathrm{Fe}}^{3 + }}(V_{{\mathrm{HS}}}^{{\mathrm{FeO}}_{1.5}} \;-\; V_{{\mathrm{LS}}}^{{\mathrm{FeO}}_{1.5}}) \\ + \;n_{{\mathrm{LS}}}^{{\mathrm{Fe}}^{2 + }}(V_{{\mathrm{HS}}}^{{\mathrm{FeO}}} \;-\; V_{{\mathrm{LS}}}^{{\mathrm{FeO}}}),$$\end{document}ΔVexact=(VHSFeO1.5−VHSFeO)−nLSFe3+(VHSFeO1.5−VLSFeO1.5)+nLSFe2+(VHSFeO−VLSFeO),where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{\mathrm{HS}}}^{{\mathrm{FeO}}_{1.5}} - V_{{\mathrm{HS}}}^{{\mathrm{FeO}}} = \Delta V$$\end{document}VHSFeO1.5−VHSFeO=ΔV has been rigorously constrained in this study. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{{\mathrm{LS}}}^{{\mathrm{Fe}}^{3 + }}$$\end{document}nLSFe3+ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{{\mathrm{LS}}}^{{\mathrm{Fe}}^{2 + }}$$\end{document}nLSFe2+ represent the fractions of Fe3+and Fe2+ in low-spin (LS) state, respectively (satisfying the relations \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{{\mathrm{HS}}}^{{\mathrm{Fe}}^{3 + }} + n_{{\mathrm{LS}}}^{{\mathrm{Fe}}^{3 + }} = n_{{\mathrm{HS}}}^{{\mathrm{Fe}}^{2 + }} + n_{{\mathrm{LS}}}^{{\mathrm{Fe}}^{2 + }} = 1$$\end{document}nHSFe3++nLSFe3+=nHSFe2++nLSFe2+=1, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{{\mathrm{HS}}}^{{\mathrm{Fe}}^{3 + }}$$\end{document}nHSFe3+ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{{\mathrm{HS}}}^{{\mathrm{Fe}}^{2 + }}$$\end{document}nHSFe2+ represent the corresponding HS fractions) and their values as a function of pressure and temperature for silicate melt with 25% Fe can be found in ref. 53. Karki et al.53 also evaluated the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{\mathrm{HS}}}^{{\mathrm{FeO}}_{1.5}} - V_{{\mathrm{LS}}}^{{\mathrm{FeO}}_{1.5}}$$\end{document}VHSFeO1.5−VLSFeO1.5 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{\mathrm{HS}}}^{{\mathrm{FeO}}} - V_{{\mathrm{LS}}}^{{\mathrm{FeO}}}$$\end{document}VHSFeO−VLSFeO to be constant with respect to pressure within the computational uncertainties. At 3000 and 4000 K, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{\mathrm{HS}}}^{{\mathrm{FeO}}_{1.5}} - V_{{\mathrm{LS}}}^{{\mathrm{FeO}}_{1.5}} \approx$$\end{document}VHSFeO1.5−VLSFeO1.5≈1.25 cm3 mol−1and 1.00 cm3mol−1, respectively, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{{\mathrm{HS}}}^{{\mathrm{FeO}}} - V_{{\mathrm{LS}}}^{{\mathrm{FeO}}} \approx$$\end{document}VHSFeO−VLSFeO≈ 1.75 cm3 mol‒1 and 1.10 cm3 mol‒1, respectively53. We calculate the difference of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V_{{\mathrm{exact}}}$$\end{document}ΔVexact and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV at 3000 and 4000 K as well as the difference of the oxygen fugacity using these two volume differences (Supplementary Fig. 11).At pressures less than 60 GPa, we find that the deviation of the volume difference caused by considering the spin transition is less than 3%, so the oxygen fugacity does not change much when spin effects are included (Supplementary Fig. 11). With increasing pressure, the magnitude of (ΔVexact − ΔV) further increases and bounces back at around 100 GPa, at which the fraction of low-spin Fe3+ reaches around 50%. Note that at pressures greater than 80 GPa, the temperature of the MO is around 3500 K for a cold thermal profile and continues increasing with pressure. Therefore, the results at 4000 K are more relevant at these pressures. Overall, neglecting the spin transition of Fe tends to overestimate the oxygen fugacity, especially at high pressures. The maximum deviation occurs around 120 GPa, which is ~0.6 log units, comparable to the uncertainties of our model prediction (~0.5). Therefore, we consider the effects of spin transition of iron on the redox state of MOs to be mostly insignificant.Supplementary information
Supplementary Information
Peer Review File | nature communications | [
"Article"
] | [
"Early solar system",
"Geochemistry",
"Inner planets",
"Geochemistry"
] | redox condition planetary bodies influences chemical differentiation governs composition atmospheres1–5 understand bio-essential volatiles carbon hydrogen incorporated near Earth’s surface requires knowledge redox state early stages history studies shown uppermost mantle Earth oxidized (IW + 3.5 3.5 log units above iron-wüstite buffer)6 Petrological evidence suggests oxidized conditions formed early 4.3–4.4 Ga ago7 Martian Lunar mantles more reduced (~IW − 1)8–11 contrasting oxidization states may have set up during early planetary formation magma oceans (MOs) could existed3 MO involved mechanism established redox controlling reactions poorly constrained in magma ocean scenarios studies inferred oxygen fugacity profile of silicate melts at high pressures applicable for shallow magma oceans13 oxygen fugacity function of pressure temperature composition likely varying within MOsstudy redox reactions in magma oceans simulating silicate melts ferrous ferric iron with-principles molecular dynamics thermodynamic modeling at pressures temperatures up to 5000 K results suggest ferric iron energetically favorable with pressure small partial molar volume in silicate melts under large compression magma oceans of Earth Mars Moon homogeneous vertical gradient in oxygen fugacity deeper magma ocean early Earth more oxidizing surface conditions Mars Moon contrasting conditions suggest early atmosphere dominated by CO2 H2O enriched H2O H2 Mars H2 CO Moon discussionEquations of state silicate meltsAt metallic melts pond before oxygen fugacity governed by between metallic silicate melts calculated compositionsbase absent rapid sinking velocity16 MO redox state controlled by redox buffering reaction3[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}\mathrm{FeO}}{melt}FeOmelt+14O2=FeO1.5meltThe thermodynamic behavior reaction informs oxygen fugacity varies with temperature pressure oxygen fugacity MO base condition obtain oxygen fugacity throughout MO thermodynamic properties reactants products Eq. 1 known key-parameter difference in molar volumes between FeO1.5 FeO in melts ΔV measured at 1 bar18 inferred from experiments up to 23 GPa ~2500 K12 conditions far expected in MOs Earth Mars goal to calculate ΔV function of pressure temperature composition constrain oxygen fugacity in redox buffering Eq. (1) conditions evaluate MO redox states of Earth Mars Moon understand oxidation conditions present-day mantle chemistry earliest atmospherepresent results FPMD simulations iron-bearing MgSiO3 liquids valence states 2000–4000 K 140 GPa calculated pressure–volume–temperature (P–V–T) relationships described equation\documentclass[12pt]{minimal{amsmath{upgreek\oddsidemargin-69pt}}$$P(V,T) P\left {V,T_0}\right + B_{TH}(T - T_0){document}P(V,T)=PV,T0+BTH[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin}-69pt}$P(V,T_0)}P(V,T0) represents reference isotherm at T0 = 3000 K fourth-order Birch-Murnaghan equation statesecond term contains thermal pressure coefficient[12pt{minimal\usepackage{amsmath\oddsidemargin-69pt}}\mathrm{TH\left V{V_0^2{document}BTHV=a−bVV0+cVV02/1000 a b c constants for melt composition bulk moduli Fe2+ -bearing melts larger than Fe3+ bearing melts Fe3+ melts more compressible at conditions investigated (Fig. 1) consistent low-pressure 1Molar volume difference (ΔV) between FeO1.5 and FeO in silicate melts ΔV function of pressure at different temperatures 12.5 mol% iron Earth Moon-like magma ocean 25 mol% iron Mars-like magma ocean show pressure–volume relationships melts 12.5 25 mol% iron Fe2+ Fe3+ Volumes plotted isotherms pressures liquid state 1σ standard deviation of ΔV ~0.2–0.5 cm3 mol−1.pressure–volume results silicate melts molar content Fe3+ Fe2+ calculate difference (ΔV between FeO1.5 FeO function pressure (Fig. 1) calculated ΔV at zero pressure agrees with experimental data18 pressure increases ΔV decreases low-pressure increases decreases higher pressures trend weakens at higher temperatures silicate melts different iron contents 12.5 25 mol%) ΔV different values weak positive trend with iron content consistent with dependency ΔV on composition18 results show ΔV remains positive at pressures up to 140 GPa irrespective of temperature composition contradicts inferences ΔV negative pressure range Earth’s mantle3 models on ΔV adopt bulk modulus derivative 4 equation state fit data limited pressure compare model values with calculated results for silicate melts 12.5 mol% iron low pressures (<10 results with recent model sharp decrease ΔV low pressures higher pressures results models13 ΔV levels out ΔV differences due to different pressure dependencies FeO1.5 FeO silicate melts4th order Birch-Murnaghan fit yields larger Kʼ FeO1.5 (4.6) FeO (3.3). Previous studies assume Kʼ FeO1.5 FeO 4 similar pressure dependency ΔV contrasting behavior ΔV 12 caused by different Kʼ values 1.3 8 for FeO1.5 FeO extreme values Kʼ consistent with studies silicate melts Kʼ 3–825,26 FeO liquid Kʼ 3–427,28 reason inconsistency unclear analysis shows Fe–O coordination increases pressure pressures beyond 40 GPa pressure trend similar to ΔV implies correlation between iron-oxygen bonding silicate ΔV first-principles results no assumption on value Kʼ FeO1.5 FeO applicable over mantle regime Earth evaluate ΔV along two magma ocean thermal profiles calculated ΔV varies remains positive over pressure temperature magma oceanindicating positive contribution pressure[12pt]{minimal}{amsmath}{wasysym{upgreek\oddsidemargin-69pt}{document}$$f\mathrm{O}}_2\end{document}fO2.Redox calculated ΔV profiles magma ocean thermal profiles assess redox state magma oceans Earth Mars Moon MOs convective well-mixed homogeneous Fe3+ total Fe ratio (Fe3+/ΣFe)3.thermodynamic relationship reaction (1)[12pt{amsmath\oddsidemargin-69pt}{document}$$-\Delta G_r^0\left( {P_0,T} \right)\mathop\smallint\nolimits_{P_0}^P \Delta V\left( {P,T} \right)dP}}{{RT}} = \mathrm{ln}}\frac{{X_{{\mathrm{FeO}}_{1.5}}{melt\mathrm{ln}}\gamma\mathrm{FeO}}_{1.5}}{melt\frac{1}{4}{\mathrm{ln}}\mathrm{O}}_2}$$\end{document}−ΔGr0P0,T+ ∫P0PΔVP,TdPRT=lnXFeO1.5meltXFeOmelt+lnγFeO1.5meltγFeOmelt−14lnfO2[12pt]{minimal}{amsmath{wasysymgreek\setlength-69pt}\Delta G_r^0\left {P_0,T}\end{document}ΔGr0P0,T free energy reaction (Eq.(1) pressure P0(1 bar) temperature T X γ molar fractions activity coefficients Fe-oxide component[12pt]{minimal\usepackage{amsmath}{wasysym{mathrsfs}{upgreek-69pt}\mathrm{O}}_2\end{document}fO2 oxygen fugacity R gas constantequation used in literatures3,12,13,23 suggests variation of\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{amssymb{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document$$f\mathrm{O}}_2}{document}fO2 pressure hinges on\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek{\oddsidemargin}{-69pt}{document$\Delta V$$\end{document}ΔV only\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amsbsy{mathrsfs{upgreek{\oddsidemargin}{-69pt}{document}$\Delta V(P,T){document}ΔV(P) depends on pressure temperature extensive properties including configuration entropy excess enthalpyevaluate\documentclass[12pt{minimal{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}\Delta G_r^0\left_0,T for FeO1.5 FeO O2 function temperature (Supplementary Fig. 5 Note 1) estimate activity ratio[12pt{minimal{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}\mathrm{ln}}{FeO}}}lnγFeO1.5meltγFeOmelt relating interaction parameters between components ref. 19. experimental results on ferric iron content (Fe3+/ΣFe) conditions Supplementary Table 2) fit Eq. (3) resolve interaction parameters (Supplementary Table 3 Supplementary Fig. 6 Note 2) explore four methods fit interaction parameters models yield similar redox profiles for MOs (Supplementary Fig. 7)choose smallest reduced chi-square best model predicted ferric iron contents Supplementary Fig. 8) consistent with 1-bar experiments18–22 recent high-pressure experiments12,13,23 Note 3) redox gradients Earth Mars Moon calculated Eq. (3) cold thermal profile 2100 K assumed temperature surface (Fig. 2) Similar results hot geotherm surface temperature 2500 K Fig. 9) uncertainties parameters Eq. (3) propagated calculate oxygen fugacity LMFIT package29 use ΔV 12.5 mol% Fe silicate melts representative Earth30 Moon31 25 mol% Fe Mars32 assumption justified mantle compositions small effects iron content on ΔV (Fig. 1) quantify redox states oxygen fugacity relative IW buffers ΔIW = log\documentclass IW reference IW ref. 33. temperatures considered higher than IW buffer calibrated extrapolate buffer equation to high temperatures12,13.assume bases MOs at depths 55 GPa34 14 GPa35 5 GPa36 redox states) −2 −1.5 −2 for Earth Mars Moon redox values representatives for terrestrial bodies molten iron ponds local equilibrium with overlaying MOs13 pressures based on single stage model equilibrium between silicate melt iron melt general consideration magma ocean depths discussed.Fig. 2Redox profiles of magma oceans for Earth Mars Moon.Redox state defined by ΔIW = log IW function of pressure along cold thermal profile MO bases at depths 55 14 5 GPa with redox states −2, −1.5 −2 for Earth Mars Moon 1σ standard deviation oxygen fugacity ~0.5 log unit study previous model results from ref. 12 23 within applicable ranges model ref. 13 extrapolate to 15 25 GPa predict oxygen fugacities of Martian Earth’s MOsredox states (upper mantle Earth Mars Moon displayed in horizontal bars55 See Supplementary Fig. 9 MO redox profiles hot thermal profile Supplementary Fig. 3 thermal profiles absolute oxygen fugacities MOs Earth Mars Moon increase with gradually at greater depths 9) expected ΔV decreases with pressure remains positive conditions relative oxygen fugacity (ΔIW) increases slightly with pressure ~0.3 log unit uppermost mantle decreases with pressure rest mantle (Fig. 2) results show upper mantle should more oxidized oxidized upper mantle natural consequence of MO pressure- temperature-dependent ΔV ferric iron stable at greater depths reduced conditions raises Fe3+/ΣFe silicate melts equilibrium with metal alloy). derived redox profiles of Earth Mars Moon nearly parallel similar ΔV values Earth ~2 log units more oxidized than Mars ~2 more oxidized Moon at same depthorder redox states MOs early Earth Mars Moon coincides with present-day mantles possible inheritance oxidation states from early MOs comparison between predicted redox profile MO present-day mantle each planet informs MO stage redox evolution oxidation state uppermost mantle Earth near IW + 3.537 constant within ∼1.0 log unit since early Archean6,7 predicted redox state uppermost MO lower bound of present-day values ferric iron content 1.0–3.5%, overlapping with lower end present-day ferric iron abundances upper mantle37 predicted redox state ferric iron content suggest Earth’s oxidizing uppermost mantle natural outcome of thermodynamic equilibrium deep MO during MO stage Secondary contributions may from disproportionation of Fe2+ lower mantle by crystallization bridgmanite38 late accretion of oxidized materials40 Martian uppermost mantle less oxidized,42 consistent with predicted redox state shallow Martian MOsimilarity negligible effects tectonic processes oxidizing mechanisms on redox state Martian Lunar basalts record oxygen fugacity IW to IW‒28–10 predicted redox state lower end values predicted redox profiles differ from previous models (Fig. 2) Previous models predicted reduced MOs Moon Mars or oxidizing MO data limited pressure temperature 3 GPa 1673 K23 7 GPa 2023 K13 23 GPa 2300 K12 effects varying MOs on redox states equilibrium ferric iron content redox states MO bases fixed at −2 −1.5 −2 for Earth Mars Moon deeper MO shifts upwards oxygen fugacity at shallower depths. redox states ferric iron contents Lunar Martian MOs marginally affected small sizes Earth’s MO 25–90 GPa siderophile elements abundances core formation oxygen fugacity surface decrease ~1.5 log units if base MO moved from 55 GPa to 25 GPa Fe3+/ΣFe drop factor two deeper magma ocean induce spin transition iron silicate melts effect spin transition on oxygen fugacity insignificant within MO thermal profilesassumption ferric iron distribution homogeneous within MO due to convection profile evolves during solidification MO evolution controlled by MO partitioning iron species between melt crystal study suggests mantles Earth Mars enriched in ferric iron upper mantle since MO stage.Fig. 3Redox state ferric iron contents surfaces magma oceans varying calculated relative redox state Fe3+/ΣFe ratio of MOs Earth Mars Moon function cold thermal profile Calculations pressure ranges redox state/ferric iron content mantle Earth Mars Moon as horizontal bars55 ferric iron contents mantle Mars Moon poorly constrained samples suffer alterations post-formation oxidations reflect ferric iron contents source 1σ standard deviation ~0.5 log unit for oxygen fugacity ~0.03–0.06 ferric iron content.Chemistry of early redox states may dictated chemical speciation early atmospheres early atmosphere chemical equilibrium with underlying MO3 approach calculate speciation of volatiles results Fig. 3 redox state at MO surface corresponds to ~IW + 2 for Earth − 0.3 for Mars − 2 for MoonAssuming C–O–H atmosphere mass H/C ratio 0.5 1 bar 1800 K Earth’s early atmosphere enriched H2O~70 mol%) CO2 (~15 mol%) depleted CO H2. Martian atmosphere H2O H2 equal ~40 mol%) ~15 mol% CO ~5 mol% CO2. lunar atmosphere enriched H2>70 mol%) CO (~20 mol%) depleted H2O (10 (Fig. 4) early atmospheres evolve planets cool speciation mass change thermodynamic re-equilibrium hydrodynamic loss degassing ingassing/regassing early atmospheric compositions influenced evolution climate magma ocean solidification evolution surficial. compositions early atmospheres redox states magma oceans dominant chemical speciation overlying atmospheres Earth Mars Moon estimated thicknesses oxidation states not scaled vertical gradient MO redox state apply other rocky planets Earth’s sister planet Venus similar size iron content redox state post-MO upper mantle Venus similar Earth tested future Venusian missions super-Earths close stars MOs extended atmospheres detected space telescope missions47detailsFirst-principles molecular dynamics simulations VASP NVT-canonical ensemble temperature controlled Nosé projector augmented wave potentials50 generalized gradient approximation exchange-correlation plane-wave cutoff 400 eV Brillouin zone sampling Gamma point Pulay stress corrections calculated pressures simulated Mg14Fe2Si16O48 Mg14Fe2Si16O49(ferric) melts 12.5 mol% iron melts 25 mol% iron Iron high-spin state system melted thermalized at 6000 K quenched temperatures 4000 3000 2500 2000 K Simulations 10–30 picoseconds time step 1 femtosecond details ref. 53.Calculation volume difference (ΔV molar volume difference between FeO1.5 FeO defined[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document} ={1.5}} -{document}ΔV=VFeO1.5−VFeO calculated volume difference ferric ferrous iron-bearing silicate melts silicate melt 12.5 mol% iron examplevolumes Mg14Fe2Si16O48 melts\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document\mathrm{Mg}}{14{Fe}}{Si}}_{16{O}}_{48}}{document}VMg14Fe2Si16O48[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{upgreek}\oddsidemargin}{-69pt}{document$V\mathrm{Mg}}_{14{Fe}}_2{Si}}_{16}{O}}_{49}}{document}VMg14Fe2Si16O49 calculated same pressure temperature conditions resolved equation state parameters (Supplementary Table 1)volumes related partial volumes components[12pt]{amsmath-69pt$V\mathrm{Mg}}_{14{Fe}}_2{48}} = 14V_{{{MgSiO}}_3} + 2V{FeO + 2V{SiO}}_2} + V\mathrm{E}}{reduced{document}VMg14Fe2Si16O48=14VMgSiO3+2VFeO+2VSiO2+VE[12pt]{minimal}{amsmath}{amsbsy{mathrsfs}{upgreek-69pt}$V\mathrm{Mg}}{14{Fe}}_2{Si}}_{16{O}}_{49}} = 14V_{{{MgSiO}}_3} + 2V_{{{FeO}}_{1.5}} + 2V{SiO}}_2} + V^{{\mathrm{E}}{oxidized}}}{document}VMg14Fe2Si16O49=14VMgSiO3+2VFeO1.5+2VSiO2+VE,oxidized[12pt]{minimal}{amsmath}{wasysym}\oddsidemargin}{-69pt}{document}$$V^{{\mathrm{E}}{reduced}}}{document}VE,reduced[12pt]{minimal}{amsmath}{wasysym}}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$$V^{{\mathrm{E}}{oxidized{document}VE,oxidized excess volumes for reduced oxidized systems sensitive to ironprevious low-pressure experiments show excess terms small for silicate melts if Na2O Al2O3 TiO2 absent18 approximate ΔV by6\documentclass[12pt]{minimal}\usepackage{amsmath-69pt}\mathrm\Delta }}V = V_{{\mathrm{FeO}}_{1.5}} - V\left {V\mathrm{Mg}}_{49}} - V{48}}}{document}ΔV=VFeO1.5−VFeO≈VMg14Fe2Si16O49−VMg14Fe2Si16O48/2Similarly for 25 mol% iron content use7\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek-69pt\mathrm{\Delta }}V = V_{{\mathrm{FeO}}_{1.5}} - V_{{\mathrm{FeO}}} {V\mathrm{Mg}}{12{16{50}} - V{16{48.\end{document}ΔV=VFeO1.5−VFeO≈VMg12Fe4Si16O50−VMg12Fe4Si16O48/4.equation calculate ΔV assume\documentclass[12pt]{minimal}{amsmath}{wasysym{upgreek}\oddsidemargin}{-69pt}{document}$$V^{{\mathrm{E}}\mathrm{oxidized}}}\end{document}VE,oxidized[12pt{minimal{amsmath{wasysym}{upgreek}\oddsidemargin}{-69pt}{document}$$V^{{\mathrm{E}}\mathrm{reduced}}}{document}VE,reduced similar values[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$V^{{\mathrm{E}}\mathrm{oxidized}}}\mathrm{E}}\mathrm{reduced}}} 0\end{document}VE,oxidized−VE,reduced≈0.\documentclass[12pt{minimal\usepackage{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}{document}\mathrm{E}}{oxidized{reduced{document,reduced large non-zero value ΔV differs compositions (12.5 25 mol% iron in silicate melts). calculated results show ΔV values 12.5 25 mol% iron differ slightly difference diminishes at high pressures justifies assumptions small excess volume not conflicted with large Margules interaction parameters silicate meltsexcess volume thermodynamically defined as\documentclass[12pt{minimal}\usepackage{amsmath{wasysym{amsfonts{amssymb{mathrsfs{upgreek\setlength\oddsidemargin-69pt}\begin{document}\mathrm{E}}\left G{mix{P{T}}\left H{mix{P{T}}\end{document}VE=∂Gmix∂PT=∂Hmix∂PT Gmix Hmix Gibbs free energy enthalpy of mixing P pressure T temperature Hmix function of interaction parameter (W) composition54 binary system with endmember components A B Hmix = WXAXB XA XB molar fractions of A and B small VE requires pressure derivative interaction parameter small not value W small study W assumed pressure independent with assumption VE smallEffects spin transition iron\documentclass[12pt]{minimal{amsmath{wasysym{amsfonts{mathrsfs{upgreek-69pt\Delta}ΔV oxygen ferric ferrous irons undergo spin transitions high pressure predicted FPMD study53 high- to low-spin transition Fe3+ Fe2+ occurs over pressure intervals 90 110 GPa at 3000 K transition pressures higher than maximum pressures magma oceans study (Fig. 2) Fe3+ Fe2+ high-spin (HS) state at magma ocean pressures evaluate volume difference between FeO1.5 and FeO[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek-69pt\Delta V\mathrm{HS}}}{FeO}}{FeO{document}ΔV=VHSFeO1.5−VHSFeO all Fe3+ Fe2+ undergo HS-LS transition condition spin transition-induced changes volume contribute to ΔV evaluationassess spin effects ΔV phase diagrams KarkiHS LS distributions ferrous ferric irons evaluate volume difference between FeO1.5 FeO as8[12pt{minimal}{amsmath\oddsidemargin-69pt}{document}\Delta V_{{\mathrm{exact}}}\mathrm{HS}}}\mathrm{FeO}}_{1.5}}{HS{FeO\mathrm{LS}}}\mathrm{Fe}}^{3 + }}{HS}}}{FeO}}_{1.5}}\mathrm{LS}}}{FeO}}_{1.5}}\mathrm{LS}}}\mathrm{Fe}}^{2 + }}\mathrm{HS}}}{FeO}}}\mathrm{LS}}}{FeO\end{document}ΔVexact=(VHSFeO1.5−VHSFeO)−nLSFe3+(VHSFeO1.5−VLSFeO1.5)+nLSFe2+(VHSFeO−VLSFeO),where[12pt]{minimal}\usepackage{amsmath}}usepackage{mathrsfs{upgreek\setlength-69pt{document}\mathrm{HS{FeO}}{1.5}} - V =\Delta{document}VHSFeO1.5−VHSFeO=ΔV constrained study[12pt]{minimal{amsmath{wasysym{upgreek}\oddsidemargin}{-69pt}{document}\mathrm{LS}}}{Fe}}^{3 + }}{document}nLSFe3+[12pt{minimal{amsmath}{upgreek}\oddsidemargin}{-69pt}\mathrm{LS}}}{Fe}}^{2 + }}{document}nLSFe2+ fractions of Fe3+and Fe2+ in low-spin (LS) state[12pt]{minimal}{amsmath}{wasysym}}}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}\mathrm{HS}}}\mathrm{Fe}}^{3 + }} +{LS}}}\mathrm{Fe}}^{3 + }} =\mathrm{HS}}}\mathrm{Fe}}^{2 + }} + n\mathrm{LS}}}\mathrm{Fe}}{2 + }} = 1{document}nHSFe3++nLSFe3+=nHSFe2++nLSFe2+=1[12pt]{minimal{amsmath\oddsidemargin{-69pt}{document}\mathrm{HS}}}\mathrm{Fe}}^{3 + }}\end{document}nHSFe3+[12pt]{minimal}{amsmath}{upgreek}\oddsidemargin}{-69pt}{document}\mathrm{HS}}}\mathrm{Fe}}^{2 + }}{document}nHSFe2+ represent HS fractions values function pressure temperature silicate melt with 25% Fe ref. 53. Karki et al.evaluated[12pt]{minimal{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document}\mathrm{HS{FeO}}_{1.5}} -{LS}}}{FeO}}{1.5}}\end{document}VHSFeO1.5−VLSFeO1.5[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}{amsbsy}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}\mathrm{HS}}}\mathrm{FeO}}} -\mathrm{LS}}}{FeO\end{document}VHSFeO−VLSFeO constant computational uncertainties3000 4000 K[12pt\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}\mathrm{HS{FeO}}{1.5}}{LS}}}{FeO}}{1.5}}{document}VHSFeO1.5−VLSFeO1.5≈1.25 cm3 1.00 cm3mol−1[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin-69pt}{document}\mathrm{HS}}}\mathrm{FeO}}} - V\mathrm{LS}}}{FeO}}}\end{document}VHSFeO−VLSFeO≈ 1.75 cm3 mol‒1 1.10 cm3 mol‒1calculate difference of\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}\Delta V_{{\mathrm{exact{document}ΔVexact[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}\Delta V{document}ΔV at 3000 and 4000 K difference oxygen fugacity using volume differences (Supplementary Fig. pressures less than 60 GPa deviation volume difference spin transition less than 3% oxygen fugacity change spin effects increasing pressure magnitude of (ΔVexact − ΔV) increases at around 100 GPa of low-spin Fe3+ around 50% pressures greater than 80 GPa temperature MO around 3500 K increasing with pressure results at 4000 K more relevant neglecting spin transition of Fe oxygen fugacity high pressuresmaximum deviation 120 GPa ~0.6 log units uncertainties model prediction effects spin transition iron redox state MOs insignificant.Supplementary information Peer Review File | 48.4 | 1.052451 |
10.1038/s41467-020-17458-0 | PMC7367841 | Comparison of spontaneous canine cancers and human cancers may illuminate future therapeutic avenues. Here, genomic analyses of these tumors highlights a convergence on PI3K-Akt oncogenic pathways. | Genomic and precision medicine research has afforded notable advances in human cancer treatment, yet applicability to other species remains uncertain. Through whole-exome and transcriptome analyses of 191 spontaneous canine mammary tumors (CMTs) that exhibit the archetypal features of human breast cancers, we found a striking resemblance of genomic characteristics including frequent PIK3CA mutations (43.1%), aberrations of the PI3K-Akt pathway (61.7%), and key genes involved in cancer initiation and progression. We also identified three gene expression-based CMT subtypes, one of which segregated with basal-like human breast cancer subtypes with activated epithelial-to-mesenchymal transition, low claudin expression, and unfavorable disease prognosis. A relative lack of ERBB2 amplification and Her2-enrichment subtype in CMT denoted species-specific molecular mechanisms. Taken together, our results elucidate cross-species oncogenic signatures for a better understanding of universal and context-dependent mechanisms in breast cancer development and provide a basis for precision diagnostics and therapeutics for domestic dogs. | IntroductionCancer arises in dogs of all ages, just as in humans. Unlike commonly used animal models with artificial genetic modifications, canine tumors occur spontaneously with an intact immune system in ordinary living environments1. Moreover, in addition to similarities in anatomy and physiology between dogs and humans, canine tumors also exhibit the principal pathologic features of human cancers, including a long-term oncogenic setting, intratumoral heterogeneity, acquired resistance to treatment, and distant metastases2. Hence, canine tumors are invaluable representatives for human cancer research3,4. Among canine tumors, canine mammary tumors (CMTs) are the most common in female dogs5 and have been studied for a long time6. CMTs share molecular and clinical features with human breast cancers7, providing a basis for the adoption of classification systems including genetic, morphological, and prognostic elements8. Nowadays, calls for a deeper understanding of the molecular characteristics of CMTs are growing in order to uncover cross-species hallmarks of cancer and to provide better opportunities for treating cancers in dogs.Despite their apparent similarities, CMTs and human breast cancers show molecular and histological discrepancies that have perplexed veterinary and cancer researchers. For example, unlike in human breast cancers, the clinical benefits of Her2 amplifications and their association with Her2 overexpression are not straightforward in CMTs9,10, putting into question the incidence and potential clinical utility of ERBB2 amplification in CMTs. Moreover, the histological features of CMTs differ from those of human breast cancer. For example, benign tumors are more prevalent in CMTs (i.e., half of the observed cases)11. And tumors with mesenchymal origins (e.g., fibrosarcomas and carcinosarcomas) and proliferation of myoepithelial cells (e.g., complex adenomas/carcinomas) are often found in CMTs, all of which are extremely rare in human breast cancers12. These observations may imply the presence of distinctive mechanisms underlying carcinogenesis and cancer progression in and the need for more-specified therapeutic strategies for CMTs.Several studies have attempted to advance understanding of the genetic landscape underlying CMTs. Beck et al.13 documented CMT-specific gene fusions and deletions using low-depth genome sequencing of five cases. Gene expression profiling revealed genetic markers of disease progression and locoregional metastasis14,15. More recently, Liu et al.16 employed whole-exome sequencing (WES) and RNA sequencing (RNA-seq) of 12 CMT cases to identify histology-specific genetic alterations in CMTs: the authors proposed somatic alterations and epigenetic alterations as markers for simple and complex carcinomas, respectively. However, the mutational landscape of CMTs remains somewhat unclear owing to the small cohort sizes and lack of integrative analysis in these studies. We presumed that multi-omics profiling of CMTs in a large cohort, as in research into human cancers, would lead to better understanding of the underlying molecular pathogenesis of CMTs and inter-species relationships with human cancer.Here, we report our analysis of WES and transcriptome-sequencing (RNA-seq) data for 191 CMT cases, as the first cohort-level multi-omics study in canine cancers. Our study covers most of the latest genomic analyses applied in human cancer research, including the landscape of somatic mutations and involved pathways, mutational features (mutation burden and signatures), clonal selection, subtype specificity, gene expression, molecular subtyping, immune microenvironment, and survival analysis. We show a notable similarity between CMT and human breast cancers in terms of recurrent aberration in oncogenic pathways, which suggests molecular convergence of carcinogenesis, and highlight a number of novel CMT-specific mutations and their effects on tumor characteristics. Inclusion of a substantial number of benign tumors, which are usually not available in human, was relied upon to identify oncogenic characteristics in early cancer development. Finally, our study outlines molecular subtypes of prognostic relevance and suggests a need for the discovery of novel biomarkers of CMTs with which to facilitate early diagnosis for curative surgery and to develop targeted therapies.ResultsResearch cohortAll CMT specimens were obtained from 191 female dogs after curative surgery. Clinicopathological information for the cohort is summarized in Supplementary Data 1. The cohort comprised three histology types in 43 benign tumors (17 simple and 15 complex adenomas and 11 benign mixed tumors) and >5 histology types in 148 malignant tumors (78 simple and 44 complex carcinomas, 17 carcinomas in benign mixed tumors, and 9 others, including 4 osteosarcomas and 3 carcinosarcomas). The most frequent histologic type in the cohort was simple carcinoma, 63% (49/78) of which were of tubulopapillary type, resembling the natural incidence of malignant CMT8. The average age at diagnosis was 11.8 years. In total, WES data for 183 cases (all with matched blood sequencing data) and RNA-seq data for 157 cases (64 cases with matched normal tissue sequencing data) were obtained and further analyzed to ascertain the genomic and transcriptomic landscape of CMT. Sequencing information for WES and RNA-seq are available in Supplementary Data 2 and 3, respectively.Somatic mutation profiles of the CMT genomeBy comparing the tumor and matched normal sequencing data, we identified two types of somatic alterations: single-nucleotide substitution/variations (SNVs) and short insertions/deletions (indels). A total of 10,855 exonic mutations (8569 SNVs and 2286 indels, Supplementary Data 4) were identified for 183 cases with WES by a customized variant calling pipeline that includes strict filtration and canine-specific annotation (see Methods). The mutation landscape for nine recurrently mutated genes in CMT (eight genes harboring non-silent mutations for >5% of cases and AKT1) is depicted in Fig. 1a.Fig. 1Landscape of somatic mutations in CMT.a The mutational landscape of 183 CMTs (40 benign and 143 malignant CMTs) are shown for nine recurrently mutated genes. For benign and malignant CMTs, major histology types are shown at the top. Six mutation classes with respect to functional changes in the encoded amino acids are shown according to the color legend. b Non-silent mutations in four PI3K-Akt pathway genes are shown in lollipop plots. c Mutant allele frequencies of 3968 and 497 mutations in benign and malignant CMTs are compared. d dNdScv values are compared between 38 benign and 136 malignant CMTs. The significance was estimated by two-sided U tests. For all boxplots in this manuscript, the median and 1st/3rd quartiles of the data were plotted as the center line and the lower/upper boundaries. Whiskers represents the minimum and maximum of the data after removing outliers, which were defined as values smaller than 1st quartile – 1.5× IQR (interquartile range) or larger than 3rd quartile + 1.5× IQR.Among the somatic mutations, PIK3CA mutations were the most frequent (91 missense mutations and five in-frame indels in 43.1% of all 183 cases), which is consistent with activation of the PI3K-Akt pathway in breast cancers17,18. The majority of the missense mutations occurred at hotspot positions. The most frequently mutated amino-acid residue was H1047R/L (65.9% of 91 missense mutations); 12 other missense mutations (6.5%) were also observed on known PIK3CA hotspots19, suggesting that PIK3CA mutations are likely activating mutations and functional drivers of CMTs (Fig. 1b). Mutations on other genes in the PI3K-Akt pathway were also frequently observed: PTEN mutations (nine missense, two nonsense, and thee frameshift indels; 6.5% of cases), PIK3R1 mutations (two missense, one splicing and two frameshift; 6.0% of cases), and AKT1 mutations (eight missense, all on the known E17K hotspot; 4.9% of cases)20 (Fig. 1b). Overall, 55.7% of the CMTs harbored at least one non-silent mutation in four PI3K-Akt pathway genes. Of note, although most of the mutations within the PI3K-Akt pathway were widespread across different histology types, AKT1 mutations were exclusively observed in complex carcinomas (0/78 vs 8/44 in simple and complex carcinomas, respectively; P = 0.0002, Fisher’s exact test), suggesting a tissue-specific role of the mutation.Recurrent mutations in CMT outside the PI3K-Akt pathway showed intertumoral heterogeneity. KRAS mutations were found in 19 cases (10.4%), a relatively higher rate than that in human breast cancers (5%)21. All observed SNVs in KRAS were located on three known hotspots: 14 p.G12D/V/A substitutions, p.G13C, and p.E63K, with no silent mutations. This indicates KRAS mutation as another major driver of CMT. Truncating mutations in NF1 (one frameshift indel and two splicing mutations with two missense mutations) and in SF3B1 (one splicing mutation) were also observed in CMT, findings that are consistent with truncating mutations in human breast cancers (1.5% of putative driver mutations and mutations of unknown significance in NF1 and SF3B1, respectively)21.We further found that novel recurrent mutations at hotspot sites may account for species-specific mechanisms of carcinogenesis and progression in CMT. Encoding Ki-67 protein, a proliferation marker, MKI67 was frequently mutated in our cohort (6.0% cases; six missense mutations and two in-frame indels). Although Ki-67 protein levels have been proposed as a proliferation marker for discriminating luminal A and luminal B subtypes, no recurrent MKI67 mutations have been reported in human breast cancers. The recurrent nature of MKI67 mutations in CMT, as well as the presence of mutation hotspots (e.g., six non-silent mutations occurred at a single amino-acid residue of p.C1606), is indicative of a potential oncogenic role for MKI67 mutations in CMTs.Germline predisposing variants in CMTTo identify predisposing genetic events in CMT genomes, we detected 2005 germline variants (1124 SNVs and 881 indels) that are uncommon (novel or minor allele frequency <5%) and potentially damaging (i.e., truncating, likely pathogenic or pathogenic) (Supplementary Data 5). First, we found 10 cases harboring germline predisposing variants in BRCA1/2 genes, the prevalence of which (5.5%, 10/183) is slightly higher than those of human cancers (2.9–3.0%), measured in a meta-analysis of unselected breast cancer patients22. Among six BRCA1/BRCA2 mutations observed, four (two BRCA1 and two BRCA2 mutations) were stop-gain or nonsense mutations indicative of the potential functionality. Of interests, five out of the 10 BRCA1/2 germline variants harboring cases were observed in complex carcinomas suggestive of tissue type-specific mutations along with AKT1 somatic mutations (Supplementary Fig. 1). High prevalence of BRCA1/2 germline has been reported in basal-like and triple-negative breast cancer (TNBC) in human (14–15%)22, suggesting that the impact of inherited deficiency of BRCA1/2 is not universal, but context-specific in both species.Mutation enrichment analysis on nine DNA damage-repair pathways23 further identified that genes harboring germline variants are significantly enriched in homology-dependent recombination (HDR) pathway (P = 0.029; Fisher’s exact test, see Table 1). Other than BRCA1/2, 10 additional mutations were observed in the genes involved in HDR pathway (NBN, NSMCE1, POLD1, RECQL4, RMI1, RTEL1, SLX4, SMC5, TOP3B, and XRCC3), which may be also responsible for the CMT pathogenesis as shown in the example of NBN germline variants and associates breast cancer risk24.Table 1Germline variants of CMT in DNA damage-repair pathways.FunctionGenesP valueGenes with predisposing germline variantsHomology-dependent recombination (HDR)880.029BRCA1, BRCA2, NBN, NSMCE1, POLD1, RECQL4, RMI1, RTEL1, SLX4, SMC5, TOP3B, XRCC3Fanconi anemia (FA)410.081BRCA1, BRCA2, FAN1, RMI1, SLX4, TOP3BDirect repair (DR)40.266MGMTNon-homologous end joining (NHEJ)230.519NBN, RIF1Mismatch repair (MMR)240.542PMS1, POLD1Base excision repair (BER)470.689APEX1, NEIL3, POLD1Translesion synthesis (TLS)200.787POLINucleotide excision repair (NER)510.901ERCC6, POLD1Nucleotide pools (NP)51.0−The level of significance (P value) was estimated with Fisher’s exact test.Global patterns of somatic mutationsAnalyzing tumor mutation burden (TMB) (i.e., number of exonic mutations in a given genome), we recorded 10–198 exonic mutations per case of CMT (median of 30 and mean of 43.5 exonic mutations per case), except for an outlier with 2939 mutations. Compared with CMT, human breast cancers21 showed significantly higher level of TMB overall, i.e., n = 981 cases in the TCGA consortium; 45 median and 92.8 mean exonic mutations per case (P = 1.9e-9, U test; Supplementary Fig. 2).The mutation spectra and the functional consequences of mutations (e.g., changes in coding residues) were overall constant among CMTs (Supplementary Fig. 3). Considering the outlier case (CMT-033; 2939 exonic mutations) as hypermutated, the frequency of hypermutation in CMT (0.54%; 1 out of 183) was lower than that in human breast cancer (2.03%, 20 out of 981; cutoff for hypermutation >10 mutations per Mb). Although no mutations were observed in proofreading DNA polymerases (POLD1 and POLE) in CMT-033, we observed one truncating somatic mutation in MUTYH and missense mutations in genes encoding DNA damage-repair pathways, such as LIG1, LIG3, XRCC5, BRCA2, and XPC (Supplementary Data 4), which may contribute to a mutator phenotype for the CMT genome. In addition, 43 germline predisposing variants were observed in CMT-033 but no variants were found in DNA damage-repair pathways (Supplementary Data 5).Next, we examined the base substitution patterns of the somatic mutations based on the frequencies of 96 trinucleotides. We found that the trinucleotide frequencies were mostly similar across the examined CMT cases (Supplementary Fig. 4a). We first employed de novo mutation signature deconvolution based on non-negative matrix factorization (NMF) and Wellcome Trust Sanger Institute (WTSI) mutation signature framework (https://kr.mathworks.com/matlabcentral/fileexchange/38724-sigprofiler)25,26. Both methods revealed mutation signatures similar to Signature #1 (Sanger ver. 2 30 COSMIC mutation signatures #1 to #30, cosine similarity of 0.74–0.86) (Supplementary Fig. 4b, c) suggesting Signature #1 represents a major mutation signature in CMT genomes. We further employed mutation signature assignment analyses to estimate the levels of 30 COSMIC mutation signatures for all CMT genomes. Consistently, we observed that a single mutation signature of Signature #1, which represents age-related mutations, were prevalent in all CMT genome profiles (Supplementary Fig. 4d). These findings indicate that the mutation forces active during the carcinogenesis of CMT are largely uniform across individual cases and highlight spontaneous deamination leading to C-to-T transitions at CpG dinucleotides as the major contributor of somatic mutations therein (Supplementary Fig. 4c).Mutations in early cancer development and progressionInclusion of multi-type benign tumors in the cohort (43 out of 191, 22.5%) allowed us to directly compare genomic mutation profiles between benign and malignant tumors, which has rarely been undertaken in human cancer studies. As shown in the mutation profiles (Fig. 1a), PIK3CA mutations were frequently found in the benign tumors, suggesting that this key driver mutation is commonly acquired in advance of malignant progression. In contrast, TP53 mutations were found only in malignant CMTs (0 out of 43 benign vs 16 out of 148 malignant CMTs; P = 0.025, Fisher’s exact test). This suggests that TP53 mutations may arise as late evolutionary events in malignant progression, instead of CMT-initiating drivers. Likewise, KRAS mutations were over-represented in malignant tumors, particularly in complex carcinoma, being indicative of late events in the transformation to mesenchymal phenotypes27.We further examined and compared mutation abundance and composition between the benign and malignant tumors. The TMBs of the benign and malignant tumors were similar (P = 0.60, U test; Supplementary Fig. 2). Given the tumor purity has been proposed as a confounding factor in evaluating TMB, we further adjusted TMB with respect to the estimated tumor purity28. Consistently, no statistically significant difference was observed between the purity-adjusted TMB of benign and malignant CMTs (P = 0.44, U test). These findings suggest that the abundance of mutations largely remain the same during malignant progression. However, two mutation-based measures associated with tumor evolution (mutant allele frequencies (MAFs) and sample-wise dNdScv scores29 (dN/dS ratio)) provided clues on potential transforming events during benign-to-malignant progression. We found that the MAFs of malignant CMTs were significantly higher than those of benign CMTs (P = 7.95e-154; U test) (Fig. 1c), indicating that the malignant progression of CMT may accompany clonal selection events affecting subclonal mutation architecture, such as clonal sweeps. In addition, sample-wise dNdScv scores reflective of degrees of positive selection on individual cases29 were also substantially higher in malignant CMTs than in benign CMTs (mean dNdScv score of benign and malignant CMTs being 0.73 and 0.91, respectively) although not significantly different (P = 0.069; U test) (Fig. 1d). This demonstrated that the mutation composition of malignant CMTs shifts towards non-synonymous mutations, making them more likely to endure positive selection, compared with benign CMTs29. Gene-wise application of dNdScv scores revealed that the top 14 genes with positive selection (false discovery rate or FDR < 0.3; Supplementary Table 1) included four genes in the PI3K-Akt pathway (PIK3CA, PIK3R1, PTEN, and AKT1) and frequently mutated genes, such as KRAS and TP53. In particular, no silent or synonymous mutations were observed in the four PI3K-Akt pathway genes. Taken together, the molecular mechanisms underlying the progression of CMT can be modeled as (i) the emergence of early oncogenic mutations (e.g., PIK3CA mutation) in benign tumors, (ii) subsequent acquisition of additional drivers accompanying malignant transformation (e.g., TP53 and KRAS mutations), and (iii) clonal domination of malignant subclones with genomic footprints (e.g., MAF and dN/dS ratios).Somatic copy number aberrations in CMTSomatic copy number alterations (SCNAs) were profiled according to the read depth ratios of tumor and matched normal sequencing data. Genome-wide profiles of chromosomal copy number gains and losses in CMT are depicted across histologic types in Fig. 2a (shown in the order of the cases in Fig. 1a). Among the histologic types, SCNA-frequent cases were enriched in simple carcinoma, which is consistent with a previous report16. Genomic fractions with copy number imbalances were significantly higher in malignant CMTs than in benign CMTs (P = 0.0011; U test, Fig. 2b), suggesting that the genomic instability leading to SCNA is a late evolutionary event occurring after malignant progression. We also observed that genomes harboring TP53 mutations more commonly had SCNAs than genomes that did not (P = 8.8e-06; U test, Fig. 2b), supporting the notion that TP53 mutations lead to the accumulation of genomic instability and aneuploidy in cancer genomes16.Fig. 2Somatic copy number alterations (SCNAs) in CMT genome profiles.a A genome-wide heatmap of SCNAs is depicted, with red and blue representing chromosomal gains and losses, respectively. The cases are listed in order as in Fig. 1a. b Genome fractions with SCNAs were compared between 40 benign and 143 malignant CMTs, as well as between CMTs with or without TP53 mutations (n = 16 and 167, respectively). c Amplification and deletion peaks identified by GISTIC algorithms are shown across canine genomes with cancer-related genes belonging to the peaks shown. The significance was estimated by two-sided U tests.In regards to specific genes, recurrent SCNAs were observed on AKT1 (amplified in 2.7%) and PTEN (deleted in 10.9%). Approximately two-thirds of the cases (113 CMTs, 61.7% of the cohort) showed genomic aberrations in PI3K-Akt pathway genes, harboring either somatic mutations or SCNAs in PIK3CA, PTEN, PIK3R1, and AKT1. We further performed GISTIC analysis to identify 18 recurrent amplification and 49 recurrent deletion peaks across the CMT genomes (Supplementary Table 2). The identified GISTIC peak regions mirrored known cancer-related genes (i.e., Cancer Gene Census30) (Fig. 2c). Among the recurrently amplified and deleted loci, we noted that canonical oncogenes, such as EGFR and HRAS, were recurrently amplified in CMTs; however, amplification of ERBB2 and MYC, which is common in human breast cancers, was not reflected in the GISTIC peaks of CMTs. Thus, we further examined the genomic amplification patterns of EGFR in comparison to those of ERBB2 and MYC (Supplementary Fig. 5). In the case of EGFR, minimal amplification patterns around the genetic locus were observed and exhibited typical alteration patterns of functional oncogenes (Supplementary Fig. 5A). However, ERBB2 amplification was often observed as arm-level events or separate from pter-located GISTIC peaks, including ASPSCR1 and RNF213 (Supplementary Fig. 5B), which is in stark contrast to the minimal amplification of ERBB2 observed in human breast cancers (Supplementary Fig. 5B, inlet). Together with the low concordance between the copy numbers and gene expression levels of ERBB2 in CMT (r = 0.09, measured in this cohort), unlike in human breast cancers (r = 0.857), we can assume a reduced role and limited clinical value for ERBB2 in CMT, as was previously proposed9,10. Similarly, MYC amplification was frequently present at the chromosomal level, which resulted in an inability to detect MYC by GISTIC peaks (Supplementary Fig. 5C). Also, low correlation between copy numbers and gene expression levels (r = 0.093) was observed for MYC in CMT, in contrast to human breast cancers (r = 0.247). Regarding other genes, recurrent TERT amplification was noted; the prognostic implications thereof have been described for human breast cancers31. We also observed recurrent loss of genes with roles in genomic instability, including ATM and TP53, which are well-recognized cancer-associated genes in human cancers.Transcriptome analysis of CMTRNA-seq data were available for 157 tumors and 64 matched normal CMTs. We first applied NMF for the CMT transcriptome data set to delineate key transcriptional features or metagene signatures at the bulk level. Cophenetic scores showed that at least five metagene signatures were present in the transcriptome data set (Supplementary Fig. 6). Among the five metagene signatures (annotated NMF1–NMF5), three were relatively specific to CMTs (NMF1, NMF2, and NMF3); the other two signatures (NMF4 and NMF5) were specific to adjacent normal breast tissues (black and gray, respectively; Fig. 3a). We further performed gene set enrichment analysis to identify molecular functions associated with the five metagene signatures (Supplementary Table 3). According to the enriched molecular terms (based on the Hallmark gene set of MSigDB), we could annotate three tumor-specific metagene signatures as “mitosis” (NMF1), “DNA repair” (NMF2), and “epithelial-to-mesenchymal transition” (EMT) (NMF3). Likewise, the other two normal tissue-specific signatures were annotated as “estrogen-late” (NMF4) and “estrogen-early” (NMF5). The enrichment plots and top-enriched genes (i.e., leading edge genes) for the selected functions of the five metagene signatures are shown in Supplementary Fig. 6. Marker gene expression analysis revealed that NMF3 CMTs showed upregulation of the EMT markers SNAI1/2, ZEB1/2, and TGFB1. Of note, NMF3 CMTs further showed down-regulation of claudin-encoding genes (CLDN3, CLDN4, and CLDN7), as well as E-cadherin (CDH1). Along with a high degree of correlation between gene expression in NMF3 CMTs and that in tumor-initiating cells (“Signature correlation” in Fig. 3a), these findings suggest that NMF3 CMTs recapitulate the molecular features of claudin-low breast cancer subtypes32.Fig. 3Molecular taxonomy and tumor microenvironments of CMT based on transcriptomic analyses.a Five NMF metagene signatures are identified (NMF1–NMF5) with functional annotations. A heatmap shows that three NMF metagene signatures (NMF1–NMF3) are upregulated to varying degrees in CMT tumors (black), whereas NMF4 and NMF5 signatures are upregulated in normal tissue transcriptomes (gray). NMF clusters for individual CMTs are assigned according to the level of five NMF metagene signatures. The expression of selected EMT markers and claudin genes is also shown in a heatmap. The degrees of correlation between the expression signatures of tumor-initiating cells and CMT transcriptomes are depicted in signature correlation. b CMT tumors are classified into three subtypes (NMF1, NMF2, and NMF3), and their respective Kaplan–Meier survival curves are shown. The significance of survival differences across NMF clusters was estimated by log-rank test (two-sided). c Three NMF CMT subtypes were compared with histology types. d ESTIMATE-derived stromal and immune scores are plotted against the three NMF CMT clusters (92 NMF1, 35 NMF2, and 18 NMF3 CMTs, respectively). e Relative abundance of 12 immune cell subtypes estimated by CIBERSORT algorithms are shown for 92 NMF1, 35 NMF2, and 18 NMF3 CMTs.We further assessed whether the metagene signatures held any prognostic significance. Re-classification of the 157 CMT tumors with the five metagene signatures assigned 145 tumors to one of the three tumor signatures (97, 35, and 18 to NMF1, NMF2, and NMF3, respectively), which were subjected to survival analysis. Log-rank tests revealed significant differences in overall survival across the three CMT clusters (P = 0.010, log-rank test; Kaplan–Meyer survival curves in Fig. 3b), with the least favorable prognosis for the NMF3 CMTs with activated EMT. Although evidence of EMT has been reported in a few canine tumors33,34, our analysis establishes the presence of an EMT subtype and its association with poor prognosis in CMT, as in multiple human cancers35. Interestingly, the NMF3 cluster was enriched with rare CMT subtypes, including carcinosarcomas and osteosarcomas (Fig. 3c). As claudin-low subtypes are known to be enriched with tumor-initiating cells or stem cells that can differentiate into either myoepithelial or luminal progenitors in the hierarchy of mammary epithelial development36, the molecular features of NMF3 CMTs resembling those of claudin-low subtypes may be responsible for the presence of rare CMT histology subtypes with mixed epithelial and mesenchymal components, such as carcinosarcomas. To further evaluate the clinical relevance of NMF3 CMTs in an extended cohort of human breast cancers, we obtained four expression profiles in public database (GSE17907, GSE20711, GSE25066, and GSE31519). Patients with high NMF3 metagene scores showed unfavorable clinical outcomes, i.e., significantly different survival was observed for two of four cohorts (Supplementary Fig. 8). In addition, high EMT scores were observed for those with high NMF3 metagene scores.Next, we examined relationships among the metagene signatures with intrinsic molecular classification (e.g., luminal A, luminal B, Her2-enriched, and basal-like subtypes) for human breast cancer37. To apply the intrinsic human breast cancer subtypes to CMT, we focused on the expression of genes belonging to PAM50 in the expression profiles of human breast cancers and CMT. After merging the expression profiles, hierarchical clustering thereof largely segregated CMT tumors into two classes, one of which segregated with luminal A (non-basal CMTs) and the other with basal-like subtypes of human breast cancers (basal CMT) (Supplementary Fig. 7). We noted that ERBB2 overexpression was exclusive to Her2-enriched human breast cancers; it was not observed in other human breast cancer subtypes and CMTs. Basal CMT showed significant unfavorable clinical outcome in terms of survival, compared to non-basal CMTs (P = 0.004, log-rank test). In addition, basal CMTs were enriched with NMF3 CMTs and rare histological subtypes of CMT. Altogether, application of human breast cancer molecular subtyping to CMTs revealed that a subset of CMTs sharing the clinical behavior and histologic presentation of NMF3 CMTs is transcriptionally similar with basal-like human breast cancer. However, the results did not support the presence of Her2-enriched human breast cancer subtypes in our CMT cohort.The presence of EMT-related NMF signatures and their associations with CMT suggested that the tumor microenvironment affects CMT pathogenesis. Accordingly, we applied the ESTIMATE algorithm38 to estimate the relative fraction of stromal and immune cells in the CMT microenvironment. The analysis revealed significantly different ESTIMATE stromal scores among the three NMF classes (P = 5.18e-18, analysis of variance; ANOVA) (Fig. 3d). Higher stromal scores for NMF3 CMTs than for NMF1-NMF2 CMTs suggested higher degrees of stromal cell infiltration in NMF3 tumors. Accordingly, we deemed that the EMT-representing transcripts in NMF3 tumors may likely be derived from tumor-infiltrating stromal cells instead of tumor cells, consistent with a recent observation in mouse xenograft models39. Although no significant differences in ESTIMATE immune scores were observed among the three NMF classes (P = 0.388, ANOVA), it is possible that activated EMT may still impact the immune contexture or the immune cell composition in CMT, as only total immune cell counts are reflected in ESTIMATE immune scores. Thus, we further applied the CIBERSORT algorithm to estimate and compare the relative abundance of 22 immune cells among the NMF classes40. Twelve immune cell types showed significant differences (P < 0.05, ANOVA) among three NMF tumor types, as shown in Fig. 3e. We noted that NMF3 CMTs were relatively depleted of tumor-infiltrating immune cells, such as naive B cells, CD8 T cells, nature killer cells, and monocytes, compared with NMF1 and NMF2 CMTs. The enrichment of M0 and M2 macrophages was observed for NMF3 CMTs, suggesting that macrophages are the major immune component in the microenvironments thereof. We presume that NMF3 CMTs favor the polarization of M0 macrophages to M2 macrophages, which are important in inflammation and tissue repair, instead of M1 proinflammatory macrophages.Cross-species genomic alterations of canine and human breast cancersFor cross-species comparison of pathway-level mutations and SCNAs between CMT and human breast cancer, we examined 13 genes belonging to two signaling pathways (the PI3K-Akt and p53 pathways), particularly in regards to the frequencies of activation or inactivation thereof in benign and malignant CMTs and in human breast cancer (Fig. 4a). Four genes in the PI3K-Akt pathway, PIK3CA (55%, 38%, and 39% activated in benign CMT, malignant CMT and human breast cancer, respectively), PTEN (4%, 20%, and 13%), PIK3R1 (2%, 10%, and 8%), and AKT1 (0%, 9%, and 4%), showed comparable alteration frequencies between benign/malignant CMTs and human breast cancers, indicating that mutations in the PI3K-Akt signaling pathway are conserved across species in breast cancer pathogenesis. In addition, the higher mutation frequencies of PIK3CA in benign CMTs, compared with malignant CMTs and human breast cancers, highlight the early oncogenic roles of PIK3CA mutations. Other genes showed relatively lower alteration frequencies in CMTs than in human breast cancers, including TP53 (inactivated in 0%, 15%, and 48% of benign CMTs, malignant CMTs, and human breast cancers, respectively), as well as EGFR, ERBB2, ATM, and CHEK2. In addition, AKT3, MDM2, and MDM4 were rarely altered in CMTs. The relative lack of alterations in genes other than PI3K-Atk pathway genes may indicate that a limited repertoire of mutations is sufficient to give rise to CMT in a relatively short time, compared with human breast cancer development. Nevertheless, additional studies are required to investigate whether mutations less frequent in CMT genomes, such as ERBB2 amplifications, have similar biological or clinical implications as those in human breast cancers or whether they merely represent redundant alterations arising in the background of neutral mutations. The non-silent mutation frequencies of CMT genomes are compared with those of human breast cancers in Supplementary Data 6.Fig. 4Cross-species comparison of mutations and SCNAs.a For two major signaling pathways (PI3K-Atk and p53 pathways), the respective levels of activation and inactivation (red and blue, respectively) for 13 genes are shown. For each gene, activity levels are shown for benign CMTs, malignant CMTs, and human breast cancers, respectively. b CMT SCNAs are aligned onto the human reference genome (hg19) and compared with SCNA profiles of human breast cancers in the linear space of hg19 (chr1 to chr22 with sex chromosomes). The degrees of amplification and deletion are presented as the sum of log-ratios in hg19-aligned CMT and human breast cancer cohorts. c Chromosome 5 is shown, where human breast cancer shows arm-level 5p gains and 5q losses. The hg19-aligned CMT genomes show a deletion peak harboring PIK3R1. d Chromosome 8 shows peaks in six genes of hg19-aligned CMT SCNAs (CSMD, PSD3, IDO1, STK3, MYC, and PTK2), along with FGFR1 in an amplification peak of human breast cancer SCNAs.Finally, we transformed the CMT SCNA profiles by synteny realignment (canFam3.1) onto the human reference genome (hg19) using the blastz alignment algorithm41 (see Methods). SCNA profiles of hg19-aligned CMT genomes and human breast cancers are shown in Fig. 4b. We noted that cross-species correlations between segment-level amplifications and deletions were not strong (r = 0.225 and 0.037 for amplifications and deletions, respectively). However, major peaks in CCND1 (11q gain), CDKN2A (9p loss), and PTEN (10q loss) were concordant between dogs and humans. Owing to the different chromosomal constitutions of the two species, a single large (e.g., arm-level) SCNA event in human breast cancers may be represented by multiple SCNA events in CMT, helping to pinpoint regions of functional relevance. In chromosome 5, we observed a narrow deletion peak involving PIK3R1 in the hg19-aligned CMT SCNAs, which has been shown to reflect arm-level losses of 5q in human breast cancer SCNAs, suggesting functional relevance in terms of PI3K-Akt signaling (Fig. 4c). Likewise, in chromosome 8, arm-level gains of 8q and losses of 8p in human breast cancer SCNAs were further segmented into three amplification peaks involving STK3, MYC, and PTK2 and three deletion peaks involving CSMD1, PSD3, and IDO1 in hg19-aligned CMT SCNAs, respectively (Fig. 4d). These cross-species comparative oncogenomic results exemplify how genomic analysis of CMT can lead to the better understanding of human breast cancers.DiscussionIn this study, we conducted a comprehensive genomic analysis of CMT at the cohort level. Exome- and transcriptome-sequencing based molecular characterization revealed somatic mutations and SCNAs in CMTs at an unprecedented scale. Compared with the molecular characteristics of human breast cancers21, notable similarity in terms of core oncogenic signatures including key genes of the PI3K-Akt and p53 pathways were identified. We were also able to uncover species-specific molecular characteristics, such as uncertain role for ERBB2 amplification in CMTs, and mutations prevalent in benign tumors that may reflect the early genetic basis underlining the initiation of CMT pathogenesis. Finally, gene expression-based molecular taxonomy revealed the presence of an EMT-associated subtype in CMTs with an unfavorable prognosis. Overall, our study highlights the molecular convergence of key oncogenic pathways and supports the potential use of therapeutics for human breast cancer in dogs with CMTs42.Domestic dogs have a much shorter life expectancy than humans (10–13 vs 79 years), and tumorigenesis in dogs is accomplished within a shorter period (~10 years). The relatively shorter period for cancer onset may be responsible for the unique features of CMT. For example, the relative paucity of aneuploidy and SCNA drivers, such as ERBB2 amplifications in CMT, compared with human breast cancers, may be attributed to the shorter development time of the disease, as aneuploidy has often been considered as a late-stage marker43. Less-abundant mutation burdens of CMT compared with human breast cancers suggests that the mutational composition for breast cancer development may be different across species and those of CMT are relatively simpler compared with human breast cancers. It will require further investigation to see whether the additional mutations acquired by human breast cancer genomes merely represent the driver-accompanying neutral alterations or confer additional benefits. In addition, we found that CMT genomes showed relatively uniform TMB levels and sequence compositions (e.g., mutation signatures). This indicated that the mutational settings giving rise to CMT may be achieved at similar ages.Of interest, TMB levels were comparable between benign and malignant CMTs in this study. This has also been demonstrated for other tumor types, including colorectal cancers44, raising two possibilities. The first possibility is that an optimal or tolerable TMB level is fixed for a given cancer cell such that malignant progression allows for only an essential, but limited, number of additional mutations to be acquired. This assumption may be supported by our finding of cross-species similarities in CMT and human breast cancers in terms of PI3K-Akt pathway aberrations. The second possibility is that genetic programs favoring benign and malignant disease are determined early instead of following traditional stepwise acquisition of mutations during disease progression. Nevertheless, since the presence of mutations enriched in malignant disease, such as TP53 and KRAS mutations, supports that the malignant progression of CMT may follow traditional stepwise evolution, determining whether the malignant progression of CMT proceeds in accordance with traditional stepwise or parallel evolution will require further investigation with an extended longitudinal setting (e.g., comparison of multiple samples from the same individual). In addition, we found that measures representative of clone- and gene-level selective forces, such as MAF and dNdScv, indicated that the malignant progression of CMT involves selection events that alter the frequencies and compositions of somatic mutations. Thus, whether such selection events are responsible for or are independent of malignant progression need to be determined.Our research also raised questions as to whether the intrinsic mechanisms for suppressing cancer development (e.g., DNA repair, cell cycle arrest, or immunosurveillance) are intact or not in CMT. The predominance of a single type of mutation signature (Signature #1) in the CMT genomes suggested that the majority of somatic mutations may be those accumulated during the lifetime of a host, with a limited impact of other mutagenic sources in human breast cancers45 (e.g., APOBEC overactivity and BRCA deficiency). However, no correlation was observed (r = 0.04) between host age and the levels of Signature #1 and it is possible that somatic mutations of CMT genomes have arisen in a limited time period such as a mutational burst, or at least have accumulated in a non-gradual manner probably associated with a loss of DNA repair mechanisms. The identification of susceptibility factors for somatic mutations may also lead to developing means of reinforcing tumor-suppressing systems in both dogs and humans.As the fierce battle with cancer is now expanding to companion animals, treatment of canine cancer itself is becoming an important issue. Although anticancer agents originally developed to treat human cancers may be applied to dogs, little evidence has been given in terms of their therapeutic efficacy, especially in relation to cost effectiveness. Recently, a study of drug sensitivity showed trametinib (a MEK1/2 inhibitor) to be effective in canine cancer cell lines46, and we expect more lines of evidence will accumulate on trans-species use of more drugs of these kinds. In spite of a concern regarding the discordance between animal and human in drug efficacy and toxicities47, treatment of canine cancer may benefit from the development of novel human cancer drugs that target shared oncogenic mutations (e.g., alpelisib for metastatic breast cancer with PIK3CA mutations48). We envision that genomic studies of different cancer types, further stratification, and companion diagnostics will lead to more efficient treatment of canine cancers, just as they have done for human cancers over last 10 years.MethodsDetailed information on the study design, sample collection, data generation, and quality control strategies has been described in a separate data descriptor paper49. Here, we have provided a brief overview of the data that are essential to understanding the presented study: some parts of this section may contain overlapping descriptions with the data descriptor paper, especially for the conventional protocols.Cohort design and sample collectionThe cohort was designed as a tumor and matched normal control cohort to facilitate the investigation and comparison of genomic and transcriptomic features. In available cases, blood (buffy coat) and adjacent normal mammary tissues were used as controls for tumor DNA and RNA, respectively. In total, 191 dogs with mammary tumors were recruited via private veterinary clinics in Korea, with informed consent from their owners. Tumor tissues, adjacent normal tissues, and blood were collected from the dogs following the guidelines of the Institutional Animal Care and Use Committee of Konkuk University (KU16106 and KU17162) upon availability. Fresh tissue samples were immediately transferred to RNAlater (Thermo Fisher Scientific, Vilnius, Lithuania), refrigerated overnight at 4 °C, and stored at −80 °C. Genomic DNA was extracted from tumor tissue and buffy coats using QIAamp DNA mini kits (Qiagen, Germany). Total RNA was extracted from tumor and adjacent normal tissues using RNeasy mini kits (Qiagen).HistopathologyFor histological examination, sections (4-μm thick) from formalin-fixed paraffin-embedded blocks were stained with hematoxylin and eosin and were diagnosed by two researchers (B.J.S. and J.H.S.). Histological subtyping was based on the World Health Organization classification50. The diagnosis of malignancy, which included ambiguous subtypes (e.g., simple adenoma vs simple carcinoma (grade 1), complex adenoma vs complex carcinoma (grade 1), benign mixed tumor vs carcinoma in benign mixed tumor (grade 1)) was determined in accordance with that described by Rasotto et al.51. Histological grade was assessed according to the Peña system52, exclusively on the neoplastic epithelial component. In cases of mammary osteosarcoma and mammary fibrosarcoma, histological grade was assessed according to the grading system for canine osteosarcoma53 and the grading system for cutaneous and subcutaneous soft tissue sarcoma in dogs54, respectively. Lymphatic invasion, defined as the presence of tumor cells in peritumoral lymphatic vessels (all cases) and/or regional lymph nodes (only available cases), was also determined.WES and RNA-seqAmong 191 samples, 183 cases with tumor DNA and matched normal DNA were subjected to WES. Two hundred nanograms of fragmented DNA was prepared to construct libraries with the SureSelect Canine All Exon Kit (Agilent, Inc., USA) using the manufacturer’s protocol. In brief, qualified genomic DNA samples were randomly fragmented by Covaris, followed by adapter ligation, purification, hybridization, and PCR. Captured libraries were then examined on an Agilent 2100 Bioanalyzer to evaluate quality and were loaded on an Illumina HiSeq sequencer, according to the manufacturers’ recommendations. In addition, 157 tumor tissues and 64 matched, normal, adjacent tissues with RNA available were also subjected to RNA-seq. Before library construction, RNA 6000 Nano kits (Agilent Technologies, CA) were used to assess RNA quality. For cDNA library construction, 1 μg of RNA was obtained and purified with oligo-dT magnetic beads. Fragmentation was performed with purified mRNA, and double-stranded cDNAs were synthesized. The cDNAs were primed with poly-A, and sequencing adapters were connected using TruSeq RNA sample prep kits (Illumina, CA). Fragments were filtered to a specific length using BluePippin 2% agarose gel cassettes (Sage Science, MA), and PCR amplification was conducted. Fragment lengths and quality were electrophoretically verified with Agilent High Sensitivity DNA kits (Agilent Technologies, CA). Libraries were observed with a window spanning an average of 392 bp, standard deviation of 66 bp. WES and RNA-seq were performed using Illumina HiSeq 2500 (Illumina, CA) with the protocol out-sourced to Theragenetex Inc.Processing of sequencing dataWES reads were aligned to the CanFam3.1 (Canis lupus familiaris) reference genome with BWA-MEM255. Duplicate fragments were marked and eliminated with Picard (version 2.2) (http://picard.sourceforge.net). After assessing mapping quality and filtering out low-quality mapped reads, paired read information was evaluated to ensure that all mate-pair information was in sync between each read. Then, processes of removing PCR duplicates, indel realignment, fixing mate information, base quality score recalibration, and variant quality score recalibration on putative SNVs and indels were performed using GATK4.0 following GATK Best Practices recommendations56 with CanFam3.1 (Ensembl Release 91) as a reference. The whole pipeline was implemented in-house49. RNA-Seq data of the 157 tumor samples and 64 normal, adjacent samples were mapped to the canine reference genome CanFam3.1 using splice-aware aligner of TopHat57 (v.2.0.9), with Ensembl gene annotation and fr-firststrand library type. FPKM (fragments per kilobase of transcript per million) values were calculated by Cufflinks58 (v2.1.1) using aligned bam files.Germline variants calling and annotationGATK-HC59 (v4.1.5.0) was used to call germline variants in paired bulk data and filtered by VariantFiltration of GATK4 with the criteria recommended for germline variants; excluding candidates with QD < 2.0, FS > 60.6, MQ < 40.0, ReadPosRankSum < −8.0, MQRanksum <2.5 or MQRankSum >2.5. Variants with ≥10 depth in both or only in normal sample were used for further analysis. Variants were liftovered from CanFam3.1 to hg38 with Crossmap60 (v0.2.9) using UCSC chain file (camFam3ToHg38.over.chain.gz). Variants successfully liftovered to hg38 were annotated using SnpEff61 (v4.3t) and ClinVar62 (build 2020-03-10). We only kept variants that are tagged as truncated (stop gained, splice variant, frameshift), pathogenic, or likely pathogenic. Only novel or uncommon candidates with MAF < 0.05 in Dog Genome SNP Database (DogSD) (release 2017-06-10)63 from iDOG and within the cohort were included in the final germline variant list.Somatic variant calling, filtration, and annotationWe detected single-nucleotide variations (SNVs) and small insertions/deletions (indels) using Mutect264 from GATK4 v4.0.10.1. The VCF file produced by the pipeline utilizes reference bases on the positive strand of CanFam3.1 in the REF field, and variants are shown in the ALT field. We filtered out falsely detected variants using FilterMutectCalls from GATK4 and selected PASS variants from the VCF files. Variant Effect Predictor was used to annotate identified variants65. To estimate dNdScv scores, we used R packages (https://github.com/im3sanger/dndscv)29.We examined the possibility of the somatic mutations being falsely detected by alignment errors. In the absence of utility for alignment error assessment in canine genome, we built our in-house workflow to strictly rule out variant sites with dubious alignment patterns using the aligner-generated alignment scores at optimal and suboptimal mappings. For each SNV position, passing sequence reads with an alternative allele were collected to calculate the alignment score at optimal site (AS), and the alignment score at suboptimal (or secondary) site (XS), which are averaged over all samples to derive the mean alignment scores at optimal site (meanAS) and suboptimal site (meanXS). We judged that the candidate SNVs are likely from alignment errors if the meanXS is greater than or equal to the 80% of the meanAS (meanXS ≥ 0.8×meanAS).Tumor purity estimation and TMB adjustmentTumor purity of CMT genomes were visually estimated by histological examination of H&E stained slides by a pathologist (B-J Seung). Only samples with >70% of the judged proportion of neoplastic cells were used for analysis. In the case of human breast cancers of TCGA consortium, we obtained consensus purity estimates from a literature66. The TMB of individual genomes were multiplied with the correction factor corresponding to the purity of the given case, to derive the purity-corrected TMB28.Copy number variant callingWe calculated the depth of coverage using GATK and then followed the typical XHMM workflow. SAMTOOLS67 v1.9 and VARSCAN68 v.2.4.3 were used to identify SCNAs following the recommended workflow. First, we ran the mpileup function in SAMTOOLS to estimate the bin-level sequencing read depth both for tumor and normal BAM files. The ratios of tumor/normal sequencing read depth were then calculated, and the normalized read depth ratios were further GC-corrected. We applied the circular binary segmentation (CBS)69 algorithm for segmentation and used the IGV browser for visualization of SCNAs70. We also used GISTIC2 to identify recurrent chromosomal gains and losses of CMT genomes71. For cross-species comparison of SCNAs between CMTs and human breast cancer, we used the cross-species alignment information of canFam3 and hg19 assemblies generated by the blastz algorithm (http://hgdownload.cse.ucsc.edu/goldenpath/hg19/vsCanFam3/). The segment-level log2 ratios of CMT genomes (canFam3.1) were aligned onto hg19 using the blastz chained alignment. The hg19-aligned SCNA profiles were further smoothed and segmented using the CBS algorithm.Transcriptome-based CMT subtypesWe performed NMF deconvolution to identify latent features in CMT expression profiles by decomposing the log-transformed CMT expression matrix into a basis matrix (hereafter, NMF metagene signatures) and metagene expression profiles72. For deconvolution, CMT expression profiles along with those of normal, adjacent mammary tissues were subjected to NMF. To determine the optimal number of NMF metagene signatures, we measured the cophenetic score for 2–10 NMF metagene signatures as a stability measure. Five NMF metagene signatures were derived including three signatures representing CMT tumors. We further performed pre-ranked version of gene set enrichment analysis with functional gene sets (MSigDB, Hallmark category) for functional annotation of metagene signatures73. For prognostic evaluation of NMF clusters, Kaplan–Meier survival curves were drawn and log-rank tests were applied for three tumor-specific NMF clusters. We also performed molecular classification of CMT transcriptomes using PAM50 genes. The expression levels of PAM50 genes for TCGA human breast cancer and CMT tumors were merged and subjected to hierarchical clustering. To estimate correlations between individual CMT transcriptomes and the expression of tumor-initiating cells, we analyzed 154 up- and 339 downregulated genes of CD44+/CD24− tumor-initiating cells, with signature correlation levels calculated as previously described32.Validation of CMT subtypes in human cohortFour expression profiles of human breast cancers in public database (GSE17907, GSE20711, GSE25066, and GSE31519) were obtained with clinical outcomes. Patients in the individual cohorts were discriminated into high and low NMF3 with the median of NMF3 scores (i.e., average expression level of genes in NMF3 metagenes). Patient survival of high and low NMF3 was compared using log-rank tests and also for the EMT scores as average expression of genes in EMT/Hallmark MSigDB gene set.Tumor microenvironment profilingWe used the ESTIMATE R package to estimate scores representative of the relative proportion of immune and stromal cells in the admixture of CMT transcriptome data38. To infer the relative abundance of tumor-infiltrating immune cells, CIBERSORT was used, with the LM22 set representing 22 immune cell subtypes40.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Data 1
Supplementary Data 2
Supplementary Data 3
Supplementary Data 4
Supplementary Data 5
Supplementary Data 6
Reporting Summary | nature communications | [
"Article"
] | [
"Breast cancer",
"Cancer genomics",
"Genome informatics"
] | arises in dogs as humans Unlike canine tumors occur spontaneously intact immune system canine tumors exhibit pathologic features human cancers long oncogenic setting intratumoral heterogeneity acquired resistance to treatment distant metastases2. canine tumors invaluable for human cancer canine mammary tumors (CMTs most common in female studied long CMTs share molecular clinical features with human breast cancers7 basis for classification systems genetic prognostic for deeper understanding of molecular characteristics CMTs growing uncover cross-species hallmarks cancer treating cancers CMTs human breast cancers show molecular histological discrepancies researchers clinical benefits of Her2 amplifications association with overexpression not straightforward in CMTs9 incidence utility amplification histological features of CMTs differ from human breast cancer benign tumors prevalent in CMTs half of cases tumors with mesenchymal origins proliferation of myoepithelial cells often found in CMTs rare in human breast cancers12observations imply mechanisms carcinogenesis cancer progression need strategies CMTs studies genetic landscape CMTs Beck et al documented CMT-specific gene fusions deletions sequencing five cases Gene expression profiling revealed markers disease progression locoregional Liu et al employed sequencing RNA sequencing 12 CMT cases histology-specific genetic alterations proposed somatic epigenetic alterations markers simple complex carcinomas mutational landscape CMTs unclear small cohort sizes lack integrative analysis presumed multi-omics profiling CMTs large cohort molecular pathogenesis inter-species relationships cancer report analysis WES 191 CMT cases first cohort-level multi-omics study in canine cancers study covers genomic analyses somatic mutations pathways mutational features clonal selection subtype specificity gene expression molecular subtyping immune microenvironment survival analysis similarity between CMT human breast cancers recurrent aberration oncogenic pathways molecular convergence carcinogenesis novel CMT-specific mutations effects tumor characteristicsbenign tumors not human oncogenic characteristics early cancer development study outlines molecular subtypes prognostic suggests need novel biomarkers CMTs early diagnosis curative surgery targeted therapies CMT specimens 191 female dogs after surgery Clinicopathological information Supplementary Data 1. three histology types 43 benign tumors (17 simple 15 complex adenomas 11 mixed tumors >5 histology types 148 malignant tumors (78 simple 44 complex 17 9 4 osteosarcomas 3 carcinosarcomas). most frequent type simple carcinoma 63% (49/78) tubulopapillary type malignant CMT8 average age 11.8 years WES data 183 cases RNA-seq data 157 cases obtained analyzed genomic transcriptomic landscape CMT Sequencing information WES RNA-seq Supplementary Data 2 3mutation profiles CMT tumor normal sequencing data identified somatic alterations single-nucleotide substitution/variations short insertions/deletions 10,855 exonic mutations (8569 SNVs 2286 indels identified 183 cases WES variant pipeline filtration canine-specific annotation mutation landscape nine recurrently mutated genes CMT non mutations >5% Fig. 1a mutations CMT landscape 183 CMTs (40 benign 143 malignant nine mutated genes major histology types top Six mutation classes functional changes amino acids Non-silent mutations four PI3K-Akt pathway genes lollipop plots Mutant allele frequencies 3968 mutations benign malignant CMTs compared dNdScv values compared 38 benign 136 malignant CMTs significance estimated two-sided U tests median 1st/3rd quartiles center line lower/upper boundaries Whiskers minimum maximum data removing outliers smaller 1st – 1.5× larger 3rd quartile + 1.5× IQR PIK3CA mutations most frequent (91 missense mutations five in-frame indels 43.1% 183 consistent activation PI3K-Akt pathway breastmissense mutations at hotspot positions most frequently mutated amino H1047R/L (65.9% of 91 12 mutations (6.5%) observed on PIK3CA hotspots19 PIK3CA mutations drivers CMTs Mutations other genes PI3K-Akt pathway frequently observed PTEN mutations (nine two 6.5% PIK3R1 mutations (two 6.0% mutations (eight E17K hotspot 4.9% 55.7% CMTs harbored non-silent mutation in four PI3K-Akt pathway genes mutations observed in complex carcinomas (0/78 vs 8/44 tissue-specific role mutations in CMT outside PI3K-Akt pathway intertumoral heterogeneity KRAS mutations in 19 cases (10.4%), higher rate human breast cancers (5% SNVs in KRAS on three hotspots 14 p.G12D/V/A substitutions, p.G13C p.E63K no silent mutations KRAS mutation major driver CMTTruncating mutations NF1 two SF3B1 observed in CMT consistent with human breast cancers (1.5% putative driver mutations unknown significance in SF3B1 novel recurrent mutations hotspot sites may account for species-specific carcinogenesis progression in CMT Encoding Ki-67 protein proliferation marker MKI67 frequently mutated cohort (6.0% cases six missense mutations two in-frame Ki-67 protein levels proliferation marker luminal A B subtypes no recurrent MKI67 mutations reported human breast cancers recurrent mutations mutation hotspots six non-silent mutations single amino-acid residue p.C1606) potential oncogenic role MKI67 mutations predisposing variants detected 2005 germline variants (1124 SNVs 881 indels) uncommon frequency <5% potentially damaging likely pathogenic 10 cases germline predisposing variants in BRCA1/2 genes prevalence (5.5%, 10/183) higher than human cancers (2.9–3.0%) six BRCA1/BRCA2 mutations four stop-gain nonsense potential functionalityfive 10 BRCA1/2 variants complex carcinomas High prevalence BRCA1/2 basal-like triple-negative breast cancer (14–15% inherited deficiency BRCA1/2 context-specific enrichment analysis nine DNA damage-repair genes enriched homology-dependent recombination) pathway (P = 0.029 10 mutations genes HDR pathway (NBN NSMCE1 POLD1 RECQL4 RMI1 RTEL1 SLX4 SMC5 TOP3B XRCC3) CMT pathogenesis breast cancer variants CMT DNA damage-repair pathways variantsHomology-dependent recombination)880.029BRCA1 BRCA2 NBN NSMCE1 POLD1 RECQL4 RMI1 RTEL1 SLX4 SMC5 TOP3B XRCC3Fanconi anemia)410.081BRCA1 BRCA2 RMI1 SLX4 TOP3BDirect repair)40.266MGMTNon-homologous end joining)230.519NBN RIF1Mismatch repair)240.542PMS1 POLD1Base excision repair)470.689APEX1 NEIL3 POLD1Translesion synthesis)200.excision repair)510.901ERCC6 POLD1Nucleotide pools significance estimated Fisher’s exact test patterns somatic tumor burden recorded 10–198 mutations per case CMT (median 30 43.5 outlier 2939 mutations human breast higher TMB n = 981 cases TCGA consortium 45 median 92.8 mean mutations per case (P = 1.9e-9 mutation spectra functional consequences constant CMTs outlier case (CMT-033 2939 mutations hypermutated frequency CMT (0.54% 1 out of 183 lower human breast cancer (2.03% 20 out of 981 no mutations DNA polymerases (POLD1 POLE) CMT-033 observed one truncating somatic mutation MUTYH missense mutations genes DNA damage-repair pathways LIG1 LIG3 XRCC5 BRCA2 XPC contribute mutator phenotype CMT genome 43 germline predisposing variants CMT-033 no variants DNA damage-repair pathways examined base substitution patterns somatic mutations frequencies 96 trinucleotidestrinucleotide frequencies similar across CMT cases Fig employed mutation signature deconvolution non-negative matrix factorization Wellcome Trust Sanger Institute) mutation signature framework revealed mutation signatures similar Signature #1 (Sanger. 2 30 COSMIC mutation signatures #1 to #30 cosine similarity 0.74–0.86) 4b major mutation CMT genomes mutation signature assignment analyses 30 COSMIC mutation signatures CMT genomes single mutation signature Signature #1 age-related mutations prevalent all CMT genome profiles 4d). findings indicate mutation forces carcinogenesis CMT uniform across cases highlight spontaneous deamination C-to-T transitions CpG dinucleotides major contributor mutations Fig early cancer development multi-type benign tumors cohort (43 out of 191, 22.5% genomic mutation profiles benign malignant tumors PIK3CA mutations frequently benign tumors acquired malignant progression TP53 mutations malignant CMTs (0 out 43 benign 16 out of 148 malignant P = 0.025 TP53 mutations late evolutionary events malignant progressionKRAS mutations over-represented in malignant tumors particularly complex carcinoma indicative of late events transformation mesenchymal phenotypes27 compared mutation abundance composition benign malignant tumors TMBs similar (P = 0.60 tumor purity factor TMB adjusted TMB estimated purity28 no statistically significant difference between purity-adjusted TMB benign malignant CMTs (P = 0.44 suggest abundance mutations same during malignant progression mutation-based measures (mutant allele frequencies (MAFs) sample-wise dNdScv scores29 potential transforming events during benign-to-malignant progression MAFs of malignant CMTs higher than benign CMTs (P = 7.95e-154 malignant progression may accompany clonal selection events subclonal mutation sample-wise dNdScv scores higher in malignant CMTs benign CMTs 0.73 0.91 not significantly different (P = 0.069 (Fig mutation composition malignant CMTs shifts towards non-synonymous mutations more likely endure positive selectiondNdScv scores top 14 genes positive selection < 0.3 included four genes PI3K-Akt pathway (PIK3CA PIK3R1 PTEN AKT1) mutated genes KRAS TP53 no silent synonymous mutations in four PI3K-Akt genes molecular mechanisms progression CMT modeled emergence early oncogenic mutations PIK3CA mutation in benign tumors additional drivers TP53 KRAS clonal domination malignant subclones copy number aberrations profiled read depth ratios tumor matched normal sequencing data Genome-wide profiles chromosomal copy number gains losses CMT depicted across histologic types in Fig. 2a SCNA-frequent cases enriched in simple carcinoma consistent Genomic fractions with copy number imbalances higher in malignant CMTs than benign CMTs (P = 0.0011 instability leading SCNA late evolutionary event after malignant progression genomes TP53 mutations commonly had SCNAs (P = 8.8e-06 TP53 mutations lead to genomic instability aneuploidy in cancercopy alterations CMT genome profiles genome-wide heatmap SCNAs red blue chromosomal gains losses cases listed Fig. 1a Genome SCNAs compared between 40 benign 143 malignant CMTs without TP53 mutations (n = 16 167 Amplification deletion peaks GISTIC canine genomes cancer-related genes significance estimated two-sided U tests recurrent SCNAs observed on AKT1 (amplified 2.7% PTEN (deleted 10.9%). two-thirds cases (113 CMTs 61.7% cohort showed genomic aberrations PI3K-Akt pathway genes mutations SCNAs PIK3CA PTEN PIK3R1 AKT1 GISTIC analysis 18 amplification 49 deletion peaks CMT genomes 2) GISTIC peak regions mirrored cancer-related genes amplified canonical oncogenes EGFR HRAS amplified CMTs amplification ERBB2 MYC common breast cancers not reflected GISTIC peaks examined genomic amplification patterns EGFR ERBB2 MYC EGFR minimal amplification patterns genetic locus typical alteration patterns functional oncogenesERBB2 amplification observed arm-level events separate from GISTIC peaks including ASPSCR1 RNF213 contrast to minimal amplification ERBB2 in human breast cancers low concordance between copy numbers gene expression levels ERBB2 in CMT (r = 0.09 human breast cancers = reduced role limited clinical value for ERBB2 in CMT MYC amplification present at chromosomal level inability detect MYC by GISTIC peaks low correlation between copy numbers gene expression levels (r = 0.093) for MYC in CMT contrast to human breast cancers (r = 0.247) recurrent TERT amplification noted prognostic implications described for human breast cancers31 recurrent loss of genes genomic instability including ATM TP53 cancer-associated genes in.Transcriptome analysis CMTRNA-seq data for 157 tumors 64 normal CMTs applied NMF for CMT transcriptome data transcriptional features level five metagene signatures present three specific to CMTs two specific to normal breast tissuesperformed gene set enrichment analysis functions five metagene signatures (Supplementary Table 3) enriched molecular terms Hallmark gene set three tumor-specific metagene signatures (NMF1) (NMF2) “epithelial-to-mesenchymal (NMF3) two normal tissue signatures annotated “estrogen (NMF4) “estrogen (NMF5) enrichment plots-enriched genes Supplementary Fig. 6. analysis NMF3 CMTs upregulation EMT markers SNAI1/2 ZEB1/2 TGFB1 down-regulation claudin-encoding genes (CLDN3 CLDN7) E-cadherin correlation gene expression NMF3 CMTs tumor-initiating cells recapitulate molecular features claudin-low breast cancer. 3Molecular taxonomy tumor microenvironments CMT transcriptomic analyses Five NMF metagene signatures identified (NMF1–NMF5) functional annotations three signatures upregulated in CMT tumors NMF4 NMF5 upregulated normal tissue transcriptomes NMF clusters CMTs assigned level five NMF metagene signaturesexpression EMT markers claudin genes shown in heatmap correlation tumor cells CMT transcriptomes correlation CMT tumors classified three subtypes (NMF1 Kaplan–Meier survival curves survival differences clusters estimated log-rank test Three NMF CMT subtypes compared with histology types ESTIMATE stromal immune scores plotted against three NMF clusters (92 NMF1 35 NMF2 18 NMF3 abundance 12 immune cell subtypes CIBERSORT algorithms for 92 NMF1 35 NMF2 18 NMF3 CMTs assessed metagene signatures prognostic significance Re-classification 157 CMT tumors assigned 145 tumors signatures survival Log-rank tests differences survival across three CMT clusters (P = 0.010 curves least favorable prognosis NMF3 CMTs activated EMT evidence EMT canine analysis establishes EMT subtype poor prognosis CMT NMF3 cluster enriched with rare CMT subtypes carcinosarcomas osteosarcomasclaudin-low subtypes enriched with tumor-initiating cells into myoepithelial or luminal progenitors mammary molecular features NMF3 CMTs resembling claudin-low subtypes for rare CMT histology subtypes mixed epithelial mesenchymal carcinosarcomas clinical relevance NMF3 CMTs human breast cancers obtained four expression profiles (GSE17907 GSE20711 GSE25066 GSE31519) high NMF3 metagene scores showed unfavorable clinical outcomes different survival two of four cohorts high EMT scores observed for high NMF3 metagene scores examined relationships among metagene signatures with intrinsic molecular classification luminal A B Her2-enriched basal-like subtypes for human breast focused on expression genes PAM50 expression profiles clustering segregated CMT tumors into two classes luminal A basal-like ERBB2 overexpression exclusive to Her2-enriched breast cancers not observed in other subtypes CMTsBasal CMT unfavorable outcome non-basal CMTs (P = 0.004 CMTs enriched with NMF3 CMTs rare subtypes CMT human breast cancer molecular subtyping CMTs NMF3 CMTs similar with basal human breast cancer results support Her2-enriched breast cancer subtypes CMT cohort EMT-related NMF signatures associations CMT tumor microenvironment affects CMT pathogenesis applied ESTIMATE stromal immune cells CMT microenvironment different stromal scores among three NMF classes (P = Higher stromal scores NMF3 CMTs suggested higher stromal cell infiltration tumors EMT-representing transcripts in NMF3 tumors derived from tumor-infiltrating stromal cells tumor consistent mouse xenograft no significant differences in immune scores among three NMF classes (P = 0.388 activated EMT may impact immune contexture composition CMT applied CIBERSORT algorithm abundance 22 immune cells among NMF Twelve immune cell types showed differences (P < 0.05 among three NMF tumor types Fig. 3eNMF3 CMTs depleted of tumor-infiltrating immune cells CD8 T killer monocytes compared with NMF1 NMF2 CMTs enrichment M0 M2 macrophages NMF3 CMTs major immune component NMF3 CMTs favor polarization M0 to M2 important tissue repair M1 macrophages-species genomic alterations canine human breast mutations CMT human breast cancer examined 13 genes PI3K-Akt p53 frequencies activation benign malignant CMTs human breast cancer. genes PI3K-Akt pathway PIK3CA (55% 38% 39% PTEN (4% 20% 13%), PIK3R1 (2% 10% 8%) AKT1 (0% 9% comparable alteration frequencies between benign/malignant CMTs breast cancers mutations PI3K-Akt conserved across species breast cancer pathogenesis higher mutation frequencies PIK3CA in benign CMTs highlight early oncogenic roles PIK3CA mutationsgenes lower alteration frequencies in CMTs breast cancers including TP53 (inactivated in 0% 15% 48% benign malignant EGFR ERBB2 ATM CHEK2. AKT3 MDM2 MDM4 rarely altered in CMTs lack of alterations in PI3K-Atk limited repertoire mutations CMT short cancer additional studies required mutations frequent CMT genomes ERBB2 implications cancers redundant alterations mutations non-silent mutation frequencies CMT genomes compared with human breast cancers Supplementary Data 6.Fig. 4Cross-species comparison mutations SCNAs signaling (PI3K-Atk p53 levels activation inactivation 13 genes activity levels benign CMTs malignant CMTs human breast cancers CMT SCNAs aligned human reference genome compared with SCNA profiles breast cancers degrees amplification deletion sum log-ratios in hg19-aligned CMT human breast cancer cohorts Chromosome 5 human breast cancer shows-level 5p gains 5q losses hg19-aligned CMT genomes show deletion peak PIK3R1.Chromosome 8 peaks genes hg19 CMT SCNAs (CSMD PSD3 IDO1 STK3 MYC PTK2) human breast cancer SCNAs transformed CMT SCNA profiles human reference genome (hg19) blastz alignment SCNA profiles breast cancers Fig. 4b cross-species correlations segment-level amplifications deletions strong (r = 0.225 0.037 peaks CCND1 (11q CDKN2A (9p PTEN (10q loss concordant between dogs humans chromosomal constitutions large SCNA event breast cancers represented multiple events functional relevance chromosome 5 narrow deletion peak PIK3R1 hg19-aligned CMT SCNAs arm-level losses 5q breast cancer functional relevance chromosome 8 arm-level gains 8q losses 8p cancer SCNAs segmented three amplification peaks STK3 MYC PTK2 deletion peaks CSMD1 PSD3 IDO1 hg19-aligned CMT SCNAs (Fig. cross-species comparative oncogenomic results genomic analysis CMT understanding human breast cancersstudy conducted comprehensive genomic analysis CMT cohort level Exome- transcriptome-sequencing molecular characterization revealed somatic mutations SCNAs in CMTs unprecedented scale Compared with human breast cancers21 similarity core oncogenic signatures key genes PI3K-Akt p53 pathways identified species-specific characteristics uncertain role ERBB2 amplification CMTs mutations in benign tumors early genetic basis CMT pathogenesis gene expression-based molecular taxonomy revealed EMT-associated subtype in CMTs unfavorable prognosis study highlights molecular convergence oncogenic pathways supports potential use for human breast cancer in dogs with CMTs42.Domestic dogs shorter life expectancy (10–13 vs 79 tumorigenesis shorter (~10 shorter period cancer onset for unique features CMT paucity of aneuploidy SCNA drivers in CMT shorter development time late-stage Less-abundant mutation burdens CMT suggests mutational composition cancer development different across species simpler further investigation additional mutations cancer genomes represent neutral alterations or benefits CMT genomes showed uniform TMB levels sequence compositionsmutation indicated mutational settings CMT achieved at similar ages TMB levels comparable between benign malignant CMTs demonstrated for other tumor types colorectal two possibilities first optimal TMB level fixed for cancer cell malignant progression allows limited additional mutations supported by cross-species similarities in CMT human breast cancers PI3K-Akt pathway aberrations second possibility genetic programs favoring benign malignant disease determined early traditional stepwise mutations presence mutations enriched in malignant disease TP53 KRAS supports malignant progression CMT follow traditional stepwise evolution determining further investigation samples measures clone- gene-level selective forces MAF dNdScv malignant progression CMT involves selection events frequencies compositions somatic mutations independent malignant progression need research raised questions mechanisms for suppressing cancer development DNA repair cell cycle arrest immunosurveillance intact in CMT predominance of single type mutation signature (Signature #1) in CMT genomes majority of somatic mutations accumulated during lifetime host limited impact of other mutagenic sources in human breast cancers45APOBEC overactivity BRCA deficiency). no correlation observed (r = 0.04) between host age levels Signature #1 possible somatic mutations of CMT genomes arisen limited time period accumulated non-gradual with loss of DNA repair mechanisms identification of susceptibility factors for somatic mutations may lead to tumor-suppressing systems in dogs humans battle with cancer expanding to companion animals treatment canine cancer important anticancer agents dogs little evidence therapeutic efficacy cost effectiveness study showed trametinib MEK1/2 inhibitor effective in canine cancer cell expect more evidence on trans-species use drugs concern discordance between animal human drug efficacy toxicities47 treatment canine cancer may benefit from novel human cancer drugs shared oncogenic mutations alpelisib for metastatic breast cancer PIK3CA envision genomic studies cancer stratification companion diagnostics lead to efficient treatment canine cancers information study design sample collection data generation quality control strategies in separate data descriptor paper49 brief overview of data.Cohort design sample designed as tumor matched normal control cohort comparison genomic transcriptomic featuresblood coat mammary tissues controls for tumor DNA RNA 191 dogs with mammary tumors recruited veterinary clinics Korea consent owners Tumor tissues normal tissues blood collected guidelines Institutional Animal Care Committee Konkuk University (KU16106 KU17162 tissue samples transferred to RNAlater refrigerated overnight 4 °C stored at −80 °C Genomic DNA extracted from tumor tissue buffy coats using QIAamp DNA mini kits Total RNA extracted from tumor normal tissues RNeasy mini kits (Qiagen).HistopathologyFor sections (4-μm from formalin-fixed paraffin-embedded blocks stained with hematoxylin eosin diagnosed by researchers (B.J.S. J.H.S Histological subtyping World Health Organization malignancy ambiguous subtypes determined Rasotto et al Histological grade assessed Peña neoplastic epithelial component mammary osteosarcoma fibrosarcoma grade assessed system canine osteosarcoma53 cutaneous subcutaneous soft tissue sarcomaLymphatic invasion tumor cells in peritumoral lymphatic vessels regional lymph nodes determined.WES RNA 191 samples 183 tumor normal DNA subjected WES Two hundred nanograms fragmented DNA SureSelect Canine All Exon Kit DNA samples fragmented by Covaris adapter ligation purification hybridization PCR libraries examined Agilent 2100 Bioanalyzer loaded Illumina HiSeq sequencer 157 tumor tissues 64 normal tissues subjected RNA-seq RNA 6000 Nano kits RNA quality library 1 μg RNA purified oligo-dT magnetic beads Fragmentation purified mRNA double-stranded cDNAs synthesized primed with poly-A sequencing adapters connected TruSeq RNA sample prep kits Fragments filtered BluePippin 2% agarose gel cassettes PCR amplification Fragment lengths quality verified with Agilent High Sensitivity DNA kits Libraries observed average 392 bp standard deviation 66 bp WES RNA-seq Illumina HiSeq 2500 protocol out-sourced to Theragenetex Incsequencing reads aligned CanFam3.1 (Canis lupus familiaris genome BWA-MEM255 Duplicate fragments marked eliminated Picard 2.2 mapping low-quality paired information evaluated removing PCR duplicates indel realignment mate information base recalibration variant recalibration GATK4.0 CanFam3.1 (Ensembl Release 91) reference pipeline implemented in RNA-Seq data 157 tumor 64 normal mapped canine genome CanFam3.1-aware aligner TopHat57 (v Ensembl gene annotation fr-firststrand library type values calculated Cufflinks58 (v2.1.1 files variants calling annotationGATK-HC59 (v4.1.5.0 germline variants filtered VariantFiltration GATK4 excluding QD < 2.0 FS > 60.6 MQ < 40.0 ReadPosRankSum < −8.0 MQRanksum <2.5 >2.5 Variants ≥10 depth used analysis Variants liftovered from CanFam3.1 to hg38 Crossmap60 (v0.2.9 UCSC chain fileVariants liftovered to hg38 annotated using SnpEff61 ClinVar62 (build 2020-03-10) kept variants tagged truncated pathogenic likely pathogenic novel uncommon candidates with MAF < 0.05 in Dog Genome SNP Database) included in final germline variant list variant calling filtration detected single-nucleotide variations small insertions/deletions using Mutect264 GATK4 v4.0.10.1 VCF file reference bases CanFam3.1 REF field variants shown ALT field filtered falsely detected variants FilterMutectCalls selected PASS variants from VCF files Variant Effect Predictor dNdScv scores used R packages examined somatic mutations falsely detected by alignment errors rule out variant sites dubious alignment patterns-generated alignment scores optimaleach SNV position passing sequence reads alternative allele collected calculate alignment score optimal suboptimal site averaged samples mean alignment scores candidate SNVs likely alignment errors if meanXS greater 80% meanAS ≥ 0.8×meanAS).Tumor purity estimation TMB adjustmentTumor purity CMT genomes estimated histological examination H&E stained slides pathologist-J samples >70% neoplastic cells used analysis human breast cancers TCGA consortium obtained consensus purity estimates TMB genomes multiplied with correction factor purity purity-corrected TMB28 variant calculated depth of coverage using GATK XHMM workflow SAMTOOLS67 v1.9 VARSCAN68 v.2.4.3 used identify SCNAs mpileup function SAMTOOLS estimate bin-level sequencing read depth tumor normal BAM files ratios calculated GC-corrected applied circular binary segmentation (CBS)69 algorithm segmentation IGV browser visualization used GISTIC2 identify recurrent chromosomal gains losses CMT genomes71cross-species comparison SCNAs CMTs breast cancer alignment canFam3 hg19 assemblies blastz algorithm segment-level log2 ratios CMT genomes aligned hg19 blastz alignment hg19-aligned SCNA profiles smoothed segmented CBS algorithm-based CMT NMF deconvolution latent features CMT expression profiles CMT expression matrix basis matrix NMF signatures CMT expression profiles mammary tissues subjected NMF measured cophenetic score 2–10 NMF metagene signatures Five NMF metagene signatures derived three CMT tumors performed pre-ranked gene set enrichment analysis functional sets annotation prognostic evaluation NMF Kaplan–Meier survival curves drawn log-rank tests three tumor-specific NMF clusters molecular classification CMT transcriptomes PAM50 genes expression levels cancer CMT tumors merged hierarchical clustering correlations CMT transcriptomes expression tumor-initiating cells analyzed 154 up- 339 downregulated genes CD44+/CD24− tumor-initiating cells signature correlation levels calculatedValidation CMT subtypes human cohortFour expression profiles breast cancers (GSE17907 GSE20711 GSE25066 GSE31519) obtained clinical outcomes Patients cohorts discriminated high low NMF3 median NMF3 scores average expression level NMF3 survival high low NMF3 compared log-rank tests EMT scores average expression EMT/Hallmark MSigDB gene set.Tumor microenvironment ESTIMATE R package estimate scores proportion immune stromal cells CMT transcriptome abundance tumor-infiltrating immune cells CIBERSORT used set 22 immune cell subtypes40 Nature Research Reporting Summary.Supplementary information Peer Review File Additional Supplementary Files Data Summary | 49.7 | 1.006344 |
10.1038/s41467-020-18739-4 | PMC7532154 | Song et al. inferred that stridulatory wings and tibial ears co-evolved in a sexual context among crickets, katydids, and their allies, while abdominal ears evolved first in a non-sexual context in grasshoppers, and were later co-opted for courtship. They found little evidence that the evolution of these organs increased lineage diversification. | Acoustic communication is enabled by the evolution of specialised hearing and sound producing organs. In this study, we performed a large-scale macroevolutionary study to understand how both hearing and sound production evolved and affected diversification in the insect order Orthoptera, which includes many familiar singing insects, such as crickets, katydids, and grasshoppers. Using phylogenomic data, we firmly establish phylogenetic relationships among the major lineages and divergence time estimates within Orthoptera, as well as the lineage-specific and dynamic patterns of evolution for hearing and sound producing organs. In the suborder Ensifera, we infer that forewing-based stridulation and tibial tympanal ears co-evolved, but in the suborder Caelifera, abdominal tympanal ears first evolved in a non-sexual context, and later co-opted for sexual signalling when sound producing organs evolved. However, we find little evidence that the evolution of hearing and sound producing organs increased diversification rates in those lineages with known acoustic communication. | IntroductionAcoustic communication is one of the most conspicuous modes of signalling among animals. The use of acoustic signalling has been well documented in bony fishes, frogs, birds, cetaceans, terrestrial mammals and insects. Moreover, the intricate interplay and co-evolution between signal sender and receiver in the context of mating, prey location, predator avoidance and other interactions has led to the amazing diversity and complexity of the soundscape we know today1–4.Signal emission and reception are the two major components of acoustic communication, and are enabled by dedicated sound-producing organs and matching hearing sensory organs. Across the animal kingdom, different vertebrate and ‘invertebrate’ lineages have independently evolved diverse structures and mechanisms for hearing and sound production4–7. For example, although all inner ear structures of vertebrates can be traced to the same structure found in Silurian ostracoderms8, tympanal ears have evolved independently in frogs, mammals, and reptiles6. As for the sound-producing organs, vocal cords in the larynx have evolved several times within tetrapods6, while birds have evolved a unique organ called the syrinx9. As for insects, the ability to hear using tympanal ears has independently evolved at least in seven different orders (Orthoptera, Mantodea, Hemiptera, Neuroptera, Coleoptera, Lepidoptera and Diptera), involving at least 15 body locations10–13. Although the lack of tympanal ears does not necessarily mean that other insect orders cannot hear, as it has been shown that internal sensory organs can be sensitive to sound without external tympana14,15, the tympanal ears clearly enable far-field hearing over a broad frequency range and at high sensitivity14. The ability to produce sound that can travel over a long distance using specialised organs, such as stridulatory (vibration-producing) apparatus or tymbals, has evolved at least in six insect orders (Blattodea, Coleoptera, Hemiptera, Lepidoptera, Mantodea and Orthoptera), also involving many body parts6,11,16.While many studies have focused on the proximate mechanisms of hearing and sound production and the evolutionary processes driving the diversity of acoustic signalling1–3,10,17–20, questions about when, how, and in what context hearing and sound-producing organs evolved in the first place, and how these organs have co-evolved along the phylogeny remain inadequately addressed7,11,12. For insects that use acoustic signalling, there are at least two prevailing views on how these structures might have evolved originally11,12. The first view is that they could have evolved as an adaptation to detect and escape vertebrate predators7,11,12,21–23. Tympanal hearing may have evolved in the context of general auditory surveillance of the environment for predator movements, as it has been demonstrated in moths24,25, mantises26, and grasshoppers27. Likewise, early forms of stridulatory organs could have evolved as a defensive mechanism6, as part of a deimatic behaviour. These hearing and sound-producing organs could have also led to the evolution of sexual signalling via the so-called ‘sensory bias’ mechanism, in which male sexual signals evolve from structures originally involved in a non-sexual context that females already have perception for, also in a non-sexual context11. A phylogenetic pattern consistent with this sensory bias mechanism would be that, in a given lineage, the evolution of one component (e.g. hearing organ) would precede the evolution of its counterpart (e.g. sound-producing organ). The second view is that hearing and sound-producing organs could have evolved jointly as female perception and male signalling devices, co-evolved via a Fisherian mechanism11. It has been suggested that cicadas, crickets, and katydids evolved acoustic communication in this way7,11. A predictable phylogenetic pattern would be that the origin of both hearing and sound-producing organs would be traced to a single common ancestor. Thus, in order to gain deeper understanding of the evolution of acoustic communication, it is important to trace the evolution of hearing and sound-producing organs in a phylogenetic framework and in a lineage including both species lacking the ability to hear or produce sound and species with diverse acoustic communication strategies.Among animal groups that exhibit acoustic communication, the insect order Orthoptera (crickets, katydids, grasshoppers and allies) stands out as an ideal model to address these evolutionary questions7,11. With about 16,000 species primarily using acoustic signalling as a main mode of sexual communication, it is the most species-rich clade of all acoustically communicating animals, outnumbering frogs, birds, mammals28 or any of the known acoustically-active insect lineages. Furthermore, within Orthoptera, there are lineages that do not use acoustic signalling for mating but for defensive signalling29,30, and others lacking specialised structures for hearing or sound production11. Orthoptera is also the earliest known lineage of animals to have evolved complex acoustic communication, as evidenced by fossil forewings possessing a stridulatory apparatus homologous to that of present-day crickets, which is known from as early as the Triassic31,32. Therefore, Orthoptera is an excellent group to study the evolution of acoustic communication, but the lack of robust, time-calibrated phylogeny has been a major challenge for inferring the complex and dynamic patterns of how hearing and sound-producing organs originated and evolved over time.In this study, we reconstruct the evolution of hearing and sound-producing organs in Orthoptera, which can provide a bird’s-eye view of how acoustic communication originated and diversified during several hundred million years of evolution. We first establish reliable phylogenetic relationships among major lineages within Orthoptera by combining 4986 multiple sequence alignments of protein-coding genes selected from the transcriptomes of 60 taxa (50 orthopterans and 10 polyneopteran outgroups) and 249 previously and newly generated mitochondrial genomes (mtgenomes). We employ carefully selected fossils and rigorous topology testing to produce a robust, time-calibrated phylogeny of the order. This framework is then used to trace the evolution of tympanal ears and associated internal sensory organs, as well as that of diverse sound-producing mechanisms known in Orthoptera. This allows us to test evolutionary hypotheses regarding the origins of these organs and whether diversification patterns were influenced by these innovations. We find lineage-specific and dynamic patterns of evolution for hearing and sound-producing organs. Specifically, we infer that these two organs co-evolved in a sexual context in crickets, katydids and their allies, but we find little evidence that the evolution of these organs increased diversification rates in the singing lineages. Contrastingly, we find that the hearing organs evolved first in a non-sexual context in grasshoppers, and later co-opted for sexual signalling when sound-producing organs evolved.ResultsPhylogenetic relationships and divergence times of major orthopteran lineagesWe thoroughly explored the signal in the phylogenomic data by creating and analysing six phylogenomic data sets differing in the level of matrix saturation, character coding (amino acid vs. nucleotide), and data size (nuclear genes only vs. combined) in a maximum likelihood framework (see Supplementary Methods 1.1-1.6). The six data sets resulted in largely congruent topologies in terms of family-level relationships (see Supplementary Fig. 3), but the phylogenetic placements of Rhaphidophoridae, Gryllotalpidae and Pamphagidae varied among the resulting trees. We applied four-cluster likelihood mapping (FcLM)33 and permutation tests for these specific relationships using all six data sets to check for confounding signal, such as among-lineage heterogeneity that violates globally stationary, reversible and homogeneous conditions, non-random distribution of missing data, and a mixture of both (Supplementary Methods 1.7). We found that the placement of Rhaphidophoridae was robust and unbiased, but the placements of Pamphagidae and Gryllotalpidae were potentially biased by the confounding signal and our small taxon sampling for these families was not sufficient to make unambiguous conclusions about their relationships (see Supplementary Methods 1.7). Nevertheless, the ambiguous placements of these two latter families had little impact in inferring the evolution of hearing and sound-producing organs.Our analyses confirmed the monophyly of Orthoptera and its two suborders, Ensifera and Caelifera (Figs. 1, 2). Moreover, we recovered a comparatively ancient age for crown-Orthoptera, at ~355 million years ago (Mya) [95% credibility interval (CI), 393.8–320.0 million years (My)] (Fig. 1), which is ~63 My earlier than a previous estimate for this group34. We estimated crown-Ensifera to have appeared during the Late Carboniferous (308 Mya; CI, 348.0–267.4 My) (Fig. 1), which is consistent with the known fossil record, with the earliest stem-Ensifera being 272 million-years-old31,32. Our analyses recovered two monophyletic infraorders within this group, Gryllidea and Tettigoniidea, the former consisting of Grylloidea (including Gryllidae, Phalangopsidae, Trigonidiidae and Mogoplistidae), Gryllotalpidae and Myrmecophilidae, and the latter consisting of the remaining families (Figs. 1, 2). We inferred that crown-Gryllidea originated in the late Triassic or early Jurassic (200 Mya; CI, 247.5–154.1 My) (Fig. 1). Crown-Tettigoniidea originated in the Permian (268 Mya; CI, 308.1–227.7 My) and diverged into its major extant lineages throughout the Mesozoic (Fig. 1). Within Tettigoniidea, we recovered the following family-level relationships: (Rhaphidophoridae (Schizodactylidae ((Gryllacrididae (Stenopelmatidae + Anostostomatidae)) + (Prophalangopsidae + Tettigoniidae)))) (Fig. 2). We estimated that crown-Caelifera originated in the Carboniferous (320 Mya; CI, 359.5–282 My), and our analyses recovered two monophyletic infraorders (Figs. 1,2), Tridactylidea and Acrididea, the former consisting of Cylindrachetidae, Ripipterygidae and Tridactylidae, which diverged in the late Carboniferous, and the latter consisting of the remaining families. The more diverse Acrididea originated in the Late Permian (263 Mya; CI, 301.5–224.6 My) and split into two monophyletic groups, Tetrigidae and the superfamily group Acridomorpha (grasshopper-like insects) (Fig. 1). Most modern grasshopper diversity arose in the Cenozoic (Fig. 1). Additional details regarding the specific relationships within Orthoptera are described in Supplementary Methods 1.9.Fig. 1Dated phylogeny of Orthoptera based on the phylogenomic data.This chronogram is a result of a divergence time estimate analysis based on the most decisive data set (Daa,trans,strict) consisting of 436,488 aligned amino acids. Bootstrap support (BS) values are indicated by coloured nodes (green: BS = 100; yellow: BS = 96); values below 90 are not shown. Divergence time estimates were calculated using 86,043 amino-acid sites and 11 fossil calibrations (species names and dates listed on figure). Blue bars indicate 95% credibility intervals of node ages. Geological timescale is shown at the bottom. Additional details on data generation and analyses can be found in Supplementary Methods 1 and 2.Fig. 2Comprehensive phylogeny of Orthoptera.This phylogeny is estimated based on analyses of data from transcriptomes and mitochondrial genomes (Dnt,trans+mito,strict). The tree is derived from a maximum likelihood analysis of 448,861 aligned sites of nucleotides. Bootstrap support (BS) values are indicated by coloured nodes (green: BS = 100; yellow: BS = 90-99; orange: BS = 80-89). Red branches indicate the suborder Ensifera and blue branches indicate Caelifera. The red and blue clade names indicated by arrows (Gryllidea, Tettigoniidea, Tridactylidea and Acrididea) are infraorder names. The names in white, in red or blue bars are superfamily names. Broad circular bars are colour-coded by superfamily. TRIDAC Tridactyloidea, TETRI Tetrigoidea, EUMAST Eumastacoidea, PRO Proscopioidea, TA Tanaoceroidea, TR Trignopterygoidea, PN Pneumoroidea, GRYLLOTA Gryllotalpoidea, RHAPHID Rhaphidophoroidea, SCH Schizodactyloidea, HAG Hagloidea.Evolution of hearing and sound-producing organs in OrthopteraUsing ancestral character state reconstruction, we found lineage-specific patterns of evolution for hearing and sound-producing organs in Orthoptera (Fig. 3). In Ensifera, our analysis found that tegmino-tegminal stridulation likely evolved in the common ancestor of all extant lineages (Fig. 3). It is secondarily absent in Myrmecophilidae and Rhaphidophoridae, but as a consequence of the complete loss of wings in these families. Genuine loss occurred in Schizodactylidae and the common ancestor of Gryllacrididae, Stenopelmatidae and Anostostomatidae (Fig. 3), but in these families, abdomino-femoral stridulation evolved instead, known to produce defensive signalling against predators29,30. As for the hearing organs, we inferred that tibial tympana evolved at least three times in Ensifera (Fig. 3), once in the common ancestor of Gryllidea, once in the common ancestor of Anostostomatidae, and once in the common ancestor of Prophalangopsidae and Tettigoniidae. However, our analysis did recover a small probability that tibial tympana evolved in the common ancestor of Ensifera (Fig. 3), and thus, we could not completely rule out the possibility that the presence of tibial tympana was the ground plan for the suborder as well. We also examined the evolution of the complex tibial organ in the forelegs based on character mapping (Fig. 3), and found that the ancestral Ensifera had the tibial organ consisting of the subgenual organ (SGO) and the intermediate organ (IO), and the common ancestor of Gryllidea gained far-field hearing by evolving the tympanal organ (TO), which was modified from the IO7, and the tibial tympana. Rhaphidophoridae retained the ancestral configuration, but in the common ancestor of Tettigoniidea, a third component known as the crista acustica homologue (CAH) evolved. With the modification of this third component as the crista acustica (CA) and with the evolution of tibial tympana, the common ancestor of Prophalangopsidae and Tettigoniidae gained far-field hearing.Fig. 3Ancestral character state reconstruction of hearing and sound-producing organs.The topology used for this analysis is the comprehensive phylogeny based on Dnt,trans+mito,strict (presented in Fig. 2). The coloured circle at each branch tip indicates the character state of the corresponding species, with grey circles indicating the absence. The coloured circle at each node shows the probability of each ancestral character state. On the left, the character evolution of hearing organs is shown and the character states are colour-coded. In addition to the ancestral character state reconstruction, two additional traits are mapped. The first trait is the internal sensory organs in the ensiferan foretibia, shown in red. The ancestral condition for Ensifera is the presence of the subgenual organ (SGO) and the intermediate organ (IO). In the common ancestor of Gryllidea, IO was modified to tibial organ (TO), Rhaphidophoridae retains the ancestral SGO + IO. In the common ancestor of Schizodactyloidea, Stenopelmatoidea, Hagloidea and Tettigonioidea, a novel third component known as crista acustica homologue (CAH) evolved. In the common ancestor of Hagloidea and Tettigonioidea, CAH was modified to an auditory sensory organ called crista acustica (CA). The second trait is the loss of wings, which is indicated by black circles. Often, the species that lack tympanal hearing also have lost wings. On the right, the character evolution of sound-producing organs, in the form of stridulatory apparatus, is shown, and the character states are colour-coded. We used a specific naming convention in which the first-named structure has the stridulatory file and the second named structure has the scraper. For example, abdominal-femoral stridulation would have the stridulatory files on the abdomen and the scraper on the inner side of hind femora. Different mechanics of tegmino-tegminal stridulation are mapped onto the phylogeny. The common ancestor of Gryllidea evolved “left-over-right” stridulation, the common ancestor of Hagloidea evolved “ambidextrous” stridulation, and the common ancestor of Tettigonioidea evolved “right-over-left” stridulation. OG Outgroups, GRYT Gryllotalpoidea, GRYL Grylloidea, RHAP Rhaphidophoroidea, SCHI Schizodactyloidea, STEN Stenopelmatoidea, HAG Hagloidea, TETT Tettigonioidea, TRID Tridactyloidea, TETR Tetrigoidea, EUMAS Eumastacoidea, PROS Proscopioidea, TANA Tanaoceroidea, TRIG Trignopterygoidea, PNEU Pneumoroidea, PYRG Pyrgomorphoidea, ACRI Acridoidea.Within Caelifera, stridulatory organs evolved at least 10 times across the phylogeny based on our current taxon sampling, involving many different body parts (Fig. 3). However, the ability of these structures to produce sound remains largely unconfirmed, except for a few species that use acoustic signalling for mating or defence11,35,36, and thus, we must consider them putative for now. Definitive sound-producing organs used for mating evolved at least three times in Caelifera (Fig. 3), once in the common ancestor of Pneumoridae using abdomino-femoral stridulation, once in the common ancestor of Pamphagidae using Krauss’s organ-femoral stridulation, and once in the common ancestor of acridid subfamilies Acridinae, Gomphocerinae and Oedipodinae using hind femora and tegmina, although the location of the stridulatory file varies within these insects36. Our analysis showed that abdominal tympana likely evolved at least three times (Fig. 3), once in the common ancestor of Pyrgomorphidae, once in the common ancestor of Pamphagidae, and once in the common ancestor of Romaleidae, Ommexechidae and Acrididae. Similar to the case of tibial tympana in Ensifera, our ancestral character state reconstruction recovered a small probability that abdominal tympana evolved in the common ancestor of all these tympanate lineages (Fig. 3), and thus the presence of this structure could have been the ground plan for them as well.We performed Pagel’s37 binary character correlation test to determine whether hearing and sound-producing organs co-evolved within Orthoptera and within each of its suborders (Fig. 4). For all comparisons, we recovered statistically significant correlation between the two organs, but their co-evolutionary dynamics were different depending on the lineages (Fig. 4, Supplementary Methods 3.2). Considering Orthoptera as a whole, the best-supported model was that the evolution of hearing organs depended on the evolution of sound-producing organs (weighted AIC = 0.7685). Specifically, there were much higher instances of the absence of hearing organs when sound-producing organs were absent, and of the presence of hearing organs when sound-producing organs were present (Fig. 4). However, because taxon sampling is known to affect correlation analyses38 and because the known patterns of acoustic communication are very different between Ensifera and Caelifera11, we examined the patterns of character correlation for each suborder, which delivered highly contrasting patterns (Fig. 4). For Ensifera, the model that the evolution of sound-producing organs depended on that of hearing organs (weighted AIC = 0.4490) and the model that the evolution of hearing organs depended on that of sound-producing organ (weighted AIC = 0.4343) similarly explained the pattern. Of all possible interactions between the two organs, we found magnitudes higher instances of the presence of sound-producing organs when hearing organs were present (Fig. 4), which indicates that nearly all ensiferans that can hear also produce sound, suggesting an extremely high correlation between the two traits. For Caelifera, the model that the evolution of hearing organs depended on that of sound-producing organs (weighted AIC = 0.4361) best explained the data, but the model that the evolution of sound-producing organs depended on that of hearing organs (weighted AIC = 0.3567) also reasonably explained the data. We found that there were higher instances of the absence of sound-producing organs regardless of the presence or absence of hearing organs in Caelifera (Fig. 4), almost an opposite pattern from what we found in Ensifera.Fig. 4Evolutionary correlation between hearing and sound production in Orthoptera.Pagel’s test for evolutionary correlation was calculated between hearing and sound production in Orthoptera and its two suborders, Ensifera and Caelifera. The thickness of arrows corresponds to the rate of change from one combination of trait states (i.e. no hearing and no sound production) to another combination (i.e. no hearing and sound production). The higher the rate, the thicker the arrow. In each of the three analyses, there is a strong evolutionary correlation between hearing and sound production, but the patterns are different. In Orthoptera as a whole, the strongest transition rate is from hearing present & sound production absent to both hearing and sound production absent. In Ensifera, the strongest rate is from hearing present & sound production absent to both hearing and sound production present. In Caelifera, the strongest transition rate is from hearing present and & sound production absent to both hearing and sound production absent. The differences between Ensifera and Caelifera show that the co-evolutionary dynamics between hearing and sound production differ between the two lineages.Rates of lineage diversification in relation to acoustic communicationThe Bayesian analysis of macroevolutionary mixtures (BAMM)39 on our larger data set found three episodes of rate shift along the phylogeny of Orthoptera (Fig. 5). The first episode of rate shift took place in the common ancestor of Tettigoniidae, during the Cretaceous, with a mean clade-specific evolutionary rate for the family (0.08186125) nearly double the background rate for Orthoptera (0.04820052) as well as for Ensifera (0.04996942) (Fig. 5). The second episode of rate shift took place in the common ancestor of Pamphagidae, during the late Cretaceous and the early Paleogene, with a mean clade-specific rate (0.1290903) almost tripling the background rate for Orthoptera as well as for Caelifera (0.04905859), which was retrieved as the highest evolutionary rate among all orthopteran lineages (Fig. 5). The third episode of rate shift took place in the common ancestor of Romaleidae, Ommexechidae and Acrididae, during the late Cretaceous and throughout the Paleogene, with a mean clade-specific rate (0.07159634) slightly higher than the background rate (Fig. 5). Interestingly, however, we found that other singing lineages within Ensifera, namely Grylloidea, Gryllotalpidae and Prophalangopsidae, did not show any discernible rate shift (Fig. 5). We tried not to over-interpret the recovered patterns from this analysis, because the appropriateness of BAMM in diversification analyses has been questioned, especially concerning its ability to accurately estimate diversification rates40 although the developers of BAMM have argued that these criticisms were unjustified41.Fig. 5Bayesian analysis of macroevolutionary mixtures for Orthoptera.The ultrametric tree used in this analysis is the dated phylogeny based on the combined data (Dnt,trans+mito,strict). a Phylorate plot showing speciation rates (cool colours = slow, warm = fast; specific rate range shown in the vertical colour legend) along each branch of the Orthoptera phylogeny. The three clades indicated by the black circled nodes are the clades with increased rate shifts. Lineages that show acoustic communication are indicated with vertical lines near the terminals. Branches are coloured according to the rate shifts. b About 95% credible sets of macroevolutionary shift configurations. f value of 0.21 indicates that 21% of the samples in posterior can be assigned to shift configuration shown in the upper left plot. These four shift configurations collectively account for 51.5% of the posterior distribution. c Clade-specific evolutionary rate variation through time for Orthoptera and the three lineages identified to have rate shifts.We fitted various models of trait-dependent and trait-independent diversification using HiSSE (Hidden State Speciation and Extinction)42 to test whether the evolution of hearing and sound-producing organs affected speciation and extinction rates of different orthopteran lineages (Fig. 6). For the hearing organs, the best-fitting model, according to AIC scores, was one of the HiSSE models, which suggests character-dependent diversification where all the diversification parameters are free and where transitions between hidden states of hearing-organ-absent and hearing-organ-present were disallowed (HiSSE, q0B1B = 0, q1B0B = 0, all other q’s equal). We found a higher net diversification rate associated with the presence of hearing organs, which is likely due to a higher diversification rate of the hidden state (Fig. 6). The net diversification rate associated with the absence of hearing organs was relatively lower. For the sound-producing organs, the best-fitting model was one of the CID (trait-independent) models, which assumes that the evolution of a binary trait (presence or absence of sound-producing organ) is independent of the diversification process without forcing the diversification process to be constant across the tree (CID-4: q’s equal) (Fig. 6). In other words, the selected model suggested that the evolution of sound-producing organs did not affect the net diversification rate. When acoustic communication was coded as a binary trait, the best-fitting model was the identical trait-independent model selected for the sound-producing organs (CID-4: q’s equal), which suggested that diversification process was independent from the evolution of acoustic communication (Fig. 6).Fig. 6Models of trait-dependent diversification.Character reconstruction of states and net diversification rates estimated using multimodel inference methods implemented in hisse. Shown here are the best-fitting models for each tested trait (presence/absence of hearing organs, of sound hearing organs and of acoustic communication) from the 24 models of trait-dependent and trait-independent diversification models. All clades that are characterised by having sexual communication using acoustic signalling are labelled in the circular trees, and estimates of the most likely state and rate are based on the model-averaged marginal reconstructions inferred under the best-fitting models. The histograms inside the trees show the location of the rates on a gradient of rates, as well as the frequency of both these rates and states for each contemporary tip taxa. For hearing organs, the best-fitting model was one of the HiSSE models, but for both sound-producing organs and acoustic communication, the best-fitting model was one of the trait-independent diversification models (CID-4).DiscussionOrthopteran insects, such as crickets, katydids and grasshoppers, have been model systems for studying acoustic communication for decades2,7,11,18,30,43–46, but how hearing and sound-producing organs originated and evolved throughout the diversification of these insects has remained elusive due to the lack of a well-resolved phylogeny. This work firmly establishes phylogenetic relationships among the major lineages and divergence time estimates within Orthoptera based on phylogenomic data and carefully selected fossil calibration points. We find that crown-Orthoptera likely originated 355 million years ago, and diverged into Ensifera and Caelifera in the Carboniferous (Fig. 1). Our study suggests that these two suborders have each followed very different, lineage-specific patterns of evolution for hearing and sound-producing organs (Figs. 3, 4).Hearing and sound-producing organs co-evolved in EnsiferaEnsifera, the larger of the two suborders, encompasses ~15,500 extant described species, many of which are nocturnal and use acoustic signalling as a primary mode for sexual communication. The singing ensiferans include four extant lineages (crickets [Grylloidea, including Gryllidae, Phalangopsidae, Trigonidiidae and Mogoplistidae], mole crickets [Gryllotalpidae], katydids [Tettigoniidae] and grigs [Prophalangopsidae]), which have specialised hearing sensory organs in the form of tympanal ears located on front tibiae and a stridulatory apparatus on male tegmina (forewings)3. They account for nearly 85% of the ensiferan diversity3,17,47,48. The remaining ensiferan lineages have neither tibial tympana nor stridulatory tegmina (ant-loving crickets [Myrmecophilidae] and cave crickets [Rhaphidophoridae]), or lack tibial tympana but possess a stridulatory apparatus on the abdomen, used for defensive signalling29,30,49 and present in both sexes and in nymphs (splay-footed crickets [Schizodactylidae], raspy crickets [Gryllacrididae], Jerusalem crickets [Stenopelmatidae] and some king crickets and wetas [Anostostomatidae]). The monophyly of Ensifera has been consistently supported by all modern cladistic analyses34,43,50,51, and most researchers agree that the suborder consists of two monophyletic infraorders, Gryllidea and Tettigoniidea34,51, which our study also confirmed (Figs. 1,2). However, there has not been a consensus on the internal relationships among families and superfamilies, as different phylogenetic studies utilising different character systems (morphology, ribosomal RNAs or mtgenomes) disagreed with each other34,43,50,52, leading to conflicting inferences about the evolution of acoustic communication7,11,43,53. Especially, whether the stridulatory apparatus evolved once or multiple times has been contentious32,43,53. Our phylogenomic analysis recovered strongly supported relationships among the families (Figs. 1,2), which are more congruent with a morphology-based phylogeny43 than with the previous molecular studies34,50–52. Based on the recovered topology and divergence time estimates (Figs. 1,2), as well as ancestral character state reconstruction (Fig. 3), we can infer the following evolutionary scenario regarding how hearing and sound-producing organs might have evolved in Ensifera.Between the late Carboniferous and the early Permian, crown-Ensifera diverged (Fig. 1) and male-specific tegmino-tegminal stridulation evolved in the common ancestor of Ensifera (Fig. 3), which represents one of the earliest occurrences of airborne sound generation in animals. Although the oldest fossil ensiferans (such as Gryllavus and Protogryllus) with a well-preserved stridulatory apparatus homologous to the present-day structure are known from the Triassic31,32, our finding suggests that a similar mechanism of sound production could have evolved much earlier. The earliest insectivorous tetrapods appeared in the early Carboniferous, and these animals did not have tympanic ears6. It has been hypothesised that these predators would have been deterred by stridulation produced by insect prey upon seizure, which would have stimulated their tactile receptors that caused them to release the prey6,54,55. In a sense, stridulation could have originally evolved as part of a deimatic behaviour56. Fossil evidence shows that specialised sound-producing organs involving wings were present among the Permian and Triassic stem-Orthoptera31,32,57,58. If we accept the possibility that the ability to move wings to produce sound was an ancient invention during the early diversification of Orthoptera, it is conceivable that this behaviour could have been co-opted for sexual communication, possibly in parallel within several lineages. For instance, the stem-orthopteran lineage Titanoptera had modified veins in the forewings highly indicative of sound production, present in both sexes57,58 and possibly used for pair formation via reciprocal duetting31,59. This group evolved from the Permian ‘tcholmanvissiids’59, which lack specialised forewing sound-producing organs, if any. Another contemporaneous lineage of stem-orthopterans, the Mesoedischiidae, had male-specific tegmino-tegminal stridulation, although the specific veins modified for sound production were not homologous to those in extant Ensifera57,58. Among the four extant singing ensiferan lineages, the specific mechanics of tegmino-tegminal stridulation are known to differ43. Crickets and mole crickets stridulate by moving the left forewing over the right, and katydids stridulate in the opposite way by moving the right forewing over the left43. Grigs are able to stridulate by moving their wings both ways43. Moreover, a recent comparative morphological analysis proposed that the stridulatory apparatus involved different veins of the forewing in these four lineages53, although the case remains debated. Regardless, it can be argued that the muscular mechanics and associated neurocircuit enabling male-specific tegmino-tegminal stridulation are phylogenetically conserved and potentially plesiomorphic in Ensifera, but different lineages independently evolved different ways of creating audible sound, building on the same physiological mechanism.Our analysis suggests that tegmino-tegminal stridulation was secondarily lost in several ensiferan lineages (Fig. 3), and this loss is often associated with adaptations to novel environments that promote the loss of wings. For example, extant members of Rhaphidophoridae are completely apterous and often associated with caves60. Similarly, members of Myrmecophilidae are wingless and intimately associated with ant colonies61. Many members of Schizodactylidae, Gryllacrididae, Stenopelmatidae and Anostostomatidae are specialists on subterranean habitats and wingless as well62. However, each of these four latter families includes some species with fully functional tegmina lacking stridulatory apparatus63. It has been documented that several cricket and katydid species have secondarily lost the ability to sing47,48, and one well-documented case, that of the Hawaiian cricket Teleogryllus oceanicus, demonstrates that the loss of stridulatory apparatus repeatedly and convergently evolved due to a strong selective pressure from an introduced phonotactic parasitoid fly, Ormia ochreacea64. This loss has a genetic basis in the form of a simple alteration of a master regulatory switch during early development that can lead to the dramatic change in the adult phenotype65. While it is difficult to attribute the same process to explain the loss of tegmino-tegminal stridulation in the non-singing ensiferans, we conclude that this loss of complex trait could have been achieved easily multiple times during the diversification of Ensifera.Interestingly, many of the non-singing ensiferans are known to engage in some type of intraspecific communication using substrate-borne vibration, drumming using abdomen or legs, or tremulation (shaking without any substrate)29,49,63,66,67, and have well-developed chordotonal organs for sensing vibration7,68. Our analysis also shows that abdomino-femoral stridulation likely evolved at least twice (Fig. 3), once in the common ancestor of Schizodactylidae and once in the common ancestor of Gryllacrididae, Stenopelmatidae and Anostostomatidae. This mechanism is found in both sexes as well as in nymphs, and it is not used for sexual communication, but for producing defensive signal against predators29,30. These patterns collectively suggest that the loss of tegmino-tegminal stridulation could have promoted the evolution of both vibratory signalling in a sexual context, and an alternative acoustic signalling in a non-sexual context in these non-singing lineages.For hearing, it is less clear whether the first hearing organs also evolved in the common ancestor of Ensifera based on our current data. Hearing organs in the forelegs are complex organs consisting of external tympana as well as internal complex tibial organs7,14,69. It is unclear what the original form of sound detection was in the ancestral ensiferans, but it is conceivable that thin cuticle of foretibia could have initially functioned as a resonator for the internal sensory organs to pick up sound wave. We infer that thinning of the cuticle evolved at least three times to give rise to tympanal membrane within this group (Fig. 3). However, the neurophysiological mechanisms underlying hearing independently evolved twice (Fig. 3), leading to two different types of hearing sensory organs, SGO + TO found in crickets and mole crickets, and SGO + IO + CA found in katydids and grigs. These results are consistent with the idea that the common ancestor of Ensifera probably did not have the structures enabling far-field hearing, but different lineages independently evolved far-field hearing. Although it is generally hypothesised that the early form of hearing in insects evolved in the context of detecting and avoiding predators11,12,25, the specific position where tympana evolved in the singing ensiferans raises an intriguing possibility that hearing in Ensifera could have evolved in another context. Extant ensiferan ears usually have two auditory inputs, with sound arriving at the external surface of the tibial tympana, and also internally via the acoustic trachea, which open on the acoustic spiracles at the side of the pronotum44,70–72. These ears are pressure difference receivers73,74, as the sound travelling internally on the trachea travels slower and a longer distance than that reaching the external surface of the tympanum from the outside at the normal speed of sound propagation in air. This causes differences in gain between sound arriving externally and internally72,74,75. This complex acoustic tracheal system also shows lineages-specific differences. In crickets and mole crickets, acoustic trachea connect all four sound inputs with an enlarged part in its midline, accompanied by two thin septa originating from each trachea72,76. In katydids and grigs, the acoustic trachea starting at the acoustic spiracles do not connect in the middle, such that the trachea starting from the right and left acoustic spiracles connect to the right and left tibial tympana, respectively70,75,77. In katydids, the tracheae are enlarged as acoustic bullae at the acoustic spiracles and gradually narrow as they approach the tympanal ears47,70. Therefore, we conclude that this elaborate directional hearing mechanism evolved independently in the context of accurately locating the source of conspecific calls.Sound-producing organs and hearing organs evolved separately in CaeliferaCaelifera is the other of the two orthopteran suborders, with ~12,200 extant species, and consisting of familiar insects such as grasshoppers and locusts, as well as the lesser-known pygmy mole crickets, pygmy grasshoppers, monkey grasshoppers, stick grasshoppers, and their relatives34. Sexual communication using acoustic signalling is a relatively rare feature across Caelifera, which has only been documented in a small number of divergent families (bladder grasshoppers [Pneumoridae], pamphagid grasshoppers [Pamphagidae], tooth-legged grasshoppers [Acrididae: Gomphocerinae] and banded-wing grasshoppers [Acrididae: Oedipodinae])11,78. Our literature survey shows that these lineages each use different stridulatory mechanisms to produce sound (Fig. 3), but they all involve rubbing hind femora up and down against other body parts, such as thickened veins on tegmina or specialised areas on the abdomen. We find that most of the early-diverging caeliferan lineages do not have hearing organs, and tympanal hearing is only found in a few grasshopper families (Pamphagidae, Pyrgomorphidae, Romaleidae, Ommexechidae and Acrididae) (Fig. 3), which originated in the Cretaceous and the Paleogene (Fig. 1). When present, tympana are located on both sides of the first abdominal segment, which usually have large tympanal membranes that are innervated with the auditory sensory organs, and externally encircled by sclerotised rings, with air-filled tracheal sacs internally positioned between the tympanal membranes45. According to our phylogenomic analysis (Fig. 2), which recovered relationships that are largely congruent with previous studies34,51,79,80, hearing and sound-producing organs in Caelifera did not evolve jointly, but followed different evolutionary trajectories (Fig. 3). There is no fossil evidence to suggest the antiquity of hearing or sound production in Caelifera, and we deduce that acoustic communication is generally a more recent invention in Caelifera compared to Ensifera.Our study shows that, throughout the diversification of Caelifera, several lineages evolved paired structures equipped with a stridulatory file on one body part and a scraper on another body part (Fig. 3), which involve mouthparts, forewings and hindwings, middle legs and hind legs and abdomen. However, it is largely unconfirmed whether these paired structures are actually used for sound production, except for the aforementioned families that use acoustic signalling. It is also not clear in what context these structures evolved. For example, these structures are found in both sexes and in nymphs in some lineages (e.g. mandibulo-maxillary stridulation found in Cylindrachetidae)81, which could have evolved in the context of defence. Similarly, these putative sound-producing organs are found only in males in some lineages (e.g. abdomino-femoral stridulation found in Tanaoceridae)82, which could have evolved in a sexual context. In other words, there is much to be learned in terms of the diversity, mechanisms and functions of sound production in Caelifera. Intriguingly, none of the caeliferans is known to engage in tegmino-tegminal stridulation, which is the primary and phylogenetically conserved sound-producing mechanism in Ensifera. This implies that the neurophysiological machinery enabling tegmino-tegminal stridulation was never part of the caeliferan ground plan.Our study finds that the first form of sexual communication using acoustic signalling in Caelifera likely evolved in the common ancestor of the South African family Pneumoridae, in the Jurassic (Figs. 2,3). By this time, complex acoustic signalling was already well-established in Ensifera. Extant bladder grasshopper males, which are fully winged, have an inflated abdomen that functions as a resonating chamber to produce loud low-frequency calls that can travel up to 2 km using abdomino-femoral stridulation83. In response to male calling, receptive females, which are flightless, indicate their willingness to mate by acoustically responding, which leads to pair formation via reciprocal dueting84. Female sound-producing organs are not homologous to those in males and different species use different body parts to create sound (V. Couldridge, personal communication). Interestingly, both males and females lack tympanal ears, and instead have chordotonal organs innervating each abdominal segment, and as such, the entire abdomen functions as a hearing organ46. This pattern suggests that there could have been a selective pressure for evolving acoustic communication as early as the Jurassic, but perhaps because dedicated directional hearing organs did not yet evolve. These lineages never radiated like their ensiferan counterparts did.It was not until the Cretaceous that abdominal tympana appeared in Caelifera (Figs. 1,3). Our finding is more consistent with the idea of multiple origins of abdominal tympana, although we did recover a small probability that the common ancestor of Pyrgomorphoidea and Acridoidea could have evolved abdominal tympana once (Fig. 3). An intermediate option would involve a rather unspecialised, early form of abdominal hearing organ which might have then undergone parallel evolution, towards proper abdominal tympana, within the different lineages. The context in which these hearing organs evolved is not clear. Grasshoppers with abdominal tympana generally show jumping or flying behaviour upon hearing approaching sound27, which indicates that its current function is most likely for detecting predators or disturbances, and this is indeed the most commonly invoked hypothesis on the origin of grasshopper ears11,23. However, insectivorous predators were already well-diversified by the Cretaceous6 and it is unlikely that a sudden and strong selective pressure triggered the evolution of predator-detection hearing. There were also other caeliferan lineages that radiated without evolving hearing, such as Tetrigoidea and Eumastacoidea, and these insects faced predators, yet succeeded without tympana. Given that most grasshopper species with abdominal tympana do not have sexual communication using acoustic signalling, it is also difficult to think that hearing evolved in a sexual context. One alternative explanation comes from our observation that secondary loss of abdominal tympana is often found in those species that evolved wing reduction or loss85, which suggests that there could be a connection between flight and hearing. The physiological mechanisms of the auditory pathway in grasshoppers and locusts have been intensely studied45, and it has been shown that auditory information processing through abdominal tympana is in fact intimately influenced by thoracic muscle movement and wingbeat noise during flight86. Although the ability to fly is a plesiomorphy for Orthoptera, Pyrgomorphoidea and Acridoidea are the first large-bodied caeliferans with an exceptionally strong dispersal capacity, which raises an intriguing possibility that abdominal tympana could have originally evolved in the context of modulating flight, rather than detecting disturbances or locating mates. This idea is indirectly bolstered by the pattern that many brachypterous katydids and crickets still retain the ability to hear through tibial tympana47,48, which are probably not involved in flight modulation.The evolution of abdominal tympana in early grasshoppers could have led to the evolution of sexual signalling under the ‘sensory-bias’ mechanism11, which we think was achieved by the independent evolution of sound-producing organs in two grasshopper lineages, Pamphagidae and a monophyletic group within Acrididae consisting of Acridinae, Gomphocerinae and Oedipodinae (Fig. 3). However, we find that the path to evolving acoustic communication differed considerably between the two. Pamphagidae is a large-bodied family that originated in the Cretaceous (Fig. 1). Similar to the bladder grasshoppers, pamphagid grasshopper engage in pair formation via reciprocal dueting78, and males are often fully winged and females are flightless, although the loss of wings is quite common in this family35. Our study finds that Krauss’s organ-femoral stridulation is a phylogenetically conserved mechanism of sound production for the family (Fig. 3). The Krauss’s organ is a specialised plate located on the lower anterior corners of the second abdominal tergite, which is rubbed by the ridges inside hind femora87. This mechanism is present in both males and females, and the sound produced by this mechanism is species-specific35. Although not included in our taxon sampling, many pamphagids are also known to utilise other types of sound-producing mechanisms for mating, involving abdomen, hind femora, forewings, hindwings, middle tibiae and thorax35,78. These collectively suggest that the evolution of sound production occurred in the common ancestor of Pamphagidae, which already had the ability to hear, and this could have led to the elaboration of acoustic communication in the entire lineage.On the other hand, sound production evolved much later in Acrididae, after the lineage has already diversified (Fig. 3). We find that the presence of abdominal tympana is plesiomorphic for the family (Fig. 3), and the male-specific stridulatory mechanism using tegmina and hind femora likely evolved between the Eocene and the Oligocene in the common ancestor of Acridinae, Gomphocerinae and Oedipodinae, likely in a sexual context. However, even within this lineage, the sound-producing organs followed different evolutionary trajectories in terms of specific modifications of the stridulatory apparatus. For example, in Gomphocerinae, stridulatory pegs are located on the hind femora, which rub against the thick veins in the forewings, whereas in Oedipodinae, a row of stridulatory files on the intercalary veins in the forewing rubs against the scrapers in the hind femora36. In addition to the stridulatory signalling, Oedipodinae and some members of Acridinae evolved an alternative and non-stridulatory acoustic mechanism, called crepitation, which produces sound by snapping wings when they fold and unfold36. In all these grasshoppers, acoustic signalling is often complemented with visual signalling, such as leg movements, characterising a multimodal sexual selection36. Thus, acoustic signalling found in Acrididae represents the most recently evolved form of sexual communication within Orthoptera.Evolution of acoustic communication did not influence diversification rates in OrthopteraWe have shown that the evolution of sexual communication using acoustic signalling in Ensifera and Caelifera each followed a very different trajectory (Figs. 3, 4). In Ensifera, we infer that tegmino-tegminal stridulation was an ancestral feature that could have evolved as defensive signalling in crown-Orthoptera, and different lineages independently evolved tibial tympana in a sexual context. In each common ancestor of the singing lineages, both hearing and sound-producing organs were present, allowing the Fisherian mechanism to shape the co-evolution between female perception and male signalling devices11. Our Pagel’s binary character correlation test found overwhelming evidence that hearing and sound-producing organs co-evolved in Ensifera (Fig. 4), supporting this hypothesis. In Caelifera, abdominal tympana evolved later in the lineage diversification (Fig. 3), possibly in the context of modulating flight in large-bodied grasshoppers, which was later co-opted for detecting predators, and again co-opted for sexual communication when male-specific sound-producing organs evolved independently in different lineages. This pattern fits well with the ‘sensory bias’ mechanism. The Pagel’s test found little support for the co-evolution between hearing and sound-producing organs (Fig. 4), thus supporting the alternative hypothesis. Having established these evolutionary mechanisms, we now ask whether the evolution of hearing and sound-producing organs affected diversification rates in different lineages that use acoustic signalling in a sexual context.It is generally accepted that sexual selection is a major driving evolutionary force shaping the diversification of singing insects88,89, and theory predicts that sexually selected traits tend to evolve rapidly88,90. Especially, if the inferred mechanism for the evolution of hearing and sound-producing organs is the Fisherian mechanism, we would expect an elevated diversification rate in a clade that is characterised by sexual communication using acoustic signalling89. This idea was recently tested in tetrapods but, surprisingly, it was found that acoustic communication did not increase diversification rates in these animals4. To test this proposal in Orthoptera, we first performed a diversification analysis using BAMM39 to determine clade-specific evolutionary rates. Among the lineages with known acoustic communication, we find that Tettigoniidae was the only lineage within Ensifera to show an increased mean clade-specific evolutionary rate, while other lineages (Grylloidea, Gryllotalpidae and Prophalangopsidae) did not show any discernible rate shifts. Likewise, Pamphagidae was the only lineage within Caelifera with an increased mean clade-specific evolutionary rate, and neither Pneumoridae nor the monophyletic group consisting of Acridinae, Gomphocerinae and Oedipodinae showed any rate shifts. Rate shifts are usually associated with key innovations leading to increased diversification rates39, which would indicate that the evolution of sexual communication using acoustic signalling was not necessarily the major key innovation for all of these singing lineages. It is conceivable that both Tettigoniidae and Pamphagidae did experience the increased diversification rates due to their evolution of acoustic signalling, which is the most widespread and dominant mode of communication in these lineages35,47 and other forms of signalling (visual or chemical) are not known. However, it is also possible that, at least for Tettigoniidae, there could have been other key innovations, such as impressive leaf masquerade and diverse feeding habits91, that might have led to the rate shift possibly related to the contemporaneous rise of angiosperms. These findings are bolstered by a more direct analysis of trait-dependent diversification using HiSSE42. Regardless of models used, the lineages that evolved hearing and sound-producing organs, as well as the lineages with confirmed acoustic communication had higher net diversification rates. However, when the multimodel inference method was used, the best-fitting models collectively suggest that the evolution of hearing organs affected the net diversification rate, but both the evolution of sound-producing organs and the evolution acoustic communication were independent of the diversification processes, and thus did not affect the net diversification rate. Therefore, our study finds a pattern consistent with what was shown in tetrapods4 in that we find little evidence that acoustic communication alone increased net diversification. Our results have a major implication in enhancing our understanding of signal sender-receiver co-evolution and diversification in Orthoptera, revealing more general insights about the evolution and mechanisms of animal communication.MethodsPhylogenomic analyses and divergence time estimationOur taxon sampling consisted of 239 species of Orthoptera and 10 polyneopteran outgroups, totalling 249 species. Together, these data represented all 16 superfamilies and 36 families of extant Orthoptera. We included 60 transcriptomes, of which 39 orthopteran species were newly generated either by the 1K Insect Transcriptome Evolution (1KITE) consortium or by the Song Lab at Texas A&M University. The remaining 21 transcriptomes (11 orthopteran and 10 polyneopteran) were from the previous publications (see Supplementary Methods 1.1). To increase taxon sampling, we then combined the transcriptome data with 169 previously and 80 newly generated mtgenomes from 249 taxa. RNA extraction, cDNA library preparation and transcriptome sequencing and assembly were performed within the 1KITE project using the protocols detailed in Supplementary Methods 1.2 and 1.5. Protocols used for the Song Lab samples, as well as for mtgenome data generation are detailed in Supplementary Methods 1.3 and 1.4. A detailed list of all species, including their collection data and National Center for Biotechnology Information (NCBI) accession numbers, is presented in Supplementary Data 1 and 2.For the transcriptome data, a custom-made orthologous gene set was designed with OrthoDB v7 (ref. 92) using four hemimetabolous (Zootermopsis nevadensis, Pediculus humanus, Acyrthosiphon pisum and Rhodnius prolixus) and one holometabolous (Nasonia vitripennis) species, which resulted in 5414 protein-coding genes. We used Orthograph v0.5.3 (ref. 93) to generate a profile hidden Markov model (pHMM) from the amino-acid sequences of each reference gene, which was used to search for ortholog candidates in transcript libraries. Orthograph ran protein BLAST (blastp) search using the translated query protein sequences against a database of all amino-acid sequences from all the reference orthologous genes/groups (OGs). For each pHMM hit transcript, the corresponding BLAST result was checked whether the best hit sequence belonged to the OG that the pHMM is based on. Only if the sequence matched, the best-reciprocal hit criterion was fulfilled and the OG was extended with the candidate transcript. Using these methods, we identified on average 3700 OGs. The amino-acid sequences of these OGs were individually aligned on amino-acid level using MAFFT v7.130b94 with the L-INS-i algorithm, and the quality of the multiple sequence alignment (MSA) was checked using the pipeline described in Supplementary Methods 1.5.For downstream phylogenetic analyses, we considered regions identified as protein clans, families, single domains or non-annotated regions as evolutionary units in the partitioned analyses. The methods for identifying these evolutionary units are detailed in Supplementary Methods 1.5. Using custom Perl scripts, the results from the protein domain identification step and the identified randomised MSA sections were merged into a masked supermatrix. The total alignment length spanned 1,647,472 amino-acid positions, and a back-translated nucleotide supermatrix was created using several custom-made Perl scripts. MARE v0.1.2-rc95 was used to assess the information content (IC) of each data block, and all identified data blocks showing an information content of 0 (IC = 0) were removed from the supermatrices. From these data, we created four transcriptome data sets: (i) Daa,trans,complete, 1,541,865 aligned amino acids with 1743 domain-based metapartitions; (ii) Daa,trans,strict, 436,488 aligned amino acids with 102 metapartitions with 100% matrix saturation; (iii) Dnt,trans,complete, a corresponding data set of Daa,trans,comple, comprising 1,541,865 aligned sites of second codon positions only; and (iv) Dnt,trans,strict, a corresponding data set of Daa,trans,strict, comprising 436,488 aligned sites of second codon positions only. In order to select the most appropriate number of partitions for these data sets, we used PartitionFinder 2.0 (ref. 96) in combination with the provided RAxML version. For the mtgenome data, we created a concatenated matrix of nucleotide sequences consisting of 13 protein-coding genes aligned based on the conservation of reading frames using MUSCLE97 and divided the data matrix into a total of 39 data blocks (13 mitochondrial protein-coding genes divided into individual codon positions). We used PartitionFinder to search for the best-fit scheme as well as to estimate the model of nucleotide evolution for each partition. We then combined the transcriptome data with the mtgenome data by concatenating the Dnt,trans,complete and Dnt,strans,strict data sets each with the aligned matrix of mtgenomes of 249 taxa, 60 of which overlapped with the transcriptome data: (v) Dnt,trans+mito,complete, comprising 1,554,238 aligned sites of nucleotides with 1766 metapartitions; and (vi) Dnt,trans+mito,strict, comprising 448,861 aligned sites of nucleotides with 125 metapartitions. Additional details on data set preparation are presented in Supplementary Methods 1.5.We analysed these six data sets in a maximum likelihood framework using IQ-TREE v1.5.4 (ref. 98) with the best-scoring substitution matrix for each partition. We performed 50 independent tree searches for each data set and node support was estimated via non-parametric bootstrapping of 100 bootstraps replicates in IQ-TREE and mapped onto the ML tree with the best log-likelihood. We also determined support for specific phylogenetic relationships using four-cluster likelihood mapping (FcLM)33 by selecting incongruent nodes based on the tree inferences of the six data sets and additionally checking for confounding signal due to among-lineage heterogeneity, non-random substitution processes and/or distribution of missing data with permuted data sets with phylogenetic signal destroyed.To estimate divergence times, we first conducted a thorough review of available fossils to identify potential calibration points (see Supplementary Methods 2). We applied a rigorous set of criteria to select the most reliable ones. In total, we included 5 polyneopteran and 6 orthopteran fossils to time-calibrate for the analysis. All the calibrations, including the root age, were set to soft maximum bound at 412 million years ago (the oldest age of Rhynie Chert99) using uniform priors. We estimated divergence times using MCMCTree implemented in the software package PAML v.4.9 (ref. 100) based on the modified matrix of the Daa,trans,strict data set as it represented the most decisive data set. This modified matrix containing only sites with unambiguous data for at least 80% of the 60 taxa was necessary to overcome computational limitations when estimating node ages resulting from the large size of the data set. Previous studies have shown that results of dating analysis are robust to missing data patterns and this data set reduction101. In addition, to further reduce computational effort, we chose an unpartitioned dating analysis. We set the model LG (aaRatefile = lg.dat) + G with 5 rate categories, empirically estimated base frequencies (model = 2) and allowed rates to be inferred from individual sites (RateAncestor = 1). We conducted Hessian matrix calculations according to the above specifications with CODEML as implemented in PAML using empirical +F base frequencies estimated from the respective data set. MCMC chains ran for 1,000,000 generations (sfreq = 50) while discarding a burn-in of 100,000 generations. A total of four independent runs were done at the University of Memphis HPC cluster and using Texas A&M HPC cluster. Additional details on phylogenetic analysis, topology testing, and divergence time estimate analysis are presented in Supplementary Methods 1.6, 1.7, and 1.8.Phylogenetic comparative methodsTo trace the evolution of hearing and sound-producing organs along the phylogeny, we first conducted a thorough literature review and physical examination of the specimens to characterise these organs in all species included in this study. For hearing organs, we coded whether tympanum was absent, present on thorax (Mantidae), on fore tibiae (Ensifera) or on abdomen (Caelifera). We also included atympanate hearing found in Pneumoridae as one of the states. For sound-producing organs, we used a specific naming convention in which the first-named structure has the stridulatory file and the second named structure has the scraper. For example, abdominal-femoral stridulation would have the stridulatory files on the abdomen and the scraper on the inner side of hind femora. The possible combinations used were: absent, tegmino-pronotal, tegmino-femoral, tegmino-alary, tegmino-tegminal, abdomino-alary, abdomino-femoral, Krauss’s organ-femoral and femoro-tegminal stridulation. In addition, we included another type of sound-producing mechanism, only found in Acrididae, known as crepitation36, which produces sound by snapping wings when grasshoppers fold and unfold. The complete list of characters used for this analysis is presented in Supplementary Data File 14.We performed ancestral character state reconstruction of hearing and sound-producing organs in a maximum likelihood framework using the topology resulting from the Dnt,trans+mito,strict. We fitted a continuous-time Markov chain (Mk) single-rate (ER) model to our data to infer character evolution using the R package phytools102. Using the same data set, we also performed Pagel’s37 binary character correlation test for the evolutionary correlation between hearing and sound production, using phytools. We pruned the phylogenetic tree to create Orthoptera-only, Ensifera-only and Caelifera-only data sets to compare and contrast lineage-specific patterns. We recoded different types of tympanal and stridulatory mechanisms as simple presence-absence binary characters for both hearing and sound production to reveal the general co-evolutionary dynamics of these two traits. For each data set, we fitted four models of co-evolution between hearing and sound production and compared the results using the Akaike Information Criterion (AIC): (i) hearing and sound production evolve independently; (ii) the evolution of hearing depends on the evolution of sound production; (iii) the evolution of sound production depends on the evolution of hearing; and (iv) hearing and sound production evolve interdependently.For Ensifera, we also examined the evolution of the complex tibial organ in the forelegs. Although detailed neuroanatomical studies of auditory sensory organs are limited to only a small number of species7,68, it has been suggested that the complex tibial organ consisting of the subgenual organ (SGO) and the intermediate organ (IO) was the ancestral condition in Ensifera7. Extant Grylloidea and Gryllotalpidae have the sensory organ consisting of the SGO and the tympanal organ (TO), which is presumed to be modified from the IO as the auditory receptor cells7. By contrast, Tettigoniidae and Prophalangopsidae have the sensory organ consisting of the SGO, the IO, and the sensory neurons in the crista acustica (CA) responding to the auditory signal7. Atympanate ensiferans are also known to vary in terms of the configuration of the complex tibial organ. Rhaphidophoridae have the SGO and the IO, with no obvious trace of specialised auditory receptor cells103. Schizodactylidae, Gryllacrididae, Stenopelmatidae and Anostostomatidae all have the sensory organ similar to that of Tettigoniidae, consisting of the SGO, the IO and the sensory neurons that are homologous to the CA, but with no auditory specialisation, called the crista acustica homologue (CAH)69,104. Because we did not have detailed neuroanatomical data for our taxon sampling, we could not perform ancestral character state reconstruction of the complex tibial organ in Ensifera, but we were able to map this character on the phylogeny based on the assumption that the configuration would be conserved at the taxonomic family level.To estimate the rates of lineage-specific diversification, we performed a diversification analysis using the program Bayesian analysis of macroevolutionary mixtures (BAMM)39 and the R package BAMMtools105. Because BAMM required a comprehensive time-calibrated ultrametric tree, we performed a divergence time estimate analysis using the 249-taxa Dnt,trans+mito,strict data set with the same 11 fossil calibration points using MCMCTree as described in Supplementary Methods 1.8. To accurately represent species diversity and to account for incomplete taxon sampling, we specified sampling fraction for each family based on the number of described species recorded in the Orthoptera Species File106. We set priors using setBAMMpriors function in BAMMtools before the analysis and modified the default setting to achieve convergence. The priors used for the analysis were expectedNumberOfShifts=1.0; lambdaInitPrior=17.0512659943593; lambdaShiftPrior=0.00279913753644403; muInitPrior=17.0512659943593; lambdaIsTimeVariablePrior=0. We used “speciationextinction” as a model for the diversification analysis in BAMM, and ran for 10 million generations with a sampling frequency of 1000. Convergence assessment, analysis of rate shifts, and calculation of clade-specific rates were performed using BAMMtools.To test whether the evolution of hearing and sound production has affected speciation and extinction rates, we fitted models of trait-dependent diversification using the R package hisse42. Because it has been shown that the presence of unmeasured factors (or hidden states) could impact estimation of diversification rates for any observed trait when analysed under the framework of BiSSE (Binary State Speciation and Extinction) methods107, we adopted a multimodel inference method, implemented in HiSSE (Hidden State Speciation and Extinction)42. We first pruned the time-calibrated ultrametric tree to only include Orthoptera (239 terminals), and used the binary character data sets for hearing and sound-producing organs previously used for the Pagel’s test. Because the presence of these organs does not necessarily indicate the presence of acoustic communication, we created an additional data set to code acoustic communication as a binary character to test if its evolution affected the diversification rate. We fitted 24 different models, used in Beaulieu and O’Meara42, to both the hearing data set and the sound production data set for Orthoptera. These models included four models corresponding to BiSSE models, four models corresponding to trait-independent models (described as CID models), and 16 models corresponding to different HiSSE models that assumed a hidden state associated with both the observed states. Detailed descriptions of these models are included in Supplementary Methods 3.4. For all cases, we included the sampling fraction for the observed states in the models by calculating the proportion of the known 0’s (i.e. absence) represented and the proportion of the known 1’s (i.e. presence) represented in our tree. The resulting models were compared using the AIC. All analyses were carried out in hisse.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer ReviewReporting SummaryDescription of Additional Supplementary FilesSupplementary Data 1-15 | nature communications | [
"Article"
] | [
"Phylogenetics",
"Sexual selection",
"Animal behaviour",
"Entomology"
] | IntroductionAcoustic communication signalling among animals documented in fishes frogs birds cetaceans terrestrial mammals insects interplay-evolution between signal sender receiver mating prey location predator avoidance led to diversity complexity soundscape.Signal emission reception major components acoustic communication enabled by sound-producing organs hearing sensory organs animal vertebrate lineages evolved structures mechanisms for hearing sound inner ear structures Silurian ostracoderms8 tympanal ears evolved in frogs mammals reptiles6 vocal cords in larynx evolved within tetrapods6 birds evolved unique organ syrinx9 insects hear using tympanal ears evolved in seven orders (Orthoptera Mantodea Hemiptera Neuroptera Coleoptera Lepidoptera 15 body locations10–13 lack of tympanal ears other insect orders hear internal sensory organs sensitive to sound without tympanal ears enable far-field hearing broad frequency range highability produce sound long distance using specialised organs stridulatory apparatus tymbals evolved in six insect orders (Blattodea Coleoptera Hemiptera Lepidoptera Mantodea Orthoptera), involving many body parts6,11 studies focused on mechanisms hearing sound production evolutionary processes acoustic signalling1–3 questions about when context hearing sound organs evolved inadequately addressed7 For insects acoustic signalling two prevailing views on structures first view evolved adaptation to detect escape predators7 Tympanal hearing may evolved auditory surveillance for predator movements demonstrated in moths24 mantises26 grasshoppers27 early stridulatory organs could evolved as defensive mechanism6 deimatic behaviour hearing sound-producing organs could led to evolution sexual signalling via ‘sensory bias’ mechanism male sexual signals evolve from phylogenetic pattern consistent with sensory bias mechanism evolution of one component hearing organ evolution counterpartsound-producing second view hearing sound-producing organs evolved as female perception male signalling devices Fisherian suggested cicadas crickets katydids evolved acoustic communication predictable phylogenetic pattern origin of hearing sound-producing organs to single common ancestor understanding evolution acoustic communication important to trace evolution in phylogenetic framework lineage including species lacking hear sound species with diverse acoustic communication strategies animal groups communication insect order Orthoptera ideal model evolutionary 16,000 species using acoustic signalling sexual communication most species-rich clade acoustically communicating animals outnumbering frogs birds mammals28 acoustically-active insect lineages Orthoptera lineages use acoustic signalling mating defensive signalling29 others lacking structures hearing sound Orthoptera earliest lineage evolved complex acoustic communication fossil forewings stridulatory apparatus homologous to-day crickets from Triassic31 Orthoptera excellent group study evolution acoustic communication lack of robust time-calibrated phylogeny major challenge for inferring complex patterns hearing sound-producing organsstudy reconstruct evolution hearing sound-producing organs in Orthoptera bird’s-eye view acoustic communication diversified hundred million years evolution establish phylogenetic relationships lineages combining 4986 alignments protein-coding genes 60 taxa (50 orthopterans 10 polyneopteran outgroups 249 mitochondrial genomes employ selected fossils topology testing robust time-calibrated phylogeny framework evolution tympanal ears internal sensory organs diverse sound-producing mechanisms test evolutionary hypotheses origins organs diversification patterns influenced innovations lineage-specific dynamic patterns evolution for hearing sound-producing organs infer organs co-evolved sexual context in crickets katydids allies little evidence increased diversification rates singing lineages hearing organs evolved non-sexual context in grasshoppers co-opted for sexual signalling sound-producing organs evolved.ResultsPhylogenetic relationships divergence times of major orthopteran explored signal phylogenomic data six phylogenomic data sets differing matrix saturation character coding size maximum likelihood framework six data sets congruent topologies family-level relationships Figphylogenetic placements of Rhaphidophoridae Gryllotalpidae Pamphagidae varied among trees applied four-cluster likelihood mapping permutation tests for relationships six data sets confounding signal among-lineage heterogeneity non-random distribution missing data mixture Methods placement Rhaphidophoridae robust unbiased Pamphagidae Gryllotalpidae potentially biased small taxon sampling not sufficient unambiguous conclusions ambiguous placements little impact evolution of hearing sound-producing organs analyses confirmed monophyly of Orthoptera suborders Ensifera Caelifera (Figs. 1 2) recovered ancient age for crown-Orthoptera ~355 million years ago credibility interval 393.8–320.0 million ~63 My earlier than previous estimate estimated crown-Ensifera appeared during Late Carboniferous.0–267.4 consistent with fossil record earliest stem-Ensifera 272 million-years-old31analyses recovered monophyletic infraorders Gryllidea Tettigoniidea former Grylloidea Gryllotalpidae Myrmecophilidae latter remaining families 1 crown-Gryllidea originated late Triassic early Jurassic (200 Mya 247.5–154.1 My-Tettigoniidea originated Permian (268 Mya 308.1–227.7 My diverged lineages Mesozoic Tettigoniidea recovered family-level relationships (Rhaphidophoridae (Schizodactylidae ((Gryllacrididae Anostostomatidae (Prophalangopsidae crown-Caelifera originated Carboniferous (320 Mya 359.5–282 recovered two monophyletic infraorders Tridactylidea Acrididea former Cylindrachetidae Ripipterygidae Tridactylidae diverged late Carboniferous latter remaining families Acrididea originated Late Permian (263 Mya 301.5–224.6 My split groups Tetrigidae Acridomorpha modern grasshopper diversity arose Cenozoic details relationships Orthoptera Supplementary Methods 1.9.Fig. Orthoptera datachronogram divergence time estimate analysis data set 436,488 amino acids Bootstrap support values coloured nodes 100 yellow below 90 not estimates calculated 86,043 amino-acid sites 11 fossil calibrations Blue bars indicate 95% credibility intervals node ages Geological timescale bottom details data Supplementary Methods 1 2.Fig. Orthoptera estimated transcriptomes mitochondrial genomes tree derived likelihood analysis 448,861 aligned sites nucleotides Bootstrap support values coloured nodes 100 yellow 90-99 orange 80-89) Red branches Ensifera blue branches Caelifera red blue clade names (Gryllidea Tettigoniidea Tridactylidea Acrididea infraorder names white red blue bars superfamily names circular bars colour-coded superfamily TRIDAC Tridactyloidea TETRI Tetrigoidea EUMAST Proscopioidea Tanaoceroidea Trignopterygoidea Pneumoroidea Schizodactyloidea Hagloidea.Evolution hearing sound-producing organs lineage-specific patterns evolution hearing sound organs Orthoptera (Fig. 3)In Ensifera analysis found tegmino-tegminal stridulation likely evolved in common ancestor lineages. absent in Myrmecophilidae Rhaphidophoridae consequence loss of wings loss occurred in Schizodactylidae common ancestor Gryllacrididae Stenopelmatidae Anostostomatidae abdomino-femoral stridulation evolved defensive signalling against hearing organs tibial tympana evolved three times in Ensifera Gryllidea Anostostomatidae Prophalangopsidae Tettigoniidae analysis small probability tibial tympana evolved in common ancestor Ensifera rule out possibility presence tibial tympana ground plan for suborder examined evolution complex tibial organ in forelegs ancestral Ensifera had tibial organ subgenual organ intermediate organ ancestor Gryllidea gained far-field hearing tympanal organ modified from tibial tympana Rhaphidophoridae retained ancestral configuration Tettigoniidea third component crista acustica homologue) evolved modification evolution of tibial tympana common ancestor Prophalangopsidae Tettigoniidae gained far-field hearingcharacter state reconstruction hearing sound-producing organs topology comprehensive phylogeny Dnt,trans+mito,strict Fig. 2) coloured circle branch tip indicates character state species grey circles absence coloured circle node probability ancestral character state character evolution hearing organs states colour-coded two traits mapped first internal sensory organs ensiferan foretibia red ancestral condition Ensifera subgenual organ intermediate organ ancestor Gryllidea IO modified tibial organ Rhaphidophoridae retains SGO + IO Schizodactyloidea Stenopelmatoidea Hagloidea Tettigonioidea third component crista acustica homologue) evolved Hagloidea Tettigonioidea CAH modified auditory sensory organ crista acustica second trait loss of wings black circles species tympanal hearing lost wings character evolution sound-producing organs stridulatory apparatus character states colour-coded specific naming convention first-named structure stridulatory file second named scraper abdominal-femoral stridulation stridulatory files abdomen scraper hind femora mechanics tegmino-tegminal stridulation mappedGryllidea “left-over-right” Hagloidea “ambidextrous” Tettigonioidea-over-left” OG Outgroups GRYT Gryllotalpoidea GRYL Grylloidea RHAP Rhaphidophoroidea SCHI Schizodactyloidea STEN Stenopelmatoidea HAG Hagloidea TETT Tettigonioidea TRID Tridactyloidea TETR Tetrigoidea EUMAS Eumastacoidea PROS Proscopioidea TANA Tanaoceroidea TRIG Trignopterygoidea PNEU Pneumoroidea PYRG Pyrgomorphoidea ACRI Acridoidea Caelifera stridulatory organs evolved 10 times sound unconfirmed few species acoustic signalling mating defence11 consider putative sound-producing organs mating evolved three times Caelifera Pneumoridae abdomino-femoral Pamphagidae Krauss’s organ-femoral stridulation acridid Acridinae Gomphocerinae Oedipodinae hind femora tegmina location varies abdominal tympana evolved three times Pyrgomorphidae Pamphagidae Romaleidae Ommexechidae Acrididaetibial tympana in Ensifera ancestral character reconstruction recovered probability abdominal tympana evolved common ancestor tympanate lineages (Fig. 3) structure could ground plan performed Pagel’s37 binary character correlation test hearing sound-producing organs co-evolved within Orthoptera suborders (Fig. 4) recovered significant correlation organs co-evolutionary dynamics different depending lineages (Fig. 4 Supplementary Methods 3.2). Orthoptera model evolution hearing organs depended on sound-producing organs AIC = 0.7685). higher instances hearing organs when sound-producing organs absent presence when present (Fig. 4) taxon sampling correlation patterns acoustic communication different between Ensifera Caelifera11 examined character correlation each suborder delivered contrasting patterns (Fig. 4) Ensifera evolution sound-producing organs depended on hearing organs AIC = 0.4490) hearing sound-producing AIC = 0.4343) explained pattern higher instances presence sound-producing organs when hearing organs present all ensiferans hear produce sound high correlation traitsCaelifera model evolution hearing on sound-producing organs AIC = 0.4361 explained data sound-producing organs hearing organs AIC = 0.3567) explained data higher instances absence sound-producing organs Caelifera (Fig. 4) opposite pattern from Ensifera.Fig. 4Evolutionary correlation between hearing sound production in Orthoptera.Pagel’s test evolutionary correlation calculated hearing sound production Orthoptera Ensifera Caelifera thickness of arrows corresponds to rate change from to higher rate thicker arrow analyses strong evolutionary correlation between hearing sound production patterns different Orthoptera strongest transition rate from hearing present sound production absent to absent Ensifera hearing present absent to both present Caelifera hearing present absent to absent differences between Ensifera Caelifera show co-evolutionary dynamics hearing sound production differ between lineages.Rates lineage diversification acoustic Bayesian analysis macroevolutionary mixtures (BAMM)39 found three episodes of rate shift along phylogeny Orthoptera (Fig. 5)first episode rate shift Tettigoniidae Cretaceous mean clade-specific rate (0.08186125) double Orthoptera (0.04820052) Ensifera (0.04996942) second episode rate shift Pamphagidae late Cretaceous early Paleogene mean clade-specific rate (0.1290903) tripling rate Orthoptera Caelifera (0.04905859) highest evolutionary rate orthopteran lineages third episode rate shift Romaleidae Ommexechidae Acrididae late Cretaceous Paleogene mean clade-specific rate (0.07159634) slightly higher background rate other singing lineages Ensifera Grylloidea Gryllotalpidae Prophalangopsidae show rate shift over-interpret patterns appropriateness BAMM diversification analyses questioned estimate diversification developers BAMM criticisms unjustified41.Fig. 5Bayesian analysis macroevolutionary mixtures Orthoptera ultrametric tree dated phylogeny combined data (Dnt,trans+mito Phylorate plot speciation rates slow warm fast vertical colour legend each branch Orthoptera phylogeny three clades black circled nodes increased rate shifts Lineages acoustic communication indicated vertical lines near terminalsBranches coloured rate shifts 95% credible macroevolutionary shift configurations value 0.21 21% samples posterior assigned shift configuration upper left plot four shift configurations account 51.5% posterior distribution Clade-specific evolutionary rate variation for Orthoptera three lineages rate shifts fitted models trait-dependent trait-independent diversification using HiSSE test evolution hearing sound-producing organs speciation extinction rates orthopteran lineages (Fig. 6) hearing organs best-fitting model HiSSE character-dependent diversification diversification parameters free transitions between hidden states disallowed (HiSSE q0B1B = 0 q1B0B 0 q’s higher net diversification rate presence hearing organs likely due to higher rate hidden state net diversification rate absence hearing organs lower sound-producing organs best-fitting model CID (trait-independent) models evolution binary trait independent of diversification process constant tree model suggested evolution sound-producing organs affect net diversification rate acoustic communication coded as binary trait best-fitting model identical trait-independent model sound-producing organs (CID-4 diversification independent from evolution acoustic communication. 6Models trait-dependent diversification reconstruction states net diversification rates estimated using multimodel inference methods best-fitting models for each trait hearing acoustic communication from 24 models trait-dependent clades sexual communication acoustic signalling labelled in circular trees estimates likely state rate based on model-averaged marginal reconstructions-fitting models histograms show location rates gradient frequency for each contemporary tip taxa hearing organs best-fitting model HiSSE models sound-producing organs acoustic communication trait-independent diversification models (CID-4).DiscussionOrthopteran insects model systems acoustic communication hearing sound-producing organs originated evolved elusive lack phylogeny work establishes phylogenetic relationships lineages divergence time estimates Orthoptera based data fossil calibration points crown-Orthoptera originated 355 million years ago diverged into Ensifera Caelifera in Carboniferous (Fig. 1) study suggests suborders followed different lineage-specific patterns evolution for hearing sound-producing organs (Figs. 3 4)Hearing sound-producing organs co-evolved in larger suborders encompasses ~15,500 species many nocturnal use acoustic signalling sexual communication singing ensiferans include four lineages (crickets mole crickets katydids grigs specialised hearing organs tympanal ears front tibiae stridulatory apparatus on male tegmina account 85% ensiferan diversity3 remaining lineages neither tibial tympana stridulatory tegmina (ant-loving crickets [Myrmecophilidae cave crickets lack tibial tympana possess stridulatory apparatus abdomen defensive signalling29 present in both sexes-footed crickets raspy crickets Jerusalem crickets king crickets wetas monophyly of Ensifera supported by cladistic analyses34 agree suborder consists two monophyletic infraorders Gryllidea Tettigoniidea34 study confirmedconsensus on internal relationships among families superfamilies phylogenetic studies disagreed,43 to conflicting inferences evolution acoustic communication7 stridulatory apparatus evolved once or multiple times contentious32 phylogenomic analysis recovered relationships among families (Figs. congruent with morphology-based phylogeny43 than previous molecular studies34 recovered topology divergence time estimates ancestral character state reconstruction (Fig. 3) infer evolutionary scenario hearing sound-producing organs in Ensifera late Carboniferous early Permian crown-Ensifera diverged male-specific tegmino-tegminal stridulation evolved in ancestor Ensifera (Fig. 3) earliest airborne sound generation in oldest fossil ensiferans Gryllavus Protogryllus with-preserved stridulatory apparatus known from Triassic31 suggests similar mechanism sound production could evolved earlier earliest insectivorous tetrapods in early Carboniferous tympanic ears6 hypothesised predators deterred by stridulation insect prey tactile receptors stridulation could evolved deimatic behaviour56evidence shows sound-producing organs wings present among Permian Triassic stem-Orthoptera31 wings sound ancient invention behaviour co-opted for sexual communication lineages-orthopteran lineage Titanoptera had modified veins forewings indicative sound production both sexes57 used for pair formation reciprocal evolved from Permian ‘tcholmanvissiids’59 forewing sound-producing organs lineage Mesoedischiidae had male-specific tegmino-tegminal stridulation veins not homologous Ensifera57 four singing ensiferan lineages mechanics tegmino-tegminal stridulation Crickets mole crickets stridulate left forewing right katydids right forewing Grigs stridulate wings both comparative analysis proposed stridulatory apparatus involved different veins forewing in four lineages53 debated muscular mechanics neurocircuit enabling male-specific tegmino-tegminal stridulation conserved potentially plesiomorphic in Ensifera different lineages evolved different ways creating audible sound mechanism analysis suggests tegmino-tegminal stridulation secondarily lost in several ensiferan lineagesloss associated with adaptations to environments wings members Rhaphidophoridae apterous associated with caves60 Myrmecophilidae wingless associated with ant colonies61 Schizodactylidae Gryllacrididae Stenopelmatidae Anostostomatidae specialists on subterranean habitats wingless families species functional tegmina lacking stridulatory apparatus63 cricket katydid species lost ability to Hawaiian cricket Teleogryllus oceanicus loss stridulatory apparatus evolved due to pressure from phonotactic parasitoid fly Ormia ochreacea64 loss genetic basis alteration of master regulatory switch during early development to change in adult phenotype65 difficult to process loss of tegmino-tegminal stridulation in non-singing ensiferans loss could achieved multiple times during diversification Ensifera non-singing ensiferans engage in intraspecific communication using substrate-borne vibration drumming tremulation well-developed chordotonal organs for sensing vibration7,68 abdomino-femoral stridulation evolved twice in Schizodactylidae Gryllacrididae Stenopelmatidae Anostostomatidaemechanism found in both sexes nymphs not used for sexual communication defensive signal against predators29 suggest loss of tegmino-tegminal stridulation promoted evolution vibratory signalling sexual acoustic signalling non less clear first hearing organs evolved in ancestor Ensifera Hearing organs forelegs complex external tympana internal tibial organs7 unclear original form sound detection in ensiferans thin cuticle of foretibia could functioned as resonator for organs sound thinning cuticle evolved three times tympanal membrane neurophysiological mechanisms hearing evolved twice to two different types hearing sensory organs SGO + TO in crickets mole crickets SGO + IO + CA in katydids grigs consistent with common ancestor Ensifera structures far-field hearing different lineages evolved far-field hearing early hearing in insects evolved detecting avoiding predators11 specific position tympana evolved in singing ensiferans possibility hearing Ensifera evolved another contextensiferan ears have two auditory inputs sound external tibial tympana internally via acoustic trachea on spiracles side pronotum44 ears are pressure difference receivers73 sound internally slower longer than external surface tympanum normal causes differences gain between sound externally internally72 complex acoustic tracheal system shows lineages-specific differences In crickets mole crickets trachea connect four sound inputs enlarged part midline two thin septa each In katydids grigs trachea spiracles connect right left connect to right left tibial tympana In katydids tracheae enlarged as acoustic bullae at spiracles narrow tympanal ears47 directional hearing mechanism evolved independently locating source of calls.Sound-producing organs hearing organs evolved separately in CaeliferaCaelifera orthopteran suborders ~12,200 species insects grasshoppers locusts pygmy mole crickets grasshoppers monkey grasshoppers stick grasshoppers relatives34Sexual communication acoustic signalling rare across Caelifera documented in small divergent families (bladder grasshoppers pamphagid tooth-legged banded-wing shows lineages use different stridulatory mechanisms produce sound involve rubbing hind femora against body parts thickened veins abdomen early-diverging caeliferan lineages have hearing organs tympanal hearing found in few grasshopper families (Pamphagidae Pyrgomorphidae Romaleidae Ommexechidae Acrididae) originated in Cretaceous Paleogene present tympana located on both sides first abdominal segment large tympanal membranes innervated with auditory sensory organs encircled by sclerotised rings air-filled tracheal sacs between tympanal membranes45 phylogenomic analysis hearing sound-producing organs in Caelifera evolve jointly followed different evolutionary trajectories no fossil evidence antiquity of hearing sound production in Caelifera acoustic communication recent invention in Caelifera shows lineages evolved paired structures stridulatory file on one body scraper anothermouthparts forewings hindwings middle hind legs abdomen unconfirmed paired structures for sound production except families acoustic signalling not clear context structures evolved structures found in both sexes lineages mandibulo-maxillary stridulation Cylindrachetidae could evolved defence sound-producing organs in males lineages abdomino-femoral stridulation Tanaoceridae could evolved sexual context diversity mechanisms functions sound production in Caelifera caeliferans engage tegmino-tegminal stridulation primary sound-producing mechanism Ensifera implies machinery stridulation never part of caeliferan ground plan finds first sexual communication acoustic signalling in Caelifera evolved in ancestor South African family Pneumoridae Jurassic. complex acoustic signalling-established in Ensifera bladder grasshopper males fully winged have inflated abdomen resonating chamber produce loud low-frequency calls 2 km abdomino-femoral stridulation83 male calling receptive females willingness mate acoustically leads to pair formation reciprocal dueting84Female sound-producing organs homologous males species use different body parts sound males females lack tympanal ears chordotonal organs abdominal segment entire abdomen functions as hearing pattern suggests selective pressure for acoustic communication Jurassic directional hearing organs evolve lineages radiated like ensiferan counterparts Cretaceous abdominal tympana appeared in Caelifera (Figs. consistent with multiple origins abdominal tympana small probability Pyrgomorphoidea Acridoidea evolved abdominal tympana intermediate option unspecialised early form abdominal hearing organ parallel evolution tympana lineages context organs clear Grasshoppers with abdominal tympana show jumping flying behaviour hearing function detecting predators commonly invoked hypothesis origin grasshopper insectivorous predators-diversified by Cretaceous6 unlikely selective pressure triggered predator-detection hearing other caeliferan lineages radiated without hearing Tetrigoidea Eumastacoidea insects faced predators succeeded without tympana grasshopper species with abdominal tympana sexual communication acoustic signalling difficult to think hearing evolved sexual contextalternative explanation secondary loss of abdominal tympana in species wing reduction suggests connection between flight hearing mechanisms auditory pathway in grasshoppers locusts studied45 auditory information processing through abdominal tympana influenced by thoracic muscle movement wingbeat noise during fly plesiomorphy for Orthoptera Pyrgomorphoidea Acridoidea first large-bodied caeliferans strong dispersal capacity possibility abdominal tympana evolved modulating flight detecting disturbances locating mates brachypterous katydids crickets retain hear through tibial tympana47 not involved in flight modulation evolution of abdominal tympana in early grasshoppers could led to sexual signalling ‘sensory-bias’ achieved independent evolution of sound-producing organs in two grasshopper lineages Pamphagidae Acrididae path to acoustic communication differed Pamphagidae large-bodied family originated in Cretaceous pair formation via reciprocal dueting78 males fully winged females flightless loss of wings common Krauss’s organ-femoral stridulation phylogenetically conserved mechanism of sound productionKrauss’s organ is plate lower anterior corners second abdominal tergite rubbed by ridges hind femora87 mechanism present in males and females sound species-specific35 not pamphagids utilise sound-producing mechanisms for mating abdomen hind femora forewings hindwings middle tibiae thorax35 evolution sound production in ancestor Pamphagidae led to acoustic communication lineage sound production evolved later in Acrididae after lineage diversified abdominal tympana plesiomorphic family male-specific stridulatory mechanism using tegmina hind femora evolved between Eocene Oligocene Acridinae Gomphocerinae Oedipodinae sexual context sound-producing organs followed different evolutionary trajectories modifications stridulatory apparatus in Gomphocerinae stridulatory pegs on hind femora rub against veins forewings Oedipodinae files on intercalary veins forewing rubs against scrapers hind femora36 Oedipodinae Acridinae evolved alternative non acoustic mechanism crepitation produces sound snapping wingsgrasshoppers acoustic signalling complemented with visual signalling leg movements multimodal sexual acoustic signalling in Acrididae recently evolved sexual communication Orthoptera acoustic communication influence diversification rates in evolution sexual communication acoustic signalling in Ensifera Caelifera different trajectory (Figs 3 4) Ensifera tegmino-tegminal stridulation ancestral defensive signalling lineages evolved tibial tympana sexual context common ancestor hearing sound-producing organs present Fisherian mechanism co-evolution between female perception male signalling Pagel’s binary character correlation test evidence hearing sound-producing organs co-evolved in Ensifera hypothesis Caelifera abdominal tympana evolved later in lineage diversification modulating flight co-opted for detecting predators for sexual communication pattern fits with ‘sensory bias’ mechanism Pagel’s test little support for co-evolution between hearing sound-producing organs alternative hypothesis evolution hearing sound-producing organs affected diversification rates in lineages acoustic signalling sexual selection major force diversification singing sexually selected traits evolve rapidly88if inferred mechanism evolution hearing sound organs Fisherian mechanism expect elevated diversification rate in clade sexual communication using acoustic signalling89 tested in tetrapods acoustic communication increase diversification rates proposal in Orthoptera performed diversification analysis using BAMM39 clade-specific evolutionary rates acoustic communication Tettigoniidae only Ensifera increased mean rate other lineages (Grylloidea Gryllotalpidae Prophalangopsidae rate shifts Pamphagidae only Caelifera increased mean clade rate neither Pneumoridae Acridinae Gomphocerinae Oedipodinae showed rate shifts Rate shifts associated with key innovations increased diversification sexual communication acoustic signalling not major key innovation for singing lineages Tettigoniidae Pamphagidae increased diversification rates due to acoustic signalling communication other forms signalling (visual chemical known other key innovations leaf masquerade diverse feeding habits91 led to rate shift related to of angiosperms findings bolstered by analysis of trait-dependent diversification using HiSSE42lineages hearing sound-producing organs acoustic communication had higher net diversification rates multimodel inference models suggest evolution hearing organs affected net diversification sound-producing acoustic communication independent affect study little evidence acoustic communication increased net diversification results understanding signal sender-receiver co-evolution diversification in Orthoptera insights evolution animal communication.MethodsPhylogenomic analyses divergence time taxon sampling 239 species Orthoptera 10 polyneopteran outgroups 249 species data represented 16 superfamilies 36 families extant Orthoptera included 60 transcriptomes 39 orthopteran species newly generated Song Lab Texas A&M University remaining 21 transcriptomes (11 orthopteran 10 polyneopteran from previous publications combined transcriptome data with 169 previously 80 newly generated mtgenomes from 249 taxa. RNA extraction cDNA library preparation transcriptome sequencing assembly performed 1KITE project protocols Supplementary Methods 1.2 1.5. Protocols Song Lab samples mtgenome data generation Supplementary Methods 1.3 1.4detailed list species collection data National Center Biotechnology Information accession numbers in Supplementary Data 1 transcriptome data custom-made orthologous gene set designed OrthoDB v7 (ref. 92) four hemimetabolous (Zootermopsis nevadensis Pediculus humanus Acyrthosiphon pisum Rhodnius prolixus one holometabolous (Nasonia vitripennis species 5414 protein-coding genes Orthograph v0.5.3 (ref. 93) hidden Markov model) amino-acid sequences gene ortholog candidates protein BLAST) search database orthologous result checked best hit sequence matched best-reciprocal hit criterion fulfilled OG extended candidate transcript identified 3700 OGs amino-acid sequences aligned level using MAFFT v7.130b94 L-INS-i algorithm quality multiple sequence alignment) checked Supplementary Methods 1.5 downstream analyses considered regions protein clans families domains non-annotated as evolutionary units partitioned analyses methods detailed in Supplementary Methods 1.5. Perl scripts results protein domain identification randomised MSA sections merged into masked supermatrixalignment length spanned 1,647,472 amino-acid positions back-translated nucleotide supermatrix created using Perl scripts MARE v0.1.2-rc95 information content (IC) data block data blocks removed from supermatrices created four transcriptome data sets Daa,trans,complete, 1,541,865 aligned amino acids 1743 domain-based metapartitions Daa,trans,strict, 436,488 aligned amino acids 102 metapartitions 100% matrix saturation Dnt,trans,complete 1,541,865 aligned second codon positions Dnt,trans,strict 436,488 aligned sites second codon positions partitions used PartitionFinder 2.0 (ref 96) RAxML version mtgenome data created concatenated matrix of nucleotide sequences 13 protein-coding genes aligned MUSCLE97 divided into 39 data blocks codon used PartitionFinder best-fit scheme estimate nucleotide evolution for each partitioncombined transcriptome data mtgenome data Dnt,trans,complete,strict sets mtgenomes 249 taxa 60 overlapped with transcriptome Dnt,trans+mito,complete 1,554,238 sites 1766 metapartitions Dnt,trans+mito,strict 448,861 sites 125 metapartitions details Supplementary Methods 1.5 analysed six data sets maximum likelihood framework IQ-TREE v1.5.4 (ref. 98) best-scoring substitution matrix each partition performed 50 tree searches node support estimated via non-parametric bootstrapping 100 replicates IQ-TREE mapped ML tree best log-likelihood determined support relationships four-cluster likelihood mapping selecting incongruent nodes checking for confounding among-lineage heterogeneity non-random substitution processes missing data permuted data sets estimate divergence times review fossils calibration points applied criteria reliable included 5 polyneopteran 6 orthopteran fossils time-calibrate calibrations root age set maximum 412 million years ago oldest age Rhynie Chert99) estimated divergence times using MCMCTree PAML v.4.9modified matrix Daa,trans,strict data set decisive unambiguous data 80% 60 taxa computational limitations estimating node ages large size data studies dating analysis robust missing data patterns data set reduce computational effort chose unpartitioned dating analysis set model LG + G 5 rate categories estimated base frequencies = 2) allowed rates inferred sites (RateAncestor = 1) conducted Hessian matrix calculations CODEML PAML empirical +F base frequencies data MCMC chains ran 1,000,000 generations (sfreq = 50 burn-in 100,000 generations four independent runs University of Memphis HPC cluster Texas A&M HPC cluster details phylogenetic analysis topology testing divergence time estimate analysis Supplementary Methods 1.6, 1.7 1.8.Phylogenetic comparative evolution hearing sound-producing organs conducted literature review physical examination specimens species hearing organs coded tympanum absent thorax fore tibiae abdomen included atympanate hearing Pneumoridae sound-producing organs specific naming convention first-named structure stridulatory file second named scraperabdominal-femoral stridulation files abdomen scraper inner side hind femora possible combinations absent tegmino-pronotal tegmino-femoral tegmino-alary-tegminal Krauss’s organ-femoral femoro-tegminal stridulation included sound-producing mechanism Acrididae crepitation36 produces sound snapping wings when grasshoppers fold unfold complete list characters Supplementary Data File performed ancestral character state reconstruction of hearing sound-producing organs maximum likelihood framework topology Dnt,trans+mito,strict fitted continuous-time Markov chain) single-rate) model infer character evolution R package phytools102 performed Pagel’s37 binary character correlation test correlation hearing sound production pruned phylogenetic tree Orthoptera-only Ensifera-only Caelifera-only data sets compare contrast lineage-specific patterns recoded tympanal stridulatory mechanisms as presence-absence binary characters for hearing sound production co-evolutionary dynamics fitted four models co-evolution hearing sound production compared results using Akaike Information Criterion evolve independently interdependently Ensifera examined evolution complex tibial organ in forelegs.neuroanatomical studies auditory organs limited small suggested complex tibial organ subgenual intermediate ancestral in Ensifera7 Grylloidea Gryllotalpidae have SGO tympanal organ modified from IO auditory receptor Tettigoniidae Prophalangopsidae have SGO IO sensory neurons in crista acustica auditory Atympanate ensiferans vary configuration complex tibial organ Rhaphidophoridae have SGO IO no specialised auditory receptor Schizodactylidae Gryllacrididae Stenopelmatidae Anostostomatidae have sensory organ similar Tettigoniidae SGO IO sensory neurons homologous CA no auditory specialisation crista acustica homologue (CAH detailed neuroanatomical data perform ancestral character state reconstruction of complex tibial organ in Ensifera map character on assumption configuration conserved at taxonomic family level lineage-specific diversification performed diversification analysis using Bayesian R package BAMMtools105 time-calibrated ultrametric tree performed divergence time estimate analysis using 249-taxa Dnt,trans+mito,strict data set 11 fossil calibration points MCMCTree 1.8represent species diversity account incomplete taxon sampling specified sampling each family based species Orthoptera Species File106 set priors setBAMMpriors BAMMtools before analysis modified default setting convergence priors analysis expectedNumberOfShifts=1.0 lambdaInitPrior=17.0512659943593 lambdaShiftPrior=0.00279913753644403 lambdaIsTimeVariablePrior=0 used “speciationextinction” model diversification analysis BAMM ran 10 million generations sampling frequency 1000 Convergence assessment analysis rate shifts calculation clade-specific rates performed using BAMMtools test evolution hearing sound production speciation extinction rates fitted models trait-dependent diversification R package hisse42 unmeasured factors impact diversification rates adopted multimodel inference method pruned time-calibrated ultrametric tree include Orthoptera (239 used binary character data sets for hearing sound-producing organs Pagel’s test acoustic communication created additional data set code acoustic communication binary character test evolution diversification rate fitted 24 models to hearing sound production data set for Orthopteramodels included four BiSSE models four trait-independent models CID 16 models HiSSE models hidden state descriptions in Supplementary Methods 3.4. included sampling fraction for observed states calculating proportion known 0’s absence known 1’s presence in tree models compared using AIC analyses carried out in hisse.Reporting summaryFurther information research design in Nature Research Reporting Summary.Supplementary InformationPeer ReviewReporting SummaryDescription Additional Supplementary FilesSupplementary Data 1-15 | 50.1 | 1.127675 |
10.1038/s41467-021-21307-z | PMC7955114 | Some bacterial pathogens release NADase enzymes into the host cell that deplete the host’s NAD+ pool, thereby causing rapid cell death. Here, Strømland et al. identify NADases on the surface of fungal spores, and show that the enzymes display unique biochemical and structural properties. | Nicotinamide adenine dinucleotide (NAD) is a key molecule in cellular bioenergetics and signalling. Various bacterial pathogens release NADase enzymes into the host cell that deplete the host’s NAD+ pool, thereby causing rapid cell death. Here, we report the identification of NADases on the surface of fungi such as the pathogen Aspergillus fumigatus and the saprophyte Neurospora crassa. The enzymes harbour a tuberculosis necrotizing toxin (TNT) domain and are predominately present in pathogenic species. The 1.6 Å X-ray structure of the homodimeric A. fumigatus protein reveals unique properties including N-linked glycosylation and a Ca2+-binding site whose occupancy regulates activity. The structure in complex with a substrate analogue suggests a catalytic mechanism that is distinct from those of known NADases, ADP-ribosyl cyclases and transferases. We propose that fungal NADases may convey advantages during interaction with the host or competing microorganisms. | IntroductionSince its discovery in the 1930s1, nicotinamide adenine dinucleotide (NAD+) has emerged as one of the fundamental metabolic molecules of life and as a central player in cellular signalling processes owing to its function as substrate for various ADP-ribose transfer reactions2,3. Not long after NAD+ was discovered, enzymatic hydrolysis of the dinucleotide was detected. The enzymes that break the N-glycosidic bond between the terminal ribose and the nicotinamide (Nam) moiety are termed NAD+ glycohydrolases or NADases4–6. In addition to cleavage of NAD+ to Nam and ADP-ribose, NADases found in animals act as ADP-ribosylcyclases, generating cyclic ADP-ribose (cADPR), a potent intracellular Ca2+ mobilizing agent7. However, in microbial species, NADases appear to have evolved as powerful weapons to mediate infections5,6,8. For example, Streptococcus pyogenes and Mycobacterium tuberculosis produce enzyme toxins that trigger necrotic cell death of host macrophages by rapid NAD+ depletion. Recently, it was discovered that certain Toll/interleukin-1 receptor (TIR) domains in animals and plants constitute a new family of NADases, which are involved in cell death pathways and defence against infections9–11. For example, mammalian SARM1 is an NADase that mediates axon degeneration through rapid NAD+ degradation9.Even though ubiquitously found in bacteria, plants and animals no NADase has so far been identified in fungi. Earlier reports indicated the presence of NADase activity on conidia and hyphae of the filamentous fungus Neurospora crassa12,13. However, the molecular identity of this activity has remained obscure. Given the widespread occurrence of NADases and their prominent role in microbial infection mechanisms we wondered whether pathogenic fungi may also produce similar enzymes.In this work, we identify NADases on the surface of fungi and show that these enzymes are predominately found in pathogenic species. The A. fumigatus and N. crassa enzymes cleave NAD+ and NADP+ but not their reduced counterparts, NADH and NADPH. Moreover, the enzymes lack both ADP-ribosyl cyclase and base exchange activity. The structure of the dimeric A. fumigatus enzyme revealed the presence of a Ca2+ binding site whose occupancy partially regulates enzymatic activity. In addition, each protomer is N-linked glycosylated at three asparagine residues and stabilized by two disulphide bridges. We also solved the structure of the enzyme bound to the reaction products, nicotinamide and ADP ribose, and the non-hydrolysable substrate analogue benzamide adenine dinucleotide (BAD). The structure bound to the substrate analogue along with mutagenesis of predicted critical amino acid residues led us to suggest a reaction mechanism distinct from all known NADases. We propose that these NADases may represent hitherto unrecognized factors that convey advantages for the fungi during interaction with the host or competing microorganisms in the environment.Results and discussionIdentification of fungal surface NADasesSince conidia from N. crassa have been known to exhibit NADase activity, we tested whether conidia from the human opportunistic pathogen A. fumigatus may have a similar activity13. Using the fluorescent NAD+ analogue nicotinamide 1, N6-ethenoadenine dinucleotide (εNAD+) we detected robust cleavage on the surface of A. fumigatus conidia from the strain CEA17ΔakuB, indicating the presence of one or more NADases (Fig. 1A). NADase activity was present at different growth stages suggesting that one or several surface NADase(s) are expressed during A. fumigatus development (Fig. 1B). NADase activity was also present on conidia from the clinical strains Af293, D141 and ATCC46645 (Supplementary Fig. 1A). Since the surface proteome of A. fumigatus is dependent on the nutritional source14, we investigated whether the activity was influenced by the growth medium. NAD+ cleavage activity was present on conidia cultivated in all media tested, with higher activity found using Aspergillus minimal media and malt agar compared to Saboraud and RPMI agar (Supplementary Fig. 1B). 1H NMR showed that the conidial enzyme(s) cleave(s) NAD+, yielding ADP ribose (ADPR) and Nam as reaction products (Fig. 1C). Thereby, we established the presence of one or more NADases on the surface of conidia from A. fumigatus. To identify the conidial protein(s) of N. crassa and A. fumigatus responsible for NAD+ cleavage, we first visualized their activity in a polyacrylamide gel following SDS-PAGE in non-reducing conditions using ɛNAD+
15. Conidia from both N. crassa and A. fumigatus exhibited strong NADase activity with a migration corresponding to a molecular mass of ~50 kDa (Fig. 1D). Strikingly, NADase activity was still detectable in the A. fumigatus conidia following heat treatment at 95 °C for 5 min, albeit with the majority of the detectible activity migrating as a protein with a molecular mass of ~30 kDa (Fig. 1D). LC-MS/MS-based proteomic analyses of the bands exhibiting NADase activity identified 27 proteins in untreated A. fumigatus conidia compared to eight in heat-treated conidia (Supplementary Data 1). The only overlapping hit between the two samples was the predicted gene product AFUA_6G14470, a protein with unknown function and a theoretical molecular mass of 26 kDa. Of note, this protein has already been detected previously on the surface of conidia14,16. Analyses of conidia from N. crassa yielded peptides corresponding to a hypothetical protein homologous to that identified in A. fumigatus (Fig. 1E). Bioinformatic analyses of the sequences predicted three N-linked glycosylated asparagine residues with high confidence (Fig. 1E). This is in line with studies that showed glycosylation of N. crassa NADase and could explain the discrepancy between the observed and theoretical molecular masses13. A predicted N-terminal secretory signal peptide in the A. fumigatus protein was confirmed by LC-MS/MS (highlighted in blue in Fig. 1E). To verify the identity of the suspected NADase gene, we generated an A. fumigatus knock-out strain lacking the gene AFUA_6G14470. Indeed, we did not observe any NADase cleavage in conidia of this knock-out strain, whereas in-locus complementation restored NADase activity (Fig. 1F, G). These results confirmed that the gene AFUA_6G14470, now named nadA, encodes a conidial NADase in A. fumigatus (GenBank accession number: MT276230). Moreover, we have established the molecular identity of the long-sought N. crassa NADase, encoded by the gene NCU07948 (GenBank accession number: MT316195).Fig. 1Identification of fungal conidial NADases.A NADase activity of A. fumigatus conidia demonstrated by a fluorometric assay using ɛNAD, n = 3. B NADase activity of different A. fumigatus growth stages demonstrated by a flurorometric assay using ɛNAD, n = 3. C Top: identification of nicotinamide (Nam) and ADP ribose (ADPR) as the NAD+ cleavage products following incubation with A. fumigatus conidia by 1H NMR. Bottom: Assignment of relevant NMR signals for NAD+ and its cleavage products. Hydrolysis of NAD+ leads to the formation of Nam and ADPR, the protons giving rise to the NMR signals are labelled blue for NAD+ and red for Nam and ADPR. D Identification of possible fungal NADases. Top: Enzyme activity gels of N. crassa and Aplysia californica cyclase, which was used as a positive control (left) and A. fumigatus (right) conidia developed with ɛNAD. In the right panel, the yellow and cyan asterisks denote the bands obtained with and without heat treatment, respectively. Fluorescent bands were excised and subjected to proteomics analysis. Bottom: Top hits of A. fumigatus proteins identified by mass spectrometry sorted by their peptide spectral matches score (PSM). The only overlapping hit between the two samples was an uncharacterized, predicted protein (highlighted in red). E Global pairwise alignment of predicted NADase sequences from A. fumigatus and N. crassa deduced from the genes identified based on the results shown in (D). The peptides that were detected in A. fumigatus and N. crassa conidia are highlighted in red. Bioinformatic analyses predict the presence of a secretory signal peptide and N-linked glycosylated asparagine residues, shown in blue and green, respectively. F In-gel ɛNADase assay of conidia from A. fumigatus wild type (WT), knockout mutant (KO) strain ΔnadA which lacks the gene predicted to encode the conidial NADase (AFUA_6G14470) and conidia from a complementation strain (Comp) ΔnadA:nadA. G Fluorometric assay of NADase activity of the samples described in (F), n = 3. Experiments in (A, B, D, F, G) were performed independently three times with similar results. Source data are provided as a Source Data file.Fungal NADases hydrolyse NAD+ and NADP+, but do not mediate synthesis of calcium messengers (cADPR or NAADP)Next, we investigated the catalytic and proteo-chemical properties of the fungal NADases. Recombinant A. fumigatus NADase (AfNADase) was expressed in stably transfected human 293 cells as well as recombinant baculovirus-infected Sf9 insect cells. In both systems, the overexpressed enzyme was secreted into the medium and exhibited robust NADase activity (Fig. 2A and Supplementary Fig. 2A, B). The purified proteins, expressed in Sf9 and 293 cells, exhibited a size of ~30 kDa and 40 kDa, respectively, based on their migration in SDS-PAGE, presumably owing to different extent of glycosylation. Size-exclusion chromatography suggested the enzyme to be homodimeric (Supplementary Fig. 2C–F). Subsequent experiments were performed with the protein expressed in and purified from Sf9 cells. As observed with conidia, AfNADase activity was unusually heat-resistant (Tm = 78.5 °C) and could be detected following both 5 and 10 min of incubation at 95 °C (Supplementary Fig. 2G, H). However, the enzyme was sensitive to the reducing agent DTT and the metal chelator EGTA. Inhibition by EGTA was reversed by Ca2+ (Fig. 2B), but not by other bivalent metal ions such as Mn2+, Mg2+ or Zn2+. EGTA titration showed that the enzyme is only partially dependent on Ca2+, as activity was still detected even in the presence of 5 mM EGTA. Interestingly, N. crassa NADase (NcNADase) activity was not affected by either EGTA or Ca2+ (Supplementary Fig. 2I). Both AfNADase and NcNADase cleaved NAD+ and NADP+, but not their reduced counterparts, NADH and NADPH (Fig. 2C, Supplementary Figs. 3, 4). Besides hydrolysis of NAD(P)+ to (phospho)ADPR and Nam, animal NADases (ADP-ribosylcyclases) also catalyse the formation of the Ca2+ messenger cyclic ADPR (cADPR) as well as the exchange of the nicotinamide moiety by nicotinic acid (Fig. 2E). Both AfNADase and NcNADase did not produce any detectable cADPR and did not mediate base-exchange with nicotinic acid (NA), reactions readily catalysed by Aplysia californica cyclase (Fig. 2F). AfNADase also failed to hydrolyse nicotinic acid adenine dinucleotide (NAAD), the deamidated form of NAD+ (Supplementary Fig. 5A). The absence of ADP-ribosyl cyclase activity from the fungal NADase was further substantiated by its inability to convert the NAD+ analogue nicotinamide hypoxanthine dinucleotide (NHD+) to Nam and fluorescent N7-cyclic inosine diphosphoribose (N7-cIDPR), a reaction specifically observed for cyclases (Supplementary Fig. 5B–D). These results demonstrate that fungal NADases are pure NAD glycohydrolases as they lack both cyclase and base-exchange activity. We determined the kinetics of AfNADase-mediated NAD+ and NADP+ hydrolysis by 1H NMR. The hydrolysis of NAD(P)+ was followed by inspecting the resonance decay of NAD(P)+ protons and the corresponding resonance increase in Nam and (2’-phospo-) ADPR protons (Fig. 2D). The KM were calculated to be 119.7 ± 40.8 µM and 106 ± 27.1 µM for NAD+ and NADP+, respectively. The turnover rates were determined to be 1962 ± 133 s−1 and 418 ± 108 s−1 for NAD+ and NADP+, respectively.Fig. 2AfNADase is a pure NAD(P)+ glycohydrolase.A NADase activity of recombinant AfNADase purified from Sf9 insect cells measured by the fluorescence assay using ɛNAD, n = 3. B left: NADase activity of recombinant AfNADase from Sf9 insect cells treated with DTT, EGTA or EGTA and calcium chloride, n = 3 Right: NADase activity of AfNADase from Sf9 insect cells titrated with EGTA and CaCl2. C identification of AfNADase substrate specificity by HPLC. The HPLC chromatograms display AfNADase mediated reactions using the indicated substrates. D Kinetics of NAD+ and NADP+ hydrolysis by AfNADase determined by 1H NMR. The integral of the resonances corresponding to NAD+ N-2 and N-6 were plotted and used for curve fitting. E Absence of ADP-ribosyl cyclase activity in AfNADase. In the presence of an excess of NA some NADases (namely ADP-ribosyl cyclases) can catalyse the exchange of the Nam moiety in NAD+ for NA producing NAAD (“base-exchange reaction”). Cyclases can also produce cyclic ADPR or simply cleave NAD+ to ADPR and Nam. F
AfNADase does not possess base-exchange or ADPR cyclase activity, identifying it as pure NADase. HPLC chromatograms of AfNADase and Aplysia californica cyclase base exchange reactions. Experiments in (A, B, C, F) were performed independently three times with similar results. Source data are provided as a Source Data file.The crystal structure of AfNADase reveals the presence of a TNT domain and distinct structural properties including a regulatory calcium-binding siteNext, we sought to get insights into the structural assembly and the catalytic mechanism of AfNADase. We solved the crystal structure of AfNADase to a resolution of 1.6 Å (Fig. 3A and Supplementary Table 1) (PDB: 6YGE). The protein crystallizes in the space group P3221 with two molecules in the asymmetric unit confirming the homodimeric assembly suggested from the chromatographic analysis (Supplementary Fig. 2C, D). The dimeric assembly is formed via an interface of 2260 Å2 of the total solvent-accessible area of 11030 Å2 (calculated with PDBePISA server), involving 66 residues, including the C-terminus which is intertwined with the other protomer (Fig. 3A). The protomers consist of two domains, an N-terminal ‘thumb’ and a C-terminal ‘palm’ domain. The thumb domain, residues 20-117, is folded containing five α-helices connected with loop regions, and the fold is stabilized by two disulfide bridges, C33-80 and C38-50. The palm domain consists of a seven-stranded central β-sheet flanked by two short α-helices and two 310 helices. As described below, we identified the palm domain as TNT domain, based on the 3D structure. In addition, the C-terminus contains a metal binding site. The location of this binding site is at the dimerization interface, just before the C-terminus turns to intertwine with the other protomer. Noteworthy, an acetate ion, originating from the crystallization solution, is trapped in a cavity located near the domain boarder and facing the central β-sheet (strands β2, 3 and 5).Fig. 3Crystal structure of AfNADase.A Left panel, cartoon representation of AfNADase. The protomers forming the dimer are coloured yellow and blue. The secondary structure elements are sequentially labelled, α and β for helical elements and for beta strands, respectively. The N-linked glycosylated asparagine residues are shown in stick representation. The green sphere represents the bound calcium ion and the residues that are involved in the coordination are shown in stick representation (pink). Right panel, cartoon representation of domain structure (rotated 70° on X-axis from the left panel orientation). The thumb domain is shown in grey and palm domain in blue, and the dimerization interface through the 2-fold symmetry axis is displayed as a dotted line. The disulfide bridges are shown in stick representation. B Stick representation of the observable N-linked glycosylation at asparagine 45, 95 and 118. C Cartoon and stick representations of the calcium binding site showing the residues and water molecules that are involved in the pentagonal bipyramidal coordination of the metal ion. D NADase activity of A. fumigatus calcium binding site mutant (D119A/E220A) measured by the fluorometric assay using ɛNAD, n = 3. The experiment was performed independently three times with similar results. Source data are provided as a Source Data file.The pentagonal bipyramidal coordination of the metal ion in the crystal structure, the refined density as well as the B-factors indicate that the bound metal ion is Ca2+, in line with the observed Ca2+ activation of the enzyme following EGTA treatment (Fig. 2B). The Ca2+ ion binding site is comprised of the side chains of D219, E220 and E223 as well as the main chain of S216. In addition, two water molecules are involved in metal ion coordination (Fig. 3A, C). Each protomer is glycosylated at three asparagine residues, N45, N95 and N118 (Fig. 3B) and these sites coincide with those originally predicted (Fig. 1E). N-linked glycosylation of NcNADase expressed in 293 cells was also observed (Supplementary Fig. 6). Taken together, these observations provide the structural basis for the biochemical findings regarding DTT and EGTA sensitivity, as well as migration in SDS-PAGE according to a molecular mass higher than predicted from the polypeptide alone (Fig. 1D). To elucidate a possible role of the Ca2+ binding site of AfNADase in catalysis, we created a Ca2+-free mutant by substituting D219 and E220 with alanine residues. Disruption of the Ca2+ binding site by mutagenesis led to a sevenfold reduction in ɛNAD+ hydrolysis as compared to the WT enzyme, highlighting the importance of Ca2+ coordination for efficient hydrolysis (Fig. 3C, D). To get a better understanding of the Ca2+ ion in catalysis we attempted to crystallize the Ca2+ binding site mutant of AfNADase, however the protein did not yield diffraction quality crystals. To circumvent this issue, we crystallized the native protein treated with EGTA. However, Ca2+ was still present in the crystal structure, albeit with a lower occupancy. Interestingly, when these crystals were soaked with a tenfold molar excess of NAD+, the Ca2+ ions were no longer present in the crystal, and in molecule A of the asymmetric unit the reaction products, Nam and ADPR, were bound to a putative active site (Fig. 4A, B) (PDB: 6YGF). Nam is located in the deep cavity, where acetate was trapped in the apo structure, and is hydrogen-bonded to the enzyme with the side chains of R129 and R148, the main chain of F130 and a water molecule, from here on referred to as water I (Fig. 4C). These residues are highly conserved, suggesting they play a role in hydrolysis of NAD(P)+ (Supplementary Fig. 7). ADPR is found in close proximity to Nam and rests in a crevice on the surface of the enzyme. The ribose moiety that was bound to Nam is flipped out of the cavity and makes hydrophobic contact with the aromatic ring of Y100, a residue that is also conserved. The two phosphate moieties make extensive contacts with the enzyme through hydrogen bonds with R129, S132, L138, N154 and water I. Finally, the adenosine moiety is bound to the enzyme via π-stacking with the side chain of F158. These observations indicate that the residues interacting with Nam and ADPR indeed constitute the active site of AfNADase. However, given that the distance between the Ca2+ ion and the active site is more than 40 Å, the role of Ca2+ in catalysis cannot be easily explained. Since the activity is lower in the Ca2+ binding site mutant, it is tempting to speculate that the Ca2+ ion plays a role in regulation of the enzymatic activity to prevent suicidal NAD(P)+ depletion as the intracellular concentration of Ca2+ is in the nanomolar range in conidia17.Fig. 4Crystal structure of AfNADase in complex with reaction products or the non-hydrolysable NAD+ analogue BAD.A Complex of AfNADase and the reaction products Nam and ADPR. The protein is shown as a cartoon embedded into a transparent surface map. The palm domain is coloured blue and the reaction products Nam and ADPR are displayed in stick representation. B Calculated mFo-DFc POLDER electron density map contoured at 3.0 σ to confirm the binding of ADPR and nicotinamide. C Residues of AfNADase involved in the interaction with the reaction products Nam and ADPR are shown in stick representation. The water molecule participating in the interaction is labelled WI. The dotted lines represent hydrogen bonds between the protein, water molecule and the reaction products. D In comparison with NAD+ the nitrogen in the pyridine rings has been substituted with a carbon atom in the non-hydrolysable analogue BAD. E Complex of AfNADase and the substrate analogue BAD. The protein is shown as a cartoon embedded into a transparent surface map. The palm domain is coloured blue and BAD is shown in stick representation. F Calculated mFo-DFc POLDER electron density map contoured at 3.0 σ to confirm the binding of BAD. G Residues of AfNADase involved in the interaction with BAD are shown in stick representation. The water molecules participating in the interaction are labelled WI and WII. The dotted lines represent hydrogen bonds between the protein, water molecules and the substrate analogue. H Proposed reaction mechanism of AfNADase-mediated NAD+ hydrolysis. In an SN2-like reaction mechanism, water molecule I is hydrogen-bonded to the in-ring oxygen of the proximal ribose, the nicotinamide moiety, the phosphodiester backbone and R129. This acts as a bridge to stabilize a positive change on the in-ring oxygen. The nucleophilic attack is further prepared by Q194, which interacts with the 2” OH of the proximal ribose and induces a δ-negative charge. The attack on the C1’ position from water II coincides with the nicotinamide leaving the active site and the formation of an oxonium intermediate. I activity of recombinant AfNADase WT and active site mutants R129A, F130A, F137A, Q194A and Q194K measured by the fluorometric assay using ɛNAD, n = 3. The experiment was performed independently three times with similar results. Source data are provided as a Source Data file.Identification of the catalytic mechanism of fungal and other TNT-containing NADases based upon the crystal structure in complex with a non-cleavable substrate analogueTo understand the catalytic mechanism of AfNADase and related NADases such as TNT, we attempted to solve the structure of the enzyme in complex with NAD+ by soaking the crystals with the dinucleotide. However, due to the fast conversion rate of NAD+ to Nam and ADPR, it was not possible to trap the substrate in the crystals. We therefore used the non-cleavable NAD+ analogue BAD, differing from NAD+ by the substitution of the nitrogen atom in the pyridine ring with a carbon atom (Fig. 4D). We were able to co-crystallize AfNADase with BAD bound to the active site of molecule B of the asymmetric unit (Fig. 4E, F) (PDB: 6YGG). Residual electron density can be observed in molecule A but not enough to model BAD into the active site. The structure of the complex clearly illustrates the binding mode of the substrate to the active site, and many of the residues are the same as those involved in product binding (Fig. 4G). The interaction with BAD includes two water molecules; one is found in the same position as water I in the structure of AfNADase bound to the products and the other, water II, makes contact with the proximal ribose (Fig. 4G). In the complex, F137 makes contact with the proximal ribose of BAD positioning the scissile bond into the active site. The benzamide (nicotinamide, by analogy) is also held in place by hydrogen bonding to R129, F130 and R148. The sidechain of Q194 makes contact with the 2” OH group of the proximal ribose. In other NADases, ADPR transferases (ARTs), which catalyse the transfer of an ADPR moiety from NAD+ to an acceptor, as well as in ADP-ribosyl cyclases, a catalytic acidic residue, glutamate or aspartate, is found in the corresponding position. In these enzymes NAD(P)+ is cleaved through the formation of an oxocarbenium intermediate. However, this is unlikely to be the case in AfNADase as the pull from Q194 on the bond would not be strong enough. Since the formation of an oxocarbenium reaction intermediate is required for both cyclase and base exchange activity, this observation explains why AfNADase cannot catalyse these reactions. In order to achieve cleavage of NAD(P)+, a nucleophilic attack on the anomeric carbon (C1’) is needed. In AfNADase, the attack cannot come from the β-face because this is the location of the nicotinamide moiety and the β-face is shielded by its hydrophobic interaction with F137. We propose a mechanism where water molecule I is hydrogen-bonded to the in-ring oxygen of the proximal ribose, the nicotinamide moiety, the phosphodiester backbone and R129. This could act as a bridge to stabilize a positive charge on the in-ring oxygen. The nucleophilic attack is further prepared by Q129 which interacts with the 2” OH of the proximal ribose and can potentially induce a δ-negative charge. This mechanism would resemble a SN2 reaction with the formation of a transition state, in which the attack on the α-face on the C1’ position from water II coincides with the nicotinamide leaving and the formation of a partial oxocarbenium ion (oxonium intermediate) (Fig. 4H). The proposed mechanism precludes NAAD cleavage since the interaction between the carboxylate of the nicotinic acid moiety, watermolcule I and the side chain of R129 is less likely to lead to catalysis, in line with our observation that AfNADase does not cleave NAAD (Supplementary Fig. 5A). In support of the proposed catalytic mechanism, mutagenesis of R129, F137 and Q194 inactivated the enzyme (Fig. 4I). Mutagenesis of F130 did not completely inactivate the enzyme. This is not surprising given that F130 interacts with the product/substrate with the carbonyl and amide backbone atoms (Fig. 4H). Substitution of R148, which interacts with the nicotinamide (moiety), was also attempted. However, neither Western blotting nor activity measurements indicated the presence of the recombinant protein expressed in 293 cells, suggesting that the protein is unstable and rapidly degraded. In line with this supposition, R148 forms a salt bridge with D128 that may be vital for protein stability. Indeed, mutation of the corresponding R780 in TNT also resulted in an unstable protein (18). These residues are conserved in all TNT domain-containing proteins in both fungi and bacteria (Supplementary Fig. 7).Fungal NADases are predominantly present in pathogenic speciesNext, we investigated the phylogenetic relationship between AfNADase and other NADases. Initial BLAST searches using the primary structure of AfNADase and standard parameters did not return any known NADases. However, a Pfam search revealed that the palm domain of AfNADase, residues 117-234, harbours a tuberculosis necrotizing toxin (TNT) domain (Fig. 5A). Intriguingly, the TNT domain was recently identified in the M. tuberculosis (Mtb) protein CpnT as a toxin with NADase activity, which mediates toxicity towards host macrophages by depleting NAD(P)+. The structure of Mtb CpnT TNT, from here on referred to as TNT, has been solved in complex with its immunity factor, an endogenous inhibitory protein which prevents suicidal NADase activity (PDB: 4QLP)8. No homologue of the gene encoding the M. tuberculosis immunity factor was found in the genomes of N. crassa and A. fumigatus. An immunity factor may not be needed in fungi since the enzyme enters the secretory pathway and secretion precludes it from interacting with the major cellular NAD+ pools. Another possible explanation is that AfNADase activity is regulated by Ca2+ and therefore an immunity factor is not required. However, we noted certain structural similarities between AfNADase and TNT (Supplementary Fig. 8). The palm domain is similar between AfNADase and TNT and they can be superimposed with an RMSD of 0.83 Å (Supplementary Fig. 8). However, the thumb domain of AfNADase differs markedly from the thumb domain of TNT. In AfNADase, this region is stabilized by two disulfide bridges and contains two N-linked glycosylation sites, N45 and N95; features that are not present in TNT. The most distinct feature of the AfNADase palm domain is the C-terminal extension harbouring the Ca2+ binding site. In AfNADase the C-terminus is intertwined with the other protomer, whereas the C-terminus of TNT is located immediately after the last helix in the core fold (helix 10 in AfNADase). The active site of TNT coincides with that of AfNADase and critical residues and sequence motifs are conserved (Supplementary Figs. 7 and 8). AfNADase and TNT exhibit the same substrate specificity and both lack cyclase and base exchange activity (Fig. 2 and Supplementary Fig. 5)18. Therefore, the fungal NADases can be classified as TNT domain-containing proteins, which are widely found in bacteria and fungi. Moreover, the proposed reaction mechanism of AfNADase is likely valid for all TNT domains.Fig. 5AfNADase is the founding member of a family of fungal NADases.A Domain architecture of AfNADase, NcNADase and Mtb CpnT TNT. B Structure-based alignment of active site motifs of NADases and ADP-ribosyltransferases. The residues are found in three distinct regions and are coloured according to their location in the sequence. C Comparison of the active sites of AfNADase in complex with BAD, Mtb CpnT TNT, Tse6 and cholera toxin. Amino acids are coloured based on their position in the active site motif, as in (B). D Phylogenetic distribution of NADase within the kingdom fungi. Classes with the NADase are coloured red or green, red for classes without the calcium binding site and green for those with the calcium binding site, and the number in parentheses represents the number of species within the class that harbour the NADase. The calcium binding site emerged (blue star) in the Eurotiomycetes, and is only present in the order Eurotiales (inset).The fungal NADases also share commonalities with more distantly related NADases and ADP-ribosyltransferases (ARTs), such as Tse6, cholera toxin and diphtheria toxin, as their active sites can be aligned (Fig. 5B, C and Supplementary Table 2). ARTs are divided into two groups based on active site motifs distributed across three regions. The cholera toxin has an R-S-E motif, whereas diphtheria toxin has an H-Y/Y-E motif, which is also present in poly (ADP-ribose) polymerases (PARPs). NADases contain similar sequence motifs and in AfNADase the motif consists of R129, F137 and Q194 residues that are vital for NADase activity.Bacteria and fungi with a predicted TNT domain protein mainly belong to the phyla Firmicutes, Actinobacteria and Ascomycota. However, the fungal NADases exhibit specific structural and functional properties, distinct from their bacterial counterparts. These properties include the expression on the outer surface of the conidia and hyphae, the homodimeric nature, a structurally unique thumb domain, posttranslational glycosylation as well as the presence of critical disulfide bonds. In addition, in some species such as A. fumigatus, the catalytic activity is regulated by Ca2+ ions.The Ca2+ binding site found in AfNADase emerged in the order Eurotiales, which includes Aspergillus spp (Fig. 5D). The predicted 3D structure of NcNADase is highly similar to that of AfNADase (Supplementary Fig. 9), except that it lacks the residues involved in Ca2+ binding found in AfNADase (Fig. 1D). A comprehensive phylogenetic analysis of the distribution of the fungal NADases revealed that they are predominantly present in families and species that are known to be pathogenic suggesting a potential role of the protein in fungal virulence (Fig. 5D). The conidia of A. fumigatus are ubiquitously present in the environment and normally cleared from the lungs by the immune system19. However, aspergilloses are an increasing problem in animal husbandry and represent a considerable threat for immune-compromised humans20. A. fumigatus spores harbour an extensive array of virulence factors to establish an infection which could potentially be therapeutic targets21. However, the spectrum of antifungal drugs is rather limited, and development of resistance has become a major challenge for the treatment of these infections22–26. Interestingly, the gene encoding AfNADase is upregulated during conidiation of A. fumigatus strains displaying high adherence to pulmonary epithelial cells27.In summary, we have discovered a group of fungal surface NADases that have distinct structural and functional features, and have characterised the NADase from A. fumigatus. Importantly, no homologues in mammals have been found. Given their functional similarity to known toxins with NADase activity, and predominant presence in pathogenic fungi, we speculate that fungal NADases may convey advantages during interaction with the host or competing microorganisms in the environment.MethodsFungal strains, media and growth conditionsAspergillus fumigatus strain CEA17ΔakuB28 was used as parental strain and for generation of the ΔnadA knockout and complementation strains. For conidiation, strains were cultivated on Aspergillus minimal medium 1.5% (w/v) agar plates at 37 °C (AMM; 70 mM NaNO3, 11.2 mM KH2PO4, 7 mM KCl, 2 mM MgSO4, (pH 6.5) with 1% (w/v) glucose and 1 µl/ml trace element solution: 18 mM FeSO4, 171 mM EDTA, 77 mM ZnSO4, 180 mM H3BO3, 25 mM MnCl2, 6.7 mM CoCl2, 6.4 mM CuSO4, 0.9 mM (NH4)6Mo7O2429,30. After 3–7 days, conidia were harvested in sterile, ultra-filtrated water using a cell scraper and counted in a Neubauer chamber. For isolation of genomic DNA, A. fumigatus was grown in liquid AMM for 24 h.Isolation and manipulation of A. fumigatus nucleic acidStandard techniques for manipulation of DNA were carried out using standard procedures. Chromosomal DNA of A. fumigatus was prepared using the Master Pure Yeast DNA Purification Kit (Epicentre Biotechnologies, USA). For Southern blot analysis, DNA fragments were separated on an agarose gel and blotted onto Hybond N+ nylon membranes (GE Healthcare Bio-Sciences, Germany). DNA probes were labelled using the DIG labelling mix (Roche Applied Science, Germany) following manufacturers recommendations. Hybridization and detection of DNA-DNA hybrids were performed using DIG Easy Hyb and a CDP-star ready-to-use kit (Roche Applied Science, Germany) following manufacturers recommendations31.Generation of A. fumigatus mutant strainsDeletion of nadA was done by using a PCR-based strategy. Upstream and downstream flanking regions of gene nadA (AFUA_6G14470) were amplified by PCR using primer pairs 6G14470_5for (GGTCATTGTAAATATCTGGG) and 6G14470_ptrArev (GGCCTGAGTGGCCATCGAATTCCGCCGTGTAATACTGAGAAG) and 6G14470_ptrAfor (GAGGCCATCTAGGCCATCAAGCCTTATGGGAAGTGGATCTTG) and 6G14470_3rev (GTAGTGGATAACGAAGATTCG), respectively (Supplementary Table 3). By this reaction overlapping ends to the pyrithiamine resistance cassette were introduced at the 3’-end of the upstream flanking region and at the 5’-end of the downstream flanking region of the nadA gene. The ptrA resistance cassette was amplified from plasmid pSK27532 with primers ptrA-for (GAATTCGATGGCCACTCAGGCC) and ptrA-rev (GCTTGATGG CCTAGATGGCCTC). All PCR reactions were performed with Phusion Flash Polymerase Master Mix (Thermo Scientific, Germany) according to the manufacturer’s recommendations and PCR fragments were purified by gel extraction. The final deletion construct was generated by a three fragment PCR employing primers 6G14470_5for and 6G14470_3rev. The resulting PCR product was purified and used for transformation of A. fumigatus protoplasts as by electroporation33. Pyrithiamine (1 mg/mL, Merck, Germany) resistant transformants were analyzed for deletion of nadA by Southern blot analysis.To create the complemented strain, the deletion strain ΔnadA was complemented in-locus using the same approach and reagents as described above. The nadA gene was amplified from genomic DNA of A. fumigatus strain CEA17ΔakuB using primers 14470_fwd (AGGCGTATCACGAGGCCCTTTCGTCGGTCATTGTAAATATCTGGG) and 14470_rev (CAATAGTGCCACGCTATTGGGATCACTGGC). The hph resistance cassette was amplified from plasmid pUChph34 using primers hph_fwd (TGATCCCAATAGCGTGGCACTATTGATCATCC) and hph_rev (GGCCATCGAATTCGCCAGTGTGCTGGAATTC). The ptrA resistance cassette was amplified using primers compl_ptrA_fwd (CAGCACACTGGCGAATTCGATGGCCACTCAG) and compl_ptrA_rev (TCACCGTCATCACCGAAACGCGCGAGCTTGATGGCCTAGATGG). The complementation construct consisting of the amplified nadA gene (including promoter and terminator), hph and ptrA resistance cassettes was generated by a multifragment PCR, purified and transformed as described above.Transformants resistant to hygromycin (150 µg/ml, Roche Applied Science, Germany) were analyzed for complementation of nadA by Southern blot analysis.Fluorometric determination of NADase activityThe NADase activity of conidia, purified proteins and medium from cells transiently overexpressing AfNADase or mutants were determined by measuring the increase in fluorescence upon cleavage of the fluorescent NAD analogue nicotinamide 1, εNAD+. Reactions were prepared in 200 μl reaction buffer (50 mM Na-Acetate pH 5.5, 150 mM NaCl, 0.5 mM CaCl2) supplemented with 80 µM εNAD+. The reaction was started by adding conidia, medium from transiently transfected 293 cells or purified AfNADase. The initial reaction rate was monitored by measuring the change in fluorescence over time at 410 nm produced by excitation at 300 nm. The experiments were performed using a Cary Eclipse fluorescence spectrophotometer (Varian)Fluorometric determination of NADase activity of A. fumigatus at different growth stagesA. fumigatus conidia in Aspergillus minimal medium were incubated in a 96 well plate at 37 °C for the indicated time. The reaction was started by adding εNAD+ to a final concentration of 80 µM and the initial reaction rate was monitored by measuring the change in fluorescence over time at 410 nm triggered by excitation at 310 nm. The experiments were performed using an Infinite M200Pro plate reader (Tecan).In gel NADase activity assaySamples of N. crassa conidia and A. fumigatus conidia were prepared by mixing them with 1X non-reducing SDS-PAGE sample buffer (50 mM Tris-HCl pH 6.8, 2 % (v/v) SDS, 10 % (v/v) glycerol, 0,005 % bromophenol blue). In addition, samples of heat-treated A. fumigatus conidia were prepared by incubating the spores at 95 °C for 5 min. The samples were run on a 12 % non-reducing SDS polyacrylamide gel at 10 mA, following electrophoresis the gels were washed twice in washing buffer (50 mM Tris-HCl pH 7, 0.5 CaCl2 0.5 % (v/v) NP-40). The gel was developed by incubation for 5 min in developing buffer (50 mM Tris-HCl pH 7, 0.5 CaCl2, 0.5 % (v/v) NP-40, 80 μM εNAD+). The fluorescent bands were visualized with a UV-transilluminator and excised for mass spectrometry analysis.Sample preparation for LC-MS/MS analysisThe bands showing activity in the fluorometric assay (Fig. 1D) were excised from the gel, washed and destained by repetitive alternating incubation in acetonitrile and 50 mM NH4HCO3. Disulfide bonds in the proteins were reduced using 20 mM Tris(2-carboxyethyl)phosphine for 30 min at 55 °C. Reduced cysteine residues were carbamidomethylated using 25 mM iodoacetamide for 30 min at RT before gel pieces were washed and dried. For tryptic digestion, gel pieces were reconstituted with Trypsin/LysC (50 ng/µl, Promega) solution and incubated at 37 °C for 18 h. Peptides were extracted by three steps of sonication (first extraction 1/49/50 trifluoroacetic acid/water/ acetonitrile, second extraction 1/29/70, third extraction 1/9/90). Extracts were pooled, dried and dissolved in 0.05% TFA in 2/98 acetonitrile/water for LC-MS/MS analysis.LC-MS/MS analysisLC-MS/MS analysis of tryptic peptides was performed on an Ultimate 3000 RSLC nano instrument coupled to a QExactive Plus mass spectrometer (both Thermo Fisher Scientific). Tryptic peptides were trapped for 4 min on an Acclaim Pep Map 100 column (2 cm × 75 µm, 3 µm) at a flow-rate of 5 µL/min. The peptides were then separated on an Acclaim Pep Map column (50 cm × 75 µm, 2 µm) using a binary gradient (A: 0.1% (v/v) formic acid in H2O; B: 0.1% (v/v) formic acid in 90:10 (v/v) ACN/H2O): 0–4 min at 4% B, 10 min at 7% B, 40 min at 10% B, 60 min at 15% B, 80 min at 25% B, 90 min at 30% B, 110 min at 50% B, 115 min at 60% B, 120–125 min at 96% B, 125.1-150 min at 4% B. Positively charged ions were generated by a Nanospray Flex Ion Source (Thermo Fisher Scientific) using a stainless steel emitter with 2.2 kV spray voltage. Ions were measured in data-dependent MS2 (Top10) mode: Precursor ions were scanned at m/z 300-1500 (R: 70,000 FWHM; AGC target: 1 ∙ 106; max. IT: 120 ms). Fragment ions generated in the HCD cell at 30% normalized collision energy using N2 were scanned (R: 17,500 FWHM; AGC target: 2e5; max. IT: 120 ms) using a dynamic exclusion of 30 s.Protein database searchThe MS/MS data were searched against the Uniprot database of Aspergillus fumigatus Af293 / Neosartorya fumigata Af293 using Proteome Discoverer 2.2 and the algorithms of Mascot 2.4.1, Sequest HT, and MS Amanda 2.0. Two missed cleavages were allowed for tryptic peptides. The precursor mass tolerance was set to 10 ppm and the fragment mass tolerance was set to 0.02 Da. Dynamic modifications were set as oxidation of Met and acetylation of the protein N-terminus. The static modification was set to carbamidomethylation of Cys. One unique rank 1 peptide with a strict target false discovery (FDR) rate of <1% on both peptide and protein level (compared against a reverse decoy database) were required for positive protein hits.Determination of substrate specificity by HPLCReactions were prepared in 1 ml reaction buffer (50 mM Tris-HCl pH 8.0, 0.5 mM CaCl2, 150 mM NaCl) supplemented with 200 µM substrate (NAD+, NADH, NADP+, NADPH). After adding 10 ng purified AfNADase and subsequent incubation at 20 °C for 1 h 50 µl of the reaction were analysed by HPLC using a CC 250/3 nucleosil 100-3 C18 HD column (Macherey-Nagel cat.no. 721492 30). Samples were run with a gradient between buffer A (10 mM ammonium acetate pH 7.5, 2 mM TBA-bromide, 10% (v/v) acetonitrile) and buffer B (10 mM ammonium acetate pH 7.5, 2 mM TBA-bromide, 90% (v/v) acetonitrile). The following gradient was used: 0 min: 0.1% B; 1 min: 0.1% B; 7 min: 4% B; 19 min: 22% B; 21 min: 90% B; 23.5 min: 90% B; 26 min: 0.1% B; 35 min: 0.1 % B. The flow rate was set to 0.6 ml/min and the column compartment was kept at 30 °C during the run. The progress of the run was visualised using a UV-VIS detector at the wavelengths of 259 and 340 nm.ADPR cyclization and base exchange assayed by HPLCReactions were prepared in 1 ml reaction buffer (Na-PO4 pH 6.8 or Na-Acetate pH 4, 0.5 mM CaCl2) supplemented with 1 mM NA and 100 µM NAD+. After adding 10 ng purified AfNADase and subsequent incubation at 20 °C for 1 h 50 µl of the reaction were analysed by HPLC as described above. Aplysia californica ADP-ribosyl cyclase (Sigma-Aldrich, CAS: 135622-82-1) was used as positive control.Fluorometric cyclization assayThe cyclization activity of purified AfNADase was assayed using the NAD+ analogue nicotinamide hypoxanthine dinucleotide (NHD+) which is converted to the fluorescent N7-cIDPR and nicotinamide by ADPR cyclases. Reactions were prepared in 200 µl reaction buffer (50 mM Na-Acetate pH 5.5, 150 mM NaCl, 0.5 mM CaCl2) supplemented with 40 µM NHD+. The reaction was started by addition of 10 ng purified AfNADase and followed by measuring the change in fluorescence over time at 410 nm produced by excitation at 300 nm. The experiments were performed using a Cary Eclipse fluorescence spectrophotometer (Varian).Expression and purification of AfNADase using Sf9 insect cellsThe sequence encoding A. fumigatus NADase (gene: AFUA_6G14470) was amplified from cDNA and inserted in-frame with a C-terminal C3 cleavage site and 6xHis-Tag in the pFastBacNKI-ORF-3C-His insect expression vector (NKI Protein Facility, The Netherlands). Briefly, the empty vector, linearized with KpnI, and the amplified insert were separately treated with T4 polymerase in the presence of dTTP to generate single stranded overhangs. The reaction was terminated by adding 25 mM ethylenediaminetetraacetic acid (EDTA) and by heat inactivation at 75 °C for 20 min. The linearized vector and insert were gel purified with the NucleoSpin® Gel and PCR Clean-up kit (Machery-Nagel). The linearized vector and insert were annealed and used for transformation of One Shot™ TOP10 chemically competent E. coli cells (ThermoFischer Scientific). Positive clones were identified by colony PCR and plasmids isolated using the NucleoSpin® Plasmid kit (Machery-Nagel), and the sequence of the plasmid was controlled by Sanger sequencing. DH10EMBacY E. coli cells (EMBL, Grenoble) were used for bacmid generation following the protocol described by35. Sf9 cells (Invitrogen) in solution at a density of 0.5 million cells/ml were transfected with the AfNADase bacmid using the Cellfectin™ II Reagent (ThermoFischer Scientific), and the virus particles were harvested after seven days. Subsequently the recombinant primary virus was amplified to a high-titre viral stock.For protein expression Sf9 cells were cultivated in Sf-900™ II SFM (Gibco™) medium and infected with high titre viral stocks at a density of 1.5-2 million cells/ml. Three days past growth arrest the cells were pelleted by centrifugation and the medium filtered using a 0.22 µm filter (Merck Millipore). The secreted enzyme was purified by immobilized metal affinity chromatography using a HisTrap™ excel (GE Healthcare) column connected to an ÄKTA pure chromatography system (GE Healthcare) following manufacturers recommendations. The column was washed with washing buffer (50 mM Tris-HCl, pH 8, 300 mM NaCl) and the protein was eluted with elution buffer (50 mM Tris-HCl, pH 8, 300 mM NaCl, 500 mM Imidazole). The purified protein was concentrated using 10 kDa MWCO Amicon® Ultra Centrifugal filters (Merck Millipore) and further purified by size exclusion chromatography using a Superdex 200 16/60 HiLoad prepgrade column (GE Healthcare) connected to an ÄKTA pure chromatography system (GE Healthcare). The size exclusion purification was performed using a SEC buffer (50 mM Tris-HCl pH 8, 300 mM NaCl, 2 mM TECEP) and typically the yield was 20-25 mg per litre of medium.Expression and purification of NcNADase and AfNADase using 293 cellsThe sequence encoding NcNADase (gene: NC_026504.1 in N. crassa OR74A) was codon-optimized for mammalian expression, synthesised and inserted into the vector pUC57 by GenScript. The synthesised sequence was flanked by a 5’-end HindIII restriction site and a Kozak sequence, and at the 3’-end by a KpnI restriction site. The resulting pUC57 vector was restriction digested with HindIII and KpnI, prior to ligation into the vector pCMV-Flag5a via the HindIII and KpnI restriction sites. The newly generated plasmid pCMV-NcNADase-Flag was isolated and sequenced. 293 cells were transiently transfected using the Effectene transfection reagent (Qiagen) following manufacturers recommendations, and the cell medium was harvested 96 h post transfection. The medium containing the overexpressed NcNADase was concentrated with 10 kDa MWCO Amicon® Ultra Centrifugal filters and subsequently the enzyme was purified using Anti-FLAG M2 affinity gel (Sigma-Aldrich) following manufacturers recommendations with the following modifications: Triton X-100 and NaCl was added to the medium to the final concentrations of 0,5 % (v/v) and 300 mM, respectively. 40 µL of the affinity gel was added to the medium and incubated for 2 h on a rotating wheel at 4 °C. Subsequently the affinity gel was washed twice with TBS (Tris-HCl pH 7.4, 150 mM NaCl) and then four times with TBS containing 1 M NaCl. The protein was eluted by incubation with 3X FLAG-peptides (5 µg/ul) in TBS for 30 min on a rotating wheel at 4 °CThe sequence encoding A. fumigatus NADase (gene: AFUA_6G14470) was amplified from cDNA, using the forward primer 5’-GTTGGATCCCCACCATGATCTTCACCAAC-3’ and the reverse primer 5’-GCATAGAATTCCTAGTGATGGTGATGGTGATGCTGATTCGGCCCCGGAGTATAC-3’. The resulting PCR amplicon is endowed with a 3’-end BamHI restriction site followed by a Kozak sequence and a 5’-end EcoRI restriction site followed by a sequence encoding a C-terminal hexahistidine tag. The PCR product was restriction digested with the aforementioned restriction enzymes and inserted into the vector pcDNA3.1(+). The resulting plasmid pcDNA3.1(+)-AfNADase-6XHis which encodes a C-terminal his tagged AfNADase was isolated and sequenced. Stably transfected 293 cells were generated by transfection using the calcium phosphate precipitation method followed by two rounds of selection in the presence of 550 µg/ml G418. Monoclonal cell lines were adapted to grow in the chemically defined medium Gibco™ FreeStyle™ 293 Expression Medium (Thermo Fisher). The overexpressed protein secreted into the medium was purified as described above for the Sf9 insect cells.Generation of AfNADase mutantsMutants for expression in Sf9 insect cells and 293 cells were generated by mutating the parental plasmids using the Q5® Site-Directed Mutagenesis Kit (New England Biolabs Inc.) following manufacturers recommendations. The primers were designed using the NEB base exchanger web tool (New England Biolabs Inc.) The F137A mutant was generated using the primes 5’-GTATGGCACCGCTCTGGCGCCGC-3’ and 5’-TCCGATCCGAAACGGTCAAG-3’. The Q194A and Q194K was produced with the primers 5’-GATGGGGACGGCTTTCGTGACATATACCAATG-3’, 5’-CCTGGCTGCTCAAACCAA-3’, 5’-GATGGGGACGAAGTTCGTGACAT-3’ and 5’-CCTGGCTGCTCAAACCAAG-3’, respectively. The R129A mutant was generated using the primers 5’-GAAGCTTGACGCGTTCGGATCGGAGTATGG-3’ and 5’-ATGCCAACCGGTAAGGTC-3’. The F130A mutant was produced with the primers 5’-GCTTGACCGTGCGGGATCGGAGTATG-3’ and 5’-TTCATGCCAACCGGTAAG-3’. The calcium site binding mutant (D219A/E220A) was generated with the primers 5’-GAGCGAGTATGCTGCCAAGGTGGAATACTC-3’ and 5’-TCATCCAACCGTCGCAAG-3’.Crystallization and structure determinationCrystallization screening for native AfNADase was performed in a high throughput manner using a sparse matrix approach utilizing the JCSG + and PACT premier HT screens from Molecular Dimensions. Sitting drop vapour diffusion crystallization trials were set up on Swiss CI SD3 plates at 20 °C and 8 °C against 40 µl reservoir, with ratios of 2:1, 1:1 and 1:2 of protein and precipitant in the drop, respectively. The total drop size was 600 nl and the plates were set up using Mosquito LCP crystallization robot (TTP Labtech). Crystallization was followed with a ROCK IMAGER (Formulatrix) instrument at 20 °C, and manually for the 8 °C plates. The first crystals appeared within a few days and growth continued for up to one-week. Initial crystallization conditions for native AfNADase were 0.1 M sodium acetate pH 5, 0.2 M CaCl2 and 25 % (w/v) polyethylene glycol (PEG) 6000. Optimized crystals used for data collection were grown at 20 °C using hanging-drop vapour diffusion in conditions containing 0.1 M sodium acetate pH 5, 0.3 M CaCl2 and 20-25 % (w/v) PEG 8000. Prior to the data collection, crystals were cryoprotected by briefly soaking them in cryo-protectant, consisting of the crystallization solution supplemented with 20 % glycerol (v/v).Co-crystallization of AfNADase and BAD was performed by adding BAD to a final concentration of 5 mM in the protein solution prior to setting up the crystallization. BAD was a kind gift from John Pascal or synthesised enzymatically from benzamide riboside. The crystallization condition was the same as for the native protein. The cryo-protectant used for these crystals was supplemented with 10 mM BAD.EGTA-treated AfNADase was crystallized in 0.04 M potassium phosphate monobasic, 16% PEG 8000 and 20% glycerol. Because of the glycerol in the crystallization condition cryoprotection was not needed but crystals were soaked for 1 min in the crystallization condition supplemented with 10 mM NAD prior to mounting in Litho loops (Molecular Dimensions) and flash freezing in liquid N2.Crystals used for single-wavelength anomalous dispersion (SAD) phasing were co-crystallized with 100 mM NaI using the optimized condition for native AfNADase. In addition, the crystals were soaked in cryoprotectant supplemented with 20% glycerol and 500 mM NaI before being mounted in Litho loops and flash frozen in liquid N2.Phasing was done at P13, operated by EMBL Hamburg at the PETRA III storage ring (DESY, Hamburg, Germany), utilizing iodide derivatives of AfNADase and the SAD method. The data were collected with 6 kEV energy to reach anomalous scattering coefficient of 11 e− (f″). 720° with 0.2° oscillation of data were collected to maximize the multiplicity. The resulting autoprocessed data (XDS and aimless), showing significant anomalous signal, was used for determining the phases with AutoSol36 and further building of the initial model with AutoBuild37, both from the PHENIX package38. 35 heavy atom sites with partial occupancy were found, and the space group was determined to be P3221. The initial model was then used as a search model for molecular replacement in Phaser39 with high-resolution native data also collected at P13 using 12.7 keV energy.Data for the EGTA and NAD treated crystal were collected at the P11 end station operated by DESY at the PETRA III storage ring (DESY, Hamburg, Germany) using 12 keV energy, and the BAD co-crystallization data were collected at the I04-1 at the Diamond Light Source (Ditcot, Great Britain) using 13.5 keV energy.Data sets were processed using XDS40 and scaled with Aimless41. Crystals structures were refined using Phenix.refine38 and manual inspection and model building was performed using Coot42. Structures were validated with MolProbity38 during the refinement cycles. Ligand restraints were created using eLBOW43. POLDER maps were calculated using the POLDER maps program in the Phenix package44. Details for the data collection and refinement are shown in Supplementary Table 1. Structures were illustrated using PyMol and UCSF Chimera. Structures are deposited in the Protein Data Bank under ID: 6YGE, 6YGF and 6YGG.Determination of A. fumigatus conidia NAD+ cleavage products by 1H NMRA. fumigatus conidia were incubated at room temperature for 1 h with 500 uM NAD+ in NMR reaction buffer (25 mM sodium phosphate pH 5.8, 50 mM NaCl, 5 % (v/v) D2O) in a final volume of 500 µl. Subsequently the conidia were removed by centrifugation followed by filtration through a 10 kDa MWCO Amicon® Ultra Centrifugal filter (Merck Millipore). The reaction products where identified by NMR and data were collected on a Bruker Ascend 850 MHz instrument fitted with a cryogenically cooled triple resonance 5 mm TCI probe with pulse filed gradients along the z-axis at 23 °C. The reaction products were identified by 1H NMR using the pulse sequence zgesgppe allowing water suppression using excitation sculpting with pulse field gradients and perfect echo. The spectra were acquired with 16 scans and a recovery delay of 3.9 secondsDetermination of AfNADase kinetics by 1H NMRNMR data were collected on a Bruker Ascend 850 MHz instrument fitted with a cryogenically cooled triple resonance 5 mm TCI probe with pulse filed gradients along the z-axis at 23 °C. The hydrolysis of NAD+ and NADP+ was measured by 1H NMR using the pulse sequence zgesgppe allowing water suppression using excitation sculpting with pulse field gradients and perfect echo. The specta were acquired with eight scans and a recovery delay of 3.9 seconds. The reactions consisted of 500 µM NAD+ or NADP+ in NMR reaction buffer (25 mM sodium phosphate pH 5.8, 50 mM NaCl, 5 % (v/v) D2O) in a final volume of 500 µl. Reference spectrums of NAD+ and NADP+ in the absence of AfNADase were acquired. The hydrolysis of NAD+ and NADP+ were measured by adding AfNADase to a final concentration of 0.2 nM, the time of enzyme addition was recorded, and spectra were acquired until the substrate resonances declined to the baseline level. The resulting spectra were manually phased and base line corrected. The resonances were assigned using standard correlation methods. The resonances of interest were integrated using the program Dynamics centre 2.5 (Bruker) and compared to the reference spectra of NAD+ and NADP+, respectively. The kinetic parameters of AfNADase were determined by progress curve analysis as described by Golicnik, M45 using MatLab.Phylogenetic analysisTo gain further insight into the role and evolutionary origin of AfNADase we scanned the NCBI protein database for potential fungal homologues. For this, we obtained the fungal phylogeny from the timetree database at the “class” rank compromising 15 fungal classes46. We then performed standard BLAST searches47. Our search database contained all NCBI non-redundant protein sequences (nr) and was restricted taxonomically to each of the 15 fungal classes (taxid:4751). We used default parameter settings of blastp but filtered for low complexity regions and set the number of target sequences to 500 and the expected value to 1. Homology to the identified calcium binding motif DE[KV]E from the A. fumigatus sequence was used as an indicator for the possibility of calcium binding. We did not identify any homologous sequence of non-Eurotiales species containing this or a highly similar motif, suggesting it is specific to this clade.Synthesis of benzamide ribosideFig. 6 Synthesis of 3-(1’-β-D-Ribofuranosyl)benzamide (benzamide riboside). Synthesis adapted from48.Step 1: 3-(2,3,5-Tri-O-benzyl-1-β-D-ribofuranosyl)benzonitrileUnder nitrogen atmosphere, a stirred solution of 3-iodobenzonitrile (1.50 g, 6.5 mmol) in anhydrous THF (97 mL) was cooled to −78°C and a solution of isopropylmagnesium chloride (3.70 mL, 2 M in THF) was added thereto. The mixture was stirred at −78 °C for 2 h at −78°C. The above reaction mixture was transferred under nitrogen through a cannula into a solution of 2,3,5-tri-O-benzyl-D-ribono-1,4-lactone (2.73 g, 6.5 mmol) in anhydrous THF (18 mL) stirred at −78 °C. The stirred reaction mixture was allowed to react at −78 °C for 2 h. It was allowed to reach ambient temperature (25 °C) and left at this temperature and at stirring overnight to give a pink solution. Next day, a saturated aqueous solution of sodium bicarbonate (60 mL) was added, and the mixture was extracted with ether (ca. 300 mL). The organic layer was separated, washed with brine, separated again, dried over Na2SO4, filtered, and concentrated under vacuum to give yellow oil (3.41 g). The crude product was dissolved in toluene and volatiles were evaporated (repeated 3 times) to remove any possible traces of water. The residue was diluted with anhydrous CH2Cl2 (16 mL), and the solution was cooled under nitrogen atmosphere to −78 °C. Boron trifluoride etherate (1.65 mL, 13.1 mmol) was added from a syringe by drops (over ca. 5 min period) to the solution, followed by addition of triethylsilane (2.1 mL, 13.1 mmol). The reaction solution was stirred at −78 °C for 1 h. The solution was allowed to reach ambient temperature (25°C) and stirred overnight. The reaction was quenched with a saturated aqueous solution of NaHCO3 and the reaction mixture was extracted with CH2Cl2. Organic layer was separated, dried over Na2SO4, filtered, and evaporated under reduced pressure to give pink-yellow oil (2.98 g). The crude product was twice purified by MPLC column chromatography (100 g SNAP Ultra SiO2 Biotage column, l = 15.5 cm; d = 3.7 cm) with UV detection (254 and 280 nm) with elution in step-wise gradient mode from hexanes to hexanes/EtOAc mixtures (containing from 15% to 25% of EtOAc) to give 3-(2,3,5-tri-O-benzyl-1-β-D-ribofuranosyl)benzonitrile (1.708 g, 51%) as a colourless oil. 1H NMR (δ, CDCl3, 400 MHz): 3.58 and 3.65 (AB part of ABX system, JAB = 10.3 Hz, JAX = 4.0 Hz, JBX = 3.6 Hz, 2H, H5’A and H5’B), 3.73 (dd, J = 7.7 Hz, 5.1 Hz, 1H, H2’), 4.01 (dd, J = 5.1 Hz, 3.2 Hz, 1H, H3’), 4.34 − 4.37 (m, 2H, H4’ and 1/2 CH2 from Bn), 4.51–4.61 (m, 5H, 2×CH2 and 1/2 CH2 from Bn), 4.97 (d, J = 7.4 Hz, 1H, H1’), 7.12–7.15 (m, 2H), 7.23–7.38 (m, 14H), 7.53 (d, J = 7.7 Hz, 1H), 7.59 (d, J = 7.7 Hz, 1H), 7.66 (s, 1H). 13C NMR (δ, CDCl3, 100 MHz): 70.36 (C5’), 72.01 (CH2Ph), 72.52 (CH2Ph), 73.57 (CH2Ph), 77.32 (C3’), 81.13 (C1’), 82.36 (C4’), 83.81 (C2’), 112.35 (C–CN), 118.87 (CN), 127.72, 127.82, 127.90, 127.92, 127.95, 128.13, 128.38, 128.46, 128.51, 128.93, 129.71, 130.82, 131.28, 137.31, 137.71, 137.82, 142.12.Step 2: 3-(2,3,5-Tri-O-benzyl-1-β-D-ribofuranosyl)benzamideTo a stirred solution of 3-(2,3,5-tri-O-benzyl-1-β-D-ribofuranosyl)benzonitrile (1.50 g, 3.0 mmol) in acetone (8 mL), hydrogen peroxide−urea complex CO(NH2)2•H2O2 (1.12 g, 12.0 mmol) was added, followed by addition of water (3 mL) and potassium carbonate (0.041 g, 3.0 mmol). The reaction mixture was stirred at room temperature (25°C) overnight. Next day, additional amounts of CO(NH2)2•H2O2 (1.00 g), potassium carbonate (0.04 g) and acetone (2 mL) were added, and the reaction was allowed to proceed under the same conditions for 3 days. After this time, the reaction mixture transformed into a clear colourless solution and TLC indicated completion of the reaction. The reaction mixture was diluted with water, dichloromethane was added. Organic material was extracted with EtOAc. Organic phase was separated, dried over Na2SO4, filtered, and evaporated under reduced pressure to give colourless viscous substance. Dichloromethane was added to the product and evaporated to remove traces of EtOAc. The procedure was repeated to give sticky product gradually transformed into a white crystalline substance (1.45 g, 93%). 1H NMR (δ, CDCl3, 400 MHz): 3.65 and 3.75 (AB part of ABX system, JAB = 10.5 Hz, JAX = 3.3 Hz, JBX = 3.0 Hz, 2H, H5’A and H5’B), 3.82 (t, J = 5.7 Hz, 1H, H2’), 4.06 (t, J = 4.4 Hz, 1H, H3’), 4.35 − 4.39 (m, 1H, H4’), 4.44–4.65 (m, 6H, 3×CH2 from Bn), 5.06 (d, J = 6.3 Hz, 1H, H1’), 5.66 − 5.92 (br m, 2H, NH2), 7.18–7.20 (m, 2H), 7.25–7.38 (m, 14H), 7.51 (d, J = 7.6 Hz, 1H), 7.76–7.79 (m, 2H). 13C NMR (δ, CDCl3, 100 MHz): 70.32 (C5’), 72.05 (CH2Ph), 72.36 (CH2Ph), 73.54 (CH2Ph), 77.17 (C3’), 81.85 (C1’), 82.08 (C4’), 83.68 (C2’), 124.60, 127.31, 127.61, 127.82, 127.85, 127.89, 128.05, 128.37, 128.43, 128.51, 128.73, 130.21, 132.83, 137.53, 137.79, 137.98, 141.09, 169.49 (C = O). MS, m/z: 524.14 [M + H]+ (Fig. 6).Step 3: 3-(1-β-D-Ribofuranosyl)benzamideIn a 100 mL flask, under nitrogen atmosphere, 3-(2,3,5-tri-O-benzyl-1-β-D-ribofuranosyl)benzamide (1.30 g, 2.5 mmol) was dissolved in anhydrous dichloromethane (50 mL) at stirring. The solution was cooled to −78°C and a 1 N solution of BBr3 in CH2Cl2 (10 mL; 10 mmol; 4 equiv.) was slowly added dropwise from a syringe. The reaction mixture was stirred at −78°C for 1 h 20 min. Cooling bath was removed and the reaction mixture was allowed to reach room temperature and left at stirring overnight at room temperature. Next day, methanol (20 mL) and dichloromethane (20 mL) were added, and the reaction mixture was evaporated to dryness under reduced pressure. The residue was chromatographed on a silica gel column, starting with dichloromethane as eluent and subsequently using mixtures of dichloromethane/MeOH with gradient from 100/10 to 100/30. Appropriate fractions containing 3-(1-β-D-ribofuranosyl)benzamide were collected, combined and evaporated to give almost colourless viscous substance that was dried under vacuum to give white solid foam (0.50 g). 1H NMR showed that compound contained ca. 1/2 molecule of MeOH per 1 molecule of 3-(1-β-D-ribofuranosyl)benzamide. Yield: 68%. 1H NMR (δ, DMSO-d6, 400 MHz): 3.17 (s, 1.5H, MeOH), 3.53 − 3.61 (m, AB part of ABX system, JAB = 11.8 Hz, JAX = 4.5 Hz, JBX = 4.7 Hz, 2H, H5’A and H5’B), 3.71 (t, J = 6.3 Hz, 1H, H2’), 3.81 − 3.84 (m, 1H, H3’), 3.90 (t, J = 4.3 Hz, 1H, H4’), 3.98–4.25 (br s, 3H, OH overlapped with water in DMSO-d6), 4.60 (d, J = 7.1 Hz, 1H, H1’), 7.33 (br s, 1H, 1/2 NH2), 7.40 (t, J = 7.8 Hz, 1H), 7.55 (d, J = 7.5 Hz, 1H), 7.77 (d, J = 7.5 Hz, 1H),7.86 (s, 1H), 7.33 (br s, 1H, 1/2 NH2). 13C NMR (δ, DMSO-d6, 100 MHz): 49.03 (MeOH), 62.47 (C5’), 71.83 (C3’), 77.94 (C1’), 83.26 (C4’), 85.68 (C2’), 125.89, 126.82, 128.35, 129.55, 134.55, 141.91, 168.42 (C = O). MS, m/z: 254.16 [M + H]+, 276.13 [M + Na]+, 295.16 [M + H + CH3CN]+, 317.13 [M + Na+CH3CN]+.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Data 1Reporting Summary | nature communications | [
"Article"
] | [
"Hydrolases",
"X-ray crystallography",
"Fungal biology",
"Pathogens"
] | discovery 1930s1 nicotinamide adenine dinucleotide (NAD+) fundamental metabolic life central player cellular signalling processes ADP-ribose transfer reactions2,3 NAD+ enzymatic hydrolysis dinucleotide detected enzymes break N-glycosidic bond between terminal ribose nicotinamide NAD+ glycohydrolases NADases act as ADP-ribosylcyclases cyclic ADP-ribose intracellular Ca2+ mobilizing microbial species NADases mediate infections5,6 Streptococcus pyogenes Mycobacterium tuberculosis produce toxins trigger necrotic cell death NAD+ depletion Toll/interleukin-1 receptor (TIR) domains animals plants constitute new family NADases involved cell death pathways defence against mammalian SARM1 NADase mediates axon degeneration rapid NAD+ degradation9 found bacteria plants animals no NADase identified in fungi reports NADase activity on conidia hyphae fungus Neurospora crassa12 molecular identity obscure occurrence role infection pathogenic fungi may produce similar enzymesNADases surface fungi pathogenic species A. fumigatus N. crassa enzymes cleave NAD+ NADP+ not NADH NADPH lack ADP-ribosyl cyclase base exchange activity A. fumigatus enzyme Ca2+ binding site regulates enzymatic activity each protomer N-linked glycosylated three asparagine residues stabilized two disulphide solved structure enzyme reaction products nicotinamide ADP ribose non-hydrolysable substrate analogue benzamide adenine dinucleotide structure mutagenesis critical amino acid residues reaction mechanism distinct NADases propose NADases represent unrecognized factors advantages fungi interaction host microorganisms discussionIdentification fungal surface conidia N. crassa exhibit NADase activity tested conidia A. fumigatus similar fluorescent NAD+ analogue nicotinamide 1 N6-ethenoadenine dinucleotide detected cleavage surface A. fumigatus conidia strain CEA17ΔakuB presence NADases (Fig. NADase activity different growth stages A. fumigatus developmentNADase activity on conidia strains Af293 D141 ATCC46645 Fig surface proteome A. fumigatus dependent nutritional investigated activity growth medium NAD+ cleavage activity present conidia all media higher activity Aspergillus minimal malt agar Saboraud RPMI agar NMR conidial enzyme(s cleave NAD+ ADP ribose) Nam reaction products (Fig. established presence NADases surface conidia A. fumigatus conidial protein) NAD+ cleavage visualized activity in polyacrylamide gel SDS-PAGE Conidia N. crassa A. fumigatus strong NADase activity molecular mass ~50 kDa (Fig. NADase activity detectable in A. fumigatus conidia following heat treatment 95 °C 5 min majority activity protein molecular mass ~30 kDa-MS/MS-based proteomic analyses identified 27 proteins in untreated A. fumigatus conidia eight heat-treated conidia overlapping hit gene product AFUA_6G14470 protein unknown function molecular mass 26 kDa protein detected previously on surface conidia14 Analyses conidia N.crassa yielded peptides protein A. fumigatus (Fig. Bioinformatic analyses predicted three N-linked glycosylated asparagine residues studies glycosylation N. crassa NADase discrepancy observed theoretical molecular predicted N-terminal secretory signal peptide A. fumigatus confirmed by LC-MS/MS verify NADase gene generated A. fumigatus knock-out strain lacking gene AFUA_6G14470 NADase cleavage in conidia in-locus complementation restored NADase activity (Fig. 1F results confirmed gene AFUA_6G14470 encodes conidial NADase in A. fumigatus established molecular identity N. crassa NADase encoded gene NCU07948. 1Identification fungal conidial NADases NADase activity A. fumigatus conidia fluorometric assay NADase activity A. fumigatus growth stages nicotinamide ADP ribose NAD+ cleavage products incubation A. fumigatus conidia NMR NMR signals NAD+ cleavage productsHydrolysis NAD+ Nam ADPR protons NMR signals labelled blue NAD+ red Nam ADPR Identification fungal NADases Enzyme activity gels N. crassa Aplysia californica cyclase A. fumigatus conidia ɛNAD yellow cyan asterisks bands without heat treatment Fluorescent bands excised subjected proteomics analysis A. fumigatus proteins mass spectrometry peptide spectral matches score overlapping hit uncharacterized predicted protein pairwise predicted NADase sequences A. fumigatus N. crassa genes peptides detected A. fumigatus N. crassa conidia highlighted red Bioinformatic analyses predict secretory signal peptide N-linked glycosylated asparagine residues blue green In-gel ɛNADase assay conidia A. fumigatus wild type strain gene NADase complementation strain ΔnadA Fluorometric assay NADase activity samples n = 3. Experiments (A B D F G) performed three times similar results Source data fileFungal NADases hydrolyse NAD+ NADP+ mediate synthesis calcium messengers investigated catalytic proteo-chemical properties fungal NADases Recombinant A. fumigatus NADase) expressed in human 293 cells baculovirus-infected Sf9 insect cells overexpressed enzyme secreted medium robust NADase activity (Fig. 2A purified proteins Sf9 293 cells size ~30 kDa 40 kDa glycosylation Size-exclusion chromatography enzyme homodimeric experiments with protein Sf9 cells AfNADase activity heat-resistant = 78.5 °C detected 5 10 min incubation at 95 °C sensitive to agent DTT metal chelator EGTA Inhibition EGTA reversed by Ca2+ not metal ions Mn2+ Mg2+ Zn2+ EGTA titration partially dependent on Ca2+ activity detected 5 mM EGTA N. crassa NADase (NcNADase) activity affected by EGTA Ca2+ AfNADase NcNADase cleaved NAD+ NADP+ not NADH NADPHFigs. 3 4) hydrolysis NAD(P)+ (phospho)ADPR Nam animal NADases-ribosylcyclases catalyse Ca2+ messenger cyclic ADPR exchange nicotinamide nicotinic acid. AfNADase NcNADase produce cADPR mediate base-exchange acid Aplysia californica cyclase. AfNADase hydrolyse acid adenine dinucleotide deamidated form NAD+ Fig. ADP-ribosyl cyclase NADase convert NAD+ nicotinamide hypoxanthine dinucleotide Nam N7-cyclic inosine diphosphoribose Fig. fungal NADases NAD glycohydrolases lack cyclase base-exchange activity determined kinetics AfNADase NAD+ NADP+ hydrolysis 1H NMR NAD resonance decay NAD protons increase Nam (2’-phospo- ADPR protons (Fig. KM 119.7 ± 40.8 μM 106 ± 27.1 μM NAD+ NADP+ turnover rates 1962 ± 133 s−1 418 ± 108 s−1 NAD+ NADP+pure NAD(P)+ glycohydrolase NADase activity AfNADase Sf9 insect cells measured fluorescence assay ɛNAD n = 3. NADase activity Sf9 insect cells treated DTT EGTA calcium chloride 3 NADase activity Sf9 cells EGTA CaCl2. AfNADase substrate specificity HPLC chromatograms display reactions substrates Kinetics NAD+ NADP+ hydrolysis determined 1H NMR resonances NAD+ N-2 N-6 plotted curve fitting Absence ADP-ribosyl cyclase activity AfNADase excess NA catalyse exchange Nam moiety NAD+ NA NAAD produce cyclic ADPR cleave NAD+ to ADPR Nam AfNADase base-exchange ADPR cyclase activity pure NADase HPLC chromatograms Aplysia cyclase base exchange reactions Experiments B C F performed three times similar results Source data file crystal structure AfNADase TNT domain regulatory calcium-binding structural assembly catalytic mechanism solved crystal structure resolution 1.6 Å3A Supplementary Table 1) (PDB protein crystallizes space group P3221 two molecules asymmetric unit homodimeric assembly chromatographic analysis Fig. 2C dimeric assembly interface 2260 Å2 solvent-accessible area 11030 Å2 PDBePISA 66 residues C-terminus intertwined protomer (Fig. protomers two domains N-terminal ‘thumb’ C ‘palm’ domain thumb domain residues 20-117 folded five α-helices loop regions stabilized disulfide bridges C33-80 C38-50 palm domain seven-stranded central β-sheet two α-helices 310 helices palm domain TNT domain C-terminus metal binding site dimerization interface C-terminus protomer acetate ion crystallization solution trapped cavity domain boarder central β-sheet β2 3 5). 3Crystal structure AfNADase protomers dimer coloured yellow blue secondary structure elements labelled α β helical beta strands N-linked glycosylated asparagine residues stick representationgreen sphere calcium ion residues stick cartoon domain structure 70° thumb domain grey palm domain blue dimerization interface 2-fold symmetry axis dotted line disulfide bridges stick N-linked glycosylation asparagine 45, 95 118 Cartoon stick calcium binding site residues water molecules pentagonal bipyramidal coordination NADase activity A. fumigatus calcium binding site mutant (D119A/E220A) measured fluorometric assay n = 3. experiment performed three times similar results Source data file pentagonal bipyramidal coordination crystal structure refined density B-factors bound metal ion Ca2+ activation EGTA treatment (Fig. 2B). Ca2+ ion binding site side chains D219 E220 E223 main chain S216 two water molecules (Fig. 3A protomer glycosylated at asparagine residues N45 N95 N118 (Fig. 3B sites coincide predicted (Fig. 1E). N-linked glycosylation NcNADase 293 cells observed Fig. 6)observations provide basis for biochemical findings DTT EGTA sensitivity migration in SDS-PAGE molecular mass higher polypeptide (Fig. 1D). role Ca2+ binding site AfNADase catalysis created Ca2+-free mutant substituting D219 E220 with alanine residues Disruption Ca2+ binding site led sevenfold reduction in ɛNAD+ hydrolysis importance of Ca2+ coordination for efficient hydrolysis (Fig. 3C, D). attempted crystallize Ca2+ mutant AfNADase protein yield crystals crystallized native protein treated with EGTA Ca2+ present in crystal structure lower occupancy crystals soaked with tenfold excess NAD+ Ca2+ ions present reaction products Nam ADPR bound to active site (Fig. 4A, B) Nam in deep cavity acetate hydrogen-bonded to enzyme with side chains R129 R148 main chain F130 water molecule (Fig. 4C). residues conserved role in hydrolysis of NAD(P)+ Fig. 7) ADPR to Nam in crevice on surface enzymeribose moiety bound to Nam flipped contact with aromatic ring Y100 conserved two phosphate moieties enzyme hydrogen bonds R129 S132 L138 N154 water adenosine moiety bound to enzyme π-stacking side chain F158 residues with Nam ADPR active site AfNADase distance between Ca2+ ion active site more than 40 Å role Ca2+ catalysis activity lower in Ca2+ binding site mutant Ca2+ enzymatic activity NAD(P)+ depletion intracellular concentration Ca2+ nanomolar range.Fig. 4Crystal structure AfNADase reaction products non-hydrolysable NAD+ analogue BAD Complex AfNADase reaction products Nam ADPR protein cartoon transparent surface map palm domain coloured blue reaction products Nam ADPR stick representation Calculated mFo-DFc POLDER electron density map 3.0 σ binding of ADPR nicotinamide Residues AfNADase Nam ADPR stick water molecule labelled WI dotted lines represent hydrogen bonds between protein nitrogen in pyridine rings substituted with carbon atom non-hydrolysable analogue BAD Complex AfNADase substrate analogue BAD protein cartoon transparent surface mappalm domain blue BAD stick Calculated mFo-DFc POLDER electron density map 3.0 σ binding BAD Residues AfNADase BAD stick water molecules labelled WI WII dotted lines hydrogen bonds protein water molecules substrate analogue mechanism AfNADase NAD+ hydrolysis water molecule I hydrogen-bonded in-ring oxygen proximal ribose nicotinamide moiety backbone R129 bridge positive change oxygen nucleophilic attack prepared Q194 interacts 2” OH proximal ribose induces δ-negative charge attack C1’ position water nicotinamide site formation oxonium intermediate activity recombinant AfNADase WT active site mutants R129A F130A F137A Q194A Q194K measured fluorometric assay n = 3. experiment performed three times similar results Source data file catalytic mechanism fungal TNT-containing NADases crystal structure complex non-cleavable substrate catalytic mechanism AfNADase attempted solve structure enzyme NAD+ crystals dinucleotide fast conversion rate NAD+ Nam ADPR trap substrate crystalsused non-cleavable NAD+ analogue BAD differing nitrogen atom pyridine with carbon atom (Fig. 4D). co-crystallize AfNADase with BAD bound active site molecule B unit (Fig. 4E F Residual electron density in molecule A not enough model BAD active site structure complex illustrates mode substrate active site residues same as product binding (Fig. 4G). interaction BAD includes two water molecules one same water I other water II proximal ribose (Fig. F137 proximal ribose BAD bond active site benzamide (nicotinamide) held hydrogen bonding to R129 F130 R148 sidechain Q194 with 2” OH group proximal ribose other NADases ADPR transferases catalyse transfer from NAD+ acceptor ADP-ribosyl cyclases catalytic acidic residue glutamate aspartate corresponding position NAD(P)+ cleaved oxocarbenium intermediate unlikely in AfNADase pull from Q194 bond strong enough oxocarbenium intermediate required for cyclase base exchange activity AfNADase catalyse reactionscleavage NAD(P)+ nucleophilic attack on anomeric carbon (C1’) needed AfNADase attack from β-face nicotinamide moiety shielded by hydrophobic interaction F137 propose mechanism water molecule I hydrogen-bonded to in-ring oxygen proximal ribose nicotinamide moiety phosphodiester backbone R129 positive charge oxygen nucleophilic attack prepared by Q129 interacts with 2” OH proximal ribose induce δ-negative charge SN2 reaction transition state attack on α-face C1’ coincides with nicotinamide leaving formation partial oxocarbenium ion) (Fig. 4H). mechanism precludes NAAD cleavage interaction between carboxylate nicotinic acid moiety watermolcule I side chain R129 less likely catalysis AfNADase cleave NAAD mutagenesis of R129 F137 Q194 inactivated enzyme Mutagenesis F130 inactivate enzyme interacts with carbonyl amide backbone atoms Substitution of R148 interacts with nicotinamide attempted Western blotting activity measurements recombinant protein in 293 cells protein unstable rapidly degradedR148 forms salt bridge with D128 vital for protein stability mutation R780 in TNT unstable protein residues conserved in TNT domain proteins fungi bacteria 7) NADases present in pathogenic investigated relationship between AfNADase NADases BLAST searches return NADases Pfam search revealed palm domain of AfNADase residues 117-234 harbours tuberculosis necrotizing toxin (TNT) domain (Fig. TNT domain identified in M. tuberculosis) protein CpnT toxin with NADase activity mediates macrophages NAD(P)+ structure Mtb CpnT TNT solved with immunity factor prevents suicidal NADase activity No homologue gene M. tuberculosis immunity factor in genomes N. crassa A. fumigatus immunity factor needed in fungi enzyme enters secretory pathway precludes cellular NAD+ pools AfNADase activity regulated by Ca2+ immunity factor not required structural similarities between AfNADase TNT palm domain similar with RMSD of 0.83 Å thumb domain AfNADase differs from TNTAfNADase stabilized by two disulfide bridges two N-linked glycosylation sites N45 N95 not present in TNT distinct C-terminal extension Ca2+ binding site C-terminus intertwined with protomer TNT after last helix core fold active site TNT coincides with AfNADase critical residues sequence motifs conserved 7 AfNADase TNT same substrate specificity lack cyclase base exchange activity 2 fungal NADases TNT domain-containing proteins in bacteria fungi reaction mechanism AfNADase valid for all TNT domains 5AfNADase founding member family fungal NADases Domain architecture of AfNADase NcNADase Mtb CpnT TNT alignment active site motifs NADases ADP-ribosyltransferases residues in three regions coloured location sequence Comparison active sites AfNADase with BAD Mtb CpnT TNT Tse6 cholera toxin Amino acids coloured based on position active site motif Phylogenetic distribution of NADase fungi Classes with coloured red or green number represents speciescalcium binding site emerged Eurotiomycetes present order Eurotiales fungal NADases share related Tse6 cholera toxin diphtheria toxin active sites (Fig. 5B C Table 2) ARTs divided groups active site motifs three regions cholera toxin R-S-E diphtheria toxin H-Y/Y-E motif present-ribose polymerases NADases similar sequence motifs AfNADase motif R129 F137 Q194 residues vital for NADase activity fungi TNT domain protein belong Firmicutes Actinobacteria Ascomycota fungal NADases exhibit specific structural functional properties distinct bacterial expression conidia homodimeric nature unique thumb domain posttranslational glycosylation critical disulfide bonds A. fumigatus catalytic activity regulated by Ca2+ ions Ca2+ site AfNADase emerged Eurotiales Aspergillus spp (Fig. 5D). 3D structure NcNADase similar AfNADase lacks residues Ca2+ bindingphylogenetic analysis fungal NADases present in families species pathogenic potential role protein fungal virulence (Fig. 5D). conidia A. fumigatus present cleared lungs by immune aspergilloses problem animal husbandry threat for immune-compromised A. fumigatus spores harbour virulence factors spectrum antifungal drugs limited resistance gene AfNADase upregulated during conidiation A. fumigatus strains high adherence to pulmonary epithelial cells27 discovered fungal surface NADases distinct structural functional features characterised A. fumigatus no homologues in mammals functional similarity to toxins NADase activity presence in pathogenic fungi fungal NADases may convey advantages interaction host competing microorganisms strains media growth conditionsAspergillus fumigatus strain CEA17ΔakuB28 parental strain generation ΔnadA knockout complementation strainsstrains cultivated Aspergillus 1.5% agar plates 37 °C 70 mM NaNO3 11.2 mM KH2PO4 7 mM KCl 2 mM MgSO4 6.5 1% glucose 1 μl trace element solution 18 mM FeSO4 171 mM EDTA 77 mM ZnSO4 180 mM H3BO3 25 MnCl2 6.7 CoCl2 6.4 mM CuSO4 0.9 mM (NH4)6Mo7O2429 3–7 days conidia harvested ultra-filtrated water counted Neubauer chamber DNA A. fumigatus grown liquid AMM 24 h Chromosomal DNA Yeast DNA Purification Kit DNA fragments separated agarose gel blotted Hybond N+ nylon membranes probes labelled DIG labelling mix Hybridization detection DIG Easy Hyb CDP-star kit A. fumigatus mutant strainsDeletion PCR strategyUpstream downstream regions gene nadA (AFUA_6G14470) amplified by PCR primer pairs 6G14470_5for_ptrArev (Supplementary Table 3) overlapping ends pyrithiamine resistance cassette introduced at 3’-end upstream 5’-end downstream nadA gene ptrA resistance cassette amplified from plasmid pSK27532 primers ptrA-for-rev PCR reactions with Phusion Flash Polymerase Master Mix PCR fragments purified by gel extraction final deletion construct generated three fragment PCR primers 6G14470_5for 6G14470_3rev resulting PCR product purified used for transformation A. fumigatus protoplasts Pyrithiamine (1 mg/mL Merck Germany) resistant transformants analyzed for deletion nadA Southern blot analysis deletion strain ΔnadA complemented in-locus nadA gene amplified from DNA A.fumigatus strain CEA17ΔakuB primers 14470_fwd_rev hph resistance cassette amplified plasmid pUChph34 primers_fwd_rev ptrA resistance cassette amplified primers_ptrA_fwd_rev complementation construct amplified nadA gene promoter hph ptrA resistance cassettes generated multifragment PCR purified transformed.Transformants resistant hygromycin (150 μg/ml Roche Applied Science Germany analyzed complementation nadA Southern blot analysis.Fluorometric determination NADase activity conidia purified proteins medium cells overexpressing AfNADase mutants increase fluorescence cleavage NAD Reactions 200 μl reaction buffer (50 mM Na-Acetate pH 5.5 150 mM NaCl 0.5 mM CaCl2) 80 μM εNAD+ reaction adding conidia medium 293 cells purified AfNADase initial reaction rate change fluorescence 410 nm excitation 300 nmexperiments Cary Eclipse fluorescence spectrophotometer determination NADase activity A. fumigatus growth stagesA fumigatus Aspergillus medium incubated 96 well plate 37 °C reaction εNAD+ 80 μM monitored fluorescence 410 nm excitation 310 nm Infinite M200Pro plate reader NADase activity assaySamples N. crassa conidia A. fumigatus conidia 1X non-reducing SDS-PAGE sample buffer (50 Tris-HCl pH 6.8 2 % SDS 10 % glycerol 0,005 % bromophenol samples heat-treated A. fumigatus conidia spores 95 °C 5 min samples run 12 % non-reducing SDS polyacrylamide gel 10 mA washed buffer-HCl pH 7 CaCl2 gel developed incubation 5 min buffer NP-40 80 μM εNAD+) fluorescent bands visualized UV-transilluminator excised for mass spectrometry analysis preparation LC-MS bands excised from gel washed destained incubation acetonitrile 50 mM NH4HCO3Disulfide bonds reduced 20 mM Tris(2-carboxyethyl)phosphine 30 min 55 °C Reduced cysteine residues carbamidomethylated 25 mM iodoacetamide 30 min gel washed dried gel reconstituted Trypsin/LysC (50 ng/μl incubated 37 °C 18 h Peptides extracted three steps sonication acid acetonitrile second third 1/9/90) Extracts pooled dried dissolved 0.05% TFA 2/98 acetonitrile/water-MS/MS analysis Ultimate 3000 RSLC nano instrument QExactive Plus mass spectrometer Tryptic peptides trapped 4 min Acclaim Pep Map 100 column flow-rate 5 μL/min separated Acclaim Pep Map column binary gradient 0–4 min 4% B 10 7% B 40 10% 60 15% B 80 25% B 90 30% B 110 50% B 115 60% B 120–125 96% B 4% B Positively charged ions generated Nanospray Flex Ion Source stainless steel emitter 2.2 kV spray voltageIons measured MS2 Precursor ions scanned/z 300-1500 70,000 FWHM AGC 106 120 Fragment ions HCD cell 30% collision energy scanned 17,500 FWHM AGC 120 ms exclusion 30 s database Uniprot database Aspergillus fumigatus Af293 Neosartorya fumigata Af293 Proteome Discoverer 2.2 Mascot 2.4.1 Sequest HT MS Amanda 2.0 missed cleavages tryptic peptides precursor mass tolerance 10 ppm fragment mass tolerance 0.02 Da modifications oxidation Met acetylation N-terminus static modification carbamidomethylation Cys rank 1 peptide rate <1% positive protein hits substrate specificity HPLCReactions 1 ml buffer (50 Tris-HCl 0.5 CaCl2 150 NaCl 200 μM substrate NADH 10 ng purified AfNADase incubation 20 °C 1 h 50 μl analysed HPLC 250/3 nucleosil 100-3 C18 columnSamples run buffer A (10 ammonium acetate 2 TBA-bromide 10% acetonitrile B 90% acetonitrile). gradient 0 0.1% B 1 0.1% B 7 4% B 19 22% B 21 90% B 23.5 90% B 26 0.1% 35 0.1 % B flow rate 0.6 ml/min column compartment 30 °C progress visualised UV-VIS detector 259 340 nm.ADPR cyclization base exchange assayed HPLCReactions 1 ml buffer (Na-PO4 pH 6.8 Na-Acetate pH 4 0.5 mM CaCl2) 1 mM NA 100 μM NAD+ 10 ng AfNADase incubation 20 °C 1 h 50 μl reaction analysed HPLC Aplysia californica ADP-ribosyl cyclase positive control.Fluorometric cyclization AfNADase NAD+ N7-cIDPR Reactions 200 μl buffer (50 Na-Acetate pH 5.5 150 mM NaCl 0.5 mM CaCl2) 40 μM NHD+ 10 ng AfNADase change fluorescence 410 nm 300 nmexperiments Cary Eclipse fluorescence spectrophotometer purification AfNADase Sf9 insect A. fumigatus NADase_6G14470) amplified cDNA inserted in-frame C-terminal C3 cleavage site 6xHis-Tag-ORF-3C vector Protein Facility empty vector linearized KpnI amplified insert treated T4 polymerase single stranded overhangs 25 mM ethylenediaminetetraacetic acid heat inactivation 75 °C 20 min vector insert gel purified NucleoSpin® Gel PCR Clean-up kit annealed transformation One ShotTM TOP10 E. coli cells Positive clones identified PCR plasmids isolated NucleoSpin® Plasmid sequence controlled Sanger sequencing DH10EMBacY E. coli cells bacmid generation Sf9 cells 0.5 million cells/ml transfected AfNADase bacmid virus particles harvested after seven days recombinant virus amplified high-titre viral stock Sf9 cells cultivated Sf-900TM II SFM medium infected high titre viral stocks 1.5-2 million cells/mldays growth arrest cells pelleted medium filtered 0.22 μm filter secreted enzyme purified immobilized metal affinity chromatography HisTrapTM excel column ÄKTA pure chromatography system column washed buffer (50 Tris-HCl pH 8 300 NaCl protein eluted buffer 500 mM purified protein concentrated 10 kDa MWCO Amicon® Ultra Centrifugal filters purified size exclusion chromatography Superdex 200 16/60 HiLoad column size exclusion purification SEC buffer (50 mM Tris-HCl 8 300 mM NaCl 2 mM TECEP yield 20-25 mg per litre purification NcNADase AfNADase 293 sequence NcNADase_026504.1 N. crassa codon-optimized mammalian expression synthesised inserted vector pUC57 GenScript 5’-end HindIII site Kozak sequence 3’ KpnI restriction site pUC57 vector digested HindIII KpnI ligation pCMV-Flag5a plasmid pCMV-NcNADase-Flag isolated sequenced293 cells transfected Effectene transfection reagent cell medium harvested 96 h post transfection medium overexpressed NcNADase concentrated 10 kDa MWCO Amicon® Ultra Centrifugal filters purified Anti-FLAG M2 affinity gel-Aldrich Triton X-100 NaCl added 0,5 % 300 mM 40 μL affinity gel incubated 2 h 4 °C washed twice TBS-HCl pH 7.4 150 mM NaCl four times TBS 1 M NaCl protein eluted incubation 3X FLAG-peptides (5 μg/ul) TBS 30 min 4 sequence encoding A. fumigatus (gene AFUA_6G14470) amplified from cDNA primer PCR amplicon 3’-end BamHI restriction site Kozak sequence 5’-end EcoRI restriction site C-terminal hexahistidine tag PCR product digested restriction enzymes inserted into vector pcDNA3.1(+) plasmid pcDNA3.1(+)-AfNADase-6XHis C-terminal AfNADase isolated sequencedtransfected 293 cells generated transfection calcium phosphate precipitation method two rounds selection 550 μg/ml G418 Monoclonal cell lines adapted grow GibcoTM FreeStyleTM 293 Expression Medium (Thermo Fisher). overexpressed protein purified Sf9 insect cells.Generation AfNADase mutantsMutants expression Sf9 insect cells 293 cells generated mutating parental plasmids Q5® Site-Directed Mutagenesis Kit (New England Biolabs Inc recommendations primers designed NEB base exchanger web tool (New England Biolabs F137A mutant generated primes 5’-GTATGGCACCGCTCTGGCGCCGC-3’ Q194A Q194K produced primers 5’ R129A mutant generated primers 5’ F130A mutant produced primers 5’-GCTTGACCGTGCGGGATCGGAGTATG-3’ calcium site binding mutant (D219A/E220A) generated primers 5’Crystallization structure screening native AfNADase high throughput sparse matrix approach JCSG + PACT HT screens drop vapour diffusion trials Swiss CI SD3 plates 20 °C 8 °C 40 μl reservoir ratios 2:1 1:1 1:2 protein precipitant drop size 600 nl Mosquito LCP crystallization robot followed ROCK IMAGER instrument 20 °C manually 8 °C first crystals days growth one-week Initial crystallization conditions 0.1 M sodium acetate pH 5 0.2 M CaCl2 25 % polyethylene glycol (PEG) 6000 Optimized crystals grown 20 °C hanging-drop vapour diffusion 0.1 M sodium acetate pH 5 0.3 M CaCl2 20-25 % PEG 8000 crystals cryoprotected-protectant 20 % glycerol-crystallization AfNADase BAD BAD 5 mM protein solution benzamide riboside crystallization condition same native protein cryo-protectant 10 mM BAD-treated AfNADase crystallized 0.04 M potassium monobasic 16% PEG 8000 20% glycerolglycerol crystallization cryoprotection needed crystals soaked 1 min supplemented 10 mM NAD mounting Litho loops flash freezing liquid N2.Crystals single-wavelength dispersion phasing co-crystallized with 100 mM NaI optimized AfNADase crystals soaked cryoprotectant 20% glycerol 500 mM NaI before Litho loops flash frozen liquid N2.Phasing at P13 EMBL Hamburg PETRA III storage ring iodide derivatives AfNADase SAD method data collected 6 kEV energy anomalous scattering coefficient 11 e− 0.2° oscillation multiplicity autoprocessed data anomalous signal determining phases AutoSol36 initial model AutoBuild37 35 heavy atom sites partial occupancy found space group P3221 initial model model for molecular replacement Phaser39 high-resolution native data P13 12.7 keV energy.Data EGTA NAD treated crystal collected P11 end station 12 keV BAD co-crystallization data I04-1 Diamond Light Source 13.5 keV processed XDS40 scaled Aimless41 Crystals structures refined Phenix.refine38 manual inspection model building Coot42.Structures validated with MolProbity38 Ligand restraints created eLBOW43 POLDER maps calculated Phenix package44 data collection refinement in Supplementary Table 1. Structures illustrated using PyMol UCSF Chimera deposited in Protein Data Bank under ID: 6YGE 6YGF 6YGG A. fumigatus conidia NAD+ cleavage products by 1H NMRA incubated room temperature 1 h with 500 uM NAD+ in NMR reaction buffer (25 mM sodium phosphate pH 5.8 50 mM NaCl 5 % D2O) volume 500 μl conidia removed by centrifugation filtration 10 kDa MWCO Amicon® Ultra Centrifugal filter reaction products identified by NMR collected Bruker Ascend 850 MHz instrument triple resonance 23 °C 1H NMR spectra acquired with 16 scans recovery delay 3.9 AfNADase kinetics by 1H NMRNMR data Bruker Ascend 850 MHz hydrolysis of NAD+ NADP+ measured by 1H NMR acquired eight scans recovery delay 3.9 secondsreactions 500 μM NAD+ or NADP+ in NMR buffer (25 mM sodium phosphate pH 5.8 50 mM NaCl 5 % D2O volume 500 μl spectrums NAD+ NADP+ AfNADase acquired hydrolysis NAD measured adding AfNADase 0.2 nM time enzyme addition recorded spectra acquired until substrate resonances declined baseline level spectra phased corrected resonances assigned standard correlation methods resonances integrated Dynamics centre 2.5 compared to spectra NAD+ NADP+ kinetic parameters AfNADase determined by progress curve analysis MatLab.Phylogenetic scanned NCBI protein database for fungal homologues fungal phylogeny 15 fungal performed BLAST database contained NCBI non-redundant protein sequences restricted taxonomically to 15 fungal classes used default parameter settings filtered low complexity regions target sequences to 500 expected value 1. Homology to calcium binding motif DE[KV]E from A. fumigatus sequence homologous sequence non motif specific clade.Synthesis of benzamide ribosideFigSynthesis 3-(1’-β-D-Ribofuranosyl)benzamide 3-(2,3,5-Tri-O-benzyl-1-β-D-ribofuranosyl 3-iodobenzonitrile (1.50 g 6.5 mmol anhydrous THF (97 mL cooled −78°C isopropylmagnesium chloride (3.70 mL 2 added stirred −78 °C 2 h transferred 2,3,5-tri-O-benzyl-D-ribono-1,4-lactone (2.73 g, 6.5 mmol) anhydrous THF (18 mL) −78 °C −78 °C 2 h ambient temperature (25 °C left overnight pink solution Next day sodium bicarbonate (60 mL) added extracted ether 300 organic layer separated washed brine dried Na2SO4 filtered concentrated vacuum yellow oil (3.41 g). crude product dissolved toluene volatiles evaporated residue diluted anhydrous CH2Cl2 (16 cooled −78 °C Boron trifluoride etherate (1.65 mL, 13.1 mmol) added triethylsilane (2.1 mL 13.1 mmol). stirred −78 °C 1 hsolution temperature (25°C stirred overnight reaction quenched NaHCO3 extracted CH2Cl2. Organic layer separated dried Na2SO4 filtered evaporated pressure pink-yellow oil (2.98 crude product purified MPLC column chromatography (100 g SNAP Ultra SiO2 Biotage column l 15.5 cm d cm UV detection (254 280 nm elution gradient/EtOAc mixtures 15% to 25% EtOAc 3-(2,3,5-tri-O-benzyl-1-β-D-ribofuranosyl)benzonitrile (1.708 g 51%) colourless oil 1H NMR CDCl3 400 3.58 3.65 10.3 Hz 4.0 Hz JBX Hz 2H 3.73 Hz 5.1 Hz 1H 4.01 5.1 3.2 Hz 4.34 4.37 2H 1/2 CH2 4.51–4.61 5H 1/2 4.97 7.4 Hz 1H 7.12–7.15 7.23–7.38 7.53 7.59 7.66 13C NMR CDCl3 100 70.36 72.01 72.52 73.57 77.32 81.13 82.36 83.81 112.35 118.87 127.72 127.82.90.92.95 128.13.38.46.51.93 129.71 130.82 131.28 137.31.71.82 142.12.Step 3-(2-Tri-O-benzyl-1-β-D-ribofuranosyl (1.50 g 3.0 mmol acetone (8 hydrogen peroxide−urea CO(NH2)2•H2O2 (1.12 g 12.0 mmol water (3 mL potassium carbonate (0.041 g 3.0 stirred room temperature (25°C overnight Next day CO(NH2)2•H2O2 (1.00 potassium carbonate (0.04 g acetone (2 mL 3 days colourless solution completion diluted water dichloromethane added Organic material extracted EtOAc separated dried Na2SO4 filtered evaporated colourless viscous substance Dichloromethane added evaporated remove EtOAc repeated product white crystalline substance (1.45 g 93%)1H NMR CDCl3 400 3.65 3.75 ABX JAB 10.5 Hz JAX 3.3 Hz JBX 3.0 Hz 2H H5’A 3.82 Hz 1H 4.06 4.4 Hz 4.35 4.39 1H 4.44–4.65 6H 3×CH2 5.06 J 6.3 Hz 1H 5.66 5.92 2H NH2) 7.18–7.20 7.25–7.38 7.51 J Hz 7.76–7.79 13C NMR CDCl3 100 70.32 72.05.36 73.54 77.17 81.85 82.08 83.68 124.60 127.31 127.61.82.85.89 128.05.37.43.51.73 130.21 132.83 137.53 137.79 137.98 141.09 169.49 524.14 H (Fig. 6) 3-(1-β-D-Ribofuranosyl 100 mL flask.30 g 2.5 mmol dissolved anhydrous dichloromethane (50 mL cooled −78°C 1 N BBr3 CH2Cl2 mLadded dropwise syringe mixture stirred −78°C 1 h 20 min Cooling bath removed room temperature left overnight Next day methanol (20 mL dichloromethane (20 mL added evaporated dryness reduced pressure residue chromatographed silica gel column dichloromethane eluent mixtures dichloromethane/MeOH 100/10 to 100/30 3-(1-β-D-ribofuranosyl)benzamide collected combined evaporated colourless viscous substance dried vacuum white solid foam (0.50 1H NMR 1/2 molecule MeOH per 1 molecule 3-(1-β-D-ribofuranosyl)benzamide Yield 68% 1H NMR DMSO-d6 400 3.17 1.5H 3.53 − 3.61 11.8 Hz JAX 4.5 Hz JBX 4.7 Hz 3.71 6.3 Hz 1H 3.81 − 3.84 3.90 4.3 Hz 3.98–4.25 water DMSO-d6) 4.60 7.1 Hz 1H 7.33 1/2 NH2) 7.40 Hz 7.55 7.5 7.77 7.5 7.3313C NMR DMSO-d6 100 49.03 62.47 71.83 77.94 83.26 85.68 125.89 126.82 128.35 129.55 134.55 141.91 168.42 MS m 254.16 + H 276.13 + Na 295.16 + H CH3CN 317.13 + Na+CH3CN Nature Research Reporting Summary Review Data Summary | 49.8 | 1.138842 |
10.1038/s41467-020-17585-8 | PMC7414839 | Triacylglycerols (TG) are synthesized at the endoplasmic reticulum (ER) bilayer and packaged into monolayer lipid droplets (LDs), but how proteins partition between ER and LDs is poorly understood. Here authors use synthetic model systems and find that proteins containing hydrophobic membrane association domains strongly prefer monolayers and that returning to the bilayer is unfavorable. | Triacylglycerols (TG) are synthesized at the endoplasmic reticulum (ER) bilayer and packaged into organelles called lipid droplets (LDs). LDs are covered by a single phospholipid monolayer contiguous with the ER bilayer. This connection is used by several monotopic integral membrane proteins, with hydrophobic membrane association domains (HDs), to diffuse between the organelles. However, how proteins partition between ER and LDs is not understood. Here, we employed synthetic model systems and found that HD-containing proteins strongly prefer monolayers and returning to the bilayer is unfavorable. This preference for monolayers is due to a higher affinity of HDs for TG over membrane phospholipids. Protein distribution is regulated by PC/PE ratio via alterations in monolayer packing and HD-TG interaction. Thus, HD-containing proteins appear to non-specifically accumulate to the LD surface. In cells, protein editing mechanisms at the ER membrane would be necessary to prevent unspecific relocation of HD-containing proteins to LDs. | IntroductionLipid droplets (LDs) are lipid storage organelles primarily functioning in cellular energy metabolism1. LD biogenesis occurs at the endoplasmic reticulum (ER) membrane during energy rich or stress conditions. LD biogenesis starts with the synthesis of neutral lipids, such as triacylglycerols (TG) or sterol esters, which, at low concentration, are dissolved in the ER bilayer2. Upon increase in concentration, neutral lipids demix from membrane phospholipids to form an oil lens or a nascent droplet within the bilayer3 (Fig. 1a). The lens grows and emerges in the cytosol as a mature LD: an oil-in-water droplet covered by a phospholipid monolayer with proteins embedded. Indeed, throughout the steps of LD emergence, many proteins target to the surface and around the LD4–7. Proteins targeting the LD surface essentially come from the ER membrane or from the cytosol7, and ensure proper LD budding8. How proteins bind and accumulate to LDs is not well understood but the neutral lipid chemistry is determinant to these processes9. Specificity of protein targeting to LDs is at the heart of LD biology, and understanding its principles will provide fundamental knowledge on lipid metabolism and cellular proteostasis5–7,10,11.Fig. 1Characterization of droplet interface bilayers.a Schematic representation of the ER phospholipid bilayer contiguous with the monolayer of a nascent LD (left side); the corresponding DIB system reproducing contiguous bilayer and monolayers is shown on the right side. The water phase is represented in light blue and the oil phase in yellow (neutral lipid, e.g. triglycerides (TG)). b Drawings of a DIB bilayer of DOPE (top) and DOPC/DOPE (1:1) (bottom). c The thickness of the hydrophobic region of the DIB bilayer in DOPE (white) and DOPC/DOPE (1:1) (gray) is determined by capacitance measurement. Results are shown as box-plots (box limits, upper and lower quartiles; middle line, median; whiskers, minimum and maximum value; the mean is indicated) from n = 5 independent experiments. Each point is represented as a black dot. d Distribution of Rh-PE between the bilayer and the monolayers in DOPE (white) and DOPC/DOPE (1:1) (gray) DIBs. The results are the mean ± SD of respectively n = 10 and n = 5 independent measurements. Each point is represented as a black dot. Significance was determined by Welch’s t-test (unpaired parametric test, two-tailed p-value) and is indicated by ns (not significant): p > 0.05. Source data are provided as a Source Data file.Membrane physicochemical properties regulate the protein distribution at bilayer-encircled organelles12–14. The LD-water interface is distinguishable from a bilayer-water interface by several features: it can sustain a loose lipid packing9,15,16; the thickness of the underlying hydrophobic region, up to hundreds nm, is much larger than the hydrophobic thickness of a bilayer (~3 nm)17,18; the hydrophobic core consists of neutral lipids, instead of phospholipid acyl chains. Considering these discrepancies in physical chemistry, it may not be surprising that proteins show preference for one interface over the other.Most proteins physically associating with LD surfaces are either peripheral or monotopic6,7,11 and do not fully cross bilayer membranes. Proteins moving from the ER to LD surface, contain helical hydrophobic domains (HDs), which are monotopic integral membrane domains embedded only in one face of the membrane. These HDs include helical hairpins, hydrophobic helices, and possibly transmembrane domains not fully crossing a bilayer11,19–21. In contrast, soluble proteins often use amphipathic helices (AHs) for binding to LDs.The binding of AHs to LDs is more documented both in vitro and in vivo9,16,19,22–24: AHs act as surfactants, favorably adsorbed to the oil/water interface of LDs to decrease the interfacial energy. AHs recognizes a variety of membrane features, such as surface charges, curvature, phospholipid packing defects, and neutral lipids9,16,19,22,23. In contrast, much less is known about HDs which target to LDs mostly from the ER membrane through ER-LD connecting bridges25–27. Neither the energetics involved in their binding to LDs nor the parameters controlling their ER-to-LD partitioning are known.The inclusion of HD-containing proteins into lipid bilayers can cause local perturbation to the bilayer properties, which translates into an energy penalty28–31. For instance, proteins can locally perturb the organization of the phospholipids and enhance exposure of the hydrophobic core of the bilayer to water28–32. The extent of membrane perturbation depends on the amino acid sequence, and is for instance important when the mismatch between the bilayer thickness and the HD length is significant31,32. As for protein insertion into LD surfaces, no information is available regarding the energy cost of the process, nor the type and the extent of the perturbation generated in the surrounding lipids.Here, we study how LD proteins, and particularly monotopic HD-containing proteins, partition between a bilayer and an LD in contiguity. We employ the droplet interface bilayer (DIB) system33 (Fig. 1a) to study the partitioning of proteins and peptides bearing HDs, as compared with AH-containing proteins. We find that all proteins investigated partition preferentially to the LD monolayer surface, but HD-containing proteins display a higher enrichment in the monolayer than AH-containing ones. Relocation of HD proteins to the bilayer is unfavorable, while moving from the bilayer to the monolayer is spontaneous. We also found that protein distribution is altered by the ratio between PC and PE phospholipids by regulating the extent of HD-TG contact at the LD surface.ResultsCharacterization of the droplet interface bilayer systemTo determine the partition coefficient of proteins capable of binding a monolayer and a bilayer in contiguity, we decided to employ the droplet interface bilayer (DIB) system33,34. DIBs consist of two micrometric buffer-in-oil droplets covered by a phospholipid monolayer (Fig. 1a). The oil phase used here was trioctanoate, a triglyceride with similar interfacial energy as triolein8, the major cellular neutral lipid. Contact of the droplets induces the formation of a bilayer in contiguity with the two monolayers (Fig. 1a). Thus DIBs mimic ER-LD contiguity (Fig. 1a) without curvature considerations; the different interfaces are flat at the protein scale and the concavity of the monolayer surfaces is irrelevant with respect to curvature. For phospholipids, we used dioleoyl phosphatidylethanolamine (termed PE) and dioleoyl phosphatidylcholine (termed PC) (Fig. 1b). Phospholipids were added to the oil phase and were recruited to the surface of the aqueous droplets whose contact generates within 5 min an equilibrated DIB34,35.DIBs can be generated with almost any phospholipids35. In the case of non-bilayer phospholipids, such as DOPE, a PE-DIB bilayer is made thanks to the presence in the bilayer of TG molecules whose level is decreased by the addition of PC35. To get insight into the amount of TG present in a PE-DIB bilayer, we measured the thickness of the hydrophobic region of the bilayer by capacitance measurements36 (Fig. 1c, Supplementary Fig. S1a). The thickness measured in PE-DIBs was 2.68 nm, only ~7% above the thickness of a PC/PE (1:1) DIB, 2.52 nm (Fig. 1c). Importantly, these values are comparable to the thickness of the hydrophobic region in phospholipid vesicles devoid of oil, between 2.3–2.7 nm18. Additionally, all-atom molecular dynamics simulations indicate that adding PE to a PC bilayer devoid of oil is sufficient to increase bilayer thickness up to 10% (Supplementary Fig. S1b). Altogether, these data indicate that the thickness of the DIBs made here is similar to that of phospholipid bilayer vesicles and is not significantly affected by the presence of oil.Since the PC/PE mixture was added to the oil phase, we wanted to know whether this bulk ratio reflects the monolayer composition. We had previously measured the surface tension of monolayers made of PC/PE and found a linear decrease as this ratio increased in bulk oil35 (from ~2 mN m−1 for PE at 100% to ~0.6 mN m−1 for 100% PC). This supports that the bulk PC/PE composition reflects the one at the monolayer, as otherwise a plateau of surface tension against PC/PE should be observed. We next asked whether the PC/PE ratio in the monolayer and in the DIB bilayer are identical. To address this, we measured the partitioning of Rhodamine-PE (Rh-PE) between the DIB monolayers and bilayer, in the case of PE and PC/PE (1:1) DIBs. In pure PE-DIB, Rh-PE was uniformly distributed, indicating that the distributions of Rh-PE and PE are similar. In PC/PE, Rh-PE was also almost uniformly distributed, suggesting that the monolayer and the bilayer have a similar PC/PE composition (Fig. 1d, Supplementary Fig. S1c). To confirm this finding, we investigated lipid distribution in model nascent LDs using molecular dynamics simulations, in three systems containing TG and (a) pure DOPC or DOPC/DOPE mixtures, (b) 80/20 and (c) 60/40. We found that DOPE and DOPC mix ideally and their distribution was approximately homogeneous (Supplementary Fig. S1d). DOPC was only slightly enriched in the monolayer compared to the bilayer, while DOPE was slightly enriched in the bilayer compared to the monolayer—the differences being minor in both cases (Supplementary Fig. S1d). Overall, the data confirm that PE/PC mixtures are ideal mixtures, with an approximately even distribution of both lipids between the monolayer and bilayer interfaces.The above characterizations indicate that DIBs recapitulate sufficiently well conditions of a bilayer containing an oil droplet, as previously shown35. We subsequently use DIBs to study protein partitioning.Monotopic proteins strongly bind to TG-covering monolayersWe screened the monolayer-bilayer partitioning of two classes of proteins or peptides: soluble proteins, targeting to LDs from the cytosol, and monotopic integral membrane proteins (moving from the ER bilayer to LD surface. Soluble peptides were directly added to the buffer droplets. Monotopic membrane proteins were added to the buffer droplets from purified LDs or from proteoliposomes (Fig. 2a); mixing relocalized the proteins from LDs, or proteoliposomes, to the interface between the buffer droplet and the oil phase. Phospholipids from liposomes or LDs also relocalized to this new interface, but their total amount was always much less than the amount of the exogenous phospholipids we added; the latter would control the interfacial lipid composition in all of our systems. In practice, buffer-in-oil droplets containing the proteins at the interface were prepared before adding phospholipids to the oil phase (Fig. 2a). Two droplets were then brought together to form a DIB. The protein partition coefficient was determined 10 min after contact, at equilibrium, by quantifying the enrichment level of the protein in the bilayer relative to the monolayers (Fig. 2b).Fig. 2Partitioning of hydrophobic and amphipathic helix-containing proteins to droplet interface bilayers.a Formation of protein-containing DIBs: soluble protein, proteoliposome, or purified LDs are added to the buffer droplet (left). Mixing the droplet in a TG-containing phospholipids solution allows the relocalization of proteins to the buffer droplet interface (middle). When two drops come close together, their monolayers zip to form a bilayer. The proteins are thus offered to relocalize to the bilayer (right). b Protein distribution between the bilayer and the monolayer is determined by the partition coefficient P, which is the protein signal at the bilayer divided by sum of its signals at the monolayers. When a protein partitions preferentially to the bilayer, p > 1; if preference is for the monolayer then p < 1. c, d Distribution of GMAP-210-AH (soluble protein) and ACSL3 (monotopic membrane protein) respectively, in DOPE or DOPC/DOPE (1:1) DIBs. Scale bar: 20 µm. The partition coefficient is represented for each condition as box-plots from n = 5 independent measurements (excepted for GMAP-210-AH in PE, n = 4). e Enrichment parameter in DOPC/DOPE (1:1) membranes for AH- (green) and HD- (pink) containing proteins, shown as floating bars (bar limits, min to max values; central line, mean), 2 ≤ n ≤ 8 independent measurements were done for each protein. HD-containing proteins coming from LDs are Plin 1, Plin 1C, ACSL3, CG2254, CG9186, Oleosin 1, Caveolin 1, HPos; those coming from proteoliposomes are Syt1 57-421, t-snare, Vamp2; AH-containing proteins coming from LDs are Plin3, Plin2, Plin1N; the other AHs are added soluble. f Average partition coefficient of the groups of soluble or monotopic membrane proteins in DOPE or DOPC/DOPE (1:1). Results presented in (Supplementary Fig. S2e) were used to generate (f). Box-plots are defined as follow: box limits, upper and lower quartiles; middle line, median; whiskers, minimum and maximum value. In c, d, e, f each point is represented as a black dot. Source data are provided as a Source Data file.For monotopic membrane proteins, we tested Plin1, ACSL3, CG2254, CG9186, oleosin 1, Hpos, and caveolin1 (Fig. 2d, Supplementary Fig. S2a, c), most of which contain helical hairpin and hydrophobic or amphipathic helix motifs responsible for their localization to LDs37–40. These proteins were tagged with fluorescent proteins and expressed in cells that were subsequently loaded with oleate to induce LDs. LDs bound by the proteins were purified and added to the DIB system. One limitation of this approach is that other proteins contained in the LDs would also relocalize to the DIB interfaces, although not visible. Furthermore, proteins with single transmembrane domains, not fully crossing the ER bilayer, could target to the LD surface, but this has never been shown clearly so far. To test this hypothesis, we prepared proteoliposomes containing some of the SNARE components bearing a transmembrane helix, but not crossing the bilayer. Finally, we studied a group of soluble proteins, including Plin2–3, Plin1 AH-containing domains19,24, and the lipid packing sensors ArfGap1-AH41 and GMAP-210-AH42 (Fig. 2c, Supplementary Fig. S2c, d).For all of the tested proteins, we found a stronger partitioning to the monolayers than to the bilayer, independently of PC/PE ratio (Fig. 2e, f, Supplementary Fig. S2e, f). Additionally, HD-containing proteins showed on average a higher LD enrichment than AH proteins (Fig. 2e, Supplementary Fig. S2f), supporting that proteins coming from the ER bilayer better associate with LDs than soluble proteins. For a subgroup of HD proteins, we measured the partitioning in both PC/PE (1:1) and a more biologically relevant composition (DOPE/DOPC/liverPI/cholesterol, 5:3:1:1), and found very similar results (Supplementary Fig. S2a, b).Finally, the bilayer localization of AH-containing proteins was increased by addition of PE in most cases, but partitioning to the monolayer region was still more favorable (Fig. 2f, Supplementary Fig. S2e, f). The negative spontaneous curvature of PE is known to cause lipid packing defects, which can be sensed by AHs43. This is well illustrated by the highest partition coefficient (close to 1) obtained in PE with the AH domains of GMAP-210 and ArfGap1, which are lipid packing sensors43 (Fig. 2c, Supplementary Fig. S2e). In contrast to AHs, the dependence of HD-containing protein on PC/PE was less clear (Supplementary Fig. S2e, f).In summary, both AH- and HD-containing proteins localized preferentially to the monolayer interface over the bilayer. HD-proteins more strongly partitioned to the monolayer and barely relocated to the bilayer.KWALP peptides recap the global behavior of HD proteinsAt this stage, it is difficult to explain the partitioning trend of the full-length HD proteins. This is in part because most of the proteins, coming from purified LDs, may interact with other unidentified proteins in the system. Also, LD proteins can bear multiple HDs and/or AHs; this is the case for Oleosin1, Caveolin1, HPos, and ACSL337,38,40, which possess an AH motif adjacent to their HD motif. To better understand the determinants of partitioning for pure HD domains, we focused on model peptides of the KWALP family. KWALP peptides consist of a repeated leucine-alanine motif (Fig. 3a), bounded by two tryptophan residues at the C-terminus and three lysine residues at the N-terminus; to this N-terminus we added a glycine linked to a rhodamine-B dye. This peptide features a strong tendency towards helical conformation and transmembrane partitioning, as reviewed from numerous previous studies28; therefore they represent a valid model for transmembrane domains of proteins, including those localizing to the Golgi and plasma membranes17,28,44. Moreover, KWALP represents an excellent model for the minimal basic hydrophobic sequences, commonly found in HD domains of LD proteins (Supplementary Fig. S3). We used KWALP20, with 16 hydrophobic amino acids (Fig. 3a) and a length (when folded in an α-helix) close to the ER bilayer thickness (requiring ~ 20 hydrophobic amino acids28). For comparison, we also studied the partitioning of an AH motif derived from the 11-mer repeat of Perilipin1—termed here PL108 (Fig. 3a)24. KWALP20 and PL108 represent useful models for the two classes of HD and AH proteins tested above.Fig. 3PC/PE ratio modulates the partitioning of model AH and HD peptides.a Helical wheel representation of KWALP20 and PL108-AH, generated from HeliQuest69. b Distribution of KWALP20 and PL108 in DOPE or DOPC/DOPE (1:1) DIBs. KWALP20 is labeled with and PL108 with NBD. Scale bar: 50 µm. The partition coefficient is plotted for both peptides as box-plots (box limits, upper and lower quartiles; middle line, median; whiskers, minimum and maximum value), from n = 4 independent measurements for each condition. Individual data are shown as black dots. c FRAP experiment shows that KWALP20 (purple) and lipid (green) signals are mobile at the monolayer. NBD-PE reports for phospholipids. Yellow arrows indicate the area bleached. Scale bar: 50 µm. Recovery half-time was obtained using one-phase association fitting in GraphPad software and is shown in the upper right box. d KWALP20 distribution in DIBs of different PC/PE ratio. Line profiles (not displayed) are drawn perpendicular to the bilayer and monolayers (as described in Fig. 2b); the thickness of the line is 30-40% of the bilayer size. The corresponding signals are shown in the box (right); black arrows depict the bilayer signal. Scale bar: 20 µm. e Partition coefficient of KWALP20 in DIBs of different DOPC/DOPE ratios represented as box-plots (box limits, upper and lower quartiles; middle line, median; whiskers, minimum and maximum value). Sample size was n = 31 for 0% PC, n = 14 for 10% PC, n = 18 for 20% PC, n = 41 for 30% PC, n = 16 for 40% PC and n = 28 for 50% PC. Each data point is plotted. f Relocalization of KWALP20 from the monolayer to the bilayer after addition of DOPC to DOPE DIBs. The bilayer signal is plotted over time. Image brightness is enhanced to improve bilayer viewing. Scale bar: 50 µm. Source data are provided as a Source Data file.We prepared DIBs containing both KWALP20 and PL108 to compare their distribution under identical experimental conditions. When the DIB contained PE exclusively, PL108 partitioned almost equally between the monolayer and the bilayer (Fig. 3b), while KWALP20 was surprisingly absent from the bilayer (Fig. 3b). When PE/PC (1:1) was used, KWALP20 signal in the bilayer increased, while PL108 bilayer concentration significantly decreased (Fig. 3b). These observations are consistent with the behavior of most HD- and AH-containing proteins (Fig. 2e, Supplementary Fig. S2f): both peptides partition more favorably to the monolayer, especially for KWALP20; in PE lipids, PL108 partitions more evenly, like GM210-AH, and KWALP20 is barely detectable to the bilayer, like Oleosin1, Caveolin1 or Hpos (Supplementary Fig. S2c).Since KWALP recapitulated the partitioning of most of the full-length HD-proteins (Fig. 2e, f, Supplementary Fig. S2c, e, f), we further investigated the driving forces for its distribution to establish general principles underpinning the enrichment of HDs to LD surface.PC/PE ratio regulates the partitioning of KWALPKWALP20 was almost totally absent from the bilayer in PE, while it was well folded in the monolayer (Supplementary Fig. S4a). Importantly, the peptide was laterally mobile, as shown by the rapid recovery of fluorescence following photobleaching (Fig. 3c, Supplementary Fig. S4c); this recovery was indeed due to in-plane diffusion, because recovery from bulk did not happen within 10 minutes (Supplementary Fig. S4b). Since the hydrophobic thickness of the PE DIB bilayer is comparable to the peptide length, it is unlikely that KWALP20 localization to the bilayer was prevented by hydrophobic mismatch. We next increased further the PC/PE ratio, and observed that the concentration of the peptide in the bilayer increased with PC level, but it still remained lower than in the monolayer (Fig. 3d, e). In the bilayer, the KWALP peptide was also mobile but showed a significant tendency to cluster as the PC level was increased (Fig. 3f, Supplementary Fig. S4f). To verify if the bilayer localization was dynamic, and if diffusion was not prevented by peptide aggregation, we followed the peptide signal while changing the phospholipid composition. Starting from a pure PE DIB, where KWALP20 was absent from the bilayer, we added a TG solution containing PC to the oil phase surrounding the droplets (Fig. 3f). The recruitment of PC to the interface of the droplets was demonstrated by the increase in the contact angle between the droplets35 (Supplementary Fig. S4e) and it was concomitant with an increase of KWALP20 signal in the bilayer (Fig. 3f). We also noticed the appearance of KWALP clusters after PC recruitment (Fig. 3f, Supplementary Fig. S4f), suggesting that clustering is an inducible equilibrium state. For comparison, PL108 followed the opposite trend, as it was excluded from both the bilayer and the monolayer by PC recruitment (Supplementary Fig. S4g). These results suggest that the system has lower free energy when the peptide is at the monolayer, and the free energy gap between configurations where KWALP is at the monolayer or at the bilayer is decreased by PC. This energy gap does not come from a hydrophobic mismatch since when we repeated the above experiments with a longer KWALP version, namely KWALP28 (Supplementary Fig. S3a), which should be longer than the bilayer thickness, the peptide behave almost exactly as KWALP20, within the resolution limits of our measurements (Supplementary Fig. S4c, d).In conclusion, our results show that KWALP20 partitions dynamically between the monolayer and the bilayer, but it has a clear preference for the monolayer, especially in high PE levels. In cells, if there is no regulation of ER-to-LD partitioning, HDs would be favorably adsorbed at the LD surface as a result of free energy minimization.Phospholipid shape defines KWALP partitioningSince changing PC/PE ratio varied partitioning, we hypothesized that the affinity of HDs for lipids may be a driving force for partitioning. An HD protein can interact with phospholipid acyl chains, TG, and water, although interactions with the latter are unfavorable. We wanted to know which interactions would be responsible for the accumulation of the peptides to the monolayer.We prepared phospholipid-free buffer-in-TG droplets containing KWALP20. The protein signal at the interface was uniform (Fig. 4a, b). When the interface was lined by PE, the signal was also uniform in most cases (Fig. 4a, b). Instead, when PC alone lined the interface, the protein formed clusters (Fig. 4a, b) in which the peptide was mobile (Supplementary Fig. S5a). Similar clustering was observed when PC was added to DIBs (Fig. 3f, Supplementary Fig. S4f), and never observed for AHs. Apparently, protein-protein interactions become more favorable at the PC monolayer interface, suggesting that KWALP has more affinity for TG than for phospholipid acyl chains. We thus hypothesize that the relative contact of an HD with TG and phospholipids determines HD monolayer-bilayer partitioning.Fig. 4The partitioning of KWALP is altered by phospholipid shape.a KWALP20 surface distribution in bare TG-buffer droplets or in TG-buffer droplets covered by DOPE or DOPC. Red arrow highlights peptide clustering (in DOPC condition). The yellow square regions are enlarged on the right side of each image. Scale bar: 100 µm. b Quantification of KWALP pattern, i.e., uniform (light green) or clustering (red) signal, in TG (n = 8), TG + DOPE (n = 47) and TG + DOPC (n = 19), from n independent measurements. c Schematic representation of the difference in phospholipid packing, and thus in HD-TG contact, when DOPE (cone shape) or DOPC (cylinder shape) are present. Increasing DOPC concentration in a DOPC/DOPE monolayer increases the lipid packing and decreases the contact between HDs and TG. d Distribution of KWALP20 in DOPE, N-methyl-PE, N,N-dimethyl-PE, and DOPC DIBs. These phospholipids have incremental curvature between that of DOPE and DOPC. KWALP20 is labeled with Rh-B. Line profiles (not displayed) are drawn perpendicular to the bilayer and monolayers (as described in Fig. 2b); the thickness of the line is 30–40% of the bilayer size. Arrows indicate the bilayer signal. Scale bar: 20 µm. e Partition coefficient of KWALP20 in DIBs of different compositions is represented as box-plots (box limits, upper and lower quartiles; middle line, median; whiskers, minimum and maximum value). Sample size is n = 29 and 20 for KWALP20 in N-methyl-PE and N,N-dimethyl-PE respectively. Previous results of varying PC/PE ratios (Fig. 3e) are reported in light gray. Individual data points are indicated. Source data are provided as a Source Data file.The monolayer of a droplet contiguous with a bilayer is less packed with phospholipids than the bilayer leaflets9. In a bilayer, HD peptides would be in contact with phospholipid acyl chains along their full length while, in a monolayer, a significant fraction of the peptide would be in contact with TG. Thus, the higher affinity of HD proteins for TG over phospholipids can explain why HD proteins partition more favorably to monolayers. Why would the PC/PE ratio matter in this picture? Very likely because PC/PE can modulate the probability for HDs to contact with TG. PC and PE do not differ in their acyl chain composition (two oleoyl chains in both cases) but they differ in their average shape: PC has a cylindrical shape while PE is conical (Fig. 4c). Therefore, PC proffers a higher phospholipid monolayer packing which, in turn, reduces the probability of contact of an HD with TGs, and increases the probability of contact with phospholipid acyl chains (Fig. 4c). As a consequence, HD would less efficiently partition to the monolayer when the PC/PE ratio is increased.Our model suggests that phospholipid shape modulates HD monolayer-bilayer partitioning. To test this, we used dioleoyl phosphatidic acid (PA), which has a negative spontaneous curvature, like PE. We found that, in a PA DIB, KWALP20 was almost excluded from the bilayer (Supplementary Fig. S5b), as observed in PE. To further challenge our hypothesis, we repeated the KWALP20 partitioning experiments in N-methyl-PE and in N,N-dimethyl-PE phospholipids; these are, from a structural standpoint, intermediates between PE and PC (PC is N,N,N-trimethyl PE) (Fig. 4d, Supplementary Fig. S5c). Increasing methylation increased KWALP signal in the bilayer (Fig. 4d, e, Supplementary Fig. S5c, d), an effect similar to increasing PC/PE ratio, in agreement with our prediction.In conclusion, our data indicate that HDs prefer mixing with TG instead of being in contact with membrane phospholipids. Because phospholipid packing is less compact in a monolayer compared with a bilayer9, and because a monolayer thickness is half the thickness of a bilayer, partitioning toward monolayers is favored as they expose HD proteins to TG (Fig. 4c). Increasing the monolayer phospholipid packing, for example by increasing the PC/PE ratio, increases HD-phospholipid interaction at the expense of HD-TG interaction. In this case, the peptide less efficiently discriminates the monolayer of the bilayer from the monolayer covering TG and therefore its monolayer accumulation is dampened.TG is responsible for the accumulation of HDs in monolayersOur model postulates that KWALP accumulates to LD monolayers because it mixes with TG more favorably than with membrane phospholipids (Fig. 4a, b). To challenge this model, we altered the relative affinity of the peptide for phospholipids over the oil phase by changing the chemical nature of the oil. We chose an oil phase very different from TG, namely silicone oil, in an attempt to trigger major changes in oil-protein affinity. Silicone oil is chemically very different from TG but they both have a high surface tension with water9.In PE DIBs made in TG, KWALP20 was absent from the bilayer, as described above (Fig. 5a, c). In contrast, by replacing TG with silicone oil, we systematically observed that KWALP20 was in the bilayer (Fig. 5b, d). Moreover, phospholipid clusters regularly appeared at the monolayer interface and they were enriched in the peptide (Fig. 5b, d). Outside these areas, the peptide signal was weaker at the oil-water monolayer interface. Our interpretation is that the peptide has a higher affinity for phospholipids than for silicone oil, and therefore it preferentially distributes to phospholipid-rich regions, i.e., the bilayer and the phospholipid clusters.Fig. 5The distribution of KWALP HD depends on the oil chemistry.a, b Distribution of KWALP20 in DOPE DIBs. KWALP20 is labeled with Rh-B. Oil phase is TG (a) or silicone oil (b). Blue arrows indicate the monolayers, yellow ones indicate the bilayer, and red ones indicate cluster areas. Plot profiles are determined using the yellow lines. The bilayer signal is indicated by a black arrow. Scale bar: 20 µm. c, d Partition coefficient is reported in gray in TG (c) and silicone oil (d), as mean ± sd (n = 2 and 8 independent measurements respectively). A cluster enrichment coefficient (red) is determined for the experiment in silicone oil, and is shown as mean ± SD (n = 8 independent measurements). Each data point is shown as a black dot. Source data are provided as a Source Data file.To further validate our findings, we repeated the same experiment with VAMP2, one of the SNARE components that binds to membranes with a transmembrane domain. In TG, VAMP2 was completely absent from the bilayer (Supplementary Fig. S5e, g) while in silicone oil it was in the bilayer and clustered with phospholipids at the oil-water interface (Supplementary Fig. S5f, h), exactly like KWALP20.Altogether, our results suggest that HD-containing proteins partition to regions where they find the highest molecular affinity. They have more affinity for TG than for phospholipids, and therefore get enriched in sites offering easier access to TG.KWALP egresses membranes to accumulate in model LDsThe DIB system revealed the existence of an energy gap favoring the higher enrichment of HDs to monolayers, due to their preferential mixing with TG over bilayer phospholipids. Thus, when a nascent LD is formed in a bilayer, as during the early step of LD biogenesis (Fig. 1a), HDs would all preferentially relocalize to the nascent LD. Such behavior has been reported for many HD-containing proteins, including HPos, LiveDrop, or Oleosins27,37,39,40,45. We tested this hypothesis.To mimic the situation of a forming LD, we used the droplet-embedded vesicle system (DEV), which is a giant unilamellar vesicle (GUV) with TG droplets incorporated in between the bilayer leaflets8 (Fig. 6a). We incorporated KWALP into PC/PE (7/3) GUVs, during GUV electroformation or after, by mixing the GUVs with the peptide, (Fig. 6a, Supplementary Fig. S6a). Next, the KWALP-containing GUVs were mixed with TG-in-water droplets in order to generate DEVs (Fig. 6a). We found that the peptide was massively enriched onto the monolayer side, consistent with our predictions and with results obtained in DIBs (Fig. 6a, b, Supplementary Fig. S6b, c). Next, we used molecular dynamics simulations to account for possible size and curvature effects which are not recapitulated in DEVs or DIBs. We generated bilayers in which 16 or 32 copies of the KWALP20 peptides were all incorporated from one side of a DOPC bilayer. In the absence of TG, the peptides were randomly distributed (Fig. 6c). When TG was incorporated into the bilayer, first it spontaneously nucleated a lens, then almost all the peptides moved to the surface of the lens, as quantified by the peptide distribution profile (Fig. 6d). The peptides remained mobile and were able to transiently move to the bilayer region, indicating that the equilibrium is dynamic and no major kinetic barrier traps the peptides in the monolayer. The same result was obtained for KWALP28 peptides (Supplementary Fig. S6d). These results are consistent with the previous ones in the DEV and DIB systems.Fig. 6Nascent LDs are attractive to monotopic proteins.a Formation of DEVs-containing KWALP20 labeled with Rh-B: (left) KWALP20 is inserted in GUVs, during GUV electroformation or by mixing GUVs with the peptide; an example of the resulting KWALP20-interted GUV is shown. (right) KWALP20-containing GUVs are mixed with a TG-in-buffer droplet to generate droplet-embedded vesicles; several examples of the resulting DEVs are displayed: The peptide is labeled with Rh-B. The KWALP signals on the monolayer and the bilayer are respectively depicted by blue and yellow arrows. Strong accumulation happens at the monolayer. Image brightness is enhanced to improve monolayer viewing. Scale bar: 2 µm. b The ratio between monolayer and bilayer signals is plotted as mean ± SD (n = 5 independent measurements). Individual points are indicated. Source data are provided as a Source Data file. c, d Left: snapshot of molecular dynamics simulations of a bilayer with 32 KWALP20 peptides in parallel orientation, in the absence (c) and the presence (d) of a TG lens. Hydrophobic amino-acids are represented in blue and charged ones (lysines) in red. On the right of each simulation is displayed the average protein density profiles in the bilayer plane, averaged over the entire MD simulation (20 μs), in the absence (d) and in the presence (e) of the TG lens. e Basic model of how the different interactions of an HD favor its LD monolayer accumulation. HD interacts with TG, phospholipids or water; the monolayer packing regulates the contact between these species and tunes the reaction constant kon/koff.Overall, these two sets of data indicate that the free energy of the system is lower when HD proteins are at the monolayer. These results reinforce the idea that HD-containing proteins can sense TG and accumulate at TG hotspots.DiscussionProtein-lipid interactions has a key role in membrane biology by controlling protein localization and functionality14,28,31. In particular, a variety of protein-phospholipid interactions are responsible for the localization of many proteins to specific organelles or membranes regions28. TG is not a membrane lipid but a bulk lipid. Our findings support that most HD proteins have a higher affinity for TG over phospholipids in a membrane environment (Fig. 6e). Consequently, under LD biogenesis conditions, HD-containing proteins would more favorably be recruited to sites of TG accumulation, instead of remaining in the ER bilayer (Fig. 6e). Helical hairpins, hydrophobic helices, and transmembrane domains not fully crossing the ER bilayer, would accumulate to nascent LDs. Even AH-containing proteins would do so, but to a lesser extent (Fig. 2f). In short, emerging nascent LDs in the ER would be hotspots that attract nearby HD-containing proteins. Controlling these stages of LD formation will be critical for defining the proteome of the emerging LDs8,45,46,47 and for keeping ER homeostasis.While our results predict that HD proteins preferentially accumulate to LD monolayers, clearly not all ER HD-containing proteins target to LDs. Hence, there must be counteracting mechanisms that reduce and prevent unspecific HD-protein targeting to emerging LDs5,6. Amino acid composition of an HD might determine the HD-TG affinity and hence the ER-to-LD partitioning extent. For instance, the presence of charged residues in a HD may hamper HD-TG interaction, since embedding charges in a low dielectric milieu such as of a TG phase is unfavorable, and generally requires conformational changes to the protein or interaction with a protein of opposite charge. More generally, as we recently proposed for AHs9, there could exist sequence motifs tailored with an optimal affinity with TG.The presence of HDs can perturb the structure of lipid bilayers, generating stresses that tend to reduce protein-phospholipid interactions, for example by clustering proteins, as predicted by theoretical studies and confirmed by molecular simulations17,28,31. LD monolayers have more phospholipid packing defects than bilayers9, and allow exposing TG to water molecules. During LD formation, the relocalization of HDs from the bilayer to a forming LD would reduce the stresses caused by the protein to the bilayer and possibly mask phospholipid packing defects at the LD monolayer. Such partitioning would be highly favorable as it would minimize energy on both bilayer and monolayer interfaces. Actually, even prior to LD assembly, transient TG clusters appearing in the ER bilayer2 may attract HD-proteins or, inversely, HD-proteins can trigger the clustering of TG molecules around them48, thereupon promoting LD nucleation and alleviating ER stress.Finally, LD formation is stimulated by diverse physiological conditions such as excess nutrients or ER stress11. During ER stress, the formation of LDs may be stimulated in order to sequester damaged HD-containing proteins to be degraded, by macrolipophagy for example49. Indeed, in this case, proteins tend to expose hydrophobic sequences that would be favorably adsorbed to LDs. In this context, LD formation would serve as a protein quality control pathway maintaining ER proteostasis, a function different from the primary role of LDs in metabolism50. Accordingly, another mechanism triggering ER bilayer stress is the alteration of ER phospholipid composition51, especially when PC/PE ratio is decreased10,52,53. Here, our data bring important insights on how this ratio can modulate the partitioning of HD-containing proteins between ER and LDs: decreased PC/PE favors HD targeting and retention to the LD monolayer. Thus, by tuning PC/PE ratio, cells may be able to shift more HD-containing proteins from the ER bilayer to LDs, or vice versa, for degradation for instance. Along this analysis, increased PE/PC levels in liver is caused by dysfunctions of the phosphatidylethanolamine N-methyltransferase and associated with steatohepatitis54,55, a condition linked to LD formation. Based on our data, such PE/PC-induced steatosis may be related to abnormal ER-to-LD protein trafficking. In contrast to mammalain cells, Drosophila cells present high PE/PC levels under normal conditions56; therefore, the ER-to-LD partitioning extent of HD-proteins in these cell lines may strongly differ from mammalian cells.In conclusion, our data connect various fields involving protein-lipid interactions, from basic membrane biophysics to membrane biology, lipid metabolism, and cellular proteostasis. Our findings highlight the attractiveness of LD surface for HD-containing proteins. Accumulating neutral lipids would be rapidly detected by proteins bearing these domains. Such non-selective detection is clearly prevented by cells by means to be discovered.MethodsMaterialHEPES, Kacetate, MgCl2, sodium phosphate monobasic, sodium phosphate dibasic, choloroform, trifluoroethanol, Octyl-ß-D-glucopyranoside were bought from Sigma Aldrich. DOPC (1,2dioleyl-sn-glycero-3-phosphocholine), DOPE (1,2dioleyl-sn-glycero-3-phosphoethanolamine), N-methyl-PE, N,N-dimethyl-PE, liver PI, Rhodamine-DOPE and NBD-DOPE were purchased from Avanti Polar Lipids, Inc. CAV1-GFP plasmid was purchased from Sino Biological (catalog no. HG11440-ACG). The following plasmids were gifts: YFP-CG2254 and YFP-CG9186 from Dr. Mathias Beller; GFP-Plin 1, GFP-Plin 1N, GFP-Plin 1C, mcherry-Plin 2 and mcherry-Plin 3 from Dr. David Savage; EGFP-ACSL3 plasmid from Dr. Joachim Füllekrug57; GFP-HPos from Prof. Albert Pol40. Cells were obtained from American Type Culture Collection and no contamination for mycoplasma was detected.Peptides and proteins preparationRhB-KWALP peptides, RhB-ArfGAP1, RhB-GMAP-210 and NBD-CAV1-AH were synthesized by peptide 2.0 Inc., NBD-PL108 was made by Proteogenix SAS, and RhB-NS5A was synthesized by Eric Diesis. All the peptides were chemically synthesized and purified by reverse phase high-performance liquid chromatography (HPLC). Their purity was higher than 95%, as determined by analytical HPLC and their mass was confirmed by mass spectrometry. The amino-acid sequences of the peptides are:KWALP20: RhB-GKKKLALALALALALALWWA-AmideKWALP28: RhB-GKKKLALALALALALALALALALALWWA-AmideArfGAP1-AH: RhB-FLNSAMSSLYSGWSSFTTRAKKFAKGMAP-210-AH: RhB-MSSWLGGLGSGLGQSLGQVGGSLASLTGQISNFTKDMLCAV1-AH: NBD-LFEAVGKIFSNVRINLQKEIPL108: NBD-PPEKIASELKDTISTRLRSARNSISVPIASNS5A: RhB-SGSWLRDVWDWVCTILTDFKNWLTSKLFPKL-AmidePlin proteins, CG2254, CG9186, CAV1, HPos, NS5A and ACSL3 were obtained from purified lipid droplets be using the following protocol. LD purification from Huh7 cells expressing fluorescently tagged LD proteins: cells from five 15 cm dishes were harvested, washed once in ice-cold PBS, and lysed using a 30 G needle in 1 ml of homogenization buffer containing 20 mM Tris and completeTM protease and phosphatase inhibitors, at pH 7.5; for LD isolation, 1 ml of cell lysates was mixed with 1 ml of 60% sucrose in Tris-EDTA buffer supplemented with protease inhibitors, successively overlaid with 20, 10, and 0% buffered sucrose in an 5 ml Ultra-Clear centrifuge tubes (Beckman). The tube was centrifuged for 16 h at 100,000 G and 4 °C, using an SW60 rotor in a Beckman L8-70 centrifuge. The upper 300 µl fraction was collected from as the LD fraction.Fluorescently labeled Arf1 was generated using an Arf1-variant in which the single cysteine residue of Arf1 was replaced with serine, and the C-terminal lysine was replaced with cysteine, yielding Arf1- C159S-K181C. In short, human Arf1- C159S-K181C and yeast N-myristoyltransferase were coexpressed in Escherichia coli supplied with BSA-loaded myristate. Cell lysates were subjected to 35% ammonium sulfate, and the precipitate, enriched in myristoylated Arf1, was further purified by DEAE-ion exchange. Eluted fractions of interest were concentrated in spin-column filters with a 10-kD cutoff (Millipore), and fluorescently labeled using Cy3-maleimide (GE Healtcare) according to the manufacturer’s protocol. To remove excess dye, samples were purified by gel filtration using a Superdex 75 column (GE Healthcare).Oleosin1 lipid droplets were obtained from Arabidopsis seeds, provided by Dr. Martine Miquel.Vamp2, Tsnare (complex of syntaxin1a and SNAP25) and Synaptotagmin 1 57–421 C277A, E265C (Syt1) were produced and purified by the team of Frédéric Pincet. The proteins Vamp2, Tsnare and Syt1 (solubilized in Octyl-ß-d-glucopyranoside (OG) micelles) were labeled with a fluorescent probe Atto-565 maleimide (Atto-tec, GmbH), according to the manufacturer’s instructions. Free-dye was removed by gel-filtration, using a Sephadex G25 column (PD-minitrap G25, GE Healthcare). Labeled-proteins were then purified in DOPC/DOPE 1:1 proteoliposomes (P/L 1:1000): DOPE and DOPC were mixed in a glass tube, then the chloroform was removed under an argon flow and the glass tube was left under vacuum for at least one hour. The resulting lipid film was rehydrated with the Atto565-protein solution during 30 minutes. The sample was then diluted 3 times to decrease OG concentration below its cmc and a dialysis was performed overnight in a 10 kDa Slide-A-Lyzer dialysis cassette (Thermo Scientific) in order to remove OG and keep the protein into liposomes. Final buffer was the following: 25 mM HEPES pH 7.4, 120 mM KCl, 1 mM DTT (with 0.5 mM CaCl2 for Syt1).Droplet interface bilayer formationUnless mentioned, in vitro experiments were performed in HKM buffer: 50 mM HEPES, 120 mM Kacetate, 1 mM MgCl2 at pH 7.4. KWALP peptides were dissolved in trifluorethanol at 200 μM, and then diluted in HKM to get a final concentration of 10 μM. PL108 was solubilized at 50 µM in HKM, CAV1-AH at 10 µM in HKM (with 0.1% DMSO), GMAP-210 at 2 µM (with 0.1% DMSO, 16 mM urea, 80 µM DTT), ArfGAP1 at 8 µM in HKM (with 0.1% DMSO). NS5A was diluted in HKM to obtain a final concentration of 1 µM and 10% of trifluoroethanol was added to ensure of its folding. The proteins in LD (Plin 1, Plin 1 C, ACSL3, CG2254, CG9186, Oleosin 1, CAV1, HPos) or proteoliposomes (Syt1 57–421, t-snare, VAMP2) were used directly. DTT was added at a final concentration of 2.5 mM in case of aggregation.Phospholipids (eventually with 0.2% of labeled-PE) were evaporated under a stream of argon to remove the chloroform. The resulting lipid film was then re-solubilized to the desired concentration (0.2% w/w) in trioctanoate (or silicone oil), strongly vortexed and sonicated for 10 min to ensure a complete dissolution. To form DIBs, buffer-in-oil emulsions were made using 10 μl HKM (or peptide/protein solution) dispersed in 100 μl of trioctanoate. This emulsion was strongly vortexed in order to let the protein relocalize to the surface of the droplets. Then, the same volumes of peptide/protein emulsion and lipids in oil phase were put together and the resulting emulsion was placed on a hydrophobic coverslip (glass coverslip #0 from Menzel Glaser, Braunschweig, Germany, which was covered by PDMS). The sample was let to equilibrate for 10 min and was then observed by confocal fluorescence microscopy (LSM 800, Carl Zeiss, Oberkochen, Germany), with a ×10 or oil-immersed ×63 objective depending of the size of the droplets. When the emulsion is poured onto the observation glass, droplets which are closer to each other spontaneously adhere, because of the poor solubility of the phospholipids in the oil phase, and form a bilayer. The final lipid concentration in the oil phase is then 0.1% w/w and the interfacial lipid composition is determined by lipid composition in this oil phase.To study the effect of PC on the localization of KWALP or PL108 peptide in a dynamic way, the two peptides were both used at the same concentration of 50 µM in DOPE DIBs. Then, 5 µl of DOPC 0.2% in trioctanoate (containing 10% CHCl3) was added to the sample.Giant unilamellar vesicles formationPhospholipids (DOPC/DOPE (7:3 or 6:4)) in chloroform at 2.5 mM were spread on an indium tin oxide (ITO)-coated glass plate. After chloroform evaporation, the resulting lipid film was then placed under vacuum for 1 h. The chamber was sealed with another ITO-coated glass plate. GUVs were grown by electroformation in a sucrose solution (0.1 g ml−1, ≈280 mosmol kg−1) with the following settings: 100 Hz, 1.25 V, for 1.5 to 2 h. They were then collected carefully with a Pasteur pipette, placed in a Eppendorf® tube and stored at 4 °C.Droplets embedded vesicles formationDroplets were made using an oil-in-water emulsion: 20 μl of trioctanoate were mixed with 100 μl of HKM buffer. The solution is then sonicated to form small droplets. 10 µl of 20 µM KWALP peptide solution was added to 40 µl of a GUV solution, which were then incubated with 20 µl of droplets for 5 min. We also added KWALP to dried phospholipids prior the electroformation and this also led to the incorporation of KWLAP to GUVs, which were subsequently used to make DEV (Fig. 6). With both approaches, the DEV/KWALP sample was then placed on a glass coverslip pretreated with 10% (w/w) BSA and washed three times with buffer, and it was then observed by confocal fluorescence microscopy (LSM 800, Carl Zeiss, Oberkochen, Germany), with an oil-immersed ×63 objective.Electrical measurementAqueous droplets in oil were blown at the tip of micropipettes containing Ag/AgCl electrodes (connected to an Axopatch 200B amplifier—Molecular Device) and filled with electrolyte buffer. Micropipettes are made from borosilicate capillary (Harvard Apparatus, 1.0 mm OD×0.50 mm, ID×150mm) pulled with a micropipette puller (Sutter Instrument) to obtain tip with inner diameter of 2 µm. Before any use, tip of the micropipettes was treated dipping in a dimethyldichlorosilane solution to avoid capillary wetting by the aqueous droplets. Micropipettes were manipulated through MP225 and MP285 micromanipulators (Sutter Instrument). After blowing droplets at each micropipettes tip, 5 min are waited to allow monolayer formation, then micropipettes are moved to put into contact the two droplets to allow formation of the bilayer. Once DIB is stable, the electrical measurement was performed. This consisted in repeatedly imposing a 20 mV voltage step for 300 ms between the two sides of the DIB and measuring the resulting current. At the same time as capacitance measurements, images of droplets were acquired using a IDS camera mounted on an Olympus IX71 inverted microscope with a 20x objective to measure bilayer area of the DIB.Thickness calculationThe capacitance value C was obtained from the fitting by an exponential of the transient capacitive current at the beginning of the voltage step to determine its time constant. The thickness of the bilayer was then calculated assuming that the bilayer can be assimilated to a dielectric material using the relation: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e = \frac{{\varepsilon _r.\varepsilon _0.S}}{C}$$\end{document}e=εr.ε0.SC where ε0 is the permittivity of vacuum, εr the dielectric constant of the material (εr = 2.8)58 and S the surface of the bilayer calculated from images treated on ImageJ.Molecular dynamics simulationsTo study protein distribution in nascent LDs, we carried out molecular dynamics (MD) simulations at the coarse-grained level using the MARTINI force field50,59,60 (version 2.2). First, we generated a system containing 2016 DOPC lipids, 625 triolein (TO) molecules, and approximately 83,000 water particles; the approximate size was ca. 27 × 27 × 18 nm. TO molecules were initially dispersed homogeneously in the DOPC bilayer, and phase-separated spontaneously to form an oil lens in the bilayer. The system was simulated for 20 μs, and its shape and properties did not change during the last 10 μs. Then, protein-containing systems were generated from the equilibrated lens system, inserting 16 or 32 copies of different transmembrane peptides. Peptides were always inserted in the bilayer region of the system. We used 2 similar peptide sequences, KWALP20 and KWALP28, both in parallel and anti-parallel orientation (i.e., half of the peptides pointing up and half pointing down). In total, we built 8 different protein-containing lens systems. For each system, MD simulations were carried out for 20 μs, and the last 10 μs were used for analysis.To study the distribution of phospholipids in LDs, we carried out simulations of large nascent LDs, containing 18144 phospholipids (DOPC and/or DOPE), 7500 TO molecules, and ca. 1.9 million water particles; the system size was approximately 78 × 78 × 40 nm. We carried out 3 simulations: one with 100% DOPC, one with DOPC:DOPE 80:20, and one with DOPC:DOPE 60:40. Each simulation was carried out for 30 μs, and the last 20 μs were used for analysis.All coarse-grained MD simulations were carried out with GROMACS (v2016.4) software61, using the leap-frog integrator and a time step of 20 fs. Non-bonded interactions were calculated with the Verlet neighborlist algorithm, with a Verlet buffer tolerance of 10−6 kJ mol−1 ps−1 and a cutoff of 1.1 nm; electrostatic interactions were shifted to zero from 0 nm, long-range electrostatics were calculated with the reaction-field method (εR = 15, εRF = ∞); Lennard-Jones potential was shifted to zero at the cutoff. The stochastic velocity rescaling thermostat62 with a time constant of 1 ps was used to maintain the temperature of the membrane (lipids and proteins) and the solvent separately at 300 K. Pressure was controlled semi-isotropically using the Parrinello–Rahman barostat63 with a reference pressure of 1 bar, compressibility of 4×10−4 bar−1, and a time constant of 12 ps.Analysis of protein density was carried out with in-house software17 after re-centering the trajectory, using the center of mass of the largest TO cluster as the center of the simulation box.Analyses of TO content in the bilayer, DOPC:DOPE contact fraction and mixing, and phospholipid distribution between bilayer and monolayer region were carried out with in-house software, freely available on our web site (https://mmsb.cnrs.fr/en/team/mobi-en/softwares/).To study the effect of DOPE lipids on membrane thickness, we carried out all-atom simulations of pure DOPC and DOPC:DOPE 1:1 mixtures, using the CHARMM36 force field64 and the TIP3P water model65. Simulation boxes contained 100 lipids (50 per leaflet) and 5000 water molecules, and simulation time was 400 ns.Simulations were carried out with the GROMACS 2020 software, using the leap-frog integrator and a time step of 2 fs. Non-bonded interactions were calculated with the Verlet neighborlist algorithm, with a Verlet buffer tolerance of 10−6 kJ mol−1 ps−1 and a cutoff of 1.2 nm. The PME algorithm66,67 was used for long-range electrostatics. The temperature was maintained at 298 K using the stochastic velocity rescaling thermostat62 with a time constant of 1 ps. Pressure was controlled with the semi-isotropic Parrinello–Rahman barostat63, with a reference pressure of 1 bar, compressibility of 4.5 × 10−4 bar−1, and a time constant of 10 ps. Analysis of mass density was carried out over the last 300 ns of the trajectories, with standard GROMACS tools.Circular dichroismCD spectra were recorded over the wavelength range 185–250 nm, at 0.2 nm intervals and 20 nm min−1 scan speed, on a Jasco 815 spectropolarimeter (Jasco Inc., Easton, MD). Temperature was kept at 20 °C. Spectra measurements were performed in a 1 mm path length quartz cells from Hellma GmbH (internal volume 200 µl). Experiments were done either in TFE, 10 mM phosphate buffer pH 7.4, DOPC small unilamellar vesicles or TG emulsions with or without phospholipids (DOPC, DOPE, DOPC/DOPE 1:1). To prepare small unilamellar vesicles, DOPC in chloroform was put in a glass tube and the chloroform was removed using a stream of argon. Then the resulting lipid film was dried under vacuum for at least 30 min, and it was rehydrated with phosphate buffer and vortexed strongly. Finally, the lipid solution was sonicated to reduce the size of the vesicles. The oil-in-buffer emulsions were done by mixing 30 µl trioctanoate (eventually 0.2% w/w phospholipids) with 500 µl of buffer, and then sonicating the mixture. KWALP concentration was 20 µM and DOPC liposomes concentration was 1 mM. Data obtained were collected and processed using the software Spectra Manager®, and are then reported as molar ellipticity per residue (degree dmol−1 cm2 residue−1), given by:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[\theta ]_{{\mathrm{molar}}} = \frac{{100 \, \times \theta }}{{c \, \times l \, \times N}}$$\end{document}[θ]molar=100×θc×l×Nwhere θ is the recorded ellipticity in degrees, c is the peptide concentration in mol l−1, l is the cell path-length in cm and N is the number of residues of the peptide. In order to estimate the peptide secondary structure content, an analysis of CD spectra was done using CDPro software68.StatisticsData analysis and representation were performed in Prism 7 (GraphPad Software, US). Information about sample size, errors bars and statistical tests are reported in each figure legend.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationReporting Summary | nature communications | [
"Article"
] | [
"Membrane biophysics",
"Biological physics"
] | droplets (LDs) are lipid storage organelles cellular energy LD biogenesis occurs at endoplasmic reticulum membrane during energy stress conditions starts neutral lipids low concentration dissolved in ER neutral lipids demix from phospholipids form oil lens or nascent droplet (Fig. lens grows emerges in cytosol as mature LD oil-in-water droplet covered by phospholipid monolayer with proteins embedded proteins target surface around Proteins targeting LD surface from ER membrane cytosol7 ensure LD budding8 proteins bind accumulate to LDs neutral lipid chemistry determinant Specificity of protein targeting to LDs of LD biology understanding knowledge on lipid metabolism cellular. of droplet interface bilayers of ER phospholipid bilayer contiguous with monolayer nascent LD DIB system water phase in light blue oil phase in yellow of DIB bilayer of DOPE DOPC/DOPE thickness of hydrophobic region of DIB bilayer in determined by capacitance measurementResults box-plots limits upper quartiles middle line median whiskers minimum maximum value mean from = 5 experiments Each point black dot Distribution of Rh-PE between bilayer monolayers in DOPE DOPC/DOPE (1:1) DIBs results mean ± SD of n = 10 5 measurements Each point black dot Significance determined by Welch’s t-test two-value indicated by ns p > 0.05. Source data file.Membrane properties regulate protein distribution at bilayer-encircled LD-water interface from bilayer-water interface loose lipid hydrophobic region larger than bilayer hydrophobic core neutral lipids phospholipid acyl chains discrepancies proteins preference for one interface proteins associating with LD surfaces peripheral or cross bilayer membranes Proteins moving from ER to LD contain helical hydrophobic domains monotopic one include hairpins hydrophobic helices transmembrane domains not crossing soluble proteins use amphipathic helices for binding to LDsbinding AHs to LDs documented in vitro vivo9 AHs act as surfactants adsorbed to oil/water interface LDs interfacial energy AHs membrane features surface charges curvature phospholipid packing defects neutral lipids9 less known HDs LDs ER membrane-LD energetics binding LDs parameters controlling ER-to-LD partitioning known HD-containing proteins into lipid bilayers perturbation bilayer energy proteins perturb organization phospholipids enhance exposure hydrophobic core bilayer to extent membrane perturbation depends on amino acid sequence mismatch bilayer thickness HD length protein insertion into LD surfaces no information energy cost type extent perturbation lipids study LD proteins HD-containing partition between bilayer LD droplet interface bilayer (DIB) partitioning proteins peptides bearing HDs AH-containing proteins proteins partition preferentially to LD monolayer surface HD-containing proteins display higher enrichment in monolayer than AH Relocation HD proteins to bilayer unfavorable moving bilayer to monolayer spontaneousfound protein distribution altered by ratio PC PE phospholipids regulating HD-TG contact at LD surface droplet interface bilayer partition coefficient proteins binding monolayer bilayer droplet interface bilayer (DIB DIBs two micrometric buffer-in-oil droplets phospholipid monolayer oil phase trioctanoate triglyceride similar energy triolein8 cellular neutral lipid Contact droplets induces formation bilayer monolayers DIBs mimic ER-LD contiguity without curvature interfaces flat at protein scale concavity monolayer surfaces irrelevant curvature phospholipids used dioleoyl phosphatidylethanolamine PE dioleoyl phosphatidylcholine PC Phospholipids added to oil phase recruited to surface aqueous droplets contact generates 5 min equilibrated DIB34 generated with any phospholipids35 non-bilayer phospholipids PE-DIB bilayer made TG molecules level decreased by addition measured thickness hydrophobic region bilayer by capacitance measurements36thickness in PE-DIBs 2.68 nm ~7% above PC/PE (1:1) DIB 2.52 nm (Fig. values comparable to hydrophobic region phospholipid vesicles devoid oil 2.3–2.7 nm18 simulations adding PE to PC bilayer devoid oil bilayer thickness to 10% Fig data indicate thickness DIBs similar to phospholipid bilayer vesicles not affected by oil PC/PE mixture added to oil phase bulk ratio reflects monolayer composition measured surface tension monolayers linear decrease as ratio increased bulk ~2 mN PE 100% to ~0.6 mN 100% supports bulk PC/PE composition reflects monolayer plateau surface tension against PC/PE PC/PE ratio in monolayer DIB bilayer identical measured partitioning Rhodamine-PE (Rh-PE) between DIB monolayers bilayer PE-DIB Rh-PE uniformly distributed PC/PE Rh-PE almost uniformly distributed monolayer bilayer similar PC/PE composition (Fig. 1d Supplementary Figinvestigated lipid distribution in nascent LDs molecular dynamics simulations three systems TG pure DOPC or/DOPE mixtures 80/20 60/40 DOPE DOPC mix ideally distribution homogeneous Fig DOPC slightly enriched monolayer DOPE slightly enriched bilayer differences minor both data confirm PE/PC mixtures ideal even distribution lipids between monolayer bilayer interfaces DIBs recapitulate conditions bilayer containing oil droplet use DIBs study protein partitioning.Monotopic proteins bind to TG-covering monolayersWe screened monolayer-bilayer partitioning of soluble proteins monotopic integral membrane proteins Soluble peptides added to buffer droplets Monotopic membrane proteins added from purified LDs or proteoliposomes (Fig. mixing relocalized proteins to interface between buffer droplet oil phase Phospholipids from liposomes LDs relocalized to new interface total amount less than exogenous phospholipids added interfacial lipid composition buffer-in-oil droplets proteins prepared before adding phospholipids oil phaseTwo droplets DIB protein partition coefficient determined 10 min after contact enrichment level protein bilayer monolayers (Fig. 2Partitioning hydrophobic amphipathic helix proteins to droplet bilayers Formation protein DIBs soluble protein proteoliposome purified LDs added buffer droplet Mixing droplet TG solution relocalization proteins buffer droplet two drops monolayers zip form bilayer proteins relocalize bilayer Protein distribution between bilayer monolayer determined by partition coefficient P protein signal divided protein partitions to bilayer p > 1 monolayer p < 1. Distribution GMAP-210-AH ACSL3 (monotopic membrane protein in DOPE (1:1) DIBs Scale bar 20 μm partition coefficient as box-plots from n = 5 measurements Enrichment parameter DOPC/DOPE (1:1) membranes for AH- HD-) proteins floating bars 2 ≤ n ≤ 8 measurements each proteinHD-containing proteins from LDs Plin 1 1C ACSL3 CG2254 CG9186 Oleosin 1 Caveolin 1 HPos proteoliposomes Syt1 57-421 t-snare Vamp2 AH-containing proteins LDs Plin3 Plin2 Plin1N other AHs soluble Average partition coefficient membrane proteins DOPE (1:1) Results Fig. S2e Box-plots box limits quartiles middle line median whiskers minimum maximum value c d e f point black dot Source data file monotopic membrane proteins tested Plin1 ACSL3 CG2254 CG9186 oleosin 1 Hpos caveolin1 (Fig. 2d helical hairpin hydrophobic amphipathic helix motifs localization LDs37–40 proteins tagged with fluorescent proteins expressed cells loaded with oleate induce LDs LDs purified added DIB system other proteins relocalize DIB interfaces proteins single transmembrane domains could target LD surface prepared proteoliposomes SNARE components notstudied soluble proteins Plin2–3 Plin1 AH-containing domains19,24 lipid packing sensors ArfGap1-AH41 GMAP-210-AH42 (Fig. 2c stronger partitioning to monolayers than bilayer independently PC/PE ratio (Fig. 2e HD-containing proteins higher LD enrichment than AH proteins proteins ER bilayer associate with LDs HD proteins measured partitioning in PC/PE (1:1) relevant composition (DOPE/DOPC/liverPI/cholesterol 5:3:1:1) similar results bilayer localization of AH-containing proteins increased by addition PE partitioning monolayer more favorable (Fig. 2f negative curvature of PE lipid packing defects by highest partition coefficient (close 1) in PE with AH domains of GMAP-210 ArfGap1 lipid (Fig. 2c HD-containing protein on PC/PE less clear AH- HD-containing proteins localized to monolayer interface over HD-proteins strongly partitioned to monolayer barely relocated to bilayerKWALP peptides recap behavior HD difficult explain partitioning trend full-length HD proteins proteins from purified LDs interact with proteins LD proteins bear multiple HDs AHs Oleosin1 Caveolin1 HPos ACSL337,38,40 AH motif adjacent to HD motif partitioning HD domains focused on model peptides KWALP family peptides repeated leucine-alanine motif two tryptophan residues C-terminus three lysine residues N-terminus added glycine linked to rhodamine-B dye peptide tendency towards helical conformation transmembrane partitioning valid model for transmembrane domains proteins Golgi plasma KWALP model for minimal basic hydrophobic sequences in HD domains proteins used KWALP20 16 hydrophobic amino acids length close to ER bilayer thickness 20 hydrophobic amino studied partitioning AH motif from 11-mer repeat PL108 KWALP20 PL108 models for HD AH proteins 3PC/PE ratio modulates partitioning AH HD peptides Helical wheel representation of KWALP20 PL108-AH HeliQuest69Distribution KWALP20 PL108 DOPE DOPC (1:1) DIBs KWALP20 labeled PL108 NBD Scale bar 50 μm partition coefficient plotted peptides box-plots = 4 measurements data black dots FRAP experiment KWALP20 lipid signals mobile monolayer NBD-PE reports Yellow area bleached Scale 50 μm Recovery half-time one-phase association fitting GraphPad upper right box KWALP20 distribution DIBs different PC/PE ratio Line profiles drawn perpendicular bilayer monolayers. thickness line 30-40% bilayer size signals shown box black bilayer signal Scale bar 20 μm Partition coefficient KWALP20 DIBs DOPC/DOPE ratios box-plots Sample size 31 0% PC 14 10% PC 18 20% PC 41 30% PC 16 40% PC 28 50% PC Each data point plotted Relocalization KWALP20 monolayer to bilayer after addition DOPC DOPE DIBs bilayer signal plotted over time Image brightness enhanced bilayer viewing Scale bar 50 μm Source data file prepared DIBs KWALP20 PL108 compare distribution identical conditionsDIB contained PE PL108 partitioned equally between monolayer bilayer (Fig. KWALP20 absent from bilayer PE/PC (1:1) used KWALP20 signal bilayer increased PL108 bilayer concentration decreased observations consistent with behavior HD- AH-containing proteins. 2e peptides partition favorably to monolayer especially KWALP20 PE lipids PL108 partitions evenly KWALP20 barely detectable to bilayer Oleosin1 KWALP recapitulated partitioning full-length HD-proteins 2e investigated driving forces distribution principles enrichment HDs surface.PC/PE ratio regulates partitioning KWALPKWALP20 absent from bilayer in PE folded in monolayer peptide laterally mobile rapid recovery fluorescence following photobleaching. 3c recovery due to in-plane diffusion bulk 10 minutes hydrophobic thickness PE DIB bilayer comparable to peptide length unlikely KWALP20 localization to bilayer prevented by hydrophobic mismatch increased PC/PE ratio concentration in bilayer increased with PC level lower than monolayerIn bilayer KWALP peptide mobile tendency to cluster as PC level increased (Fig. 3f S4f). verify bilayer localization dynamic diffusion not prevented by peptide aggregation followed peptide signal phospholipid composition PE DIB KWALP20 absent added TG solution PC to oil phase droplets (Fig. 3f). recruitment PC to interface droplets contact angle between droplets35 Fig S4e concomitant with increase KWALP20 signal bilayer noticed KWALP clusters after PC recruitment clustering inducible state PL108 opposite trend excluded from bilayer monolayer by PC recruitment Fig S4g). results suggest system lower free energy when peptide at monolayer energy gap between KWALP decreased by PC energy gap from hydrophobic mismatch experiments with longer KWALP version KWALP28 longer bilayer thickness peptide as KWALP20 within resolution limits measurements Fig. S4c results show KWALP20 partitions dynamically between monolayer bilayer preference for monolayer especially in high PE levelscells no regulation ER-to-LD partitioning HDs adsorbed at LD surface free energy minimization.Phospholipid shape defines KWALP changing PC/PE ratio partitioning hypothesized affinity HDs for lipids partitioning HD protein interact with phospholipid acyl chains TG water latter unfavorable interactions accumulation peptides monolayer prepared phospholipid-free buffer-in-TG droplets KWALP20 protein signal interface uniform interface lined by PE PC lined protein formed clusters peptide mobile Similar clustering observed PC added to DIBs AHs protein-protein interactions favorable at PC monolayer interface KWALP for TG than phospholipid acyl chains contact HD with TG determines HD monolayer partitioning.Fig partitioning KWALP altered by phospholipid shape KWALP20 surface distribution in bare TG-buffer droplets covered DOPE DOPC Red arrow highlights peptide clustering yellow square regions enlarged right side Scale bar: 100 μm Quantification KWALP patternuniform green clustering signal in TG 8) + DOPE 47) + DOPC measurements Schematic representation difference phospholipid packing HD-TG contact DOPE DOPC present Increasing DOPC concentration monolayer increases lipid packing decreases contact between HDs TG Distribution KWALP20 in DOPE N-methyl-PE N,N-dimethyl-PE DOPC DIBs incremental curvature between DOPE DOPC KWALP20 labeled with Rh-B Line profiles drawn perpendicular to bilayer monolayers Fig. thickness line 30–40% bilayer size Arrows indicate bilayer signal Scale bar 20 μm Partition coefficient KWALP20 in DIBs box-plots Sample size n = 29 20 for KWALP20 in N-methyl-PE N,N-dimethyl-PE results varying PC/PE ratios. 3e light gray data points indicated Source data file monolayer droplet contiguous bilayer less packed phospholipids bilayer HD peptides phospholipid acyl chains monolayer TG higher affinity HD proteins for TG over phospholipids to monolayersPC/PE ratio probability HDs contact with TG PC and PE differ in acyl chain composition chains average shape PC cylindrical PE conical PC higher phospholipid monolayer packing reduces probability contact HD with TGs increases contact with phospholipid acyl chains HD less efficiently partition to monolayer when PC/PE ratio increased model suggests phospholipid shape modulates HD monolayer-bilayer partitioning used dioleoyl phosphatidic acid (PA), negative spontaneous curvature like PE PA DIB KWALP20 almost excluded from bilayer repeated KWALP20 partitioning experiments in N-methyl-PE N,N-dimethyl-PE phospholipids intermediates between PE and PC N-trimethyl PE Increasing methylation increased KWALP signal in bilayer similar to increasing PC/PE ratio data indicate HDs prefer mixing with TG with membrane phospholipids phospholipid packing less compact in monolayer monolayer thickness half bilayer partitioning toward monolayers favored HD proteins to TGIncreasing monolayer phospholipid packing increases HD-phospholipid interaction HD-TG interaction peptide less discriminates monolayer from TG monolayer accumulation dampened.TG responsible for accumulation HDs in model postulates KWALP accumulates to LD monolayers mixes with TG than membrane phospholipids. altered affinity peptide for phospholipids changing chemical nature oil chose silicone oil changes oil-protein affinity Silicone oil chemically different from TG high surface tension with PE DIBs TG KWALP20 absent from bilayer replacing TG with silicone oil KWALP20 in bilayer phospholipid clusters appeared at monolayer interface enriched in peptide peptide signal weaker at oil-water monolayer interface peptide higher affinity for phospholipids than silicone oil preferentially distributes to phospholipid-rich regions distribution of KWALP HD depends on oil chemistry Distribution KWALP20 in DOPE DIBs labeled with Rh-B Oil phase is TG or silicone oil Blue arrows indicate monolayers yellow bilayer red cluster areasPlot profiles determined yellow lines bilayer signal indicated black arrow Scale bar 20 μm Partition coefficient reported gray TG silicone oil mean ± sd (n = 2 8 measurements cluster enrichment coefficient (red) determined silicone oil shown mean ± SD (n = 8 measurements). data point black dot Source data provided file repeated experiment with VAMP2 SNARE membranes TG VAMP2 absent from bilayer silicone oil in bilayer clustered with phospholipids oil-water interface KWALP20 results suggest HD-containing proteins partition highest molecular affinity more affinity for TG phospholipids enriched sites access TG.KWALP egresses membranes accumulate in model DIB system energy gap favoring higher enrichment HDs monolayers mixing with TG over bilayer phospholipids nascent LD formed bilayer HDs relocalize to nascent LD behavior reported many HD-containing proteins HPos LiveDrop Oleosins27 tested hypothesis.mimic forming LD used droplet-embedded vesicle system giant unilamellar vesicle (GUV) TG droplets between bilayer leaflets8 (Fig. 6a). incorporated KWALP into PC/PE (7/3) GUVs mixing GUVs with peptide (Fig. 6a KWALP-containing GUVs mixed with TG-in-water droplets generate DEVs peptide enriched monolayer side consistent with predictions results DIBs (Fig. 6a used molecular dynamics simulations size curvature effects DIBs generated bilayers 16 or 32 copies KWALP20 peptides incorporated one side DOPC bilayer TG peptides randomly distributed (Fig. TG incorporated bilayer nucleated lens peptides moved surface lens peptide distribution profile peptides remained mobile move bilayer region dynamic no kinetic barrier traps peptides monolayer same result for KWALP28 peptides results consistent with DEV DIB systems.Fig. 6Nascent LDs attractive to monotopic proteinsFormation DEVs-containing KWALP20 labeled Rh-B KWALP20 inserted in GUVs electroformation mixing peptide example KWALP20-interted GUV KWALP20 GUVs mixed with TG-in-buffer droplet generate droplet-embedded vesicles examples resulting DEVs peptide labeled Rh-B KWALP signals on monolayer bilayer depicted blue yellow arrows accumulation at monolayer Image brightness enhanced monolayer viewing Scale bar 2 μm ratio between monolayer bilayer signals plotted mean ± SD (n = 5 measurements). Individual points indicated Source data file molecular dynamics simulations bilayer with 32 KWALP20 peptides parallel orientation TG lens Hydrophobic amino-acids blue charged ones (lysines red average protein density profiles bilayer plane simulation model interactions HD LD monolayer accumulation HD interacts with TG phospholipids water monolayer packing regulates contact tunes reaction constant indicate free energy lower when HD proteins at monolayer HD-containing proteins sense TG accumulate at TG hotspots-lipid interactions in membrane biology protein localization functionality14 protein-phospholipid interactions responsible for localization proteins to organelles membranes TG not membrane bulk lipid support HD proteins higher affinity for TG over phospholipids in membrane (Fig. under LD biogenesis conditions HD-containing proteins recruited to sites TG accumulation ER bilayer Helical hairpins hydrophobic helices transmembrane domains not crossing ER bilayer accumulate to nascent LDs AH-containing proteins lesser extent (Fig emerging nascent LDs in attract HD-containing proteins Controlling stages LD formation critical for defining proteome emerging LDs8 keeping ER homeostasis HD proteins accumulate to LD monolayers not all HD proteins target to LDs counteracting mechanisms HD-protein targeting to emerging LDs5,6 Amino acid composition HD HD-TG affinity ER-to-LD partitioning charged residues in HD hamper HD-TG interaction requires conformational changes interaction opposite sequence motifs optimal affinity with TGHDs perturb lipid bilayers protein-phospholipid interactions clustering proteins predicted by confirmed molecular simulations17 LD monolayers have more phospholipid packing defects TG to water molecules During LD formation relocalization of HDs from bilayer to LD stresses phospholipid packing defects partitioning energy on bilayer monolayer interfaces transient TG clusters in ER bilayer2 attract HD-proteins or HD-proteins trigger clustering TG promoting LD nucleation alleviating ER stress LD formation stimulated by conditions excess nutrients ER stress11 During ER stress formation of LDs stimulated sequester damaged HD proteins macrolipophagy proteins expose hydrophobic sequences adsorbed to LDs LD formation protein quality control ER proteostasis LDs triggering ER bilayer stress alteration of ER phospholipid composition51 PC/PE ratio decreased10 data partitioning of HD-containing proteins between ER LDs decreased PC/PE favors HD targeting retention to LD monolayertuning PC/PE ratio cells shift HD proteins ER to LDs increased PE/PC levels caused dysfunctions N-methyltransferase linked LD formation PE/PC-induced steatosis abnormal ER-to-LD protein trafficking Drosophila cells high PE/PC levels ER-to-LD partitioning HD-proteins data connect protein-lipid interactions membrane biophysics biology lipid metabolism cellular proteostasis findings highlight attractiveness LD surface for HD-containing proteins Accumulating neutral lipids detected by proteins non-selective detection prevented by cells Kacetate MgCl2 sodium phosphate monobasic dibasic choloroform trifluoroethanol Octyl-ß-D-glucopyranoside from Sigma Aldrich DOPC-glycero-3-phosphocholine), DOPE N-methyl-PE-dimethyl-PE liver PI Rhodamine-DOPE NBD-DOPE purchased from Avanti Polar Lipids CAV1-GFP plasmid Sino Biological plasmids YFP-CG2254 YFP-CG9186 DrBeller GFP-Plin 1 2 3 David Savage EGFP-ACSL3 plasmid Joachim GFP-HPos Prof. Albert Cells American Type Culture Collection no contamination mycoplasma proteins preparationRhB-KWALP peptides-ArfGAP1-GMAP-210 NBD-CAV1-AH peptide 2.0 Inc. NBD-PL108 Proteogenix SAS RhB-NS5A Eric Diesis peptides synthesized purified liquid chromatography purity higher 95% HPLC mass confirmed mass spectrometry amino-acid sequences peptides RhB-GKKKLALALALALALALWWA-AmideKWALP28-AmideArfGAP1-AH-210-AH NBD-LFEAVGKIFSNVRINLQKEIPL108-PPEKIASELKDTISTRLRSARNSISVPIASNS5A RhB-AmidePlin proteins CG2254 CG9186 CAV1 HPos NS5A ACSL3 obtained purified lipid dropletspurification Huh7 cells proteins cells 15 cm dishes harvested washed PBS lysed 30 G needle 1 ml buffer 20 mM Tris protease phosphatase inhibitors pH 7.5 isolation 1 ml cell lysates 1 ml 60% sucrose Tris-EDTA buffer protease inhibitors 20 10 0% buffered sucrose 5 ml Ultra-Clear centrifuge tubes centrifuged 16 h 100,000 G 4 °C SW60 rotor Beckman L8-70 centrifuge upper 300 μl fraction collected LD fraction labeled Arf1 generated replaced serine lysine cysteine Arf1- C159S-K181C Arf1--K181C yeast N-myristoyltransferase coexpressed Escherichia coli lysates 35% ammonium sulfate precipitate Arf1 purified DEAE-ion exchange Eluted fractions concentrated spin-column filters 10-kD labeled Cy3-maleimide samples purified gel filtration Superdex 75 column lipid droplets Arabidopsis seeds Dr. Martine MiquelVamp2 Tsnare syntaxin1a SNAP25) Synaptotagmin purified Frédéric Pincet proteins Vamp2 Tsnare Syt1 Octyl-ß-d-glucopyranoside labeled Atto-565 maleimide Free-dye removed gel-filtration Sephadex G25 column purified DOPC/DOPE 1:1 proteoliposomes mixed glass tube chloroform removed argon vacuum one hour lipid film rehydrated Atto565-protein solution 30 minutes diluted 3 times OG concentration dialysis overnight 10 kDa Slide-A-Lyzer dialysis cassette remove OG liposomes buffer 25 mM HEPES pH 7.4 120 mM KCl 1 mM DTT 0.5 mM CaCl2 Syt1) in vitro experiments HKM buffer 50 mM HEPES 120 mM Kacetate 1 mM MgCl2 pH 7.4. KWALP peptides dissolved trifluorethanol 200 μM diluted HKM concentration 10 μMPL108 solubilized 50 μM HKM CAV1-AH 10 μM HKM 0.1% GMAP-210 2 μM 0.1% DMSO 16 mM urea 80 μM ArfGAP1 8 μM HKM 0.1% NS5A diluted HKM concentration 1 μM 10% trifluoroethanol added folding proteins LD (Plin 1 C ACSL3 CG2254 CG9186 Oleosin 1 CAV1 HPos proteoliposomes (Syt1 57–421 t-snare VAMP2) used DTT added 2.5 mM.Phospholipids 0.2% labeled-PE evaporated argon remove chloroform lipid film re-solubilized concentration (0.2% w/w trioctanoate silicone vortexed sonicated 10 min dissolution DIBs buffer-in-oil emulsions 10 μl HKM peptide/protein solution 100 μl trioctanoate emulsion vortexed protein relocalize droplets volumes peptide/protein emulsion lipids oil emulsion hydrophobic coverslip sample equilibrate 10 min observed confocal fluorescence microscopy ×10 ×63 objective size dropletsemulsion poured observation glass droplets adhere form bilayer final lipid concentration oil phase 0.1% interfacial lipid composition determined effect PC KWALP PL108 peptide concentration 50 μM in DOPE DIBs 5 μl DOPC 0.2% trioctanoate 10% CHCl3) added unilamellar vesicles formationPhospholipids (DOPC/DOPE (7:3 or 6:4) chloroform 2.5 mM spread indium tin oxide-coated glass plate lipid film under vacuum 1 h sealed ITO glass plate GUVs grown electroformation sucrose solution (0.1 g mosmol 100 Hz 1.25 V 1.5 to 2 h collected Pasteur pipette Eppendorf® tube stored 4 °C.Droplets embedded vesicles oil-water emulsion 20 μl trioctanoate mixed 100 μl HKM buffer solution sonicated droplets 10 μl 20 μM KWALP peptide solution added to 40 μl GUV solution incubated 20 μl droplets 5 min added KWALP to dried phospholipids KWLAP GUVs used DEV (Fig. 6)DEV/KWALP sample placed on glass coverslip pretreated with 10% BSA washed three times with buffer observed by confocal fluorescence microscopy (LSM 800 oil-immersed ×63 objective.Electrical measurementAqueous droplets in oil blown at micropipettes Ag/AgCl electrodes Axopatch 200B filled with electrolyte buffer Micropipettes from borosilicate capillary 1.0 mm OD×0.50 mm ID×150mm pulled with micropipette puller (Sutter Instrument tip inner diameter 2 μm treated dimethyldichlorosilane solution avoid capillary wetting Micropipettes manipulated through MP225 MP285 micromanipulators 5 min waited monolayer formation micropipettes moved bilayer electrical measurement performed 20 mV voltage step for 300 ms measuring resulting current images of droplets acquired using IDS camera Olympus IX71 inverted microscope 20x objective bilayer area.Thickness capacitance value C obtained exponential transient capacitive current voltage stepthickness bilayer calculated assuming dielectric material using relation\documentclass[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt} =\varepsilon _0.S}}}e=εr.ε0.SC ε0 permittivity vacuum εr dielectric constant material (εr = 2.8)58 S surface bilayer calculated from images ImageJ.Molecular dynamics protein distribution nascent LDs molecular dynamics simulations coarse-grained level MARTINI force field50 (version 2.2). generated system 2016 DOPC lipids 625 triolein (TO) molecules 83,000 water particles size 27 × 27 × 18 nm TO molecules dispersed DOPC bilayer phase-separated form oil lens system simulated 20 μs shape properties change last 10 μs protein-containing systems generated equilibrated lens system inserting 16 or 32 copies transmembrane peptides bilayer region used 2 peptide sequences KWALP20 and KWALP28 parallel anti-parallel built 8 protein-containing lens systemsMD simulations 20 μs last 10 μs analysis distribution phospholipids LDs large nascent LDs 18144 phospholipids 7500 TO molecules 1.9 million water particles system size 78 × × 40 nm 3 simulations 100% DOPC:DOPE 80:20 60:40 30 μs last 20 μs analysis MD simulations GROMACS (v2016.4) leap-frog integrator time step 20 fs Non-bonded interactions calculated Verlet neighborlist algorithm buffer tolerance 10−6 kJ mol−1 ps−1 cutoff 1.1 nm electrostatic interactions shifted zero nm long-range electrostatics reaction-field method (εR = 15 εRF = Lennard-Jones potential shifted zero cutoff stochastic velocity rescaling thermostat62 time constant 1 ps temperature membrane solvent 300 K Pressure controlled Parrinello–Rahman barostat63 reference pressure 1 bar compressibility 4×10−4 bar−1 time constant 12 ps.Analysis protein density in-house center largest TO cluster simulation TO content bilayer DOPC:DOPE contact fraction mixing phospholipid distribution between bilayer monolayer in-house softwareeffect DOPE lipids membrane thickness-atom simulations DOPC:DOPE 1:1 mixtures CHARMM36 force TIP3P water boxes 100 lipids 5000 water molecules time 400 ns GROMACS 2020 software leap-frog integrator time step 2 fs Non-bonded interactions calculated Verlet neighborlist algorithm buffer 10−6 kJ mol−1 cutoff 1.2 nm PME long-range electrostatics temperature maintained K stochastic velocity rescaling time constant 1 ps Pressure controlled semi Parrinello–Rahman barostat63 reference pressure 1 bar compressibility 4.5 × 10−4 bar−1 time constant 10 ps Analysis mass density 300 ns GROMACS tools dichroismCD spectra recorded 185–250 nm 0.2 nm intervals 20 nm min−1 scan speed Jasco 815 spectropolarimeter Temperature 20 °C measurements 1 mm path length quartz cells Hellma GmbH 200 Experiments TFE 10 mM phosphate buffer pH 7.4 DOPC small unilamellar vesicles TG emulsions DOPC chloroform glass tube removed argonlipid film dried under vacuum 30 min rehydrated with phosphate buffer vortexed lipid solution sonicated reduce size vesicles oil-in-buffer emulsions 30 μl trioctanoate 0.2% phospholipids with 500 μl buffer sonicating KWALP concentration 20 μM DOPC liposomes 1 mM Data collected processed Spectra Manager® reported as molar ellipticity per residue (degree dmol−1 cm2 residue−1)\documentclass[12pt{amsmath-69pt\theta{molar = \theta l N}}[θ]molar=100×θc×l×Nwhere θ ellipticity degrees c peptide concentration mol l−1 l cell path-length cm N number of residues peptide peptide secondary structure content analysis of CD spectra using CDPro software68.StatisticsData analysis representation in Prism 7 (GraphPad Software sample size errors bars statistical tests in figure legend information Nature Research Reporting SummarySupplementary Summary | 49 | 1.13795 |
10.1038/s41467-020-18643-x | PMC7547083 | The underlying mechanisms regulating the mouse hair cycle remain poorly understood. Here, the authors find that Fgf and Wnt signaling pathways interact in the mesenchymal niche of the hair follicle to regulate the molecular clock that dictates the duration of hair growth. | Tissue growth in the adult is an orchestrated process that often requires biological clocks to time stem cell and progenitor activity. Here, we employed the hair follicle, which cycles between growth and regression in a timely-restricted mode, to show that some components of the hair cycle clock reside within the mesenchymal niche of the hair follicle, the dermal papilla (DP), and both Fgf and Wnt signaling pathways interact within the DP to regulate the expression of these components that include Wnt agonists (Rspondins) and antagonists (Dkk2 and Notum). The levels of Wnt agonists and antagonists in the DP are progressively reduced and elevated during the growth phase, respectively. Consequently, Wnt signaling activity in the overlying epithelial progenitor cells decreases, resulting in the induction of the regression phase. Remarkably, DP properties allow Wnt activity in the DP to persist despite the Wnt-inhibiting milieu and consequently synchronize the induction and progression of the regression phase. This study provides insight into the importance of signaling crosstalk in coupling progenitors and their niche to regulate tissue growth. | IntroductionTissue growth and regeneration in the adult often require the activity of specialized stem cell and progenitor populations, but also need to be constrained within a time frame to allow the formation of proper size, shape, and structure1,2. Furthermore, tissue removal may also accompany tissue growth to achieve normal function3. The hair follicle periodically cycles between growth and regression in a timely coordinated mode4–6, providing a powerful model system to explore the molecular mechanisms that time stem cell activity and progenitor behavior.The mature hair follicle during the growth phase (anagen) of the hair cycle is largely composed of concentric layers of differentiated keratinocytes that comprise the outer root sheet (ORS), inner root sheet (IRS) and hair shaft (HS). At the bottom of the follicle, the dermal papilla (DP), a mesenchymal component of the follicle, is embedded in the hair bulb and surrounded by keratinocytes collectively called matrix cells. Matrix keratinocytes in direct contact with the DP behave as progenitor cells of the hair matrix7. Asymmetric divisions renew this progenitor cell compartment and persistently provide vertical flux of transit amplifying progeny that undergo a limited number of divisions and differentiate to form the constituents of the IRS and HS. The persistent addition of these new constituents to the base of the inner layers drives the growth of the hair and its emergence through the skin. At the end of the anagen phase, proliferation in the hair matrix ceases and the regression phase (catagen) ensues. During catagen, apoptosis initiates in the matrix around the DP and spreads from the matrix upwards. Consequently, follicles shorten and regress8,9. Concomitantly, the DP is withdrawn with the regressing epithelial strand to lie beneath the permanent portion of the follicle during quiescence (telogen).Numerous models and theories have been proposed during the last five decades to explain the cyclic nature of the hair follicle10–18, and in spite of intensive research, the components and the molecular mechanisms that underlie the hair cycle remain poorly understood. Remarkably, the duration of anagen within a specific mouse strain is largely constant and strictly maintained from individual to individual and from cycle to cycle, suggesting the existence of a biological clock that dictates the periodicity of the hair cycle. Previous studies have shown that Fgf signaling may play an important role in regulating the hair cycle clock19–23. Fgf5 knockout mice display abnormally long hairs as a result of prolonged anagen, while Fgf5 administration during mid anagen results in premature induction of catagen21–23. Expression analysis of Fgf5 in the hair follicle revealed that Fgf5 is expressed during anagen in the lower ORS all the way to the base of the bulb region including matrix cells adjacent to the bottom of the DP21. As catagen induction starts when matrix cells undergo apoptosis in the bulb region8,9, Fgf5 may act directly on matrix cells to inhibit their proliferation or/and induce apoptosis. Alternatively, Fgf5 may affect the matrix cells indirectly by signaling to the DP21,24. As the proximity of Fgf5-expressing cells to the DP is limited to only few cells of the lower part of the DP, the direct mechanism by which Fgf5 acts on matrix cells has been considered more likely. However, the mechanism by which Fgf5 acts to regulate the transition from anagen to catagen remains unknown.In addition to Fgf signaling, canonical Wnt signaling pathway may also play a role in regulating the hair cycle clock. Ablation of beta-catenin in the matrix or DP during mid anagen results in premature induction of catagen25,26. While this clearly illustrates that Wnt signaling activity is required in both the matrix and DP to maintain the anagen phase, it remains unclear whether activation of the Wnt signaling pathway in both compartments converges into a single biological process that regulates the duration of anagen, or Wnt signaling transduction in the matrix and DP independently retain the anagen-maintaining activity of each compartment. Interestingly, forced expression of the constitutive active form of beta-catenin specifically in the DP does not alter the hair cycle27. This suggests that while Wnt signaling activity in the DP is required to sustain anagen, it is not sufficient to counteract the catagen-inducing signals. In the current study, we show that Fgf and Wnt signaling pathways interact in the DP to regulate the hair cycle clock by orchestrating the expression of Wnt agonists (Rspondins) and antagonists (Dkk2 and Notum).ResultsFgf signaling in the DP modulates components of the hair cycle clock that regulate anagen durationTo unravel the role of Fgf signaling in the DP, we first determined which Fgf receptors mediate Fgf signaling in the DP. In situ hybridization was employed to assess the expression of all four Fgfr receptors during the growth phase. All four receptors were readily detected in the epithelial part of the follicle as previously reported28. However, while Fgfr1 and Fgfr2 are abundantly expressed in the DP throughout the anagen and catagen phase, only very low levels of Fgfr3 and Fgfr4 are detected in this compartment (Fig. 1a, b, Supplementary Fig. 1a–d), suggesting that Fgfr1 and Fgfr2 are the predominant transducers of Fgf signaling in the DP. Subsequently, Fgfr1 and Fgfr2 were specifically ablated in the DP during the hair cycle by crossing conditional knockout mouse lines of Fgfr1 and Fgfr2 (refs. 29,30) with the DP-specific Cor-cre mouse line that expresses the cre recombinase in the DP postnatally25. Three genotypically distinct types of mutants were generated; single Fgfr1, single Fgfr2, and double Fgfr1/2 mutants (designated dMF1/2). While the hair coat of both Fgfr1 and Fgfr2 single mutants appears normal (Supplementary Fig. 2a, b), the double Fgfr1/2 mutant exhibits extremely long hairs (Fig. 2a, b).Fig. 1Fgfr1 and Fgfr2 are expressed in the DP throughout the anagen and catagen phase.In situ hybridizations for Fgfr1 (a) and Fgfr2 (b) are shown. Note that the transcripts (blue) for both receptors become restricted to the DP during catagen (P17). All images were taken within the same scale and chronologically ordered clockwise. Unpigmented FVB mice were used. n > 3 mice per time point. Scale bar, 50 μm.Fig. 2Ablation of Fgf signaling in the DP results in extended anagen.In the upper panels, pictures of a wild type control (a) and a Fgfr1/2 double mutant (b), taken at postnatal day 50 (P50), are shown (n > 10 mice per genotype). The panels below depict HE staining of dorsal skin sections derived from the posterior region of wild-type and mutant mice at different stages of the hair cycle (on the right side, P indicates postnatal day). Note the prolonged anagen in the double mutant. At P19, a wild-type skin section (a) is shown to illustrate mid-to-late catagen, while P24 and P26 are examples of Fgfr1/2 double mutant mice (b) that are in the transition from anagen to catagen. By contrast, wild-type follicles at P26 (a) have already completed regeneration and are in the mid anagen of the second cycle. n > 3 mice per genotype per stage. Scale bar, 200 μm.To explore the underlying mechanism of the extended hair length, back skins were harvested from wild-type control and dMF1/2 double mutant mice at different time points during the first and second hair cycles. As the hair cycle gradiently progresses along the anterior–posterior axis, dorsal skins were collected from the posterior region of this axis. Skin sections were used for HE staining to illustrate follicle morphology (Fig. 2a, b). In wild-type mice, catagen is induced around postnatal day 18 (P18), telogen commences at P20, and a new hair cycle is initiated around P22 (Fig. 2a, Supplementary Fig. 3a–e). By contrast, catagen in double mutant mice is initiated between P24 to P26, 6–8 days later than wild type (Fig. 2b). Of note, while catagen induction in wild-type mice in this particular genetic background strictly occurs at P18, catagen induction in the double mutant substantially varies; even at P26, mutant mice are occasionally found in anagen (Fig. 2b). To test whether this extension in anagen is restricted only to the first cycle or also occurs in subsequent cycles, the anagen duration of the second cycle was evaluated. Note that the extended anagen of the first cycle and the large variance in catagen induction of the mutant preclude the direct comparison between control and mutant littermates and thus complicate the analysis. Therefore, the duration of anagen in the second cycle was individually estimated by exploiting the active process of hair pigmentation that occurs from Anagen III to catagen induction (Supplementary Fig. 4). During this period, the skin appears dark, and thus the duration of this period can be visually measured for each individual mouse. As this time frame constitutes most of the anagen phase, its measurement provides a good estimation for the length of anagen. Consistently, the anagen phase is dramatically extended in the absence of Fgf signaling in the DP also during the second cycle (Supplementary Fig. 4). Together these data clearly suggest that some key components of the hair cycle clock that regulate the duration of anagen reside in the DP, and Fgf signaling pathway in the DP modulates the expression and/or activity of these components.Profiling the DP transcriptome reveals that two predominant expression programs operate during the growth phaseWe hypothesized that at least some of the components that constitute the hair cycle clock are progressively accumulated or degraded in the DP during the anagen phase, and once these molecules surpass a certain threshold, catagen is induced. To test this hypothesis, gene expression profiling of DP cells from control and double mutant mice at multiple time points during anagen of the first cycle was performed by RNA-seq. To isolate DP cells from control and dMF1/2 mutant mice, DP cells were endogenously labeled with YFP by activating the cre-dependent YFP-reporter allele31, designed to be included in both control and dMF1/2 mutant mice (Supplementary Fig. 3a–e). Following the dissociation of skin into a single-cell suspension, DP cells were FACS sorted from mice at P10, P12, P14, P16, and P18 (Supplementary Fig. 3f). This spans successive time points during anagen (P10–P16) in both wild-type and mutant mice (Supplementary Fig. 3a–d) and includes a time point during the transition from anagen to catagen (P18) in wild-type mice (Supplementary Fig. 3e). Total RNA was purified from sorted cells and used to prepare sequencing libraries (Supplementary Fig. 3g).Note that anagen of the first cycle is often considered late follicle morphogenesis, and anagen of subsequent cycles is thought to be the genuine anagen of the hair cycle. While follicle morphogenesis and the anagen phase of the hair cycle were detailedly described and anatomically classified into multiple stages (Stage1–8 and AnaI-VI, respectively)32,33, most of these stages refer to early steps of follicle formation (Stage1–7)32 or the very early events of anagen during regeneration (AnaI-III) and immediately post regeneration (AnaIV–V)33. The last stage of both follicle morphogenesis and the anagen of the hair cycle (Stage8 and AnaVI, respectively) begins when the tip of the HS emerges through the epidermis and lasts most of the growth (anagen) phase of the first and second hair cycles respectively. Furthermore, follicle morphology remains largely unaltered throughout this last and long stage of hair growth. Therefore, Stage8 and AnaVI are often considered equivalent and “monotonous”. Our hypothesis suggests that this “monotonous” stage is molecularly dynamic, and to simplify the analysis, this monotonous stage of anagen during the first cycle, which is the focus of our analysis, was chronologically subclassified into three time frames; early-to-mid anagen (P10–P12), mid-to-late anagen (P14–P16), and late anagen to early catagen (P16–P18).Principle component analysis (Fig. 3a) revealed that the DP transcriptome during early-to-mid anagen (P10–P12) remains relatively unaltered regardless of the genotype. By contrast, the DP transcriptome of wild-type mice during mid anagen to catagen induction (P14–P18) is highly dynamic and progressively undergoes substantial changes. The DP transcriptome of mutant mice during P14–P18 also evolves but is highly distinctive than control. To test the anagen progression hypothesis, we further explored the expression behavior of DP cells during anagen by focusing on genes found to be differentially expressed between control and mutant mice at P16 and analyzing their expression pattern from P10. The focus on this selected set of genes relies on the hypothesis that genes involved in the regulation of catagen induction should be differentially expressed between controls and mutants during late anagen before the initiation of catagen in the control. Samples were chronologically ordered along a timescale and presented as a heatmap for control and mutant mice separately (Fig. 3b). This clearly illustrates the presence of two predominant expression programs in control mice, opposing in their behavior and dynamics. The transcript levels of one program during mid anagen start elevated and progressively decrease during late anagen until they stabilize on certain low levels. Conversely, the expression levels of the other program begin low and gradually incline to its maximum at the transition between anagen and catagen. By contrast, while the same two transcriptional programs do occur in the mutant, the rate of the transcriptional change in both programs is substantially slower. Together, these data are in line with our hypothesis and suggest that the part of the hair cycle clock that orchestrates anagen duration is composed of both accumulating and degrading components.Fig. 3Abrogation of Fgf signaling in the DP results in altered expression of Wnt agonists and antagonists.a PCA analysis. Each colored triangle or dot represents the DP transcriptome of a single control or mutant mouse, respectively. Black and red circles demarcate control and mutant mice at a given time point, respectively. n = 3 mice per genotype per time point. b Heatmap analysis. Two classes of genes were revealed based on their expression pattern: genes whose expression during anagen gradually accumulate in control mice while remains low in the mutant and genes whose expression is progressively reduced in the control while persists at higher levels in mutant mice. c Graphic representation of the sequence reads of the RNA-seq analysis for Wnt agonists (Rspo1–4) and antagonists (Notum and Dkk2). Note the different scales of the reads. Data are mean ± SEM. Two-sided modified t-test adjusted for multiple comparisons using the Benjamin–Hochberg correction was employed. *1 Padj = 0.001, *2 Padj = 7.14E − 11, *3 Padj = 3.26E − 08, *4 Padj = 3.18E − 57, *5 Padj = 3.48E − 24, *6 Padj = 4.91E − 26, *7 Padj = 4.05E − 38, *8 Padj = 0.0003, *9 Padj = 0.02, *10 Padj = 3.2E − 06, *11 Padj = 4.14E − 07. n = 3 mice per genotype per time point. d, e In situ hybridization of Rspo3 and Dkk2 at P10 and P16 is shown to illustrate the specific and dynamic expression of these genes in the DP and their altered expression in the mutant. RNA transcripts are in dark blue and pigmented hair shafts in black. n > 5 mice per genotype per time point. Scale bar, 25 μm.The expression of Wnt agonists and antagonists in the DP are progressively altered during anagen in a cooperative mannerKEGG and GO analysis identified numerous signaling pathways that are significantly altered in the mutant DP. Wnt signaling was amongst these pathways. Remarkably, agonists and antagonists of the pathway are combinedly altered in a cooperative mode (Fig. 3c). In wild-type mice, all four Rspondins, known to play an important role in augmenting Wnt signaling34, progressively decrease during mid-to-late anagen, while the antagonists Notum and Dkk2 gradually accumulate. By contrast, the transcriptional levels of these genes in the mutant initially remain closer to their initial baseline and subsequently are altered in a similar manner to the dynamics of the wild type but at a slower pace. The expression pattern and dynamics of Rspondins, Dkk2 and Notum were further explored in skin sections by in situ hybridization (Fig. 3d, e and Supplementary Fig. 5). All four Rspondins are either predominantly expressed in the DP with substantially lower levels in the matrix (Rspo1, Rspo2, and Rspo4; Supplementary Fig. 5a) or their expression is restricted only to the DP (Rspo3; Fig. 3d), suggesting the DP is the major source of Rspondins in the bulb region. Note that while the transcripts of all four Rspondins are readily observed in the DP of P10 mice regardless of the genotype, expression of Rspo2 and Rspo3 in the DP of wild-type mice at P16 is reduced beneath the detection levels (Fig. 3d and Supplementary Fig. 5a). Expression of Rspo1 and Rspo4 in wild-type mice at P16 is still sufficiently high to be readily detected by in situ hybridization at this stage but become undetectable during catagen (Supplementary Fig. 5a, b). In contrast to wild-type mice, the transcripts of all four Rspondins are abundantly detected in the DP of P16 dMF1/2 mutant mice, consistent with the RNA-seq analysis.DP-confined expression is also observed for Notum and Dkk2 during late anagen (Fig. 3e, Supplementary Fig. 5c). At P10, the expression levels of Dkk2 and Notum in both wild-type and dMF1/2 mutant mice are too low to be detected by in situ hybridization. At P16, however, the expression of Dkk2 and Notum in wild-type mice dramatically increases to the levels that are readily noticed by in situ hybridization. These high levels of Dkk2 and Notum in the wild-type DP persist through early catagen and decline to undetectable levels during the end of catagen or mid catagen, respectively (Supplementary Figs. 6a and 7a). By contrast, the expression of Dkk2 and Notum in the mutant at P16 remains beneath the detection levels (Fig. 3e and Supplementary Fig. 5c).The persistent expression of Dkk2 and Notum in wild-type DP during catagen suggests that both play an instructive role in regulating the normal execution of catagen and its progression. Furthermore, their DP expression at P16 may coincide with but not necessarily reflect the regulation of catagen induction. To precisely map whether the expression of Dkk2 and Notum in the DP precedes catagen initiation, both proliferation and apoptosis were evaluated at P16 (Supplementary Fig. 8). During catagen induction, proliferation in matrix cells ceases and subsequently programed cell death is initiated within the matrix in close proximity to the DP8. Immunostaining for Ki67 of skin sections from control and dMF1/2 mutant mice at P16 revealed that matrix cells are highly proliferative regardless of the genotype. Furthermore, Tunel staining failed to detect apoptotic cells within the matrix, clearly indicating that wild-type follicles at P16 represent genuine late anagen. Together these data demonstrate that expression of Dkk2 and Notum in the DP is normally elevated prior to catagen induction and thus suggest that both may participate in regulating catagen initiation.Persistent expression of Rspo3 in the DP of the dMF1/2 mutant contributes to the extended anagenThe expression pattern and dynamics of the Wnt agonists and antagonists in the DP suggests that Wnt signaling activity in the bulb of wild-type mice decreases during mid-to-late anagen, and this reduction promotes catagen induction. This hypothesis predicts that abrogation of Rspondins specifically in the DP should result in premature induction of catagen. However, the co-expression of all four Rspondins predominantly in the DP is likely to complicate the analysis of a single ablation as a result of functional redundancy or accumulative effects that are proportional to the levels of each Rspondin. Consistently, single ablation of Rspo3 in the DP does not result in a measurable premature induction of catagen during the first and the second cycle. Note however that based on our RNA-seq and in situ analyses, the most significant changes in gene expression between control and dMF1/2 mutant mice were observed for Rspo2 and Rspo3, and thus, their persistent expression in the DP of the dMF1/2 mutant is likely to predominantly contribute to the extended anagen. To explore this proposition, a triple mutant of Rspo3, Fgfr1 and Fgfr2 (designated tM/F1/2/R3) was generated and the duration of anagen during the first and second cycle was evaluated. While the prolonged anagen of the double dMF1/2 mutant is comparable to that of the triple tMR3/F1/F2 mutant during the first cycle, anagen duration of the triple mutant during the second cycle shortens relatively to the double mutant but still longer than controls (Supplementary Fig. 4). This suggests that (1) persistent expression of Rspo3 in the absence of Fgfr1 and Fgfr2 in the DP contributes to the extended anagen and (2) Rspondin expression in the DP is required to regulate the duration of anagen.The Wnt agonists and antagonists from the DP predominantly act on matrix cells despite their autocrine potentialTo explore the proposition that Wnt signaling activity in the bulb is normally reduced during late anagen, the Axin2-lacZ Wnt-reporter allele35 was introduced in both control and dMF1/2 mutant mice. Remarkably, the dynamics of Wnt activity in the matrix differ substantially than those in the DP during anagen of wild-type mice. Wnt signaling activity in wild-type matrix is reduced at P16 as compared to early anagen of wild-type mice at P10, while Wnt activity in wild-type DP during late anagen persists and even intensifies (Fig. 4a, b, upper panels). Furthermore, the reduction of Wnt activity in wild-type matrix during mid-to-late anagen occurs progressively (Supplementary Fig. 9). While the intensification of Wnt activity in the DP also occurs in the dMF1/2 mutant, Wnt activity in the matrix of mutant mice at P16 remains high and comparable to P10 (Fig. 4a, b, lower panels). Immunostaining for beta-catenin at P16 revealed activated beta-catenin localized to the nuclei of DP cells in both wild-type and mutant mice (Fig. 4c, d), corroborating the persistent activity of Wnt signaling specifically in the DP during late anagen. Together these data suggest that DP cells are resistant to the activity of Rspondins, Dkk2 and Notum, and matrix cells are the cellular target for these agonists and antagonists.Fig. 4Wnt signaling activity during mid-to-late anagen is progressively decreased only in the matrix to promote catagen induction.a X-Gal staining in skin sections from wild-type (upper panel) and Fgfr1/2 double mutant (lower panel) mice shows LacZ expression at P10. Note that Wnt signaling activity is readily detected in both the matrix and DP regardless of the genotype. Also note the comparable signal between the DP and the matrix. n = 3 mice per genotype. b X-Gal staining during late anagen (P16) revealed that Wnt activity in the wild type (upper panel) is diminished in the matrix but persists in the DP. By contrast, Wnt activity in the mutant (lower panel) persists in both the matrix and DP. Note however the signal in the DP is higher than the matrix. The dashed black line demarcates the boundaries of the follicle. n = 5 mice per genotype. c, d Immunostaining for beta-catenin during late anagen is shown to illustrate the presence of activated beta-catenin in the nuclei of DP cells in both wild-type (c) and mutant (d) littermates. For each genotype, the same follicle is shown twice: on the left, only beta-catenin fluorescent image is presented, and on the right, a merge of beta-catenin and nuclear staining (DAPI) is displayed. n = 6 mice per genotype. e Wild-type skin sections from P10 mice were used to immunostain all the components of the machinery required to transduce Rspondin activity. Nuclei are in blue (DAPI). Note that none of these components is detected in the DP. n > 3 mice. Scale bar in (a, b), 50 μm. Scale bar in (c–d), 25 μm.Note that while Notum is a secreted enzyme that hydrolyzes the palmiteoylate adducts from Wnt ligands and thus inactivates the ability of Wnts to bind Fzd receptors36, Rspondins and Dkk2 activities require a whole molecular machinery within the target cells. We hypothesized that the confined action of these Wnt agonists and antagonists on matrix cells is predominantly achieved by restricting the expression of the molecular machinery, required to mediate the activity of Rspondins and Dkk2, to matrix cells. To explore this hypothesis, the expression pattern of the molecular machinery required to mediate the activity of Rspondins and Dkk2 was assessed in wild-type mice (Fig. 4e, Supplementary Figs. 10 and 11). Znrf3 and Rnf43 are transmembrane E3 ubiquitin ligases that remove Wnt receptors from the membrane by targeting the Fzd receptors for degradation37,38. Rspondins bind both the E3 ubiquitin ligase and a member of Lgr family (Lgr4, Lgr5, and Lgr6) to form a ternary structure that abolishes the ubiquitination activity of the E3 ligase and thus augments Wnt activity by allowing surface accumulation of Fzd receptors. Immunostaining and in situ hybridization for Lgr4, Lgr5, Lgr6, Znrf3, and Rnf43 revealed that all are abundantly expressed in the epithelial part of the follicle including the matrix but are not detected in the DP (Fig. 4e, Supplementary Fig. 10). Dkk2 inhibits Wnt signaling by binding to both the Wnt co-receptor Lrp5/6 and the transmembrane receptor Krm1/2, which results in internalization of Lrp5/6 and consequently reduction in the ability of the cell to respond to Wnt ligands39. In situ hybridization for both Krm1 and Krm2 revealed that both are undetectable in the DP but highly abundant in the epithelial part of the follicle (Supplementary Fig. 11). Of note, while Krm1 is widely expressed in the follicular epithelium, Krm2 is asymmetrically localized to a small group of matrix cells adjacent to the DP. Together these data suggest that only matrix cells are appropriately equipped to respond to the activity of Rspondins and Dkk2 and provide molecular insight to the reduction in Wnt signaling activity only in the matrix during late anagen.The persistent expression of Dkk2 and Notum during catagen suggests that inhibition of Wnt activity within the epithelial part of the follicle may be required also for catagen progression, and the apparent resistance of the DP to the inhibition activity of these Wnt antagonists suggests that Wnt signaling activity in the DP is retained during catagen. The Axin2-lacZ Wnt-reporter allele35 was further used to follow Wnt activity during catagen. Consistently, Wnt activity within the lower part of the regressing follicle was observed only within the DP and is reduced during mid-to-late catagen (Fig. 5a–c).Fig. 5Wnt signaling activity during catagen.LacZ staining of dorsal skin sections from adjacent regions along the anterior–posterior axis of a wild-type mouse at P18 are shown. a, a′ Posterior dorsal region is displayed to illustrate the high levels of Wnt activity in the DP and the lack of Wnt activity in the epithelial part of the follicle adjacent to the DP during early catagen. In (a′), a higher magnification of the field demarcated by the dashed rectangle in (a) is presented. n = 3 mice. b Middle dorsal region is shown to demonstrate Wnt activity during early-to-mid catagen. n = 3 mice. c, c′ Anterior region is depicted to show Wnt activity during mid catagen. In (c′), a higher magnification of the field outlined by the dashed rectangle in (c) is shown. n = 3 mice. Scale bar, 50 μm.Wnt signaling in the DP is required for Wnt activity in the matrix and is stimulated by Wnt ligands from the matrixThe reduction of Wnt signaling in the matrix during late anagen (Fig. 4) is consistent with the observation that ablation of beta-catenin in the matrix results in premature induction of catagen26 and suggests that this reduction is the physiological mechanism to induce catagen. However, ablation of beta-catenin in the DP also results in premature induction of catagen25. We hypothesized that Wnt signaling in the matrix directly maintains the anagen phase while Wnt activity in the DP indirectly regulates the duration of anagen by modulating Wnt signaling in the matrix. To explore this hypothesis, the gene for beta-catenin (Ctnnb1) was specifically ablated in the DP by crossing the Cor-cre line with a mouse line that harbors a floxed allele of beta-catenin as previously described25,27. In line with these prior reports, catagen in the Ctnnb1 mutant is prematurely induced at P12, 4 days earlier than wild type (Supplementary Fig. 12a–e). Note that catagen induction in wild-type mice of this genetic background occurs at P16 (ref. 25). Immunostaining for beta-catenin at P10 when both wild-type and Ctnnb1 mutant follicles are in anagen confirmed the efficient deletion of beta-catenin specifically in the DP of the Ctnnb1 mutant (Fig. 6a, b). Furthermore, in addition to its membrane localization, beta-catenin in wild-type mice was readily detected also in the nuclei of matrix cells, a clear indication of Wnt signaling activity in these cells. Intriguingly, nuclear localization of beta-catenin in the matrix of Ctnnb1 mutant mice is reduced, suggesting that Wnt signaling activity in the matrix is decreased as a consequence of beta-catenin ablation in the DP. The Axin2-LacZ Wnt-reporter allele was further used to corroborate that Wnt signaling activity is abolished in both the DP and the matrix when beta-catenin was ablated in the DP (Fig. 6c) and supported that Wnt activity in the matrix depends on Wnt activity in the DP.Fig. 6Ablation of beta-catenin specifically in the DP results in reduction of Wnt signaling activity in the matrix.a Immunostaining for beta-catenin during mid anagen (P10) of a wild-type follicle is displayed. Nuclei are in blue (DAPI). A higher magnification of the field outlined with the dashed white square in the left panel is presented on the right panel. Note that in addition to the membrane localization of beta-catenin in matrix cells, activated beta-catenin is readily detected also in the nuclei of these cells (white arrowheads). n = 3 mice. b beta-catenin mutant mice were used to immunostain for beta-catenin at P10 to illustrate the efficient deletion of beta-catenin in the DP and the dramatic reduction of nuclear beta-catenin in matrix cells (white arrowheads). A higher magnification of the field outlined with the dashed white square in the left panel is shown on the right panel. n = 3 mice. c X-Gal staining in skin sections from P10 wild-type (left panel) and mutant (right panel) littermates demonstrates that Wnt activity in the mutant is abolished in both the DP and matrix. The internal dashed black line outlines the DP and the external black dashed line demarcates the follicle boundaries. n = 3 mice per genotype. d–g In situ hybridization for Rspondins on skin sections from P10 wild-type and mutant littermates illustrates the absence of Rspondin transcripts in both the DP and matrix of the mutant. n = 3 mice per genotype. Scale bar in (a, b), 10 μm. Scale bar in (c–g), 25 μm.Previous reports have shown that during mid-to-late anagen Wnt ligands are expressed in the matrix and are not detected in the DP40–42, suggesting the matrix is the only functional source of Wnt ligands in the bulb region that autocrinally and paracrinally activate Wnt signaling in the matrix and DP, respectively. To test this hypothesis and to exclude the possibility of undetectable but functional expression of Wnt ligands in the DP during late anagen, the ability of DP cells to secrete Wnt ligands was curtailed by ablation of the Wls gene specifically in the DP. For this, a mouse line that harbors a floxed allele of Wls43 was crossed with the Cor-cre mouse line. Such genetic manipulation neither altered the first hair cycle nor affected the HS, corroborating that the matrix is the predominant functional source of Wnt ligands in the bulb region.Wnt signaling activity in the DP activates the expression of Rspondins and is essential for the extended anagen in the dMF1/2 mutantPrevious reports demonstrated that Rspondins in some settings are transcriptional targets of Wnt signaling44. This suggests that Wnt activity in the DP regulates Wnt signaling in the matrix by activating the expression of Rspondins in the DP. Indeed, in situ hybridization of skin sections from P10 wild-type and Ctnnb1 mutant littermates unveiled not only that expression of all Rspondins in the DP is abolished, but also the low expression of Rspondins in the matrix is diminished in Ctnnb1 mutant mice (Fig. 6d–g). This is consistent with Rspondins being target genes of Wnt signaling and with the dramatic reduction in Wnt activity in the matrix when beta-catenin is ablated in the DP. Furthermore, this suggests that the extended anagen, observed when Fgfr1 and Fgfr2 were ablated in the DP, requires the presence of active Wnt signaling in the DP. To test this proposition, a triple mutant of Ctnnb1, Fgfr1, and Fgfr2 (designated tMF1/2/b) was generated. Consistently, follicles within the triple mutant enter prematurely to catagen similar to a single Ctnnb1 mutant (Fig. 7a–f), corroborating the functional interaction between Fgf and Wnt signaling pathways in the DP.Fig. 7Concomitant ablation of Fgfr1, Fgfr2, and beta-catenin specifically in the DP results in premature induction of catagen.a–d HE staining of skin sections from control, dMF1/2, Ctnnb1, and tMF1/2/b mice at P16 is shown. Note the unsynchronized nature of catagen in both Ctnnb1 and tMF1/2/b mutant mice. n > 3 mice per genotype. e Tunel staining of P13 skin sections from control, dMF1/2 double mutant, Ctnnb1 mutant, and tMF1/2/b triple mutant is shown to illustrate the normal apoptosis around the DP during the premature catagen. Pcad immunostaining was also included to outline the DP. Nuclei are in blue (DAPI). n = 3 mice per genotype. f Higher magnifications of the fields surrounding the DP of Ctnnb1 and tMF1/2/b mutant follicles in e are displayed. Scale bar in (a–d), 200 μm. Scale bar in (e, f), 25 μm.Wnt signaling activity in the DP activates the expression of Dkk2 and Notum in a timely dependent mannerThe absence of Fgf signaling in the DP of the dMF1/2 mutant at P16 prevents the elevation of both Dkk2 and Notum, suggesting that Dkk2 and Notum are transcriptionally activated by Fgf signaling in the DP. However, Dkk2 and Notum were previously shown to be target genes of Wnt signaling as part of a negative feedback loop45–47. To explore whether Dkk2 and Notum are target genes of Wnt signaling in the DP, in situ hybridization was performed to compare their expression between wild-type and Ctnnb1 mutant littermates. We first tested the hypothesis that Wnt activity in the DP suppresses the expression of Dkk2 and Notum, while Fgf signaling in the DP activates the expression of Dkk2 and Notum by derepressing the activity of Wnt signaling. This predicts dramatic elevation of Dkk2 and Notum levels in the DP of Ctnnb1 mutant. However, at P10, both Dkk2 and Notum are undetectable regardless of the genotype (Supplementary Fig. 13a, b), refuting the hypothesis that Wnt activity in the DP suppresses the expression of Dkk2 and Notum. At P12, when catagen is induced in the Ctnnb1 mutant, Dkk2 and Notum remain undetectable, and this lack of expression persists throughout the premature catagen phase of the Ctnnb1 mutant (Supplementary Fig. 13a, b), in contrast to the normal expression of these genes and the restricted activity of Wnt signaling in the DP during catagen of wild-type mice (Fig. 5, Supplementary Figs. 6a and 7a). This suggests that Dkk2 and Notum are transcriptionally activated in the DP by Wnt signaling. Of note, this activation occurs only during late anagen despite the presence of Wnt activity at earlier times. The latter and the observation that Wnt activity in the DP is intensified during late anagen suggest that activation of Dkk2 and Notum requires high levels of Wnt signaling activity, or alternatively, a longer period of Wnt signaling reception is required to activate the expression of these genes. Furthermore, together these data support that (1) while both Dkk2 and Notum may participate in regulating the initiation of catagen, they are not required to induce catagen; and (2) the reduction of Rspondin levels during mid-to-late anagen is the preponderant driving force that promotes and induces catagen.Catagen induction and progression in the dMF1/2 mutant are apparently normalThe dMF1/2 mutant follicles remain longer in anagen but eventually enter catagen. To explore whether the molecular clock that induces catagen in wild-type mice also operates in the double mutant and therefore catagen initiation and progression in the double mutant are normal but only delayed, in situ hybridization for Rspondins, Dkk2 and Notum was performed during the late anagen and catagen of the double mutant. At P22, the expression of Rspo2, Rspo3, and Rspo4 in mutant DP is almost completely abolished, and this extinct expression persists during the catagen phase of the mutant (Supplementary Fig. 14a). The transcript levels of Rspo1 in the mutant remain high and are readily detected at P22 but become undetectable during catagen. Furthermore, immunostaining for Ki67 and Tunel labeling at P22 revealed that mutant follicles at this stage still sustain the proliferative state of the matrix, demonstrating that P22 in the double mutant is a genuinely late anagen stage (Supplementary Fig. 8). Together these data suggest that reduction in Rspondin expression in the DP during mid-to-late anagen is the driving force that induces catagen also in the double mutant. In contrast to Rspondins, the expression levels of Dkk2 and Notum in the double mutant are elevated during late anagen at P22 similar to control mice at P16 (Fig. 3e, Supplementary Figs. 5c and 14a). Note however, while the drop of Dkk2 during catagen in the dMF1/2 mutant is similar to that of wild-type mice (Supplementary Fig. 6), the drop of Notum in the dMF1/2 mutant appears to occur earlier as the levels of Notum in the DP of the dMF1/2 mutant during catagen is consistently lower than control mice (Supplementary Fig. 7). Together these data suggest that catagen induction and progression in the dMF1/2 double mutant are largely normal. This is further corroborated by the observation that Wnt activity during late anagen to mid catagen of the dMF1/2 mutant becomes restricted to the DP and is comparable to that of late anagen to mid catagen of control mice (Fig. 5, Supplementary Fig. 14a, b).In the absence of Fgfr1 and Fgfr2 in the DP, expression of Rspondins is maintained. However, this maintenance lasts for a certain period, and eventually, Rspondin expression is extinguished to allow catagen induction, suggesting the involvement of additional mechanisms in regulating the suppression of Rspondins. While the transcripts of Fgfr3 and Fgfr4 are barely detected in the DP of wild-type mice throughout the anagen phase (Supplementary Fig. 1c, d), the expression of these genes may be upregulated in the double mutant during late anagen to compensate for the loss of Fgfr1 and Fgfr2. To test this hypothesis, in situ hybridization for Fgfr3 and Fgfr4 was performed on skin sections from controls at P16 and dMF1/2 mutants at P22 (Supplementary Fig. 15a, b). Regardless of the genotype, while the transcripts of Fgfr3 and Fgfr4 were abundantly observed in the matrix, very low levels were detected in the DP, refuting the compensation hypothesis and suggesting that additional tyrosine receptor kinases and/or signaling pathways are involved in regulating the duration of anagen and catagen induction.DiscussionHair follicles undergo cycles of growth (anagen), regression (catagen), quiescence (telogen), and regeneration. In each phase of the cycle, follicles adopt a different morphology and structure, and consequently, the transitions between these phases involve dramatic morphological alterations that require not only the presence of epithelial stem and progenitor cell populations but also entail the coordination of multiple biological processes. Interestingly, such complexity is orchestrated within relatively precise time frames, suggesting that the hair follicle possesses a biological clock that dictates the periodicity of the hair cycle. Numerous models and theories have been proposed during the last five decades to explain the cyclic nature of the hair follicle10–17, and the hair cycle as a clock was previously postulated on the ground of systems biology approach17,18. In the current study, we revealed that the crosstalk between Fgf and Wnt signaling pathways in the DP generates adjustable positive and negative feedback loops that molecularly couple the epithelial and mesenchymal compartments in order to regulate the duration of anagen. Furthermore, the distinct properties of matrix and DP cells uncouple Wnt activity in the DP from Wnt activity in the matrix and thus allow the synchronization of catagen induction. At the mechanistic level, this type of control is consistent with the concept of the hair cycle clock, and a graphical summary of suggested model based on the data of the current study is shown in Fig. 8.Fig. 8Graphical summary of the positive and negative feedback loops that regulate the duration of anagen.Wnt ligands from the matrix activate Wnt signaling in both matrix and DP cells. Wnt activity in the DP induces the expression of Rspo that act on matrix cells to augment Wnt signaling by suppressing the inhibitory action of Znrf3/Rnf43 on Fzd receptors. This positive feedback loop maintains anagen. Fgf signaling in the DP uncoils this positive loop by repressing Rspo and consequently induces catagen. Concomitantly to the reduction in Rspo expression, a negative feedback loop is activated by Wnt activity in the DP that upregulates the Wnt antagonists Dkk2 and Notum. The inhibitory action of these antagonists further dampens Wnt signaling in the matrix to synchronize catagen induction. Note that the absence and presence of multiple Wnt inhibitory pathways in the DP and matrix respectively allow Wnt activity in the DP to be uncoupled from Wnt activity in the matrix and thus enable the simultaneous incline and decline of Wnt activity in the DP and matrix, respectively. This illustration is created with BioRender.com.A previous work demonstrated that Wnt signaling activity in the matrix maintains the anagen phase26, suggesting that reduction of Wnt signaling activity in the matrix to a certain threshold is the underlying mechanism to physiologically induce catagen. Consistently, our study illustrated that such decline of Wnt activity in the matrix does occur during late anagen, and further demonstrated that Wnt signaling in the matrix depends on Wnt activity in the DP. This dependency stems from (1) the ability of Wnt signaling in the DP to activate the expression of all four Rspondins, (2) the predominant expression of Rspondins in the DP, and (3) the intrinsic features of matrix cells that require Rspondins to augment their susceptibility to Wnt ligand activity (Fig. 8). This allows the DP to regulate the duration of anagen. Furthermore, Wnt ligands during mid-to-late anagen are largely expressed in the matrix and not in the DP40–42, suggesting that the matrix is the predominant source of Wnt ligands that activate the pathway in both the DP and the matrix. Specific ablation of Wls in the DP, performed in the current study, corroborated this proposition. Together, these create a unique positive feedback loop that intertwines the Wnt signaling pathway within the epithelial–mesenchymal interactions to regulate Wnt activity in the matrix and thus control the duration of the growth phase.Fgf signaling in the DP modulates this positive loop by progressively suppressing the expression of Rspondins (Fig. 8). The kinetics by which this reduction occurs define the period required to reduce Wnt activity in the matrix and to promote the induction of the regression phase, and consequently, control the duration of anagen. Remarkably, this decline in expression of Rspondins in wild-type mice starts ~4 days before the appearance of any morphological signs of catagen induction. This contradicts the current paradigm that strictly distinguishes between anagen and catagen and suggests that catagen at the molecular level is induced during mid-to-late anagen. Note that suppression of Rspondins by Fgf signaling in the DP occurs in the presence of active Wnt signaling in DP cells (Figs. 3d and 4b–d, Supplementary Fig. 5a), suggesting that Fgf signaling downregulates the expression of Rspondins downstream to beta-catenin activation. This is in line with a previous report illustrating that forced expression of the constitutive active form of beta-catenin specifically in the DP does not alter the hair cycle27. Mechanistically, Fgf signaling in the DP may antagonize beta-catenin activity in the DP at the promoter level of Rspondins, predicting a wide range of promotor activity that depends on the particular promoter. This is corroborated by the observation that ablation of beta-catenin in the DP results in complete abolishment of all Rspondins at the transcript level (Fig. 6d–g), while abrogation of Fgf signaling in the DP preferentially affects the expression of Rspo2 and Rspo3 (Fig. 3d, Supplementary Fig. 5a).It is noteworthy that while ablation of Fgfr1 and Fgfr2 in the DP results in anagen extension, mutant follicles eventually enter catagen. Furthermore, the reduction in Rspondins expression in the DP of the Fgfr1/2 double mutant does occur during the extended anagen but apparently at a slower pace, corroborating that catagen induction at the molecular level occurs long before the actual regression of the follicle and suggesting that additional molecular components or pathways are involved in regulating the same molecular pacemaker. While elevation of Fgfr3 and Fgfr4 in the DP of the dMF1/2 mutant during the extended anagen to compensate for the loss of Fgfr1 and Fgfr2 was not observed, the very low levels of Fgfr3 and Fgfr4 may be sufficient for contributing to the regulation of the hair cycle clock in the DP but require a longer time to suppress the expression of Rspondins. Alternatively, other receptor tyrosine kinases are also involved in regulating the hair cycle clock in the DP through cooperative activation of the same intracellular signal transduction pathways activated by Fgfr, and ablation of Fgfr receptors in the DP reduces the activation levels of these pathways but does not abolish them completely.Concomitant to the decline in Rspondins expression, the Wnt antagonists Dkk2 and Notum in the DP are elevated during late anagen and persist at least throughout early-to-mid catagen, suggesting a cooperative mode of action between agonists and antagonists to regulate catagen induction (Fig. 8). However, ablation of beta-catenin specifically in the DP results in premature induction of catagen despite the lack of elevation in both Dkk2 and Notum. This suggests not only that Dkk2 and Notum are target genes of Wnt signaling in the DP but also that Dkk2 and Notum are dispensable for catagen induction, raising the question of what is the precise role of these antagonists in catagen induction and progression. In addition to premature induction of catagen, catagen induction and progression are extremely unsynchronized in the absence of beta-catenin in the DP despite the apparently normal apoptosis observed in the regressing epithelial strand25. This supports the notion that Wnt signaling activity in the DP is also required to synchronize catagen induction and progression in addition to its role in the maintenance of the anagen phase. We hypothesize that this function of Wnt signaling activity is mediated by activating the expression of Dkk2 and Notum in the DP during late anagen to overcome fluctuations between follicles in the pace by which Wnt activity in the matrix is reduced and thus robustly coordinates catagen induction. Furthermore, the current analysis revealed that Wnt activity in the DP operates by orchestrating both positive and negative feedback loops. While the positive loop preserves Wnt activity in the matrix to maintain anagen, the negative loop assures that differences in the rate by which Wnt signaling activity in the overlying matrix decreases during late anagen are overrode to synchronize catagen induction.The suppression of Rspondins by Fgf signaling and the activation of Wnt antagonists by Wnt signaling only during late anagen and not at earlier times when both pathways are active suggest that signaling activation per se is not sufficient to repress or activate the expression of these genes, respectively. Instead, the duration of signaling reception may also contribute to these functions of Fgf and Wnt signaling in the DP. Alternatively, signaling levels must surpass a certain high threshold to repress or induce these genes. Unfortunately, sensitive methods to directly evaluate Fgf signaling levels in the DP are lacking and therefore preclude us from experimentally testing the latter hypothesis for Fgf signaling. By contrast, the observation that Wnt signaling activity is elevated specifically in the DP during mid-to-late anagen is consistent with this hypothesis. Future studies that manipulate the levels of both signaling pathways along a continuum are required to directly explore these hypotheses.While the current study did not aim to address the regulation of Fgf signaling in the DP upstream to the Fgf receptors, previous works suggest that Fgf5 is likely to contribute to this regulation. Fgf5 is expressed in the lower ORS and lower matrix21. Furthermore, Fgf5 knockout mice exhibit extended anagen. Combined with our analysis, these data together suggest that Fgf5 acts on DP cells to regulate the duration of anagen. Alternatively, Fgf5 function is relayed to the DP through activation of other Fgf ligands in the matrix that trigger Fgf transduction in the DP. However, anagen in the Fgf5 knockout mice is extended by 3 days22 while ablation of Fgfr1 and Fgfr2 in the DP results in anagen extension between 6 and 8 days. This suggests that a combined action of a few Fgf ligands is involved. Consistently, several Fgf ligands are known to be expressed during anagen in both the epithelial compartment and the DP28. Future functional analysis of these ligands will decipher their role in this process and dissect their relative contribution.MethodsMiceFgfr1Flox/Flox (Fgfr1tm5Sor)29, Fgfr2Flox/Flox (Fgfr2tm1Dor)30, Ctnnb1Flox/Flox (Ctnnb1tm2Kem)48, ROSA26 YFP reporter (Gt(ROSA)26Sortm1Cos)31, and Axin2-lacZ Wnt-reporter (Axin2tm1Wbm)35 mouse lines were obtained from Jackson Lab. The Rspo3Flox/Flox line49 was previously generated by Christof Niehrs laboratory. The DP-specific Corin-cre (Corintm2Bamo) mouse was previously generated25 and kindly provided by Bruce Morgan (Harvard medical school).Crosses for the Fgfr analysis were designed to obtain the following genotypes:Controls: [Cor-cre/+; Fgfr1Flox/+; Fgfr2Flox/+] OR [+/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox]Double mutants (dMF1/2): [Cor-cre/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox]Crosses for the beta-catenin analysis were designed to obtain the following genotypes:Controls: [+/+; Ctnnb1Flox/Flox]Mutants: [Cor-cre/+; Ctnnb1Flox/Flox]Crosses for the triple mutant (tMF1/2/b) analysis were designed to obtain the following genotypes:Controls: [+/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox; Ctnnb1Flox/+] OR [+/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox; Ctnnb1+/+]Double mutants: [Cor-cre/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox; Ctnnb1Flox/+] OR [Cor-cre/+;Fgfr1Flox/Flox;Fgfr2Flox/Flox;Ctnnb1+/+]Triple mutants: [Cor-cre/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox; Ctnnb1Flox/Flox]Crosses for the triple mutant (tMF1/2/R3) analysis were designed to obtain the following genotypes:Controls: [+/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox; Rspo3Flox/+] OR [+/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox; Rspo3+/+]Double mutants: [Cor-cre/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox; Rspo3Flox/+] OR [Cor-cre/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox; Rspo3+/+]Triple mutants: [Cor-cre/+; Fgfr1Flox/Flox; Fgfr2Flox/Flox; Rspo3Flox/Flox]The Federation of Laboratory Animal Science Associations guidelines were followed to house mouse colonies and experimental animals. Animals were maintained and bred in a temperature-controlled room. Food and water were available ad libitum, and a cycle of 12 h light/dark was employed. The Institutional Animal Care and Use Committee of Bar Ilan University approved the experimental protocols.Hair cycle variation and skin collectionAnagen duration varies between mouse lines and strains. In most lines used in this study, catagen of the first cycle in wild-type and control mice is induced around P16. This includes all lines with the conditional allele of beta-catenin and the FVB strain. In the genetic background of the Fgfr1/2 double mutant, however, catagen of the first cycle in wild-type and control mice is induced around P18.Dorsal skins for the analysis of follicle morphology by HE staining shown in Fig. 2 were harvested from the posterior region. For the lacZ staining shown in Fig. 5, adjacent dorsal regions along the anterior–posterior axis from the same mouse were harvested. For all the rest of the study, middle dorsal region along the anterior–posterior axis was used.Histology and immunofluorescenceFor immunostaining, dorsal skins were collected, fixed for 16 h at 4 °C in 4% paraformaldehyde (PFA)/PBS, serially dehydrated in sucrose (10–15–20%), embedded in optimal cutting temperature (OCT) compound, frozen on liquid nitrogen, and cryosectioned (7–10 µm). Sections were fixed in ice-cold acetone/methanol (1:1) for 5 min. Subsequently, antigen retrieval was performed by boiling the sections for 5 min in citrate buffer, pH 6 in a microwave. Sections were then cooled to room temperature, permeabilized in methanol for 10 min, washed for 10 min three times in PBS and blocked in 10% heat inactivated sheep serum (HISS) in PBS for 2 h. Sections were incubated overnight at 4 °C with primary antibodies (Abs), washed three times for 10 min in PBS, and incubated with secondary Abs for 1 h at room temperature. Sections were then washed three times for 10 min in PBS and mounted with DAPI FluoroMount-G (Electron microscopy sciences).To analyze morphology, hematoxylin and eosin (HE) was utilized to stain fixed sections using standard methods.Primary Abs: Rabbit anti-Lgr4 (1:100, Abcam Ab224480), Rabbit anti-Lgr5 (1:100, Abcam Ab75850), Rabbit anti-Lgr6 (1:100, Abcam Ab214325), Rabbit anti-Rnf43 (1:100, Abcam Ab217787), Rabbit anti-Znrf3 (1:100, Bioss Abs, 9141-R), mouse anti-Bcat (1:100, Abcam Ab6301), Rat anti-Pcad antibody (1:100; BD, #MAB761), Rabbit anti-Ki67 (1:100; Abcam Ab15580).Secondary Abs: Donkey anti-Rabbit FITC-conjugated, 1:1000 (Jackson ImmunoResearch cat#711-095-152), Donkey anti-Rabbit TRITC-conjugated, 1:1000 (Jackson ImmunoResearch cat#711-025-152), Donkey anti-mouse AlexaFluor488-conjugated, 1:500 (Jackson ImmunoResearch, cat#715-545-150).In situ hybridizationFixed sections from dorsal skins were used for nonradioactive in situ hybridization with Dig labeled RNA probes corresponding to nucleotides (nts) 1127–1689 of Fgfr1 (GenBank Acc. No. NM_010206), nts 1690–2105 of Fgfr2 (NM_010207), nts 2082–2586 of Fgfr3 (NM_001163215), nts 469–869 of Fgfr4 (NM_469-869), nts 958–1395 of Lgr4 (NM_172671), nts 1227–1823 of Lgr5 (NM_010195), nts 1018–1423 of Lgr6 (NM_001033409), nts 720–1313 of Rnf43 (NM_172448), nts 509–773 of Znrf3 (NM_001080924), nts 717–1212 of Rspo1 (NM_138683), nts 1033–1469 of Rspo2 (NM_001357957), nts 921–1348 of Rspo3 (NM_028351), nts 300–760 of Rspo4 (NM_001040689), nts 1168–1481 of Notum (NM_175263), nts 998–1503 of Dkk2 (NM_020265), nts 522–1040 of Krm1 (NM_032396), nts 193–619 of Krm2 (NM_028416).For signal detection, BM purple substrate (Roche) was used.X-gal stainingFor X-Gal staining, fresh frozen dorsal skins were used. To keep the harvested skin flat while embedding in OCT compound, skins were stretched on a membrane and immediately frozen on liquid nitrogen. 20-µm cryosections were fixed for 10 min in 0.2% glutaraldehyde (Sigma Aldrich), washed three times in PBS for 5 min, and incubated overnight at 37 °C with 1 mg/ml X-gal (Sigma Aldrich). Sections were washed in PBS three times for 5 min and mounted with Immu-Mount mounting medium (Thermo scientific).Tunel staining and co-immunostaining for Pcad10-µm skin sections were washed twice with PBS, fixed with 4% PFA for 10 min and again washed twice with PBS-0.1% Tween. Subsequently, sections were incubated with ProtK (10 µg/ml) for 5 min and washed twice with PBS-0.1% Tween. For Tunel staining, the “In situ Cell Death Detection Kit-TMR red” (Roche) kit was used according to the manufacturer’s instructions. Briefly, 50 µl Enzyme solution and 550 µl Label solution were mixed and applied on the sections, incubated for 60 min at 37 °C in a humidified box in the dark. Slides were washed three times in PBS then mounted with DAPI. When co-stained for P-Cadherin, the ProtK step was omitted, and after the last three washes in PBS, slides were blocked for 2 h in 10% HISS in PBS. Sections were overnight incubated at 4 °C with Rat anti-Pcad antibody (1:100; BD, #MAB761), washed in PBS three times for 10 min, and incubated for 1 h at room temperature with Donkey anti-Rat Cy5 antibody (1:500; Jackson immunoresearch, #712-175-753). Sections were then washed in PBS three times for 10 min and mounted with DAPI FluoroMount-G (Electron microscopy sciences).MicroscopyImaging of in situ hybridization and X-gal staining was performed with Zeiss upright AxioImagerM2 through a 20× objective with tiling mode using Zen Blue 2.3 software. HE staining was imaged with Zeiss upright AxioImagerM2 or Zeiss slide scannerZ1 through a 20× objective. For immunofluorescence, Zeiss LSM780 inverted confocal microscope was used to acquire images through a 20× objective using the Zen Black 11 (service pack 7) software. Adobe Photoshop CS5.1 was employed to process all images.Cell sortingSingle-cell suspension from whole skin was obtained by placing the dermis side down in 0.25% Trypsin (GIBCO) at 4 °C overnight, minced and stirred in 0.2% collagenase for 1 h at 37 °C. Strainers (100, 70, and 40 µM) were used to serially filter the dissociated cells. YFP-positive cells were twice FACS sorted on MoFlo Astrios (Backman Coulter): enrichment 1–2 mode was applied for the first sort, and purify 1 mode was employed for the subsequent sort. FACS analyses were performed using Summit program.RNA sequencingFACS-sorted cells were utilized to purify total RNA using the RNeasy Plus Micro kit (QIAGEN) according to manufacturer’s instructions. RNA integrity was tested using the Agilent RNA Pico Kit and Bioanalyzer at the Genome Technology center of the Faculty of Medicine Bar Ilan University. In total, 100 ng of total RNA were used to deplete ribosomal RNA, and the Nebnext Ultra Directional RNA kit (NEB, #E7420L) was utilized to generate libraries for Illumina sequencing. The dsDNA HS Assay Kit and QUBIT (Molecular Probes, Life Technologies) were employed to quantify the sequencing libraries. Additional measurements to test the quality of the sequencing libraries were performed by qPCR analysis using the illumina P7 and P5 primers. A standard library was used for optimal load on the Illumina HiSeq 2500 instrument. 2 nM of the library was denatured in 0.1 M NaOH for 5 min at room temperature. In total, 10 pM was loaded onto the Flow Cell with 1% Phix library control and sequenced using a 61-cycles single-read sequencing mode.BioinformaticsTrimmomatic was applied to trim and quality filter the 61-base single-end reads, and then Tophat (version 2) was utilized to map the reads to the mouse genome (NCBI38/mm10). Mapped reads for each annotated ENSEMBL gene (GRCm38.p4) were counted using HTSeq-count tool. DESeq2 was used to perform read count normalization and differential gene expression analysis.Statistics and reproducibilityStatistical methods to predetermine sample size were not used. Sample size and biological replicates are indicated in the figure legends and at least three biological replicates per genotype were used. Data are presented as mean ± SEM. A Padj value of <0.05 was considered significant.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationReporting Summary | nature communications | [
"Article"
] | [
"Cell signalling",
"Developmental biology",
"Stem-cell niche",
"Skin stem cells"
] | IntroductionTissue growth regeneration adult require specialized stem cell progenitor populations constrained time frame formation proper size shape structure1,2 tissue removal accompany growth normal hair follicle cycles between growth regression model molecular mechanisms stem cell activity progenitor behavior mature hair follicle growth phase (anagen composed of concentric layers differentiated keratinocytes outer root sheet inner root sheet hair shaft dermal papilla (DP), mesenchymal component embedded in hair bulb surrounded by matrix cells Matrix keratinocytes behave as progenitor cells hair matrix7 Asymmetric divisions renew progenitor cell compartment provide vertical flux transit progeny limited divisions form constituents IRS HS addition new constituents drives growth hair emergence through skin end anagen phase proliferation ceases regression phase (catagen) ensues apoptosis initiates matrix around DP spreads upwards follicles shorten regress8,9DP withdrawn with regressing epithelial strand beneath permanent follicle during quiescence models theories proposed explain cyclic nature hair follicle10–18 components molecular mechanisms hair cycle poorly understood duration of anagen mouse strain constant individual cycle suggesting biological clock periodicity hair cycle studies Fgf signaling hair cycle clock19–23 Fgf5 knockout mice display long hairs prolonged anagen Fgf5 administration during mid anagen premature induction catagen21–23 Expression analysis Fgf5 in hair follicle expressed during anagen in lower ORS to base bulb region including matrix cells adjacent DP21 catagen induction starts when matrix cells apoptosis Fgf5 may act on cells proliferation induce apoptosis may affect cells indirectly to DP21 proximity Fgf5-expressing cells to DP limited to lower direct mechanism Fgf5 matrix more likely mechanism transition from anagen to catagen unknown canonical Wnt signaling pathway may hair cycle clock Ablation of beta-catenin in matrix DP during mid anagen results in premature induction catagen25Wnt signaling activity required in matrix DP maintain anagen phase unclear whether activation into single process duration anagen or anagen-maintaining activity forced expression beta-catenin in DP alter hair cycle27 suggests Wnt signaling activity required sustain anagen not counteract catagen-inducing signals current study Fgf Wnt signaling pathways interact DP regulate hair cycle clock expression of Wnt agonists (Rspondins antagonists (Dkk2 Notum).ResultsFgf signaling in DP modulates hair cycle clock anagen durationTo determined Fgf receptors mediate In situ hybridization expression four Fgfr receptors during growth phase receptors detected in epithelial part follicle Fgfr1 Fgfr2 expressed in DP anagen catagen phase low levels of Fgfr3 Fgfr4 detected (Fig. 1a b suggesting Fgfr1 Fgfr2 predominant transducers of Fgf signaling in DP Fgfr1 Fgfr2 ablated in DP during hair cycle by crossing conditional knockout mouse lines of Fgfr1 Fgfr2DP-specific Cor-cre mouse line recombinase Three distinct mutants generated single Fgfr1 Fgfr2 double Fgfr1/2 mutants hair coat Fgfr1 Fgfr2 normal double Fgfr1/2 long hairs 1Fgfr1 Fgfr2 expressed DP anagen catagen phase hybridizations Fgfr1 Fgfr2 transcripts receptors restricted DP during catagen (P17) images same scale ordered clockwise Unpigmented FVB mice used > 3 mice per time point Scale bar 50 μm 2Ablation Fgf signaling extended anagen pictures wild type control Fgfr1/2 double mutant postnatal day 50 > 10 mice per panels HE staining dorsal skin sections wild-type mutant mice different stages hair cycle prolonged anagen double mutant P19 wild-type skin section mid-to-late catagen P24 P26 Fgfr1/2 double mutant mice transition from anagen to catagen wild-type follicles P26 completed regeneration mid anagen second cycle n > 3 mice per genotype per stage Scale bar 200 μmextended hair length back skins harvested from wild-type control dMF1/2 double mutant mice first second hair cycles hair cycle progresses dorsal skins collected from posterior Skin sections used for HE staining follicle (Fig. 2a wild-type mice catagen induced postnatal day 18 (P18) telogen commences at P20 new hair cycle around P22 (Fig. 2a catagen in double mutant mice initiated between P24 to P26 6–8 days later catagen induction in wild-type mice occurs at P18 double mutant varies at P26 mutant mice occasionally found in anagen test extension anagen first cycle anagen duration second cycle evaluated extended anagen first cycle large variance in catagen induction mutant preclude comparison control mutant complicate analysis duration anagen second cycle estimated active process hair pigmentation from Anagen III to catagen induction Fig. 4) skin appears dark duration visually measured each time frame most anagen phase estimation length anagen anagen phase extended in of Fgf signaling during second cycledata suggest key components hair cycle clock duration anagen in DP Fgf signaling pathway modulates expression DP transcriptome reveals two predominant expression programs operate during growth components hair cycle clock accumulated or degraded in DP during anagen phase molecules surpass threshold catagen induced test hypothesis gene expression profiling of DP cells from control double mutant mice during anagen performed by RNA-seq. DP cells cells labeled with YFP by activating cre-dependent YFP-reporter allele31 dissociation DP cells FACS sorted from mice at P10 P12 P14 P16 P18 spans time points during anagen in wild-type mutant mice includes time point during transition from anagen to catagen (P18) in wild-type mice Total RNA purified from sorted cells used sequencing libraries anagen of first cycle late follicle morphogenesis anagen subsequent cycles genuine anagen of hair cyclefollicle morphogenesis anagen phase hair cycle classified stages (Stage1–8 AnaI-VI stages refer early steps follicle formation early events anagen regeneration (AnaI-III post regeneration (AnaIV–V last stage morphogenesis anagen (Stage8 AnaVI begins tip HS emerges epidermis lasts growth first second hair cycles follicle unaltered stage growth Stage8 AnaVI equivalent “monotonous”. hypothesis suggests stage molecularly dynamic anagen subclassified into three time frames early-to-mid anagen (P10–P12) mid-to-late anagen (P14–P16) late anagen to early catagen (P16–P18) analysis (Fig. 3a DP transcriptome early-to-mid anagen (P10–P12) unaltered regardless genotype DP transcriptome wild-type mice mid anagen to catagen (P14–P18) dynamic changes DP transcriptome mutant mice P14–P18 evolves distinctive control anagen progression hypothesis explored expression behavior DP cells during anagen genes expressed control mutant mice at P16 expression pattern from P10focus on genes hypothesis catagen induction differentially expressed between controls mutants during late anagen before Samples ordered presented heatmap for control mutant mice (Fig. 3b). two expression programs in control mice transcript levels one program mid anagen elevated decrease late anagen low other begin low incline maximum at transition between anagen catagen same programs in mutant rate change slower data hypothesis suggest hair cycle clock anagen duration accumulating degrading components.Fig. 3Abrogation of Fgf signaling in DP results in altered expression of Wnt agonists antagonists PCA analysis colored triangle represents DP transcriptome control mutant mouse Black red circles demarcate control mutant mice n = 3 mice per genotype per time point Heatmap analysis Two classes of genes revealed expression pattern expression during anagen in control low in mutant expression reduced in control higher in mutant Graphic sequence reads RNA-seq analysis for Wnt agonists (Rspo1–4) antagonists (Notum Dkk2) different scales reads Data mean ± SEM.Two-sided t-test comparisons Benjamin–Hochberg correction Padj = 0.001 7.14E − 11 3.26E − 08 3.18E − 57 3.48E − 24 4.91E − 26, 4.05E − 38 0.0003 0.02 3.2E − 06 4.14E − 07. 3 mice per genotype per time point In situ hybridization Rspo3 Dkk2 P10 P16 expression genes DP altered expression mutant transcripts dark blue pigmented hair shafts black 5 mice per genotype per time point Scale bar 25 μm expression Wnt agonists antagonists DP altered during anagen analysis signaling pathways altered mutant DP Wnt signaling agonists antagonists altered cooperative (Fig. 3c). wild-type mice four Rspondins Wnt decrease mid-to-late anagen antagonists Notum Dkk2 accumulate transcriptional levels genes mutant remain closer baseline altered slower pace expression pattern dynamics Rspondins Dkk2 Notum explored skin sections in situ hybridization (Fig. 3d e Supplementary Fig. 5)four Rspondins predominantly expressed in DP lower levels matrix (Rspo1 or expression restricted to DP (Rspo3 DP major source Rspondins bulb region transcripts four Rspondins observed in DP P10 mice genotype expression Rspo2 Rspo3 DP wild-type mice at P16 reduced beneath detection levels 3d Expression Rspo1 Rspo4 wild-type at P16 high situ hybridization undetectable during catagen 5a transcripts all four Rspondins detected in DP P16 dMF1/2 mutant mice-seq analysis.DP-confined expression observed for Notum Dkk2 during late anagen (Fig. 3e At P10 expression levels Dkk2 Notum wild-type dMF1/2 mutant mice low detected situ hybridization At P16 expression Dkk2 Notum increases situ hybridization high levels persist early catagen decline to undetectable levels end catagen mid catagen Figs 6a 7a). expression Dkk2 Notum mutant at P16 remains beneath detection levels (Fig. 3epersistent expression Dkk2 Notum in wild-type DP during catagen suggests catagen DP expression at P16 may coincide regulation catagen induction expression Dkk2 Notum catagen initiation proliferation apoptosis evaluated at P16 Fig. During catagen induction proliferation matrix ceases programed cell death initiated DP8 Immunostaining for Ki67 dMF1/2 mutant mice at P16 revealed matrix cells highly proliferative regardless genotype Tunel staining detect apoptotic cells matrix wild-type follicles at P16 represent late anagen data expression Dkk2 Notum in DP elevated prior catagen induction both catagen initiation.Persistent expression Rspo3 in DP dMF1/2 mutant contributes to extended expression Wnt agonists antagonists Wnt signaling activity decreases during mid-to-late anagen promotes catagen induction hypothesis predicts abrogation of Rspondins in DP premature induction catagen co-expression of four Rspondins in DP analysis single ablation single ablation of Rspo3 in DP result in premature induction catagen during first second cycleRNA-seq situ analyses significant changes gene expression control dMF1/2 mutant mice observed for Rspo2 Rspo3 persistent expression DP dMF1/2 mutant to extended anagen triple mutant of Rspo3 Fgfr1 Fgfr2/F1/2/R3) generated duration anagen first second cycle evaluated prolonged anagen double dMF1/2 mutant comparable to triple tMR3/F1/F2 mutant first cycle anagen duration second cycle shortens longer than controls 4) suggests persistent expression Rspo3 Fgfr1 Fgfr2 contributes extended anagen Rspondin expression duration anagen Wnt agonists antagonists DP act on matrix cells autocrine Wnt signaling activity reduced during late anagen Axin2-lacZ Wnt-reporter allele35 introduced in control dMF1/2 mutant mice dynamics Wnt activity matrix differ DP anagen wild-type mice Wnt signaling activity wild-type matrix reduced at P16 P10 Wnt activity DP late anagen persists intensifies reduction Wnt activity matrix during mid-to-late anagen occurs progressivelyintensification Wnt activity DP dMF1/2 mutant Wnt activity matrix mutant mice at P16 high comparable to P10 (Fig. 4a Immunostaining beta-catenin at P16 activated beta-catenin DP cells in wild-type mutant mice. persistent Wnt signaling DP late anagen data suggest DP cells resistant to Rspondins Dkk2 Notum matrix cells target agonists antagonists. 4Wnt signaling activity mid anagen decreased in matrix catagen induction X-Gal staining shows LacZ expression at P10 Wnt signaling detected in matrix DP genotype comparable signal matrix n = 3 mice per genotype X-Gal staining late anagen (P16) Wnt activity wild type diminished matrix persists DP Wnt mutant persists matrix DP signal DP higher than matrix line boundaries follicle n = 5 mice per genotype Immunostaining for beta-catenin late anagen presence activated beta-catenin in DP cells wild-type mutant littermateseach genotype same follicle shown twice left beta-catenin fluorescent image right merge beta-catenin nuclear staining (DAPI) n = 6 mice per genotype Wild-type skin sections P10 mice immunostain Rspondin activity Nuclei blue none detected DP. n > 3 mice Scale bar (a b), 50 μm (c–d), 25 μm Notum enzyme hydrolyzes palmiteoylate adducts Wnt ligands inactivates Wnts bind Fzd Rspondins Dkk2 require molecular machinery target cells confined action Wnt agonists antagonists matrix cells restricting expression molecular machinery Rspondins expression assessed wild-type mice (Fig. 4e Figs. 10 11). Znrf3 Rnf43 transmembrane E3 ubiquitin ligases remove Wnt receptors membrane Fzd receptors Rspondins bind E3 ubiquitin ligase Lgr family (Lgr4 structure abolishes ubiquitination activity E3 ligase augments Wnt activity surface accumulation Fzd receptorsImmunostaining hybridization Lgr4 Lgr5 Lgr6 Znrf3 Rnf43 expressed epithelial follicle not detected DP (Fig. 4e Fig. 10). Dkk2 inhibits Wnt signaling Wnt co-receptor Lrp5/6 transmembrane receptor Krm1/2 internalization Lrp5/6 reduction cell Wnt hybridization Krm1 Krm2 undetectable in DP abundant epithelial follicle Fig. 11). Krm1 expressed follicular epithelium Krm2 asymmetrically localized small matrix cells adjacent DP data suggest matrix cells respond Rspondins Dkk2 insight reduction Wnt signaling activity matrix during late anagen persistent expression Dkk2 Notum during catagen inhibition Wnt activity catagen progression resistance DP inhibition Wnt signaling activity retained during catagen Axin2-lacZ Wnt-reporter allele35 Wnt activity during catagen Wnt activity lower regressing follicle observed DP reduced during mid-to-late catagen (Fig. 5a–c).Fig. 5Wnt signaling activity during catagen.LacZ staining dorsal skin sections wild-type mouse at P18a Posterior dorsal region high Wnt activity DP lack Wnt epithelial part follicle adjacent DP early catagen higher magnification field dashed rectangle n = 3 mice. b Middle dorsal region Wnt activity early-to-mid catagen n = 3 mice. c Anterior region Wnt activity mid catagen higher magnification field dashed rectangle n = 3 mice. Scale bar 50 μm.Wnt signaling DP required for Wnt activity matrix stimulated by Wnt ligands reduction Wnt signaling matrix late anagen (Fig. 4) ablation beta-catenin matrix premature induction catagen26 physiological ablation beta-catenin DP premature induction catagen25 Wnt signaling matrix maintains anagen phase Wnt activity DP regulates duration anagen Wnt signaling gene beta-catenin (Ctnnb1) ablated DP crossing Cor-cre line mouse line floxed allele beta-catenin catagen Ctnnb1 mutant prematurely induced at P12 4 days earlier wild type Fig. 12a–e). catagen induction wild-type mice at P16Immunostaining for beta-catenin at P10 wild-type Ctnnb1 mutant follicles in anagen confirmed deletion of beta-catenin in DP Ctnnb1 mutant (Fig. 6a beta-catenin in wild-type mice detected in nuclei matrix cells indication Wnt signaling activity nuclear localization beta-catenin in matrix Ctnnb1 mutant mice reduced Wnt signaling activity decreased beta-catenin ablation DP Axin2-LacZ Wnt-reporter allele Wnt signaling activity abolished in DP matrix beta-catenin ablated DP (Fig. 6c Wnt activity matrix depends on Wnt activity DP.Fig. 6Ablation of beta-catenin in DP Wnt signaling activity matrix Immunostaining for beta-catenin during mid anagen (P10) wild-type follicle Nuclei in blue activated beta-catenin detected in nuclei n = 3 mice beta-catenin mutant mice immunostain for beta-catenin at P10 efficient deletion beta-catenin in DP reduction of nuclear beta-catenin in matrix cells higher magnification n = 3 miceX-Gal staining skin P10 wild-type mutant littermates Wnt activity abolished in DP matrix internal black line outlines DP external black line demarcates follicle boundaries n = 3 mice per genotype In situ hybridization for Rspondins skin P10 Rspondin transcripts in DP matrix mutant. n = 3 mice per genotype Scale bar (a, b), 10 μm (c–g), 25 μm reports mid-to-late anagen Wnt ligands expressed in matrix not detected matrix only functional source Wnt ligands bulb region expression late anagen DP cells Wnt curtailed by ablation Wls gene DP mouse line floxed allele Wls43 crossed with Cor-cre mouse line genetic manipulation altered first hair cycle affected HS matrix predominant functional source Wnt ligands bulb region.Wnt signaling activity DP activates expression Rspondins essential for extended anagen dMF1/2 mutantPrevious reports Rspondins transcriptional targets of Wnt signaling44 Wnt activity DP regulates Wnt signaling matrix expression Rspondinsin situ hybridization skin P10 Ctnnb1 mutant expression Rspondins DP abolished low expression diminished Ctnnb1 mutant mice. consistent Rspondins target genes Wnt signaling reduction Wnt activity beta-catenin ablated DP extended anagen Fgfr1 Fgfr2 ablated requires active Wnt signaling triple mutant Ctnnb1 Fgfr1 Fgfr2 tMF1/2/b generated follicles enter prematurely catagen Ctnnb1 mutant (Fig. interaction Fgf Wnt signaling pathways ablation Fgfr1 Fgfr2 beta-catenin DP premature induction catagen staining skin sections control dMF1/2 Ctnnb1 tMF1/2/b mice P16 unsynchronized catagen Ctnnb1 tMF1/2/b mutant mice > 3 mice per genotype Tunel staining P13 skin sections dMF1/2 Ctnnb1 tMF1/2/b triple mutant normal apoptosis DP premature catagen Pcad immunostaining included Nuclei blue n = 3 mice per genotypeHigher magnifications fields surrounding DP Ctnnb1 tMF1/2/b mutant follicles displayed Scale bar (a–d), 200 μm (e, f), 25 μm.Wnt signaling activity DP activates expression Dkk2 Notum Fgf signaling DP dMF1/2 mutant P16 prevents elevation Dkk2 Notum suggesting activated Fgf signaling Dkk2 Notum target genes Wnt signaling negative feedback explore target in situ hybridization expression wild-type Ctnnb1 mutant littermates tested hypothesis Wnt activity DP suppresses expression Dkk2 Notum Fgf signaling activates expression Dkk2 Notum Wnt predicts elevation Dkk2 Notum levels DP Ctnnb1 mutant P10 Dkk2 Notum undetectable regardless genotype Fig. 13a refuting hypothesis Wnt activity suppresses expression P12 catagen induced Ctnnb1 mutant Dkk2 Notum remain undetectable lack expression persists premature catagen phase Ctnnb1 contrast normal expression restricted activity Wnt signaling DP during catagen wild-type mice (Fig. 56a 7a). suggests Dkk2 Notum activated DP by Wnt signaling activation occurs during late anagen despite Wnt activity earlier times Wnt activity intensified during late anagen suggest activation Dkk2 Notum requires high Wnt signaling activity or longer period Wnt signaling reception expression data support Dkk2 Notum initiation catagen not required to induce catagen reduction of Rspondin levels during mid-to-late anagen driving force catagen.Catagen induction progression in dMF1/2 mutant follicles remain in anagen enter catagen in double mutant in situ hybridization for Rspondins Dkk2 Notum performed during late anagen catagen At P22 expression of Rspo2, Rspo3 Rspo4 in mutant DP almost abolished expression persists during catagen phase 14a). transcript levels of Rspo1 remain high detected at P22 undetectable during catagen immunostaining for Ki67 Tunel labeling at P22 mutant follicles sustain proliferative state P22 late anagen stage 8). data suggest reduction in Rspondin expression DP during mid-to-late anagen induces catagen in double mutantcontrast Rspondins expression levels Dkk2 Notum double mutant elevated late anagen at P22 similar control mice P16 (Fig. 3e Figs. 5c 14a). drop Dkk2 catagen dMF1/2 mutant similar wild-type mice drop Notum dMF1/2 mutant earlier lower than control mice 7) data suggest catagen induction progression dMF1/2 double mutant normal Wnt activity late anagen mid catagen restricted to DP comparable to control mice (Fig. 5 14a of Fgfr1 Fgfr2 DP expression Rspondins maintained maintenance lasts Rspondin expression extinguished catagen induction suggesting additional mechanisms regulating suppression Rspondins transcripts Fgfr3 Fgfr4 barely detected in DP wild-type mice anagen 1c expression genes may upregulated in double mutant late anagen compensate loss Fgfr1 Fgfr2. in situ hybridization for Fgfr3 Fgfr4 performed on skin sections controls P16 dMF1/2 mutants at P2215a, b). Regardless genotype transcripts of Fgfr3 Fgfr4 observed in matrix low levels detected in DP refuting compensation hypothesis suggesting additional tyrosine receptor kinases signaling pathways regulating duration anagen catagen induction follicles undergo cycles growth regression quiescence regeneration each adopt different structure transitions involve alterations epithelial stem progenitor cell populations coordination biological processes complexity orchestrated precise time frames hair follicle possesses biological clock dictates periodicity cycle models theories proposed explain cyclic nature hair hair cycle as clock postulated systems biology current study revealed crosstalk between Fgf Wnt signaling pathways in DP generates adjustable positive negative feedback loops epithelial mesenchymal compartments regulate duration anagen properties of matrix DP cells uncouple Wnt activity DP matrix allow synchronization of catagen induction control consistent with concept hair cycle clock graphical summary model shown in Fig. positive negative feedback loops regulate duration anagen.Wnt ligands from matrix activate Wnt signaling in matrix DP cellsWnt activity DP induces expression Rspo matrix cells Wnt signaling inhibitory action Znrf3/Rnf43 Fzd receptors positive feedback loop maintains anagen Fgf signaling DP uncoils loop Rspo induces catagen reduction Rspo expression negative feedback loop activated by Wnt activity DP upregulates Wnt antagonists Dkk2 Notum inhibitory action antagonists dampens Wnt signaling matrix catagen induction multiple Wnt inhibitory pathways in DP matrix allow Wnt activity matrix enable simultaneous incline decline Wnt activity illustration BioRender work demonstrated Wnt signaling activity matrix maintains anagen phase26 reduction Wnt signaling catagen study decline Wnt activity matrix during late anagen Wnt signaling depends on Wnt activity DP dependency stems from Wnt signaling DP activate expression four Rspondins predominant expression Rspondins DP intrinsic features matrix cells Rspondins augment susceptibility Wnt ligand activity (Fig. allows DP regulate duration anagenWnt ligands during mid-to-late anagen expressed in matrix not DP40–42 matrix predominant source of Wnt ligands DP matrix ablation of Wls in DP study proposition positive feedback loop Wnt signaling pathway epithelial–mesenchymal Wnt activity matrix duration growth phase.Fgf signaling in DP modulates loop expression Rspondins (Fig. reduction define period reduce Wnt activity matrix induction regression phase control duration anagen decline in expression Rspondins in wild-type mice starts ~4 days before signs catagen induction contradicts current paradigm anagen catagen catagen induced during mid-to-late anagen suppression of Rspondins by Fgf signaling in DP occurs in presence active Wnt signaling in DP cells (Figs. 3d 4b–d signaling downregulates expression Rspondins beta-catenin activation previous report forced expression beta-catenin in DP alter hair cycle27 Fgf signaling DP may antagonize beta-catenin activity at promoter level Rspondins range promotor activity promotercorroborated ablation beta-catenin DP Rspondins transcript level (Fig. 6d–g), abrogation Fgf signaling affects expression Rspo2 Rspo3 (Fig. 3d Fig. ablation of Fgfr1 Fgfr2 results anagen extension mutant follicles enter catagen reduction in Rspondins expression DP Fgfr1/2 mutant during extended anagen slower pace catagen induction occurs before regression follicle additional molecular components pacemaker elevation of Fgfr3 Fgfr4 in DP dMF1/2 mutant loss Fgfr1 Fgfr2 observed low levels of Fgfr3 Fgfr4 hair cycle clock require longer time to suppress expression Rspondins other receptor tyrosine kinases hair cycle clock intracellular signal transduction pathways Fgfr ablation of Fgfr receptors reduces activation abolish completely decline Rspondins Wnt antagonists Dkk2 Notum DP elevated during late anagen persist early-to-mid catagen cooperative action agonists antagonists catagen induction (Fig. 8). ablation of beta-catenin DP results in premature induction catagen despite lack elevation Dkk2 Notumsuggests Dkk2 Notum target genes Wnt signaling DP dispensable for catagen induction question role antagonists in catagen induction progression catagen induction progression unsynchronized in beta-catenin DP despite normal apoptosis in regressing epithelial supports Wnt signaling activity DP synchronize catagen induction progression anagen phase hypothesize function Wnt signaling mediated by activating expression Dkk2 Notum during late anagen overcome fluctuations follicles Wnt activity coordinates catagen induction analysis Wnt activity operates positive negative feedback loops positive loop preserves Wnt activity anagen negative loop assures differences Wnt signaling activity during late anagen synchronize catagen induction suppression of Rspondins by Fgf signaling activation of Wnt antagonists by Wnt signaling during late anagen not earlier active suggest signaling activation not to repress activate expression genes duration of signaling reception may contribute to functions Fgf Wnt signaling signaling levels must surpass high threshold to repress or induce genes sensitive methods evaluate Fgf signaling levels DP lacking preclude testing latter hypothesisWnt signaling activity elevated in DP during mid-to-late anagen consistent with hypothesis Future studies signaling pathways explore hypotheses current study regulation Fgf signaling DP previous works suggest Fgf5 regulation expressed in lower ORS matrix21 Fgf5 knockout mice exhibit extended anagen data suggest Fgf5 acts on DP cells duration anagen Fgf5 relayed to DP through activation other Fgf ligands anagen in Fgf5 knockout mice extended 3 ablation of Fgfr1 Fgfr2 DP anagen extension 6 8 days suggests combined action Fgf ligands involved several Fgf ligands expressed during anagen in epithelial compartment Future functional analysis ligands decipher role contribution.MethodsMiceFgfr1Flox Fgfr2Flox Ctnnb1Flox ROSA26 YFP reporter Axin2-lacZ Wnt-reporter)35 mouse lines obtained from Jackson Lab Rspo3Flox/Flox line49 generated by Christof Niehrs laboratoryDP-specific Corin-cre mouse Bruce Morgan Fgfr analysis genotypes-cre Fgfr1Flox Fgfr2Flox Fgfr1Flox mutants-cre Fgfr1Flox Fgfr2Flox beta-catenin analysis genotypes+ Ctnnb1Flox/Flox-cre+ Ctnnb1Flox triple mutant (tMF1/2/b genotypes+ Fgfr1Flox/Flox Fgfr2Flox Ctnnb1Flox Fgfr1Flox Fgfr2Flox Ctnnb1 mutants [Cor-cre Fgfr1Flox Fgfr2Flox/Flox Ctnnb1Flox-cre;Fgfr1Flox/Flox;Fgfr2Flox/Flox+/ mutants [Cor-cre/+ Fgfr1Flox/Flox Fgfr2Flox Ctnnb1Flox triple mutant (tMF1/2/R3) analysis genotypes+ Fgfr1Flox/FloxFgfr2Flox/Flox Rspo3Flox/+ Fgfr1Flox/Flox Fgfr2Flox/Flox Rspo3+]Double mutants Fgfr1Flox/Flox Fgfr2Flox Rspo3Flox Fgfr1Flox/Flox Fgfr2Flox/Flox Rspo3+/]Triple mutants Fgfr1Flox/Flox Fgfr2Flox/Flox Rspo3Flox Federation Laboratory Animal Science Associations guidelines mouse colonies experimental animals temperature-controlled room Food water ad libitum 12 h light/dark Institutional Animal Care Use Committee Bar Ilan University approved protocols.Hair cycle variation skin collectionAnagen duration varies between mouse lines strains catagen first cycle induced around P16 lines beta-catenin FVB strain Fgfr1/2 double mutant catagen induced around P18.Dorsal skins follicle harvested posterior region lacZ staining Fig. 5 adjacent dorsal regions mouse harvested middle dorsal region usedHistology immunofluorescenceFor immunostaining dorsal skins collected fixed 16 h 4 °C 4% paraformaldehyde/PBS dehydrated sucrose embedded optimal cutting temperature compound frozen liquid nitrogen cryosectioned (7–10 μm). fixed-cold acetone/methanol (1:1) 5 min antigen retrieval boiling 5 min citrate buffer pH 6 microwave cooled room temperature permeabilized methanol 10 min washed 10 min three times PBS blocked 10% heat inactivated sheep serum) 2 h incubated overnight 4 °C primary antibodies washed three times 10 min incubated secondary Abs 1 h room temperature washed three times 10 min DAPI FluoroMount-G hematoxylin eosin (HE) stain fixed sections Abs Rabbit anti-Lgr4-Lgr5-Lgr6-Rnf43-Znrf3 mouse anti-Bcat Rat anti-Pcad antibody Rabbit anti-Ki67Donkey anti-Rabbit FITC-conjugated 1:1000#711-095-152) anti-Rabbit TRITC-conjugated 1:1000#711-025-152) Donkey anti-mouse AlexaFluor488 1:500#715-545-150).In situ dorsal skins nonradioactive hybridization labeled RNA probes 1127–1689 Fgfr1 1690–2105 Fgfr2 2082–2586 Fgfr3 469–869 Fgfr4 958–1395 Lgr4_172671) 1227–1823 Lgr5 1018–1423 Lgr6_001033409) 720–1313 Rnf43 509–773 Znrf3 717–1212 Rspo1 1033–1469 Rspo2 921–1348 Rspo3 300–760 Rspo4_001040689) 1168–1481 Notum 998–1503 Dkk2 522–1040 Krm1 193–619 Krm2_028416) signal detection BM purple substrate used-gal frozen dorsal skins usedharvested skin OCT skins stretched frozen liquid nitrogen 20-μm cryosections fixed 10 min 0.2% glutaraldehyde washed PBS 5 min incubated overnight 37 °C 1 mg/ml X-gal Sections washed PBS three 5 min Immu-Mount medium staining co-immunostaining Pcad10-μm sections washed twice fixed 4% PFA 10 min PBS-0.1% Tween incubated ProtK (10 μg/ml 5 min washed PBS-0.1% Tween Tunel staining situ Cell Death Detection Kit 50 μl Enzyme 550 μl Label solution mixed sections incubated 60 min 37 °C humidified washed three times PBS DAPI co-stained P-Cadherin ProtK step omitted slides blocked 2 h 10% HISS overnight incubated 4 °C Rat anti-Pcad antibody washed PBS three 10 min incubated 1 h room temperature Donkey anti-Rat Cy5 antibodySections washed PBS 10 min mounted DAPI FluoroMount-G hybridization X-gal staining Zeiss AxioImagerM2 20× objective Zen Blue 2.3 software HE staining imaged Zeiss AxioImagerM2 Zeiss slide scannerZ1 20× objective immunofluorescence Zeiss LSM780 confocal microscope 20× Zen Black 11 software Adobe Photoshop CS5.1 sortingSingle-cell suspension 0.25% Trypsin) 4 °C overnight minced stirred 0.2% collagenase 1 h 37 °C Strainers (100 70 40 μM) filter dissociated cells YFP-positive cells FACS sorted MoFlo Astrios enrichment 1–2 mode purify 1 mode subsequent FACS analyses Summit program.RNA sequencingFACS-sorted cells RNA RNeasy Plus Micro kit RNA integrity tested Agilent RNA Pico Kit Bioanalyzer Genome Technology center Faculty Medicine University 100 ng RNA deplete ribosomal RNA Nebnext Ultra Directional RNA kit libraries Illumina sequencing dsDNA HS Assay Kit QUBIT sequencing librariesmeasurements sequencing libraries qPCR analysis illumina P7 P5 primers standard library used load Illumina HiSeq 2500 instrument 2 nM denatured 0.1 M NaOH 5 min room temperature 10 pM loaded Flow Cell 1% Phix library control sequenced 61-cycles single-read sequencing mode 61-base-end reads Tophat (version 2) map reads mouse genome (NCBI38/mm10) Mapped reads ENSEMBL gene (GRCm38.p4) counted HTSeq-count tool DESeq2 read count normalization differential gene expression analysis.Statistics sample size Sample size biological replicates indicated figure legends three per genotype Data mean SEM Padj value <0.05 significant Nature Research Reporting Summary.Supplementary | 50.6 | 1.007759 |
10.1038/s41467-021-21625-2 | PMC7935925 | The process of thin sheet crumpling is characterized by high complexity due to an infinite number of possible configurations. Andrejevic et al. show that ordered behavior can emerge in crumpled sheets, and uncover the correspondence between crumpling and fragmentation processes. | As a confined thin sheet crumples, it spontaneously segments into flat facets delimited by a network of ridges. Despite the apparent disorder of this process, statistical properties of crumpled sheets exhibit striking reproducibility. Experiments have shown that the total crease length accrues logarithmically when repeatedly compacting and unfolding a sheet of paper. Here, we offer insight to this unexpected result by exploring the correspondence between crumpling and fragmentation processes. We identify a physical model for the evolution of facet area and ridge length distributions of crumpled sheets, and propose a mechanism for re-fragmentation driven by geometric frustration. This mechanism establishes a feedback loop in which the facet size distribution informs the subsequent rate of fragmentation under repeated confinement, thereby producing a new size distribution. We then demonstrate the capacity of this model to reproduce the characteristic logarithmic scaling of total crease length, thereby supplying a missing physical basis for the observed phenomenon. | IntroductionCrumpling is a complex, non-equilibrium process arising in diverse systems across a wide range of length scales, from the microscopic crumpling of graphene membranes1, to the macroscopic folding of Earth’s viscoelastic crust2. Crumpled structures are highly porous, providing function for applications such as high-performance batteries and supercapacitors by increasing the electrochemical surface area3,4. Controlled crumpling has also been used to tune electronic, optical, and surface properties in graphene films1. Further, understanding the mechanics of crumpling is essential as flexibility and shape conformation become integral considerations in the design of thin, wearable devices5,6. Despite its ubiquity, a complete understanding of crumpling dynamics remains elusive due to the complexity of the disordered process. Nevertheless, statistical properties of crumpled geometries are highly reproducible in experiment7–13 and confirmed via simulation14,15, which suggests that this complex process is strongly dictated by universal aspects of thin sheets such as topology and self-avoidance13.Similarly adopting a coarse-grained perspective, Gottesman et al.16 revealed an unexpected order to ridge network evolution in crumpled sheets. By performing a protocol of repeated compaction and unfolding, as in the schematic of Fig. 1a, they demonstrated that the intricate details of ridge networks in crumpled sheets could be reduced to a single collective quantity, the total crease length, which evolves robustly as a logarithm in the number of crumpling repetitions across varying degrees of compaction. Notably, the incremental damage added upon re-crumpling the sheet was found to be independent of the sheets’ crumpling history -the sequence of prior compactions performed to produce the current crease network. Rather, the added crease length is determined only by the current total crease length and the new compaction depth. While processes that evolve logarithmically in time are readily observed in a variety of disordered physical systems, including stress or strain relaxation of a compacted sheet17–19, conductance relaxation of disordered electronic systems20, and creep dynamics of granular suspensions21, the emergence of a logarithmic model in the specific context of damage evolution in crumpled sheets is clearly distinct, and has had limited physical justification thus far.Fig. 1Data processing.a An L0 × L0 Mylar sheet is uniaxially compressed to a compaction ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}=L/{L}_{0}$$\end{document}Δ~=L/L0, unfolded, and its height profile scanned using a laser profilometer, for n iterations. b The mean curvature obtained from the height profiles of two distinct sheets at different \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}$$\end{document}Δ~. Red and blue colors denote folds in opposite directions. c The facet segmentation of b, colored randomly to visually discern facets.In this work, we take a novel approach to characterize crumpling and offer explanation for the logarithmic model by drawing a correspondence between crumpling and fragmentation processes. Fragmentation models have a rich history of theoretical development22–24 as well as industrial applications22,25 and use in modeling collision and fracture phenomena26. Here we concentrate on a theoretical, physically based rate equation for modeling time-dependent fragmentation detailed by Cheng and Redner27, which provides a general framework for processes that may be treated as successive, homogeneous breakups instigated by non-local stresses. The model has been flexibly applied to describe polymer degradation28 and volcanic fragments expelled in an eruption29, for example, though to the best of our knowledge this is the first application of such concepts to describe crumpling.Our work is organized as follows: we derive a scaling solution to the rate equation presented in Cheng and Redner27 which decomposes into a time-invariant distribution of scaled facet area and a time-dependent evolution of mean facet size. We demonstrate that the derived area distribution effectively reproduces key statistical features of experimental crumpled patterns. Fragment distributions are a natural point of comparison between theory and experiment; however, in this work we go a step further to draw additional correspondence in the temporal evolution of the patterns. The temporal parameter that chronicles the evolution of mean facet size serves as an intrinsic clock measuring the maturity of the fragmentation process. We connect this to experimental parameters driving fragmentation forward, namely the number of crumpling iterations and compaction strength which characterize the experiments of Gottesman et al.16. To do so, we construct a simple geometric model that likens crumpling to a random walk and is informed by the statistical properties of the derived area distribution. We derive an analytical relation for how geometric frustration occurring in a confined random walk instigates new damage and advances the temporal measure of fragmentation maturity. We demonstrate how this approach allows one to recover the logarithmic evolution of damage in ridge networks observed in Gottesman et al.16 and explain the history independence of damage formation, thereby furnishing a missing physical basis for this unexpected result.The key idea behind our model is the extension of fragmentation theory to incorporate a feedback loop: as facets become smaller, they make the sheet more compliant and therefore lower the rate of subsequent fragmentation. This idea may extend to many physical systems where the accumulation of damage inhibits further damage from occurring. Our work therefore shows how fragmentation theory could be applied more generally, and suggests that the universal damage evolution seen in crumpling may have analogs in other physical systems.ResultsThe collection and processing of experimental crumpling data used to verify analytical results presented in this work is fully detailed in the Methods section. Crease patterns obtained from uniaxially compressed Mylar sheets as shown in Fig. 1b are carefully segmented into individual facets as in Fig. 1c. The samples collected vary in compaction ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}$$\end{document}Δ~, the ratio of final to initial height, and in the number of successive crumples of the same sheet, n. A total of 24 segmented crease patterns is analyzed spanning 7 different compaction ratios and including n = 1, 2, 3, and 24 crumpling iterations.Throughout this work, we will refer to fragmentation in the context of crumpling as the successive partitioning of a thin sheet into smaller, flat facets separated by ridges or creases. To facilitate the construction of a model for this process, we begin with the general theory of fragmentation kinetics outlined in Cheng and Redner27. Let x represent facet area and c(x, t) the concentration of facets of area x at time t; then the linear integro-differential equation describing the evolution of c(x, t) is given by1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{\partial c(x,t)}{\partial t}=-r(x)c(x,t)+\int_{x}^{\infty }c(y,t)r(y)f(x| y){\mathrm{d}}y,$$\end{document}∂c(x,t)∂t=−r(x)c(x,t)+∫x∞c(y,t)r(y)f(x∣y)dy,where the effective time t measures the progress or maturity of the fragmentation process, r(x) is the overall rate at which a facet of area x fragments, and f(x∣y) is the conditional probability that x is produced from the breakup of y, with y ≥ x. Inferred from this formulation are the assumptions that fragmentation occurs via a homogeneously applied external force, and independently of a facet’s shape.Breakup ratesIn order to assess the correspondence between crumpling and a fragmentation process as described by Eq. (1), two relationships must be specified: the overall breakup rate r(x), and conditional breakup probability f(x∣y), which characterize fragmentation at the scale of an individual facet. Two principles help shape our formulations of the two: firstly, a common choice of r(x) consistent with physical breakup processes is the homogeneous kernel r(x) = xλ27. Furthermore, the conditional probability f(x∣y) must satisfy area conservation:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int_{0}^{y}xf(x| y){\mathrm{d}}x=y.$$\end{document}∫0yxf(x∣y)dx=y.We use the collection of facets within each sheet as representative samples from which breakup rates may be determined. Figure 2a–c shows a typical example over three crumpling repetitions and traces the progressive fragmentation of selected facets. From such sequences, we estimate r(x) by determining the fraction of facets which fragment between two successive crumples as a function of their area x. The rates are computed separately for each sheet to ensure the same change in t elapses for all facets considered at a time. Without loss of generality, the values of x in all results are scaled so that 10 cm × 10 cm, the size of one sheet, corresponds to unit area. A breakup rate of the form r(x) = xλ appears consistent with experimental breakup data, as shown in Fig. 2d. Results for samples at other compaction ratios \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}$$\end{document}Δ~ are provided in Supplementary Fig. 6. We note one limitation of this analysis: sheets crumpled at a low compaction ratio may have too few facets for a robust sample size from which to infer a strong statistical trend; in the opposing extreme, sheets at high compaction likely undergo a cascade of multiple fragmentation events in a single crumpling iteration, and thus obscure the statistics of single breakup events. The power law relationship is motivated both by its consistency predominantly at low compaction, as well as the simplicity it affords later in our model.Fig. 2Estimation of breakup rates.a Segmentation of a sample sheet crumpled once at compaction \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}=0.27$$\end{document}Δ~=0.27, with four selected facets outlined and emphasized in white. Shown in b are the new facets that subdivide those regions after crumpling a second time (n = 2). In c, the subdivision of the facets highlighted in b is shown after n = 3. d For n = 1 (left), the proportion of facets r(x)Δt present in a as a function of their area x, which have fragmented into at least two distinct facets in b over the elapsed Δt between n = 1 and n = 2. For n = 2 (right), the fraction of facets from b which have fragmented in c. Error bars denote the standard deviation of the fragmentation probability if the fragmentation of each facet is regarded as a Bernoulli trial, with the fraction of fragmented facets taken as the success probability within each histogram bin. The dashed line corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt{x}$$\end{document}x. e The probability density function ρ(x/y) of facet areas x normalized by their parent facet’s area y from the previous crumpling iteration. The n = 1 (left) panel is the distribution of area fractions for facets in b relative to their parent facets in a, and n = 2 the corresponding distribution for facets in c relative to b. The dashed line corresponds to a fit of Eq. (3) with the fitted exponent β given.To deduce f(x∣y), it is helpful to first examine the distribution ρ(x/y) of the area fraction x/y that a child facet occupies relative to its parent. That is, if x is the area of a facet at crumpling iteration n, and y the area of its enclosing facet at iteration n−1, then ρ(x/y)d(x/y) is the probability that a facet breaks to produce a fragment that is between x/y and x/y + d(x/y) of its initial area, for a small differential element d(x/y). To account for minor misalignment between successive scans, a child facet is identified if at least half of its area lies within the contour of the candidate parent facet. The area fractions display a power law distribution, as shown in Fig. 2e, and suggest a fit to a probability density function of the form3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \left(\frac{x}{y}\right)=(\beta +1)\left(\frac{x}{y}\right)^{\beta },$$\end{document}ρxy=(β+1)xyβ,supported on x/y ∈ [0, 1]. This formulation introduces the assumption that fragmentation is a scale invariant process; while this is consistent with the present data, we note that a physical lower limit on facet area exists, and would expect deviation from scale invariant behavior as facet areas become comparable to the sheet thickness. Nevertheless, we observe clear indication of a power law relationship within our data, as shown in Fig. 2e. Extended results are provided in Supplementary Fig. 7; as previously noted, samples at high compaction undergo a succession of fragmentation events between crumples, and thus their distributions begin to depart from the power law dependence toward the more mature facet distributions of repeatedly crumpled sheets we present later, in our discussion of ridge length statistics. Taking f(x∣y) proportional to ρ(x/y) and obtaining the appropriate normalization which satisfies Eq. (2), we arrive at our final forms for the breakup rates:4a\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r(x)={x}^{\lambda },$$\end{document}r(x)=xλ,4b\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f(x| y)=\frac{1}{y}\left(\frac{\beta +2}{\beta +1}\right)\rho \left(\frac{x}{y}\right)=\frac{1}{y}(\beta +2)\left(\frac{x}{y}\right)^{\beta }.$$\end{document}f(x∣y)=1yβ+2β+1ρxy=1y(β+2)xyβ.It will prove useful to express the free parameter β as5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta =\frac{a}{2}-1.$$\end{document}β=a2−1.With this definition, we demonstrate in a following subsection that the new free parameter a corresponds to the shape parameter for the distribution of crease length.Scaling solutionWith r(x) and f(x∣y) specified, we pursue an analytical solution to the fragmentation rate equation, Eq. (1). Specifically, we seek a scaling solution independent of initial conditions, a property that allows us to solve analytically and proves compatible with the chosen form of homogeneous breakup kernels27. We thereby test a scaling ansatz c(x, t) = ϕ(ξ)/s(t)2 as proposed in Cheng and Redner27, where ξ = x/s(t), and the mean area, s(t), carries all explicit dependence on t. The scaling function ϕ(ξ) satisfies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathop{\int}\nolimits_{0}^{\infty }\phi (\xi ) {\mathrm{d}}\xi =1$$\end{document}∫0∞ϕ(ξ)dξ=1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathop{\int}\nolimits_{0}^{\infty }\xi \phi (\xi ) {\mathrm{d}}\xi =1$$\end{document}∫0∞ξϕ(ξ)dξ=1 such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathop{\int}\nolimits_{0}^{\infty }c(x,t){\mathrm{d}}x=1/s(t)$$\end{document}∫0∞c(x,t)dx=1/s(t) gives the average number of fragments, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathop{\int}\nolimits_{0}^{\infty }xc(x,t){\mathrm{d}}x=1$$\end{document}∫0∞xc(x,t)dx=1 is the total area, conserved by construction. We note that ϕ(ξ) is a valid probability density function and represents the distribution of the scaled facet area ξ. The rate equation may be solved following the procedure in Cheng and Redner27 as detailed in Supplementary Note 1; by this approach we arrive at a solution c(x, t) = ϕ(ξ)/s(t)2, valid at large t, with6a\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi (\xi )=\frac{\lambda }{{{\Gamma }}\left(\right.\frac{a}{2\lambda }\left)\right.}G(a,\lambda )\left(\right.G(a,\lambda )\xi {\left)\right.}^{\frac{a}{2}-1}{e}^{-\left(\right.G(a,\lambda )\xi {\left)\right.}^{\lambda }},$$\end{document}ϕ(ξ)=λΓa2λG(a,λ)G(a,λ)ξa2−1e−G(a,λ)ξλ,6b\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s(t)=G(a,\lambda ){t}^{-1/\lambda },$$\end{document}s(t)=G(a,λ)t−1/λ,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G(a,\lambda )={{\Gamma }}\left(\right.\frac{a+2}{2\lambda }\left)\right./{{\Gamma }}\left(\right.\frac{a}{2\lambda }\left)\right.$$\end{document}G(a,λ)=Γa+22λ/Γa2λ, and Γ(z) is the gamma function. We motivate a fixed choice of the breakup rate parameter λ = 1/2 both by its consistency with breakup statistics at low compaction, which more accurately reflect single breakup events as discussed earlier, as well as the simplification it provides to obtain an analytically tractable model. We thereby obtain the final forms7a\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi (\xi )=\frac{a(a+1)}{2{{\Gamma }}(a)}\left(\right.a(a+1)\xi {\left)\right.}^{\frac{a}{2}-1}{e}^{-\sqrt{a(a+1)\xi }},$$\end{document}ϕ(ξ)=a(a+1)2Γ(a)a(a+1)ξa2−1e−a(a+1)ξ,7b\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s(t)=\frac{a(a+1)}{{t}^{2}}.$$\end{document}s(t)=a(a+1)t2.Ridge length statisticsTo facilitate comparison with ϕ(ξ) in Eq. (7a), the area of individual facets is scaled by the mean area for that sheet and plotted as a histogram using logarithmically spaced bins. Figure 3 shows the mean curvature, hand segmentation, and scaled area distributions for a typical example from our dataset at four different crumpling iterations n. By our preliminary observations from Fig. 2e, we notice sample-to-sample variation in the parameter β (correspondingly a), which suggests a is a function of t; however, we expect weak dependence on t such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{lim}\,}_{t\to \infty }{\mathrm{d}}a/{\mathrm{d}}t=0$$\end{document}limt→∞da/dt=0, to uphold the assumptions made in solving Eq. (1). Indeed, an individual fit of a to each distribution of facet areas reveals a dependence of the form8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a(t)=\sqrt{t/\tau }$$\end{document}a(t)=t/τwith a universal parameter τ, as shown in Fig. 4a. Thus, Fig. 3c additionally shows a best fit curve to Eq. (7a) with τ ≈ 24.041 as a universal fitting parameter across all samples, and individual a and t for each sample computed by solving Eqs. (7b) and (8) self-consistently. The complete set of segmented crease patterns and fitted area distributions for all data samples is provided in Supplementary Figs. 1 and 2.Fig. 3Facet area distributions for a sample sheet.a The map of mean curvature for iterations n = 1, 2, 3, and 24 of a sample sheet crumpled with compaction ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}=0.27$$\end{document}Δ~=0.27, and b the corresponding facet segmentation. c Experimental distributions of scaled facet area ξ = x/s for each sample (scattered points) and best fit curve to Eq. (7a) (solid line). The parameter a for each sample is obtained via self-consistent calculation by Eqs. (7b) and (8), and only the universal parameter τ ≈ 24.041 is collectively fit for all samples.Fig. 4Model parameters and preliminary comparison to empirical result.a Individual fits of the shape parameter a from Eq. (7a) for each facet distribution (scattered points) alongside the best fit to Eq. (8) (dashed line), corresponding to τ ≈ 24.041. b The measured total crease length ℓmeas. of each segmented sheet plotted against the quantity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(1-\tilde{{{\Delta }}})t/(a+1)$$\end{document}(1−Δ~)t/(a+1) (scattered points). By Eq. (14), we expect the slope of this plot to correspond to ne/2, or half the average number of facets per facet. A best fit line reveals ne/2 ≈ 2.1, or ne ≈ 4.2 (dashed line). c With the results of a and b, we can make a comparison of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ell }_{\,\text{model}\,}^{(t)}\equiv \ell (t,\tilde{{{\Delta }}})$$\end{document}ℓmodel(t)≡ℓ(t,Δ~) as given by the derived relation Eq. (14), with the experimental model \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ell }_{\text{empir.}}\equiv \ell (n,\tilde{{{\Delta }}})$$\end{document}ℓempir.≡ℓ(n,Δ~) of Eq. (11). The parameters of Eq. (11) are set to c1 = 52 (normalized by 100 mm sheet size) and c2 = 0.1, comparable to the best fit values reported in Gottesman et al.16: c1 = 5200 mm, c2 = 0.063. A 1:1 reference line (dashed) is provided as a guide to the eye, and shows good agreement between the two models. Marker colors in all panels correspond to different values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}$$\end{document}Δ~, as indicated by the colorbar.The close correspondence between Eq. (7a) and experimental data supports the hypothesis that successive partitioning of the sheet’s surface into facets during crumpling evolves according to the fragmentation process described by Eq. (1). We can study the further implications of this statistical description on attributes such as the distribution of crease length, which has been explored in previous studies7–10,14,15,30. Let X be the random variable representing the area of a single facet. Following from Eq. (7a), X is distributed as9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{X}(x)=\frac{1}{2{\theta }^{2}{{\Gamma }}(a)}\left(\frac{x}{{\theta }^{2}}\right)^{\frac{a}{2}-1}{e}^{-\sqrt{x}/\theta },$$\end{document}fX(x)=12θ2Γ(a)xθ2a2−1e−x/θ,with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta =\sqrt{s/a(a+1)}=1/t$$\end{document}θ=s/a(a+1)=1/t by consequence of Eq. (7b), and mean area s. Let Y be a random variable representing the edge length of a facet in the ridge network. If Y scales as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt{X}$$\end{document}X, the consequent distribution of Y is a gamma distribution,10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{Y}(y)=\frac{1}{\theta {{\Gamma }}(a)}\left(\frac{y}{\theta }\right)^{a-1}{e}^{-y/\theta },$$\end{document}fY(y)=1θΓ(a)yθa−1e−y/θ,with θ the scale and a the shape parameter, alluded to in our discussion of breakup rates, and with mean edge length aθ. The distributions of facet area and edge length provided by Eqs. (9) and (10) allow us to formulate an expression for the typical total crease length as a function of t, in tandem with the evolution of mean area s(t). First, we briefly restate the key empirical result of Gottesman et al.16 to which we will compare our model. The total crease length ℓ was found to vary according to a logarithm of the number of crumpling and unfolding repetitions n:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ell }_{\text{empir.}}\equiv \ell (n,\tilde{{{\Delta }}})={c}_{1}(1-\tilde{{{\Delta }}}){\mathrm{log}}\,\left(1+\frac{{c}_{2}n}{\tilde{{{\Delta }}}}\right),$$\end{document}ℓempir.≡ℓ(n,Δ~)=c1(1−Δ~)log1+c2nΔ~,with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}$$\end{document}Δ~ the compaction ratio, and c1 and c2 fitting parameters. A striking property of this model is its implication that the rate at which new damage accumulates, as measured by added crease length per crumpling iteration δℓempir. ≡ ∂ℓ/∂n, is independent of the details of the sheet’s preparation:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta {\ell }_{\text{empir.}}=\frac{{c}_{1}{c}_{2}\left(\right.1-\tilde{{{\Delta }}}\left)\right.}{\tilde{{{\Delta }}}}\exp \left(-\frac{\ell }{{c}_{1}(1-\tilde{{{\Delta }}})}\right).$$\end{document}δℓempir.=c1c21−Δ~Δ~exp−ℓc1(1−Δ~).We observe from Eq. (12) that the added crease length δℓempir. is uniquely determined by a sheet’s instantaneous state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\ell ,\tilde{{{\Delta }}})$$\end{document}(ℓ,Δ~); moreover, the model is independent of the details of the crease network, such as the spatial homogeneity of damage across the sheet. The fitting parameters c1 and c2 are universal to all values of n and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}$$\end{document}Δ~. The facet segmentation of each crease pattern provides a second measurable quantity, d, equal to the sum of all interior perimeters of facets; i.e., the total length of all edges shared between two facets. We expect d and ℓ to be proportional, with differences arising due to incomplete scarring around facet perimeters as regions of the sheet restore elastically, particularly for mild compression. We find that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(1-\tilde{{{\Delta }}})d$$\end{document}(1−Δ~)d accomplishes the desired proportionality, and define \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ell }_{\text{meas.}}\equiv (1-\tilde{{{\Delta }}}){d}_{\text{meas.}}$$\end{document}ℓmeas.≡(1−Δ~)dmeas. to be the measured total crease length obtained from our segmented data. Next, working with the moments of our derived facet area and edge length distributions, we can estimate d analytically as the average length of an edge, aθ, times the average number of edges. The latter may be expressed as the average number of facets, or the total sheet area divided by the typical facet area s, times the number of edges per facet ne, halved to account for shared edges, which yields13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{\text{model}}=\frac{a\theta (t)}{s(t)}\times \frac{{n}_{e}}{2}=\frac{{n}_{e}t}{2(a+1)}.$$\end{document}dmodel=aθ(t)s(t)×ne2=net2(a+1).Thus, we obtain that14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ell }_{\,\text{model}\,}^{(t)}\equiv \ell (t,\tilde{{{\Delta }}})=(1-\tilde{{{\Delta }}})\frac{{n}_{e}t}{2(a+1)}.$$\end{document}ℓmodel(t)≡ℓ(t,Δ~)=(1−Δ~)net2(a+1).Here, the superscript (t) denotes the explicit dependence of ℓmodel on t; in a following subsection, we develop a connection between t and n that allows ℓmodel to be expressed in terms of n and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}$$\end{document}Δ~, mirroring Eq. (11). A fit of Eq. (14) to the measured length ℓmeas. reveals a value of ne ≈ 4.2 as shown in Fig. 4b, which suggests an average of 4–5 sides per facet. Finally, Fig. 4c demonstrates the agreement between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ell }_{\,\text{model}\,}^{(t)}$$\end{document}ℓmodel(t) of Eq. (14), and ℓempir. of Eq. (11).Numerical evidence for the insensitivity to initial preparationNow that the connection between the statistical model of facet area and total crease length has been presented, we briefly note on the insight that may be gained by additionally solving Eq. (1) numerically. A numerical integration scheme is implemented using second-order composite trapezoid rule for discretization in x, and second-order implicit multi-step discretization in t. The sample numerical result in Fig. 5 reveals a rapid convergence to the steady state analytical solution given by Eqs. (7a) and (7b), and thereby relative insensitivity to the initial state. To demonstrate the significance of this behavior, we reiterate the observed history independence of total crease length. As discussed in Gottesman et al.16, sheets with different loading histories—one hand-crumpled and another deliberately folded along straight lines—yet nearly equal total crease lengths exhibited the same subsequent accumulation of damage when subjected to the protocol of Fig. 1a. Such sheets had clearly distinct initial facet area distributions: The facet areas of the deliberately folded sheet were sharply peaked near two different values, while those of the hand-crumpled sheet were broadly distributed. Thus, signatures of initial preparation appear to be quickly eclipsed by the strong attractor of the crumpled state, echoed in the rapid convergence to steady state seen numerically.Fig. 5Numerical validation of the analytical solution to Eq. (1).a Selected snapshots of the numerically calculated ϕnum(ξ) with initial condition c(x, 0) = δ(x − 1) and with a = 1, revealing a rapid convergence to the steady state distribution. The dashed line corresponds to the analytical form of Eq. (7a) valid at large t. b The corresponding evolution of mean area s(t), with the analytical solution at large t given by Eq. (7b) shown by the dashed line. c The mean area of the experimental samples as a function of t computed from Eq. (7b) (scattered points). The dashed line corresponds to Eq. (7b) with a(t) as given by Eq. (8). Marker colors correspond to different values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}$$\end{document}Δ~, as indicated by the colorbar.Thus far, we have established that an estimate of total crease length constructed from moments of the derived facet area and ridge length distributions shows consistency with the logarithmic scaling of Eq. (11). In the following section, we propose a simple mechanism for how the geometric incompatibility of a folded sheet and its confinement leads to further fragmentation, driving t forward. This argument establishes the evolution of t in accordance with n, and thus supplies the missing link to a physically based model that corroborates experimental findings.One-dimensional modelTo offer an explanation for the observed logarithmic scaling, we develop a simple one-dimensional model that proposes how additional fragments may form when a crumpled sheet is re-crumpled, relying on the statistical descriptions of facet area and segment length formulated in the previous sections. Our goal can be summarized by the following two questions: (1) Given its current state and prescribed confinement, with what probability does a sheet undergo further fragmentation? (2) How does this probability relate to the continuous variables in the fragmentation model of Eq. (1)? First, we appeal to the axial symmetry of our confinement to simplify our view of crumpling to a 1D strip of length L0, as shown in Fig. 6. The strip is characterized by a sequence of folds in alternating directions which divide the strip into random segments. The lengths r of the segments, which are equal to the cross-sections of the intercepted facets, are distributed according to the derived gamma distribution of Eq. (10), weighted by the horizontal facet width, which increases the likelihood of a facet’s occurrence within a randomly selected vertical strip. For facets of ~1:1 aspect ratio, the distribution of segment length is thereby15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{R}(r)=\frac{1}{\theta {{\Gamma }}(a+1)}\left(\frac{r}{\theta }\right)^{a}{e}^{-r/\theta },$$\end{document}fR(r)=1θΓ(a+1)rθae−r/θ,with the average segment length given by (a + 1)θ. A comparison of Eq. (15) with experimental data is provided for a strongly compacted sample in Fig. 6a, b, with extended results for all samples presented in Supplementary Fig. 5.Fig. 6A folded cross-section considered as a one-dimensional random walk.a A sample segmented sheet with dashed line indicating a vertical cross section. b The distribution of segment lengths from all such cross-sections of the sheet in a (filled points), with Eq. (15) plotted as a solid curve. No additional fit is performed; the value of the shape parameter a which appears in Eq. (15) is uniquely determined from Eq. (8) and the best fit τ to the facet area distributions. c A schematic of the analog between the folding of a one-dimensional strip in an axially confined sheet and a one-dimensional random walk whose time axis is extended vertically for clarity. The filled curve represents the distribution of the walker’s final displacement, with darker shaded regions denoting the fraction of walks which lie outside a given confinement. d Simplified illustration of one-dimensional folding which facilitates a geometric estimate of the critical confinement w, further detailed in Supplementary Note 4.As a preliminary step, we derive the final displacement of the strip when folded at each break, in the absence of confinement. This problem can be mapped to the displacement of a walker performing a one-dimensional random walk with gamma-distributed steps. To enforce the concept of folding, the walker’s steps occur in alternating directions. The distribution fZ(z) of position Z after 2k steps accurate for all k is derived in full in Supplementary Note 2. However, the salient trends may be likewise observed by applying the central limit theorem and considering the position Z valid for large k, or small step size, which gives16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{Z}(z;\theta )=\frac{1}{\sqrt{2\pi {L}_{0}\theta }}\exp \left(-\frac{{z}^{2}}{2{L}_{0}\theta }\right),$$\end{document}fZ(z;θ)=12πL0θexp−z22L0θ,and describes a normal distribution of zero mean and variance L0θ.If a confinement is now introduced at the locations ∣z∣ = w, we next ask with what likelihood the walker steps beyond this confinement. One approach to approximate this probability is to integrate Eq. (16) for all ∣z∣ > w, producing a two-sided survival function of Eq. (16). Although this is not equivalent to our initial question, as intermediate steps may also have landed past ∣z∣ > w, it proves an acceptable estimate as the last step has the greatest variance. A more accurate calculation would be to evaluate the likelihood that a given walk escapes the confinement at any step; however, looking at the last step is useful for its simplicity in analytical form, and still captures the anticipated behavior. A comparison to the more accurate formulation is made numerically and provided in Supplementary Fig. 9. Once again, we pursue here the simpler form of the survival function valid for large k, and refer to Supplementary Note 2 for the exact derivation valid at all k. The survival function of Eq. (16), SZ(w; θ) = P(∣Z∣ > w; w≥0), for a threshold confinement w, is given by17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${S}_{Z}(w;\theta )=1-\,{\text{erf}}\,\left(\frac{w}{\sqrt{2{L}_{0}\theta }}\right),$$\end{document}SZ(w;θ)=1−erfw2L0θ,where erf(z) is the error function. In order for walkers at ∣z∣ > w to be restored within the limits of confinement, one or more of their steps must fragment, thereby increasing the number of steps taken and decreasing the overall average, which drives the evolution of fragmentation. This articulates our key claim: considering our original, cylindrically shaped sheets as a statistical ensemble of one-dimensional random walks, we suggest that the progression of fragmentation measured by a change dt, over a single crumpling iteration dn, should be proportional to the fraction of walks in the ensemble which leave the confinement imposed at ∣z∣ = w: dt/dn ~ SZ(w; θ). Equivalently, this is the likelihood that a single random walk leaves the critical confinement. We note that this resulting fragmentation rate describes an average fragmentation likelihood given only a confinement w and current temporal parameter t = 1/θ describing the maturity of the fragmentation process thus far; it does not enforce direct correlations between successive crumpling iterations, whereby new folds should occur preferentially along previous ones. Instead, the decrease in fragmentation rate with n is encoded through the decreasing mean facet area with t. Moreover, while stronger correlation is expected between walks representing nearby transects of the sheet, here we consider the statistical behavior of the sheet as a whole, and account for the increased fragmentation likelihood for facets with larger horizontal extent through the weighting introduced in Eq. (15). At present, Eq. (17) gives the likelihood that new creases will form; however, it does not yet describe how much new damage is created, for which two additional factors should be considered: (1) When the sheet is strongly confined in closely packed layers, the layers tend to collectively fragment, as alluded to by Sultan and Boudaoud15 and Gottesman et al.16, thus contributing a factor p ~ 1/L such that halving the final height doubles the number of additional ridges. (2) In the opposite limit of low compaction, facets are not in close proximity and need not behave cooperatively; thus, new damage scales linearly with the amount of compression L0−L, as argued in Gottesman et al.16. With these additional considerations, we propose that the evolution of the fragmentation process with crumpling iteration behaves as18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta t\equiv \frac{\partial t}{\partial n}=\alpha \frac{1-\tilde{{{\Delta }}}}{\tilde{{{\Delta }}}}{S}_{Z}(w;t),$$\end{document}δt≡∂t∂n=α1−Δ~Δ~SZ(w;t),where α is a fitted constant of proportionality. We indicate the explicit dependence on t here, as t and θ are inversely related. The critical width w is determined by the geometry of the imposed confinement, as illustrated in Fig. 6d; a complete derivation is provided in Supplementary Note 4:19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w(\tilde{{{\Delta }}})\approx \frac{R}{\sqrt{1-{\tilde{{{\Delta }}}}^{2}}},$$\end{document}w(Δ~)≈R1−Δ~2,where R is the radius of the container. By consequence of Eq. (14) we can directly relate Eqs. (18) and (12) as20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta {\ell }_{\text{model}}=\frac{d{\ell }_{\text{model}\,}^{(t)}}{dt}\delta t=\frac{(1-\tilde{{{\Delta }}}){n}_{e}}{2(a+1)}\delta t$$\end{document}δℓmodel=dℓmodel(t)dtδt=(1−Δ~)ne2(a+1)δtand obtain a fit to the proportionality constant α. By performing an asymptotic approximation in the limit of large t, detailed in Supplementary Note 3, Eq. (18) may be analytically integrated to provide a scaling relation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t(n,\tilde{{{\Delta }}})$$\end{document}t(n,Δ~) which bears similarity to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ell (n,\tilde{{{\Delta }}})$$\end{document}ℓ(n,Δ~) of Eq. (11):21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t(n,\tilde{\Delta })={\tilde{c}}_{1}(1-{\tilde{\Delta }}^{2}){\mathrm{log}}\,\left(1+\frac{{\tilde{c}}_{2}n}{\tilde{\Delta }(1+\tilde{\Delta })}\right),$$\end{document}t(n,Δ~)=c~1(1−Δ~2)log1+c~2nΔ~(1+Δ~),where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tilde{c}}_{1}=2{L}_{0}/{R}^{2}$$\end{document}c~1=2L0/R2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tilde{c}}_{2}=\alpha {R}^{2}/{L}_{0}\sqrt{2\pi }$$\end{document}c~2=αR2/L02π, and L0 and R are the sheet length (equivalently the confining container height) and container radius, respectively. Taken together, Eqs. (14) and (21) thereby provide a theoretically motivated expression \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ell }_{\,\text{model}\,}^{(n)}\equiv \ell (t(n,\tilde{{{\Delta }}}),\tilde{{{\Delta }}})$$\end{document}ℓmodel(n)≡ℓ(t(n,Δ~),Δ~) based on properties of fragmentation kinetics and a simple mechanism for re-fragmentation formulated as a random walk. Figure 7 compares the agreement of the empirical relations δℓempir. and ℓempir., as well as the derived models δℓmodel and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ell }_{\,\text{model}\,}^{(n)}$$\end{document}ℓmodel(n), with the measured quantities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta {\ell }_{\text{meas.}}\equiv {\ell }_{\text{meas.}\,}^{(n)}-{\ell }_{\,\text{meas.}\,}^{(n-1)}$$\end{document}δℓmeas.≡ℓmeas.(n)−ℓmeas.(n−1) and ℓmeas. for various n. Collectively, the results of Figs. 4 and 7 demonstrate clear consistency of the fragmentation model with the anticipated logarithmic growth.Fig. 7Validation of empirical and derived models for crease length evolution with measurement.a Predicted change in total crease length δℓempir. given by Eq. (12) plotted against the measured change in crease length between two successive crumples, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta {\ell }_{\text{meas.}}\equiv {\ell }_{\text{meas.}\,}^{(n)}-{\ell }_{\,\text{meas.}\,}^{(n-1)}$$\end{document}δℓmeas.≡ℓmeas.(n)−ℓmeas.(n−1). Open markers correspond to manually segmented data consistent with prior results presented in this work, while filled circles correspond to data which was processed using the automated segmentation as detailed in the Supplementary Methods. A 1:1 reference line (dashed) is provided as a guide to the eye. b The total crease length ℓempir. given by Eq. (11) plotted against the measured total crease length ℓmeas.. c The change in total crease length δℓmodel as predicted by Eq. (20), and d the total crease length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\ell }_{\,\text{model}\,}^{(n)}$$\end{document}ℓmodel(n) obtained from Eqs. (14) and (21), against their corresponding measured values, in direct comparison to a and b. Marker colors in all panels correspond to different values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}$$\end{document}Δ~, as indicated by the colorbar. We see that both the empirical and derived relations for δℓ and ℓ serve as strong models of measured data, and affirm the suitability of a logarithmic relationship to describe damage evolution in this system.DiscussionBy pursuing a correspondence between the crumpling of a thin sheet and a general fragmentation process, we have derived a physically based framework for the evolution of statistical properties of intricate crumpled patterns. Equipped with theoretical models in close agreement with experimental data, we have proposed a simple model of one-dimensional folding in which further fragmentation ensues due to a geometric incompatibility between the sequence of folds and the imposed confinement, likened to a random walk exceeding a critical allowed displacement. The predicted accrual of damage, quantified by added crease length, shows strong consistency with the logarithmic model of Gottesman et al.16, and thereby supplies a possible physical basis for the puzzling origin of logarithmic scaling in repeated crumpling experiments. Furthermore, our model explains the history independence of the logarithmic scaling, since the area distribution of the crumpled state is such a strong attractor in the fragmentation process.The consistency of crumpling with fragmentation theory hints at the possibility of universal behavior uniting more diverse fragmenting systems. For example, the activation of defects in the fragmentation of ceramics can locally slow down subsequent fracture, and may bear semblance to the slowing of damage accumulation as a re-crumpled sheet exploits its existing folds31. Thus, studies of crumpled systems might offer a new lens through which to interpret other complex processes. An immediate extension of this work would be a validation of the results on sheets of varied thicknesses and material parameters, as well as those prepared according to different compaction protocols. One simplifying assumption of our analysis is that fragmentation of facets is a scale invariant process over the range of areas considered; however, this assumption starts to break down particularly for large crumpling iterations n. The work of Lechenault et al.32 offers a compelling approach for identifying this limit by considering the energetic competition between bending of facets and rigid folding along existing creases, with energy cost of the former proportional to the sheet’s bending rigidity, and the latter proportional to crease stiffness. The energy balance of these competing deformations provides a characteristic length scale which varies in proportion to the sheet thickness. This improvement to the current work would strongly benefit from further studies over a range of material parameters. Length scales of folds in crumpled systems have also been studied in the context of thermally driven dynamics, and it may thus be useful to draw possible connections to statistical mechanical models of crumpling33–35. Moreover, it may be of value to explore slight generalizations of proposed functional forms introduced in this study, such as the breakup rates; this could allow variations across other experimental results to be explained, such as those arising between low and high compaction regimes12,15, thereby providing a unifying framework for such observations.Additionally, deeper understanding of crumpling dynamics can assist data-driven approaches to predicting damage network formation. Though machine learning methods are capable of unveiling hidden structure in complex, disordered systems36,37, prior work has demonstrated the importance of preserving physical properties in making faithful predictions: for example, preserving vertex angle constraints in synthetic fold patterns to assist the task of ridge network reconstruction in crumpled sheets38. In addition to encoding physical rules implicitly through data, future machine learning approaches may explicitly enforce constraints such as facet area and crease length statistics in predicting ridge network evolution. Strategies which couple detailed spatial data with coarse-grained theoretical insight could thus enable more comprehensive predictions of crumpling dynamics in future studies.MethodsThe data analyzed in this work was collected for the study of Gottesman et al.16; here, we briefly summarize the experimental protocol for reference. In total, 10 cm × 10 cm Mylar sheets are rolled into a 3-cm diameter cylindrical container and compressed uniaxially to a specified compaction ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}=L/{L}_{0}$$\end{document}Δ~=L/L0, the ratio of final to initial height, with L0 = 10 cm (Fig. 1a). The resulting ridge network inscribed on each sheet is extracted by carefully unfolding and scanning the sheet using a custom laser profilometer, which produces a height map of the sheet. A two-dimensional map of mean curvature is determined from the spatial gradients of the height profile; sharp peaks in curvature mark the signature of a ridge (Fig. 1b). Successive re-crumpling and scanning of a single sheet is performed n times up to n = 24. Individual facets, characterized as contiguous regions of near-zero curvature, are delineated as shown in Fig. 1c. Due to noise and artifacts in data collection, not all facets are completely enclosed by a contour of ridges; breaks along a ridge, or smoothing out and softening of ridges, occur inevitably during re-crumpling and unfolding. Automated methods of crease detection and facet labeling were initially tested to perform the segmentation; however, these methods proved sensitive to noise and thus were prone to over-fragmenting the sheets. Each sample presented and analyzed in this work was digitally labeled by hand. Additional details of automated segmentation are provided in the Supplementary Methods, and resulting segmentations and facet area distributions are shown in Supplementary Figs. 3 and 4. With manual segmentation, care was taken to identify not only the dominant lines of each pattern as seen in the examples, but also the less pronounced softer scarring. The segmentation was performed for sheets after iterations n = 1, 2, and 3 at seven different compaction ratios: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}=0.63,0.45,0.36,0.27,0.18,0.09$$\end{document}Δ~=0.63,0.45,0.36,0.27,0.18,0.09, and 0.045. Each series of successive crumples was compared across all iterations n for consistency, to ensure that labeled facets from earlier iterations persist in later ones. Samples with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{{{\Delta }}}=0.63,0.45$$\end{document}Δ~=0.63,0.45, and 0.27 were also labeled after n = 24 crumples, for a total of 24 samples overall. We acknowledge that samples at n = 24 are more prone to missing detail as older scarring is obscured by newer ridges, but are nonetheless valuable to the study. The results of manual segmentation and corresponding facet area distributions are provided in Supplementary Figs. 1 and 2.Supplementary informationSupplementary InformationPeer Review File | nature communications | [
"Article"
] | [
"Applied mathematics",
"Scientific data",
"Statistical physics, thermodynamics and nonlinear dynamics"
] | IntroductionCrumpling complex non-equilibrium process in diverse systems length scales microscopic crumpling graphene membranes1 to macroscopic folding viscoelastic crust2. Crumpled structures porous for high-performance batteries supercapacitors increasing electrochemical surface area3,4 Controlled crumpling electronic optical surface properties in graphene films1 understanding mechanics crumpling essential flexibility shape conformation integral in design thin wearable devices5,6 complete understanding crumpling dynamics elusive complexity statistical properties of crumpled geometries reproducible in experiment7–13 confirmed via simulation14 complex process dictated by thin sheets topology self-avoidance13 Gottesman et al revealed unexpected order ridge network evolution in crumpled sheets repeated compaction unfolding intricate details ridge networks reduced to single total crease length evolves crumpling repetitions degrees compaction incremental damage added upon re-crumpling independent of crumpling history added crease length determined by current total crease length new compaction depthprocesses evolve logarithmically observed in disordered systems including stress strain relaxation compacted sheet17–19 conductance relaxation disordered electronic systems20 creep dynamics granular suspensions21 emergence logarithmic model damage evolution in crumpled sheets distinct limited physical justification.Fig. 1Data processing L0 × L0 Mylar sheet uniaxially compressed to compaction ratio\documentclass[12pt{minimal}{amsmath{wasysym}-69pt=L/{L}_{0{document}Δ~=L/L0 unfolded height profile scanned using laser profilometer for n iterations. mean curvature from height profiles of two sheets at different[12pt]{minimal}{amsmath{wasysym{upgreek-69pt Red blue colors denote folds in opposite directions facet segmentation of b colored randomlywork novel approach characterize crumpling explanation logarithmic model correspondence between fragmentation processes Fragmentation models theoretical industrial modeling collision fracture phenomena26 concentrate on theoretical rate equation time-dependent fragmentation Cheng and Redner27 framework for successive homogeneous breakups non-local stresses model applied polymer degradation28 volcanic fragments eruption29 first application crumpling work derive scaling solution to rate equation Cheng Redner27 decomposes time-invariant distribution of scaled facet area time-dependent evolution of mean facet size derived area distribution reproduces statistical features experimental crumpled patterns Fragment distributions natural point comparison between theory experiment additional correspondence in temporal evolution patterns temporal parameter evolution mean facet size clock maturity fragmentation process connect to experimental parameters fragmentation number crumpling iterations compaction strength experiments Gottesman et al construct geometric model crumpling to random walk informed by statistical properties derived area distribution analytical relation geometric frustration confined random walk instigates new damage advances temporal fragmentation maturity approach recover logarithmic evolution of damage in ridge networks Gottesman et al.explain history of damage formation missing physical basis for unexpected result key idea model is extension of fragmentation theory feedback loop facets smaller make sheet compliant lower fragmentation idea may to systems accumulation damage inhibits further damage work shows fragmentation theory suggests universal damage evolution in crumpling analogs in other systems.ResultsThe collection processing of experimental crumpling data results detailed in Methods section Crease patterns from uniaxially compressed Mylar sheets in Fig. 1b segmented into facets Fig. 1c samples vary in compaction ratio ratio of final to initial height number of successive crumples of same sheet 24 segmented crease patterns analyzed 7 compaction ratios n = 1 2 3 24 crumpling iterations to fragmentation in crumpling as partitioning of thin sheet into smaller flat facets separated by ridges or creases construction model begin with general theory of fragmentation kinetics Cheng and Redner27x facet area c(x, t concentration of facets time t linear integro-differential equation evolution of c(x, t) given\documentclass[12pt]{minimal\usepackage{amsmath\oddsidemargin-69pt{document}\partial c(x,t)=-r(x)c(x,t)+{x\infty }c(y,t)r(y)f(x y)\mathrm{d}}y{document}∂c(x,t=−r(x)c(x,t)+∫x∞c(y,t)r(y)f(x∣y)dy effective time t progress fragmentation process r(x) rate facet x fragments f(x∣y) conditional probability x produced from breakup y y ≥ x fragmentation occurs via homogeneously external force independently of facet’s shape.Breakup assess correspondence crumpling fragmentation process two relationships overall breakup rate r(x), conditional breakup probability f(x∣y), characterize fragmentation individual facet principles choice r(x) homogeneous kernel r(x) = xλ27.conditional probability f(x∣y) satisfy area conservation\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}}xf(x y=y{document}∫0yxf(x∣y)dx=y collection facets each sheet as representative samples breakup rates Figure 2a–c shows example three crumpling repetitions progressive fragmentation of facets estimate r(x) facets between two crumples function of area x rates computed separately for each sheet same change all facets values x scaled 10 cm × 10 cm corresponds to unit area breakup rate r(x) = xλ consistent with experimental breakup data in Fig. 2d. Results for samples at other compaction ratios\documentclass[12pt]{minimal}{amsmath}{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}{document}\tilde\Delta\end{document provided in Supplementary Fig.limitation analysis sheets crumpled low compaction few facets for sample strong statistical trend sheets high compaction multiple fragmentation events obscure statistics single breakup events power law relationship motivated by consistency at low compaction simplicity later in model.Fig. 2Estimation breakup rates Segmentation sample sheet crumpled at compaction\documentclass[12pt{minimal{amsmath{upgreek\oddsidemargin-69pt}{document}.27~=0.27 four facets outlined emphasized in white new facets after crumpling second time (n = 2) c subdivision facets after n = 3. n = 1 proportion of facets r(x)Δt fragmented into two distinct facets in b over Δt between n = 1 and n = 2. n = 2 (right), fraction of facets from b fragmented in c Error bars denote standard deviation fragmentation probability Bernoulli trial fraction of fragmented facets success probability each histogram bindashed line corresponds to[12pt{minimal{amsmath\oddsidemargin-69pt} probability density function ρ(x/y) of facet areas x normalized by parent area y previous iteration n = 1 panel distribution area fractions for facets b parent a n = 2 facets c relative b dashed line corresponds to Eq. (3) fitted exponent β deduce f(x∣y), examine distribution ρ(x/y) of area fraction child facet relative to parent x area facet at crumpling iteration n y enclosing facet at iteration n−1 ρ(x/y)d(x/y) probability facet breaks produce fragment between x/y and + d(x/y) initial area small differential element d(x/y). child facet identified if half area within contour parent facet area fractions display power law distribution in Fig.suggest fit probability density function form3\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{upgreek\setlength\oddsidemargin-69pt}{document}$$\rho{x=\beta +1)\end{document}ρxy=(β+1)xyβ x/y ∈ [0, 1] formulation assumption fragmentation scale invariant process consistent present data physical lower limit facet area exists expect deviation scale invariant behavior facet areas comparable sheet thickness power law relationship data Fig. 2e Extended results Supplementary Fig. 7 samples high compaction undergo fragmentation events between crumples distributions depart power law dependence mature facet distributions repeatedly crumpled sheets ridge length statistics f(x∣y) proportional to ρ(x/y) appropriate normalization satisfies Eq.final forms breakup rates:4a\documentclass[12pt]{minimal{amsmath{wasysym\setlength\oddsidemargin{-69pt}\begin{document}$$r(x)={x}\lambda\end{document}r(x)=xλ,4b\documentclass[12pt{amsmath{wasysym{mathrsfs}{upgreek}{-69pt}\begin{document}$$f(x y)=\frac{1}{y}\left(\frac{\beta +2} +1}\right)\frac{x}{y})=\frac{1}{y} +2)\end{document}f(x∣y)=1yβ+2β+1ρxy=1y(β+2)xyβ express free parameter β as5\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek}\setlength\oddsidemargin}{-69pt}\begin{document}$$\beta =\frac{a}{2}-1.}β=a2−1 demonstrate new free parameter a corresponds shape parameter distribution crease length.Scaling r(x) f(x∣y) pursue analytical solution fragmentation rate equation seek scaling solution independent initial conditions compatible homogeneous breakup kernels27 test scaling ansatz c(x, t) = φ(ξ)/s(t)2 Cheng Redner27 ξ = x/s(t), mean area s(t), dependence tscaling function φ(ξ) satisfies[12pt]{minimal\usepackage{amsmath{wasysym\oddsidemargin}{-69pt}{document}\mathop\nolimits\infty\phi\xi\mathrm{d}}\xi =1\end{document}∫0∞φ(ξ)dξ=1[12pt]{minimal}{amsmath{wasysym{upgreek}\oddsidemargin}{-69pt}{document}\mathop\nolimits\infty\xi \phi\mathrm{d}}\xi =1\end{document}∫0∞ξφ(ξ)dξ=1[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}\mathop{\nolimits\infty(x,t\mathrm{d}}x=1/s(t)\end{document}∫0∞c(x,t)dx=1/s(t)average number fragments\documentclass[12pt]{minimal\usepackage{amsmath-69pt}{document}$$\mathop{\int}\nolimits{0\infty }xc(x,t)\mathrm{d}}x=1\end{document}∫0∞xc(x,t)dx=1 total area conserved construction φ(ξ) valid probability density function distribution scaled facet area ξ rate equation solved procedure Cheng and Redner27 Supplementary Note 1 solution c(x, t) = φ(ξ)/s(t)2 valid large t\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\setlength-69pt}{document}$$\phi (\xi )=\frac{\lambda\Gamma }}\left(\right.\frac{a\lambda\right.(a,\lambda )\xi\right\frac{a}{2}-1}\left(\right.G(a\lambda )\xi\right.{\lambda }}\end{document}φ(ξ)=λΓa2λG(a)ξa2−1e−G(a,λ)ξλ\documentclass[12pt]{minimal}\usepackage{amsmath\oddsidemargin{-69pt}{document}$$s(t)=G(a,\lambda ){-1/\lambda }\end{document}s(t)=G(a,λ)t−1/λ[12pt]{minimal}{amsmath\oddsidemargin}{-69pt}{document}$$G(a,\lambda )={{\Gamma }}\left.\frac{a+2}{2\lambda }\right\Gamma\end{document}G(a,λ)=Γa+22λ/Γa2λ Γ(z) is gamma function motivate fixed choice breakup rate parameter λ = 1/2 consistency with breakup statistics low compaction single breakup events simplification analytically tractable model.obtain final forms7a\documentclass[12pt]{minimal{amsmath-69pt}}\phi\xi=\frac{a(a+1)}\Gamma\left.a(a+1)\xi\frac{a{2}-1}\sqrt{a(a+1)\xi\end{document}φ(ξ)=a(a+1)2Γ(a)a(a+1)ξa2−1e−a(a+1)ξ\documentclass[12pt]{minimal{amsmath\oddsidemargin-69pt}{document$s(t)=\frac{a(a+1)}{{t}{2}}\end{document}s(t)=a(a+1)t2.Ridge length comparison with φ(ξ) Eq. (7a), area facets scaled by mean area plotted histogram logarithmically spaced bins Figure 3 shows mean curvature hand segmentation scaled area distributions typical example dataset four crumpling iterations preliminary observations Fig.notice sample-to-sample variation in parameter β suggests function of t expect weak dependence on t\documentclass[12pt]{minimal{amsmath-69pt{lim}limt→∞da/dt=0 uphold assumptions solving Eq. (1) individual fit of a to each distribution facet areas reveals of form8[12pt{amsmath-69pt(t)={t/}a(t)=t/τwith universal parameter τ Fig. 4a Fig. 3c shows best fit curve to Eq. (7a) with τ ≈ 24.041 universal fitting parameter across all samples individual a and t for each sample computed by solving Eqs. (7b) and (8) complete segmented crease patterns fitted area distributions for data samples in Supplementary Figs. 1 and 2.Fig. 3Facet area distributions for sample sheetmap mean curvature iterations 1 2 3 24 sample sheet crumpled compaction ratio[12pt{minimal{amsmath{wasysym-69pt.27.27 corresponding facet segmentation Experimental distributions scaled facet area ξ = x/s each sample points best fit curve to Eq. (7a) (solid parameter a sample obtained self calculation Eqs. (7b) (8) universal parameter τ ≈ 24.041 collectively fit all samples.Fig. 4Model parameters comparison empirical result Individual fits shape parameter a Eq. (7a) each facet distribution best fit Eq. (8) corresponding τ ≈ 24.041 measured total crease length lmeas. each segmented sheet plotted against quantity[12pt]{minimal{amsmath{wasysym-69pt)t/(a+1){document}(1−Δ~)t/(a+1) (scattered points). Eq.expect slope plot ne/2 half average facets best fit line reveals ne/2 ≈ 2.1 or ne ≈ 4.2 (dashed line). results a b comparison\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek-69pt{document}\ell{model}\equiv \ell\tilde\Delta\end{document}lmodel(t)≡l(t,Δ~) derived relation Eq. (14) experimental model[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin-69pt{document}\ell{empir\equiv \ell (n\tilde\Delta\end{document.≡l(n,Δ~) Eq. (11) parameters Eq. (11) c1 = 52 (normalized 100 mm sheet size c2 = 0.1 comparable best fit values Gottesman et al.16 c1 = 5200 mm, c2 = 0.063 1:1 reference line (dashed) guide shows agreement modelsMarker colors panels correspond values[12pt{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek{\oddsidemargin{-69pt}{document}\Delta\end{document}Δ~ indicated colorbar correspondence between Eq. (7a) experimental data supports hypothesis partitioning surface facets during crumpling evolves fragmentation process Eq. (1) study implications statistical description distribution crease length explored previous studies7–10,14,15,30. X random variable representing area single facet Eq.(7a), X distributed\documentclass[12pt]{minimal{amsmath\oddsidemargin-69pt}{document}{X}(x)=\frac{1}{2{\theta^{2\Gamma(a\left\frac{x}{{\theta{2}}\right\frac{a}{2}-1}{e}{-\sqrt{x}/\theta\end{document}fX(x)=12θ2Γ(a)xθ2a2−1e−x/θ[12pt]{minimal}{amsmath}\oddsidemargin{-69pt}\begin{document}\theta =\sqrt{s/a(a+1)}=1/t\end{document}θ=s/a(a+1)=1/t consequence Eq. (7b), mean area s Y random variable edge length facet ridge networkY scales\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}\end{document distribution Y gamma distribution[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}-69pt}{document}${f}{Y}(y)=\frac{1}{\theta\Gamma(a)\left{y\theta^{a-1}{e}^{-y/\theta }\end{document}fY(y)=1θΓ(a)yθa−1e−y/θ θ scale a shape parameter breakup rates mean edge length aθ distributions of facet area edge length Eqs. (9) (10) formulate expression typical total crease length function of t evolution mean area s(t). restate empirical result Gottesman et al.16 compare modeltotal crease length crumpling unfolding repetitions\documentclass[12pt]{minimal}{amsmath}{wasysym{upgreek\oddsidemargin}-69pt}\begin{document}\ell\text{empir.}}\equiv\tilde{{{\Delta)={c}_{1}(1-\tilde\Delta)\mathrm{log}}\left(1+\frac{{c}_{2}n}{\tilde{{{\Delta\right),\end{document}lempir.≡l(n,Δ~)=c1(1−Δ~)log1+c2nΔ~\documentclass[12pt]{minimal}{amsmath}{wasysym}{upgreek}\oddsidemargin}{-69pt}{document}\tilde{{{\Delta }}}\end{document}Δ compaction ratio c1 c2 fitting parameters model new damage accumulates measured added crease length per crumpling iteration∂l/∂n independent preparation:12\documentclass[12pt]{minimal}{amsmath{wasysym\oddsidemargin}{-69pt}\begin{document}$$\delta {\ell\text{empir.}}=\frac{{c}_{1}}_{2}\left(.1-\tilde{{{\Delta }}}\Delta(-\frac{\ell_{1}(1-\tilde{{{\Delta }}}\right).\end{document}δlempir.=c1c21−Δ~Δ~exp−lc1(1−Δ added crease length determined by sheet’s instantaneous state[12pt]{minimal}{amsmath}{wasysym}{mathrsfs{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$(\ell\tilde{{{\Delta }}}\end{document}(l,Δ model independent details crease network spatial homogeneity damagefitting parameters c1 c2 universal to values n\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin}-69pt}\begin{document\Delta\end{document~ facet segmentation crease pattern provides second measurable quantity d equal to sum interior perimeters facets total length edges facets expect d l proportional differences due to incomplete scarring facet perimeters sheet restore mild compression\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin}{-69pt}{document}\Delta{document accomplishes desired proportionality\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts{amsbsy{mathrsfs{upgreek}{\oddsidemargin}{-69pt}\begin{document}$$\text{meas.\equiv (1-\tilde\Delta){d}{meas{document.≡(1−Δ~)dmeas measured total crease length segmented data facet area edge length distributions estimate d average length edge times average edges expressed average number facets total sheet area divided facet area times edges facet halved shared edges yields13\documentclass[12pt]{minimal}\usepackage{amsmath{upgreek{\oddsidemargin}{-69pt}\begin{document}${d}\text{model}}=\frac{a\theta (t)}{s(t)} \frac{{n}{2=\frac{{n}_{e}t}{2(a+1)}.\end{document}dmodel=aθ(t)s(t)×ne2=net2 obtain\documentclass[12pt]{minimal}\usepackage{amsmath\oddsidemargin-69pt}\begin{document}\ell{model}\equiv (t\tilde\Delta)=(1-\tilde\Delta)\frac{{n}_{e}t}{2(a+1)}\end{document}lmodel(t)≡l(t,Δ~)=(1−Δ~)net2 superscript (t) denotes lmodel on t develop connection between t and n lmodel n[12pt]{minimal}{amsmath}{upgreek\oddsidemargin-69pt}{document}\tilde{{{\Delta\end{document}Δ~ mirroring Eq. (11) fit Eq. (14) to measured length lmeas. reveals value ne ≈ 4.2 Fig. 4b average 4–5 sides per facet. Fig.4c agreement between[12pt{minimal{amsmath\oddsidemargin-69pt}{document}{model}lmodel(t Eq. (14) lempir Eq. (11).Numerical evidence for insensitivity to initial connection between statistical model facet area total crease length presented insight solving Eq. (1) numerically numerical integration scheme second-order composite trapezoid rule discretization x implicit multi-step discretization in t. sample numerical result Fig. 5 reveals rapid convergence to steady state solution Eqs. (7a) (7b), insensitivity to initial state observed history independence of total crease length Gottesman et al.16 sheets with different loading hand-crumpled deliberately folded equal total crease lengths same accumulation damage protocol Fig. 1a sheets distinct initial facet area distributions areas deliberately folded sheet peaked values hand-crumpled sheet broadly distributed signatures initial preparation eclipsed by crumpled state rapid convergence to steady state.Fig.5Numerical validation analytical solution Eq. (1) snapshots numerically calculated φnum(ξ) initial condition c(x, 0) = δ(x − 1) a = 1 rapid convergence steady state distribution dashed line analytical form Eq. (7a) large t evolution mean area s(t), analytical solution large t Eq. (7b) dashed line mean area experimental samples function t computed from Eq. (7b) dashed line corresponds Eq. (7b) a(t) Eq. (8) Marker colors correspond values\documentclass[12pt{minimal{amsmath-69pt colorbar estimate total crease length derived facet area ridge length distributions consistency logarithmic scaling Eq. (11) propose mechanism geometric incompatibility folded sheet confinement leads fragmentation driving t forward argument establishes evolution t accordance n supplies missing link physically based model experimental findingsOne-dimensional explanation logarithmic scaling develop simple one-dimensional model additional fragments crumpled sheet re-crumpled relying on statistical descriptions facet area segment length goal summarized two questions current state probability sheet further fragmentation? probability to continuous variables fragmentation model Eq. (1)? appeal to axial symmetry confinement crumpling to 1D strip length L0 Fig. 6. strip characterized directions into random segments lengths r segments equal to cross-sections intercepted facets distributed according to derived gamma distribution Eq. (10) weighted by horizontal facet width increases likelihood occurrence randomly selected vertical stripfacets ~1:1 aspect ratio distribution segment length\documentclass[12pt{minimal{amsmath\setlength\oddsidemargin-69pt}(r\frac{1}\theta(a+1)\theta-r\theta(r)=1θΓ(a+1)rθae−r/θ average segment length (a + 1)θ comparison Eq. (15) experimental data strongly compacted sample Fig. 6a b extended results samples Supplementary Fig. 5.Fig. 6A folded cross-section one-dimensional random walk sample segmented sheet dashed line vertical cross section distribution segment lengths cross-sections sheet (filled Eq. (15) solid curve No additional fit value shape parameter a Eq (15) determined Eq. (8) best fit τ facet area distributions schematic analog folding one-dimensional strip axially confined sheet one-dimensional random walk time axis extended vertically filled curve represents distribution walker’s final displacement darker shaded regions walks outside confinementillustration one-dimensional folding geometric estimate critical confinement w detailed Supplementary Note final displacement strip folded at each break absence confinement problem mapped to displacement walker one-dimensional random walk gamma-distributed steps walker’s steps occur alternating directions distribution fZ(z) of position Z after 2k steps k derived Supplementary Note 2. salient trends observed applying central limit theorem considering position Z valid for large k small step size\documentclass[12pt]{minimal}\usepackage{amsmath-69pt}{document}{Z}\theta={1}_{0\theta}fZ(z;θ)=12πL0θexp−z22L0θ normal distribution of zero mean variance L0θ confinement introduced at locations ∣z∣ = w likelihood walker steps beyond confinement probability integrate Eq. (16) for all ∣z∣ > w two-sided survival function of Eq. (16)not equivalent to initial question intermediate steps past ∣z∣ > w acceptable estimate last step greatest variance accurate calculation evaluate likelihood walk escapes confinement any step last step useful simplicity captures anticipated behavior comparison to accurate formulation in Supplementary Fig. 9. pursue simpler form survival function for large k refer Supplementary Note 2 for exact derivation all k survival function of Eq. (16), SZ(w; θ) = P(∣Z∣ > w; w≥0) threshold confinement w given\documentclass[12pt]{minimal}\usepackage{amsmath}-69pt}}(w;\theta )=1-}SZ(w;θ)=1−erfw2L0θ erf(z) error function walkers at ∣z∣ > w restored within limits confinement steps must fragment increasing number steps taken decreasing average drives evolution fragmentationarticulates claim considering original shaped sheets random walks progression fragmentation measured change dt over single iteration dn proportional to walks confinement at ∣z∣ = w: dt/dn ~ SZ(w; θ). likelihood single random walk leaves critical confinement fragmentation rate describes average fragmentation likelihood confinement w temporal parameter t = 1/θ maturity fragmentation enforce direct correlations between iterations new along previous ones decrease in fragmentation rate n encoded through decreasing mean facet area with t stronger correlation expected between walks nearby transects sheet consider statistical behavior sheet account increased fragmentation likelihood for facets larger horizontal extent weighting Eq. (15) Eq. (17) gives likelihood new creases form describe new damage two additional factors considered sheet strongly confined in closely packed layers collectively fragment factor ~ 1/L halving final height doubles additional ridges opposite limit low compaction facets not close proximity behave cooperatively new damage scales linearly with compression L0−Lconsiderations propose evolution fragmentation process crumpling iteration\documentclass[12pt]{minimal}{amsmath{wasysym\oddsidemargin-69pt}{document}$$\delta t\equiv\partial t n}=\alpha{1-\tilde{{{\Delta }}}}{S}{Z}(w{document}δt≡∂t∂n=α1−Δ~Δ~SZ(w α fitted constant proportionality explicit dependence t t θ inversely related critical width w determined geometry imposed confinement illustrated Fig. 6d complete derivation Supplementary Note 4:19\documentclass[12pt]{minimal}{amsmath}{upgreek}{{-69pt}{document}$$w(\tilde{{{\Delta }}}\approx \frac{R}{{1-\tilde\Delta{2}}}{document}w(Δ~)≈R1−Δ~2 R radius container Eq. (14) relate Eqs.(18) (12)\documentclass[12pt{minimal{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}\begin{document}\delta\ell{model}}\frac{d(t\delta t=\frac{(1-\tilde\Delta{2(a+1)}\delta\end{document}δlmodel=dlmodel(t)dtδt=(1−Δ~)ne2(a+1 fit proportionality constant α asymptotic approximation limit large t Supplementary Note 3 Eq.(18) analytically integrated scaling relation[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin}{-69pt}\begin{document}(n\tilde\Delta\end{document}t(n,Δ~) similarity[12pt]{minimal}{amsmath}{wasysym}{amsfonts{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\begin{document}$$\ell (n\tilde{{{\Delta }}}\end{document}l(n,Δ~) Eq.(11):21[12pt]{minimal{amsmath{wasysym{amsfonts{mathrsfs{upgreek{-69pt}{document}\tilde{\Delta{\tilde{c}}_{1}(1\Delta{2})\mathrm{log}}\left(1+\frac{{\tilde{c}}_{2}n\tilde{\Delta }(1\Delta\end{document}t(n,Δ~)=c~1(1−Δ~2)log1+c~2nΔ~(1[12pt]{minimal}{amsmath{wasysym{upgreek}\oddsidemargin{-69pt}{document}{\tilde{c}}_{1}=2{L}_{0}/{R}{2}{document}c~1=2L0/R2[12pt]{minimal}{amsmath}{wasysym{upgreek\oddsidemargin}{-69pt}{document}{\tilde{c}}_{2}=\alpha {R{2}{L}{0}\sqrt{2\pi\end{document}c~2=αR2/L02π L0 R sheet length container height radius Eqs. (14) (21) provide theoretically motivated expression\documentclass[12pt]{minimal}{amsmath{upgreek\oddsidemargin{-69pt}{document}$${\ell{model}{(n)}\equiv (t(n\tilde\Delta\end{document}lmodel(n)≡l(t(n,Δ~),Δ~) fragmentation kinetics mechanism re-fragmentation random walk Figure 7 compares empirical relations δlempir. lempir.derived models δlmodel\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym\oddsidemargin}{-69pt}\begin{document}\ell{model}\end{document}lmodel(n), measured quantities[12pt]{minimal}{amsmath}{wasysym}{upgreek}{\oddsidemargin}{-69pt}{document}\delta {\ell\text{meas.}}\equiv {\ell{meas.(n)(n-1)\end{document}δlmeas.≡lmeas.(n)−lmeas.(n−1) lmeas various n results Figs. 4 7 consistency fragmentation model anticipated logarithmic growth.Fig. 7Validation empirical derived models crease length evolution measurement Predicted change in total crease length δlempir by Eq.(12) plotted against change crease length between crumples\documentclass[12pt]{minimal}{amsmath{wasysym\oddsidemargin}{-69pt}{document}$$\delta {\ell\text{meas.}}\equiv(n)(n-1)\end{document}δlmeas.≡lmeas.(n)−lmeas.(n−1) Open markers manually segmented data filled circles data automated segmentation Supplementary Methods 1:1 reference line) guide eye total crease length Eq. (11) plotted against measured total crease length lmeas. change in total crease length δlmodel predicted by Eq. (20) d total crease length[12pt]{minimal}{amsmath{wasysym{upgreek{\oddsidemargin}{-69pt}{document}\ell\text{model}{(n)\end{document}lmodel(n) obtained from Eqs. (14) and (21) against measured values comparison to a and b.Marker colors panels correspond values of\documentclass[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document\Delta\end{document}Δ~ indicated colorbar empirical derived relations for δl l strong models measured data affirm logarithmic relationship damage evolution correspondence between crumpling thin sheet fragmentation process derived framework for evolution statistical properties intricate crumpled patterns theoretical models experimental data proposed model of one-dimensional folding fragmentation due to geometric incompatibility between confinement random walk exceeding critical displacement predicted accrual damage quantified by added crease length consistency with logarithmic model Gottesman et al.16 basis for logarithmic scaling in crumpling experiments model explains history independence logarithmic scaling area distribution crumpled state strong attractor fragmentation consistency crumpling with fragmentation theory hints universal behavior uniting diverse fragmenting systemsactivation of defects in fragmentation ceramics can slow fracture damage accumulation as re-crumpled sheet exploits folds31 studies of crumpled systems offer new lens interpret complex processes extension validation of results on sheets varied thicknesses material parameters different compaction protocols assumption fragmentation of facets is scale invariant assumption down for large crumpling iterations Lechenault et al.32 approach identifying limit energetic competition between bending of facets rigid folding along creases energy cost former proportional to bending rigidity latter proportional to crease stiffness energy balance provides characteristic length scale varies to sheet thickness improvement benefit from further studies over material parameters Length scales of in crumpled systems studied thermally driven dynamics to draw connections to statistical mechanical models of crumpling33–35 explore generalizations of functional forms breakup rates variations across experimental results compaction unifying framework for observations deeper understanding of crumpling dynamics can assist damage network formationmachine learning structure in complex systems36 prior work importance preserving physical properties predictions preserving vertex angle constraints in synthetic fold patterns ridge network reconstruction in crumpled sheets38 future machine learning may enforce facet area crease length statistics ridge network evolution spatial data with theoretical insight enable comprehensive predictions of crumpling dynamics future data analyzed collected for study Gottesman et al.16 experimental protocol 10 cm × 10 cm Mylar sheets rolled into 3-cm diameter cylindrical container compressed uniaxially to specified compaction ratio//L0 ratio of final to initial height L0 = 10 cm (Fig. 1a). ridge network on each sheet extracted by scanning sheet using laser profilometer produces height map two-dimensional map of mean curvature determined from spatial gradients height profile sharp peaks in curvature mark signature of ridge (Fig1b). re-crumpling scanning single sheet performed n times up to n = 24. facets contiguous regions near-zero curvature delineated in Fig. 1c. to noise artifacts data collection not all facets enclosed by ridges breaks occur during re-crumpling unfolding. Automated methods crease detection facet labeling tested segmentation sensitive to noise to over-fragmenting sheets Each sample digitally labeled by hand details of automated segmentation in Supplementary Methods segmentations facet area distributions shown in Supplementary Figs. 3 and 4. manual segmentation dominant lines pattern less pronounced softer scarring segmentation performed for sheets after iterations n = 1, 2 3 at seven compaction ratios:\documentclass{minimal{amsmath-69pt{document=0.63,0.45,0.36,0.27,0.18,0.09 crumples compared across iterations n for consistency labeled facets earlier persist in laterSamples\documentclass[12pt{minimal\usepackage{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}{document}\tilde\Delta=0.63,0.45\end{document=0.63,0.45 0.27 labeled after n = 24 crumples total 24 samples samples at n = 24 missing detail older scarring ridges valuable study results manual segmentation facet area distributions Supplementary Figs. 1 2.Supplementary Review File | 48.5 | 1.266665 |
10.1038/s41467-020-19703-y | PMC7679449 | The development of molecular electronics at single molecule level calls for new tools beyond electrical characterisation. Kos et al. show an optical probe of molecular junctions in a plasmonic nanocavity geometry, which supports in situ interrogation of molecular configurations. | Molecular electronics promises a new generation of ultralow-energy information technologies, based around functional molecular junctions. Here, we report optical probing that exploits a gold nanoparticle in a plasmonic nanocavity geometry used as one terminal of a well-defined molecular junction, deposited as a self-assembled molecular monolayer on flat gold. A conductive transparent cantilever electrically contacts individual nanoparticles while maintaining optical access to the molecular junction. Optical readout of molecular structure in the junction reveals ultralow-energy switching of ∼50 zJ, from a nano-electromechanical torsion spring at the single molecule level. Real-time Raman measurements show these electronic device characteristics are directly affected by this molecular torsion, which can be explained using a simple circuit model based on junction capacitances, confirmed by density functional theory calculations. This nanomechanical degree of freedom is normally invisible and ignored in electrical transport measurements but is vital to the design and exploitation of molecules as quantum-coherent electronic nanodevices. | IntroductionConstructing electronic junctions with molecules is a potential avenue to achieve new device functionality, further junction miniaturisation and reduce energy consumption1–4. Metal-molecule-metal junctions have been extensively studied using scanning probe techniques and break junctions, typically through statistical analysis of electrical conductance in repeated transient pull-off experiments which shed little light on individual junction morphologies. Experiments which contact self-assembled molecular monolayers (SAMs) by pulling away a sharp tip5–7, until one final molecule bridges the tip-surface gap (Fig. 1a, b), probe different metal atomic configurations and molecular binding sites at each realisation8,9. Combining many thousands of such pull-off conductance measurements delivers systematic results which depend on the molecular electronic states. Alternative in situ averaging over many molecules through large-area contacts can also access molecular electronic properties10–12, with damage minimised by using nano-particulate or liquid metal contacts13–15 (Fig. 1c). Inferring the exact configuration of molecules in situ from these current-voltage measurements however is not easy, because the integrated device geometries prevent external probing of the junction during operation. Optical spectroscopies with the capability to follow molecular charging have been attempted by indirectly contacting molecules inside electrochemical cells16,17 (Fig. 1d), but suffer interference from ions in solution18 while probing regions with >1015 molecules.Fig. 1Schemes for contacting molecular junctions.a, b Junctions formed by scanning tunnelling microscope (STM) tip pull-off from a SAMs or b single molecules. c Eutectic GaIn alloy contacts. d Large area electrochemical cells indirectly contact AuNPs on SAMs and e scheme here using single plasmonic nanoparticle contact. f, g Detail of Au nanoparticle on SAM (NPoM geometry) which is electrically contacted by a conductive transparent cantilever, giving real-time optical access to the junction. Scale bar in g is 10 µm. h Raman spectroscopy of BPDT molecular layer in NPoM through the cantilever shows molecules are unperturbed. Laser power at sample is 0.2 mW, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}λ = 633 nm, focussed to sub-μm spot.In our devices, molecular monolayers are deposited on a flat Au substrate with single Au nanoparticles used as the top electrode, where the junction area can be scaled by using different diameters D (Fig. 1e, f). The extreme plasmonic coupling between nanoparticle and Au surface for such nm gaps allows in situ optical access to the behaviour of junctions under bias. Our measurements show that the molecules undergo conformational changes in the V < 1 V tunnelling regime. This new molecular nanomechanical mechanism shrinks MEMS functionality from the micro- to nano-domain.This scheme exploits the tight light confinement inside d = 1–2 nm gaps (lateral optical full-width \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt {Dd} \sim$$\end{document}Dd~ 7 nm) within nanoparticle-on-mirror (NPoM) structures17,19 to optically investigate ~100 junction molecules (Fig. 1f, g). Solution-deposited low-density D = 100 nm Au nanoparticles (AuNP) are conformally coated with an insulating parylene layer, and dry-etched to expose the AuNP crown (Methods section). NPoMs are individually contacted using a conductive transparent Si3N4 tip-less cantilever, giving negligible contact resistance (Supplementary Note 4).Each NPoM is optically accessed through the cantilever for imaging and spectroscopy in real time during electrical measurements. The enhanced (E > 500) optical field within the NPoM plasmonic gap enables strong surface enhanced Raman spectroscopy (SERS ∝ E4) of the molecular spacer17, giving >100 kcounts/mW/s thus allowing short integration times for tracking dynamics. The collected Raman from the λ = 633 nm focussed laser (dashed Fig. 1h) originates only from molecules underneath the AuNP facet, initially biphenyl-dithiol (BPDT), showing their Raman-active vibrational modes which remain unchanged when contacted (Fig. 1h). Biphenyl SAMs are widespread in molecular electronics, known to form closely-packed uniform SAMs20 (giving consistent SERS), and provide large Raman cross-sections.Changes in refractive index, thickness or conductivity of the molecular spacer can be tracked in real time through dark-field (DF) scattering spectroscopy of the electrically contacted NPoMs (seen red in circled DF image, Fig. 1g). Dark-field spectra of individual NPoMs with BPDT (Supplementary Fig. 2) are dominated by a strong peak at ∼680 nm from coupled plasmon oscillations between the AuNP and underlying Au mirror (Fig. 1f). Conductance measurements show that ∼50 molecules are contacted (Supplementary Note 4), while optical probing tracks a similar number of molecules under the facet17.Using these optically accessible and reconfigurable but non-disruptive contacts for molecular electrical junctions we reveal that molecules twist under bias voltage, modifying their conductance. The applied bias triggers a redistribution of charge and potential across the junction, and molecules react by changing their conformation to minimise the overall dipolar and capacitive energies. This change is largely invisible in the electrical response, and only revealed by direct optical access to the molecules in the junction.ResultsTwisting of molecular rings upon applied biasThe cantilever used to approach an individual NPoM junction is lowered until flat and parallel to the surface (Supplementary Fig. 1) to make an ohmic contact (Methods section), avoiding significant forces by bending (evidenced by the absence of vibrational Raman shifts from the 50 gap molecules probed). Electrical measurements are performed simultaneously with continuous Raman or dark-field spectral acquisitions. A constant voltage during each acquisition (Fig. 2a) gives highly repeatable spectra (Supplementary Figs. 4 and 5, molecules absorb only in UV thus giving no photocurrent). Strong reduction in SERS is seen whenever \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| V \right|$$\end{document}V > 0.5 V, saturating above 1.0 V (Fig. 2b, c) with over 10-fold decrease. In dark-field no changes are detected for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| V \right|$$\end{document}V < 1 V (Fig. 2d–f). The collapse in SERS strength by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| V \right|$$\end{document}V = 0.5 V occurs before any significant variation in scattering spectrum (Fig. 2f), showing cantilever artefacts cannot be responsible. No effects are seen when the SAM is not included in the device. The conductance G < 1 nS for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| V \right|$$\end{document}V < 1.5 V remains in the linear direct tunnelling regime G < 10−4
G0, becoming nonlinear only for higher bias (Fig. 2a, d) where non-reversible redshifts of the NPoM coupled mode are seen (saturating at Δλ ∼ 80 nm). Similar effects are seen with both positive and negative bias (Supplementary Fig. 4).Fig. 2Raman (SERS) and dark-field switching under bias in single junction of BPDT.a, b Real-time SERS for increasingly negative bias voltage. c SERS spectra decrease 10-fold by −1 V bias. d–f Dark-field scattering intensity under bias, with decrease in amplitude and coupled plasmon redshift for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|V|$$\end{document}∣V∣ > 1 V but f no change for −1 V bias.Although the light intensity in the nano-gap does not change (shown by unchanging dark-field spectra), the 20-fold decrease in SERS shows that the molecular Raman cross-sections reduce (at slightly different rates, Supplementary Fig. 4). Biphenyl molecules have a delocalised π electron distribution21 due to near-alignment of the π orbitals across the C atoms connecting the two rings. Twisting of π-orbitals across this C–C link disrupts the delocalisation, reducing the molecule polarizability. Previous work however suggested that bias charges these molecules, which aligns the two rings and thus leads to an increased Raman cross section and a shift in the peak positions, as opposed to the decrease in signal observed here with no peak shifts (Supplementary Note 10)16,22.We explore this twisting through density functional theory (DFT) on a BPDT molecule bound to Au atoms at both thiol terminal groups. The molecule is progressively twisted by changing the dihedral angle θ between the two ring planes from 0° (in-plane) to 90° (rings perpendicular to each other). The energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{{\mathrm{DFT}}}(\theta )$$\end{document}UDFT(θ) and Raman signal intensity are computed for each configuration (Methods section). The energy minimum at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta \sim 35^\circ$$\end{document}θ~35∘ sets the initial molecular state within the junction. The simulated Raman intensity decreases with increasing twist angle (Fig. 3a), and is minimised at 90° because the extended dipole across both rings is broken23 decreasing the Raman cross-section.Fig. 3Comparison of Raman vs twist θ in theory and experiment.a DFT calculations show decrease in Raman intensity with increasing dihedral angle for BPDT. b Experimental reduction in SERS intensity as voltage increases from 0 V to 1 V. c SERS intensity reduction quantified as ratio at ±1.5 V vs 0 V for different biphenyl molecules and NPT (using Raman peak intensity at 1590 cm−1, 1070 cm−1 for NPT). Line shows model predictions.Our experiments reproduce this Raman suppression with increasing voltage (Fig. 3b), suggesting we indeed observe molecular twisting. This is supported by experiments on 2-naphthalene-thiol, which cannot twist and gives no voltage dependence (NPT in Fig. 3c, and Supplementary Fig. 5). Other biphenyl molecules with different functional groups still show SERS switching, but with decreased on/off SERS ratios and larger voltage thresholds when only one thiol group is present (Fig. 3c and Supplementary Note 3). While the 20 nm wide facets of 100 nm AuNP nanogaps accommodate ∼100 molecules in a single junction17, we can also observe twist switching of individual molecules (Supplementary Fig. 8). Given the sub-ps lifetime of the torsional mode, molecular NEMS switching potentially accesses the THz regime, promising for devices. Modulating the applied voltage in our experiments gives both a low frequency contribution to the switching (∼10 kHz) alongside a high-frequency contribution which persists above 1 MHz.Origin of molecular twisting behaviourPrevious models for bi-phenyl systems considered redox effects22 that modify the Raman cross-sections (flattening the molecule upon reduction, as replicated in DFT here, Supplementary Note 10), or changes in binding of the terminal group with Au adatoms24, that cannot explain our results. Instead, to investigate the origin of molecular twisting in our system, we implement detailed non-equilibrium calculations based on DFT modelling of the junction under applied bias (see Methods section and Supplementary Note 11). These calculations show that local charges and electrostatic potential distribute differently across the gap region depending on molecular configuration, and that a twisted ring conformation is energetically favoured as bias increases.The junction is modelled with a BPDT molecule bound to two identical Au leads via linker-Au atom protrusions (Fig. 4a), for fixed twist θ of the rings. Leads are one atom thick for faster convergence of calculations, but this agrees with the full three dimensional lattice of Au atoms (Supplementary Note 11). We calculate the energy of the system under bias for θ = 15° and θ = 90° (Fig. 4b), and observe that the energy of the θ = 90° configuration has a weaker voltage dependence, dropping below the θ = 15° energy for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{\mathrm{t}}$$\end{document}Vt > 1.5 V. To identify the origin of this trend, we extract the electrostatic potential profile along the junction including the leads and molecule in the two cases along with the Muliken charge distributions (Fig. 4c and Supplementary Fig. 19). In the θ = 90° case, around half of the applied potential is dropped across the central C–C bond (V2, black dashed arrow). From the charge stored vs potential drop across this bond, a capacitance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_2$$\end{document}C2 = 0.015 aF is obtained, nearly independent of bias and matching that expected for a bond-sized capacitor (Supplementary Note 9). In a simple electrostatic picture, the total energy of this configuration is then1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{90^\circ } = U_{{\mathrm{twist}}} + \frac{1}{2}C_2V^2$$\end{document}U90∘=Utwist+12C2V2where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{{\mathrm{twist}}}$$\end{document}Utwist = 0.55 eV is the energy required to twist the rings to θ = 90° (Supplementary Note 11). This model provides a surprisingly good account of the DFT calculation (Fig. 4b black dashed). By contrast for θ = 15° twists, the central C–C bond has higher conductance and gives negligible potential drop (Fig. 4c). Instead, the potential drop is almost entirely concentrated at one of the molecule-lead interfaces and is accompanied by a build-up of positive charge on the linker-Au and negative charge on the S. This decreases the dipole moment at the Au–S interface formed when the molecule binds. The same behaviour is reproduced at the opposite interface when the bias polarity is inverted (Supplementary Fig. 19). The total energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{15^\circ }$$\end{document}U15∘ is then the sum of a linear contribution \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{{\Delta}}}q\left| V \right|$$\end{document}ΔqV, symmetric in voltage, given by the surface dipole induced by the field, and a quadratic capacitive term accounting for the charge on the S and surrounding Au atoms2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{15^\circ } = {\mathrm{{\Delta}}}q\left| V \right| + \frac{1}{2}C_1V^2$$\end{document}U15∘=ΔqV+12C1V2with contact capacitance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_1\sim$$\end{document}C1~ 0.04 aF and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{{\Delta}}}q$$\end{document}Δq = 0.25e extracted from the DFT (Supplementary Note 11). Again this simplified \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{15^\circ }$$\end{document}U15∘ accounts for the DFT energies well (Fig. 4b), in particular showing the low-field reversal of the induced surface dipole for opposite bias, supporting our interpretation based on local charges and potentials. The transition voltage \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{\mathrm{t}}$$\end{document}Vt for triggering switching to θ = 90° when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{15^\circ } > U_{90^\circ }$$\end{document}U15∘>U90∘ is then controlled by the relative contact energy3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{\mathrm{c}} = {\mathrm{{\Delta}}}q.V_{\mathrm{c}} = \frac{1}{2}\frac{{{\mathrm{{\Delta}}}q^2}}{{\left( {C_1 - C_2} \right)}}\sim 0.20\,{\mathrm{eV}}$$\end{document}Uc=Δq.Vc=12Δq2C1−C2~0.20eV4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{\mathrm{t}} = 2V_{\mathrm{c}}\left\{ {\sqrt {1 + \frac{{U_{{\mathrm{twist}}}}}{{U_{\mathrm{c}}}}} - 1} \right\}\sim 1.5{\mathrm{V}}$$\end{document}Vt=2Vc1+UtwistUc−1~1.5Vclose to the observed value of 1 V. This expression for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{\mathrm{t}}$$\end{document}Vt can be adapted to all molecules with a dihedral degree of freedom, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{{\mathrm{twist}}}$$\end{document}Utwist is modulated by side groups that determine the steric properties of the molecule, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_1 - C_2$$\end{document}C1−C2 is set by the difference in capacitance between the twisting C–C bond and the molecule-lead interface, in turn related to metal type and molecular terminal group. The asymmetry of the current for negative and positive voltages (Fig. 4d) is due to asymmetry of the junctions, while we also observe a small rectification in our experiment (Supplementary Fig. 9). This is related to the asymmetry of the junctions imposed by the twisted biphenyl rings and their relative orientation to the Au–S–C angle. The energy profile changes slightly when the configuration to the electrodes changes (Supplementary Fig. 21), however the twisting behaviour is not affected.Fig. 4Modelling of junction under bias.a Model DFT geometry shown for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}θ = 15°, Au and S atoms remain fixed when changing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}θ. b DFT energy for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}θ = 90° is lower than \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}θ = 15° for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V$$\end{document}V > 1.5 V, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{15^\circ }$$\end{document}U15∘, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{90^\circ }$$\end{document}U90∘ calculated from bond capacitances and dipoles (dashed) follow DFT, predicting configuration switch at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{\mathrm{t}}\sim$$\end{document}Vt~ 1.5 V. c Potential distributions in atoms at the molecule-lead interface for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}θ = 15°. Applied potential \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{{\Delta}}}V$$\end{document}ΔV drops predominantly across central C–C bond for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}θ = 90°, and across one Au–S interface for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}θ = 15°. d Calculated tunnelling between the leads, showing lack of twisting signature at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{\mathrm{t}}$$\end{document}Vt.Circuit model of molecular junctionWe translate the localised charges and potentials into an intuitive circuit model linked directly to the spectroscopic data. We divide the molecular junction into three sections, each with its own conductance and capacitance (Fig. 5a). Conductive AFM and STM break-junction experiments in liquid already showed that the tunnelling conductance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_2$$\end{document}G2 through fixed-twist biphenyl molecules is controlled by θ as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_2 = G_{{\mathrm{CC}}}(1 + g{\mathrm{cos}} ^{2}\theta )$$\end{document}G2=GCC(1+gcos2θ) where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_{{\mathrm{CC}}}\sim$$\end{document}GCC~ 76 μS and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g\sim$$\end{document}g~ 50 (refs. 25–27) (Supplementary Note 9). The molecule-Au contact regions are equivalent to Schottky diodes oriented back-to-back (Fig. 5a), so that when one is reverse biased the other is forward-biased (and thus conducting). The applied potential is then divided between the central bond (with variable barrier height depending on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}θ) and the linker Au–S at the negative contact. Since the current \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I = V_1G_1 = V_2G_2 = VG_t = (V_1 + V_2)\left[ G_1^{ - 1} + G_2^{ - 1} \right]^{ - 1}$$\end{document}I=V1G1=V2G2=VGt=(V1+V2)G1−1+G2−1−1, the electrostatic energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{\mathrm{Q}} = \frac{1}{2}\left[ C_1V_1^2 + C_2V_2^2 \right] + {\Delta}q.V_1$$\end{document}UQ=12C1V12+C2V22+Δq.V1 is simply obtained (noting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_3\sim$$\end{document}V3~ 0 from Fig. 4d; for negative bias \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_1$$\end{document}V1,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_1$$\end{document}C1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_3$$\end{document}V3,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_3$$\end{document}C3 are swapped). The interface capacitances dominate the twist capacitance (Fig. 4e) because of the larger area at the contact. Adding the configuration energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{{{{\mathrm{twist}}}}}(\theta )$$\end{document}Utwist(θ) required to twist the molecule by an angle \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}θ calculated from DFT, we obtain the total energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U = U_{{\mathrm{twist}}}\left( \theta \right) + U_{\mathrm{Q}}(\theta )$$\end{document}U=Utwistθ+UQ(θ). Plotting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U$$\end{document}U for different bias levels (Fig. 5b) indeed predicts that the stable twist angle \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta _{{\mathrm{eq}}}\left( V \right)$$\end{document}θeqV increases as voltage is applied, reaching 90° ~ 1 V.Fig. 5Circuit model compared to SERS.a Circuit model for BPDT molecules in the junction gap, with each section 1–3 of the junction characterised by its conductance and capacitance; central CC bond has variable conductance. b Calculated energy profile \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U = U_{{\mathrm{DFT}}}\left( \theta \right) + U_{\mathrm{Q}}(\theta )$$\end{document}U=UDFTθ+UQ(θ) as voltage is increased, shifting stable angle towards larger twists. c Voltage dependence of DFT twist angle \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta _{{\mathrm{eq}}}\left( V \right)$$\end{document}θeqV (black) and resulting SERS intensity (at 1590 cm−1) from DFT (dashed) compared to experiment (points). d Voltage-induced shift of central CC twist vibration to smaller wavenumbers (arrows). SERS normalised using a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T$$\end{document}T = 320 K thermal ratio between Stokes and anti-Stokes.The DFT Raman intensity at each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta _{{\mathrm{eq}}}\left( V \right)$$\end{document}θeqV from Fig. 3a then predicts the SERS vs voltage, and matches our data well (Fig. 5c, see Supplementary Notes 9 and 11 for model details) providing a direct link between the experimentally observed modulation of SERS and the molecular twist. Crucially, although the switching is readily detected optically, in electrical transport we find that the full DFT calculations predict only smooth changes in conductance during switching (Fig. 4d and Supplementary Fig. 20), which is indeed observed (Supplementary Fig. 9). This therefore motivates the use of nano-optics in molecular electronics.Inserting an extra carbon into the chain (BMMBP, Supplementary Note 3) reduces the molecular conductance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_2$$\end{document}G2 which increases the voltage threshold for twisting, while removing the thiol from one end (BPT, CN-BPT) decreases the upper junction \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_1$$\end{document}C1, which also increases this threshold (Supplementary Note 9). Conduction in the |V| < 1 V regime is small (<nA, Fig. 2a, d) via direct tunnelling (Supplementary Note 9), but sets the potentials across each molecular section. At higher voltages significant currents flow, heating the junction and disrupting the molecular and Au structure resulting in DF spectral shifts17,28.Our model is based on individual molecules, but cooperative twisting within the SAM may also be important29. The voltage-dependent torque, 61\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V^2$$\end{document}V2 pN.nm, exceeds the DFT-estimated counter-torque of 35 pN.nm at a threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{\mathrm{t}}$$\end{document}Vt = 0.8 V in good agreement with experiment. The twisting energy per molecule, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{\mathrm{Q}}\sim 14k_BT\sim 56{\mathrm{zJ}}$$\end{document}UQ~14kBT~56zJ is robust to thermal fluctuations, despite being only 10x the lowest possible switching energy. Plasmonic nanocavities which allow us to extract Raman signatures, thus give in situ NEMS-based molecular torque measurements with a resolution of ~5 pN.nm.To further confirm this molecular twisting, we directly observe the central C–C twist of BPDT at low wavenumbers (Methods section) in both Stokes and anti-Stokes SERS (Fig. 5d). The 0 V peak at 65 cm−1 corresponds well to the DFT potential \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{{\mathrm{DFT}}}(\theta )$$\end{document}UDFT(θ) (Fig. 5b). Compared to other Lorentzian SERS lines (<10 cm−1 FWHM), the C–C twist is a broad Gaussian (100 cm−1 FWHM), suggesting each molecule is sensitive to its subtly different molecular environment within the layer, which supports the idea that Fermi energy varies slightly across the molecules in the gap. As bias is applied, a 10-cm−1 shift to smaller wavenumbers of this C–C twist is seen, in agreement with the 27% reduction in curvature of the full potential \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U(\theta )$$\end{document}U(θ) at 0.5 V in our circuit model (Fig. 5b) that should lead to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt {0.27}$$\end{document}0.27 = 15% or 10 cm−1 reduction in vibrational energy. Adjacent SERS lines corresponding to the S–Au bond stretch show no detectable shift, while the Stokes: anti-Stokes ratio17 shows no change in temperature with bias. This nano-device measures real-time molecular twist angles, suggesting the capability for detecting binding events, for instance sensing trace-gas molecules.DiscussionWe conclude that the energies obtained from the full quantum calculation of our molecular junction can be understood as being dominated by energies from electrostatics, and modelled as a classical electrical circuit based on surface dipoles and capacitance (Supplementary Fig. 21). Their energy cost as the bias increases drives a switch to the twisted molecular state. Surprisingly, we find that a dynamic conformational change in the molecule affects the coupling of the molecule itself to the leads, triggering a redistribution of charges and potential across the junction, in contrast with common views. Our findings demonstrate that the electrostatic potential profile changes significantly with bias voltage in a given molecular junction – a large potential drop at one interface between the planar BPDT molecule and gold electrode changes in the twisted BPDT to mostly develop across the C–C bond between the two rings (Fig. 4c). We also demonstrate that transport remains coherent at higher bias voltages and the molecule is not charged as suggested previously. This opens fascinating routes to utilise quantum interference at higher bias voltages for various nanoelectronic applications.Our model emphasises that theories of molecular electronics treating leads and molecules as separate objects are incomplete, and inclusion of local potentials, charges, and interfaces is needed even for simple junctions. In the tunnelling regime here, the capacitive energy saved by dropping potential across the rings yields elastic potential energy to twist the molecular torsion spring. Compared to conformational changes in molecules from STM22, our direct contacting of NPoM structures is performed in ambient conditions and without feedback currents passing through the molecule for tip positioning, and much closer to realistic device configurations. Future work to access ac conductivities through smaller area cantilevers would be useful to determine circuit model parameters directly. Our versatile contacting technique can be applied to a wide variety of other nanostructures such as plasmonic dimers or metasurfaces. Our findings have critical implications in the design of novel devices based on quantum interference that rely on molecular conjugation, since the molecule-lead interactions can disrupt conjugation and modify device functionality. Promising molecules to explore include para vs meta OPE3, anthracene, anthraquinone and dihydropyrene derivatives. Coupling molecular twisting, tunnelling electronic transport and plasmonics may also lead to novel modes of light emission30.MethodsSample preparationDevice fabrication starts with 10 nm of Cr for adhesion followed by 100 nm of Au deposited by evaporation onto a 250-nm-thick SiO2 on Si substrate in a pattern defined using shadow-mask evaporation. Samples are left overnight in a 1-mM solution of the desired molecule (BPDT, BPT, BMMBP, CN-BPT and NPT) in ethanol to deposit a self-assembled monolayer, and are then rinsed with ethanol. Gold nanoparticles (AuNPs, 100 nm diameter) from BBI solutions are drop-cast on the surface to obtain a number density of ~0.001 µm−2 on the Au pattern for contacting individual AuNPs through the cantilever. Sample are post-coated with a ~350-nm thick uniform layer of parylene-C (SCS Labcoter 2) at room temperature, and the parylene-C layer is then selectively etched using O2 plasma to expose the top of the AuNPs while leaving ~20–30 nm insulating layer around the sides and base (see Supplementary Note 1).Conductive cantilever preparationTransparent tip-less SiN AFM cantilevers (200 µm long, 35 µm wide and 600 nm thick) are coated with a 3-nm Cr adhesion layer and 6 nm Au by e-beam evaporation. The quality of the conductive film is verified with scanning electron microscopy and a uniform film without pinholes is obtained, with a sheet resistance 3Ω/square.Optical microscopy and spectroscopyOptical dark-field images are recorded in a custom-modified Olympus BX51 microscope. Samples are illuminated with a focussed white light source (halogen lamp). The scattered light is collected through a ×100 dark-field objective (Olympus LMPLFLN100xBD, NA 0.8) and analysed with a fibre-coupled (50 µm core diameter optical fibre) Ocean Optics QE65000 cooled spectrometer. We use a standard diffuser as a reference to normalise white light scattering.Atomic force microscopySample topography is performed using an Asylum Research MFP-3D atomic force microscope in non-contact AC mode. Asylum Research AFM Software version 15 is used for instrument control and data post-processing.Raman spectroscopyRaman spectroscopy is performed in the same custom-modified Olympus BX51, using a 633-nm excitation laser with 200 µW power focussed into a diffraction limited spot. The Raman emission is detected after separation through a dichroic beamsplitter and laser line filter using an Andor grating monochromator and EMCCD. Low wavenumber measurements are performed with the same system using a 785 nm excitation laser with 200 µW power.DFT calculationsComputational spectra are calculated after geometry optimisation with B3LYP hybrid functional employing Grimme’s D3 dispersion correction with Becke-Johnson damping. The basis set of def2SVP was used including the pseudopotential definitions for gold atoms. Potential bias was modelled as a homogeneous dipole field. Raman intensities were recalculated from the polarizability derivatives according to the experimental setup, including the temperature and frequency of the excitation laser. Numerical integrals were computed on an ultrafine grid, as method implemented in Gaussian09 Rev. E (Supplementary Note 6).Non-equilibrium transport calculations using DFT under biasTo investigate the quantum transport through BPDT molecules between gold electrodes, we first carried out geometry optimisation to the force tolerance of 10 meV/Å using the SIESTA implementation of density functional theory (DFT), with a single-ζ basis set and the Local Density Approximation (LDA) functional with CA parameterisation. A real-space grid was defined with an equivalent energy cut-off of 250 Ry. We applied bias voltage under non-equilibrium conditions and calculated the Hamiltonian and electrostatic potential for each bias voltage. From the converged DFT calculation, the underlying mean-field Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H$$\end{document}H was combined with our quantum transport code, Gollum to calculate transmission coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T(E,V_{\mathrm{b}})$$\end{document}T(E,Vb) for electrons of energy E passing from the source to the drain. We then calculate the room-temperature current \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I(V_{\mathrm{b}})$$\end{document}I(Vb) from the obtained \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T(E,V_{\mathrm{b}})$$\end{document}T(E,Vb) using the Landauer formula \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I(V_{\mathrm{b}}) = \frac{{2e}}{h}\mathop {\smallint }\nolimits_{\!\!- \infty }^{ + \infty } dET\left( {E,V_{\mathrm{b}}} \right)\left\{ {f\left( {E,T,V_{\mathrm{b}}/2} \right) - f\left( {E,T, - V_{\mathrm{b}}/2} \right)} \right\}$$\end{document}I(Vb)=2eh∫−∞+∞dETE,VbfE,T,Vb/2−fE,T,−Vb/2 (see Supplementary Note 11 for more details).Supplementary informationSupplementary Information | nature communications | [
"Article"
] | [
"Molecular electronics",
"Nanophotonics and plasmonics",
"Nanoparticles",
"Nanocavities",
"Nanophotonics and plasmonics"
] | electronic junctions with molecules device functionality junction miniaturisation energy Metal-molecule-metal junctions studied scanning probe techniques junctions statistical analysis electrical conductance pull-off experiments junction morphologies Experiments contact molecular monolayers (SAMs pulling until molecule tip-surface gap probe metal atomic configurations molecular binding sites Combining pull-off conductance measurements delivers results molecular electronic states averaging molecules large-area contacts molecular electronic damage nano-particulate liquid metal Inferring configuration molecules easy integrated device geometries prevent probing junction during operation Optical spectroscopies molecular charging molecules inside electrochemical suffer interference from ions >1015 molecules. 1Schemes contacting molecular junctions Junctions formed scanning tunnelling microscope) tip pull-off from molecules Eutectic GaIn alloy contacts Large area electrochemical cells contact AuNPs on SAMs single plasmonic nanoparticle contact Au nanoparticle on SAM electrically contacted by conductive transparent cantilever real-time optical access junction Scale bar 10 μmRaman spectroscopy BPDT layer NPoM shows molecules unperturbed Laser power sample 0.2 mW[12pt]{minimal{amsmath = nm sub-μm spot molecular monolayers deposited flat Au substrate single Au nanoparticles top electrode junction area scaled different diameters (Fig. 1e, f). extreme plasmonic coupling between nanoparticle Au surface nm gaps allows optical access behaviour junctions under bias measurements show molecules undergo conformational changes V < 1 V tunnelling regime molecular mechanism shrinks MEMS functionality micro- to nano-domain scheme exploits tight light confinement inside d = 1–2 nm gaps optical full-width[12pt]{minimal{amsmath 7 nm nanoparticle-on-mirror (NPoM) structures17 optically investigate ~100 junction molecules (Fig.Solution-deposited low-density D = 100 nm Au nanoparticles coated parylene dry-etched AuNP crown NPoMs contacted transparent Si3N4 tip-less cantilever negligible contact resistance NPoM accessed cantilever for imaging spectroscopy real enhanced (E > 500) optical field NPoM plasmonic gap enables Raman spectroscopy molecular >100 kcounts/mW/s short integration times collected Raman λ = nm laser originates from molecules underneath AuNP facet biphenyl-dithiol Raman-active vibrational modes unchanged contacted Biphenyl SAMs molecular electronics form closely-packed uniform provide large Raman cross-sections refractive index thickness conductivity molecular spacer tracked real time dark-field scattering spectroscopy contacted NPoMs Dark-field spectra NPoMs BPDT dominated peak ∼680 nm plasmon oscillations between AuNP mirror Conductance measurements ∼50 molecules contacted optical probing tracks similar molecules under facet17optically accessible reconfigurable non-disruptive contacts for molecular electrical junctions molecules twist under bias voltage modifying conductance bias triggers redistribution of charge potential across junction molecules conformation minimise dipolar capacitive energies change invisible in electrical response revealed by optical access to molecules.ResultsTwisting of molecular rings upon cantilever NPoM junction lowered until flat parallel to surface ohmic contact avoiding forces bending absence of vibrational Raman shifts from 50 gap molecules Electrical measurements performed with continuous Raman dark-field spectral acquisitions constant voltage during acquisition gives repeatable spectra molecules absorb in UV no photocurrent). reduction in SERS seen > 0.5 V saturating above 1.0 V over 10-fold decrease.dark-field no changes detected for\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek-69pt}}V < 1 V (Fig. 2d–f). collapse in SERS strength by[12pt{amsmath{wasysym{mathrsfs{upgreek-69pt}V = 0.5 V occurs before significant variation scattering spectrum (Fig. 2f), cantilever artefacts responsible No effects SAM not included conductance G < 1 nS for[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek\oddsidemargin-69pt}}V < 1.5 V remains linear direct tunnelling regime G < 10−4 G0 nonlinear for higher bias (Fig.2a d non-reversible redshifts NPoM mode (saturating Δλ ∼ 80 nm). Similar effects positive negative bias Fig. 4).Fig. 2Raman (SERS) dark-field switching bias junction BPDT Real-time SERS negative bias voltage SERS spectra decrease 10-fold −1 V bias d–f Dark-field scattering intensity bias decrease amplitude plasmon redshift no change −1 V bias light intensity nano-gap change unchanging dark-field 20-fold decrease SERS molecular Raman cross-sections reduce different rates Supplementary Fig. 4) Biphenyl molecules delocalised π electron near-alignment π orbitals C atoms Twisting π-orbitals disrupts delocalisation reducing molecule polarizability bias charges molecules aligns rings leads increased Raman cross section shift peak positions decrease signal no peak shifts (Supplementary Note 10 explore twisting density functional theory) BPDT molecule bound Au atoms thiol terminal groupsmolecule twisted changing dihedral angle θ between ring planes from 0° to 90° energy[12pt{minimal{amsmath{DFT(θ) and Raman signal intensity computed for each configuration (Methods section). energy minimum at[12pt-69pt sets initial molecular state junction simulated Raman intensity decreases with increasing twist angle (Fig. minimised at 90° extended dipole across rings broken23 decreasing Raman cross-section.Fig. 3Comparison of Raman vs twist θ in theory experiment DFT calculations show decrease in Raman intensity with increasing dihedral angle for BPDT Experimental reduction in SERS intensity as voltage increases from 0 V to 1 VSERS intensity reduction ±1.5 V vs 0 V biphenyl molecules NPT Raman peak intensity 1590 cm−1 1070 cm−1 Line model predictions experiments reproduce Raman suppression increasing voltage (Fig. molecular twisting supported by experiments 2-naphthalene-thiol twist no voltage dependence 3c 5) biphenyl molecules functional groups show SERS switching decreased on/off SERS ratios larger voltage thresholds one thiol group (Fig. 3c Note 3) 20 nm 100 nm AuNP nanogaps accommodate ∼100 molecules single twist switching molecules Fig. sub-ps lifetime torsional mode molecular NEMS switching accesses THz regime devices Modulating applied voltage low frequency switching high above 1 MHz.Origin molecular twisting models bi-phenyl redox Raman cross-sections changes terminal group Au explain results origin molecular twisting detailed non-equilibrium calculations DFT modelling junction bias local charges electrostatic potential distribute differently gap region molecular configuration twisted ring conformation favoured as bias increasesjunction modelled BPDT molecule bound two Au leads linker-Au atom protrusions (Fig. fixed twist θ rings Leads one atom thick faster convergence agrees three dimensional lattice Au atoms (Supplementary Note 11). calculate energy system under bias θ = 15° θ = 90° (Fig. energy θ = 90° configuration weaker voltage dependence dropping below θ = 15° energy{amsmath-69pt}Vt > 1.5 V identify origin trend extract electrostatic potential profile junction leads molecule Muliken charge distributions (Fig. 4c Supplementary Fig.θ = 90° case half applied potential dropped central C–C bond (V2 black dashed arrow). charge stored vs potential drop bond capacitance[12pt{minimal}{amsmath\oddsidemargin-69pt}_2}C2 = 0.015 aF obtained independent bias matching expected bond-sized capacitor (Supplementary Note 9)electrostatic picture total energy configuration\documentclass[12pt{minimal}{amsmath{wasysym{upgreek\oddsidemargin-69pt}$U_{90^\circ } = U_{{\mathrm{twist}}} + \frac{1}{2}C_2V^2}U90∘=Utwist+12C2V2where[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin{-69pt}\mathrm{twist}Utwist = 0.55 eV energy twist rings θ = 90° (Supplementary Note 11). model DFT calculation (Fig. 4b θ = 15° twists central C–C bond higher conductance negligible potential drop (Fig. 4c). potential drop concentrated molecule-lead interfaces build-up positive charge linker-Au negative charge S decreases dipole moment Au–S interface molecule binds same behaviour opposite interface bias polarity inverted Fig.total energy[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}{document}$$U_{15^\circ{document}U15∘ sum linear contribution[12pt]{minimal}{amsmath{wasysym{upgreek}\oddsidemargin}-69pt}{document}$\mathrm{{\Delta}}}q\left V \right|{document}ΔqV voltage surface dipole field quadratic capacitive term charge S surrounding Au atoms2[12pt]{minimal}{amsmath}{wasysym{amsfonts}}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$U_{15^\circ }\mathrm{{\Delta}}}q\left V \right|\frac{1}{2}C_1V^2{document}U15∘=ΔqV+12C1V2with contact capacitance[12pt]{minimal}{amsmath}{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}{document}$$C_1\sim{document}C1~ 0.04 aF[12pt]{minimal}{amsmath}{wasysym{mathrsfs{upgreek}\oddsidemargin{-69pt}{document}$${\mathrm{{\Delta{document}Δq = 0.25e extracted DFT (Supplementary Note 11). simplified[12pt]{minimal}{amsmath}{wasysym}{amsfonts{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$$U_{15^\circ{document}U15∘ DFT energies (Fig. 4b), low-field reversal induced surface dipole opposite bias interpretation local charges potentialstransition voltage[12pt]{minimal{amsmath{wasysym\oddsidemargin-69pt}}$$V_{\mathrm{t}}\end{document}Vt triggering switching to θ = 90°[12pt{amsmath}-69pt}$$U_{15^\circ > U_{90^\circ }{document}U15∘>U90∘ controlled by relative contact[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$$U_{\mathrm{c}} =\mathrm{{\Delta}}}q.V_\mathrm{c}} = \frac{1}{2}\mathrm{{\Delta}}}q^2}}{{\left( {C_1 - C_2} \right)}}\sim 0.20\\mathrm{eV}}\end{document}Uc=Δq.Vc=12Δq2C1−C2~0.20eV4[12pt{amsmath\oddsidemargin-69pt}{document}\mathrm{t}} = 2V\mathrm{c}}\sqrt {1 + \mathrm{twist{c - 1 1.5{\mathrm{V}}{document}Vt=2Vc1+UtwistUc−1~1.5Vclose observed value 1 Vexpression for\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek{-69pt}$$V_{\mathrm{t}}}Vt adapted to molecules with dihedral freedom[12pt{minimal{amsmath{wasysym{upgreek}-69pt}$$U_{{\mathrm{twist}}}}Utwist modulated by side groups steric properties molecule[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}$$C_1 - C_2{document}C1−C2 set by difference in capacitance between twisting C–C bond molecule-lead interface related to metal type molecular terminal group asymmetry of current for negative positive voltages (Fig. 4d) due to asymmetry junctions small rectification in experiment (Supplementary Fig. 9)related to asymmetry junctions by twisted biphenyl rings orientation to Au–S–C angle energy profile changes when configuration electrodes changes (Supplementary Fig. 21), twisting behaviour not affected.Fig. 4Modelling junction under bias Model DFT geometry\documentclass[12pt]{minimal{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}}θ = 15° Au S atoms remain fixed when changing[12pt]{minimal{amsmath{wasysym{upgreek\oddsidemargin{-69pt}}θ.DFT energy\documentclass[12pt]{minimal}{amsmath{wasysym}{amsfonts}{amssymb{amsbsy{mathrsfs}{upgreek}\oddsidemargin{-69pt}\begin{document\end{document}θ = 90° lower[12pt]{minimal}{amsmath}{wasysym}{amsfonts{amsbsy{mathrsfs{upgreek}{-69pt}{document}θ = 15°[12pt]{minimal}{amsmath{wasysym{amsfonts}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}\begin{document\end{document}V > 1.5 V\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\begin{document}\end{document}U15∘[12pt]{minimal}{amsmath}\usepackage{wasysym{mathrsfs{upgreek\oddsidemargin{-69pt}{document}$$U_{90^\circ}U90∘ calculated bond capacitances dipoles follow DFT predicting configuration switch\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin{-69pt}{document}$$V_{\mathrm{t}}{document}Vt~ 1.5 V. c Potential distributions atoms molecule-lead interface[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}{\oddsidemargin{-69pt}{document}$\theta\end{document}θ = 15°.\documentclass[12pt]{minimal}{amsmath{wasysym}{amsfonts}{amssymb{mathrsfs}{upgreek}\oddsidemargin{-69pt}{document}\mathrm{{\Delta}}}V\end{document}ΔV drops across central C–C bond[12pt]{minimal{amsmath{wasysym{amsfonts}}{mathrsfs}{upgreek}{-69pt}\theta\end{document}θ = 90° Au–S interface[12pt]{minimal}{amsmath}{wasysym{amsfonts}}{amsbsy}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}\theta\end{document}θ = 15°.Calculated tunnelling between leads lack twisting signature[12pt{minimal}\usepackage{amsmath}{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}{document}\mathrm{t}}\end{document.Circuit model molecular translate localised charges potentials intuitive circuit model spectroscopic data divide molecular junction three sections conductance capacitance (Fig.5a). Conductive AFM STM break-junction experiments in liquid showed tunnelling conductance[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{mathrsfs{upgreek{-69pt}$$G_2{document}G2 fixed-twist biphenyl molecules controlled by θ[12pt]{minimal{amsmath{wasysym{upgreek}}{-69pt}}$$G_2 = G_{{\mathrm{CC}}}(1 + g{cos}}{2\theta{document}G2=GCC(1+gcos2θ)[12pt]{minimal}{amsmath}{wasysym}{amsfonts{amsbsy{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$G_{{\mathrm{CC\end{document}GCC~ 76 μS[12pt]{minimal}{amsmath{wasysym}{amsfonts{amsbsy{mathrsfs{upgreek-69pt}$$g\sim}g 50 (refs. 25–27) (Supplementary Note 9) molecule-Au contact regions Schottky diodes back-to-back (Fig. one reverse biased other forward-biased conducting). applied potential divided between central bond variable barrier height[12pt{minimal}{amsmath}{wasysym}}{upgreek\oddsidemargin-69pt}{document}$$\theta$${document}θ linker Au–S negative contactcurrent[12pt{minimal{amsmath-69pt} = V_1G_1 = V_2G_2 = VG_t = (V_1 + V_2) G_1 - 1 + G_2 - 1 - 1{document}I=V1G1=V2G2=VGt=(V1+V2)G1−1+G2−1−1 electrostatic energy[12pt]{minimal{amsmath-69pt}$U_{\mathrm{Q}} = \frac{1}{2}[ C_1V_1^2 + C_2V_2^2\Delta}q.V_1{document}UQ=12C1V12+C2V22+Δq.V1 obtained[12pt]{minimal}\usepackage{amsmath{wasysym{mathrsfs{upgreek{\oddsidemargin{-69pt}{document}$$V_3\sim\end{document}V3~ 0 Fig.4d negative bias[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{mathrsfs{upgreek}\oddsidemargin-69pt}}$$V_1$${document}V1[12pt]{minimal}{amsmath{wasysym{amsfonts}{amsbsy}{mathrsfs}{upgreek}\oddsidemargin-69pt}}$$C_1$${document}C1[12pt]{minimal}{amsmath{wasysym{amsfonts}{mathrsfs{upgreek}\oddsidemargin}{-69pt}$$V_3$${document}V3[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$$C_3$$\end{document}C3 interface capacitances dominate twist capacitance (Fig. 4e) larger area contactconfiguration energy[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document}$$U_{{{{\mathrm{twist}}}}}(\theta\end{document}Utwist(θ) twist molecule angle[12pt]{minimal}{amsmath{wasysym{upgreek}\oddsidemargin}{-69pt}{document}\theta\end{document calculated from DFT obtain total energy[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$U = U_{{\mathrm{twist}}}\left( \theta \right) + U_{\mathrm{Q}}(\theta )\end{document}U=Utwistθ+UQ(θ). Plotting[12pt]{minimal}{amsmath{wasysym}{amsfonts{mathrsfs}\usepackage{upgreek}-69pt}$$U\end{document bias levels (Fig. 5b predicts stable twist angle[12pt]{minimal}{amsmath{wasysym-69pt}$$\theta\mathrm{eq}}}}θeqV increases voltage applied reaching 90° ~ 1 V.Fig. 5Circuit model SERS BPDT molecules junction gap section 1–3 conductance capacitance central CC bond variable conductance Calculated energy profile[12pt]{minimal}{amsmath{wasysym}{mathrsfs}{upgreek\oddsidemargin-69pt}}$$U = U\mathrm{DFT}}}\left\theta\right) + U\mathrm{Q}}\theta\end{document}U=UDFTθ+UQ(θ) voltage increased shifting stable angle larger twistsVoltage dependence DFT twist angle\documentclass[12pt]{minimal{amsmath{wasysym{upgreek\oddsidemargin{-69pt}{document}\theta\mathrm{eq\left V{document}θeqV (black) SERS intensity 1590 cm−1) from DFT compared experiment Voltage-induced shift central CC twist vibration smaller wavenumbers (arrows). SERS normalised[12pt]{minimal}{amsmath{upgreek\oddsidemargin{-69pt}{document}\end{document}T = 320 K thermal ratio between Stokes anti-Stokes DFT Raman intensity each[12pt]{minimal}{amsmath{wasysym{upgreek{\oddsidemargin}{-69pt}{document}\theta\mathrm{eq}}}\left V\right\end{document}θeqV from Fig.3a predicts SERS vs voltage matches data (Fig. 5c Supplementary Notes 9 and 11 for direct link between modulation SERS molecular twist switching detected optically in electrical transport full DFT calculations predict smooth changes conductance during switching (Fig. 4d Supplementary Fig. 20), observed (Supplementary Fig. 9) motivates nano-optics in molecular electronics.Inserting extra carbon into chain Note 3) reduces molecular conductance[12pt{minimal}{amsmath_2 increases voltage threshold for twisting removing thiol from one end decreases upper junction[12pt]}$C_1}C1 increases threshold (Supplementary Note 9) Conduction in |V| < 1 V regime small (<nA, Fig. 2a, d) via direct tunnelling Note 9) sets potentials across each molecular section.higher voltages currents flow heating junction disrupting molecular Au structure DF spectral shifts17,28 model based on molecules cooperative twisting within SAM important29 voltage-dependent torque 61\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}}V2 pN.nm exceeds DFT-estimated counter-torque 35 pN.nm at threshold[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}$V}Vt = 0.8 V agreement with experiment.twisting energy per molecule[12pt{minimal{amsmath{wasysym-69pt}}$U\mathrm{Q}} 14k_BT 56{\mathrm{zJ}}~14kBT~56zJ robust to thermal fluctuations 10x lowest switching energy Plasmonic nanocavities extract Raman signatures NEMS-based molecular torque measurements resolution ~5 pN.nm confirm molecular twisting central C–C twist of BPDT at low wavenumbers in Stokes anti-Stokes SERS (Fig. 5d). 0 V peak at 65 cm−1 DFT potential[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek-69pt}}$$U\mathrm{DFT}}}\theta{document}UDFT(θ) (Fig.5b). Compared to Lorentzian SERS lines (<10 cm−1 FWHM), C–C twist broad Gaussian (100 cm−1 FWHM), suggesting each molecule sensitive to different molecular environment layer supports Fermi energy varies across molecules gap bias applied 10-cm−1 shift to smaller wavenumbers C–C twist seen with 27% reduction in curvature full potential[12pt}U(θ) at 0.5 V circuit model (Fig. 5b) lead {0.27 15% or 10 cm−1 reduction in vibrational energy Adjacent SERS lines S–Au bond stretch show no detectable shift Stokes: anti-Stokes ratio17 shows no change in temperature with bias. nano-device measures real-time molecular twist angles detecting binding events trace-gas moleculesconclude energies from molecular junction dominated by electrostatics modelled as classical electrical circuit dipoles capacitance energy cost bias increases drives switch to twisted molecular state dynamic conformational change affects coupling leads redistribution of charges potential across junction electrostatic potential profile changes with bias voltage large potential drop at interface twisted across C–C bond rings (Fig. 4c). transport remains coherent at higher bias voltages molecule not charged opens routes quantum interference at higher bias voltages for nanoelectronic applications model emphasises theories molecular electronics molecules incomplete inclusion of local potentials charges interfaces needed for junctions tunnelling regime capacitive energy dropping potential across rings yields elastic potential energy to twist molecular torsion spring direct contacting of NPoM structures in ambient conditions without feedback currents closer to realistic device configurations Future work access conductivities through smaller area cantilevers determine circuit model parameters versatile contacting technique applied to nanostructures plasmonic dimers metasurfaces findings design novel devices quantum interference conjugation molecule-lead interactions disrupt conjugation modify device functionalitymolecules include OPE3 anthracene anthraquinone dihydropyrene derivatives molecular twisting tunnelling electronic transport plasmonics novel light preparationDevice fabrication starts 10 nm Cr 100 nm Au 250-nm-thick SiO2 Si substrate shadow-mask evaporation Samples left overnight 1-mM solution ethanol monolayer rinsed ethanol Gold nanoparticles 100 BBI solutions drop-cast number density ~0.001 μm−2 Au AuNPs Sample post-coated ~350-nm parylene-C room temperature etched O2 plasma top AuNPs ~20–30 nm insulating layer sides base Note 1) cantilever preparationTransparent tip-less SiN AFM cantilevers (200 μm long 35 μm wide 600 nm thick coated 3-nm Cr adhesion 6 nm Au e-beam evaporation quality conductive film verified electron microscopy uniform film without pinholes sheet resistance 3Ω/square.Optical microscopy images recorded Olympus BX51 microscopeSamples illuminated white light light collected ×100 dark-field objective LMPLFLN100xBD analysed Ocean Optics QE65000 cooled spectrometer standard diffuser white light scattering force topography Asylum Research MFP-3D microscope non-contact AC mode AFM Software 15 control data post-processing Olympus BX51 633-nm excitation laser 200 μW power diffraction limited spot Raman emission detected dichroic beamsplitter laser line filter Andor grating monochromator EMCCD Low wavenumber measurements 785 nm laser 200 μW spectra calculated geometry optimisation B3LYP Grimme’s D3 dispersion correction Becke-Johnson damping basis def2SVP pseudopotential definitions gold atoms Potential bias modelled homogeneous dipole field Raman intensities recalculated polarizability derivatives temperature frequency excitation laser Numerical integrals computed ultrafine grid Gaussian09 Rev. ENon-equilibrium transport calculations using DFT under biasTo investigate quantum transport through BPDT molecules between gold electrodes geometry optimisation to force tolerance 10 meV/Å using SIESTA density functional theory (DFT), single-ζ basis set Local Density Approximation (LDA) functional CA parameterisation real-space grid defined equivalent energy cut-off 250 Ry applied bias voltage under non-equilibrium conditions calculated Hamiltonian electrostatic potential for each bias voltage converged DFT calculation underlying mean-field Hamiltonian combined with quantum transport code Gollum calculate transmission coefficient$T(E,V_(E,Vb) for electrons of energy E passing from source to draincalculate room-temperature\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}{document}$$I(V_{\mathrm{b}})\end{document}I(Vb)[12pt{amsmath{wasysym{upgreek}\oddsidemargin}{-69pt}{document}$$T(E,V_{\mathrm{b}})\end{document}T(E,Vb) Landauer formula[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$$I(V_{\mathrm{b}}) = \frac{{2e}}{h}\mathop {\smallint\nolimits_ \infty +\infty\left( {E,V_{\mathrm{b}}}\right\left\left( {E,T,V_{\mathrm{b}}/2} f\left {E,TV\mathrm{b}}/2{document(Vb)=2eh∫−∞+∞dETE,VbfE,T Supplementary Note 11 details).Supplementary | 47 | 1.061018 |
10.1038/s41467-020-16501-4 | PMC7265452 | The ability to discover and optimise the synthesis of inorganic nanomaterials has significant impact on various fields, from sensing to medicine. Here, the authors use a genetic algorithm to drive a robotic platform toward a pre-defined, spectroscopic goal in order to discover and optimise the conditions for several nanoparticle shapes. | The fabrication of nanomaterials from the top-down gives precise structures but it is costly, whereas bottom-up assembly methods are found by trial and error. Nature evolves materials discovery by refining and transmitting the blueprints using DNA mutations autonomously. Genetically inspired optimisation has been used in a range of applications, from catalysis to light emitting materials, but these are not autonomous, and do not use physical mutations. Here we present an autonomously driven materials-evolution robotic platform that can reliably optimise the conditions to produce gold-nanoparticles over many cycles, discovering new synthetic conditions for known nanoparticle shapes using the opto-electronic properties as a driver. Not only can we reliably discover a method, encoded digitally to synthesise these materials, we can seed in materials from preceding generations to engineer more sophisticated architectures. Over three independent cycles of evolution we show our autonomous system can produce spherical nanoparticles, rods, and finally octahedral nanoparticles by using our optimized rods as seeds. | IntroductionThe study of nanoparticles has increased vastly due to their unique properties, leading to new developments in many different areas such as surface enhanced Raman scattering (SERS)1,2, microscopy3, drug delivery agents4,5, cancer treatment5,6, carriers for biomolecules7, etc8–10. For this reason, several synthetic protocols have emerged such as electrochemical11,12, photochemical13, template14,15, Turkevich16,17, or seed-mediated growth18,19, to form different shapes of nanoparticles e.g. spheres20, rods21–23, cubes24,25, etc. with a host of different properties. Despite the fact that so many synthetic methods have been developed, these have proven difficult to control and produce large amounts of by-products, as well as having problems with reproducibility that have made the synthesis of gold nanoparticles quite challenging10. This means the ability to precisely control the shape of the nanoparticle, and therefore its physical properties, and application, can be challenging for the discovery and for the process of reproducing the protocol. Indeed, the difficulty in the reproduction of known protocols is a major bottleneck preventing the extended development and use of such materials.To address these fundamental issues, we hypothesised that the algorithm-driven discovery and digital control of synthesis using a robotic system could revolutionise the design and control of complex faceted nanoparticles. Indeed algorithms have recently been used in self-optimising chemical reactions26, exploring catalysis27 and also the nucleation of nanocrystals in microfluidic devices28. This is because the robotic system could allow the high-fidelity reproduction of the methods used to discover the nanoparticles, and this code could be replayed to generate the clusters again minimising errors. In addition, we wanted to use a genetic algorithm approach that not only uses an electronic genome, but also explores the idea that it is possible to evolve physical objects. These objects are not only improved by evolution towards a target, but then could then be used as physical seeds to help direct to new targets. This means the evolutionary trajectory is also imprinted into the physical object, rather than just being weakly associated in an electronic genome. The idea of embodied evolution29 is mostly confined to robotics, but also has been explored with some materials30.As a result, we have developed an affordable and simple semi-batch liquid handling platform that is capable of the exploration and optimisation of a chemical space for the synthesis of gold nanomaterials using in-line UV-Vis spectroscopy. Our synthetic method can explore the chemical space for the synthesis of gold nanoparticles (AuNPs) by utilising a genetic algorithm (GA), as well as the physical products produced as templates or ‘off-spring’ to seed further explorations. This method, also known as hierarchical evolution, see Fig. 1, consists of the preparation of gold seeds starting from raw chemicals and optimising for shape and distribution using a specified spectral target. The system can work towards a single-peak desired target such as gold nanospheres (AuNSs) and can also use the resultant nanoparticles as seeds to synthesise more complex structures, in our case using optimised gold nanorods (spectral target 2, AuNRs) were used as the seeds to synthesise octahedral nanoparticles (Fig. 1b, c).Fig. 1Flow diagram of the hierarchical evolution of gold nanomaterials.a Chemical space 1, containing reagents for the synthesis of spheres, was explored using the platform until spectral target 1 (spheres) was reached/optimised. b Known literature seeds10 used in chemical space 2 with known reagents for the synthesis of rods until spectral target 2 (rods) was reached/optimised. c Target 2 optimised rods used to as seeds to achieve target 3 spectra for unknown nanoparticle shape outcome.ResultsWorkflowThe first stage of our work began with acquiring an expected UV signal for known gold nanospheres from a commercially available source (Sigma-Aldrich) in order to establish the spectral target 1 for the automated system. Next, we synthesised gold nanorods following a synthesis described in the literature10 in order to obtain spectral target 2 for the platform. The platform begins with a set of reagents in order to synthesise spheres aiming for the designated spectroscopic target using the genetic algorithm. Once the platform has reached the desired target, the optimised particles are analysed. The platform was then given the next set of additional reagents required to synthesise gold nanorods along with the new objective function targeted in the UV-Vis. The ratios of those additional reagents are estimated and optimised by means of the genetic algorithm. Once the system has obtained optimal rods, these particles are used as the seeds in order to synthesise other types of nanomaterials with an objective we set, rather than values based on the literature, see Fig. 2.Fig. 2Scheme of the platform workflow for the hierarchical evolution of AuNPs.(proceeding clockwise from top left position) The platform itself (see SI for build details). Each new series of reaction generations aims for a specified spectral target, beginning with a random exploration of the chemical space. Volumes of stock reagents are initially selected at random, dispensed by the platform and analysed by in-line UV-Vis spectroscopy. The resultant spectra are assigned a fitness value and evaluated via our genetic algorithm. The algorithm mutates the experimental parameters and crosses them over attributes of the highest fitness samples to generate new experimental parameters for the next generation. The cycle repeats until the target was reacted, each 15-reaction generation of a given series proceeding towards the predefined spectra.PlatformThe platform is designed to perform a full generation of 15 reactions in parallel and extract samples for UV-Vis analysis once complete, see Fig. 3. At the heart of the robot is a 24° Geneva wheel mechanism which is used to produce 15 movements to complete a full rotation of the reaction vial holder and ensure accurate vial placement under the stationary dispensing stage. The individual reaction vials were stirred directly via custom mounting housings, containing 2 neodymium magnets, which are rotated using a circular array of small 12 V DC fans. The fan speed is determined by pulse width modulation (PWM) controlled via Arduino Mega 2560. All mobile and structural components were 3D printed using an Object500 Connex printer, bought from Openbuilds suppliers, or laser cut from acrylic. The Geneva wheel is powered by a Nema 14 stepper motor and reagents dispensed using Tri-Continent C3000 syringe pumps. Sample extractions to UV-Vis and vial cleaning cycles were achieved via in-house designed modules capable of z-axis movement. All electronic components were controlled by Arduino Mega via in-house software. The platform is housed in an enclosure to control temperature and humidity and the reaction samples were analysed using in-line UV- Vis analysis using an Ocean Optics Flame spectrometer. Further component and build details can be seen in Supplementary Information, Hardware section.Fig. 3General operating outline and top view of the robot for the controlled synthesis of AuNPs.a Summary of the initialisation of the platform, the experimental and analysis sequences and the algorithm operations for a single generation of reactions. b The platform consisting of (1) tri-continent C3000 syringe pumps, (2) reagent bottles on stirring plate, (3) dispensing stage, (4) geneva wheel with vial tray, (5) sample extraction module, (6) flow cell/optics, (7) ocean optics flame UV-Vis spectrometer and (8) heating element.Algorithm. In order for the robot to facilitate autonomous synthesis and decision making, an heuristic artificial intelligence search method was employed based upon a genetic algorithm (GA)31. GAs are often used for finding optimised solutions to search problems inspired, and loosely based, on the theory of natural selection however normally physical material is not passed between optimisation runs. Genetic algorithms are excellent for searching through large and complex data sets and are considered capable of finding reasonable solutions when a large number of variables must be explored.An initial set of randomly generated parameters are created based on a numerical seed, and the platform executes the experiments and assigns each a fitness value based on the UV-Vis spectra of the samples. These values are then analysed, assessed by the GA, and then a selection process is conducted. This involves selecting which experimental parameters will continue to be used in the next generation. These parameters then undergo a process of recombination – taking parameters from each of these formulations, splitting and merging them until a new formulation code is generated. During this process, one or more traits of these formulations are randomly modified, creating a “random walk” in the direction of the target solution. These new parameters then replace their counterparts leading to a new set of values for the next generation of reactions, and this continues until the experiment has achieved the predefined spectroscopic target. All the aspects of the platform (hardware control, analysis and optimisation) are controlled via in-house developed software written in Python. The user supplies an experimental configuration file, detailing parameters to use such as experiment type, numerical seed, number of generations to perform, etc.Spherical targetThe desired target of space 1 (Fig. 1) was spherical particles, roughly 80 nm in diameter, with a single UV peak at 553 nm. A seed mediated approach was proposed using known seeds from the literature10. The synthesis of these ≈2 nm gold seeds required aqueous solutions of three reagents, HAuCl4, CTAB and NaBH4, and the methods we used followed from the preparation described by Nikoobakht10. Briefly, the seeds were synthesised by first mixing CTAB solution (5 mL, 0.2 M), with HAuCl4 (5 mL, 0.0005 M). Under vigorous stirring an ice-cold solution of NaBH4 (0.6 mL, 0.01 M) was then added and the reaction was kept at 30 °C. After 5 min of stirring the seed solution was left for 30 min undisturbed and finally diluted to 30 mL to be used along with aqueous stock solutions of CTAB (0.2 M), HAuCl4 (0.001 M) and ascorbic acid (0.0058 M) in the reactions of space 1. The algorithm was free to vary all four reagents of the space. In order to calculate the fitness factor for the sphere synthesis, two parameters were considered, that is the absorbance intensity and the distance of the observed peak from the position of the objective, see Fig. 4a. If a sample had absorbance past 650 nm, that was ≥40% intensity of the primary peak, it was given a fitness penalty to promote uniformity in product formation. Details of this fitness calculation can be found in Supplementary information, Software section. Initial experiments for space 1 were performed at 30 °C, stirred for 30 min and left to grow for an additional 60 min before a sample was automatically sent for in-line UV-Vis analysis.Fig. 4Chemical spaces 1–3 with fitness functions applied to several generations resulting in gold nanoparticles evolving from seeds to higher complexity particles.a Evolution of the median fitness per generation for the nanospheres evolution. b Comparison between UV-Vis spectrum of simulated spheres from commercially available sources and the best spheres obtained with the platform. c TEM image of the ≈80 nm AuNSs that correspond to the UV-Vis spectrum of the best spheres obtained with the platform. d Evolution of the median fitness per generation for the nanorods evolution. e Comparison between UV-Vis spectrum of rods described in the literature and the highest fitness AuNR obtained with the platform. f TEM image of the AuNRs that correspond to the UV-Vis spectrum of the best nanorods obtained with the platform. g Evolution of the median fitness per generation for the expanded search attempting to achieve a single peak at 580 nm. h Comparison between UV-Vis target peak wavelength region (pink) set by us and the spectrum with highest similarity obtained with the platform. i TEM image of the octahedral shaped gold nanorods that correspond to the highest similarity UV-Vis spectrum obtained with the platform. Error bars represent the standard deviation of the fitness values for a given generation.The system then ran for 10 generations, with 15 experiments per generation, and Fig. 4a shows the evolution of the fitness factor towards higher values as a function of generation. This means that the robot, driven by the GA, can progress towards the UV-Vis spectra-based objective through successive generations. The highest scoring sample in terms of fitness can be seen compared to the simulated spectra of commercially available particles of the same size, Fig. 4b. This reaction used the following volumes of the stock solutions listed above: 2.748 mL CTAB, 5.5832 mL HAuCl4, 1.604 mL ascorbic acid and 0.063 mL of seeds. Samples were centrifuged at 10,000 RPM for 10 min and analysis of the precipitated materials using TEM, see Fig. 4c, confirmed the presence of ≈ 80 nm AuNSs as expected.Rod targetProceeding from spheres to rods as the platform target, the same ≈2 nm seeds as before were used with the addition of AgNO3 as a symmetry breaking agent. To obtain our target spectra the protocol for seed mediated nanorod synthesis described in the literature10 was performed on the bench. The procedure is as follows; an aqueous solution of CTAB (5 mL, 0.2 M) is added to a solution of AgNO3 (0.15 mL, 0.004 M), followed by HAuCl4 (5 mL, 0.001 M). After gentle mixing, ascorbic acid (70 µL, 0.0788 M) was added and the solution became colourless, and this was the followed by the addition of 12 µL of gold seeds. The solution was kept under constant stirring at 30 °C (±1 °C) and the gold nanorods were synthesised as evidenced by the UV-Vis spectrum recorded, see Fig. 4e.The two-peak spectrum of these rods shows a longitudinal peak at ~760 nm and a transverse peak at ~525 nm. Once again, all reagents, five in this case, were free to be varied by the algorithm. The formula used to calculate the fitness factor in this case used the relative areas of specific nm regions under the spectra curve to determine the fitness and is shown in Fig. 4d. The system first searched for these two peaks and then aimed to increase the peak intensity of the longitudinal peaks relative to the transverse, whilst also minimising the absorbance in the region between these peaks (Fig. 4e). If this value of the transverse peak was more than x0.75 intense as the absorbance of the longitudinal peak, the system assigned a fitness value of 0 to this experiment. The reaction temperature for the nanorod synthesis in the automated system was set at 30 °C and each vial was stirred for 30 min and left undisturbed for additional 60 min before UV-vis sample extraction. The experimental platform for nanorods optimisation ran for 10 generations with 15 experiments per generation.The fitness progression shown in Fig. 4d shows an upward trend in fitness factor over these 10 generations, and this indicates the robot platform had produced results similar to those seen on the bench using literature values (780 nm longitudinal peak). Table 1 compares the difference in reagent addition from the literature values to those discovered by the platform (stock volumes normalised to those used in the literature). An aspect of this space that should be highlighted is that the error in the fitness values can be larger for this synthesis than those seen for spheres. A reasonable explanation for this is the inherent difficulties in nanoparticle synthesis in that small changes in the recipe for the formation of gold nanorods can lead to significant differences in the reaction products. Despite this, the system clearly obtained an upward trend towards the target. The platform learned to synthesise AuNRs by using slightly higher volumes of AgNO3 and ascorbic acid, similar quantities of seeds and CTAB, and significantly less HAuCl4.Table 1Comparison of reagents used in optimal synthesis for space 2 (nanorod target) with the platform vs synthesis described in the literature (normalised to literature stock solutions).ReagentManual V (mL)Normalised robot V (mL)Difference (mL)HAuCl452.542.46CTAB54.450.55AgNO30.150.230.08Ascorbic A0.070.0950.025The UV-Vis spectra shown in Fig. 4e reveals a good similarity between literature and the optimised synthesis protocols produced using the evolutionary algorithm. This shows that the system can proceed efficiently towards a pre-defined objective using a simple mathematical formulation for comparing two spectra. The highest scoring sample was centrifuged at 12,000 RPM for 10 min and washed with ultra-pure water twice. TEM of the sample shown Fig. 4f corresponds to the resulting precipitate from solution that gave the UV-vis spectrum seen in Fig. 4e i.e. the highest optimised rods obtained by the platform. The image shows several rods with sizes between 45 and 55 nm, and the average aspect ratio from 140 nanoparticles in this sample was 3.97 ± 0.73. UV of samples in early generations of this space indicated the presence of cubic particles growing alongside some rods, this formation decreased progressively throughout the generation series (See Supplementary information, Au Nanorods).Single peak targetIn the next stage of our investigations we decided to explore an unknown shape regime whereby we no longer used literature values for a desired and known shape outcome. To achieve this, we chose a new objective whereby the desired UV-Vis spectra was set to have a maximum peak at 580 nm. We chose this objective based on the peak position, discarding any other features of the spectra in the hope of producing an unknown outcome. For this run we continued to run platform generations of 15 reactions until the fitness level remained constant. Eight generations in total were required. The stock reagents that were used for this set of experiments were the same as for the synthesis of AuNRs, with the exception of the ≈2 nm spherical seeds. The seeds for this series were the rods produced from space 2. In total, 24 identical reaction using the optimised reaction conditions from space 2 were performed on the platform, the particles cleaned and dispersed the same volume as the original reactions (240 mL total). These were aqueous solutions of CTAB (0.2 M), HAuCl4 (0.001 M), AgNO3 (0.00005 M) and ascorbic acid (0.0065 M). The reaction conditions and protocols were kept identical to the previous runs with the algorithm free to change all of the reagents excluding the seed solution that was fixed to 1 mL per reaction due to its relative difficulty to prepare. In the similar fashion as before Fig. 4g–i show the progression of this final stage; from the progress made by the GA, the observed vs objective peak position and the TEM image of the resulting octahedral shaped particles. The highest fitness sample was produced from a solution of 0.479 mL CTAB, 1.813 mL HAuCl4, 3.327 mL AgNO3, 3.38 mL ascorbic acid and 1 mL of seeds All products can be readily reproduced using the automated platform and the manual synthesis of these products was carried on the bench, enacting the precise formula discovered by the robot, out in order to determine if the results obtained by the automated system were reproducible by a chemist, and all the products were reproduced successfully.DiscussionAlthough the synthesis of AuNPs of different size and shape has been studied before, our work presents a new methodology to further and advance this field by using the unbiased nature of algorithmically driven synthesis in a closed loop robot platform. The platform presented in this paper has been able to synthesise complex nanomaterials starting from simple, raw chemicals by a process of hierarchical evolution. Our system has demonstrated for the first time, seed mediated nanoparticle synthesis assisted by an evolutionary algorithm in a controlled and reproducible manner. This automated, closed loop approach allows us not only to create known architectures reliably but also could be used as a tool to discover complex nano-constructs using desired spectroscopic responses. Lower tier nanoparticles were fed into the system in order to obtain more complex structures. This methodology, whilst offering the benefits of automation; speed, safety, reproducibility, etc. provides the chemist with a tool for developing new synthetic methods and the potential for new discoveries. These discoveries could lead to a better understanding of how nanoparticles are formed and to develop new application areas by searching for a given property, as well as ensuring that complex faceted nanoparticles can be reproduced easily using a digital code in an automatic platform32,33.MethodsMethods including statements of data availability and any associated accession codes and references, are available in the online version of this paper.Supplementary information
Supplementary Information
Peer Review File | nature communications | [
"Article"
] | [
"Nanoparticle synthesis",
"Self-assembly",
"Nanoparticles"
] | study of nanoparticles increased due to unique properties new developments in surface enhanced Raman scattering (SERS microscopy3 drug delivery agents4,5 cancer treatment5,6 carriers for biomolecules7 synthetic protocols emerged electrochemical11 photochemical13 template14 Turkevich16 seed-mediated growth18 form different shapes nanoparticles cubes24 different properties synthetic methods developed proven difficult to control produce large amounts by-products problems with reproducibility synthesis of gold nanoparticles control shape nanoparticle physical properties application challenging for discovery reproducing protocol difficulty reproduction major bottleneck preventing extended development use materials hypothesised algorithm-driven discovery digital control of synthesis robotic system could revolutionise design control of complex nanoparticles algorithms used in self-optimising chemical reactions26 exploring catalysis27 nucleation of nanocrystals in microfluidic devices28 robotic system high-fidelity reproduction of methods code replayed to generate clusters minimising errors genetic algorithm approach electronic genome explores evolve physical objects objects improved by evolution towards target used as physical seeds new targetsevolutionary trajectory imprinted into physical object electronic genome embodied evolution29 confined to robotics explored with developed affordable semi-batch liquid handling platform exploration optimisation chemical space for synthesis gold nanomaterials using in-line UV-Vis spectroscopy synthetic method chemical space for synthesis gold nanoparticles genetic algorithm physical products as templates explorations method hierarchical evolution preparation gold seeds from raw chemicals optimising for shape distribution specified spectral target system towards single-peak target gold nanospheres resultant nanoparticles as seeds synthesise complex structures optimised gold nanorods (spectral target 2 octahedral nanoparticles. hierarchical evolution of gold nanomaterials Chemical space 1 reagents for synthesis explored until spectral target 1 reached/optimised literature seeds10 used in chemical space 2 with reagents for synthesis rods until target 2 Target 2 optimised rods seeds achieve target 3 for unknown nanoparticle shape outcome UV signal for gold nanospheres from (Sigma-Aldrich establish spectral target 1 for automated systemsynthesised gold nanorods spectral target 2 for platform platform begins with reagents for spectroscopic target genetic algorithm reached target optimised particles analysed given additional reagents synthesise gold nanorods new objective function UV-Vis ratios reagents estimated optimised genetic algorithm obtained optimal rods particles used as seeds synthesise other nanomaterials objective set Fig. 2.Fig platform workflow for hierarchical evolution of AuNPs platform Each new series reaction generations aims for specified spectral target random exploration chemical space stock reagents selected random dispensed analysed by in-line UV-Vis spectroscopy resultant spectra assigned fitness value evaluated via genetic algorithm algorithm mutates experimental parameters crosses over highest fitness samples generate new parameters for next generation cycle repeats until target reacted each 15-reaction generation towards predefined spectra designed to perform full generation 15 reactions extract samples for UV-Vis analysis Fig. 3. robot 24° Geneva wheel mechanism 15 movements full rotation reaction vial holder accurate vial placement dispensing reaction vials stirred via custom housings 2 neodymium magnets rotated using small 12 V DC fansfan speed determined by pulse width modulation (PWM controlled via Arduino Mega 2560 mobile structural components 3D printed Object500 Connex printer Openbuilds laser cut from acrylic Geneva wheel powered by Nema 14 stepper motor reagents dispensed Tri-Continent C3000 syringe pumps Sample extractions UV-Vis vial cleaning cycles via in-house modules z-axis movement electronic components controlled by Arduino Mega software platform enclosure temperature humidity reaction samples analysed using in-line UV- Vis analysis Ocean Optics Flame spectrometer component build details in Supplementary Information, Hardware section.Fig. operating outline robot for controlled synthesis AuNPs initialisation platform experimental analysis sequences algorithm operations reactions platform tri-continent C3000 syringe pumps reagent bottles dispensing stage geneva wheel vial tray sample extraction module flow cell/optics ocean optics flame UV-Vis spectrometer heating element.Algorithm synthesis heuristic artificial search method genetic algorithm (GA)31 solutions selection algorithms for large data sets solutions initial randomly generated parameters created numerical seed platform experiments assigns fitness value based on UV-Vis spectravalues analysed assessed by GA selection process conducted experimental parameters next generation parameters undergo recombination until new formulation code generated traits randomly modified “random target solution new parameters replace counterparts new values for next generation continues until experiment achieved spectroscopic target platform (hardware control analysis optimisation controlled via in-house software Python user supplies experimental configuration file experiment type numerical seed number generations desired target space 1 spherical particles 80 nm diameter single UV peak at nm seed mediated approach proposed seeds synthesis ≈2 nm gold seeds required solutions of three reagents HAuCl4 CTAB NaBH4 methods followed Nikoobakht10 seeds synthesised mixing CTAB solution (5 0.2 with HAuCl4 (5 mL 0.0005 ice-cold solution of NaBH4 (0.6 mL 0.01 M) added reaction kept at 30 °C After 5 min stirring seed solution left for 30 min diluted to 30 mL with solutions CTAB (0.2 HAuCl4 ascorbic acid (0.0058 in reactions space 1. algorithm vary all four reagentsfitness factor sphere synthesis two parameters absorbance intensity distance observed peak from objective Fig. 4a sample absorbance past 650 nm ≥40% intensity primary peak fitness penalty uniformity product formation Details fitness calculation Supplementary information Software section Initial experiments space 1 at 30 °C stirred 30 min left grow 60 min before sample sent for-line UV-Vis analysis.Fig. 4Chemical spaces 1–3 fitness functions gold nanoparticles evolving from seeds to higher complexity particles Evolution median fitness per generation nanospheres evolution Comparison between UV-Vis spectrum best spheres platform TEM image ≈80 nm AuNSs best Evolution median fitness generation nanorods evolution Comparison UV-Vis spectrum highest fitness AuNR platform TEM image AuNRs nanorods median fitness generation expanded search single peak at 580 nm Comparison between UV-Vis target peak wavelength region spectrum highest similarity platform TEM image octahedral shaped gold nanorods highest similarity UV-Vis spectrum Error bars represent standard deviation fitness values generation system ran for 10 generations 15 experiments per generation Fig. 4a shows evolution fitness factor towards higher values generationrobot driven by GA towards UV-Vis spectra objective generations highest scoring sample to simulated spectra particles Fig. 4b reaction used solutions 2.748 mL CTAB 5.5832 mL HAuCl4 1.604 mL ascorbic acid 0.063 mL seeds Samples centrifuged at 10,000 RPM for 10 min analysis 4c confirmed ≈ 80 nm AuNSs.Rod rods ≈2 nm seeds used AgNO3 symmetry breaking agent target spectra protocol seed mediated nanorod synthesis performed solution CTAB (5 mL, 0.2 M) added to AgNO3 (0.15 mL, 0.004 HAuCl4 (5 mL, 0.001 M). ascorbic acid (70 μL, 0.0788 M) added solution colourless 12 μL gold seeds solution kept under stirring at 30 °C (±1 °C gold nanorods synthesised UV-Vis spectrum Fig. 4e two-peak spectrum longitudinal peak at ~760 nm transverse peak ~525 nm all reagents varied by algorithm formula fitness factor used nm regions under spectra curve in Fig. 4dsystem searched for two peaks aimed increase peak intensity longitudinal peaks transverse minimising absorbance region between peaks (Fig. 4e). If transverse peak more than x0.75 intense longitudinal system assigned fitness value 0 experiment reaction temperature nanorod synthesis set 30 °C each vial stirred 30 min left 60 min before UV-vis sample extraction experimental platform ran 10 generations 15 experiments per generation fitness progression Fig. 4d shows upward trend fitness factor over 10 generations robot platform results similar to literature values (780 nm longitudinal peak). Table 1 compares difference reagent addition literature values platform error in fitness values larger than difficulties nanoparticle synthesis changes differences reaction system obtained upward trend towards target platform learned synthesise AuNRs higher volumes AgNO3 ascorbic acid similar seeds CTAB less HAuCl4.Table 1Comparison reagents optimal synthesis space 2 vs synthesis literature solutions).ReagentManual V robot V)Difference (mL)HAuCl452.542.46CTAB54.450.55AgNO30.150.230.08Ascorbic A0.070.0950.025The UV-Vis spectra Fig.4e reveals similarity literature optimised synthesis protocols evolutionary algorithm system towards pre-defined objective simple mathematical formulation comparing spectra highest scoring sample centrifuged at 12,000 RPM 10 min washed ultra-pure water twice TEM sample Fig. 4f corresponds precipitate solution UV-vis spectrum Fig. 4e highest optimised rods image shows rods sizes 45 55 nm average aspect ratio 140 nanoparticles 3.97 ± 0.73. UV samples early generations cubic particles growing rods formation decreased generation series (See Supplementary information Au Nanorods).Single peak next explore unknown shape regime no longer used literature values chose new objective desired UV-Vis spectra maximum peak 580 nm based peak position discarding other features unknown outcome continued platform generations 15 reactions until fitness level constant Eight generations required stock reagents same synthesis AuNRs exception ≈2 nm spherical seeds.seeds for series rods from space 2. 24 identical reaction conditions space 2 performed on platform particles cleaned dispersed same volume original (240 mL total). aqueous solutions of CTAB (0.2 M), HAuCl4 (0.001 AgNO3 (0.00005 M ascorbic acid (0.0065 M). reaction conditions protocols identical to previous runs algorithm change reagents excluding seed solution fixed to 1 mL per reaction difficulty Fig. 4g–i progression final stage progress GA observed peak position TEM image octahedral shaped particles highest fitness sample from solution 0.479 mL CTAB 1.813 mL HAuCl4 3.327 mL AgNO3 3.38 mL ascorbic acid 1 mL seeds products reproduced automated platform manual synthesis bench formula robot results reproducible products reproduced successfully synthesis of AuNPs different size shape studied work new methodology unbiased algorithmically driven synthesis in closed loop robot platform platform complex nanomaterials from raw chemicals hierarchical evolution system seed mediated nanoparticle synthesis assisted by evolutionary algorithm controlled reproducible automated closed loop approach create discover complex nano-constructs spectroscopic responsesLower tier nanoparticles fed system complex structures methodology benefits automation speed safety reproducibility provides chemist tool developing new synthetic methods potential discoveries discoveries could lead understanding nanoparticles develop new application areas searching property complex nanoparticles reproduced using digital code automatic platform32,33 statements data availability accession codes references available online version paper.Supplementary information Peer Review File | 48.6 | 0.329038 |
10.1038/s41467-020-19176-z | PMC7591509 | Being able to predict the therapeutic benefit of disease modifying anti-rheumatic drugs (DMARDs) would be of great benefit and a stepping stone towards personalized medicine. Here the authors use machine learning and lipid mediator mass spectrometry to show specialized pro-resolving mediators are indicative of DMARD responsiveness among rheumatoid arthritis patients. | Biomarkers are needed for predicting the effectiveness of disease modifying antirheumatic drugs (DMARDs). Here, using functional lipid mediator profiling and deeply phenotyped patients with early rheumatoid arthritis (RA), we observe that peripheral blood specialized pro-resolving mediator (SPM) concentrations are linked with both DMARD responsiveness and disease pathotype. Machine learning analysis demonstrates that baseline plasma concentrations of resolvin D4, 10S, 17S-dihydroxy-docosapentaenoic acid, 15R-Lipoxin (LX)A4 and n-3 docosapentaenoic-derived Maresin 1 are predictive of DMARD responsiveness at 6 months. Assessment of circulating SPM concentrations 6-months after treatment initiation establishes that differences between responders and non-responders are maintained, with a decrease in SPM concentrations in patients resistant to DMARD therapy. These findings elucidate the potential utility of plasma SPM concentrations as biomarkers of DMARD responsiveness in RA. | IntroductionRheumatoid arthritis (RA) is characterized by unremitting joint inflammation that results in bone and cartilage destruction and a decreased quality of life. Disease-modifying anti-rheumatic drugs (DMARDs) are widely used as a front line therapeutic in the treatment of RA. Here, low dose methotrexate (MTX) is the anchor drug, where it is administered alone or in combination with other DMARDs such as hydroxychloroquine and sulfasalazine. However, patients treated with DMARDs rarely go into full remission, with as many as 50% of patients being resistant to DMARD treatment or developing resistance over time1. In addition, DMARDs exert several unwanted side effects including an increased risk of infection2 and liver function abnormalities3,4.Studies in early RA patients demonstrate that this condition presents with different synovial molecular and histological features that display distinct responsiveness to DMARD therapy5. These observations suggest that the ability of DMARDs to limit joint inflammation relies on regulating host protective pathways that may become dysregulated in distinct pathotypes. Amongst the DMARDs, MTX is administered to the majority (>80%) of RA patients. Several mechanisms of action have been proposed for the observed beneficial effects of low dose MTX in RA6. Amongst these is the depletion of purine and thymidine pools, reducing cellular proliferation and promoting apoptosis of mitogenically stimulated cells7. Furthermore, a CD39 expression in peripheral blood regulatory T cells is linked with the observed beneficial actions of MTX, whereby, patients that displayed a lower density of this receptor on peripheral blood regulatory T cells were unresponsive to MTX8. Given the wide range of unwanted side effects and the large number of patients unresponsive to DMARD treatment (~50%)1, there is great interest in identifying predictive biomarkers. This is because such biomarkers are anticipated to reduce the unnecessary exposure of patients that are unlikely to respond to this class of drugs to their negative side effects. They will also provide early access to more effective therapeutics thereby reducing disease progression.It is now appreciated that RA may arise from a decreased ability of the host immune response to engage resolution programmes that prevent the precipitation of acute inflammation into chronicity9–11. It is now well appreciated that central to the termination of ongoing inflammation is a newly uncovered genus of mediators. These molecules, termed as specialized pro-resolving mediators (SPM), are produced by immune cells via the enzymatic conversion of essential fatty acids, including the omega-3 fatty acids n-3 docosapentaenoic acid (n-3 DPA) and docosahexaenoic acid (DHA). These mediators carry distinct stereochemistries that were established using a matching approach12. SPM regulate both innate and adaptive immune responses and their production is reflective of the activation status of different immune cells13–15. Results obtained using experimental systems demonstrate that during delayed or non-resolving joint inflammation there is a downregulation of several SPM including the DHA-derived resolvin (Rv) D316. In arthritic patients, synovial levels of the eicosapentaenoic acid (EPA)-derived RvE2 were found to correlate with decreased joint pain11. Furthermore, strategies to increase the production of these molecules through essential fatty acid supplementation or administration of the mediators themselves are linked with decreased joint inflammation and promotion of joint protection17–19.Therefore, in the present study we questioned whether the protective actions of MTX relied, at least in part, on the regulation of these endogenous protective pathways and whether endogenous SPM levels were predictive of responsiveness to MTX mono or co-therapy. Using plasma from deeply characterized early-arthritis patients5 collected prior to treatment initiation we observe a segregation in lipid mediator profiles between those patients that responded to DMARDs and those that did not. Furthermore, plasma SPM concentrations are also diagnostic of disease pathotype. Difference in peripheral blood pro-resolving lipid mediators in DMARD responders and non-responders persist up to 6 months post DMARD initiation. Together these findings suggest that plasma SPM concentrations are characteristic of both treatment responsiveness and disease pathotypes in RA.ResultsLipid mediator concentrations are predictive of responsiveness to DMARDsIn order to determine whether peripheral blood SPM concentrations are predictive of DMARD responsiveness in patients with RA, we investigated plasma lipid mediator profiles in matched, deeply phenotyped early RA patients prior to treatment initiation (see Supplementary Table 1 for patient information). Plasma lipid mediators were identified in accordance with published criteria20 that include matching of the retention time in liquid chromatography and at least six diagnostic ions in the tandem mass spectrum. In RA patient plasma, we identified mediators from all four major essential fatty acid metabolomes, i.e. arachidonic acid (AA), EPA, n-3 DPA and DHA (Supplementary Tables 2 and 3). These included the EPA, n-3 DPA and DHA-derived resolvins and the n-3 DPA and DHA-derived protectins and maresins (Supplementary Figs. 1–4 and Supplementary Tables 2 and 3). We next used orthogonal projections to latent structures discriminant analysis (OPLS-DA), which generates a regression model based on concentrations of lipid mediators differently expressed between two groups21, to assess the concentrations of identified mediators between DMARD responders and DMARD non-responders. Here we observed two distinct clusters representing each of these patient groups (Fig. 1a, b). Since circulating peripheral blood cells are significant contributors to plasma lipid mediator profiles we next assessed whether there were differences between peripheral blood cell counts in these two patient groups. This analysis revealed that circulating platelet and phagocyte counts were essentially identical in the two groups (Supplementary Fig. 5).Fig. 1Baseline lipid mediator profiles are predictive of DMARD responsiveness in RA patients.Plasma was collected from RA patients prior to the initiation of treatment with DMARDs and lipid mediator concentrations established using LC-MS/MS-based lipid mediator profiling (see ‘Methods’ for details). a, b OPLS-DA analysis of peripheral blood lipid mediator concentrations for DMARD responders (Resp) and DMARD non-responders (Non-Resp). a Two-dimensional score plot. Grey circle represents the 95% confidence regions. b Two-dimensional loading plots. Lipid mediators with VIP score >1 are upregulated in Non-Resp and denoted in blue. Results are representative of n = 30 Resp and n = 22 Non-Resp. c Percent accuracy score of prediction models based on the combination of all lipid mediators identified and quantified (AL LM) or individual fatty acid metabolomes as indicated. Clin. Score = clinical score (see ‘Methods’ for parameters included). d ROC curves and AUC values (provided in brackets) for predictive models. e Classification predictions for each class (sensitivity and specificity) of the n-3 DPA model. Green indicates the samples that were predicted as Resp while blue indicates those patients predicted Non-Resp. Percentages indicate true positives (Resp class) and true negatives (Non-Resp class). f Relevance of lipid mediators in the prediction performance of the “ALL LM” model based on decreasing accuracy. g Percent accuracy score of models using the indicated SPM. h ROC curves and AUC values for predictive models based on the indicated SPM. All the models were created using the random forest methodology (“randomForest” package from R). Source data are provided as a Source data file.We next used the machine-learning method random forests to build models based on plasma lipid mediator concentrations to further evaluate whether pre-treatment levels of these mediators were linked with DMARD responsiveness. Using this approach, we first assessed whether specific lipid mediator metabolomes were predictive of treatment responsiveness using plasma lipid mediator profiles from 30 DMARD responders and 22 DMARD non-responders. Here we found that cumulative concentrations of the DHA (that includes the D-series resolvins, protectins and maresins) and n-3 DPA (that includes the 13-series resolvins, D-series resolvins, protectins and maresins) metabolomes were the most accurate at predicting whether a patient would respond to treatment or not. The accuracy for the DHA metabolome at predicting outcome was of ~81% and that of the n-3 DPA metabolome was of ~69% (Fig. 1c and Table 1). Of note, these values were also higher than those obtained using a combination of clinical parameters that included the DAS28-ESR and rheumatoid factor concentrations (Fig. 1c and Table 1). In order to validate the robustness of our model we obtained peripheral blood lipid mediator profiles from a second cohort of DMARD naive patients composed of 36 responders and 22 non-responders, and tested whether the models generated using the different mediator metabolomes predicted outcome for patients in this cohort (see Supplementary Tables 4 and 5 for patient clinical parameters and lipid mediator concentrations). For this purpose, we assessed the receiver operating characteristic (ROC) curve, which evaluates the diagnostic potential of a classifier by varying its discrimination threshold. Assessment of the area under the ROC curve demonstrated that the DHA metabolome gave an AUC of 0.44, whereas the n-3 DPA metabolome gave an AUC of 0.58 (Fig. 1d, Supplementary Fig. 6 and Table 1). Similar findings were made using support vector machines, a different machine-learning strategy. Here, the DHA metabolome gave the highest accuracy score of ~62%, an AUC of 0.54, while the n-3 DPA metabolome gave an accuracy score of 61%, an AUC of 0.66 (Table 1). We further evaluated the ability of the n-3 DPA metabolome-based model to accurately categorize patients using the resulting confusion matrix of the model. Here we found that the model based on concentrations of n-3 DPA-derived mediators was able to correctly classify ~83% of responders in the appropriate category (Fig. 1e). Thus, these results indicate that baseline peripheral blood lipid mediator profiles are linked with DMARD treatment outcome.Table 1Summary of prediction models created using support vector machine and random forest.Machine learning methodologySamplesModelNumber of variables% Accuracy scoreModel validationModel evaluationTPRTNRTPFPTNFNTPRTNRTPFPTNFNAUCrandomForest (RF)All samples (1st cohort)Four metabolomes54730.870.55874555130.580.64583664420.72randomForest (RF)All samples (1st cohort)DHA metabolome23810.90.68903268100.390.55394555610.44randomForest (RF)All samples (1st cohort)n-3 DPA metabolome10690.830.5835050170.780.32786832220.58randomForest (RF)All samples (1st cohort)EPA metabolome3650.80.45805545200.530.32536832470.6randomForest (RF)All samples (1st cohort)AA metabolome18650.770.5775050230.830.77832377170.89randomForest (RF)All samples (1st cohort)Clin. Score11500.60.36606436400.030.8831288970.53randomForest (RF)All samples (2nd cohort)RvD4, 10S, 17S-diHDPA, 15R-LXA4, MaR1n-3 DPA4830.920.6892326880.830.59834159170.8randomForest (RF)All samples (2nd cohort)RvD4, 10S, 17S-diHDPA, 15R-LXA4, 5S,12S-diHETE, 4S,14S-diHDHA, MaR1n-3 DPA6860.940.7394277360.870.64873664130.79randomForest (RF)Fibroid samples (1st & 2nd cohort)RvD4, 10S, 17S-diHDPA, 15R-LXA4, MaR1n-3 DPA4880.890.8789138711N/AN/AN/AN/AN/AN/AN/ArandomForest (RF)Lymphoid samples (1st & 2nd cohort)RvD4, 10S, 17S-diHDPA, 15R-LXA4, MaR1n-3 DPA4830.890.789307011N/AN/AN/AN/AN/AN/AN/ArandomForest (RF)Myeloid samples (1st & 2nd cohort)RvD4, 10S, 17S-diHDPA, 15R-LXA4, MaR1n-3 DPA4700.860.4786534714N/AN/AN/AN/AN/AN/AN/AClassyfire (SVM)All samples (1st cohort)Four metabolomes54610.630.54634654370.940.059495560.53Classyfire (SVM)All samples (1st cohort)DHA metabolome23620.650.56654456350.780.18788218220.54Classyfire (SVM)All samples (1st cohort)n-3 DPA metabolome10610.630.54634654370.940.2394772360.66Classyfire (SVM)All samples (1st cohort)EPA metabolome3600.620.52624852380.920.059295580.58Classyfire (SVM)All samples (1st cohort)AA metabolome18580.60.48605248400.920.099291980.66TPR sensitivity, TNR specificity, TP true positives, FP false positives, TN true negatives, FN false negatives, AUC area under the curve.Identification of specific SPM that are predictive of treatment outcomeHaving found that lipid mediator profiles are linked with responsiveness to DMARD treatment in RA patients we next investigated whether we could identify specific lipid mediators that may be useful as biomarkers for treatment responsiveness. For this purpose, we conducted a random forest “importance” analysis that identifies the relevance of every mediator in the performance of the model based on the prediction accuracy. Here we found that the DHA-derived RvD4 (4S,5R,17S-trihydroxy-6E,8E,10Z,13Z,15E,19Z-docosahexaenoic acid) and 10S, 17S-diHDPA (10S,17S-dihydroxy-7Z,11E,13Z,15E,19Z-docosapentaenoic acid) were the most important mediators in predicting treatment responsiveness, with 15R-LXA4 (5S,6R,15R-trihydroxy-7E,9E,11Z,13E-eicosatetraenoic acid), 5S,12S-diHETE (5S,12S-dihydroxy-6E,8Z,10E,14Z-eicosatetraenoic acid), 4S, 14S-diHDHA (4S,14S-dihydroxy-5E,7Z,10Z,12E, 16Z, 19Z-docosahexaenoic acid) and n-3 DPA-derived Maresin 1 (MaR1n-3 DPA) (7R,14S-dihydroxy-8E,10E,12Z,16Z,19Z-docosapentaenoic acid) also displaying a marked contribution, although to a lesser extent than the RvD4 and 10S, 17S-diHDPA (Fig. 1f). Having identified potential candidate biomarkers, we next built machine-learning models using the random forest methodology and concentrations of either RvD4, 10S, 17S-diHDPA, 15R-LXA4 and MaR1n-3 DPA or RvD4, 10S, 17S-diHDPA, 15R-LXA4, 5S,12S-diHETE, 4S,14S-diHDHA and MaR1n-3 DPA. Using a second group of DMARD naive patients we then tested the ability of this model to assign patients to the correct outcome group (see Supplementary Tables 4 and 5 for patient clinical parameters and lipid mediator concentrations). Here we found that the combination of the top six mediators predicted treatment outcome in ~86% of the cases while the model build using the four mediators gave a prediction score of ~83% (Fig. 1g). We next validated the accuracy of these two models using mediator concentrations from a different group of DMARD naive patients. Results from these analyses demonstrated that the model built using the four mediators gave an AUC of 0.80, whereas the model built with the six mediators gave an AUC score of 0.79 (Fig. 1h). Of note, the AUC for these mediators were markedly better than those obtained using mediator concentrations from the n-3 DPA metabolome and disease scores (Fig. 1d).Increased lipid mediators in plasma from DMARD non-respondersTo gain insights into mechanisms determining the responsiveness of patients to DMARD treatment, we conducted lipid mediator pathway analysis to identify which pathways were differentially regulated between the two patient groups. This demonstrated that there was an upregulation of SPM biosynthetic pathways, including the DHA-derived RvD4 and the n-3 DPA-derived MaR1n-3 DPA in DMARD non-responders. These increases were coupled with an upregulation of pro-inflammatory eicosanoids, including the nociceptive mediators PGD2 and PGE2, in these patients when compared with DMARD responders (Fig. 2). To determine whether the differences in SPM expression were linked with a distinct transcriptional regulation of enzymes involved in SPM biosynthesis we assessed the transcript expression of ALOX enzymes in peripheral blood from these two patient groups. ALOX5, ALOX12, ALOX15 and ALOX15B transcript levels were similar between the DMARD responders and DMARD non-responders (Supplementary Fig. 7). These results suggest that regulation of SPM biosynthetic pathways may be via either the regulation of protein translation or post-translational modification of the enzymes to regulate their activity22,23. Thus, we next tested whether the activity of these enzymes was altered. For this purpose, we measured plasma levels of monohydroxylated fatty acids from all four fatty acid metabolomes to gain insights into their activity in the two patient groups. Assessment of plasma concentrations of 5-HETE, 5-HEPE, 7-HDPA and 7-HDHA, markers of ALOX5 activity, revealed a significant upregulation of 7-HDHA, 5-HEPE and 5-HETE in DMARD non-responders when compared with responders. Concentrations of markers for ALOX12 (14-HDPA and 14-HDHA) and ALOX15 (17-HDPA, 17-HDHA, 15-HEPE and 15-HETE) indicated an increase in activity for these enzymes in non-responders, given the increased levels of these molecules in plasma from these patients when compared with those found in responders (Supplementary Fig. 8). To further evaluate the origin of these proposed pathway markers we conducted chiral liquid chromatography-tandem mass spectrometry, which permits the separation the R and S isomers of a given hydroxylated fatty acid. Here we found that in plasma of both DMARD responders and DMARD non-responders the most abundant isomer for all monohydroxylated fatty acids tested was that carrying the alcohol in the S conformation (Supplementary Figs. 9–12 and Supplementary Table 6). Given that all four ALOX enzymes involved in SPM biosynthesis preferentially oxygenate fatty acids in the S conformation24,25, these results indicate an increased ALOX activity in non-responders.Fig. 2Upregulation of baseline peripheral blood lipid mediators in DMARD non-responders.Peripheral blood was collected in patients DMARD responders (Resp) and DMARD non-responders (Non-Resp) prior to DMARD treatment initiation. Peripheral blood lipid mediator profiles were established in accordance with published criteria including matching retention time and MS/MS fragmentation spectra. Pathway analysis for the differential expression of mediators from the (top panel) DHA and n-3 DPA, and (bottom panel) EPA and AA bioactive metabolomes in Non-Resp when compared to Resp. Statistical differences between the normalised concentrations (expressed as the fold change) of the lipid mediators from the Non-Resp and Resp groups were determined using a two-sided t test followed by a multiple comparison correction using Benjamini–Hochberg procedure. Up- or downregulated mediators are denoted with using upward and downward facing triangles, respectively, and on changes of the node’s size. Bolded mediators represent statistical differences between the two groups when adjusted p value <0.05. Results are representative of n = 66 for Resp and n = 44 Non-Resp. Source data are provided as a Source data file.Baseline lipid mediators are linked with distinct disease pathotypesSynovial molecular and histological features patients with RA can be classified into three categories lympho-myeloid, diffuse-myeloid and pauci-immune-fibroid. These pathotypes are associated with distinct disease evolution and responses to DMARD treatment5. Therefore, we next questioned whether immune-related features in each of these groups extended beyond the synovium into the systemic circulation. To address this question, we conducted lipid mediator profiling to assess whether peripheral blood lipid mediator concentrations in RA patients from each of these three groups were distinct prior to the initiation of DMARD therapy. Using PLS-DA we found that plasma lipid mediators were indeed characteristic for different disease pathotypes, i.e. distinct lipid mediator profiles clustered with each category (Fig. 3a). Assessment of the variable importance in projection (VIP) scores, which identify the contribution of each mediator in the observed separation between groups demonstrated an upregulation of pro-resolving mediators, including 15R-LXA4 and MCTR2 (13R-cysteinylglycinyl, 14S-hydroxy-4Z,7Z,9E,11E,13R,14S,16Z,19Z-docosahexaenoic acid), in peripheral blood from patients with the pauci-immune-fibroid pathotype. In plasma from these patients we also found an upregulation of pro-inflammatory and immunosuppressive mediators including PGD2 and TxA2, measured as its stable further metabolite TxB2 (Fig. 3b).Fig. 3Combining disease pathotypes and select SPM concentrations enhances model predictiveness.Plasma was collected from RA patients prior to the initiation of treatment with DMARD and lipid mediator concentrations established using LC-MS/MS-based lipid mediator profiling. a, b PLS-DA analysis of peripheral blood lipid mediator concentrations for lympho-myeloid (lymphoid), diffuse-myeloid (myeloid) and pauci-immune-fibroid (fibroid) pathotypes. a 3-dimensional score plot. b Variable importance in projection (VIP) scores of 15 lipid mediators with the greatest differences in concentrations between the three groups. Results are representative of n = 18 for Fibroid n = 17 for myeloid and n = 19 for lymphoid. c Pathway analysis for the differentially expressed mediators from the DHA and n-3 DPA bioactive metabolomes in DMARD non-responders (Non-Resp) when compared to DMARD responders (Resp) for each pathotype. Statistical differences between the normalised concentrations (expressed as the fold change) of the lipid mediators from the Non-Resp and Resp groups were determined using two-sided t test followed by a multiple comparison correction using Benjamini–Hochberg procedure. Up- or downregulated mediators are denoted with using upward and downward facing triangles, respectively, and on changes of the node’s size. Bolded mediators represent statistical differences between the two groups when adjusted p value <0.05. Results are representative of n = 18 for fibroid Resp, n = 15 for fibroid Non-Resp, n = 19 for lymphoid Resp, n = 10 for lymphoid Non-Resp, n = 22 for myeloid Resp, n = 15 for myeloid Non-Resp. d Classification accuracies for each class (sensitivity and specificity) of the RvD4, 10S,17S-diHDPA, 15R-LXA4 and MaR1n-3 DPA model created using the specific dataset for the different pathotypes (fibroid, lymphoid and myeloid). Green indicates the samples that were predicted as Resp while blue indicates predicted Non-Resp. Percentages indicate true positives (Resp class) and true negatives (Non-Resp class). All the models were created using the random forest methodology (“randomForest” package from R). Source data are provided as a Source data file.We next investigated the differential regulation of lipid mediator profiles between DMARD responders and non-responders for each of these three pathotypes. Here we found an increase in ALOX5 products from both the n-3 DPA and DHA metabolomes in non-responders with lympho-myeloid and those with a pauci-immune-fibroid pathotype when compared with responders with the respective pathotypes. These included significant increases in the DHA-derived RvD4 and PDX. In these patients we also found a significant increase in n-3 DPA-derived MaR1n-3 DPA (Fig. 3c). Assessment of mediators from the AA and EPA metabolomes demonstrated an increase in ALOX5 products in non-responders with lympho-myeloid and pauci-immune-fibroid pathotypes that reached statistical significance in patients with a pauci-immune-fibroid pathotype for the leukotriene (LT) B4 pathway, including LTB4 and its further metabolite 20-COOH-LTB4. In these patients, we also found a statistically significant increase in concentrations of the pro-inflammatory and nociceptive mediator PGE2 (Supplementary Fig. 13). These results demonstrate that differences in peripheral blood lipid mediator profiles between DMARD responders and non-responders are common to different RA pathotypes.Having found that lipid mediator concentrations were different between these patient groups, we next assessed whether combining disease pathotypes with the biomarkers identified above would further enhance the predictiveness of our machine-learning model. Results from this analysis demonstrate a marked increase in the predictiveness of RvD4, 10S, 17S-diHDPA, 15R-LXA4 and MaR1n-3 DPA at identifying responders, when separate machine-learning models were built for each of the RA pathotypes, with the ability of the model to correctly classify responders increasing to ~89% (Fig. 3d).Differences in SPM concentrations are maintained after treatment initiationHaving observed a differential regulation in peripheral blood SPM concentrations between DMARD responders and DMARD non-responders prior to treatment initiation, we next investigated whether differences in peripheral blood lipid mediator concentrations were also present in patients 6 months after treatment initiation. OPLS-DA analysis demonstrated that plasma lipid mediator profiles from DMARD responders 6 months after the initiation of treatment were distinct from those of DMARD non-responders, as demonstrated by a separation between the two patient clusters (Fig. 4a, Supplementary Table 7). Assessment of VIP scores identified 22 mediators and SPM pathway markers that were differentially expressed between the two patient groups (Fig. 4b). Amongst these mediators we found SPM that are involved in coordinating the host response during ongoing inflammation such as PCTR2 (16-cysteinylglycinyl, 17S-hydroxy-4Z,7Z,11,13,15E,19Z-docosahexaenoic acid), RvD2 (7S,16R,17S-trihydroxy-4Z,8E,10Z,12E,14E,19Z-docosahexaenoic acid) and RvD3 (4S,11R,17S-trihydroxy-5Z,7E,9E,13Z,15E,19Z-docosahexaenoic acid)16,22,26 as well as mediators linked with pain modulation e.g. RvE211 (Fig. 4b).Fig. 4Decreased SPM levels in DMARD non-responders 6 months after treatment initiation.Peripheral blood was collected in patients that displayed reduced joint disease (DMARD responders, Resp) and those that did not (DMARD non-responders; Non-Resp) 6 months after treatment initiation. Peripheral blood lipid mediator profiles were established using LC-MS/MS-based lipid mediator profiling. a, b OPLS-DA of lipid mediator profiles from Resp and Non-Resp. a Score plot, b Represents the loading plot with mediators displaying VIP score >1 highlighted in green or blue and correspond with either Resp or Non-Resp, respectively. c, d Pathway analysis for the differential expression of mediators from the c DHA and n-3 DPA and d EPA and AA bioactive metabolomes in Non-Resp when compared to Resp. Statistical differences between the normalised concentrations (expressed as the fold change) of the lipid mediators from the Non-Resp and Resp groups were determined using a two-sided t test followed by multiple comparison correction using Benjamini–Hochberg procedure. Up- or downregulated mediators are denoted with using upward and downward facing triangles, respectively, and on changes of the node’s size. Bolded mediators represent statistical differences between the two groups when adjusted p value <0.05. Results are representative of n = 27 for Resp and n = 17 for Non-Resp. Source data are provided as a Source data file.In order to gain further insights into the mediator pathways that were differentially regulated between these patient groups, we interrogated the biosynthetic pathways for each of the essential fatty acid metabolomes. This analysis demonstrated significant increases in concentrations of select ALOX5 and ALOX15-derived mediators from the DHA metabolome that included RvD1 and 17R-PD1 (10R,17R-dihydroxy-7Z,11E,13E,15Z,19Z-docosahexaenoic acid) in plasma of non-responders when compared to responders (Fig. 4c). Pathway analysis of EPA and AA-derived lipid mediator concentrations demonstrated that while ALOX5 derived products of EPA were also reduced, AA-derived ALOX5 products, including those of the potent leucocyte chemoattractant LTB4 and the ionotropic cysteinyl leukotrienes27, were markedly increased in plasma from non-responders when compared with responders (Fig. 4d).Having observed significant changes in SPM concentrations, we next investigated whether the activity of ALOX enzymes and the conversion of DHA and n-3 DPA was altered in peripheral blood cells from the two patient groups. For this purpose, we measured plasma levels of monohydroxylated fatty acids from the DHA and n-3 DPA metabolomes to gain insights into both enzyme activity and substrate conversion. Plasma concentrations of the ALOX5 products 7-HDPA and 7-HDHA were either similar (7-HDPA) between the two patient groups, or upregulated (7-HDHA) in DMARD non-responders. Concentrations of the ALOX12 (14-HDPA and 14-HDHA) and ALOX15 (17-HDHA and 17-HDPA) products were increased in non-responders (Supplementary Fig. 14). Of note, as observed in baseline plasma, chiral analysis of monohydroxylated fatty acids demonstrated that the predominant isomer for these products was the S-isomer (Supplementary Table 8). These findings indicate that the observed reduction in plasma DHA and n-3 DPA-derived SPM in DMARD non-responders was not due to a decrease in ALOX activity and/or substrate availability/conversion in peripheral blood cells from these patients. Together these observations demonstrate that 6 months after treatment initiation plasma SPM concentrations in DMARD responders were higher than those measured in DMARD non-responders. Given that enzyme activity was elevated in non-responders when compared with responders, this suggests that uncoupling of the SPM biosynthetic pathways may be responsible for the reductions in plasma SPM concentrations.DiscussionThe present findings uncover a previously unappreciated role for SPM as predictive biomarkers to DMARD responsiveness in RA. Assessment of baseline plasma SPM demonstrated that the concentrations of select mediators were predictive of DMARD treatment responsiveness, identifying novel functional biomarkers. Differences in plasma SPM concentrations were found to persist to 6 months after the initiation of DMARD treatment in non-responders when compared with responders.Mounting evidence implicates a role for altered resolution mechanisms in the onset and propagation of RA. Increasing synovial RvE2 concentrations were found to correlate with decreased joint pain, whereas plasma SPM concentrations were negatively related to erythrocyte sedimentation rate11. In experimental systems 17R-RvD1 (7S,8R,17R-trihydroxy-4Z,9E,11E,13Z,15E,19Z-docosahexaenoic acid; also referred to as aspirin-triggered-RvD1) attenuates arthritis severity, cachexia, hind-paw oedema, and paw leucocyte infiltration, shortening the remission interval19. RvD3 concentrations are reduced in inflamed joints from mice with delayed-resolving arthritis when compared with self-resolving inflammatory arthritis. Administration of this mediator to arthritic mice reduced joint leucocyte trafficking, joint eicosanoid concentrations, and joint inflammation16. RvD1 (7S,8R,17S-trihydroxy-4Z,9E,11E,13Z,15E,19Z-docosahexaenoic acid) and its precursor 17-HDHA were also found to display anti-hyperalgesic properties28. In the present study, we found that peripheral blood concentrations of both SPM and inflammatory eicosanoids were increased in DMARD non-responders at baseline (Figs. 1 and 2). These changes were independent of differences in overall circulating platelet and leucocyte numbers, suggesting that they may reflect a differential activation status in peripheral blood leucocytes as previously reported for other leucocyte subsets13. Furthermore, the concentrations of select SPM were higher in patients that were non-responsive to DMARDs when compared with those that were responsive 6 months after treatment initiation (Fig. 4). Given that one of the key biological actions of SPM is to counter-regulate eicosanoid production6,10,15,16,19,22,29, these findings suggest that SPM activity in DMARD non-responders may be compromised. This observation is in line with findings made in diabetic patients where the signalling downstream of the RvE1 receptor, chemerin chemokine-like receptor 1, was found to be altered reducing the ability of RvE1 to regulate peripheral blood leucocyte responses from these patients30.It is widely believed that a precision medicines approach is likely to be more effective in the treatment of patients with chronic inflammatory disorders, including those with RA, than the current process where patients are treated in a prescriptive manner using a one size fits all approach31. Unfortunately, the lack of robust biomarkers to determine treatment responsiveness in many chronic inflammatory conditions, including RA, has hindered the development of this approach in clinical settings32. Results from the present study demonstrate that plasma lipid mediator concentrations prior to the initiation of treatment are different in patients that respond to DMARDs when compared with those that do not. Using machine-learning methodologies, we found that the concentrations of a select group of mediators were predictive of treatment responsiveness in two RA cohorts, with prediction accuracy of up to ~89% (Figs. 1 and 3). Furthermore, plasma SPM concentrations prior to DMARD treatment initiation were also diagnostic of distinct joint disease pathotypes (Fig. 3). Importantly, while patients enrolled in this study were DMARD naive, most of the patients in both cohorts were on a wide range of other medications for a number of co-morbidities, although there were no significant differences in any of these parameters between the two patient groups (see Supplementary Tables 1 and 4). Therefore, the identification of a specific lipid mediator signature that is predictive of DMARD responsiveness suggests that changes in these lipid mediators are specific for this group of therapeutics. In addition, since SPM regulate host innate and adaptive immune responses and their production is reflective of leucocyte activation status13–15, the present findings indicate that peripheral blood SPM concentrations are potential functional biomarkers for both patient stratification and predicting treatment responsiveness to DMARDs.The biosynthesis of SPM involves the stereoselective conversion of essential fatty acids by distinct enzymes, with the successful formation of the bioactive product relying on the expression and activity of the enzyme as well as their appropriate subcellular localization. In this context studies investigating mechanisms regulating SPM biosynthesis demonstrated that activation of MAPK leads to enzyme phosphorylation at serine 271 that is subsequently translocated to the nuclear membrane where it couples with phospholipase A2 and Leukotriene A4 Hydrolase to produce leukotrienes27,33. On the other hand, in the absence of phosphorylation the enzyme is retained in the cytosol where it couples with ALOX15 to produce SPM33. During ongoing inflammation, for example in atherosclerosis, an increase in the expression of phosphorylated ALOX5 and a decrease in the RvD1 to LTB4 ratio was observed33. These findings indicate that in addition to expression of the enzyme, post-translational modifications of the protein are central in determining the product profile of the enzyme, that is, whether the enzymes produce SPM or pro-inflammatory eicosanoids. In the present study, we found that concentrations for most SPM identified in the plasma of DMARD non-responders were either similar to those found in responders or reduced. This observation was coupled with an increase in ALOX activity (Fig. 4, Supplementary Fig. 14 and Supplementary Table 8), suggesting that the SPM biosynthetic pathways become uncoupled post DMARD treatment initiation in non-responders.In summation, the present study identifies novel functional biomarkers, including RvD4, 10S, 17S-diHDPA, 15R-LXA4 and MaR1n-3 DPA, that predict both treatment response to DMARDs as well as joint disease pathotype. Thus, these biomarkers may be clinically useful in identifying patients who are unlikely to respond to conventional DMARD therapy and would benefit from being fast-tracked to the next level of RA therapeutics. This would in turn help minimise or even prevent further structural damage to the joints together with disease progression and disability, thereby improving quality of life.MethodsMaterialsLiquid chromatography (LC)-grade solvents were purchased from Fisher Scientific (Pittsburgh, PA, USA); Poroshell 120 EC-C18 column (100 mm × 4.6 mm × 2.7 µm) was obtained from Agilent (Cheshire, UK); C18 SPE columns were from Biotage (Uppsala, SE); synthetic standards for LC-tandem mass spectrometry (MS-MS) quantitation and deuterated (d) internal standards (d8-5S-HETE (Cat no: CAY334230); d5-RvD2 (Cat no: CAY11184); d5-LXA4 (Cat no: CAY10007737); d4-PGE2 (Cat no: CAY314010); d4-LTB4 (Cat no: CAY320110); d5-LTC4 (Cat no: CAY10006198); d5-LTD4 (Cat no: CAY10006199); d5-LTE4 (Cat no: CAY10007858)) and synthetic lipid mediator standards (RvD1, CAY10012554; 17R-RvD1 (Cat no: CAY13060); RvD2 (Cat no: CAY10007279); RvD3 (Cat no: CAY13834); 17R-RvD3 (Cat no: CAY9002880); RvD4 (Cat no: CAY13835); RvD5 (Cat no: CAY10007280); MaR1 (Cat no: CAY10878); MaR2 (Cat no: CAY16369); MCTR1 (Cat no: CAY17007); MCTR2 (Cat no: CAY17008); MCTR3 (Cat no: CAY19067); PDX (Cat no: CAY10008128); PCTR1 (Cat no: CAY19064); PCTR2 (Cat no: CAY19065); PCTR3 (Cat no: CAY19066); 4-HDHA (Cat no: CAY33200); 7-HDHA (Cat no: CAY33300); 14-HDHA (Cat no: CAY33550); 17-HDHA (Cat no: CAY33650); RvE1 (Cat no: CAY10007848); 5-HEPE (Cat no: CAY32210); 12-HEPE (Cat no: CAY32540); 15-HEPE (Cat no: CAY32700); 18-HEPE (Cat no: CAY32840); RvD5n-3 DPA, CAY10546; LXA4 (Cat no: CAY90410); 15-epi-LXA4 (Cat no: CAY90415); LXB4 (Cat no: CAY90420); 5S,15S-diHETE (Cat no: CAY35280); PGD2 (Cat no: CAY12010); PGE2 (Cat no: CAY14010); PGF2α (Cat no: CAY16010); TXB2 (Cat no: CAY19030); LTB4 (Cat no: CAY20110); 6-trans-LTB4 (Cat no: CAY35250); 6-trans,12-epi-LTB4 (Cat no: CAY35265); 20-OH-LTB4 (Cat no: CAY20190); 20-COOH-LTB4 (Cat no: CAY20180); LTC4 (Cat no: CAY20210); LTD4 (Cat no: CAY20310); LTE4 (Cat no: CAY20410); 5-HETE (Cat no: CAY34210); 12-HETE (Cat no: CAY34550); 15-HETE (Cat no: CAY34700)) were purchased from Cambridge Bioscience (Cambridge, UK) or provided by Charles N. Serhan (Harvard Medical School, Boston, Massachusetts, USA; supported by NIH-funded P01GM095467); Dulbecco’s phosphate-buffered saline (DPBS, without calcium and magnesium, Sigma (Cat no: D8537)).Pathobiology of Early Arthritis CohortPlasma samples were taken at baseline and 6 months from 112 and 44 patients, respectively. These were obtained from the Pathobiology of Early Arthritis Cohort (PEAC). The PEAC cohort study was approved by the King’s College Hospital Research Ethics Committee (REC 05/Q0703/198). Patients provided informed consent. Peripheral blood samples and synovial tissue were obtained from patients recruited at Barts Health NHS Trust into the Pathobiology of Early Arthritis Cohort (PEAC, http://www.peac-mrc.mds.qmul.ac.uk) undergoing ultrasound (US)-guided synovial biopsy of the most inflamed joint (knee, wrist or small joints of hands or feet)5. All patients were DMARDs and steroid-naive, had symptoms duration <12 months and fulfilled the ACR/EULAR 2010 classification criteria for RA. RA individuals were categorised into three pathotypes based on histological classification of synovial tissue: lympho-myeloid, diffuse-myeloid and pauci-immune-fibroid (for more details see ref. 5). Patients were treated with DMARDs. Response status after 6 months of mixed DMARD therapy was determined by EULAR response criteria based on DAS28-ESR.Targeted lipid mediator profilingPlasma was obtained from peripheral blood following centrifugation at 1500 × g for 10 min at room temperature. All samples were extracted using solid-phase extraction columns as in refs. 20,34. A step-by-step description of the extraction, analysis and quantitation procedures are detailed in the following protocol found in Protocol Exchange35. Prior to sample extraction, deuterated internal standards, representing each region in the chromatographic analysis (500 pg each) were added to facilitate quantification. Samples were kept at −20 °C for a minimum of 45 min to allow protein precipitation. Supernatants were subjected to solid-phase extraction, methyl formate fraction collected, brought to dryness and suspended in phase (methanol/water, 1:1, vol/vol) for injection on a Shimadzu LC-20AD HPLC and a Shimadzu SIL-20AC autoinjector, paired with a QTrap 5500 or QTrap 6500+ (Sciex). An Agilent Poroshell 120 EC-C18 column (100 mm × 4.6 mm × 2.7 µm) was kept at 50 °C and mediators eluted using a mobile phase consisting of methanol/water/acetic acid of 20:80:0.01 (vol/vol/vol) that was ramped to 50:50:0.01 (vol/vol/vol) over 0.5 min and then to 80:20:0.01 (vol/vol/vol) from 2 min to 11 min, maintained till 14.5 min and then rapidly ramped to 98:2:0.01 (vol/vol/vol) for the next 0.1 min. This was subsequently maintained at 98:2:0.01 (vol/vol/vol) for 5.4 min, and the flow rate was maintained at 0.5 ml/min. QTrap 5500 or QTrap 6500+ were operated using a multiple reaction monitoring method as in refs. 20,34. Supplementary Tables 9 and 10 report instrument source parameters and Supplementary Tables 11–12 report coefficient of variation for sMRM transitions employed in the quantitation of lipid mediators. Each lipid mediator was identified using established criteria, these included: (1) presence of a peak with a minimum area of 2000 counts, (2) matching retention time to synthetic or authentic standards with maximum drift between the expected retention time and the observed retention time of 0.05 s, (3) ≥4 data points, and (4) matching of at least 6 diagnostic ions to that of reference standard, with a minimum of one backbone fragment being identified in reperesntative samples13,20,34. Calibration curves were obtained for each mediator using lipid mediator mixtures at 0.78, 1.56, 3.12, 6.25, 12.5, 25, 50, 100 and 200 pg that gave linear calibration curves with an r2 values of 0.98–0.99. Signal-to-noise ratio was calculated using the Signal-to-Noise script from Analyst (version 1.6.3, Sciex). Here the application the intensity value for the region denoted as the signal/peak of interest by the intensity value for the highest peak in the region denoted as the noise.Chiral LC-MS/MS analysisStep-by-step description of the extraction, analysis and quantitation procedures are detailed in the following protocol found in Protocol Exchange35. Briefly, a Chiralpak AD-RH column (150 mm × 2.1 mm × 5 μm) was used with isocratic methanol/water/acetic acid 95:5:0.01 (v/v/v) at 0.15 ml/min. To monitor isobaric monohydroxy fatty acid levels, a multiple reaction monitoring (MRM) method was developed using signature ion fragments for each molecule as in ref. 22.Description of data used for model buildingThe data used for the machine-learning models and network analyses consisted of the lipid mediator profiles of patients with RA who responded (n = 30) or did not (n = 24) to the treatment with DMARDs for the first PEAC-derived patient cohort. The lipid mediator profile included DHA-derived resolvins (RvD1, RvD2, RvD3, RvD4, RvD5, RvD6, 17-RvD1 and 17-RvD3), protectins (PD1, 17-PD1, 10S,17S-diHDHA, also known as PDX, and 22-OH-PD1), PCTRs (PCTR1, PCTR2 and PCTR3), maresins (MaR1, MaR2, 7S,14S-diHDHA, 4S,14S-diHDHA, 14-oxo-MaR1 and 22-OH-MaR1), MCTRs (MCTR1, MCTR2 and MCTR3), 13-series resolvins (RvT1, RvT2, RvT3 and RvT4), n-3 DPA-derived resolvins (RvD1n-3 DPA, RvD2n-3 DPA and RvD5n-3 DPA), n-3-DPA-derived protectins (PD1n-3 DPA and 10S, 17S-diHDPA), n-3 DPA-derived maresins (MaR1n-3 DPA), E-series resolvins (RvE1, RvE2 and RvE3), leukotrienes (LXA4, LXB4, 5S,15S-diHETE, 20-OH-LTB4, 20-COOH-LTB4, 6-trans-LTB4 and 12-epi-6-trans-LTB4), cysteinyl leukotrienes (LTC4, LTD4 and LTE4), prostaglandins (PGD2, PGE2 and PGF2α) and thromboxane (TXB2). A Clinical Score model was obtained using the following clinical parameters: disease duration, erythrocyte sedimentation rate (ESR), rheumatoid factor (RF titre), tiredness visual analogue scale (VAS), pain VAS, patient global health VAS, physician global assessment VAS, swollen joints number, disease activity score-28 (DAS28) and 12 max US Synovial Thickness and US Power Doppler scores. A second patient cohort of 58 patients (36 responders and 22 non-responders) was obtained from the PEAC study and was used as the test dataset for the lipid mediator profiling and Clinical Score models, and also as the training dataset for improved models based on specific biomarkers. Age, sex and clinical parameters not mentioned before were not considered for this first approximation of creating a model able to classify the response of RA patients to DMARD treatment.Model buildingData were pre-processed and analysed using R Software (v3.5.1; https://www.r-project.org/) and RStudio environment (v1.1.456; https://www.rstudio.com/).From the exploratory analysis, two samples were removed for showing outlier concentrations of TXB2, which likely reflected coagulation during sample collection and an additional sample was removed due to lack of clinical records. Although no normalization was required since all the lipid mediator concentrations were calculated based on the same amount of standard, the concentrations were scaled by subtracting the mean and dividing by the standard deviation of each feature.Two supervised machine-learning methodologies were used to create the classifier models: support vector machine (SVM)36 and randomForest37. SVM separates groups by organizing the samples in two spaces divided by a hyperplane in a way that the distances between the samples in the same group are not too wide and the distance between the groups is as large as possible38. A nonlinear kernels radial basis function SVM was used. In order to identify the best model, we created models testing different times of the resampling and different number of ensembles (fusion of the individual classifiers created during the bootstrapping step) with 70 bootstrap iterations and 70 individual classifiers in each ensemble that gave stable models for all groups tested. Furthermore, we also used the inbuilt automatic optimization step that includes minimization of the bootstrapping error36 in the R Package “classyfire” (https://cran.r-project.org/src/contrib/Archive/classyfire/), to improve and validate the models (see Supplementary Fig. 6 for representative outcomes).RandomForest operates by getting the consensus of weak decision tree classifiers39. The decision trees are created using the features as vertices and classes as leaves; each tree is designed using a different set of randomly chosen features40. In the present studies using the R package “randomForest” (https://cran.r-project.org/package=randomForest), which uses bootstrapping as the test method, we first used a small loop to test the different mtry values, the number of variables randomly sampled as candidates at each split. Then using the mtry value that gave the best classification performance for each model we tested a number of ntree values, the number of decision trees that are created before creating the consensus classifier tree, to obtain the most stable models.Here we found that an ntree of 10,000 gave us the most stable models for all the lipid mediator groups and for the majority of the variables tested. Increasing the number of ntrees beyond this value did not markedly improve the outcomes (see Supplementary Fig. 6 for representative outcomes).Model evaluationReceiver operating characteristic curves (ROC curves) were built to evaluate the prediction accuracy of the models when predicting between DMARDs responder and DMARDs non-responder in a test dataset. ROC curves are created by plotting the true positive rate against the false-positive rate, showing the sensitivity and specificity of the model, when the discrimination threshold changes. The area under the curve (AUC) is calculated as the prediction performance of the models. ROC curves were created using the R package “pROC” (https://cran.r-project.org/web/packages/pROC/index.html). AUC values close to 1 (AUC > 0.8) refer to good classifier models.Alongside ROC curves, other statistics such as the percentage of correctly classified samples (% accuracy score), specificity and sensitivity were also calculated.Feature selection and model improvementAs random forests showed the best validation scores during the testing step, the model improvement was based on the RandomForest methodology. The lipid mediators were separated in groups based on their precursors (DHA, n-3 DPA, EPA and AA) or the distinct clusters of mediators. Different models were created using only the most relevant lipid mediators and the “importance” function of ‘randomForest’ package. This function organizes the model’s features by relevance based on the model’s decreased mean accuracy when the specific feature is not present. The % accuracy score and AUC (ROC) were calculated for all the models and, according to the results, the best models and the most relevant biomarkers for the classification of the DMARD-responder and DMARD non-responder patients were selected.Pathotypes analysesAll the data (training and test cohorts) was separated based on the specific pathotype shown for the patients: pauci-immune/fibroid (n = 28), lympho-myeloid (n = 27) and diffuse-myeloid (n = 31). This was made with the purpose of seeking better classification models and seeing if specific lipid mediators were responsible for the different manifestation of the disease. The models were build using RandomForest and different statistic scores were calculated for the validation of each model.Network analysesStatistical differences between the normalised concentrations (expressed as the fold change) of the lipid mediators from the DMARD non-responder and DMARD-responder groups were determined using a two-sided t test followed by a multiple comparison correction using Benjamini–Hochberg procedure. Based on these differences, lipid mediator biosynthesis networks were constructed using Cytoscape v3.7.1. The different pathways were illustrated using different colours and line shapes, while up- or downregulated mediators were denoted with using upward and downward facing triangles, respectively, and on changes of the node’s size. Bolded mediators represent statistical differences between the two groups. The comparison between DMARD non-responders and DMARD responders were made with pre and post-treatment data.Enzyme transcript expressionRNA was extracted from whole blood samples in RNALater solution using the Ambion Ribo-Pure Blood kit (ThermoFisher Scientific). Total RNA-sequencing (RNA-seq) was performed on an Illumina HiSeq2500 platform. Raw data were quality-controlled using FastQC, trimmed or removed with Cutadapt. Transcript abundance was derived from paired sample FASTQ files over GENCODE v24/GRCh38 transcripts using Kallisto v0.43.0. Normalization of the raw data and differential gene expression analysis between DMARD-responder and DMARD non-responder were performed using the quasi-likelihood method of the Bioconductor R package “edgeR” (https://bioconductor.org/packages/release/bioc/html/edgeR.html). Results are expressed as the log counts per million of each gene.Statistical analysisWe performed all statistical analyses and data derivation using R v3.5.141, MetaboAnalyst v4.021, Prism v8 and Microsoft Excel Professional Plus 2016. Results presented in the figures are expressed as means and those displayed in the tables are displayed as mean ± sem.Differences between groups were determined using two-sided t test (normalized data) or Mann–Whitney test (2 groups). Sample sizes for each experiment were determined on the variability observed in prior experiments. Partial least squares-discrimination analysis (PLS-DA) and orthagonal partial least squares-discrimination analysis (OPLS-DA) were performed using MetaboAnalyst v4.021 or SIMCA v14.1 software (Umetrics, Umea, Sweden) after mean centring and unit variance scaling of lipid mediator concentrations. PLS-DA is based on a linear multivariate model that identifies variables that contribute to class separation of observations (e.g. treatment response) on the basis of their variables (lipid mediator concentrations). During classification, observations were projected onto their respective class model. The score plot illustrates the systematic clusters among the observations (closer plots presenting higher similarity in the data matrix).Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationPeer Review FileSupplementary InformationReporting Summary | nature communications | [
"Article"
] | [
"Lipids",
"Machine learning",
"Rheumatology",
"Rheumatic diseases"
] | IntroductionRheumatoid arthritis (RA) characterized by joint inflammation bone cartilage destruction decreased quality of life Disease-modifying anti-rheumatic drugs (DMARDs) used RA low dose methotrexate (MTX) anchor drug administered alone or with other DMARDs hydroxychloroquine sulfasalazine patients DMARDs rarely full remission 50% resistant resistance DMARDs exert unwanted side effects increased risk infection2 liver function abnormalities3,4 early RA patients different features distinct responsiveness to DMARD therapy5 DMARDs limit joint inflammation regulating host protective pathways MTX administered to majority (>80%) of RA patients for beneficial effects of low dose MTX depletion of purine thymidine pools reducing cellular proliferation apoptosis CD39 expression in peripheral blood regulatory T cells linked with beneficial actions MTX patients lower density unresponsive to MTX8 wide range unwanted side effects large number patients unresponsive to DMARD treatment (~50%)1 interest in identifying predictive biomarkers reduce exposure of patients unlikely negative side effectsprovide early access to effective therapeutics disease progression RA from decreased host immune response resolution programmes inflammation into chronicity9–11 central to termination inflammation genus mediators specialized pro-resolving mediators produced by immune cells via conversion fatty acids omega-3 n-3 docosapentaenoic docosahexaenoic acid mediators carry distinct stereochemistries established matching approach12 SPM regulate innate adaptive immune responses production of activation status immune during delayed non-resolving joint inflammation downregulation of SPM including DHA-derived resolvin (Rv) D316 In arthritic patients synovial levels eicosapentaenoic RvE2 with decreased joint pain11 increase production supplementation administration mediators linked with decreased joint inflammation joint protection17–19 study questioned protective actions MTX on regulation endogenous protective pathways endogenous SPM levels predictive of responsiveness to MTX mono co-therapy plasma from early-arthritis segregation in lipid mediator profiles between patients to DMARDs plasma SPM concentrations diagnostic of disease pathotypeperipheral blood lipid mediators in DMARD responders non-responders 6 months post initiation findings suggest plasma SPM concentrations characteristic treatment responsiveness disease pathotypes in RA mediator concentrations predictive responsiveness blood SPM concentrations DMARD responsiveness RA investigated plasma lipid mediator profiles in matched early RA patients prior treatment initiation Table 1 lipid mediators identified retention time liquid chromatography six diagnostic ions in tandem mass spectrum identified mediators from essential fatty acid metabolomes arachidonic acid EPA n-3 DPA DHA included EPA n-3 DPA DHA-derived resolvins n-3 DPA DHA-derived protectins maresins used orthogonal projections discriminant analysis assess concentrations between DMARD responders non-responders observed two clusters (Fig. 1a circulating peripheral blood cells contributors to plasma lipid mediator profiles assessed differences between blood cell counts groups circulating platelet phagocyte counts identical in two groups lipid mediator profiles predictive of DMARD responsiveness in RA patientsPlasma collected from RA patients treatment with DMARDs lipid mediator concentrations established LC-MS/MS lipid mediator profiling OPLS-DA analysis peripheral blood lipid mediator concentrations for DMARD responders non-responders Two-dimensional score plot Grey circle 95% confidence regions loading plots Lipid mediators with VIP score >1 upregulated in Non-Resp denoted blue of n = 30 Resp 22 Non-Resp Percent accuracy score of prediction models lipid mediators fatty acid metabolomes Clin Score ROC curves AUC values for predictive models Classification predictions class n-3 DPA model Green Resp blue Non-Resp Percentages positives negatives Relevance of lipid mediators in prediction performance “ALL LM” model Percent accuracy score models using indicated SPM ROC curves AUC values for predictive models models created random forest methodology Source data used machine-learning method models plasma lipid mediator concentrations evaluate pre-treatment DMARD responsiveness assessed lipid mediator metabolomes treatment responsiveness from 30 DMARD responders 22 non-respondersfound cumulative concentrations DHA and n-3 DPA metabolomes accurate predicting patient to treatment accuracy for DHA metabolome outcome ~81% n-3 DPA metabolome ~69% (Fig. 1c Table 1) values higher than clinical parameters DAS28-ESR rheumatoid factor concentrations (Fig Table validate model obtained peripheral blood lipid mediator profiles from second cohort of DMARD naive patients 36 responders 22 non-responders tested models different mediator metabolomes predicted outcome (see Supplementary Tables 4 5 for clinical parameters lipid assessed receiver operating characteristic (ROC) curve DHA metabolome AUC 0.44 n-3 DPA metabolome AUC 0.58 (Fig. 1d Fig. 6 Table 1) Similar findings using support vector machines different machine-learning strategy DHA metabolome highest accuracy score ~62%, AUC 0.54 n-3 DPA metabolome accuracy score 61% AUC 0.66 (Table 1) evaluated ability n-3 DPA metabolome-based model to categorize patients confusion matrix model based on concentrations n-3 DPA mediators ~83% of responders in appropriate category (Fig.results baseline peripheral blood lipid mediator profiles linked DMARD treatment outcome prediction models vector machine random forest learning variables Accuracy metabolomes54730.870.580.64583664420 metabolome23810.90.68903268100.390.55394555610 DPA metabolome10690.830.5835050170.780.32786832220 metabolome3650.80.45805545200 metabolome18650.770.5775050230.830.77832377170. Score11500.60.36606436400.030.8831288970 (2nd cohort)RvD4 10S 17S-diHDPA 15R-LXA4 MaR1n-3.6892326880.830 10S 17S-diHDPA 15R-LXA4-diHETE-diHDHA MaR1n-3 DPA6860.940.7394277360.870.64873664130 samples (1st 2nd cohort)RvD4 10S 17S-diHDPA 15R-LXA4 MaR1n-3)Lymphoid samples (1st 2nd cohort)RvD4 10S 17S-diHDPA 15R-LXA4 MaR1n-3.890)Myeloid samples (1st 2nd cohort)RvD4 10S 17S-diHDPA 15R-LXA4 MaR1n-3 DPA4700.860.4786534714N samples metabolomes54610.630.54634654370.940 metabolome23620.650.56654456350.780.18788218220 DPA metabolome10610.630.54634654370.940.2394772360 metabolome3600.620.52624852380.920.059295580 metabolome18580.60.48605248400.920.099291980.66TPR sensitivity TNR specificity TP positives FP false positives TN negatives FN false negatives AUC area curve SPM predictive treatment lipid mediator profiles linked responsiveness DMARD treatment RA lipid mediators biomarkers treatment responsiveness random forest “importance” analysis relevance mediator modelDHA-derived RvD4 (4S-trihydroxy-6E-docosahexaenoic acid 10S 17S-diHDPA important treatment responsiveness 15R-LXA4-trihydroxy-7E-eicosatetraenoic 5S,12S-diHETE 4S 14S-diHDHA 19Z DPA-derived Maresin 1 (MaR1n-3 DPA) (7R,14S-dihydroxy-8E-docosapentaenoic acid contribution lesser RvD4 10S 17S-diHDPA identified biomarkers built machine-learning models random forest methodology concentrations RvD4 10S 17S-diHDPA 15R-LXA4 MaR1n-3 DPA RvD4 10S 17S-diHDPA 15R-LXA4 5S,12S-diHETE 4S,14S-diHDHA MaR1n-3 DPAsecond group DMARD naive patients tested model correct outcome group Supplementary Tables 4 5 clinical parameters lipid mediator top six mediators predicted treatment outcome ~86% model four mediators ~83% (Fig. 1g). validated accuracy models using mediator concentrations different group DMARD patients model four mediators AUC 0.80 six mediators AUC 0.79 (Fig. 1h). AUC mediators better than concentrations n-3 DPA metabolome disease scores (Fig. 1d).Increased lipid mediators in plasma DMARD non-respondersTo responsiveness DMARD treatment conducted lipid mediator pathway analysis groups upregulation SPM biosynthetic pathways DHA-derived RvD4 n-3 DPA-derived MaR1n-3 DPA in DMARD non-responders increases upregulation pro-inflammatory eicosanoids nociceptive mediators PGD2 PGE2 responders (Fig. 2) differences SPM expression transcriptional regulation enzymes assessed transcript expression ALOX enzymes peripheral blood ALOX5 ALOX12 ALOX15 ALOX15B transcript levels similar between DMARD responders non-responders (Supplementary Fig. 7)results suggest regulation SPM biosynthetic pathways via protein translation or modification enzymes tested activity enzymes measured plasma levels monohydroxylated fatty acids four metabolomes activity patient groups Assessment plasma concentrations 5-HETE 5-HEPE 7-HDPA 7-HDHA markers ALOX5 activity revealed significant upregulation 7-HDHA in DMARD non-responders responders Concentrations ALOX12 (14-HDPA 14-HDHA ALOX15 (17-HDPA 17-HDHA 15-HEPE 15-HETE indicated increase activity in non-responders conducted chiral liquid chromatography-tandem mass spectrometry found plasma DMARD responders non-responders most abundant isomer for monohydroxylated fatty acids carrying alcohol in S conformation 9–12 four ALOX enzymes SPM biosynthesis oxygenate fatty acids in S conformation24 results indicate increased ALOX activity in non-responders.Fig. 2Upregulation baseline peripheral blood lipid mediators in DMARD non-responders blood collected in DMARD responders non-responders prior to DMARD treatment initiationPeripheral blood lipid mediator profiles established criteria retention time MS fragmentation spectra Pathway analysis differential expression mediators DHA n-3 DPA EPA AA metabolomes Non-Resp Resp Statistical differences concentrations lipid mediators Non-Resp Resp determined two-sided t test multiple comparison correction Benjamini–Hochberg procedure Up- downregulated mediators denoted upward downward facing triangles changes node’s size Bolded mediators represent differences adjusted p value <0.05. Results representative n = 66 Resp n = 44 Non-Resp Source data provided lipid mediators linked disease pathotypesSynovial RA lympho-myeloid diffuse-myeloid pauci-immune-fibroid associated disease evolution responses DMARD questioned immune-related features extended beyond synovium systemic circulation conducted lipid mediator profiling peripheral blood lipid mediator concentrations RA patients distinct DMARD therapy plasma lipid mediators characteristic different disease pathotypes distinct profiles each categoryAssessment variable importance scores upregulation pro-resolving mediators 15R-LXA4 MCTR2 (13R-cysteinylglycinyl 14S-hydroxy-4Z-docosahexaenoic peripheral blood patients pauci-immune-fibroid pathotype plasma upregulation pro-inflammatory immunosuppressive mediators PGD2 TxA2 TxB2 (Fig. disease pathotypes SPM concentrations enhances model predictiveness collected RA patients treatment DMARD lipid mediator concentrations LC-MS lipid mediator profiling PLS-DA analysis peripheral blood lipid mediator concentrations lympho-myeloid diffuse-myeloid pauci-immune-fibroid pathotypes 3-dimensional score plot Variable importance scores 15 lipid mediators differences concentrations groups representative n = 18 Fibroid 17 myeloid 19 lymphoid Pathway analysis differentially expressed mediators DHA n-3 DPA bioactive metabolomes DMARD non-responders responders Statistical differences normalised concentrations lipid mediators determined two-sided t test multiple comparison correction Benjamini–Hochberg procedureUp downregulated mediators denoted upward downward facing triangles changes node’s size Bolded mediators represent statistical differences groups adjusted p value <0.05. Results representative n = 18 fibroid 15 Non 19 lymphoid 10 22 myeloid 15 Classification accuracies class RvD4 10S,17S-diHDPA 15R-LXA4 MaR1n-3 DPA model dataset pathotypes (fibroid lymphoid myeloid). Green indicates samples predicted Resp blue Non-Resp Percentages indicate positives (Resp negatives (Non-Resp models created random forest methodology Source data provided investigated differential regulation lipid mediator profiles DMARD responders non-responders pathotypes found increase ALOX5 products n-3 DPA DHA metabolomes non lympho-myeloid pauci-immune-fibroid pathotype responders significant increases DHA-derived RvD4 PDX increase n-3 DPA-derived MaR1n-3 DPAAssessment mediators AA metabolomes increase ALOX5 products non-responders lympho-myeloid pauci-immune-fibroid pathotypes statistical significance leukotriene (LT) B4 pathway LTB4 20-COOH-LTB4 significant increase concentrations pro-inflammatory nociceptive mediator PGE2 Fig results differences peripheral blood lipid mediator profiles between DMARD responders non common to RA pathotypes lipid concentrations assessed combining disease pathotypes biomarkers predictiveness machine-learning model Results increase predictiveness RvD4, 10S 17S-diHDPA 15R-LXA4 MaR1n-3 DPA identifying responders-learning models model classify responders to ~89% (Fig. 3d).Differences SPM concentrations maintained after treatment differential regulation blood SPM concentrations between DMARD responders non-responders treatment investigated differences concentrations 6 months after treatment OPLS-DA analysis plasma lipid mediator profiles DMARD responders 6 months distinct from non-responders clusters (Fig. 4a Table Assessment VIP scores identified 22 mediators SPM pathway markers differentially expressed between two patient groupsmediators found SPM response PCTR2 (16-cysteinylglycinyl 17S-hydroxy-4Z,7Z,11,13,15E,19Z-docosahexaenoic RvD2 (7S,16R,17S-trihydroxy-4Z RvD3 (4S,11R,17S-trihydroxy-5Z,7E,9E)16,22,26 mediators pain modulation RvE211 (Fig. 4Decreased SPM levels DMARD non 6 months after treatment blood collected patients reduced joint disease non 6 months treatment lipid mediator profiles established LC-MS/MS lipid mediator profiling lipid mediator profiles Resp Non-Resp Score plot mediators VIP score >1 green blue Resp Non-Resp Pathway analysis differential expression mediators DHA n-3 DPA EPA AA metabolomes Non-Resp Statistical differences concentrations lipid mediators Non-Resp Resp determined two-sided t test multiple comparison correction Benjamini–Hochberg procedure Up- downregulated mediators denoted upward downward facing triangles changes node’s sizeBolded mediators represent differences groups adjusted p value <0.05. Results representative of n = 27 Resp 17 Non-Resp. Source data provided mediator pathways interrogated biosynthetic pathways essential fatty acid metabolomes analysis demonstrated increases concentrations ALOX5 ALOX15 mediators DHA metabolome RvD1 17R-PD1 acid in non-responders responders (Fig. 4c). analysis EPA AA-derived lipid mediator concentrations ALOX5 products EPA reduced AA-derived ALOX5 products increased in plasma non-responders responders (Fig. changes concentrations investigated activity ALOX enzymes conversion DHA n-3 DPA in peripheral blood cells groups measured plasma levels monohydroxylated fatty acids DHA n-3 DPA metabolomes enzyme activity substrate conversion Plasma concentrations ALOX5 products 7-HDPA 7-HDHA similar or upregulated in non-responders Concentrations ALOX12 (14-HDPA 14-HDHA ALOX15 (17-HDHA 17-HDPA products increased in non-respondersbaseline plasma monohydroxylated fatty acids predominant isomer S-isomer Table reduction in plasma DHA n-3 DPA SPM in DMARD non-responders not due to ALOX activity substrate availability peripheral blood cells 6 months after treatment plasma SPM concentrations in DMARD responders higher than non-responders enzyme activity elevated in non uncoupling SPM biosynthetic pathways for reductions findings uncover role SPM predictive biomarkers DMARD responsiveness RA baseline plasma SPM concentrations mediators predictive DMARD treatment responsiveness Differences in plasma SPM concentrations 6 months after DMARD treatment non-responders evidence implicates altered resolution mechanisms in onset propagation RA Increasing synovial RvE2 concentrations with decreased joint pain plasma SPM concentrations negatively related to erythrocyte sedimentation 17R-RvD1 acid-triggered-RvD1) attenuates arthritis cachexia hind-paw oedema leucocyte infiltration remissionRvD3 concentrations reduced in inflamed joints delayed-resolving arthritis compared self-resolving inflammatory arthritis Administration mediator mice reduced joint leucocyte trafficking eicosanoid concentrations RvD1 (7S-trihydroxy-4Z-docosahexaenoic acid) precursor 17-HDHA anti-hyperalgesic study peripheral blood concentrations of SPM inflammatory eicosanoids increased in DMARD non-responders at baseline (Figs. 1 2) changes independent of differences circulating platelet leucocyte numbers differential activation status leucocytes concentrations SPM higher in patients non-responsive to DMARDs responsive 6 months after treatment initiation (Fig. 4) SPM eicosanoid findings suggest SPM activity in DMARD non-responders may compromised with findings in diabetic patients signalling RvE1 receptor altered reducing RvE1 peripheral blood leucocyte responses precision medicines approach more effective in treatment chronic inflammatory disorders RA than current one size fits alllack of biomarkers treatment responsiveness in chronic inflammatory conditions including RA hindered development approach Results present study plasma lipid mediator concentrations prior treatment different in patients DMARDs machine-learning concentrations select mediators predictive of treatment responsiveness in RA cohorts prediction accuracy to ~89% (Figs. 1 3) plasma SPM concentrations DMARD treatment diagnostic of joint disease pathotypes (Fig patients DMARD naive on other medications for co-morbidities no significant differences in parameters between groups Tables 1 4) specific lipid mediator signature predictive DMARD responsiveness suggests changes specific for SPM regulate immune responses production reflective of leucocyte activation findings indicate peripheral blood SPM concentrations potential biomarkers for patient stratification treatment responsiveness to DMARDs biosynthesis of SPM conversion essential fatty acids by enzymes formation bioactive product on expression activity enzyme subcellular localization activation of MAPK leads to enzyme phosphorylation at serine 271 to nuclear membrane couples with phospholipase A2 Leukotriene A4 Hydrolase produceabsence phosphorylation enzyme retained in cytosol couples with ALOX15 produce inflammation atherosclerosis increase expression phosphorylated ALOX5 decrease RvD1 to LTB4 ratio findings indicate post-translational modifications protein product profile produce SPM or pro-inflammatory eicosanoids study concentrations SPM in plasma DMARD non-responders similar to responders or reduced with increase ALOX activity (Fig. 4 14 8) SPM biosynthetic pathways uncoupled post DMARD treatment in non-responders study identifies novel functional biomarkers RvD4, 10S, 17S-diHDPA 15R-LXA4 MaR1n-3 DPA predict treatment response DMARDs joint disease pathotype biomarkers useful identifying patients unlikely to respond conventional DMARD therapy benefit fast next level RA therapeutics structural damage joints disease progression disability quality of lifechromatography solvents Fisher Scientific Poroshell 120 EC-C18 column (100 mm mm 2.7 μm Agilent (Cheshire C18 SPE columns Biotage (Uppsala synthetic standards-tandem mass spectrometry quantitation deuterated standards (d8-5S-HETE CAY334230); d5-RvD2 d5-LXA4 CAY10007737) d4-PGE2 d4-LTB4 d5-LTC4 CAY10006198) d5-LTD4 CAY10006199) d5-LTE4 CAY10007858) synthetic lipid mediator standards (RvD1 CAY10012554 17R-RvD1 CAY13060) RvD2 CAY10007279) RvD3 CAY13834) CAY9002880) RvD4 CAY13835) RvD5 CAY10007280) MaR1 CAY10878) MaR2 CAY16369) MCTR1 MCTR2 MCTR3 CAY19067) PDX CAY10008128) PCTR1 PCTR3 4-HDHA 7-HDHA 14-HDHA 17-HDHA RvE1CAY10007848) 5-HEPE 12-HEPE 15-HEPE 18-HEPE CAY32840); RvD5n-3 DPA CAY10546 LXA4 CAY90410); 15-epi-LXA4 CAY90415) LXB4 CAY90420); 5S,15S-diHETE CAY35280); PGD2 CAY12010); PGE2 PGF2α CAY16010); CAY19030); LTB4 CAY20110) 6-trans-LTB4 CAY35250); 6-trans,12-epi-LTB4 CAY35265); 20-OH-LTB4 CAY20190) 20-COOH-LTB4 CAY20180) LTC4 CAY20210); LTD4 LTE4 5-HETE 12-HETE 15-HETE CAY34700 Cambridge Bioscience Charles N. Serhan Medical School NIH-funded P01GM095467) Dulbecco’s phosphate-buffered saline D8537)).Pathobiology Early Arthritis CohortPlasma samples baseline 6 months 112 44 patients approved King’s College Hospital Research Ethics Committee 05/Q0703/198) Patients informed consentPeripheral blood samples synovial tissue obtained from patients Barts Health NHS Trust Early Arthritis Cohort ultrasound synovial biopsy inflamed joint patients DMARDs steroid-naive symptoms duration <12 months fulfilled ACR/EULAR 2010 classification criteria RA RA categorised three pathotypes lympho-myeloid diffuse-myeloid pauci-immune-fibroid treated with DMARDs Response status after 6 months mixed DMARD determined by EULAR response criteria.Targeted lipid mediator profilingPlasma obtained from peripheral blood centrifugation 1500 × g 10 min temperature samples extracted solid-phase extraction columns 20,34 step extraction analysis quantitation procedures protocol Protocol Exchange35 deuterated internal standards added Samples kept at −20 °C 45 min protein precipitationSupernatants solid-phase extraction methyl formate fraction collected suspended (methanol/water 1:1 injection Shimadzu LC-20AD HPLC Shimadzu SIL-20AC autoinjector QTrap 5500 or 6500+ Agilent Poroshell 120 EC-C18 column (100 mm × 4.6 mm 2.7 μm kept 50 °C mediators eluted mobile phase methanol/water/acetic acid 20:80:0.01 ramped to 50:50:0.01 0.5 min 80:20:0.01 2 to 11 min maintained till 14.5 min ramped 98:2:0.01 0.1 min maintained 98:2:0.01 5.4 min flow rate 0.5 ml/min QTrap 5500 6500+ operated multiple reaction monitoring method Supplementary Tables 9 10 instrument source parameters Tables 11–12 coefficient variation sMRM transitions lipid mediators identified peak 2000 counts retention time standards 0.05 s ≥4 data points matching 6 diagnostic ions to reference standard one backbone fragment in Calibration curves obtained mediator mixtures at 0.78 1.56 3.12 6.25 12.5 25 50 100 200 pg linear calibration curves r2 values 0.98–0.99 Signal-to-noise ratio calculated script Analyst 1.6.3 intensity value highest peak LC-MS/MS extraction analysis quantitation procedures protocol Exchange35 Chiralpak AD-RH column (150 mm × 2.1 mm 5 μm) isocratic methanol/water/acetic acid 95:5:0.01 0.15 ml/min monitor isobaric monohydroxy fatty acid levels multiple reaction monitoring (MRM) method signature ion fragments molecule ref. data model lipid mediator profiles patients RA responded = 30 = 24) treatment DMARDs first PEAC-derived patient cohortlipid mediator profile DHA-derived resolvins protectins (PD1 10S,17S-diHDHA 22-OH-PD1) PCTRs maresins (MaR1 7S-diHDHA 14-oxo-MaR1 22-OH-MaR1) MCTRs 13-series resolvins (RvT1 n-3 DPA-derived resolvins-DPA protectins (PD1n-3 17S DPA maresins E-series resolvins (RvE1 leukotrienes (LXA4 5S,15S-diHETE 20-OH-LTB4 20-COOH-LTB4 6-trans-LTB4 12-epi-6-trans-LTB4) cysteinyl leukotrienes (LTC4 LTD4 LTE4) prostaglandins (PGD2 PGE2 PGF2α thromboxane (TXB2) Clinical Score model parameters disease duration erythrocyte sedimentation rate rheumatoid factor tiredness pain patient global health physician assessment swollen joints disease activity 12 US Synovial Thickness Power Doppler scoressecond patient cohort 58 patients (36 responders 22 non-responders obtained from PEAC study used test dataset for lipid mediator profiling Clinical Score models training dataset for improved models biomarkers Age sex clinical parameters not considered for model response RA patients to DMARD treatment buildingData pre-processed analysed using R Software (v3.5.1 RStudio environment (v1.1.456 two samples removed for outlier concentrations TXB2 additional sample removed lack clinical records no normalization required lipid mediator concentrations calculated concentrations scaled by subtracting mean dividing by standard deviation.Two machine-learning methodologies classifier models support vector machine (SVM randomForest37 SVM separates groups samples in spaces divided hyperplane distances not wide large nonlinear kernels radial basis function SVM used best model created models testing different times resampling ensembles 70 bootstrap iterations 70 classifiers each ensemble stable models for all groups used inbuilt automatic optimization step minimization bootstrapping error36 in R Package “classyfire” to improve validate modelsrepresentative outcomes).RandomForest operates consensus weak decision tree classifiers39 decision trees created using features vertices classes leaves each designed different randomly chosen present studies R package “randomForest” bootstrapping test method used small loop test mtry values variables value best classification performance tested ntree values decision trees before consensus tree stable models ntree of 10,000 most stable models for lipid mediator groups majority variables tested Increasing ntrees beyond improve outcomes Supplementary Fig. 6 representative outcomes).Model evaluationReceiver operating characteristic curves (ROC curves) evaluate prediction accuracy responder non-responder test created plotting true positive rate against false-positive rate sensitivity specificity model discrimination threshold changes area under curve (AUC) calculated as prediction performance curves created R package “pROC” AUC values close to 1 (AUC > 0.8) refer to good classifier models other statistics percentage of correctly classified samples specificity calculated.Feature selection model random forests showed best validation scores model improvement based on RandomForest methodologylipid mediators separated precursors (DHA n-3 DPA EPA AA models created using relevant mediators “importance” function ‘randomForest’ organizes features relevance decreased accuracy % accuracy score AUC (ROC) calculated for models best models relevant biomarkers for DMARD-responder non-responder patients selected.Pathotypes data separated pathotype pauci-immune/fibroid 28), lympho-myeloid 27) diffuse-myeloid 31). better classification models lipid mediators for manifestation disease models build using RandomForest statistic scores calculated for validation.Network analysesStatistical differences between normalised concentrations lipid mediators DMARD non-responder DMARD-responder groups determined using two t test multiple comparison correction Benjamini–Hochberg procedure lipid mediator biosynthesis networks constructed using Cytoscape v3.7.1. pathways illustrated using colours line shapes up- downregulated mediators denoted upward downward facing triangles changes node’s size Bolded mediators represent statistical differences comparison between DMARD non-responders responders with pre post-treatment dataEnzyme transcript expressionRNA extracted from blood samples solution Ambion Ribo-Pure Blood kit (ThermoFisher RNA-sequencing Illumina HiSeq2500 platform data quality-controlled FastQC trimmed removed Cutadapt Transcript abundance derived from paired FASTQ files GENCODE v24/GRCh38 transcripts Kallisto v0.43.0 Normalization data differential gene expression analysis DMARD-responder non-responder quasi-likelihood method Bioconductor R package Results expressed log counts per million each gene analyses data derivation R v3.5.141 MetaboAnalyst v4.021 Prism v8 Microsoft Excel Professional Plus 2016. Results means mean ± sem.Differences groups determined two-sided t test Mann–Whitney test Sample sizes determined variability prior experiments Partial-discrimination performed MetaboAnalyst v4.021 SIMCA v14.1 software (Umetrics mean centring unit variance scaling lipid mediator concentrations PLS-DA linear multivariate model variables class separation observations projected class modelscore plot illustrates systematic clusters observations plots higher similarity data matrix).Reporting research design Nature Research Reporting Summary.Supplementary informationPeer Review Summary | 50.5 | 0.890479 |
10.1038/s41467-020-15065-7 | PMC7060318 | Parkinson’s disease (PD) is a common neurodegenerative disorder with a complex etiology involving genetics and the environment. Here, Vallerga et al. identify two CpG probes associated with PD in a blood cell type-corrected epigenome-wide meta-analysis, implicating the SLC7A11 gene as a plausible biological target. | An improved understanding of etiological mechanisms in Parkinson’s disease (PD) is urgently needed because the number of affected individuals is projected to increase rapidly as populations age. We present results from a blood-based methylome-wide association study of PD involving meta-analysis of 229 K CpG probes in 1,132 cases and 999 controls from two independent cohorts. We identify two previously unreported epigenome-wide significant associations with PD, including cg06690548 on chromosome 4. We demonstrate that cg06690548 hypermethylation in PD is associated with down-regulation of the SLC7A11 gene and show this is consistent with an environmental exposure, as opposed to medications or genetic factors with effects on DNA methylation or gene expression. These findings are notable because SLC7A11 codes for a cysteine-glutamate anti-porter regulating levels of the antioxidant glutathione, and it is a known target of the environmental neurotoxin β-methylamino-L-alanine (BMAA). Our study identifies the SLC7A11 gene as a plausible biological target in PD. | IntroductionParkinson’s disease (PD) is a debilitating neurodegenerative disorder characterized by cytoplasmic and axonal aggregations of alpha-synuclein, known as Lewy bodies, and the progressive loss of dopaminergic neurons in the substantia nigra pars compacta (hereafter, substantia nigra) of the midbrain. The prevalence of PD is estimated to be ~1% in people over the age of 60, increasing to 3–4% at the age of 801. With rising life expectancy worldwide, the number of individuals with PD is expected to more than double by 20402, leading to an increasingly heavy social and economic burden. The etiology of PD is complex, involving both genetic and environmental factors, but the specific molecular mechanisms contributing to pathogenesis remain poorly understood. Identifying epigenetic modifications associated with PD may provide insights into disease etiology.DNA methylation at CpG dinucleotides is the most widely characterized epigenetic mechanism and is essential to normal development and maintenance of cell- and tissue-specific gene expression patterns3. Variation in DNA methylation can arise from environmental4, stochastic5, or genetic6 perturbations, and there is growing evidence that DNA methylation could mediate the relationship between these processes in influencing risk of complex disease7. In the most recent epigenetic study of PD involving 289 PD cases and 219 controls8, Chuang et al. reported 82 epigenome-wide significant CpG probes associated with disease, although none remained significant after accounting for differences in blood cell composition between cases and controls. Larger-scale studies of DNA methylation from PD case–control cohorts will be needed for identification of methylation probes robustly associated with disease risk, independent of cell type composition.Here, we report results from analyses of genome-wide blood-based DNA methylation data in 1638 unrelated individuals (851 PD cases, 787 controls) of European descent from the System Genomics of Parkinson’s Disease (SGPD) consortium (Supplementary Table 1) and 493 European individuals from the Parkinson’s disease, Environment and Gene (PEG) cohort9 (281 PD cases and 212 controls). We use the Houseman et al.10 algorithm to impute blood cell count proportions in both the SGPD and PEG samples and find consistent differences in cell type composition between PD cases and controls. We show that a substantial proportion of the variance in PD status is captured by all methylation probes jointly and we use the PEG cohort to evaluate a DNA methylation-based classifier for PD based on the aggregated effect of all CpG probes in the SGPD discovery cohort. We perform methylome-wide association studies (MWAS) of PD in the SGPD and PEG data sets using recently developed mixed linear model-based MWAS methods11, and we identify two previously unreported DNA methylation associations with PD in a meta-analysis of the SGPD and PEG MWAS summary data. We use Summary data-based Mendelian Randomization (SMR)12 to demonstrate that one of these associations, with CpG probe cg06690548 in the SLC7A11 gene, is consistent with an environmental exposure, as opposed to genetic factors influencing DNA methylation or gene expression.ResultsAnalysis of predicted cell type proportionsWe first used logistic regression to determine whether predicted cell type proportions (CTPs) were associated with PD status in the SGPD data set, after adjusting for sex and predicted age (Eq. (1), Methods). Compared with controls, we found that PD cases had more granulocytes (calculated as the sum of eosinophil and neutrophil proportions, p < 2 × 10−16), fewer B cells (p = 2.16 × 10−10) and fewer helper T cells (CD4T p = 2.97 × 10−15, CD8T p = 5.29 × 10−03), consistent with findings in a previous report13 (Fig. 1). In addition, we observed a deficiency of natural killer cells (NK, p = 7.53 × 10−05) in PD cases compared with controls (Fig. 1). To investigate whether significant differences in CTPs between PD cases and controls could be owing to medication, the relationship between levodopa equivalent daily dosage (LEDD) and CTPs was investigated in 494 unrelated PD cases from the SGPD cohort for whom data on LEDD were available. We observed significant correlations (Bonferroni-adjusted p value threshold = 8.3 × 10−03, correcting for testing of six cell types) between LEDD and the proportion of granulocytes (Pearson’s correlation = 0.17, p = 9.38 × 10−05), CD4T cells (Pearson’s correlation = −0.14, p = 2.08 × 10−03) and CD8T cells (Pearson’s correlation = −0.15, p = 1.08 × 10−03; Supplementary Fig. 1), although none of these associations survived after adjusting for disease duration (Supplementary Fig. 2), which exhibited a strong and significant association with LEDD (Supplementary Fig. 3). Moreover, the case–control difference in CTPs was similar in analyses comparing controls with the most exposed cases (top 10% of LEDD), as to the least exposed cases (bottom 10% of LEDD) (Supplementary Table 2). These analyses suggest that medication exposure is unlikely to explain the CTP differences between PD cases and controls, although we are unable to exclude a small effect of PD medications on CTPs. For this reason, and because CTP strongly influences DNA methylation measures, we conservatively corrected for CTPs in our subsequent analyses.Fig. 1Distribution of predicted blood cell type proportions (CTPs) in PD and controls.Boxplots of predicted blood CTPs in 1638 unrelated European individuals from the SGPD data set, stratified by PD status (851 PD cases, 787 controls). B cells, p = 2.16 × 10−10; CD4T cells, p = 2.97 × 10−15; CD8T cells, p = 5.29 × 10−03; eosinophils (eos) cells, p = 1.43 × 10−01; monocytes (mono) cells, p = 2.22 × 10−02; neutrophils (neu) cells, p < 2 × 10−16; natural killer (NK) cells, p = 7.53 × 10−05; granulocytes, p < 2 × 10−16. P values from logistic regression of PD status on predicted age, sex, and individual cell types. Granulocytes are calculated as the sum of eosinophil and neutrophil proportions. Boxplot center lines show the median, box limits denote upper and lower quartiles, whiskers represent 1.5 × interquartile range and individual points show outliers.Phenotypic variance for PD attributable to all probesNext, we used the omics restricted maximum likelihood (OREML) method implemented in OSCA11 to estimate the proportion of phenotypic variance for PD captured by all the probes (ρ2) (Eqs. (2) and (3), Methods). For the SGPD and PEG cohorts, these estimates were 0.28 (se = 0.05, p = 4.0 × 10−20) and 0.24 (se = 0.12, p = 1.4 × 10−05) on the observed scale (i.e., 0–1), respectively, in models adjusted for sex, predicted age, predicted smoking exposure, and predicted CTPs.MWAS of PD in the SGPD dataWe then used OSCA to perform MOA (mixed linear model-based omic association) and MOMENT (multi component mixed linear model-based omic association excluding the target) testing of 263,264 CpG probes across 1638 unrelated PD cases (N = 851) and controls (N = 787) of European ancestry from the SGPD study (Eqs. (4) and (5), Methods). No inflation of the test statistics was observed in either MOA (λ = 1.01; Supplementary Fig. 4) or MOMENT (λ = 1.04; Supplementary Fig. 5), indicating that the mixed models efficiently corrected for potential confounding factors, consistent with simulations11. In the MOA analysis, we identified two epigenome-wide significant associations with PD (Bonferroni-adjusted significance threshold, p < 1.9 × 10−07; Fig. 2; Table 1), the strongest of which was for cg16001422 on chromosome 8 (p = 2.3 × 10−08, PLEC gene; Supplementary Fig. 6; Supplementary Data 1). Using the MOMENT method, which has been shown to be more robust and reliable than MOA for probes that are strongly associated with an unknown confounding factor11, we did not identify any genome-wide significant associations (Supplementary Fig. 7; Supplementary Data 1).Fig. 2Manhattan plots of MOA MWAS of PD.a MOA MWAS meta-analysis of PD in the SGPD and PEG cohorts (N = 2131 individuals); b MOA MWAS of PD in the SGPD data set (N = 1638 individuals); c MOA MWAS for PD in the PEG data set (N = 493 individuals).Table 1Epigenome-wide significant probes for PD identified in SGPD.ChrProbeBPGeneCohortbMOAseMOApMOAbMOMseMOMpMOM8cg16001422145022842PLECSGPD−3.050.542.3 × 10−08−1.040.608.3 × 10−02PEG−0.440.960.65−8.1 × 10−030.910.9910cg2603352074004071ASCC1SGPD−1.550.291.3 × 10−07−1.090.291.9 × 10−04PEG−1.170.560.04−0.770.530.15Results are shown for the two epigenome-wide significant probes in the MOA MWAS of SGPD, together with results for these probes from the MOMENT (MOM) MWAS of SGPD and the MOA and MOMENT MWAS’s of the PEG replication data. Effect sizes (b) and standard errors (se) from OSCA are not standardized.MWAS of PD in the PEG dataNext, we performed mixed linear model-based association testing of 242,205 CpG probes in European samples (281 PD cases and 212 controls) from the PEG cohort (Methods), with no single CpG probe surpassing genome-wide significance in either the MOA or MOMENT analyses (p < 2.06 × 10−07, Fig. 2, Supplementary Figs. 4, 5, 7; Supplementary Data 2). Focusing on the MOA results, neither of the two epigenome-wide significant CpG’s identified in the SGPD cohort replicated in PEG (prep = 2.5 × 10−02; Table 1), but 17 of the 21 approximately independent (pairwise R2 < 0.1) CpGs with p < 1 × 10−04 in the SGPD MWAS showed the same direction of effect in the PEG analysis (one-sided binomial test, p = 3.6 × 10−03; empirical p ≤ 6.2 × 10−03 based on 10,000 random samples of 21 CpG probes; Supplementary Fig. 8). These observations are consistent with post hoc power analyses, which suggest that samples sizes in excess of 2000–3000 would be required for 80% power to replicate associations identified in SGPD (Supplementary Fig. 9). As expected, the effect sizes of CpG probes with p < 5 × 10−06 in the Chuang et al.8 PEG analysis in which methylation was adjusted for blood CTPs were highly correlated with those in our analysis of the PEG cohort (n = 10 probes, Pearson’s correlation = 0.93, p = 8.9 × 10−05, Supplementary Fig. 10), as were the effect sizes of genome-wide significant probes reported by Chuang et al. in their analysis without adjustment for blood CTPs (N = 78 shared probes, Pearson’s correlation = 0.77, p < 2.2 × 10−16, Supplementary Fig. 11). However, none of the 78 shared probes was genome-wide significant in our analysis of the PEG data set, and nor did any single shared probe replicate in the SGPD data (p < 0.05/78).MWAS meta-analysis of PDWe next conducted PD MWAS meta-analyses of MOA and MOMENT results for 229,071 CpG probes in 1132 PD cases and 999 controls from the SGPD and PEG cohorts (Methods). We identified two epigenome-wide significant probes in the MOA meta-analysis (Bonferroni-adjusted significance threshold, p < 2.18 × 10−07; Fig. 2; Table 2; Supplementary Fig. 4; Supplementary Data 3), including one not previously identified in the SGPD MWAS (cg06690548 on chromosome 4; Supplementary Fig. 12). In the MOMENT meta-analysis, no single probe was epigenome-wide significant (Supplementary Figs. 5, 7; Supplementary Data 3), but cg06690548 was the most strongly associated probe (beta = 0.92, se = 0.20, p = 2.3 × 10−06) and the effect sizes of CpGs with p < 1 × 10−04 in the MOA meta-analysis (N = 58) were highly correlated with those in the MOMENT meta-analysis (Pearson’s correlation = 0.93, p < 2.2 × 10−16, Supplementary Fig. 13). Neither of the two epigenome-wide significant probes in the MOA meta-analysis harbored a common SNP in the 3′ 5bp-subsequence of the probe that could have influenced our results. There was no evidence for strongly associated probes in known PD genes (e.g., SNCA) in either the MOA or MOMENT meta-analyses (Supplementary Data 4).Table 2Epigenome-wide significant probes identified in the meta-analysis of SGPD and PEG.ChrProbeBPGenebMOAseMOApMOAbMOMseMOMpMOM4cg06690548139162808SLC7A111.090.206.2 × 10−080.920.192.3 × 10−0610cg2603352074004071ASCC1−1.470.261.6 × 10−08−1.010.257.5 × 10−05Results are shown for the two epigenome-wide significant probes identified in the MOA MWAS meta-analysis, together with results for these probes from the MOMENT (MOM) MWAS meta-analysis. Effect sizes (b) and standard errors (se) from OSCA are not standardized.Summary data-based Mendelian randomizationNext, we used SMR12 to identify specific genes associated with epigenome-wide significant methylation probes in the MOA analyses, and to evaluate evidence for genetic association between DNA methylation, gene expression and PD (Methods). Of the two significant probes in the MWAS meta-analysis based on MOA, and the additional epigenome-wide significant probe in the discovery MWAS, the only probe with a significant methylation quantitative trait locus (mQTL) (p < 5 × 10−08) in either blood or brain was cg06690548 on chromosome 4. SMR applied to this locus identified a significant association (Bonferroni-adjusted significance threshold correcting for 12 tests = 4.2 × 10−03) between cg06690548 hypermethylation and downregulation of the neighboring gene SLC7A11 (SMR: p = 3.59 × 10−03, Fig. 3 and Supplementary Table 3), but there was no evidence for a genetic association between PD and either cg06690548 methylation or SLC7A11 expression (i.e., there was no evidence that the association of cg06690548 hypermethylation with PD was owing to genetic factors that influence cg06690548 methylation or SLC7A11 expression). The HEterogeneity In Dependent Instrument (HEIDI) test12 (p = 0.16) enabled us to rule out the possibility of an association between cg06690548 and SLC7A11 expression owing to linkage disequilibrium between two distinct causal variants.Fig. 3Summary data-based Mendelian randomization analyses at the SLC7A11 locus.a The uppermost plot shows −log10(P values) of SNPs from the PDWBS GWAS meta-analysis63. The red diamonds and blue circles represent −log10(P values) from SMR tests for association of gene expression and DNA methylation probes with PD, respectively. Neither SLC7A11 nor cg06690548 show evidence for a genetic association with PD. The middle plot shows −log10(P values) of the SNP associations for gene expression probe ENSG00000151012 (tagging SLC7A11) from the Brain-eMeta eQTL study57. The bottom plot shows −log10(P values) of the SNP associations for DNA methylation probe cg06690548 from the Brain-mMeta mQTL study57. b Relationship between effect sizes of mQTLs for cg06690548 from the Brain-mMeta mQTL study57 and the corresponding effect sizes for gene expression probe ENSG00000151012 from the Brain-eMeta eQTL study57 (SMR, p = 3.59 × 10−03). The red triangle shows the top cis-mQTL, blue circles indicate cis-mQTLs. Error bars show the standard errors of the SNP effects.DNA methylation-based classification of PDFinally, we examined the efficacy of a DNA methylation-based classifier for PD, using SGPD as the discovery sample and PEG as the target sample (Methods). We obtained best linear unbiased prediction-based estimates of effect sizes for 263,264 probes from a mixed linear model-based analysis in SGPD (adjusted for sex, predicted age, predicted smoking exposure, and predicted CTPs) and used these to generate methylation profile scores for each European individual in the PEG sample. As expected, methylation profile scores were positively associated with PD in PEG (Fig. 4a), and in a logistic regression model adjusted for sex, predicted age, predicted smoking exposure and predicted CTPs, were found to capture 2.8% (Nagelkerke R2, p = 7.0 × 10−04) of the variance in case–control status (on the observed scale). The area under the receiver operator characteristic curve (AUC) was 0.70 (95% CI = 0.66–0.75; Fig. 4b).Fig. 4DNA methylation-based classification of PD.a Density distribution of the SGPD-derived DNA methylation-based classifier in the PEG cohort. Orange and gray represent the distributions of the z-scaled classifier in PD cases and controls in the PEG cohort (respectively). b Area under the receiver operator characteristic curve (AUC) of the SGPD-based DNA methylation classifier in the PEG cohort (AUC = 0.70, 95% CI = 0.66–0.75).DiscussionIn our analysis of blood-based DNA methylation in 1132 PD cases and 999 controls, we identified two DNA methylation probes associated with PD using MOA: cg26033520 on chromosome 10 in the vicinity of the ASCC1 (Activating Signal Co-integrator 1 Complex subunit 1) gene and cg06690548 on chromosome 4 in the promoter of the SLC7A11 (Solute Carrier Family 7 member 11) gene. The latter was also the most strongly associated probe in the MOMENT meta-analysis, suggesting that it is unlikely to be a false positive, despite not surpassing epigenome-wide significance using that method. Our results are consistent with previous simulations showing that MOMENT is more reliable and robust but slightly less powerful than MOA11.A particularly interesting finding from the SMR analysis was that hypermethylation of cg06690548 in PD was associated with downregulation of SLC7A11 but there was no evidence for a genetic association between PD and either cg06690548 methylation or SLC7A11 expression. To our knowledge the SMR method has not previously been used to rule out a genetic effect underlying a disease association. The SMR results imply that the association of cg06690548 with PD is not owing to genetic factors and so may reflect a PD-related environmental exposure or be a consequence of disease (e.g., a medication effect). We investigated the latter possibility by applying MOA (with sex, predicted age, predicted smoking exposure, and predicted CTPs as covariates) to data on LEDD in 494 unrelated European PD cases in the SGPD cohort. We found no evidence for an association of cg06690548 methylation with PD medication dosage (beta = 0.50; s.e. = 0.37, p = 0.17; unpublished data), although we cannot rule out the possibility of a brain-specific association.The SLC7A11 gene encodes a sodium-independent cysteine-glutamate antiporter known as system Xc- (or xCT). The antiporter couples the release of one molecule of intracellular glutamate, which is necessary for excitatory signaling between neurons, with the uptake of one molecule of extracellular cystine, which is rate limiting for synthesis of glutathione, the primary antioxidant in the brain. Downregulation of system Xc- results in decreased intracellular levels of glutathione14,15, which in turn leads to increased oxidative stress. Disruption of glutamate signaling has been associated with multiple neurodegenerative diseases16, and for reasons that are not yet understood, dopaminergic neurons in the substantia nigra appear to be especially susceptible to damage by reactive oxygen species17. This is notable given that loss of dopaminergic neurons in the substantia nigra is the hallmark pathology of PD and reduced levels of glutathione have been reported in the substantia nigra, but not other brain regions, and in human olfactory neural stem cells in PD patients compared with aged-matched controls15,18,19. In this context, our SMR findings suggest that the mechanism underlying the association of cg06690548 hypermethylation with PD is that this causes downregulation of system Xc- (Supplementary Table 3), resulting in reduced glutathione levels and increased oxidative stress, thereby triggering degeneration of dopaminergic neurons in the substantia nigra.Intriguingly, system Xc-, coded by the SLC7A11 gene, is a target of the environmental neurotoxin β-methylamino-l-alanine (BMAA)20, a non-protein amino acid produced by cyanobacteria (i.e., blue green algae)21. Chronic dietary exposure to BMAA is believed to be the cause of the amyotrophic lateral sclerosis/parkinsonism-dementia complex (ALS/PDC) that in the 1950′s had a 50 to 100-fold higher prevalence in the Chamorro people of Guam than in developed countries22–25. Support for the BMAA hypothesis in neurodegeneration has waxed and waned in the 30 years since the hypothesis was first proposed26–30, but there has been renewed interest in recent years based on diverse evidence, including that dietary exposure to BMAA can induce neurofibrillary tangles and β-amyloid plaques similar to those in brain tissues from Chamorros with ALS/PDC31 and that regional variation in ALS prevalence is associated with cyanobacterial algal blooms32. The mechanism of BMAA neurotoxicity is thought to involve competition between BMAA and cystine at the cystine/glutamate antiporter, leading to reduced uptake of cystine, a depletion of intracellular glutathione and increased oxidative stress33. Concurrently, this competition may also lead to excitotoxicity via elevated glutamate release and activation of metabotropic glutamate receptor 5 (mGluR5)34, and a third possibility is that BMAA transported by system Xc- may be mis-incorporated in human proteins in place of l-serine, leading to misfolding and aggregation35.Our findings, although not directly implicating BMAA in risk of PD, are notable for being entirely independent of previous reports—spanning several decades—for an association of this molecule with neurodegenerative disease. More research is warranted to determine whether dietary exposure to BMAA results in hypermethylation of cg06690548, if altered methylation at this CpG also occurs in the substantia nigra, and whether these methylation changes are associated with an increase in oxidative stress in these cells via downregulation of system Xc-. Confirmation of a causal link between BMAA exposure and risk of PD would have significant public health implications, as monitoring and control of BMAA in the food chain may be an efficient and cost-effective intervention for Parkinson’s disease and other neurodegenerative disorders. We acknowledge that environmental factors other than BMAA, such as pesticide exposure, could also plausibly influence cg06690548 methylation and SLC7A11 expression in PD.The gene closest to the most significant probe (cg26033520) identified in our MOA MWAS meta-analysis (ranked 35th out of >219 K probes in the MOMENT meta-analysis) was ASSC1, which encodes a subunit of the activating signal co-integrator 1 complex. ASCC1 is a compelling candidate for PD because a recessive mutation in this gene has been associated with spinal muscular atrophy with congenital bone fractures-2 (SMABF2)36, and studies in animal models have demonstrated that ASCC1 knockdown is associated with impaired axonal outgrowth of alpha-motor neurons, impaired neuromuscular junction formation and compromised motor functioning36. More generally, ASCC1 is a transcriptional activator that has an important role in gene transactivation via multiple transcription factors including nuclear factor kappa-B (NF-kB). NF-kB has a key role in regulating the immune response to infection and dysregulated NF-κB activation has been associated with cancer, inflammatory, and autoimmune diseases and septic shock37. This is noteworthy given recent findings from in vivo studies of PD, supporting a role for immune system dysfunction in disease onset and progression, including via cytokine activation and neuroinflammation38.The strongest signal in the SGPD MOA MWAS was for cg16001422, located in the proximity of the PLEC gene. This probe failed to replicate in MOA analysis of PEG, did not achieve epigenome-wide significance in the MOA meta-analysis and was not supported by MOMENT in SGPD, PEG or the meta-analysis. Nonetheless, it is noteworthy that PLEC, which encodes a large polypeptide acting as a crosslinker between actin microfilaments, microtubules, and intermediate filaments in the cell cytoskeleton39, has previously been associated with multiple autosomal recessive forms of muscular dystrophy, including epidermolysis bullosa simplex with muscular dystrophy, and muscular dystrophy, limb-girdle, autosomal recessive 1740–43. Our results suggest that further investigations are merited on the role of PLEC in risk of PD.The proportion of variance in PD status associated with all the methylation probes (ρ2) was consistently high in both the SGPD and PEG cohorts (0.28 and 0.24, respectively), and similar in magnitude to the variance in PD status explained by common SNPs (0.28 in SGPD, estimated using the genomic-relatedness-based restricted maximum likelihood method implemented in the GCTA software44). Notably, when the two epigenome-wide significant probes identified in the MOA SGPD MWAS were fitted as fixed effects (in addition to sex, predicted age, predicted smoking exposure, and predicted CTPs) in an OREML analysis of the PEG data, ρ2 decreased from 0.24 to 0.21. These results imply that the majority of variance in PD status associated with DNA methylation is still unaccounted for, and that larger PD MWAS are likely to identify additional genomic regions associated with PD.A DNA methylation classifier based on best linear unbiased prediction-based effect sizes for all probes from an mixed linear model analysis of SGPD captured ~2.8% of the variance in PD case–control status in the PEG data set. This estimate was highly statistically significant but lower than previously reported adjusted R2 estimates for methylation-based classifiers of other complex traits (e.g., BMI, R2 = 12.2% in the Lothian Birth Cohorts;45,46). The receiver operator characteristic curve is used in clinical epidemiology to quantify how well medical diagnostic tests discriminate between affected and non-affected (or controls) individuals. Accuracy is measured by the area under the receiver operator characteristic curve (AUC). An AUC of 1 represents a perfect test, whereas an AUC of 0.5 represents a test with no discriminatory power. In this study, the AUC of the methylation-based classifier for PD was estimated at ~ 0.70. This value is comparable to that of the latest GWAS-based predictors for PD47, but we do not know if PD-associated methylation differences are a cause or consequence of the disease. For instance, methylation differences may be owing to medication effects or to some other physiological change that occurs as a consequence of having PD. This has implications for the utility of methylation-based classifiers, as they may differentiate PD cases from controls, but have no predictive value in terms of identifying individuals at high risk of PD in the population (i.e., before they develop the disease).Finally, prior studies of PD13 have reported disease-associated differences in blood CTPs. We replicated a number of these CTP differences in analysis of PD cases and controls in our SGPD data, but nonetheless chose to adjust for CTPs in our MWAS analyses, owing to the fact that we could not rule out the possibility that PD medications contributed to these differences. We acknowledge that this approach is conservative and that true positive methylation associations with PD may have been overlooked as a result. Further investigations will be needed to establish the extent to which CTP differences associated with PD are part of the disease process, as opposed to a consequence of the disease. We also acknowledge that analysis of whole blood-based DNA is a limitation, given that PD is a disorder of the central and peripheral nervous system. Analysis of DNA methylation in post-mortem brain tissues from PD cases and controls should be a future priority for the field.In conclusion, we describe two previously unreported associations between DNA methylation and PD, including with the ASCC1 gene on chromosome 10 and the SLC7A11 gene on chromosome 4. These associations provide support for glutamate signaling disruption, oxidative stress, and neuroinflammation as potential etiological mechanisms contributing to PD. The SLC7A11 association is especially notable given prior evidence that this gene is a target of the neurotoxin BMAA. An important caveat of our study is that epigenome-wide significant associations were only identified with MOA, which has a higher false positive rate than MOMENT11. However, we have shown that the two probes achieving epigenome-wide significance in the MOA meta-analysis were also among the most strongly associated probes in the MOMENT meta-analysis (ranks 1 and 35 for cg06690548 and cg26033520, respectively). This suggests that these associations are unlikely to be false positives, and that their failure to achieve epigenome-wide significance in MOMENT is likely owing to the slightly reduced power of this method relative to MOA11. Our linear mixed model analyses revealed that a surprisingly high proportion of variance in PD status was associated with DNA methylation measured in whole blood, suggesting that studies of DNA methylation in larger PD case–control cohorts are likely to yield additional discoveries relevant to the disorder. A DNA methylation classifier based on the SGPD data captured a significant proportion of the variance in PD status in the independent PEG cohort, but an improved understanding of the mechanisms of cause and effect in relation to DNA methylation probes altered in PD is required prior to the application of DNA methylation-based classifiers as a diagnostic tool.MethodsStudy populationsThe System Genomics of Parkinson’s Disease (SGPD) cohort comprises genotype, phenotype, and DNA methylation data for a total of 2333 participants (1292 PD cases, 1041 controls) recruited from three different studies across Australia and New Zealand: (1) the Queensland Parkinson’s Project (QPP), (2) the New Zealand Brain Research Institute PD case–control cohort (NZBRI), and (3) the Sydney PD case–control cohort (SYD).The QPP cohort includes 1791 participants (867 PD cases, 924 controls) mostly of European ancestry. PD was diagnosed according to standard criteria48 and controls consisted of healthy community-based, age-matched volunteers residing in the same area and from the same ethnic background as the PD patients (N = 507), together with patients’ spouses (N = 266) and siblings (N = 151). Whole blood or saliva (N = 4) samples were collected at the time of recruitment. A total of 1692 individuals completed the Parkinsonism and related neurological disorders survey and underwent the same evaluations, which included questionnaires on demographics, medical history, environmental exposures and the Geriatric Depression Scale (GDS). In addition, cases undertook a cognition test summarized by the Unified Parkinson’s Disease Rating Scale (UPDRS) score.The NZBRI cohort comprises 210 participants (151 PD cases, 59 matched controls) recruited by the NZBRI. Exclusion criteria for PD patients were prior history of learning disability, severe head injury, stroke, or other neurological impairment and major psychiatric complications at the point of study entry. Whole blood samples were collected at the same time as phenotypic measurements, which included demographic, medical, and environmental exposure information for all participants. In addition, cases underwent periodic cognition tests, which included the Parkinson’s Disease Dementia (PDD) criteria49, the Parkinson’s Disease and Mild Cognitive Impairment (PD-MCI) criteria50, and the UPDRS score.In the SYD cohort, 332 participants (274 PD cases, 57 matched controls, 1 individual with missing phenotype) were recruited from the Parkinson’s Disease Research Clinic, Brain and Mind Research Institute at the University of Sydney. PD was diagnosed according to the UKPDS Brain Bank clinical diagnostic criteria51.The PEG study is a large population-based study of PD of mostly rural and township residents of California’s central valley9. The PEG study comprises of 508 European (289 PD cases, 219 controls) and 64 Hispanic individuals (46 PD cases, 18 controls) for a total of 334 PD cases and 237 controls. Cohort details and DNA extraction methods are described elsewhere9,13.Ethics approval and consent to participateAll participants gave written consent. The QPP study was approved by the Griffith University Human Research Ethics Committee (ESK0411HREC). The Southern Health and Disability Ethics committee (New Zealand) approved the NZBRI study protocol (URA/11/08/042; URB/09/08/037). The SYD study protocol was approved by the University of Sydney Human Research Ethics Committee (10963; 2013/945). Approval for analysis of SGPD samples was granted by the University of Queensland Human Research Ethics Committee (2011001173). The PEG study was approved by the UCLA Institutional Review Board (IRB 11–001530).Genotyping and quality controlDNA samples from SGPD participants and 26 plate controls were genotyped using the Illumina PsychArray-B.bpm (571,054 SNPs) at the QIMR Berghofer Molecular Epidemiology Laboratory. Standard individual-level and SNP-level quality control was applied to the data. A total of 39 unique samples were excluded from the SGPD data set on the basis of unresolved sex discrepancies (20 samples), high missing data rate (genotype failure rate 0.03; 17 samples) and sample duplications (PIhat > 0.9; 10 samples). SNP markers were excluded if they had missing genotype rate >5%, Hardy–Weinberg Equilibrium (HWE) test p value < 1 × 10−05 in controls, different missing genotype rates between cases and controls (p value < 1 × 10−05) or low minor allele frequency (MAF < 0.01). We additionally dropped a total of 5293 duplicated SNPs. This screen reduced the total number of analyzed SNPs by ~49% (predominantly owing to the MAF filter), leaving 288,452 SNPs and 2227 individuals (1688 from QPP, 210 from NZBRI, 329 from SYD). We performed multidimensional scaling analysis on the cleaned SNP data to establish the genetic ancestry of study participants. We merged the SGPD sample with the HapMap3 data set comprising 988 individuals across 11 populations and defined Europeans as those falling within ±5 SD of the mean of the HapMap3 European cluster based on the top two principal components. A total of 78 individuals (62 from QPP, 4 from NZBRI and 12 from SYD) showed evidence for non-European genetic ancestry and were removed. The cleaned SGPD genotypes were imputed to the Haplotype Reference Consortium52 and then additional filtering was performed to remove SNPs with low MAF < 0.01, HWE test p value in controls <1 × 10−05 and INFO score <0.3. This screen reduced the total number of imputed SNPs (~38 million) by 81%, leaving 7,582,086 SNPs and 2138 individuals of European ancestry (1169 PD cases, 65% male; and 968 controls, 46% male, 1 individual with missing phenotype). These data were used solely for the purpose of identifying unrelated individuals for inclusion in DNA methylation analyses, as described below.DNA methylation and quality controlWhole blood-derived genomic DNA from a total of 1974 SGPD individuals was bisulphite-converted using the EZ-96 DNA Methylation kit (Zymo Research). The Human Methylation 450 K BeadChip was used to assess methylation status at 485,512 CpG sites across the genome. Data were available on 1704 (820 cases, 884 controls) QPP participants, 210 (151 cases, 59 controls) NZBRI participants, and 60 (30 cases, 30 controls) SYD participants. Methylation arrays were run in two batches: the first comprising 940 samples (835 QPP, 105 NZBRI) and the second comprising 1034 samples (869 QPP, 105 NZBRI, 60 SYD). Samples were randomly placed with respect to array and array position in order to minimize the potential for batch effects. Low-quality probes and samples were excluded from further analysis as described below. Methylation scores for each CpG site, obtained as a ratio of the intensities of fluorescent signals, are represented as β-values, which range between 0 and 1; a value of 0 indicates that all copies of the CpG site in the sample (i.e., all cells) were completely un-methylated, and a value of 1 indicates that every copy of the site was methylated. Raw intensity data were background-corrected and normalized using internal controls, and methylation β-values were generated using the R meffil package. Quality control (QC) was performed to remove probes with a low (<95%) detection rate at p < 0.01 and those that failed the minimum threshold for the number of beads (N = 3). The R meffil package was also used to perform sample QC using Illumina recommended thresholds. Samples were dropped if call rate was low (<450,000 probes detected at p < 0.001), if predicted sex, based on XY probes, did not match reported sex, and if predicted median methylated signal was more than three standard deviations from the expected. A total of four individuals whose DNA was extracted from saliva were also excluded. After these QC steps, methylation β-values were quantile-normalized with respect to 20 principal components (PCs) generated from the control matrix and the most variable probes. In addition, normalization was adjusted for batch, slide, cohort, sentrix row/column, sex, and age. A total of 485,237 probes and 1889 samples (959 PD cases, 930 controls) were retained following QC. We used the GCTA software44 to generate a genetic relationship matrix (GRM) using all ~7.5 M imputed QC-pass SNPs, and removed one of each pair of individuals (total N = 190) with an estimated pairwise genetic relationship >0.05. Subsequently, the Omics-data-based Complex trait Analysis (OSCA) software11 (see http://cnsgenomics.com/software/osca) was used to perform additional filtering of cross-reactive probes (29,218), X/Y-chromosomes probes (10,711) and lowly variable probes (sd < 0.02), leaving 263,264 probes and 1638 unrelated European individuals (851 PD cases, 787 controls) available for analysis.We downloaded Illumina Infinium Human Methylation 450 K BeadChip data from 508 European (289 PD cases, 219 controls) and 64 Hispanic individuals (46 PD cases, 18 controls) in the PEG study from GEO (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE111629), with the purpose of replicating epigenome-wide significant probes identified in the SGPD EWAS and performing meta-analysis. We performed DNA methylation QC on the European samples from the PEG data using the same R meffil pipeline applied to the SGPD data set. Overall, 15 unique European samples were filtered out owing to low call rate (<450,000 probes detected at p < 0.001; N = 10) or because predicted sex did not match reported sex (N = 6). After these QC steps, methylation β-values were quantile-normalized with respect to 15 PCs generated from the control matrix and the most variable probes. In this study, we chose 15 PCs as opposed to 20 because the variance captured by PCs 16–20 was minimal. In addition, normalization was adjusted for slide, sentrix row/column, sex, and age. Finally, cross-reactive probes, X/Y-chromosomes probes and lowly variable probes (sd < 0.02) were also filtered out leaving 242,205 probes and 493 European samples available for analysis.Cell type proportionsThe R Meffil package was used to impute blood CTPs according to the Houseman et al.10 method for each individual in both the SGPD and the PEG cohorts. Differences in CTPs between PD cases and controls were then singularly tested through logistic regression of PD status on covariates (predicted age, sex) to account for possible confounding factors. Specifically, we used the following model:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{y}} = \beta _0 + \mathop {\sum}\limits_{\mathrm{i}} {\beta _i{\mathbf{X}}_{\boldsymbol{i}} + {\mathbf{e}}}$$\end{document}y=β0+ ∑iβiXi+ewhere y is an n × 1 vector of n individuals representing the log odds that an individual is a case, i ∈ [1, k] with k being the number of fitted covariates, βi is the effect of the ith covariate on the phenotype, Xi is the ith independent variable or covariate and e is a n × 1 vector of residuals. Finally, individual CTPs in 494 unrelated PD cases from the SGPD cohort were regressed on LEDD calculated accordingly to Tomlinson et al.53. LEDD is a quantitative measure of PD-specific medication measured as mg/day. In a few instances, exact LEDD was not available but was reported as a range (e.g., 530–1800 mg); in these cases, LEDD was calculated as the mid-point between the lowest and highest values in the range. Prior to the regression analysis, LEDD was log-transformed owing to the right skewness of its distribution and standardized.OREML analysesThe OREML method implemented in the OSCA11 software is a model in which a single random effect component is used to estimate the proportion of variance in a specific trait captured by all the DNA methylation probes. The model can be written as:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{y}} = {\mathbf{C\beta }} + {\mathbf{Wu}} + {\mathbf{e}}$$\end{document}y=Cβ+Wu+ewhere y is an n × 1 vector of phenotype values of n individuals (coded as 1 for PD and 0 for unaffected), C is an n × p matrix for covariates (e.g., sex, age) with p being the number of fitted covariates, β is a p × 1 vector of the effects of the fixed covariates on the phenotype, W is an n × m matrix of m standardized DNA methylation values, where m is the number of DNA methylation sites, u is an m × 1 vector of the joint effects of all the probes on the phenotype, and e is an n × 1 vector of residuals. The variance-covariance matrix for y can be written as:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{var}}\left( {\mathbf{y}} \right) = {\mathbf{V}} = {\mathbf{WW}}^\prime {\upsigma}_{\mathrm{u}}^2 + {\mathbf{I}}{\upsigma}_{\mathrm{e}}^2 = {\mathbf{A}}{\upsigma}_o^2 + {\mathbf{I}}{\upsigma}_{\mathrm{e}}^2$$\end{document}vary=V =WW′σu2+Iσe2=Aσo2+Iσe2where A (the omics-data-based relationship matrix or ORM) is defined as WW′/m and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\upsigma}_o^2 = m{\upsigma}_u^2$$\end{document}σo2=mσu2. In such a mixed linear model, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\upsigma}_o^2$$\end{document}σo2 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\upsigma}_e^2$$\end{document}σe2 are the variance components and the amount of variance in the phenotype attributable to all the probes is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho ^2 = \sigma _o^2/\left( {\sigma _o^2 + \sigma _e^2} \right)$$\end{document}ρ2=σo2∕σo2+σe2. Estimation of the variance components can be computed using REML54, which in OSCA nomenclature is OREML. In the context of DNA methylation data, ρ2 parallels the concept of SNP-based heritability as implemented in GCTA44, with the caveat that association signals of DNA methylation data may reflect both cause and consequence of disease. Here, estimates of ρ2 were from a model that included both confounder covariates (sex, predicted age, predicted smoking exposure) and predicted CTPs (with the exclusion of eosinophils).Mixed linear model-based MWAS of PDWe used two mixed linear model-based approaches implemented in the OSCA11 software to test for association between disease status and DNA methylation: MOA and MOMENT. Specifically, the MOA model is:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{y}} = {\mathbf{w}}_{\boldsymbol{i}}b_i + {\mathbf{C\beta }} + {\mathbf{Wu}} + {\mathbf{e}}$$\end{document}y=wibi+Cβ+Wu+ewhere wi is an n × 1 vector of standardized DNA methylation measures of the target probe i and bi is the effect of probe i on the phenotype (fixed effect). Important properties of the MOA model are that the target probe is fitted both as a fixed effect (bi) and a random effect (the ith element of the vector u) and effect sizes of probes in the random effect term are assumed to come from a single distribution. The MOMENT approach differs by partitioning probes, on the basis of their statistical association with the phenotype in an initial linear regression, into two random effects with different effect size distributions. This method has been shown to be more robust to potential confounders than MOA but at the cost of slightly reduced power11. The MOMENT model is:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbf{y}} = {\mathbf{w}}_{\boldsymbol{i}}b_i + {\mathbf{C\beta }} + \mathop {\sum}\limits_{\mathrm{j}} {{\mathbf{W}}_{\mathbf{j}}{\mathbf{u}}_{\mathbf{j}} + {\mathbf{e}}}$$\end{document}y=wibi+Cβ+ ∑jWjuj+ewhere Wj is an n × mj matrix of standardized DNA methylation probe values in the jth group, and mj is the number of DNA methylation sites in the group (excluding the DNA methylation sites <50 Kb from the probe being tested).Application of MOA and MOMENT to the SGPD data set involved analysis of 263,264 DNA methylation probes in 1638 (851 PD cases, 787 controls) unrelated (estimated pairwise relationship <0.05 calculated using GCTA44) European individuals, and for the PEG data set involved analysis of 242,205 probes in 493 European individuals (281 PD cases, 212 controls). Prior to association analyses we regressed DNA methylation measures on known batch effects, which included sentrix row, column and slide in SGPD, and sentrix row and column only in PEG, because slide was significantly associated with disease status in that sample. Sex, predicted age55, predicted smoking exposure (unpublished work) and predicted CTPs (with the exception of eosinophils) were fitted as fixed effects. We checked strongly associated probes for the presence of common SNPs, using the masking manifests provided by Illumina (https://zwdzwd.github.io/InfiniumAnnotation).MWAS meta-analysis of PDWe performed a MWAS meta-analysis of MOA or MOMENT results for PD using summary data from the SGPD and PEG cohorts, comprising analysis of 229,071 common methylation probes in a total of 1132 PD cases and 999 controls. The analyses were conducted in OSCA, which implements the conventional inverse-weighted-variance analysis assuming independence among cohorts. Novel signals of association were those surpassing the Bonferroni-adjusted significance threshold (p = 0.05/229, 071 = 2.2 × 10−07).Summary data-based Mendelian randomizationThe SMR and HEIDI methods12 were used to identify putative target genes regulated by PD-associated CpG sites, and to examine evidence for causal relationships between DNA methylation, gene expression and PD. Specifically, to infer the regulatory role of epigenome-wide significant probes on PD, we applied the SMR approach to test the hypothesis that DNA methylation is associated with PD through regulation of gene expression56. This can be done in the SMR framework by testing the pairwise associations between epigenome-wide CpG probes and PD, epigenome-wide CpG probes and genes within 2 Mb distance in either directions, and between these genes and PD. These analyses involved summary-level SNP data, including Brain-eMeta57 eQTL data (neff = 1194) obtained from the meta-analysis of GTEx brain58, CMC59, and ROSMAP60 correcting for sample overlap by the MeCS method57, Brain-mMeta mQTL57 data (neff = 1160) obtained from the meta-analysis of ROSMAP60, Hannon et al.61, and Jaffe et al.62, and PD GWAS summary data (neff = 308,518) from the PDWBS meta-analysis63 of 6476 PD cases and 302,042 controls. Consistent with assumptions in the SMR method, we included only genes with at least one cis-eQTL at PeQTL < 5 × 10−08 in Brain-eMeta, leaving 12 genes for analysis.DNA methylation-based out-of-sample classificationIn the SGPD cohort, we estimated the aggregated effect of all the probes on disease status using a mixed linear model, which included sex, predicted age, predicted smoking exposure, and predicted CTPs as fixed effects, as implemented in OSCA. Subsequently, the best linear unbiased prediction solutions for the probe effects were calculated and used to generate DNA methylation profile scores for European individuals in the PEG study. Using logistic regression adjusted for sex, predicted age and predicted CTPs, we estimated the proportion of variance in PD status (Nagelkerke R2) in the PEG study captured by the standardized methylation profile scores. In addition, the classification accuracy of the methylation profile scores was evaluated by calculating the AUC.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Supplementary Dataset 1
Supplementary Dataset 2
Supplementary Dataset 3
Supplementary Dataset 4
Peer Review File
Reporting Summary
Description of Additional Supplementary Files | nature communications | [
"Article"
] | [
"DNA methylation",
"Epigenomics",
"Parkinson's disease",
"Risk factors"
] | IntroductionParkinson’s disease (PD) neurodegenerative disorder cytoplasmic axonal aggregations of alpha-synuclein Lewy bodies loss of dopaminergic neurons in substantia nigra compacta midbrain prevalence PD ~1% over 60 3–4% at 801 rising life expectancy PD expected to double by 20402 social economic burden etiology PD complex genetic environmental factors molecular mechanisms pathogenesis poorly understood epigenetic modifications PD may etiology methylation at CpG dinucleotides epigenetic mechanism essential to development cell tissue-specific gene expression Variation in DNA methylation stochastic5 perturbations evidence could risk complex recent epigenetic study PD 219 Chuang et reported 82 epigenome-wide significant CpG probes associated disease none significant after differences blood cell composition between controls Larger-scale studies DNA methylation PD case–control needed for identification methylation probes disease risk independent cellreport results genome-wide blood DNA methylation data 1638 individuals (851 PD cases 787 controls European descent System Genomics Parkinson’s Disease consortium 493 European Parkinson’s Environment Gene cohort9 (281 PD cases 212 Houseman et al algorithm impute blood cell count proportions SGPD PEG samples consistent differences cell type composition between PD cases controls substantial proportion variance PD status captured by methylation probes PEG cohort evaluate DNA methylation classifier PD effect CpG probes cohort perform methylome-wide association studies) PD SGPD PEG mixed linear model-based MWAS identify two unreported DNA methylation associations with PD meta-analysis SGPD PEG MWAS data Mendelian Randomization CpG probe cg06690548 SLC7A11 gene consistent with environmental exposure genetic factors DNA methylation predicted cell type logistic regression PD status SGPD data sex predicted ageCompared controls PD cases had more granulocytes eosinophil neutrophil proportions p < 2 × 10−16) fewer B cells (p = 2.16 × 10−10) fewer T cells (CD4T p 2.97 × 10−15 CD8T p 5.29 × 10−03) consistent with previous observed deficiency of natural killer cells (NK p = 7.53 × 10−05) in PD cases differences CTPs medication relationship levodopa dosage CTPs investigated in PD cases SGPD cohort observed correlations (Bonferroni-adjusted p threshold = 8.3 × 10−03 between LEDD granulocytes 0.17 = 9.38 × 10−05) CD4T cells −0.14 2.08 × 10−03) CD8T cells −0.15 1.08 × none associations survived after adjusting for disease duration strong association with LEDD case–control difference in CTPs similar exposed least exposed cases medication exposure unlikely to explain CTP differences between PD cases controls exclude small effect of PD medications on CTPsCTP influences DNA methylation corrected CTPs analyses. blood cell proportions PD controls CTPs 1638 European individuals SGPD data stratified PD status (851 PD cases controls). B cells 2.16 × 10−10 CD4T 2.97 × 10−15 CD8T 5.29 × 10−03 eosinophils 1.43 × 10−01 monocytes 2.22 × 10−02 neutrophils < 2 × 10−16 natural killer cells 7.53 × 10−05 granulocytes p < 2 × 10−16 values regression PD status age sex cell types Granulocytes calculated sum eosinophil neutrophil proportions lines median limits upper quartiles whiskers 1.5 × interquartile range points outliers.Phenotypic variance PD used likelihood) method variance PD probes SGPD PEG cohorts estimates 0.28 0.05 p 4.0 × 10−20) 0.24 0.12 1.4 × 10−05) scale adjusted for sex age smoking exposure CTPsPD SGPD used OSCA MOA MOMENT testing 263,264 CpG probes 1638 PD cases (N = 851) controls (N = 787) European SGPD study No inflation test statistics MOA = 1.01 MOMENT = 1.04 models corrected confounding factors MOA analysis identified two epigenome-wide significant associations PD p < 1.9 × 10−07 strongest cg16001422 on chromosome 8 (p = 2.3 × 10−08 MOMENT method more robust reliable than MOA genome-wide significant associations.Fig. plots MOA MWAS PD meta-analysis PD SGPD PEG cohorts (N = 2131 MOA MWAS PD SGPD data (N = 1638 MOA MWAS PEG data set (N = 493 individuals).Table 1Epigenome-wide significant probes PD SGPD.ChrProbeBPGeneCohortbMOAseMOApMOAbMOMseMOMpMOM8cg16001422145022842PLECSGPD−3.050.542.3 × 10−08−1.040.608.3 × 10−02PEG−0.440.960.65−8.1 × 10−030.910.9910cg2603352074004071ASCC1SGPD−1.550.291.3 10−07−1.090.291.9 10−04PEG−1.170.560.04−0.770.530.15Results two epigenome-wide probes MOA MWAS SGPD MOMENT MWAS SGPD PEG replication data Effect sizes standard errors OSCA not standardized PEG mixed linear association testing 242,205 CpG probes European samples (281 PD cases 212 controls PEG cohort no CpG probe surpassing genome-wide significance MOA MOMENT analyses (p < 2.06 × 10−07 two epigenome-wide significant CpG’s SGPD cohort replicated PEG 2.5 × 17 21 independent R2 < 0.1) CpGs p < 1 × 10−04 SGPD MWAS same direction PEG analysis p = × 10−03 empirical p 6.2 × 10−03 10,000 samples analyses samples sizes 2000–3000 80% power replicate associations SGPD effect sizes CpG probes p < 5 × 10−06 Chuang et alPEG analysis adjusted blood CTPs correlated PEG cohort 10 probes correlation 0.93 p × 10−05 effect sizes genome-wide significant probes Chuang. without adjustment 78 probes correlation 0.77 p < 2.2 × 10−16 none 78 probes genome-wide significant PEG replicate SGPD data (p < 0.05/78) meta-analysis MWAS meta-analyses MOA MOMENT 229,071 CpG probes 1132 PD cases 999 controls SGPD PEG cohorts identified two epigenome-wide significant probes MOA meta-analysis threshold p < 2.18 × 10−07 one not SGPD (cg06690548 chromosome 4 MOMENT meta-analysis no probe epigenome-wide significant cg06690548 strongly associated probe (beta = 0.92 se = 0.20 p = 2.3 × 10−06) effect sizes CpGs p < 1 × 10−04 MOA meta-analysis (N = correlated MOMENT meta-analysis correlation = 0.93 p < 2.2 × 10−16Neither two probes in MOA meta-analysis common SNP results no evidence for associated probes in PD genes in MOA or MOMENT meta-analyses.Table 2Epigenome probes SGPD and PEG.090.206.2 × 10−080.920.192.3 × 10−0610cg2603352074004071ASCC1−1.470.261.6 × 10−08−1.010.257.5 × 10−05Results two probes MOA MWAS meta-analysis MOMENT MWAS meta-analysis. Effect sizes standard errors OSCA not standardized data-based Mendelian used SMR12 identify genes significant methylation probes MOA evaluate genetic association between DNA methylation gene expression PD two significant probes in MWAS meta-analysis only probe significant methylation trait locus) (p < 5 × 10−08) was cg06690548 on chromosome 4. SMR identified significant association × between cg06690548 hypermethylation downregulation gene SLC7A11 p = 3.59 × 10−03Supplementary Table no evidence genetic association PD cg06690548 methylation SLC7A11 expression no cg06690548 hypermethylation PD genetic factors HEterogeneity Dependent Instrument) test12 (p = 0.16) out association cg06690548 SLC7A11 expression linkage disequilibrium causal variants. Mendelian randomization analyses SLC7A11 locus uppermost plot shows −log10(P values SNPs PDWBS GWAS meta-analysis63 red diamonds blue circles represent(P values SMR tests association gene expression DNA methylation probes PD Neither SLC7A11 cg06690548 genetic association PD middle plot values SNP associations gene expression probe ENSG00000151012 bottom plot DNA methylation probe cg06690548 Relationship effect sizes mQTLs cg06690548 gene expression probe ENSG00000151012 p = 3.59 × 10−03) red triangle shows top cis-mQTL blue circles cis-mQTLs Error bars standard errors SNP effectsDNA methylation classification examined efficacy DNA classifier PD SGPD discovery PEG target sample obtained estimates effect sizes 263,264 probes mixed linear model analysis SGPD sex age smoking exposure CTPs methylation profile scores European individual PEG sample methylation profile scores positively associated PD PEG (Fig. logistic regression model capture 2.8% × 10−04) variance case–control status area under receiver operator characteristic curve (AUC) 0.70 (95% CI = 0.66–0.75 Fig.. 4DNA methylation-based classification PD Density distribution SGPD-derived DNA methylation classifier PEG cohort Orange gray distributions-scaled classifier cases Area under receiver operator characteristic curve (AUC) SGPD DNA methylation classifier PEG cohort (AUC = 0.70 95% CI 0.66–0.75) analysis blood DNA methylation 1132 PD cases 999 controls identified two DNA methylation probes associated PD cg26033520 chromosome 10 gene cg06690548 chromosome 4 SLC7A11 genelatter strongly associated in MOMENT meta-analysis unlikely false positive not surpassing epigenome-wide significance results consistent simulations MOMENT more reliable robust less powerful than MOA11 SMR analysis hypermethylation cg06690548 in PD associated with downregulation SLC7A11 no evidence genetic association between PD cg06690548 methylation SLC7A11 expression SMR not used rule out genetic effect disease results imply association cg06690548 with PD not genetic may reflect PD-related environmental exposure or consequence disease medication investigated latter MOA sex age smoking exposure CTPs to LEDD in European PD cases SGPD cohort no evidence association cg06690548 methylation with PD medication dosage (beta = 0.50 s. = 0.37 p = 0.17 rule out brain-specific association SLC7A11 gene encodes sodium-independent cysteine-glutamate antiporter system Xc- antiporter couples release intracellular glutamate extracellular cystine synthesis glutathione Downregulation Xc- decreased intracellular levels increased oxidative stressDisruption glutamate signaling associated with neurodegenerative diseases16 dopaminergic neurons in substantia nigra susceptible to damage by reactive oxygen loss of dopaminergic neurons hallmark PD reduced levels glutathione reported in not other brain regions human olfactory neural stem cells in PD patients SMR findings suggest association cg06690548 hypermethylation with PD causes downregulation system Xc- reduced glutathione levels increased oxidative stress triggering degeneration dopaminergic neurons system Xc- coded by SLC7A11 gene target of environmental neurotoxin β-methylamino-l-alanine (BMAA produced by cyanobacteria Chronic dietary exposure to BMAA cause amyotrophic lateral sclerosis/parkinsonism-dementia complex) 50 to 100-fold higher prevalence in Chamorro people of Guam Support for BMAA hypothesis neurodegeneration renewed interest evidence dietary exposure BMAA induce neurofibrillary tangles β-amyloid plaques ALS regional variation in ALS prevalence associated with cyanobacterial algal blooms32BMAA neurotoxicity competition between BMAA cystine antiporter reduced uptake cystine depletion intracellular glutathione increased oxidative competition excitotoxicity elevated glutamate release activation metabotropic glutamate receptor 5 BMAA mis-incorporated in human proteins l-serine misfolding findings not implicating BMAA risk PD independent of previous association neurodegenerative disease More research dietary exposure to BMAA hypermethylation cg06690548 altered methylation methylation changes increase oxidative stress downregulation system Xc- Confirmation causal link between BMAA exposure risk PD public health implications monitoring control BMAA food chain efficient cost intervention for Parkinson’s disease neurodegenerative disorders environmental factors BMAA exposure influence cg06690548 methylation SLC7A11 expression in PD gene closest to significant probe (cg26033520) MOA MWAS meta-analysis 35th out >219 probes ASSC1 encodes subunit activating signal co-integrator 1 complexASCC1 candidate for PD recessive mutation associated with spinal muscular atrophy congenital bone fractures-2 ASCC1 knockdown impaired axonal outgrowth alpha-motor neurons impaired neuromuscular junction formation compromised motor ASCC1 transcriptional activator gene transactivation transcription factors (NF immune response infection dysregulated activation associated with cancer inflammatory autoimmune diseases septic findings in vivo studies PD role immune system dysfunction in disease onset progression cytokine activation strongest signal SGPD MOA MWAS for cg16001422 PLEC gene failed replicate in MOA analysis PEG epigenome-wide significance MOA meta-analysis not supported by MOMENT in SGPD PEG meta-analysis PLEC encodes large polypeptide crosslinker between actin microfilaments microtubules intermediate filaments cell associated with autosomal recessive muscular dystrophy epidermolysis bullosa simplex-girdle results further investigations role PLEC risk PD proportion variance in PD status methylation probes high in SGPD PEG cohorts (0.28 0.24 similar to variance PD status SNPs (0.28 in SGPD estimated using genomic-relatedness-based restricted maximum likelihood method GCTA two epigenome-wide probes in MOA SGPD MWAS fitted fixed effects smoking analysis PEG ρ2 decreased from 0.24 to 0.21 results imply majority variance PD status DNA methylation unaccounted for larger PD MWAS likely identify additional genomic regions PD DNA methylation classifier mixed model analysis SGPD captured ~2.8% variance in PD case–control status in PEG data set estimate significant lower than adjusted R2 estimates for methylation-based classifiers BMI, R2 = 12.2% in Lothian Birth Cohorts receiver operator characteristic curve diagnostic tests between affected non-affected individuals Accuracy measured by 1 perfect test 0.5 no discriminatory power study AUC of methylation-based classifier for PD estimated at ~ 0.70 comparable to latest GWAS-based predictors for PD47 if PD-associated methylation differences cause or consequence disease differences medication effects or physiological change PDimplications for methylation-based classifiers differentiate PD cases from controls no predictive value identifying high risk PD before prior studies of PD13 reported disease-associated differences in blood CTPs replicated CTP differences in PD cases controls SGPD data for CTPs in MWAS analyses PD medications approach conservative positive methylation associations with PD may overlooked Further investigations needed establish CTP differences PD part of disease process analysis of whole blood-based DNA limitation PD disorder central peripheral nervous system Analysis of DNA methylation in post-mortem brain tissues PD cases controls future priority two unreported associations between DNA methylation PD ASCC1 gene on chromosome 10 SLC7A11 gene on chromosome 4. associations support glutamate signaling disruption oxidative stress neuroinflammation potential PD SLC7A11 association notable target of neurotoxin BMAA epigenome-wide significant associations only identified with MOA higher false positive rate than MOMENT11.two probes epigenome-wide significance in MOA meta-analysis strongly associated in MOMENT meta-analysis (ranks 1 35 cg06690548 cg26033520 associations unlikely false positives failure significance in MOMENT likely reduced power method MOA11 linear mixed model analyses high proportion variance PD status with DNA methylation whole blood studies DNA methylation in larger PD case–control cohorts likely yield additional discoveries DNA methylation classifier SGPD data captured variance PD status independent PEG cohort improved understanding cause effect required DNA methylation classifiers System Genomics of Parkinson’s Disease (SGPD) cohort genotype phenotype DNA methylation data for 2333 participants (1292 PD cases, 1041 controls from three studies Zealand Queensland Parkinson’s Zealand Sydney QPP cohort 1791 participants (867 PD cases 924 controls mostly European PD diagnosed standard controls healthy community-based age-matched volunteers same ethnic background PD patients (N = 507) patients’ spouses 266) siblings = 151). Whole blood or saliva (N = 4) samples collected at time recruitment1692 individuals completed Parkinsonism disorders survey evaluations questionnaires demographics medical history environmental exposures Geriatric Depression Scale cognition test Unified Parkinson’s Disease Rating Scale (UPDRS) score NZBRI cohort 210 participants (151 PD cases 59 controls Exclusion criteria prior learning disability severe head injury stroke neurological impairment major psychiatric complications blood samples collected measurements demographic medical environmental exposure information cognition tests Parkinson’s Disease Dementia) Mild Cognitive Impairment UPDRS score SYD cohort 332 participants (274 PD cases 57 matched controls 1 missing phenotype Parkinson’s Disease Research Clinic University of Sydney PD diagnosed UKPDS Brain Bank clinical diagnostic criteria51 PEG study PD rural residents California’s central 508 European (289 PD cases 219 controls 64 Hispanic (46 PD cases 18 controls 334 PD cases 237 controls Cohort details DNA extraction methods described participants consent study approved Griffith University Human Research Ethics Committee Southern Health Disability Ethics committee approved NZBRI study protocolSYD study protocol approved University of Sydney Ethics Committee SGPD University of Queensland Committee PEG study approved UCLA Institutional Review Board.Genotyping quality controlDNA samples SGPD participants 26 plate controls genotyped Illumina PsychArray-B.bpm (571,054 SNPs QIMR Berghofer Molecular Epidemiology Laboratory individual SNP-level quality control 39 samples excluded SGPD unresolved sex discrepancies high missing data rate sample duplications > 0.9 10 SNP markers excluded missing genotype rate >5% Hardy–Weinberg Equilibrium) test p value < 1 × 10−05 different missing genotype rates low minor allele frequency dropped 5293 duplicated SNPs reduced analyzed SNPs ~49% leaving 288,452 SNPs 2227 individuals (1688 from QPP 210 NZBRI 329 SYD). multidimensional scaling analysis SNP data genetic ancestry merged SGPD sample HapMap3 data set defined Europeans within ±5 SD mean HapMap3 European cluster 78 individuals (62 QPP 4 NZBRI 12 SYD non-European genetic ancestry removedcleaned SGPD genotypes imputed Haplotype Reference Consortium52 filtering SNPs low MAF < 0.01 HWE test p value <1 × 10−05 INFO score <0.3 reduced imputed SNPs (~38 million 81% leaving 7,582,086 SNPs 2138 individuals European (1169 PD cases 65% male 968 controls 46% male 1 missing phenotype). data used identifying unrelated individuals DNA methylation analyses methylation quality blood DNA from 1974 SGPD individuals bisulphite-converted EZ-96 DNA Methylation kit (Zymo Human Methylation 450 K BeadChip methylation status 485,512 CpG sites Data on 1704 (820 884 controls QPP 210 (151 59 controls NZBRI 60 SYD Methylation arrays run two batches first 940 105 second 1034 (869 105 60 Samples randomly placed batch effects Low-quality probes samples excluded from analysis Methylation scores CpG site fluorescent signals represented β-values range between 0 and 1 0 un-methylated 1 every methylatedintensity data background-corrected normalized controls methylation β-values generated R meffil package Quality control probes low<95% detection rate p < 0.01 minimum threshold = 3) R meffil package sample QC Illumina thresholds Samples dropped call rate low<450,000 probes p < 0.001) predicted sex match median methylated signal more than three standard deviations four individuals DNA extracted saliva excluded methylation β-values-normalized 20 components control matrix variable probes normalization adjusted batch slide cohort sentrix row/column sex age,237 probes 1889 samples (959 PD cases 930 controls retained GCTA genetic relationship matrix ~7.5 M QC-pass SNPs removed one N = 190 pairwise genetic relationship >0.05. Omics-data-based Complex trait Analysis) filtering cross-reactive probes X/Y-chromosomes lowly variable probes < 0.02) 263,264 probes 1638 unrelated European individuals (851 PD cases 787 controls available analysisdownloaded Illumina Infinium Human Methylation 450 K BeadChip data 508 European PD cases 219 controls 64 Hispanic PD 18 controls PEG study GEO replicating epigenome probes SGPD EWAS meta-analysis performed DNA methylation QC European samples R meffil pipeline SGPD 15 European samples filtered out low call rate<450,000 < 0.001 10 predicted sex match reported sex = 6) methylation β-values quantile-normalized 15 PCs control matrix variable probes chose 15 PCs variance minimal normalization adjusted slide sentrix/column sex age cross-reactive X/Y-chromosomes lowly variable probes (sd < 0.02) filtered out 242,205 probes 493 European samples analysis R Meffil package blood CTPs Houseman et al.10 method SGPD PEG cohorts Differences CTPs PD cases controls tested logistic regression PD status covariates age sex confounding factorsused model:1\documentclass[12pt{minimal\usepackage{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}\begin{document}\mathbf{y}} = \beta _0 + \mathop\sum\limits\mathrm{i}}\mathbf{X}}\boldsymbol{i}} +\mathbf{e}}}\end{document}y=β0+ ∑iβiXi y n × 1 vector of n individuals log odds case i ∈ [1, k] k fitted covariates βi effect ith covariate phenotype Xi ith independent variable e n × 1 vector of residuals CTPs in 494 unrelated PD cases SGPD cohort regressed on LEDD calculated Tomlinson et al.53 LEDD PD-specific medication mg/day exact LEDD range 530–1800 calculated mid-point between lowest highest values range LEDD log-transformed standardized.OREML method OSCA11 software single random effect component estimate proportion variance in specific trait DNA methylation probesmodel written[12pt{minimal{amsmath{upgreek\oddsidemargin-69pt}{document\mathbf{y}} ={C\beta +{Wu}} +{e}}{document}y=Cβ+Wu y n × 1 vector phenotype values n individuals 1 for PD 0 unaffected), C n × p matrix covariates sex age p number fitted covariates β p × 1 vector effects fixed covariates W n × m matrix standardized DNA methylation values m number DNA methylation sites u m × 1 vector joint effects probes e n × 1 vector residualsvariance-covariance matrix y written[12pt]{minimal}{amsmath{wasysym\oddsidemargin-69pt}{document}\mathbf{var}}\left\mathbf{y}}\right) =\mathbf{V}} =\mathbf{WW}}\upsigma\mathrm{u}}^2 +\mathbf{I}}\upsigma{e}}^2 =\mathbf{A}}\upsigma}_o^2 +^2\end{document}vary=V =WW′σu2+Iσe2=Aσo2+Iσe2where A omics-data-based relationship matrix) defined WW′/m[12pt]{minimal}{amsmath{wasysym{upgreek}\setlength{\oddsidemargin}{-69pt}{document}\upsigma}_o^2 = m{\upsigma}_u^2$\end{document}σo2=mσu2. mixed linear model[12pt]{minimal}{amsmath{wasysym{amsbsy}{mathrsfs}{upgreek}\oddsidemargin-69pt}\begin{document\upsigma}_o^2${document}σo2[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{upgreek-69pt}\upsigma}_e^2$${document}σe2 variance components variance phenotype probes defined\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$\rho ^2 = \sigma _o^2/\left\sigma _o^2 +\sigma _e^2}\right\end{document}ρ2=σo2∕σo2+σe2. variance components computed using REML54 OSCA nomenclature OREML. DNA methylation data ρ2 parallels SNP-based heritability GCTA44 association signals reflect cause consequence diseaseestimates of ρ2 from model confounder covariates (sex age smoking exposure CTPs exclusion of eosinophils).Mixed linear model-based MWAS used two approaches OSCA11 software association between disease status DNA methylation: MOA MOMENT MOA model\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}\begin{document}\mathbf{y}} ={w}}\boldsymbol{i}}b{C\beta{Wu}}\mathbf{e}}\end{document}y=wibi+Cβ+Wu wi n × 1 vector of standardized DNA methylation measures of target probe i bi effect of probe i on phenotype MOA model target probe as fixed effect random effect effect sizes of probes assumed from single distribution MOMENT approach partitioning probes into two random effects different effect size distributions method more robust to confounders than MOA reduced power11MOMENT model\documentclass[12pt\usepackage{amsmath{upgreek\oddsidemargin-69pt}{document\mathbf{y}} =\mathbf{w}}\boldsymbol\mathbf{C\beta\sum\limits\mathrm{j}}\mathbf{W}}{j}}{u}}{j}}\mathbf{e}}}\end{document}y=wibi+Cβ+ ∑jWjuj Wj n × mj matrix standardized DNA methylation probe values jth group mj number DNA methylation sites <50 Kb MOA MOMENT SGPD data set 263,264 DNA methylation probes 1638 (851 PD cases controls) unrelated pairwise relationship <0.05 GCTA44) European individuals PEG data set 242,205 probes 493 European individuals (281 PD cases 212 controls). regressed DNA methylation measures batch effects sentrix row column slide SGPD sentrix row column PEG slide associated disease status Sex age55 smoking exposure CTPs fixed effectschecked associated probes for common SNPs using masking manifests Illumina meta-analysis performed MWAS meta-analysis results for PD data SGPD PEG cohorts,071 common methylation probes 1132 PD cases 999 controls analyses conducted in OSCA inverse-weighted-variance analysis independence signals association surpassing Bonferroni-adjusted significance threshold (p = 0.05/229, 071 = 2.2 × 10−07).Summary data-based Mendelian SMR HEIDI identify target genes regulated by PD-associated CpG sites causal relationships between DNA methylation gene expression PD regulatory role epigenome-wide probes on PD applied SMR approach DNA methylation PD gene pairwise associations between epigenome-wide CpG probes PD genes within 2 Mb distance analyses involved SNP data Brain-eMeta57 eQTL data 1194) GTEx CMC59 ROSMAP60 sample overlap MeCS Brain-mMeta mQTL57 data 1160) PD GWAS summary data = 308,518) PDWBS meta-analysis63 of 6476 PD cases 302,042 controlsSMR method included genes cis-eQTL PeQTL < 5 × 10−08 in Brain-eMeta 12 genes for analysis.DNA methylation-based out-of-sample classificationIn SGPD cohort estimated effect probes on disease status mixed linear model sex age smoking exposure CTPs best linear unbiased prediction solutions probe effects calculated DNA methylation profile scores European individuals PEG study logistic regression adjusted sex age CTPs estimated variance PD status (Nagelkerke R2) PEG study standardized methylation profile scores classification accuracy evaluated AUC.Reporting Nature Research Reporting Summary.Supplementary information 1 2 3 4 Peer Review File Reporting Summary Additional Supplementary Files | 49.9 | 0.900135 |
10.1038/s41467-020-16358-7 | PMC7270101 | Age-associated metabolic changes include lipid accumulation. Here, the authors show that with replicative aging yeast accumulate lipid droplets which protect cells from cold stress and can be modulated through Biosynthesis of NAD+ 2 (BNA2). | Age-dependent changes in metabolism can manifest as cellular lipid accumulation, but how this accumulation is regulated or impacts longevity is poorly understood. We find that Saccharomyces cerevisiae accumulate lipid droplets (LDs) during aging. We also find that over-expressing BNA2, the first Biosynthesis of NAD+ (kynurenine) pathway gene, reduces LD accumulation during aging and extends lifespan. Mechanistically, this LD accumulation during aging is not linked to NAD+ levels, but is anti-correlated with metabolites of the shikimate and aromatic amino acid biosynthesis (SA) pathways (upstream of BNA2), which produce tryptophan (the Bna2p substrate). We provide evidence that over-expressed BNA2 skews glycolytic flux from LDs towards the SA-BNA pathways, effectively reducing LDs. Importantly, we find that accumulation of LDs does not shorten lifespan, but does protect aged cells against stress. Our findings reveal how lipid accumulation impacts longevity, and how aging cell metabolism can be rewired to modulate lipid accumulation independently from longevity. | IntroductionAs most organisms age, metabolism slows, which can manifest as a gradual accumulation of neutral lipid (i.e., fat), an important energy source stored in lipid droplets (LDs). Aspects of metabolic decline during aging have been attributed to an age-associated decline in nicotinamide adenine dinucleotide (NAD+) levels1,2. Intense research has focused on interventions to increase cellular NAD+, such as with NAD+ precursors (e.g., niacin, nicotinamide mononucleotide, and nicotinamide riboside), and to mitigate age-associated changes in metabolism, including reducing fat. While such interventions have been shown to improve facets of metabolism and reduce fat, they do not always increase NAD+ or extend lifespan3–7. Recent studies present alternative mechanisms that modulate fat accumulation during aging. For example, it was shown that during aging in mice, the frequency of DNA double stranded breaks increases and this activates a muscle specific enzyme (DNA-PK) that is responsible for 40% of the weight gained by mice fed high-fat diets8. In aging flies, specific muscles accumulate LDs, and the overexpression of a cytosolic histone deacetylase (HDAC6) suppresses this accumulation9. Thus, many questions remain concerning how lipid accumulation is regulated during aging and specifically, whether this accumulation during aging reduces longevity.Here, we explore whether LDs change in Saccharomyces cerevisiae cells as they replicatively age, and find that LDs accumulate. We describe a new link between the biosynthesis of NAD+ (kynurenine) pathway and lipid droplets during aging. Specifically, our genetic and metabolomic approaches reveal that increasing the BNA pathway (by overexpressing BNA2, the first gene of the pathway) reduces LD accumulation during aging. This reduction is achieved by pulling metabolic flux, likely from glycolysis, through the shikimate and aromatic amino acid biosynthesis (SA) pathways, which are upstream of the BNA pathway and which synthesize and supply tryptophan, the major substrate of the BNA pathway. By using genetic approaches to dissect the role of the SA-BNA pathway on longevity, we find, surprisingly, that LD accumulation during aging can be modulated independently from longevity. Thus, LD accumulation during normal aging does not shorten lifespan. Rather, we find LD accumulation protects aging cells against cold stress. These findings reveal how lipid-droplet accumulation impacts longevity, and provide a new strategy for lessening lipid accumulation during aging independently from longevity.ResultsLipid droplets accumulate during yeast replicative agingReplicative aging in the budding yeast, Saccharomyces cerevisiae, is defined as the number of times an individual cell divides before death10. We previously showed that replicatively aged yeast cells exhibit characteristics of age-induced decline observed in metazoa, including reduced mitochondrial function and increased genomic instability11,12. In this study we screened cells for age-associated changes in organelles and vesicles12,13, and found that lipid droplets (LDs) accumulate (Fig. 1a). Quantification of LDs using BODIPY 493/503 revealed LDs increase 7.2-fold from median age 0 (young) to 23 (old) (Fig. 1c, d, Supplementary Fig. 1), which was corroborated biochemically (Fig. 1b).Fig. 1Age-associated lipid droplet accumulation is suppressed by BNA2 overexpression.a, c Replicative age below panels represents the median age of mother cells determined by budscar counting (see Methods). a Lipid droplets (LDs) in Control (WT, AB18-07) cells visualized by mCherry tagged Erg6p, a known yeast LD protein (magenta) and BODIPY 493/503 (neutral lipid stain, green) (black scale bar: 4 μm). Representative images from four independent experiments. b Total neutral lipid and phospholipid fractions extracted from young (black dots) and aged (blue dots) UCC492512, 13, 19 cells were analyzed by gas chromatography mass spectrometry (n = 6 independent experiments). Paired t-test (two-tailed): **p = 0.0065. c BODIPY stained LDs in Control (WT, AB18-07) and BNA2 overexpression (BNA2-OE) cells (black scale bar: 4 μm). Representative images were taken from cells aged and quantified in one experiment from (d) below. d LDs were stained with BODIPY 493/503 and quantified in Control (WT, AB18-07, black dots) and BNA2-OE (blue squares) cells by flow cytometry. The mean, normalized BODIPY intensities for each time point and strain were determined as in Supplementary Fig. 1 for each individual experiment, and these mean, normalized intensities were averaged for n = 13 independent experiments and plotted ±SEM. Two-way ANOVA multiple comparisons test: ****p ≤ 0.0001. p > 0.05 indicated as ns (not significant). e Replicative lifespan (RLS) determined using microdissection10. 25 Control cells (WT, AB18-07, black line, median lifespan 25, maximum lifespan 55) and 39 BNA2-OE cells (blue line, median 34, maximum 66) analyzed in n = 1 independent experiment. Log-rank test: p = 0.0116. f RLS determined using the Mother Enrichment Program19 (see Methods). 2300 Control cells (WT, AB18-07, black line) and 2450 BNA2-OE cells (blue line) analyzed over n = 16 independent experiments. Average median and maximum lifespan (±SEM): Control (25.2 ± 0.3, 41.6 ± 0.6), BNA2-OE (28.9 ± 0.5, 49.9 ± 1.3). Log-rank test: ****p ≤ 0.0001. g, h Cellular NAD+ and NADH extracted from Control (WT, AB18-07, black dots) and BNA2-OE (blue squares) cells were quantified by absorbance at 450 nm, averaged, and plotted ±SEM (n = 5 independent experiments). Source data for (b), (d–h) provided as a Source data file.A link between the biosynthesis of NAD+ and lipid dropletsTo gain insights into how this LD accumulation might be regulated, we performed a limited over-expression screen for reduced LDs that included genes previously associated with yeast glycerolipid and LD dynamics (see Supplementary Data 1)14–17. We identified BNA2, which was not previously associated with regulation of yeast LDs, and encodes an indoleamine 2,3-dioxygenase that utilizes tryptophan in the first step of the biosynthesis of NAD+ (or kynurenine) pathway18. We made homozygous, BNA2 overexpression diploid cells (BNA2-OE cells), which we compared with diploid WT (control) cells throughout this study. In young BNA2-OE and control cells, LD levels were similar, whereas in old BNA2-OE cells, LD levels were significantly decreased: a 40% reduction compared with control cells not over-expressing BNA2 (Fig. 1c, d, Supplementary Fig. 1), indicating BNA2-OE reduces LD levels in aged cells. To determine whether BNA2-OE impacts replicative lifespan, we used complementary approaches [microdissection10 and a new protocol applying the Mother Enrichment Program19 (see Methods)], and found in each case that BNA2-OE cells exhibit significant increases of lifespan (>15%) (Fig. 1e, f, Supplementary Fig. 2a). These data are consistent with the simple hypothesis that reducing LD accumulation during aging promotes longevity.Given Bna2p’s role, it was anticipated that BNA2-OE would affect NAD+ levels; however, we observed no significant difference in NAD+ levels between BNA2-OE and control cells (Figs. 1g–h, 4f). To determine how the BNA pathway affects LD accumulation and longevity, we made homozygous deletions in core BNA pathway genes that act “downstream” of BNA2 (BNA1, BNA2, BNA5, BNA6, BNA7) and a branch point gene (BNA3) (Figs. 2a, 3a)18,20. Deleting core BNA pathway genes significantly reduced lifespan (Fig. 2h, j–l, Supplementary Figs. 2b,d,e, 3b), whereas deleting BNA3 had no effect (Fig. 2i, Supplementary Fig. 2c). These data indicated that the core BNA pathway, but not the BNA3 branch point, was important for longevity. However, even though deleting core BNA pathway genes reduced lifespan (Fig. 2h, j–l, Supplementary Figs. 2b,d,e, 3b), these deletions did not affect the normal accumulation of LDs during aging (Fig. 2b, d–f, Supplementary Fig. 3a), which is inconsistent with the simple hypothesis noted above, but rather suggests that LD accumulation during aging does not impact lifespan.Fig. 2Lipid droplet accumulation is separable from longevity.a Shikimate, aromatic amino acid, BNA (SA-BNA) pathways. b–g LD quantification by flow cytometry (±SEM, see Supplementary Fig. 1). Two-way ANOVA multiple comparisons: p > 0.05, not significant; p ≤ 0.05 indicated in figure panels; ****p ≤ 0.0001. All strains: Control (WT, AB18-07), black dots. BNA2-OE, blue squares. Deletions, red diamonds. BNA2-OE + deletions, orange triangles. n=number of independent experiments. b
n = 5. c
n = 4. d
n = 5. e
n = 4. f
n = 4. g
n = 4. h–m RLS analysis (see Fig. 1f, Methods). For each panel n = number of independent experiments. Strain: total number of cells analyzed over n; average median RLS ± SEM; average maximal RLS ± SEM. Line colors: Control (WT, AB18-07), black; BNA2-OE, blue; deletions, red; BNA2-OE + deletion, orange. Log-rank tests: p > 0.05, not significant); p ≤ 0.05 indicated in figure panels; ****p ≤ 0.0001. h
n = 3. Control: 400; 24 ± 1.2; 39.3 ± 0.7. BNA2-OE: 400; 27.5 ± 0.8; 47.7 ± 2.2. bna1Δ: 400; 19 ± 2.3; 38 ± 1.7. BNA2-OE bna1Δ: 400; 25.6 ± 1.8; 42.6 ± 5.2. i
n = 3. Control: 350; 24 ± 1.2; 38 ± 1.2. BNA2-OE: 350; 27.2 ± 0.9; 45 ± 0.6. bna3Δ: 350; 23.7 ± 0.9; 38.7 ± 2.7. BNA2-OE bna3Δ: 350; 27 ± 0.6; 47.3 ± 4.4. j
n = 3. Control: 475; 25.3 ± 0.3; 46.7 ± 4.7. BNA2-OE: 475; 30 ± 0.6; 49 ± 1.2. bna5Δ: 475; 21.7 ± 0.6; 36.7 ± 0.7. BNA2-OE bna5Δ: 475; 25.3 ± 0.3; 47 ± 4.4. k
n = 5. Control: 775; 24.6 ± 0.2; 42.4 ± 0.8. BNA2-OE: 925; 28.6 ± 0.5; 53.2 ± 2.1. bna6Δ: 700; 21.4 ± 0.5; 37.8 ± 1.4. BNA2-OE bna6Δ: 725; 22 ± 0.5; 39 ± 2.1. l
n = 3. Control: 325; 22 ± 1; 38.7 ± 1.7. BNA2-OE: 325; 28.7 ± 0.3; 49.7 ± 0.3. bna7Δ: 325; 17 ± 0.6; 30.7 ± 1.2. BNA2-OE bna7Δ: 325; 15.3 ± 0.9; 33 ± 2.5. m
n = 5. Control: 725; 24.6 ± 0.5; 41.8 ± 1.1. BNA2-OE: 725; 28.6 ± 0.8; 48.6 ± 2.9. aro1Δ: 775; 17.2 ± 2.0; 41.2 ± 2.6. BNA2-OE aro1Δ: 750; 25.0 ± 1.3, 48.2 ± 1.7. Source data for (b–m) provided as a Source data file.Fig. 3SA pathway metabolite levels anti-correlate with lipid droplet levels during aging.a The SA-BNA pathway, including metabolite names, genes, and sources of crosstalk (purple dashed arrows) between the aromatic amino acid synthesis and BNA pathways. b–i Ratios of metabolites from glycolysis, the shikimate and aromatic amino acid pathways, and the BNA pathway are plotted black, red, or blue, respectively. b Middle-aged Control (WT, AB18-07) and BNA2-OE cells were analyzed by global metabolomics (see Methods). Measurements of specific metabolites (arbitrary units) detected from n = 3 independent experiments were normalized to cell counts within each experiment. The normalized measurements from BNA2-OE cells were divided by the normalized measurements from Control (WT) cells for each experiment, and these ratios were then averaged and plotted as a ratio of metabolite levels in BNA2-OE versus control cells ± SEM (see Supplementary Fig. 4 and source data). c–i Cells as indicated were analyzed by targeted metabolomics (see Methods and source data). Measurements of specific metabolites detected in the shikimate (red symbols) or BNA (blue symbols) pathways in Control (WT, AB18-07, dots), BNA2-OE (squares), BNA2-OE aro1 (triangles), and BNA2-OE bna6 (diamonds) cells were normalized to cell counts, averaged from n = 4 independent experiments and reported as arbitrary units ± SEM. #Below limit of detection. Paired t-test (two-tailed): p ≤ 0.05 indicated in figure panels; ****p ≤ 0.0001; p > 0.05 indicated as ns (not significant). j Model for the mechanism by which BNA2-OE suppresses the accumulation of lipid droplets during aging and extends lifespan. Aged Control (WT) cells accumulate lipid droplets during normal aging. BNA2 overexpression (BNA2-OE) decreases lipid-droplet accumulation by pulling some substrates away from lipid droplets through the shikimate and aromatic amino acid (SA) pathways toward the BNA pathway. A deletion in ARO1 prevents BNA2-OE from pulling flux through the SA pathway and from suppressing lipid-droplet accumulation during aging, whereas lifespan extension by BNA2-OE correlates with increasing the branch point metabolite xanthurenate. Source data for (b–i) are provided as a Source data file.Lipid droplet accumulation is separable from longevityWe investigated whether BNA pathway genes were required for the ability of BNA2-OE to reduce LD accumulation during aging. Strikingly, cells with overexpressed BNA2 still displayed reduced LD accumulation during aging regardless of whether BNA pathway core or branch point genes were deleted (bna1Δ, bna3Δ, bna5Δ, bna6Δ, or bna7Δ) (Fig. 2b–f). These results suggested that the flux of metabolites from Bna2p toward NAD+ synthesis may not be important for suppressing the age-associated accumulation of LDs by BNA2-OE.We next investigated the dependency of the BNA pathway on longevity in BNA2-OE cells. Cells with BNA2-OE combined with either bna1Δ, bna3Δ, or bna5Δ deletions had significantly increased lifespan compared with the cells with only the respective BNA pathway genes deleted, i.e., bna1Δ, bna3Δ, or bna5Δ (Fig. 2h–j, Supplementary Fig. 2b–d). Interestingly, the increase in lifespan of BNA2-OE cells was to the same degree (above WT (control) cells) with or without BNA3 (Fig. 2i), but the increase in lifespan from BNA2-OE only achieved control levels when either BNA1 or BNA5 were eliminated (Fig. 2h, j, Supplementary Fig. 2b,d). In further contrast, deleting BNA6 or BNA7 blocked the ability of BNA2-OE to extend lifespan beyond the shortened lifespan of bna6Δ and bna7Δ cells, respectively (Fig. 2k–l, Supplementary Fig. 2e). Thus, BNA2-OE can extend lifespan without BNA1, BNA3, or BNA5 but not without BNA6 or BNA7. These data indicate that the BNA pathway is linked to lifespan extension, but is not necessary for suppressing LD accumulation during aging, by BNA2-OE. Taken together, our results indicate that LD accumulation and longevity are genetically separable and can be modulated independently.Since BNA pathway genes downstream of BNA2 were not required for the suppression of age-associated LD accumulation by BNA2-OE, we examined whether the shikimate and aromatic amino acid synthesis (SA) pathways upstream of BNA2 were critical for this phenotype. Substrates for the shikimate pathway are supplied by glycolysis (phosphoenolpyruvate) and the pentose phosphate pathway (erythrose-4-phosphate)21,22. These substrates are converted into chorismate, the precursor of the aromatic amino acids phenylalanine, tyrosine, and the Bna2p substrate, tryptophan (Figs. 2a, 3a). There is also crosstalk between the aromatic amino acid synthesis and BNA pathways, through a shared intermediate, anthranilate23, and a gene, ARO924 (Fig. 3a, purple). ARO1 is a gene upstream of aromatic amino acid synthesis that encodes the rate limiting enzyme of the shikimate pathway (Fig. 3a). When ARO1 was deleted, this blocked the ability of BNA2-OE to suppress LD accumulation during aging (Fig. 2g), indicating that ARO1 is essential for BNA2-OE to suppress LDs in aging cells. With respect to lifespan, deleting ARO1 was more nuanced. In the absence of ARO1, cells had a dramatically reduced median lifespan compared with WT (control) cells (Fig. 2m). Yet BNA2-OE increased this shortened lifespan of aro1Δ cells to longer than WT lifespan levels, albeit not as long as BNA2-OE alone (Fig. 2m, Supplementary Fig. 2f). Taken together, these results suggested that BNA2-OE reduced LD accumulation during aging by pulling upstream substrates away from LDs through the upstream SA pathway, and extended lifespan by increasing flux downstream toward the BNA pathway (Fig. 3j).Metabolic rewiring reduces lipid droplet levels in old cellsTo test this idea about how BNA2-OE impacts LD accumulation, metabolites within the SA-BNA pathway (Fig. 3a) were examined in some of the aforementioned genetically altered cells (see Supplementary Figure 4 and Source Data file). Although BNA2-OE minimally affected phosphenolpyruvate levels (Fig. 3b), BNA2-OE dramatically increased SA pathway metabolite levels, including shikimate, shikimate-3-phosphate, and chorismate (by ~16 to ~60-fold) (Fig. 3b, c), and to a lesser extent tyrosine, phenylalanine, and tryptophan levels (by ~4-fold, ~1.9-fold, and ~47%, respectively) (Fig. 3b, e). Chorismate and anthranilate were also increased markedly in BNA2-OE bna6Δ cells (Fig. 3c,d), which have reduced levels of LDs (Fig. 2e); however, these metabolites were at low or background levels in BNA2-OE aro1Δ cells (Fig. 3c,d), which have high levels of LDs (Fig. 2g). Thus, BNA2-OE requires ARO1 to increase SA pathway metabolites, and importantly, SA pathway metabolite levels inversely correlate with LD accumulation during aging.In the early steps of the core BNA pathway, BNA2-OE greatly increased kynurenine and 3-hydroxykynurenine (3-HK) levels (by >170-fold) (Fig. 3b, f, h). These metabolites were also detected at high levels in BNA2-OE bna6Δ and BNA2-OE aro1Δ cells (Fig. 3f, h), though there was less kynurenine (~20-fold less) and 3-HK (~2.7-fold less) in BNA2-OE aro1Δ than BNA2-OE cells (Fig. 3f, h), consistent with kynurenine and 3-HK being derived from exogenous tryptophan in the media. Importantly, these results are consistent with BNA2-OE increasing core BNA pathway metabolite levels (through 3-HK) by pulling substrates through the upstream SA pathway. Notably, BNA2-OE minimally altered metabolite levels at the last steps of the core BNA pathway [3-hydroxyanthranilate (3-HA, increased ~2.9-fold) (Fig. 3b) and quinolinic acid (minimal change) (Fig. 3b)] and NAD+ (minimal change) (Figs. 1e, f, 3i). Thus, BNA2-OE greatly increases core BNA pathway metabolite levels through the 3-HK step of the pathway, but has much less effect on the steps that proceed thereafter in the production of NAD+. Thus, our results suggest that through 3-HK, the metabolic flux can be diverted into BNA branch points and not necessarily toward NAD+.Consistent with this idea, BNA2-OE induced changes also increased branch point metabolites associated with kynurenine and 3-HK (Fig. 3a). Specifically, BNA2-OE increased kynurenic acid and xanthurenate ~350- and 49-fold, respectively (Fig. 3b). Notably, kynurenic acid was also high in BNA2-OE bna6Δ cells (Fig. 3g), but not in BNA2-OE aro1Δ cells (Fig. 3g). This latter finding is consistent with increased branch point metabolite levels requiring increased levels of tryptophan, formylkynurenine, or kynurenine and/or crosstalk between the aromatic amino acid synthesis and BNA pathways (Fig. 3a, purple)23,24. Taken together, these data strongly support the model that BNA2-OE reduces LD accumulation during aging by pulling substrates through the SA pathway toward the BNA pathway and away from LDs (Fig. 3j).Lipid droplets may protect aging cells against stressLDs have been implicated in offering protection from several types of cellular stresses25. This led us to explore whether the accumulation of LDs in aging cells provides any benefit to old cells. This idea was tested with a cold-shock regimen: cell viability was determined in middle-aged cells after exposure to cold (4 °C). Interestingly, cold exposure slightly increased the lifespan of WT (control) cells (Fig. 4a, Supplementary Fig. 3g), and minimally affected BNA2-OE aro1Δ cells (Fig. 4b, Supplementary Fig. 3h). However, cold exposure significantly decreased the lifespan of BNA2-OE cells (Fig. 4c, Supplementary Fig. 3i). Because aged control and BNA2-OE aro1Δ cells accumulate more LDs, and aged BNA2-OE cells accumulate fewer LDs (Fig. 2g), we speculate that LD accumulation during aging offers a level of protection against stress.Fig. 4Increased survival to cold exposure correlates with LD accumulation during aging.a–c Cells as indicated were aged in 30 °C media until median age 16 (middle-age), and then exposed to 4 °C (blue lines) or 30 °C (black lines) media at regular intervals until death (see Methods), and lifespan was analyzed as in Fig. 1f (see Methods). Log-rank test: p > 0.05, not significant; ****p ≤ 0.0001. a Control (WT, AB18-07, 550 cells at 30 °C, and 525 cells at 4 °C), b
BNA2-OE aro1Δ (525 cells at 30 and 4 o C), and c
BNA2-OE (550 cells at 30 and 4 °C) were analyzed over n = 4 independent experiments. Average median and maximal lifespans at 30 °C (±SEM): Control 25.8 ± 1.3, 41 ± 1.9; BNA2-OE aro1Δ 25.1 ± 2.2, 44.3 ± 4.2; BNA2-OE 29.5 ± 1.7, 48.8 ± 2.5. Average median and maximal lifespans at 4 °C (+SEM): Control 27.5 ± 1.3, 43.3 ± 2.5; BNA2-OE aro1Δ 25 ± 1.8, 45 ± 4.6; and BNA2-OE 24.8 ± 1.3, 40 ± 3.7. Source data for (a–c) are provided as a Source data file.DiscussionHigh-fat diets contribute to fat accumulation and disease (e.g., obesity, cancer)26, but how fat accumulated during normal aging impacts health and longevity is less understood. The present study shows that yeast normally accumulate fat in the form of LDs during aging, similar to metazoans8,9. We find that LD accumulation during aging does not simply correlate with longevity (Fig. 2), but does associate with longevity under stress (Fig. 4). Similar context specific correlations between LDs and aging have been observed in Drosophila and mice27. This leaves open the possibility that neutral lipid accumulation provides a selective advantage to aging cells experiencing variable environmental conditions.We provide evidence that aging cell metabolism can be rewired by BNA2 overexpression to increase flux through the SA pathway and thus suppresses LD accumulation during aging. In addition to the pathways reported here, a wider network of genes and reactions are linked to yeast lipid metabolism28. It remains to be determined how this network may become skewed toward neutral lipid synthesis and LDs during aging. However, our findings may provide insight: the substrates of the shikimate pathway are supplied by glycolysis and the pentose phosphate pathway. Glycolysis also supplies the pentose phosphate pathway, and importantly, acetyl-CoA (via pyruvate), a precursor of all lipids. Furthermore, altering glucose levels is known to alter the lipid profile of yeast29–31. Thus, we propose that as cells age, glycolytic flux is skewed toward neutral lipid and LDs—presumably through pyruvate and acetyl-CoA (Fig. 3j). By contrast, in cells overexpressing BNA2, glycolytic flux is skewed from pyruvate, acetyl-CoA, and neutral lipid toward the SA-BNA pathway, which would reduce the level of LD accumulation during aging (Fig. 3j).We show here that BNA2-OE increases kynurenic acid, but how exactly this metabolite suppresses neutral lipid accumulation in aging cells remains to be determined. In this light, it is noteworthy that links have been described between kynurenine metabolites (e.g., kynurenic acid) and lipid metabolism in mammalian cells32–34. In particular, the regulation of IDO1 (the homolog of BNA2) activity and its expression is under complex regulation in metazoan cells34–37. In certain cancers, IDO1 is overexpressed, and is thought to promote cancer progression in part by decreasing exogenous levels of tryptophan, which serves as an immune detection signal34,38. As shown here in yeast, increased kynurenine branch point metabolites (e.g., xanthurenate) correlate with replicative lifespan extension (i.e., increased cell growth and proliferation), and may provide further insights about how IDO1 overexpression promotes cell growth and proliferation and, in turn, cancer progression.MethodsStrainsAll yeast strains used are derivatives of S. cerevisiae S288C39 and UCC877319 and listed in Supplemental Table 1. Gene deletion strains were created by one step PCR-mediated gene replacement39,40. Plasmid templates for gene deletion construction were from the pRS400 vector39. Oligonucleotides for gene replacement and mCherry tagging are listed in Supplementary Table 2. Transformation and insertion of NotI-digested plasmid pAG306-GPD-xsome1 or pAG306-GPD-BNA2-xsome1 was done to create overexpression strains12. The parental genotype for all BNA2 overexpression and deletion strains is MATa/MATα
his3△1/his3△1 leu2△0/leu2△0 ura3△0/ura3△0 lys2△0/+ ho△::SCW11pr-Cre-EBD78-NATMX/ho△::SCW11pr-Cre-EBD78-NATMX loxP-CDC20-Intron-loxP-HPHMX/loxP-CDC20-Intron-loxP-HPHMX loxP-UBC9-loxP-LEU2/loxP-UBC9-loxP-LEU2 Gene Deletion/Gene Deletion GPD-(BNA2)-URA3-chr1/GPD-(BNA2)-URA3-chr1. Strains expressing in-frame, C-terminal chromosomal fusions of ERG6 with mCherry were made by transformation of parental yeast strains with pKT127-mCherry12.Aging cell lipid-droplet suppressor screenMother enrichment program cells were individually transformed in 96-well format with 280 high-copy 2-μm plasmids from a tiled genomic DNA library as previously described12. Each plasmid contained a unique sequence-verified genomic DNA fragment that expressed at least one gene previously determined to affect lipid-droplet morphology or glycerolipid metabolism14–17 and additional genes that were not previously determined to affect these processes (see Supplementary data). To identify genes that suppressed age-induced accumulation of lipid droplets, all plasmid-expressing strains were individually grown in 2 mL of yeast extract peptone 2% glucose (YEPD) and kanamycin at 30 °C as done previously12. Cells were stained with BODIPY 493/503 (ThermoFisher #D3922) or Nile Red (Sigma #N3013), and imaged by fluorescence microscopy. Plasmid containing strains in which 50% of middle-aged cells exhibited young-cell-like levels of lipid droplets were scored as suppressors of age-induced lipid-droplet accumulation. To confirm potential suppressors (e.g., BNA2), candidate genes were transferred using LR clonase from pDONR221 Gateway entry plasmids from a previously described collection (HIP) into pAG306-GPD-ccdB-xsome112,41. Plasmids were integrated into chromosome 1 of control strains after NotI digestion, and strains were aged, purified, and examined by fluorescence microscopy as above (see Supplementary data 1). The integrated suppressor (BNA2) was confirmed by the ability to maintain young-cell-like levels of lipid droplets in at least 50% of middle-aged cells.Enrichment of aging cells from cultures using magnetic beadsCells stored in 15% (w/v) glycerol at −70 °C were struck onto YEPD agar and grown at RT < 1 day, and colonies were used to inoculate 2 mL YEPD in 16 × 150 mm culture tubes that were grown 27.5 h at 30 °C, 40 rpm (New Brunswick model TC-7 rotor). Saturated cultures were diluted 1:250,000 in 100 mL YEPD in 250-mL flasks, and grown ~15.5 h at 30 °C, 180 rpm (shaker) to ≤2.5 × 106 cells/mL. Lyophilized NHS-activated supraparamagnetic 50-nm beads (Ocean Nanotech; SN00051) were prepared according to manufacturer’s instructions, then suspended to 2.5 μg beads/mL PBS (final). 30 × 106 cells were centrifuged initially at 1000 × g, 5 min, room temperature (RT), then transferred to microcentrifuge tubes using 1 mL PBS + 2% glucose (PBS-G), and centrifuged 1500 × g, 1 min, RT. The supernatant was aspirated and cells were washed again in PBS-G. After a second wash, centrifugation, and aspiration, 600 μL 50% (v/v) PEG-3350, 400 μL PBS-G, and 10 μL bead slurry (in this order) was gently layered onto the cell pellets, which prevents the beads from damaging the cells. Cells were suspended gently by pipetting until homogenous, and incubated 10 min, with rotation, at RT. Labeled cells were centrifuged 1500 × g, 1 min, RT, the supernatant was aspirated, then the cell pellets were rinsed gently with 1 mL 30% (w/v) PEG-3350 (final) in YEPD. Cells were centrifuged as above, supernatants were aspirated, and cells were suspended in 30 °C YEPD. For aging of cells, 1 L cultures of 30 °C YEPD in 4-L flasks were inoculated with 8 × 106 labeled cells, and incubated on a shaker at 30 °C, 95 rpm, whereas the remaining labeled cell suspension was centrifuged as above, fixed in 4% paraformaldehyde in PBS (10 min at RT), and saved in PBS at 4 °C as the starting culture (median age 0 cells). At 11, 23, and 34 h after inoculation, labeled, aged cells were recovered from the aged cell cultures by centrifugation of the cultures (1000 × g, 5 min), then the pelleted cells were resuspended in 5 mL 30 °C YEPD per culture, and the cell suspension was applied to LS MACS columns (Miltenyi Biotec, 130-042-401) on magnets (columns were pre-equilibrated on the magnet using 1 column volume of 30 °C YEPD). The columns were allowed to drain by gravity flow, and were washed with three column volumes of 30 °C YEPD. Columns were then taken off the magnets, and the aged cells were eluted and collected in 30 °C YEPD. For microscopy and flow cytometry analyses, 1–2 × 106 of eluted cells were fixed in 4% paraformaldehyde and saved in PBS as above for later analyses, whereas the remaining eluted cell suspension was used to inoculate 1.5 L 30 °C YEPD to grow from 11 to 23 h, or to inoculate 0.5 L 30 °C YEPD to grow from 23 to 34 h. For NAD+/NADH and metabolomics analyses, eluted cells were instead processed as described below.NAD+/NADH analysesYoung and aged cells enriched as above were washed with ice-cold Yeast Nitrogen Base +2% glucose (final) (centrifugation at 1500 × g, 1 min, 4 °C). NAD+/NADH was analyzed using the NAD/NADH quantitation kit according to the manufacturer’s instructions (Sigma-Aldrich, MAK037). Briefly, cells were suspended in extraction buffer, subjected to two freeze thaw cycles on dry ice, and stored at −70 °C or immediately cleared by centrifugation (14,000 × g, 5 min, 4 °C). Supernatants applied to Amicon regenerated cellulose 10,000 NMWL centrifugal filters (Millipore #MRCPRT010) were centrifuged (15,000 × g, 1.5 h, at 4 °C), and the flow-through was analyzed per manufacturer’s instructions [450 nm absorbance measured using a Powerwave XS (Biotek, Winooski, VT)].Global and targeted metabolomicsAged cells were washed with ice-cold yeast nitrogenous base + 2% glucose (final) as above, then suspended in 80:20 methanol:water at −20 °C, vortexed, and incubated 3 min at −20 °C. Ammonium bicarbonate pH 6.8 was added to 25 mM final concentration to samples, then incubated at −20 °C for 20 min. Extracts were cleared using Amicon regenerated cellulose centrifugal 10,000 NMWL centrifugal filters (Millipore #MRCPRT010) at 15,000 × g, 1.5 h, at −8 °C. Flow-through was kept at −70 °C until analysis. Samples were dried down under nitrogen gas.Global mass spectrometry analysis using LC-MS was performed at Calico Life Sciences (Fig. 3b, Supplementary Fig. 4, Source data file). Samples were resuspended in water for analysis on a C18 column in negative ion mode, and diluted 1:4 into acetonitrile for analysis on a HILIC column in positive ion mode. All samples were analyzed using two separate LC-MS methods on Vanquish UPLCs coupled to a Q-Exactive Plus mass spectrometers running Thermo Xcalibur 4.0 (version 4.0.27.10) with Thermo Foundation 3.1 SP1 (version 3.1.83.0) (ThermoFisher Scientific, Waltham, MA).Metabolites analyzed in positive ionization mode were separated using a SeQuant® ZIC®-pHILIC column, 5 μm particle size, 200 Å, 150 × 2.1 mm. Mobile phase A was 20 mM ammonium carbonate in water (pH 9.2); mobile phase B was acetonitrile. The flow rate was 150 μL/min and the gradient was t = −6, 80% B; t = 0, 80% B; t = 2.5, 73% B; t = 5, 65% B, t = 7.5, 57% B; t = 10, 50% B; t = 15, 35% B; t = 20; 20% B; t = 22, 15% B; t = 22.5, 80% B; t = 24; 80% B. The mass spectrometer was operated in positive ion mode using data-dependent acquisition (DDA) mode with the following parameters: resolution = 70,000, AGC target = 3.00E + 06, maximum IT (ms) = 100, scan range = 70–1050. The MS2 parameters were as follows: resolution = 17,500, AGC target = 1.00E + 05, maximum IT (ms) = 50, loop count = 6, isolation window (m/z) = 1, (N)CE = 20, 40, 80; underfill ratio = 1.00%, Apex trigger(s) = 3–10, dynamic exclusion(s) = 25.Metabolites analyzed in negative ionization mode were separated using a reverse phase ion-pairing chromatographic method with an Agilent Extend C18 RRHD column, 1.8 μm particle size, 80 Å, 2.1 × 150 mm. Mobile phase A was 10 mM tributylamine, 15 mM acetic acid in 97:3 water:methanol pH 4.95; mobile phase B was methanol. The flow rate was 200 μL/min and the gradient was t = −4, 0% B; t = 0, 0% B; t = 5; 20% B; t = 7.5, 20% B; t = 13, 55% B; t = 15, 95% B; t = 18.5, 95% B; t = 19, 0% B; t = 22, 0% B. The mass spectrometer was operated in negative ion mode using data-dependent acquisition (DDA) mode with the following parameters: resolution = 70,000, AGC target = 1.00E + 06, maximum IT (ms) = 100, scan range = 70–1050. The MS2 parameters were as follows: resolution = 17,500, AGC target = 1.00E + 05, maximum IT (ms) = 50, loop count = 6, isolation window (m/z) = 1, (N)CE = 20, 50, 100; underfill ratio = 1.00%, Apex trigger(s) = 3–12, dynamic exclusion(s) = 20. Metabolites were identified by matching fragmentation spectra and retention times from chemical standards that were previously analyzed on the same instrumentation. Global metabolomics data RAW files were converted to mzXML files using msconvert from the open source software package ProteoWizard, version 3.0.8789 (http://proteowizard.sourceforge.net/); identity and peak integration were manually verified and quantified using the open source software program Maven, version 8.0.2 (https://github.com/eugenemel/maven)42. Quantified measurements were analyzed and plotted using Microsoft Excel v16.34 and Graphpad Prism v8.3.1.Targeted mass spectrometry analysis was performed at the University of Washington Department of Medicinal Chemistry Mass Spectrometry Center (Fig. 3c–i, Source data file). Samples were reconstituted in 50:50 (v/v) mixture of 100% methanol and 0.4% (v/v) acetic acid in water, and injected into a 1290 Agilent UPLC system coupled to an Agilent 6520 Quadrupole Time of Flight (Q-TOF). Features were extracted and analyzed using Agilent MassHunter Data Acquisition and Processing software, and the measurements above the limit of detection are reported as averages ± SEM from four independent experiments in each figure panel. The measurements (volume in arbitrary units) of each feature were normalized based on the number of cells per sample compared with the number of cells in control (WT) strain, and then analyzed and plotted using Microsoft Excel v16.34 and Graphpad Prism v8.3.1. Standards of each reported metabolite were run as a comparison to detect features in experimental samples.Fluorescence microscopyAll images were captured on a Leica Microsystems (Buffalo Grove, IL) DMI6000 running LAS X (Version 3.3.3.16958) equipped with a Leica DFC365 FX camera, GFP (excitation 470/50, emission 525/50), Texas Red 2 (excitation 560/40, emission 645/76), and QAS (filterset: DFTC excitation 350/50, 490/20, 555/26, 645/30, and emission 455/50, 525/36, 605/52, 705/72) filters, and a HC PL APO ×63 oil objective. Fluorescent images were analyzed using Adobe Photoshop CC 2019 v20.0.8.Lipid purification and analysisUCC4925 cells were grown on YEPD agar and cultured to saturation and then to log phase (<2.0 × 106 cells/mL) as described above. 20 × 106 cells were washed twice in PBS, and then incubated in 3 mg/mL (w/v) EZ-Link Sulfo-NHS-LC-Biotin (Thermo Scientific, 21335) in PBS (30 min, RT). Cells were pelleted, supernatant was aspirated, and the labeled cells were suspended in 500 mL YEPD. Cultures were incubated at 30 °C in a shaker (95 rpm). At 10.5 h, cell culture density was determined. In total, 20 × 106 log cells were harvested (centrifugation 1000 × g, 5 min, RT), supernatant was removed by aspiration, and cell pellets were flash-frozen in liquid nitrogen (stored at −70 °C) (median age 0 cells). The remaining culture was centrifuged (1000 × g, 5 min, 4 °C), and cell pellets were washed twice with ice- cold PBS, and suspended to 2 × 108 cells/mL in ice-cold PBS; 1/20th volume of azide-free Multimacs Streptavidin Microbeads (Miltenyi Biotec, 130-092-954) was added, and cells were incubated for 30 min at 4 °C with rotation. Cells were washed by centrifugation and PBS at 4 °C as above. Cells were suspended in ice-cold PBS, and suspensions were applied to LS MACS columns (Miltenyi Biotec, 130-042-401) (pre-equilibrated using 5-mL ice-cold PBS) on magnets. The columns were drained by gravity flow, and washed twice with 1 volume of ice-cold PBS. To elute NHS-biotin-labeled cells (original mother cells), columns were removed from the magnet and cells were eluted with 4-mL ice-cold PBS. Cells were pelleted, supernatant was aspirated, and cells were flash-frozen in liquid nitrogen (stored at −70 °C) (median age 8 cells).Phospholipid (PL) and neutral lipid (NL) purification was performed as described43. Total lipids were extracted with 2:1 chloroform:methanol for 1 h at room temperature. Next, 0.2 volumes of 0.9% NaCl were added and the mixture was vortexed and allowed to separate for 2 min. The top layer was removed and the bottom layer was dried down under nitrogen. Dried lipids were resuspended in 1 mL of chloroform. Calibrated phospholipid standard (1,2-Dihepatdecanoyl-sn-Glycero-3-Phosphocholine, Avanti Polar Lipids) and TAG standard (tritridecanoin, Nu-Chek Prep) were added to the total lipid mixture before extraction. PLs and NLs were purified by solid-phase exchange (SPE) chromatography. Extracted lipid was resuspended in chloroform and loaded onto SPE columns (100 mg capacity, Fisher Scientific) pre-equilibrated with 3 mL of chloroform. TAGs were eluted first with 3 ml of chloroform. Glycosphingolipids were eluted next with 5 mL of a 9:1 acetone:methanol mixture, and phospholipids were eluted last with 3 mL of methanol. Purified lipids were dried, resuspended in methanol/2.5% H2SO4, and incubated for 1 h at 80 °C to create FAMEs. FAMEs were analyzed by gas chromatography/mass spectrometry (GC/MS) (Agilent 5975GC, 6920MS) and data were collected using Agilent MSD Chemstation and analyzed using Agilent ChemStation Integrator. To quantify NL and PL yields, total PL and NL were compared with the internally added standards. Data are presented as NL:PL ratio, which was determined by measuring the sum of all fatty acids found in NL fractions versus the sum of all fatty acids found in PL fractions, and measurements were analyzed and plotted using Microsoft Excel v16.34 and Graphpad Prism v8.3.1.Flow cytometryCells were stained with BODIPY 493/503 (ThermoFisher #D3922), Wheat Germ Agglutinin Alexa Fluor 647 Conjugate (ThermoFisher #W32466), and SYTOX Blue Dead Cell Stain (ThermoFisher #S34857), and signals were quantified using a Becton Dickinson (BD) FACSCanto II running BD FACSDiva 6.1.3 software with an excitation wavelength of 488 nm and an emission wavelength of 530 nm (FITC, Bodipy), excitation 633 nm and an emission wavelength of 660 nm (APC, WGA-647), and excitation 405 nm and an emission wavelength of 450 nm (PacificBlue, Sytox Blue). The gating used in analyses was defined to include live cells based on the absence of Sytox Blue Staining, presence of WGA-647 staining in aged samples, and excluded particles that were either too small or too large to be living yeast cells, based on the side scatter (SSC-A) versus forward scatter (FSC-A) plots (see Supplementary Fig. 1). The mean BODIPY fluorescence was determined from 10,000 cells, per time point, per experiment. Data were analyzed with the FlowJo v10 software (Becton, Dickinson & Company, Franklin Lakes, NJ). For each experiment, all BODIPY measurements within an individual experiment were normalized to the average of all BODIPY measurements of median age 0 control cells (see Supplemental Fig. 1 for an example analysis of one experiment), and collected data were normalized and plotted using Microsoft Excel v16.34 and GraphPad Prism v8.3.1 (San Diego, CA).Replicative lifespan analysesTraditional replicative lifespan analysis has been performed using microdissection as previously described10. This method was used to distinguish between the lifespans of control (AB18-07) and BNA2-OE (AB18-11) cells (Fig. 1e).Lifespans were also determined using a modification to the Mother Enrichment Program (MEP), which can distinguish between lifespans19 (Fig. 1f). The following modifications were made in order to generate data with high statistical power and collect data on hundreds of cells per strain: Strains were grown on YEPD agar at RT < 24 h, then used to inoculate 2 mL of YEPD, and grown individually and exponentially for 27.5 h at 30 °C on a rotor (40 rpm as above), diluted 1:250,000, and then grown exponentially in 25 mL of YEPD (125-mL flasks) for 15.5 h to a maximum density of ~2 × 106 cells/mL. 2.0 × 106 cells were washed with PBS-GN (PBS with 2% glucose and 1 mM nicotinic acid final), and labeled with 5 μg/mL (w/v) DyLight 680 NHS-Ester (Thermo Scientific; #46419) in PBS-GN for 5 min at RT. Labeled cells were washed twice with 30 °C YEPD, and 1.0 × 106 labeled cells were used to inoculate 4.5 mL 30 °C YEPD, and grown for 2 h on a rotor (40 rpm as above). Fifteen microliters of these 2 h cultures were used to inoculate 5.0 mL YEPD with β-estradiol (1 μM final concentration—to induce the MEP) and ampicillin (50 μg/mL final), and incubated as above. The media was changed every ~9–12 h (pelleting cultures at 1000 × g, 5 min, room temperature before removing the supernatant and adding 5.0 mL 30 °C YEPD (1 μM β-estradiol, 50 μg/mL ampicillin final) for 76 h. Cells were collected by centrifugation (1000 × g, 5 min), fixed in 4% paraformaldehyde, and stained with wheat-germ agglutinin Texas Red (Thermo Fischer #W21405) according to manufacturer’s instructions. The DyLight 680-labeled cells (original mother cells) were identified by fluorescent microscopy, and wheat-germ agglutinin Texas Red-labeled bud scars were counted (>125 cells were analyzed per strain and experiment). Each lifespan curve in Figs. 1f, 2h–m, 4a–c, and Supplementary Fig. 2b was obtained by concatenating all the budscar counts for each strain from multiple (≥3) biological replicates/experiments, which only aged and compared the specific strains presented within the respective figure panel. The concatenated replicative lifespans from multiple biological replicates for each strain were then analyzed and plotted using Microsoft Excel v16.34 and Graphpad Prism v8.3.1. Independent budscar counting was accurate to within ±1.Statistics and reproducibilityStatistical analyses including two-way ANOVA (multiple comparisons) for flow cytometry, log-rank tests for replicative lifespan comparisons, and paired t-tests (two-tailed) were performed using GraphPad Prism v8.1.1 (San Diego, CA). The calculated p-values are indicated in the figure panels or legends. All figure panels represent data from indicated biological replicates experiments yielding similar results.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Information
Supplementary Data 1
Reporting Summary | nature communications | [
"Article"
] | [
"Lipids",
"Metabolomics",
"Cell biology",
"Organelles",
"Genetics"
] | organisms age metabolism slows accumulation of neutral lipid fat), energy source in lipid droplets metabolic decline during aging attributed to decline nicotinamide (NAD+) levels1,2 research on interventions increase NAD+ mitigate changes metabolism including fat interventions improve metabolism reduce fat always increase NAD+ extend studies present alternative mechanisms fat accumulation during aging aging in mice DNA double stranded breaks increases activates muscle specific enzyme (DNA-PK) responsible for 40% weight gained by high-fat diets8 In aging flies muscles accumulate LDs overexpression of cytosolic histone deacetylase (HDAC6) suppresses accumulation9 questions remain concerning lipid accumulation during aging reduces longevity explore LDs change in Saccharomyces cerevisiae cells LDs accumulate new link between biosynthesis NAD+) pathway lipid droplets during aging increasing BNA pathway reduces LD accumulation during agingreduction pulling metabolic flux from glycolysis through shikimate aromatic amino acid biosynthesis pathways upstream BNA pathway supply tryptophan substrate genetic approaches SA-BNA pathway longevity LD accumulation aging modulated independently from longevity LD accumulation shorten lifespan protects cells against cold stress findings reveal lipid-droplet accumulation impacts longevity new strategy for lessening lipid accumulation droplets accumulate during yeast replicative Saccharomyces cerevisiae defined cell divides before aged yeast cells exhibit-induced decline reduced mitochondrial function increased genomic instability11 study screened cells for age-associated changes in organelles lipid droplets (LDs accumulate (Fig BODIPY 493/503 increase 7.2-fold from age 0 to 23 corroborated biochemically 1Age-associated lipid droplet accumulation suppressed by BNA2 overexpressionReplicative age median age mother cells budscar counting Lipid droplets (LDs Control AB18-07) cells mCherry Erg6p yeast LD protein BODIPY 493/503 lipid 4 images from four experiments neutral lipid phospholipid fractions from young aged UCC492512 13, 19 cells analyzed gas chromatography mass spectrometry = 6 Paired t-test **p = 0.0065 BODIPY stained LDs BNA2 overexpression cells 4 μm). images cells aged quantified one experiment LDs stained BODIPY 493/503 quantified BNA2-OE cells flow cytometry mean normalized BODIPY intensities strain determined Supplementary Fig. 1 averaged = 13 experiments plotted ±SEM Two-way ANOVA test ****p ≤ 0.0001 p > 0.05 Replicative lifespan determined microdissection10 25 Control cells 25 maximum 55 39 BNA2-OE cells analyzed = 1 experiment Log-rank test p = 0.0116.RLS Mother Enrichment Program19 2300 Control cells 2450 BNA2-OE cells analyzed 16 experiments Average median maximum lifespan Control (25.2 ± 0.3 41.6 ± 0.6), BNA2-OE (28.9 ± 0.5 49.9 ± 1.3). Log-rank test ****p ≤ 0.0001 Cellular NAD+ NADH Control BNA2-OE cells quantified absorbance 450 nm averaged plotted ±SEM 5 experiments). Source data (b), (d–h) link biosynthesis NAD+ lipid limited over-expression screen reduced LDs genes yeast glycerolipid LD dynamics identified BNA2 not LDs encodes indoleamine 2,3-dioxygenase tryptophan biosynthesis NAD+ homozygous BNA2 overexpression diploid cells compared diploid WT cells young BNA2-OE control cells LD levels similar old BNA2-OE cells decreased 40% reduction control cells not BNA2 BNA2-OE reduces LD levels aged cellsBNA2-OE lifespan used complementary approaches [microdissection10 Mother Enrichment Program19 found BNA2-OE cells significant increases lifespan>15% (Fig. 1e f Fig. 2a). data consistent hypothesis reducing LD accumulation aging promotes longevity anticipated BNA2-OE NAD+ levels no significant difference NAD+ levels between BNA2-OE control cells (Figs. 1g–h 4f). BNA pathway LD accumulation longevity homozygous deletions core BNA pathway genes BNA2 branch point gene (Figs. 2a 3a Deleting core BNA pathway genes reduced lifespan 2h deleting BNA3 no effect. 2i data core BNA pathway not BNA3 branch point important for longevity deleting BNA genes reduced lifespan deletions affect normal accumulation LDs aging (Fig. 2b d–f inconsistent hypothesis suggests LD accumulation impact lifespan.Fig. 2Lipid droplet accumulation separable from longevity BNA pathwaysLD flow cytometry Fig. 1) Two-way ANOVA comparisons p > 0.05 ≤ 0.05 ≤ 0.0001 Control black dots BNA2-OE blue Deletions red diamonds orange triangles n experiments RLS analysis Fig. 1f experiments cells median RLS maximal RLS colors Control black BNA2-OE blue deletions red orange Log-rank tests p > 0.05 ≤ 0.05 panels ≤ 0.0001 Control 400 24 ± 1.2 39.3 ± 0.7 BNA2-OE 27.5 0.8 47.7 2.2 bna1Δ 19 2.3 38 1.7 25.6 1.8 42.6 ± 5.2 350 24 ± 1.2 38 ± 1.2. BNA2-OE 27.2 ± 0.9 45 0.6 bna3Δ 23.7 0.9 38.7 ± 2.7. 27 0.6 47.3 ± 4.4 Control 475 25.3 ± 0.3 46.7 ± 4.7. BNA2-OE 30 0.6 bna5Δ 21.7 0.6 36.7 ± 0.7475 25.3 0.3 47 4.4 775 24.6 0.2 42.4 0.8 925 28.6 0.5 53.2 ± 2.1 bna6Δ 700 21.4 0.5 37.8 1.4 725 22 ± 0.5 39 2.1 325 22 1 38.7 1.7 BNA2-OE 325 28.7 0.3 49.7 ± 0.3 bna7Δ 325 17 ± 0.6 30.7 ± 1.2. 15.3 ± 0.9 33 ± 2.5 24.6 ± 0.5 41.8 ± 1.1. BNA2-OE 725 28.6 ± 0.8 48.6 ± 2.9. aro1Δ 775 17.2 ± 2.0 41.2 ± 2.6 750 25.0 ± 1.3 48.2 ± 1.7 Source data metabolite levels-correlate lipid droplet levels aging SA-BNA pathway metabolite names BNA glycolysis BNA Middle-aged Control BNA2-OE cells analyzed metabolomics metabolites experiments normalized cell countsnormalized measurements BNA2-OE cells divided by Control (WT) cells ratios averaged plotted as metabolite levels versus control cells ± SEM Fig. 4 Cells analyzed by targeted metabolomics Measurements metabolites in shikimate BNA pathways in Control BNA2-OE aro1 bna6 cells normalized to cell counts averaged from = 4 experiments reported as arbitrary units ± SEM limit detection Paired t-test p ≤ 0.05 ≤ 0.0001 p > 0.05 BNA2-OE suppresses accumulation lipid droplets extends lifespan Control cells accumulate lipid droplets aging BNA2 decreases lipid-droplet accumulation pulling substrates lipid droplets toward BNA deletion in ARO1 prevents BNA2-OE suppressing accumulation lifespan extension by BNA2-OE with increasing branch point metabolite xanthurenate Source data for provided.Lipid droplet accumulation separable from investigated BNA pathway genes for BNA2-OE reduce LD accumulation agingcells with overexpressed BNA2 displayed reduced LD accumulation regardless BNA genes deleted (Fig. results flux metabolites from Bna2p toward NAD+ synthesis important for suppressing age accumulation by BNA2-OE investigated dependency BNA pathway on longevity in BNA2-OE cells Cells with BNA2-OE combined with bna1Δ bna5Δ deletions increased lifespan BNA genes deleted (Fig. 2h–j increase lifespan BNA2-OE cells same with without BNA3 increase BNA2-OE achieved control levels when BNA1 or BNA5 eliminated deleting BNA6 or BNA7 blocked BNA2-OE extend lifespan beyond (Fig. 2k–l BNA2-OE can extend lifespan without BNA1 BNA3 BNA5 not without BNA6 or BNA7 BNA pathway linked to lifespan extension not necessary for suppressing LD accumulation by BNA2-OE results indicate LD accumulation longevity genetically separable can modulated independentlyBNA pathway genes BNA2 required for suppression age-associated LD accumulation by BNA2-OE examined shikimate aromatic amino acid synthesis pathways upstream BNA2 critical Substrates shikimate pathway supplied by glycolysis (phosphoenolpyruvate pentose phosphate pathway (erythrose-4-phosphate substrates converted into chorismate precursor aromatic amino acids phenylalanine tyrosine tryptophan (Figs. 2a crosstalk between aromatic amino acid synthesis BNA pathways through shared intermediate anthranilate23 gene ARO924 (Fig. 3a ARO1 encodes rate limiting enzyme shikimate pathway ARO1 deleted blocked BNA2-OE suppress LD accumulation aging ARO1 essential for BNA2-OE aging cells deleting ARO1 ARO1 reduced median lifespan BNA2-OE increased lifespan cells longer WT not BNA2-OE BNA2-OE reduced LD accumulation aging pulling substrates extended lifespan increasing flux downstream toward BNA pathwayrewiring reduces lipid levels in old BNA2-OE LD accumulation metabolites SA-BNA pathway examined in genetically altered cells Figure 4 BNA2-OE minimally affected phosphenolpyruvate levels increased SA pathway metabolite levels shikimate-phosphate chorismate ~16 to ~60-fold tyrosine tryptophan levels ~4-fold ~1.9-fold ~47%, Chorismate anthranilate increased in BNA2-OE bna6Δ cells metabolites low levels in BNA2-OE aro1Δ cells 3c high levels LDs BNA2-OE requires ARO1 increase SA pathway metabolites levels correlate with LD accumulation aging BNA2-OE increased kynurenine 3-hydroxykynurenine (3-HK) levels >170-fold) metabolites detected high levels in BNA2-OE bna6Δ aro1Δ cells less kynurenine (~20-fold less 3-HK (~2.7-fold less in BNA2-OE aro1Δconsistent with kynurenine 3-HK from exogenous tryptophan results consistent with BNA2-OE increasing BNA pathway metabolite levels pulling substrates upstream SA pathway BNA2-OE minimally altered metabolite levels last steps BNA pathway [3-hydroxyanthranilate-fold quinolinic acid NAD+ BNA2-OE increases BNA metabolite levels 3-HK less effect on production NAD+ results suggest 3-HK metabolic flux diverted into BNA branch points not toward NAD+ BNA2-OE increased branch point metabolites kynurenine 3-HK increased kynurenic acid ~350- 49-fold kynurenic acid high in BNA2-OE bna6Δ cells not BNA2-OE aro1Δ cells consistent with increased branch point metabolite levels requiring increased tryptophan formylkynurenine crosstalk between aromatic amino acid synthesis BNA pathways data support model BNA2-OE reduces LD accumulation during aging pulling substrates toward BNA pathwaydroplets protect aging cells against stressLDs cellular stresses25 accumulation LDs in aging cells tested with cold-shock regimen cell viability middle-aged cells after exposure cold (4 cold exposure increased lifespan WT cells minimally affected BNA2-OE aro1Δ cells cold decreased lifespan BNA2-OE cells aged control BNA2-OE aro1Δ cells accumulate more LDs aged BNA2-OE fewer speculate LD accumulation aging protection against stress 4Increased survival to cold exposure correlates with LD accumulation Cells aged in 30 °C media until 16 exposed to 4 °C or 30 °C media until death lifespan analyzed Fig. 1f Log-rank test p > 0.05 not ****p ≤ 0.0001 Control (WT 550 cells at 30 °C 525 at 4 °C), BNA2-OE aro1Δ BNA2-OE analyzed over n = 4 experimentsmaximal lifespans at 30 °C Control 25.8 ± 1.3 41 ± 1.9; BNA2-OE aro1Δ 25.1 ± 2.2 44.3 ± 4.2 BNA2-OE 29.5 ± 1.7 48.8 ± 2.5 lifespans at 4 °C Control 27.5 ± 1.3 43.3 ± 2.5; BNA2-OE aro1Δ 25 ± 1.8 45 ± 4.6 BNA2-OE 24.8 ± 1.3 40 ± 3.7. Source data file-fat diets contribute fat accumulation disease fat aging health longevity study shows yeast accumulate fat LDs LD accumulation longevity longevity under stress correlations in Drosophila neutral lipid accumulation advantage aging cells conditions evidence aging cell metabolism rewired by BNA2 overexpression flux LD accumulation network genes reactions linked to yeast lipid network toward neutral lipid synthesis LDs during aging substrates shikimate pathway supplied by glycolysis pentose phosphate pathway Glycolysis supplies pentose phosphate pathway acetyl-CoA precursor lipidsaltering glucose levels lipid profile cells age glycolytic flux skewed toward neutral lipid through pyruvate acetyl-CoA (Fig. cells overexpressing BNA2 glycolytic flux from toward SA-BNA pathway LD accumulation during aging BNA2-OE increases kynurenic acid suppresses neutral lipid accumulation in cells links described between kynurenine metabolites lipid metabolism in mammalian regulation IDO1 BNA2) activity expression in metazoan cancers IDO1 overexpressed cancer progression decreasing levels tryptophan immune detection increased kynurenine metabolites correlate with replicative lifespan extension cell growth IDO1 overexpression promotes cell growth proliferation cancer progression yeast strains derivatives of S. cerevisiae S288C39 UCC877319 listed in Supplemental Table 1. Gene deletion strains created by one step PCR-mediated gene replacement39 Plasmid templates for gene deletion construction from pRS400 vector39Oligonucleotides gene replacement mCherry tagging Supplementary Table 2. Transformation insertion NotI-digested plasmid pAG306-GPD-xsome1 overexpression parental genotype BNA2 overexpression deletion strains MATa/MATα leu2△0-Cre-EBD78-CDC20-Intron-loxP-HPHMX-UBC9 Gene Deletion GPD-(BNA2) Strains expressing C-terminal chromosomal fusions ERG6 mCherry transformation parental yeast strains pKT127-mCherry12 cell lipid-droplet suppressor screenMother enrichment program cells transformed 96-well format 280 high-copy 2-μm plasmids tiled genomic DNA library Each plasmid unique sequence-verified genomic DNA fragment gene lipid-droplet glycerolipid additional genes genes age-induced accumulation lipid droplets plasmid-expressing strains grown in 2 mL yeast extract peptone 2% glucose) kanamycin at 30 °CCells stained BODIPY 493/503 Nile Red #N3013) imaged fluorescence microscopy Plasmid strains 50% cells lipid droplets suppressors age lipid accumulation suppressors genes transferred pDONR221 pAG306-GPD-xsome112,41 Plasmids integrated chromosome 1 strains digestion aged purified examined fluorescence microscopy suppressor (BNA2) confirmed young-cell-like levels lipid droplets 50% middle-aged cells aging cells magnetic beadsCells stored 15% glycerol −70 °C YEPD agar grown < 1 day 2 mL YEPD 16 × 150 mm culture tubes grown 27.5 h 30 °C 40 rpm Saturated cultures diluted 1:250,000 100 mL YEPD 250-mL flasks grown ~15.5 h 30 °C 180 rpm × 106 cells/mLLyophilized NHS-activated 50-nm beads (Ocean Nanotech suspended 2.5 μg/mL PBS 30 × 106 cells centrifuged 1000 × g 5 min transferred microcentrifuge tubes 1 mL PBS 2% glucose centrifuged 1500 × g 1 min supernatant aspirated washed PBS-G second 600 μL 50% PEG-3350 400 μL PBS-G 10 μL bead slurry layered cell pellets suspended homogenous incubated 10 min cells centrifuged 1500 × g 1 min supernatant aspirated pellets rinsed 1 mL 30% PEG-3350 centrifuged supernatants aspirated suspended 30 °C YEPDaging 1 L cultures 30 °C YEPD 4-L flasks inoculated 8 × 106 cells incubated shaker 30 °C 95 rpm remaining cell suspension centrifuged fixed 4% paraformaldehyde saved PBS 4 °C 11 23, 34 h after inoculation aged cells recovered centrifugation 5 resuspended 5 mL 30 °C YEPD culture suspension applied LS MACS columns Biotec magnets pre-equilibrated 1 volume 30 °C drain washed three volumes 30 °C YEPD aged cells eluted collected 30 °C YEPD microscopy flow cytometry analyses 1–2 × 106 eluted cells fixed 4% paraformaldehyde saved PBS remaining suspension inoculate 1.5 L 30 °C YEPD 11 to 23 h 0.5 L 30 °C YEPD 23 to 34 h NAD+/NADH metabolomics analyses eluted cells processedNAD+/NADH analysesYoung aged cells washed ice-cold Yeast Nitrogen Base +2% glucose 1500 × g 1 min 4 NAD+/NADH analyzed quantitation kit cells suspended extraction buffer freeze thaw cycles dry ice stored −70 °C cleared centrifugation (14,000 × g 5 min 4 Supernatants Amicon regenerated cellulose filters centrifuged (15,000 1.5 h 4 flow-through analyzed nm absorbance measured Powerwave XS metabolomicsAged cells washed-cold yeast base 2% glucose suspended 80:20 methanol:water −20 °C incubated 3 min −20 °C Ammonium bicarbonate pH 6.8 added 25 mM concentration incubated −20 °C 20 min Extracts cleared Amicon cellulose centrifugal 15,000 × g 1.5 h −8 °C Flow-through −70 °C until analysis Samples dried nitrogen gas mass spectrometry analysis LC-MS Calico Life Sciences Samples resuspended water diluted 1:4 into acetonitrilesamples analyzed LC-MS methods Vanquish UPLCs Q-Exactive Plus mass spectrometers Thermo Xcalibur 4.0 Thermo Foundation 3.1 SP1 Waltham positive ionization separated SeQuant® ZIC®-pHILIC column 5 μm particle size 200 150 2.1 mm phase A 20 mM ammonium carbonate water B acetonitrile flow rate 150 μL/min gradient −6 80% 2.5 73% 65% 7.5 57% 50% 35% 22.5 80% 24 80% spectrometer positive ion mode resolution 70,000 AGC target IT 100 scan range 70–1050 MS2 parameters resolution 17,500 AGC 1.00E 50 loop count 6 window/z 1 20 40 80 underfill ratio 1.00% Apex trigger(s) 3–10 dynamic exclusion(s) negative ionization separated reverse phase ion-pairing chromatographic method Agilent Extend C18 RRHD column 1.8 μm particle size 80 Å 2.1 150 mm phase A 10 mM tributylamine 15 mM acetic acid 97:3 water:methanol pH 4.95 B methanolflow rate 200 μL/min gradient −4, 0% 0 0% 20% 7.5 13, 55% 15 95% 18.5 95% 19, 0% 22, 0% mass spectrometer negative ion mode data-dependent parameters resolution 70,000 AGC target 1.00E + 06 maximum IT (ms 100 scan range 70–1050 MS2 parameters resolution 17,500 AGC target 1.00E + 05 maximum IT (ms 50 loop count 6 window (m/z 1 (N)CE 20 50 100 underfill ratio 1.00% Apex trigger(s) 3–12 dynamic exclusion(s) 20. Metabolites identified fragmentation spectra retention times chemical standards metabolomics RAW converted mzXML ProteoWizard identity peak integration verified Maven measurements analyzed plotted Microsoft Excel v16.34 Graphpad Prism v8.3.1 mass spectrometry analysis University of Washington Department Medicinal Chemistry Mass Spectrometry Center Samples reconstituted 50:50 100% methanol 0.4% acetic acid water injected 1290 Agilent UPLC system Agilent 6520 Quadrupole Time of Flight Features extracted analyzed Agilent MassHunter Data Acquisition Processing software measurements above limit detection reported averages ± SEM from four experiments measurements normalized cells per sample control strain analyzed plotted Microsoft Excel v16.34 Graphpad Prism v8.3.1 Standards metabolite run comparison.Fluorescence images captured Leica Microsystems (Buffalo Grove DMI6000 LAS X 3.3.3.16958) Leica DFC365 FX camera GFP Texas Red 2 QAS filters HC PL APO ×63 oil objective images analyzed Adobe Photoshop CC 2019 v20.0.8.Lipid purification analysisUCC4925 cells grown YEPD agar cultured to saturation log phase<2.0 × 106 cells/mL 20 × 106 cells washed twice PBS incubated 3 mg/mL EZ-Link Sulfo-NHS-LC-Biotin PBS Cells pelleted supernatant aspirated suspended 500 mL YEPDCultures incubated 30 °C shaker (95 10.5 h cell density determined 20 × 106 cells harvested 1000 × g 5 min supernatant removed cell pellets flash-frozen liquid nitrogen −70 °C age 0 remaining culture centrifuged g 5 min 4 pellets washed twice suspended 2 × 108 cells/mL ice-cold PBS 1/20th volume azide-free Multimacs Streptavidin Microbeads incubated 30 min 4 °C washed centrifugation PBS 4 °C suspended ice-cold PBS LS MACS columns 5-mL ice-cold columns drained washed twice 1 volume ice-cold PBS NHS-biotin-labeled cells columns removed eluted 4-mL ice-cold PBS Cells pelleted supernatant aspirated flash-frozen liquid nitrogen −70 °C age 8 neutral lipid purification lipids extracted 2:1 chloroform:methanol 1 h temperature 0.2 volumes 0.9% NaCl added vortexed 2 min top layer removed bottom layer dried nitrogenDried lipids resuspended 1 mL chloroform Calibrated phospholipid-Dihepatdecanoyl-Glycero-3 TAG (tritridecanoin added mixture before extraction PLs NLs purified by solid chromatography Extracted lipid resuspended chloroform loaded SPE columns (100 pre-equilibrated 3 mL chloroform TAGs eluted first 3 ml chloroform Glycosphingolipids 5 mL 9:1 acetone:methanol last 3 mL methanol Purified lipids dried resuspended methanol/2.5% H2SO4 incubated 1 h 80 °C FAMEs analyzed data collected Agilent MSD Chemstation analyzed Agilent ChemStation Integrator PL compared with added standards Data presented NL:PL ratio determined fatty acids NL versus PL measurements analyzed Microsoft Excel v16.34 Graphpad Prism v8Flow cytometryCells stained with BODIPY 493/503 (ThermoFisher #D3922) Wheat Germ Agglutinin Alexa Fluor 647 Conjugate SYTOX Blue Dead Cell Stain #S34857) signals quantified Becton Dickinson) FACSCanto II FACSDiva 6.1.3 software excitation 488 nm emission 530 nm (FITC 633 nm (APC, WGA 405 450 nm (PacificBlue Sytox Blue). gating live cells absence Sytox Blue Staining presence WGA-647 staining samples excluded particles small large yeast mean BODIPY fluorescence determined from 10,000 cells per experiment Data analyzed FlowJo v10 software (Becton Dickinson Company Franklin Lakes, BODIPY measurements normalized to average median age 0 control cells Supplemental Fig. 1 data normalized plotted using Microsoft Excel v16.34 GraphPad Prism v8.3.1 CA).Replicative lifespan microdissection lifespans control (AB18-07) BNA2-OE (AB18-11) cellsdetermined modification Mother Enrichment Program. modifications generate high cells strain Strains grown YEPD agar < 24 h 2 mL YEPD grown 27.5 h 30 °C diluted 1:250,000 grown 25 mL YEPD (125-mL flasks 15.5 h density ~2 × 106 cells/mL 2.0 × 106 cells washed PBS-GN glucose 1 mM nicotinic acid labeled 5 μg/mL DyLight 680 NHS-Ester PBS-GN 5 min cells washed twice 30 °C YEPD 1.0 × 106 cells inoculate 4.5 mL 30 °C YEPD grown 2 h rotor (40 Fifteen microliters cultures inoculate 5.0 mL YEPD β-estradiol ampicillin (50 μg/mL incubated media changed every ~9–12 h cultures 1000 × g 5 min 5.0 mL 30 °C YEPD (1 μM β-estradiol 50 μg/mL ampicillin 76 h Cells collected centrifugation fixed 4% paraformaldehyde stained wheat-germ agglutinin Texas RedDyLight 680-labeled cells identified by fluorescent microscopy wheat-germ agglutinin Texas Red-labeled bud scars counted>125 cells analyzed per strain lifespan curve in Figs. 1f 2h–m 4a–c Fig. 2b obtained concatenating budscar counts from multiple (≥3) replicates strains concatenated lifespans analyzed plotted using Microsoft Excel v16.34 Graphpad Prism v8.3.1. Independent budscar counting accurate within.Statistics reproducibilityStatistical analyses two-way ANOVA cytometry log-rank tests paired t-tests performed GraphPad Prism v8.1.1 (San Diego calculated p-values indicated in figure panels panels represent data from similar results.Reporting Nature Research Reporting Summary.Supplementary information | 47 | 0.989871 |
10.1038/s41467-021-21082-x | PMC7864911 | The analysis of AMPA-type glutamate receptor (AMPAR) trafficking is essential for understanding molecular mechanisms of learning and memory, but the analytical tools are currently limited. Here, the authors report a method that combines affinity-based receptor labeling and bioorthogonal click chemistry to quantify AMPAR distribution and trafficking under physiological conditions. | The regulation of glutamate receptor localization is critical for development and synaptic plasticity in the central nervous system. Conventional biochemical and molecular biological approaches have been widely used to analyze glutamate receptor trafficking, especially for α-amino-3-hydroxy-5-methyl-4-isoxazole-propionate-type glutamate receptors (AMPARs). However, conflicting findings have been reported because of a lack of useful tools for analyzing endogenous AMPARs. Here, we develop a method for the rapid and selective labeling of AMPARs with chemical probes, by combining affinity-based protein labeling and bioorthogonal click chemistry under physiological temperature in culture medium. This method allows us to quantify AMPAR distribution and trafficking, which reveals some unique features of AMPARs, such as a long lifetime and a rapid recycling in neurons. This method is also successfully expanded to selectively label N-methyl-D-aspartate-type glutamate receptors. Thus, bioorthogonal two-step labeling may be a versatile tool for investigating the physiological and pathophysiological roles of glutamate receptors in neurons. | IntroductionIn the central nervous system, ionotropic glutamate receptors (iGluRs) mediate fast excitatory neurotransmission. iGluRs are categorized into distinct classes based on their pharmacology and structural homology, including the α-amino-3-hydroxy-5-methyl-4-isoxazole-propionate (AMPA) receptor (GluA1–4), kainate receptor (GluK1–5), N-methyl-D-aspartate (NMDA) receptor (GluN1, GluN2A–D, GluN3A–B), and δ receptors (GluD1–2)1. iGluRs assemble as tetramers, and functional receptors are formed exclusively by the assembly of subunits within the same functional receptor class.AMPA receptors (AMPARs), which are mainly permeable to monovalent cations (Na+ and K+), mediate the majority of excitatory synaptic transmission. AMPARs can form homotetramers or heterotetramers, and subunit compositions are dependent on brain regions. In hippocampal CA1 neurons, the majority of AMPARs are made up of GluA1/A2 and GluA2/A3 subunit combinations, with a small contribution of GluA1 homomers2,3. Recent studies have revealed that AMPARs are constitutively cycled in and out of the postsynaptic membrane through endocytosis and exocytosis. The precise regulation of this process is critical for synaptic plasticity, which is the basis of learning, memory, and development in neural circuits2,3. Although AMPARs and kainate receptors are activated by glutamate binding, NMDA receptors (NMDARs), which have high permeability to Ca2+, require depolarization as well as agonist binding for their activation. Functional NMDARs require the assembly of two GluN1 subunits together with either two GluN2 subunits, or a combination of GluN2 and GluN3 subunits1. An NMDAR-dependent Ca2+ influx triggers intracellular signal transduction cascades, and the precise targeting of NMDARs to synapses is essential for controlling neuronal connectivity or neuroplasticity4. Thus, to understand the molecular mechanisms of learning and memory, it is critical to analyze the membrane localization and trafficking of iGluRs.Biochemical approaches, such as surface biotinylation assays or related methods, have been widely used to analyze membrane protein localizations, and these methods have been successfully applied to AMPARs57. Although they are powerful tools for the analysis of AMPAR trafficking, cell-surface proteins are randomly labeled with biotin using these methods. As a result, purification of biotin-labeled AMPAR is required, which hampers quantitative analyses of trafficking. In contrast, to selectively visualize glutamate receptors, fluorescent proteins are fused to the receptors using genetically encoded approaches. For example, a pH-sensitive variant of GFP (super-ecliptic pHluorin [SEP]) can be fused to the extracellular region of receptors to visualize cell-surface receptors in live neurons8,9. Instead of fluorescent proteins, protein tags such as SNAP- or Halo-tags are fused to the receptors for the covalent labeling of small chemical probes at the time that the probes are added10–12. The downsizing of these protein tags has been successfully demonstrated by using a short peptide tag (1–3 kDa) and its probe pair13–15. More recently, genetic code expansion in combination with bioorthogonal click chemistry has been reported for the fluorescent labeling of iGluRs in HEK293T cells, in which chemical probes are covalently attached to the side chain of an unnatural amino acid residue16,17. These genetically encoded approaches have been widely used in trafficking studies of iGluRs, especially for AMPARs. However, in most cases, these methods largely rely on the overexpression of target iGluR subunits. Given the formation of heterotetramers consisting of different subunits in iGluRs, the overexpression of a single iGluR subunit may interfere with the localization and/or trafficking of native iGluRs in neurons. Ideally, endogenously expressed iGluRs should be tagged with small chemical probes18,19.In situ chemical protein labeling is ideal for analyzing native proteins in live cells. Affinity-based protein labeling is a powerful technique for the selective modification of target proteins20–26. As a traceless affinity-based labeling method for cell-surface proteins, our group has reported ligand-directed acyl imidazole (LDAI) chemistry24,25. With this technique, small chemical probes including fluorophores are covalently attached to nucleophilic amino acid residues located near the ligand-binding site. Recently, we have developed an AMPAR-selective LDAI reagent, termed “chemical AMPAR modification 2” (CAM2) reagents, which allows us to label chemical probes to AMPARs endogenously expressed in cultured neurons or acutely prepared brain slices26. Although this technique is powerful for the selective modification of chemical probes to AMPARs, there are some restrictions for visualizing or analyzing cell-surface AMPARs. First, live cells need to be kept at low temperatures (e.g., 17 °C) during CAM2 labeling (1–4 h) to suppress the internalization of labeled AMPARs27. Second, the neuronal culture medium needs to be exchanged for serum-free medium or buffered saline during labeling to decrease non-specific labeling of serum proteins such as albumin. The relatively long-term exposure (1–4 h) to these non-physiological conditions may interfere with neuronal activity or survival28–30. Ideally, neurons should be kept under physiological conditions during chemical labeling.Here, we show a method for the rapid and selective labeling of AMPARs under physiological temperature in culture medium by combining LDAI-based protein labeling and the inverse electron demand Diels–Alder (IEDDA) reaction, a form of fast click chemistry31–33. This two-step labeling allows the quantitative analyses of distribution and/or trafficking of endogenous AMPARs from short to long periods in cultured neurons. In addition, we successfully apply this technique to chemically label and study the trafficking of endogenous NMDARs in neurons.ResultsRapid labeling of surface AMPARs by bioorthogonal two-step labelingWe propose a bioorthogonal two-step labeling technique, which combines LDAI-based protein labeling with the bioorthogonal IEDDA reaction for the rapid and selective modification of chemical probes to cell-surface iGluRs. For the first step, a strained alkene is covalently attached to iGluRs using LDAI chemistry, where the acyl substitution reaction to nucleophilic amino acid residues is facilitated by selective ligand–protein recognition (first step in Fig. 1a). Next, the labeled alkene group is rapidly modified with tetrazine-conjugated probes (Tz-probes) on the cell surface, as a result of the high selectivity and high reaction rate of the IEDDA reaction (second step in Fig. 1a).Fig. 1Rapid labeling of cell-surface iGluRs by ligand-directed two-step labeling.a Schematic illustration of the two-step labeling to iGluRs. In the 1st step, a strained alkene is covalently attached to iGluRs by LDAI chemistry. In the 2nd step, Tz probe is selectively tethered by IEDDA reaction. Lg, selective ligand for iGluRs; Nu, nucleophilic amino acid residue. Glu, glutamate. b Chemical structure of CAM2(TCO). c Chemical structure of Tz probes. The detailed chemical structures are shown in Supplementary Fig. 2. d Schematic illustration of the two-step labeling in live cell.For the selective labeling of a strained alkene to AMPARs in the first step, we designed a CAM2 reagent bearing trans-cyclooctene (TCO), which we termed CAM2(TCO) (Fig. 1b). TCO was selected as the strained alkene because of its extremely fast cycloaddition kinetics in the IEDDA reaction. Compared with the original CAM2 reagents (e.g., CAM2(Ax488)) that bear aromatic fluorophores (see Supplementary Fig. 1a, b), an ethylene glycol linker is added between the reactive acyl imidazole unit and the TCO group in CAM2(TCO) to increase its hydrophilicity. Hydrophobic or aromatic groups have high affinity to albumin abundantly contained in serum34; therefore, this improvement decreases the undesired labeling of albumin, which allows the chemical labeling of AMPARs to be conducted in cell culture medium containing serum or substitutes. In addition, the first labeling is conducted at a physiological temperature (37 °C). Although some of the labeled AMPARs are likely to be internalized in this condition, this is not problematic with the two-step labeling technique. This is because the chemical probes are selectively tethered to cell-surface AMPARs in the second step reaction (Fig. 1d).Regarding the second step (the IEDDA reaction), the reaction rate is highly dependent on the chemical structure of the tetrazine group. We selected monoaryl tetrazine, which has both a fast reaction rate and high bioorthogonality, and prepared cell-impermeable Tz-probes bearing hydrophilic and anionic fluorophores or biotin for cell-surface labeling (Fig. 1c and Supplementary Fig. 2). Shortening the reaction time of the probe labeling not only contributes to cell-surface specific labeling, but also decreases the adsorption of the chemical probes to cells, culture dishes, or coverslips. Moreover, some tetrazine-fluorophore conjugates have a “turn-on” response upon the IEDDA cycloaddition35–37, which contributes to a high signal-to-noise (S/N) ratio in fluorescence imaging.Chemical labeling of surface AMPARs ectopically expressed in HEK293T cellsThe designed two-step labeling method was initially examined in HEK293T cells transiently expressing GluA2, a main subunit of AMPARs. For the first step reaction, CAM2(TCO) was added to the culture medium, which included 10% fetal bovine serum (FBS), and the culture dish was incubated at 37 °C for 4 h. The second step reaction was performed for 5 min by adding membrane-impermeable Tz(Fl) for fluorescein labeling on the cell surface. As shown in Fig. 2a, western blotting of the cell lysate using anti-fluorescein (anti-Fl) antibodies showed a strong band around 110 kDa (lane 1). This band was not observed in the cells co-treated with a competitive ligand (NBQX) or in any other control conditions (lanes 2–5). With regard to the molecular weight of the labeled band, the anti-Fl signal corresponded to the highest signal among multiple bands that were detected using anti-GluA2/3 antibodies (Fig. 2b). The multiple GluA2 bands converged into a single lower band after treatment with peptide-N-glycosidase F (PNGase F), which is consistent with previous reports showing that GluA2 is highly glycosylated with N-linked sugars38. Importantly, in the PNGase F-treated samples, the shifted anti-GluA2 band merged with the anti-Fl signal (Fig. 2b). These findings indicate that highly glycosylated GluA2 is selectively labeled using our methods. Furthermore, in the case of direct fluorescein labeling using the original CAM2(Fl) under the same conditions (see Supplementary Fig. 1b for its structure), there was a strong band around 70 kDa as well as the 110 kDa band (lanes 6–7 in Fig. 2a and Supplementary Fig. 3). The 70 kDa band, whose intensity did not change even in the presence of NBQX, corresponds to albumin contained in serum (for details, see Fig. 2a legend). These results therefore indicate the high selectivity of the two-step labeling technique using CAM2(TCO) compared with the original CAM2(Fl) under cell culture conditions.Fig. 2The two-step labeling of cell-surface AMPARs ectopically expressed in HEK293T cells.a Western blotting analyses of HEK293T cells after the two-step labeling. HEK293T cells transfected with GluA2flip(Q) (AMPAR(+)) or vector control (AMPAR(−)) were treated with 2 μM CAM2(TCO) for 4 h followed by the addition of 1 μM Tz(Fl) for 5 min, or treated with 2 μM CAM2(Fl) for 4 h in the presence or absence of 50 μM NBQX in culture medium at 37 °C. The cell lysates were analyzed by western blotting using anti-fluorescein or anti-GluA2/3 antibody. Quantification of the band intensity is shown in Supplementary Fig. 3. When CAM2(Fl) is added in serum free medium, strong bands around 70 kDa in lane #6 and #7 disappear (for details see ref. 26). b Effects of PNGase F treatment on the western blotting of labeled AMPAR in HEK293T cells expressing GluA2. Lower image shows the overlay of anti-Fl image and anti-GluA2/3 image for lane #3 and #4. Two-step labeling was conducted as described in (a). PNGase F (1000 units/100 μL) was added to the cell lysate. For details, see Methods section. c Confocal live imaging of the HEK293T cells labeled with 2 μM CAM2(TCO) and 0.1 μM Tz(Ax488). Labeling was conducted as described in (a). mCherry-F was utilized as a transfection marker. Scale bars, 10 μm. d Reaction kinetics of tetrazine ligation on live cells by confocal live imaging of the HEK293T cells labeled with 2 μM CAM2(TCO) after addition of 0.3 μM Tz(Ax488) at 37 °C. In left, confocal images are shown. Scale bars, 20 μm. In right, time-course of the fluorescent intensity of Alexa 488 is shown (n = 6 cells). Data are represented as mean ± s.e.m.For visualizing fluorescently labeled AMPARs on the cell surface, confocal microscopic live imaging was performed after the two-step labeling process under cell culture conditions. Here, Tz(Ax488) was used in the second step of labeling. Alexa 488 has bright fluorescence that is unaffected under endosomal acidic conditions; in contrast, fluorescein has weakened fluorescence under acidic conditions. Thus, Alexa 488 is more suitable to quantify the cellular distribution or trafficking of labeled AMPARs using fluorescent imaging. As shown in Fig. 2c, prominent fluorescence was observed exclusively from the cell surface in cells co-transfected with mCherry-F, a membrane-targeted transfection marker. In contrast, fluorescent signals were not observed in control conditions, such as in CAM2(TCO)-untreated or NBQX-co-treated cells (Supplementary Fig. 4). In the case of direct Alexa 488 labeling using CAM2(Ax488) in the same cell culture conditions, labeled signals were observed not only from the cell surface but also from the intracellular space (Supplementary Fig. 1c). This suggests that the two-step labeling technique is superior for the fluorescent visualization of cell-surface AMPARs under cell culture conditions. We also determined the reaction kinetics of rapid fluorophore labeling of cell-surface AMPARs with the help of the turn-on fluorescent property of Tz(Ax488) upon the IEDDA reaction (Supplementary Fig. 5). Immediately after adding Tz(Ax488), prominent fluorescent signals were observed from the cells co-transfected with the transfection marker mCherry-F (Fig. 2d), and the fluorescent signals were saturated within 3 min. Thus, cell-surface AMPARs can be labeled by the fluorophore with fast kinetics.By taking advantage of the high bioorthogonality of the IEDDA reaction, cell-surface AMPARs were successfully labeled with various kinds of chemical probes, ranging from small molecules to middle-sized molecules such as SeTau-647, a squaraine rotaxane dye that has high photostability and a long fluorescence lifetime39,40 (Supplementary Figs. 2 and 6). This probe flexibility is another feature of the two-step labeling technique that is superior compared with the direct labeling of probes using original CAM2 reagents, where each probe-tethered CAM2 needs to be synthesized (Supplementary Fig. 1b).With regard to labeling efficacy, quantification of the remaining unlabeled GluA2 fraction showed that 35 ± 3% of surface AMPARs were visualized in the two-step labeling (Supplementary Fig. 7). Moreover, AMPAR function was not visibly affected by the two-step labeling (Supplementary Fig. 8), which is consistent with our previous analyses that showed minimal disturbance of AMPAR ion channel properties by CAM2 labeling26.Analyses of AMPAR trafficking in HEK293T cellsOnce we had a potential labeling method for cell-surface AMPARs under cell culture conditions, we analyzed receptor trafficking using both live imaging and biochemical approaches. First, we analyzed cell-surface AMPAR trafficking in HEK293T cells using confocal live imaging. After incubating the cells with CAM2(TCO) under physiological temperature in culture medium, Tz(Ax488) was added to the culture medium to selectively visualize cell-surface AMPARs and cells were incubated for each period (0–8 h) (Fig. 3a). As shown in Fig. 3b, the labeled fluorescence on the cell surface decreased after incubation at 37 °C. Fluorescent granules were instead observed in the intracellular area, and most of the fluorescent signals were from intracellular granules after 8 h of incubation. The half-time of cell-surface AMPARs (t1/2surface), which includes both the remaining and recycled fractions, was calculated to be 5.7 ± 0.7 h from the fluorescent intensity on the cell surface (Fig. 3b). In addition, the intracellular punctate signals merged with a fluorescent lysosome marker (LysoTracker) after 8 h of incubation, suggesting that internalized AMPARs were transported to lysosomes (Fig. 3c). Similar internalization behavior was observed when AMPARs were labeled with different fluorophores using Tz(Ax647) or Tz(ST647) (Supplementary Fig. 9).Fig. 3Quantitative analyses of AMPAR trafficking in HEK293T cells using the two-step labeling under the physiological cell culture condition.a Schematic illustration of trafficking analyses of cell-surface AMPARs by confocal microscopy. b Time-lapse confocal imaging of HEK293T cells after two-step labeling using CAM2(TCO) and Tz(Ax488). The HEK293T cells were transfected with GluA2flip(Q). In left, confocal images are shown. Scale bars, 10 μm. In right, time-course of the fluorescent intensity from the cell surface is shown (n = 3 biological replicates). [CAM2(TCO)] = 2 μM, [Tz(Ax488)] = 0.1 μM. c Co-staining of labeled AMPARs with LysoTracker™. The labeled HEK293T cells were incubated with LysoTracker™ Red dnd99 immediately after Tz(Ax488) labeling (upper panel) or subsequent 8 h incubation at 37 °C (lower panel), and confocal live imaging was performed. Pearson’s correlation coefficients (r) are shown in the image. Scale bars, 10 μm. d–f Determination t1/2surface or t1/2life of AMPARs by western blotting. d Schematic illustration of the procedure is shown. In e and f, t1/2surface and t1/2life are determined, respectively. In left, representative results of western blotting are shown. In right, time-course of the labeled band is shown (n = 3 biological replicates). [CAM2(TCO)] = 2 μM, [Tz(Fl)] = 1 μM. g, h Determination of TCO-labeled AMPARs on cell-surface, in intracellular area, or in whole-cell by western blotting. In g, schematic illustration of the procedure is shown. In h, intracellular and surface ratio after CAM2(TCO) labeling for 4 h are determined. In left, representative results of western blotting are shown. In right, band intensities for cell-surface and intracellular labeling were analyzed, both of which were normalized by that for whole-cell labeling (n = 3 biological replicates). [CAM2(TCO)] = 2 μM, [Tz(Fl)] = 1 μM. Data are represented as mean ± s.e.m.Quantitative analyses of the fates of cell-surface AMPARs were examined using biochemical approaches. To quantify the t1/2surface of AMPARs, HEK293T cells were incubated for each period (0–36 h) after treatment with CAM2(TCO) (path-1 in Fig. 3d). Next, Tz(Fl) was added for the selective modification of fluorescein to cell-surface AMPARs. Using western blotting of the cell lysate, the t1/2surface of AMPARs was determined to be 5.3 ± 0.3 h (Fig. 3e), which was similar to the value that was determined using confocal imaging (Fig. 3b). The half-time of the degradation (t1/2life) of cell-surface receptors was evaluated by modifying the protocols, where Tz(Fl) was added after CAM2(TCO) labeling (path-2 in Fig. 3d). The cells were then incubated for each period (0–36 h), and the cell lysates were subjected to western blotting. As shown in Fig. 3f, t1/2life was determined to be 8.1 ± 0.7 h, which was slightly longer than the t1/2surface of cell-surface AMPARs (p < 0.05). Considering the colocalization of internalized AMPARs and lysosomes (Fig. 3c), the internalized AMPARs are likely decomposed via lysosomal degradation in HEK293T cells.We next determined the intracellular versus surface percentages of TCO-labeled AMPARs (TCO-AMPARs) in each period, which can provide valuable information regarding the fate of cell-surface AMPARs. To quantify these percentages, cell-surface AMPARs in HEK293T cells treated with CAM2(TCO) were selectively labeled with fluorescein by adding cell-impermeable Tz(Fl) to the medium under live cell conditions (path-1 in Fig. 3g). To label intracellular TCO-AMPARs, surface TCO-AMPARs were first masked with cell-impermeable Tz(Ax647) (path-2 in Fig. 3g). After lysis of the cells, Tz(Fl) was added to the cell lysate to label intracellular TCO-AMPARs. The whole-cell-labeling fraction, where both cell-surface and intracellular TCO-AMPARs were labeled with fluorescein, was prepared by adding Tz(Fl) after cell lysis (path-3 in Fig. 3g). Prior to these analyses, we first investigated whether the second step reaction using Tz(Fl) proceeds rapidly and/or selectively in cell lysate. Western blotting analyses revealed that the covalent modification of fluorescein was selective to AMPARs in cell lysate (Supplementary Fig. 10), and the reaction was saturated after 15 min when either 0.1 or 0.3 μM Tz(Fl) was added. The intracellular and cell-surface percentages of TCO-AMPARs after 4 h of incubation with CAM2(TCO) were analyzed using this protocol and determined to be 4.1 ± 0.9% and 94.4 ± 3.3%, respectively (Fig. 3h), indicating that intracellular TCO-AMPAR levels were quite low.Rapid labeling of endogenous AMPARs in neuronsWe next examined the applicability of the bioorthogonal two-step labeling technique for the rapid modification of cell-surface AMPARs that are endogenously expressed in neurons. Primary cultured neurons from the cerebral cortex were incubated with CAM2(TCO) for 10 h under neuronal culture conditions, and Tz(Fl) was then added for 5 min for cell-surface labeling. Western blotting analyses of the cell lysate showed a single strong band corresponding to the molecular weight of AMPARs (see lane 1 in Fig. 4a). This band was not detected in the co-presence of the competitive ligand NBQX, or in other control conditions (see lanes 2–4 in Fig. 4a). As observed with the AMPARs expressed in HEK293T cells (Fig. 2a), smeared bands were detected using anti-GluA2 antibodies; the anti-Fl band corresponded to the highest band in the smeared anti-GluA2 signals. After the removal of N-linked sugars by PNGase treatment, the smeared anti-GluA2 bands converged into a single lower band, which merged with the anti-Fl signal (Supplementary Fig. 11). These results suggest that the highly glycosylated fraction of endogenous AMPARs were selectively labeled with fluorescein by the rapid labeling.Fig. 4The two-step labeling of cell-surface AMPARs endogenously expressed in neurons.a Western blotting analyses of cortical neurons after the two-step labeling. Primary cultured cortical neurons were treated with 2 μM CAM2(TCO) for 10 h followed by the addition of 1 μM Tz(Fl) for 5 min in the presence or absence of 50 μM NBQX in culture medium at 37 °C. The cell lysates were analyzed by western blotting using anti-fluorescein or anti-GluA2 antibody. b Confocal live imaging of the neurons labeled with 2 μM CAM2(TCO) and 0.1 μM Tz(Ax488). Labeling was conducted as described in a. Scale bars, 2 μm. c Immunostaining of cortical neurons after the two-step labeling. Labeling was conducted as described in b. The neurons were fixed, permeabilized, and immunostained using anti-MAP2 (upper) or anti-PSD95 antibody (lower). Scale bars, 5 μm. Pearson’s correlation coefficients (r) are shown in the image. Whole images are shown in Supplementary Fig. 14a. d FLIM imaging and analyses of cell-surface AMPARs in the neurons after the two-step labeling. The neurons were prepared as described in c. In left, representative confocal image (upper) and FLIM image (lower) for a lifetime fraction (τ = 2.4 ± 0.1 ns) are shown. Scale bars, 5 μm. In right, FLIM intensities in spine and dendrite were analyzed (n = 4 cells). [CAM2(TCO)] = 2 μM, [Tz(Ax488)] = 0.1 μM. *Significant difference (p < 0.05 by two-sided Student’s t-test. p = 0.048). Data are represented as mean ± s.e.m.Of the AMPAR subunits (GluA1–4), GluA1, GluA2, and GluA3 are highly expressed in cultured cortical neurons41. We next examined the efficacy of our methods for visualizing tetrameric AMPARs by quantifying the remaining unlabeled GluA2 fraction. As shown in Supplementary Fig. 12a, 44 ± 4% of GluA2-containing AMPARs were recognized by the two-step labeling method. Similarly, we calculated that 37 ± 7% of GluA1- and 43 ± 5% of GluA3-containing AMPARs were recognized. However, considering the heterotetrameric formation of AMPAR subunits, we also needed to examine whether each subunit was covalently labeled with the probe or not. In this context, the immunoprecipitation assay in the denatured condition revealed that GluA2 and GluA3, but not GluA1, were covalently labeled with CAM2(TCO) (Supplementary Fig. 12b). This selectivity is consistent with our previous results26 in HEK293T cells. With regard to the efficacy of CAM2(TCO) labeling, the time-course of the labeling clearly indicated that chemical labeling occurred more efficiently at 37 °C than in the previous condition at 17 °C (Supplementary Fig. 13a). In addition, the concentration dependency of CAM2(TCO) revealed the EC50 value (0.90 ± 0.10 µM) of two-step labeling at 37 °C in neurons (Supplementary Fig. 13b).Using the two-step labeling technique in primary hippocampal neurons, fluorescently labeled AMPARs were visualized by confocal microscopy. At 5 min after the addition of Tz(Ax488), confocal live imaging showed punctate fluorescent signals from the CAM2(TCO)-treated neurons, and these signals were not observed in neurons co-treated with NBQX (Fig. 4b). To characterize the fluorescent signals in detail, Tz(Ax488)-treated neurons were fixed with paraformaldehyde (PFA) and immunostained with anti-MAP2 or anti-PSD95 antibodies for dendritic or postsynaptic staining, respectively. As shown in Fig. 4c and Supplementary Fig. 14a, labeled Alexa 488 signals were observed alongside the anti-MAP2 signals, and merged well with the anti-PSD95 signals. Considering the short incubation time with Tz(Ax488), the Alexa 488 signal likely corresponds to cell-surface AMPARs that are endogenously expressed in neurons. In addition, the endogenous AMPARs were successfully visualized using probes such as Alexa 647 or SeTau-647, using Tz(Ax647) or Tz(ST647), respectively (Supplementary Fig. 14b, c).Next, we quantified the surface distribution of AMPARs in neurons using fluorescence lifetime imaging microscopy (FLIM). In this method, the fluorescence decay curve in each pixel is analyzed by fitting it to a multi-exponential function, and the target fluorescence lifetime (τ) component can be extracted to quantify the fluorescence of interest without any background. Here, we used Tz(ST647) to visualize surface AMPARs because SeTau-647 has a unique fluorescence lifetime and high photostability39,40. The typical FLIM image for a lifetime fraction (τ = 2.4 ± 0.1 ns) corresponding to SeTau-647 is shown in Fig. 4d, which revealed that surface AMPARs in spines are 3.3 times more concentrated than those in dendrites.Trafficking analyses of endogenous AMPARs in neuronsAMPARs are dynamically regulated at synapses, which underlie activity-dependent neuronal plasticity. Molecular biology or biochemical methods, such as the genetic incorporation of fluorescent proteins, surface biotinylation assays, and metabolic incorporation of radioisotopes, have revealed the diffusion dynamics42, recycling process5, and half-life43,44 of AMPARs, respectively. Although powerful, these methods are highly specialized for analyzing each process (see the Discussion for detail). However, we now have a rapid method to selectively label cell-surface AMPARs in neurons under physiological temperature in culture medium. We therefore applied the two-step labeling method to analyze AMPAR trafficking over a long period.Prior to analyzing the AMPAR trafficking, we examined the influence of the two-step labeling process on the viability of primary cultured neurons by comparing it with our original CAM2 labeling. In immature neurons (day in vitro [DIV] 4), neither the original CAM2 labeling nor the two-step labeling affected neuronal viability after 24 h of labeling (Supplementary Fig. 15). However, in mature neurons (DIV 12), where iGluRs are highly expressed, viability was decreased after 3 h of labeling with the original CAM2 labeling, and neurons were severely damaged after 24 h of the original labeling. The main reason for the lowered viability may be ascribed to the experimental procedure rather than the CAM2 reagents, because similar neuronal damage was observed when the growth medium was exchanged for serum-free medium. In contrast, in the case of the two-step labeling method, where CAM2(TCO) was directly added into the growth medium at 37 °C, negligible damage was observed even after 24 h of the two-step labeling, indicating that this method is suitable for analyzing AMPARs over long periods in neurons. Besides, we examined the effects of CAM2(TCO) labeling on neuronal function. Neither the surface amount nor the synaptic fraction of AMPARs and NMDARs were affected by CAM2(TCO) labeling for 10 h (Supplementary Fig. 17a, b). Association between GluA2 and the accessary protein, TARPγ8 (ref. 45) was also unaffected by CAM2(TCO) labeling (Supplementary Fig. 17c). Although the homeostatic phosphorylation of ERK and CREB was not significantly but slightly decreased, phosphorylation levels of AMPAR (GluA1) and NMDAR (GluN1) were not influenced by the CAM2(TCO) labeling (Supplementary Fig. 17d).We then analyzed the trafficking of endogenous AMPARs in neurons over a long period using the two-step labeling method. As described in the previous section, confocal microscopic imaging was able to clearly visualize cell-surface AMPARs in neurons in the two-step labeling method (Fig. 4c). However, synapses are very narrow (200–500 nm diameter), and it is thus difficult to analyze AMPAR trafficking in detail using optical microscopy. We therefore applied biochemical methods for analyzing the t1/2surface of AMPARs, including remaining and recycled components (Fig. 5a). As shown in Fig. 5b, the t1/2surface was calculated to be 33.2 ± 5.1 h, which was markedly longer than the t1/2surface in HEK293T cells (5.3 ± 0.3 h).Fig. 5Quantitative analyses of AMPAR trafficking in neurons using the two-step labeling under the physiological cell culture condition.a Schematic illustration of the procedure for determining t1/2surface for AMPARs. b Determination of t1/2surface by western blotting. In left, representative results of western blotting are shown. In right, time-course of the labeled band is shown (n = 3 biological replicates). [CAM2(TCO)] = 2 μM, [Tz(Fl)] = 1 μM. c Determination of intracellular and surface ratio after CAM2(TCO) labeling for 10 h. In left, representative results of western blotting are shown. In right, band intensities for cell-surface and intracellular labeling were analyzed, both of which were normalized by that for whole-cell labeling (n = 3 biological replicates). See also Supplementary Fig. 19 for tetrazine ligation in serum containing medium or in cell lysate. [CAM2(TCO)] = 2 μM, [Tz(Fl)] = 1 μM. d, e Analyses of recycled AMPARs by pulse-chase-type analyses using the two-step labeling in neurons. In d, schematic illustration of the procedure is shown. In e, recycled AMPARs were analyzed by western blotting. In left, representative results of western blotting are shown. In right, exocytose AMPARs were quantified, which were normalized by that for surface labeling (n = 3 biological replicates). [CAM2(TCO)] = 2 μM, [Tz(Fl) or Tz(Ax647)] = 1 μM. Data are represented as mean ± s.e.m. f, g Trafficking and distribution of cell-surface AMPARs quantified by the two-step labeling in neuron (in f) and HEK293T cells (in g). Data are represented as mean ± s.e.m.To explore the difference in the t1/2surface values, we focused on the trafficking of internalized AMPARs. We analyzed the intracellular and surface percentages of TCO-AMPARs after 10 h of incubation with CAM2(TCO). The intracellular percentage (10.3 ± 1.7%) was markedly smaller than the surface percentage (90.4 ± 0.6%), and these values were largely similar to those obtained after 10 h of labeling in HEK293T cells (Fig. 5c and Supplementary Fig. 18). We next evaluated the recycling process of internalized AMPARs, as follows. The cultured neurons were labeled with CAM2(TCO) for 10 h in the culture medium at 37 °C, and Tz(Fl) was then added to the culture medium for 5 min as a first pulse to mask the surface TCO-AMPARs (Fig. 5d). The neurons were further incubated for 15 min in the culture medium at 37 °C, and then Tz(Ax647) was added to the culture medium as a second pulse to label the recycled TCO-AMPARs. As shown in Fig. 5e, western blotting using anti-Alexa 647 antibodies clearly detected recycled AMPARs. The percentage was determined to be 12.5 ± 2.8% compared with the surface AMPARs, suggesting that most of internalized AMPARs were recycled during this short period. In contrast, the recycled fraction was almost undetectable in HEK293T cells (Supplementary Fig. 20). These data indicate that AMPARs are constantly recycled via endocytosis and exocytosis in neurons, which may be the molecular basis for the long lifetimes of AMPARs (Fig. 5f, g).Two-step labeling and trafficking analyses of NMDARsThe NMDARs are another essential family of iGluRs, and form heterotetramers composed of GluN1, GluN2A–D, and GluN3 in neurons. Pharmacological features of NMDARs are different from those of AMPARs. GluN2A–D recognize glutamate, whereas GluN1 and GluN3 recognize glycine or D-serine, as endogenous ligands4. Here, we designed a ligand-directed two-step labeling reagent targeted for GluN2A using the selective antagonist perzinfotel (Pzf) as the ligand moiety (Fig. 6a)46. This reagent was termed “chemical NMDAR modification (TCO)” (CNM(TCO)).Fig. 6The two-step labeling of cell-surface NMDARs in HEK293T cells and neurons.a Chemical structure of CNM(TCO) for two-step labeling reagent for NMDARs. Gly glycine. b Two-step labeling of cell-surface NMDARs ectopically expressed in HEK293T cells. In left, schematic illustration is shown. In right, western blotting analyses after the two-step labeling are shown. HEK293T cells transfected with NR1-1 and NR2A or vector control were treated with 10 μM CNM(TCO) for 4 h followed by the addition of 1 μM Tz(Fl) for 5 min in the presence or absence of 250 μM Pzf in culture medium at 37 °C. The cell lysates were analyzed by western blotting using anti-fluorescein or anti-NR2A antibody. c Determination of t1/2surface of NMDARs in HEK293T cells by western blotting. In left, representative results of western blotting are shown. In right, time-course of the labeled band is shown (n = 3 biological replicates). d Two-step labeling of cell-surface NMDARs endogenously expressed in cultured cortical neurons. In left, schematic illustration is shown. In right, western blotting analyses after the two-step labeling are shown. The cultured cortical neurons were treated with 10 μM CNM(TCO) for 10 h followed by the addition of 1 μM Tz(Fl) for 5 min in the presence or absence of 250 μM Pzf in culture medium at 37 °C. The cell lysates were analyzed by western blotting using anti-fluorescein or anti-NR2A antibody. e Determination of t1/2surface of NMDARs in neurons by western blotting. In left, representative results of western blotting are shown. In right, time–course of the labeled band is shown (n = 3 biological replicates). Data are represented as mean ± s.e.m.To confirm the selective labeling of GluN2A, the two-step labeling method was first examined in HEK293T cells transiently co-expressed with GluN1 and GluN2A. Western blotting analyses of cell lysate showed a prominent band around 180 kDa, corresponding to the molecular weight of GluN2A (see lane 1 in Fig. 6b). The 180 kDa band was not detected under the control conditions (see lanes 2–5 in Fig. 6b and Supplementary Fig. 21), suggesting that this band corresponds to labeled GluN2A. The t1/2surface of cell-surface NMDARs was quantitatively analyzed using western blotting by focusing on the labeled GluN2A band. As shown in Fig. 6c, the t1/2surface was calculated to be 4.3 ± 0.4 h, which was slightly shorter than that of AMPARs in HEK293T cells.Next, we examined the chemical labeling of endogenous NMDARs in cultured neurons using CNM(TCO). As shown in Fig. 6d, western blotting showed a prominent band around 180 kDa, which was negligible in the co-presence of Pzf, indicating the successful labeling of endogenous NMDARs. Similar to AMPAR labeling, the time-course of NMDAR labeling clearly indicated that chemical labeling occurred more efficiently at 37 °C than at 17 °C (Supplementary Fig. 22a). However, the concentration dependency of CNM(TCO) for NMDAR labeling was different from that of CAM2(TCO) for AMPARs. As shown in Supplementary Fig. 22b, the labeled bands were not saturated in the 0–10 µM range, indicating that the affinity of CNM(TCO) was lower than that of CAM2(TCO). Importantly, CNM(TCO) labeled NMDAR with minimal disturbance to neuronal functions, including the constitutive phosphorylation of ERK and CREB (Supplementary Fig. 23). We next analyzed NMDAR trafficking using two-step labeling via biochemical methods in neurons. As shown in Fig. 6e, the t1/2surface of endogenous NMDARs in cultured neurons was 22.6 ± 7.2 h, which was substantially longer than in HEK293T cells. However, the value was shorter than that of endogenous AMPARs in neurons (33.2 ± 5.1 h). Thus, two-step labeling was successfully applied to NMDARs, another important family of iGluRs, by changing the selective ligands. In addition, we revealed that the lifetime of surface NMDARs is shorter than that of AMPARs in neurons.DiscussionWe described a bioorthogonal two-step labeling method for the selective modification of chemical probes to cell-surface AMPARs. This method can be used under neuronal culture conditions. In our previous method, fluorescein was directly labeled to cell-surface AMPARs using CAM2(Fl) in serum-free medium at 17 °C (ref. 26). Although the dynamics of cell-surface AMPARs have been successfully tracked over a short period after labeling, the present study revealed that this labeling condition affects the viability of mature neurons after 24 h of labeling. In contrast, the present bioorthogonal two-step labeling method negligibly affected the cell viability of neurons. Our method therefore allows the quantitative analysis of AMPAR trafficking for over 120 h after labeling. The present investigation also revealed that the homeostatic phosphorylation of ERK and CREB was slightly decreased by CAM2(TCO) but not CNM(TCO) treatment. Considering the high affinity of CAM2(TCO) for AMPAR, this influence may be reduced when neurons are treated with low concentration of CAM2(TCO).Many studies have focused on analyzing AMPAR trafficking in neurons using biochemical methods. To study trafficking over short periods (less than 30 min), surface biotinylation assays, where cleavable (disulfide-linked) biotin is randomly labeled on the cell surface, have conventionally been used. To date, both endocytosis and exocytosis of AMPARs have been investigated in cultured neurons using this method. However, this method is not suitable for analyzing AMPAR trafficking over a long period, because the biotinylation reaction must be conducted in serum-free medium at 4 °C; this affects neuronal viability, as we demonstrated in the present study. The long-term process of AMPAR trafficking, including decomposition, has been previously analyzed using the metabotropic incorporation of a radioisotope [35S]-labeled amino acid, or SILAC (stable isotope labeling with amino acids in cell culture), for mass spectrometry analyses43,44,47. Although isotope labeling can be conducted under cell culture conditions, these methods are not suitable for analyzing AMPAR trafficking over short periods. In contrast, our bioorthogonal two-step labeling selectively and rapidly modifies target iGluRs under cell culture conditions, and this allows us to analyze receptor trafficking over both short and long periods. Moreover, both biotinylation assays and metabotropic isotope labeling methods require the solubilization of iGluRs by mild detergents for pull-down or immunoprecipitation assays, because cellular proteins are randomly labeled in these methods. This step is problematic for the quantitative analysis of iGluRs, because iGluRs such as NMDARs are mainly localized in the postsynaptic density (PSD), where it is difficult for proteins to be solubilized by mild detergents. This peculiar feature hampers the quantitative analysis of iGluR trafficking by conventional biochemical methods. In contrast, purification steps are not required after labeling in our two-step labeling method. That is, in our method, all labeled proteins in the neurons can be analyzed quantitatively after denaturing in Laemmli sample buffer. This allows for the quantitative analysis of iGluRs contained in the PSD fraction.The selective visualization of cell-surface iGluRs is essential for analyzing their trafficking and distribution. One potential method involves the use of antibodies that selectively recognize the extracellular regions of iGluRs. However, these antibodies are very limited, and in most cases their selectivity is insufficient. Positron emission tomography imaging would be another candidate for visualizing native AMPARs48. However, this method would not be useful for trafficking studies due to its low resolution (1–2 mm). Here, we demonstrated that various kinds of chemical probes could be used to selectively and rapidly label endogenous cell-surface iGluRs in neurons using bioorthogonal two-step labeling. The selective labeling of SeTau-647, a middle-sized molecule with a long fluorescence lifetime and high photostability, allowed the quantitative analyses of cell-surface AMPAR distribution in neurons using FLIM. These analyses revealed a three-fold concentration of cell-surface AMPARs in spines compared with dendrites. As a future direction, cell-surface SeTau-647 labeling would be utilized for single molecule tracking40 or super resolution imaging of synaptic AMPARs. Moreover, the second step IEDDA reaction can be applied to label polymers or nanoparticles, to allow the distribution of iGluRs to be analyzed in more detail using electron microscopy in the future.Previous studies using radiolabeling methods indicated that the half-lives of synaptic proteins are 1–2 days43,44. Consistent with previous results, our present investigation demonstrated that the half-lives of AMPARs and NMDARs are 33.2 and 22.6 h, respectively, in cultured neurons. However, these values were significantly longer than those obtained in HEK293T cells. A plausible explanation for the differences between neurons and HEK293T cells may involve the formation of the macromolecular protein complexes of iGluRs. In the case of AMPARs, many binding partners such as transmembrane AMPAR regulatory proteins, synapse-associated protein 97 kDa, and glutamate receptor-interacting protein, which are selectively expressed in neurons, control the recycling of AMPARs and/or their stabilization at synapses45,49,50. Differences in phosphorylation levels may be another possible explanation. Activity-dependent phosphorylation by CaMKII or PKA contributes to the recycling and cell-surface insertion of AMPARs51,52. In addition, ubiquitination or deubiquitination via Nedd4 or USP46, respectively, may be considered53,54. In most cases, the contributions of these accessory proteins and post-translational modifications to AMPAR trafficking have been evaluated using genetic approaches, such as overexpression of an AMPAR subunit tagged with pH-sensitive SEP on the N-terminus. However, in some cases, complementary genetic experiments using knock-in or knock-out mice of the target gene have not supported the data19. Considering the heterotetrameric formation of AMPARs by the GluA1–4 subunits, and considering that each subunit has inherent roles, conflicting findings may be ascribed to the formation of non-native tetramers by the overexpression of single AMPAR subunits in neurons. In contrast, the present two-step labeling method can be used to visualize native iGluRs under physiological temperature in culture medium. Thus, this method can contribute to our understanding of the physiological and pathophysiological roles of iGluR trafficking in neurons.MethodsSynthesisAll synthesis procedures and compound characterizations are described in Supplementary Methods.General methods for biochemical and biological experimentsSDS-PAGE and western blotting were carried out using a BIO-RAD Mini-Protean III electrophoresis apparatus. Samples were applied to SDS-PAGE and electrotransferred onto polyvinylidene fluoride membranes (BIO-RAD), followed by blocking with 5% nonfat dry milk in Tris-buffered saline containing 0.05% Tween 20. Primary antibody was indicated in each experimental procedure, and anti-rabbit IgG-HRP conjugate (CST, 7074S, 1:3,000) or anti-mouse IgG-HRP conjugate (CST, 7076S, 1:3,000) was utilized as the secondary antibody. Chemiluminescent signals generated with ECL Prime (GE Healthcare) were detected with a Fusion Solo S imaging system (Vilber Lourmat).AnimalsPregnant ICR mice and pregnant Sprague Dawley rats maintained under specific pathogen-free conditions were purchased from Japan SLC, Inc (Shizuoka, Japan). The animals were housed in a controlled environment (23 ± 1 °C, 12 h light/dark cycle) and had free access to food and water, according to the regulations of the Guidance for Proper Conduct of Animal Experiments by the Ministry of Education, Culture, Sports, Science, and Technology of Japan. All experimental procedures were performed in accordance with the National Institute of Health Guide for the Care and Use of Laboratory Animals, and were approved by the Institutional Animal Use Committees of Kyoto University or Nagoya University.Expression of AMPARs or NMDARs in HEK293T cellsHEK293T cells (ATCC) were cultured in Dulbecco’s modified Eagle’s medium (DMEM)-GlutaMAX (Invitrogen) supplemented with 10% dialyzed FBS (Invitrogen), penicillin (100 units ml–1), and streptomycin (100 µg ml–1), and incubated in a 5% CO2 humidified chamber at 37 °C. Cells were transfected with a plasmid encoding rat GluA2 (GluA2flip(Q))26 or the control vector pCAGGS (kindly provided by Dr. H. Niwa from RIKEN) using Lipofectamine 2000 (Invitrogen) according to the manufacturer’s instructions, and subjected to labeling experiments after 36–48 h of the transfection. For NMDAR expression, cells were transfected with plasmids encoding rat GluN1-155 and rat GluN2A55, and 30 µM MK-801 (Funakoshi) was added to the culture medium to suppress cell death.Two-step labeling of AMPARs or NMDARs in HEK293T cellsFor the first step labeling of AMPARs, HEK293T cells transfected with GluA2 were treated with 2 µM CAM2(TCO) in the absence or presence of 50 µM NBQX in the culture medium at 37 °C for 4 h. For the second step labeling, the culture medium was removed, and 1 µM Tz(Fl) in PBS was added for 5 min at room temperature. To quench excess Tz(Fl), 1 µM TCO-PEG4-COOH in PBS was added.For western blot analyses of labeled AMPARs, labeled cells were washed three times with PBS, lysed with radio immunoprecipitation assay (RIPA) buffer containing 1% protease inhibitor cocktail (Nacalai tesque), and mixed with 5× Laemmli sample buffer containing 250 mM DTT. Western blotting analyses were performed as described in “General methods for biochemical and biological experiments.” The Fl-labeled GluA2 was detected using rabbit anti-fluorescein antibody (abcam, ab19491, 1:3,000). GluA2 was detected using a rabbit anti-GluA2/3 antibody (Millipore, 07-598, 1:3,000).In the case of two-step labeling of NMDARs, HEK293T cells transfected with GluN1-1 and GluN2A were treated with 10 µM CNM(TCO) in the absence or presence of 250 µM Pzf in the culture medium at 37 °C for 4 h. The second step labeling and subsequent western blotting were performed as described above. Immunodetection of GluN2A was performed with a rabbit anti-NR2A antibody (Millipore, 07-632, 1:1,000).CAM2(TCO), CNM(TCO) and Tz-probes were stored in DMSO solution. The stock solutions were kept in deep freezer (–80 °C) to prevent decomposition.Enzymatic deglycosylation of AMPARs expressed in HEK293T cellsGluA2-expressing HEK293T cells were labeled as described in “Two-step labeling of AMPARs or NMDARs in HEK293T cells.” The labeled cells were washed three times with PBS and lysed in PBS containing 1% triton X-100, 0.6% SDS, and 1% protease inhibitor cocktail for 30 min at 37 °C. The lysates were diluted (2.0-fold) in sodium phosphate buffer (50 mM, pH 7.5) containing 2% NP40 and 100 mM DTT. PNGase F (New England Biolabs) were used at 1,000 units/100 µl of lysate and incubated overnight at 37 °C. The samples were subjected to western blotting analyses as described in “Two-step labeling of AMPARs or NMDARs in HEK293T cells.” In this experiment, after western blotting using anti-Fl antibody, the membrane was stripped with stripping buffer (250 mM glycine (pH = 2.5) and 1% SDS) and reprobed with the anti-GluA2/3 antibody.Confocal live cell imaging of labeled AMPARs in HEK293T cellsHEK293T cells were co-transfected with GluA2flip(Q) and mCherry-F26 as a transfection marker. First step labeling was performed as describe in “Two-step labeling of AMPARs or NMDARs in HEK293T cells.” For the second step labeling, after removal of the culture medium, 100 nM Tz(Ax488) was treated for 5 min in HBS (20 mM HEPES, 107 mM NaCl, 6 mM KCl, 2 mM CaCl2, and 1.2 mM MgSO4 at pH 7.4) at room temperature and washed three times with HBS. Confocal live imaging was performed with a confocal microscope (LSM900, Carl Zeiss) equipped with a 63×, numerical aperture (NA) = 1.4 oil-immersion objective. Fluorescence images were acquired by excitation at 405, 488, 561, or 640 nm derived from diode lasers.For studying the reaction kinetics of tetrazine ligation to cell surface AMPARs, after first step labeling, cells were then incubated with 300 nM Tz(Ax488) at room temperature and imaged at specified time points by confocal microscopy. To quantify the fluorescence intensity of the membrane at each time point, mCherry-F positive cells (n = 6) were selected and the average signal intensity of ROIs set on the membrane was calculated by ZEN blue software (Carl Zeiss). After subtracting background fluorescence, the averaged membrane intensity was defined as F at each time point (FMAX was defined as F at 20 min). The membrane intensity was fitted with KaleidaGraph (Synergy software) using following equation (1): F = a + b (1 – e−ct).For trafficking analyses, after first step labeling, 100 nM Tz(Ax488) was added in DMEM-GlutaMAX for 5 min. The cells were washed three times in DMEM-GlutaMAX and incubated for 0, 1, 4, and 8 h in growth medium at 37 °C. Live cell imaging was performed with a confocal microscope. To quantify the fluorescence intensity of the membrane at each time point, mCherry-F positive cells were selected and the average signal intensity of ROIs set on the membrane was calculated by ZEN blue software after subtracting background fluorescence. The averaged membrane intensity was defined as F at each time point. The membrane intensity was fitted with KaleidaGraph using following equation (2): F = a + b·e−ct, and the offset value (a) was set equal to zero. The t1/2 was defined as t1/2 = ln(2)/c.For co-staining with LysoTracker Red DND-99 (Invitrogen), after first step labeling, the cells were treated with 50 nM LysoTracker Red DND-99 for 30 min at 37 °C in culture medium, and then treated with 100 nM Tz(Ax488) in culture medium for 5 min. After washing three times with culture medium or subsequent incubation for 8 h, cells were imaged using a confocal microscope.Half-life studies of AMPARs by western blotting in HEK293T cellsSchematic illustration of the experiments is shown in Fig. 3d. For determining t1/2surface (path-1 in Fig. 3d), first step labeling was conducted as described in “Two-step labeling of AMPARs or NMDARs in HEK293T cells.” After medium exchange for removal of the labeling reagents, the cells were incubated for 0, 1,4, 6, 12, 24, and 36 h. The cells were treated with 1 µM Tz(Fl) in PBS for 5 min, and excess Tz(Fl) was quenched by addition of 1 µM TCO-PEG4-COOH in PBS.For determining t1/2life (path-2 in Fig. 3d), after first step labeling, cells were washed three times with culture medium and treated with 1 µM Tz(Fl) in culture medium for 5 min. Excess Tz(Fl) was quenched by addition of 1 µM TCO-PEG4-COOH in culture medium. Cells were then incubated for 0, 1,4, 6, 12, 24, and 36 h and washed three times with PBS. Cell lysis and western blotting were performed as described in “Two-step labeling of AMPARs or NMDARs in HEK293T cells.”The target bands were manually selected, and the intensity were calculated with Fusion software (Vilber Lourmat), background intensity was manually subtracted by cutting the minimal intensity in the selected area. The half-life was calculated by curve fitting using KaleidaGraph and following equation (3): I = a + b·e−ct, and the offset value (a) was set equal to zero. The t1/2 was defined as t1/2 = ln(2)/c.Intracellular and surface ratio of labeled AMPARs in HEK293T cellsSchematic illustration of the experiments is shown in Fig. 3g. For determining labeled AMPARs on cell surface (path-1 in Fig. 3g), after first step labeling as describe in “Two-step labeling of AMPARs or NMDARs in HEK293T cells,” the cells were treated with 1 µM Tz(Fl) for 5 min in PBS at room temperature. To quench excess Tz(Fl), 1 µM TCO-PEG4-COOH in PBS was added and lysed with RIPA buffer containing 1% protease inhibitor cocktail for 30 min at 4 °C.For determining intracellular labeled AMPARs (path-2 in Fig. 3g), after first step labeling, 1 µM Tz(Ax647) was treated for 5 min for masking of cell-surface TCO-labeled AMPARs. After cell lysis using RIPA buffer, the lysate was reacted with 0.3 µM Tz(Fl) for 30 min at room temperature. Excess Tz(Fl) was quenched by addition of 1 µM TCO-PEG4-COOH in the cell lysate.For preparing whole-cell-labeling fraction (path-3 in Fig. 3g), after first step labeling, the cells were lysed with RIPA buffer containing 1% protease inhibitor cocktail for 30 min at 4 °C. The lysate was reacted with 0.3 µM Tz(Fl) for 30 min at room temperature. Excess Tz(Fl) was quenched by addition of 1 µM TCO-PEG4-COOH in the cell lysate.Western blotting was performed as described in “Half-life studies of AMPARs by western blotting in HEK293T cells.” The target bands were manually selected, and the intensity were calculated with ImageJ software, background intensity was manually subtracted by selecting a region with no bands from the same lane. In more detail, the band intensity was determined as described below:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( {{\mathrm{target}}\;{\mathrm{intensity}}} \right)-\left( {{\mathrm{target}}\;{\mathrm{area}}} \right)/\left( {{\mathrm{background}}\;{\mathrm{area}}} \right) \times \left( {{\mathrm{background}}\;{\mathrm{intensity}}} \right)$$\end{document}targetintensity−targetarea/backgroundarea×backgroundintensityPreparation of primary cortical neuronal cultureTwenty-four-well plates (BD Falcon) were coated with poly-D-lysine (Sigma-Aldrich), and washed with sterile dH2O three times. Cerebral cortices from 16-day-old ICR mouse embryos were aseptically dissected and digested with 0.25 w/v% trypsin (Nacalai tesque) for 20 min at 37 °C. The cells were re-suspended in Neurobasal Plus medium supplemented with 10% FBS, penicillin (100 units/ml), and streptomycin (100 µg/ml) and filtered by Cell Strainer (100 µm, Falcon) and centrifuged at 1,000 rpm for 5 min. The cells were re-suspended in Neurobasal Plus medium supplemented with 2% of B-27 Plus Supplement, 1.25 mM GlutaMAX I (Invitrogen), penicillin (100 units/ml), and streptomycin (100 µg/ml) and plated at a density of 2 × 105 cells on the 24-well plate. The cultures were maintained at 37 °C in a 95% air and 5% CO2 humidified incubator. Culture medium was replaced every 3 or 4 days and the neurons were used at 12–15 DIV.Two-step labeling of AMPARs or NMDARs in cultured neuronsTo label endogenous AMPARs, 12 µM CAM2(TCO) in 100 µl of the culture medium with or without 300 µM NBQX was gently added to the cortical neurons cultured in 500 µl medium on 24-well plates to a final concentration of 2 µM CAM2(TCO) and 50 µM NBQX. The cells were incubated for 10 h at 37 °C. For the second step, the culture medium was removed and the cells were treated with 1 µM Tz(Fl) for 5 min in PBS at room temperature. To quench excess Tz(Fl), 1 µM TCO-PEG4-COOH in PBS was added. Western blot analyses of labeled AMPARs were performed as described in “Two-step labeling of AMPARs or NMDARs in HEK293T cells.”To label endogenous NMDARs, 60 µM CNM(TCO) in 100 µl of growth medium with or without 1.5 mM Pzf was gently added to the cortical neurons cultured in 500 µl medium on 24-well plates to a final concentration of 10 µM CNM(TCO) and 50 µM Pzf. The cells were incubated for 10 h at 37 °C. The second step labeling and subsequent western blotting was performed as described above.Half-life studies of endogenous AMPARs or NMDARs in cultured neuronsSchematic illustration of the experiments is shown in Fig. 5a. For determining t1/2surface, after first step labeling as describe in “Two-step labeling of AMPARs or NMDARs in cultured neurons,” the cells were incubated for 0, 2, 4, 6, 12, 24, 48, 72, and 120 h. For the second step, the culture medium was removed and the cells were treated with 1 µM Tz(Fl) for 5 min in PBS at room temperature. To quench excess Tz(Fl), 1 µM TCO-PEG4-COOH in PBS was added and washed three times with PBS. Cell lysis and western blotting were performed as described in “Two-step labeling of AMPARs or NMDARs in HEK293T cells.” The immunodetection of GluA2 was conducted with a rabbit anti-GluA2 antibody (abcam, ab20673, 1:3,000). Quantification of the band intensity and calculation of the half-time was calculated described as in “Half-life studies of AMPARs by western blotting in HEK293T cells.”Intracellular and surface ratio of labeled AMPARs in cultured neuronsSchematic illustration of the experiments is shown in Fig. 3g. For determining labeled AMPARs on cell surface (path-1 in Fig. 3g), after first step labeling as describe in “Two-step labeling of AMPARs or NMDARs in cultured neurons,” the cells were treated with 1 µM Tz(Fl) for 5 min in PBS at room temperature. To quench excess Tz(Fl), 1 µM TCO-PEG4-COOH in PBS was added and lysed with RIPA buffer containing 1% protease inhibitor cocktail for 30 min at 4 °C.For determining intracellular labeled AMPARs (path-2 in Fig. 3g), after first step labeling, 1 µM Tz(Ax647) was treated for 5 min for masking of cell-surface TCO-labeled AMPARs. After cell lysis using RIPA buffer containing 1% protease inhibitor cocktail for 30 min at 4 °C, the lysate was reacted with 0.3 µM Tz(Fl) for 30 min at room temperature. Excess Tz(Fl) was quenched by addition of 1 µM TCO-PEG4-COOH in the cell lysate.For preparing whole-cell-labeling fraction (path-3 in Fig. 3g), after first step labeling, the cells were lysed with RIPA buffer containing 1% protease inhibitor cocktail for 30 min at 4 °C. The lysate was reacted with 0.3 µM Tz(Fl) for 30 min at room temperature. Western blotting was performed as described in “Two-step labeling of AMPARs or NMDARs in HEK293T cells.” Quantification of the band intensity and calculation of the ratio was conducted as described in “Intracellular and surface ratio of labeled AMPARs in HEK293T cells.”Quantification of recycled AMPARs in cultured neuronsThe first step labeling was performed as describe in “Two-step labeling of AMPARs or NMDARs in cultured neurons.” For the second step, the culture medium was removed and the cells were treated with 1 µM Tz(Fl) for 5 min in the culture medium at 37 °C. To quench excess Tz(Fl), 1 µM TCO-PEG4-COOH in the culture medium was added. After incubation at 37 °C for 15 min, recycled AMPARs were labeled with 1 µM Tz(Ax647) for 5 min in PBS. To quench excess Tz(Ax647), 1 µM TCO-PEG4-COOH in PBS was added. Cell lysis and western blotting were performed as described in “Two-step labeling of AMPARs or NMDARs in HEK293T cells” using anti-Alexa 647 antibody. Quantification of the band intensity and calculation of the ratio was conducted as described in “Intracellular and surface ratio of labeled AMPARs in HEK293T cells.”Anti-Alexa 647 antibody was prepared from the sera of a rabbit immunized with an antigen which was a conjugate of Alexa 647-NHS and KLH (Sigma), and the antibody was affinity-purified using Alexa 647-conjugated agarose. Alexa 647-conjugated agarose was prepared from CarboxyLink Coupling Resin (Thermo Fisher) and Alexa 647 NHS ester (Invitrogen). The anti-sera (1:2,000) or the purified antibody (1:1,000) was used for the western blotting.Preparation of primary hippocampal neuronal cultureGlass bottom dishes (IWAKI) or coverslips (diameter, 13 mm, Matsunami) were coated with poly-d-lysine (Sigma-Aldrich), and washed with sterile dH2O three times. Hippocampi from 18-day-old Sprague Dawley rat embryos were aseptically dissected and digested with 0.25 w/v% trypsin (Nacalai tesque) for 20 min at 37 °C. The cells were re-suspended in Neurobasal Plus medium supplemented with 10% FBS, penicillin (100 units/ml) and streptomycin (100 µg/ml) and filtered by Cell Strainer (100 µm, Falcon) and centrifuged at 1,000 rpm for 5 min. The cells were re-suspended in Neurobasal Plus medium supplemented with 2% of B-27 Plus Supplement, 1.25 mM GlutaMAX I (Invitrogen), penicillin(100 units/ml), and streptomycin (100 µg/ml) and plated at a density of 2 × 104 cells on glass coverslips inside 24-well plates (BD Falcon) or glass bottom dishes. Cultures were maintained at 37 °C in a 95% air and 5% CO2 humidified incubator. Culture medium was replaced every 7 days and the neurons were used at 16–18 DIV.Live cell imaging of AMPARs in cultured neuronsTo label endogenous AMPARs, 12 µM CAM2(TCO) in 300 µl of growth medium with or without 300 µM NBQX was gently added to the hippocampal neurons cultured in 1.5 ml medium on glass bottom dishes to a final concentration of 2 µM CAM2(TCO) and 50 µM NBQX. After removal of the culture medium, neurons were treated with 100 nM Tz(Ax488) for 5 min in HBS at room temperature and washed three times with HBS. Confocal live imaging was performed with a confocal microscope.Immunostaining of cultured neurons after labelingPrimary cultures of hippocampal neurons were labeled by 2 µM CAM2(TCO) and followed by 100 nM Tz(Ax488) as described above. The cells were fixed with 4% PFA in PBS at room temperature for 30 min and washed three times with PBS. PFA-fixed cells were permeabilized for 15 min with PBS containing 0.1% Triton X-100 at room temperature. The cells were washed three times in PBS and incubated in 10% normal goat serum for 1 h at room temperature. After blocking, the cells were incubated overnight at 4 °C with primary antibodies in PBS containing 1% normal goat serum. The cells were then washed three times with PBS and incubated for 1 h at room temperature with secondary antibodies in PBS containing 1% normal goat serum. The following primary antibodies were used: mouse anti-PSD95 (abcam, ab2723, 1:1,000) or rabbit anti-MAP2 (Millipore, AB5622, 1:1,000). Secondary antibodies were used goat anti-mouse Alexa 647 (abcam, ab150115, 1:1,000) and goat anti-rabbit Alexa633 (Invitrogen, A21070, 1:2,000). Imaging of immunostained hippocampal neurons was performed with a confocal microscope.Fluorescence lifetime imaging of AMPARs in cultured hippocampal neuronsPrimary cultures of hippocampal neurons were labeled by 2 µM CAM2(TCO), followed by 100 nM Tz(ST647) and fixed with 4% PFA in PBS. The cells were immunolabeled with PSD95 and MAP2 primary antibodies, and stained with Alexa 488 and Alexa405 secondary antibodies, respectively. Confocal and lifetime imaging of immunostained hippocampal neurons was performed by TCS SP8 FALCON (Leica microsystems) equipped with a white light laser and 63×, NA = 1.4 oil-immersion objective. SeTau-647 was imaged using 640 nm exc. (laser power 100, emission collected at 653–700 nm, using 0–12.5 ns time gate).FLIM images were processed in LAS X 3.5.5 software (Leica microsystems) to fit the lifetime decay curves using an n-exponential reconvolution model with the number of components that χ2 value is closest to 1. In our data, a three-component fit was utilized. The component of the lifetime corresponding to SeTau-647 (τ = 2.4 ± 0.1 ns) was used for calculating the intensity of FLIM images. The synaptic or dendric region was selected ROI on PSD95 or MAP2 signals and the background intensities were subtracted by selecting a region of no cells.Statistics and reproducibilityAll graphs were generated using Microsoft Excel. All data are expressed as mean ± s.e.m. We accumulated the data for each condition from at least three independent experiments. We evaluated statistical significance with Student’s t-test for comparisons between two mean values. A value of P < 0.05 was considered significant.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Neurophysiology",
"Chemical modification",
"Chemical tools",
"Ligand-gated ion channels"
] | central ionotropic glutamate receptors (iGluRs mediate excitatory neurotransmission categorized classes structural homology α-amino-3-hydroxy-5-methyl-4-isoxazole-propionate (AMPA receptor (GluA1–4) kainate (GluK1–5) N-methyl-D-aspartate (NMDA receptor (GluN1 δ receptors iGluRs assemble tetramers functional receptors formed subunits class receptors permeable to monovalent cations (Na+ K+) mediate majority excitatory synaptic transmission form heterotetramers subunit compositions dependent brain regions hippocampal CA1 neurons majority AMPARs GluA1/A2 GluA2/A3 subunit combinations small GluA1 AMPARs cycled postsynaptic membrane endocytosis exocytosis regulation critical for synaptic plasticity learning memory development AMPARs kainate receptors activated by glutamate binding NMDA receptors+ require depolarization agonist binding require two GluN1 subunits GluN2 GluN2 GluN3NMDAR-dependent Ca2+ influx triggers intracellular signal transduction precise targeting to synapses essential for neuronal connectivity understand molecular mechanisms learning memory critical analyze membrane localization trafficking of iGluRs.Biochemical approaches surface biotinylation assays used analyze membrane protein localizations applied to AMPARs57 cell-surface proteins randomly labeled with biotin purification of biotin-labeled AMPAR required hampers analyses trafficking visualize glutamate receptors fluorescent proteins fused to receptors genetically encoded approaches pH-sensitive variant GFP fused to extracellular receptors SNAP- Halo-tags fused to receptors for labeling small chemical probes downsizing protein tags demonstrated short peptide tag) probe genetic code expansion bioorthogonal click chemistry reported for fluorescent labeling of iGluRs in HEK293T cells chemical probes attached to side unnatural amino acid residue16 genetically encoded approaches used in trafficking studies iGluRs especially AMPARs rely on overexpression of target iGluR subunitsformation heterotetramers iGluRs overexpression localization trafficking native iGluRs neurons endogenously expressed iGluRs tagged with small chemical probes18 situ chemical protein labeling analyzing native proteins live cells Affinity-based protein labeling selective modification target traceless affinity labeling reported ligand-directed acyl imidazole (LDAI) chemistry24 small chemical probes attached to nucleophilic amino acid residues near ligand-binding site developed AMPAR-selective LDAI reagent AMPAR modification chemical probes to AMPARs endogenously expressed in cultured neurons prepared brain slices26 restrictions for visualizing cell AMPARs live cells at low temperatures during CAM2 labeling suppress internalization labeled neuronal culture medium for serum-free medium or buffered saline decrease non-specific labeling serum proteins long-term exposure (1–4 h non-physiological conditions interfere neuronal activity neurons under physiological conditions during chemical labelingmethod rapid selective labeling AMPARs temperature culture medium combining LDAI protein labeling inverse electron Diels–Alder reaction two-step labeling allows analyses distribution trafficking endogenous AMPARs cultured neurons technique study trafficking endogenous neurons labeling AMPARs bioorthogonal two-step two-step labeling technique LDAI protein labeling IEDDA reaction rapid selective modification chemical probes cell-surface iGluRs first strained alkene attached iGluRs LDAI chemistry acyl substitution reaction selective ligand–protein recognition labeled alkene group modified tetrazine-conjugated probes (Tz-probes cell surface high selectivity reaction rate IEDDA reaction step Fig. 1Rapid labeling cell-surface iGluRs ligand-directed two-step labeling two-step labeling strained alkene attached iGluRs LDAI chemistry 2nd step Tz probe tethered IEDDA reaction Lg ligand iGluRs Nu nucleophilic amino acid residue Glu glutamate Chemical structure CAM2 structure Tz probes detailed chemical structures Supplementary Figillustration two-step labeling live cell selective labeling strained alkene AMPARs first step designed CAM2 reagent trans-cyclooctene termed CAM2(TCO) (Fig. TCO selected fast kinetics IEDDA reaction original CAM2 reagents aromatic fluorophores ethylene glycol linker added reactive acyl imidazole unit TCO group CAM2 increase hydrophilicity Hydrophobic aromatic groups affinity albumin improvement decreases undesired labeling albumin allows chemical labeling AMPARs cell culture medium serum first labeling physiological temperature (37 °C). labeled AMPARs internalized not problematic two-step labeling chemical probes tethered cell-surface AMPARs second step reaction. second step IEDDA reaction rate chemical structure tetrazine group selected monoaryl tetrazine fast reaction rate high bioorthogonality prepared cell-impermeable Tz-probes hydrophilic anionic fluorophores biotin cell-surface labeling 1c Shortening reaction time probe labeling contributes cell-surface specific labeling decreases adsorption chemical probes cells culture dishes coverslipstetrazine-fluorophore conjugates-on” response IEDDA high signal-to-noise ratio fluorescence imaging labeling surface AMPARs HEK293T two-step labeling method examined HEK293T cells expressing GluA2 AMPARs first step CAM2(TCO) added culture medium 10% fetal bovine serum incubated 37 °C 4 h second step reaction 5 min membrane-impermeable Tz(Fl) for fluorescein labeling cell 2a western blotting cell lysate anti-fluorescein antibodies showed strong band around 110 kDa band not observed cells co-treated competitive ligand) control conditions anti-Fl signal highest signal among multiple bands anti-GluA2/3 antibodies GluA2 bands converged into single lower band after treatment peptide-N-glycosidase F GluA2 highly glycosylated with N-linked PNGase F-treated samples shifted anti-GluA2 band merged with anti-Fl signal highly glycosylated GluA2 selectively labeled methods direct fluorescein labeling original CAM2(Fl) conditionsstrong band 70 110 kDa band (lanes 6–7 Fig. 2a 3) 70 kDa band intensity NBQX corresponds albumin serum indicate high selectivity two-step labeling CAM2(TCO) original CAM2(Fl) two labeling cell-surface AMPARs HEK293T cells Western blotting analyses HEK293T cells transfected GluA2flip(Q) treated 2 μM CAM2(TCO) 4 h 1 μM Tz(Fl) 5 min 2 μM CAM2(Fl) 4 h 50 μM NBQX culture medium 37 °C cell lysates analyzed blotting anti-fluorescein anti-GluA2/3 antibody band intensity Supplementary Fig. 3. CAM2(Fl) added serum free medium bands around 70 kDa lane #6 #7 disappear Effects PNGase F treatment western blotting AMPAR HEK293T cells GluA2. anti-Fl anti-GluA2/3 image lane #3 #4 Two-step labeling PNGase F (1000 units/100 μL) added cell lysate MethodsConfocal live imaging HEK293T cells labeled 2 μM CAM2(TCO) 0.1 μM Tz(Ax488). Labeling mCherry-F transfection marker Scale bars 10 μm Reaction tetrazine ligation live cells HEK293T 2 μM CAM2(TCO) 0.3 μM Tz(Ax488) 37 °C confocal images Scale 20 μm time-course fluorescent intensity Alexa 488 (n = 6 Data mean ± s.e.m fluorescently labeled AMPARs confocal microscopic live imaging two-step labeling cell culture Tz(Ax488) used second step labeling Alexa 488 bright fluorescence acidic conditions weakened fluorescence acidic Alexa 488 suitable cellular distribution trafficking labeled AMPARs imaging fluorescence observed cell surface cells co-transfected mCherry-F not observed control conditions CAM2(TCO)-untreated NBQX-co-treated cells direct Alexa 488 labeling CAM2(Ax488) labeled signals observed cell surface intracellular spacetwo-step labeling technique superior for fluorescent visualization cell-surface AMPARs under culture conditions determined reaction kinetics rapid fluorophore labeling AMPARs turn-on fluorescent property Tz(Ax488) IEDDA reaction 5) after adding Tz(Ax488), fluorescent signals observed from cells co-transfected with marker mCherry-F saturated within 3 min AMPARs labeled by fluorophore fast kinetics high bioorthogonality IEDDA reaction AMPARs labeled with chemical probes small to middle-sized SeTau-647 rotaxane dye high photostability long fluorescence lifetime39 2 6) probe flexibility two-step labeling superior direct labeling original CAM2 reagents labeling efficacy remaining unlabeled GluA2 showed 35 ± 3% surface AMPARs visualized in two-step labeling 7) AMPAR function not affected by two-step labeling 8) consistent previous analyses minimal disturbance AMPAR properties by CAM2 AMPAR trafficking in HEK293T analyzed receptor trafficking using live imaging biochemical approachesanalyzed cell-surface AMPAR trafficking HEK293T cells confocal imaging cells with CAM2(TCO) temperature Tz(Ax488) added visualize AMPARs (0–8 h labeled fluorescence cell surface decreased after incubation at 37 °C Fluorescent granules observed intracellular after 8 h half-time of cell-surface AMPARs 5.7 ± 0.7 h from fluorescent intensity intracellular signals merged with fluorescent lysosome marker) after 8 h internalized AMPARs transported to lysosomes Similar internalization behavior AMPARs labeled with fluorophores Tz(Ax647) or Tz(ST647).Fig. analyses AMPAR trafficking in HEK293T cells two-step labeling culture illustration trafficking analyses cell-surface AMPARs confocal microscopy Time-lapse confocal imaging HEK293T cells after two-step labeling CAM2(TCO) Tz(Ax488). cells transfected with GluA2flip(Q). confocal imagesfluorescent intensity cell surface 3 [CAM2(TCO) 2 μM [Tz(Ax488) 0.1 μM Co-staining labeled AMPARs LysoTrackerTM HEK293T cells incubated LysoTrackerTM Tz(Ax488 labeling 8 h incubation 37 °C confocal live imaging Pearson’s correlation coefficients Scale bars 10 μm Determination t1/2surface t1/2life AMPARs western blotting Schematic procedure f t1/2surface t1/2life determined results western blotting time-course labeled band 3 [CAM2(TCO)] 2 μM [Tz(Fl)] 1 μM Determination TCO-labeled AMPARs cell-surface intracellular whole-cell western blotting intracellular surface ratio CAM2(TCO) labeling 4 h results western blotting band intensities cell-surface intracellular labeling normalized whole-cell labeling 3 replicates). [CAM2(TCO)] 2 μM [Tz(Fl)] = 1 μM Data mean s.e.m.Quantitative analyses cell-surface AMPARs biochemical approachest1/2surface AMPARs HEK293T cells incubated (0–36 h after treatment CAM2(TCO) Fig. Tz(Fl) added modification fluorescein cell-surface AMPARs western blotting t1/2surface 5.3 ± 0.3 h (Fig. similar confocal imaging (Fig. half-time degradation (t1/2life) cell receptors evaluated protocols Tz(Fl) added after CAM2(TCO labeling (path-2 Fig. cells incubated (0–36 lysates subjected western blotting Fig. 3f t1/2life 8.1 0.7 h slightly longer t1/2surface AMPARs (p < 0.05) internalized AMPARs likely decomposed lysosomal degradation HEK293T cells determined intracellular versus surface percentages TCO-labeled AMPARs fate cell-surface AMPARs AMPARs HEK293T cells treated CAM2(TCO labeled fluorescein cell-impermeable Tz(Fl) medium live conditions (path-1 Fig. surface TCO-AMPARs masked cell-impermeable Tz(Ax647) (path-2After lysis Tz(Fl) added lysate intracellular TCO-AMPARs whole-cell-labeling TCO-AMPARs fluorescein prepared Tz(Fl) after lysis Fig. investigated second step reaction Tz(Fl) rapidly in cell lysate blotting analyses covalent modification fluorescein selective to AMPARs lysate reaction saturated after 15 min 0.1 or 0.3 μM Tz(Fl) added intracellular-surface percentages TCO-AMPARs after 4 h incubation CAM2(TCO) 4.1 ± 0.9% 94.4 ± 3.3% intracellular TCO-AMPAR levels low.Rapid labeling endogenous AMPARs in examined bioorthogonal two-step labeling technique rapid modification cell-surface AMPARs Primary cultured neurons incubated with CAM2(TCO) 10 h Tz(Fl) added 5 min for cell-surface labeling Western blotting analyses lysate showed single strong band molecular weight AMPARs band not detected in co-presence competitive ligand NBQX control conditions AMPARs in HEK293T cellssmeared bands detected anti-GluA2 antibodies anti-Fl band highest removal N-linked sugars PNGase treatment anti-GluA2 bands converged lower band merged anti-Fl signal results suggest glycosylated endogenous AMPARs labeled fluorescein rapid labeling two-step labeling cell-surface AMPARs neurons Western blotting analyses cortical neurons neurons treated 2 μM CAM2(TCO) 10 h 1 μM Tz(Fl) 5 min 50 μM NBQX culture medium 37 °C cell lysates analyzed anti-fluorescein anti-GluA2 antibody Confocal live imaging neurons labeled 2 μM CAM2(TCO) 0.1 μM Tz Labeling 2 Immunostaining neurons after-step labeling neurons fixed permeabilized immunostained anti-MAP2 anti-PSD95 antibody Scale bars 5 μm Pearson’s correlation coefficients images Supplementary Fig. 14a FLIM imaging analyses cell-surface AMPARs after-step labeling neuronsleft confocal FLIM image lifetime fraction = 2.4 ± 0.1 ns Scale bars 5 μm right FLIM intensities spine dendrite analyzed = 4 [CAM2(TCO) 2 μM [Tz(Ax488) 0.1 μM difference (p < 0.05 t-test p = 0.048) Data mean ± s.m AMPAR subunits (GluA1–4) GluA1 GluA2 GluA3 expressed cultured cortical examined visualizing tetrameric AMPARs quantifying unlabeled GluA2 fraction 12a 44 ± 4% GluA2-containing AMPARs recognized two-step labeling 37 ± 7% GluA1- 43 ± 5% GluA3-containing AMPARs recognized heterotetrameric formation labeled immunoprecipitation assay GluA2 GluA3 not GluA1 covalently labeled CAM2(TCO) selectivity consistent previous HEK293T cells CAM2(TCO) labeling labeling efficiently 37 °C 17 °C concentration dependency CAM2(TCO) EC50 value (0.90 ± 0.two-step labeling 37 °C neurons Fig.-step labeling hippocampal neurons fluorescently labeled AMPARs visualized confocal microscopy 5 min addition Tz imaging showed fluorescent signals CAM2(TCO)-treated neurons not observed neurons co-treated NBQX (Fig. Tz(Ax488)-treated neurons fixed with paraformaldehyde immunostained anti-MAP2 anti-PSD95 antibodies staining Fig. 4c Fig 14a Alexa 488 signals observed anti-MAP2 merged anti-PSD95 signals short incubation time Tz signal corresponds cell-surface AMPARs endogenously expressed endogenous AMPARs visualized using Alexa 647 SeTau-647 Tz(Ax647) Tz(ST647), Fig. 14b quantified surface distribution AMPARs fluorescence lifetime imaging microscopy fluorescence decay curve analyzed multi-exponential function target fluorescence lifetime (τ) component extracted quantify used Tz(ST647) visualize surface AMPARs unique fluorescence lifetime hightypical FLIM image lifetime fraction (τ = 2.4 ± 0.1 ns) SeTau-647 Fig. 4d surface AMPARs spines 3.3 times concentrated dendrites.Trafficking analyses endogenous AMPARs neuronsAMPARs regulated at synapses underlie activity-dependent neuronal plasticity Molecular biology methods genetic surface biotinylation assays metabolic incorporation radioisotopes revealed diffusion recycling half-life43 AMPARs methods specialized rapid method selectively label cell-surface AMPARs neurons physiological temperature culture medium applied two-step labeling method analyze AMPAR trafficking long period examined influence two-step labeling viability primary cultured neurons original CAM2 labeling neurons original CAM2 two-step labeling affected viability after 24 h labeling mature neurons (DIV iGluRs highly expressed viability decreased after 3 h labeling original neurons severely damaged after 24 h lowered viability experimental procedure similar neuronal damage growth medium exchanged for serum-free mediumtwo-step labeling method CAM2(TCO) added into growth medium at 37 °C negligible damage observed after 24 h suitable for analyzing AMPARs long periods examined effects CAM2 labeling on neuronal function surface amount synaptic fraction of AMPARs NMDARs affected for 10 h Association between GluA2 protein TARPγ8) unaffected by homeostatic phosphorylation of ERK CREB slightly decreased levels AMPAR (GluA1) NMDAR (GluN1) not influenced CAM2 labeling analyzed trafficking of endogenous AMPARs in neurons long two-step labeling confocal microscopic imaging cell-surface AMPARs synapses narrow (200–500 nm difficult to analyze AMPAR trafficking optical microscopy applied biochemical methods analyzing t1/2surface of AMPARs t1/2surface 33.2 ± 5.1 h longer than HEK293T cells (5.3 ± 0.3 h).Fig analyses of AMPAR trafficking in neurons two-step labeling physiological cell culture conditionprocedure t1/2surface AMPARs western blotting results time-course labeled band = 3 [CAM2(TCO)] = 2 μM [Tz(Fl)] = 1 μM intracellular surface ratio after CAM2(TCO) labeling 10 h results band intensities cell-surface intracellular labeling analyzed normalized whole-cell labeling 3 Supplementary Fig. 19 tetrazine ligation serum cell lysate [CAM2(TCO)] = 2 μM [Tz(Fl)] = 1 μM Analyses recycled AMPARs pulse-chase-type two-step labeling neurons procedure recycled AMPARs analyzed western blotting results exocytose AMPARs quantified normalized surface labeling 3 replicates). [CAM2(TCO)] = 2 μM [Tz(Fl)(Ax647)] = 1 μM Data represented mean s.e.m. Trafficking distribution cell-surface AMPARs two labeling neuron HEK293T cells mean ± s.e t1/2surface trafficking internalized AMPARs analyzed intracellular surface percentages TCO-AMPARs after 10 h incubation CAM2(TCO). intracellular percentage± 1.7%) smaller than surface percentage (90.4 ± 0 values similar after 10 h labeling HEK293T cells (Fig. 5c evaluated recycling internalized AMPARs cultured neurons labeled with CAM2(TCO) 10 h 37 °C Tz(Fl) added 5 min mask surface TCO-AMPARs incubated 15 min 37 °C Tz(Ax647) added second label recycled TCO-AMPARs Fig. 5e western blotting anti-Alexa 647 antibodies detected recycled AMPARs percentage 12.5 ± 2.8% compared surface AMPARs most AMPARs recycled recycled undetectable in HEK293T cells AMPARs recycled via endocytosis exocytosis long lifetimes (Fig. 5f g).Two-step labeling trafficking analyses family iGluRs form heterotetramers GluN1 GluN2A–D GluN3 neurons Pharmacological features different AMPARs GluN2A–D recognize glutamate GluN1 GluN3 recognize glycine D-serinedesigned ligand-directed two-step labeling reagent GluN2A antagonist perzinfotel (Pzf) ligand (Fig. 6a reagent “chemical NMDAR modification (TCO two-step labeling cell-surface NMDARs HEK293T cells neurons Chemical structure CNM(TCO)-step labeling reagent Two-step labeling cell-surface NMDARs ectopically HEK293T cells blotting analyses cells transfected NR1-1 NR2A treated 10 μM CNM(TCO) 4 h 1 μM Tz(Fl) 5 min 250 μM Pzf 37 °C cell lysates analyzed western blotting anti-fluorescein anti-NR2A antibody Determination t1/2surface NMDARs HEK293T cells western blotting results time-course labeled band = 3 Two-step labeling cell-surface NMDARs endogenously expressed cultured cortical neurons treated 10 μM CNM(TCO) 10 h 1 μM Tz(Fl) 5 min 250 μM Pzf 37 °C cell lysates analyzed western blotting anti-fluorescein anti-NR2A antibodyDetermination t1/2surface NMDARs neurons western blotting left results right time–course labeled band = 3 replicates). Data mean ± s.e.m selective labeling GluN2A two-step labeling method examined HEK293T cells co-expressed GluN1 GluN2A Western blotting analyses band around 180 kDa molecular weight GluN2A lane 1 Fig 180 kDa band not detected control conditions lanes 2–5 Fig 6b band corresponds labeled GluN2A t1/2surface NMDARs analyzed labeled GluN2A band Fig 6c t1/2surface 4.3 ± 0.4 h shorter AMPARs HEK293T cells examined chemical labeling endogenous NMDARs cultured neurons CNM(TCO). Fig. 6d western blotting showed band around 180 kDa negligible co-presence Pzf successful labeling time-course NMDAR labeling 37 °C 17 °C Fig. 22a). concentration dependency CNM(TCO) different CAM2(TCO) AMPARslabeled bands not saturated 0–10 μM affinity CNM(TCO) lower than CAM2(TCO). CNM(TCO) labeled NMDAR minimal disturbance neuronal functions phosphorylation ERK CREB analyzed NMDAR trafficking two-step labeling neurons Fig. 6e t1/2surface endogenous NMDARs cultured neurons 22.6 ± 7.2 h longer than HEK293T cells shorter endogenous AMPARs neurons (33.2 ± 5.1 h). two-step labeling applied to NMDARs iGluRs changing selective ligands lifetime surface NMDARs shorter AMPARs neurons bioorthogonal two-step labeling method chemical probes cell-surface AMPARs neuronal culture conditions fluorescein labeled AMPARs CAM2(Fl) serum-free medium at 17 °C labeling affects viability neurons after 24 h two-step labeling negligibly cell viability allows quantitative analysis AMPAR trafficking 120 h after labeling homeostatic phosphorylation of ERK CREB slightly decreased by CAM2(TCO) not CNM(TCO) treatmenthigh affinity CAM2(TCO) for AMPAR influence treated low concentration CAM2 studies analyzing AMPAR trafficking neurons biochemical methods trafficking short periods 30 surface biotinylation assays cleavable biotin labeled cell surface used endocytosis exocytosis of AMPARs investigated in cultured neurons not suitable for AMPAR trafficking long period biotinylation reaction in serum-free medium at 4 °C affects neuronal viability long process AMPAR trafficking decomposition analyzed using metabotropic incorporation radioisotope [35S]-labeled amino acid for spectrometry isotope labeling cell culture conditions not suitable for analyzing AMPAR trafficking short periods bioorthogonal two-step labeling modifies target iGluRs cell culture allows analyze receptor trafficking short long periods biotinylation assays metabotropic isotope labeling require solubilization iGluRs by mild detergents proteins randomly labeled problematic for quantitative analysis iGluRs localized in postsynaptic density difficult by mild detergentsfeature hampers analysis iGluR trafficking biochemical methods purification not required after labeling in two-step labeling method all labeled proteins analyzed after denaturing Laemmli sample buffer allows quantitative analysis iGluRs in PSD selective visualization cell-surface iGluRs essential trafficking distribution potential method antibodies extracellular regions iGluRs antibodies limited selectivity insufficient Positron emission tomography imaging candidate not useful low resolution (1–2 demonstrated chemical probes label endogenous cell-surface iGluRs neurons bioorthogonal two-step labeling selective labeling SeTau-647 middle-sized molecule long fluorescence lifetime high photostability allowed analyses cell-surface AMPAR distribution revealed three-fold concentration AMPARs in spines compared with dendrites future SeTau-647 labeling for single molecule super resolution imaging synaptic AMPARs second step IEDDA reaction label polymers nanoparticles distribution iGluRs electron microscopy studies radiolabeling half-lives of synaptic proteins 1–2investigation demonstrated half-lives of AMPARs NMDARs 33.2 22.6 h in cultured neurons values longer than HEK293T cells explanation differences neurons cells may formation macromolecular protein complexes iGluRs binding partners transmembrane AMPAR regulatory proteins synapse-associated protein 97 kDa glutamate receptor-interacting protein control recycling stabilization at synapses45 Differences in levels explanation Activity-dependent phosphorylation by CaMKII PKA contributes to recycling cell-surface insertion AMPARs51 ubiquitination deubiquitination via Nedd4 or USP46 may contributions accessory proteins post-translational modifications to AMPAR trafficking evaluated using genetic approaches overexpression AMPAR subunit pH-sensitive genetic experiments not supported heterotetrameric formation AMPARs by GluA1–4 subunits conflicting findings may formation non-native tetramers overexpression single AMPAR subunits two-step labeling method native iGluRs under physiological temperature in culture medium understanding physiological roles iGluR trafficking in neuronsprocedures compound characterizations Supplementary Methods methods biochemical biological experimentsSDS-PAGE western blotting BIO-RAD Mini-Protean III electrophoresis apparatus Samples SDS-PAGE electrotransferred polyvinylidene fluoride membranes 5% nonfat dry milk Tris-buffered saline 0.05% Tween 20. Primary antibody anti-rabbit IgG-HRP anti-mouse IgG-HRP secondary antibody Chemiluminescent signals ECL Prime detected Fusion Solo S imaging system ICR mice Sprague Dawley rats pathogen-free purchased from Japan SLC, controlled environment (23 ± 1 °C 12 h light/dark cycle free access food water procedures National Institute of Health Guide Animals approved by Institutional Animal Use Committees Kyoto University Nagoya University AMPARs NMDARs HEK293T cultured in Dulbecco’s modified Eagle’s medium)-GlutaMAX (Invitrogen) 10% dialyzed FBS (Invitrogen), penicillin (100 streptomycin (100 incubated in 5% CO2 humidified chamber at 37 °C.Cells transfected GluA2 control pCAGGS Dr. H. Niwa RIKEN Lipofectamine 2000 labeling experiments after 36–48 h NMDAR expression cells transfected plasmids GluN1-155 GluN2A55 30 μM MK-801 added suppress death-step labeling AMPARs NMDARs HEK293T first cells transfected GluA2 treated 2 μM CAM2(TCO) 50 μM NBQX 37 °C 4 h second step culture medium removed 1 μM Tz(Fl) PBS added 5 min excess 1 μM TCO-PEG4-COOH PBS added analyses AMPARs cells washed three times PBS lysed radio buffer 1% protease inhibitor cocktail mixed 5× Laemmli sample buffer 250 mM DTTWestern blotting methods Fl-labeled GluA2 detected rabbit anti-fluorescein antibody rabbit anti-GluA2/3 antibody 07-598 two-step labeling HEK293T cells transfected GluN1-1 GluN2A treated 10 μM CNM(TCO) 250 μM Pzf 37 °C 4 h second step labeling western blotting Immunodetection GluN2A rabbit anti-NR2A antibody 07-632 1:1,000).CAM2 CNM(TCO Tz-probes stored DMSO solution freezer (–80 °C decomposition deglycosylation AMPARs HEK293T cellsGluA2-expressing HEK293T cells labeled cells washed lysed 1% triton X-100 0.6% SDS 1% protease inhibitor cocktail 30 min 37 °C lysates diluted sodium phosphate buffer 7.5 2% NP40 100 mM DTT PNGase F 1,000 units/100 μl lysate incubated overnight 37 °Csamples subjected western blotting-step labeling HEK293T membrane stripped buffer (250 mM glycine 2.5 1% SDS reprobed anti-GluA2/3 antibody cell imaging AMPARs HEK293T co with GluA2flip(Q) mCherry-F26 First step labeling second step labeling 100 nM Tz(Ax488) treated 5 min HBS (20 mM HEPES 107 mM NaCl 6 mM KCl 2 mM CaCl2 1.2 mM MgSO4 pH 7.4) room temperature washed three times HBS Confocal imaging microscope (LSM900 63× 1.4 oil-immersion objective Fluorescence images excitation 405 561 640 nm diode lasers reaction kinetics tetrazine ligation cell surface AMPARs cells incubated with 300 nM Tz(Ax488) room temperature imaged confocal microscopyfluorescence intensity mCherry-F positive cells (n = 6) selected average signal intensity calculated ZEN blue software averaged membrane intensity defined F each (FMAX F 20 min). membrane intensity fitted with KaleidaGraph F = a + b (1 – e−ct).For trafficking analyses 100 nM Tz(Ax488) added DMEM-GlutaMAX 5 min cells washed three times incubated 0 1 4 8 h growth medium 37 °C Live cell imaging confocal microscope mCherry-F positive cells selected average signal intensity calculated averaged membrane intensity defined F each fitted KaleidaGraph equation F = a + b·e−ct offset value (a) equal zero t1/2 defined = ln(2)/c co-staining LysoTracker Red DND-99 cells treated 50 nM LysoTracker DND-99 30 min 37 °C treated 100 nM Tz(Ax488) 5 min. washing three times incubation 8 h cells imaged confocal microscope.Half-life studies AMPARs western blotting in HEK293T Fig. 3d.determining t1/2surface (path-1 Fig. first step labeling-step labeling HEK293T medium exchange cells incubated 0 1,4 6 12 24 36 h treated 1 μM Tz(Fl) PBS 5 min excess Tz(Fl) quenched 1 μM TCO-PEG4-COOH PBS determining t1/2life (path-2 Fig. labeling cells washed three times treated 1 μM Tz(Fl) 5 min Excess Tz(Fl quenched 1 μM TCO-PEG4-COOH Cells incubated 0 1,4 6 12 24 36 h washed three times PBS Cell lysis western blotting target bands selected intensity calculated Fusion software background intensity subtracted minimal intensity half-life calculated curve fitting KaleidaGraph equation I = a + b·e−ct offset value (a) equal zero t1/2 defined = ln(2)/c.Intracellular surface ratio labeled AMPARs HEK293T experiments Fig. 3g determining labeled AMPARs cell surface (path-1 labeling cells treated 1 μM Tz(Fl) 5 min PBS room temperatureexcess Tz 1 μM TCO-PEG4-COOH PBS lysed RIPA buffer 1% protease inhibitor 30 min 4 °C determining intracellular AMPARs 1 μM Tz(Ax647) treated 5 min masking cell-surface TCO-labeled AMPARs reacted 0.3 μM Tz(Fl) 30 min room temperature Excess Tz(Fl) quenched 1 μM TCO-PEG4-COOH whole-cell-labeling cells lysed RIPA buffer 1% protease inhibitor 30 min 4 °C lysate reacted 0.3 μM Tz(Fl) 30 min Excess Tz(Fl quenched 1 μM TCO-PEG4-COOH lysate blotting performed HEK293T target bands selected intensity calculated ImageJ software background intensity subtracted region no bandsband intensity determined[12pt{minimal{amsmath{wasysym\oddsidemargin-69pt{document}\mathrm{target}}{intensity{document}targetintensity−targetarea/backgroundarea×backgroundintensityPreparation primary cortical neuronal cultureTwenty-four-well plates (BD Falcon coated poly-D-lysine washed sterile dH2O three times Cerebral cortices from 16-day ICR mouse embryos dissected digested with 0.25 w/v% trypsin 20 min at 37 °C cells re-suspended in Neurobasal Plus medium supplemented 10% FBS penicillin (100 streptomycin (100/ml filtered by Cell Strainer (100 centrifuged at 1,000 rpm 5 min.cells-suspended Neurobasal Plus medium supplemented 2% B-27 Plus Supplement 1.25 mM GlutaMAX penicillin streptomycin (100 μg/ml plated 2 × 105 cells 24-well plate cultures maintained 37 °C 95% air 5% CO2 humidified incubator medium replaced 3 4 days neurons used 12–15 DIV.Two-step labeling AMPARs endogenous 12 μM CAM2(TCO) 100 μl medium 300 μM NBQX added neurons 500 μl medium 2 μM CAM2(TCO) 50 μM NBQX cells incubated 10 h 37 °C medium cells treated 1 μM Tz(Fl) 5 min PBS 1 μM TCO-PEG4-COOH PBS added Western blot analyses AMPARs NMDARs 60 μM CNM(TCO) 100 μl medium 1.5 mM Pzf added cortical neurons 500 μl medium 10 μM CNM(TCO) 50 μM Pzf cells incubated 10 h 37 °C second step labeling western blotting-life studies endogenous AMPARs cultured Fig. 5a.determining t1/2surface first labeling cells incubated 0 2 4 6 12 24 48 72 120 h second step culture medium removed cells treated 1 μM Tz(Fl) 5 min PBS room temperature quench excess Tz 1 μM TCO-PEG4-COOH PBS added washed three times Cell lysis western blotting-step labeling immunodetection GluA2 rabbit anti-GluA2 antibody band intensity half-time-life studies AMPARs western blotting HEK293T cells.”Intracellular surface ratio labeled AMPARs experiments Fig. 3g determining labeled AMPARs cell surface first labeling cells treated 1 μM Tz(Fl) 5 min PBS quench excess Tz 1 μM TCO-PEG4-COOH PBS added lysed RIPA buffer 1% protease inhibitor cocktail 30 min 4 °C determining intracellular labeled AMPARs 1 μM Tz(Ax647) treated 5 min masking cell-surface TCO-labeled AMPARscell lysis RIPA buffer 1% protease inhibitor 30 min 4 °C lysate reacted 0.3 μM Tz(Fl) 30 min room temperature Excess Tz(Fl) quenched 1 μM TCO-PEG4-COOH whole-cell-labeling Fig. cells lysed RIPA buffer 1% protease 30 min 4 °C lysate reacted 0.3 μM Tz(Fl 30 min room temperature Western blotting-step labeling HEK293T Quantification band intensity ratio surface ratio labeled AMPARs recycled AMPARs cultured first step labeling second step culture medium removed cells treated 1 μM Tz(Fl) 5 min 37 °C excess Tz 1 μM TCO-PEG4-COOH added incubation 37 °C 15 min recycled AMPARs labeled 1 μM Tz(Ax647) 5 min PBS quench excess Tz 1 μM TCO-PEG4-COOH PBS added Cell lysis western blotting-step labeling anti-Alexa 647 antibodyband intensity ratio “Intracellular surface ratio AMPARs HEK293T cells.”Anti-Alexa 647 antibody sera rabbit 647-NHS-purified Alexa 647-conjugated agarose CarboxyLink Coupling Resin Alexa 647 NHS ester anti-sera (1:2,000 purified antibody (1:1,000 western blotting primary hippocampal neuronal cultureGlass dishes coated poly-d-lysine washed sterile dH2O three times 18-day Sprague Dawley rat embryos dissected digested 0.25 w/v% trypsin 20 min 37 °C cells re-suspended Neurobasal Plus medium 10% FBS penicillin streptomycin filtered Cell Strainer centrifuged 1,000 5 min cells re-suspended Neurobasal Plus medium 2% B-27 Plus Supplement 1.25 mM GlutaMAX I penicillin streptomycin plated 2 × 104 cells glass coverslips 24-well plates Cultures maintained 37 °C 95% air 5% CO2 humidified incubator medium replaced every 7 days neurons used 16–18 DIV.cell imaging AMPARs cultured endogenous AMPARs 12 μM CAM2(TCO) 300 μl growth medium 300 μM NBQX added hippocampal neurons 1.5 ml medium concentration 2 μM CAM2(TCO) 50 μM NBQX neurons treated 100 nM Tz(Ax488) 5 min washed three times imaging confocal microscope cultured neurons cultures labeled 2 μM CAM2(TCO) 100 nM Tz(Ax488) cells fixed 4% PFA PBS 30 min washed three times-fixed cells permeabilized 15 min PBS 0.1% Triton X-100 washed three times incubated 10% normal goat serum 1 h incubated overnight 4 °C primary antibodies 1% normal goat serum washed three times incubated 1 h secondary antibodies 1% normal goat serum primary antibodies mouse anti-PSD95 rabbit anti-MAP2 Secondary antibodies goat anti-mouse Alexa goat anti-rabbit Alexa633 Imaging immunostained hippocampal neurons confocal microscopeFluorescence imaging AMPARs cultured hippocampal cultures labeled 2 μM CAM2 100 nM Tz(ST647) fixed 4% PFA cells immunolabeled PSD95 MAP2 antibodies stained Alexa 488 Alexa405 secondary antibodies lifetime imaging TCS SP8 FALCON (Leica microsystems white light laser 63× NA = 1.4 oil-immersion objective SeTau-647 imaged 640 nm. power 100 emission collected 653–700 nm 0–12.5 ns time gate).FLIM images processed LAS X 3.5.5 software lifetime decay curves n-exponential reconvolution model χ2 closest 1. three-component fit SeTau-647 (τ = 2.4 ± 0.1 ns intensity FLIM synaptic dendric region selected PSD95 MAP2 signals background intensities subtracted region no cells generated Microsoft Excel data expressed mean ± s.e.m. accumulated data three experiments evaluated statistical significance Student’s t-test P < 0.05 significant Nature Research Reporting Summary | 49.9 | 1.200465 |
10.1038/s41467-020-19203-z | PMC7596564 | Designing efficient and low power memristors-based neuromorphic systems remains a challenge. Here, the authors present graphene-based multi-level (>16) and non-volatile memristive synapses with arbitrarily programmable conductance states capable of weight assignment based on k-means clustering. | Memristive crossbar architectures are evolving as powerful in-memory computing engines for artificial neural networks. However, the limited number of non-volatile conductance states offered by state-of-the-art memristors is a concern for their hardware implementation since trained weights must be rounded to the nearest conductance states, introducing error which can significantly limit inference accuracy. Moreover, the incapability of precise weight updates can lead to convergence problems and slowdown of on-chip training. In this article, we circumvent these challenges by introducing graphene-based multi-level (>16) and non-volatile memristive synapses with arbitrarily programmable conductance states. We also show desirable retention and programming endurance. Finally, we demonstrate that graphene memristors enable weight assignment based on k-means clustering, which offers greater computing accuracy when compared with uniform weight quantization for vector matrix multiplication, an essential component for any artificial neural network. | IntroductionThe recent decline in complementary metal-oxide-semiconductor (CMOS) technology after almost five decades of relentless growth necessitates alternate computing methods to circumvent existing challenges1. A subject of great interest in this regard is the human brain. While powerful supercomputers can rival or even exceed the brain in number of operations performed per second, the brain is indisputably superior in terms of energy and area efficiency, capable of performing anywhere from 5 trillion to 5 quadrillion operations per Watt and only taking up 0.0012 m3 in volume2. In comparison, IBM’s supercomputer Summit can only perform approximately 10 billion operations per Watt while taking up an area of over 850 m23. Artificial neural networks (ANNs) seek to emulate the efficiency of the brain by directly mimicking its most fundamental unit: neuron-to-neuron connections via synapses. However, even the most sophisticated chips based on ANNs, such as IBM’s TrueNorth4, lack the ability to be scaled up to the full capacity of a human brain without becoming inordinately power hungry and area-inefficient5. The traditional von Neumann architecture that operates on the basis of physical separation between logic and memory is inherently incapable of scaling ANNs with millions of synaptic weights.The motivation behind biologically-inspired computing architecture lies in the ability of such systems to continuously adapt to external stimuli that varies with time6. For ANNs, such learning is obtained by modulating the synaptic weights assigned to the connections between neurons, allowing for the overall connectivity of the network to be reconfigured7. To properly reproduce this functionality from biological neurons, ANNs require a device capable of changing and retaining its synaptic weight (resistance/conductance) upon experiencing synaptic activity (the application of a current or bias) while also demonstrating analog behavior (possessing several resistance/conductance states). In this context, modern ANNs have progressed tremendously when compared to the first computational model developed by McCulloch and Pitts8,9, with different ANNs being classified according to their respective network architectures and connectivity structures. ANNs possess a large number of computational layers, and those possessing greater than three layers are often referred to as deep neural networks (DNNs). The layers of greatest interest for the purposes of this paper are fully connected (FC) layers that appears in all forms of ANNs. These are layers wherein all outputs from a single layer are connected to all inputs of the next layer, allowing for this next layer to compute the weighted sum of all the outputs. This is typically done by performing vector-matrix multiplication (VMM) upon the outputs8. This process is extremely energy inefficient using CMOS technology in conjunction with the traditional von Neumann computing architecture. Recent research has shown that higher efficiency can be achieved by exploiting a crossbar array architecture and utilizing a direct weight update scheme based on physical laws. Each crosspoint in the array is composed of a material with adjustable conductance, G, essentially making each crosspoint an analog non-volatile memory cell. By mapping the weight matrices of FC layers to the conductance matrices of the crossbar arrays, VMM can be performed at lower latency and thus avoid the von Neumann bottleneck, i.e., data shuttling between memory and compute. The development of such devices is aided by resistive random access memory (RRAM), or memristors, that display a programmable conductance capable of being changed via the application of short (<1 s), high amplitude (>1 V) voltage pulses10–13. Most memristors are, however, binary since they possess only two resistance states: a high resistance state (HRS), in which the device is considered to be off, and a low resistance state (LRS), in which the device is considered to be on. Analog operation, mentioned previously as possessing several resistance states, is far more preferred due to its enhanced accuracy (through minimization of quantization error) over binary operation, however the difficulty of operating memristors in an analog fashion is a significant limitation that may hinder its hardware implementation. One solution is to implement analog operations using binary devices. But this naturally leads to high computational and memory costs, limiting the application of ANNs in situations with limited storage and computing power, a prime example being portable devices14,15. In order to increase power, area, and computational efficiency, weights are often quantized into lower bits. By performing mathematical operations at lower-precisions (i.e., 8-bit integer operations as opposed to 32-bit floating point operations), ANNs consume less energy and increase efficiency while also requiring less memory storage. A known downside of this approach is a loss of accuracy due to non-idealities (namely quantization errors and noise) generated by the weight quantization process, which can negatively impact an ANN’s ability to converge16,17.In this article, we experimentally demonstrate a non-volatile graphene-based resistive memory device which is capable of achieving in excess of 16 conductance states. While non-volatile graphene memory is not a new concept18–23, most explorations into graphene memory are unable to realize more than 2 memory states (1-bit) on a single device. We also show that the graphene memristive synapses possess desirable retention and switching endurance and also allow for the hardware implementation of quantization through k-means clustering, resulting in enhanced accuracy when compared to the uniform weight quantization used by other synaptic devices. Overall, our demonstration of multi-bit and non-volatile graphene memristive synapses can be transformative for the realization of area and energy efficient hardware for neuromorphic computing and for the integration of ANNs with emerging technologies such as the Internet of Things (IoT)24.ResultsNon-volatile and multi-bit graphene-based memristorsWe have achieved programmable conductance in graphene field effect transistor (GFET) devices similar to that seen in oxide-based memristors. To fabricate the GFETs, large-area chemical vapor deposition (CVD) grown graphene was transferred onto a 50 nm alumina (Al2O3) substrate, which acts as a back-gate oxide, on a stack of Pt/TiN/p++-Si, which functions as a back-gate electrode. The use of 50 nm Al2O3 as the back-gate oxide, when compared to conventional 300 nm SiO2, was motivated by the high relative dielectric constant (~10) of Al2O3 that allows for better electrostatic control of the GFET. Each GFET used for the experiments was fabricated with a channel length (L) and channel width (W) of 1 µm and 0.5 μm, respectively. Further fabrication details, including the specifics of the transfer process used, can be found in the “Methods” section. Figure 1a, b, respectively, show the schematic and scanning electron microscope image of a representative GFET. Figure 1c shows the Raman spectrum of the graphene channel, taken at a wavelength of 532 nm. The peak at approximately 1600 cm−1 is known as the G-band and is found in all sp2 carbon materials as a result of C–C bond stretching. The existence of a strong peak at a Raman shift value of 2500–2800 cm−1 indicates the presence of single-layer graphene, with the peak itself being referred to as the 2D-band. Notably, the Raman spectrum shown here lacks a peak at approximately 1400 cm−1, as well as a sub-peak directly adjacent to the G-band. These peaks, known as the D-band and D’-band, respectively, are indicative of disorder/impurities in the sp2 structure of graphene. The absence of these peaks thus indicates that the graphene used in the GFETs discussed in this paper is of high quality in addition to being single-layer. Figure 1d, e display the output characteristics, i.e., the source-to-drain current (IDS) versus the drain-to-source voltage (VDS), for two p-type GFETs as they were subjected to forward and backward voltage sweeps at a constant back-gate voltage (VBG) of 0 V. Each separate curve displayed in Fig. 1d, e represents a different sweep range of VDS. In the measurements represented by Fig. 1d, the voltage was swept to a positive maximum, VDSmax, ranging from 1 V to 6.5 V in steps of 0.5 V. As demonstrated by the curves, increasing the sweep range appears to increase the hysteresis window of the GFET until VDSmax = 5 V, beyond which the direction of hysteresis reverses and the hysteresis window begins to decrease. This phenomenon was also seen in the device characterized in Fig. 1e, wherein each VDSmax value was negative. Switching occurred at a similar magnitude (VDSmax = −5.5 V) despite the difference in polarity with the device seen in Fig. 1d. Hysteresis switching behavior with the increase in VDSmax was taken to be indicative of memristive switching between states of high and low conductance, similar to the switching in oxide-based memristors caused by the initial formation of conductive filaments due to voltage application, known as the forming process25–27.Fig. 1SET process for graphene memristors.a Schematic and (b) false-colored scanning electron microscope (SEM) image of a graphene memristor. c Raman spectrum of the graphene channel, taken at a wavelength of 532 nm. Output characteristics (i.e., the source-to-drain current (IDS) versus the drain-to-source voltage (VDS)) of an as-fabricated graphene field effect transistor (GFET) at VBG = 0 V for different VDS sweep ranges (denoted by VDSmax), from (d) 1 V to 6.5 V and (e) −1 V to −6.5 V in steps of 0.5 V. The arrows denote the sweep direction (blue for the forward sweep and black for the reverse sweep). In either case, the hysteresis window initially increases with increasing VDSmax before reversing direction and starting to decrease past VDSmax = 5 V for (d) and VDSmax = −5.5 V for (e). These results indicate switching between states of lower and higher conductance in GFETs. Transfer characteristics at VDS = 10 mV following the sweeps of (f) VDSmax = 1 V and VDSmax = 6.5 V and (g) VDSmax = −1 V and VDSmax = −6.5 V. Sweeping the GFET to a higher positive VDSmax results in a large shift of the Dirac voltage from VDirac = 6.4 V to VDirac = −5.8 V, making the GFET more n-type, whereas sweeping the GFET to a higher negative VDSmax results in a smaller shift from VDirac = 6.5 V to VDirac = −0.2 V, making the GFET more ambipolar. h Difference between the conductance of two states as a function of VBG after the sequential application of positive and negative VDS pulses of magnitude 5 V for different pulse durations (t) at VDS = 10 mV. i Switching endurance. Histogram of conductance distributions following 200 cycles of SET (red) and RESET (blue) pulses of different magnitudes. Conductance states obtained using VDS pulses of magnitude 5 V display both a relatively large difference in conductance and a switching endurance >200 cycles. These experiments demonstrate the ability to SET and RESET conductance states in graphene by applying VDS pulses of opposite polarity, making it attractive for non-volatile memory (NVM) applications.To establish the presence of memristive switching mechanisms in GFETs distinct from those seen in traditional oxide-based memristors (i.e., formation/degradation of conductive filaments in the oxide), a series of programming pulses through the back-gate of different GFETs was performed, with results displayed in Supplementary Note 1. For biases equal to or greater than those utilized in programming through the source/drain, no change in the transfer characteristics or conductance states was noted. This established that no conductive filament formation/degradation was occurring as a result of the back-gate programming pulses, making it clear that the mechanism being utilized to create memory states in GFET memristors is distinct from that utilized in oxide-based memristors. Indeed, as will later be discussed, the bulk oxide (Al2O3) is not believed to play any major role in the memristive mechanisms shown, with the mechanisms instead being dominated by interactions at the graphene/Al2O3 interface.Figure 1f, g, respectively, display the transfer characteristics (i.e., IDS versus VBG) for the GFETs at VDS = 10 mV, measured immediately following the sweeps represented in Fig. 1d, e that correlate with the given VDSmax. As shown in Fig. 1f, sweeping the GFET to a higher positive VDSmax (6.5 V as opposed to 1 V) results in a large shift towards n-type characteristics, with the Dirac point (VDirac) shifting from VDirac = 6.4 V to VDirac = −5.8 V, whereas, as shown in Fig. 1g, sweeping the GFET to a higher negative VDSmax results in a much smaller shift, changing the characteristics of the device from p-type to ambipolar as VDirac shifts from VDirac = 6.5 V to VDirac = −0.2 V. Hysteresis loops of the drain-to-source current have long been noted in graphene and related materials, including graphene oxide and carbon nanotubes (CNTs). This phenomenon has been the subject of numerous studies and is generally attributed to interactions between the materials and trap sites on their substrates and/or extraneous molecules adsorbed on the material surface or at the material/substrate interface28,29. Of these adsorbates, water molecules (H2O) have seen attention in studies due to their prevalence in most ambient environments, as well as due to the use of water baths in traditional graphene transfers30–32. While surface-bound H2O can be easily removed via vacuum or the addition of a passivation layer, H2O trapped at the graphene/substrate interface requires specific treatments to remove and can have significant impact on the electrical properties of the graphene. An investigation by Cho et al.33 on the effects of water trapping at the graphene/Al2O3 interface identified two possible adsorption modes for H2O trapped at the interface: molecular adsorption (in which the oxygen atom is bound to an AlS site on the substrate surface) and dissociative adsorption (in which the water molecule is split into an OH− molecule bound to an AlS site and a H+ ion bound to an OS site). The alignment of H2O relative to graphene differs between the two modes (parallel for H2O in molecular adsorption and perpendicular for OH− in dissociative adsorption), leading to differences in the local electrical field, with the field induced by dissociative adsorption being magnitudes larger than that induced by molecular adsorption. The stronger dissociative field, in turn, leads to a higher planar-averaged charge density and p-type doping of the graphene29,32,33.Based on the distinctly p-type nature of the GFETs tested and discussed in this paper, it is reasonable to assume that it is a result of dissociative adsorption of H2O trapped at the graphene/Al2O3 interface, most likely as a result of the graphene transfer process discussed in the “Methods” section. Similar processes have been noted to result in trapped water adlayers at the interfaces of graphene and a number of different substrates30,33,34. Stemming from this, it is also reasonable to assume that the hysteresis shown in Fig. 1d, e is primarily caused by the trapped H2O as well. To explore this phenomenon further, effort was made to observe the effects of passivation upon the demonstrated GFET hysteresis switching. A separate set of GFETs was fabricated on a separate Al2O3 substrate and passivated via the deposition of 120 nm of PMMA. Following passivation, the hysteresis switching tests discussed and demonstrated in Fig. 1d, e were performed upon the passivated devices. The results for these tests are demonstrated in Supplementary Note 2. The hysteresis seen for the positive and negative sweeps in Supplementary Fig. 2a and 2b closely resembles that seen in Fig. 1d, e, respectively. This indicates that the hysteresis and hysteresis switching is not tied to any adsorbates on the free surface of the graphene channel.However, this does not rule out contributions from adsorbates trapped at the graphene/Al2O3 interface. Previous studies, such as that by Woong Kim et al.35, have established that hysteresis due to adsorption of water at the interface can persist following surface passivation. Based on these observations, the forming process discussed in this paper is believed to be a result of switching between different adsorption modes for water molecules trapped at the graphene/Al2O3 interface. Following fabrication, these molecules are believed to be dissociatively adsorbed on account of the distinctly p-type nature of the transfer characteristics for all GFETs tested, as well as the noticeable hysteresis when observing the swept output characteristics. The increase in the drain bias applied during these sweeps is believed to induce a transition to molecular adsorption of the water molecules. The OH− molecule and H+ ion bound to an AlS site and OS site, respectively, on the Al2O3 surface would recombine and bind to an AlS site, with the OH-bonds of the resulting H2O molecule lying relatively parallel to the plane of the graphene. This is supported by the transition of the GFET transfer characteristics following each sweep; as VDSmax increases in magnitude, VDirac shifts more and more negative, causing the transfer characteristics to become either ambipolar (for negative bias pulsing) or n-type (for positive bias pulsing). While the GFET is then able to demonstrate analog switching between the n-type and ambipolar states, it is unable to return to the original p-type characteristics indicative of dissociative adsorption. In addition, following the initial bias sweeping, any subsequent sweeping fails to demonstrate any significant hysteresis, a known characteristic of molecular adsorption. This can be seen in Supplementary Fig. 2c and 2d. Further discussion of the hysteresis and the potential contributions of interface defects/adsorbates can be seen in Supplementary Note 3 and 4, respectively. The characteristic switching displayed in Fig. 1f, g indicates the presence of at least two distinct conductance states for GFETs, achievable by applying high source-drain biases across the graphene channel. Since programming time is a vital factor for memory of any form, experiments were performed to observe the time needed to maximize the conductance difference between the conductance states. Figure 1h displays the difference between the conductance of two states as a function of VBG (i.e., the read gate voltage after the sequential application of positive and negative VDS write pulses voltages of magnitude 5 V applied for different pulse durations). A read voltage (VDS = 10 mV) was used to extract conductance values following each write pulse, as per the equation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = \frac{{I_{{\mathrm{DS}}}}}{{V_{{\mathrm{DS}}}}}$$\end{document}G=IDSVDS. A minimal change in the conductance was observed for difference pulse times. Nevertheless, the above experiment demonstrates the ability to program and erase conductance states in graphene simply by applying a VDS pulse of opposite polarity, making it attractive for non-volatile memory (NVM) applications.A critical qualifier for any NVM is switching endurance, which determines how many times the memory can be overwritten to store new information27,36–39. We found that when VDS pulses >6 V are applied to GFETs, the devices experience switching failure after only a few (< 10) cycles. To better analyze the effect of VDS write pulse magnitude on switching endurance, as well as its effect on the memory ratio of GFETs, 200 cycles of positive and negative VDS pulses of different magnitudes were applied to different GFETs. The pulse duration was set to 1 s. A histogram of conductance values following the positive (red) and negative (blue) pulses extracted from these tests are shown in Fig. 1i. Prior to measurement, each GFET used in this experiment was subjected to sequential positive and negative voltage pulses of magnitude 6 V. This was done to set the device characteristics into the n-type and ambipolar conductance states demonstrated in Fig. 1f, g. Supplementary Note 5 demonstrates that the conductance distributions resulting from write pulses of very large magnitude (>6 V) tend to overlap due to poor cycling endurance. Distributions resulting from write pulses of low magnitude (<4.5 V) also overlap, as these voltage pulses are of insufficient magnitude to induce a threshold shift capable of forming distinct conductance states. Conductance states obtained from positive and negative write pulses of magnitude 5 V, and, to a lesser extent, those obtained from pulses with magnitude 4.5 V, display both a relatively large difference in conductance and a switching endurance >200 cycles. Power consumption for the GFETs is approximately 5 mW for write operations at a pulse magnitude of 5 V and less than 40 nW for read operations at a read voltage of 10 mV. Using a pulse time of 1 s, this establishes a switching energy of approximately 5 mJ.A major advantage of resistive memory devices is their ability to support multiple memory states, allowing for a single device to encompass multiple bits of memory and therefore possess a higher data storage density. This, in turn, can lead to the development of smaller, more efficient devices, which are highly advantageous for applications such as the Internet of Things (IoT)24 and mobile devices capable of utilizing ANNs10. However, while all memristors are capable of realizing bi-stable (1-bit) memory cells due to their ability to switch between two (ON/OFF) conductance states, there is still significant challenge in implementing memristive devices that can be reliably programmed at a multitude of distinct conductance states27,40. Our demonstration of the electrical characteristic switching of GFETs shown in Fig. 1 indicates that the graphene devices can achieve multiple (>2) conductance states and could serve as multi-bit NVM if we can exploit write pulses of different magnitudes. Figure 2a displays the transfer curves for a GFET when negative write voltage pulses of duration 1 s with increasing magnitude are applied to the GFET, starting at 3 V and increasing to 6 V in steps of 0.2 V. The GFET characteristics clearly show a monotonic transition from n-type to ambipolar characteristics. Note that prior to the application of these write pulses, the GFET was set to n-type characteristics via the application of a positive VDS pulse of magnitude 6 V. Clearly, there exists multiple distinct Dirac points between the two end states, resulting in multiple (>2) conductance states for GFETs. For multi-bit memory it is critical to test the retention and distinguishability among the different memory states27,36,41. Figure 2b through e display the temporal variation (retention) in the conductance values for each state, measured for a total duration of 100 s, when the GFET is programmed into 2, 4, 8, and 16 conductance states, respectively, through different write pulse step sizes. The read voltage (VDS) was kept at 10 mV for all tests shown while VBG was kept at 0 V. Retention and endurance testing over longer durations that what is shown here can be seen in Supplementary Note 6. Accompanying histograms display the conductance distribution for each programming configuration. For each set of states tested, the initial states (t = 0 s) were set by applying a −5 V VDS pulse for 1 s. For each subsequent state, pulse time was kept to 1 s in order to maximize the memory ratio between each state. The maximum write pulse magnitude was restricted to ≤5 V in order to ensure high switching endurance, as indicated in Fig. 1g. As evident from Fig. 2b and the corresponding conductance histograms, 2 distinct conductance states with significant memory ratio are achieved by applying VDS write pulses with a step size of 2 V. However, as the number of conductance states increases from 2 states to 4 (Fig. 2c), 8 (Fig. 2d), and 16 (Fig. 2e) states by decreasing write pulse step size to 0.5 V, 0.25 V, and 0.125 V, respectively, the memory ratio diminishes between each state, reducing the distinguishability between the states. We performed similar experiments for positive write pulse polarity, as shown in Fig. 2f. In this case, the GFET was initially set to ambipolar characteristics via the application of a negative VDS pulse of magnitude 6 V. Figure 2f shows that the GFET can be gradually switched to n-type characteristics via the application of positive VDS pulses of increasing magnitude. The distinct Dirac points seen in the process indicate the potential for achieving multiple distinct conductance states. Figure 2g through j display the temporal variation in the conductance values for each state, measured for a total duration of 100 s, when the GFET is programmed into 2, 4, 8, and 16 conductance states, respectively, through different write pulse step sizes. Accompanying histograms display the conductance distribution for each programming configurations. A similar conclusion is drawn regarding the memory ratio and retention for positive write pulse polarity as for negative write pulse polarity. These results indicate the ability to step the conductance states of a single GFET in either direction (i.e., higher conductance to lower or vice versa), as well as the ability to return to previous conductance states by applying voltage pulses of opposite polarity.Fig. 2Memory levels, memory ratio, and memory retention of graphene memristors.a Transfer characteristics of a GFET when negative write voltage pulses of duration 1 s with increasing magnitude are applied, starting at −3 V and increasing to −6 V in steps of −0.2 V. The GFET characteristics clearly show a monotonic transition from n-type to ambipolar characteristics. Prior to the application of these write pulses, the GFET was set to n-type characteristics via the application of a positive VDS pulse of magnitude 6 V. Clearly, multiple (>2) memory levels (conductance states) are achieved in graphene memristors. Memory ratio and memory retention, measured for a total duration of 100 s at VBG = 0 V, for a graphene memristor programmed into (b) 2, (c) 4, (d) 8, and (e) 16 memory levels using different write pulse (VDS) step sizes, each of duration 1 s. Accompanying histograms display the conductance distributions for each programming configuration. The maximum write pulse magnitude was restricted to ≤5 V in order to ensure high switching endurance. Significant memory ratio is achieved when VDS step size is 2 V. However, as the number of memory level is increased by decreasing VDS step size to 0.5 V, 0.25 V, and 0.125 V, respectively, the memory ratio diminishes, reducing the distinguishability between the conductance states. f Transfer characteristics of an ambipolar GFET when positive write voltage pulses of duration 1 s with increasing magnitude are applied, starting at 3 V and increasing to 6 V in steps of 0.2 V. The GFET returns to n-type characteristics. Memory ratio, memory retention, and corresponding histograms of conductance distributions for the same graphene memristor programmed into (g) 2, (h) 4, (i) 8, and (j) 16 memory levels. These results indicate the ability to configure the GFET in precise conductance states, change it in either direction (i.e., higher conductance to lower or vice versa), and return it to previous conductance states by applying voltage pulses of opposite polarity.It should be called to attention that when operating at ≥16 states, such as what is shown in Fig. 2e, j, the memory ratio between neighboring memory states can decrease significantly to the point where the non-volatility of the devices can be called into question. Indeed, when one considers the accompanying histograms for Fig. 2e, j, it is readily apparent that there is a non-insignificant amount of crossover in the distributions of neighboring conductance (memory) states. While this can negatively affect operation of the GFETs at a higher number of memory states, it does serve to demonstrate the analog (“continuously variable”) nature of the conductance states achievable on GFETs. As demonstrated by the other subfigures of Fig. 2, when operating at ≤8 memory states, GFETs maintain non-volatility at the cost of number of memory states. However, they remain able to be programmed to any of the memory states shown in Fig. 2e, j. This provides an attractive amount of flexibility for neuromorphic applications, allowing for the GFETs to achieve targeted conductance (weight) values reliably and accurately as needed. This is exemplified through our demonstration of k-means clustering compared to uniform quantization, as will be discussed in the subsection On-Chip VMM using Graphene Memristors and k-Means Clustering. It should also be noted that for the negative and positive write pulse sequences shown in Fig. 2b through e and Fig. 2g through j, respectively, all tests shown of a given polarity were performed upon the same GFET. A slight difference in the initial IDS values can be seen for testing of both polarities, with a maximum initial current difference of 0.13 µA for the negative pulsing and 0.23 µA for the positive pulsing. These minor differences could be due to shifts in the threshold voltage as a result of varying interface trap state population/depopulation from the high electric field generated during pulsing. While the majority of dangling bonds at the graphene/Al2O3 interface are believed to be occupied by water molecules, there is no doubt a small number still capable of acting as carrier traps. However, based on the consistency of the memory ratios between states and the reduction of hysteresis following the forming process, the overall effect on the GFETs is believed to be minimal.The conductance switching demonstrated in GFETs following the forming process, as highlighted by Fig. 1d through g, is believed to be the result of dipole moment switching due to the generated electric field. Such effects have been shown to result in threshold and conductance shifting in field effect transistors with interfacial dipole monolayers, leading to the development of distinct memory states41–45. Following the forming process (the transition from dissociative adsorption of water molecules at the interface to molecular adsorption), the water molecules are randomly oriented due to the uncoordinated nature of AlS states at the Al2O3 surface33. In this state, the local electric field generated by the water molecules is far weaker than in dissociative adsorption, owing to the interference caused by the random orientation of neighboring dipoles. As a result, the graphene tends to display ambipolar transfer characteristics as opposed to its initial p-type characteristics. Previous studies have shown that interfacial water molecules at a graphene surface can be reoriented through the application of an external electric field46,47. This polarizes the water molecules and can align their dipoles due to their preference for an orientation parallel to the electric field, enhancing the local electric field and increasing conductance of the graphene channel48. Experimentally, this phenomenon is reflected by the increase in conductance observed when positive bias pulses are applied to the GFETs through the drain, as demonstrated in Fig. 2g through j. When negative bias pulses are applied, the water molecules are oppositely polarized, leading to reorientation. This is reflected by the decrease in conductance through negative bias pulsing shown in Fig. 2b through e. The ability for GFETs to switch to and from conductance states without being reset to n-type or ambipolar characteristics could allow for faster writing and erasing of data, as well as higher density data storage, when used as multi-bit memory49. Also note that the conductance values corresponding to the different memory states of the GFETs are linearly and symmetrically distributed, fostering high accuracy in ANNs that rely on backpropagation learning rule50. Furthermore, the precise control of GFET conductance states can offer tremendous benefit for on-chip training that rely on precise weight updates for faster convergence.The multi-terminal nature of the GFET-based synaptic device allows for it to be modulated by both the VDS programming pulses mentioned previously and by the VBG applied during read operations. By varying VBG, the resistance of the graphene channel can be modulated, allowing for tuning of the conductance states (weight values) programmed into the device via the VDS pulses, as well as the memory ratio between neighboring states. This tuning of weight values through the application of a separate bias could be considered reminiscent of heterosynaptic plasticity in biological neural networks, in which the stimulation of a given neuron causes a change in the strength (weight) of synaptic connections between other, inactivated, neurons51,52. The most well-known example of this mechanism is the existence of modulatory neurons, also known as interneurons. When activated, these neurons release chemicals known as neuromodulators, differentiated from typical neurotransmitters by their ability to alter synaptic efficacy instead of generating an electrical response. These alterations can last up to several minutes, providing comparatively long-term modulation of synaptic events53. In addition, repeated heterosynaptic modulation has, through study, been found to promote the growth/retraction of synaptic connections, creating persistent changes in synaptic weight and contributing to long-term memory formation/storage54. This has made the implementation of heterosynaptic plasticity an important goal for developing the next generation of novel neuromorphic systems52. The modulation of conductance states afforded by different modulatory bias, Vmod (VBG), values can be seen in Supplementary Note 7. All measurements were conducted on the same GFET using the same VDS pulsing scheme utilized in Fig. 2e. Each state was held for 100 s with no observable degradation into neighboring states, indicating good retention for all Vmod. The changes in conductance states and memory ratios between adjacent states as a result of changing Vmod indicate the ability to implement synaptic potentiation and depression by using the back-gate bias to increase or decrease conductance states (weight values) independently from the application of programming pulses across the source and drain. Thus, the extra degree of freedom offered by the multiterminal design of GFETs allows for synaptic modeling that is not possible in traditional two-terminal synaptic devices, such as those that operate using oxide-based memristors.To investigate the scalability of GFET memristors, several sets of GFETs featuring reduced channel lengths (L), ranging from 200 nm to 800 nm in steps of 200 nm, were fabricated on a separate Al2O3 substrate using the same fabrication processes discussed in the “Methods” section. Half of each set was fabricated with a channel width (W) of 1 µm, while the other half was fabricated with a channel width matching the channel length. The device layout that covered the smallest area while remaining functional was found to be that with L = 400 nm and W = 1 µm, for a total area of 0.4 µm2. In addition to the reduced area, the devices demonstrated the ability to shift conductance states at lower pulse magnitudes than the 1 μm channel length (0.5 µm2 channel area) devices detailed previously, indicating a channel length/area dependency for the conductance switching mechanism in GFETs.Supplementary Fig. 8a demonstrates the shifting from initial p-type characteristics indicative of the forming process at lower voltages than was necessary for the 1 µm channel length devices. This can be explained as an effect of the shorter channel length increasing the electric field generated by each pulse, allowing for reorientation of the water molecule dipoles trapped at the graphene/Al2O3 interface at lower biases. The shift from n-type to ambipolar characteristics when negative low magnitude voltage pulses are applied, as shown in Supplementary Fig. 8b, supports this theory. Interestingly, the 400 nm channel length devices continued to display high endurance even at high pulse magnitudes. Supplementary Fig. 8c and 8d display the shifting of the Dirac point when positive and negative pulses of magnitude 4 V and 5 V, respectively, are applied. Insets show the endurance at each pulse magnitude for 200 cycles. The 5 V pulse cycling, which was noted as the maximum sustainable voltage for the 1 µm channel length devices, shows similar memory ratio and endurance to that of the 1 µm channel length devices discussed earlier. These results indicate that a wider range of pulse magnitudes can be used for shorter channel devices, potentially increasing the number of distinct achievable memory states while retaining similar endurance. In contrast, all 200 nm channel length devices tested were found to be intrinsically ambipolar and displayed little-to-no shifting when programming pulses were applied. These results were taken to support our initial hypothesis regarding the role of water molecule adsorption and dipole alignment at the interface regarding changing conductance states. It is believed that the small area of the 200 nm channel length devices did not allow for sufficient water molecules to be trapped to demonstrate p-doping via dissociative adsorption or conductance shifting via dipole realignment. The channel scalability demonstrated also shows promise for large-scale integration of GFETs into crossbar-array architectures. As shown and discussed in Supplementary Note 9, the nature of the programming phenomenon of GFETs, bias pulsing through the drain, allows for electrostatic isolation of devices in close proximity to one another despite the presence of a global back-gate. Together with the aforementioned scaling, this indicates the potential for high integration density of GFET memristors, offering an attractive alternative for close-packed memristive device architectures such as dense crossbar-arrays.Impact of weight assignment–uniform versus k-means clusteringWeight quantization is inevitable for hardware implementation of ANNs. However, it leads to quantization error since weights must be rounded to the nearest analog value. For ANNs implemented using the traditional von Neumann architecture, weights can be stored in high precision digital memory to reduce the quantization error at the expense of latency, energy inefficiency, and area overhead. On the contrary, ANNs exploiting in-memory computing can suffer from the limited number of analog memory levels offered by the crossbar architecture in spite of high speed, low power, and area efficient design. In this section, we elucidate on the impact of number of analog memory levels on the quantization error for VMM architecture and demonstrate how it can be mitigated by adopting proper quantization techniques. Figure 3a shows a data structure with two vectors, A and B, and their product, C. Vector A contains 5000 matrices of size 1 × 2 and vector B contains 5000 matrices of size 2 × 1, with matrix elements drawn randomly from some given weight distributions. Figure 3b, c, respectively, show two such weight distributions, namely uniform distribution and normal distribution in the range of [−1,1]. Figure 3d shows the schematic of uniform quantization where the data range [−1, 1] is divided into N equally spaced bins, with N being the number of analog memory levels. Any synaptic weight that belongs to a given bin is assigned to the analog memory value associated with that bin. Figure 3e, f, respectively, show the error histogram as a function of N when the weights are drawn from the uniform and the normal distributions corresponding to Fig. 3b, c. The error histogram is computed from (CQ – C), where the elements of vector C, i.e., cn, are the product of matrix [an1 an2] from vector A and matrix [b1n; b2n] from vector B, and the corresponding elements of vector CQ are the product of quantized AQ and BQ. Figure 3g shows that the error, as expected, decreases with increasing N. However, for similar N, the error due to uniform quantization is significantly higher for normally distributed weights when compared to uniformly distributed weights. Since weight distributions in practical scenarios55–58 are more likely to follow a normal distribution, uniform quantization can lead to significantly high inference inaccuracy. In order to mitigate the challenges associated with uniform quantization, we propose k-means clustering based quantization. Figure 3h shows the schematic of k-means clustering, which is an unsupervised learning algorithm that divides the n data samples into k clusters, such that k ≤ n59. The algorithm randomly chooses the centroids, calculates the distance of each point to the centroid, and, finally, minimizes the variance of the distance iteratively to identify the centroids. These centroids are usually located near the mean of the clusters. In k-means clustering based quantization, weights in a specific cluster are quantized to their respective centroids. Figure 3i, j, respectively, show the error histogram as a function of N when the weights are drawn from the uniform and the normal distributions corresponding to Fig. 3b, c and are quantized using k-means clustering. As shown in Fig. 3k, the error decreases with increasing N. More importantly, k-means clustering offers better accuracy compared to uniform quantization and the benefits are found to be more for normally distributed weights. However, centroids of the weight distributions are not necessarily symmetric or follow linear trends. As such, hardware implementation of k-means clustering based quantization will require analog memory not only with multiple levels but also with the capability of configuring the individual memory states. In the following section we experimentally demonstrate this idea based on analog graphene memristive synapses.Fig. 3Weight assignment using uniform distribution and k-means clustering.a Data structure showing vectors A and B of sizes 1 × 2 and 2 × 1, respectively, and their product, C. Matrix elements for A and B are drawn randomly from (b) uniform and (c) Gaussian normal weight distributions in the range of [−1, 1]. d Uniform quantization where the data range [−1, 1] is divided into N equally spaced bins. Any weight that belongs to a given bin is assigned to the analog memory value associated with that bin. The error histogram computed from (CQ – C), where the elements of CQ are the product of quantized elements from A and B (i.e., AQ and BQ) as a function of N, when weights are drawn from (e) uniform and (f) normal distributions, corresponding to (b) and (c), respectively. g Box plot of the error in (e) and (f), which shows monotonic decrease as N increases. Also, for similar N, the error is significantly higher for normally distributed weights when compared to uniformly distributed weights. h The schematic of k-means clustering, which is an unsupervised learning algorithm that divides the n data samples into k clusters, such that k ≤ n. The algorithm randomly chooses the centroids, calculates the distance of each point to the centroid, and, finally, minimizes the variance of the distance iteratively to identify the centroids. These centroids are usually located near the mean of the clusters. In k-means clustering quantization, weights in a specific cluster are quantized to their respective centroids. The error histogram as a function of N when the weights are drawn from the (i) uniform and (j) normal distributions corresponding to (b) and (c), respectively. k Box plot of the error in (i) and (j), which shows significant reduction in error for both cases when compared to that shown for uniform quantization in (g), especially for normally distributed weights.On-chip VMM using graphene memristors and k-means clusteringFigure 4a depicts our graphene-based resistive memory architecture for executing VMM operations (see Supplementary Note 9 for programming of successive GFETs in an array). Note that, for any given back-gate voltage (VBG), the output current is given by the product of the conductance matrix and input voltage vector following the equation:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_{{\mathrm{OUT}}} = G_1V_1 + G_2V_2 = \left[ {\begin{array}{*{20}{c}} {G_1} & {G_2} \end{array}} \right]\left[ {\begin{array}{*{20}{c}} {V_1} \\ {V_2} \end{array}} \right]$$\end{document}IOUT=G1V1+G2V2=G1G2V1V2Fig. 4Graphene memristor based vector-matrix multiplication (VMM) using k-means clustering.a Memory architecture for executing VMM operations. Drain voltages (V1 and V2) are used as the input vector and graphene memristor conductance values (G1 and G2) are used as the weight matrix. The output current (IOUT) is used as the output vector. b Colormap of the expected output current corresponding to different input voltage vectors for G1 = 215 µS and G2 = 155 µS. Experimentally obtained output currents when these weights are rounded to the nearest conductance states offered by the respective graphene memristors with uniformly distributed memory levels for (c) N = 2 and (d) N = 4. Error between the expected and actual output current for (e) N = 2 and (f) N = 4. g Experimentally obtained output current and (h) error when the weights are rounded to the nearest conductance states following k-means clustering.We consider a situation where the elements of the weight matrix are the centroids of the weight distribution obtained through k-means clustering quantization with k = 2. The desired conductance values are, for example, G1 = 215 µS and G2 = 155 µS. Figure 4b shows the colormap of the expected output current corresponding to different input voltage vectors. Figure 4c, d show the experimentally obtained output current when these weights are rounded to the nearest conductance states offered by the respective graphene memristors with uniformly distributed memory levels for N = 2 and N = 4. For N = 2, the allowed conductance states for each GFET are 230 µS and 140 µS, corresponding to the programming voltage pulses of −3.0 V and −5.0 V, respectively. For N = 4, the allowed conductance states for each GFET are 230 µS, 200 µS, 170 µS, and 140 µS corresponding to the programming voltage pulses of −3.0 V, −3.5 V, −4.0 V, and −5.0 V, respectively. Figure 4e, f show the error between the expected and actual output current, which is relatively high since the weights can only be rounded to the nearest conductance states \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_1^{U2}$$\end{document}G1U2 = 230 µS and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_2^{U2}$$\end{document}G2U2 = 140 µS for N = 2 and to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_1^{U4}$$\end{document}G1U4 = 200 µS and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_2^{U4}$$\end{document}G2U4 = 170 µS for N = 4. Despite the increase in the value of N, the colormaps of error in Fig. 4e, f are similar. This is due to the fact that the programmed conductance values differ from their targeted values by 15 µS for both N = 2 and N = 4. This means that while the error distribution is shifted due to the change in the relative position of the conductance states, the overall accuracy of the synapse is not improved. This serves to highlight a drawback of uniform weight distribution and its implementation using devices with discrete memory states, such as oxide-based memristors. When the desired weight (conductance) value lies between set states, it can be very difficult for the system to reach it unless it utilizes a very large number of memory states. The analog nature of GFET memristors, on the other hand, allows for precise programming of any weight (conductance) value within the distribution of conductance states. The experimentally obtained output current for when the GFETs are directly programmed to the nearest achievable conductance states, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_1^K$$\end{document}G1K = 214 µS and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G_2^K$$\end{document}G2K = 156 µS, can be seen in Fig. 4g. The error between this output current and the expected output current shown in Fig. 4b is displayed by Fig. 4h. As would be expected, the error is significantly reduced. See Supplementary Note 10 for the post-programmed characteristics of individual GFETs for all cases. Our demonstration shows the benefits of graphene-based memristors over oxide-based memristors, as the former allows for reliable programming of individual GFETs in an array to specific conductance states. In ANN applications, this allows for the realization of k-means clustering, which improves the computing accuracy. A brief summary and comparison of GFET-based memristive synapses with other 2D material based memristive synapses is provided in a table in Supplementary Note 11.DiscussionIn conclusion, we have successfully demonstrated graphene-based ultra-thin resistive memory capable of achieving multiple (>16) memory states with necessary retention and endurance. We have also demonstrated that these memory states are configurable to desired conductance values, unlike conventional memristive NVMs, through the application of drain voltage pulses of differing magnitudes. Furthermore, we have discussed and shown through simulation the benefits of k-means clustering when compared to uniform quantization as a high precision method for quantizing synaptic weights in ANNs. Finally, we have experimentally demonstrated the ability of analog graphene memristive synapses to allow for the on-chip realization of k-means clustering while also establishing the ability to perform VMM operations through the use of conductance states as computational weights. We believe that these results will aid in the development of high precision, low-power, and area-efficient neuromorphic computing engines based on graphene memristors for various incarnations of ANNs.MethodsDevice fabricationCommercially grown monolayer graphene (Graphenea), procured on copper foil and with PMMA pre-spun, was used in our experiments. A wet transfer method was used to transfer the graphene onto a 1 cm2, 50 nm Al2O3 substrate with highly-doped (p++) Si as the back-gate electrode. The PMMA/graphene/Cu-foil stack was placed on a surface of iron (III) chloride (FeCl3), with the graphene-covered side facing away from the solution, in order to wet etch the copper foil. Once the etching was complete, the PMMA/graphene stack was transferred to a deionized (DI) water bath using a glass slide. A total of three DI water baths were used (10 min each), with the glass slide being cleaned with acetone and isopropyl alcohol (IPA) between each transfer. The PMMA/graphene stack was then transferred from the last water bath using the Al2O3 substrate previously mentioned. The substrates were heated for 5 min at 55 °C to evaporate any remaining water on the surface and then heated at 150 °C for 10 h to eliminate any wrinkles that may have originated from the transfer process and promote adhesion between the graphene and substrate. The PMMA was then removed using an acetone bath (10 min), which was followed by an IPA bath (5 min) to clean the sample. The graphene channels were defined using electron-beam (e-beam) lithography (Vistec EBPG5200), with the surrounding graphene being etched using O2 plasma (Vision 320 RIE) at room temperature for 15 s. The source/drain contacts were then defined using e-beam lithography. Ni (40 nm) followed by Au (30 nm) was deposited using e-beam evaporation for the contacts. Prior to each instance of e-beam lithography, photoresists MMA EL6 and PMMA A3 were spun onto the substrate at a rate of 4000 RPM for 40 s, with the MMA EL6 serving to promote adhesion and enhance liftoff of the PMMA A3. Following e-beam lithography, the substrate was developed using a 1:1 mixture of 4-methyl-2-pentanone (MIBK) and IPA followed by pure IPA for 60 s and 45 s, respectively. Liftoff was performed by submerging the substrate in acetone for approximately 10 min, with a subsequent IPA bath of approximately 5 min to clean the substrate of any residue. All devices were fabricated with a dual-channel structure possessing channel lengths of 1 µm and channel widths of 0.5 µm.Device measurementsElectrical characterization was performed at room temperature in high vacuum (≈10–6 Torr) on a Lake Shore CRX-VF probe station and using a Keysight B1500A parameter analyzer.Supplementary informationSupplementary InformationPeer Review File | nature communications | [
"Article"
] | [
"Electronic devices",
"Electronic properties and devices"
] | decline in metal-oxide-semiconductor (CMOS) technology after five decades necessitates alternate computing methods challenges1 subject of interest is human brain supercomputers rival exceed brain in operations per second, brain superior in energy and area efficiency 5 trillion to 5 quadrillion operations per Watt taking up 0.0012 m3 in volume2. IBM’s supercomputer Summit 10 billion operations per Watt area over 850 m23 Artificial neural networks (ANNs) emulate efficiency brain mimicking neuron-to-neuron connections via synapses sophisticated chips IBM’s TrueNorth4 lack ability to to full capacity human brain without becoming power hungry and area-inefficient5 traditional von Neumann architecture incapable of scaling ANNs with millions synaptic weights motivation behind biologically-inspired computing architecture in adapt to external stimuli learning by modulating synaptic weights connections between neurons connectivity ANNs require device changing synaptic weight (resistance/conductance) upon activity analog behavior modern ANNs progressed compared to first model by McCulloch and Pitts8,9 different ANNs classified according to network architectures connectivity structuresANNs possess large computational layers greater than three layers deep neural networks (DNNs). layers greatest interest for paper are fully connected (FC) layers in all forms ANNs outputs from single layer connected to inputs next layer weighted sum of outputs typically vector-matrix multiplication (VMM) upon outputs8 process energy inefficient using CMOS technology traditional von Neumann computing research higher efficiency crossbar array architecture direct weight update scheme physical laws Each crosspoint array material with adjustable conductance analog non-volatile memory cell mapping weight matrices of FC layers to conductance matrices crossbar arrays VMM at lower latency avoid von Neumann bottleneck data shuttling between memory compute development devices aided by resistive random access memory (RRAM), memristors programmable conductance via short (<1 s), high amplitude (>1 V) voltage pulses10–13 Most memristors binary two resistance states high off low on Analog operation preferred due to enhanced accuracy error over binary operation difficulty of operating memristors hinder hardware implementation solution implement analog operations using binary devices.leads to high computational memory costs limiting application ANNs in limited storage power portable devices14 increase power area computational efficiency weights quantized into lower bits performing operations at lower-precisions 8-bit ANNs consume less energy increase efficiency less memory storage downside loss of accuracy due to quantization errors noise ANN’s demonstrate non-volatile graphene-based resistive memory device 16 conductance states non-volatile graphene memory not new explorations realize more than 2 memory states (1-bit) on single device graphene synapses possess desirable retention switching endurance allow quantization through k-means clustering enhanced accuracy weight quantization demonstration of multi-bit non-volatile graphene synapses transformative for area energy efficient hardware neuromorphic computing integration ANNs with Internet of Things (IoT)24.ResultsNon-volatile multi-bit graphene-based achieved programmable conductance in graphene field effect transistor (GFET) devices similar oxide-based memristorsGFETs large-area chemical vapor deposition) graphene transferred 50 nm alumina (Al2O3) substrate back-gate oxide stack Pt/TiN/p++-Si back-gate electrode 50 nm Al2O3 300 nm SiO2 motivated high relative dielectric constant (~10) Al2O3 better electrostatic control GFET Each GFET fabricated channel length width (W) 1 μm 0.5 μm fabrication details transfer process “Methods” section. Figure 1a, b schematic scanning electron microscope image representative GFET Figure 1c Raman spectrum channel wavelength 532 nm peak 1600 cm−1 G-band sp2 carbon materials C–C bond stretching strong peak at Raman shift value 2500–2800 cm−1 indicates single-layer graphene 2D-band Raman spectrum lacks peak at 1400 cm−1 sub-peak adjacent G-band peaks D-band D’-band indicative of disorder/impurities sp2 structure graphene absence peaks indicates graphene GFETs high quality single-layer. Figure 1d, e display output characteristicssource-to-drain current (IDS versus voltage for two p-type GFETs voltage sweeps at constant voltage 0 V curve Fig. 1d represents different sweep range VDS voltage swept to positive maximum VDSmax 1 V to 6.5 V steps 0.5 V increasing sweep range hysteresis window GFET until VDSmax = 5 V reverses in device Fig. 1e VDSmax value negative Switching similar magnitude (VDSmax = −5.5 V) despite difference polarity Hysteresis switching behavior increase VDSmax of memristive switching between high low conductance oxide-based memristors conductive filaments voltage forming. 1SET process for memristors image graphene memristor Raman spectrum graphene channel wavelength 532 nm Output characteristics source-to-drain current (IDS versus voltage of as-fabricated graphene field effect transistor (GFET) at VBG = 0 V for different VDS sweep ranges 1 V to 6.5 V −1 V to −6.5 V in steps 0.5 Varrows denote sweep direction forward black reverse hysteresis window increases VDSmax past VDSmax = 5 V (d) −5.5 V (e). results indicate switching lower higher conductance GFETs Transfer characteristics VDS = 10 mV sweeps VDSmax = 1 V 6.5 V (g −1 V −6.5 V Sweeping GFET higher positive VDSmax shift voltage 6.4 to −5.8 V n-type higher negative VDSmax smaller shift 6.5 to −0.2 V ambipolar Difference conductance states VBG after positive negative VDS pulses 5 V different durations VDS = 10 mV Switching endurance Histogram conductance distributions 200 cycles SET RESET pulses magnitudes VDS pulses 5 V display large difference conductance switching endurance >200 cycles experiments demonstrate ability SET RESET conductance states VDS pulses opposite polarity attractive non-volatile memory (NVM) applications memristive switching mechanisms GFETs distinct traditional oxide-based memristorsformation/degradation conductive filaments in oxide), programming pulses through back-gate GFETs performed results in Supplementary Note 1. biases equal no change in transfer characteristics conductance states noted no conductive filament formation/degradation back-gate programming mechanism memory states GFET memristors distinct from oxide-based memristors bulk oxide (Al2O3) major role in memristive mechanisms dominated by interactions/Al2O3 interface.Figure 1f, g display transfer characteristics IDS versus VBG for GFETs at VDS = 10 mV measured following sweeps Fig. 1d 1f sweeping GFET to higher positive VDSmax (6.5 V large shift towards n-type characteristics Dirac point from 6.4 to −5.8 V Fig. 1g sweeping GFET higher negative VDSmax smaller shift characteristics from p-type to ambipolar VDirac shifts from 6.5 to −0.2 VHysteresis loops drain-to-source current noted in graphene related materials graphene oxide carbon nanotubes (CNTs). phenomenon subject studies attributed to interactions between materials trap sites extraneous molecules adsorbed water molecules (H2O) attention prevalence ambient environments use water baths in traditional graphene surface-bound H2O removed via vacuum passivation layer H2O trapped at graphene/substrate interface requires specific treatments electrical properties graphene investigation Cho et al water trapping at graphene/Al2O3 interface identified two adsorption modes for H2O molecular adsorption oxygen atom bound to AlS site dissociative adsorption water molecule split into OH− molecule bound AlS site H+ ion OS alignment of H2O relative to graphene differs between modes differences in local electrical field field induced dissociative adsorption larger than stronger dissociative field leads higher planar-averaged charge density p-type doping graphene29p-type nature GFETs tested assume dissociative adsorption H2O at graphene/Al2O3 interface likely graphene transfer process Similar processes result in trapped water adlayers at interfaces graphene different substrates30 hysteresis in Fig. 1d, e caused by trapped H2O effects passivation GFET hysteresis switching separate GFETs fabricated on separate Al2O3 substrate passivated via 120 nm PMMA hysteresis switching tests Fig. 1d performed devices results in Supplementary Note 2. hysteresis for positive negative sweeps in Supplementary Fig. 2a and 2b resembles Fig. 1d, e hysteresis switching not tied to adsorbates on free surface graphene channel rule out contributions from adsorbates trapped at graphene/Al2O3 interface Previous studies Woong Kim et al hysteresis due to adsorption water at interface can persist following surface passivation forming process result of switching between adsorption modes for water molecules trapped at graphene/Al2O3 interfacefabrication molecules believed dissociatively adsorbed p-type transfer characteristics GFETs noticeable hysteresis swept output characteristics increase drain bias sweeps transition to molecular adsorption water molecules OH− H+ ion bound to AlS OS site on Al2O3 surface recombine bind to AlS site OH-bonds H2O molecule parallel to plane supported by transition GFET transfer characteristics sweep VDSmax increases VDirac shifts negative transfer characteristics ambipolar or n-type GFET between n-type ambipolar states return to original p-type characteristics adsorption initial bias sweeping subsequent sweeping demonstrate significant hysteresis characteristic molecular adsorption in Supplementary Fig. 2c and 2d. discussion hysteresis potential contributions interface defects/adsorbates in Supplementary Note 3 and 4 characteristic switching in Fig. 1f, g indicates two distinct conductance states for GFETs high source-drain biases across graphene channel programming time vital for memory experiments maximize conductance difference between states Figure 1h displays difference between conductance two states VBGread gate voltage after positive negative VDS write pulses 5 V different pulse read voltage (VDS = 10 mV) used extract conductance values each write pulse[12pt{minimal{amsmath\oddsidemargin-69pt =\mathrm{DS=IDSVDS minimal change conductance observed for difference pulse times experiment demonstrates program erase conductance states in applying VDS pulse opposite polarity attractive for non-volatile memory (NVM) applications critical qualifier NVM switching endurance memory new VDS pulses >6 V GFETs experience switching failure after few< 10) cycles analyze effect VDS write pulse magnitude on switching endurance memory ratio GFETs 200 cycles positive negative VDS pulses different magnitudes applied to different GFETs pulse duration set to 1 s histogram of conductance values following positive negative pulses in Fig. 1i each GFET subjected to sequential positive negative voltage pulses 6 Vdevice characteristics into n-type ambipolar conductance states in Fig. 1f, g. Supplementary Note 5 conductance distributions from write pulses large magnitude (>6 V) overlap poor cycling endurance write pulses low magnitude (<4.5 V) overlap insufficient threshold shift distinct conductance states Conductance states from positive negative write pulses 5 V 4.5 V display large difference in conductance switching endurance >200 cycles Power consumption GFETs 5 mW for write operations 5 V less than 40 nW for read operations 10 mV pulse time 1 s switching energy approximately 5 mJ resistive memory devices multiple memory states higher data storage density smaller efficient devices advantageous for Internet of Things (IoT)24 mobile devices ANNs10 memristors bi-stable (1-bit) memory cells between (ON/OFF conductance states challenge implementing devices programmed at distinct conductance states27 demonstration of electrical characteristic switching of GFETs Fig. 1 devices achieve multiple (>2) conductance states serve as multi-bit NVM write pulses different magnitudesFigure 2a displays transfer curves GFET negative write voltage pulses 1 s increasing magnitude applied starting 3 V to 6 V 0.2 V show monotonic transition from n-type to ambipolar prior set to n-type positive VDS pulse 6 V multiple Dirac points between end states multiple (>2) conductance states multi-bit memory critical test retention distinguishability among memory Figure 2b through e display temporal variation conductance values each state duration 100 s GFET programmed into 2 4 8 16 conductance states write pulse step sizes read voltage (VDS) kept at 10 mV VBG 0 V Retention endurance testing over longer durations in Supplementary Note 6. histograms display conductance distribution each programming configuration initial states set −5 V VDS pulse for 1 s subsequent state pulse time 1 s maximize memory ratio maximum write pulse magnitude restricted to V high switching endurance 2 distinct conductance states significant memory ratio achieved by applying VDS write pulses step size 2 V number conductance states increases from 2 to 4 16decreasing write pulse step size to 0.5 0.25 V 0.125 V memory ratio diminishes reducing distinguishability similar experiments for positive write pulse polarity Fig. 2f GFET set to ambipolar characteristics negative VDS pulse 6 V to n-type characteristics positive VDS pulses increasing magnitude Dirac points indicate potential multiple conductance states Figure 2g j temporal variation conductance values state 100 s GFET programmed into 2 4 8 16 conductance states different write pulse step sizes histograms display conductance distribution for programming configurations similar conclusion memory ratio retention for positive negative results indicate step conductance states GFET return to previous states applying voltage pulses opposite polarity.Fig. 2Memory levels ratio retention of memristors characteristics GFET negative write voltage pulses 1 s increasing magnitude applied starting at −3 V increasing to −6 V −0.2 V monotonic transition from n-type to ambipolar set to n-type characteristics positive VDS pulse 6 V multiple (>2) memory levels achieved in graphene memristorsMemory ratio retention measured duration 100 s at VBG = 0 V graphene memristor programmed into 2 4 8 16 memory levels different write pulse (VDS step sizes duration 1 s histograms display conductance distributions each programming configuration maximum write pulse restricted ≤5 V high switching endurance memory ratio achieved when VDS step size 2 V memory level increased VDS step size 0.5 V 0.25 V 0.125 V memory ratio diminishes reducing distinguishability conductance states Transfer characteristics ambipolar GFET positive write voltage pulses duration 1 s increasing magnitude applied starting 3 V to 6 V steps 0.2 V GFET returns to n-type characteristics Memory ratio retention conductance distributions memristor programmed 2 4 8 16 memory levels results indicate configure GFET precise conductance states change return previous states applying voltage pulses opposite polarity operating ≥16 states 2e memory ratio between states decrease non-volatility non-insignificant crossover in distributions neighboring conductance statesnegatively operation GFETs at higher memory states conductance states GFETs 2 operating at ≤8 memory states GFETs maintain non-volatility cost memory states programmed to memory states Fig. 2e, j provides flexibility for neuromorphic applications GFETs achieve targeted conductance values reliably accurately exemplified through demonstration k-means clustering uniform quantization On-Chip VMM using Graphene Memristors k-Means Clustering negative positive write pulse sequences in Fig. 2b through e Fig. 2g through j tests performed same GFET slight difference in initial IDS values both polarities maximum initial current difference 0.13 μA for negative 0.23 μA positive minor differences due to shifts threshold voltage varying interface trap state population/depopulation from high electric field pulsing majority dangling bonds at/Al2O3 interface occupied by water molecules small number as carrier traps consistency memory ratios reduction of hysteresis following forming effect on GFETs minimal conductance switching in GFETs following forming Fig. 1d through g of dipole moment switching due to generated electric fieldeffects in threshold conductance shifting in field effect transistors with interfacial dipole monolayers distinct memory states41–45 forming process transition from dissociative to molecular adsorption), water molecules randomly oriented due uncoordinated AlS states at Al2O3 local electric field weaker than dissociative adsorption interference random orientation dipoles graphene ambipolar transfer characteristics initial p-type characteristics interfacial water molecules reoriented through external electric field46 polarizes water molecules dipoles parallel electric field enhancing local electric field increasing conductance graphene Experimentally reflected by increase conductance when positive bias pulses applied to GFETs Fig. 2g j negative bias pulses water molecules oppositely polarized to reorientation decrease conductance through negative bias pulsing in Fig. 2b through e GFETs switch to conductance states without ambipolar faster writing erasing data higher density data storage multi-bit memory49 conductance values memory states GFETs linearly symmetrically distributed high accuracy in ANNs backpropagation learning rule50precise control of GFET conductance states for on-chip training updates multi-terminal GFET synaptic device modulated by VDS programming pulses VBG during read operations varying VBG resistance channel modulated tuning conductance states (weight values programmed via VDS pulses memory ratio between states tuning of weight values separate bias of heterosynaptic plasticity in neural networks stimulation neuron strength synaptic connections between example modulatory neurons interneurons activated release neuromodulators synaptic efficacy alterations last several minutes long-term modulation of synaptic repeated heterosynaptic modulation growth/retraction of synaptic connections persistent changes in synaptic weight long-term memory formation/storage54 implementation heterosynaptic plasticity important goal for next generation novel neuromorphic systems52 modulation of conductance states by different modulatory bias Vmod (VBG), values in Supplementary Note 7. measurements conducted on same GFET using VDS pulsing scheme Fig. 2e Each state held for 100 s no degradation into neighboring states good retention for all Vmod.changes in conductance states memory ratios Vmod indicate synaptic potentiation depression back-gate bias conductance states independently from programming pulses freedom multiterminal design of GFETs allows synaptic modeling possible in traditional two-terminal devices scalability of GFET memristors sets GFETs reduced channel lengths 200 to 800 nm fabricated on separate Al2O3 substrate processes Half set with channel width (W 1 μm other half width matching channel length device layout smallest area functional with L = 400 nm W = 1 μm total area 0.4 μm2. devices shift conductance states at lower pulse magnitudes than 1 μm channel length channel length/area dependency for conductance switching mechanism in GFETs Fig. 8a demonstrates shifting from initial p-type characteristics at lower voltages shorter channel length increasing electric field pulse reorientation of water molecule dipoles at graphene/Al2O3 interface at lower biases shift from n-type to ambipolar characteristics when negative low magnitude voltage pulses applied supports theory 400 nm channel length devices display high endurance even at high pulse magnitudes8c 8d display shifting Dirac point positive negative pulses 4 V 5 V applied show endurance each pulse for 200 cycles 5 V pulse cycling maximum sustainable voltage for 1 μm channel length devices shows similar memory ratio endurance results indicate wider range pulse magnitudes for shorter channel devices increasing achievable memory states retaining endurance 200 nm channel devices ambipolar displayed little-to-no shifting programming pulses applied results support hypothesis water molecule adsorption dipole alignment changing conductance states small area 200 nm channel devices water molecules p-doping adsorption conductance shifting dipole realignment channel scalability shows promise for large-scale integration GFETs into crossbar-array architectures programming phenomenon GFETs allows electrostatic isolation devices global back-gate potential for high integration density GFET memristors alternative for close-packed device architectures crossbar-arrays.Impact weight assignment–uniform versus k-means clusteringWeight quantization inevitable for ANNs leads to quantization error weights rounded to nearest analog valueANNs traditional von Neumann architecture weights stored in high precision digital memory reduce quantization error latency energy inefficiency area overhead ANNs in-memory computing suffer limited analog memory levels crossbar architecture high speed low power design impact analog memory levels quantization error VMM architecture proper quantization techniques Figure 3a data structure two vectors A B product C Vector A 5000 matrices 1 × 2 B 5000 2 × 1 matrix elements drawn randomly from weight distributions Figure 3b, c weight distributions uniform normal distribution range [−1,1] Figure 3d uniform quantization data range [−1, 1] divided into N bins N analog memory levels synaptic weight bin assigned to analog memory value Figure 3e f error histogram function of N weights drawn from uniform normal distributions error histogram computed from (CQ – elements vector C product matrix [an1 an2] A [b1n b2n] B vector CQ quantized AQ BQ Figure 3g error decreases with increasing N error due uniform quantization higher for normally distributed weightsweight distributions in scenarios55–58 likely follow normal distribution uniform quantization high inference inaccuracy mitigate propose k-means clustering quantization Figure 3h k-means clustering learning algorithm divides n data samples into k clusters k ≤ n59 algorithm chooses centroids calculates distance minimizes variance identify centroids centroids near mean clusters weights in cluster quantized to centroids Figure 3i, j error histogram function of N weights drawn from uniform normal distributions error decreases with increasing N. k-means clustering offers better accuracy benefits for normally distributed weights centroids weight distributions not symmetric follow linear trends hardware implementation analog memory configuring memory states demonstrate idea based analog memristive synapses.Fig. 3Weight assignment using uniform distribution k-means clustering Data structure showing vectors A and B sizes 1 × 2 2 × 1 product C Matrix elements A B drawn randomly from uniform Gaussian normal weight distributions in range [−1, 1]. Uniform quantization data range [−1, 1] divided into N equally spaced binsweight to bin assigned to analog memory value error histogram computed from (CQ – C), elements CQ product of quantized elements A and B AQ BQ function of N when weights drawn from uniform (f normal distributions (b plot error in (e) and (f), decrease as N increases N error higher for normally distributed weights to uniformly distributed schematic of k-means clustering algorithm divides n data samples into k clusters algorithm chooses centroids calculates distance minimizes variance identify centroids centroids near mean clusters weights in cluster quantized to centroids error histogram function of N when weights drawn from uniform (j normal distributions (b) (c), Box plot error in (i) and (j), significant reduction in error for compared to uniform quantizationOn-chip VMM graphene memristors k-means clusteringFigure 4a graphene-based resistive memory architecture VMM operations Supplementary Note 9 programming GFETs back-gate voltage output current product conductance matrix input voltage vector equation:1\documentclass[12pt{minimal{amsmath\oddsidemargin-69pt}{OUT G_1V_1 + G_2V_2 {G_1_2_2}IOUT=G1V1+G2V2=G1G2V1V2Fig. 4Graphene memristor vector-matrix multiplication (VMM) k-means clustering Memory architecture VMM operations Drain voltages (V1 V2) input vector graphene memristor conductance values (G1 G2) weight matrix output current (IOUT) output vector Colormap expected output current input voltage vectors G1 = 215 μS G2 = 155 μSobtained output currents weights rounded nearest conductance states memristors memory N 2 N = 4. Error expected actual output current N = 2 N = 4. output current error weights rounded nearest conductance states k-means clustering elements weight matrix centroids weight distribution k-means clustering k 2. desired conductance values G1 215 μS G2 155 μS Figure 4b expected output current input voltage vectors Figure 4c d obtained output current weights rounded nearest conductance states memristors N = 2 4. N = 2 allowed conductance states GFET 230 μS 140 μS voltage pulses −3.0 V −5.0 V N = 4 conductance states 230 μS 200 μS 170 μS 140 μS −3.0 −3.5 −4.0 V −5.0 VFigure 4e f error expected actual output current high weights rounded nearest conductance states[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{mathrsfs{upgreek\oddsidemargin{-69pt}{document}$$G_1^{U2}{document}G1U2 = 230 μS[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin}{-69pt}$$G_2^{U2}{document}G2U2 = 140 μS N = 2[12pt]{minimal}{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$$G_1^{U4}{document}G1U4 = 200 μS[12pt]{minimal}{amsmath{wasysym{amsfonts{amsbsy{mathrsfs{upgreek}length\oddsidemargin-69pt}\begin$G_2^{U4}\end{document}G2U4 = 170 μS N = 4. Despite increase value N colormaps error in Fig. 4e, f similar programmed conductance values differ from targeted by 15 μS for N = 2 and N = 4. error distribution shifted states accuracy synapse not improved drawback of uniform weight distribution discrete memory states oxide-based memristors desired weight (conductance value between set states difficult reach unless large number memory states GFET memristors allows precise programming weight value statesexperimentally obtained output current GFETs programmed nearest conductance states\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek-69pt}$G_1^K}G1K = 214 μS-69pt_2^K}G2K = 156 μS seen in Fig. 4g error between output current expected output current Fig. 4b displayed by Fig. 4h error significantly reduced See Supplementary Note 10 post-programmed characteristics individual GFETs demonstration shows benefits graphene-based memristors over oxide-based memristors allows reliable programming GFETs to specific conductance states allows k-means clustering improves computing accuracy summary comparison of GFET-based memristive synapses with other 2D material based synapses provided in table Supplementary Note successfully demonstrated graphene-based ultra-thin resistive memory achieving multiple (>16) memory states with necessary retention endurance.demonstrated memory states configurable conductance values unlike conventional NVMs drain voltage pulses differing discussed benefits k-means clustering uniform quantization high precision quantizing synaptic weights ANNs demonstrated analog graphene memristive synapses on-chip k-means clustering VMM operations conductance states computational weights results aid development high precision low-power area-efficient neuromorphic computing engines memristors ANNs fabricationCommercially grown monolayer graphene procured copper foil PMMA pre-spun used experiments wet transfer method 1 cm2, 50 nm Al2O3 substrate highly-doped (p++) Si back-gate electrode PMMA/graphene/Cu-foil stack on iron (III) chloride (FeCl3) graphene-covered side facing wet etch copper foil transferred to deionized water bath three water baths used (10 min glass slide cleaned with acetone isopropyl alcohol transfer PMMA stack transferred last water bath Al2O3 substratesubstrates heated 5 min 55 °C evaporate water heated 150 °C 10 h eliminate wrinkles promote adhesion graphene PMMA removed acetone bath (10 IPA bath (5 min graphene channels defined electron-beam lithography (Vistec EBPG5200), surrounding graphene etched O2 plasma (Vision 320 RIE room temperature 15 s source/drain contacts defined e-beam lithography Ni (40 nm) Au (30 nm) deposited e-beam evaporation photoresists MMA EL6 PMMA A3 spun substrate 4000 RPM 40 s MMA EL6 adhesion liftoff PMMA A3 substrate developed 1:1 mixture 4-methyl-2-pentanone) IPA pure IPA 60 s 45 s Liftoff acetone 10 min IPA bath 5 min clean devices dual-channel structure channel lengths 1 μm widths 0.5 μm measurementsElectrical characterization room temperature high vacuum Lake Shore CRX-VF probe station Keysight B1500A parameter analyzer.Supplementary File | 48.4 | 1.160911 |
10.1038/s41467-020-17564-z | PMC7382469 | Spinal cord injury (SCI) often leads to immune dysfunction, but mechanistic insights are still lacking. Here the authors show that SCI alters chemokine signaling and induces long, persisting defects in hematopoietic stem and progenitor cell migration, thereby entrapping them in the bone marrow and disrupting peripheral immune homeostasis. | Spinal cord injury (SCI) causes immune dysfunction, increasing the risk of infectious morbidity and mortality. Since bone marrow hematopoiesis is essential for proper immune function, we hypothesize that SCI disrupts bone marrow hematopoiesis. Indeed, SCI causes excessive proliferation of bone marrow hematopoietic stem and progenitor cells (HSPC), but these cells cannot leave the bone marrow, even after challenging the host with a potent inflammatory stimulus. Sequestration of HSPCs in bone marrow after SCI is linked to aberrant chemotactic signaling that can be reversed by post-injury injections of Plerixafor (AMD3100), a small molecule inhibitor of CXCR4. Even though Plerixafor liberates HSPCs and mature immune cells from bone marrow, competitive repopulation assays show that the intrinsic long-term functional capacity of HSPCs is still impaired in SCI mice. Together, our data suggest that SCI causes an acquired bone marrow failure syndrome that may contribute to chronic immune dysfunction. | IntroductionIn adults, mature immune cells develop from a pool of hematopoietic stem and progenitor cells (HSPCs) through a process known as hematopoiesis. Hematopoiesis occurs primarily in bone marrow, where complex cellular interactions and molecular signaling pathways regulate the renewal of millions of immune cells each day1. Ultimately, the maintenance of bone marrow hematopoiesis is essential for effective host defense and tissue repair.Under physiological conditions, the proliferation, differentiation, and retention/release of bone marrow cells, including HSPCs, are controlled by neuroendocrine hormones and the autonomic nervous system, specifically sympathetic neurons2–12. After spinal cord injury (SCI), brain and brainstem connections that normally control spinal sympathetic preganglionic neurons are lost, creating a decentralized spinal autonomic network that includes aberrant sympathetic and neuroendocrine reflexes13. Uncontrolled sympathetic reflexes have been implicated in the overstimulation and cytotoxicity of mature leukocytes14–18 and disruption of leukocyte homing after SCI19.A decentralized bone marrow may also impair hematopoiesis and contribute to the chronic immune dysfunction that plagues individuals with SCI20–27. Indeed, data from two independent studies in which bone marrow aspirates were analyzed from small cohorts of SCI and control subjects indicate that SCI impairs human bone marrow stem cell function21,22. Notably, SCI increased the overall proliferation and total numbers of HSPCs in bone marrow of human SCI subjects; however, the ability of these stem cells to form mature immune cells was also impaired21,22.The goal of the current study is to determine the extent of hematopoietic dysfunction after acute and chronic SCI, and to identify molecular, cellular, and physiological mechanisms that may explain any SCI-induced impairments that develop in the bone marrow. We hypothesize that SCI will induce acute and chronic bone marrow failure in mice, similar to what has been described after SCI in humans. Our data show that SCI causes a rapid and chronic bone marrow failure syndrome characterized by excessive HSPC proliferation, accumulation, and impaired function. Importantly, post-injury injections of Plerixafor, an FDA-approved drug that blocks CXCR4, liberates HSPCs from bone marrow and partly reverses bone marrow failure by promoting extramedullary hematopoiesis. Treating bone marrow failure after SCI may help to reverse chronic immune dysfunction and anemia that persist indefinitely after SCI in humans.ResultsSCI increases bone marrow cell proliferationHematopoiesis, or the formation of new red and white blood cells, requires that HSPCs proliferate and differentiate in the marrow of all bones, including long bones (i.e. femur/tibia) and sternum. To investigate how SCI generally affects cell proliferation in bone marrow, we monitored post-injury changes in the sternum and femur/tibia of transgenic female mice that express luciferase in dividing cells (Mito-luc mice4,28,29; Fig. 1a). We used peak (max radiance) and average (total flux) bioluminescence as representative measures of bone marrow proliferation. To control for potential effects of surgical stress on cell proliferation, SCI mice were compared with sham-injured mice receiving a laminectomy only (Lam). Within 72 h of surgery, cellular proliferation increased in the sternum and femur of sham-injury and SCI mice; however, enhanced proliferation was maintained only in SCI mice, peaking within the first 7 days post-injury (dpi) and persisting up to 1-month post-injury (latest time evaluated) (Fig. 1b, c).Fig. 1T3 transection SCI protracts HSPC proliferation and causes HSPC accumulation in bone marrow.a Bioluminescent imaging of sternum and femur/tibia regions of interest (ROIs; red ellipses) pre-injury and 7 dpi. b, c Bioluminescence expressed as max radiance (peak intensity within ROI) and total flux (average intensity within ROI) for each region, with left and right femur/tibia averaged; data also expressed as the total (sum) of signals from 1 to 28 dpi. d Proportion of proliferating lineage low, Sca-1+, c-Kit+ (LSK) bone marrow cells, as indicated by the S–G2–M phase of the cell cycle, at 3 dpi; representative histograms of LSK proliferation at 3 dpi (closest to the mean). Quantification of LSKs (e), c-Kit+ progenitors (f), and total bone marrow cells (g) at 3, 7, and 28 dpi. h Images and heatmaps of tibia pairs from Lam and SCI mice demonstrating increased optical density (artificial units) corresponding with increased bone marrow cellularity. All data are mean ± SEM, mixed effect model with repeated measures (b, c), two-sample t-test (b–d), two-way ANOVA with Bonferroni multiple comparisons (e–g); n = 4 and 7/group in b and c, n = 10/group in d, n = 10/group (3 dpi), 5/group (7 dpi), and 5 and 6/group (28 dpi) in e–g. Dotted lines represent average data from naïve mice. dpi days post-injury, Lam laminectomy (sham injury). Source data are provided as a Source Data file.To determine whether these changes could be specifically attributed to enhanced proliferation of HSPCs, multi-color flow cytometry was used to quantify the proportion of proliferating lineage−, Sca-1+, c-Kit+ (LSK) cells in the femoral bone marrow of C57BL/6 wild-type mice after SCI (Supplementary Fig. 1a, b)30. At 3 dpi, LSK cell proliferation increased significantly (Fig. 1d). In separate cohorts of mice, the mitogenic effect of SCI on LSK cells was confirmed and was found to be long lasting (Fig. 1e); total numbers of differentiated c-Kit+ progenitors (Fig. 1f) and mature bone marrow cells increased until at least 28 dpi (Fig. 1g, h). Together, these data indicate a reactive bone marrow response after SCI marked by protracted proliferation and accumulation of HSPCs.SCI causes chronic expansion of bone marrow HSCsThe LSK fraction of bone marrow cells characterized in Fig. 1 includes all stem and multipotent progenitor cells31,32. To better define how SCI affects this pool of phenotypically and functionally heterogeneous cells, multi-color flow cytometry was used to analyze cells from another group of sham-injured and SCI mice (Fig. 2a, Supplementary Fig. 1c). Consistent with data in Fig. 1, the total number of bone marrow cells (Fig. 2b), including total LSK cells (Fig. 2c), increased after SCI. The increase in cell number affects multiple subsets of stem and progenitor cells including long-term repopulating stem cells (LT-HSCs; CD150+/CD48−/CD135− LSKs), short-term HSCs (ST-HSCs; CD150−/CD48−/CD135−LSKs), multipotent progenitor subset 2 cells (MPP2s; CD150+/CD48+/CD135−LSKs), myeloid-primed MPP3s (CD150−/CD48+/CD135−LSKs), lymphoid-primed MPP4s (CD135+ LSKs), and granulocyte–monocyte progenitors (GMPs; CD16/32+ LK cells) (Fig. 2d–i). Importantly, SCI did not cause HSPCs to increase expression of γH2AX, a marker of double-stranded DNA breaks (Fig. 2j). These data demonstrate that SCI causes excessive proliferation of all HSPCs but without evidence of long-term replication stress.Fig. 2T3 transection SCI causes the chronic accumulation of hematopoietic stem cells and differentiated progenitors.a Hematopoietic hierarchy of HSPCs and their cell surface markers. Total number of bone marrow cells (b), LSKs (c), long-term HSCs (d), short-term HSCs (e), multipotent progenitors 2–4 (f–h), and granulocyte–monocyte progenitors (i) at 28 dpi. j Mean fluorescence intensity of yH2AX expression (i.e. replication stress) within c-Kit+ HSPCs at 28 dpi. All data are mean ± SEM, two-sample t-test; n = 9 and 10/group (b–i), n = 5 and 4/group (j). Dotted lines represent average data from naïve mice. dpi days post-injury, Lam laminectomy (sham injury). Source data are provided as a Source Data file.SCI prevents HSPC mobilization from bone marrowEnhanced HSPC proliferation after sham injury or SCI may represent a compensatory response to the stress of blood loss and trauma. Indeed, both psychological and physical stressors enhance bone marrow hematopoiesis and extramedullary hematopoiesis in secondary lymphoid tissues, including the spleen4,33–37.We confirmed in the present study an acute effect of surgical stress on bone marrow hematopoiesis. Three days after sham surgery, extensive numbers of HSPCs were found in blood and spleen of sham-injured mice (occurring above circulating numbers of HSPCs in naïve mice) as assessed by MethoCult (Fig. 3a, b) and flow cytometry (Fig. 3d, e) assays, leading to an increase in spleen size (Fig. 3c). In contrast, 3 days after SCI, fewer HSPCs entered the circulation (Fig. 3a) or colonized the spleen in SCI mice (Fig. 3b, d, e). Additionally, fewer c-Kit+ HSPCs were proliferating in the spleen of SCI mice (Fig. 3e), likely as fewer proliferating HSPCs exit the bone marrow and enter the circulation after SCI.Fig. 3T3 transection SCI prevents acute HSPC mobilization into blood and trafficking to the spleen at 3 dpi.a–f Outcomes in female mice at 3 dpi. a Total numbers of CFCs per mL of whole blood. b Total numbers of CFCs per spleen. c Spleen weight and number splenocytes (% of Lam). d, e Total numbers of (d) LSK and (e) c-Kit+ cells in the spleen. f Proportion of proliferating c-Kit+ cells (S–G2–M phase) in the spleen. g–i Identical parameters as measured in a–c in male mice at 3 dpi. All data are mean ± SEM, two-sample t-test; n = 15/group in a; n = 10/group in b–d (15/group spleen weight); n = 6/group in e, f; n = 4 and 5/group g–i (8 and 9/group spleen weight). Dotted lines represent average data from naïve mice. dpi days post-injury, Lam laminectomy (sham injury), CFCs colony forming cells. Source data are provided as a Source Data file.The effects of SCI on HSPC sequestration in bone marrow were independent of sex; SCI abolished HSPC mobilization in both female and male SCI mice (compare Fig. 3a–c and g–i). Further, SCI-dependent effects on HSPCs were not influenced by either injury severity or spinal injury level; HSPCs fail to enter into the circulation after complete spinal transection injuries at either L6/S1, T9, or T3 spinal levels (Fig. 4a) and after an incomplete spinal contusion injury at either T3 or T9 spinal levels (Fig. 4b).Fig. 4Complete and incomplete SCI impairs acute mouse and human HSPC mobilization.a Number of CFCs per mL whole blood after T3, T9, or L6/S1 transection injury. b Number of CFCs per mL whole blood after T3 and T9 contusion injury. c Number of hCFCs per mL whole blood. d Total numbers of splenic hCFCs, proportion of splenic hCD34hi/hCD38low human HSPCs, and spleen weight of hNSG mice 3–5 dpi. All data are mean ± SEM, one-way ANOVA with Tukey multiple comparison test (a, b), two-sample t-test (c, d); n = 20, 15, 14, and 5/group in a; n = 6, 10, and 5/group in b; n = 8/group in c; n = 6, 6; 6, 4; 8 and 9/group in d. dpi days post-injury, Lam laminectomy (sham injury). Source data are provided as a Source Data file.To determine if the SCI-dependent effects on HSPCs were species-specific, male and female mice with human HSPCs and immune systems (i.e., humanized mice) were generated38–40. Similar to wild-type mice, SCI prevented human HSPCs from entering the circulation and trafficking to the spleen (Fig. 4c, d).Together, these data, when combined with data in Figs. 1–3, indicate that SCI triggers excessive proliferation of HSPCs but that these cells are unable to leave the bone marrow niche, causing HSPCs to progressively accumulate within the bone marrow. Importantly, HSPC sequestration in bone marrow, although SCI-dependent, is independent of sex, injury level, injury severity, or species (both mouse and human HSPCs respond identically to SCI).SCI impairs B cell development and mobilizationThe decentralized bone marrow that develops after SCI may sequester cells other than HSPCs. Indeed, in both animal models and human subjects, total numbers of circulating leukocytes decrease after SCI and stroke19,24,41–44. Here, we confirmed and expanded those data using a model of complete SCI (T3 spinal level). Specifically, at 3 dpi, T3 transection SCI reduced circulating blood lymphocytes (Fig. 5a, b) with a concomitant increase in the proportion of mature lymphocytes found in bone marrow, including B220hi B cells, CD3+/CD4− T cells, and CD3+/CD4+ T cells (Fig. 5c–e). Further, SCI reduced numbers of B220low immature B cells (confirmed to be enriched for ProB, PreB, and IgMlow immature B cells; Supplementary Fig. 1e) in the bone marrow (Fig. 5c), similar to previous data showing impaired B lymphopoiesis after SCI and stroke19,44,45.Fig. 5T3 transection SCI causes acute lymphopenia in blood with concurrent sequestration of lymphocytes in bone marrow.
a Total numbers of white blood cells (WBCs), lymphocytes, neutrophils, and monocytes per mL of whole blood at 3 dpi. b Lymphocytes (gray), neutrophils (red), monocytes (blue), and eosinophils/basophils (yellow) as a proportion of total WBCs. c Proportion of immature (i.e. developing) B cells (B220low), mature B cells (B220hi), and mature CD3+ T cells (CD4− and CD4+ subsets) in bone marrow. d Flow cytometry gating of bone marrow B220+ B cells demonstrating reduced numbers of immature (B220low) and increase numbers of mature (B220hi) B cells with SCI. Examples of B220 and forward scatter (size) characteristics of B cells in bone marrow after sham injury and SCI. e Flow cytometry gating of bone marrow CD3+/CD4− and CD3+/CD4+ mature T cell subsets demonstrating increased numbers after SCI. All data are mean ± SEM, *p < 0.05, two-sample t-test; n = 12 and 14/group in a, b; n = 3/group in c–e. Shades of blue/red (Lam/SCI) represent individual sample plots in d and e. Dotted lines represent average data from naïve mice. dpi days post-injury, Lam laminectomy (sham injury), FSC forward scatter, BM bone marrow. Source data are provided as a Source Data file.SCI increases CXCL12-CXCR4 signaling in bone marrow HSPCsImmunoregulatory cytokines and chemokines, including IL1β, TNFα, TGFβ, CCL2, CXCL12, and CXCR4 can affect the development and function of HSPCs46,47. To determine if the excessive proliferation and retention of HSPCs in bone marrow after SCI is associated with changes in these cytokines, we prepared mRNA from whole-bone marrow cells isolated from T3 transection SCI mice and sham-injured mice. Real-time PCR analyses revealed that, with the exception of Ccl2, SCI increased the expression of all bone marrow cytokines and chemokines assessed (Fig. 6a). Notably, Cxcl12, a chemokine produced by bone marrow stromal cells, and its receptor Cxcr4 were increased in parallel (fourfold and threefold, respectively).Fig. 6Plerixafor (AMD3100) mobilizes HSPCs, rescues extramedullary hematopoiesis, and boosts circulating immune cells after T3 transection SCI.a Cytokine and chemokine mRNA expression from whole-bone marrow cells of SCI mice 3 dpi (expressed as fold change of Lam). b CXCL12 protein expression from bone marrow extracellular fluid extracts. c Protein expression (MFI) of CXCR4 expression on LSKs from whole bone marrow. d Total CFCs per mL whole blood, e CFCs per spleen, f LSKs per spleen, g c-Kit+ HSPCs per spleen, and h total cells per spleen 3 days after SCI with and without daily AMD3100 treatment (once per day). i Total white blood cells (WBCs), lymphocytes, neutrophils, and monocytes per mL whole blood. j Proportion of monocytes (MO), lymphocytes (LY), and neutrophils (NE) in whole blood (n = 5/group). k Total cells per femur. All data are mean ± SEM, two-sample t-test; n = 3/group (pooled) in a; n = 6 and 7/group in b; n = 7/group in c; n = 5/group in d–k. Dotted lines represent average data from lam (a) or naïve mice (d, e, h, i, k). BM bone marrow, MFI mean fluorescence intensity, WBCs white blood cells, MO monocytes, LY lymphocytes, NE neutrophils, AMD Plerixafor (AMD3100). Source data are provided as a Source Data file.Several published studies demonstrated that CXCL12 binding to CXCR4 on HSPCs influences the maintenance and retention of HSPCs in bone marrow, a phenomenon that is dependent, in part, on the sympathetic branch of the autonomic nervous system8,10,48. Therefore, we validated the effects of high-level complete (T3 transection) SCI on this chemokine/chemokine receptor pair. In a separate cohort of T3 SCI and sham-injured mice, we found that SCI increased the amount of secreted CXCL12 protein in bone marrow extracellular fluid (Fig. 6b) and expression of CXCR4 receptor on LSK cells (Fig. 6c).To test whether the SCI-dependent increase in CXCL12/CXCR4 expression causes HSPC sequestration in bone marrow, mice were treated with Plerixafor (AMD3100), a small-molecule antagonist of CXCR4, or vehicle for 3 days, beginning 1 h post-SCI. In Plerixafor-treated mice, the number of HSPCs released into the blood increased >6-fold (Fig. 6d), with a corresponding increase in the number of HSPCs found in the spleen (Fig. 6e–g). Plerixafor treatment also reversed the normal onset of post-SCI leukopenia; both total numbers of splenocytes (Fig. 6h) and all subsets of circulating WBCs increased (Fig. 6i, j). Importantly, releasing cells from bone marrow with Plerixafor did not ablate the bone marrow. In fact, Plerixafor increased total bone marrow cellularity (Fig. 6k). These data implicate sustained CXCL12/CXCR4 signaling within the bone marrow as a potential mechanism underlying SCI-induced sequestration of HSPCs. Importantly, post-SCI sequestration of HSPCs and mature immune cells can be overcome using the FDA-approved drug, Plerixafor.Chronic SCI impairs bone marrow responses to inflammatory stimuliData above indicate that SCI causes an acquired bone marrow failure syndrome characterized by protracted HSPC proliferation with sequestration and associated peripheral lymphopenia. Bone marrow failure develops within 72 h post-injury and persists until at least 1-month post-injury. To determine whether failed bone marrow in SCI mice is able to respond to physiologically relevant stimuli, mice with chronic SCI were challenged with endotoxin (1 mg kg−1; lipopolysaccharide (LPS)), a potent inducer of HSPC proliferation and mobilization34,49–51. Mito-luc transgenic and C57BL/6J (wild-type) mice were injected with LPS daily for 3 days beginning 6 weeks after T3 transection SCI or sham surgery (Fig. 7a).Fig. 7Stimulating hematopoiesis with LPS in chronic-injured mice recapitulates acute bone marrow responses to injury.
a Experimental outline demonstrating T3 transection SCI and sham injury followed by 6–8 weeks of recovery prior to LPS stimulation (1 mg kg−1 day−1 i.p. for 3 days). Mice then underwent in vivo imaging of bone marrow proliferation (A: Mito-luc transgenic mice) or automated CBC and MethoCult assay of whole blood (B: C57BL/6J mice). b Bioluminescent images from representative Lam and SCI mice before and 11 days post-LPS. c Quantification of total flux in sternum and femur/tibia regions as a % of pre-LPS levels (dotted line). d Total number of CFCs per mL whole blood at 4 and 11 days post-LPS. e Lymphocytes (% of WBCs and total numbers) and neutrophils (total numbers) in blood 4 days post-LPS. All data are mean ± SEM, two-way ANOVA with repeated measures (c) and two-sample t-test (d, e); n = 4/group in c, n = 5/group in d, e. Dotted lines represent average data from naïve mice (d, e); days post-LPS refers to time after first LPS dose. Lam laminectomy (sham injury). Source data are provided as a Source Data file.In all sham and SCI Mito-luc transgenic mice, cellular proliferation in bone marrow decreased 2–4 days after the first LPS injection (Fig. 7c). This effect was expected since LPS stimulates the egress of proliferating cells from the bone marrow into the circulation34,52. Indeed, when compared to total number of HSPCs found in the blood of non-LPS naïve mice, LPS significantly increased total HSPCs in blood in all mice. However, fewer HSPCs were detected in the blood of SCI mice at both peak mobilization (4 days after first dose of LPS) and during recovery (11 days after first dose of LPS) (Fig. 7d). Strikingly, LPS stimulation reduced the proportion and number of circulating lymphocytes in SCI mice, but not in sham-injured mice (Fig. 7e).By 7 days after LPS injections, basal cell proliferation is restored and maintained in sham-injured mice, presumably because HSPCs have repopulated the depleted bone marrow49. In contrast, LPS-induced proliferation continues in the bone marrow of SCI mice; the luciferase signal overshoots baseline proliferation by >200% between 9 and 14 days post-LPS (Fig. 7b, c). Remarkably, the same characteristics that define bone marrow failure early after SCI are recapitulated in chronic SCI animals after LPS injections (compare Fig. 7 to Figs. 1–5). Collectively, these data indicate that SCI causes chronic, and perhaps permanent, bone marrow failure.SCI impairs the long-term function of bone marrow HSPCsData in Fig. 6 indicate that the sequestration and accumulation of HSPCs in bone marrow after SCI are due, in part, to aberrant cytokine and chemokine signaling in the bone marrow niche. To determine whether SCI also affects the intrinsic capacity of bone marrow HSPCs, notably their ability to restore hematopoiesis to a depleted bone marrow niche, we performed in vivo competitive repopulating assays.Bone marrow HSPCs were removed from SCI mice 3 days after T3 transection SCI or sham surgery. In SCI mice, this timing corresponds with a period of enhanced HSPC proliferation (Fig. 1) and sequestration (Fig. 3), but no difference in total HSPC numbers (Fig. 1). When bone marrow cells were injected into uninjured mice that had received lethal doses of irradiation (Supplementary Fig. 2a, b), SCI-derived donor cells exhibited faster engraftment in recipient mice at 8 weeks (Fig. 8a, b). By 19 weeks post-engraftment, stable engraftment was achieved in all recipient mice, regardless of the donor cell source (Fig. 8b). Notably, the enhanced engraftment potential of SCI bone marrow cells relative to sham donor cells waned after 8 weeks (arrowhead, Fig. 8a, c). However, we did not observe long-term lineage biasing of engrafted SCI donor cells through 19 weeks (Supplementary Fig. 2e). These data indicate a possible deficit in the self-renewal capacity, but not differentiation, of long-term hematopoietic stem cells from SCI donors (Fig. 8c, d).Fig. 8T3 transection SCI impairs the long-term clonogenic potential of bone marrow by 3 dpi.a Percent donor engraftment in recipient BoyJ mice 4–19 weeks after primary transplantation with 3 dpi bone marrow from naïve, sham-injured (Lam), and T3 transection SCI mice. b Percent donor engraftment after primary CRU assay at the 8- and 19-week timepoints from a. c Change in percent donor engraftment from 8 to 19 weeks post-transplantation. d Percent donor engraftment in bone marrow of recipient BoyJ mice 19 weeks after primary transplantation. e Percent donor engraftment in recipient BoyJ mice 4–20 weeks after secondary transplantation with bone marrow from primary CRU mice. f Percent donor engraftment after primary CRU assay at the 4- and 20-week timepoints from d. g Change in percent donor engraftment from 4 to 20 weeks post-transplantation. h Percent donor engraftment in bone marrow of recipient BoyJ mice 25 weeks after secondary transplantation. All data are mean ± SEM, two-way ANOVA with repeated measures followed by Bonferroni multiple comparisons (a, b, e, f), two-sample t-test (c, d, g, h); n = 5/group in a–h; dotted lines represent averaged data from naïve mice. Source data are provided as a Source Data file.To test whether SCI impairs the self-renewal capacity of long-term HSPCs, donor bone marrow was isolated from primary recipients after 19 weeks and then re-transplanted into new lethally irradiated recipient mice (Supplementary Fig. 2a). HSPC engraftment was identical between groups 4 weeks after secondary transplantation (Fig. 8e, f). However, no significant change in engraftment occurred throughout the evaluation period (6 months) in mice receiving SCI bone marrow; only mice receiving HSPCs that were originally derived from sham-injured mice demonstrated increases in donor engraftment (Fig. 8e–g). Importantly, SCI donor cells produced significantly fewer donor WBCs than sham-injured donor cells (Supplementary Fig. 2h). By 24 weeks after secondary transplantation, engraftment was significantly reduced in the bone marrow of mice receiving SCI donor cells compared with mice receiving sham-injured donor cells (Fig. 8h). These data indicate that SCI negatively affects the long-term self-renewal capacity of bone marrow HSPCs.DiscussionPrior studies in mice45 and humans22 found that SCI increases the total number of cells in bone marrow. Data in this report extend those observations and also provide new insight to help explain how/why after SCI, both humans24,41,53 and rodents42,43,54,55 exhibit prolonged hematological abnormalities marked by leukopenia and chronic immunologic dysfunction18,23,26,56. Specifically, data in this report reveal that traumatic SCI causes a bone marrow failure syndrome marked by excessive proliferation and sequestration of HSPCs, altering of cytokine and chemokine signaling within the bone marrow niche, failure to generate and mobilize mature lymphocytes, and chronic impairments in the clonogenic potential of HSPCs (Fig. 9). Bone marrow failure diseases develop when the bone marrow is unable to produce appropriate numbers of healthy mature white and red blood cells. Normal aging12,57 and various diseases including diabetes58,59, warts, hypogammaglobulinemia, infections, and myelokathexis (WHIM) syndrome60–62, glioblastoma63, and chemotherapy64 cause sequestration of mature and immature cells in the bone marrow. In each case, sequestration of bone marrow cells either causes or is associated with hematopoietic dysfunction. Both neural and humoral mechanisms undoubtedly participate in causing the aberrant sequestration phenomenon, although disease-specific mechanisms are likely.Fig. 9Summary of the effects of T3 transection SCI on bone marrow hematopoiesis.Under homeostatic conditions, HSPCs undergo highly regulated and short-lived proliferation (1). When demand dictates (i.e. stress, trauma, infection), normal bone marrow downregulates CXCL12-CXCR4 signaling (2), allowing HSPC mobilization and extramedullary hematopoiesis (3). Homeostatic bone marrow also allows for the coordinated differentiation (4) and mobilization (5) of all leukocyte lineages, maintaining normal circulating blood leukocyte levels. T3 transection SCI, however, causes chronic bone marrow failure, including protracted proliferation (1), excessive CXCL12-CXCR4 chemotactic signaling (2), and impaired mobilization leading to the accumulation of HSPCs within the bone marrow and loss of extramedullary hematopoiesis. T3 transection SCI also causes lymphopenia by reducing lymphopoiesis (4) and sequestering mature lymphocytes in bone marrow (5).HSPCs reside in a specialized perivascular bone marrow niche (i.e. microenvironment)1. It is in the niche that complex cellular and molecular cues orchestrated between HSPCs, CXCL12-expressing perivascular stromal cells and the autonomic nervous system, particularly sympathetic noradrenergic nerve fibers, control HSPC proliferation, regeneration, and differentiation8,10,11,65. Specifically, post-ganglionic sympathetic neurons release norepinephrine into the bone marrow, causing bone marrow stromal cells to decrease their expression of CXCL12 which, in turn, untethers CXCR4+ HSPCs and mature leukocytes from the niche, facilitating their egress into the circulation8,10,36. Loss of proper sympathetic tone in bone marrow, whether by experimental manipulation or by aging, causes hematopoietic dysfunction12,64. A similar neurogenic mechanism may cause bone marrow failure after SCI.After SCI, much or all of the tonic supraspinal control over the sympathetic nervous system is lost. Previously, using models of thoracic SCI, we showed that normal physiological stimuli (e.g., visceral afferent input from bladder/bowel) trigger exaggerated or uncontrolled sympathetic reflexes in spinal autonomic circuits14–17. However, similarly aberrant sympathetic reflexes may be triggered even when SCI occurs below the caudal-most sympathetic preganglionic neurons located in lower thoracic/upper lumbar spinal cord. Anatomical tracing studies indicate that femoral bone marrow is innervated by both sympathetic and sensory nerves originating as far rostral in the spinal cord as T8-9 and as far caudal as the sacral spinal cord7,66,67. This is a large segment of spinal cord through which sensory input from bone marrow could activate propriospinal relay neurons and multi-segmental sympathetic reflexes13. Depending on which relay neurons are activated and the relative integrity (anatomical or functional) of the intersegmental circuitry, bone-specific intersegmental sympathetic reflexes may be exaggerated or silenced after SCI. At the level of the bone marrow niche, either outcome would be perceived by stromal cells and HSPCs as a break in homeostasis. Under pathological conditions, when sympathetic tone to bone marrow is disrupted, CXCL12 and CXCR4 levels increase and HSPC mobilization is impaired10,12,59,64. Since proper descending modulation of spinal sympathetic reflexes is never restored after traumatic SCI and also intersegmental neuronal circuitry continues to undergo structural and functional plasticity, neurogenic control of bone marrow function may never be restored after SCI. Evidence for permanent bone marrow failure is apparent from data in Fig. 7, which revealed that a potent hematopoietic stimuli (endotoxin) does not effectively mobilize HSPCs and mature lymphocytes from the bone marrow of mice at chronic post-injury periods.SCI-induced changes in circulating hormones, glucocorticoids in particular, are also likely culprits underlying acute and chronic bone marrow failure. Normally, circulating glucocorticoids act on diverse cell types in the periphery and the nervous system to help maintain HSPCs and the bone marrow niche68. Glucocorticoids also promote HSPC homing from the bone marrow, an effect that is mediated by glucocorticoid receptor-mediated induction of CXCR4 transcription in HSPCs69. The acute physical stress of spinal trauma and the sustained activation of aberrant sympathetic–neuroendocrine reflexes cause a primary hypercortisolism after SCI14–16,19. As a result, circulating levels of glucocorticoids increase within 24 h post-injury then remain at supraphysiological concentrations indefinitely after SCI. The development of aberrant sympathetic–neuroendocrine reflexes may explain why, in the present study, SCI always causes HSPC sequestration, regardless of injury level (see Fig. 4). Future studies are needed to explore in-depth how SCI affects neural–humoral signaling in bone marrow and to what extent glucocorticoids and other blood-borne factors (e.g., microbial metabolites) contribute to acquired bone marrow failure.An important observation made in this report, and one that may have an immediate impact on people affected by SCI, is that it is possible to overcome certain aspects of SCI-induced bone marrow failure. Specifically, we found that the FDA-approved drug and CXCR4 antagonist Plerixafor, when injected post-injury, effectively liberates HSPCs and mature leukocytes from the bone marrow of SCI mice (Fig. 6). In SCI patients, Plerixafor could be a potentially safe and effective way to mobilize cells from the bone marrow niche to help restore immune function50. Indeed, Plerixafor can safely reverse immunodeficiency in WHIM patients62. Still, to minimize the risk associated with boosting the immune system and exacerbating neuroinflammatory-mediated injury or trauma-induced autoimmunity, both of which can impair neurological recovery, more research is needed to define the optimal therapeutic conditions for Plerixafor treatment after SCI.Another intriguing aspect of SCI-induced bone marrow failure was revealed during the competitive repopulating unit (CRU) assays (Fig. 8). Specifically, we found evidence that the long-term clonogenic potential of HSPCs is impaired and that these long-term effects are imprinted in HSPCs early after injury. Indeed, bone marrow cells isolated from SCI donor mice 3-day post-injury, when transplanted into irradiated naïve mice, engrafted the irradiated bone marrow faster than bone marrow cells obtained from sham-injured mice. This early advantage could be explained by improved homing to the bone marrow by HSPCs from SCI donors, perhaps because of glucocorticoid-mediated enhancement of CXCR4 on bone marrow LSK cells (Fig. 6). However, this repopulation advantage was transient as deficits in the long-term repopulation or clonogenic potential of SCI bone marrow became obvious after secondary transplantation (Fig. 8). Currently, we do not know why SCI impairs the long-term function of HSPCs or how this occurs within only 3 days. Since we did not see an increase in the DNA damage marker γH2AX after chronic SCI (Fig. 2), it is unlikely that excessive HSPC proliferation after SCI causes DNA damage70. However, HSPC exhaustion is possible as cell cycle number has been shown to inversely correlate with long-term HSPC function71,72. It is also possible that aberrant sympathetic–neuroendocrine reflexes influence the clonogenic potential of HSPCs; both catecholamines and glucocorticoids can cause epigenetic modifications in cells, imprinting then with new functional identities69,73. In the context of monocytes/macrophages, for example, exposure of these cells to certain stimuli endows them with enhanced microbicidal functions against secondary infections74. This trained immunity in monocytes/macrophages is orchestrated by epigenetic reprogramming, which also occurs in HSPCs75–77. Perhaps SCI creates an environment in the bone marrow niche that favors the induction of a form negative trained immunity. Regardless of mechanism, our data are consistent with those showing that the clonogenic potential of hematopoietic and stromal cells isolated from SCI patient bone marrow is impaired21,22,78.In conclusion, data in this report reveal that SCI-induced bone marrow failure is caused by cell-intrinsic and extrinsic mechanisms; impaired control of HSPC proliferation and sequestration is likely a cell-extrinsic phenomenon regulated in the niche while deficits in HSPC clonogenic potential are likely the result cell-intrinsic changes. Bone marrow failure develops soon after injury but has long-lasting adverse effects on the host HSPCs. For example, impaired bone marrow function may limit the development of an effective immune system, perhaps explaining the higher incidence of infectious morbidity and mortality in SCI patients. Also, SCI-induced bone marrow failure may preclude the use of bone marrow cells from SCI donors as a transplantation source. A similar limitation was recently described for bone marrow isolated from mice with experimental CNS autoimmune disease79,80. Together, these data highlight the bone marrow as a previously underappreciated therapeutic target for improving health outcomes and quality of life after SCI.MethodsMice and housingThe Institutional Animal Care and Use Committee of the Office of Responsible Research Practices at The Ohio State University approved all animal protocols for this study. All experiments were performed in accordance with the guidelines and regulations of The Ohio State University and outlined in the Guide for the Care and Use of Laboratory Animals from the National Institutes of Health. Female and male C57BL/6 mice (strain #000664; CD45.2) were purchased from The Jackson Laboratory (Bar Harbor, ME), female repTOP mitoIRE luciferase mice (Mito-luc) were purchased from Charles River Laboratories (Wilmington, MA), female B6.SJL-Ptprca Pepcb/BoyJ mice (C57BL/6-CD45.1; BoyJ) were bred in-house from adult breeding pairs originally purchased from The Jackson Laboratory (strain #002014), and male and female NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ mice (NSG mice) were bred in-house from adult breeding pairs purchased from The Jackson Laboratory (strain #005557). Animals were fed commercial food pellets and chlorinated reverse osmosis water ad libitum and housed (≤5/cage) in ventilated microisolator cages layered with corn cob bedding in a 12-h light–dark cycle at a constant temperature (20 ± 2 °C) and humidity (50 ± 20%). All mice were housed in a specific pathogen-free housing facility with routine testing of sentinel mice for specific pathogens. Generation of NSG mice with human immune systems (hNSG mice) was performed as follows38–40. Newborn NSG pups (24–72 h postnatal) received 1 Gy whole-body X-ray irradiation (RS 2000, Rad Source, Suwanee, GA), followed immediately by engrafting 1–5 × 104 human umbilical cord CD34+ stem cells (Lonza Incorporated, Walkersville, MD or Stemcell Technologies, Vancouver, BC) via intrahepatic injection. Body temperature was maintained at 37 °C using a heating pad until pups were returned to their dams for normal maturation and weaning 21–24 days postnatal.SCI and animal careAdult C57BL/6 mice (10–16 weeks old) and hNSG mice (16–24 weeks old) were used for SCI experiments. Mice were subjected to a complete spinal cord transection injury at the third thoracic, ninth thoracic, or sixth lumbar spinal levels. Mice subjected to laminectomy (Lam or sham injury) only served as controls, and naïve mice were used as a reference for outcomes. Mice were anesthetized with ketamine (120 mg kg−1, i.p.) and xylazine (10 mg kg−1, i.p.) for all surgical procedures and provided prophylactic antibiotic treatment with gentamicin sulfate (5 mg kg−1, s.q.). Aseptic conditions were maintained during all surgical procedures and mice were placed on a warming pad to regulate body temperature. Hair was shaved at the region of the thoracic spinal cord and skin was treated with a sequence of betadine, 70% ethanol, and betadine. A small midline incision was made to expose the vertebra and then a partial laminectomy was performed. The meninges were cut using micro-scissors and then spinal cord transected using micro-scissors and a sterile glass aspiration tube for suction of fluid/blood, stabilization of the spinal cord during transection, and to confirm completion of injury. Muscle overlaying the injury site was sutured, followed by closure of the wound with sutures or staples. After surgery, mice were placed in cages on heating pads and monitored frequently until they recovered consciousness and were moving spontaneously. Mice were given fluids (1–2 mL 0.9% sterile saline) to maintain hydration and softened food to eat ad libitum as they recovered. Bladders were expressed at least two times daily to maintain urinary function, and urine underwent periodic pH testing to identify bladder infections. Gentocin antibiotic was subcutaneously administered once daily at 5 mg kg−1 for 5 dpi. For contusion studies, identical surgical procedures were followed, with the exception of laminectomy level (T9). An Infinite Horizon Impactor (Precision Systems and Instrumentation, Lexington, KY) was used to generate a moderate 70 kdyne injury at either the third or ninth thoracic spinal levels.IVIS imagingMito-luc mice were injected with 80 mg kg−1 (i.p.) D-luciferin potassium salt (Cayman Chemical, Ann Arbor, MI), anesthetized with isoflurane (2.5–4% vaporized in oxygen, delivered at 1 L min−1), and placed on a heated surface within the IVIS Lumina II system (Caliper Life Sciences, Hopkinton, MA) for image acquisition. Imaging of long bones in limbs required securing forelimbs at 90° from midline, and hindlimbs 45° from midline, with small pieces of translucent medical tape. Mice were kept on 1.5–2% isoflurane throughout the duration of imaging. Baseline levels of mitosis were determined prior to SCI, and then mice were imaged at 1, 3, 7, 14, 21, and 28 dpi. Data were analyzed with the Living Image® software (v.4.3.1; Caliper Life Sciences). Bioluminescence was measured by defining regions of interest (ROIs) of defined sizes around the sternum, left femur/tibia, and right femur/tibia. Total flux (photons s−1) and the maximum radiance for each ROI was determined, and an average was calculated for both left and right femur/tibia ROIs. Data were plotted for each individual animal as either raw values (post-SCI study; Fig. 1) or percent of pre-LPS (chronic SCI plus LPS study; Fig. 7).Plerixafor treatmentAfter SCI or sham surgery, mice were injected subcutaneously with 5 mg kg−1 Plerixafor (AMD3100; Sigma-Aldrich) in 0.9% sterile saline. First dose was given 1-h post-injury, and then once a day until 3 dpi. Mice were terminally anesthetized with ketamine and xylazine 1 h after a final dose of AMD and tissues were collected as described below.Stimulating hematopoiesis with systemic endotoxinMito-luc transgenic and wild-type mice underwent SCI or sham surgery as previously described. Approximately 6–8 weeks after injury mice were injected i.p. with 1 mg kg−1 LPS (E. coli O55:B5, Sigma-Aldrich) in 0.9% sterile saline once per day for 3 days. Mito-luc mice underwent IVIS imaging at 2, 4, 7, 9, 11, and 14 days post-LPS (first dose). Wild-type mice underwent submandibular bleeds prior to LPS, 1-h post-LPS, 4 days post-LPS, and 11 days post-LPS. Blood was collected into an EDTA-coated capillary tube (Sarstedt Inc.; Thermo Fisher Scientific, Waltham, MA).Tissue collection and processingMice were terminally anesthetized with ketamine and xylazine for euthanasia and tissue collection. Blood was then collected via cardiac puncture and placed in blood collection tubes coated with EDTA. Blood was then treated with ammonium chloride-based red blood cell (RBC) lysis buffer and resuspended in Iscove’s Modified Dulbecco’s Medium (IMDM) with 2% fetal bovine serum (FBS) for downstream MethoCult assays or 0.1 M phosphate buffer saline (PBS) with 2% FBS (flow buffer) for flow cytometry. Spleens were rapidly isolated, weighed, and placed in Hank’s Balanced Salt Solution (HBSS). Spleens were minced with sterile dissection scissors, smashed through a 40-μm sterile filter using the plunger of a 3-mL syringe, and rinsed with 10 mL of HBSS or IMDM. Mouse femurs and tibiae were removed, cleaned, and placed in a small volume of HBSS. Bone marrow cells were isolated by either flushing bones with 10 mL of HBSS or crushing in a mortar and pestle and washed with media. Cell counts were obtained by a standard hemocytometer (bone marrow and spleen), or with a Hemavet 950 fs multi-species hematology (blood; Drew Scientific, Miami Lakes, FL) system capable of analyzing whole blood with 5-part white blood cell differential, platelets, and RBCs.Immunolabeling and flow cytometry2–10 × 106 bone marrow cells and splenocytes, or approximately 50 μL RBC-lysed blood, were allocated for flow cytometry analysis. All antibodies were used at a 1:100 dilution for staining purposes. BD StemflowTM Mouse Hematopoietic Stem Cell Isolation Kit (BD Biosciences, cat #560492) was used to label lineage−, c-Kit+, Sca-1+ HSPCs. Mouse antibody lineage cocktail (BD Biosciences; cat #558074) contained the following APC-conjugated antibodies: CD3 (145-2C11), CD11b (M1/70), CD45R/B220, TER-119, and Ly6G/C (RB6-8C5). Fc receptors were blocked for 15 min using rat anti-mouse CD16/32 antibody (BD Biosciences, cat #553142), followed by labeling with antibodies for 60 min. Dead cells were labeled with eFluor780 (eBioscience, cat #65-0865-14) approximately 30 min into antibody incubation. Labeled cells were fixed and permeabilized with BD Cytofix/CytopermTM solution (BD Biosciences, cat #554722) for 20 min. For cell cycle analysis, DNA was labeled with DAPI (BD Biosciences, cat #564907) in flow buffer with 0.1% Triton X-100 for 20 min after antibody labeling. For human HSPCs, antibodies for phosphorylated γH2AX (2F3; BioLegend cat #613414) were used to measure replication stress in c-Kit+ HSPCs. For total CXCR4 receptor expression in LSK cells, mature bone marrow cells were depleted using Lineage Cell Depletion Kit and MACS system as per the manufacturer’s protocol (Miltenyi Biotec, cat #130-090-858, Auburn, CA), followed by cell surface staining for LSK markers, fixation, permeabilization with BD Perm/Wash, and staining for CXCR4 (2B11; BD Biosciences, cat #562738). All incubations were performed at 4 °C, followed by a wash step using excess flow buffer, and centrifugation for 5 min at 4–10 °C. Antibodies for mouse lineage (BD Biosciences, cat #560492), human lineage (Invitrogen, cat #22-7778-72), human CD34 (581; BD Biosciences, cat #555824), and human CD38 (HIT2; BD Biosciences, cat #560677) were used to identify human HSPCs (Fig. 4d, Supplementary Fig. 1). Antibodies for CD3 (17A2; BD Biosciences, cat #564008), CD4 (RM4-5; BD Biosciences, cat #553052), CD24 (M1/69; BD Biosciences, cat #562563), CD43 (S7; BD Biosciences, cat #562865), CD45/B220 (RA3-6B2; BD Biosciences, cat #552772), IgM (II/41; BD Biosciences, cat #562032), and IgD (11-26c.2a; BD Biosciences, cat #562022) were used for analysis of bone marrow B and T cells (Fig. 5, Supplementary Fig. 1). Lineage cocktail (BioLegend, cat #133307) and antibodies for CD117 (2B8; BioLegend, cat #105827), Sca-1 (D7; BioLegend, cat #108142), CD48 (HM48.1; BioLegend, cat #103423), CD150 (TC15-12F12.2; BioLegend, cat #115916), CD135 (A2F10; BioLegend, cat #135305), and CD16/32 (93; BioLegend, .cat #101327) were used to determine LT-HSC/MPP1, ST-HSC, MPP2, MPP3, MPP4, GMP, and CMP/MEP subsets of HSPCs after LSK gating (Fig. 2). LSR II and LSR Fortessa flow cytometers (BD Biosciences) were used to analyze samples. Forward scatter and side scatter parameters were used to gate viable cell populations for phenotypic analysis (Supplementary Fig. 1). Positive and negative cell populations were selected based on staining with isotype control antibodies and fluorescent minus one control. Offline data analysis was completed with FlowJo v.10 software (Tree Star, Inc., Ashland, OR).MethoCult CFC assayThe MethoCultTM GF M3434 ex vivo culture assay (Stemcell Technologies, Vancouver, BC) consists of a methocellulose media containing cytokines and growth factors to support the development of cell colonies from single HSPCs. Bone marrow cells were plated at concentrations ranging from 7.5 × 103, splenocytes at 2 × 105, and RBC-lysed blood at 150 μL in a total of 1 mL of MethoCult media. Samples were plated in meniscus-free six-well SmartDishTM (Stemcell Technologies) and placed in an incubator at 37 °C and 5% CO2. Approximately 10–12 days after plating, colonies were quantified by standard inverted light microscopy using a StemGridTM counting underlay (Stemcell Technologies). Colonies were identified as blast forming unit-erythrocyte, granulocyte, monocyte, granulocyte/monocyte, or granulocyte/erythrocyte/monocyte/megakaryocyte based on colony composition and cell morphology. MethoCultTM H4034 Optimum was used for quantification of human CFCs isolated and purified from humanized mice.CRU assayThe CRU assay was used to assess repopulation potential of whole-bone marrow cells isolated from naïve, lam, and SCI mice. Bone marrow cells (5 × 106) from donor C57BL/6 mice (3 dpi) expressing leukocyte antigen CD45.2 were isolated and mixed with equal numbers of rescue bone marrow from BoyJ mice expressing leukocyte antigen CD45.1. This mixture was injected via lateral tail vein into lethally irradiated BoyJ recipient mice (4–6 weeks old; total 9 Gy split into two doses of 4.5 Gy approximately 24 h apart; gamma irradiation from Cesium-137). Donor chimerism was assessed every 4 weeks after engraftment via flow cytometry for CD45.1 (BoyJ) and CD45.2 (donor) peripheral blood leukocytes and expressed as a percent of total leukocytes: CD45.2/(CD45.1+CD45.2). Bone marrow, spleen, and blood were collected 19 weeks after engraftment for donor chimerism and blood phenotyping using CD4, CD8, CD45/B220, and CD11b antibodies. Secondary engraftment was performed after isolating donor CD45.2 bone marrow cells (5 × 106) from primary recipients using magnetic bead antibodies and columns. Isolated cells were mixed with equal numbers of newly isolated CD45.1 BoyJ bone marrow and engrafted into new BoyJ recipients.RT-qPCR of cytokine and chemokine mRNABone marrow was isolated and pooled from tibia and femurs of laminectomy and T3Tx mice (n = 3/group). RNA was isolated using TRIzol as per the manufacturer’s instructions. Genomic DNA was eliminated using 1 µg µL−1 DNase I (Invitrogen). One microgram DNase-treated RNA was primed with random hexamers (1 µM; Applied Biosystems) and reverse transcribed into cDNA using SuperScript II reverse transcription (Applied Biosystems) in a 20-µL reaction. RNase-free sterile water was used to dilute 1 µg cDNA 1:10, loaded onto a 96-well plate, and then primers were loaded (Supplementary Table 1). SYBR Green Master Mix (Applied Biosystems) was used to detect amplified cDNA. Melting point curves assessed the quality of each reaction. Samples were run in triplicates, gene expression was normalized to s18 control samples using the delta-delta CT method, and SCI data were expressed as fold increase of sham-injured values.Quantification of CXCL12 in bone marrow extracellular fluidBone marrow was isolated from two tibia and two femurs per mouse using 1 mL ice cold PBS. After light trituration, cells and debris were separated from extracellular fluid by centrifugation for 5 min at 400 × g. Fluid was aliquoted and frozen in liquid nitrogen. A Mouse CXCL12/SDF-1α Quantikine ELISA kit (R&D Systems) was used as per the manufacturer’s protocol. Bone marrow extracellular fluid preps were measured undiluted. Plates were read at 450 nm wavelength on a SpectraMAX190 and analyzed using the SoftMax Pro software (Molecular Devices, San Jose, CA).Statistics and reproducibilityAll data are represented as mean ± standard error of the mean (SEM), with individual data points representing independent biological replicates. Group sizes were determined by analyzing preliminary and published data; using G*Power (v3.1), an n = 4 was found to be sufficient to detect a 1.25-fold change with a coefficient of variation of 20% and >80% power for flow cytometry and MethoCult assays. To compensate for unexpected morbidity/mortality, an n = 6/group was determined. Except for the CRU assay, reference data from naïve C57BL/6J (10–16 weeks old over) were collected during development and testing of endpoint assays and are included as dotted lines on graphs. Data were excluded from analysis only if identified as a statistical outlier by Grubbs’ test with proof of deviation from standard recovery after SCI (i.e. bladder infection, post-operative autophagia, etc.). A single female T3 transection animal met these criteria and was removed from Figs. 3 and 4 due to a documented bladder infection and blood CFCs > 800 mL−1. To compare groups of SCI (T3Tx) and sham, when only one time point involved, we used two-sample t-tests (with Welch’s t-test for unequal variances). For the experiment comparing the effect among sham, low-thoracic, and high-thoracic injuries (T3Tx vs T9Tx; Fig. 4a, b), one-way ANOVA was performed followed by Tukey post hoc comparisons. For the longitudinal measures (e.g. proliferation, BM engraftment; Figs. 1, 7, and 8), mixed effect or two-way ANOVA with repeated measures was conducted followed by Bonferroni multiple comparisons. All statistical tests were two-tailed. Exact p values denoted on graphs; *p < 0.05 (Fig. 5b). Samples were blinded either during processing or prior to analysis by a separate experimenter not involved in the analysis. All attempts at replication were successful: BM (Fig. 1a–c) and HSPC (Fig. 1d) proliferation data are from two independent experimental replications. Accumulation of HSPCs was demonstrated in Fig. 1e–h, and then verified in an independent experimental replicate in Fig. 2. Accumulation of cells in tibia (Fig. 1h) has been confirmed with an independent experimental replication not included in the manuscript. Impaired HSPC mobilization after SCI was demonstrated in independent experimental replications using female and then male wild-type mice (Fig. 3) and further replicated in other SCI models (Fig. 4). Impaired HSPC mobilization was then confirmed with mice with human HSPCs (Fig. 4c, d). In total, impaired HSPC mobilization after SCI was confirmed in at least six independent experimental replications. Data demonstrating lymphopenia after T3 SCI (Fig. 5a, b) were from two independent experimental replications. Data demonstrating enhanced CXCL12-CXCR4 levels (Fig. 6b, c) are from two independent experimental replications. Data demonstrating that Plerixafor liberates HSPCs from bone marrow after SCI (Fig. 6d–k) were independently verified in two additional experiments not included in the manuscript. Data were analyzed using GraphPad Prism software v5.0 (GraphPad Software Inc., San Diego, CA). Illustrations were created with BioRender scientific illustration software paid subscription (biorender.com). Figures were generated in Adobe Photoshop CS5 v12 (Adobe Systems Inc., San Jose, CA).Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary | nature communications | [
"Article"
] | [
"Haematopoiesis",
"Neuroimmunology",
"Spinal cord injury",
"Haematopoietic stem cells"
] | adults mature immune cells develop from hematopoietic cells hematopoiesis in bone marrow cellular interactions regulate renewal of immune cells maintenance bone marrow hematopoiesis essential for host defense tissue repair proliferation differentiation retention/release of bone marrow cells controlled by neuroendocrine hormones autonomic nervous system sympathetic After spinal injury brain connections sympathetic neurons lost decentralized spinal autonomic network aberrant sympathetic neuroendocrine reflexes13 Uncontrolled sympathetic reflexes implicated in overstimulation cytotoxicity of mature disruption of leukocyte homing after SCI19 decentralized bone marrow may impair hematopoiesis to chronic immune dysfunction data from two studies SCI impairs human bone marrow stem cell SCI increased proliferation numbers HSPCs in form mature immune cells impaired21 goal current study to determine hematopoietic dysfunction after acute chronic SCI identify molecular mechanisms SCI-induced impairments hypothesize SCI induce acute chronic bone marrow failure in mice similarSCI causes rapid chronic bone marrow failure excessive HSPC proliferation accumulation impaired function post-injury injections Plerixafor CXCR4 liberates HSPCs failure extramedullary hematopoiesis Treating bone marrow failure after SCI may chronic immune dysfunction anemia increases bone marrow cell proliferationHematopoiesis requires HSPCs differentiate in bones including sternum SCI monitored post-injury changes in sternum femur/tibia of transgenic female mice used peak average bioluminescence measures bone marrow proliferation SCI mice compared with sham-injured mice laminectomy Within 72 h of surgery cellular proliferation increased in sternum femur sham-injury SCI mice enhanced proliferation maintained only in SCI mice peaking first 7 days post-injury persisting to 1-month post-injury (Fig. 1b SCI protracts HSPC proliferation causes accumulation in bone marrow Bioluminescent imaging of sternum femur/tibia pre-injury 7Bioluminescence max radiance total flux region left right femur/tibia averaged signals 1 to 28 dpi proliferating lineage Sca-1+ c-Kit+ (LSK bone marrow cells S–G2–M phase 3 dpi histograms LSK proliferation 3 dpi Quantification LSKs c-Kit+ progenitors total bone marrow cells at 3 7 28 dpi Images heatmaps tibia pairs Lam SCI mice increased optical density increased bone marrow cellularity data mean SEM mixed effect model repeated measures two-sample t-test two-way ANOVA Bonferroni comparisons n = 4 7/group 10/group 5 (7 6 (28 dpi Dotted lines average data naïve mice post-injury Lam laminectomy Source data file enhanced proliferation HSPCs multi-color flow cytometry proliferating Sca-1+ c-Kit+ (LSK cells femoral bone marrow C57BL/6 wild-type mice after SCI 3 dpi LSK cell proliferation increasedcohorts mice effect SCI on LSK cells confirmed long lasting (Fig. differentiated c-Kit+ progenitors mature bone marrow cells increased until 28 dpi (Fig. 1g data indicate reactive bone marrow response after SCI protracted proliferation accumulation of HSPCs.SCI causes chronic expansion bone marrow HSCsThe LSK bone cells Fig. 1 includes stem multipotent progenitor cells31 SCI multi-color flow cytometry cells sham-injured SCI mice (Fig. 2a total number bone marrow cells including LSK cells increased after SCI increase affects subsets stem progenitor cells including long-term repopulating stem cells short-term HSCs multipotent progenitor subset 2 cells myeloid-primed MPP3s lymphoid-primed MPP4s granulocyte–monocyte progenitors (Fig. 2d–i). SCI cause increase expression γH2AX marker of double-stranded DNA breaksdata SCI causes proliferation HSPCs long-term replication stress. SCI causes accumulation hematopoietic stem cells differentiated progenitors Hematopoietic hierarchy HSPCs cell surface markers bone marrow cells LSKs long-term HSCs short-term HSCs multipotent progenitors granulocyte–monocyte progenitors at 28 dpi fluorescence intensity yH2AX expression replication stress c-Kit+ HSPCs at 28 dpi data mean SEM two-sample t-test n = 9 10/group = 5 4/group Dotted lines average data naïve mice post-injury Source data prevents HSPC mobilization bone marrowEnhanced HSPC proliferation after SCI compensatory response blood loss trauma psychological physical stressors enhance bone marrow hematopoiesis extramedullary hematopoiesis tissues acute effect surgical stress bone marrow hematopoiesis Three days after surgery HSPCs found in blood spleen-injured mice increase spleen size 3 days after SCI fewer HSPCs entered circulationcolonized spleen SCI mice (Fig. 3b fewer c-Kit+ HSPCs spleen (Fig. circulation after SCI. 3T3 transection SCI prevents HSPC mobilization spleen at 3 dpi Outcomes female mice at 3 dpi CFCs per mL blood per spleen Spleen weight splenocytes% LSK c-Kit+ cells spleen Proportion proliferating c-Kit+ cells parameters male mice at 3 dpi data mean ± SEM two-sample t-test n = 15/group 10/group 6/group f 4 5/group (8 9 Dotted lines average data naïve mice days post-injury Lam laminectomy CFCs colony forming cells Source data effects SCI on HSPC sequestration independent sex abolished HSPC mobilization in female male SCI mice Fig. 3a–c g–i). SCI-dependent effects HSPCs influenced by injury severity spinal injury level HSPCs fail enter circulation after spinal transection injuries L6/S1 T9 T3 spinal levels incomplete spinal contusion T3 T9. SCI impairs mouse human HSPC mobilization CFCs per mL blood after T3 T9 L6/S1 transection injury after T3 T9 contusion injury hCFCs splenic hCFCs hCD34hi/hCD38low human HSPCs spleen weight hNSG mice 3–5 dpi data mean ± SEM one-way ANOVA Tukey comparison test two-sample t-test n = 20 15 14 5/group a 6, 10 5/group b 8/group c 6 6 4 8 9/group d days post-injury laminectomy Source data file-dependent effects HSPCs species-specific male female mice with human HSPCs immune systems SCI prevented human HSPCs circulation spleen (Fig. 4c data Figs. 1–3 indicate SCI triggers proliferation HSPCs cells leave bone marrow HSPCs accumulate HSPC sequestration independent of sex injury level severity species HSPCs respond impairs B cell development decentralized bone marrow after SCI sequester cells other than HSPCsanimal human circulating leukocytes decrease after SCI stroke19,24 confirmed expanded data model complete SCI (T3 spinal 3 dpi T3 transection SCI reduced blood lymphocytes. 5a mature lymphocytes bone marrow B220hi B cells CD3+/CD4− T cells CD3+/CD4+ T cells (Fig. SCI reduced B220low B cells ProB in bone marrow data impaired B lymphopoiesis after SCI stroke19.Fig. 5T3 SCI causes acute lymphopenia sequestration lymphocytes bone marrow white blood cells lymphocytes neutrophils monocytes per mL blood at 3 dpi Lymphocytes neutrophils monocytes eosinophils/basophils WBCs Proportion immature B cells mature B cells CD3+ T cells in bone marrow Flow cytometry gating bone marrow B220+ B cells reduced immature mature B cells with SCI scatter characteristics B cells bone marrow after injury SCIFlow cytometry bone marrow CD3++ T cell increased SCI data mean ± SEM *p < 0.05 two-sample t-test n = 12 14/group a b 3/group c–e Shades blue/red/SCI sample plots d e Dotted lines average data naïve mice days post-injury Lam laminectomy FSC forward scatter BM bone marrow Source data.SCI increases CXCL12-CXCR4 signaling bone marrow IL1β TNFα TGFβ CCL2 CXCL12 CXCR4 development function proliferation retention HSPCs SCI mRNA-bone marrow cells T3 SCI sham-injured mice PCR analyses SCI increased expression bone marrow cytokines chemokines Cxcl12 receptor Cxcr4 increased threefold 6Plerixafor (AMD3100 mobilizes HSPCs rescues extramedullary hematopoiesis boosts immune cells T3 SCI Cytokine chemokine mRNA expression marrow cells SCI mice CXCL12 protein expression bone marrow extracellular fluidProtein expression CXCR4 LSKs bone marrow CFCs per mL blood spleen LSKs c-Kit+ HSPCs total cells per spleen 3 days after SCI AMD3100 treatment white blood cells lymphocytes neutrophils monocytes per mL blood Proportion monocytes lymphocytes neutrophils blood = 5 Total cells per femur data mean ± SEM two-sample t-test n = 3/group a 6 7/group b 7/group c 5/group d–k Dotted lines average data lam naïve mice bone marrow MFI fluorescence intensity WBCs white blood cells MO monocytes LY lymphocytes NE neutrophils AMD Plerixafor Source data CXCL12 binding CXCR4 HSPCs influences maintenance retention HSPCs bone marrow dependent sympathetic branch autonomic nervous validated effects high-level SCI chemokine receptor pair SCI increased secreted CXCL12 protein bone marrow extracellular fluid expression CXCR4 receptor LSK cellstest SCI-dependent increase CXCL12/CXCR4 expression HSPC sequestration bone marrow mice treated Plerixafor (AMD3100), antagonist CXCR4 3 days 1 h post-SCI Plerixafor-treated mice HSPCs released blood increased >6-fold HSPCs spleen Plerixafor reversed post-SCI leukopenia splenocytes circulating WBCs increased releasing cells from bone marrow Plerixafor ablate bone marrow increased bone marrow cellularity implicate sustained CXCL12/CXCR4 signaling potential SCI-induced sequestration HSPCs post-SCI sequestration HSPCs cells overcome using FDA-approved Plerixafor.Chronic SCI impairs bone marrow responses inflammatory causes acquired bone marrow failure syndrome protracted HSPC proliferation sequestration peripheral lymphopenia develops 72 h post-injury persists 1-month post-injury failed bone marrow mice challenged with endotoxin (1 mg kg−1 inducer HSPC proliferation mobilization34Mito-luc transgenic C57BL/6J mice injected LPS daily 3 days 6 weeks after T3 SCI surgery. hematopoiesis LPS chronic-injured mice recapitulates bone marrow responses Experimental outline T3 transection SCI sham injury 6–8 weeks recovery LPS stimulation (1 mg kg−1 3 Mice vivo imaging bone marrow proliferation CBC MethoCult assay Bioluminescent images Lam SCI mice before 11 days post-LPS total flux sternum femur/tibia regions pre-LPS CFCs per mL blood 4 11 days post-LPS Lymphocytes neutrophils 4 days post-LPS data mean ± SEM two-way ANOVA repeated measures two-sample t-test n = 4/group = 5/group Dotted lines average data naïve mice days post-LPS first dose Lam laminectomy Source data Mito-luc transgenic mice cellular proliferation bone marrow decreased 2–4 days after first LPS injection. LPS stimulates cellscompared to non-LPS mice LPS increased HSPCs fewer HSPCs detected in SCI mice at peak mobilization recovery (11 LPS stimulation reduced circulating lymphocytes in SCI mice not sham-injured mice 7 days after LPS injections basal cell proliferation restored in sham-injured mice HSPCs repopulated depleted bone marrow49 LPS-induced proliferation continues in SCI mice luciferase signal overshoots baseline proliferation by >200% 9 14 days post-LPS characteristics bone marrow failure after SCI recapitulated in chronic SCI animals after LPS injections indicate SCI causes chronic permanent bone marrow failure.SCI impairs function bone marrow HSPCsData sequestration accumulation of HSPCs after SCI due to aberrant cytokine chemokine signaling bone marrow HSPCs hematopoiesis performed vivo competitive repopulating assays HSPCs removed from SCI mice 3 days after T3 SCI or sham surgery corresponds with enhanced HSPC proliferation sequestration no difference in total HSPC numbersbone marrow cells injected uninjured mice lethal doses irradiation SCI-derived donor cells faster engraftment at 8 weeks 19 weeks post-engraftment stable engraftment all mice donor cell source enhanced engraftment potential SCI cells waned after 8 weeks long-term lineage biasing engrafted SCI donor cells through 19 weeks data indicate deficit in self-renewal capacity differentiation long-term hematopoietic stem cells from SCI donors (Fig. 8c SCI impairs long-term clonogenic potential bone marrow by 3 dpi Percent donor engraftment BoyJ mice 4–19 weeks after primary transplantation with bone marrow from SCI mice engraftment after CRU assay 8- 19-week timepoints Change to 19 weeks post-transplantation engraftment BoyJ mice 19 weeks after primary transplantation engraftment 4–20 weeks after secondary transplantation CRU engraftment after 4- 20-week Change 20 post engraftment BoyJ mice 25 weeks after secondary transplantationdata mean ± SEM two-way ANOVA repeated measures Bonferroni comparisons two-sample t-test n = 5/group a–h dotted lines averaged data naïve mice Source data file SCI self-renewal long-term HSPCs donor bone marrow isolated from recipients after 19 weeks re-transplanted into new lethally irradiated mice HSPC engraftment identical groups 4 weeks after secondary transplantation no significant change engraftment mice receiving SCI bone marrow sham-injured mice increases engraftment SCI donor cells produced fewer WBCs sham-injured cells 24 weeks after secondary transplantation engraftment reduced mice SCI donor cells sham-injured data indicate SCI negatively affects self-renewal bone marrow HSPCs studies SCI increases number cells bone marrow Data report observations insight after SCI rodents42 exhibit prolonged hematological abnormalities leukopenia chronic immunologic dysfunction18traumatic SCI causes bone marrow failure syndrome excessive proliferation sequestration HSPCs altering cytokine chemokine signaling failure generate mobilize mature lymphocytes impairments clonogenic potential HSPCs Bone marrow failure diseases develop produce healthy mature white red blood cells diseases warts hypogammaglobulinemia infections myelokathexis glioblastoma63 chemotherapy64 cause sequestration mature cells sequestration causes hematopoietic dysfunction neural humoral mechanisms sequestration disease-specific mechanisms likely effects T3 SCI on bone marrow hematopoiesis homeostatic conditions HSPCs proliferation bone marrow downregulates CXCL12-CXCR4 signaling HSPC mobilization extramedullary hematopoiesis Homeostatic bone marrow allows coordinated differentiation mobilization leukocyte lineages normal leukocyte levels T3 SCI causes chronic bone marrow failure protracted proliferation excessive CXCL12-CXCR4 chemotactic signaling impaired mobilization accumulation HSPCs loss extramedullary hematopoiesis causes lymphopenia reducing lymphopoiesis mature lymphocytesHSPCs in perivascular bone marrow niche cellular molecular cues between HSPCs CXCL12-expressing cells autonomic nervous system sympathetic noradrenergic nerve fibers control HSPC proliferation regeneration differentiation8,11,65 post-ganglionic sympathetic neurons release norepinephrine into bone marrow expression CXCL12 untethers CXCR4+ HSPCs mature leukocytes egress into circulation8 Loss of sympathetic tone causes hematopoietic dysfunction12 mechanism bone marrow failure after SCI SCI tonic supraspinal control over sympathetic nervous system lost normal stimuli trigger exaggerated sympathetic reflexes aberrant sympathetic reflexes triggered SCI below caudal-most sympathetic preganglionic neurons lower thoracic lumbar spinal cord femoral bone marrow innervated by sympathetic sensory nerves T8-9 sacral spinal cord7,66,67 sensory input activate propriospinal relay neurons multi-segmental sympathetic reflexes13relay neurons activated integrity intersegmental circuitry bone-specific sympathetic reflexes exaggerated or silenced after SCI bone marrow perceived stromal cells HSPCs as break homeostasis pathological conditions sympathetic tone disrupted CXCL12 CXCR4 levels increase HSPC mobilization modulation spinal sympathetic reflexes restored after SCI intersegmental neuronal circuitry plasticity neurogenic control bone marrow function restored after SCI permanent bone marrow failure data Fig. 7 hematopoietic stimuli (endotoxin) mobilize HSPCs mature lymphocytes from bone marrow-injury-induced changes circulating hormones glucocorticoids acute chronic bone marrow failure glucocorticoids act maintain HSPCs bone marrow promote HSPC homing bone marrow CXCR4 transcription acute stress spinal trauma sustained activation aberrant sympathetic–neuroendocrine reflexes cause hypercortisolism after circulating levels glucocorticoids increase 24 h post-injury remain supraphysiological indefinitely after SCIaberrant sympathetic–neuroendocrine reflexes SCI causes HSPC sequestration regardless injury level Fig. 4) Future studies needed explore SCI affects neural–humoral signaling bone marrow glucocorticoids blood-borne factors microbial metabolites contribute acquired bone marrow failure observation possible to overcome SCI-induced bone marrow failure FDA-approved drug CXCR4 antagonist Plerixafor injected post-injury liberates HSPCs mature leukocytes from bone marrow SCI mice (Fig. 6) Plerixafor mobilize cells bone marrow restore immune can reverse immunodeficiency in WHIM neuroinflammatory injury autoimmunity more research needed conditions for Plerixafor treatment after SCI SCI-induced bone marrow failure revealed competitive repopulating unit) assays. long-term clonogenic potential of HSPCs impaired imprinted HSPCs early after injury bone marrow cells from SCI donor mice 3-day post-injury transplanted irradiated mice engrafted irradiated bone marrow faster than sham-injured miceearly advantage by improved homing to bone marrow by HSPCs from SCI donors glucocorticoid enhancement CXCR4 LSK cells repopulation advantage transient deficits in long-term repopulation clonogenic potential SCI bone marrow after secondary transplantation why SCI impairs long-term function HSPCs within 3 days increase in DNA damage marker γH2AX after chronic SCI unlikely excessive HSPC proliferation after causes DNA HSPC exhaustion possible cell cycle number with long-term HSPC aberrant sympathetic–neuroendocrine reflexes influence clonogenic potential HSPCs catecholamines glucocorticoids cause epigenetic modifications new functional identities69 exposure to stimuli enhanced microbicidal functions against trained immunity orchestrated by epigenetic reprogramming in SCI form negative trained immunity data clonogenic potential of hematopoietic stromal cells from SCI patient bone marrow impaired21data reveal SCI-induced bone marrow failure caused by cell-intrinsic extrinsic mechanisms impaired control HSPC proliferation sequestration likely cell-extrinsic deficits in HSPC clonogenic potential cell-intrinsic changes failure develops after injury-lasting adverse effects on host HSPCs impaired bone marrow function effective immune system infectious morbidity mortality in patients use bone marrow cells transplantation similar limitation for bone marrow from mice with CNS autoimmune data highlight bone marrow underappreciated therapeutic target for improving health outcomes quality of life after SCI Institutional Animal Care Use Committee Ohio State University approved animal protocols experiments performed with guidelines regulations Ohio State University Guide for Care of Laboratory Animals National Institutes of Health Female C57BL/6 mice purchased from Jackson Laboratory female luciferase mice from Charles River Laboratories female B6.SJL-Ptprca Pepcb/BoyJ mice bred in-house from adult pairs Jackson Laboratory(NSG bred in-house from breeding pairs Jackson Laboratory fed commercial food pellets chlorinated reverse osmosis water housed (≤5/cage in ventilated microisolator cages corn cob bedding 12-h light–dark cycle constant temperature (20 ± 2 °C humidity (50 ± pathogen-free housing facility routine testing for pathogens Generation NSG immune systems Newborn NSG pups (24–72 received 1 Gy X-ray irradiation 1–5 × 104 human umbilical CD34+ stem cells Stemcell Technologies intrahepatic injection temperature maintained 37 °C heating pad until pups returned for normal maturation weaning 21–24 days postnatal C57BL/6 mice (10–16 weeks old hNSG mice (16–24 used for SCI experiments subjected complete spinal transection injury third ninth sixth lumbar levels laminectomy controls naïve mice reference anesthetized with ketamine (120 mg xylazine (10 mg prophylactic antibiotic treatment gentamicin sulfate (5 mgAseptic conditions maintained surgical procedures mice on warming pad temperature Hair shaved thoracic spinal cord skin treated betadine 70% ethanol small midline incision vertebra partial laminectomy performed meninges cut spinal cord transected sterile glass aspiration tube for suction fluid/blood stabilization completion injury Muscle injury site sutured wound sutures staples After surgery mice placed in cages on heating pads monitored until given fluids (1–2 mL 0.9% sterile saline softened food Bladders expressed two times daily function urine pH testing bladder infections Gentocin antibiotic administered daily 5 mg kg−1 5 dpi contusion studies identical surgical procedures exception laminectomy Infinite Horizon Impactor moderate 70 kdyne injury third or ninth thoracic spinal levels.IVIS imagingMito mice injected with 80 mg kg−1 D-luciferin anesthetized with isoflurane (2.5–4% oxygen placed on heated surface IVIS Lumina II system for image acquisitionImaging bones limbs forelimbs 90° hindlimbs 45° translucent medical tape Mice kept on 1.5–2% isoflurane Baseline levels mitosis determined SCI imaged at 1 3 7 14 21, 28 dpi Data analyzed Living Image® software Bioluminescence measured regions interest sternum left right femur Total flux maximum radiance ROI determined average calculated Data plotted raw values-SCI or pre-LPS.Plerixafor treatmentAfter SCI surgery mice injected 5 mg kg−1 Plerixafor 0.9% sterile saline First dose 1-h post-injury once day until 3 dpi Mice terminally anesthetized with ketamine xylazine 1 h after tissues collected hematopoiesis systemic endotoxinMito transgenic wild-type mice underwent SCI surgery 6–8 weeks after injury injected 1 mg kg−1 LPS (E. O55:B5 0.9% sterile saline once day 3 daysMito-luc mice IVIS imaging 2 4 7 9 11 14 days post-LPS Wild-type mice submandibular bleeds 4 11 days Blood collected EDTA-coated capillary tube (Sarstedt Thermo Fisher Scientific Waltham collection processingMice anesthetized ketamine xylazine euthanasia tissue collection Blood collected cardiac puncture placed tubes EDTA Blood treated ammonium chloride red blood cell) lysis buffer resuspended Iscove’s Modified Dulbecco’s Medium 2% fetal bovine serum assays 0.1 M buffer saline 2% FBS cytometry Spleens isolated weighed placed Hank’s Balanced Salt Solution minced scissors smashed 40-μm filter rinsed 10 mL HBSS IMDM Mouse femurs tibiae removed cleaned placed HBSS Bone marrow cells isolated flushing bones 10 mL HBSS washed Cell counts obtained standard hemocytometer Hemavet 950 fs multi-species hematology white blood cell differentialImmunolabeling flow 106 bone marrow cells splenocytes 50 μL RBC-lysed blood cytometry analysis antibodies 1:100 dilution staining BD StemflowTM Mouse Hematopoietic Stem Cell Isolation Kit lineage− c-Kit+ Sca-1+ HSPCs Mouse antibody lineage cocktail APC-conjugated antibodies CD3 CD11b CD45R/B220 TER-119 Ly6G/C Fc receptors blocked 15 min anti CD16/32 antibody labeling antibodies 60 min Dead cells labeled eFluor780 30 min incubation cells fixed permeabilized BD Cytofix/CytopermTM solution 20 min cell cycle analysis DNA labeled DAPI buffer 0.1% Triton X-100 20 min labeling human HSPCs antibodies γH2AX measure replication stress c-Kit+ HSPCs CXCR4 receptor expression mature bone marrow cells depleted Lineage Cell Depletion Kit MACS system Biotec cell surface staining LSK markers fixation permeabilization BD Perm/Wash staining CXCR4incubations at 4 °C wash step excess flow buffer centrifugation 5 min 4–10 °C Antibodies for mouse lineage #560492), human lineage #22-7778 human CD34 (581 CD38 (HIT2) used identify human HSPCs (Fig. 4d Antibodies for CD3 (17A2 CD4 CD24 (M1/69 CD43 (S7 CD45/B220 (RA3-6B2 IgM (II/41 IgD (11-26c.2a) used for analysis bone marrow B T cells (Fig. 5 1) Lineage cocktail (BioLegend #133307) antibodies for CD117 (2B8 Sca-1 (D7 CD48 CD150 CD135 (A2F10 CD16/32 (93 determine LT-HSC/MPP1 ST-HSC MPP2 MPP3 MPP4 GMP CMP/MEP subsets of HSPCs after LSK gating (Fig. 2) LSR II LSR Fortessa flow cytometers (BD used analyze samplesForward side scatter parameters cell populations phenotypic analysis Positive negative populations selected staining isotype control antibodies fluorescent minus control Offline data analysis FlowJo v.10 software Star Ashland CFC M3434 Technologies Vancouver methocellulose media cytokines growth factors cell colonies Bone marrow cells plated 7.5 × 103 splenocytes 2 × 105 RBC-lysed blood 150 μL 1 mL MethoCult media Samples plated meniscus-free six-well SmartDishTM incubator 37 °C 5% CO2. 10–12 days after colonies quantified inverted light microscopy StemGridTM underlay Colonies identified blast forming unit-erythrocyte granulocyte monocyte composition MethoCultTM H4034 Optimum human CFCs mice repopulation potential whole-bone marrow cells naïve lam SCI mice marrow cells (5 106 donor C57BL/6 mice leukocyte antigen CD45isolated mixed rescue bone marrow BoyJ mice antigen CD45.1. mixture injected vein irradiated BoyJ mice (4–6 weeks old 9 Gy doses 4.5 Gy 24 h gamma irradiation Cesium-137). Donor chimerism assessed 4 weeks CD45.1 CD45.2 blood leukocytes Bone marrow spleen blood collected 19 weeks after engraftment chimerism blood phenotyping CD4 CD8 CD45/B220 CD11b antibodies Secondary engraftment CD45.2 bone marrow cells bead antibodies cells mixed with isolated CD45.1 BoyJ bone marrow engrafted BoyJ recipients-qPCR cytokine chemokine mRNABone marrow isolated tibia femurs laminectomy T3Tx mice (n = 3 isolated TRIzol Genomic DNA eliminated 1 μg DNase I DNase-treated RNA primed hexamers reverse transcribed into cDNA SuperScript II RNase-free water 1 μg cDNA 1:10 loaded 96-well plate SYBR Green Master Mix detect amplified cDNAMelting point curves reaction Samples run triplicates gene expression normalized to s18 control samples delta CT SCI data increase sham-injured values CXCL12 bone marrow extracellular isolated two tibia femurs per mouse 1 mL ice cold PBS trituration cells debris separated fluid centrifugation 5 min 400 × g Fluid aliquoted frozen in liquid nitrogen Mouse CXCL12/SDF-1α Quantikine ELISA kit used Bone fluid measured undiluted Plates read at 450 nm wavelength SpectraMAX190 analyzed SoftMax Pro software (Molecular Devices San Jose data represented mean ± error independent replicates Group sizes determined preliminary data n = 4 sufficient 1.25-fold change variation 20% >80% power flow cytometry MethoCult assays morbidity/mortality n = 6/group reference data from naïve C57BL/6J (10–16 weeks collected included dotted lines graphs Data excluded statistical outlier deviation standard recovery SCI post-operative single female T3 transection animal removed3 4 bladder infection blood CFCs > 800 mL−1 SCI (T3Tx) sham used two-sample t-tests Welch’s t-test unequal experiment sham low high-thoracic injuries (T3Tx vs T9Tx Fig. 4a one-way ANOVA Tukey post hoc comparisons longitudinal measures proliferation BM engraftment Figs 1 7 8) mixed effect two-way ANOVA Bonferroni comparisons statistical tests two-tailed Exact p values *p < 0.05 (Fig. 5b). Samples blinded attempts replication successful BM. 1a–c HSPC 1d proliferation data two replications Accumulation HSPCs Fig. 1e–h verified Fig. 2. Accumulation cells in tibia (Fig. 1h confirmed replication Impaired HSPC mobilization after SCI female wild-type mice. 3) SCI models (Fig. 4) confirmed human HSPCs (Fig. 4c d). impaired HSPC mobilization SCI confirmed six replications lymphopenia after T3 SCI. 5a two replications enhanced CXCL12-CXCR4 levels (Fig. 6b c two replicationsPlerixafor liberates HSPCs bone marrow after SCI (Fig. 6d–k verified two experiments analyzed GraphPad Prism software v5.0 San Diego Illustrations created BioRender software Figures generated Adobe Photoshop CS5 v12 San Jose CA).Reporting Nature Research Reporting Summary.Supplementary information Peer Review File Reporting Summary | 47.6 | 0.906174 |
10.1038/s41467-020-17648-w | PMC7403315 | Inhibiting thrombosis without inducing bleeding is a major challenge for anticoagulant agents. Here the authors describe a synthetic FXIIa inhibitor able to efficiently prevent thrombosis in mice and suppress coagulation in artificial lungs in rabbits without increasing the risk of bleeding. | Inhibiting thrombosis without generating bleeding risks is a major challenge in medicine. A promising solution may be the inhibition of coagulation factor XII (FXII), because its knock-out or inhibition in animals reduced thrombosis without causing abnormal bleeding. Herein, we have engineered a macrocyclic peptide inhibitor of activated FXII (FXIIa) with sub-nanomolar activity (Ki = 370 ± 40 pM) and a high stability (t1/2 > 5 days in plasma), allowing for the preclinical evaluation of a first synthetic FXIIa inhibitor. This 1899 Da molecule, termed FXII900, efficiently blocks FXIIa in mice, rabbits, and pigs. We found that it reduces ferric-chloride-induced experimental thrombosis in mice and suppresses blood coagulation in an extracorporeal membrane oxygenation (ECMO) setting in rabbits, all without increasing the bleeding risk. This shows that FXIIa activity is controllable in vivo with a synthetic inhibitor, and that the inhibitor FXII900 is a promising candidate for safe thromboprotection in acute medical conditions. | IntroductionCoagulation stops the bleeding at sites of vessel wall injuries, but excess coagulation is equally dangerous as it can lead to thrombosis. Blood vessel blockages can occur in the arterial and venous circulation and cause myocardial infarction, ischemic stroke, and pulmonary embolism, which collectively are the leading causes of disability and death in the industrialized world1. In all of the anticoagulants widely used for the acute and prophylactic treatment of thrombosis2,3, bleeding is a common side effect, and the initiation of any therapy to treat thrombotic disorders must always weigh these risks and benefits, considering the severity of the potential side effects. Thus, there is a growing need for the development of effective anticoagulants that ideally would not impair hemostatic ability4.A promising novel perspective for developing safe anticoagulants with reduced or no bleeding risks is the inhibition of coagulation factors XII (FXII) and XI (FXI), both proteases of the intrinsic coagulation pathway5,6. FXII is the initiating protease of the procoagulant and proinflammatory contact system, and it drives both the intrinsic pathway of coagulation and the kallikrein-kinin system7. Mice lacking FXII display reduced thrombosis in mouse models of injury-induced arterial and venous thrombosis8,9 and are protected from pathological thrombosis in cerebral ischemia10. At the same time, humans naturally lacking FXII and FXII-knockout mice have a normal hemostatic capacity and do not bleed abnormally8,11, which supports the idea that drugs targeting the protease could potentially display antithrombotic effects without significantly compromising hemostasis. Concordantly, the reduction of FXII expression by antisense oligonucleotides suppresses thrombosis in arterial and venous thrombosis mouse models and catheter thrombosis in rabbits12,13. The inhibition of FXII by protein-based inhibitors, such as antibodies14,15 or insect- and plant-derived proteins16–18, also reduces thrombosis in mouse, rat, rabbit, and primate models of induced arterial or venous thrombosis and shows potential avenues for therapeutic anticoagulation.FXII-driven blood coagulation is a major challenge in cardiopulmonary bypass (CPB) surgeries in which a heart-lung machine temporarily supports the circulation. Contact between FXII and artificial surfaces, such as the oxygenator membrane or tubing, induces a conformational change that leads to the proteolytic activation of the FXII zymogen that turns on the contact system. Activation of the procoagulant intrinsic coagulation pathway and the proinflammatory kallikrein-kinin system leads to blood clotting and inflammation19. Contact activation also poses a problem in extracorporeal membrane oxygenation (ECMO), a medical procedure in which an artificial lung system is used for longer periods of time as a life support for patients with severe cardiac and/or pulmonary failure20. The standard strategy for suppressing contact system activation in extracorporeal circuits relies on high doses of heparin, which also inhibit proteases of the extrinsic and common coagulation pathway and thus bear an inherent risk of bleeding21. Several strategies for suppressing contact activation were established in experimental models, but heparin remains the anticoagulation strategy of choice22. In a recent study, Renné and co-workers showed that a human FXIIa-inhibiting antibody, 3F7 (IC50 = 13 nM), prevents clotting and thrombosis in a cardiopulmonary bypass system in rabbits without increasing therapy-associated bleeding14, indicating the usefulness of targeting FXII.FXII is implicated in several other medical conditions23, including hereditary angioedema (HAE)24, reperfusion injury25, Alzheimer’s disease26,27, and multiple sclerosis28, meaning that potential FXIIa inhibitors could be used in a variety of treatments not limited to thrombosis prevention. In Type III HAE, for example, a mutation in FXII facilitates its activation, leading to excessive bradykinin release that causes edema29. An improved variant of the above-mentioned FXIIa inhibitory antibody 3F7 has entered a phase I clinical evaluation for the treatment of HAE30.While high-affinity protein-based FXIIa inhibitors were successfully generated over the last years31, the development of synthetic, small-molecule inhibitors has been more challenging. Small molecules have a number of strengths that make them attractive for drug development, including a uniform composition, efficient tissue penetration, high stability, low immunoreactivity, and ease of production by chemical synthesis. The best small molecule FXIIa inhibitors reported to date are the coumarin derivative 44 (IC50 = 4.4 μM)32 and H-D-Pro-Phe-Arg-chloromethylketone (PCK; IC50 = 0.18 μM)10. While they have proven to be useful as research compounds in FXIIa inhibition studies, their covalent inhibition mechanism and their moderate potency and selectivity limit their drug development potential.We recently identified high-affinity FXIIa inhibitors based on cyclic peptides using phage display, including the bicyclic peptide FXII618 (Ki = 8.1 nM)33. In previous work, the substitution of individual amino acids in FXII618 to unnatural ones further improved the affinity34,35, though due to their limited binding to animal FXIIa homologs and low proteolytic stability in plasma, these particular inhibitors could not be evaluated in vivo. For this reason, we describe herein the further improvement of the inhibitor to achieve sub-nanomolar potency towards human, mouse, and rabbit FXIIa, and we enhanced the stability in plasma to several days (half-life in human plasma ex vivo). These properties allowed for the pre-clinical evaluation of the inhibitor in two animal models. This work shows that synthetic FXIIa inhibitors can be developed to efficiently prevent thrombosis in mice and suppress coagulation in artificial lungs in rabbits without increasing the risk of bleeding.ResultsEngineered macrocycle inhibits FXIIa with sub-nanomolar affinityIn a previously developed FXII inhibitor, FXII618, we showed that the N-terminal arginine (Arg1) is rapidly cleaved by plasma proteases (t1/2 around 4 h in human plasma), causing a 40-fold reduction in the Ki of the inhibitor34. When Arg1 was substituted with diverse unnatural amino acids, the stability could be improved at the price of a weaker inhibition, wherein amino acids with positively charged side chains showed the smallest affinity losses34. In a new attempt, we replaced Arg1 with a panel of di-peptides that could potentially reach a larger surface area on FXIIa to form productive interactions (Fig. 1a, Supplementary Table 1). Both amino acids in the di-peptides were D-amino acids to prevent proteolysis, and one of the two was D-Arg to maintain a positive charge. To avoid extensive purifications for the initial screening tests, we developed a synthesis and screening approach based on coumarin-labeled peptides to compare the crude activities, as described in detail in the Methods. A stability assay showed that the di-peptides based on D-amino acids were resistant to proteolysis, prolonging the plasma half-life around five-fold to 20 h (Supplementary Fig. 1), and several di-peptides improved the Ki values around two-fold over FXII618 (Fig. 1a). The best inhibitor, Arg1→D-Arg-D-Ser (FXII850) was synthesized without the coumarin label and showed a Ki of 5 nM, which was a 1.6-fold improvement compared to FXII618 (Ki = 8.1 ± 0.7 nM; Fig. 1b and c, Supplementary Fig. 2). We also tested a panel of diverse amino acid substitutions for Arg8, a position in FXII618 that had not been systematically optimized, using the same screening strategy based on coumarin-labeled peptides (Fig. 1a, Supplementary Table 1). The substitution Arg8→His (FXII851) did not affect the Ki of human FXIIa though had a clear effect on mouse FXIIa inhibition, with Ki values of 6.5 nM and 36 nM, for human and mouse respectively, which corresponded to 1.2- and 2.4-fold improvements over FXII618 (Fig. 1b and c, Supplementary Table 2).Fig. 1Improving affinity and stability of bicyclic peptide FXIIa inhibitor.a Amino acids Arg1 and Arg8 of FXII618 were individually substituted to two D-amino acids or natural amino acids, respectively. The relative Ki values for human and mouse FXIIa are indicated as compared to the lead peptide FXII618. Average values of two measurements are shown for peptides with improved affinity. b Schematic representation of the lead peptide FXII618. Amino acid substitutions that improve the activity and/or stability are indicated. c Inhibition of human FXIIa by FXII618, variants of FXII618 containing a single amino acid substitution, and FXII900 in which four beneficial modifications were combined and Arg13 was deleted. Residual FXIIa activity was determined in two (FXII850, FXII851), three (FXII618, FXII700, FXII800) or five (FXII900) independent measurements. Mean values are indicated. For inhibitors that were tested three or five times, data are presented as mean ± SD. d Specificity profiling of FXII900. Residual activities of human FXIIa and ten homologous human proteases were measured. Inhibition of FXII was determined in five, plasmin and APC in three, and all other proteases in two independent measurements. Means values are indicated. e Proteolytic stability of FXII900 and two precursors. Bicyclic peptide was incubated in human plasma at 37 °C for the indicated time periods, and the remaining inhibitor activity was quantified in a FXIIa activity assay using fluorogenic substrate. Two independent measurements were performed for FXII850 and FXII900, and four for FXII618. Means values are indicated. f Proteolytic stability of FXII618 and FXII900. Bicyclic peptide was incubated in plasma at 37 °C for the indicated time periods and analyzed by LC-MS. Total ion count (TIC) is shown. Source data of a, c–e are provided in a Source data file.We next combined the beneficial amino acid substitutions identified in this work, the Arg1→D-Arg-D-Ser (FXII850) and Arg8→His (FXII851) with the two previously identified substitutions, Arg11→(S)-β3-homoarginine (βhArg; FXII700; Ki = 1.5 ± 0.1 nM)35 and Phe3→4-fluorophenylalanine (Phe4F; FXII800; Ki = 0.84 ± 0.03 nM)34 (Fig. 1b and c, Supplementary Table 2). In addition, we deleted the C-terminal Arg13 that did not contribute to the binding affinity. The resulting inhibitor, FXII900, blocked human and mouse FXIIa with Ki values of 0.37 ± 0.04 nM and 0.45 ± 0.11 nM, respectively, and was thus substantially more potent than any of its precursors (Fig. 1c). This final inhibitor could be efficiently synthesized in gram-scale by solid-phase peptide synthesis of the linear precursor and subsequent macrocyclization of the crude peptide with the 1,3,5-triacryloyl-1,3,5-triazinane linker (TATA). Only a single HPLC purification step was required to obtain > 95% pure product with an isolated yield greater than 50% (Supplementary Fig. 3). We had previously developed a structural model of the precursor peptide FXII618 bound to FXIIa, in which we could determine with high confidence the positions and interactions of the amino acids Phe3 to Pro633. The newly identified beneficial amino acid substitutions lie outside this region and we could thus not use the model to rationalize the molecular basis of the achieved affinity enhancements.Finally, a specificity profiling against a panel of homologous plasma proteases was performed to indicate any off-target interactions that would occur in physiological conditions. Activity assays against this chosen panel in the presence of the inhibitor showed that FXII900 is highly selective (Fig. 1d, Supplementary Table 3). All physiologically relevant proteases displayed 100,000-fold or higher selectivity (Ki values > 40 μM), with only trypsin showing any significant inhibition, which occurred at low micromolar concentrations (Ki = 1.46 μM).FXIIa inhibitor is stable in human plasma for several daysTo measure the stability of the inhibitor in plasma, FXII900 was incubated in human plasma at 37 °C for extended time periods and then the remaining FXIIa inhibitory activity was quantified (Fig. 1e). Mass spectrometric analysis of the various plasma samples showed only FXII900 and no degradation products, though the quantity of the intact inhibitor decreased over time (Fig. 1f, right panel). This suggested that after an initial cleavage event, the inhibitor was rapidly degraded and, therefore, the cleavage products were not detectable. Even so, the half-life of FXII900 was 128 h, around 25-fold longer than that of FXII618. It was also much longer than that of FXII850 carrying only the Arg1→D-Arg-D-Ser modification (t1/2 = 18.5 h; Fig. 1e), indicating that more than one of the introduced mutations contributed to the improved stability. To identify which of the modifications was most important for stability, we individually reverted the amino acid substitutions (Supplementary Fig. 4 and Supplementary Table 4) and retested the variants, revealing that the Arg11→βhArg substitution contributed the most to the stability improvement (Supplementary Fig. 5).FXII900 efficiently inhibits the intrinsic coagulation pathway ex vivoTo assess the ability of FXII900 to block FXII-driven blood plasma coagulation and to further evaluate its selectivity, we performed coagulation tests to measure activated partial thromboplastin time (aPTT) and prothrombin time (PT). aPTT and PT measure the time until coagulation upon initiation of the intrinsic and extrinsic pathways, respectively, meaning that selective FXIIa inhibition would prolong aPTT but not PT, and results are reported in terms of the concentration of the tested compound required for a 1.5x increase in coagulation time (EC1.5x). FXII900 prolonged aPTT in human plasma with an EC1.5x of 0.79 μM, thus 4.3-fold better than its precursor FXII618 (EC1.5x of 3.4 μM; Fig. 2a). FXII900 did not affect PT, even at the highest concentration tested (120 μM), confirming the selectivity for the FXIIa target (Fig. 2b). The inhibitor also prolonged the aPTT in plasma of all other species, namely mouse (EC1.5x = 2.9 μM), rabbit (EC1.5x = 0.102 μM), and pig (EC1.5x = 12.2 μM), wherein the potency varied significantly (Fig. 2a). For rabbit plasma, we observed a strong prolongation of aPTT at low inhibitor concentrations and a sharp transition to a maximal effect. This pattern was previously seen for antibody-based FXIIa inhibitors tested in rabbit plasma14, and we thus expected that this was a rabbit plasma-specific phenomenon and not based on a particularly strong affinity for rabbit FXIIa. The inhibitor did not prolong PT in any of the three animal plasma tested (Fig. 2b).Fig. 2Inhibiting FXIIa of different species.a Inhibition of the intrinsic coagulation pathway in human, mouse, rabbit, and pig plasma ex vivo. The aPTT in the presence of different concentrations of FXII900 is shown. For human and mouse plasma, the aPTT of FXII618 was measured for comparison. All coagulation times were determined in two, four or six independent measurements. Mean values are indicated. Data of individual measurements are shown as dots. b PT in the presence of different concentrations of FXII900 and FXII618 in human plasma (upper panel), and FXII900 in mouse, rabbit, and pig plasma (lower panel). PT was determined in two (human plasma) or three independent measurements (all animal plasmas). Mean values are indicated. Data of individual measurements are shown as dots. For the animal plasmas, means ± SD is indicated. c Inhibition of recombinant FXIIa (rFXIIa) from different species by FXII618 and FXII900. The inhibition of human FXIIa derived from blood is shown for comparison (data from Fig. 1c). Residual FXIIa activities were determined in two (human, mouse) or three (pig, rabbit) independent measurements. Mean values are indicated. For pig and rabbit FXIIa inhibition, the data are presented as mean ± SD. Source data of a–c are provided as a Source data file.To assess if FXII900 prolonged aPTT in the plasma of the four species to different extents due to varying Kis for the various FXII orthologues, we cloned and expressed rabbit and pig FXII because they were not commercially available (Supplementary Figs. 6 and 7) and tested their inhibition by FXII618 and FXII900 (Fig. 2c). These experiments, described in detail in the Supplementary Information, showed that the FXII orthologues were inhibited with different strengths and that the extent of aPTT prolongation correlated with FXII inhibition (Ki).Pharmacokinetics in mice, rabbits, and pigsWe assessed the pharmacokinetic properties in three species, mice, rabbits, and pigs, because suitable models for thrombotic diseases are available for these animals. In mice, we determined the pharmacokinetics following subcutaneous administration (5 mg kg−1, n = 3, different mice for each time point). FXII900 reached a plasma concentration of around 1 μM after 3 min and remained above this concentration for 30 min as determined by LC-MS (Fig. 3a, upper panel). The aPTT was prolonged as expected based on the plasma concentrations; namely, it was extended by two-fold at 15 min and remained prolonged by over 1.5-fold until 30 min (Fig. 3a, lower panel).Fig. 3Pharmacokinetics of FXII900 in mouse, rabbit, and pig.a Pharmacokinetics in mouse after subcutaneous administration (5 mg kg−1, n = 3). Concentration of FXII900 in plasma (upper panel) and aPTT (lower panel) are indicated. Means ± SD are indicated. b Pharmacokinetics in rabbit after intravenous administration (3.7 mg kg−1, n = 3). Concentration of FXII900 in plasma (upper panel) and aPTT (lower panel) are indicated. Plasma samples were analyzed in duplicate for rabbit 1 and in triplicate for rabbits 2 and 3. Mean values are indicated. For rabbit 2 and 3, data are presented as mean ± SD. c Pharmacokinetics in rabbit after subcutaneous administration (3.7 mg kg−1, n = 4). Concentration of FXII900 in plasma (upper panel) and aPTT (lower panel) was measured. Plasma samples were analyzed in duplicate. Mean values are indicated for each rabbit. d Pharmacokinetics in pig after intravenous administration (4 mg kg−1, n = 3). Concentrations of FXII900 in plasma were determined in duplicate, and mean values are indicated for each pig. Source data of a–d are provided as a Source data file.In rabbits, we determined the pharmacokinetic properties for intravenous and subcutaneous administration. Upon intravenous administration (n = 3, 3.7 mg kg−1), FXII900 showed an elimination half-life of 12 ± 2 min (Fig. 3b, upper panel; pharmacokinetic parameters provided in Supplementary Table 5). The aPTT was initially more than eight-fold prolonged and remained more than three-fold prolonged for the entire time monitored (40 min; Fig. 3b, lower panel). When applying the same dose subcutaneously (n = 4, 3.7 mg kg−1), the peptide remained in a narrower concentration range, staying above 100 nM and below 300 nM for between 10 and 80 min after administration (Fig. 3c, upper panel). The aPTT with the subcutaneous administration was prolonged by more than 2-fold for 40 min (Fig. 3c, lower panel).We further determined the pharmacokinetics of FXII900 in pigs, as this species offers models for indications in which FXII plays a role, such as ischemic reperfusion injury25,36. After intravenous administration (n = 3, 4 mg kg−1), the inhibitor reached a maximal concentration of around 7 μM and was cleared with a t1/2 of 36 ± 5 min (Fig. 3d, upper panel). The clearance rate and volume of distribution were 11.8 ± 1.8 ml kg−1 min−1 and 610 ± 140 ml kg−1 (Supplementary Table 6). In all the studies performed with the three species, we did not observe any signs of toxicity or other adverse effects.FXII900 inhibits ferric chloride-induced thrombosis in miceWe next evaluated the thromboprotective properties of FXII900 in a ferric chloride (FeCl3)-induced thrombosis mouse model. When applied directly onto the blood vessels, FeCl3 induces thrombosis by aggregating the red blood cells, which in turn activates platelets at the application site37–39. To test the ability of our compound to protect against thrombosis in this model, FXII900 or an inactive control peptide FXII901 were administered subcutaneously to two groups each containing 10 mice (5 mg kg−1). The negative control FXII901 is a variant of FXII900 with three modifications (Phe4F3→Phe, Arg4→Ala, βhArg11→Arg) that reduce its inhibitory constant 100,000-fold for FXIIa (Ki = 39 ± 14 μM; Supplementary Fig. 8). A solution of 7.5% FeCl3 was applied to the mesenteric arterioles and the blood flow was monitored by intravital microscopy for 25 min. Mice that were injured by FeCl3 could be clearly identified due to a speckled pattern that supposedly represented an accumulation of platelets (Fig. 4a)40. Seventeen out of the 20 mice showed this pattern (nine treated, eight control) and were taken for further analysis (Supplementary Figs. 9 and 10). The arterioles of the mice receiving the inactive control peptide FXII901 rapidly formed clots, and in a majority of the animals, the vessels were completely occluded after around 10 min, as exemplified in the lower micrographs in Fig. 4a and shown for all mice in Supplementary Fig. 9. The time point and extent of clot formation observed for the control peptide were as expected based on previous studies in which vessels of non-treated mice were exposed to 7.5% FeCl341. In contrast, mice that were treated with the inhibitor FXII900 showed only the characteristic speckled pattern with the occasional formation of a blood clot (Fig. 4a, upper micrographs; Supplementary Fig. 10). In this group, no complete occlusion was observed. We quantified the anticoagulant effect of FXII900 by comparing the rates for both clot formation (a clearly visible blood clot; Fig. 4b, left panel) and blood vessel occlusion (a blood clot with a diameter of the blood vessel; Fig. 4b, right panel). Of the eight mice receiving the control peptide, seven showed blood clotting and five showed full-vessel occlusion. In contrast, in the FXII900-treated mice, only three showed blood clotting, and in no mice were the vessels fully occluded (p for chi-squared test = 0.02 and 0.005, respectively). In addition, the few mice in the treated group that showed signs of coagulation developed these blood clots on average around 10 min later than the control group (p for chi-squared test = 0.006; Supplementary Table 7).Fig. 4Thromboprotection of FXII900 in mice and effect on bleeding.a Intravital fluorescence microscopy images showing mesenteric arterioles in which thrombosis was induced by topical application of FeCl3 (7.5%, 1 min). Platelets were fluorescently labeled with Rhodamine 6G for visualization. Representative images are shown for two mice, one treated with FXII900 (upper panels) and one treated with inactive control peptide FXII901 (lower panels) 15 min before the application of ferric chloride (5 mg kg−1, subcutaneous injection). Vessel walls at the FeCl3 application site are indicated with yellow markers. Three distinct morphological changes, (1) a characteristic speckled pattern, (2) clot formation, (3) and vessel occlusion, are indicated. Scale bar: 200 μm. b The percentage of mice showing either clot formation or full occlusion at different time points after ferric chloride application is indicated over time (inhibitor FXII900: 5 mg kg−1, n = 9; 25 mg kg−1, n = 8; neg. control FXII901: 5 mg kg−1, n = 8, 25 mg kg−1, n = 9; heparin: n = 10). Clot formation was defined as the appearance of an aggregate with a diameter larger than 10 μm. Full occlusion of the blood vessel was defined as a blood clot having the same diameter as the blood vessel. c Blood loss and bleeding time of mice with 2 mm tail transections. Mice were treated 5 min before clipping the tail tips with the vehicle (PBS, IV, n = 6), heparin (200 IU kg−1, IV, n = 10), FXII900 (25 mg kg−1, SC, n = 6) or FXII901 (25 mg kg−1, SC, n = 6). Mean values and standard deviations are indicated. The significance for a prolonged bleeding time was assessed with an unpaired, one-tailed t-test. The p value is indicated for significant results and n.s. stands for not significant. Mice treated with heparin showed either a small or medium-to-large loss of blood, and a mean value is thus not indicated.Repeating the experiment with a 5-fold higher dose of FXII900 (n = 8, 25 mg kg−1) and the negative control FXII901 (n = 9, 25 mg kg−1) gave comparable results to the lower dose, with no further delay in coagulation or reduction in the rate of clot formation (Fig. 4b; intravital microscopy for 20 min; by mistake 5 min shorter than in the experiment with lower dose). In this repeat study, a positive control using the current gold-standard drug heparin was added for comparison (n = 10, 200 IU kg−1). In this control group, more mice showed blood clotting (six) and vessel occlusion (one) than in the two groups treated with FXII900 (Fig. 4b).Mice treated with FXII900 have a normal hemostatic capacityWe assessed the bleeding propensity of mice treated with the higher of the two doses of FXII900 used in the above ferric chloride-induced thrombosis study (25 mg kg−1, SC) using a tail transection model (tail transection at 2 mm from the tail end, diameter > 1 mm; n = 6). As positive and negative controls, we treated mice with PBS (n = 6) and heparin (200 IU kg−1, IV, n = 6). We placed the tails into a tube with warmed PBS and measured the time of bleeding and the volume of blood lost. Mice injected with heparin bled essentially continuously throughout the monitored 30 min period, which is similar to previous reports41 and showed either a small (3 mice) or large loss of blood (3 mice). Given the heterogeneous result for the blood loss, we performed the heparin control with four additional mice, which showed medium to high blood loss. Mice treated with the FXIIa inhibitor showed an average bleeding time of 10 min and blood loss of 21 μl, which was comparable to the PBS-treated mice (average bleeding time: 7 min, average blood loss: 34 μl) and the FXII901 control (Fig. 4c, Supplementary Fig. 11), which indicated that the FXIIa inhibitor does not affect the hemostatic capacity.FXII900 provides bleeding-free anticoagulation in artificial lungsTo test the clinical potential of FXII900 for inhibiting FXII-driven coagulation in ECMO, we tested the inhibitor in an artificial lung rabbit model. New Zealand White rabbits were anesthetized, either treated intravenously with FXII900 (n = 4, 2 mg kg−1 bolus and 0.075 mg kg−1 min−1 infusion for four hours), left untreated (n = 3), or treated with heparin for comparison (n = 2, infusion of 60 IU h−1 and rate adjustments to maintain ACT within a clinical range of 220–300 s), and were mounted for a 4-h period to a veno-venous ECMO configuration using a polymethylpentene (PMP) fiber artificial lung system.Treatment with FXII900 prolonged the coagulation parameters aPTT and activated clotting time (ACT) around 10-fold during the entire course of the experiment. This was to a much larger extent than those for heparin, which were prolonged 1.5 and 2-fold, the range recommended for heparin anticoagulation (Fig. 5a and b). For most aPTT measurements using FXII900, no coagulation was observed at 240 s, the maximal aPTT value measured with this device (Supplementary Fig. 12a). In untreated rabbits, the coagulation times remained at baseline values of around 20 s (aPTT) and 170 s (ACT) throughout the experiment. Because increased pressure at the inlet of the artificial lung indicates clogging that can be caused by blood coagulation, we measured the resistance of the artificial lung, indicated by the difference in pressure at the inlet and outlet of the lung divided by the flow. In all rabbits treated with FXII900, the resistance remained at the baseline of 50 mmHg L−1 min−1 during the entire experiment, except for one rabbit in which the resistance started to increase strongly after two hours at baseline (Fig. 5c and Supplementary Fig. 12c). This was in stark contrast to the untreated animals in which the resistance increased in the first hour and doubled for all rabbits from 50 mmHg L−1 min−1 to 100 mmHg L−1 min−1 during the course of the experiment.Fig. 5Bleeding-free anticoagulation in an artificial lung model in rabbits.Rabbits were connected veno-venous to an artificial lung system for four hours. FXII900 was injected as a bolus (2 mg kg−1) before the start of the extracorporeal circulation and as a constant infusion (0.075 mg kg−1 min−1) over the full time course of the experiment (n = 4; data shown in blue). Rabbits in the control group were untreated (n = 3, data shown in red). Rabbits in the heparin group (n = 2) were initially infused with one IU per min heparin and the drip rate was adjusted to maintain ACT values between 220 and 300 s. Values indicated for the time point 0 refer to measurements made before connecting the artificial lung. Mean ± SD is indicated for the four rabbits in the FXII900-treated group and the three rabbits in the control group. Mean values are indicated for the two rabbits in the heparin group. Data from individual rabbits is shown in Supplementary Fig. 12. a aPTT. Coagulation times of 240 s indicate that plasma did not coagulate at this time. b ACT. Clotting times of 1500 s indicate that blood did not coagulate at this time. c Resistance calculated based on pressure at the inlet and outlet of the device and the flow rate. d Bleeding time measured by incision-provoked injuries at the ears. e Platelet count. f Volume of blood clots indicated in % of volume of the artificial lungs.Because excessive bleeding is a major concern with standard anticoagulation in ECMO systems, we analyzed if FXII900 affected the hematological parameters of bleeding time and platelet count. Rabbits treated with the inhibitor showed completely normal bleeding times throughout the experiment (4–5 min; Fig. 5d and Supplementary Fig. 12d). In contrast, the group treated with heparin at a clinically relevant dose showed 2-fold prolonged bleeding times (5–10 min) at all the time points measured (Fig. 5d and Supplementary Fig. 12d). The platelet count was essentially the same in the three groups, decreasing around two-fold from an initial level of 2 × 108 cells per ml (before connecting the lung) and remaining constant over the entire course of the experiment (Fig. 5e and Supplementary Fig. 12e). Analysis of the artificial lungs at the end of the experiment showed that the inhibitor- and heparin-treated rabbits had a reduced amount of clotted blood. The volume of the clot in relation to the device volume was 10 ± 6% for the FXII900-treated rabbits, which was significantly smaller than that of untreated rabbits (37 ± 10%; p = 0.03), but it was slightly larger than that of the heparin-treated rabbits (5%; Fig. 5f and Supplementary Fig. 12f).DiscussionVarious protein-based inhibitors were developed that showed that suppression of FXIIa activity provides efficient anticoagulation without increasing bleeding risks, though the development of high-affinity small-molecule FXIIa inhibitors has been lagging. Given the attraction of small, synthetic molecules as therapeutics, in this work, we improved the binding affinity and stability of FXII618, a peptide macrocycle FXIIa inhibitor that was previously developed by phage display. The generated inhibitor FXII900 blocks FXIIa with a Ki of 370 pM, shows 100,000-fold selectivity over other plasma proteases, and is stable in blood plasma.The improvement of the binding affinity of FXII618 by 21-fold and proteolytic stability by 25-fold was achieved by combining random screening and rational design strategies. Two amino acid substitutions that enhance the affinity and/or stability of the inhibitor were identified by screening a panel of singly mutated peptides, and were combined with two previously identified beneficial mutations. Individual reversion of the amino acid substitutions in FXII900 showed that each single one contributes to improving the binding affinity and that the modifications of Arg1 to D-Arg-D-Ser and Arg11 to βhArg are the main ones that improve the stability. The higher binding affinity translates into a prolongation of the aPTT in all species tested, with the EC1.5x improved 4.3-fold in human plasma and 10-fold in mouse plasma. As expected, the EC1.5x values are much higher than the Ki values because FXII needs to be inhibited nearly fully to prolong aPTT 1.5-fold (we estimate > 99%) while the Ki is by definition the concentration at which 50% of FXII is inhibited. The much-enhanced potency in mouse plasma was of interest in view of the planned pharmacologic study in mice. FXII900 did not prolong PT at 120 μM at all, which contrasted with FXII618 that prolonged the PT slightly at this highest concentration tested, indicating that modification of the four amino acids had also improved the selectivity of the inhibitor. The enhancement of affinity, selectivity, and stability achieved in this work is owed mostly to unnatural amino acids that could be incorporated due to the synthetic nature of the inhibitor, which is an advantage over protein-based inhibitors that need to be produced by recombinant expression and do not easily allow the incorporation of unnatural building blocks.FXII900 is composed of nine natural amino acids, four non-natural ones, and a central core structure that connects the side chains of three cysteines. The inhibitor can be conveniently synthesized by automated solid-phase peptide synthesis in large quantities. All applied non-natural amino acids are inexpensive, allowing for production of inhibitor for a moderate cost. Crude linear peptide was cyclized efficiently by the chemical linker TATA and only a single HPLC purification step was required to obtain > 95% pure product with isolated yields greater than 50%. We provide a supplementary figure that describes the exact peptide sequence and the conditions recommended for peptide cyclization, which may be used for ordering the peptide from companies offering custom peptide synthesis (Supplementary Fig. 3).FXII900 is rapidly cleared upon intravenous administration, most likely by renal filtration, as expected based on the small size and the polar structure. The elimination half-life ranges from 12 min in rabbits to 36 min in pigs. The prolongation of the aPTT correlated with the PK data for all species. Based on allometric scaling, the half-life in humans can be expected to be in the range of one hour. No degradation products were detected in plasma samples of all species, suggesting that the inhibitor fully resists proteases in the blood circulation and that it is filtered out as an intact molecule, which is in line with the high stability and long half-life in plasma found ex vivo. Despite the fast clearance, a single-dose injection of FXII900 prolonged the aPTT in mice and rabbits for 30 min. The relatively short circulation time of these small molecules, especially as opposed to protein therapeutics, allows flexible adjustment of the duration for which the body is exposed to the inhibitor in therapeutic applications. The pharmacokinetic profile of FXII900 makes it attractive for acute disease treatment such as CPB, hemodialysis and ECMO, or anticoagulation for short time periods after ischemic stroke or surgery, wherein a patient is hospitalized and could be given infusions of the inhibitor for the desired time period. For hemodialysis applications, the inhibitor would potentially need to be tethered to a large protein or polymer to prevent rapid removal by the filtration unit. For controlling the activity of FXII over several days, weeks or months, FXII900 is not suited and other therapeutic modalities such as antibody-based FXII inhibitors or antisense oligonucleotides are clearly better options. Given its high proteolytic stability, FXII900 could potentially resist to some extent proteases of the stomach and intestines, but due to the multiple charges, it would unlikely be able to cross the gastrointestinal epithelium, preventing oral administration.We evaluated the pharmacologic activity of FXII900 in a ferric chloride-induced thrombosis mouse model. FXII-deficient mice show strong reduction of coagulation in this model, indicating that FXII plays a central role in this process8,9. We found that a bolus injection of FXII900 efficiently protected mice from thrombosis, while bleeding time and blood loss were not affected, as assessed in a tail transection bleeding model. This was in strong contrast to a control experiment with heparin (200 IU kg−1) in which mice were constantly bleeding throughout the 30 min period monitored and lost a large volume of blood. It is important to say, however, that thrombosis in myocardial infarction, ischemic stroke, or pulmonary embolism is more complex than in the applied mouse thrombosis model, as thrombosis can also be caused by triggers that are not dependent on FXII activation. Independent of this, the results obtained show that a synthetic FXIIa inhibitor can efficiently suppress coagulation in cases where it is mediated via FXII activation.Finally, we further tested if FXII900 could suppress the coagulation activated by the membrane of artificial lungs using a rabbit model. Heart-lung machines are used for heart surgery because of the difficulty of operating on the beating heart such that approximately half a million cardiac operations performed per year in the US use CPB. Similar, more compact systems are also used to support patients with cardiac and respiratory dysfunction, with more than 7000 patients treated annually with extracorporeal life support42. The contact of blood with the hydrophobic polymer surfaces found in these machines induces FXII contact activation and thus coagulation. This coagulation is currently suppressed using high doses of heparin, which bears bleeding risks. An analysis of nearly 80,000 ECMO patients showed that hemorrhage is the major adverse event in these situations, with bleeding from surgical sites occurring in 6.3% to 29.3% of patients depending on the patient group42. Moreover, both device thrombosis and patient bleeding lead to significant increases in patient mortality43. We found that continuous infusion of FXII900 efficiently prolonged the coagulation parameters ACT and aPTT over four hours and thus the entire course of the experiment. The inhibitor reduced coagulation in an artificial lung as measured by the resistance of the device, which remained low and indicated reduced clogging by coagulation and a reduction in the total volume of the blood clots in the device, which was more than three-fold lower in treated rabbits. Importantly, all rabbits treated with FXII900 showed normal bleeding times. These findings indicated that a small molecule FXIIa inhibitor is suitable for reducing blood coagulation in CPB. While heparin may still be required to suppress coagulation triggered through routes other than FXII, the application of a small molecule FXIIa inhibitor may allow for a reduced heparin dose, which would reduce the bleeding risks. In addition to its role in thrombosis in ECMO and CPB, FXIIa initiates the inflammatory kallikrein-kinnin system19. Bradykinin plasma concentrations are largely elevated in CPB and ECMO patients, so targeting FXIIa may provide an additional anti-inflammatory benefit to these patients, to reduce or prevent adverse effects, such as organ damage.In conclusion, a synthetic FXIIa inhibitor with sub-nanomolar affinity, high selectivity, and good stability has been developed that allows for efficient anticoagulation in relevant animal disease models without increasing bleeding risks, addressing several concerns surrounding the current gold-standard treatment of heparin. The synthetic nature and the small size allows for efficient production of the compound. With its excellent binding properties and stability as demonstrated herein, the inhibitor may be readily applicable in its current form for acute procedures or conditions associated with an increased risk of thrombosis, such as CPB during heart surgery or ECMO without compromising hemostasis at the wound site.MethodsStudy designThe objectives of this study were to improve the pharmacologically relevant properties of a peptide macrocycle inhibitor of FXIIa, to assess the pharmacokinetic properties of the resulting inhibitor in three species, and to evaluate the inhibitor in two clinically relevant animal models. The inhibitor was engineered by synthesizing variants of the lead peptide and testing their inhibitory activity, specificity, and stability in various assays in vitro. Inhibitors showing improved activities were prepared with greater than 95% purity and their activities were tested at least in duplicate. The affinity of the final peptide FXII900 was determined in quintuplicate from three different synthetic batches. The pharmacokinetic properties in mice, rabbits, and pigs were assessed by quantifying the inhibitor in plasma samples using liquid chromatography and mass spectrometry, and by measuring aPTT in plasma samples in duplicate or triplicate, as indicated. The pharmacologic effects of the inhibitor were tested in a FeCl3-induced thrombosis model in mice and in an artificial lung model in rabbits. In all animal experiments, subjects were randomly assigned to groups.Screening method based on coumarin-labeled peptidesIn order to test the inhibitory activity of peptides without prior purification, we developed the following method. Variants of the lead FXII618 peptide were synthesized with the C-terminal amino acid Fmoc-β-(7-methoxy-coumarin-4-yl)-Ala-OH. The fluorophore 7-methoxy-coumarin allowed precise quantification of the crude peptide. After solid-phase peptide synthesis, cleavage, and ether precipitation, the peptides were chemically cyclized and their Ki values were determined in a FXIIa activity assay using a fluorogenic substrate. Peptides synthesized and characterized with this strategy displayed an around 2.5-fold higher apparent Ki than analogous HPLC-purified peptides without the tag. For example, the reference peptide FXII618 carrying the coumarin amino acid had an around 2.5-fold weaker Ki (20 ± 3 nM) compared to the purified FXII618 without the tag (8.1 ± 0.7 nM). Despite the difference, this method allowed for a comparison between peptides and the identification of the most active variants of FXII618. The best peptides from the screen were synthesized without the coumarin label, purified, and characterized.Plasma stability assaysPeptide (2 μl of 2 mM in H2O) was added to 398 μl of citrated human plasma (Innovative Research) to obtain a final peptide concentration of 10 μM. The mixture was incubated in a water bath at 37 °C. At different time points (0, 0.5, 1, 2, 4, 8, 12, 24, 48, and 96 h), samples of 30 μl were removed, diluted to 200 μl with aqueous buffer (10 mM Tris-HCl, pH 7.4, 150 mM NaCl, 10 mM MgCl2, 1 mM CaCl2), and incubated for 20 min at 65 °C to inactivate plasma proteases. The peptide/plasma samples were stored at −20 °C until the residual inhibitory activity of the peptides was measured in a FXIIa inhibition assay. For the activity assay, the peptide/plasma samples were centrifuged for 5 min at 16,000 × g, serial two-fold dilutions of the supernatant were prepared (peptide concentration ranges from 0.5 nM to 0.5 μM), and the residual activity of 0.5 nM human β-FXIIa was measured using 50 μM Boc-Gln-Gly-Arg-AMC substrate. IC50 values were derived from the fitted curve using an equation described in the Supplementary Methods. Residual inhibition in % was calculated using the equation IC50,0h/IC50,xh*100, wherein IC50,0h is the functional strength of the inhibitor at time point 0 and IC50,xh the functional strength of the inhibitor after one of the different plasma incubation periods mentioned above.Plasma degradation assaysPlasma stability and peptide cleavage sites were assessed by incubating the bicyclic peptides in mouse plasma and analyzing the products with an LC-MS system (LCMS-2020, Shimadzu). Peptide (2 μl of 2 mM in H2O) was added to 48 μl of citrated mouse plasma (Innovative Research) to obtain a final peptide concentration of 80 μM. The mixture was incubated in a water bath at 37 °C. At different time points (0, 15, 24, 72, 96, 120, and 144 h), samples of 5 μl were removed, mixed with 5 μl of 6 M guanidinium hydrochloride, and incubated for 30 min at RT. Plasma proteins were precipitated by incubation with 200 μl of ice cold EtOH and 0.1% (v/v) formic acid for 30 min and centrifuged at 9000 × g for 20 min at 4 °C. The supernatant was evaporated in a Speedvac at 50 °C and reduced pressure. The residue was dissolved in 40 µl of deionized water containing 0.1% (v/v) CHOOH and analyzed by LC-MS. The samples were analyzed using an analytical C18 column (Phenomenex C18 Kinetex column, 50 × 2.1 mm, 2.6 μm, 100 Å) and a linear gradient of 5–35% solvent B (MeCN, 0.05% [v/v] CHOOH) in solvent A (H2O, 0.05% [v/v] CHOOH) in 5.5 min at a flow of 1 ml per min. The masses of the intact peptide and degradation products were measured on a single quadrupole mass spectrometer in positive ion mode using electrospray ionization. Peptides were quantified based on the absolute intensities of the detected mass peaks (M3+ and M4+).Cloning of vectors for expression of FXII in mammalian cellsThe protein sequences for human, rabbit and pig factor XII were taken from the following database entries: human FXII: UniProtKB - P00748; rabbit FXII: NCBI RefSeq - XP_008253687.1; pig FXII: UniProtKB - O97507. The sequences are shown in the Supplementary Methods. DNA encoding the full-length proteins were ordered from Eurofins Genomics. The codons in these sequences were optimized for mammalian expression using the codon optimization tool from Integrated DNA Technologies (IDT). In addition to the FXII gene, the ordered DNA sequences encode a C-terminal GSGS-linker, a His6-tag and a stop codon, and they are flanked by NheI (GCTAGC) and a HindIII (AAGCTT) restriction sites. The entire DNA sequences are provided in the supporting materials. The DNA sequences were cloned into the pEXPR-IBA42 vector downstream of a BM40 signal sequence for secreted expression in mammalian cells. The ligated vector was transformed into DH5 alpha electrocompetent E. coli cells. Plasmid DNA from single clones was sequenced by Sanger sequencing.Recombinant expression, purification, and activation of FXII1.5 mg plasmid DNA was transfected into 500 ml CHO cells (Thermo Fisher Scientific) in cell suspension culture using polyethylenimine (PEI). Cells were incubated for seven days at 37 °C, 5% CO2 under shaking conditions. The cells were removed by centrifugation and the secreted protein was purified from the supernatant using a nickel-charged immobilized metal affinity chromatography (IMAC) column (5 ml HisTrap FF Crude, GE Healthcare). The column was equilibrated with buffer containing 15 mM imidazole, 100 mM NaCl, 20 mM Tris-HCl, pH 7.4. The pH of the cell culture supernatant was adjusted to 8.0 using NaOH. The supernatant was run through the column at a flow of 5 ml per min. The column was washed with 20 column volumes of equilibration buffer at a flow of 5 ml per min. The protein was eluted with 500 mM imidazole, 100 mM NaCl, 20 mM Tris-HCl, pH 7.4 at a flow of 5 ml per min. The buffer was subsequently exchanged to 100 mM NaCl, 20 mM Tris-HCl, pH 7.4 by three iterative steps of 10-fold dilution and 10-fold concentration using a 10,000 MWCO centrifugal concentrator. The concentration was determined by measuring absorption at 280 nm (ε = 100,000). For the activation of recombinant FXII, 10 μg protein was diluted to a total volume of 10 μl 100 mM NaCl, 20 mM Tris-HCl, pH 7.4 and further diluted with 10 μl of 2-fold concentrated assay buffer (300 mM NaCl, 20 mM MgCl2, 2 mM CaCl2, 20 mM Tris-HCl, 0.02% (v/v) Triton-X100, pH 7.4). Dextran sulfate 500 kDa (DXS500) was added to a final concentration of 0.2 μg μl−1 and incubated for 1 hour at 37 °C. 5 μg protein was analyzed by SDS-PAGE under reducing conditions.Animal study authorizationAll experiments in mice and pigs were conducted in accordance with the terms of the Swiss animal protection law and were approved by the animal experimentation committee of the cantonal veterinary service (Canton of Berne, Switzerland). The pharmacokinetic studies in rabbits were performed at Washington Biotech Inc. following ethical standards for animal studies of the Office for Laboratory Animal Welfare (OLAW), a division of the US Public Health Service as administered by the National Institutes for Health. The extracorporeal circulation studies in rabbits were performed in compliance with the Allegheny Health Network Institutional Animal Care and Use Committee. The studies in rabbits were approved by the American Association for Laboratory Animal Science (IACUC).Pharmacokinetic study in miceMice for all experiments were kept at ambient temperature (around 20 °C), around 60% humidity, half-day light/half-day dark cycles, and in groups of 2–5 mice per cage. C57BL/6J wild-type mice (male, 10–20 weeks old, 25–30 g, Charles River) were injected subcutaneously over the shoulders with 5 mg kg−1 of FXII900 (0.5 mg ml−1 in PBS, pH 7.4). The mice were anaesthetized 3 min before the scheduled blood collection time point (40 mg kg−1 pentobarbital). An abdominal midline incision was performed and 450 μl of whole blood was drawn from the inferior vena cava into a syringe containing sodium citrate (50 µl 3.2% [w/v] sodium citrate). The mice were euthanized by cervical dislocation. The blood was immediately processed to plasma by centrifugation for 10 min at 2000 × g and 4 °C, and stored at −80 °C. The concentration of FXII900 in the plasma samples was determined by LC-MS as described below. Differenct mice were used for each time point.Pharmacokinetic study in rabbitsThe pharmacokinetic properties of FXII900 applied intravenously were determined as follows44. Female New Zealand White rabbits (10–20 weeks old) were injected with 3.7 mg kg−1 FXII900 dissolved in 1 ml PBS, pH 7.4 via the ear vein. Blood samples (2.7 ml) were collected at different time points into sodium citrate tubes (BD Vacutainer ref # 363083) and immediately processed to plasma by centrifugation at 1400 × g for 15 min. The concentration of FXII900 in the plasma samples was determined by LC-MS as described below. The pharmacokinetic properties of FXII900 applied subcutaneously were determined as follows. Female New Zealand White rabbits (2.5–2.9 kg) were injected subcutaneously over the shoulders with 3.7 mg kg−1 of FXII900 dissolved in 1 ml PBS pH 7.4. Blood samples (1.8 ml) were collected at different time points into sodium citrate tubes (BD Vacutainer ref # 363080) and immediately processed to plasma by centrifugation at 1800 × g for 10 min. The plasma was stored at −80 °C. The concentration of FXII900 in the plasma samples was determined by LC-MS as described below.Pharmacokinetic study in pigsSwiss large white pigs, both sexes, 3–4 months old (30 ± 5 kg) were anesthetized and prepared following procedures used to study myocardial ischemia/reperfusion injury45. In this procedure, the ACT is monitored and 2500 IU of heparin are injected when ACT values fall below 180 s. Pigs were injected intravenously with 4 mg kg−1 of FXII900 dissolved in 1 ml PBS, pH 7.4. Blood samples (2.9 ml) were collected at different time points into sodium citrate tubes (Sarstedt S-Monovette ref # 04.1902.001) and immediately processed to plasma by centrifugation at 1400 × g for 15 min at RT. The concentration of FXII900 in the plasma samples was determined by LC-MS as described below.Quantification of inhibitor in plasma samplesThe concentration of FXII900 in the plasma samples was quantified based on peak intensities of total ion current (TIC) chromatograms acquired by LC-MS (LCMS-2020, Shimadzu). To 15 μl of plasma sample, 1 μM of internal standard peptide and 5 μl of 6 M guanidinium hydrochloride solution were added and mixed. Plasma proteins were precipitated by the addition of 400 µl of ice cold ethanol (99.9% [v/v] EtOH, 0.1% [v/v] TFA) and incubated on ice for one hour. Precipitate was removed by centrifugation (9000 × g, 20 min, 4 °C) and the supernatant dried by centrifugal evaporation under vacuum. Dried samples were dissolved by sequentially adding 2 μl of DMSO and 18 μl of H2O containing 0.1% (v/v) CHOOH and analyzed by LC-MS. The samples were analyzed using an analytical C18 column (Phenomenex C18 Kinetex column, 50 × 2.1 mm, 2.6 µm, 100 Å) and a linear gradient of 5–30% solvent B (MeCN, 0.05% [v/v] CHOOH) in solvent A (H2O, 0.05% [v/v] CHOOH) in 4.5 min at a flow of 1 ml per min. The mass was measured on a single quadrupole mass spectrometer in positive ion mode using electrospray ionization. Peptides were quantified based on the absolute intensities of the detected mass peaks (M3+ and M4+).FeCl3 injury thrombosis model in mesenteric arterioles in miceA model of thrombosis in mesenteric arterioles using intravital microscopy was performed according to Angelillo-Scherrer A. et al.46 with minor modifications. C57BL/6J wild-type mice (male, 10–20 weeks old, 25–30 g, Charles River) were injected intravenously with Rhodamine 6G (100 μl, 1 mM, ACROS Organics product 41902) to fluorescently label the platelets and leucocytes. The mice were injected subcutaneously over the shoulders with FXII900 (5 or 25 mg kg−1, 0.5 mg ml−1 in PBS, pH = 7.4) or the negative control FXII901 (5 or 25 mg kg−1, 0.5 mg ml−1 in PBS, pH = 7.4), or intravenously (retro-orbital) with 200 IU kg−1 heparin, and subsequently anesthetized with ketamine (80 mg kg−1) and xylazine (16 mg kg−1) via intraperitoneal injection. An abdominal midline incision was made to expose the mesenteric arterioles which were imaged by intravital microscopy using a Mikron IVM500 microscope (Mikron Instruments) coupled with a 50 W mercury lamp (HBO 50 microscope illuminator, Zeiss) attached to combined blue (exciter 455DF70, dichroic 515DRLP, and emitter 515ALP) and green (exciter 525DF45, dichroic 560DRLP, and emitter 565ALP) filter blocks. Thrombus formation was induced by the application of a 1 × 2 mm filter paper saturated with FeCl3 solution (7.5% [w/v], Roth, art no 5192.1) onto the blood vessel for 1 min. The blood flow, clot formation and vessel occlusion were monitored for 20 or 25 min wherein images were recorded every minute using a digital video cassette recorder (DSR-11, Sony) and analyzed using ImageJ software (version 1.52). Mice were euthanized by final bleeding and cervical dislocation. Time to clot formation and full occlusion of the blood vessel were determined as follows. Clot formation was defined as the formation of a clear aggregation with a diameter of around 10 µm or larger. Occlusion was defined as a clot covering the full diameter of the vessel. Mice showing no speckled pattern at the site of FeCl3 application were assumed to be not injured at the blood vessel and were excluded from the analysis.Tail bleeding time and volumeMice (8- to 10-week-old) were anesthetized with pentobarbital (50 mg kg−1, IP), treated with PBS (IV, retro-orbital), 200 IU kg−1 heparin (IV, retro-orbital), 25 mg kg−1 FXII900 (SC) or 25 mg kg−1 FXII901 (SC), and put on a 37 °C heating pad. After 5 min, the distal tail was transected at 2 mm with a disposable surgical blade, and the diameter measured to confirm that it is > 1 mm. The tail was placed in a 50 ml falcon tube filled with phosphate-buffered saline (warmed to 37 °C) and the bleeding time was recorded. The total blood loss was determined by measuring the absorption at 540 nm.Artificial lungsMini-lungs were created that allowed for blood exposure to the fibers in the artificial lung, but for construction simplicity, do not transfer gas. Polymethylpentene (PMP) fiber (3 M) with 50% porosity, a 380 μm outer diameter, and two layers at 30° cross angles were cut into 1.78 × 1.78 cm squares. Five of these layers were put together, making sure the fibers all ran in the same direction, and they were sealed together into a fiber bundle using a hot plate on each side of the square sheets. These fiber bundles were melted such that they had a 1.57 × 1.57 cm square frontal area perpendicular to the flow. Eight 5-layer “chiclets” were placed together to make up the fiber bundle of one mini lung device, giving a final surface area of 263 cm2. These fiber bundles were placed in a square plastic housing of 3.05 cm in length, making sure that the fiber bundles fit tightly in the device to prevent shunting around the fiber bundle. This housing was attached to plastic end caps that were placed on each end with a 1/8″-barbed tube fitting for 3/16″-ID tubing to allow the device to be connected to tubing in the circuit. These end connectors were coated with Teflon tape before connecting them to the plastic end caps to prevent leakage. The device was held together with two screws going from end cap to end cap. The entire device was secured with silicone to eliminate leakage. The silicone was left to dry for 24 h, and the device leak tested as follows. Filtered deionized water was run through the device to check for leaks within the device. If no leaks were found, the device was left to dry with filtered air running through the device for 24 h.Extracorporeal circulation in rabbitsNew Zealand white rabbits of 3.2–4.2 kg (Charles River) were anesthetized via intramuscular injections of ketamine (30 mg kg−1) and xylazine (5 mg kg−1). One of the ear veins was catheterized via a 24G winged catheter, and the rabbits were intubated with a 3.0 endotracheal tube. The animals were kept anesthetized throughout the four-hour experiment via inhaled isoflurane (2%) and were ventilated with a peak inspiratory pressure (PIP) of < 20 cmH2O, positive end expiratory pressure of 5 mmHg, tidal volume of 4–6 ml kg−1, and a respiratory rate of 22–60 breaths per min. Tidal volume and respiratory rate were adjusted to maintain normal arterial blood gases and the listed PIP. Phenylephrine was applied intravenously at a rate of 0.5–5 μg kg−1 min−1 to maintain blood pressure. For monitoring blood pressure and collecting blood samples, the rabbits’ right or left carotid arteries were cannulated using a 16-gauge angiocatheter (Becton Dickinson) and were secured with silk ties. The device and circuit were first primed with filtered CO2 and then with saline (NaCl, 0.9% [w/v]) containing solumedrol (30 mg kg−1). At this point, rabbits of the inhibitor-treated group were injected with 2 mg kg−1 of FXII900 as a 2 mM solution in PBS (pH 7.4) via the ear vein and the circuit was connected to the rabbits via a venous/venous ECMO configuration using a 14-gauge angiocatheter (Becton Dickinson) and a 6″ pressure tubing that was cut to length from a 24″ pressure tubing (Edwards Lifesciences) in the right and left internal jugular veins, respectively. The circuit was placed in a roller pump (Cobe), and the blood flow was set to 45 ml per min. After connecting the extracorporeal circuit, rabbits of the inhibitor-treated group were infused with 0.075 mg kg−1 min−1 FXII900 via the ear vein for the entire duration of the study. Rabbits treated with heparin were initially infused with 60 IU h−1 (100 IU ml−1) starting before connection to the circuit. The drip rate was adjusted based on the following nomogram to maintain the ACT in a range between 220 and 300 s. ACT > 320 s: decrease infusion rate by 12 IU h−1, 320 s > ACT > 300 s: decrease infusion rate by 6 IU h−1, 300 s > ACT > 220 s: maintain heparin infusion rate, 220 s > ACT > 200 s: increase infusion rate by 6 IU h−1, 200 s > ACT > 180 s: increase infusion rate by 12 IU h−1, ACT < 180 s bolus 100 U (1 ml). After four hours, the animals were euthanized by potassium chloride (2 mg kg−1, IV). At the end of the experiment, 5 ml of heparin was run through the artificial lung while it was still connected to the rabbits. Then the circuit was removed, and the device was washed with saline carefully so that no clot shedding occurred. Saline was run through the device until the drained portion the effluent was clear. Clot volume was measured by measuring the volume of the device prior to the experiment and again at the end of the experiment after washing. This was done by completely filling the device with saline and recording this amount of saline as the device volume. The difference between the beginning and ending volume was determined to be the clot volume. From the clot volume, the percent of clot within the device was calculated.Data acquisition during extracorporeal circulationPlatelet and white blood cell counts, hematocrit, arterial blood gases (ABG), ACT, aPTT, fibrinopeptide A (FPA), device resistance, and bleeding time were measured prior to circuit attachment and at 10, 30, 120, and 240 min following the initiation of ECMO. For the platelet counts, a syringe with 0.05 ml of 3.2% sodium citrate (w/v) was used to draw 0.45 ml of blood for a total volume of 0.5 ml. This was then centrifuged at 60 × g for 10 min, and 20 µl of the plasma was placed in 20 ml of ISOTON® diluent and counted using a Coulter Counter (Beckman Coulter, Inc. Brea, CA) with a 50 μl aperture tube. For counting, cells were considered platelets if their diameters were 1.8–5.6 μm. For white blood cell counts, 40 μl of whole blood was placed in 20 ml of the ISOTON® diluent. Six drops of ZAP-OGLOBIN lysing solution were added to the mixture and mixed gently. This was allowed to sit for 2 min. The white blood cells were also counted using the Coulter Counter (Beckman Coulter) with a 50 μl aperture tube where any particle above 3.6 μm in diameter was considered a white blood cell. Arterial blood gases were measured by drawing 0.4 ml of blood into a heparinized syringe and run using an arterial blood gas analyzer (ABL800 FLEX, Radiometer). The ACT and hematocrit were measured by collecting 0.5 ml of blood. The ACT was measured using a Hemochron analyzer with tubes containing glass beads as the activator. The hematocrit was measured via capillary centrifugation. For aPTT measurements, blood samples (1.8 ml) were collected at different time points into tubes containing sodium citrate (0.2 ml, 3.2% [w/v] sodium citrate) and immediately processed to plasma by centrifugation at 2000 × g for 15 min at 4 °C. The samples were then analyzed as described above. All platelet counts, WBC counts, FPA, and FXIIa levels were corrected for hemodilution by adjusting the raw values based on the hematocrit. The inlet and outlet pressure and blood flow rate were measured using a Biopac system (Aero Camino Goleta, CA) and pressure transducers at the inlet and outlet of the device (Edwards Lifesciences), and the resistance was calculated with the standard R = (Pi − Po)/Q where Pi is the inlet pressure in mmHg, Po is the outlet pressure in mmHg, and Q is the flow in L min−1. The bleeding time was measured by cutting small incisions of 4–5 mm at different sites of the right or left ear in each animal. Blood from the incision was removed with gauze every 30 s, and the time until the bleeding stopped was measured.Statistical analysisFor all experiments, mean values are indicated for independent replicates. For experiments performed in triplicate or more, means and standard deviations are indicated. All statistical analyses were performed using GraphPad Prism (version 5) or Microsoft Excel software (version 2016). A chi-squared test was used to determine whether mice were protected against clot formation and full occlusion in the FeCl3 thrombosis model. A one-tailed student’s t-test was applied to determine the significance of the difference in the time until clot formation occurred between the two groups. A one-tailed test is appropriate since the effect is only expected in one direction. A one-tailed Student’s t-test was applied to assess the significance for the bleeding time prolongation in the mouse tail bleeding experiment.Reporting summaryFurther information on experimental design is available in the Nature Research Reporting Summary linked to this paper.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary | nature communications | [
"Article"
] | [
"Recombinant peptide therapy",
"Drug discovery and development",
"Thrombosis"
] | IntroductionCoagulation stops bleeding at vessel wall injuries excess dangerous to thrombosis Blood vessel blockages cause myocardial infarction ischemic stroke pulmonary embolism leading causes of disability death in industrialized anticoagulants for thrombosis2,3 bleeding common side effect weigh risks benefits need for effective anticoagulants not hemostatic perspective anticoagulants reduced bleeding risks inhibition of coagulation factors XII (FXII) XI proteases intrinsic coagulation pathway5,6 FXII initiating protease drives intrinsic pathway coagulation kallikrein-kinin Mice lacking FXII display reduced thrombosis in protected from pathological thrombosis in cerebral ischemia10 humans lacking FXII-knockout mice have normal hemostatic capacity not bleed drugs targeting protease antithrombotic effects without compromising hemostasis reduction of FXII expression by oligonucleotides suppresses thrombosis in arterial venous thrombosis mouse catheter thrombosis in rabbits12inhibition FXII by protein-based inhibitors insect plant reduces thrombosis in mouse rat rabbit primate models potential for therapeutic anticoagulation.FXII-driven blood coagulation in cardiopulmonary bypass surgeries heart-lung machine Contact between FXII artificial surfaces induces conformational change proteolytic activation FXII zymogen contact system Activation procoagulant intrinsic coagulation pathway proinflammatory kallikrein-kinin system leads to blood clotting inflammation19 Contact activation problem in extracorporeal membrane oxygenation artificial lung severe cardiac standard strategy for suppressing contact activation high doses heparin inhibit proteases coagulation risk of bleeding21 strategies for suppressing contact activation established heparin remains anticoagulation strategy Renné human FXIIa-inhibiting antibody 3F7 prevents clotting thrombosis in cardiopulmonary bypass system in rabbits without increasing bleeding14 usefulness targeting FXIIFXII implicated in medical hereditary angioedema reperfusion Alzheimer’s multiple potential FXIIa inhibitors treatments Type III HAE mutation in FXII activation bradykinin release improved FXIIa inhibitory antibody 3F7 phase I clinical evaluation for treatment high-affinity protein-based FXIIa inhibitors generated development synthetic small-molecule inhibitors challenging Small molecules strengths uniform composition efficient tissue penetration high stability low immunoreactivity ease production best small molecule FXIIa inhibitors coumarin derivative 44 4.4 H-D-Pro-Phe-Arg-chloromethylketone 0.18 μM useful covalent inhibition mechanism moderate potency selectivity limit drug development potential identified high-affinity FXIIa inhibitors based peptides bicyclic peptide FXII618 8.1 nM substitution of amino acids FXII618 to unnatural ones improved affinity34 limited binding to animal FXIIa homologs low proteolytic stability plasma inhibitors evaluated in vivoimprovement inhibitor sub-nanomolar potency human mouse rabbit FXIIa enhanced stability plasma several days (half-life human plasma properties pre-clinical evaluation animal models synthetic FXIIa inhibitors prevent thrombosis mice suppress coagulation artificial lungs rabbits without risk bleeding.ResultsEngineered macrocycle inhibits FXIIa sub-nanomolar affinityIn FXII inhibitor FXII618 N-terminal arginine (Arg1) cleaved by plasma proteases 4 h 40-fold reduction Ki Arg1 substituted with unnatural amino acids stability improved weaker inhibition amino acids positively charged side chains smallest affinity losses34 new attempt replaced Arg1 with di-peptides larger surface area FXIIa productive interactions (Fig. 1a Table 1) amino acids di-peptides D-amino acids prevent proteolysis one D-Arg positive charge developed synthesis screening approach coumarin-labeled peptides activities stability assay di-peptides D resistant to proteolysis plasma half-life five-fold to 20 h di-peptides improved Ki values two-fold over FXII618best inhibitor Arg1→D-Arg-D-Ser (FXII850) synthesized without coumarin label Ki 5 nM 1.6-fold improvement compared FXII618 (Ki 8.1 ± nM tested diverse amino acid substitutions for Arg8 FXII618 not optimized screening strategy substitution Arg8→His (FXII851) affect Ki human FXIIa mouse FXIIa inhibition Ki values 6.5 nM 36 nM 1.2- 2.4-fold improvements over FXII618 1Improving affinity stability bicyclic peptide FXIIa inhibitor Amino acids Arg1 Arg8 FXII618 substituted to two D-amino acids relative Ki values human mouse FXIIa compared FXII618 Average values for improved affinity Schematic lead peptide FXII618 Amino acid substitutions activity stability Inhibition human FXIIa by FXII618 variants single amino acid substitution FXII900 modifications Arg13 deleted Residual FXIIa activity determined in two three five) independent measurements Mean values inhibitors tested three or five times data mean ± SD.Specificity profiling FXII900 activities human FXIIa ten proteases measured Inhibition FXII five plasmin APC three other proteases two Means values Proteolytic stability FXII900 two precursors Bicyclic peptide incubated human plasma 37 °C inhibitor activity quantified FXIIa activity assay fluorogenic substrate Two measurements FXII850 FXII900 four FXII618 Means values Proteolytic stability FXII618 FXII900 peptide incubated plasma 37 °C analyzed LC-MS Total ion count (TIC) shown Source data file combined amino acid substitutions Arg1→D-Arg-D-Ser (FXII850) Arg8→His (FXII851) Arg11→(S)-β3-homoarginine Phe3→4-fluorophenylalanine (Fig. 1b c Supplementary Table 2) deleted C-terminal Arg13 binding resulting inhibitor FXII900 blocked human mouse FXIIa Ki values 0.37 ± 0.04 nM 0.45 ± 0.11 nM more potent precursorsfinal inhibitor synthesized by solid-phase peptide synthesis macrocyclization crude peptide with 1,3,5-triacryloyl-1,3,5-triazinane linker single HPLC purification step > 95% pure product isolated yield than 50% Fig. 3) developed structural model precursor peptide FXII618 to FXIIa positions interactions amino acids Phe3 to Pro633 beneficial amino acid substitutions outside region model affinity enhancements specificity profiling against homologous plasma proteases performed off-target interactions Activity assays showed FXII900 highly selective (Fig. 1d relevant proteases displayed 100,000-fold or higher selectivity (Ki values > 40 only trypsin significant inhibition low micromolar concentrations (Ki = 1.46 μM).FXIIa inhibitor stable in human plasma FXII900 incubated in human plasma at 37 °C remaining inhibitory activity quantified Mass spectrometric analysis showed FXII900 no degradation products quantity intact inhibitor decreased over time inhibitor degraded cleavage products not detectable half-life of FXII900 128 h 25-fold longer than FXII618longer than FXII850 carrying Arg1→D-Arg-D-Ser modification (t1/2 = 18.5 h Fig. more contributed to improved stability important for stability reverted amino acid substitutions 4 retested variants Arg11→βhArg substitution contributed most to stability improvement 5).FXII900 inhibits intrinsic coagulation pathway ex FXII900 FXII blood plasma coagulation selectivity performed coagulation tests activated partial time prothrombin time coagulation selective FXIIa inhibition aPTT not PT results concentration for 1.5x increase coagulation time FXII900 prolonged aPTT in human plasma EC1.5x 0.79 μM 4.3-fold better than precursor FXII618 (EC1.5x 3.4 μM FXII900 affect PT at highest concentration (120 selectivity for FXIIa target (Fig. 2b). inhibitor prolonged aPTT in other species mouse 2.9 rabbit 0.102 pig 12.2 potency varied rabbit plasma strong prolongation of aPTT at low inhibitor concentrations transition to maximal effectpattern seen for antibody-based FXIIa inhibitors in rabbit expected rabbit plasma-specific not strong affinity for rabbit FXIIa inhibitor prolong PT in three animal plasma tested (Fig. 2b).Fig. 2Inhibiting FXIIa different species Inhibition intrinsic coagulation pathway in human mouse rabbit pig plasma ex vivo aPTT different concentrations FXII900 shown aPTT of FXII618 measured coagulation times determined in two four six measurements Mean values indicated PT different concentrations FXII900 FXII618 in human mouse rabbit pig plasma determined in two three measurements Mean values indicated means ± SD indicated Inhibition of recombinant FXIIa from different species by FXII618 FXII900 inhibition human FXIIa from blood shown Fig. 1c). Residual FXIIa activities determined in two three measurements Mean values indicated pig rabbit FXIIa inhibition data as mean ± SD Source data provided file assess FXII900 prolonged aPTT in plasma four species cloned expressed rabbit pig FXII tested inhibition by FXII618 FXII900experiments showed FXII orthologues inhibited aPTT prolongation correlated with FXII inhibition in mice rabbits assessed properties three species mice rabbits pigs models thrombotic diseases mice determined pharmacokinetics following subcutaneous administration (5 mg kg−1 n = 3 mice FXII900 reached plasma concentration 1 μM after 3 min remained above 30 min-MS (Fig. 3a aPTT prolonged extended two-fold at 15 min over 1.5-fold until 30 min 3Pharmacokinetics FXII900 in mouse rabbit pig mouse after subcutaneous administration (5 mg kg−1 n = 3) Concentration FXII900 plasma Means ± SD rabbit after intravenous administration (3.7 mg kg−1 n = 3) Concentration FXII900 Plasma samples analyzed duplicate rabbit 1 triplicate rabbits 2 3. Mean values indicated data mean ± SD Pharmacokinetics rabbit after subcutaneous administration (3.7 mg kg−1 n = 4) Concentration FXII900 plasma measured Mean values each rabbitPharmacokinetics pig intravenous administration (4 mg kg−1 n = 3) Concentrations FXII900 plasma determined mean values each pig Source data rabbits pharmacokinetic properties intravenous subcutaneous administration intravenous administration = 3 3.7 mg kg−1) FXII900 elimination half-life 12 ± 2 min parameters Supplementary Table 5) aPTT eight-fold prolonged three-fold prolonged (40 min dose subcutaneously (n = 4 3.7 mg kg−1) peptide narrower concentration range above 100 nM below 300 nM 10 80 min aPTT subcutaneous administration prolonged 2-fold 40 min determined pharmacokinetics FXII900 pigs intravenous administration (n = 3, 4 mg kg−1) inhibitor maximal concentration 7 μM cleared t1/2 36 ± 5 min clearance rate volume distribution 11.8 ± 1.8 ml kg−1 min−1 610 ± 140 ml kg−1 Table 6) toxicity adverse effects.FXII900 inhibits ferric chloride-induced thrombosis evaluated thromboprotective properties ferric chloride-induced thrombosis mouse modelapplied blood vessels FeCl3 induces thrombosis red blood cells activates platelets at test compound thrombosis FXII900 or inactive control peptide FXII901 administered subcutaneously to two groups 10 mice (5 mg kg−1) negative control FXII901 variant of FXII900 with three modifications (Phe4F3→Phe Arg4→Ala βhArg11→Arg inhibitory constant 100,000-fold for FXIIa (Ki = 39 ± 14 μM solution 7.5% FeCl3 applied to mesenteric arterioles blood flow monitored by intravital microscopy for 25 min Mice injured by FeCl3 identified due to speckled pattern accumulation of platelets 4a Seventeen out of 20 mice showed pattern (nine treated eight control taken for further analysis arterioles mice FXII901 formed clots vessels occluded after 10 min extent clot formation for control peptide expected mice treated with inhibitor FXII900 showed speckled pattern occasional formation blood clot no complete occlusion observed quantified anticoagulant effect of FXII900 rates for clot formation4b blood vessel occlusion blood clot Fig. 4b right eight mice control peptide seven blood clotting five full-vessel occlusion FXII900-treated mice three clotting no vessels fully occluded chi-squared test 0.02 0.005 few mice coagulation developed blood clots 10 min later control chi-squared test = 0.006 Table 7). 4Thromboprotection FXII900 mice effect bleeding Intravital fluorescence microscopy images mesenteric arterioles thrombosis induced topical application FeCl3 (7.5% 1 Platelets labeled Rhodamine 6G images two mice treated FXII900 inactive control peptide FXII901 15 min before ferric chloride (5 mg kg−1 Vessel walls FeCl3 application site indicated yellow markers Three changes speckled pattern clot formation vessel occlusion indicated Scale bar 200 μm percentage mice clot formation full occlusion after ferric chloride application time (inhibitor FXII900 5 9 25 8 control FXII901 5 heparin Clot formation diameter larger than 10 μmocclusion blood vessel blood clot same diameter Blood loss bleeding time mice 2 mm tail transections treated 5 min before clipping tips vehicle 6) heparin (200 IU kg−1 10), FXII900 (25 FXII901 6) Mean values standard deviations indicated significance prolonged bleeding time assessed unpaired one-tailed t-test p value significant results not significant Mice treated heparin showed small medium-to-large loss blood mean value not indicated experiment 5-fold higher dose FXII900 8 negative control FXII901 (n 9 comparable results no delay coagulation reduction clot formation (Fig. 4b intravital microscopy 20 min 5 min shorter positive control heparin added (n = 10 200 IU kg−1) more mice showed blood clotting (six vessel occlusion (one than groups FXII900 treated FXII900 normal hemostatic assessed bleeding propensity mice higher doses FXII900 ferric chloride-induced thrombosis study (25 mg tail transection model 2 mm diameter > 1 mm n = 6)treated mice with PBS 6) heparin (200 IU kg−1 placed tails tube warmed PBS measured bleeding volume blood lost Mice injected heparin bled continuously 30 min similar previous small or large loss blood (3 heterogeneous result heparin control four additional mice medium to high blood loss Mice treated FXIIa inhibitor average bleeding time 10 min blood loss 21 μl comparable to PBS-treated mice 7 min blood loss 34 μl FXII901 control FXIIa inhibitor affect hemostatic capacity.FXII900 bleeding-free anticoagulation artificial FXII coagulation ECMO tested artificial lung rabbit model White rabbits anesthetized treated intravenously with FXII900 (n = 4 2 mg kg−1 bolus 0.075 mg kg−1 min−1 infusion four left untreated 3) or treated heparin 2 60 IU h−1 rate adjustments 220–300 4-h veno-venous ECMO configuration polymethylpentene (PMP) fiber artificial lung system FXII900 prolonged coagulation parameters activated clotting time) 10-fold larger than heparin prolonged 1.5 2-foldb). aPTT measurements FXII900 no coagulation at 240 s maximal value Fig. 12a). untreated rabbits coagulation times baseline 20 s 170 s (ACT increased pressure inlet artificial lung indicates clogging blood coagulation measured resistance difference pressure divided flow rabbits treated FXII900 resistance 50 mmHg L−1 min−1 except one rabbit resistance after two hours (Fig. 5c contrast untreated animals resistance increased first hour doubled from 50 to 100 mmHg L−1.Fig. 5Bleeding-free anticoagulation artificial lung model rabbits.Rabbits connected artificial lung four hours FXII900 injected bolus (2 mg kg−1) constant infusion (0.075 mg kg−1 min−1) (n = 4 data blue). control group untreated (n = 3 data red). Rabbits heparin group (n = 2) infused with one IU per min heparin drip rate adjusted maintain ACT values between 220 and 300 s Values time point 0 measurements before connecting lung Mean ± SD for four rabbits FXII900-treated group three control group Mean values for two rabbits heparin groupData rabbits in Supplementary Fig. 12. aPTT. Coagulation times 240 s plasma coagulate ACT. Clotting times 1500 s blood coagulate Resistance pressure inlet outlet device flow rate Bleeding time incision injuries ears Platelet count Volume blood clots % volume artificial lungs excessive bleeding concern anticoagulation ECMO systems analyzed FXII900 affected bleeding time platelet count Rabbits treated inhibitor normal bleeding times (4–5 min 5d group treated with heparin 2-fold prolonged bleeding times (5–10 min) platelet count same three groups decreasing two-fold from 2 × 108 cells per ml constant experiment (Fig. 5e 12e). Analysis artificial lungs inhibitor- heparin-treated rabbits reduced clotted blood volume clot device volume 10 6% for FXII900-treated rabbits smaller than untreated rabbits (37 ± 10%; p = 0.03) slightly larger than heparin-treated rabbits (5% 5f protein-based inhibitors developed suppression FXIIa activity efficient anticoagulation without bleeding risks development high-affinity small-molecule FXIIa inhibitors laggingsynthetic molecules improved binding affinity stability FXII618 peptide macrocycle FXIIa inhibitor developed phage display inhibitor FXII900 blocks FXIIa Ki 370 pM 100,000-fold selectivity over plasma proteases stable in blood plasma improvement binding affinity 21-fold proteolytic stability 25-fold random screening rational design strategies Two amino acid substitutions affinity stability identified mutated peptides combined with two beneficial mutations amino acid substitutions FXII900 binding affinity modifications Arg1 to D-Arg-D-Ser Arg11 to βhArg improve stability higher binding affinity prolongation aPTT species EC1.5x improved 4.3-fold human plasma 10-fold mouse plasma EC1.5x values higher than Ki values FXII inhibited prolong aPTT 1.5-fold Ki 50% FXII inhibited-enhanced potency in mouse plasma planned pharmacologic study mice FXII900 prolong PT at 120 μM contrasted FXII618 prolonged PT slightly concentration modification four amino acids improved selectivityenhancement affinity selectivity stability owed to unnatural amino acids synthetic nature inhibitor advantage over protein-based inhibitors.FXII900 nine natural amino acids four non-natural central core three cysteines synthesized by automated solid-phase peptide synthesis large quantities non-natural amino acids inexpensive production moderate cost Crude linear peptide by chemical linker TATA single HPLC purification step required > 95% pure product isolated yields 50% supplementary figure peptide sequence conditions for peptide cyclization ordering custom peptide synthesis.FXII900 rapidly cleared intravenous administration likely by renal filtration small size polar structure elimination half-life 12 min in rabbits to 36 min in pigs prolongation aPTT correlated with PK data species half-life in humans one hour No degradation products detected in plasma samples inhibitor resists proteases blood filtered out as intact molecule high stability long half-life plasma vivo fast clearance single-dose injection of FXII900 prolonged aPTT in mice rabbits for 30 min.short circulation time of small molecules allows flexible adjustment duration to inhibitor pharmacokinetic profile FXII900 attractive for acute disease treatment hemodialysis ECMO anticoagulation after ischemic stroke surgery For hemodialysis inhibitor tethered to large protein or polymer prevent rapid removal For controlling activity FXII over days weeks months FXII900 not suited other antibody-based FXII inhibitors or antisense oligonucleotides better high proteolytic stability FXII900 could resist proteases stomach intestines multiple charges unlikely cross gastrointestinal epithelium preventing oral administration evaluated pharmacologic activity FXII900 in ferric chloride-induced thrombosis mouse model FXII-deficient mice show reduction of coagulation FXII central role bolus injection of FXII900 protected mice from thrombosis bleeding time blood loss not affected contrast to control experiment with heparin (200 IU kg−1) mice constantly bleeding lost large volume blood thrombosis in myocardial infarction ischemic stroke pulmonary embolism complex caused by triggers not dependent on FXII activationresults show synthetic FXIIa inhibitor suppress coagulation tested if FXII900 suppress coagulation membrane artificial lungs rabbit model Heart-lung machines used for surgery difficulty beating heart half a million cardiac operations US use CPB compact systems support cardiac respiratory dysfunction 7000 patients treated annually with extracorporeal life contact blood with hydrophobic polymer surfaces induces FXII activation coagulation suppressed using high doses heparin bleeding risks analysis 80,000 ECMO patients hemorrhage major adverse event bleeding from surgical in 6.3% to 29.3% device thrombosis patient bleeding lead to mortality43 continuous infusion of FXII900 prolonged coagulation parameters ACT aPTT over four hours inhibitor reduced coagulation in artificial lung resistance low reduced clogging reduction in total volume blood clots three-fold lower in treated rabbits all rabbits treated with FXII900 showed normal bleeding times small molecule FXIIa inhibitor suitable for reducing blood coagulation in CPB heparin required to suppress coagulation may reduced heparin dose bleeding risksthrombosis ECMO CPB FXIIa initiates inflammatory kallikrein-kinnin Bradykinin plasma concentrations elevated in CPB ECMO patients targeting FXIIa anti-inflammatory benefit adverse effects organ damage synthetic FXIIa inhibitor sub-nanomolar affinity high selectivity good stability developed efficient anticoagulation animal models without bleeding risks concerns treatment heparin synthetic nature small size efficient production excellent binding properties stability inhibitor applicable for acute procedures risk thrombosis CPB ECMO without compromising hemostasis objectives improve properties peptide macrocycle inhibitor FXIIa assess pharmacokinetic properties in three species evaluate in two animal models inhibitor engineered synthesizing variants lead peptide inhibitory activity specificity stability in vitro Inhibitors prepared 95% purity tested in duplicate affinity of final peptide FXII900 determined in quintuplicate from three synthetic batches pharmacokinetic properties in mice rabbits pigs assessed inhibitor plasma samples liquid chromatography mass spectrometry measuring aPTTeffects inhibitor tested FeCl3-induced thrombosis mice artificial lung rabbits experiments subjects randomly assigned groups method coumarin-labeled test inhibitory activity peptides without purification method FXII618 peptide synthesized-β-(7-methoxy-coumarin-4-yl)-Ala-OH fluorophore 7-methoxy-coumarin quantification peptide synthesis peptides chemically cyclized Ki values determined FXIIa activity assay fluorogenic substrate Peptides synthesized characterized displayed 2.5-fold higher Ki HPLC-purified peptides without tag peptide FXII618 coumarin 2.5-fold weaker Ki (20 ± 3 nM) purified FXII618 without (8.1 0.7 comparison identification active variants FXII618 best peptides synthesized without coumarin label purified stability assaysPeptide (2 μl H2O added 398 μl citrated human plasma final peptide concentration 10 μM mixture incubated water bath 37 °Cpoints 24 48 96 samples 30 μl removed diluted 200 μl buffer (10 mM Tris-HCl pH 7.4 150 mM NaCl 10 mM MgCl2 1 mM CaCl2) incubated 20 min 65 °C inactivate plasma proteases samples stored −20 °C until residual inhibitory activity measured FXIIa inhibition assay centrifuged 5 min 16,000 × g two-fold dilutions prepared 0.5 nM to 0.5 residual activity 0.5 nM β-FXIIa measured 50 μM Boc-Gln-Gly-Arg-AMC substrate IC50 values derived curve Residual inhibition % calculated IC50,0h/IC50,xh*100 IC50,0h 0 IC50,xh after incubation degradation stability peptide cleavage assessed incubating bicyclic peptides mouse plasma-MS system (2 μl H2O added 48 μl citrated mouse plasma final peptide concentration 80 μM mixture incubated water bath 37 °C24 72 96 120 144 samples 5 μl removed mixed 6 M guanidinium hydrochloride incubated 30 min RT Plasma proteins precipitated 200 μl ice cold EtOH 0.1% formic acid 30 min centrifuged 9000 × g 20 min 4 °C supernatant evaporated Speedvac 50 °C reduced pressure residue dissolved 40 μl deionized water 0.1% CHOOH analyzed LC-MS samples analyzed C18 column 50 2.1 mm 2.6 μm 100 Å gradient 5–35% solvent B 0.05% CHOOH solvent A (H2O 0.05% CHOOH 5.5 min flow 1 ml per min peptide degradation products measured quadrupole mass spectrometer electrospray ionization Peptides quantified intensities mass peaks+ M4+) FXII mammalian protein sequences human rabbit pig factor XII human rabbit pig sequences Supplementary Methods proteins ordered Eurofins Genomicscodons optimized for mammalian expression Integrated DNA Technologies FXII gene sequences encode C-terminal GSGS-linker His6-tag stop codon flanked by NheI HindIII) restriction sites DNA sequences in supporting materials cloned into pEXPR-IBA42 vector BM40 expression mammalian vector transformed into DH5 alpha electrocompetent E. cells Plasmid DNA clones sequenced Sanger sequencing FXII1.5 mg plasmid DNA transfected into 500 ml CHO cells incubated seven days 37 °C 5% CO2 shaking cells removed centrifugation secreted protein purified supernatant nickel-charged affinity chromatography) column equilibrated with buffer 15 mM imidazole 100 mM NaCl 20 mM Tris-HCl pH 7.4. pH adjusted to 8.0 NaOH supernatant run column 5 ml per min washed with 20 volumes equilibration buffer protein eluted with 500 mM imidazole 100 mM NaCl 20 mM Tris-HCl pH 7.4buffer exchanged to 100 mM NaCl 20 mM Tris-HCl pH 7.4 10-fold dilution concentration 10,000 MWCO centrifugal concentrator concentration determined absorption 280 nm activation recombinant FXII 10 μg protein diluted 10 μl 100 NaCl 20 Tris-HCl pH 7.4 diluted 10 μl 2-fold concentrated assay buffer (300 mM NaCl 20 mM MgCl2 2 mM CaCl2 20 mM Tris-HCl 0.02% Triton-X100 pH 7.4) Dextran sulfate 500 kDa) added 0.2 μg μl−1 incubated 1 hour 37 °C 5 μg protein analyzed SDS-PAGE reducing conditions experiments mice pigs Swiss animal protection law approved cantonal veterinary service studies rabbits Washington Biotech Inc. Laboratory Welfare extracorporeal circulation studies rabbits Allegheny Health Network Animal Care Committee approved Association Laboratory Animal Science study ambient temperature 60% humidity half-day light dark 2–5 mice per cageC57BL/6J wild-type mice 10–20 weeks old 25–30 g Charles River injected subcutaneously 5 mg kg−1 FXII900 (0.5 mg ml−1 PBS pH 7.4) anaesthetized 3 min before blood collection (40 mg kg−1 abdominal midline incision 450 μl blood drawn inferior vena cava syringe sodium citrate (50 μl 3.2% euthanized cervical dislocation blood processed plasma centrifugation 10 min 2000 4 °C stored −80 °C concentration FXII900 determined LC-MS Differenct mice study properties FXII900 Female New Zealand White rabbits (10–20 weeks old injected 3.7 mg kg−1 FXII900 1 ml PBS pH 7.4 Blood samples (2.7 ml collected tubes processed plasma centrifugation 1400 15 min concentration FXII900 determined LC-MS properties Female New Zealand White rabbits (2.5–2.9 kg) injected 3.7 mg FXII900 1 ml PBS pH 7.4. Blood samples (1.8 ml collected processed plasma centrifugation 1800 10 min stored −80 °Cconcentration FXII900 plasma determined by LC-MS study pigsSwiss large white pigs 3–4 months old (30 ± 5 kg anesthetized prepared myocardial ischemia/reperfusion ACT monitored 2500 IU heparin injected below 180 injected intravenously 4 mg kg−1 FXII900 1 ml PBS pH 7.4. Blood samples (2.9 ml collected sodium citrate tubes processed plasma centrifugation 1400 × g 15 min concentration FXII900 determined LC-MS inhibitor concentration FXII900 quantified peak intensities) chromatograms LC-MS 15 μl plasma sample 1 μM internal standard peptide 5 μl 6 M guanidinium hydrochloride added mixed proteins precipitated 400 μl ice cold ethanol (99.9% EtOH 0.1% TFA incubated one hour removed centrifugation dried centrifugal evaporation Dried samples dissolved 2 μl DMSO 18 μl H2O 0.1% CHOOH analyzed by LC-MSsamples analyzed C18 column 50 × 2.1 mm 2.6 μm 100 Å gradient 5–30% solvent B (MeCN 0.05% CHOOH solvent A (H2O 0.05% CHOOH 4.5 min flow 1 ml per min mass measured quadrupole mass spectrometer electrospray ionization Peptides quantified intensities mass peaks (M3+ M4+).FeCl3 injury thrombosis model mesenteric arterioles Angelillo-Scherrer A. al.46 modifications C57BL/6J wild-type mice 10–20 weeks old 25–30 g injected Rhodamine 6G (100 μl 1 mM ACROS Organics platelets leucocytes injected subcutaneously FXII900 25 7.4) negative control FXII901 intravenously 200 IU kg−1 heparin anesthetized ketamine (80 mg xylazine (16 mg kg−1) intraperitoneal injectionabdominal midline incision mesenteric arterioles imaged by intravital microscopy Mikron IVM500 microscope 50 W mercury lamp blue green filter blocks Thrombus formation induced 1 × 2 mm filter paper FeCl3 solution (7.5% blood vessel for 1 min blood flow clot formation vessel occlusion monitored 20 or 25 min images recorded every minute digital video cassette recorder analyzed ImageJ software 1.52) Mice euthanized by final bleeding cervical dislocation Time to clot formation full occlusion determined Clot formation clear diameter 10 μm or Occlusion clot covering full diameter Mice no speckled pattern FeCl3 application not injured excluded from analysis.Tail bleeding time volumeMice (8- to 10-week-old) anesthetized with pentobarbital (50 mg treated with PBS 200 IU kg−1 heparin 25 mg kg−1 FXII900 37 °C heating pad After 5 min distal tail transected at 2 mm with surgical blade diameter measured > 1 mmtail placed in 50 ml falcon tube phosphate-buffered saline (warmed to 37 °C bleeding time recorded total blood loss determined absorption at 540 nm.Artificial lungsMini-lungs created blood exposure to fibers transfer gas Polymethylpentene (PMP) fiber (3 M) with 50% porosity 380 μm outer diameter two layers at 30° cross angles cut into 1.78 × 1.78 cm squares Five layers sealed into fiber bundle hot plate fiber bundles melted 1.57 × 1.57 cm square frontal area perpendicular to flow Eight 5-layer “chiclets” mini lung final surface area 263 cm2. fiber bundles in plastic housing 3.05 cm length shunting housing attached to plastic end caps 1/8′′-barbed tube fitting for 3/16′′-ID tubing end connectors coated with Teflon tape before prevent leakage device held with two screws secured with silicone eliminate leakage silicone left to dry for 24 h leak tested Filtered deionized water check for leaks no leaks left to dry with filtered air for 24 hExtracorporeal circulation rabbitsNew Zealand white rabbits 3.2–4.2 kg River anesthetized intramuscular injections ketamine (30 mg xylazine (5 mg kg−1) ear catheterized 24G winged catheter intubated 3.0 endotracheal tube anesthetized inhaled isoflurane (2%) ventilated peak inspiratory pressure < 20 cmH2O end expiratory pressure 5 mmHg tidal volume 4–6 ml kg−1 respiratory rate 22–60 breaths per min adjusted normal arterial blood gases PIP Phenylephrine applied intravenously 0.5–5 μg kg−1 maintain blood pressure right carotid arteries cannulated 16-gauge angiocatheter secured silk ties device circuit primed filtered CO2 saline (NaCl 0.9% solumedrol (30 mg kg−1) inhibitor-treated injected 2 mg kg−1 FXII900 2 mM solution ear vein circuit connected ECMO 14-gauge angiocatheter 6′′ pressure tubing internal jugular veins circuit roller pump blood flow set 45 ml per min.extracorporeal circuit rabbits inhibitor-treated group infused 0.075 mg kg−1 min−1 FXII900 ear vein duration study Rabbits treated heparin infused 60 IU h−1 (100 IU ml−1) before connection drip rate adjusted maintain ACT 220 300 s > 320 s decrease infusion rate 12 IU h−1 > 300 decrease 6 IU h−1 > 220 maintain heparin infusion rate 220 > > 200 s increase infusion rate 6 IU h−1 200 > 180 s increase infusion 12 IU h−1 < 180 100 U (1 After four hours animals euthanized by potassium chloride (2 mg kg−1, end 5 ml heparin run through artificial lung circuit removed device washed with saline no clot shedding Saline until Clot volume measured saline difference volume clot volume percent clot within device calculated.Data acquisition extracorporeal circulationPlatelet white blood cell counts hematocrit arterial blood gases ACT, aPTT fibrinopeptide A device resistance bleeding time measured prior circuit 10 30 120 240 min following initiation ECMOplatelet counts syringe 0.05 ml 3.2% sodium citrate 0.45 ml blood volume 0.5 ml centrifuged at 60 × g 10 min 20 μl plasma 20 ml ISOTON® diluent counted Coulter Counter 50 μl aperture tube cells platelets diameters 1.8–5.6 μm white blood cell counts 40 μl blood in 20 ml ISOTON® diluent Six drops ZAP-OGLOBIN lysing solution mixed 2 min white blood cells counted Counter particle above 3.6 μm white blood cell Arterial blood gases measured 0.4 ml blood heparinized syringe arterial blood gas analyzer ACT hematocrit measured 0.5 ml blood ACT Hemochron analyzer glass beads hematocrit capillary centrifugation aPTT measurements blood samples (1.8 ml) collected tubes sodium citrate ml processed to plasma centrifugation 2000 × g 15 min 4 °C samples analyzed platelet counts WBC counts FPA FXIIa levels corrected hemodilution values hematocritinlet outlet pressure blood flow rate measured Biopac system (Aero Camino Goleta, CA pressure transducers (Edwards Lifesciences), resistance calculated with standard R = (Pi − Po)/Q Pi inlet pressure Po outlet pressure Q flow L min−1 bleeding time measured cutting incisions 4–5 mm right left ear Blood removed with gauze every 30 s time until bleeding measured.Statistical mean values independent replicates experiments triplicate more means standard deviations indicated analyses using GraphPad Prism 5) Microsoft Excel software 2016) chi-squared test mice protected against clot formation full occlusion FeCl3 thrombosis model one-tailed student’s t-test difference time until clot formation groups effect expected one direction bleeding time prolongation mouse bleeding experiment.Reporting information experimental design Nature Research Reporting Summary.Supplementary information Peer Review File Reporting Summary | 47.9 | 1.185983 |
10.1038/s41467-020-18917-4 | PMC7603343 | In non-Hermitian systems, fundamental concepts like bandgaps and locality cannot be applied as in Hermitian systems. Here, the authors introduce a class of non-Hermitian critical scenarios where the eigenstates and energies jump discontinuously across a critical point, with anomalous scaling properties | Critical systems represent physical boundaries between different phases of matter and have been intensely studied for their universality and rich physics. Yet, with the rise of non-Hermitian studies, fundamental concepts underpinning critical systems - like band gaps and locality - are increasingly called into question. This work uncovers a new class of criticality where eigenenergies and eigenstates of non-Hermitian lattice systems jump discontinuously across a critical point in the thermodynamic limit, unlike established critical scenarios with spectrum remaining continuous across a transition. Such critical behavior, dubbed the “critical non-Hermitian skin effect”, arises whenever subsystems with dissimilar non-reciprocal accumulations are coupled, however weakly. This indicates, as elaborated with the generalized Brillouin zone approach, that the thermodynamic and zero-coupling limits are not exchangeable, and that even a large system can be qualitatively different from its thermodynamic limit. Examples with anomalous scaling behavior are presented as manifestations of the critical non-Hermitian skin effect in finite-size systems. More spectacularly, topological in-gap modes can even be induced by changing the system size. We provide an explicit proposal for detecting the critical non-Hermitian skin effect in an RLC circuit setup, which also directly carries over to established setups in non-Hermitian optics and mechanics. | IntroductionLying at the boundary between distinct phases, critical systems exhibit a wide range of interesting universal properties from divergent susceptibilities to anomalous scaling behavior. They have broad ramifications in conformal and statistical field theory1–4, Schramm–Loewner evolution5,6, entanglement entropy (EE)7–14, and many other contexts. Recently, concepts crucial to criticalities—like band gaps and localization—have been challenged by studies of non-Hermitian systems15,16 exhibiting exceptional points17–27 or the non-Hermitian skin effect (NHSE), which are characterized by enigmatic bulk-boundary correspondence (BBC) violations, robust-directed amplifications, discontinuous Berry curvature, and anomalous transport behavior28–40.We uncover here a class of criticality, dubbed the “critical non-Hermitian skin effect (CNHSE)”, where the eigenenergies and eigenstates in the thermodynamic limit “jump” between different skin solutions discontinuously across the critical point. This is distinct from previously known phase transitions (Hermitian and non-Hermitian) (Fig. 1), where the eigenenergy spectrum can be continuously interpolated across the two bordering phases. A CNHSE transition, by contrast, is characterized by a discontinuous jump between two different complex spectra along with two different sets of eigenstates. As elaborated below, this behavior appears generically whenever systems of dissimilar NHSE localization lengths are coupled, no matter how weakly. Importantly, at experimentally accessible finite system sizes, the jump smooths out into an interpolation between the two phases in a strongly size-dependent manner, such that the system may exhibit qualitatively different properties, i.e., real vs. complex spectrum or presence/absence of topological modes at different system sizes. Being strongly affected by minute perturbations around the critical point, such behavior may prove useful in sensing applications41,42.Fig. 1Four different types of critical transitions.Hermitian phase transitions (a) are marked by gap closures along the real line. In non-Hermitian cases (2nd to 4th rows, axis labels omitted), spectral phase transitions can take more sophisticated possibilities in the 2D complex energy plane. For instance, the spectral topology can change under line gap closures (b) or shrink continuously to a point and re-emerge in a different topological configuration (c), without the gap ever closing38. The spectrum continuously passes through a gapless or point-like regime in the first three cases, as indicated by the black arrows. The critical non-Hermitian skin effect (d), however, is special in that OBC spectrum in the thermodynamic limit, denoted E∞, jumps discontinuously from one configuration (left), to a different configuration (middle), and to another (right) as certain parameter changes from −ϵ to 0 (critical border), and to ϵ, for an arbitrarily small ϵ, without ever interpolating between the configurations even though the parameter is continuously tuned.ResultsHints of the critical non-Hermitian skin effect from the general Brillouin zoneIn non-Hermitian systems with unbalanced gain and loss, the spectra under periodic boundary conditions (PBCs) and open-boundary conditions (OBCs) can be very different28,29,31,43–45. Indeed, under OBC, eigenstates due to NHSE can exponentially localize at a boundary, in contrast to Bloch states under PBCs. This also explains the possible violation of the BBC, taken for granted in Hermitian settings.The celebrated GBZ formalism aims to restore the BBC through a complex momentum deformation29–31,36–38. Rigorously applicable for bounded but infinitely large systems, it has however been an open question whether the GBZ can still accurately describe finite-size systems. The GBZ of a momentum-space Hamiltonian H(z), z = eik can be derived from its characteristic Laurent polynomial (energy eigenequation)1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f(z,E):= \det [H(z)-E]=0,$$\end{document}f(z,E):=det[H(z)−E]=0,where E is the eigenenergy. While the ordinary BZ is given by the span of allowed real quasimomenta k, the GBZ is defined by the complex analytically continued momentum k → k + iκ(k), with the NHSE inverse decay length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\kappa (k)=-\mathrm{log}\,| z|$$\end{document}κ(k)=−log∣z∣ determined by the smallest complex deformation z → eike−κ(k) such that f(z, E) possesses a pair of zeros zμ, zν satisfying ∣zμ∣ = ∣zν∣ for the same E29,31,38. Due to the double degeneracy of states with equal asymptotic decay rate at these E, there exist a pair of eigenstates ψμ, ψν that can superpose to satisfy OBCs, i.e., zero net amplitude at both boundaries. As such, provided that the characteristic polynomial is not reducible, the OBC spectrum in the thermodynamic limit (denoted as E∞) can be obtained from the PBC spectrum via E(eik) → E(eike−κ(k)), apart from isolated topological modes. Thus it is often claimed that the BBC is “restored” in the GBZ defined by k → k + iκ(k) or, at the operator level, with the surrogate Hamiltonian H(eik) → H(eike−κ(k))38. In general, different E (energy band) solutions can admit different functional forms of κ(k), leading to band-dependent GBZs that have recently also been described with the auxiliary GBZ formalism37. Since eike−κ(k) is generically non-analytic, it represents effectively non-local hopping terms38. As such, the GBZ description challenges the very notion of locality, which is central to critical systems, by effectively “unraveling” the real-space eigenstate accumulation through replacing local hoppings with effectively non-local ones.Due to the robustness of the NHSE, eigenspectra predicted from the GBZ typically are approached rapidly by the exact numerically obtained OBC spectra even for small system sizes (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal{O}}(1{0}^{1})$$\end{document}O(101) sites). In principle, the convergence should be exact in the thermodynamic limit, but in practical computations, floating-point errors ϵ0 are continuously amplified as they propagate across the system. We hence expect accurate numerical spectra only when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L\,<-\mathrm{log}\,({\epsilon }_{0})/\max (\kappa )$$\end{document}L<−log(ϵ0)/max(κ), a condition always checked to be satisfied here to ensure that physical phenomena presented below are not due to numerical errors common in computations with non-reciprocal systems. However, the numerical agreement in eigenspectra between finite-size systems and the GBZ predictions fails spectacularly near a critical point where f(z, E) changes from being reducible to irreducible. To understand the significance of this algebraic property of reducibility, consider a set of coupled irreducible subsystems described by the characteristic polynomial2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f(z,E)={f}_{0}+\mathop{\prod }\limits_{i}{f}_{i}(z,E),$$\end{document}f(z,E)=f0+∏ifi(z,E),where fi(z, E) is the characteristic polynomial of the i-th subsystem, and f0 is a constant that represents the simplest possible form for the subsystem coupling. When f0 = 0, f(z, E) completely factorizes into irreducible polynomials, as expected from a Hamiltonian H(z) that block-diagonalizes into irreducible sectors associated with the individual fi(z, E)’s. In particular, the OBC spectrum of this completely decoupled scenario is derived from the independent κi(k)’s of each subsystem, each determined by zμ, zν from the same subsystem.Yet, a nonzero coupling f0, no matter how small, can have marked physical consequences by hybridizing different sectors of fi significantly. Indeed, such hybridization is inevitable in the thermodynamic limit, with OBC eigenstates formed from superpositions of eigenstates ψμ, ψν from dissimilar subsystems, each corresponding non-Bloch momenta \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-i\mathrm{log}\,{z}_{\mu /\nu }$$\end{document}−ilogzμ/ν. Hence the GBZs, i.e., κ(k)’s of the coupled system, which are defined in the thermodynamic limit, are thus determined by all pairs of ∣zμ∣ = ∣zν∣ not necessarily from the same subsystem. Therefore, the GBZs in the coupled case, no matter how small is f0, can differ from the decoupled GBZs at f0 = 0. That is, the thermodynamic limit and the f0 → 0 limit are not exchangeable. However, since an actual finite physical system cannot possibly possess very different spectrum and band structure upon an arbitrarily small variation in its system parameter, the GBZ picture becomes inapplicable when describing finite systems (small or large) in the presence of CNHSE.Anomalous finite-size scaling from CNHSEFor illustration, we turn to a minimal example of two coupled non-Hermitian 1D Hatano–Nelson chains46 each containing only non-reciprocal (unbalanced) nearest neighbor (NN) hoppings (Fig. 2a). Its Hamiltonian reads as3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H}_{2-{{chain}}}(z)=\left(\begin{array}{ll}{g}_{a}(z)&{t}_{0}\hfill\\ {t}_{0}\hfill&{g}_{b}(z)\end{array}\right),$$\end{document}H2−chain(z)=ga(z)t0t0gb(z),with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${g}_{a}(z)={t}_{a}^{+}z+{t}_{a}^{-}/z+V$$\end{document}ga(z)=ta+z+ta−/z+V and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${g}_{b}(z)={t}_{b}^{+}z+{t}_{b}^{-}/z-V$$\end{document}gb(z)=tb+z+tb−/z−V, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{a/b}^{\pm }={t}_{1}\pm {\delta }_{a/b}$$\end{document}ta/b±=t1±δa/b being the forward/backward hopping of chains a and b. This model can be also realized with a reciprocal system with the NHSE in a certain parameter regime (Supplementary Note 1). When t0 = 0, the two chains are decoupled, and the characteristic polynomial is reducible as f(z, E) = [ga(z) − E][gb(z) − E]. Each factor fa/b(z, E) = ga/b(z) − E determines the skin eigensolutions of its respective chain. However, even an infinitesimal coupling t0 ≠ 0 generically makes f(z, E) irreducible, providing that the two chains correspond to different GBZs (see “Methods” section). Specifically, consider the simple case of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{a}^{+}={t}_{b}^{-}=1$$\end{document}ta+=tb−=1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{a}^{-}={t}_{b}^{+}=0$$\end{document}ta−=tb+=0. Without couplings (t0 = 0), the two chains under OBC respectively yields a Jordan-block Hamiltonian matrix in real space, with the spectrum given by E = ±V. Because the eigenstates of the decoupled chains are exclusively localized at the first or the last site, their GBZs collapse45. By contrast, for any t0 ≠ 0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f(z,E)={E}^{2}-E(z+{z}^{-1})+(z+V)({z}^{-1}-V)-{t}_{0}^{2}$$\end{document}f(z,E)=E2−E(z+z−1)+(z+V)(z−1−V)−t02 is irreducible (here \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-{t}_{0}^{2}={f}_{0}$$\end{document}−t02=f0 from Eq. (2)), insofar as the eigenenergy roots \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E=\cos k\pm \sqrt{{t}_{0}^{2}+{(V+i\sin k)}^{2}}$$\end{document}E=cosk±t02+(V+isink)2 are no longer Laurent polynomials in z = eik that can be separately interpreted as de facto subsystems with local hoppings. In fact, in higher degree polynomials, an algebraic expression for z may not even exist as implied by the Abel–Ruffini theorem. Importantly, the corresponding OBC E∞ spectrum and the GBZ for t0 ≠ 0 are now qualitatively different. As derived in the Methods section, setting ∣za∣ = ∣zb∣ gives OBC spectrum (in the thermodynamic limit): \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{\infty }^{2}=\frac{1-{\eta }^{2}}{1+{\eta }^{2}}+{V}^{2}+{t}_{0}^{2}\pm 2\sqrt{{t}_{0}^{2}-{\eta }^{2}+{\eta }^{2}{t}_{0}^{2}}/(1+{\eta }^{2})$$\end{document}E∞2=1−η21+η2+V2+t02±2t02−η2+η2t02/(1+η2), with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\eta \in {\mathbb{R}}$$\end{document}η∈R. Clearly, even one now takes the t0 → 0 limit, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{\infty }^{2}$$\end{document}E∞2 only simplifies to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{\infty }^{2}\to {V}^{2}+\frac{1\pm i\eta }{1\mp i\eta }$$\end{document}E∞2→V2+1±iη1∓iη, which is not the above-mentioned OBC spectrum of the two decoupled chains. Likewise, the t0 → 0 limit of the coupled GBZ, which can be shown to be the locus of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$z=\pm \sqrt{{V}^{2}+{e}^{i\theta }}-V$$\end{document}z=±V2+eiθ−V and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$z=1/[V\pm \sqrt{{V}^{2}+{e}^{i\theta }}]$$\end{document}z=1/[V±V2+eiθ], θ ∈ [0, 2π], has nothing in common with the collapsed GBZs of the decoupled case.Fig. 2Critical non-Hermitian skin effect and finite-size scaling in the model of two coupled Hatano–Nelson chains.a The two-chain model [Eq. (3)] with hopping asymmetry in chains a, b denoted by δa/b, and on-site energy offset ±V. A small inter-chain t0 can cause significant coupling when δa ≠ δb. b Open-boundary spectra (black dots) and eigenstate profiles (insets) with N = 10, 20, and 80 unit cells and the coupling parameter t0 = 0.01, showing very different spectral behavior at different system sizes N. At small N ≈ 10, coupling effects are negligible, with the spectrum coinciding with the real-value open-boundary E∞ spectrum (green) in the decoupled thermodynamic limit. As N increases, the spectrum gradually approaches the open-boundary E∞ spectrum (red) for the coupled thermodynamic limit, with hybridization becoming sharper. c The generalized Brillouin zones of the two chains in the decoupled limit (yellow and green), and that of the two bands with t0 = 0.01 (red circles, numerically obtained at N = 80). The Brillouin zone with ∣z∣ = 1 is given by the gray dash circle. Other parameters are t1 = 0.75, δa = − δb = 0.25, and V = 0.5.This paradoxical singular behavior of GBZs leads to anomalous scaling behavior in finite-size systems that are more relevant to experimental setups. The discontinuous critical transition illustrated above becomes a smooth crossover between the different OBC E∞ solutions. As the size N of a coupled system is varied, its physical OBC spectrum interpolates between the decoupled and coupled OBC E∞ solutions. As illustrated in Fig. 2b for the two-chain model Eq. (3) at small coupling t0 = 0.01 (with t1 = 0.75 and δa = −δb = 0.25 for well-defined skin modes), the OBC spectrum (black dots) changes markedly from N = 10 to 80 unit cells. For small N = 10, the spectrum approximates the OBC E∞ (green) for t0 = 0 (which lies on the real line), with the associated GBZs given by two perfect circles in the complex plane (Fig. 2c). At large N = 80, the spectrum converges toward the true OBC E∞ (red curve) with nonzero coupling, where the associated respective GBZs of the two bands (also shown in Fig. 2c) are much different from the two circles as decoupled GBZs. Indeed, the eigenstates for N = 10 are almost entirely decoupled across the two chains, while those for N = 80 are maximally coupled/decoupled depending on whether they approach the red/green E∞ curves. In the intermediate N = 20 case, the OBC spectrum lies far between the two E∞’s, and cannot be characterized by their GBZs. The size-dependent behavior of the OBC spectrum is further elaborated through a spectral-flow study31 in Supplementary Note 2.Let us now explain the above-observed marked size-dependent spectra via the competition between dissimilarly accumulated skin modes and the couplings across them. The general conditions for such are unveiled in the “Methods” section. In our model (Eq. 3), the inverse decay lengths in chains a, b are given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\kappa }_{a/b}=\frac{1}{2}\mathrm{log}\,({t}_{a/b}^{+}/{t}_{a/b}^{-})$$\end{document}κa/b=12log(ta/b+/ta/b−), which will be dissimilar as long as δa ≠ δb. After performing a similarity transformation that rescales each site j by a factor of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{j{\kappa }_{b}}$$\end{document}ejκb, chain b becomes reciprocal with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\kappa }_{b}^{\prime}=0$$\end{document}κb′=0 while chain a has a rescaled inverse decay length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\kappa }_{a}^{\prime}={\kappa }_{a}-{\kappa }_{b}$$\end{document}κa′=κa−κb. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\kappa }_{a}^{\prime}\,\ne\,0$$\end{document}κa′≠0, chain a always possesses exponentially growing skin modes scaling like \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\,{e}^{{\kappa }_{a}^{\prime}N}$$\end{document}eκa′N at one end. As such, the coupling t0, even if being extremely small, still affects the spectrum and eigenstates markedly as the system size N increases, as further elaborated in the “Methods” section.Scale-free exponential wavefunctionsA hallmark of conventional critical systems is scale-free power-law behavior, particularly in the wavefunctions. Interestingly, such scale-free behavior can also be found in the exponentially decaying wavefunctions, i.e., skin modes. Shown in Fig. 3a are the profiles of the slowest decaying eigenstates ψ(x) of H2-chain at different system sizes N = 20, 40, 60, and 80, with the horizontal axis normalized by N. These featured eigenstates belong to the top of the central black ring in Fig. 2b, with their distance from the coupled OBC E∞ ring (red) decreasing as ~N−1. Unlike usual exponentially decaying wavefunctions with fixed spatial decay length, here ∣ψ(x)∣ ~ e−κx with κ ~ N−1 (Fig. 3b), such that the overall profile ψ(x) has no fixed length scale. Such unique scale-free eigenmodes result from the slow critical migration of the eigenstates between E∞ solutions (Fig. 2a, inset).Fig. 3Anomalous scaling behaviors of the critical non-Hermitian skin effect.a Scale-free open-boundary skin eigenstate of the largest \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{Im}}[E]$$\end{document}Im[E] eigenenergy of H2-chain at system sizes N = 20, 40, 60, and 80 (red, purple, blue, green, respectively). Its rescaled profile, despite decaying exponentially rather than power-law, remains invariant across different N. This scale invariance persists in the N > 20 regime, and is due to the N−1 decay (dashed line) of the inverse skin depth (red dots), as plotted in the inset. Parameters follow Fig. 2’s, except with t0 = 10−3. b Entanglement entropy S (blue) of a half-filled open-boundary H2-chain at odd system sizes N, with real-space cut at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lfloor \frac{N}{2}\rfloor$$\end{document}⌊N2⌋ (the floor function of N/2) and parameters t1 = 0.58, V = 1, t0 = 0.4, and δa = − δb = 0.25. It saturates near zero in the gapless decoupled small N regime, but scales like \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim \frac{1}{3}\mathrm{log}\,N$$\end{document}~13logN (yellow) in the gapless coupled large N regime.Anomalous correlations and entanglement entropyThe OBC spectra can be gapped for certain system sizes where EE obeys an area law scaling, and then become gapless at other sizes where the EE scaling is replaced by a logarithmic dependence on system size47. This indicates that the CNHSE can lead to an unusual scaling behavior of the EE. Consider for instance the OBC H2-chain (Eq. 3) with parameters chosen to gap out the OBC spectrum at small system sizes N. With all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{Re}}[E]\,<\,0$$\end{document}Re[E]<0 states occupied by spinless-free Fermions, the real-space entanglement entropy S (blue curve in Fig. 3b) exhibits a crossover from the decoupled gapped regime at N ≤ 5 where it remains a constant due to the size-independent area of boundaries (two ends), to the gapless regime N > 20 where it approaches the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{1}{3}\mathrm{log}\,N$$\end{document}13logN behavior of a gapless system (yellow line). In generic CNHSE scenarios with multiple competing OBC E∞ loci, S can scale differently at different system size regimes, choices of fillings, and entanglement cuts, challenging the notion of single well-defined scaling behavior. As further shown in Supplementary Note 3, the two-Fermion correlator 〈ψ(1)ψ(x)〉 characterizing the EE also crossovers from rapid exponential decay at small N to 1/x power-law decay at large N. Remarkably, the probability of finding another Fermion nearby generally increases drastically when the system is enlarged (with filling fraction maintained).Size-dependent topological modesTopological modes are usually associated with bulk invariants in the thermodynamic limit, with finite-size effects having a diminishing role in the face of topological robustness. This intuition is not necessarily true in non-Hermitian systems, as hinted from ref. 34, where an infinitesimal instability can cause a Z2 topological transition in the thermodynamic limit34. Remarkably, the CNHSE here can cause topological edge modes to appear only at certain system size regimes. Consider replacing the non-reciprocal intra-chain couplings of our H2-chain model with inter-chain couplings with non-reciprocity ±δab between adjacent unit cells (Fig. 4a), as described by the following CNHSE Su–Schrieffer–Heeger (SSH) model48:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H}_{{\rm{CNHSE}}-{\rm{SSH}}}(z)={h}_{y}(z){\sigma }_{y}+{h}_{z}(z){\sigma }_{z}+{h}_{0}(z){\mathbb{I}},$$\end{document}HCNHSE−SSH(z)=hy(z)σy+hz(z)σz+h0(z)I,where hy(z) = iδab(z + 1/z), hz(z) = V + δ−(z − 1/z), and h0(z) = t1(z + 1/z) + δ+(z − 1/z), with δ± = (δa ± δb)/2. HCNHSE-SSH is so named because interestingly, at δ− = δab, it can be transformed via a basis rotation σz → σx into an extended SSH model with non-reciprocal inter-cell couplings given by ±2δ− and a uniform non-reciprocal next-nearest neighbor hopping given by t1 ± δ+ (Supplementary Note 4), which is known to possess a Z-type topologically nontrivial phase.Fig. 4Topological transition induced by changing the system’s size.a Sketch of the HCNHSE-SSH model with cross inter-chain non-reciprocal couplings ±δab. b Open-boundary spectra (black dots) at N = 20, 30, and 40 unit cells and coupling δab = 0.5 × 10−3. The main part of the spectrum in each case behaves similarly as the model in Fig. 2b, except for a pair of topological edge states blue emerging within the point gap at zero energy. The open-boundary E∞ spectrum is given by green and red colors in the decoupled and coupled thermodynamic limits respectively. Other parameters are δa = −δb = 0.5, t1 = 0.75, and V = 1.2. c
κ solutions (red, blue, green, and yellow surfaces) of f(z, E) = 0 as a function of the complex energy, with the same parameters in b. Intersecting regions (green and red dotted lines) give the open-boundary skin solutions of the system in the thermodynamic limit. Among them, green lines correspond to the skin solutions of two decoupled chains at δab = 0. The solutions of red curves emerge at a small but nonzero δab, and the skin solutions of the weakly coupled system are given by the intersecting regions with the smallest ∣κ∣, i.e., the red loop in the center and green lines at the two ends with large and small \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{Re}}[E]$$\end{document}Re[E]. d Emergence of in-gap degenerate modes as a function of δab/δa and N with δa = − δb = 0.5, t1 = 0.75, and V = 1.2, with the plotted boundary scaling logarithmically with N.When δab = 0, the system is decoupled into two Hatano–Nelson chains, which must be topologically trivial. The OBC spectrum E∞ in the decoupled case and the associated inverse decay length κ are shown in Fig. 4b, c (green curves), with positive/negative κ corresponding to skin modes accumulating population at opposite boundaries. Also shown in Fig. 4b, c (red curves) are E∞ in the coupled case and the corresponding κ for the hybridized skin modes. With small N = 20 unit cells in Fig. 4b, the finite-size OBC spectrum (gray dots) qualitatively agrees with the decoupled E∞ (green), with a real-valued gap at E = 0 along the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{Im}}[E]=0$$\end{document}Im[E]=0 axis (inset). Upon the size increase to N = 30 and then to N = 40, such a gap first closes on the complex plane and then develops into a point gap with two zero-energy degenerate modes lying in its center. The Z-type topological origin of such in-gap modes is also carefully verified in Supplementary Note 5. The gap closure and then the emergence of in-gap topological modes resemble the typical behavior of a topological phase transition. Yet, here it is an intriguing size-induced effect. Further, the emergence of in-gap modes only requires exponentially weaker inter-chain coupling (i.e., smaller δab/δa) for larger N, as shown in the “phase” diagram shown in Fig. 4d.Proposal for circuit demonstrationThe CNHSE is most simply realized when the two subsystems have equal and opposite κ values, since the system is then net reciprocal. Consider the RLC circuit as illustrated in Fig. 5. It is governed by Kirchhoff’s law I = JV, where I, V are the input currents and potentials at nodes 1A, 1B, 2A, 2B, ... and J is the circuit Laplacian given, at AC frequency ω = (LC)−1/2, by5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J(k)= \, (2i\omega C\sin k){\sigma }_{z}+{r}^{-1}({\mathbb{I}}-{\sigma }_{x})+2{R}^{-1}(1-\cos k){\mathbb{I}}\\ \to \, \left(\begin{array}{cc}2\omega C{e}^{ik}+\Delta (k)&-{r}^{-1}\hfill\\ -{r}^{-1}\hfill&2\omega C{e}^{-ik}+\Delta (k)\end{array}\right),$$\end{document}J(k)=(2iωCsink)σz+r−1(I−σx)+2R−1(1−cosk)I→2ωCeik+Δ(k)−r−1−r−12ωCe−ik+Δ(k),where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta (k)={r}^{-1}+2{R}^{-1}-2({R}^{-1}+\omega C)\cos k$$\end{document}Δ(k)=r−1+2R−1−2(R−1+ωC)cosk. The second line was obtained via a unitary basis transformation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{y}\to {\tilde{\sigma }}_{y}=U{\sigma }_{y}{U}^{-1}={\sigma }_{z}$$\end{document}σy→σ~y=UσyU−1=σz that transforms the circuit Laplacian into a form similar to Eq. (3) (with V = 0), which is susceptible to the CNHSE. In this rotated basis, we evidently have two effective chains coupled by −r−1, each with unbalanced gain/loss couplings that give rise to equal and opposite NHSE. Note that RLC components are all reciprocal and cannot realize the non-reciprocal effective chains individually. However, with the basis transformation given above, the two effective chains become entangled in a way such that they are net reciprocal and hence easy to realize with RLC components, as illustrated in Fig. 5.Fig. 5A circuit lattice realizing the critical non-Hermitian skin effect.The Laplacian of it is given by Eq. (5) with N unit cells with open-boundary conditions. The nodes A and B of unit cells 1, 2, 3, … are labeled accordingly. Grounded R resistors are added to make the boundary nodes experience the same outdegree.One can experimentally demonstrate the CNHSE by building copies of the circuit with different numbers of unit cells N (or alternatively by adjusting its length with appropriately placed switches), and mapping their Laplacian (admittance) spectra via established approaches49–52. For instance, one can systematically connect a current source I to each node α, one node at a time (the current exits through the ground), and measure the resultant electrical potentials Vβ,α at each node β. The spectrum of J is given by the inverse of the eigenvalues of the matrix Vβ,α/I. In the presence of the CNHSE, the spectral plots for different N should qualitatively resemble that in Fig. 2b, since Eq. (5) is of the form of Eq. (3). Due to the robustness of the skin effect, component uncertainties in an actual experiment should minimally affect the resultant spectrum, as verified by simulation results presented in Supplementary Note 6. In particular, the circuit Laplacian spectra, which manifest the CNHSE, are almost undisturbed by uncertainty tolerances of up to 20%.DiscussionIn mathematical terms, the CNHSE arises when the energy eigenequation exhibits an algebraic singularity that leads to inequivalent auxiliary GBZs across the transition. The CNHSE heralds a whole class of discontinuous critical phase transitions with rich anomalous scaling behavior, challenging traditional associations of criticality with scale-free behavior. Even a vanishingly small coupling between dissimilar skin modes can be consequential as the system size increases. This insight is much relevant to sensing and switching applications. Beyond our two-chain models, there are other scenarios that can engineer coupling between subsystems of dissimilar NHSE length scales and hence yield CNHSE (e.g., see “Methods” section for a discussion of general two-band models). In particular, we anticipate fruitful investigations in various experimentally feasible settings such as electric circuits53–56, cold atom systems57,58, photonic quantum walks59, and metamaterials41,60, all of which are investigated with finite-size systems and hence highly relevant to the CNHSE.MethodsDiscontinuous transition of GBZ in two-chain modelsThe discontinuous transition induced by an infinitesimal transverse coupling in the thermodynamic limit, and also the crossover in a finite system, exist only when the two decoupled chains have different κ of their OBC skin solutions. To see this, we consider a general two-chain model described by Hamiltonian6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(z)=\left(\begin{array}{cc}{g}_{a}(z)+{V}_{a}&{t}_{0}\hfill\\ {t}_{0}\hfill&{g}_{b}(z)+{V}_{b}\\ \end{array}\right),$$\end{document}h(z)=ga(z)+Vat0t0gb(z)+Vb,where ga,b(z) only contains terms with nonzero order of z. When decoupled, the two chains correspond to the polynomials ga,b(z) + Va,b, respectively, and possess the same κ solutions when and only when gb(z) = cga(z), with c a nonzero coefficient. When a nonzero transverse coupling t0 is introduced, the characteristic polynomial of the two-chain system can always be written in the form of7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{rcl}{P}_{c}(z)&=&({g}_{a}(z)+{V}_{a}-E)({g}_{b}(z)+{V}_{b}-E)-{t}_{0}^{2}\\ &=&(c{g}_{a}(z)-A)({g}_{a}(z)-B),\hfill\end{array}$$\end{document}Pc(z)=(ga(z)+Va−E)(gb(z)+Vb−E)−t02=(cga(z)−A)(ga(z)−B),where A, B are two coefficients determined by other parameters. Therefore for two chains with the same κ solutions, a transverse coupling t0 only modifies the energy offset between them, without inducing a transition of skin solutions.Nevertheless, the above factorization does not hold when the coupling term t0 is z-dependent, corresponding to inter-chain couplings between different unit cells. Under this condition, Pc(z) cannot be factorized into two sub-polynomials of ga(z) and gb(z) = cga(z), meaning that the skin solution is changed for the system.GBZ solutions E∞ for the two-chain modelFor analytic tractability, we consider the case of Eq. 3 of the main text with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{a}^{+}={t}_{b}^{-}=1$$\end{document}ta+=tb−=1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{a}^{-}={t}_{b}^{+}=0$$\end{document}ta−=tb+=0 (i.e., t1 = δa = −δb = 0.5), but nonzero b and V. We obtain8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H}_{2-{\rm{chain}}}(z)=\left(\begin{array}{cc}z+V&{t}_{0}\hfill\\ {t}_{0}\hfill&1/z-V\end{array}\right),$$\end{document}H2−chain(z)=z+Vt0t01/z−V,with the characteristic polynomial given by9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{rcl}f(z,E)&=&{E}^{2}-E({z}^{-1}+z)+[(z+V)({z}^{-1}-V)-{t}_{0}^{2}]\\ &=&\hskip-1.8pc\frac{V-E}{z}-z(V+E)+[{E}^{2}-{V}^{2}-{t}_{0}^{2}+1].\end{array}$$\end{document}f(z,E)=E2−E(z−1+z)+[(z+V)(z−1−V)−t02]=V−Ez−z(V+E)+[E2−V2−t02+1].To find the GBZ solutions E∞ for comparison with the actual OBC solutions, we solve for roots ∣z+∣ = ∣z−∣ of f(z, E) = 0 (with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Sigma ={E}^{2}-{V}^{2}-{t}_{0}^{2}+1$$\end{document}Σ=E2−V2−t02+1):10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{rcl}{z}_{\pm }&=&\frac{\left(\Sigma \pm \sqrt{{\Sigma }^{2}+4({V}^{2}-{E}^{2})}\right)}{2(V+E)}\\ &=&\frac{\Sigma \pm \sqrt{{(\Sigma -2)}^{2}-4{t}_{0}^{2}}}{2(V+E)}.\end{array}$$\end{document}z±=Σ±Σ2+4(V2−E2)2(V+E)=Σ±(Σ−2)2−4t022(V+E).For ∣z+∣ = ∣z−∣ to hold, the square root quantity must differ from Σ by a complex argument of π/238 i.e.,11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt{{(\Sigma -2)}^{2}-4{t}_{0}^{2}}=i\eta \Sigma,$$\end{document}(Σ−2)2−4t02=iηΣ,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\eta \in {\mathbb{R}}$$\end{document}η∈R. Simplifying, we obtain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Sigma =\frac{2}{1+{\eta }^{2}}\left(1\pm \sqrt{{t}_{0}^{2}+{\eta }^{2}({t}_{0}^{2}-1)}\right)$$\end{document}Σ=21+η21±t02+η2(t02−1) or, in terms of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}^{2}\to {E}_{\infty }^{2}$$\end{document}E2→E∞2,12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{\infty }^{2}=\frac{1-{\eta }^{2}\pm 2\sqrt{{t}_{0}^{2}-{\eta }^{2}+{\eta }^{2}{t}_{0}^{2}}}{1+{\eta }^{2}}+{V}^{2}+{t}_{0}^{2}.$$\end{document}E∞2=1−η2±2t02−η2+η2t021+η2+V2+t02.as in the main text, with η tracing out a one-parameter continuous spectrum. The GBZ can be numerically obtained by substituting Eq. (12) into the expression for z± in Eq. (10) with E = E∞. From that, we obtain two momentum values \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{\pm }={\rm{Re}}[-i\mathrm{log}\,{z}_{\pm }]$$\end{document}k±=Re[−ilogz±] with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\kappa ({k}_{+})=\kappa ({k}_{-})=-\mathrm{log}\,| {z}_{+}| =-\mathrm{log}\,| {z}_{-}|$$\end{document}κ(k+)=κ(k−)=−log∣z+∣=−log∣z−∣ as the inverse length scales. Note however that because of the proximity to the t0 = 0 critical point, this value of κ(k±) is significantly different from the actual inverse OBC skin depth for a large range of finite system sizes.Dissimilar skin modes in general two-band modelsIn a more general picture, the CNHSE and the size-dependent variation may exist when different parts of the system have dissimilar skin accumulation of eigenmodes. In the two-chain model, we mainly consider regimes with small inter-chain couplings, thus the two energy bands (overlapped or connected in most cases) with dissimilar skin modes are mostly given by one of the two chains respectively. To unveil the condition of having dissimilar skin modes in a general two-band system, we consider an arbitrary two-band system described by a non-Bloch Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H(z)={h}_{0}(z){\mathbb{I}}+{\sum }_{n = 1,2,3}{h}_{n}(z){\sigma }_{n}$$\end{document}H(z)=h0(z)I+∑n=1,2,3hn(z)σn, with z = eike−κ(k), and κ(k) a complex deformation of momentum k describing the NHSE. Its characteristic polynomial is given by13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f(z,E)={[E-{h}_{0}(z)]}^{2}-P(z)=0,$$\end{document}f(z,E)=[E−h0(z)]2−P(z)=0,with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(z)={\sum }_{n = 1,2,3}{h}_{n}^{2}(z)$$\end{document}P(z)=∑n=1,2,3hn2(z). NHSE can be described by a GBZ where the solutions of f(z, E) = 0 satisfy Eα(zμ) = Eα(zν) with ∣zμ∣ = ∣zν∣ and α = ± the band index, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\kappa (k)=-\mathrm{log}\,| z|$$\end{document}κ(k)=−log∣z∣ gives the inverse decay length. Conventionally, NHSE is studied mostly for a system with only nonzero h0(z) (i.e., a one-band model) or P(z) (e.g., the non-reciprocal SSH model), where the zeros of f(z, E) lead to E± = h0(z) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{\pm }^{2}=P(z)$$\end{document}E±2=P(z), respectively. In either case, we can see that the two bands of E± must have the same inverse skin localization depth κ(k), as Eα(zμ) = Eα(zν) must be satisfied for α = ± with the same zμ,ν. To have dissimilar skin modes for the two bands, h0(z) and P(z) must both be non-vanishing, and possess different skin solutions. That is, although h0(zμ) = h0(zν) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P({z}_{\mu ^{\prime} })={h}_{0}({z}_{\nu ^{\prime} })$$\end{document}P(zμ′)=h0(zν′) can still be satisfied with ∣zμ∣ = ∣zν∣ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {z}_{\mu ^{\prime} }| =| {z}_{\nu ^{\prime} }|$$\end{document}∣zμ′∣=∣zν′∣, we cannot have \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${z}_{\mu }=z^{\prime}$$\end{document}zμ=z′ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${z}_{\nu }=z^{\prime}$$\end{document}zν=z′ at the same time, otherwise the same κ(k) can be obtained for the two bands.Competition between skin localization and inter-chain couplingAs mentioned in the main text, if two coupling chains have inverse NHSE decay lengths (non-Hermitian localization length scales) κa, κb, a change of basis will bring their coupling to be effective between a chain with no NHSE, and another with an effective skin depth κa − κb. Since that entails exponentially growing skin modes scaling like \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{({\kappa }_{a}-{\kappa }_{b})N}$$\end{document}e(κa−κb)N at one end, we expect the effect of even an infinitesimally small inter-chain coupling t0 to scale exponentially with N, and eventually change the OBC spectrum substantially.Consider increasing the inter-chain coupling t0 in our two-chain model (Eq. 3 of main text) from zero. At sufficiently small t0, we have two practically independent OBC Hatano–Nelson chains with real spectra. Their infinitesimal coupling only shifts their eigenenergies slightly along the real line. But at a critical t0 = tc, the OBC spectrum is rendered complex as one or more pairs of eigenenergies coalesce and repel along in the imaginary direction. Shown in Fig. 6a is the inverse exponential scaling of the critical t0 = tc with N. We observe that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{\mathrm{c}}^{2}{e}^{({\kappa }_{a}-{\kappa }_{b})N} \sim {\mathcal{O}}(1)$$\end{document}tc2e(κa−κb)N~O(1), in agreement with the intuitive expectation that tc should scale inverse exponentially with N because the effect of t0 scales exponentially with N. Yet, the fact that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{\mathrm{c}}^{2} \sim {e}^{-({\kappa }_{a}-{\kappa }_{b})N}$$\end{document}tc2~e−(κa−κb)N signifies that the CNHSE is fundamentally a non-perturbative effect since it differs from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{\mathrm{c}} \sim {e}^{-({\kappa }_{a}-{\kappa }_{b})N}$$\end{document}tc~e−(κa−κb)N as expected from first-order perturbation theory with left and right eigenstates that are oppositely exponentially localized spatially.Fig. 6Inverse exponential scaling of the critical bare coupling t0 = tc required for the open-boundary spectrum of H2-chain to make the transition from real to complex.a, b
tc Versus the system’s size N and effective skin depth κa − κb, respectively. The numerical data (blue) fits very well with the predicted scaling law \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{\mathrm{c}} \sim {e}^{-({\kappa }_{a}-{\kappa }_{b})N/2}$$\end{document}tc~e−(κa−κb)N/2 (dashed lines) with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\kappa }_{a}-{\kappa }_{b}=\mathrm{log}\,2$$\end{document}κa−κb=log2 in a and N = 40 in b. Unless specified in the figure, the parameters are t1 = 0.75, δa = − δb = 0.25 as in Fig. 2 of the main text. In b, κa − κb is obtained from Eq. (14) with δa = −δb varying from 0.1 to 0.4.The scaling behavior of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{({\kappa }_{a}-{\kappa }_{b})N}$$\end{document}e(κa−κb)N also suggests that increasing N has similar consequences as increasing the non-reciprocity in the system, the strength of which is reflected by the absolute value of (κa − κb). Therefore it is also expected that the CNHSE shall emerge when we enhance the non-reciprocity but fix N. In Fig. 6b, we show the inverse exponential scaling of the critical t0 = tc with κa − κb, where the inverse NHSE decay lengths are given by14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{{\kappa }_{a,b}}=\sqrt{\frac{{t}_{1}+{\delta }_{a,b}}{{t}_{1}-{\delta }_{a,b}}}$$\end{document}eκa,b=t1+δa,bt1−δa,bfor the two decoupled chains. The scaling behavior versus κa − κb further confirms that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${t}_{\mathrm{c}}^{2} \sim {e}^{-({\kappa }_{a}-{\kappa }_{b})N}$$\end{document}tc2~e−(κa−κb)N.Supplementary informationSupplementary InformationPeer Review File | nature communications | [
"Article"
] | [
"Condensed-matter physics",
"Phase transitions and critical phenomena"
] | between phases critical systems exhibit properties divergent susceptibilities to anomalous scaling behavior ramifications in conformal statistical field Schramm–Loewner evolution5,6 entanglement entropy other contexts concepts band gaps challenged by studies non-Hermitian systems15 non skin effect by bulk-boundary correspondence violations robust-directed amplifications discontinuous Berry curvature anomalous transport criticality “critical non-Hermitian skin effect (CNHSE eigenenergies eigenstates limit “jump” between skin solutions discontinuously across critical point distinct from phase transitions (Hermitian non spectrum interpolated across phases CNHSE transition by discontinuous jump between complex spectra eigenstates behavior appears whenever systems of dissimilar NHSE localization lengths coupled at finite system sizes jump smooths into interpolation between phases size-dependent system different properties affected by minute perturbations around critical point behavior useful in sensing applications41.Fig. 1Four types of critical transitions.Hermitian phase transitions marked by gap closures along real linenon-Hermitian cases (2nd to 4th rows axis labels spectral phase transitions sophisticated 2D complex energy plane spectral topology change under line gap closures shrink point re-emerge different configuration without gap spectrum passes through gapless point-like regime first three cases indicated black arrows critical non-Hermitian skin effect special OBC spectrum thermodynamic limit E∞ jumps configuration parameter changes from −ε to 0 small without interpolating configurations parameter tuned critical non-Hermitian skin effect Brillouin non-Hermitian systems unbalanced gain loss spectra under periodic open-boundary conditions (OBCs different28,31,43–45 OBC eigenstates due NHSE localize at boundary contrast Bloch states under PBCs explains violation BBC Hermitian settings GBZ formalism restore BBC complex momentum deformation29–31 applicable bounded large systems open question GBZ describe finite-size systemsGBZ momentum-space Hamiltonian H(z), z = eik derived from characteristic Laurent polynomial (energy eigenequation[12pt]{minimal{amsmath-69pt(z,E): [H(z)-E]=0}f(z[H(z)−E]=0 E eigenenergy ordinary BZ given span quasimomenta k GBZ defined by complex continued momentum k → k + iκ(k), NHSE inverse decay length[12pt]{minimal{amsmath-69pt}{document} (k\mathrm{log}{document}κ(k)=−log∣z∣ determined by complex deformation z → eike−κ(k) f(z, E) possesses zeros zμ, zν satisfying ∣zμ∣ = ∣zν∣ E29,31,38 double degeneracy states equal asymptotic decay rate E eigenstates ψμ, ψν superpose satisfy OBCszero net amplitude at boundaries characteristic polynomial not reducible OBC spectrum in thermodynamic limit E∞) obtained from PBC spectrum via E(eik) → E(eike−κ(k)), isolated topological modes BBC “restored” in GBZ by k → k + iκ(k) or surrogate Hamiltonian H(eik) → H(eike−κ(k))38 different E (energy band solutions admit different functional forms of κ(k), to band-dependent GBZs described with auxiliary GBZ formalism37 eike−κ(k) non-analytic represents non-local hopping terms38 GBZ description challenges locality real-space eigenstate accumulation replacing local hoppings with non-local ones robustness NHSE eigenspectra predicted from GBZ approached rapidly by exact obtained OBC spectra for small system sizes convergence should be exact in thermodynamic limit floating-point errors ε0 amplified across systemexpect accurate numerical spectra[12pt]{minimal{amsmath{wasysym\oddsidemargin-69pt}{document}$$L\mathrm{log}(\epsilon_{0})/\max{document}L<−log(ε0)/max(κ), condition checked ensure physical phenomena due numerical errors non-reciprocal systems numerical agreement eigenspectra between finite-size systems GBZ predictions fails critical point f(z, E) changes reducible to irreducible understand reducibility consider coupled irreducible subsystems characteristic polynomial2[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin-69pt}{document}$$f(z,E)={f}_{0}+\mathop{\prod }\limits_{i}{f}_{i}(z,E),{document}f(z,E)=f0+∏ifi(z fi(z, E) characteristic polynomial of i-th subsystem f0 constant simplest form subsystem couplingf0 = 0 f(z, E) factorizes into irreducible polynomials Hamiltonian H(z) block-diagonalizes irreducible sectors fi(z, E)’s OBC spectrum decoupled scenario derived from independent κi(k)’s each subsystem determined by zμ, zν same subsystem nonzero coupling f0 sectors fi hybridization inevitable in thermodynamic limit OBC eigenstates formed from superpositions eigenstates ψμ, ψν from subsystems non-Bloch momenta{minimal{amsmath-69pt GBZs κ(k)’s coupled system defined thermodynamic limit determined by all pairs of ∣zμ∣ = ∣zν∣ not same subsystem GBZs in coupled case f0 differ from decoupled GBZs at f0 = 0 thermodynamic limit and f0 → 0 limit not exchangeablefinite system different spectrum band structure small variation system parameter GBZ picture inapplicable describing finite systems CNHSE.Anomalous finite-size scaling from CNHSEFor two non-Hermitian 1D Hatano–Nelson chains46 non-reciprocal (unbalanced nearest hoppings (Fig.Hamiltonian reads\documentclass[12pt]{minimal\usepackage{amsmath{wasysym\oddsidemargin{-69pt}{document}${H}_{2-{{chain(z)=\left{array}{ll}{g}_{a}(z){t}_{0} {t}_{0}\hfill{g}_{b}(z\end{array}\end{document}H2−chain(z)=ga(z)t0t0gb[12pt]{minimal}{amsmath}}\oddsidemargin}{-69pt}{document}${g}_{a}(z)={t}_{a}{+}z+{t}-/z+V\end{document}ga(z)=ta+z+ta−/z+V[12pt]{minimal}\usepackage{amsmath}{wasysym}\setlength{\oddsidemargin}{-69pt}{document}$${g}_{b}(z)={t}+}z+{t/z-V{document}gb(z)=tb+z+tb−/z−V[12pt{minimal{amsmath\oddsidemargin-69pt}{document{t{a/b{1}\pm\delta/b}ta/b±=t1±δa/b forward/backward hopping chains a b model reciprocal system NHSE parameter regime (Supplementary Note 1) t0 = 0 two chains decoupled characteristic polynomial f(z, E) = [ga(z) − E][gb(z) − E] factor fa/b(z E) = ga/b(z) − E determines skin eigensolutions chaininfinitesimal coupling t0 ≠ 0 makes f(z, E) irreducible two chains correspond to different GBZs (see “Methods” section). consider case of \documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts{upgreek-69pt-}=1}ta+=tb−=1[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek-69pt{document+}=0{document}ta−=tb+=0. Without couplings (t0 = 0), two chains under OBC Jordan-block Hamiltonian matrix in real space spectrum given by E = ±V. eigenstates of decoupled chains localized at first or last site GBZs collapse45.t0 ≠ 0[12pt]{minimal{amsmath{wasysym\oddsidemargin-69pt}{document}(z,E)={E}^{2}-E(z+{z{-1})+(z+V)(-V)-{t_{0}^{2}\end{document}f(z,E)=E2−E(z+z−1)+(z+V)(z−1−V)−t02 irreducible[12pt]{minimal}{amsmath{upgreek}\oddsidemargin}{-69pt}\begin{document}$-{t}_{0}^{2}={f}_{0}\end{document}−t02=f0 Eq.eigenenergy roots[12pt{minimal\usepackage{amsmath\oddsidemargin-69pt}{document}$E=\cos k\pm\sqrt{{t}{0}^{2}+(V+i\sin k)}^{2}}\end{document}E=cosk±t02+(V+isink)2 polynomials z = eik subsystems local hoppings higher degree polynomials algebraic expression z may not exist Abel–Ruffini theorem OBC E∞ spectrum GBZ for t0 ≠ 0 qualitatively different.Methods section setting ∣za∣ = ∣zb∣ gives OBC spectrum thermodynamic[12pt]{minimal}{amsmath{wasysym\oddsidemargin-69pt}{document}{E}\infty{2}{1-{\eta{2}}+\eta+{V{2}+{t{0^{2}{2-\eta+{2}}/(1+{\eta{2}\end{document}E∞2=1−η21+η2+V2+t02±2t02−η2+η2t02/(1+η2)[12pt]{minimal}{amsmath}{wasysym}{amssymb}{amsbsy}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$$\eta\mathbb{R}}\end{document}η∈R.one takes t0 → 0 limit[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek{\oddsidemargin}{-69pt}{document}{E}\infty\end{document}E∞2 simplifies[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}{E}{\infty{2} {V}^{2}+\frac{1\pm i\eta\end{document}E∞2→V2+1±iη1iη not OBC spectrum of two decoupled chainst0 → 0 limit of coupled GBZ locus of\documentclass[12pt]{minimal}{amsmath{upgreek-69pt$z=\pm{2+{i\theta-V}z=±V2+eiθ−V[12pt]{minimal}{upgreek{-69pt}$z=1/[V\pm{2{i\theta{document}z=1/[V±V2+eiθ] θ ∈ [0, 2π] with collapsed GBZs decoupled case.Fig. non-Hermitian skin effect finite-size scaling in model two coupled Hatano–Nelson chains two-chain model [Eq. (3)] hopping asymmetry in chains a, b by δa/b, on-site energy offset ±V. small inter-chain t0 significant coupling when δa ≠ δb.Open-boundary spectra eigenstate profiles N = 10 20 80 cells coupling parameter t0 = 0.01 different behavior at system sizes small N ≈ 10 coupling effects negligible spectrum with open-boundary E∞ spectrum decoupled limit N increases spectrum approaches-boundary E∞ hybridization sharper Brillouin zones chains decoupled limit bands t0 = N = Brillouin zone with ∣z∣ = 1 gray dash circle parameters t1 = 0.75 δa = δb = 0.25 V = 0.5 behavior GBZs leads anomalous scaling finite-size systems discontinuous transition crossover between OBC E∞ solutions size N system varied OBC spectrum interpolates between decoupled coupled Fig. small coupling t0 = 0.01 t1 = 0.75 δa = 0.25 OBC spectrum changes from N = 10 to 80 cells small N = 10 spectrum approximates OBC E∞ (green) for t0 = 0 GBZs two circles complex planeN = 80 spectrum converges toward OBC E∞ (red curve) with nonzero coupling GBZs two bands Fig. 2c different from circles decoupled GBZs eigenstates for N = 10 decoupled across chains N = 80 maximally coupled/decoupled depending red/green E∞ curves intermediate N = 20 OBC spectrum between two E∞’s characterized by GBZs size-dependent behavior OBC spectrum elaborated through spectral-flow study31 Supplementary Note explain size-dependent spectra via competition between dissimilarly accumulated skin modes couplings conditions in “Methods” section model (Eq. 3) inverse decay lengths in chains a, b given by\documentclass[12pt{minimal{amsmath-69pt{document/b={1}{2}\mathrm{log/b=12log(ta/b+/ta/b−), dissimilar as δa ≠ δbsimilarity transformation rescales site j factor[12pt]{minimal}{amsmath}{wasysym}{mathrsfs{upgreek}{\oddsidemargin}{-69pt}{document}{e\kappa\end{document}ejκb chain b becomes reciprocal[12pt]{minimal}{amsmath{wasysym}}{upgreek}{\oddsidemargin}{-69pt}{document}\kappa{b\prime}=0\end{document}κb′=0 chain a rescaled inverse decay length[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}\kappa\prime}\kappa\kappa{b\end{document}κa′=κa−κb.\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin}{-69pt}{document}$${\kappa\prime}\end{document}κa′≠0 chain a growing skin modes[12pt{minimal{amsmath{wasysym{mathrsfs{upgreek{\oddsidemargin}{-69pt}{document}$$\kappa\prime}N\end{document}eκa′N one end coupling t0 small affects spectrum eigenstates as system size N increases “Methods” section.Scale-free exponential wavefunctionsA hallmark conventional critical systems scale-free power-law behavior wavefunctions decaying wavefunctions skin modes Fig. 3a profiles slowest decaying eigenstates ψ(x) of H2-chain at system sizes N = 20 40 60 80 horizontal axis normalized by N eigenstates belong top central black ring in Fig.distance from OBC E∞ ring decreasing ~N−1 Unlike exponentially decaying wavefunctions fixed decay length ∣ψ(x ~ e−κx κ ~ N−1 (Fig. profile ψ(x) no fixed length scale unique scale-free eigenmodes from slow migration eigenstates between E∞ solutions (Fig. 2a. 3Anomalous scaling behaviors non-Hermitian skin effect Scale-free open-boundary skin eigenstate largest H2-chain system sizes N = 20 40 60 80 green rescaled profile decaying exponentially remains invariant across N scale invariance persists N > 20 regime due to N−1 decay inverse skin (red Parameters follow Fig. 2’s except t0 = 10−3.Entanglement entropy (blue half-filled open-boundary H2-chain odd system sizes N real-space cut[12pt]{minimal\usepackage{amsmath{wasysym{upgreek-69pt\lfloor}N2 floor function N/2 parameters t1 = 0.58 V = 1 t0 = 0.4 δa = − δb = 0.25 saturates near zero gapless decoupled small N regime scales like[12pt]{minimal}{amsmath-69pt}\sim{1}{3\mathrm{log}~13logN (yellow gapless coupled large N regime.Anomalous correlations entanglement OBC spectra gapped certain system sizes EE area law scaling gapless other sizes EE scaling replaced logarithmic dependence system size47 CNHSE unusual scaling behavior EE OBC H2-chain (Eq. 3) parameters gap OBC spectrum small system sizes N\documentclass[12pt]{minimal{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}[E states spinless-free Fermions real-space entanglement entropy S (blue curve Fig. 3b) crossover from decoupled gapped regime N ≤ 5 to gapless regime N > 20[12pt]{minimal{amsmath\oddsidemargin}{-69pt}}{1}{3}\mathrm{log}}13logN behavior gapless system (yellow CNHSE scenarios multiple OBC E∞ loci S scale differently at system size regimes choices fillings entanglement cuts challenging single scaling behavior Supplementary Note 3 two-Fermion correlator 〈ψ(1)ψ(x)〉 EE crossovers from rapid exponential decay at small N to 1/x power-law decay at large Nprobability finding Fermion increases when system enlarged filling maintained).Size-dependent topological associated with bulk invariants thermodynamic limit finite-size effects topological robustness not true in non-Hermitian systems infinitesimal instability Z2 topological transition CNHSE topological edge modes at certain system size regimes replacing non-reciprocal intra-chain couplings H2-chain with inter-chain couplings non-reciprocity cells (Fig.CNHSE Su–Schrieffer–Heeger (SSH) model48:4[12pt{amsmath-69pt(z=(z}HCNHSE−SSH(z)=hy(z)σy+hz(z)σz+h0(z hy(z = iδab(z + 1 hz(z = V + δ−(z − 1 h0(z = t1(z + 1/z) + δ+(z − 1 δ± = (δa ± δb)/2 HCNHSE-SSH δ− = δab transformed rotation σz → σx extended SSH model non inter-cell couplings ±2δ− uniform non-reciprocal next-nearest neighbor hopping t1 ± δ+ 4) Z-type topologically nontrivial phase.Fig. 4Topological transition changing size Sketch HCNHSE-SSH model cross inter-chain non-reciprocal couplings ±δabOpen-boundary spectra dots at N = 20 30 40 cells coupling δab = 0.5 × 10−3 spectrum Fig. 2b topological edge states blue point gap at zero energy open-boundary E∞ spectrum green red colors decoupled coupled thermodynamic limits parameters δa = −δb = 0.5 t1 = 0.75 V = 1.2. solutions (red blue green yellow surfaces f(z, E) = 0 complex energy same parameters Intersecting regions (green red lines give open-boundary skin solutions green lines skin solutions decoupled chains at δab = 0 solutions red curves small nonzero δab skin solutions weakly coupled system intersecting regions smallest ∣κ∣ red loop center green lines two ends in-gap degenerate modes function of δab/δa N δa = − δb = 0.5, t1 = 0.75 V = 1.2 plotted boundary scaling with N.δab = 0 system decoupled into two Hatano–Nelson chains topologically trivial OBC spectrum E∞ decoupled case inverse decay length κ Fig. 4b c (green positive/negative κ skin modes opposite boundaries Fig. 4b c (red curves E∞ coupled case κ hybridized skin modes small N = 20 unit cells Fig. 4b finite-size OBC spectrum agrees with decoupled E∞ real-valued gap at E = 0 size increase to N = 30 N = 40 gap closes complex plane develops into point gap two zero-energy degenerate modes center Z-type topological origin in-gap modes verified in Supplementary Note 5. gap closure emergence of in-gap topological modes resemble topological phase transition size-induced effect emergence in-gap modes requires weaker inter-chain coupling smaller δab/δa) for larger N diagram Fig. 4d.Proposal circuit CNHSE realized two subsystems equal opposite κ values system net reciprocal RLC circuit Fig.governed by Kirchhoff’s law I = JV I, V input currents potentials at nodes 1A 1B 2A J circuit Laplacian AC frequency ω = (LC)−1/2\documentclass[12pt]{amsmath\oddsidemargin-69pt}{document}$$J(k)= \ (2i\omega C\sin k){\sigma{z}+{r}^{-1}\sigma{x})+2{R}^{-1}(1-\cos k)\mathbb{array\omega C{e^{ik+\Delta (k{r}^{-1}\Delta\end{array}{document}J(k)=(2iωCsink)σz+r−1(I−σx)+2R−1(1−cosk)I→2ωCeik+Δ(k)−r−1−r−12ωCe−ik+Δ[12pt]{minimal}{amsmath\oddsidemargin}{-69pt(k)={r{-1}+2{R}-2(+\omega C)}Δ(k)=r−1+2R−1−2(R−1+ωC)cosk second line obtained unitary basis transformation[12pt{minimal{amsmath\oddsidemargin-69pt}\sigma\sigma}σy→σ~y=UσyU−1=σz transforms circuit Laplacian form similar Eq. (3) V = susceptible to CNHSE rotated basis two effective chains coupled by −r−1 unbalanced gain/loss couplings equal opposite NHSE RLC components reciprocal realize non-reciprocal effective chains basis transformation two effective chains net reciprocal easy to realize with RLC components Fig. 5.Fig. 5A circuit lattice realizing non-Hermitian skin effect Laplacian given by Eq. (5) N unit cells open-boundary conditions nodes A B of unit cells 1 2 3 labeledR resistors added nodes experience same outdegree demonstrate CNHSE building copies circuit different unit cells N adjusting length mapping Laplacian spectra connect current source I to each node α current exits measure electrical potentials Vβ,α each node β spectrum J inverse eigenvalues matrix Vβ,α/I CNHSE spectral plots N resemble Fig. 2b Eq. (5) form Eq. (3) skin effect component uncertainties minimally affect spectrum verified simulation results Supplementary Note 6. circuit Laplacian spectra manifest CNHSE undisturbed by uncertainty up to 20% CNHSE arises when energy eigenequation algebraic singularity leads inequivalent auxiliary GBZs across transition CNHSE heralds discontinuous critical phase transitions anomalous scaling behavior challenging associations small coupling between dissimilar skin modes consequential as system size increases relevant to sensing switching applications other scenarios coupling between subsystems dissimilar NHSE length scales yield CNHSEsee “Methods” section discussion two-band anticipate investigations in settings electric circuits53–56 cold atom systems57,58 photonic quantum walks59 metamaterials41,60 investigated finite-size systems relevant to CNHSE.MethodsDiscontinuous transition of GBZ in two-chain discontinuous transition infinitesimal transverse coupling thermodynamic limit crossover finite system exist when two decoupled chains have different κ OBC skin solutions consider two-chain model Hamiltonian6\documentclass[12pt{minimal\usepackage{amsmath-69pt$h(z)=\left+}h(z)=ga(z)+Vat0t0gb(z)+Vb ga,b(z) contains terms with nonzero order z decoupled two chains correspond to polynomials ga,b(z) + Va,b possess same κ solutions when gb(z) = cga(z), with c nonzero coefficientnonzero transverse coupling t0 introduced characteristic polynomial two-chain system written\documentclass[12pt{minimal}\usepackage{amsmath{wasysym{upgreek}\oddsidemargin-69pt}\begin{document}{array}{rcl{P}_{c}(z)_{a}(z)+{V}_{a}-E)(-{2{a(z-A)(}(z-B),\end{array}}Pc(z)=(ga(z)+Va−E)(gb(z)+Vb−E)−t02=(cga(z)−A)(ga(z A, B coefficients determined parameters two chains same κ solutions transverse coupling t0 modifies energy offset transition skin solutions factorization hold coupling term t0 z-dependent inter-chain couplings cells Pc(z) factorized sub-polynomials ga(z) gb(z) = cga(z), skin solution changed system.GBZ solutions E∞ two-chain consider case Eq.main text[12pt]{minimal{amsmath{upgreek\oddsidemargin}{-69pt}{document}{t}_{a}^{+}={t}_{b}^{-}=1{document}ta+=tb−=1[12pt]{minimal}{amsmath{wasysym}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}{t}_{a}^{-}={t}_{b}^{+}=0{document}ta−=tb+=0 t1 = δa = −δb = nonzero b V.obtain8\documentclass[12pt{amsmath\oddsidemargin-69pt}\begin{document}${H}{chain(z)=\left(\begin{array}z+V{t{0}-V\end{array\end{document}H2−chain(z)=z+Vt0t01/z−V characteristic polynomial\documentclass[12pt]{minimal{amsmath}\oddsidemargin-69pt}\begin{document}{array}{rcl}f(z,E){E}^{2}-E({z{-1}+z)+[(z+V)(-V)-{t}_{0}^{2}\hskip-1.8pc\frac{V-E}{z}-z(V+E)+[{E}^{2}-{V}^{2}-{t}_{0}^{2}+1].{array(z,E)=E2−E+z)]=V−Ez−z(V+E)[E2−V2−t02+1find GBZ solutions comparison OBC solutions solve roots ∣z+∣ = ∣z−∣ f(z E) = 0[12pt]{minimal\usepackage{amsmath-69pt}\begin{document}\Sigma ={E}^{2}-{V}^{2}-{t}_{0}^{2}+1\end{document}Σ=E2−V2−t02+1):10\documentclass[12pt]{minimal}{amsmath}-69pt}{document}{array}{rcl}{z}_{\pm\Sigma \pm \sqrt{{^{2}+4({V}^{2}-{E}^{2})}\right{2(V+E)\Sigma -2)}^{2}-4{t}_{0}^{2}}}{2(V+E)}{array}{document}z±=Σ±Σ2+4(V2−E2)2(V+E)=Σ±(Σ−2)2−4t022(V+E).For ∣z+∣ = ∣z−∣ square root quantity differ from Σ complex argument π/238\documentclass[12pt]{minimal{amsmath-69pt}{document}\sqrt{{(\Sigma -2)}^{2}-4{t}_{0}^{2}}=i\eta \Sigma\end{document}(Σ−2)2−4t02=iηΣ[12pt]{minimal}{amsmath}{-69pt}{document}$\eta\mathbb{R}}\end{document}η∈R.obtain[12pt]{minimal{amsmath{-69pt}{document}\Sigma =\frac{2}{1+{\eta }^{2}}\left(1 }_{0}^{2}+{\eta }^{2}(}-1)}\right\end{document}Σ=21+η21±t02+η2(t02−1)[12pt]{minimal}{amsmath}\oddsidemargin}{-69pt}\begin{document}$${E}^{2} {E}_{\infty }^{2}\end{document}E2→E∞2,12\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek}{\oddsidemargin}{-69pt}\begin{document}$${E}{\infty{2}=\frac{1-{\eta }^{2} 2\sqrt{{t}_{0}^{2}-{\eta }^{2}+{^{2}{t{2{document}E∞2=1−η2±2t02−η2+η2t021+η2+V2 main text η one-parameter continuous spectrum GBZ obtained substituting Eq. (12) z± Eq (10) E = E∞.obtain two momentum values\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin}{-69pt}\begin{document}{k}_{\pm\rm{Re}}\mathrm{log}{z\pm\end{document}k±=Re[−ilogz±][12pt]{minimal}{amsmath}{wasysym}}{upgreek}\oddsidemargin}{-69pt}{document}\kappa ({k}_{+})=\kappa ({k}_{-})=-\mathrm{log} {z}{+{log\end{document}κ(k+)=κ(k−)=−log∣z+∣=−log∣z−∣ inverse length scales proximity t0 = 0 critical point value κ(k±) different from inverse OBC skin depth finite system sizes.Dissimilar skin modes two-band CNHSE size-dependent variation exist dissimilar skin accumulation eigenmodestwo-chain model regimes small inter-chain couplings two energy bands dissimilar skin modes given by one dissimilar skin modes two-band system consider arbitrary two-band system non-Bloch Hamiltonian\documentclass[12pt]{minimal}\usepackage{amsmath{upgreek\oddsidemargin-69pt}\begin{document}$H(z)={h}_{0}(z)\mathbb{I}}+\sum{n = 1,2,3}(z\sigma{document}H(z)=h0(z)I+∑n=1,2,3hn(z)σn z = eike−κ(k), κ(k complex deformation momentum k NHSEcharacteristic polynomial given by13\documentclass[12pt]{minimal{amsmath\oddsidemargin-69pt}{document}$f(z,E)=[E-{h{0}(z{2}-P(z)=0\end{document}f(z,E)=[E−h0(z)]2−P(z)=0[12pt]{minimal{amsmath{wasysym\oddsidemargin-69pt}{document}$P(z)={n = 1,2,3}{h}{2}(z)\end{document}P(z)=∑n=1,2,3hn2(z). NHSE described by GBZ solutions of f(z, E) = 0 satisfy Eα(zμ) = Eα(zν) ∣zμ∣ = ∣zν∣ α = ± band index[12pt]{minimal}{amsmath\oddsidemargin}{-69pt}{document}$\kappa (k)=-\mathrm{log}{document}κ(k) gives inverse decay length NHSE studied for system nonzero h0(z) one-band model or P(z) non-reciprocal SSH zeros of f(z, E) lead to E± = h0(z)[12pt{minimal{amsmath\oddsidemargin-69pt}{document}{2}=P(z){document}E±2=P(z), two bands of E± same inverse skin localization depth κ Eα(zμ) = Eα(zν) satisfied for α = ± same zμ,ν dissimilar skin modes bands h0(z) and P(z) non-vanishing different skin solutionsh0(zμ) = h0(zν)\documentclass[12pt]{minimal}\usepackage{amsmath}{upgreek\setlength\oddsidemargin}{-69pt}\begin{document}$$P({z}_={h}_{0}{z{document}P(zμ′)=h0(zν′) with ∣zμ∣ = ∣zν∣[12pt]{minimal}\usepackage{amsmath}{wasysym{upgreek}\oddsidemargin}{-69pt}{document}$$| {z}_{\mu\prime {z}_{\end{document}∣zμ′∣=∣zν′∣\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}}}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$${z}=z{\prime\end{document}zμ=z′[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin{-69pt}{document}${z}=z\prime{document}zν=z′ same κ(k) for two bands.Competition between skin localization inter-chain two coupling chains inverse NHSE decay lengths κa κb change of basis coupling effective between chain no NHSE effective skin depth κa − κb growing skin modes scaling\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin{-69pt{document}$${e}^{\kappa-\kappa_{b})N{document}e(κa−κb)N one end small inter-chain coupling t0 to scale with N change OBC spectrum increasing inter-chain coupling t0 in two-chain model (Eq. 3 from zero small t0 two independent OBC Hatano–Nelson chains with real spectra infinitesimal coupling shifts eigenenergies slightly real linecritical t0 = tc OBC spectrum complex eigenenergies coalesce repel imaginary direction Fig. 6a inverse exponential scaling critical t0 = tc with N\documentclass[12pt]{minimal}\usepackage{amsmath\oddsidemargin{-69pt}{document}$${t}\mathrm{c}}^{2}{e}\kappa-{b})N}\mathcal{O}}(1)\end{document}tc2e(κa−κb)N~O(1) tc scale inverse exponentially with N effect t0 scales exponentially with N\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}{document}$${t}\mathrm{c}}{2}tc2~e−(κa−κb)N CNHSE non-perturbative effect differs from[12pt]{minimal{amsmath{wasysym{upgreek}\oddsidemargin}{-69pt}{document}$${t}\mathrm{c}}^{{b})N\end{document}tc~e−(κa−κb)N first-order perturbation theory left right eigenstates oppositely localized spatially.Fig. 6Inverse exponential scaling of critical bare coupling t0 = tc for open-boundary spectrum of H2-chain transition from real to complex.a b tc Versus system’s size N effective skin depth κa − κbnumerical data (blue fits predicted scaling law[12pt]{minimal\usepackage{amsmath{wasysym\oddsidemargin-69pt}{document}\mathrm{c}}\kappa)N/2\end{document}tc(κa−κb)N/2 (dashed lines[12pt]{minimal}{amsmath{wasysym{upgreek\setlength\oddsidemargin-69pt}{document}\kappa{b}\mathrm{log}\end{document}κa−κb=log2 a N = 40 in b parameters t1 = 0.75, δa = − δb = 0.25 Fig. 2 main text b κa − κb obtained from Eq. (14) δa = −δb varying 0.1 to 0.4.scaling behavior of\documentclass[12pt]{minimal{amsmath{wasysym{mathrsfs{upgreek-69pt}{document}\kappa{a-\kappa{b})N(κa−κb)N suggests increasing N similar consequences increasing non-reciprocity strength reflected by absolute value of (κa − κb). expected CNHSE emerge enhance non-reciprocity fix N Fig. 6b inverse exponential scaling of critical t0 = tc with κa − κb inverse NHSE decay lengths given\documentclass[12pt]{minimal{amsmath{wasysym{mathrsfs{upgreek-69pt}{document}\kappa{a,b}}=\sqrt{\frac{{t}_{1}+\delta{a{document}eκa,b=t1+δa,bt1−δa,bfor two decoupled chainsscaling behavior versus κa − κb confirms[12pt{minimal\usepackage{amsmath{wasysym\oddsidemargin{-69pt}{document}$${t}\mathrm{c}}^{2} {e}^{\kappa{a}{b})N\end{document}tc2~e−(κa−κb)N.Supplementary Review File | 48.8 | 1.49403 |
10.1038/s41467-021-20895-0 | PMC7878916 | The exciton–phonon coupling (EXPC) affects the opto-electronic properties of atomically thin semiconductors. Here, the authors develop two-dimensional micro-spectroscopy to determine the EXPC of monolayer MoSe2. | Single-layer transition metal dichalcogenides are at the center of an ever increasing research effort both in terms of fundamental physics and applications. Exciton–phonon coupling plays a key role in determining the (opto)electronic properties of these materials. However, the exciton–phonon coupling strength has not been measured at room temperature. Here, we use two-dimensional micro-spectroscopy to determine exciton–phonon coupling of single-layer MoSe2. We detect beating signals as a function of waiting time induced by the coupling between A excitons and A′1 optical phonons. Analysis of beating maps combined with simulations provides the exciton–phonon coupling. We get a Huang–Rhys factor ~1, larger than in most other inorganic semiconductor nanostructures. Our technique offers a unique tool to measure exciton–phonon coupling also in other heterogeneous semiconducting systems, with a spatial resolution ~260 nm, and provides design-relevant parameters for the development of optoelectronic devices. | IntroductionLayered materials (LMs)1–4, such as single-layer (1L) transition metal dichalcogenides (1L-TMDs)5–8, are a promising platform for new photonic and optoelectronic devices. Bulk semiconducting TMDs consist of covalently bound layers of type MX2, where M is a metal (e.g., Mo, W) and X is a chalcogen atom (e.g., S, Se), held together by van der Waals interactions3. When they are exfoliated, or grown as 1L, quantum confinement induces an indirect-to-direct bandgap transition5,6. The reduced dimensionality is also responsible for high exciton binding energies (hundreds of meV)7,8, making 1L-TMDs excellent candidates for optoelectronic devices at room temperature (RT)2.Exciton–phonon coupling (EXPC) plays a key role in determining the temperature-dependent optoelectronic and transport properties of 1L-TMDs9–11. It is responsible for, e.g., non-radiative exciton decay9,10,12, limiting the fluorescence quantum yield11, the formation of dark-exciton phonon replicas13, and it mediates spin-flip processes, thus decreasing the lifetime of spin/valley-polarized charge carriers14. For temperature < 100 K, the interaction between excitons and acoustic phonons induces linewidth broadening and dominates the excitonic resonances of 1L-TMDs9,15,16. The situation is different for higher temperature. Ref. 17 suggested that the coupling between excitons and optical phonons induces sidebands in the absorption spectrum of 1L-MoSe2 at RT. Yet the spectral signature of this coupling is obscured by inhomogeneous broadening17. The presence of EXPC was inferred from resonant Raman scattering18,19, as well as time-resolved transmission measurements20,21, where the A′1 optical phonon mode was observed to couple with the A excitonic resonance. While the exciton energies can be obtained from photoluminescence (PL) and those of (ground state) phonons from Raman measurements, this does not fully characterize the system. To obtain the complete Hamiltonian, one also requires the displacement along the phonon coordinate of the exciton-state potential energy minimum versus the ground state. This displacement is the EXPC strength, and determines how strongly phonons are excited upon an optical transition to the exciton state. To the best of our knowledge, the EXPC strength was not measured for any 1L-TMDs at RT, because overtone bands of the optical phonon mode were not detectable18–21. We determine this missing quantity in the present work.Optical four-wave-mixing experiments in semiconductors provide access to coherent dynamics of excitons10,22–24. In photon echo experiments the polarization state of incident photons (circular or linear) allows one to uncover different mechanisms behind the signal formation25,26. Different levels can be distinguished by the polarization dependence26–28. Two-dimensional electronic spectroscopy (2DES) is a powerful tool to analyze light-induced coherences in molecular systems29–32 and semiconductors33,34. It is a generalized version of transient absorption spectroscopy, providing frequency resolution not only for the probe step, but also for the pump35–39. Coherent broadband excitation of several quantum energy levels leads to wave packets that may be detected as oscillations of specific peaks in the 2d maps as a function of waiting time T30,40. Analysis of frequency, decay time, and the position of such oscillations allows one to explore the underlying energy structure and the coupling mechanism leading to level splittings30,40,41. Ref. 41 theoretically proposed that an additional Fourier transform along T and cutting the resulting 3d spectrum at certain beating frequencies could lead to 2d maps that are sensitive to the EXPC strength.It is challenging to apply 2DES on micro-scale samples or heterogeneous materials with localized structural domains on a μm lateral scale, because the standard phase-matching geometry requires the exciting beams to be non-collinear with respect to each other37. This cannot be realized simultaneously when focusing with a high-numerical-aperture (NA) objective, in which all incident light arrives at the sample from the same solid angle. As a result, if one chooses to employ phase matching, this necessarily requires longer focal lengths, leading to larger spot sizes and unwanted averaging over different spatial regions or crystal orientations42. Instead, one can also select the signal by phase cycling43–45, which relies on detecting population-based signals as a function of inter-pulse phase combinations43,45,46. The collinear geometry accessible with phase cycling enables 2d micro-spectroscopy, i.e., the combination of 2DES with fluorescence microscopy, to gain additional spatial resolution42,47.Here, we develop 2d micro-spectroscopy to resolve the spectral features of the phonon sidebands in 1L-MoSe2 at RT and determine the EXPC. We observe oscillations in 2d maps that arise from the coupling between the A′1 optical phonon mode and the A exciton. From comparison with simulated 2d beating maps, we deduce a Huang–Rhys factor S ~ 1. This implies a large EXPC strength for 1L-MoSe2, when compared with other inorganic semiconductor nanostructures, such as CdSe quantum dots48 and rods49, ZnSe quantum dots50, and single-wall carbon nanotubes51, most of which fall in the range ~ 0–0.552, providing design-relevant information for the development of photonic devices based on 1L-MoSe2. Our method can be extended to other 1L-TMDs and LMs and to other important semiconducting systems, for which the ~260-nm spatial resolution of micro-spectroscopy is required, e.g., single-wall carbon nanotubes, LM heterostructures, layered perovskites, bulk heterojunctions, or microcavities with embedded semiconductors.Results and discussionThe experimental setup is sketched in Fig. 1a. A Ti:sapphire oscillator emits 12-fs pulses at 80 MHz repetition rate. A pulse shaper generates a collinear four-pulse sequence, focused by a high-NA = 1.4 objective, so that a spatial resolution ~ 260 nm is achieved. To image the sample, the laser focus is mapped by a piezo scanning stage, and the PL signal is detected by an avalanche photodiode (APD). For the 2d map, the PL intensity is detected while scanning a first coherence time τ (delay between the first two pulses), a waiting time T (delay between second and third pulse), and a second coherence time t (delay between third and fourth pulse, Fig. 1a). Fourier transformation over τ and t results in a 2d map for every T (see Methods for data acquisition details). Nonlinear signals are obtained by systematically scanning through a number of discrete phase steps for each pulse and for each pulse-delay combination. Rephasing and nonrephasing signals are retrieved as linear superpositions of differently phase-modulated data45.Fig. 1Overview of setup and the sample.a Fluorescence-detected 2d micro-spectroscopy setup. Four collinear laser pulses are generated by a pulse shaper with controllable inter-pulse time delays (τ, T, t) and phases (φi, i = 1, 2, 3, 4) and focused by a high-NA objective (Obj). The position of the sample is controlled by a piezo scanning stage (PSS). The dichroic mirror (DM) under the objective transmits the excitation beam (red) and reflects the PL signal (yellow). A long-pass filter (LP) is used to block the remaining excitation beam. The PL signal is detected by an avalanche photodiode (APD). b PL map obtained with the setup of panel a. c PL and d Raman spectrum for 514 nm excitation. The peak in the PL spectrum corresponds to the A exciton. The Raman spectrum shows the out-of-plane A′1 mode ~ 241 cm−1, and the in-plane E′ mode ~ 288 cm−1.We investigate mechanically exfoliated 1L-MoSe2 on a 200-μm fused silica substrate (see Methods for details). Figure 1b is a PL map, taken with the setup of Fig. 1a, for a representative sample. 1L-MoSe2 has a direct bandgap at the K point of the Brillouin zone leading to two excitonic transitions A and B ~1.57 and 1.75 eV53. The PL spectrum (Fig. 1c) shows a single peak ~1.57 eV, due to the radiative recombination of A excitons54. The signal of the trion is much weaker than that of the neutral exciton at RT24,55. In our experiment we detect predominantly the neutral exciton. This is confirmed by the linear PL spectrum of our sample (Fig. 1c), in which the main peak is located at a position that agrees with that found for neutral excitons54. The Raman spectrum measured at 514 nm (Fig. 1d) shows the out-of-plane A′1 mode ~241 cm−1 with full width at half maximum (FWHM) ~ 4 cm−1, and the in-plane E′ mode ~288 cm−1 (FWHM ~ 6 cm−1). Both PL and Raman spectra confirm that the sample is 1L-MoSe218,54.The rephasing 2d maps in the region around the A exciton are shown in Fig. 2a for various T, while the nonrephasing and absorptive 2d maps are in Supplementary Figs. 1 and 5, respectively. The peak linewidth along the diagonal direction of the rephasing map (orange double arrow in upper left panel) is plotted versus T in Fig. 2b. Closer analysis of the systematic variation of this linewidth with T (Supplementary Note 2) indicates that there are three components along the diagonal, marked with purple crosses in the lower left panel of Fig. 2a, whose amplitudes oscillate, but not in phase. Thus, when T ~ 1500 fs, the amplitude of the middle component is much higher than the other two, minimizing the effective diagonal linewidth (minimum in Fig. 2b). The measured 2d maps capture the fourth-order nonlinear optical response, as sixth-order contributions are negligible (see Supplementary Note 3).Fig. 2Beating signal in the rephasing 2d maps.a Rephasing 2d maps at different T, normalized to the maximum absolute value of the real part of the map at T = 500 fs. b Diagonal linewidth (FWHM, indicated by the orange double arrow at T = 50 fs in panel a) versus T. The error bars depict 95% confidence bounds from fitting the diagonal slices by a Gaussian function. c Amplitude evolution (green diamonds) of one pixel (marked by the green diamond at T = 50 fs in panel a) and fit (solid green curve). The error bars are evaluated by calculating the fluctuations within a region containing background noise (see Supplementary Note 4).We then extract the amplitude evolution of a typical pixel (marked by the green diamond in the 2d map at 50 fs) as a function of T (Fig. 2c). The number of points is restricted, due to the long measurement time (26 h for one point). A long-lived (>2 ps) oscillation with amplitude above the noise level is observed. The reproducibility of the data is confirmed by a second measurement for the same T in Supplementary Note 4.We now analyze the origin of the oscillations in the 2d maps with the goal to deduce the EXPC strength. Previous experiments reported that the trion signal in 1L-MoSe2, located ~0.03 eV below the neutral exciton peak55, gradually dies out both in PL and absorption, when the temperature increases from 15 to 295 K, while the signal intensity of neutral excitons remains nearly unchanged24,55. Thus, the signal of the trion is much weaker than the neutral exciton at RT and in our experiment we detect predominantly the neutral exciton. This implies that wave packets involving trions can be excluded as a source of the long-lived (>2 ps) RT oscillations in Fig. 2c. Biexciton signals can be excluded in our 2d measurements due to their thermal dissociation at RT and cancellation of excited-state absorption pathways in fluorescence-detected 2d spectroscopy (see Supplementary Note 5). Vibrational wave packets were reported at RT in Refs. 20,21, with a dephasing time ~4.5 ps for 1L- and few-layer WSe220 and ~1.7 ps for 1L-MoS221. Therefore, EXPC can explain the oscillations in our 2d maps. We extract the phonon energy from a fit (Fig. 2c, solid green curve) and obtain, even for our undersampled data (less than one data point for each oscillation period as a result of a compromise arising from finite available data acquisition time), an oscillation period ~136 ± 2 fs (see Supplementary Note 6 for the fitting procedure). This corresponds to an energy splitting between the participating states ~30.4 ± 0.4 meV, matching the optical A′1 phonon mode’s energy ~29.9 meV, i.e., 241 cm−1, as measured in the Raman spectrum of Fig. 1d.We define the EXPC strength using the Huang–Rhys factor, S, in the framework of the Franck–Condon coupling model56 (see Supplementary Note 7 for a definition of S), with the minimum number of states needed to describe the observed data (Fig. 3a). The model of Fig. 3a delivers 3 transition energies, as observed experimentally (purple crosses in Fig. 2a). We assign component 1 (with the lowest energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar \omega _1$$\end{document}ħω1) to the transition between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_1} \right\rangle$$\end{document}g1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_0} \right\rangle$$\end{document}e0 (blue in Fig. 3), component 2 (with a higher energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar \omega _2$$\end{document}ħω2) to the two degenerate transitions between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_0} \right\rangle$$\end{document}g0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_0} \right\rangle$$\end{document}e0 and between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_1} \right\rangle$$\end{document}g1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_1} \right\rangle$$\end{document}e1 (black and green, respectively), and component 3 (with the highest energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar \omega _3$$\end{document}ħω3) to the transition between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_0} \right\rangle$$\end{document}g0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_1} \right\rangle$$\end{document}e1 (red).Fig. 3Analysis of beating signals.a Schematic diagram of displaced harmonic oscillators (Franck–Condon coupling model56) with two vibrational levels (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_0} \right\rangle$$\end{document}g0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_1} \right\rangle$$\end{document}g1) in the electronic ground state, and two in the electronic excited state (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_0} \right\rangle$$\end{document}e0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_1} \right\rangle$$\end{document}e1). The horizontal shift between the two potential minima, d, characterizes the EXPC strength. Transitions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_0} \right\rangle$$\end{document}g0–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_0} \right\rangle$$\end{document}e0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_0} \right\rangle$$\end{document}g0–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_1} \right\rangle$$\end{document}e1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_1} \right\rangle$$\end{document}g1–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_0} \right\rangle$$\end{document}e0, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_1} \right\rangle$$\end{document}g1–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_1} \right\rangle$$\end{document}e1 are color-coded in black, red, blue, and green, respectively. b Dependencies of Franck–Condon amplitudes χij (i, j = 0 or 1) on S, which scales, d2/2. c, d Feynman pathways giving rise to the beating signals with c negative beating frequency −ωB, and, d positive frequency +ωB. e Beating-map locations of numbered Feynman pathways from panel c. f Beating-map locations of numbered Feynman pathways from panel d.Transitions between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_0} \right\rangle$$\end{document}g0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_{i \ge 2}} \right\rangle$$\end{document}ei≥2 states are not observed in the 2d maps. This agrees with resonance Raman scattering18,19 and their time-domain analogs20,21, where the A′1 overtone was not detected. This may imply an efficient non-radiative decay channel for the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_2} \right\rangle$$\end{document}e2 state, which results in a fast dephasing time for the hot vibronic band transitions. Transitions between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_{i \ge 2}} \right\rangle$$\end{document}gi≥2 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_0} \right\rangle$$\end{document}e0 are also not observed in the 2d maps, which can be explained as a negligible thermal population of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_{i \ge 2}} \right\rangle$$\end{document}gi≥2 due to a small Boltzmann factor at RT. The transition amplitudes between different vibronic sublevels (blue, black, green, and red arrows in Fig. 3a) are proportional to the overlap of the vibrational wave functions of initial and final state, i.e., the Franck–Condon amplitudes χ57, plotted as a function of S in Fig. 3b. At S = 0, the red and blue curves are zero, indicating that it is not possible to excite \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_1} \right\rangle$$\end{document}e1 starting from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_0} \right\rangle$$\end{document}g0 or to reach \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_1} \right\rangle$$\end{document}g1 from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_0} \right\rangle$$\end{document}e0, thus the electronic/excitonic excitation is decoupled from vibrations.We now correlate S with the oscillatory signals. We perform an additional Fourier transformation of the 2d maps with respect to T. This gives rise to a 3d spectrum, which is a hypercube as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar$$\end{document}ħωτ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar$$\end{document}ħωT and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar$$\end{document}ħωt. 2d cuts at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar \omega _T$$\end{document}ħωT =\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\,+ \hbar \omega _B$$\end{document}+ħωB and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar \omega _T$$\end{document}ħωT =\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\,- \hbar \omega _B$$\end{document}−ħωB result in two 2d beating maps, where ωB is the beating frequency induced by EXPC.Figure 3c lists all possible rephasing Feynman pathways that can result in contributions at negative beating frequency −ωB. Their individual positions in the 2d beating maps are in Fig. 3e. Figure 3d contains the contributions at positive ωB, and Fig. 3f their positions in the 2d map. The determination of all peak positions of individual Feynman pathways in 2d beating maps is introduced in Supplementary Note 8. Adding all pathways, we expect the beating map to be located on the lower right of the diagonal for negative beating frequency (Fig. 3e), and on the upper left for positive (Fig. 3f). The precise shape of the overall beating map depends on the relative amplitudes of the individual Feynman pathways. Those depend on the initial populations of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_0} \right\rangle$$\end{document}g0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_1} \right\rangle$$\end{document}g1, hence on the sample temperature, and on the products of the Franck–Condon amplitudes of the four involved transitions (colored arrows in Fig. 3c, d) that in turn depend on S (Fig. 3b). Thus, analyzing the shape of the beating maps allows us to estimate S.For a quantitative evaluation, we simulate the 2d beating maps by numerically solving a Lindblad master equation58 for a system described by the Franck–Condon model illustrated in Fig. 3a (see Methods for details). S is varied from 0.25 to 2 with a step size of 0.25. Figure 4a plots the simulation for S = 0.5, 1, 1.5 from top to bottom. Data for other S are in Supplementary Fig. 13. We recognize the expected features of Fig. 3e, f. The pathway contributions overlap with each other, due to line broadening along the diagonal and anti-diagonal directions. For S = 1.5, the four underlying subpeaks create a square lineshape. For smaller S, the anti-diagonal linewidth changes strongly because of the varying relative contributions of the different Feynman pathways, leading to one asymmetric peak in each 2d beating map, whose center is located below (above) the diagonal line for negative (positive) beating frequency as predicted in Fig. 3e, f. The change in linewidth can be understood by considering that χ11 (Fig. 3b, solid green curve) crosses zero (the dashed gray line) for S = 1, such that only Feynman pathways 1, 7, 11, 13, e.g., without \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{g}}_1} \right\rangle$$\end{document}g1–\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_1} \right\rangle$$\end{document}e1 transition (green arrow in Fig. 3c, d), contribute. Therefore, the anti-diagonal linewidth reaches a minimum for S = 1.Fig. 42d beating maps.a Simulated 2d beating maps for –ωB (left) and +ωB (right) and S = 0.5, 1, 1.5 from top to bottom rows. b Measured 2d beating maps with −ωB (left) and +ωB (right). c
D between measured and simulated 2d beating maps versus S used in the simulation.Figure 4b shows the experimental 2d beating maps at –ωB (left) and +ωB (right), obtained as cuts through the rephasing 3d spectrum at the same beating frequency as in the simulations, ωB = 4.6 × 1013 s−1. The asymmetry with respect to the diagonal is visible, and the elliptical shape [rather than roundish (small S) or squarish (large S)] points at an intermediate S by comparison with simulations. The lowest contour lines of experimental and simulated beating maps in Fig. 4a, b show some “jagged” behavior. The factors that could contribute to this are discussed in Supplementary Note 9.To determine the EXPC strength quantitatively, we calculate the deviation D between measured and simulated 2d beating maps:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D = \sqrt {\frac{1}{{N^2}}\mathop {\sum }\limits_{i = 1}^N \mathop {\sum }\limits_{j = 1}^N \left( {A_{ij} - \tilde A_{ij}} \right)^2}$$\end{document}D=1N2∑i=1N∑j=1NAij−A~ij2where N is the pixel number in each dimension of the 2d beating maps, Aij (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde A_{ij}$$\end{document}A~ij) is the amplitude of the pixel in column i and row j of the simulated (experimental) 2d beating map. Figure 4c plots D versus S. We find the best agreement for S = 1. We then compare the experimental absorptive, rephasing, and nonrephasing 2d maps for T = 50 fs (Fig. 5a) with the simulation using the optimal S (Fig. 5b) and find good agreement, confirming the reliability of our Franck–Condon model.Fig. 5Absorptive (left), rephasing (middle), and nonrephasing (right) real-valued 2d maps at T = 50 fs.a Experiment. b Simulation using the deduced optimal S = 1.We note that large S on the order of 1 in 1L-TMDs are supported by theory59–61, but were never previously experimentally measured, to the best of our knowledge. The exciton coupling with longitudinal optical phonons in 1L-TMDs was studied by ab initio calculations59,60. These found that polar LO phonon vibrations give rise to a macroscopic electric field that couples to the charge carriers. Such a coupling, named “Fröhlich interaction”, is fundamentally affected by the dimensionality of the system. When the dimensionality of the system decreases from 3d to 2d, a 3-fold increase of S is predicted, see, e.g., Fig. 7 in Ref. 62. Taking into account Fröhlich interactions in a 2d model, Ref. 61 calculated S for LO phonons as a function of the polarization parameter for 1L-MoSe2, finding ~1.93–2.24. Defects may also have a strong influence on S63,64. The electric fields induced by local charges at interfaces increase the non-vanishing part of the electron and hole polaron clouds in the exciton state64 and, as a result, S as large as ~1 can be found64.In conclusion, we carried out spatially resolved, fluorescence-detected 2d micro-spectroscopy on 1L-MoSe2. We identified phonon sidebands upon excitation of the A exciton, due to coupling to the optical phonon mode A′1. While the phonon is not resolved in linear absorption or PL spectra at RT, analysis of the 2d beating frequency as a function of waiting time allowed us to assign the phonon mode via comparison with Raman data. We determined the exciton–phonon coupling strength, i.e., the displacement along the phonon coordinate of the excited-exciton oscillator potential with respect to the ground state, and found a Huang–Rhys factor, S ~ 1, by comparison with simulations of 2d beating maps. The measured S ~ 1 is larger than most reported values (S ~ 0–0.5) of other inorganic semiconductor nanostructures52, such as CdSe quantum dots48 and rods49, ZnSe quantum dots50, single-wall carbon nanotubes51, indicating a strong EXPC. This finding may benefit, amongst others, the development of TMD-based polariton devices65, in which the polariton-relaxation process strongly depends on the EXPC strength66.Our space-, time-, and excitation/detection-frequency-resolved spectroscopy provides a unique tool to measure EXPC also in other TMDs. hBN encapsulation can lower inhomogeneous broadening of 1L-TMDs67,68, we thus expect better resolved peaks for hBN-encapsulated samples. S may be influenced by the substrate, by changing the macroscopic electric field induced by the polar LO phonon at the interface60. E.g., SiO2 increases the screening of the Fröhlich interaction strongly at small momenta60. Therefore, we expect that a different substrate might result in a different S, hence, hBN encapsulation might also influence it. Our method can be extended to other semiconducting systems for which phonon-induced subbands are expected in the excitonic lineshape, such as single-wall carbon nanotubes69, layered perovskites70, bulk heterojunctions71, or other organic crystals. Because of the high spatial resolution ~260 nm, our technique can also be used to study excitonic coupling in layered materials heterostructures or microcavities with embedded semiconductors. The determination of EXPC will provide design-relevant parameters for the development of photonic and optoelectronic devices based on these semiconducting systems.MethodsSamples fabricationThe samples are prepared by micromechanical cleavage72 of bulk MoSe2 from HQ Graphene. This is performed with polydimethylsiloxane (PDMS) and, after inspection under an optical microscope, 1L-MoSe2 is dry transferred in ambient conditions to a 200-μm fused silica substrate73. After transfer, the samples are characterized by Raman and PL with a Renishaw Invia spectrometer at 514 nm and with a ×50 objective. Metallic frames (Cr/Au) are fabricated around selected 1L-MoSe2 flakes on fused silica by laser-writer lithography to facilitate the identification of the samples’ position for subsequent 2DES characterization.Data acquisitionA femtosecond oscillator (Venteon Laser Technologies GmbH, Pulse One PE) provides a laser spectrum ranging from 650 to 950 nm, confined by a hard aperture in the Fourier plane of a 4f-based pulse shaper in front of the liquid-crystal display (LCD, Jenoptik Optical Systems GmbH, SLM-S640d). The aperture acts as a short-pass filter at 808 nm, so that the longer-wavelength PL can be detected without scattering from the pump light. A Schott KG5 color filter further modulates the spectrum into a smooth shape, which ensures the absence of pronounced side peaks and other irregularities in the temporal pulse profile. The laser focus in the microscope is mapped by a piezo scanning stage (P-517.3CL, PI, Germany). Excitation occurs through a focusing objective (Nikon Plan Apo, 100×/1.40). PL is collected through the same objective, transmitted through a dichroic beam splitter (DBS, AHF Analysentechnik, F48-810) and an additional emission filter (EF, AHF Analysentechnik, F76-810), and detected by an APD (Perkin Elmer, SPCM-CD 2801).We compress the laser pulses by (1) using chirped mirrors to pre-compensate some second-order phase dispersion; (2) employing the pulse shaper to compensate any remaining dispersion. A two-photon photodiode (TPPD) is placed in the focus of the microscope objective to generate a nonlinear feedback signal that is a measure of pulse intensity and pulse duration. We then utilize the algorithm of Ref. 74 to maximize the peak intensity, leading to a transform-limited laser pulse. To characterize the result of pulse compression, an autocorrelation trace is measured using the same TPPD, as shown in Supplementary Fig. 15. This agrees well with a simulated one assuming the experimentally measured laser spectrum and a flat spectral phase. This correspondence indicates successful phase-dispersion compensation and ~12 fs pulses at the sample position, as discussed in Refs. 47,74.Linearly polarized light, acting as a superposition of left- and right-handed circularly polarized light, is used to simultaneously excite both the transitions in the K and K’ valleys. The pump fluence is ~2 μJ/cm2. We estimate the heating through laser irradiation during the experiment as discussed in Supplementary Note 10. The sample temperature increases from 300 to ~308 K within the first 100 ns, then remains constant. Thus there is no unwanted heating, thermal instabilities or damage.We obtain the 2d maps by scanning τ and t in steps of 3 fs each from 0 to 99 fs, for T = 50, 250, 500, 750, 1000, 1250, 1500, 1750, 2000 fs, using the spectral modulation function75:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M\left( \omega \right) = \, {\mathrm{exp}}\left[ {i\left( {\omega - \omega _0\left( {1 - \gamma } \right)} \right)\left( { - \tau - T} \right)} \right] + {\mathrm{exp}}\left[ {i\left( {\omega - \omega _0\left( {1 - \gamma } \right)} \right)\left( { - T} \right) + i\varphi _{12}} \right]\\ + {\mathrm{exp}}\left[ {i\varphi _{13}} \right] + {\mathrm{exp}}\left[ {i\left( {\omega - \omega _0\left( {1 - \gamma } \right)} \right)t + i\varphi _{14}} \right]$$\end{document}Mω=expiω−ω01−γ−τ−T+expiω−ω01−γ−T+iφ12+expiφ13+expiω−ω01−γt+iφ14at a center frequency ω0 = 2.5 × 1015 s−1. We avoid undersampling with time steps of 3 fs by employing a partially rotating frame with γ = 0.2. The third pulse is fixed at time 0, so that when 2d maps are measured at a certain T, only the first and fourth pulses are delayed. By setting the phase of the first pulse to 0, three relative phases, i.e., φ12, φ13, and φ14, are scanned in a 27-step phase-cycling scheme, where each relative phase takes values of 0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{2\pi }}{3}$$\end{document}2π3, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{4\pi }}{3}$$\end{document}4π3. This allows us to select rephasing and nonrephasing contributions individually from the complete raw data47,76. We obtain absorptive 2d maps by summing the real parts of the rephasing and nonrephasing 2d ones, canceling dispersion terms, leaving a pure absorptive lineshape39. Due to the finite response time of the liquid crystals of our pulse shaper, we wait ~500 ms after changing the phase mask before taking data. PL is averaged over ~1 ms for each APD acquisition period. Including additional averaging (2000 times for each pulse shape), the total measurement time for one 2d map is ~26 h. During the measurements the PL intensity of the sample is constantly monitored every ~80 s. We observe no systematic decay during the measurement time. This indicates a long-term chemical, thermal, and photostability of the sample. The group delay dispersion at the sample position is compensated by adding an additional phase to the modulation function47.SimulationsTo simulate the 2d maps, we solve the Lindblad quantum master equation583\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{\partial }{{\partial t^{\prime} }}\rho \left( {t^{\prime} } \right) = - \frac{i}{\hbar }\left[ {{\cal{H}}\left( {t^{\prime} } \right),\rho \left( {t^{\prime} } \right)} \right] + \mathop {\sum }\limits_j \frac{1}{{T_j}}\left( {{\cal{L}}_j\rho \left( {t^{\prime} } \right){\cal{L}}_j^\dagger - \frac{1}{2}{\cal{L}}_j^\dagger {\cal{L}}_j\rho \left( {t^{\prime} } \right) - \frac{1}{2}\rho \left( {t^{\prime} } \right){\cal{L}}_j{\cal{L}}_j^\dagger } \right)$$\end{document}∂∂t′ρt′=−iħHt′,ρt′+∑j1TjLjρt′Lj†−12Lj†Ljρt′−12ρt′LjLj†where the time evolution of the density matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \left( {t^{\prime} } \right)$$\end{document}ρt′ of the quantum system under a Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\cal{H}}\left( {t^{\prime} } \right)$$\end{document}Ht′ is treated in the Liouville–von Neumann formalism, with the extension of dissipative and pure dephasing effects, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\cal{H}}\left( {t^{\prime} } \right)$$\end{document}Ht′ is expressed as the sum of a time-independent Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\cal{H}}_0 = \hbar \omega _m\mathop {\sum}\nolimits_m {\left| m \right\rangle \left\langle m \right|}$$\end{document}H0=ħωm∑mmm and an interaction Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\cal{H}}_I\left( {t^{\prime} } \right) = \gamma _{{\mathrm{ex}}}E\left( {t^{\prime} } \right)\mathop {\sum}\nolimits_{m \ne n} {\mu _{m,n}\left( {\left| m \right\rangle \left\langle n \right| + \left| n \right\rangle \left\langle m \right|} \right)}$$\end{document}HIt′=γexEt′∑m≠nμm,nmn+nm, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| m \right\rangle$$\end{document}m (or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| n \right\rangle$$\end{document}n) are the unperturbed eigenstates with eigenenergies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar \omega _m$$\end{document}ħωm (or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hbar \omega _n$$\end{document}ħωn), γex is the field coupling strength for excitation with external electric field \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E\left( {t^{\prime} } \right)$$\end{document}Et′, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _{m,n}$$\end{document}μm,n is the transition dipole moment between states \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| m \right\rangle$$\end{document}m and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| n \right\rangle$$\end{document}n, Tj represents the time associated with a pure dephasing or population relaxation process, and the Lindblad operators \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\cal{L}}_j$$\end{document}Lj are defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\cal{L}}_j = a_n^\dagger a_n$$\end{document}Lj=an†an for pure dephasing and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\cal{L}}_j = a_m^\dagger a_n$$\end{document}Lj=am†an, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m\, \ne\, n$$\end{document}m≠n for a population relaxation process, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_m^\dagger$$\end{document}am†. and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_n$$\end{document}an denote the creation and annihilation operators, respectively.We assume a four-level system, with two vibrational levels in the ground electronic state (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {{\mathrm{g}}_0} \rangle$$\end{document}∣g0⟩ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {{\mathrm{g}}_1} \rangle$$\end{document}∣g1⟩) and two vibronically excited states (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_0} \right\rangle$$\end{document}e0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_1} \right\rangle$$\end{document}e1), as in Fig. 3a. The splittings within the subbands are taken to be identical, i.e., we use the same energy separations (30 meV20) between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {{\mathrm{g}}_0} \rangle$$\end{document}∣g0⟩ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {{\mathrm{g}}_1} \rangle$$\end{document}∣g1⟩ as well as between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_0} \right\rangle$$\end{document}e0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left| {{\mathrm{e}}_1} \right\rangle$$\end{document}e1. The Franck–Condon amplitudes between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {{\mathrm{g}}_i} \rangle$$\end{document}∣gi⟩ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {{\mathrm{e}}_j} \rangle$$\end{document}∣ej⟩, i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi _{ij}$$\end{document}χij. (i, j = 0 or 1) depend on S as for Fig. 3b. The initial populations of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {{\mathrm{g}}_0} \rangle$$\end{document}∣g0⟩ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left. {{\mathrm{g}}_1} \right\rangle$$\end{document}g1 are determined by the temperature, according to the Boltzmann distribution. In Supplementary Note 10 we estimate the heating through laser irradiation during the experiment. We find the sample to remain close to RT.The excitation laser field is calculated from the experimentally utilized laser spectrum assuming a flat phase and then adding the transfer function:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M\left( \omega \right) = \; {\mathrm{exp}}\left\{ {i\left[ {\omega - \omega _0\left( {1 - \gamma } \right)} \right] {T_{{\mathrm{off}}}}} \right\} + {\mathrm{exp}}\left\{ {i\left[ {\omega - \omega _0\left( {1 - \gamma } \right)} \right]\left( {T_{{\mathrm{off}}} + \tau } \right) + i\varphi _{12}} \right\} \\ + {\mathrm{exp}}\left\{ {i\left[ {\omega - \omega _0\left( {1 - \gamma } \right)} \right]\left( {T_{{\mathrm{off}}} + \tau + T} \right) + i\varphi _{13}} \right\} \\ + {\mathrm{exp}}\left\{ {i\left[ {\omega - \omega _0\left( {1 - \gamma } \right)} \right]\left( {T_{{\mathrm{off}}} + \tau + T + t} \right) + i\varphi _{14}} \right\}$$\end{document}Mω=expiω−ω01−γToff+expiω−ω01−γToff+τ+iφ12+expiω−ω01−γToff+τ+T+iφ13+expiω−ω01−γToff+τ+T+t+iφ14where Toff is an offset of the position of the first pulse in time domain, set at 100 fs to avoid cutting off the first pulse at time zero. In the experimental modulation function of Eq. (2), time zero is set at the maximum of the third pulse, leading to a different mathematical expression. However, this difference does not affect the resulting 2d maps, since only relative time delays between the pulses are relevant. τ and t in the simulation are scanned with the same parameters as in the experiment, from 0 to 99 fs in steps of 3 fs with γ = 0.2, whereas T is scanned from 0 to 200 fs in steps of 25 fs.Inhomogeneous broadening due to a Gaussian distribution of excitonic transition frequencies is taken into account by obtaining the inhomogeneously broadened response function, SI(τ, t), from the homogeneous response, S(τ, t), by solving Eq. (3), via:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_I\left( {\tau ,t} \right) = S\left( {\tau ,t} \right) \cdot {\mathrm{exp}}\left[ { - \Delta ^2 \cdot \left( {\tau \mp t} \right)^2} \right]$$\end{document}SIτ,t=Sτ,t⋅exp−Δ2⋅τ∓t2where Δ is a parameter linearly proportional to the inhomogeneous linewidth broadening,—is applied for the rephasing signal, and + for the nonrephasing signal. Equation (5) is used under two assumptions. (1) Spectral diffusion can be ignored within the T = 2 ps window of the measurements. Typically, spectral diffusion is caused by environmental fluctuations around the transition dipoles, inducing a broadening along the anti-diagonal direction for the absorptive 2d maps as T increases39. This is not observed in our experiments (see Supplementary Fig. 5), indicating a much slower than 2 ps modulation time constant of the environment, justifying the use of Eq. (5). (2) The vibrational frequency does not change with the excitonic transition energy, also assumed for the model of Fig. 3a and Eqs. (1–3). If this was not fulfilled, a tilt of elongated peaks in the 2d beating maps relative to the diagonal would be observed40, unlike in our measurements (Fig. 4b).Supplementary informationSupplementary Information | nature communications | [
"Article"
] | [
"Two-dimensional materials",
"Fluorescence spectroscopy"
] | materials single-layer (1L transition metal dichalcogenides (1L-TMDs platform for photonic optoelectronic devices Bulk semiconducting TMDs covalently bound layers MX2 M metal X chalcogen van der Waals 1L quantum confinement induces indirect-to-direct bandgap reduced high exciton energies 1L-TMDs excellent candidates optoelectronic devices room temperature)2.Exciton–phonon coupling temperature-dependent optoelectronic transport properties 1L responsible for non-radiative exciton limiting fluorescence quantum dark-exciton phonon mediates spin-flip processes lifetime-polarized charge temperature < 100 K interaction excitons phonons induces linewidth broadening dominates excitonic resonances 1L-TMDs9 different higher temperature coupling excitons phonons induces sidebands in absorption spectrum 1L-MoSe2 at RT spectral signature obscured by inhomogeneous broadening17presence EXPC inferred from Raman scattering18 time-resolved transmission measurements20 A′1 optical phonon mode with excitonic resonance exciton energies from photoluminescence phonons Raman measurements characterize system complete Hamiltonian requires displacement along phonon coordinate of exciton-state potential energy minimum versus ground state displacement EXPC strength determines optical transition to exciton state EXPC strength not measured for 1L-TMDs at RT overtone bands optical phonon mode missing quantity in present work.Optical four-wave-mixing experiments in semiconductors dynamics excitons10 photon echo experiments polarization state of photons mechanisms signal levels distinguished by polarization Two-dimensional electronic spectroscopy (2DES) light coherences in molecular semiconductors33 generalized transient absorption spectroscopy frequency resolution probe step broadband excitation quantum energy levels leads to wave packets detected as oscillations peaks 2d maps waiting time Analysis of frequency decay time position oscillations energy structure coupling mechanism level splittings30proposed additional Fourier transform along T cutting 3d spectrum frequencies lead to 2d maps sensitive EXPC strength challenging apply 2DES on micro-scale samples heterogeneous materials localized structural domains μm lateral scale standard phase-matching geometry requires beams non-collinear high-numerical-aperture (NA) objective light same angle phase matching requires longer focal lengths larger spot sizes unwanted averaging spatial regions select signal by phase detecting population-based signals inter-pulse phase collinear geometry phase cycling enables 2d micro-spectroscopy additional spatial develop 2d micro-spectroscopy resolve spectral features phonon sidebands in 1L-MoSe2 at RT determine EXPC observe oscillations in 2d maps from coupling A′1 optical phonon mode A excitoncomparison 2d beating maps deduce Huang–Rhys factor S ~ 1. implies large EXPC strength 1L-MoSe2 compared inorganic semiconductor nanostructures CdSe ZnSe dots50 single-wall carbon nanotubes51 range ~ 0–0.552 design-relevant information development photonic devices 1L-MoSe2. method extended 1L-TMDs LMs semiconducting systems ~260-nm spatial resolution micro-spectroscopy required single-wall carbon nanotubes LM heterostructures layered perovskites bulk heterojunctions microcavities semiconductors experimental setup Fig. 1a Ti:sapphire oscillator emits 12-fs pulses 80 MHz repetition rate pulse shaper generates four-pulse sequence focused high-NA = 1.4 objective spatial resolution ~ 260 nm laser focus mapped piezo scanning stage PL signal detected avalanche photodiode 2d map PL intensity detected scanning first coherence time transformation τ results 2d map T Nonlinear signals obtained scanning discrete phase steps pulse Rephasing nonrephasing signals retrieved linear superpositions phase-modulated. 1Overview setup sampleFluorescence 2d micro-spectroscopy setup Four collinear laser pulses generated pulse shaper controllable inter delays phases 1 4) focused by high-NA objective position sample controlled by piezo scanning stage dichroic mirror under transmits excitation beam reflects PL signal long-pass filter excitation beam PL signal detected by avalanche photodiode PL map Raman spectrum 514 nm excitation peak A exciton spectrum out-of-plane A′1 mode 241 cm−1 in-plane E′ mode cm−1 mechanically exfoliated 1L-MoSe2 on 200-μm fused silica substrate Figure 1b PL map representative sample 1L-MoSe2 direct bandgap at K point Brillouin zone two excitonic transitions A B ~1.57 1.75 PL spectrum single peak ~1.57 eV radiative recombination A excitons54 signal trion weaker than neutral exciton at RT24,55 predominantly neutral exciton confirmed linear PL spectrum main peak neutral excitons54 Raman spectrum measured at 514 nmshows out-of-plane A′1 mode ~241 cm−1 full half maximum ~ 4 cm−1 in-plane E′ mode ~288 cm−1 (FWHM 6 cm−1) PL Raman spectra confirm sample 1L-MoSe218,54 rephasing 2d maps A exciton Fig. 2a T nonrephasing absorptive 2d maps Supplementary Figs. 1 5 peak linewidth diagonal versus T Fig. 2b variation T three components diagonal purple crosses amplitudes oscillate not phase T ~ 1500 fs amplitude middle component higher minimizing effective diagonal linewidth measured 2d maps capture fourth-order nonlinear optical response sixth-order contributions negligible. 2Beating signal rephasing 2d maps T normalized maximum value T = 500 fs Diagonal linewidth (FWHM orange arrow T = 50 fs versus T error bars depict 95% confidence bounds fitting diagonal slices Gaussian function Amplitude evolution (green diamonds) one pixel T = 50 fs fit green curve). error bars evaluated fluctuations region background noise Note 4)extract amplitude evolution pixel green diamond 2d map 50 fs function T (Fig. 2c). number points restricted long measurement time (26 h long-lived>2 ps oscillation above noise level observed reproducibility confirmed second measurement T Supplementary Note analyze origin oscillations deduce EXPC strength experiments trion signal 1L-MoSe2 ~0.03 eV below neutral exciton dies out temperature 15 to 295 K signal intensity neutral signal trion weaker than neutral exciton at RT predominantly neutral exciton wave packets trions excluded long-lived>2 ps RT oscillations 2c Biexciton signals excluded thermal dissociation RT cancellation excited-state absorption pathways spectroscopy Note 5) Vibrational wave packets reported at RT Refs. 20,21 dephasing time ~4.5 ps 1L- WSe220 ~1.7 ps 1L-MoS221 EXPC explain oscillations 2d maps extract phonon energy from fit (Fig.solid green curve obtain undersampled data point oscillation period finite acquisition oscillation period ~136 ± 2 fs Supplementary Note 6 energy splitting states ~30.4 ± 0.4 meV optical A′1 phonon energy ~29.9 meV 241 cm−1 measured Raman spectrum Fig. 1d define EXPC strength Huang–Rhys factor Franck–Condon coupling Supplementary Note 7 definition minimum states observed data (Fig. 3a). model delivers 3 transition energies (purple crosses Figassign component 1 lowest energy\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek\oddsidemargin{-69pt}{document}\end{document}ħω1) transition[12pt]{minimal{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin{-69pt}{document}\left|\mathrm{g}}_1}\rangle\end{document}g1[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb{amsbsy{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}{document}\left|\mathrm{e}}_0}\rangle\end{document}e0 (blue Fig.component 2 higher energy[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{amsbsy{mathrsfs{upgreek\oddsidemargin}{-69pt}\begin{document}\hbar \omega\end{document}ħω2) two degenerate transitions[12pt]{minimal{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin}{-69pt}\begin{document\left|\mathrm{g}}_0\rangle\end{document}g0[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs}{upgreek}\oddsidemargin}{-69pt}\begin{document}\left|\mathrm{e}}_0\rangle\end{document}e0[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}{document}\left\mathrm{g}}_1\rangle{document}g1[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin}{-69pt}\left\mathrm{e}}_1}\rangle{document}e1 (black green component 3 highest energy[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}\omega{document}ħω3) transition[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}\left\mathrm{g}}_0\rangle{document}g0[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbs\usepackage{mathrsfs-69pt}{document}\left\mathrm{e}}_1\rangle{document}e1 (red).Fig. 3Analysis beating signalsSchematic diagram displaced harmonic oscillators (Franck–Condon coupling model56) two vibrational levels\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts}{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin}{-69pt}\mathrm{g}}_0{document}g0[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin}{-69pt}\mathrm{g}}_1{document}g1) electronic ground state two electronic excited state[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\mathrm{e}}_0{document}e0[12pt]{minimal}{amsmath{wasysym}{amsfonts}{amssymb{amsbsyage{mathrsfs{upgreek-69pt{document}\left\mathrm{e}}_1\rangle{document horizontal shift between potential minima characterizes EXPC strengthTransitions[12pt]{minimal{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek\oddsidemargin{-69pt}{document\left\mathrm{g}}_0{document}g0–[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb}{amsbsy{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document\left\mathrm{e}}_0}{document}e0[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document\left\mathrm{g}}_0\rangle{document}g0–[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb{amsbsy{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document\mathrm{e}}_1\rangle{document}e1[12pt]{minimal{amsmath{wasysym{amsfonts}{amssymb}{amsbsy{mathrsfs{upgreek}\oddsidemargin{-69pt}{document\mathrm{g}}_1}\rangle{document}g1–[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document\left\mathrm{e}}_0}\rangle{document}e0[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document\left\mathrm{g}}_1}\rangle{document}g1–[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb}{amsbsy}{mathrsfs}age{upgreek}\setlength\oddsidemargin-69pt}\left\mathrm{e}}_1}\rangle{document}e1 color-coded black red blue green Dependencies Franck–Condon amplitudes χij (i, j = 0 or 1) on S scales d2/2 c d Feynman pathways beating signals c negative frequency −ωB d positive frequency +ωB. Beating-map locations Feynman pathways panel c f panel d.Transitions between\documentclass[12pt]{minimal}{amsmath}}}\oddsidemargin-69pt}\left\mathrm{g}}_0}\rangle\end{document}g0[12pt]{minimal}\usepackage{amsmath-69pt}\left|\mathrm{e}}_{i 2}}\rangle\end{document}ei≥2 states not observed in 2d mapsagrees resonance Raman scattering18,19 time-domain analogs20,21 A′1 overtone not detected imply efficient non-radiative decay channel[12pt]{minimal}\usepackage{amsmath}\oddsidemargin-69pt}{document}\left\mathrm{e}}_2\rangle\end{document}e2 state fast dephasing time hot vibronic band transitionsTransitions between\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek-69pt}\mathrm{g}} \ge 2}}\rangle{document}gi≥2[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek{-69pt}\mathrm{e}}_0}e0 not observed in 2d maps negligible thermal population of[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin}{-69pt}\left\mathrm{g}} \ge 2}}\rangle{document}gi≥2 due to small Boltzmann factor at RT transition amplitudes between vibronic sublevels (blue black green red arrows Fig. 3a proportional to overlap of vibrational wave functions of initial and final stateFranck–Condon amplitudes χ57 S Fig. 3bS = 0 red blue curves zero not possible excite\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts}{amssymb{amsbsy{mathrsfs{upgreek\oddsidemargin{-69pt}\left\mathrm{e}}_1\rangle{document}e1 from[12pt]{minimal}{amsmath{wasysym{amsbsy}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}\left\mathrm{g}}_0\rangle\end{document}g0 reach\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\left\mathrm{g}}_1\rangle\end{document}g1 from[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb{amsbsy{mathrsfs}{upgreek\setlength\oddsidemargin-69pt}\begin{document}\left\mathrm{e}}_0\rangle\end{document}e0 electronic/excitonic excitation decoupled from vibrations correlate S with oscillatory signals Fourier transformation 2d maps T 3d spectrum hypercube\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{amssymb{amsbsy}{mathrsfs}{upgreek}\oddsidemargin-69pt}\hbar\end{document}ħωτ[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin{-69pt}\begin\end{document}ħωT[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts{amssymb{amsbsy{mathrsfs}{upgreek}\oddsidemargin{-69pt}\end{document}ħωt.2d cuts[12pt]{minimal{amsmath{wasysym{amsfonts{amssymb{mathrsfs{upgreek\oddsidemargin}{-69pt}{document}$$\hbar\omega\end{document}ħωT[12pt]{minimal{amsmath{wasysym{amsfonts}}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$\omega\end{document}+ħωB[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}\omega\end{document}ħωT[12pt]{minimal}{amsmath{wasysym{amsfonts}{amssymb{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\begin{document}$$\\omega$\end{document}−ħω2d beating maps ωB induced EXPC.Figure 3c Feynman pathways negative beating frequency −ωB positions 2d beating maps Fig. 3e Figure 3d contributions positive ωB Fig. 3f positions 2d map determination peak positions Feynman pathways 2d Supplementary Note 8. pathways beating map lower right diagonal negative beating frequency (Fig. upper left positive (Fig. 3f). shape map depends amplitudes Feynman pathwaysdepend initial populations\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}\mathrm{g}}_0{document}g0[12pt]{minimal}{amsmath{wasysym{upgreek{-69pt}\mathrm{g}}_1{document}g1 sample temperature products Franck–Condon amplitudes four transitions Fig. 3c, d depend on S (Fig. 3b). analyzing shape beating maps estimate S evaluation simulate 2d beating maps solving Lindblad master equation58 system Franck–Condon model Fig. 3a S varied from 0.25 to 2 step size 0.25. Figure 4a plots simulation for S = 0.5, 1, 1.5 top to bottom Data other S in Supplementary Fig. 13. expected features Fig. 3e, f pathway contributions overlap due line broadening diagonal anti-diagonal directionsS = 1.5 four subpeaks create lineshape smaller S anti-diagonal linewidth changes varying contributions Feynman pathways one asymmetric peak each 2d beating map center below diagonal line negative beating frequency Fig. 3e, f change linewidth χ11 (Fig. 3b green curve) crosses zero gray line S = 1 Feynman pathways 1 7 11 13,\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}{g}}_1{document[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}\oddsidemargin-69pt{e}}_1\rangle{document}e1 transition (green arrow Fig. 3c, anti-diagonal linewidth reaches minimum S = 1.Fig. 42d beating mapsSimulated 2d beating maps –ωB +ωB S = 0.5 1 1.5 top bottom rows Measured 2d beating maps −ωB) +ωB D measured simulated 2d beating maps versus S simulation.Figure 4b experimental 2d beating maps –ωB +ωB cuts 3d spectrum same beating frequency ωB = × 1013 s−1 asymmetry diagonal visible elliptical shape points intermediate S lowest contour lines Fig. 4a show “jagged” behaviorfactors Supplementary Note determine EXPC strength calculate deviation D between measured simulated 2d beating maps:1\documentclass[12pt]{minimal{amsmath-69pt{document}$D = \sqrt{1}{{N^2}}\mathop\sum{i = 1}^N{j = 1}^N\left {A_{ij} -\tilde A_{ij}}^2{document}D=1N2∑i=1N∑j=1NAij−A N pixel number each 2d beating maps Aij[12pt]{minimal-69pt A_{ij{document~ij amplitude pixel in column i row j simulated 2d beating map Figure 4c plots D versus S best agreement for S = 1. compare experimental absorptive rephasing nonrephasing 2d maps for T = 50 fs (Fig. 5a) with simulation optimal S (Fig.agreement confirming reliability Franck–Condon model.Fig. 5Absorptive rephasing nonrephasing real-valued 2d maps at T = 50 fs Experiment Simulation optimal S = large S 1 in 1L-TMDs supported never experimentally measured exciton coupling longitudinal optical phonons studied polar LO phonon vibrations macroscopic electric field charge carriers “Fröhlich affected system dimensionality decreases 3d to 2d 3-fold increase S predicted Fig. 7 Ref. 62 Ref. 61 calculated S for LO phonons polarization parameter 1L-MoSe2 ~1.93–2.24. Defects influence electric fields local charges interfaces increase non-vanishing electron hole polaron clouds exciton S large ~1 spatially resolved fluorescence-detected 2d micro-spectroscopy on 1L-MoSe2. identified phonon sidebands excitation A exciton due coupling optical phonon mode A′1 phonon not resolved linear absorption PL spectra RT analysis 2d beating frequency waiting time assign phonon mode Raman data determined exciton–phonon coupling strengthdisplacement phonon coordinate excited-exciton oscillator potential ground state found Huang–Rhys factor S ~ 1 2d beating maps measured S ~ 1 larger than values ~ 0–0.5 inorganic semiconductor CdSe ZnSe single-wall carbon strong EXPC benefit development TMD-based polariton polariton-relaxation depends EXPC space- time excitation/detection-frequency-resolved spectroscopy measure EXPC TMDs hBN encapsulation inhomogeneous broadening 1L-TMDs67 better resolved peaks hBN-encapsulated samples S influenced substrate electric field polar LO phonon SiO2 increases screening Fröhlich interaction small different substrate different S hBN encapsulation influence method extended semiconducting systems phonon-induced subbands single-wall carbon nanotubes69 layered bulk heterojunctions71 organic crystals high spatial resolution ~260 nm technique study excitonic coupling layered materials heterostructures microcavities semiconductors determination EXPC design-relevant parameters development photonic optoelectronic devices semiconducting systemsprepared by micromechanical cleavage72 bulk MoSe2 from HQ Graphene performed with polydimethylsiloxane 1L-MoSe2 transferred to 200-μm fused silica characterized by Raman PL with Renishaw Invia spectrometer at 514 nm ×50 objective Metallic frames (Cr/Au fabricated around 1L-MoSe2 flakes on fused silica laser-writer lithography position for 2DES characterization femtosecond oscillator (Venteon Laser Technologies GmbH provides laser spectrum 650 to 950 nm confined by hard aperture 4f-based pulse shaper liquid-crystal display aperture short-pass filter at 808 nm longer-wavelength PL without scattering Schott KG5 color filter modulates spectrum smooth shape absence side peaks irregularities pulse laser focus mapped by piezo scanning stage Excitation through focusing objective (Nikon Plan Apo 100×/1.40) PL collected transmitted through dichroic beam splitter emission filter detected by APD (Perkincompress laser pulses using chirped mirrors phase dispersion pulse shaper compensate remaining dispersion two-photon photodiode (TPPD) in focus microscope nonlinear feedback signal pulse intensity duration algorithm Ref. 74 peak intensity transform-limited laser pulse autocorrelation trace measured TPPD Supplementary Fig. 15. agrees with simulated one laser spectrum flat spectral phase indicates successful phase-dispersion compensation ~12 fs pulses sample position Refs. 47,74.Linearly polarized light excite transitions K K’ valleys pump fluence ~2 μJ/cm2. estimate heating through laser irradiation Supplementary Note 10. sample temperature increases 300 to ~308 K first 100 ns remains constant no unwanted heating thermal instabilities damageobtain 2d maps scanning τ t 3 fs each 0 to 99 fs T = 50, 250 500 750 1000 1250 1500 1750 2000 fs spectral modulation function75:2\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$M\left( \omega \right) = \\mathrm{exp}}\left\omega - {1 - \gamma T - - T i\varphi _{12}} \right{exp}}\left _{13}} \right{exp}}\omega - -\gamma + i\varphi _{14}} \right\end{document}Mω=expiω−ω01−γ−τ−T+iφ12+expiφ13+iφ14at center frequency ω0 = 2.5 × 1015 s−1.avoid undersampling time steps 3 fs partially rotating frame γ = 0.2 third pulse fixed at time 0 2d maps measured T first fourth pulses delayed setting phase first pulse to 0 three relative phases φ12 φ13 φ14 scanned in 27-step phase-cycling scheme each phase takes values 0\documentclass[12pt]{minimal{amsmath{wasysym-69pt[12pt{minimal-69pt\pi}4π3 allows select rephasing nonrephasing contributions individually from raw data47 obtain absorptive 2d maps summing real parts rephasing nonrephasing 2d ones canceling dispersion terms leaving pure absorptive lineshape39 finite response time liquid crystals pulse shaper wait ~500 ms after changing phase mask before taking data PL averaged over ~1 ms for each APD acquisition periodaveraging (2000 times each pulse total measurement time 2d map ~26 h PL intensity sample monitored every ~80 s no systematic decay long-term chemical thermal photostability group delay dispersion sample position compensated additional phase modulationsimulate 2d maps solve Lindblad quantum master equation583[12pt{minimal}\usepackage{amsmath{wasysym{upgreek}\oddsidemargin-69pt}{document}\frac{\partial t^{\prime}\rho \left{\prime}\right\frac{i}\hbar }\left\cal{H}}\left^{\prime}{\prime \mathop {\sum }\limits_j \frac{1}{{T_j}}\left\cal{L}}{\prime}\cal{L}}\dagger -\frac{1}{2}\cal{L}}\dagger{L}}\left {t^{\prime} \right) -\frac{1}{2}\rho^{\prime}\right{\cal{L}}\cal{L}}_j^\dagger\right\end{document}∂∂t′ρt′=−iħHt′ time evolution density matrix[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb{mathrsfs{upgreek}\oddsidemargin-69pt}\begin{document}$\rho\left{\prime}\right$\end{document}ρt′ quantum system Hamiltonian[12pt]{minimal}{amsmath{wasysym}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}{\cal{H}}\left{\prime}\right\end{document}Ht′ treated Liouville–von Neumann formalism extension dissipative dephasing effects\documentclass[12pt]{minimal}{amsmath}{wasysym}}{mathrsfs}{upgreek}\setlength\oddsidemargin}{-69pt}{document}{\cal{H}}\left {t^{\prime}\right\end{document}Ht′ expressed sum time-independent Hamiltonian\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{mathrsfs}{upgreek}\setlength{oddsidemargin-69pt}\cal{H}}_0 = \hbar\omega\mathop\nolimits\left|\right\rangle \left\langle\right|\end{document}H0=ħωm∑mmm interaction Hamiltonian[12pt]{minimal}\usepackage{amsmath}{wasysym}}{amssymb}{amsbsy}{mathrsfs}{upgreek}-69pt}\cal{H}}\left\prime\right = \gamma\mathrm{ex\left\mathop}\nolimits{m \ne n}\mu{m,n}\left|\right\rangle\langle|\rangle\langle\end{document}HIt′=γexEt′∑m≠nμm,nmn+nm\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\oddsidemargin-69pt}\left| m\right\rangle\end{document}m[12pt]{minimal}\usepackage{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}\begin{document}$\left\right\rangle\end{document unperturbed eigenstates with eigenenergies\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}{\oddsidemargin}{-69pt}{document}$$\hbar \omega\end{document}ħωm[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$\hbar \omega\end{document}ħωn), γex field coupling strength for excitation external electric field\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$E\left( {t{\prime}\end{document[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}{m,n{document}μm transition dipole between states[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}\left|\rangle\end{document}m[12pt]{minimal}{amsmath}{wasysym}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}{document}$\left|\rangle\end{document}n Tj represents time dephasing population relaxation process Lindblad operators[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{amsbsy}{mathrsfs}{upgreek}\setlength{\oddsidemargin-69pt}\begin\cal{L}}}Lj defined[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts}}}{upgreek}{\oddsidemargin}{-69pt}\begin\cal{L}}_j =_n\dagger{document}Lj=an†an dephasing[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\begin{document}\cal{L}}_j = a_m^\dagger_n{document}Lj=am†an[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$m\\end{document}m≠n population relaxation process[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{amssymb}usepackage{mathrsfs}{upgreek}\oddsidemargin-69pt}{document}$$a_m^\dagger\end{document}am†[12pt]{minimal}{amsmath}}}}}{mathrsfs}{upgreek}\oddsidemargin-69pt}{document}$$a_n$$\end{document creation annihilation operatorsassume four-level system two vibrational levels ground electronic state\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek\oddsidemargin}{-69pt}\begin{document}\mathrm{g}}_0\end{document}∣g0⟩[12pt{minimal{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}{document}\mathrm{g}}_1}\end{document}∣g1⟩ two vibronically excited states\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}\mathrm{e}}_0\end{document}e0[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin-69pt\begin{document}\left\mathrm{e}}_1\rangle\end{document}e1) Fig. 3a splittings subbands identicaluse energy separations (30 meV20) between\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin}{-69pt}{document} {{\mathrm{g}}_0}\end{document}∣g0⟩[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document} {{\mathrm{g}}_1}\end{document}∣g1⟩\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document} {{\mathrm{e}}_0}\rangle\end{document}e0[12pt]{minimal}{amsmath{wasysym}{amsfonts}{amssymb{amsbsy{mathrsfs}{upgreek}{\oddsidemargin}{-\begin{document}\left\mathrm{e}}_1\right\rangle\end{document}e1 Franck–Condon amplitudes between\documentclass[12pt]{minimal}{amsmath}{wasysym\oddsidemargin}{-69pt}$\mathrm{g}}_i}\rangle\end{document[12pt]{minimal}{amsmath{wasysym{upgreek}\oddsidemargin}{-69pt}\begin{document}$ {{\mathrm{e}}_j}\rangle\end{document}∣ej⟩\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\begin{document}$\end{document}χij. (i, j = 0 or 1) depend on S Fig. 3b.initial populations\documentclass[12pt]{minimal{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin}{-69pt}\begin{document}\mathrm{g}}_0}\end{document}∣g0⟩[12pt]{minimal}{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\mathrm{g}}_1}\rangle{document}g1 determined by temperature Boltzmann distribution Supplementary Note 10 heating through laser irradiation sample close to RT.excitation laser field calculated laser spectrum flat phase adding transfer function\documentclass[12pt{minimal\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}$M\left( \omega \right) =\mathrm{exp}}\left\omega -\left {1 - \gamma\right {T_{{\mathrm{off \right\mathrm{exp}}\left\omega - - \gamma\mathrm + \tau + i\varphi _{12}} \right\mathrm{exp}}\left\omega - _0\left {1 - \gamma\right\mathrm + \tau + T} {13}} \right\mathrm{exp}}\left\omega - _0\left {1 - \gamma\right\mathrm{off + \tau + T + t} \right) +\varphi _{14}} \right\end{document}Mω=+τ+iφ12 Toff offset position first pulse time domain set 100 fs avoid cutting first pulse zero experimental modulation Eq (2) time zero set maximum third pulse different mathematical expression difference affect 2d maps relative time delays between pulses relevant τ t scanned same parameters 0 to 99 fs steps 3 fs γ = 0.2 T scanned 0 to 200 fs steps 25 fs.Inhomogeneous broadening Gaussian distribution excitonic transition frequencies inhomogeneously broadened response function SI(τ, from homogeneous response S(τ, solving Eqvia\documentclass[12pt{minimal\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}_I\left(\tau \right = S\mathrm{exp}}\Delta ^2^2\end{document}SIτ,t=Sτ,t⋅exp−Δ2⋅τt2where Δ proportional to inhomogeneous linewidth broadening applied rephasing + nonrephasing signal Equation (5) two assumptions Spectral diffusion ignored within T = 2 ps window environmental fluctuations transition dipoles broadening anti-diagonal absorptive 2d maps T not observed in experiments Supplementary Fig. 5) slower than 2 ps modulation time constant environment justifying use Eq. (5) vibrational frequency change with excitonic transition energy model Fig. 3a Eqs. (1–3) tilt of elongated peaks in 2d beating maps relative to diagonal unlike measurements (Fig. 4b).Supplementary | 50.9 | 1.449769 |
10.1038/s41467-020-16405-3 | PMC7239926 | Underpotential deposition (UPD) is important to modify the surface properties of nanocrystals. Here, the authors show the application of in situ electrochemical dark field spectroscopy in identifying the UPD processes of silver on different facets of gold nanocrystals at the single nanoparticle level. | Underpotential deposition offers a predominant way to tailor the electronic structure of the catalytic surface at the atomic level, which is key to engineering materials with a high activity for (electro)catalysis. However, it remains challenging to precisely control and directly probe the underpotential deposition of a (sub)monolayer of atoms on nanoparticle surfaces. In this work, we in situ observe silver electrodeposited on gold nanocrystals surface from sub-monolayer to one monolayer by designing a highly sensitive electrochemical dark field scattering setup. The spectral variation is used to reconstruct the optical “cyclic voltammogram” of every single nanocrystal for understanding the underpotential deposition process on nanocrystals, which cannot be achieved by any other methods but are essential for creating novel nanomaterials. | IntroductionEngineering nanomaterials of high catalytic activity and stability with a low cost has been an eternal subject for electrocatalysis1–4. Altering the electronic structure of atoms at the topmost layer where the reaction occurs can be the most efficient way for this subject. However, it remains challenging to find a technique that can precisely tune the surface atoms at the atomic level without introducing strong binding ligands. Underpotential deposition (UPD) provides a predominant way to accomplish such a purpose, since it enables controllable deposition of foreign metal atoms at the submonolayer level to create a clean surface with unique physical and chemical properties5–8. The UPD strategy has been intensely applied to assist wet chemical synthesis of nanoparticles (NP) with a controllable shape and size or modify over the existing NPs to create novel catalysts9–13. However, the practice is essentially empirical since a clear understanding of the UPD mechanism is still limited to bulk single crystals even after decades’ investigation6. It is still a great challenge to directly study the UPD process on individual NPs with a complex morphology, adsorbed species, and unique surface energy, even for in situ transmission electron microscopy (TEM)14 and scanning probe microscopies (SPM)15.Here, we develop a highly sensitive electrochemical dark field scattering (EC-DFS) setup for in situ monitoring the UPD process of Ag on Au nanocrystals (NCs). It enables identification of small spectral variations induced by the deposition of submonolayer atoms in the electrochemical environment. More importantly, we reconstruct the optical “cyclic voltammogram (CV)” from the spectral variation, which not only precisely gives the potential and band width of UPD process of different individual NCs but also displays the energy difference of different facets on a NC. This work provides unprecedented information for understanding the UPD process on NCs, and allows the control of the UPD process at the single NP level that are of both fundamental and technological importance for electrocatalysis.ResultsThe setup for EC-DFS techniqueWe designed a highly sensitive EC-DFS setup for more challenging systems with weak signals while maintaining the accurate and stable potential control. As depicted in Fig. 1a, an oil immersion dark field condenser with a high NA of 1.4–1.6 was utilized to illuminate an ITO electrode from the bottom, and a water immersion objective (with NA of 1.0) covered with a polyethylene film was directly dipped into the electrolyte. Such a working mode can not only effectively suppress the mismatch of the refractive index in the optical path but also significantly increase the excitation and collection efficiency. Consequently, the sensitivity of the EC-DFS technique can be dramatically improved (detailed descriptions of the setup and novelty compared with the conventional setup can be found in Supplementary Note 1). Figure 1b, c compare the dark field images of a sample in the same area obtained by the new and conventional setups, respectively. It can be found that the octahedral Au NCs (bright spots) in the image of Fig. 1b exhibit much better visibility even with a lower exposure time and gain compared with Fig. 1c. The improvement in the detection sensitivity is clear by comparing the relative intensity of the scattering spectra from a single NC and the background spectra of the substrate in the two setups (Fig. 1d, e). The spectrum acquired with a 1 s acquisition time by the new setup exhibits not only much higher signal but also lower background than that by the conventional setup even with 10 s acquisition time (normalized to 1 s), resulting in a 50-fold improvement of the signal to background ratio (see Supplementary Fig. 2). Such a great improvement enables the detection of the spherical Au NPs with a diameter as low as 10–15 nm (compared with that of 50 nm in the literature16) without using a special light source or equipment, as we demonstrated in Supplementary Fig. 3. The high sensitivity is also essential for identifying the small spectral variation and tracking dynamic processes during the electrochemical sensing as well as other applications, such as biosensing and single NP catalysis.Fig. 1A novel setup for the electrochemical dark field scattering technique.a A schematic of the dark field microscopy and spectroscopy on the basis of a water immersion objective (WE: working electrode, CE: counter electrode, RE: reference electrode). Dark field image (b) and spectra (d) obtained on the new setup. Dark field image (c) and spectra (e) obtained on the conventional setup. The exposure times for spectra in d and e are 1 and 10 s (normalized to 1 s), respectively.Ag UPD on single octahedral Au NCsAg UPD on Au is a widely used route for tailoring the electrocatalytic activity and assisting the synthesis of nanomaterials3,17–19. Although electrochemical scanning tunneling microscopy (EC-STM) has been widely used to characterize the UPD process on the bulk single crystal electrode surface20, it is technically difficult to characterize UPD on NPs surfaces. We used the cyclic voltammetry to investigate the UPD process. The CV curve of Ag deposition on a Au nanooctahedrons-coated (with exposed Au (111) facet) glassy carbon electrode (Fig. 2a, b) is shown in Fig. 2c. Unfortunately, we could not identify any apparent UPD peaks from the CV curve, even when the NPs were coated on the electrode surface in a uniform closely packed structure and had an extreme narrow size distribution (Fig. 2a). The main reason for this can be the NPs with a high coverage compresses the interparticle electrolyte into an extreme thin layer. The silver ions within the nanogap quickly deplete during the deposition, which leads to the continuous negative shift of UPD peaks. The variation of the gap distances between NPs further results in a broad distribution of the UPD potential and the broadened UPD peaks. However, such an effect is not as significant in the silver dissolution process as that in the UPD process since the silver atoms are already on the NPs surface. Consequently, the corresponding dissolution peak of UPD can still be well distinguished from OPD, although is already broader than that on the single crystal surface (see Fig. 2c). Single particle electrochemistry can avoid such an effect and provide intrinsic electrochemical features of individual NCs21–26. Therefore, we used the above EC-DFS technique to track the Ag UPD process at the single NP level (Fig. 1). An ITO electrode with a low density of Au NCs was used as the working electrode. The potential of the electrode was scanned from 0.500 to 0.362 V using cyclic voltammetry (1 mV/s), and the potential-dependent scattering spectra (see Fig. 2d) from a single NC were simultaneously acquired. For clarity, the peak position and intensity are plotted as a function of potential (Supplementary Fig. 4a). The peak position blue shifts from 0.413 V and remains constant at potentials between 0.390 and 0.370 V. A more significant blue shift is observed at the potentials lower than 0.370 V along with the increase of the spectral intensity. We confirm such spectral shifts are induced by the Ag deposition from our control experiments in solutions with electrolyte only and a lower concentration of Ag ion (Supplementary Note 2). If we take the first-order derivative of the spectral shift with time and plot it with potential, we obtain a curve shown in Fig. 2e with a similar shape to the CV, which is called optical “CV”. The curve consists of two obvious deposition peaks that may correspond to the UPD and OPD processes on the single NC. We performed the potential step experiment while simultaneously acquiring the scattering spectra to verify these two deposition processes. The spectra almost remain the same at 0.500 and 0.450 V (Fig. 2f). Whereas, once the potential reaches the first deposition peak (see Fig. 2e) of 0.400 V, the peak blue shifts. The shift stops in just a few seconds. Afterwards, the spectra remain almost the same with the further negative movement of the potential. Such a phenomenon agrees well with the feature of the UPD process, i.e., once a full monolayer has been achieved, no further deposition can occur. Moreover, the scanning electron microscopy (SEM) images show that the morphology of Au nanooctahedrons after UPD potential remains unchanged (see Supplementary Note 3), which excludes the possibilities of heterogeneous selective growth of Ag on the vertex or morphological change of NCs. When the potential reaches the second deposition process (Fig. 2e) of 0.360 V, the spectra show a fast and continuous blue shift, which agrees with the feature of the overpotential deposition (OPD) process. More direct evidence can be seen from the CV of Ag deposition on the Au (111) single crystal electrode (Fig. 2g). The CV of a clean Au (111) electrode (Fig. 2g, black curve) clearly distinguishes the UPD process from the bulk deposition by showing a current peak at 0.412 V, in good agreement with the literatures20. However, the peak potential is 12 mV more positive than that of the first deposition peak in Fig. 2e, which may indicate the lowering of the surface energy of the facet due to the presence of the surfactant on NCs. To verify this effect, we modified the Au (111) electrode surface with the same surfactant, then repeated the CV measurements. As shown in Fig. 2g (red curve), the UPD potential shifts from 0.412 to 0.402 V after the modification with surfactant, agreeing with the peak potential (0.401 V) in Fig. 2e for single NC. The result convincingly demonstrates that the first (0.401 V) and second (0.362 V) deposition peaks in the optical “CV” of a single Au nanooctahedron correspond to the UPD and OPD processes, respectively. It is interesting to see that the peak width of the UPD peaks of single Au nanooctahedron is as narrow as that of the Au (111) electrode, indicating that the facet quality of the fresh NCs is as good as that of the single crystal electrode. In Fig. 2h, we present a histogram of the UPD peak positions of the “CV”s from the 20 individual NCs, which shows a distribution of the deposition potential from 0.393 to 0.401 V with a peak width of 5 mV. This result indicates different surface states of different NCs, which cannot be observed in ensemble systems. We believe the UPD potential distribution is mainly induced by the variation in the size of the NC, which results in the small surface energy difference of the facets (see Supplementary Note 4). Consequently, one can observe the UPD potential follows a similar distribution to the size (see Supplementary Fig. 8).Fig. 2Study of Ag electrodeposit on the octahedral Au nanocrystals surface.SEM images of a large and b (top) small scale along with b (bottom) structural schematic of Au nanooctahedrons to show their highly ordered geometry on the electrode surface and the structural details. c A cyclic voltammogram (CV) of Ag underpotential deposition (UPD) on a Au nanooctahedrons-coated glassy carbon electrode in the electrolyte of 1 mM Ag2SO4 and 0.05 M H2SO4. d The potential-dependent scattering spectra of a single Au nanooctahedron. The inset is the zoom-in spectra showing the details of spectral variation. e The optical “CV” of a single Au nanooctahedron reconstructed using the peak position of the scattering spectra as a function of the applied potential. f The time-dependent peak position change of the scattering spectra of a single Au nanooctahedron under different potential controls (from 0.500 to 0.400 V with a step of 50 mV and from 0.400 to 0.360 V with a step of 10 mV). g CVs for the Ag deposition on a Au (111) electrode before (black curve) and after (red curve) the adsorption of the surfactant, CTAB. h The peak potential statistics of UPD peaks obtained from the optical “CV”s of 20 individual Au nanooctahedrons. The scale bars are 200 nm in a and 50 nm in b.Ag UPD on single cubic Au NCs surfaceTo demonstrate the generality of our method, we performed the same experiments on cubic Au NCs enclosed by {100} facets (Fig. 3b). Figure 3c shows the CV of Ag deposition on a Au nanocubes-coated electrode. Similar to Au nanooctahedrons (Fig. 2c), the UPD peak is still not discernible due to the average effect. The change in the single NC scattering spectra can be clearly observed after Ag deposition (Supplementary Note 2). Compared with the Au nanooctahedrons, we found the peak intensity change occurs prior to the spectral shift, which indicates that the peak intensity is more sensitive to the Ag deposition than the peak position in the case of Au nanocubes (in Supplementary Fig. 4b). The scattering spectra of nanooctahedrons and nanocubes respond differently upon the silver deposition. We believe it is the aspect ratio change upon deposition that determines the different responses in nanooctahedron and nanocube systems (see Supplementary Fig. 9). For example, for the nanooctahedron, the aspect ratio decreases after the silver deposition, which will induce a significant spectral shift, similar to nanorod. However, for nanocubes, the aspect ratio will not change after the silver deposition due to its isotropic morphology. Meanwhile, the increased size of nanocubes leads to the obvious increase of the scattering intensity. Therefore, we reconstruct the electrochemical “CV” of single nanocube by taking the first-order derivative of the intensity change with time. As shown in Fig. 3e, the “CV” curve is also similar to the electrochemical CV and shows two deposition peaks. The first deposition peak may be attributed to the UPD process on the Au (100) facet as verified by the potential step experiment (Fig. 3f). Similar to Au nanooctahedrons, the negative shift of the UPD peaks for single Au nanocube (0.376 V) compared with the bulk Au (100) single crystal electrode (0.386 V) is due to the adsorbed surfactants (Fig. 3g) on NCs. The histogram of the UPD potentials of 20 individual nanocubes (Fig. 3h) reveals a broader distribution compared with the nanooctahedrons, which may indicate a better quality of {111} facet than {100} on the NCs surface. However, the onset potential of the second deposition peak (OPD) is almost the same for the two types of NCs, as in both cases the Ag is deposited over the existing Ag layer. One may argue that the first deposition peak is the preferential deposition of Ag on the vertex of NCs, since the vertex may have a higher energy than the facets. However, up to now it is still extremely challenging to see the single atomic layer deposition on the vertex for such an active metal. Instead, we simulated the scattering spectra of both Au nanooctahedrons and nanocubes before and after the deposition of Ag with different thicknesses (see Supplementary Fig. 10) to see the dependence of the spectral change during the Ag deposition on the vertex. It shows that the Ag deposits on the vertex of both NCs will lead to a red shift of the spectra, whilst on the facet a blue shift. The latter agrees with our experimental results, indicating that the first deposition peak is from the deposition on the facet rather than the vertex.Fig. 3The electrodeposition of Ag on the cubic Au nanocrystals surface.SEM images of a large and b (top) small-scale reveal the highly ordered geometry of Au nanocubes on the electrode along with b (bottom) structural schematic of the Au nanocube. c The CV of Ag UPD on the Au nanocubes-coated electrode in the electrolyte of 1 mM Ag2SO4 and 0.05 M H2SO4. d The scattering spectra of single Au nanocube under different potentials. The inset shows the zoom-in spectra to clearly present the spectral variations. e The reconstructed optical “CV” of single Au nanocube, showing two different deposition processes. f The time-dependent peak intensity change of the scattering spectra of single Au nanocube under different potential controls (from 0.500 to 0.400 V with a step of 50 mV, from 0.400 to 0.380 V with a step of 10 mV and from 0.380 to 0.360 V with a step of 15 mV). g The CVs of Ag deposition on a Au (100) electrode before (black curve) and after (dark cyan curve) the adsorption of the surfactants, CTAB. h The peak potential statistics of UPD bands obtained from the optical “CV”s of 20 individual Au nanocubes. The scale bars are 200 nm in a and 50 nm in b.The UPD process of Ag deposits on truncated octahedral Au NCs with two facetsThe above studies demonstrate that applicability of our method on the NCs enclosed by only one type of facet. It will be more important if the UPD processes on NCs with different facets can also be identified. We then applied our method to study the UPD process of truncated octahedral (TO) Au NCs (Fig. 4a–c) which consist of both {111} and {100} facets. A plot of spectral peak positions and intensities with the potential for single TO NC shows that the spectral shift is about 35 mV ahead of the intensity change (Supplementary Fig. 11a). Thus we used the spectral shift to reconstruct the optical “CV” of single NCs, as shown in Fig. 4d. Interestingly, three apparent deposition peaks can be seen at 0.400, 0.374, and 0.364 V in the optical “CV”. The first and second peaks can be identified as UPD processes on the facet of {111} and {100} according to the optical “CV” of single Au nanooctahedrons and nanocubes (also confirmed by the potential step experiment, Fig. 4h), and the third peak corresponds to the OPD process. The histogram analysis shows UPD peak potentials located at 0.400 and 0.376 V for {111} and {100} facets (Fig. 4e), respectively. The potentials are 0.397 V for nanooctahedrons and 0.378 V for nanocubes, which indicates a possible charge transfer between the neighboring facets. The minor difference in the UPD potentials of different facets on single NCs may allow us to selectively modify over a certain facet with foreign atoms on purpose to achieve the special electronic property and chemical activity.Fig. 4Study of Ag electrodeposit on truncated octahedral Au nanocrystals.a A schematic of truncated octahedral Au NC showing the structural details. SEM images of truncated octahedral Au NCs coated on the electrode obtained with b high and c low magnification, showing a highly ordered structure. d The optical “CV” of a single truncated octahedral Au NC reconstructed using the spectral shift of the scattering spectra at different potentials. e The peak potential statistics of UPD bands of {111} (red) and {100} (dark cyan) facets on the truncated octahedral Au NCs obtained from the optical “CV”s of 20 individual NCs. f–g The distribution of the area ratio of {111}/{100} on different individual NCs estimated from f the optical “CV”s and g SEM image. h The time-dependent peak position change of the scattering spectra of a single truncated octahedral Au NC under different potential controls, along with the corresponding schematics showing UPD processes on different facets (top graphs). The scale bars are 200 nm in b and 50 nm in c.We further calculated the area ratios of {111} to {100} using the spectral shift of Peak 1 to Peak 2 in Fig. 4d (calculation details can be found in Supplementary Note 5). The ratios are plotted with the number of NCs, which is fit with the Gaussian function (Fig. 4f). The obtained facet area ratio of 3.15 is surprisingly close to the 3.45 obtained by SEM (Fig. 4g). With the known facet ratio and particle size, the facet areas can be quantified at the single NP level with a high accuracy by EC-DFS free from the interference of the double layer charging and oxygen reduction reaction (see the obscured CV of assembled TO Au NCs in Supplementary Fig. 11b). This method can be potentially used for characterizing the NCs with multiple complex facets (such as the nanorod27 or bipyramid28) and fast screening the structure of catalyst, which is still very challenging for other methods like electron microscopy (SEM and TEM).DiscussionWe have designed a highly sensitive EC-DFS technique allowing us to track small spectral variation during the UPD process at the single NP level. The results indicate that the EC-DFS technique is sensitive enough for direct observation of monolayer atoms deposited on different facets of single NCs under the electrochemical environment. Moreover, the spectral change can be used for qualitatively reconstructing the optical “CV” curve of single NC, which even allows the accurate quantification of the facet area of NCs. From the optical “CV” curve, the UPD potential and peak width of every single NC can be precisely obtained, which is impossible to obtain in conventional ensemble measurements but is extremely important to understand and precisely control the UPD on NCs. We believe the highly sensitive EC-DFS technique offers itself a powerful tool for studying the surface chemistry at the atomic level. If this method can be further combined with a wide field imaging technique29–31, it may allow characterization of the activity and structure of NCs in a high throughput way.MethodsPreparation of Au NCsThree types of Au NCs were all synthesized by the wet chemical method with the same procedure and surfactant to ensure the only variable is the facet32. The procedure started with the synthesis of the seeds by adding 10 ml of 0.1 M CTAB and 103 μl of 1% HAuCl4 in the flask followed by stirring for 5 min while keeping the temperature at 30 °C. Then, 0.6 ml of ice-cooled 0.01 M NaBH4 was rapidly added into the flask. The color of the solution turned from yellow to dark brown. The seed was kept undisturbed for 2 h, then 1 ml of seed sol was diluted to 100 ml with the ultrapure water for further use.The growth solution was prepared by mixing 4 ml of 0.1 M CTAB, 1.5 ml of 0.1 M ascorbic acid, and 19.5 ml of ultrapure water into a colorimetric tube. After the solution has been stirred for 10 min at 30 °C, 41.2 μl of 1% HAuCl4 was added into the tube, and the solution were further stirred for 5 min. Then, 0.3 ml of diluted seed sol was rapidly added into the growth solution while stirring, and the solution was kept undisturbed at 30 °C for 12 h to obtain the seed sol of small octahedral Au NCs for the final step.Twenty microliters of 1% HAuCl4 was added rapidly into the 8 ml octahedral Au seed sol at 30 °C while stirring, and the mixture was left for 1 h to ensure the completion of the reaction. The procedure was repeated for seven times to obtain the octahedral Au NCs of 50 nm. The volume of 1% HAuCl4 is 50 μl to obtain the truncated octahedral Au NCs of 60 nm and was repeated three times. The volume of 1% HAuCl4 is 150 μl to obtain the cubic Au NCs of 50 nm without repetition.The size distribution of the three type of NCs can be found in Supplementary Fig. 8. The NCs were cleaned twice with ultrapure water before the electrochemical and spectral measurements. For the dark field measurements, the Au NCs sol were diluted 1000 times before dropped on the ITO electrode to obtain the well-dispersed sample with a large number of isolated single NCs. For the electrochemical measurements, the concentrated Au NCs sol were dropped on glassy carbon electrodes and dried in a vacuum chamber.Electrochemical measurementsSingle crystal electrodes were fabricated following the Clavilier method33. The obtained single crystal beads were further polished into half-beads to obtain large exposed single crystal surfaces. The single crystal electrodes were further electrochemically polished and flame-annealed before performing the electrochemical measurements.The NCs coated electrodes were prepared by dispersing NCs on glassy carbon electrodes. The NCs sols were first washed twice with ultrapure water by centrifugation. Afterwards, the concentrated NCs sols were dropped on glassy carbon electrodes, which were dried in a vacuum chamber. The electrochemical measurements were carried out right after the electrodes were taken out of the chamber to avoid the contamination.The ITO electrodes were cleaned by sonication with acetone (30 min), isopropanol (30 min), ethanol (30 min), and ultrapure water (10 min, three times). A solution of 1 mM Ag2SO4 and 0.05 M H2SO4 was used as the electrolyte for Ag UPD on Au NCs and single crystal electrodes.Dark field scattering spectra measurementsThe dark field measurements were performed on a home-modified confocal Raman microscope (Renishaw Invia, UK) that is equipped with upright (Leica DM2500) and inverted (Leica DMI3000B) microscopes. The annular white light (halogen lamp, 100 W) illumination was enabled by using a dark field condenser (oil immersion, NA 1.4–1.6). The scattering signal was collected by a water immersion objective (Nikon, ×60, NA 1.0), which were either selectively imaged on a color digital camera (Q-image EXI) or further dispersed to obtain the scattering spectra with a spectral CCD (Renishaw inVia) by a switchable mirror.Supplementary information
Supplementary Information | nature communications | [
"Article"
] | [
"Electrochemistry",
"Materials chemistry",
"Physical chemistry",
"Optical spectroscopy",
"Surface chemistry"
] | IntroductionEngineering nanomaterials high catalytic activity stability low cost eternal subject for electrocatalysis1–4 Altering electronic structure atoms topmost layer efficient challenging technique surface atoms level without strong binding ligands Underpotential deposition (UPD) enables controllable deposition foreign metal atoms submonolayer level clean surface unique physical chemical properties5–8 UPD strategy applied wet chemical synthesis nanoparticles controllable shape size existing NPs novel catalysts9–13 practice empirical understanding UPD mechanism limited to bulk single crystals challenge to study UPD process on individual NPs complex morphology adsorbed species unique surface energy situ transmission electron microscopy scanning probe microscopies develop sensitive electrochemical dark field scattering) setup for monitoring UPD process Ag on Au nanocrystals enables identification small spectral variations deposition submonolayer atoms reconstruct optical “cyclic voltammogram from spectral variation potential band width UPD process NCs displays energy difference facets work provides unprecedented information understanding UPD process NCs allows control UPD process single NP level for electrocatalysissetup for EC-DFS designed sensitive EC-DFS setup for challenging systems weak signals accurate potential control Fig. 1a oil immersion dark field condenser high NA 1.4–1.6 ITO electrode water immersion objective NA 1.0) polyethylene film dipped into electrolyte mode mismatch refractive index optical path excitation collection efficiency sensitivity EC-DFS technique improved descriptions in Supplementary Note 1) Figure 1b, c compare dark field images new conventional setups octahedral Au NCs (bright spots) Fig. 1b better visibility lower exposure time gain 1c improvement detection sensitivity clear comparing intensity scattering spectra from single NC background spectra substrate two setups (Fig. 1d, e). spectrum 1 s acquisition time new setup higher signal lower background conventional setup 10 s acquisition time 50-fold improvement signal to background ratio Supplementary Fig. 2) improvement enables detection spherical Au NPs diameter low 10–15 nm 50 nm without special light source equipment Supplementary Fig. 3. high sensitivity essential for identifying small spectral variation tracking dynamic processes electrochemical sensing other applications biosensing single NP catalysis.Fig.novel setup electrochemical dark field scattering technique schematic dark field microscopy spectroscopy water immersion objective working CE counter electrode RE reference electrode). Dark field image (b spectra (d new setup (c spectra (e conventional setup exposure times spectra d e 1 10 s 1 UPD on single octahedral Au tailoring electrocatalytic activity synthesis nanomaterials3 electrochemical scanning tunneling microscopy UPD bulk single crystal electrode difficult characterize UPD NPs surfaces used cyclic voltammetry investigate UPD process CV curve Ag deposition on Au nanooctahedrons-coated exposed Au (111) facet glassy carbon electrode (Fig. 2a b Fig. 2c identify apparent UPD peaks NPs coated narrow size distribution NPs high coverage compresses interparticle electrolyte thin layer silver ions deplete deposition continuous negative shift UPD peaks variation gap distances between NPs broad distribution UPD potential broadened UPD peaks effect not significant silver dissolution process UPD silver atoms on NPs surfacedissolution peak UPD from OPD broader on single crystal surface Fig. 2c). Single particle electrochemistry electrochemical features of NCs21–26 used EC-DFS technique Ag UPD process at single NP level (Fig. 1) ITO electrode with low density Au NCs used potential scanned from 0.500 to 0.362 V using voltammetry potential-dependent scattering spectra Fig. 2d from single NC acquired peak position intensity plotted function of potential peak position blue shifts from 0.413 V constant at potentials between 0.390 and 0.370 V significant blue shift at potentials lower than 0.370 V increase spectral intensity spectral shifts induced by Ag deposition from control experiments in solutions electrolyte lower concentration Ag ion derivative spectral shift time with potential curve Fig. 2e optical “CV”. two deposition peaks UPD OPD processes on single NC performed potential step experiment scattering spectra spectra same at 0.500 and 0.450 V potential reaches first deposition peak 0.400 V peak blue shifts seconds spectra remain same with negative movement potentialagrees with UPD process full monolayer achieved no further deposition scanning images show of Au nanooctahedrons after UPD potential unchanged excludes heterogeneous selective growth Ag morphological change potential reaches second deposition 0.360 V spectra show fast blue shift agrees with overpotential deposition (OPD process evidence from CV Ag deposition on Au (111) single crystal electrode (Fig. clean electrode distinguishes UPD from bulk deposition peak at 0.412 V with peak potential 12 mV more positive than first deposition peak lowering surface energy due surfactant modified Au (111) electrode surface with same surfactant repeated CV measurements UPD potential shifts from 0.412 to 0.402 V after modification surfactant agreeing with peak potential (0.401 V). 2e for single NC demonstrates first (0.401 second (0.362 deposition peaks in nanooctahedron correspond to UPD and OPD processes peak width of UPD peaks single Au nanooctahedron narrow as Au (111) electrode facet quality of fresh NCs good as single crystal electrodeFig. 2h histogram UPD peak positions 20 NCs distribution deposition potential 0.393 to 0.401 V peak width 5 mV different surface states NCs ensemble systems UPD potential distribution induced variation size NC small surface energy difference Supplementary Note 4) UPD potential similar distribution size Supplementary Fig. 8).Fig. 2Study Ag electrodeposit octahedral Au nanocrystals surface images large scale structural schematic Au nanooctahedrons ordered geometry electrode surface structural details voltammogram Ag underpotential deposition Au nanooctahedrons-coated glassy carbon electrode electrolyte 1 mM Ag2SO4 0.05 M H2SO4 potential-dependent scattering spectra single Au nanooctahedron zoom-in spectra variation optical “CV” Au nanooctahedron peak position scattering spectra applied potential time-dependent peak position change scattering spectra potential controls 0.500 to 0.400 V 50 mV 0 to 0.360 V 10 mV). Ag deposition Au (111) electrode before after adsorption surfactant peak potential statistics UPD peaks optical “CV”s 20 Au nanooctahedronsscale bars 200 nm a 50 nm b.Ag UPD cubic Au NCs performed experiments cubic Au NCs enclosed {100 facets (Fig. 3b). Figure 3c shows CV Ag deposition Au nanocubes-coated electrode UPD peak not discernible average effect change single NC scattering spectra observed after Ag deposition peak intensity change occurs prior spectral shift more sensitive Ag deposition Au nanocubes scattering spectra nanooctahedrons nanocubes respond differently silver deposition aspect ratio change determines responses nanooctahedron nanocube systems nanooctahedron aspect ratio decreases after silver deposition spectral shift nanocubes aspect ratio change silver deposition isotropic increased size nanocubes leads increase scattering intensity reconstruct electrochemical “CV” single nanocube first-order derivative intensity change with time Fig. 3e “CV” curve similar shows two deposition peaks first deposition peak attributed UPD process Au (100) facet verified potential step experiment 3f). negative shift UPD peaks single Au nanocube (0.376 V) compared bulk Au (100) single crystal electrode (0.due to adsorbed surfactants (Fig. 3g on NCs histogram UPD potentials 20 nanocubes (Fig. 3h broader distribution nanooctahedrons better quality {111} facet {100} NCs surface onset potential second deposition peak same for two NCs Ag deposited over Ag layer first deposition peak preferential Ag on vertex NCs higher energy challenging to see single atomic layer deposition vertex metal simulated scattering spectra Au nanooctahedrons nanocubes before after deposition Ag different thicknesses Supplementary Fig. 10 spectral change Ag deposition vertex Ag deposits on vertex NCs lead red shift spectra facet blue shift experimental results first deposition peak from facet vertex.Fig. electrodeposition Ag on cubic Au nanocrystals surface images ordered geometry Au nanocubes electrode structural schematic nanocube CV Ag UPD on Au nanocubes-coated electrode electrolyte 1 mM Ag2SO4 0.05 M H2SO4 scattering spectra single Au nanocube under different potentials zoom-in spectra variations reconstructed optical “CV” single Au nanocube two different deposition processestime-dependent peak intensity change scattering spectra single Au nanocube potential controls 0.500 to 0.400 V 50 mV 0.380 10 mV 0.380 to 0.360 V 15 CVs Ag deposition Au (100) electrode before after cyan curve adsorption surfactants peak potential statistics UPD bands optical “CV”s 20 Au nanocubes scale bars 200 nm a 50 nm b UPD process Ag deposits truncated octahedral Au NCs two studies demonstrate applicability method NCs one type facet UPD processes NCs different facets applied method UPD process truncated octahedral Au NCs (Fig. 4a–c {111} {100} facets spectral peak positions intensities single TO NC spectral shift 35 mV ahead intensity change used spectral shift reconstruct optical “CV” single NCs Fig. 4d three deposition peaks at 0.400 0.374 0.364 V optical first second peaks identified UPD processes facet {111} {100} third peak OPD process histogram analysis shows UPD peak potentials 0.400 0.376 V {111} {100} facetspotentials 0.397 V nanooctahedrons 0.378 V nanocubes possible charge transfer facets minor difference UPD potentials facets NCs modify foreign atoms special electronic property chemical activity.Fig. 4Study Ag electrodeposit truncated octahedral Au nanocrystals schematic truncated octahedral Au NC structural details SEM images high low magnification ordered structure optical “CV” single octahedral Au NC reconstructed spectral shift scattering spectra different potentials peak potential statistics UPD bands {111} (red) {100}) facets optical “CV”s 20 NCs distribution area ratio {111}/{100} NCs optical “CV”s SEM image time-dependent peak position change scattering spectra octahedral Au NC potential controls schematics UPD processes facets scale bars 200 nm b 50 nm c calculated area ratios {111} to {100} spectral shift Peak 1 to Peak 2 Fig. 4d Supplementary Note 5) ratios plotted number NCs Gaussian function (Fig. facet area ratio 3.15 close 3.45 SEMknown facet ratio particle size facet areas quantified single NP level high accuracy by EC-DFS interference double layer charging oxygen reduction reaction obscured CV assembled TO Au NCs Supplementary Fig. 11b). method characterizing NCs with complex facets nanorod27 bipyramid28) screening structure catalyst challenging electron microscopy designed sensitive EC-DFS technique small spectral variation UPD process single NP level results sensitive observation monolayer atoms facets single NCs electrochemical environment spectral change reconstructing optical “CV” curve single NC accurate quantification facet area NCs UPD potential peak width NC obtained impossible conventional ensemble measurements important understand control UPD NCs highly sensitive EC-DFS technique powerful tool studying surface chemistry atomic level combined with wide field imaging characterization activity structure NCs high throughput.MethodsPreparation Au NCsThree types NCs synthesized wet chemical method same procedure surfactant only variable facet32 procedure synthesis seeds adding 10 ml 0.1 M CTAB 103 μl 1% HAuCl4 flask stirring 5 min temperature 30 °C 0.6 ml ice-cooled 0.01 M NaBH4 added flaskcolor solution turned yellow to dark brown seed kept undisturbed 2 h 1 ml diluted to 100 ml ultrapure water growth solution prepared 4 ml 0.1 M CTAB 1.5 ml 0.1 M ascorbic acid 19.5 ml ultrapure water colorimetric tube stirred 10 min 30 °C 41.2 μl 1% HAuCl4 added stirred 5 min 0.3 ml diluted seed sol added kept 30 °C 12 h small octahedral Au NCs microliters 1% HAuCl4 added 8 ml octahedral Au seed sol 30 °C left 1 h repeated seven times octahedral Au NCs 50 nm 1% HAuCl4 50 μl truncated octahedral Au NCs 60 nm repeated three times 1% 150 μl cubic Au NCs 50 nm size distribution NCs Supplementary Fig. 8. NCs cleaned twice ultrapure water before electrochemical spectral measurements measurements diluted 1000 times ITO electrode sample single NCs electrochemical concentrated Au NCs dropped glassy carbon electrodes dried vacuum chamber crystal electrodes fabricated Clavilier beads polished into half-beads large exposed surfacessingle crystal electrodes polished flame-annealed before measurements NCs coated electrodes prepared dispersing NCs on glassy carbon electrodes washed twice with ultrapure water concentrated dropped on carbon electrodes dried in vacuum chamber electrochemical measurements after electrodes contamination ITO electrodes cleaned by with acetone isopropanol ethanol ultrapure water solution 1 mM Ag2SO4 0.05 M H2SO4 used electrolyte for Ag UPD on NCs crystal electrodes field scattering spectra performed on home-modified confocal Raman microscope upright inverted DMI3000B microscopes annular white light (halogen lamp 100 W illumination enabled dark field condenser scattering signal collected by water immersion objective imaged on color digital camera or dispersed with spectral CCD switchable mirror.Supplementary information | 49.3 | 0.53312 |
10.1038/s41467-020-18974-9 | PMC7576607 | Functional connectivity measured from fMRI data is widely used in neuroscience. Here the authors report an association between two types of breathing signature and obtained BOLD data, and associated sex differences. | Resting state functional connectivity magnetic resonance imaging (fMRI) is a tool for investigating human brain organization. Here we identify, visually and algorithmically, two prevalent influences on fMRI signals during 440 h of resting state scans in 440 healthy young adults, both caused by deviations from normal breathing which we term deep breaths and bursts. The two respiratory patterns have distinct influences on fMRI signals and signal covariance, distinct timescales, distinct cardiovascular correlates, and distinct tendencies to manifest by sex. Deep breaths are not sex-biased. Bursts, which are serial taperings of respiratory depth typically spanning minutes at a time, are more common in males. Bursts share features of chemoreflex-driven clinical breathing patterns that also occur primarily in males, with notable neurological, psychiatric, medical, and lifespan associations. These results identify common breathing patterns in healthy young adults with distinct influences on functional connectivity and an ability to differentially influence resting state fMRI studies. | IntroductionFunctional magnetic resonance imaging (fMRI) scanning of subjects at rest has become a major neuroimaging paradigm, termed functional connectivity or resting state MRI1. In these scans, subjects lie quietly, often staring at a crosshair, for 5–15 min or more, performing no particular instructed task while fMRI data are acquired. Correlations in task-free fMRI signals are thought to reflect the functional relatedness of the tissues producing those signals, and the spatial topography of signal correlations has been leveraged to yield new and increasingly refined macro-scale maps of the human brain2–4. These scans have the potential to deliver diagnostic and prognostic information, and large studies are now underway that use resting-state fMRI scans as cornerstones of the datasets, e.g., the ABCD study scanning 10,000 children for biomarkers of developmental trajectories5.Breathing modifies the concentration of carbon dioxide in arterial blood, which is a potent modulator of cerebral blood flow, and thus the fMRI signal6,7. When subjects breathe deeply or quickly (i.e., hyperpnea) they exhale more CO2, arterial pCO2 drops, cerebral blood flow decreases, and fMRI signals decrease; conversely, if breathing is shallow or slow (or stopped) (i.e., hypopnea or apnea), less CO2 is released, arterial pCO2 rises, cerebral blood flow increases, and fMRI signals increase. In this manner, breathing patterns can influence resting-state fMRI scans.Breathing occurs in multiple forms. The basic respiratory rhythm is a cyclic rhythm termed eupnea, which moves a tidal volume of air into and out of the lungs in each breath. But a variety of deviations from eupnea exist, from the sighs exhibited by all humans that reinflate collapsed portions of the lung, to yawns of boredom or sleepiness, to more marked forms of disordered breathing, including cluster breathing, ataxic breathing, or periodic breathing (e.g., Hunter–Cheyne–Stokes), forms of respiration often associated with heart disease or neurological injury6. Beyond having different generative neural mechanisms8, different kinds of breathing may have distinct biophysical correlates and consequences for neuroimaging.Little is known about the breathing characteristics of healthy young adults lying at rest in an MRI scanner, the kind of subject that forms the backbone of the functional connectivity literature, despite the potential for breathing to systematically influence fMRI signals. To address this issue, we jointly examined respiration and fMRI signals in a large, publicly available data set of healthy young adults with large amounts of scan time per subject, the Young Adult release of the Human Connectome Project (HCP). In this report, we describe effects seen in 440 h of scanning in 440 subjects (ages 22–36, mean 28.6, 228 males, 212 females). Such quantities of data stand in contrast to the prior fMRI literature on respiration, which usually involved small numbers of relatively short recordings9–12.The sheer size of the Young Adult HCP data set provides an unprecedented window into the respiratory behavior of humans quietly resting in scanners. In these subjects, beyond eupnea, we find two prevalent patterns in respiration with distinct correlates in fMRI signals and distinct influences on functional connectivity. One pattern represents isolated deep breaths and is not sex-biased. The other, termed bursts, is sex-biased, and we link this pattern to sex-biased breathing patterns in the respiratory literature. Patterns were congruently recognized by human raters and an algorithmic scoring system. Collectively, these results demonstrate a prevalent and sex-biased form of breathing in healthy young adults with substantial influence on functional connectivity measures that resemble a form of breathing traditionally studied in older, medically ill patients. The clinical literature suggests that these breathing patterns will be influenced by sex hormones, by age, cardiovascular, neurological, and psychiatric illness, among other factors.ResultsTwo distinct breathing patterns in subjects at restWe begin by presenting individual instances of respiratory patterns, then group descriptions, then demographic differences in pattern prevalence, and then the spatiotemporal effects of the breathing patterns and their distinct consequences for fMRI signal covariance. We focus initially on visual presentations, for these were how we first recognized the patterns.To detect influences of respiration on resting-state fMRI signals, we created and viewed plots of 1760 scans representing 440 h of scanning in 440 young adults. In addition to eupnea, we came to recognize two common patterns of respiration. One pattern was known to us, which we term a single deep breath, a lone breath considerably larger than the surrounding breaths. The other pattern was unfamiliar to us, and is undescribed in the neuroimaging literature to our knowledge; we call it a burst pattern.The two patterns are illustrated in Fig. 1. The fMRI scans are flattened into grayscale heat maps, with all in-brain voxels defining the Y axis and time on the X axis. Signals from gray matter voxels are above the bright green lines, and from white matter and ventricles below. The respiratory belt trace is shown in blue, and several deep breaths are marked by orange arrows in an otherwise eupneic scan in Fig. 1a. Three respiratory measures derived from the respiratory trace often used to model respiratory effects in the fMRI literature, are also shown (ENV, RV, and RVT, respectively, gauging the envelope of the belt trace, windowed variance in the trace, and the rate of air movement), often displaying abnormalities at these deep breaths. In the fMRI signals, throughout the gray matter, there are brief signal increases (vertical white bands) just after the deep breath is taken, followed by prominent signal decreases (vertical black bands). Figure 1b illustrates a burst respiratory pattern: a serial, rhythmic set of tapers in breathing depth (this example has apnea between bursts), with rhythmic correlates in fMRI signals. Fuller versions of these images, including fMRI signals before and after denoising, are shown in Supplementary Figs. 1 and 2, illustrating that signal effects are present both before and after FIX-ICA denoising and that the patterns are also linked to head motion and image quality measures. These scans were chosen for their stark examples of the breathing patterns, but there are many forms of deep breaths and bursts, illustrated in Fig. 1c, d and later figures. In something as straightforward as a single deep breath (e.g., Fig. 1c), a variety of waveforms are possible, including floor and ceiling effects and slippage of the respiratory belt at peak inspiration; readers wishing to gain skill in recognizing patterns are encouraged to consult Supplementary Note 1 and to view all 1760 gray plots in Supplementary Movie 1 (1.4 GB, download at https://osf.io/u35f8/).Fig. 1Gray plots of scans containing deep breaths and bursts.Gray plots containing a deep breaths and b bursts. Upper panels show the z-scored respiratory belt traces in blue (y-ticks at left are z = −1 and 1), and 3 commonly derived respiratory measures (ENV, RV, and RVT, with vertical offsets to enable non-overlapping visualization; scales are identical in all figures). In the grayscale heat maps, all in-brain fMRI signals are shown organized by anatomical compartment, with a green line separating gray matter from white matter and ventricle signals. In a, three deep breaths are indicated by arrows, with major decreases (vertical black bands) in fMRI signal following each of the breaths. In b, over two dozen bursts are present (arrows mark several), with accompanying modulations of fMRI signals. Supplementary Figs. 1 and 2 show comprehensive versions of each scan. Additional exemplars of deep breaths (c) and bursts (d) are shown at the bottom. The code to produce gray plots was published in ref. 49, and comprehensive gray plots of all 17,640 scans are shown in Supplementary Movie 1.To help convey the variety of respiratory waveforms denoting single deep breaths, five instances are shown in Fig. 2a (fuller images are shown in Supplementary Fig. 3, see also dozens of scans with deep breaths marked in Supplementary Movie 2). These isolated, deep breaths are often accompanied by subsequent breathing pauses of variable duration (often just a few seconds but sometimes lasting 10 or 20 s; brief central apneas are known sequelae of deep breaths13). In each of these instances, after an initial delay, a black band in the gray plot reflects a pan-brain decrease in fMRI signals lasting until ~30 s after the breath, consistent with a cerebral blood flow decrease after a transient increase in ventilation.Fig. 2Examples of deep breaths and bursts.Plots are formatted as in Fig. 1. Note in deep breaths a the slowness of the breaths, the possibility of transient apnea afterward, the incongruence of the respiratory measures ENV, RV, and RVT, and the presence of fMRI signal decreases in each instance. Note in bursts b the repeated, serial modulation of breathing amplitude, the congruency of ENV, RV, and RVT, and the rhythmic correlates in fMRI signals.To help convey the variety of respiratory waveforms denoting bursts, five instances are shown in Fig. 2b (fuller images are in Supplementary Fig. 4, see also dozens of scans with bursts marked in Supplementary Movie 3). In this pattern, a burst of deep breaths occurs which tapers into shallow breaths, often serially followed by additional bursts. These burst respiratory patterns differ from single deep breath patterns in several respects. First, whereas single deep breaths often occur in isolation, series of bursts often span several minutes at a time, with individual bursts often lasting ~30–50 s. Second, the burst patterns are usually quite evident in ENV, RV, and RVT traces, which all tend to concordantly display large wavelike modulations (in contrast, note the lack of concordance in some deep breaths of Fig. 2). Third, the typical fMRI signal response is an initial signal increase followed by a prolonged decrease (see orange lines marking brief white then longer black bands), with durations approximately matching those of the respiratory burst with an added lag of signal decrease. The subject at the bottom is shown for the entirety of a scan, illustrating how eupnea evolves into a burst pattern, with the emergence of fMRI signal correlates.Each subject had four 14.4-min long scans, and it was plain that, within a subject, one scan could display normal tidal breathing, but a different scan could contain markedly different breathing, with accompanying differences in fMRI signals (see two examples in Supplementary Fig. 5). It was also plain that, within scans, different breathing patterns could dominate at different times (see two examples in Supplementary Fig. 6). Because deep breaths and bursts were both prevalent, and organized, we focused on these respiratory events (though other forms of more disorganized breathing exist, see an example in Supplementary Fig. 5, upper right).Properties of respiratory patternsTo more formally characterize deep breaths and bursts, the onsets of 35 bursts, 35 deep breaths, and 35 non-respiratory head motions were visually identified (Supplementary Data 1 lists onsets; Supplementary Fig. 7 illustrates 6 onsets of each kind in gray plots, see Supplementary Movies 2–4 for all onsets marked in gray plots). The non-respiratory motion onsets were identified to address the possibility that the fMRI signal changes during deep breaths or bursts related to head motion from breathing. A random onset in a random scan of the motion subjects was also set as a control condition. Relevant signals were extracted from 30 s prior to 60 s after onsets, illustrated in Fig. 3a, with statistical contrast to random onsets shown as a thin gray-red heatmap under the main heat maps (coloring only t-tests of p < 0.001). The basis of burst and deep breath identification is evident in the respiratory belt heat maps. Plots in Fig. 3b show mean values across certain events. Bursts show marked, congruent signatures in ENV, RV, and RVT (Fig. 3b). Deep breaths also have marked signatures, but they differ across measures, with RVT having no mean positive deflection (explained in ref. 14). Global fMRI signals differ across patterns, with deep breaths on average displaying a brief, steep signal increase then a marked signal trough with nadir near 15 s after onset, and resolution 30 s after onset. Bursts have slower trajectories on average, peaking later and higher, and exhibiting prolonged troughs with nadir over 20 s after onset, with a resolution around 40 s after onset. Individual event durations are naturally modulated by the depth of breathing, the presence and duration of apnea after deep breaths, and the duration of a burst taper. We selected a non-respiratory-motion group as a control because deep breaths display considerable motion at the onset of the breath, evident in the head motion and DVARS heat maps. However, neither the motion-displaying group nor the random group showed any global fMRI signal fluctuations, effectively ruling out motion as a cause of the patterned global signals. That motion did not produce global changes is consistent with the fact that multi-echo studies show that global fMRI signals are overwhelmingly T2* signals (compatible with respiration), not S0 artifacts caused by head motion15. Heart rate is routinely elevated for several seconds after deep breaths, whereas no average effect is noted for bursts. Because deep breaths very reliably elevate heart rates, we were surprised that deepened breathing in bursts did not produce much modulation; review of individual scans indicates that some subjects reliably display cardiac modulation by bursts, but others display no modulation (Supplementary Fig. 8). Other non-HCP resting-state fMRI datasets also exhibit bursts, and, in those datasets, as in the HCP data, there is not a reliable link between bursts and heart rate modulation (see Supplementary Note 2 and Supplementary Fig. 9).Fig. 3Properties of deep breaths and bursts.a Heat maps at the top illustrate 90-s segments surrounding visually identified events. In different subjects, 35 examples each of bursts, deep breaths, and isolated non-respiratory motions were identified from respiratory belt traces and motion traces. In the motion-exhibiting subjects, a random set of timepoints was also selected. Illustrated are respiratory belt traces, three respiratory measures (ENV, RV, and RVT), the global (gray matter average) fMRI signal, head motion (framewise displacement, following filtering, and 4-TR calculation as in ref. 16), and DVARS (z-scored). A gray/red heatmap represents statistically significant differences from the random events (two-sample t-test, two-tailed) beyond p < 0.001, illustrated on a logarithmic scale capped at p < 1e−10. The basis of bursts and deep breaths are apparent in the respiratory belt images. b Mean signals of ENV, RV, and RVT congruently mark bursts, but deep breaths display differences across the respiratory measures, with RVT having little positive deflection. c Mean global fMRI signals differ across patterns: deep breaths have brief signal increases and marked signal decreases with nadir near 15 s, and 30 s to resolution, on average. Bursts have more marked positive deflections, and slower timecourses on average, resolving near 40 s on average. Motion produces no global fMRI signal changes. Shade plots reflect mean and std. d Deep breaths display considerable motion and DVARS changes time-locked to event onsets; bursts have smaller time-locked modulations that do not achieve significance. Source data are provided as a source data file.Group contrasts reveal a sex bias in burst patternsTo better understand why the respiratory patterns occur, we sought subject-level factors that scaled with these two respiratory patterns. The HCP data set has hundreds of behavioral, demographic, physiologic, and imaging measures for each subject. To mine such information, we needed either groups to contrast displaying different breathing patterns, or numeric indices of patterns (e.g., for regression). We pursued both paths simultaneously but prioritized the group contrast approach because we could carefully select and thus confidently characterize breathing patterns in the groups.For exploratory purposes, we defined three groups with clearly different breathing patterns: subjects with unambiguous, marked bursts in most or all scans and few or no single deep breaths (the burst group), subjects with unambiguous, marked single deep breaths in most or all scans but few or no bursts (the deep breath group), and subjects whose scans displayed neither bursts nor single deep breaths (the clean group). Subjects were selected based on respiratory belt traces and signal heat maps alone, without knowledge of other properties of the subject. This procedure identified three groups of 21 subjects, all unrelated. Subject identity is restricted due to associations with psychiatric instrument scores and substance use in the groups (described below); investigators with HCP Restricted Access will find a Subject Key associated with this paper identifying the groups. Full details of the statistical contrasts of these groups are reported in Supplementary Note 3, Supplementary Fig. 10, and Supplementary Data 2. Instructions to access Subject Keys are in the “Methods” section.The three groups were associated with far more HCP variables than expected by chance, including alcohol use (bursts), cigarette use (deep breaths), and strikingly included dozens of structural imaging variables that differed uniformly by group, distinguishing the burst group (thinner cortex) from other groups (smaller brains) (Supplementary Fig. 10). These findings were all subsumed, and nearly all explainable, by the following fact: males were 6/21 of the clean group, 5/21 of the deep breath group, and 14/21 of the burst group. The groups were formed without knowledge of the sex of participants, and it is very unlikely (joint probability p = 3.3e−5) that such unbalanced sex compositions would emerge three times in random group formation.Ratings of gray plots reveal a sex bias in burst patternsIn parallel with the group analyses, authors J.D.P. and C.J.L. independently rated 1596 scans (subjects 1–399) after training together on subjects 400–440, making binary decisions on the presence of deep breaths and bursts in each scan. Ratings were made purely in terms of gray plots without knowledge of any demographics, including sex. The group results above were discovered after 100 subjects had been rated (with good-to-excellent inter-rater reliability, Cohen’s kappas were 0.79 for bursts and 0.73 for deep breaths). Significant sex differences in bursts but not deep breaths were present within these first 100 subjects for both raters, were again separately present in the next 299 subjects rated for both raters, and were also present when subjects in the above-defined groups were excluded and/or when only one subject per family contributed. For simplicity, we report ratings of the entire 399-subject cohort.Identical rater decisions were made in 87% of scans on bursts and in 89% of scans on deep breaths, yielding Cohen’s kappa values of 0.73 and 0.78 overall. Due to the number of ambiguous decisions that must be made in subtle instances of patterns, or amidst disorganized and chaotic breathing styles, we prioritized the totals over four scans in each subject, referred to as pattern scores. These scores correlated at r = 0.86 (p < 1e−20) for bursts and 0.90 (p < 1e−20) for deep breaths between raters, illustrated in Fig. 4a. These numbers indicate that human raters reliably recognize the breathing patterns; interested readers may use the Supplementary pattern training module to learn the patterns (download at https://osf.io/u35f8/).Fig. 4Rater scorings and algorithmic indices detect sex effects and global functional connectivity influences.a Plots of total scans (of 4) with bursts and deep breaths for both raters, with score correlations and Cohen’s kappas inset, for N = 399 subjects (all panels display results from N = 399 subjects except d and e, which concern the three 21-subject groups). b Histograms of scores across subjects for both patterns, showing raters by color. c Bar graph of percent scans of each sex displaying patterns. Chi-squared tests of bursts yield p = 3.4e−8 and 8.8e−7 for J.D.P. and C.J.L. (denoted by ***), effects unchanged by excluding members of the three groups. No significant differences are seen by sex in deep breath scores. d Bar plots showing mean values with std error bars of the ratings in a clean, burst, and deep breath groups (each with N = 21 unrelated subjects). B and D denote burst and deep breath. Desired respiratory properties are found in each group. e Algorithm indices of the three 21-subject groups, corroborating rater scores and confirming desired breathing patterns (compare with d directly above). Box plots show median and 25th and 75th percentiles as boxes, whiskers encompass 99% of normally distributed data, outliers are individually marked (all box plots in later panels follow this format). f Algorithm indices of breathing patterns by sex, with significant differences by two-sample t-test in bursts but not deep breaths (compare with c directly above). g Box plots of algorithm indices for each pattern as a function of mean rater score, demonstrating significant Pearson correlations of humans and algorithm ratings. h Box plots of gFC as a function of mean pattern scores, showing much stronger effects of bursts on gFC, quantified by Pearson correlation. i Betas of multiple linear regression of pattern scores in gFC (gFC = b0 + b1*burst_score + b2*deep_breath_score), performed in each sex separately, showing much stronger effects of bursts. Bars show mean values, error bars show 95% confidence intervals; fits do not differ by sex (both n.s. by two-sample two-sided t-test, uncorrected for multiple comparisons). j Box plots of gFC and head size (intracranial volume, ICV) by sex, both significantly different by sex by two-sample two-sided t-test (p = 9.1e−9 and <1e−20, respectively). k Color chart of significance of main effects of multiple ANCOVA models. Sex effects become insignificant when both head size and respiratory variables are modeled. Source data are provided as a source data file, though group identity is redacted.Deep breaths occurred in about 85% of subjects, and it was common for subjects to display deep breaths in most or all scans (Fig. 4b). Bursts, on the other hand, were absent in about 30% of subjects, and it was relatively uncommon for subjects to display bursts in all scans. In both sets of ratings, chi-squared tests for sex differences were significant for bursts (p = 3.4e−8 and 8.8e−7) but not for deep breaths (Fig. 4c), effects unchanged when excluding the groups mentioned above. Overall, bursts were identified in 45% of male scans and 35% of female scans, whereas deep breaths were identified in 54% of male scans and 52% of female scans. Scores of each pattern were uncorrelated across all subjects, for each rater, and within each sex (r < 0.1 for all). Thus, the breathing patterns appear to occur independently across subjects. The pattern scores within the 3 previously contrasted groups accord with the desired group breathing properties (Fig. 4d).Automated detectionTo begin to move beyond rater decisions, which are tedious and time-consuming, we devised an algorithm to index breathing patterns based on joint information in respiratory traces and global fMRI signals. The algorithm creates probabilities of respiratory patterns from the respiratory belt traces and multiplies these probabilities with the match of global fMRI time series to templates of deep breaths and bursts, thus requiring simultaneous evidence from both sources (respiratory trace and global fMRI signal) to begin indexing patterns. This approach performs well in many situations, but haphazard, disorganized breathing can cause high indices on either pattern, thus causing us to algorithmically discount scans with markedly variable respiratory rates. This algorithm recaptures the significant differences in group breathing styles (Fig. 4e) and the significant sex difference in bursts but not deep breaths (Fig. 4f). Indices significantly correlate with rater scores on both patterns (Fig. 4g). These results collectively give good confidence in ratings and group formations and demonstrate a proof-of-principle algorithmic approach to this issue. A fuller description of the algorithm, with illustrations in single scans, is in Supplementary Note 4.Influence of respiratory patterns on global covarianceRespiratory events, by influencing cerebral blood flow, add broadly shared variance to all voxel signals, seen repeatedly in this paper as the white and black vertical bands in gray plots. To index such global effects on functional connectivity, in each subject, the median correlation of all gray matter voxel signals in each of the four scans was calculated in the minimally preprocessed data, and the mean of these values over all scans was computed (termed global functional connectivity, gFC).Robust increases of gFC are seen with increasing rater scores, and the scaling is stronger for bursts (r = 0.53, p < 1e−20) than for deep breaths (r = 0.18, p = 4e−4) (Fig. 4h). Similarly, algorithm indices scale with gFC, more for bursts (r = 0.59, p < 1e−20) than for deep breaths (r = 0.29, p = 7e−9). The higher values for the indices relative to scores may reflect their ability to scale with prevalence within-run (rather than binary rater decisions), or the fact that the indices incorporate template fits to global fMRI signals. Multiple linear regression of scores and indices yielded betas twice as high for bursts as for deep breaths and additionally demonstrated that fits to gFC for each pattern did not differ by sex (Fig. 4i shows betas (slopes) for score fits to gFC by sex).Though bursts and deep breaths produce the same effects in gFC in each sex, because bursts are more common in males, gFC may be increased in males relative to females. As Fig. 4j shows, males do have higher gFC. However, males also have larger heads with brains closer to scanner receive coils, meaning signal-to-noise ratios may differ by sex as well, providing an additional potential explanation for gFC differences. We therefore modeled gFC via ANCOVA as a function of sex, head size, and respiratory variables, and only eliminated sex differences when both head size and respiration were accounted for (Fig. 4k). Though motion does not cause global signals (which should largely drive gFC), as a precaution we also added motion covariates (FDoriginal, FDfiltered, and FDfiltered,4-TR, following16) and the data quality covariate DVARS to models 1–4; these additions did not eliminate significant gFC sex differences in any model and often failed to significantly fit as main effects when respiratory variables were present.Spatiotemporal effects of bursts and deep breaths in fMRIWe next asked whether there were non-global profiles of the breathing patterns in functional connectivity and whether such profiles differed by pattern. We first focused on covariance during the breathing patterns. Using the sets of 35 events from Fig. 3, we extracted the time series spanning −10 to +40 s about the onsets and computed mean correlation matrices in a commonly used parcellation scheme17, illustrated in Fig. 5a. Permutation tests among patterns yielded significant differences of each pattern from random onsets (only cells significant at p < 0.05 by 10,000 permutation tests are colored; nearly all cells are significant), and from each other, shown in Fig. 5b for several versions of signal processing. Mean signals within the resting state networks are plotted in Fig. 5c, illustrating the basis of the correlation matrices. Several spatially specific effects are present. For the present purposes, two points are emphasized. First, in all forms of signal processing, significant spatially specific effects exist. Second, in the minimally preprocessed data (and in FIX-ICA-denoised data), which best represents the original respiratory effects, there is a striking elevation of correlations in a primary sensory and motor distribution encompassing visual, auditory, motor, and somatosensory cortex (dotted ovals in Fig. 5b). Signals in these networks peak high and early (dotted ovals in Fig. 5c), and have deep and early troughs, relative to other networks. For completeness, time series are also shown on a brain surface in Fig. 5d (and in Supplementary Movie 5), comprehensively illustrating both global and focal effects in each pattern.Fig. 5Spatiotemporal properties of bursts and deep breaths.a Color legend of network locations and colors from ref. 17, with text labeling of the networks of particular interest for this paper (full legend in Supplementary Fig. 11). b Correlation matrices are derived from spans of −10 to 40 s about the event onsets shown in Fig. 3 in minimally preprocessed data (red dotted lines in c), and show mean differences of 35 bursts and 35 deep breaths compared to 35 random onsets, only coloring cells significant at p < 0.05 by 10,000 permutation tests (nearly all cells are significant; non-significant cells are colored gray). In grayscale matrices at right, matrices of bursts were contrasted to deep breaths via 10,000 permutation tests, and the top and bottom 2.5% of actual differences among permutation ranks are illustrated (in white and black; gray is insignificant) in matrices for minimally preprocessed (MP), FIX-ICA-denoised (Post-FIX), and minimally preprocessed time series plus global signal regression (GSR). Differences exist under each processing strategy, prominently including visual, auditory, and somatomotor cortex (blue, pink, orange, and cyan). c Mean signals of 35 patterns from each kind of time series in b, with mild smoothing, colored by the legend above. Dotted ovals encircle the peaks and troughs of the aforementioned sensorimotor networks. d Surface representations of the events, the same data in b and c, in minimally preprocessed time series. Supplementary Movie 5 animates these patterns and those of non-respiratory motion onsets and random onsets.Covariance associated with bursts and deep breathsWhereas Fig. 5 focused on effects during breathing patterns, Fig. 6 focuses on correlation structures of entire scans associated with (but not necessarily caused by) the breathing patterns. Effects in minimally preprocessed data are the focus, but some major data processing strategies are also shown. As in Fig. 5, all matrices only color effects significant at p < 0.05 by 10,000 permutation tests (gray is used for insignificant cells).Fig. 6Functional connectivity associated with breathing patterns.All images color only contrasts or differences significant at p < 0.05 by 10,000 permutation tests (gray cells are insignificant). a Contrasts of the three groups. The top row shows mean differences between groups. The bottom row shows the rank of mean group correlations amidst random, unrelated groups drawn from the entire cohort. b Mean within-subject differences between scans without breathing patterns (B−D−) and scans with bursts (B+D−, top row) or deep breaths (B−D+, bottom row) (using only scans where both raters fully agreed). c Betas of multiple linear regression performed separately in each sex, using unrelated, non-group-member subjects only. Regressors were z-scored, and all betas are shown for minimally preprocessed (MP) data. Betas for bursts alone are shown for several other data processing strategies (Supplementary Fig. 11 shows full sets of betas). In each of these three main analyses (a, b, c), bursts strongly associated with an increase in sensorimotor correlations (yellow dotted circles), and deep breaths lack elevation in these regions (blue dotted circles).First, we examined the correlation structures of the three groups (Fig. 6a). The burst group had correlations broadly elevated above both the clean and deep breath groups, with especially high elevations in the sensorimotor distribution (dotted yellow circles). The deep breath group had broadly higher correlations compared to the clean group, notably avoiding the sensorimotor distribution (dotted blue circles), despite bursts having no role in these scans or contrasts. Comparing the groups to randomly formed groups drawn from all unrelated subjects recapitulated these findings; these latter analyses can be viewed as contrasts of relatively pure breathing patterns with typical breathing (i.e., randomly selected baseline mixtures of the breathing patterns).Second, we examined within-subject differences between scans with neither respiratory pattern and scans with either bursts (Fig. 6b, top row) or deep breaths (bottom row). Within-subject burst effects are widespread elevations with sensorimotor emphasis (dotted yellow circles) and are present in minimally preprocessed and FIX-ICA-denoised data, and also in data that undergoes motion regression and censoring (third column). When global signals are removed, significant effects persist but in an altered distribution, reflecting the fact that a lagged signal structure exists in bursts that is not captured in the mean signal. This respiratory effect becomes more pronounced when motion regression and censoring are performed along with global signal regression, underscoring that these patterns are not motion-caused signals. Within-subject deep breath effects, on the other hand, are widespread elevations without global signal regression, which are largely eliminated with global signal regression, reflecting the fact that the major modulation is tightly time-locked and similar in all networks. There is a hint of the sensorimotor non-elevation seen in the group contrasts in certain processing strategies (dotted blue circles).Third, we examined betas of multiple linear regressions across subjects, separately in each sex (males in the top row of Fig. 6c, females in the bottom row, only unrelated, non-group subjects used). Very similar spatial beta structures were seen in males and females (compare top and bottom rows). Again, bursts associated with global elevations that were especially pronounced in sensorimotor distributions (dotted yellow circles), and deep breaths associated with milder widespread elevations that conspicuously avoided sensorimotor distributions (dotted blue circles). Nuisance effects of DVARS and head motion are congruent by sex and distinct from those of respiratory patterns in all processing strategies (best illustrated in Supplementary Fig. 11).For convenience, matrices from Figs. 5 and 6 are arranged by respiratory pattern pattern in Supplementary Fig. 12, along with related effects in instructed breathing paradigms.Potential influences of sleep or arousalHere, we attempt to document relationships of the respiratory patterns to arousal or sleep, factors that are well-known to modify breathing. Our ability to address these questions is limited in HCP data, which includes no hard measure of arousal or sleep, but several expected associations of breathing and sleep can be (indirectly) tested. It should first be stated that studies in young adults, whether by polysomnography or by survey, do not find sex differences in delay to sleep in the age range of the HCP subjects (though sex differences in delay to sleep do emerge later in life, after menopause)18,19. Large studies of excessive daytime sleepiness also routinely find no sex effect20,21. And studies of sleep onset during fMRI in young adults do not report sex effects22. There is thus little a priori reason to expect for a sex difference in tiredness or sleep onset to be the cause of sex-biased breathing effects in young adults.One prediction is that deep breaths should associate with sleep. In the respiratory literature, deep breaths are associated with many factors including tiredness and sleep13. No hard measure of sleep exists in HCP data, but a list of sleepy subjects was kept by scanner technicians, which includes 37% of all HCP subjects. Of our three groups, 71% of the deep breath group was on that list (p = 0.0014), compared to 38% of the clean group (n.s.), and 41% of the burst group (n.s.), giving face plausibility to the validity of the list and indicating that subjects exhibiting many deep breaths are likely enriched for people yawning. Relatedly, we examined gray plots of individual scans in which subjects were documented as definitely sleeping (outside the groups) and we could discern no visual signature of sleep, certainly not by respiratory pattern.Another testable proposition is that sex-biased bursts are merely snores or obstructive sleep apnea. Obstructive sleep apnea is strongly potentiated by obesity23, and there was no correlation between burst scores and body mass index (BMI) in either sex (male r = 0.07, p = 0.32; females r = −0.03, p = 0.66), nor did males and females significantly differ in BMI, which was in the mid-20s for both sexes. Collectively, these results support central, not obstructive, causes of bursts (though some instances could be obstructive).Another testable proposition is that the breathing patterns become more likely as scans progress, perhaps reflecting an influence of arousal (if not necessarily sleep). Our visual impressions from examining all scans were that deep breaths seemed to occur at any point of a scan, including the very beginning, and were not notably concentrated at the end, whereas bursts seemed to be uncommon at the beginning of scans and to emerge later in scans. More formally, in both males and females, algorithm indices for both patterns rose as scans progressed, and t-tests of indices in minutes 1–4 compared to 11–14 of scans were significant for both patterns in both sexes (both p’s < 1e−16 for bursts, both p’s < 0.02 for deep breaths). These statistical effects accord with our visual impressions.Collectively, these observations indicate that deep breaths are associated with sleepiness, that bursts are probably central rather than obstructive phenomena, and that sex biases in sleep onset or tiredness are unlikely to be causes for sex differences in bursts. They also indicate that bursts become more likely as scans progress in both sexes.DiscussionIn this paper, we described two respiratory patterns that differentially bias functional connectivity, both commonly seen in young adults lying in MRI scanners. This work represents our first major effort toward developing an event-related framework for understanding effects in spontaneous fMRI time series, following argumentation in ref. 24. Distinct respiratory effects can occur within-scan, between scans in subjects, and across subjects and populations. This discussion links respiratory patterns to effects in the neuroimaging literature and lays out a potential mechanistic basis of bursts that could explain why bursts occur more in males. It concludes by noting settings in which differential expression of respiratory patterns may be anticipated. A Supplementary Discussion elaborates on concepts described only briefly in the main text, especially regarding mechanisms and clinical associations of periodic breathing.Both deep breaths and bursts had brain-wide effects on fMRI signals, causing increased global functional connectivity, and the effects were most marked for bursts. The increased prevalence of bursts in males contributed to higher average fMRI signal correlations compared to females. These differences were eliminated once both brain volume and respiratory patterns were accounted for. To clarify, we are not broadly asserting that all functional connectivity sex differences are due to respiration or even necessarily that the specific differences we removed were due to respiration (a third shared variable could be at work); we are asserting that respiration is capable of causing sex differences in functional connectivity.The respiratory patterns had distinct spatiotemporal effects, with bursts strongly elevating correlations among sensorimotor brain regions, whereas deep breaths had their weakest effects in such regions. Signal modulations were fairly time-locked across the brain in deep breaths, but bursts unfolded with multi-second lags between different parts of the brain. These distinctions, joined with distinct cardiac effects, indicate that distinct cardiopulmonary and neurophysiological events are occurring during the patterns. An important topic of future work will be to disentangle pCO2-related cerebral blood flow effects during each respiratory pattern from the blood flow consequences of the neural activity that triggers and tracks and guides each pattern.It is noteworthy that the sensorimotor pattern has emerged in numerous areas of fMRI research, including studies localizing global fMRI signals at rest25,26, and studies of arousal and sleep27–30. The extent to which bursts (or deep breaths) define what is measured as the global fMRI signal is presently unknown. Given the lively debates over retaining or discarding global fMRI signals, it will be important to understand how various denoising or data processing strategies impact one or both of these respiratory patterns. Similarly, it will be of interest to examine these phenomena in the context of caffeine administration, vigilance tasks, lag structure, or quasiperiodic patterns (e.g., refs. 31–34). As detailed below, there are reasons to suspect that psychiatric symptoms will also scale with these patterns; indeed, our clean group without bursts or deep breaths reported a colloquially “clean” lifestyle, with unusually little substance use or psychiatric symptomatology (Supplementary Note 3).By our estimate, in young adults, deep breaths occurred in about half of all scans, and in well over three-quarters of subjects, numbers that are unsurprising from a respiratory perspective. Adult humans inhale deeply several times an hour to counter mild changes in blood gas tensions and to reinflate collapsed alveoli, and such physiologic sighs are especially common in the supine position used for scanning13. In addition, yawns are likely to occur in some subjects.The burst pattern occurred in over one-third of scans and in two-thirds of subjects, with a tendency to occur in males. The lack of correlation between BMI and burst prevalence in either sex suggests a central rather than obstructive origin of the pattern, though some instances might be obstructive. A centrally oriented explanation is that the patterns arise via sex-biased respiratory control parameters that are unmasked as subjects relax in the scanner. Our leading hypothesis is that the pattern emerges via parameters governing the chemoreflex, which could explain both the waveforms and the sex bias of bursts, as detailed below.The generation of waxing and waning patterns of breathing occurs via interactions among the mechanisms controlling breathing (this discussion is simplified, see the Supplementary Discussion and ref. 35 for detail); when such patterns attain cyclic stability they are called periodic breathing (Fig. 7 shows periodic breathing in several situations in which it is routinely seen). Breathing rhythms are generated in the medulla and are under three kinds of control: volition, a waking neural drive during wakefulness and rapid eye movement (REM) sleep, and a chemoreflex loop that centrally senses pH and thereby pCO28. The chemoreflex is central to this discussion. Every person has preferred set points for arterial pCO2, called the resting pCO2, and fluctuations about resting pCO2 engage a negative feedback loop: at high arterial pCO2, the chemoreflex stimulates breathing in order to reduce pCO2, but below a certain value of pCO2—the apneic threshold—the chemoreflex ceases to stimulate breathing. Cyclic waxing and waning of respiration generally occur in the following manner: (excessive) ventilation pushes pCO2 below the apneic threshold, causing chemoreflex respiratory drive to fade or cease, after which pCO2 rises and pO2 falls, eventually triggering the resumption of (excessive) ventilation and the start of the next cycle.Fig. 7Comparison of periodic breathing waveforms and bursts.At left, single-subject waveforms of periodic breathing in opioid use, stroke, heart failure, at high altitude, and in newborns. These are all conditions and situations in which periodic breathing is commonly encountered. At right, illustrations of bursts in 11 HCP subjects. All plots are on the same time scale, and the green lines measure 1 min. Note the long cycle times seen in patients with heart failure (in the Stroke and Heart failure sections), reflecting, in part, an exaggerated delay in central detection of changes in arterial gas tensions (see Supplementary Discussion for more detail). The stroke example illustrates shows the added effect of delayed circulatory time in heart failure. Images at left modified from46,54–58 with permission. Images from ref. 55 adapted with the permission of the American Thoracic Society. Copyright © 2020 American Thoracic Society. All rights reserved. The American Journal of Respiratory and Critical Care Medicine is an official journal of the American Thoracic Society. Readers are encouraged to read the entire article for the correct context at https://europepmc.org/article/med/15665317. The authors, editors, and The American Thoracic Society are not responsible for errors or omissions in adaptations.Key to initiating and propagating such cycles is the CO2 reserve – the difference between resting pCO2 and the apneic threshold – which exhibits sex differences due to gonadal hormones. Women have the same resting pCO2 levels as men, but lower apneic thresholds, and thus larger CO2 reserves36. Sex hormones influence these parameters: administration of testosterone to women raises the apneic threshold without altering resting pCO2, thus reducing CO2 reserve37. Smaller CO2 reserves mean a smaller ventilatory perturbation is needed to trigger apnea and cycles of waning and waxing breathing, and thus larger CO2 reserves make apneic events less likely, leading to the expectation that women should be less likely to initiate and perpetuate periodic breathing than males. In a recent study, when healthy adults were taken to high altitude (Mount Everest), forcing a lowered resting pCO2 (due to hyperventilation), women displayed far less central apnea, and periodic breathing specifically, than men38. These manipulations (testosterone, hypoxia) highlight the role of sex hormones in shaping chemoreflex responses, which is the basis of a well-established tendency for females to display less apnea than males. Waxing and waning cycles of breathing are usually studied during sleep and measured via the apnea/hypopnea index (AHI), indexing the number of events per hour, and numerous large respiratory studies detect robust sex differences in AHI scores, with women displaying fewer episodes than men39–42. Further evidence of the role of sex hormones in apnea is that AHI sex biases are reduced after menopause, that hypogonadal males have lower AHI than weight-matched peers43, and that women with testosterone-producing tumors have higher AHI than weight-matched peers44.In short, the sex biases of the burst pattern could potentially arise via sex-hormone dependent properties of the chemoreflex loop that potentiate sex differences in apnea and hypopnea. Such disordered breathing is—by far—most commonly studied in older, clinical populations and during NREM sleep, when the chemoreflex properties are maximally exposed (although recent studies have begun to document such breathing in the daytime45,46). Though lessening of the wakefulness drive would help to reveal instabilities in the chemoreflex loop, the extent to which sleep or reduced arousal is contributing to bursts in the HCP data is unclear, and new datasets with concurrent monitoring of respiration and arousal will be needed to properly address the issue. The issue is important because brief arousals can result from hypoxia after apnea, and can perpetuate new cycles of central apnea by transiently (excessively) increasing ventilation. The variety of forms that periodic breathing can take, and their relation to multiple parameters of the chemoreflex loop and resemblances to bursts, are discussed at greater length in the Supplementary Discussion.The influence of breathing on fMRI signals is marked, and our results suggest situations where breathing biases may be expected. The Supplementary Discussion reviews literature suggesting the following associations. Yawns will be more prevalent in tired subjects, and in subjects on medications causing sedation. To the extent that bursts share the respiratory mechanisms that drive increased AHI in the respiratory literature, one would anticipate increases in bursts (1) over the lifespan in both sexes (especially after age 60), (2) in males relative to females at all ages (a bias lessening after menopause, probably with onset around menarche), (3) in subjects on opioids and other respiratory suppressants, (4) in subjects with brain injuries (e.g., strokes), and (5) in subjects with medical conditions like heart failure. Large psychiatric studies have reported longitudinal dose-response associations of AHI with depressive symptoms47, and community samples routinely obtain cross-sectional associations of depression and AHI39. Fluctuations in gonadal hormones over pregnancy, over menstrual cycles, or by sex-hormone therapies are also likely to influence burst prevalence. Medications influencing cholinergic, noradrenergic, and serotonergic inputs to wakefulness respiratory drive may influence the emergence of bursts, and may have a role in the association of depression and AHI scores. Given the relative ease of using the respiratory belts provided with many scanners, it would be prudent to collect such physiology records during fMRI scans (see ref. 14 for discussion of implementations).MethodsData and subject selectionThe 900-subject public release of the Human Connectome Project (HCP) Young Adult data was obtained. Subjects were young adults drawn from the geographic region about St. Louis, Missouri, and were identified via a Missouri state registry of siblings sets that include twins. Subjects were healthy in a broad sense: study criteria excluded sibships containing siblings with diabetes, hypertension, or neurological or psychiatric disorders. The HCP-YA protocol was approved by The Washington University in St. Louis Institutional Review Board, WUSTL DHHS Federalwide Assurance #FWA00002284 BJH DHHS Federalwide Assurance #FWA00002281 SLCH DHHS Federalwide Assurance #FWA00002282, and the study was conducted in accord with the Declaration of Helsinki. Subjects provided written consent to participate and were compensated monetarily for participation. Each subject underwent four 14.4-min fMRI scans (two scans on two days each) while staring quietly at a crosshair at an altitude of ~450 feet above sea level. During scanning, physiology data was acquired via the Siemens Physiology Monitoring Unit (PMU), which is standard equipment that accompanies the scanner for purposes of cardiac and physiological gating. The signals acquired were 400 Hz recordings of an abdominal belt and a finger pulse oximetry waveform. The Siemens respiratory record is obtained via a pressure hose connected to a respiratory cushion placed under an elastic belt strapped around the subject’s abdomen, and output is in arbitrary units. All resting-state fMRI scans were obtained, including minimally preprocessed and FIX-ICA-denoised images, along with their accompanying head position and physiology files.Of the HCP 900-subject release, based on visual assessment of the physiology traces, 440 subjects had four 14.4-min resting-state fMRI scans with complete accompanying physiological data in which we believed we could reliably identify peaks in all cardiac and respiratory traces (all physiological data and decisions about quality can be seen in the Supplementary material of ref. 48). Only these 440 subjects were analyzed further. Characteristics of the subjects were: age 28.6 ± 3.8 (range 22–36), 228 males and 212 females; BMI 26.5 ± 5.0 (range 16.5–43.9).For the 440 subjects believed to have high-quality physiology data, the following files were obtained: four resting-state fMRI scans transformed to atlas space (in each subject’s/MNINonLinear/Results folder): [RUN] = REST1_LR, REST1_RL, REST2_LR, REST2_RL (this order is runs 1–4 in the text). rfMRI_[RUN].nii.gz and rfMRI_[RUN]_hp2000_clean.nii.gz scans were obtained, representing minimally preprocessed and FIX-ICA-denoised data. For each of these scans, the [RUN]_Physio_log.txt and Movement_Regressors_dt.txt files were also obtained. Structural scans transformed to atlas space were also obtained (in each subject’s /MNINonLinear/folder): the T1w.nii.gz and the aparc+aseg.nii.gz files, representing the anatomical T1-weighted scan and its FreeSurfer segmentation.Image and parameter processingThe aparc+aseg.nii.gz file for each subject underwent a set of serial erosions within white matter and ventricle segments, exactly as in ref. 25. Masks of cortical gray matter, the cerebellum, and subcortical nuclei were extracted, as were serially eroded layers of superficial, deep, and deepest (with respect to distance from the gray matter) masks of the white matter and ventricles. These masks, together, include all in-brain voxels of these tissue types and are used to extract certain signals and to order signals for gray plots49. For the purpose of making useful gray plots, because of the considerable thermal noise in HCP scans, a within-mask 6 mm FWHM Gaussian kernel was applied to the data using the above masks (illustrated for HCP data in ref. 49). This blurring does not mix tissue compartments, due to the use of masks, beyond partial volume effects present in the voxels themselves.Respiratory belt and pulse oximeter traces (sampled at 400 Hz) first underwent visual inspection in their entirety to determine if the quality was sufficient for reliable peak detection since traces are often partially or fully corrupted. Only subjects with traces deemed likely to successfully undergo peak detection in all runs were analyzed48. After selection, for respiratory traces, an outlier replacement filter was used to eliminate spurious spike artifacts (Matlab command: filloutliers(resp_trace,‘linear’,‘movmedian’,100)) and the traces were then gently blurred to aid peak detection (Matlab command: smoothdata(resp_trace,‘sgolay’,400)) (a 1-s window for a 400 Hz signal). These treated respiratory traces are the ones shown in Figures.Following prior literature, several respiratory measures were derived from the treated respiratory belt trace. First, the envelope of the trace over a 10-s window (at 400 Hz) was calculated after ref. 15 (Matlab command: envelope(zscore_resp_trace,4000,‘rms’)). Second, the RV measure, defined as the standard deviation of the treated respiratory trace within a 6-s window, was calculated following12 (Matlab command: movstd(zscore_resp_trace, 2400,‘endpoints’,‘shrink’)). Finally, an RVT measure, defined for all peaks as ((peak-prior trough)/(time between peaks)), was calculated. Peak detection on the trace yielded peaks (and troughs, using the inverted trace) for calculations (Matlab command: (findpeaks(zscore_resp_trace,‘minpeakdistance’,800,‘minpeakprominence’,.5))). The minimum peak distance presumes breaths occur more than 2 s apart. If a peak did not have a preceding trough prior to the previous peak, no value was scored at that peak. All traces and derived measures were visually checked to ensure that outliers and abnormalities would not drive results. These three measures were termed ENV, RV, and RVT in figures. The RVT calculated is the “core” computation studied in ref. 14, which behaves like the full computation of ref. 10.Pulse oximeter traces underwent z-scoring then peak detection (Matlab command: findpeaks(zscore_pulseox,‘minpeakdistance’,180,‘minpeakprominence’,.5)). Heart rate was calculated from the interval between peaks. The minimum peak distance presumes heart rates are under 133 beats per minute. Peak amplitude was calculated from the height of the peak relative to the previous trough. Cardiac traces are prone to transient disruptions when fingers move, and it is laborious to check and correct cardiac measures due to the large numbers of peaks and troughs. A limited number of cardiac records are therefore used in this report, but those select traces and their derived measures were visually checked to ensure accuracy.The data quality measure DVARS was calculated after refs. 50,51 as the root mean squared value in the brain at each timepoint of all voxel time series differentiated in time by backward differences. DVARS by convention is 0 at the first timepoint.Head position was taken from the Movement_Regressors_dt.txt files. In gray plots, these position parameters are displayed after subtracting the first timepoint value from the time series (so that all traces start at zero). Head motion was represented by Framewise Displacement (FD) measures, following ref. 51, wherein all position measures were differentiated in time by backward differences, rotational measures were converted to arc displacement at 5 cm radius, and the sum of the absolute value of these measures was calculated. To suppress tidal respiratory motion, head position traces were filtered with a stopband of 0.2–0.5 Hz following ref. 16. FD is typically calculated by backward differences to the preceding timepoint (here 720 ms prior), but historically FD measures using sampling rates of 2–4 s were common; for comparison to such measures, FD was also calculated by backward differences over four timepoints (4 * 720 ms = 2.88 s effective sampling rate) where indicated, exactly as in ref. 52. Thus, where indicated, head motion measures of FDoriginal, FDfiltered, and FDfiltered,4-TR are examined.Gray plot formationGray plots were created of each scan, exactly as shown in Figs. 1 and 2, and exactly following procedures outlined in refs. 14,16. Examination of these 4 * 440 = 1,760 gray plots led to the recognition of the burst pattern described throughout the paper, as well as the single deep breaths also characterized.The 35 bursts, 35 deep breaths, and 35 non-respiratory motions shown in Fig. 3 were chosen by visual inspection of respiratory and motion records, with visually marked onsets shown in Supplementary Movies 2–4 and onsets listed in Supplementary Data 1. Random onsets were defined in the motion subjects using randomly selected runs and onsets. Properties of scans in the 90 s surrounding the onsets are shown in heat maps. Thin gray bars with red heat maps show the significance of unpaired two-sample t-tests at each timepoint compared to the random onset group. Heart rate was derived from peak-to-peak intervals in pulse oximetry data and was visually verified. Plots show the mean properties of the 35 onsets, and shade plots show mean and standard deviation.Group formationAuthor JDP examined all 440 subjects, noting subjects with many scans with only deep breaths, many scans with only bursts, or many scans with only normal tidal breathing patterns. These lists were then screened for any siblings, and siblings were discarded in a manner yielding the largest equally-sized remaining groups, resulting in three groups of 21 subjects, all unrelated. Statistical contrasts of the group demographic and other properties are described in the Supplementary Materials. Because of associations reported to medical and psychiatric variables, the identities of these groups are restricted to registered HCP users who have been granted Restricted Access, via a Subject Key. To access the Keys: (1) Sign in to https://db.humanconnectome.org; (2) Navigate to and open the WU-Minn HCP data set by clicking the “Open Dataset” tab; (3) Click the “Subject Keys” tab; (4) There are three subject keys listed under “Published Subject Keys for This Dataset” associated with this manuscript. Click on a subject key to obtain a description and access the data.“Lynch_etal_2020_NatureCommunications_BurstGroup”“Lynch_etal_2020_NatureCommunications_DeepBreathGroup”“Lynch_etal_2020_NatureCommunications_CleanGroup”Note: Only HCP users with restricted data access will be able to use subject keys. If you receive an error message (e.g., “Restricted Data Access Required!”) you must request restricted data access.Ratings of breathing patterns in scansTwo authors (J.D.P. and C.J.L.) jointly examined the scans of subjects 400–440 to discuss breathing patterns, then independently rated scans 1–399 with binary decisions in each scan about the presence (1) or absence (0) of deep breaths and bursts in each scan. Cohen’s kappas were calculated for the ratings, shown in Fig. 4, yielding high inter-rater reliability. The likelihood of obtaining sex differences in each breathing pattern was determined by Chi-squared tests in each rater, and both raters obtained sex effects of bursts but not of deep breaths. The sum over the four scans in each subject of each type of breathing pattern was called the pattern score.Algorithmic indexing of breathing patternsThe automated algorithm is described in detail in the Supplementary Materials. In brief, priors were derived from respiratory belt traces indicating, separately, the likelihood of deep breaths and bursts, and these priors were multiplied into the fits of fMRI templates of each breathing pattern to global fMRI signals. The total values over each scan for each breathing pattern were obtained and averaged in each subject. Both indices were downweighed by the variation in breathing rate of a scan, to dampen outliers due to haphazard breathing. Indices were correlated to rater pattern scores and were compared in each breathing pattern by sex using unpaired two-sample t-tests in Fig. 4.Functional connectivity measuresWe first describe global functional connectivity (gFC) analyses. The pairwise correlations of all voxels within a subject’s gray matter mask were computed in each fMRI scan, and the median value in each scan was taken, followed by Fisher-Z transformation. These values represent the central tendency of gray matter functional connectivity, which is what one would expect for respiratory phenomena to most directly influence, since cerebral blood flow changes have brain-wide consequences for fMRI signals. These values were averaged over a subject’s scans for the purposes of cross-subject correlations, linear regressions, and ANOVA/ANCOVA in Fig. 4. gFC and rater scores were compared by correlation. Differences by sex were compared by unpaired two-sample t-test. gFC was fit to pattern scores separately in males and females using the formula gFC = b0 + b1*burst_score + b2*deep_breath_score, yielding similar fits in each sex. In Fig. 4k, ANOVA (model 1) and ANCOVA (models 2–6) was used to model gFC as a function of sex alone or sex plus other variables, and main effects of the explanatory variables are color-coded in Fig. 4k. Model 1 is an ANOVA with only sex, Model 2 adds head size in ANCOVA, and the other models are ANCOVAs with the indicated terms present, always significantly fit, with the exception of the sex variable, which becomes insignificant in the final two models (green cells). DVARS and the three versions of head motion were added to models 1–4 but failed to negate the sex effect in any model, and often failed to even fit as main effects when respiratory variables were present.Here we describe network analyses. The 333-parcel scheme of Gordon et al.53 was used to sample images, and the cluster assignments of that paper define the resting-state networks in this paper, illustrated in Fig. 5a (see Supplementary Fig. 11 for a full list of networks). Correlation computations were performed via Fisher-Z transforms but were converted to Pearson r values for figures and reporting. For all matrices in Figs. 5 and 6, and Supplementary Figs. 11 and 12, only cells significant at p < 0.05 are colored (always determined by 10,000 permutation tests specific to the matrix; gray cells are insignificant).Minimally preprocessed (MP) and FIX-ICA-denoised time series were sampled, and additionally several signal processing steps were applied to generate a few additional commonly used kinds of signals: MP with the mean gray matter signal removed (MP+GSR, global signal regression), MP data with six motion regressors and their derivatives regressed out and censoring timepoints with z-scored DVARS (from minimally preprocessed data) values >2 (MP+mot+cens), or both of those steps applied to MP data along with GSR (MP+GSR+mot+cens). These maneuvers simply give readers a broad sense of when, where, and how severely certain effects manifest in different signal processing regimes.In Fig. 5, using the sets of 35 onsets defined for Fig. 3, parcel time series from −10 to +40 s about the onset were used to generate correlation matrices for each kind of onset (burst, deep breath, motion, random), and the mean values for breathing patterns are shown, masked by significance in permutation tests with the random matrices. Contrasts of breathing patterns are shown in several data processing strategies (permutations between pattern matrices). Mean parcel time series are shown in Fig. 5c, with red dotted lines denoting the span used for correlations. Mean values of grayordinates are shown in MP data in Fig. 5d and Supplementary Movie 5.In Fig. 6, correlations from entire scans are used. Figure 6a illustrates the subtraction of mean matrices of the groups, screened by permutation tests between groups. The bottom row of Fig. 6a shows the values of the observed group matrices among random groups drawn from all subjects. Figure 6b illustrates the within-subject differences between scans with and without certain breathing patterns. The top row shows scans with bursts and not deep breaths (B+D−) compared to scans without either pattern (B−D−), the bottom row a comparable contrast for deep breaths. Only scans in which raters had completely agreed on the presence or absence of patterns were used, and all unrelated qualifying subjects possessing the needed scan subtypes were used in each kind of contrast, with 50–70 subjects in each contrast. Mean differences are shown in several data processing strategies, masked by permutation tests with randomly swapped scan subtype labels. Figure 6c illustrates betas in multiple linear regression of breathing pattern scores with head size and nuisance variables, all scaled identically and performed separately in each sex, with all regressors standardized prior to running the model, and masked by significance among betas from randomly permuted subjects. Betas for bursts are shown for several processing strategies. Full fits to all variables are shown in Supplementary Fig. 11 for several processing strategies.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationDescription of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2Supplementary Movie 2Supplementary Movie 3Supplementary Movie 4Supplementary Movie 5Reporting Summary | nature communications | [
"Article"
] | [
"Cognitive neuroscience",
"Respiration"
] | (fMRI scanning at rest major neuroimaging paradigm functional connectivity resting state MRI1 subjects lie quietly staring crosshair 5–15 min performing no task fMRI data acquired Correlations in task-free fMRI signals reflect functional relatedness tissues spatial topography macro-scale maps human scans deliver diagnostic prognostic information large studies resting-state fMRI scans ABCD study 10,000 children developmental.Breathing modifies carbon dioxide arterial blood cerebral blood flow fMRI deeply quickly hyperpnea exhale more CO2 arterial pCO2 drops cerebral blood flow decreases fMRI signals decrease breathing shallow slow hypopnea less CO2 released arterial pCO2 rises blood flow increases fMRI signals increase breathing patterns influence resting-state fMRI scans.Breathing occurs multiple forms basic respiratory rhythm cyclic rhythm eupnea tidal volume air lungs each breathdeviations from eupnea exist sighs collapsed lung yawns of boredom sleepiness disordered breathing cluster ataxic periodic Hunter–Cheyne–Stokes), associated with heart disease neurological injury6 breathing have biophysical correlates consequences for neuroimaging.Little known about breathing characteristics of healthy young adults in MRI scanner functional connectivity literature fMRI signals examined respiration fMRI signals in large data set of healthy young adults scan time Young Adult Human Connectome Project report effects in 440 h of scanning in 440 subjects (ages 22–36 mean 28.6, 228 males 212 females). data contrast to prior fMRI literature small short data unprecedented window into respiratory behavior of resting in scanners eupnea two prevalent patterns in respiration with distinct correlates in fMRI signals influences on functional connectivity One isolated deep breaths not sex-biased other bursts sex-biased to sex-biased breathing patterns literature Patterns recognized by human raters algorithmic scoring systemresults demonstrate sex-biased breathing in healthy young adults influence on functional connectivity measures in older ill patients clinical literature suggests breathing patterns influenced by sex hormones age cardiovascular neurological psychiatric illness distinct breathing patterns in subjects at individual group descriptions demographic differences spatiotemporal effects consequences for fMRI signal covariance focus on visual presentations recognized patterns influences respiration on resting-state fMRI signals created viewed plots of 1760 scans 440 h in 440 young adults two common patterns of respiration One known single deep breath other pattern unfamiliar burst pattern two patterns illustrated in Fig. 1. fMRI scans flattened into grayscale heat maps in-brain voxels defining Y axis time X axis Signals from gray matter voxels above green lines white matter ventricles below respiratory belt trace shown in blue deep breaths marked by orange arrows in eupneic scan in Fig. 1a Three respiratory measures from shown (ENV RV RVT trace windowed variance rate of air displaying abnormalities at deep breathsfMRI signals increases white bands after deep breath decreases black bands). Figure 1b illustrates burst respiratory pattern serial tapers breathing depth apnea between correlates fMRI signals Fuller images before after denoising Supplementary Figs. 1 2 signal effects before after FIX-ICA denoising patterns linked to head motion image quality measures scans chosen breathing patterns deep breaths bursts Fig. 1c, d later figures single deep breath waveforms floor ceiling effects slippage respiratory belt peak inspiration consult Supplementary Note 1 view 1760 gray plots Supplementary Movie 1 (1.4 GB. 1Gray plots scans deep breaths bursts breaths bursts panels z-scored respiratory belt traces blue-ticks z = −1 1) 3 respiratory measures (ENV RV RVT vertical offsets non-overlapping visualization scales identical figures). grayscale heat maps in-brain fMRI signals organized by anatomical compartment green line separating gray matter white matter ventricle signalsthree deep breaths indicated by arrows major decreases black bands in fMRI signal over two dozen bursts present (arrows mark modulations fMRI signals Figs. 1 2 each scan Additional deep breaths bursts (d bottom code gray plots published in ref. 49 comprehensive plots 17,640 scans in Supplementary Movie deep breaths five instances in Fig. 2a (fuller images Supplementary Fig. 3 deep breaths Movie 2) breaths accompanied breathing pauses variable duration 20 brief central apneas sequelae deep black band gray plot reflects pan-brain decrease in fMRI signals ~30 s after breath cerebral blood flow decrease increase ventilation.Fig. 2Examples deep breaths bursts.Plots formatted Fig. 1. deep breaths slowness possibility transient apnea incongruence respiratory measures ENV, RV RVT fMRI signal decreases bursts repeated modulation of breathing amplitude congruency of ENV RV RVT correlates in fMRI signals bursts five instances in Fig. 2b (fuller images Supplementary Fig.see scans bursts in Supplementary Movie 3) burst deep breaths tapers into shallow breaths followed by additional bursts burst respiratory patterns differ single deep breath patterns single deep breaths bursts span minutes lasting ~30–50 s burst patterns evident in ENV RV RVT traces display large wavelike modulations lack concordance in deep breaths Fig. 2) typical fMRI signal response initial increase prolonged decrease orange lines white black durations matching respiratory burst added lag signal decrease subject bottom scan eupnea evolves into burst pattern fMRI signal subject had four 14.4-min scans one scan normal tidal breathing different different breathing differences in fMRI signals examples in Supplementary Fig. 5) different breathing patterns dominate at different times examples Supplementary Fig. 6) deep breaths bursts prevalent organized focused on events other disorganized breathing exist example Supplementary Fig. 5 respiratory 35 bursts 35 deep breaths 35 non-respiratory head motions identified (Supplementary Data 1 lists onsets Supplementary Fig.6 onsets gray plots Supplementary Movies 2–4 non-respiratory motion onsets identified fMRI signal changes during deep breaths bursts head motion random onset scan control condition signals extracted 30 s prior to 60 s after onsets Fig. 3a statistical contrast onsets thin gray-red heatmap maps t-tests p < 0.001) burst deep breath identification respiratory heat maps Plots Fig. 3b show mean values Bursts show marked signatures in ENV RV RVT Deep breaths signatures differ RVT no positive deflection Global fMRI signals differ deep breaths brief steep signal increase marked signal trough nadir 15 s after resolution 30 s after Bursts slower trajectories peaking later prolonged troughs nadir over 20 s after onset resolution 40 s after event durations modulated by breathing presence duration apnea after deep breaths duration burst taper selected non-respiratory-motion group control deep breaths display motion onset motion-displaying group random group showed global fMRI signal fluctuations ruling out motion signalsmotion global changes studies show global fMRI signals T2* with not S0 artifacts head Heart rate elevated after deep breaths no average effect for bursts deep breaths elevate heart rates deepened breathing in bursts modulation some subjects display cardiac modulation by bursts others no modulation non-HCP resting-state fMRI datasets exhibit bursts reliable link between bursts heart rate modulation Supplementary Note 2 Fig. 9). 3Properties of deep breaths bursts Heat maps illustrate segments events 35 examples of bursts deep breaths non-respiratory motions identified from respiratory belt traces motion traces subjects random timepoints selected Illustrated respiratory belt traces three respiratory measures global fMRI signal head motion DVARS gray/red heatmap represents differences from random events beyond p < 0.001 capped at p < 1e−10 basis of bursts deep breaths apparent in respiratory belt imagessignals ENV RV RVT mark bursts deep breaths display differences RVT little positive deflection global fMRI signals differ deep breaths brief increases decreases nadir near 15 s 30 s to resolution Bursts more positive deflections slower timecourses resolving near 40 s Motion no global fMRI signal changes Shade plots reflect mean std Deep breaths display motion DVARS changes time-locked to event onsets bursts smaller time-locked modulations significance Source data provided file.Group contrasts reveal sex bias in burst patternsTo patterns sought subject-level factors scaled with patterns HCP data set has hundreds behavioral demographic physiologic imaging measures subject needed groups contrast numeric indices patterns pursued prioritized group contrast approach breathing patterns defined three groups with different breathing patterns subjects unambiguous marked bursts few no single deep breaths burst single deep breaths breaths clean Subjects selected based on respiratory traces signal heat maps without other properties identified three groups of 21 subjects all unrelatedSubject identity restricted to associations with psychiatric instrument scores substance use investigators with HCP Restricted Access find Subject Key identifying groups statistical contrasts in Supplementary Note 3 Fig. 10 Data 2. Instructions access Subject Keys in “Methods” section three groups associated with more HCP variables alcohol use cigarette use structural imaging variables differed by group distinguishing burst group from Fig findings by males 6/21 clean 5/21 deep breath 14/21 burst group groups formed without knowledge sex unlikely probability p = 3.3e−5) unbalanced sex compositions three times in random group formation.Ratings gray plots reveal sex bias in burst authors J.D.P. C.J.L. rated 1596 scans 1–399) 400–440 binary decisions on deep breaths bursts Ratings gray plots without demographics sex group results discovered after 100 subjects rated Cohen’s kappas 0.79 for bursts 0.deep breaths). sex differences in bursts not deep breaths present first 100 subjects both raters next 299 subjects present excluded one subject per family contributed report ratings entire 399-subject cohort.Identical rater decisions 87% bursts 89% deep breaths Cohen’s kappa values 0.73 0.78 ambiguous decisions prioritized four scans pattern scores scores correlated r = 0.86 (p < 1e−20) for bursts 0.90 < deep breaths Fig. 4a numbers indicate human raters recognize breathing patterns use Supplementary pattern training module learn patterns. 4Rater scorings algorithmic indices detect sex effects global functional connectivity influences Plots total scans bursts deep breaths both raters score correlations Cohen’s kappas N = 399 subjects except d e three 21-subject Histograms scores both patterns raters color Bar percent scans each sex patterns Chi-squared tests bursts yield p = 3.4e−8 8.8e−7 for J.D.P. C.J.L. unchanged excluding three groupsdifferences by sex in deep breath scores Bar plots mean values error bars clean burst deep breath groups N = 21 B D denote burst deep breath Desired respiratory properties each group Algorithm indices three 21-subject groups corroborating scores desired breathing patterns Box plots show median 25th 75th percentiles whiskers encompass 99% data outliers marked Algorithm indices breathing patterns by sex significant differences two t-test in bursts not deep breaths Box plots algorithm indices pattern mean rater score Pearson correlations Box plots gFC mean pattern scores stronger effects bursts gFC multiple linear regression pattern scores gFC*burst_score_breath sex stronger effects of bursts Bars show mean values error bars 95% confidence intervals fits differ by sex Box plots gFC head size by sex different by sex two-sample-sided t-test (p = 9.1e−9 <1e−20 Color chart main effects multiple ANCOVA models Sex effects insignificant when head size respiratory variables modeled Source data group identity redactedDeep breaths 85% subjects common scans (Fig. 4b). Bursts absent in 30% uncommon in all scans chi-squared tests sex differences significant for bursts (p = 3.4e−8 8.8e−7) not deep breaths (Fig. unchanged excluding groups bursts identified in 45% male 35% female scans deep breaths in 54% male 52% female Scores uncorrelated across subjects each rater sex (r < 0.1 breathing patterns occur independently pattern scores 3 contrasted groups accord with desired group breathing properties (Fig. 4d).Automated detectionTo algorithm index breathing patterns respiratory traces global fMRI signals algorithm creates probabilities respiratory patterns multiplies with global fMRI time series deep breaths bursts simultaneous evidence approach performs haphazard disorganized breathing cause high indices discount scans with variable respiratory rates algorithm recaptures differences in group breathing styles (Fig. 4e sex difference in bursts not deep breaths (Fig. 4f). Indices correlate with rater scores both patterns (Fig.results give confidence in ratings group formations demonstrate algorithmic approach fuller description algorithm illustrations in Supplementary Note 4.Influence respiratory patterns on global covarianceRespiratory events blood flow add variance to voxel signals white black bands in gray plots functional connectivity median correlation of voxel signals four scans calculated mean over computed global functional connectivity increases gFC with rater scores scaling stronger for bursts (r = 0.53 p < 1e−20) deep breaths (r = 0.18 p = 4e−4) (Fig. algorithm indices scale with gFC more for bursts = 0.59 p < 1e−20) deep breaths = 0.29 = higher values indices reflect scale prevalence incorporate template fits to fMRI signals regression of yielded betas high for bursts deep breaths fits to gFC differ by sex (Fig. 4i shows by bursts deep breaths produce effects gFC sex common in males gFC may increased in males males have higher gFCmales have larger heads brains closer to scanner coils signal-to-noise ratios differ by sex explanation gFC differences modeled gFC via ANCOVA sex head size respiratory variables eliminated sex differences when head size accounted (Fig. 4k). motion cause global signals added motion covariates data quality covariate DVARS to models 1–4 eliminate gFC sex differences failed fit effects respiratory variables present.Spatiotemporal effects of bursts deep breaths in asked non-global profiles breathing patterns functional connectivity by pattern focused on covariance during breathing patterns 35 events from Fig. 3 extracted time series −10 to +40 s computed mean correlation matrices parcellation Fig. 5a Permutation tests patterns yielded significant differences from random onsets cells significant p < 0.05 each other Fig. 5b Mean signals state networks plotted in Fig. 5c correlation spatially specific effects present all forms signal processing significant spatially specific effects existminimally preprocessed FIX-ICA-denoised original respiratory effects elevation correlations primary sensory motor distribution visual auditory motor somatosensory cortex ovals Fig. Signals peak high early Fig. deep early troughs time series shown brain surface Fig. 5d Supplementary Movie 5) illustrating global focal effects. 5Spatiotemporal properties of bursts deep breaths Color legend network locations colors ref. 17, text labeling networks Supplementary Fig. Correlation matrices from −10 to 40 s event onsets Fig. 3 minimally preprocessed data show mean differences of 35 bursts 35 deep breaths 35 random onsets coloring cells significant p < 0.05 10,000 permutation tests non cells colored grayscale matrices bursts contrasted to deep breaths 10,000 tests top bottom 2.5% differences permutation ranks illustrated minimally preprocessed FIX-ICA-denoised time series global signal regression Differences each processing strategy visual auditory somatomotor cortex Mean signals 35 patterns each time series mild smoothing colored by legendovals encircle peaks troughs sensorimotor networks Surface representations events data b c minimally preprocessed time series Movie 5 animates patterns non-respiratory motion random onsets.Covariance bursts deep Fig. 5 effects breathing patterns Fig. 6 correlation structures scans breathing Effects minimally preprocessed data major data processing strategies matrices color effects p < 0.05 by 10,000 permutation tests (gray insignificant cells).Fig. 6Functional connectivity breathing patterns images color contrasts differences p < 0.05 10,000 permutation tests (gray cells insignificant). Contrasts three groups top mean differences groups bottom group correlations groups within-subject differences between scans without breathing patterns with bursts deep breaths Betas multiple linear regression separately each sex unrelated non-group-member subjects Regressors z-scored betas minimally preprocessed data Betas for bursts other data processing strategiesshows sets betas). three analyses bursts associated sensorimotor correlations (yellow deep breaths lack elevation (blue examined correlation structures three groups (Fig. 6a). burst group correlations elevated above clean deep breath groups high elevations sensorimotor distribution yellow deep breath group higher correlations avoiding sensorimotor distribution blue bursts no role Comparing groups to randomly formed groups recapitulated findings analyses contrasts of pure breathing patterns with typical breathing examined within-subject differences between scans with respiratory pattern bursts. 6b or deep breaths Within-subject burst effects widespread elevations sensorimotor emphasis yellow present in minimally preprocessed FIX-ICA-denoised data data motion regression global signals removed effects persist altered distribution lagged signal structure in bursts not captured mean signal effect pronounced when motion regression censoring with global signal regression not motion-caused Within-subject deep breath effects widespread elevations without global signal regression eliminated with global signal regression major modulation time-locked similar in all networkshint sensorimotor non-elevation in group contrasts processing strategies (dotted blue examined betas multiple linear regressions across subjects separately each sex (males top females bottom row unrelated non-group subjects similar spatial beta structures in males females bursts global elevations pronounced in sensorimotor distributions yellow deep breaths milder elevations avoided sensorimotor distributions (dotted blue Nuisance effects DVARS head motion congruent by sex distinct from respiratory patterns processing strategies illustrated Supplementary Fig. matrices from Figs. 5 6 arranged by respiratory pattern Supplementary Fig. 12 effects breathing paradigms influences sleep document relationships respiratory patterns to arousal sleep breathing limited in HCP data no measure arousal sleep expected associations breathing sleep tested studies young adults find sex differences in delay to sleep sex differences after menopause studies excessive daytime sleepiness find no sex studies sleep onset during fMRI report sex little reason expect sex difference in tiredness sleep onset sex-biased breathing effects young adults deep breaths associate with sleeprespiratory literature deep breaths with factors tiredness sleep13 No hard measure sleep in HCP data list of sleepy subjects kept by scanner technicians includes 37% HCP subjects three groups 71% deep breath group on list (p = 0.0014) compared to 38% clean group 41% burst group plausibility validity list subjects deep breaths likely enriched for yawning examined gray plots scans sleeping no visual signature sleep not sex-biased bursts snores or obstructive sleep apnea potentiated by obesity23 no correlation between burst scores body mass index (BMI) sex (male r = 0.07 p = 0.32 r −0.03 p = 0.66) differ in BMI mid-20s both sexes results support central not obstructive causes of bursts breathing patterns likely as scans progress influence arousal visual impressions deep breaths occur at any point scan not concentrated at end bursts uncommon at beginning emerge latermales females algorithm indices rose scans progressed t-tests minutes 1–4 11–14 scans significant p’s < 1e−16 for bursts < 0.02 deep statistical effects accord with visual impressions indicate deep breaths associated with sleepiness bursts central sex biases sleep unlikely sex differences bursts bursts more likely as scans progress both sexes described two respiratory patterns bias functional connectivity seen in young adults MRI scanners first effort event-related framework understanding effects spontaneous fMRI time series respiratory effects within-scan between scans across subjects populations links respiratory patterns to effects neuroimaging literature potential mechanistic basis of bursts males settings differential expression respiratory patterns Supplementary Discussion concepts mechanisms clinical associations of periodic breathing deep breaths bursts effects fMRI signals increased global functional connectivity most marked for bursts increased prevalence bursts in males higher fMRI signal correlations differences eliminated once brain volume respiratory patterns accountednot asserting all functional connectivity sex differences due to respiration or specific differences removed third variable could asserting respiration sex differences respiratory patterns distinct spatiotemporal effects bursts elevating correlations among sensorimotor brain regions deep breaths weakest Signal modulations time-locked across brain in deep breaths bursts multi-second lags between parts brain distinct cardiac effects indicate distinct cardiopulmonary neurophysiological events occurring during patterns future work disentangle pCO2-related cerebral blood flow effects respiratory pattern from neural activity sensorimotor pattern emerged in fMRI research studies localizing global fMRI signals at arousal bursts deep breaths define global fMRI signal unknown debates over retaining or discarding global fMRI signals important to understand denoising data processing strategies impact respiratory patterns examine phenomena caffeine administration vigilance tasks lag structure quasiperiodic patterns reasons suspect psychiatric symptoms scale with patterns clean group without bursts or deep breaths reported “clean” lifestyle little substance use psychiatric symptomatologyyoung adults deep breaths half scans over three-quarters subjects unsurprising respiratory Adult humans inhale deeply counter blood gas tensions reinflate collapsed alveoli sighs common in supine position scanning13 yawns likely occur some subjects burst pattern over one-third scans two-thirds subjects tendency in males lack correlation between BMI burst prevalence sex suggests central obstructive origin pattern centrally oriented explanation patterns arise via sex-biased respiratory control parameters unmasked as subjects relax scanner hypothesis pattern emerges via parameters governing chemoreflex explain waveforms sex bias of bursts generation waxing waning patterns breathing via interactions mechanisms controlling breathing Supplementary Discussion 35 patterns cyclic stability called periodic breathing (Fig. 7 shows periodic breathing Breathing rhythms generated in medulla under control volition waking neural drive sleep chemoreflex loop senses pH pCO28 chemoreflex central discussionperson arterial pCO2 resting pCO2 fluctuations engage negative feedback loop high pCO2 chemoreflex stimulates breathing reduce pCO2 below apneic ceases Cyclic respiration occur (excessive ventilation pushes pCO2 below apneic threshold respiratory pCO2 rises pO2 falls triggering resumption start next cycle.Fig. 7Comparison periodic breathing waveforms bursts single-subject waveforms in opioid use stroke heart failure high altitude newborns illustrations bursts in 11 HCP subjects plots same time scale green lines measure 1 min. long cycle times in patients heart failure exaggerated delay in detection changes arterial gas tensions stroke example added effect delayed circulatory time heart failure Images modified from46,54–58 permission Images 55 adapted permission American Thoracic Society Copyright © 2020 American Thoracic Society American Journal of Respiratory and Critical Care Medicine official journal Thoracic Society read entire article for correct context https://europepmc.org/article/med/15665317 authors American Thoracic Society not responsible for errors omissionsKey initiating cycles CO2 reserve difference between resting pCO2 apneic threshold sex differences due to gonadal hormones Women have same resting pCO2 levels lower apneic thresholds larger CO2 reserves36 Sex hormones influence parameters testosterone raises apneic threshold pCO2 CO2 Smaller CO2 reserves mean smaller ventilatory perturbation trigger apnea make apneic events less likely women less initiate periodic breathing study adults high altitude lowered pCO2 women displayed less central apnea periodic breathing highlight role sex hormones chemoreflex responses females less apnea Waxing waning cycles breathing studied during sleep measured via apnea/hypopnea index studies detect sex differences in AHI scores women fewer episodes evidence sex hormones apnea AHI sex biases reduced after menopause hypogonadal males have lower AHI women with testosterone-producing tumors have higher AHI sex biases of burst pattern could arise via sex-hormone dependent properties chemoreflex loop sex differences in apnea hypopneadisordered breathing commonly studied in older populations during NREM sleep chemoreflex properties exposed recent studies document breathing in daytime45 lessening wakefulness drive instabilities chemoreflex loop sleep reduced arousal to bursts data unclear new datasets monitoring respiration arousal needed brief arousals result from hypoxia after apnea perpetuate cycles central apnea increasing ventilation forms periodic breathing relation to chemoreflex loop bursts discussed in Supplementary Discussion influence breathing on fMRI signals marked results suggest breathing biases expected Supplementary Discussion reviews literature associations Yawns prevalent in tired subjects subjects on medications sedation bursts share mechanisms increased AHI anticipate increases in bursts over lifespan both sexes after males after menopause subjects on opioids respiratory suppressants brain injuries medical conditions heart failure studies reported dose-response associations of AHI with symptoms47 community samples obtain cross-sectional associations of depression and AHI39 Fluctuations in gonadal hormones over pregnancy menstrual cycles sex-hormone therapies likely influence burst prevalenceMedications influencing cholinergic noradrenergic serotonergic influence bursts depression AHI scores ease respiratory belts prudent to collect physiology records during fMRI scans (see ref. 14 implementations).MethodsData subject 900-subject Human Connectome Project) Young Adult data obtained Subjects young adults from St. Louis, Missouri identified via Missouri state registry of siblings twins Subjects healthy excluded with diabetes hypertension neurological psychiatric disorders HCP-YA protocol approved by Washington University in St. Louis Institutional Review Board WUSTL study conducted with Declaration of Helsinki Subjects provided consent compensated monetarily Each subject underwent four 14.4-min fMRI scans two days crosshair at altitude ~450 feet above sea level physiology data acquired via Siemens Physiology Monitoring Unit signals were 400 Hz recordings of abdominal belt finger pulse oximetry waveform Siemens respiratory record obtained via pressure hose to respiratory cushion under elastic belt output in arbitrary units.resting-state fMRI scans obtained including preprocessed FIX-ICA-denoised images head position physiology files HCP 900-subject release 440 subjects had four 14.4-min fMRI scans complete physiological data cardiac respiratory traces data in Supplementary material ref. 48). 440 subjects analyzed Characteristics age 28.6 ± 3.8 (range 22–36) 228 males 212 females BMI 26.5 ± 5.0 (range 16.5–43.9).For 440 subjects high data files obtained four resting-state fMRI scans transformed atlas space REST1_LR_RL_LR_[RUN].nii.gz_hp2000_clean.nii.gz scans obtained preprocessed FIX-ICA-denoised data [RUN]_Physio_log.txt Movement_Regressors_dt.txt files obtained Structural scans transformed atlas space obtained T1w.nii.gz aparc+aseg.nii.gz files anatomical T1-weighted scan FreeSurfer segmentation.Image aparc+aseg.nii.gz file underwent serial erosions white matter ventricle segmentsMasks of cortical gray matter cerebellum subcortical nuclei extracted eroded layers of superficial deep masks of white matter ventricles masks include in-brain voxels used to extract signals order signals for gray plots49 For plots thermal noise in HCP scans within-mask 6 mm FWHM Gaussian kernel applied to data using masks data ref. blurring mix tissue compartments beyond partial volume effects in voxels.Respiratory belt pulse oximeter traces (sampled at 400 Hz underwent visual inspection quality for peak detection subjects with traces likely peak detection analyzed48 After selection outlier replacement filter used to eliminate spurious spike artifacts traces blurred to peak detection for 400 Hz treated respiratory traces shown in Figures respiratory measures derived from treated respiratory belt trace envelope of trace over 10-s window (at 400 Hz) calculated after ref.(Matlab command envelope(zscore_resp_trace,4000 RV measure standard deviation respiratory trace within 6-s window calculated(zscore_resp_trace 2400 RVT measure ((peak-prior trough/(time between peaks calculated Peak detection yielded peaks troughs inverted trace (Matlab (findpeaks(zscore_resp_trace,‘minpeakdistance’,800,‘minpeakprominence’.5))) minimum peak distance presumes breaths 2 s apart preceding trough no value scored traces measures visually checked outliers abnormalities measures ENV, RV RVT figures RVT calculated computation ref. 14 ref. 10.Pulse oximeter traces z-scoring peak detection (Matlab findpeaks(zscore_pulseox,180, Heart rate calculated from interval between peaks minimum peak distance presumes heart rates under 133 beats per minute Peak amplitude calculated from height peak previous trough Cardiac traces prone to disruptions laborious to check large peaks troughs limited cardiac records traces derived measures visually checked accuracydata quality measure DVARS calculated after refs. 50,51 root mean value brain time series differentiated by backward differences 0 first timepoint.Head position from Movement_Regressors_dt.txt files parameters after subtracting first timepoint Head motion represented by Framewise Displacement (FD) measures ref. 51 differentiated by backward differences rotational measures converted to arc displacement at 5 cm radius sum absolute value calculated motion head position traces filtered stopband 0.2–0.5 Hz ref. 16. FD calculated by backward differences preceding timepoint sampling rates 2–4 s calculated by backward differences over four timepoints (4 * 720 ms = 2.88 s sampling rate ref. 52 head motion measures FDoriginal FDfiltered,4-TR examined plot created scan Figs. 1 2 procedures refs. 14,16 4 * 440 = 1,760 plots burst pattern single deep breaths 35 bursts deep breaths non-respiratory motions Fig. 3 chosen by visual inspection respiratory motion records marked onsets Supplementary Movies 2–4 Supplementary Data 1. Random onsets defined selected runs onsetsProperties scans 90 shown in heat maps gray bars red heat maps show unpaired two-sample t-tests random onset group Heart rate derived from peak-to-peak intervals pulse oximetry data visually verified Plots show mean properties 35 onsets shade plots show mean standard deviation.Group JDP examined 440 subjects deep breaths bursts normal tidal breathing patterns screened for siblings discarded largest equally-sized groups three groups of 21 subjects unrelated contrasts demographic properties described in Supplementary Materials associations medical psychiatric variables identities restricted to registered HCP users Restricted Access Subject Key access Keys Sign in to https://db.humanconnectome open WU-Minn HCP data set Click “Subject Keys” tab three subject keys under “Published Subject Keys Dataset” Click subject key access data_etal_2020_NatureCommunications HCP users restricted data access use subject keys error message “Restricted Data Access Required!” request restricted data access.Ratings breathing patterns in scansTwo authors (J.D.PC.J.L. examined scans subjects 400–440 breathing patterns rated scans 1–399 binary decisions presence deep breaths bursts Cohen’s kappas calculated ratings Fig. 4 high inter-rater reliability likelihood sex differences breathing pattern determined by Chi-squared tests raters obtained sex effects bursts not deep breaths sum four scans pattern score.Algorithmic indexing breathing algorithm described Supplementary Materials priors derived from respiratory traces likelihood deep breaths bursts multiplied into fMRI templates global fMRI signals total values each scan obtained averaged subject indices downweighed by variation breathing rate haphazard breathing Indices correlated to pattern scores compared by sex using unpaired two t-tests Fig. 4.Functional connectivity global functional connectivity (gFC) analyses pairwise correlations voxels gray matter mask computed each fMRI scan median value taken followed Fisher-Z transformation values represent central tendency gray matter functional connectivity cerebral blood flow changes-wide fMRI signals values averaged over scans for cross-subject correlations linear regressions ANOVA/ANCOVA in Fig.gFC rater scores compared by correlation Differences sex compared unpaired two-sample t-test gFC scores males females formula gFC = b0 + b1*burst_score + b2*deep_breath_score similar fits each sex Fig. 4k ANOVA ANCOVA 2–6) gFC sex variables main effects color-coded Fig 4k 1 ANOVA sex 2 adds head size ANCOVA other models ANCOVAs terms fit exception sex variable insignificant final two models DVARS three versions head motion added models 1–4 failed negate sex effect fit main effects respiratory variables present network analyses-parcel scheme of Gordon et al.53 images cluster assignments define resting-state networks illustrated Fig. 5a 11 Correlation computations via Fisher-Z transforms converted to Pearson r values for matrices in Figs. 5 611 12 cells significant at p < 0.05 colored determined by 10,000 permutation tests gray cells insignificant).Minimally preprocessed (MP FIX-ICA-denoised time series sampled signal processing steps signals MP mean gray matter signal removed MP data six motion regressors derivatives timepoints with z-scored DVARS values >2 (MP+mot MP data GSR sense effects signal processing regimes Fig. 5 35 onsets parcel time series −10 to +40 s correlation matrices mean values breathing patterns masked by significance permutation tests random matrices Contrasts of breathing patterns shown in data processing strategies Mean parcel time series in Fig. 5c red dotted lines span correlations Mean values grayordinates in MP data Fig. 5d Supplementary Movie Fig. 6 correlations from scans used Figure 6a subtraction of mean matrices groups screened by permutation tests bottom row values observed group matrices random groups Figure 6b within-subject differences between scans with without breathing patternstop row shows scans with bursts deep breaths+D− compared without bottom row contrast for deep breaths scans raters agreed patterns used unrelated subjects scan subtypes used each contrast 50–70 subjects each contrast differences in data processing strategies masked by permutation tests swapped scan subtype labels Figure 6c illustrates betas linear regression breathing pattern scores head size nuisance variables scaled performed separately each sex regressors standardized masked by significance among betas permuted subjects Betas for bursts shown several processing strategies Full fits to all variables in Supplementary Fig. 11 processing strategies Nature Research Reporting Summary.Supplementary | 48 | 1.138135 |
10.1038/s41467-021-21656-9 | PMC7935868 | The cost and complexity of whole genome sequencing limits its use in identifying and validating sequences used for genetic engineering and synthetic biology. Here the authors present Prymetime, an integrated workflow to sequence engineered strains and identify engineering in metagenomes. | Yeast whole genome sequencing (WGS) lacks end-to-end workflows that identify genetic engineering. Here we present Prymetime, a tool that assembles yeast plasmids and chromosomes and annotates genetic engineering sequences. It is a hybrid workflow—it uses short and long reads as inputs to perform separate linear and circular assembly steps. This structure is necessary to accurately resolve genetic engineering sequences in plasmids and the genome. We show this by assembling diverse engineered yeasts, in some cases revealing unintended deletions and integrations. Furthermore, the resulting whole genomes are high quality, although the underlying assembly software does not consistently resolve highly repetitive genome features. Finally, we assemble plasmids and genome integrations from metagenomic sequencing, even with 1 engineered cell in 1000. This work is a blueprint for building WGS workflows and establishes WGS-based identification of yeast genetic engineering. | IntroductionComplete and accurate detection of genetic engineering is needed to validate strain engineering, protect intellectual property, monitor for release events, and detect engineered organisms in unknown samples. Whole genome sequencing (WGS) is an attractive detection method because it does not depend on specific sequence features and captures all sequences – including intended and unintended modifications. Yet, precise resolution of genetic engineering places strict requirements on a WGS workflow – genetic engineering signatures must be clearly identified within accurate, complete, and contiguous sequences.Thus, a WGS workflow is needed for engineered organisms. In this work, we focus particularly on engineered yeasts. Yeasts are a crucial testbed for genome-scale design1,2, and accurate WGS will be necessary for validating synthesized eukaryotic genomes. Yeast are also cell factories for medicines3,4, fuels5,6, materials7,8, and chemicals9,10. These are derived from several species of baker’s yeast Saccharomyces cerevisiae11–13 and nonconventional yeasts like Yarrowia lipolytica14–16 and Komagataella phaffii (formerly Pichia pastoris)17,18. Given the economic importance and increasing use of engineered yeast cell factories, it is crucial that WGS methods are developed that can efficiently validate the presence of intended engineering and confirm the absence of unintended variation. Without WGS, the majority of yeast strains are currently validated with less comprehensive methods like PCR and targeted sequencing. These methods do not capture the unintended secondary mutations common in engineered organisms19–23. There are also many unpublished accounts of WGS revealing unexpected sequences and genome structures in engineered industrial strains. Taken together, this evidence challenges the assumption that an observed phenotype is the direct result of intended engineering, illuminating a possible explanation for variation between replicates and irreproducible findings – a common problem for biology-related disciplines24. Clearly, WGS must be used more broadly to detect and validate genetic engineering in yeasts.Yeast engineering leaves many predictable sequence features in the genome, like standard plasmid sets with known replication origins and expression parts25–28, integrations29–32, gene knockouts33, and genome edits using RNA-guided endonucleases34–39. Many of these can be identified in a genome sequence with a tool such as BLAST40. Yet, engineered yeast present several obstacles to complete, accurate genome assembly. The high sequence identity in many engineered constructs, such as common plasmid elements or parts derived from the host genome, can cause identical sequences to be omitted41,42. Engineered yeast also have complex genome features like multiple deletions13, multiple plasmids with varying copy numbers30, many insertions36, and SCRaMbLEd chromosomes43,44. Furthermore, the scale of yeast engineering is increasing both in the fraction of a genome that may be rewritten45,46, and in the numbers of engineered strains created through adaptive laboratory evolution47–49 and combinatorial pathway engineering50–54. These iterative approaches result in many strains that are costly to sequence. These obstacles are in addition to typical complexities like naturally repetitive regions (telomeres and ribosomal DNA), rearrangements, and polyploidy. Each make accurate, complete, and contiguous yeast genomes difficult to attain without a significant allocation of resources.A WGS workflow consists of five steps: DNA isolation, library preparation, sequencing, assembly, and annotation. First, genomic DNA is purified using one of a variety of methods, including phenol-choloroform, bead beating, or enzymatic lysis55. Second, the sequencing library is prepared by attaching adapters and barcodes. This can be done via ligation, which involves shearing the DNA to create free ends for DNA ligase to attach adapters, or tagmentation, which randomly inserts adapter attachment points without shearing56. Third, the library is sequenced with a next-generation sequencing (NGS) platform that either generates short reads (150–300 base pairs long) with high nucleotide accuracy56 or long reads (1.5 kilobases to megabases long) with lower accuracy57. The average read length and the number of reads (genome coverage) output by the NGS platform is dependent on sequencing technology and the preceding DNA isolation and adapter attachment steps58. Fourth, the reads are computationally assembled into a final genome sequence with software that uses either an overlap-layout-consensus (OLC) or De Bruijn graph (DBG) algorithm59. OLC and DBG assemblers are further classified into short read only, hybrid (both short and long reads), or long read with error correction. Both hybrid and long read with error correction assembly approaches currently hold the most promise to achieve accurate genome sequence and structure at low read depths, primarily because two independent technologies validate basecalls. However, this entails the use of two sequencing technologies, thereby increasing costs and time. Fifth, an annotation is performed. Eukaryotic annotation involves first predicting genes in the genome sequence, followed by functional annotation60. However, features like genetic engineering parts, telomeres, centromeres, mitochondrial DNA, and natural plasmids are often not annotated, and several are by convention not included in the final assembly.In this work, we develop an inexpensive WGS workflow designed to detect genetic engineering in pure and mixed samples of engineered yeast. To accomplish this, we optimized each of the five steps in the WGS workflow in order to correctly resolve all engineering sequences in a heavily engineered yeast strain. We first improved DNA isolation and sequencing library preparation to increase representation of reads from circular DNA molecules. We then used a combination of long- and short-read sequencing from inexpensive sequencing platforms to achieve high coverage at low cost. We integrated two different assemblers to resolve both circular plasmids and linear chromosomes accurately. We developed an annotation approach based on a user-input list of genetic parts to clearly identify common signatures of engineering. Using this approach, we also annotated centromeres, telomeres, origins of replication, and mitochondrial DNA in order to put observed genetic engineering in context with the rest of the genome. The resulting workflow is named Prymetime, "Pipeline for Recombinant Yeast genoMEs That Identifies Markers of Engineering." Through a variety of demonstrations, we show that Prymetime can validate genetic engineering, produce high quality whole genome sequences, and detect engineering in metagenomic samples. This tool is broadly useful for strain validation, release monitoring, protecting intellectual property, and investigating engineering in unknown samples.ResultsOptimizing nanopore sequencing library preparation for engineered yeastsAt the beginning, we set a standard that a genome assembly workflow must be able to resolve chromosomal integrations and multiple plasmids used in yeast engineering. Therefore, we built a S. cerevisiae CEN.PK113 strain, FEY_2, containing an integrated carotenoid pathway, the native 2μ plasmid, a dCas9 plasmid, and a gRNA plasmid, shown in Fig. 1a. Initially, we prepared sequencing libraries of FEY_2 with the Oxford Nanopore Technologies (ONT) ligation kit. Sequencing these initial libraries had low average read length that varied from run to run, possibly because of differential DNA shearing during isolation. To limit this, we developed a gentle genomic DNA isolation protocol which increased average nanopore read length and reduced variance (Supplementary Figure S1). However, the sequencing results contained few reads from plasmids, as determined by comparing the average normalized mapped reads of the plasmid antibiotic selection markers to those of the ACT1 genomic locus using Minimap261. We could isolate plasmids from FEY_2 using a yeast miniprep kit, so we reasoned that the sequencing library preparation step was so gentle that it was not linearizing circular plasmids for adapter ligation. Thus, we turned to a tagmentation library preparation method, the ONT Rapid kit. The improvement in average normalized mapped plasmid reads is shown in Fig. 1b. We were reassured that the 2:1 and 20:1 marker to ACT1 read coverage ratios for each plasmid are equivalent to the approximate plasmid copy number in yeast for each origin12,62. Furthermore, tagmentation also increased the representation of other circular elements like the native 2μ plasmid and mitochondrial DNA. These results indicate that tagmentation is a key to achieving long average read lengths while also generating linear molecules from small circular DNA so that they can pass through the nanopore flow cell. Whereas tagmentation may have a slight AT sequence bias and perform poorly in extreme GC genomes, this is not the case with our yeasts. Thus, with gentle isolation and tagmentation, nanopore sequencing of FEY_2 resulted in adequate representation of plasmid reads.Fig. 1Detection of engineering signatures in S. cerevisiae FEY_2.a Photograph of FEY_2 streaked onto an agar plate, showing a functional carotenoid pathway. The illustration shows the engineering signatures comprising FEY_2, which included a carotenoid pathway chromosomal integration, a low copy plasmid expressing dCas9, and a high copy plasmid expressing gRNA. b Approximate copy number from genomic DNA libraries prepared by Oxford Nanopore Technologies’ (ONT) Ligation and Rapid kit and Illumina’s Nextera kit for the low copy and high copy plasmids in FEY_2. c BLASTN results from querying known engineering signatures against assemblies. The genome assemblers were categorized as short-read only, hybrid, or long read with error correction. The underlying algorithm type of each assembler, De Bruijn graph (DBG) or overlap-layour-consensus (OLC), is also shown. Failure modes of genome assemblies were shown as red lines (contig break) and white spaces (missing fragment). The colored pathways and plasmids represent assemblies where all engineering signatures were found in contiguous sequences.Developing a de novo assembly workflow for complete, contiguous plasmids and integrationsOnce we achieved appropriate read representation, we evaluated nine assembly algorithms with the stringent requirement that all chromosomes and plasmids must be complete, accurate, and contiguous. This is particularly stringent for the three plasmids in FEY_2 because they each have significant sequence identity between each other and the genome. The assemblers tested included the short-read only OLC assembler Edena63, the short-read only DBG assemblers ABySS64 and Velvet65, the hybrid OLC assembler Masurca66, hybrid DBG assembler HybridSPAdes67, the long-read OLC assemblers MiniASM68, Canu69, and SMARTdenovo70, and the long-read DBG assembler Flye71. The long-read assemblers, because of higher error rates57, only provide a "skeleton" for mapping additional reads72–76. Therefore, all long-read assemblies were polished with Medaka77, Racon78, and Pilon79.We used the optimized library preparation to obtain long reads at 60X genome coverage from the ONT MinION and short reads at 125X genome coverage from the Illumina iSeq 100. This common set of reads was used by each assembler, and the resulting genome assembly was analyzed using BLASTN for the presence of the integrated pathway, both plasmids, and the native 2μ plasmid. A visual representation of the BLASTN results is shown in Fig. 1a. The engineering features were rarely complete or contiguous. The short-read only de novo assemblers ABySS, Edena, and Velvet returned a fragmented, incomplete pathway and plasmids. The hybrid assemblers SPAdes and Masurca produced more complete sequences than the short-read only assemblers, but the genome integration was fragmented, and Masurca also omitted portions of the three plasmids. The long-read de novo assemblers MiniASM, Canu, Flye, and SMARTdenovo each returned a single contiguous sequence for the genome integration, yet, MiniASM, Canu, and SMARTdenovo omitted sections of the three plasmids. Only Flye eventually returned the genome integration and each plasmid correctly in contiguous sequences.Of the assemblies missing large portions of at least one of the three plasmids, almost all were generated with an OLC assembler. These algorithms use an All-versus-All consensus step that may discard highly identical sequences in order to reach consensus. To investigate this further, we used BLASTN at each step in the OLC-based Canu pipeline to determine when sequences were omitted. We noted that the complete low-copy plasmid was initially present before the consensus step, and was then lost in the final assembly. It seems that Canu discarded the plasmid at a certain threshold during the consensus step, likely because of high sequence identity with the other plasmids. In contrast, the DBG assemblers Flye, ABySS, and SPAdes did not omit sections of plasmids. DBG algorithms split reads into shorter k-mers followed by a Eulerian walk approach to construct contigs, thus DBG may be less prone to discarding highly identical sequences80. Though complete, the plasmid sequences from ABySS and SPAdes were fragmented, while Flye assembled each plasmid into a single contiguous sequence. This is possibly because ABySS and SPAdes are hybrid assemblers that assemble short reads first, then use long reads to piece together contigs. This makes them subject to the same pitfalls that plague short-read assemblers, in that the reads do not effectively span sequences with high identity. Thus, Flye, as a DBG assembler that assembles long reads first, produces higher contiguity and better resolution of sequences with high identity. These findings reinforce that genome assembly quality is dependent on high-quality long-read data and a de novo assembly approach.While the plasmid contigs from Flye were complete and contiguous, they were longer than expected. Further inspection revealed that the contigs consisted of several repeats of the expected plasmid sequence. This is a common problem for long-read assemblers, as they use linear logic to merge contigs69,81. To obtain structurally representative plasmid contigs, we chose to re-assemble them with Unicycler, software that was built to assemble circular contigs from bacterial sequencing data82. To do this, we sent contigs either identified by Flye as circular or identified by mummer as repetitive to Unicycler. Reassembly of plasmids with Unicycler improved the accuracy as measured by BLASTN and length of the contigs for the three plasmids in FEY_2 (Supplementary Fig. S2).Resolving engineering signatures in a collection of engineered yeastsWe next validated our assembly approach on a collection of engineered laboratory and nonconventional yeast. We constructed 15 strains from S. cerevisiae S288C, CEN.PK113-7D, W303-α, BY4741, BY4742, and K. phaffii ATCC 76273 (CBS 7435)83,84 and Y. lipolytica ATCC MYA-2613 (Po1f)85. Plasmids designed to construct transcriptional units for this study are described in Supplementary Fig. S3. A description of each strain is shown in Fig. 2a, with more details in Supplementary Table S1 and Supplementary Fig. S4. Engineering signatures were inserted into the genome or maintained on episomal plasmids. S. cerevisiae integrations were targeted to the HO locus26 or between NRT1 and GYP7 in chromosome XV38,51. S. cerevisiae plasmids consisted of custom TypeIIS-compatible yeast shuttle vectors with either S. cerevisiae replicon (2μ or CEN6/ARSH4). Engineering was broadly categorized into biosynthetic pathways, gene editing components, deletions, and synthetic biology elements. Biosynthetic pathways included propane86, 2β-carotene87, prespatane88, carnosic acid89, and limonene90,91. Genome editing associated tools included SpCas934, dCas935, LbCpf138, FnCpf137, and Cre recombinase33. Deletions included the synthetic auxotrophies already present in S. cerevisiae W303-α, BY4741, BY4742, and Y. lipolytica Po1f. Synthetic biology elements included fluorescent proteins92,93 and the 2A sequence94. The engineered Y. lipolytica strain "FEY_74" contained the CRISPR-Cas9 plasmid pCRISPRyl39. The engineered K. phaffii strain "FEY_75" contained a recombinase-integrated red fluorescent protein (RFP) cassette28.Fig. 2Resolving signatures of engineering from the panel of engineered yeast strains.a Visual representation of the BLASTN results from querying known engineering signatures against Prymetime-assembled genome assemblies of the 15 engineered strains. Failure modes of genome assemblies were shown as red lines (contig break) and white spaces (missing fragment). The colored pathways and plasmids represent assemblies where all engineering signatures were found in contiguous sequences. b The expected CRISPR-Cas9 expression vector for FEY_74, an engineered Y. lipolytica strain, and the actual plasmid from the Prymetime genome assembly. The DNA agarose gel confirms the missing Cas9 cassette from the FEY_74 strain in comparison to the original pCRISPRyl plasmid. The agarose gel represents one experiment, where the PCR products of the pCRISPRyl plasmid and FEY_74 plasmid were processed in parallel. c Illustration showing the expected location of the RFP integration cassette into chromosome II of FEY_75, an engineered K. phaffii strain, and the actual location of the cassette into chromosome IV.We sequenced this collection with the ONT MinION and the Illumina iSeq 100 systems using our optimized library preparation protocols. The combined assembly approach using Flye and re-assembly of circular contigs with Unicycler captured each engineering signature in each S. cerevisiae genetic background as measured by BLASTN of the reference sequence against the assembly. Shown in Fig. 2a, the approach resolved seven different genome integrations in two genome loci and eleven different plasmids. The BLASTN metrics are in Supplementary Table S2. To further demonstrate the necessity of a combined assembly approach, we repeated assembly with Flye alone. The additional Unicycler step improves plasmid length and accuracy in every strain, not just FEY_2 (Supplementary Fig. S5). No sequence complexities, like the type of gene (metabolic, selective, editing, or reporter), parts identical to the genome (Ptef1, Pgal10), or plasmid copy number, affected the accuracy or structural completeness of the assemblies.The genome assemblies from the two engineered nonconventional yeasts – Y. lipolytica strain FEY_74 and K. phaffii strain FEY_75 – revealed unintentional edits. FEY_74 was intended to contain the pCRISPR-yl plasmid39, yet the contig from the genome assembly was missing the entire Cas9 transcription unit and a portion of the E. coli origin of replication, shown in Fig. 2b. Inspection of the raw reads failed to identify a single read with the missing sequence. We performed a genomic DNA isolation and a yeast plasmid miniprep on FEY_74 and transformed the resulting DNA back into E. coli, yet did not observe any colonies. This indicates that the disrupted origin of replication in the assembly reflects an actual unintended loss rather than an assembly error. This was further confirmed by PCR of DNA isolated from FEY_74 with primers spanning the missing region of the plasmid. The length of the PCR product indicated that the sequence was indeed missing (Fig. 2b). Similarly, FEY_75 was designed to have an RFP transcription unit integrated into chromosome II (Fig. 2c). The entire pathway was found by BLASTN in the FEY_75 genome, but analysis revealed that the pathway was actually integrated into chromosome IV. This was further confirmed by PCR of the integration site in chromosome II, which was negative, yet the strain remained nourseothricin resistant and RFP positive. These results indicate that a combined assembly approach can be used to find and accurately reproduce engineering, which is useful for both strain quality control and identification of engineering in unknown samples.Whole genome assembly qualityAfter achieving assembly of all engineering sequences, we then assessed whole genome quality of the 15 engineered assemblies and genomes from the parent nonconventional yeasts Y. lipolyitica Po1f and K. phaffii CBS7435. Each genome had high contiguity, sequence accuracy, and genome completeness as measured by Benchmarking Universal Single-Copy Orthologs (BUSCO) score95, calculated using the Saccharomycetales dataset (Fig. 3a) and percent aligned reads to the parent genome (Fig. 3b). Percent unmapped reads for each genome are provided in Supplementary Table S3. Whole genome alignments for each genome compared to the parent with Mauve96 are presented in Supplementary Figs. S6 and S7. The resequenced Y. lipolytica Po1f and K. phaffii CBS7435 strains were improved compared to the publicly available genomes16,84 by several metrics (Supplementary Fig. S8). Notably, there are 6 more essential genes recovered in the resequenced K. phaffii assembly and 13 more essential genes recovered in the resequenced Y. lipolytica assembly.Fig. 3Whole genome assembly quality for the panel of engineered yeast strains.a BUSCO genome completeness score for all engineered yeast genome assemblies and their respective reference parent strain genome assemblies. b Percentage of aligned bases for each chromosome of the reference parent strain assemblies captured by the engineered yeast assemblies. We could not determine the 16 chromosomes for the reference BY4741 assembly due to its discontiguity, so the FEY_55 assembly was compared to the BY4742 reference assembly. c The number of long terminal repeats (LTRs) predicted by LTRpred for all engineered S. cerevisiae genome assemblies and their respective reference parent strain genome assemblies. d The approximate copy number of CUP1 repeats in the genome assembly, raw Nanopore reads, and raw Illumina reads for all engineered S. cerevisiae strains, along with the CUP1 copy number in the respective reference parent strain assemblies. e The approximate copy number of rDNA repeats in the genome assembly, raw Nanopore reads, and raw Illumina reads for all engineered S. cerevisiae strains, along with the rDNA copy number in the respective reference parent strain assemblies. These are also tabulated in Supplementary Table S4.The final test of completeness is whether a chromosome is resolved from a telomere, through the centromere, to the other telomere. We compared each engineered S. cerevisiae assembly to its respective reference assembly to quantify the number of complete telomere-to-telomere contigs (Supplementary Fig. S9). We found that 76% of chromosomes are complete, except the telomeres. Analysis of several smaller contigs in the assemblies reveals them to be telomeric or ribosomal DNA (rDNA) sequences. This result shows that the genomes are essentially complete, save misassembly of highly repetitive genomic sequences and telomeres.Next, we further assessed repetitive DNA elements in each S. cerevisiae genome, finding that repetitive elements like long terminal repeats (LTRs), CUP1 repeats, and rDNA are resolved with comparable copy number to the reference genomes (Fig. 3c, d, e, respectively). However, the CUP1 and rDNA repeat copy numbers were underrepresented in both our assemblies and the reference assemblies when compared to the approximate copy number of the raw Nanopore and Illumina reads. The S. cerevisiaeCUP1 copy number is highly variable, ranging from zero to 7997, while the rDNA copy number is known to range between 100 and 20098. Tandem repeats such as CUP1 and rDNA are a common problem for all de novo assemblers and are often collapsed during assembly99.Every strain investigated in the above collection is haploid. Therefore, we sequenced the heterozygous diploid strain S. cerevisiae BY4743. The resulting assembly is similar to S. cerevisiae S288C (Fig. 4a). Thus, this assembly approach cannot resolve ploidy. However, the LYS2 and MET15 heterozygous deletions can be clearly resolved by mapping average read count (Fig. 4b, c, respectively).Fig. 4Genome assembly analysis of the heterozygous diploid S. cerevisiae strain BY4743.a Mauve genome alignment between the S. cerevisiae S288C reference assembly and the S. cerevisiae BY4743 genome assembly. The colored blocks represent regions of the genomes that align, while the vertical red lines indicate a new contig. b Nanopore read coverage around the heterozygous LYS2 gene. c Nanopore read coverage around the heterozygous MET15 gene.Taken together, these results indicate that the genome assemblies generated by the combined assembly approach are structurally correct, accurate, and complete, although telomeres, repeat elements, and ploidy remain a challenge to accurately reproduce. This is currently a challenge in the field of de novo genome assembly.Annotating and visualizing engineering and genome featuresThe last step in WGS, annotation, does not usually identify genetic engineering sequences. Therefore, we developed an engineering annotation step and applied it to the collection of 15 engineered yeasts. We first wrote an automated BLASTN script to find standard yeast genetic engineering parts and genome features. Standard parts include the CEN6/ARSH4 and 2μ replication origins, selection markers, promoters, and terminators. Genome features include centromeres, telomeres, and mitochondrial DNA, which were sourced for each parent strain from the Saccharomyces Genome Database100. This list of parts and features is simply a FASTA file, which can be easily modified and updated to find any sequence of interest in genome assemblies.We then fed the BLASTN results to two interactive genome viewers – chromoMap101 and AliTV102. ChromoMap highlights the parts and features within each contig in the assembly. AliTV does the same, but also aligns the assembly to the parent strain using lastz103. This can highlight potential unintended changes like chromosomal rearrangements. The chromoMap visualization for FEY_2 (Fig. 5a) shows the integration in scaffold_3, and the two engineered plasmids in scaffold_18 and scaffold_23. The output is interactive, so hovering over the engineering blocks will display which parts were identified. Using this approach, the plasmids can be differentiated from other small contigs by the presence of the origins of replication and other engineering sequences. In the AliTV visualization, the high sequence identity and contiguity of the engineered as compared to unengineered S. cerevisiae CEN.PK is apparent. The AliTV visualization is also interactive and customizable, and is particularly useful to determine how contigs from the assembly align to the reference assembly.Fig. 5Visualizing engineering and genome features, and the Prymetime pipeline.a chromoMap interactive visualization displaying engineering signatures and structural elements identified in the FEY_2 genome assembly. b AliTV interactive visualization of the FEY_2 genome assembly aligned against its parent CEN.PK113-7D genome assembly. Engineering signatures and structural elements are also annotated. c Overview of Prymetime genome assembly pipeline.Creating an automated pipelineOptimization of each of the five steps of genome assembly led to a final set of methods and software that can accurately reproduce and visualize genetic engineering in highly accurate yeast genomes. We integrated each of the software steps into a single Dockerized tool that we call Prymetime, "Pipeline for Recombinant Yeast genoMEs That Identifies Markers of Engineering." The final pipeline is depicted in Fig. 5c. The software accepts long reads and short reads, and optionally accepts a list of sequences of interest and a reference genome. It outputs two interactive visualizations of the genome. Each visualization of the 15 engineered strains is depicted in Supplementary Figs. S10–S17.As a final demonstration, we tested each step in the Prymetime workflow with a set of publicly available raw reads for S. cerevisiae CEN.PK113-7D74,104, assessing the quality at each step (Supplementary Fig. S18). First, we evaluated the contigs from Flye step, determining that 40X long-read genome coverage is sufficient to match the reference assembly. Then, we evaluated the polishing step, which demonstrated that at least 40X short-read genome coverage is needed to achieve high identity to the reference, BUSCO, and percentage of S. cerevisiae S288C CDSs (Supplementary Tables S5–S8). Using the chromoMap visualization output from Prymetime, the CEN.PK113-7D assembly correctly captures the centromeric sequences, but not the telomeric sequences (Supplementary Fig. S19). This corroborates the observations from the engineered genomes. Using different de novo assemblers still does not solve this problem (Supplementary Table S9), thus Flye remains the best underlying assembly software for assembly of accurate, complete, and contiguous genetic engineering sequences. A detailed illustration of the full Prymetime workflow is shown in Supplementary Fig. S20. These results confirm the Prymetime software workflow is as accurate as possible and show that at least 40X genome coverage for both long- and short-read sequencing data is needed to achieve the highest quality genomes.Resolving signatures of engineering in an in silico metagenome assemblyTo demonstrate a use case for Prymetime, we attempted to resolve engineering signatures in a metagenome. Publicly available reads from the Zymo mock metagenome were combined with reads from the FEY_15 strain to simulate detection of an engineered strain in a mixed sample. The mock metagenome consists of eight bacteria species – Bacillus subtilis, Enterococcus faecalis, Escherichia coli, Lactobacillus fermentum, Listeria monocytogenes, Pseudomonas aeruginosa, Salmonella enterica, and Staphylococcus aureus – and two yeast species – Cryptococcus neoformans and Saccharomyces cerevisiae105. To simulate different abundance levels, the FEY_15 nanopore reads were diluted with increasing numbers of Zymo metagenome reads at approximate ratios of 1:1, 1:10, 1:100, and 1:1000 based on number of base pairs. All of the FEY_15 and Zymo metagenome Illumina reads were combined together at an approximate ratio of 1:20 (Fig. 6a). These read sets were then used for Prymetime assembly. In each read set, the integration and plasmid of FEY_15 were completely resolved (Fig. 6b). This shows that synthetic biology parts, and their context, can be resolved in mixed samples by Prymetime.Fig. 6Resolving signatures of engineering in an in silico metagenome assembly.a Publicly available reads from Zymo’s mock metagenome were combined with reads from the engineered S. cerevisiae strain FEY_15. b Visual representation of the BLASTN results for the in silico metagenome. Failure modes of genome assemblies were shown as red lines (contig break) and white spaces (missing fragment). The colored pathways and plasmids represent assemblies where all engineering signatures were found in contiguous sequences.DiscussionThis work develops an integrated workflow for WGS of engineered yeasts which may be extensible to all eukaryotes with a mixture of linear and circular sequences. The workflow consists of gentle gDNA isolation, tagmentation, long- and short-read NGS, accurate de novo assembly of both linear and circular elements, and annotation of genetic engineering parts and genome features. Using this, diverse engineering signatures can be resolved in complete, contiguous sequences even with multiple similar plasmids in one strain. The resulting whole genome quality is comparable to high-quality reference assemblies, therefore, it is possible to generate accurate genome assemblies both before and after engineering. This permits verification of genetic engineering in yeasts with WGS to validate strain engineering. Further, the workflow performs well using metagenomic data, permitting detection of yeast engineering in mixed samples.This work demonstrates the challenges in making effective WGS-based workflows. Interestingly, we found that only the Flye assembly algorithm supported accurate resolution of genetic engineering in complete, contiguous sequences. We observed that sequence omission commonly occurred with assemblers built around OLC algorithms, which struggle to reproduce the expected representation and resolution of repeats68,69,106. Furthermore, we observed that short-read and hybrid assemblers commonly produced fragmented sequences. Thus, Flye, as the only long-read DBG assembler, was consistently the best at resolving genetic engineering signatures. These observations highlight the difficulty of applying otherwise effective genome assembly software to engineered yeasts, which have highly identical genetic engineering signatures and repetitive genome features. Furthermore, all assemblers collapsed repetitive genome features and struggled to resolve telomeres. This limits the ability of the tool to detect variations in rDNA, SNPs, and rare variants. Based on our results, assemblers aiming to improve these areas should be benchmarked against the overall performance of Flye.To date, WGS has rarely been used in strain engineering cycles due to the barriers of cost, time, and required bioinformatics expertise. The WGS workflow we developed with the inexpensive ONT MinION and Illumina iSeq 100 platforms and the integrated, dockerized Prymetime software package overcomes these barriers. With Prymetime, we were able to achieve high-quality genomes at relatively low read depth, finding that 40X for both long and short reads was sufficient for high accuracy, completeness, and contiguity of genetic engineering sequences, and quality whole genomes. With 40X read depth, up to 30 S. cerevisiae genomes can be sequenced on one MinION flow cell and up to 4 genomes can be sequenced on one Illumina iSeq flow cell. This is because approximately 0.5 Gb is needed for 40X read depth of the 13.4 Mb S. cerevisiae genome (factoring in collapsed rDNA repeats107) and our typical yield is approximately 15 Gb from the MinION and 2.4 Gb from the iSeq 100. Not accounting for labor, this level of multiplexing would cost around $200 per genome. The entire workflow is fast – it takes under a week to start from a single colony and acquire a genome assembly, requiring only 15 h of hands-on time. Our workflow requires only a few coding steps – future users can simply load NGS reads and run the Prymetime script to detect and validate genetic engineering.MethodsStrains and mediaParent strains for all engineered strains are shown in Supplementary Table S1. All yeast strains were grown in yeast extract-peptone-dextrose (YPD) or synthetic complete (SC)+glucose media. YPD consisted of 30 g/L YEP (10 g/L yeast extract + 20 g/L peptone, Sunrise Science, 1877-1KG) and 20 g/L glucose (Alfa Aesar, A16828). SC + glucose media consisted of 6.71 g/L of YNB+Nitrogen (1.71 g/L yeast nitrogen base + 5 g/L ammonium sulfate, Sunrise Science, 1501-250), 20 g/L glucose, and a formulation of complete synthetic media (CSM). CSM formulations were (1) CSM-Leu: 0.65 g/L CSM-His-Leu-Ura (Sunrise Science, 1015-010) + 0.02 g/L Histidine (Sunrise Science, 1978-010) + 0.02 g/L Uracil (Sunrise Science, 1906-010) and (2) CSM-Trp-Ura: 0.62 g/L CSM-Leu-Trp-Ura (Sunrise Science, 1017-010) + 0.1 g/L Leucine (Sunrise Science, 1980-010). For the K. phaffii transformation, 2xYPD was prepared with 75 g/L YEP plus 20 g/L glucose, and YPDS plates were prepared by supplementing YPD agar with 1M sorbitol (Acros, 132730010). If appropriate, antibiotic selection was performed with nourseothricin at 0.1 g/L for S. cerevisiae and K. phaffii (Jena Bioscience, AB-101-10ML), geneticin at 0.2 g/L for S. cerevisiae and 0.3 g/L for K. phaffii (Life Technologies Gibco, 10131-035), and/or hygromycin B at 300 mg/L for S. cerevisiae (Thermo Fisher, 10687010). Routine growth conditions were as follows: inoculation in 5 mL media in a 14 mL Falcon tube (Corning, 352059), incubation at 30∘, and shaking at 220 rpm or agitation on a rotating drum.Chemically competent E. coli DH5α (NEB, C2987H) was used as a cloning strain and grown in 25 g/L LB Miller broth (10 g/L tryptone + 5 g/L yeast extract + 10 g/L sodium chloride, Fisher Scientific, BP1426-2). Antibiotic selection was performed with 100 mg/L ampicillin (Alfa Aesar, J63807), 25 mg/L chloramphenicol (Alfa Aesar, B20841), or 50 mg/L kanamycin (Alfa Aesar, J61272). Solid media was supplemented with 20 g/L agar (Sunrise Science, 1910-1KG).Polymerase chain reaction (PCR)PCR reactions were performed using the Q5 2X Master Mix (NEB, M0492L). Primers for PCR were designed with Benchling (https://benchling.com/, quality controlled with the New England Biolabs Tm Calculator (https://tmcalculator.neb.com/), and ordered from IDT (Integrated DNA Technologies, Inc., Skokie, Illinois). Reactions were performed in a total volume of 50 μL, with 25 μL of the Q5 Master Mix, 2.5 μL of both the forward and reverse primers (at 10 μM), X μL of template DNA (1 ng plasmid DNA, 100 ng genomic DNA), and 20-X μL of nuclease free water (VWR 02-0201-0500). PCR settings were determined based on instructions from NEB:98C for 30 sec30 PCR cycles: 98C for 10 secannealing temp. for 15 sec72C for 20 sec per kbp72C for 2 min10C holdModular cloningModular cloning with TypeIIS restriction enzymes was used to assemble genetic designs. Modular cloning uses a hierarchical assembly process to make parts, transcription units, and pathways. TypeIIS cloning reactions were based on the TypeIIS enzymes BbsI (10 U/μL, Thermo Scientific, ER1011) or BsaI (10 U/μL, NEB, R0535). DNA parts were diluted to 20 fmol/μL, with 1 μL of each part used in the reaction (2 parts for L0 assembly, 4 parts for L1 assembly). 7.9 - N parts (2 or 4) μL of nuclease free water was added to a PCR tube (USA Scientific, 1402-4700). Next, 1 μL of 10X Ligase Buffer and 0.4 μL of 20 U/μL T4 DNA ligase (Promega, M1794) was added to the tube. Finally, 1 1 μL of either the BbsI (L0 assembly) or BsaI (L1 assembly) enzymes was added to the reaction, yielding a total reaction of 10.3 μL. The reaction was then run on a thermocycler with the following conditions: 37 ∘C for 5 h, 50 ∘C for 15 min, 80 ∘C for 20 min, and a hold at 10 ∘C.Gibson cloningGibson assembly reactions followed instructions from the NEBuilder HiFi DNA Assembly Master Mix (NEB, E2621S). Briefly, PCR was used to amplify fragments with overlapping sequences (20–30 bp overlaps). When appropriate, the DpnI enzyme was used to digest template plasmid (NEB, R0176S) per instructions. The fragments were diluted to 0.2 pmols for 2–3 fragments or 0.5 pmols for 4 or more fragments in nuclease-free water, and transferred to a PCR tube. 10 μL of the HiFi master mix was added along with nuclease free water to reach a total reaction volume of 20 μL. The reaction was then run on a thermocycler at 50 ∘C for 60 min, followed by a hold at 10 ∘C.Yeast transformationsS. cerevisiae transformations were done based on the lithium acetate method108. S. cerevisiae cells from a glycerol stock were inoculated in 5 mL of YPD in a 14 mL Falcon tube and shaken overnight on a rotating drum at 30 ∘C. In the morning, these cells were used to inoculate 5 mL of fresh YPD to a density of OD = 0.25. The cells were incubated at 30 ∘C on a rotating drum until OD = 1.0 (approximately 4 h). The cells were then pelleted at 500 × g for 5 min, washed with 2.5 mL of sterile water, and centrifuged again at 500 × g for 5 min. The cells were resuspended in 100 μL of 100 mM lithium acetate (TCI, L0191) and transferred to a 1.5 mL microcentrifuge tube (USA Scientific, 1615-5500). The cells were pelleted at 500 × g for 30 s, resuspended to a total volume of 50 μL in about 40 μL of 100 mM lithium acetate, and then flicked to mix. The following were then added to the cell mixture: 240 μL PEG 3350 (VWR, 0955), 36 μL 1.0 M lithium acetate, 5 μL boiled salmon sperm DNA (Invitrogen AM9680, 10 μg/μL), and 50 μL transforming DNA. Between each addition to the cell mixture, the microcentrifuge tube was flicked to completely mix. The salmon sperm DNA was prepared by boiling for 5 min on a thermocycler at 100 ∘C. The tube was then incubated at 30 ∘C for 30 min, followed by the addition of 35 μL of dimethyl sulfoxide (DMSO, Sigma, D8418). The heat shock step followed at 42 ∘C for 15 min. For auxotrophic selection, the cells were plated onto CSM knockout agarose plates. For antibiotic selection, the cells were pelleted at 500 × g for 30 s followed by removal of the transformation mixture. 1 mL of YPD was used to gently resuspend the cell pellet and transferred to 4 mL of fresh YPD in a Falcon tube. The cells were allowed to incubate overnight at 30 ∘C, and then plated onto YPD agarose plates with the appropriate antibiotic. For both auxotrophic and antibiotic selections, the plates were incubated at 30 ∘C until transformants appeared (typically 2–4 days).Transformation of K. phaffi was performed by electroporation28. A 10 mL preculture in a 100 mL flask was inoculated from a glycerol stock and grown overnight at 30 ∘C with shaking at 200 rpm. The next morning, 50 μL of preculture was tranferred into 100 mL fresh YPD in a 250 mL flask, and this culture was incubated overnight again to OD600 = 1.3–1.5. This culture was harvested into three 50 mL conical tubes and pelleted at 4 ∘C, 1,500 × g for 5 min. The media was decanted and the pellet was resuspended by tapping firmly. The three pellets were resuspended in ice-cold sterile water and combined into one tube to a total of 40 mL. The cells were pelleted again, decanted, and resuspended in 20 mL ice-cold water. The cells were pelleted again and resuspended in 20 mL ice-cold 1 M sorbitol. The cells were pelleted again, the sorbitol was decanted, and the pellet was loosened by tapping firmly. 500 μL of ice-cold sorbitol was added to the pellet and mixed by flicking. These electrocompetent cells were stored on ice. Electroporation cuvettes (2 mm gap, Molecular BioProducts, 5520) were stored on ice during the centrifugation steps, and DNA was added to the bottom (5–10 μg for plasmid DNA, 5–20 μg linearized DNA for integration, or 10 μg circular transfer vector plus 10 μg recombinase expression vector for recombinase-based transformations). 80 μL of competent cells were added to the DNA-containing electroporation cuvettes and incubated on ice for 5 min. Cells were electroporated at 1500 V, then transferred into a round-bottom Falcon tube containing 1 mL 2xYPD at room temperature. Cells were recovered overnight at 30 ∘C with shaking at 200 rpm and 100–200 μL was plated onto YPD antibiotic plates. Plates were incubated at 30 ∘C until colonies appeared (2–4 days).Transformation of Y. lipolytica was performed by chemical transformation109. A 10 mL preculture in a 250 mL flask was inoculated from a glycerol stock and incubated at 30 ∘C with shaking at 200 rpm overnight. The next day, 25 mL of fresh YPD was inoculated from the preculture to OD6O00 = 0.5, and incubated for at 30 ∘C with shaking. After 3 h, 250 μL of 5 M hydroxyurea (Sigma H8627) was added to the culture, and incubation was continued for another 2 h. The cells were then transferred to a 50 mL conical tube (Greiner bio-one, 227261), centrifuged at 1500 × g for 5 min, and washed twice with 10 mL sterile deionized water. The pellet was resuspended to OD600 = 50 in 0.1M lithium acetate. For each transformant, 100 μL was transferred to a 1.5 mL microcentrifuge tube, which was centrifuged at 1500 × g for 5 min. The supernatant was removed and the following were added: 90 μL of 50% PEG-3350, 5 μL of 2 M ditriothreitol (G Bioscience, 277D-E), 5 μL of 2 M lithium acetate, and 2.5 μL of sheared, boiled salmon sperm DNA (10 μg per μL). This cocktail was mixed well with the cells by vortexing, then 5–10 μg of plasmid DNA in less than 40 μL was added and mixed by flicking. The cell and DNA mixture was then heat shocked at 39 ∘C for 1 h. The entire transformation mixture was plated onto SC media without leucine and incubated at 30 ∘C until colonies appeared (4 days).Parts and plasmidsAll genetic parts used in this study and their sources are detailed in Supplementary Table S10. Parts made for this study were synthesized by Integrated DNA Technologies (IDT). Design included codon optimization using IDT’s proprietary algorithm and elimination of BsaI and BbsI restriction sites.This study used cloning plasmids, integrating plasmids, and shuttle vectors. Plasmids built for this study are depicted in Supplementary Fig. S3. In modular cloning, plasmids that maintain transcriptional parts are referred to as Level 0 (L0) plasmids and plasmids maintaining transcription units are referred to as Level 1 (L1) plasmids. The L0 plasmids used in this study – pJHC07AB (Supplementary Fig. S3a), pJHC07BC (Supplementary Fig. S3b), pJHC07CD (Supplementary Fig. S3c) – are derived from pEMY07AB, pEMY07BC, and pEMY07CD51. These were constructed using Gibson assembly (NEB, E2611S) to replace the lacZ selection gene with the ccdb selection gene110. Plasmid pJHC07AB maintains promoters, pJHC07BC maintains ORFs (genes), and pJHC07CD maintains terminators. Integrating L1 plasmids built for this study included pJHC15HR1 (Supplementary Fig. S3d), pJHC15HR2 (Supplementary Fig. S3e), pJHC15HR3 (Supplementary Fig. S3f), pJHC15HR4 (Supplementary Fig. S3g), pJHC15HR5 (Supplementary Fig. S3h), and pJHC15HR6 (Supplementary Fig. S3i). These were constructed using Gibson assembly, and included two connector sequences, the ccdB selection gene, the chlR cassette, and ColE1 replicon. The connector sequences are sequentially homologous 60bp spacers, such that the 3′ spacer of pJHC15HR1 is homologous to the 5′ spacer of pJHC15HR2 and so forth. Once a transcription unit was assembled into these plasmids, PCR was used amplify the transcription unit fragment and the flanking connectors. These fragments were integrated into the S. cerevisiae genome using the native homologous recombination pathway, similar to DNA assembler29. We targeted two S. cerevisiae loci – ChrXV and HO (definitions and ref to supplement). The shuttle vector pCY112 built for this study is depicted in Supplementary Fig. S3j. It was constructed using Gibson assembly, and contains the ccdB selection gene, ColE1 replicon, chlR cassete, the low copy yeast replicon CEN6/ARSH4, and the Klleu2 auxotrophic cassette. Parts and plasmids specific to each strain are described in the next section.Yeast strain design and constructionFEY_1The parent strain was S. cerevisiae S288C hap1:HAP1111. The design was a metabolic pathway for synthesis of valine-derived chemicals (Supplementary Fig. S4a). The sequences for acetolactate synthase(ahas1), ketol-acid reductoisomerase (ilv6), and dihydroxy-acid dehydratase (ilvD1) were derived from Penicillium chrysogenum112. The sequence for aldehyde decarbonylase (ado) was derived from Prochlorococcus marinus113. The sequence for alpha-ketoisovalerate decarboxylase (kivD) was derived from Lactococcus lactis114. These CDSs were cloned into pJHC07BC using TypeIIS assembly. Transcription units were then built by combining L0 promoter, CDS, and terminator plasmids into a L1 integrating plasmid. The resulting transcription unit plasmids were pJHC15HR1-Ptef1-ahas1-Ttip1, pJHC15HR2-Psmtef1-ilv6-Tprm9, pJHC15HR3-Phta1-ilvD1-Tyhi9, pJHC15HR4-Pagtef1-nat-Tagtef1, pJHC15HR5-Psptdh3-ado-Trpl41b, and pJHC15HR6-Ptdh3-kivD-Trpl15a. Each level 1 vector was linearized by PCR and transformed into S.cerevisiae strain S228c, along with homology arms for the ChrXV integration locus (with the 5′ arm containing the spacer homologous to the pJHC15HR1 5′ spacer and 3′ homology arm containing the spacer homologous to the pJHC15HR6 3′ spacer). The primers used to amplify the homology arms from genomic DNA are included in Supplementary Table S10. The linearized fragments then assembled by yeast assembly. Transformants were selected on YPD with nourseothricin and verified by PCR.FEY_2The parent strain was S. cerevisiae CEN.PK113-7D115. The design was a metabolic pathway for the synthesis of β-carotene (Supplementary Fig. S4b). The sequences for geranylgeranyl diphosphate synthase (crtE), bifunctional lycopene cyclase/phytoene synthase (crtYB), and phytoene desaturase (crtI) were sourced from a previous study87. These coding sequences were synthesized, cloned into L0 pEMY07BC vectors with a BbsI type IIs reaction, assembled into level 1 transcription units (Psbtdh3-crtE-Trpl41b, Psptdh3-crtYB-Tyol036w, Phta2-hyg-Tagtef1, and Psmtef1-crtI-Trpl15a) with BsaI type IIs reactions, and integrated into the ChrXV locus as described above; however, the construct was integrated into strain CEN.PK113-7D and selected on YPD with hygromycin B. Plasmids pAG700 and pAG22-2 are dCas9 and gRNA expression plasmids, respectively, and were provided by Amar Ghodasara. Each plasmid was sequentially transformed into the above-described crt pathway integration strain with selection on YPD with hygromycin B, geneticin, and nourseothricin to yield FEY_2.FEY_5The parent strain was CEN.PK113-7D. The design was a fluorescent protein integrated into a genomic locus with an antibiotic selection marker (Supplementary Fig. S4c). The sequence for the fluorescent protein encoding gene yEmCitrine was sourced from a previous study92 and cloned into L0 vector pEMY07BC with a BbsI type IIs reaction. Level 1 transcription units Pact1-yEmCitrine-Tadh1 and Pagtef1-Nat-Tagtef1 were assembled into pJHC15HR1 and pJHC15HR2 with BsaI type IIs reactions and were integrated into the ChrXV locus as described above.FEY_15The parent strain was CEN.PK113-7D. The design was an inducible deactivated Cas9 expression cassette integrated into a chromosomal locus along with a high-copy replicating vector containing the guide RNA expression cassette (Supplementary Fig. S4d).The deactivated Cas9 (dCas9Mx1) sequence was sourced from a previous study35 and cloned into pEMY07BC. Transcription units Pgal10-dCas9Mx1-Tspo1 and Psptef1-nat-Ttip1 were assembled in pJHC15HR1 and pJHC15HR2 with BsaI type IIs reactions and integrated into the HO locus (HO locus homology arm primers included in Supplementary Table S10). Yeast shuttle vector pY128 containing the insert Psnr52-gRNAscr-TtracrSUP4 was subsequently transformed into the integration strain to yield FEY_15. The pY128 vector was sourced from a previous study51.FEY_18The parent strain was S. cerevisiae strain W30311. The design was a Cas9 expression cassette on a low copy number plasmid and a guide RNA expression cassette on a high copy number plasmid (Supplementary Fig. S4e). Plasmids p414-Tef1p-Cas9-Cyc1t and p426-Snr52p-gRNA.CAN1.Y-Sup4t30 were purchased from Addgene (43802 and 43803).FEY_27The parent strain was S288C hap1:HAP1. The design was a Cpf1 expression strain on a low copy number plasmid (Supplementary Fig. S4f). Plasmid pCSN06738 was purchased from Addgene (101748).FEY_29The parent strain was S288C hap1:HAP1. The design was a Cpf1 programming crRNA expression strain on a high copy number plasmid (Supplementary Fig. S4g). Plasmid pUDE72237 was purchased from Addgene (103022).FEY_30The parent strain was S288C hap1:HAP1. The design was a Cre expression strain on a low copy number plasmid (Supplementary Fig. S4h). Plasmid pSH6633 was purchased from Euroscarf (P30672).FEY_37The parent strain was S288C hap1:HAP1. The design was a single enzyme expression strain on a high copy number plasmid (Supplementary Fig. S4i). The sequence for prespatane sesquiterpene synthase (pst) was derived from Laurencia pacifica88. This coding sequence was synthesized, cloned into level 0 vector pEMY07BC with a BbsI type IIS reaction, and cloned with Ppgk1 and Ttdh1 by BsaI type IIS reactions into the level 1 shuttle vector pY128. Shuttle vector pY128 contains the endogenous yeast 2μ plasmid origin of replication.FEY_43The parent strain was S288C hap1:HAP1. The design was a three enzyme pathway integrated into a chromosomal locus with an antibiotic selection marker (Supplementary Fig. S4j). The sequence for bifunctional diterpene synthase (mdst) was derived from Selaginella moellendorffii116. The sequence for bifunctional ferruginol, 11-hydroxyferruginol synthase (hfst) was derived from Salvia pomifera117. The sequence for 11-hydroxyferruginol C20-oxidase (cast) was derived from Salvia rosmarinus89. These coding sequences were synthesized and cloned into level 0 vector pEMY07BC with BbsI. Transcription units Psktef1-mdst-Tecm10, Psptef1-nat-Ttip1, Psmtdh3-hfst-Ttdh3, and Psmtef1-cast-Teno1 were assembled into pJHC15HR1-4 with BsaI, linearized by PCR, and integrated into the HO locus as described above.FEY_45The parent strain was S288C hap1:HAP1. The design was a two enzyme pathway integrated into a chromosomal locus with an antibiotic selection marker (Supplementary Fig. S4k). The sequence for dimethylallylcistransferase (nppst) was derived from Solanum lycopersicum118. The sequence for limonene synthase was derived from Citrus limon119. These coding sequences were synthesized and cloned into level 0 vector pEMY07BC with BbsI. Transcription units Psbtef1-nppst-Tecm10, Psptef1-nat-Ttip1, and Psktdh3-lst-Ttdh1 were assembled into pJHC15HR1-3, linearized by PCR, and integrated into the HO locus as described above.FEY_48The parent strain was S. cerevisiae BY474213. The design was a monocistronic dual fluorescent protein construct integrated into a chromosomal location as well as an empty yeast shuttle vector with a low copy number origin and auxotrophic complementation marker (Supplementary Fig. S4l). The yEGFP-2A-mRuby sequence was designed by combining the yEGFP and mRuby sequences from Sheff et al.92 and Lee et al.93, respectively, with a self-cleaving 2A sequence94. This coding sequence was cloned into level 0 vector pEMY07BC with BbsI. Transcription units Psptef1-yEGFP-2A-mRuby-Trps9a and Psptef1-nat-Ttip1 were assembled into pJHC15HR1-2 with BsaI, linearized by PCR, and integrated into the HO locus of BY4742 as described above. The lacZα insert of pEMY11251 was substituted with a ccdB insert to form pCY112, which was then transformed into the EGFP-2A-mRuby integrated strain above. The plasmid pCY112 contains the CEN6/ARSH4 yeast plasmid origin of replication.FEY_55The parent strain was S. cerevisiae BY474113. The design was a low copy number yeast shuttle vector expressing a fluorescent protein (Supplementary Fig. S4m). Plasmid pKK1112(Prev1-Venus-Teno2//LEU2//CEN6//KanR-ColE1) was generated using parts from the MoClo Yeast Toolkit27. The following level 0 parts were combined with BsaI into an eight-part level 1 shuttle vector: pYTK084, pYTK002, pYTK027, pYTK033, pYTK055, pYTK067, pYTK075, and pYTK081.FEY_73The S. cerevisiae strain BY4743120 was used without modification.FEY_74The parent strain was Y.lipolytica strain Po1f16. The design was a plasmid expressing Cas9 (Supplementary Fig. S4n). Plasmid pCRISPRyl39 was purchased from Addgene (103022).FEY_75The parent strain was K.phaffii strain CBS 7435 (ATCC 76273). The design was a fluorescent protein expression cassette integrated into a genomic locus by site-specific recombination (Supplementary Fig. S4o). The strain containing an attP site for BxbI-mediated recombination was created by transforming plasmid PP74 linearized by AccI (NEB, R0161S) into the parent strain as described by Perez-Pinera28. An integrating vector pKK2147(Pgap-aMFnoEAEA-RFPsec-Taox1) was constructed by combining pYTK084, pYTK002, pYTK067, and pYTK07827 with pPTK002, pPTK006, pPTK018, pPTK019, and pPTK020 in a 9-part BsaI type IIs reaction as described by Obst121. This integration vector was co-transformed with BxBI expression plasmid PP43 into the attP-containing strain and selected on YPD plates with nourseothricin and G418 to yield FEY_75.High-molecular weight genomic DNA isolationGenomic DNA was isolated using Promega’s Genomic DNA Isolation Kit (Promega, A1120). A modified version of Promega’s protocol for yeast gDNA isolation was used to limit shearing of DNA, with added insight from Josh Quick’s Ultra-long read sequencing protocol122. No vortexing and limited pipetting/mixing steps were used to maximize Nanopore read lengths. 5 mL of cells were grown overnight (or until saturation) at 30. The cells were pelleted at 500 × g for 5 min, and resuspended in 1.5 mL of 50 mM EDTA (Millipore, 324506) and 37.5 μL of 5 U/μL zymolyase (Zymo, E1004). The samples were incubated at 37 °C for 1 h to allow the Zymolyase to digest the cell wall. The cells were pelleted at 500 × g for 5 min, re-suspended in 1.5 mL of the Nuclei Lysis Solution (mix by inversion, flicking), and incubated at room temperature for 30 min. 7.5 μL of RNAse A Solution was then added and incubated for 15 min at 37 °C. Once cooled to room temperature, 500 μL of the protein precipitation solution was added (invert to mix). The samples were put on ice for 5 min, and subsequently centrifuged for 10 min at 3000 × g. 700 μL of the supernatant was added to a fresh microcentrifuge tube with 700 μL of isopropanol (Sigma, I9516). The microcentrifuge tubes were gently mixed by inversion and centrifuged at 4000 × g for 1 min. The DNA pellet was washed with 70% ethanol (Sigma, E7023) and centrifuged at 4000 × g for 1 min. The ethanol was carefully pipetted off the DNA pellet, and the tube cap was left open at room temperature for 20 min to allow residual ethanol to evaporate. 50 μL of 10 mM Tris-HCl (Alfa Aesar, J67233) and 0.02% Triton X-100 (Sigma-Aldrich, X100-500ML) was added to resuspend the DNA pellet and incubated overnight at 4 °C. DNA quality was evaluated using a Nanodrop, and the concentration was calculated using a Qubit.Nanopore DNA library preparation and MinION loadingThe Rapid Barcoding Kit was used to tagment the DNA libraries for sequencing (ONT, SQK-RBK004). Up to four genomes were multiplexed on each MinION flow cell. Library preparation closely followed the protocol provided by ONT. Briefly, 400 ng of template DNA for each isolate was diluted to 7.5 μL, mixed with 2.5 μL of the Fragmentation Mix, and then incubated at 30 ∘C for 1 min and 80 ∘C for 1 min on a thermal cycler. The barcoded samples were then pooled together and concentrated using AMPure XP beads in 10 μL of 10 mM Tris-HCl, 50 mM NaCl. The pooled sample was next mixed with 1 μL of RAP for 5 min at room temperature, and stored on ice until ready to load. R9.4 MinION Flow Cells (ONT, FLO-MIN106) were used for all sequencing runs. The flow cells were first primed per ONT’s instructions. The 11 μL of prepped DNA was mixed with 4.5 μL of nuclease-free water, 34 uL of SQB, and 25.5 μL of LLB and loaded onto the MinION flow cell. Sequencing runs were executed using ONT’s MinKNOW software (v8.3.1) with the default settings.Read processingNanopore fast5 files were basecalled using Guppy v2.3.5 (Oxford Nanopore base caller). The subsequent fastq files were demultiplexed using the EPI2ME interface (Metrichor, Oxford, UK). Illumina reads were demultiplexed using the native software on the iSeq machine. Random subsets of Illumina and Nanopore reads at a specific genome coverage were generated using a custom python script (https://github.com/aseetharam/common_scripts/blob/master/sample_fastq.py).For the metagenome experiment, nanopore and illumina reads from FEY_15 and the zymo mock metagenome were combined into the same nanopore and Illumina fastq file. FEY_15 Nanopore reads at 10X genome coverage were used for each read dilution experiment. The number of zymo mock metagenome nanopore reads was based off the estimated number of base pairs in the FEY_15 read library. The genome size of S. cerevisiae is 12.1 Mb, so 10X genome coverage is 121 Mb. Therefore, the nanopore read library sizes of the zymo mock metagenome were 121 Mb (1:1), 1210 Mb (1:10), 12100 Mb (1:100), and 121000 Mb (1:1000). All of the Illumina reads from FEY_15 and the zymo mock metagenome were simply combined into one file and used for each of the four dilution experiments.Illumina DNA library preparation and iSeq 100 loadingThe Nextera DNA Flex Library Prep Kit (Illumina, 20018704) along with the Nextera DNA CD Indexes (Illumina, 20018707) were used to tagment the DNA libraries for sequencing. Library preparation closely followed the instructions provided by Illumina, and up to four genomes were multiplexed on one Illumina sequencing cartridge. Briefly, tagmentation was first performed with 500 ng of genomic DNA in 30 μL of nuclease-free water. The reaction was stopped by adding 10 μL of both the BLT and TB1 reagents and incubating at 55∘C for 15 min on a thermal cycler. Index adapters for each sample along with EPM were then added to barcode and amplify the genomic DNA. The DNA libraries were amplified using the following PCR program:68C for 3 min98C for 3 min5 PCR cycles: 98C for 45 sec62C for 30 sec68C for 2 min68C for 1 min10C holdThe DNA libraries were cleaned using subsequent steps with the SPM reagent and 80% ethanol, and concentrated in 32 μL of the RSB reagent. Assuming four genomes were multiplexed on one flow cell, 25 pM of each DNA library were pooled together in 100 μL of the RSB reagent and stored on ice until ready to load. The pooled libraries were loaded onto the sequencing cartridges according to Illumina’s instructions. The Local Run Manager on the iSeq 100 machine was used to initiate sequencing runs. A GENERATEFASTQ run was started, and run with the parameters Read Type: Paired End, Read Lengths: 151, and Index Reads: 2.Nanopore de novo genome assemblyFor the MiniASM68 assembly, reads were first mapped using minimap2 (v2.17-r941)61 with the parameters "-x ava-ont -t8”. MiniASM (v0.3) was then subsequently run with the default parameters. Canu69 (v1.8) was run with the parameters "minReadLength=2500 mhapSensitivity=high corMhapSenstivity=high corOutCoverage=500”. SMARTdenovo70 (v1.0) was run using the parameters "-c 1 -k 14 -J 2500 -e zmo”. Flye71 (v2.4) was run with the parameters "–meta –plasmids". ABySS64 (v 2.1.5) was run with the abyss-pe option and the parameter "k=96". Edena63 (v3.131028) was run with the default parameters. Velvet65 (v1.2.10) was run with a hashlength of 21 bp. MaSuRCA (v3.3.4) was run with the parameter "JF_SIZE = 242000000 FLYE_ASSEMBLY=1". SPAdes (v3.13.1) was run with the parameters "–sc –nanopore –pe < # > -1 –pe < # > -2".Nanopore and Illumina read polishingThe de novo Nanopore genome assemblies were first polished with Nanopore reads using Medaka (v0.4) with the default parameters. The assembly was then polished with Illumina reads, first with Racon (v1.3.1) followed by Pilon (v1.22). For Racon, the Illumina reads were first mapped to an assembly using minimap2 with the parameter "-ax sr”. Racon78 was then run using the default parameters. For Pilon, assemblies were first indexed using bwa (v0.7.17-r1188)123. Illumina reads were then mapped to the assembly using bwa with the parameter "mem -t 14”. Pilon79 was then run using the parameter "-Xmx160G”.Prymetime genome assembly workflowVisualization of the full Prymetime workflow is shown in Supplementary Fig. S20. First, Flye (v2.4) was run with the parameters "–meta –plasmids" on a Nanopore fastq file to generate the initial genome assembly. Contigs in the resulting assembly file were separated into circular or linear contig files. This was accomplished using a custom python script and the assembly_info.txt file resulting from Flye. The linear contigs were polished first with Medaka (Nanopore reads), followed by Racon and Pilon (Illumina reads). Medaka was run with the default parameters. In preparation for Racon polishing, the Illumina reads were mapped using minimap2 with the parameter "-ax sr". Racon was then run with the default parameters. For Pilon preparation, the assembly was first indexed with bwa, followed by mapping with bwa and the parameter "mem -t 14". Pilon was then run with the parameter "-Xmx160G". To find potential circular contigs that Flye may have missed, a custom python script was used on the polished linear contigs file. The script used the Mummer option nucmer (v3.1)124 with the parameters "maxmatch = True, simplify = False, mincluster = 2000, min_id = 99, min_length = 2000, coords_header = True" on contigs that were less than 50,000 bp to identify repetitive contigs. The repetitive contigs were extracted and combined with the circular contigs from the initial Flye assembly, and sent to be re-assembled with Unicycler (v0.4.8). In order to do this, the contigs were first separated into separate fasta files using awk (v4.0.2). Nanopore and Illumina reads were then mapped to each individual contig, with matches extracted into fastq files. The Nanopore reads were mapped using minimap2 with the parameters "-ax map-ont" followed by extraction of hits with samtools (v1.9) and the parameters "fastq -n -F 4 -". Each paired-end Illumina file was mapped using minimap2 with the parameter "-ax sr" followed by extraction of hits with samtools and the parameters "fastq -n -F 4 -". The resulting two Illumina files were paired using fastq_pair (v1.0) with the default parameters125. Unicycler was then run with the mapped and paired Illumina files along with the mapped Nanopore file with the default parameters82. The re-assembled circular and repetitive contigs resulting from Unicycler were combined with the polished linear contigs, yielding the final assembly.Prymetime annotation and visualization of engineering and genome featuresNon-native engineering signatures were detected in the genome assemblies using BLASTN (v2.5.0) with the parameters "-perc_identity 98 -qcov_hsp_perc 98". The query for this BLASTN search was a curated list of all non-native engineering signatures used to engineer the yeast strains in this study, and are included in the Prymetime package. The genome features telomeres, centromeres, and mitochrondrion were detected using BLASTN with the parameters "-max_target_seqs 1 -max_hsps 1". The genome feature sequences were downloaded from the Saccharomyces Genome Database100, and are included in the Prymetime package. The genome plotter chromoMap (v0.2)101 was run with the parameters "data_based_color_map = T, data_type = "categorical" to show the engineering signatures and genome elements hits from the BLASTN search in the context of the entire genome assembly. The genome alignment and visualization software AliTV (v1.0.6) was run with the default parameters102.Genome assessment toolsQUAST126 (v5.0.0) was run with the default parameters, yielding the metrics number of contigs, maximum contig length, and N50. For accuracy-related metrics, the nucmer command was run as part of the MUMmer package124. The command "dnadiff -d" was used on the resulting delta file to find the average identity to the reference and the number of SNPs. Genome assemblies were evaluated for genome completeness using BUSCO (v4.0.6)95 with the saccharomycetales_odb9 datasets, as well as a BLASTN40 search of ORFs from S. cerevisiae S288C. Engineered signatures were searched for in-genome assemblies using BLASTN with the expect threshold set at 0.0001.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationReporting Summary | nature communications | [
"Article"
] | [
"Next-generation sequencing",
"Genome assembly algorithms",
"Metabolic engineering",
"Synthetic biology"
] | detection genetic engineering needed strain engineering intellectual property monitor release events detect engineered organisms unknown samples Whole genome sequencing) attractive sequence features captures all sequences including intended unintended modifications precise resolution genetic engineering requirements WGS workflow signatures identified within accurate contiguous sequences WGS workflow needed for engineered organisms on engineered yeasts Yeasts crucial for genome-scale accurate WGS necessary validating synthesized eukaryotic genomes cell factories for derived from baker’s yeast Saccharomyces nonconventional yeasts Yarrowia Komagataella phaffii economic importance increasing use engineered yeast cell factories crucial WGS methods validate intended engineering confirm unintended variation Without WGS majority yeast strains validated with methods PCR targeted sequencing capture unintended secondary mutations engineered unpublished accounts WGS revealing unexpected sequences genome structures in engineered industrial strains evidence challenges assumption phenotype direct result intended engineering possible explanation variation between replicates irreproducible findings WGS detect validate genetic engineering in yeastsYeast engineering leaves predictable sequence features in genome like standard plasmid sets replication integrations29–32 gene knockouts33 genome edits using RNA-guided endonucleases34–39 identified in genome sequence with BLAST40 engineered yeast obstacles to complete accurate genome assembly high sequence identity constructs cause identical sequences complex genome features like multiple deletions13 plasmids varying copy insertions36 SCRaMbLEd chromosomes43 scale of yeast engineering increasing fraction of genome rewritten45 numbers of engineered strains through adaptive laboratory combinatorial pathway iterative approaches result in strains costly to sequence obstacles to complexities like repetitive regions rearrangements polyploidy make accurate complete contiguous yeast genomes difficult without significant allocation resources WGS workflow five steps DNA isolation library preparation sequencing assembly annotation DNA purified phenol-choloroform bead beating enzymatic lysis55 sequencing library prepared by attaching adapters barcodes via ligation tagmentationlibrary sequenced with next-generation (NGS platform generates short reads (150–300 high accuracy56 or long reads (1.5 to lower accuracy57 average read length number coverage on sequencing technology DNA isolation adapter attachment steps58 reads computationally assembled into final genome sequence with software overlap-layout-consensus (OLC or De Bruijn graph (DBG OLC DBG assemblers classified into short read hybrid long read with error correction hybrid long read with error correction assembly approaches accurate genome sequence structure at low read depths two technologies costs time annotation performed Eukaryotic annotation genes functional annotation60 features like genetic engineering parts telomeres centromeres mitochondrial DNA natural plasmids not annotated not included in final assembly inexpensive WGS workflow detect genetic engineering in mixed samples engineered yeast optimized steps resolve engineering sequences in heavily engineered yeast strain improved DNA isolation sequencing library preparation increase representation used long- short-read sequencing from high coverage at low costintegrated assemblers resolve circular plasmids linear chromosomes developed annotation approach user-input list genetic parts identify signatures engineering annotated centromeres telomeres origins replication mitochondrial DNA genetic engineering workflow named Prymetime Recombinant Yeast Markers validate genetic engineering genome sequences detect engineering metagenomic samples useful strain validation release monitoring protecting intellectual property investigating engineering unknown samples nanopore sequencing library preparation engineered standard genome assembly workflow resolve chromosomal integrations multiple plasmids yeast engineering built S. cerevisiae CEN.PK113 strain FEY_2 integrated carotenoid pathway native 2μ plasmid dCas9 plasmid gRNA plasmid Fig. 1a prepared sequencing libraries FEY_2 Oxford Nanopore Technologies) ligation kit libraries low read length differential DNA shearing developed gentle genomic DNA isolation protocol increased nanopore read length reduced variance sequencing results few reads plasmids comparing reads plasmid antibiotic selection markers ACT1 genomic locusplasmids FEY_2 yeast miniprep kit sequencing plasmids ligation turned tagmentation library method ONT Rapid kit improvement plasmid reads Fig. 1b 2:1 20:1 marker ACT1 read coverage ratios equivalent plasmid copy number yeast tagmentation increased representation circular elements native 2μ plasmid mitochondrial DNA tagmentation key long read lengths linear molecules small circular DNA nanopore tagmentation sequence bias extreme GC genomes not yeasts gentle tagmentation nanopore sequencing FEY_2 adequate representation plasmid reads. 1Detection engineering signatures S. cerevisiae FEY Photograph FEY_2 functional carotenoid pathway engineering signatures carotenoid pathway chromosomal integration low copy plasmid expressing dCas9 high copy plasmid expressing gRNA Approximate copy number genomic DNA libraries Oxford Nanopore Technologies’ Ligation Rapid kit Illumina’s Nextera kit low high copy plasmids FEY_2. BLASTN engineering signatures assemblies genome assemblers categorized short-read hybrid long read with error correctionalgorithm De Bruijn graph or overlap-consensus shown Failure modes shown red lines (contig break white spaces (missing colored pathways plasmids represent engineering signatures contiguous sequences.Developing assembly workflow for complete contiguous plasmids evaluated nine assembly algorithms chromosomes plasmids complete accurate contiguous three plasmids FEY_2 tested included short-read OLC Edena63 DBG ABySS64 Velvet65 hybrid OLC Masurca66 DBG HybridSPAdes67 long-read OLC MiniASM68 Canu69 SMARTdenovo70 DBG Flye71 long-read assemblers higher error provide "skeleton for mapping additional reads72–76 assemblies polished with Medaka77 Racon78 Pilon79 used optimized library preparation long reads 60X ONT MinION short reads 125X Illumina iSeq 100 common set reads assembler resulting genome assembly analyzed using BLASTN for integrated pathway plasmids native 2μ plasmid BLASTN results Fig. 1a. engineering features rarely complete contiguousshort-read assemblers ABySS Edena Velvet returned fragmented pathway plasmids hybrid assemblers SPAdes Masurca produced more complete sequences genome integration fragmented omitted three plasmids long-read assemblers MiniASM Canu Flye SMARTdenovo returned single contiguous sequence omitted sections three plasmids Flye returned genome each plasmid correctly contiguous assemblies missing portions plasmids generated with OLC assembler algorithms use All-versus-All consensus step discard identical sequences consensus used BLASTN at OLC Canu pipeline determine sequences complete low-copy plasmid present before consensus step lost in final assembly Canu discarded plasmid during consensus step high sequence identity DBG assemblers Flye ABySS SPAdes omit sections plasmids DBG algorithms split reads into shorter k-mers Eulerian walk approach less prone discarding identical sequences80 plasmid sequences from ABySS SPAdes fragmented Flye assembled each plasmid into single contiguous sequence SPAdes hybrid assemblers short reads long readssubject to pitfalls short-read assemblers span sequences high identity Flye DBG assembler assembles long reads first produces higher contiguity better resolution sequences high identity findings genome assembly quality on high long-read data de novo assembly approach plasmid contigs Flye complete contiguous longer than expected inspection repeats expected sequence common problem for long-read linear representative plasmid contigs with Unicycler sent contigs Unicycler improved accuracy length for three plasmids FEY_2.Resolving engineering signatures in engineered validated assembly approach on engineered laboratory nonconventional yeast constructed 15 strains from S. cerevisiae S288C CEN.PK113-7D W303-α BY4741 BY4742 K. phaffii ATCC 76273 (CBS Y. lipolytica ATCC MYA-2613 (Po1f)85 Plasmids transcriptional units described in Supplementary Fig. S3 description each strain in Fig. 2a details Supplementary Table S1 Fig. S4 Engineering signatures inserted into genome maintained on episomal plasmidsintegrations HO NRT1 GYP7 chromosome XV38 plasmids TypeIIS-compatible yeast shuttle vectors S. cerevisiae replicon (2μ CEN6/ARSH4) Engineering categorized biosynthetic pathways gene editing deletions biology elements pathways 2β carnosic Genome editing tools SpCas934 dCas935 LbCpf138 FnCpf137 Deletions synthetic auxotrophies S cerevisiae W303-α BY4741 BY4742 Y. lipolytica Po1f Synthetic biology elements fluorescent 2A sequence94 engineered Y. lipolytica strain "FEY_74 CRISPR-Cas9 plasmid K. phaffii strain "FEY_75 recombinase red fluorescent protein) signatures engineered yeast strains engineering signatures Prymetime genome assemblies strains Failure modes red lines white spaces colored pathways plasmids signatures contiguous sequences CRISPR-Cas9 expression vector FEY_74 Y. lipolytica strain plasmidDNA agarose gel confirms missing Cas9 cassette FEY_74 strain original pCRISPRyl plasmid gel represents experiment PCR products FEY_74 processed parallel Illustration location RFP integration cassette chromosome II FEY_75 K. phaffii actual location chromosome IV sequenced collection with ONT MinION Illumina iSeq 100 systems library preparation protocols combined assembly approach Flye Unicycler captured engineering signature S. cerevisiae genetic background BLASTN 2a resolved seven genome integrations two loci eleven plasmids BLASTN metrics Supplementary Table S2. repeated assembly Flye additional Unicycler step improves plasmid length accuracy every strain No sequence complexities parts identical plasmid copy number affected accuracy structural completeness genome assemblies engineered nonconventional yeasts Y. lipolytica strain FEY_74 K. phaffii strain FEY_75 revealed unintentional edits FEY_74 intended contain pCRISPR-yl plasmid39 missing Cas9 transcription unit portion E. origin replication Fig. 2bInspection raw reads identify missing sequence performed genomic DNA isolation yeast plasmid miniprep on FEY_74 transformed DNA into E. coli colonies disrupted origin replication unintended loss assembly error confirmed by PCR DNA FEY_74 missing region PCR sequence missing (Fig. 2b). FEY_75 RFP transcription unit integrated chromosome II (Fig. pathway found BLASTN FEY_75 genome analysis integrated into chromosome IV confirmed PCR integration site chromosome II negative strain remained nourseothricin resistant RFP positive results indicate combined assembly approach find reproduce engineering useful strain quality control engineering unknown samples genome assembly assessed genome quality 15 engineered assemblies genomes parent nonconventional yeasts Y. lipolyitica Po1f K. phaffii CBS7435 high contiguity sequence accuracy genome completeness measured Single-Copy Orthologs (BUSCO) score95 Saccharomycetales dataset percent aligned reads parent genome Percent unmapped reads Supplementary Table S3 Whole genome alignments parent Mauve96 Supplementary Figs.S6 S7 resequenced Y. lipolytica Po1f K. phaffii CBS7435 strains improved compared genomes16,84 metrics (Supplementary Fig. S8) 6 more essential genes recovered in K. phaffii 13 genes Y. lipolytica.Fig. 3Whole genome assembly quality engineered yeast strains genome completeness score yeast genome assemblies parent Percentage aligned bases each chromosome reference parent strain assemblies determine 16 chromosomes BY4741 assembly discontiguity FEY_55 assembly compared to BY4742 assembly number long terminal repeats (LTRs) predicted by LTRpred S. cerevisiae genome assemblies approximate copy number CUP1 repeats raw Nanopore reads Illumina reads. strains number copy number rDNA repeats Nanopore reads Illumina reads copy number tabulated in Supplementary Table S4 final test of completeness chromosome resolved from telomere to other telomere compared each S. cerevisiae assembly to reference assembly number complete telomere-to-telomere contigs (Supplementary Fig.found 76% chromosomes complete except telomeres Analysis smaller contigs telomeric DNA sequences genomes complete save misassembly repetitive sequences telomeres assessed repetitive DNA elements S. cerevisiae genome long terminal repeats CUP1 repeats rDNA resolved comparable copy number reference genomes (Fig. 3c d e CUP1 rDNA numbers underrepresented reference S. cerevisiaeCUP1 copy number variable zero to 7997 rDNA 100 and 20098 Tandem repeats CUP1 rDNA common problem often collapsed during strain investigated haploid sequenced heterozygous diploid strain S. cerevisiae BY4743 resulting assembly similar S. cerevisiae S288C (Fig. resolve ploidy LYS2 MET15 heterozygous deletions resolved mapping average read count (Fig. 4b c. 4Genome assembly analysis diploid S. cerevisiae strain BY4743 genome alignment S. S288C assembly BY4743 colored blocks represent regions red lines indicate new contig Nanopore read coverage around heterozygous LYS2 geneNanopore coverage heterozygous MET15 gene results indicate genome assemblies combined approach structurally correct accurate complete telomeres repeat elements ploidy challenge reproduce challenge de novo genome assembly.Annotating visualizing engineering genome WGS annotation genetic engineering sequences developed engineering annotation step applied 15 engineered yeasts wrote automated BLASTN script find standard yeast genetic engineering parts genome features parts include CEN6/ARSH4 2μ replication origins selection markers promoters terminators Genome features include centromeres telomeres mitochondrial DNA sourced from Saccharomyces Genome Database100 FASTA file modified updated fed BLASTN results to interactive genome viewers chromoMap101 AliTV102 ChromoMap highlights parts features AliTV assembly to parent strain lastz103 unintended changes chromosomal rearrangements chromoMap visualization FEY_2 shows integration scaffold_3 engineered plasmids scaffold_18_23 output interactive hovering blocks parts identified plasmids differentiated origins replication engineering sequences AliTV visualization high sequence identity contiguity engineered unengineered S. cerevisiae CEN.PK apparentAliTV visualization interactive customizable useful contigs reference assembly.Fig. engineering genome features Prymetime pipeline chromoMap visualization engineering signatures structural elements FEY_2 genome assembly AliTV visualization FEY_2 genome assembly parent CEN.PK113-7D genome assembly Engineering signatures structural elements annotated Overview Prymetime genome assembly pipeline automated pipelineOptimization steps genome assembly methods software genetic engineering yeast genomes integrated software tool Prymetime "Pipeline for Recombinant Yeast Identifies Markers final pipeline Fig. 5c software accepts long short reads list sequences interest reference genome outputs two interactive visualizations genome 15 engineered strains Figs. S10–S17 tested each step Prymetime raw reads S. cerevisiae CEN.PK113-7D74,104 quality each step S18) evaluated Flye step 40X long-read genome coverage match reference assembly evaluated polishing step 40X short-read genome coverage needed high identity reference BUSCO percentage S. cerevisiae S288C CDSs Tables S5–S8) chromoMap visualization Prymetime CEN.PK113-7D assembly captures centromeric not telomeric Fig. S19) corroborates observations engineered genomes different solve problem Flye remains best software for accurate contiguous genetic engineering sequences Prymetime workflow Fig. S20 results confirm Prymetime 40X genome coverage needed highest quality genomes.Resolving signatures engineering in metagenome engineering signatures in metagenome reads Zymo mock metagenome combined with FEY_15 strain simulate detection engineered strain mixed sample mock metagenome consists eight bacteria species Bacillus subtilis Enterococcus faecalis Escherichia coli Lactobacillus fermentum Listeria Pseudomonas aeruginosa Salmonella enterica Staphylococcus aureus two yeast species Cryptococcus neoformans Saccharomyces cerevisiae105 FEY_15 reads diluted with Zymo metagenome reads 1:1 1:10 1:100 1:1000 FEY_15 Zymo metagenome reads combined 1:20 (Fig. 6a). read sets used for Prymetime assembly integration plasmid of FEY_15 resolvedsynthetic biology parts resolved mixed samples by Prymetime.Fig. 6Resolving signatures engineering silico metagenome assembly reads Zymo’s mock metagenome combined with engineered S. cerevisiae strain FEY_15 Visual representation BLASTN results silico metagenome Failure modes red lines break white spaces (missing colored pathways plasmids represent engineering signatures contiguous sequences work develops integrated workflow for WGS engineered yeasts extensible eukaryotes linear circular sequences workflow gDNA isolation tagmentation long- short-read NGS de novo assembly annotation genetic engineering parts genome features diverse engineering signatures resolved in contiguous sequences plasmids genome quality comparable to high-quality reference assemblies accurate assemblies before after engineering permits verification genetic engineering WGS workflow metagenomic data detection yeast engineering in mixed samples demonstrates challenges WGS-based workflows Flye assembly algorithm supported accurate resolution genetic engineering contiguous sequences sequence omission OLC algorithms resolution short-read hybrid assemblers produced fragmented sequencesFlye long-read DBG assembler best resolving genetic engineering signatures observations highlight difficulty applying genome assembly software to engineered yeasts identical genetic engineering signatures features assemblers collapsed repetitive genome features struggled resolve telomeres limits tool detect variations rDNA SNPs rare variants against performance Flye WGS rarely used in strain engineering cycles cost time bioinformatics expertise WGS workflow with ONT MinION Illumina iSeq 100 platforms Prymetime software overcomes barriers Prymetime high genomes low read depth 40X sufficient for high accuracy completeness contiguity genetic engineering quality genomes 40X read depth 30 S. cerevisiae genomes on MinION cell 4 Illumina iSeq flow cell 0.5 Gb needed for 40X read depth 13.4 Mb S. cerevisiae genome collapsed rDNA typical yield 15 Gb from MinION 2.4 Gb from iSeq 100 multiplexing $200 per genome workflow fast under week 15 h hands-on time workflow requires few coding steps users load NGS reads run Prymetime script detect validate genetic engineeringMethodsStrains strains engineered strains Supplementary Table S1 yeast strains grown yeast extract-peptone-dextrose (YPD synthetic (SC+glucose media YPD 30 g/L YEP (10 yeast extract 20 g/L peptone 20 g/L glucose SC + glucose 6.71 g/L YNB+Nitrogen (1.71 g/L yeast nitrogen base 5 g/L ammonium sulfate 20 g/L glucose complete synthetic media CSM formulations CSM-Leu 0.65 g/L + 0.02 g/L Histidine Uracil CSM-Trp-Ura 0.62 g/L + 0.1 g/L Leucine K. phaffii transformation 2xYPD prepared 75 g/L YEP 20 g/L glucose plates YPD agar 1M sorbitol antibiotic selection nourseothricin 0.1 g/L S. cerevisiae K. phaffii geneticin 0.2 g/L 0.3 g/L hygromycin B 300 mg/L S. cerevisiaegrowth conditions inoculation 5 mL media 14 mL Falcon tube incubation 30∘ shaking 220 rpm agitation rotating drum competent E. coli DH5α C2987H) cloning strain grown 25 g/L LB Miller broth (10 tryptone 5 yeast extract sodium chloride Fisher Scientific Antibiotic selection 100 mg/L ampicillin Aesar 25 mg/L chloramphenicol 50 mg/L kanamycin Solid media supplemented 20 g/L agar (Sunrise Science 1910-1KG).Polymerase chain reaction (PCR Q5 2X Master Mix Primers designed Benchling quality controlled New England Biolabs Tm Calculator ordered IDT Technologies Skokie Reactions volume 50 μL 25 μL Q5 Master Mix 2.5 μL forward reverse primers X μL template DNA 100 genomic 20-X μL nuclease free water PCR settings NEB:98C 30 sec30 PCR cycles 98C 10 15 sec72C 20 sec 2 TypeIIS restriction enzymes genetic designs hierarchical assembly processTypeIIS cloning reactions enzymes BbsI or BsaI DNA parts diluted to 20 fmol/μL 1 μL each (2 L0 4 L1 (2 or 4) μL nuclease free water added PCR tube (USA Scientific 1 μL 10X Ligase Buffer 0.4 μL 20 U/μL T4 DNA ligase added μL BbsI or BsaI enzymes added total 10.3 μL run thermocycler 37 ∘C 5 h 50 ∘C 15 min 80 ∘C 20 min hold 10 ∘C.Gibson assembly reactions NEBuilder HiFi DNA Assembly Master Mix PCR fragments overlapping sequences (20–30 DpnI enzyme template plasmid fragments diluted to 0.2 pmols 2–3 or 0.5 pmols 4 more fragments nuclease-free water transferred to PCR tube 10 μL HiFi master mix added nuclease free water total reaction volume 20 μL run thermocycler 50 ∘C 60 min hold 10 ∘C transformationsS. cerevisiae lithium acetatecerevisiae cells glycerol stock inoculated 5 mL YPD 14 mL Falcon tube shaken overnight drum 30 ∘C morning 5 mL fresh YPD OD = 0.25 incubated 30 ∘C until OD = 1.0 4 pelleted 500 × g 5 min washed 2.5 mL sterile water centrifuged 500 × g 5 min resuspended in 100 μL mM lithium acetate transferred 1.5 mL microcentrifuge tube pelleted 500 × 30 s resuspended 50 μL 40 μL 100 mM lithium acetate mix added 240 μL PEG 3350 36 μL 1.0 M lithium acetate 5 μL boiled salmon sperm DNA 50 μL transforming DNA microcentrifuge mix salmon sperm DNA 5 min 100 ∘C incubated 30 ∘C 30 min 35 μL dimethyl sulfoxide heat shock 42 ∘C 15 min auxotrophic selection cells plated CSM knockout agarose plates antibiotic selection pelleted at 500 × g 30 s mixture 1 mL YPD transferred to 4 mL fresh YPD Falcon tubecells overnight 30 ∘C plated YPD agarose plates antibiotic auxotrophic antibiotic selections plates incubated 30 ∘C until transformants appeared 2–4 days).Transformation K. phaffi 10 mL preculture 100 mL flask inoculated from glycerol stock grown overnight 30 ∘C shaking 200 rpm morning 50 μL tranferred into 100 mL fresh YPD 250 mL flask incubated overnight to OD600 = 1.3–1.5 harvested into three 50 mL tubes pelleted at 4 ∘C 1,500 × g 5 min media decanted pellet resuspended resuspended ice-cold water combined into one tube 40 mL cells pelleted decanted resuspended 20 mL ice-cold water 20 mL ice-cold 1 M sorbitol pellet loosened 500 μL ice-cold sorbitol added pellet mixed electrocompetent cells stored on ice Electroporation cuvettes stored ice DNA added (5–10 μg plasmid linearized 80 μL competent cells added cuvettes incubated on ice 5 min.Cells electroporated 1500 V transferred Falcon tube 1 mL 2xYPD room temperature recovered overnight 30 ∘C shaking 200 rpm 100–200 μL plated YPD antibiotic plates incubated 30 colonies (2–4 Y. lipolytica chemical 10 mL preculture 250 mL flask inoculated glycerol incubated 30 ∘C 200 overnight 25 mL YPD inoculated OD6O00 = 0.5 incubated 30 3 h 250 μL 5 M hydroxyurea added incubation 2 h cells transferred 50 mL tube centrifuged 1500 g 5 min washed 10 mL water resuspended OD600 = 50 0.1M lithium acetate 100 μL transferred 1.5 mL microcentrifuge tube centrifuged 1500 5 min added 90 μL 50% PEG-3350 5 μL 2 M ditriothreitol 5 2 M lithium acetate 2.5 μL salmon sperm DNA (10 μg mixed cells 5–10 μg plasmid DNA 40 μL heat shocked 39 ∘C 1 htransformation mixture plated SC media without leucine incubated at 30 ∘C until colonies appeared (4 days).Parts genetic parts sources detailed in Supplementary Table S10 synthesized by Integrated DNA Technologies Design included codon optimization elimination of BsaI BbsI restriction sites study used cloning plasmids integrating plasmids shuttle vectors Plasmids in Supplementary Fig. S3 modular cloning plasmids transcriptional parts Level 0 (L0) plasmids maintaining transcription units Level 1 (L1) plasmids L0 plasmids pJHC07AB derived from pEMY07AB pEMY07BC pEMY07CD51 constructed using Gibson assembly lacZ selection gene with selection pJHC07AB maintains promoters pJHC07BC ORFs pJHC07CD maintains terminators Integrating L1 plasmids included pJHC15HR1 constructed using Gibson assembly included two connector sequences ccdB selection gene chlR cassette ColE1 repliconconnector sequences homologous 60bp spacers 3′ pJHC15HR1 5′ pJHC15HR2 transcription unit plasmids PCR fragment connectors fragments integrated S. cerevisiae genome homologous recombination pathway DNA targeted S. cerevisiae loci ChrXV HO shuttle vector pCY112 Supplementary Fig. S3j constructed Gibson assembly contains ccdB selection gene ColE1 replicon chlR cassete low copy yeast replicon CEN6/ARSH4 Klleu2 auxotrophic cassette Parts plasmids strain strain design parent strain S. cerevisiae S288C hap1:HAP1111 metabolic pathway valine-derived chemicals sequences acetolactate synthase ketol-acid reductoisomerase dihydroxy dehydratase Penicillium aldehyde decarbonylase Prochlorococcus alpha-ketoisovalerate decarboxylase Lactococcus CDSs cloned pJHC07BC TypeIIS assembly Transcription units L0 promoter CDS terminator plasmids L1 plasmidtranscription plasmids pJHC15HR1-Ptef1-ahas1-Ttip1-Tprm9-Tagtef1-Trpl41b-Ptdh3-kivD-Trpl15a level 1 vector linearized PCR transformed into S.cerevisiae strain S228c homology arms ChrXV integration locus 5′ arm pJHC15HR1 3′ primers arms DNA Supplementary Table S10 linearized fragments assembled by yeast assembly Transformants selected on YPD nourseothricin verified PCR parent strain S. cerevisiae CEN.PK113-7D115 design metabolic pathway synthesis β-carotene Fig. S4b). sequences geranylgeranyl diphosphate synthase bifunctional lycopene cyclase/phytoene synthase desaturase) sourced from previouscoding sequences synthesized cloned pEMY07BC vectors BbsI assembled transcription units-Trpl41b-Tyol036w Phta2-hyg-Tagtef1-Trpl15a integrated ChrXV locus strain CEN.PK113-7D selected YPD hygromycin B Plasmids pAG700 pAG22-2 dCas9 gRNA expression plasmids provided Amar Ghodasara transformed crt pathway integration strain hygromycin B geneticin nourseothricin FEY parent strain CEN.PK113-7D fluorescent protein genomic locus antibiotic selection marker yEmCitrine sourced cloned L0 vector pEMY07BC BbsI IIs reaction Level 1 transcription units Pact1-yEmCitrine-Tadh1 Pagtef1-Nat-Tagtef1 pJHC15HR1 pJHC15HR2 BsaI IIs reactions integrated ChrXV locus parent strain CEN.PK113-7D inducible deactivated Cas9 expression cassette chromosomal locus high-copy replicating vector RNA cassette deactivated Cas9 sequence sourced previous cloned pEMY07BCTranscription units Pgal10-dCas9Mx1-Tspo1 Psptef1-nat-Ttip1 assembled pJHC15HR1 BsaI IIs reactions integrated HO locus Supplementary Table S10) Yeast shuttle vector pY128 Psnr52-gRNAscr-TtracrSUP4 transformed integration strain FEY_15 pY128 vector sourced previous study51 parent strain S. cerevisiae strain W30311 Cas9 expression cassette low copy plasmid guide RNA cassette high copy plasmid Plasmids p414-Tef1p-Cas9-Cyc1t p426-Snr52p-gRNA.CAN1.Y-Sup4t30 purchased Addgene (43802 parent strain S288C hap1:HAP1 Cpf1 low plasmid Plasmid pCSN06738 Addgene S288C hap1:HAP1 Cpf1 crRNA high copy plasmid Plasmid pUDE72237 Addgene S288C hap1:HAP1 Cre low copy plasmid Plasmid pSH6633 Euroscarf S288C hap1:HAP1 single enzyme expression high copy plasmidsequence prespatane sesquiterpene synthase Laurencia synthesized cloned level 0 vector pEMY07BC BbsI IIS cloned Ppgk1 Ttdh1 level 1 shuttle vector pY128 endogenous yeast 2μ plasmid replication parent strain S288C hap1:HAP1 three enzyme pathway chromosomal locus antibiotic selection marker bifunctional diterpene synthase Selaginella moellendorffii116 ferruginol 11-hydroxyferruginol synthase Salvia 11-hydroxyferruginol C20-oxidase Salvia rosmarinus89 synthesized cloned level 0 vector pEMY07BC BbsI units Psktef1-mdst-Tecm10 Psptef1-nat-Ttip1-Ttdh3-Teno1 pJHC15HR1-4 BsaI linearized PCR integrated HO locus parent strain S288C hap1:HAP1 two enzyme pathway chromosomal locus antibiotic selection marker dimethylallylcistransferase Solanum lycopersicum118 limonene synthase Citrus limon119coding sequences cloned vector pEMY07BC units Psptef1-nat-Ttip1 Psktdh3-lst-Ttdh1 assembled pJHC15HR1-3 linearized PCR integrated HO locus parent strain S. cerevisiae BY474213 monocistronic dual fluorescent protein chromosomal location empty yeast shuttle vector low copy number origin auxotrophic complementation marker yEGFP-2A-mRuby sequence Sheff self-cleaving 2A cloned level 0 vector pEMY07BC BbsI units Psptef1-yEGFP-2A-mRuby Psptef1-nat-Ttip1 pJHC15HR1-2 linearized PCR integrated HO locus BY4742 lacZα insert pEMY11251 substituted ccdB insert pCY112 transformed EGFP-2A-mRuby strain plasmid pCY112 CEN6/ARSH4 yeast plasmid origin replication parent strain S. cerevisiae BY474113 low copy number yeast shuttle vector fluorescent protein Plasmid pKK1112 generated MoClo Yeast Toolkit27level 0 parts combined with BsaI into eight-part level 1 shuttle vector pYTK084 S. cerevisiae strain BY4743120 used without modification parent strain Y.lipolytica strain Po1f16 plasmid expressing Cas9 Plasmid pCRISPRyl39 purchased from Addgene parent strain K.phaffii strain CBS 7435 fluorescent protein expression cassette genomic locus site recombination strain attP site BxbI-mediated recombination transforming plasmid PP74 parent strain integrating vector pKK2147(Pgap-aMFnoEAEA-RFPsec-Taox1) combining pYTK084 pYTK07827 with pPTK002 9-part BsaI type IIs reaction co-transformed with BxBI plasmid PP43 into attP-containing strain selected YPD plates with nourseothricin G418 FEY_75-molecular weight genomic DNA isolated Promega’s Genomic DNA Isolation Kit modified protocol gDNA limit shearing DNA Josh Quick’s Ultra-long read sequencingvortexing limited pipetting Nanopore lengths 5 mL cells grown overnight 30 pelleted 500 × g 5 min resuspended 1.5 mL 50 mM EDTA 37.5 μL 5 U/μL zymolyase incubated 37 °C 1 h Zymolyase cell wall pelleted 500 × g 5 min re-suspended 1.5 mL Nuclei Lysis Solution incubated room temperature 30 min 7.5 μL A Solution incubated 15 min 37 °C cooled room temperature 500 μL protein precipitation solution added samples 5 min centrifuged 10 min 3000 × g 700 μL supernatant microcentrifuge tube 700 μL isopropanol mixed centrifuged 4000 × g 1 min DNA pellet washed 70% ethanol centrifuged 4000 g 1 min ethanol pipetted tube cap left open room temperature 20 min ethanol 50 μL 10 mM Tris-HCl 0.02% Triton X-100 added incubated overnight 4 °C DNA quality evaluated Nanodrop concentration calculated QubitNanopore DNA library preparation MinION Rapid Barcoding Kit DNA libraries sequencing four genomes multiplexed MinION flow cell preparation protocol ONT 400 ng template DNA diluted 7.5 μL mixed 2.5 μL Fragmentation Mix incubated 30 ∘C 1 min 80 ∘C 1 min thermal cycler barcoded samples pooled concentrated AMPure XP beads 10 μL mM Tris-HCl 50 mM NaCl pooled sample mixed 1 μL RAP 5 min stored load R9.4 MinION Flow Cells sequencing runs primed 11 μL prepped DNA mixed 4.5 μL nuclease-free water 34 uL SQB 25.5 μL LLB loaded MinION flow cell Sequencing runs MinKNOW software (v8.3.1) processingNanopore files basecalled Guppy v2.3.5 files demultiplexed EPI2ME interface Illumina reads demultiplexed native software Random subsets Illumina Nanopore genome coverage generated custom python script metagenome experiment nanopore reads combined nanopore fastq fileFEY_15 Nanopore reads 10X genome coverage each dilution experiment zymo reads estimated base pairs FEY_15 library genome size S. cerevisiae 12.1 Mb 10X coverage 121 Mb sizes zymo metagenome 121 Mb (1:1) 1210 Mb 12100 (1 121000 (1:1000). Illumina reads FEY_15 zymo mock metagenome combined file four dilution experiments DNA library preparation iSeq 100 Nextera DNA Flex Library Prep Kit Nextera DNA CD Indexes DNA libraries sequencing four genomes multiplexed one sequencing cartridge tagmentation 500 ng genomic DNA 30 μL nuclease-free water 10 μL BLT TB1 reagents incubating 55∘C 15 min thermal Index adapters sample EPM added barcode amplify DNA libraries amplified PCR program:68C 3 min98C 3 PCR cycles 98C 45 sec62C 30 sec68C 2 1 libraries cleaned SPM reagent 80% ethanol concentrated 32 μL RSB reagent four genomes multiplexed cell 25 pM each DNA pooled 100 μL RSB reagent stored until ready loadpooled libraries loaded sequencing cartridges instructions Local Run Manager iSeq 100 sequencing GENERATEFASTQ run started Read Type Paired End Lengths 151 Index Reads 2.Nanopore genome MiniASM68 reads mapped minimap2-x ava-ont -t8”. MiniASM (v0.3) default parameters Canu69 (v1.8) "minReadLength=2500 SMARTdenovo70 (v1.0)-c 1 -k 14 -J 2500 -e Flye71 (v2.4) "–meta –plasmids". ABySS64 2.1.5 abyss-pe option "k=96". Edena63 default parameters Velvet65 (v1.2.10) hashlength 21 bp MaSuRCA (v3.3.4) "JF_SIZE 242000000 FLYE_ASSEMBLY=1". SPAdes (v3.13.1) "–sc –nanopore –pe -2".Nanopore Illumina Nanopore genome assemblies polished Nanopore Medaka (v0.4) default parameters polished Illumina reads Racon (v1.3.1) Pilon (v1.22) reads mapped minimap2-ax Racon78 default parameters Pilon assemblies indexed bwa (v0.7Illumina reads mapped assembly bwa "mem -t Pilon79 run "-Xmx160G”.Prymetime genome assembly Supplementary Fig. S20 Flye (v2.4) run parameters "–meta –plasmids Nanopore fastq file initial genome assembly separated circular linear files custom python script assembly_info.txt file Flye linear contigs polished Medaka (Nanopore Racon Pilon (Illumina Medaka run default parameters Racon Illumina reads mapped minimap2 "-ax sr". Racon run default parameters Pilon assembly indexed bwa bwa "mem -t 14". Pilon run "-Xmx160G". circular contigs custom python script contigs Mummer option nucmer (v3.1)124 parameters "maxmatch True simplify False mincluster 2000, min_id = 99 min_length 2000, coords_header True contigs less than bp repetitive contigs repetitive contigs extracted combined circular contigs Flye re Unicycler (v0.4.8) contigs separated fasta files awk (v4.0.2) Nanopore Illumina reads mapped each contig matches extracted fastq filesNanopore reads mapped minimap2-ax map-ont extraction hits samtools (v1.9 "fastq -n -F 4 paired-end Illumina file mapped minimap2-ax sr extraction samtools -n -F 4 two Illumina files paired fastq_pair (v1.0) default parameters125 Unicycler run mapped paired Illumina files Nanopore default parameters82 re-assembled circular repetitive contigs combined polished linear contigs final assembly.Prymetime annotation visualization engineering genome featuresNon-native engineering signatures detected BLASTN (v2.5.0) parameters-perc_identity 98 -qcov_hsp_perc non-native engineering signatures Prymetime genome features telomeres centromeres mitochrondrion detected BLASTN-max_target_seqs 1 -max_hsps 1". genome feature sequences downloaded Saccharomyces Genome included Prymetime package genome plotter chromoMap (v0.2 parameters_based_color_map T data_type "categorical engineering signatures genome genome visualization software AliTV (v1.0.6) default parameters102.Genome assessment toolsQUAST126 (v5.0.run default parameters metrics contigs maximum length N50 accuracy-related metrics nucmer command run MUMmer command "dnadiff -d" used delta file average identity reference number SNPs Genome assemblies evaluated completeness (v4.0.6)95 saccharomycetales_odb9 datasets BLASTN40 search ORFs S. cerevisiae S288C Engineered signatures searched in-genome assemblies BLASTN expect threshold 0.0001.Reporting Nature Research Reporting Summary.Supplementary | 49.7 | 1.274571 |
10.1038/s41467-020-19206-w | PMC7585417 | Atomic structure of alkali metal is difficult to be revealed at room temperature because of the chemical reactivity and irradiation sensitivity. Here the authors show that electron beam-induced in situ growth of alkali metals enables the investigation of atomic structure and growth kinetics at high spatiotemporal resolution. | Alkali metals are widely studied in various fields such as medicine and battery. However, limited by the chemical reactivity and electron/ion beam sensitivity, the intrinsic atomic structure of alkali metals and its fundamental properties are difficult to be revealed. Here, a simple and versatile method is proposed to form the alkali metals in situ inside the transmission electron microscope. Taking alkali salts as the starting materials and electron beam as the trigger, alkali metals can be obtained directly. With this method, atomic resolution imaging of lithium and sodium metal is achieved at room temperature, and the growth of alkali metals is visualized at atomic-scale with millisecond temporal resolution. Furthermore, our observations unravel the ambiguities in lithium metal growth on garnet-type solid electrolytes for lithium-metal batteries. Finally, our method enables a direct study of physical contact property of lithium metal as well as its surface passivation oxide layer, which may contribute to better understanding of lithium dendrite and solid electrolyte interphase issues in lithium ion batteries. | IntroductionIt has been a long history for the research on alkali metals1, the most excitement it has generated in recent years is for energy conversion and storage2–6. As the “holy grail”, lithium metal is considered to be the ultimate choice of anode in battery as it has the highest theoretical capacity and lowest electrochemical potential2,3. However, for alkali metals, less is known about their structural information on the atomic-level. An important reason is that they are so reactive and can never exist in air in elemental form. So far, vacuum or cryogenic transfer seems to be the only reliable way to take alkali metal samples into the microscopes for observation. However, contaminations and degradations cannot be entirely ruled out during the transfer process. Moreover, the electron beam sensitive nature of alkali metals disables high-resolution transmission electron microscopy (HRTEM) imaging. Until recently, atomic-scale imaging of cryo-transferred lithium has been reported at low temperature7,8, and room temperature imaging is achieved by intercalating lithium into graphene sheets9, whereas a direct room temperature atomic resolution imaging of bare alkali metals has never been demonstrated. More importantly, direct in situ observation of the growth of alkali metals at high spatiotemporal resolution has not been achieved yet. The missing link in the microstructure of lithium metals hinders further understanding and development of lithium-ion batteries.In the present work, we propose a simple and versatile method to form alkali metals inside the transmission electron microscope (TEM) directly. Taking the observing electron beam as the trigger, alkali metals can be produced from their corresponding alkali salts. Here, two types of alkali salts, i.e., alkali carbonates and alkali fluorides, are chosen as demonstrations. Both alkali carbonates and fluorides are alkali compounds used in a wide field of industrial, technical and medical applications1,10–14. Taking Li2CO3 as an example, it is the first chemical in the lithium production chain and also the key component in lithium-air batteries11,12. In addition, it is an essential medicine for the treatment of manic depression and bipolar disorder13. Moreover, both Li2CO3 and LiF are the starting materials for preparing of various ceramics, glass and cements1 in broad applications. Therefore, fundamental research on the stability and decomposition of lithium carbonates and fluorides can also be interesting in general.For the electron beam-induced decomposition reaction observed here, we found pure alkali metals as the products. Compared to the conventional multistep reactions to produce solid alkali metals15,16, the observed one-step reaction here, on the other hand, provides an elegant way to prepare and study the intrinsic structure and structural kinetics of alkali metals. During observation, the electron beam damage is minimized and atomic resolution is obtained by using low dose aberration-corrected HRTEM (AC-HRTEM)17. Atomic resolution imaging of lithium and sodium is achieved at room temperature. In situ fast camera enables millisecond temporal resolution so that the process of the alkali metal growth can be tracked. This work provides a paradigm to investigate such chemically reactive and electron beam sensitive materials. The immediate advantage is that an ambiguity about lithium metal growth on garnet-type solid electrolytes for lithium-metal batteries can be clarified, and physical contact property of lithium metals and surface passivation oxides can be revealed.ResultsIn situ formation of alkali metals from alkali saltsFigure 1a shows the schematic of our configuration. The reaction can be triggered by the observing electron beam while the reaction rate is controlled by the intensity of the irradiation. The key point is to control the dose-rate of the electron beam. Typically, 10~1000 e Å−2 s−1 is a good choice. By focusing the high energy electron beam onto the edge of alkali carbonate/fluoride particles, alkali metal particles grow out from the illuminated spot and scale to hundreds of nanometers. The reaction rate is carefully controlled by tuning the beam intensity so that the reaction process is slow enough to be captured by the detector, and atomic resolution imaging of both lithium and sodium is demonstrated at room temperature. Details are provided in the following. To begin, pure alkali carbonates/fluorides were drop-casted onto the TEM grids. The basic structural and compositional characterizations of the as-drop-casted alkali salts are shown in Supplementary Figs. 1–3, confirming pure-phase alkali salts acting as the starting materials. Next, three examples of the alkali metal formation and growth process are demonstrated. Figure 1b–d show the selected sequential snapshots of in situ growth process of lithium from lithium carbonate (Supplementary Movie 1, ~100 e Å−2 s−1), sodium from sodium carbonate (Supplementary Movie 2, ~10 e Å−2 s−1) and lithium from lithium fluoride (Supplementary Movie 3, ~20 e Å−2 s−1), respectively. During the growth process, sharp edges and corners can be observed which indicate the crystalline feature of these particles. In general, the particles would expand along several growth directions (denoted by white arrows). It is concluded that the electron beam-induced formation of alkali metals from alkali salts could be a common phenomenon.Fig. 1In situ formation and growth of alkali metal particles.a Schematic of the experimental process. An alkali (Li, Na) metal particle formed from their corresponding alkali containing materials under beam irradiation. The inset shows the b.c.c. structure of alkali metals. Atomic resolution images of both lithium and sodium metals are shown at room temperature condition. b In situ growth of Li particles from lithium carbonate. c In situ growth of Na particles from sodium carbonate. d In situ growth of Li particles from lithium fluoride.In order to confirm the composition and structure of the particles, selected area electron diffraction (SAED) was performed on lithium and sodium particles, respectively. Figure 2a shows two neighbor particles grown from lithium carbonates and the black circle represents the selected area for diffraction. Figure 2b, c shows the SAED pattern and its rotational average spectrum, respectively. Figure 2d, g shows the crystal structures of body-center cubic (b.c.c.) lithium and face-center cubic (f.c.c.) lithium oxide, respectively. Their corresponding simulated polycrystalline diffraction patterns and rotational average spectra are shown in Fig. 2e, f, h, i, respectively. By comparison, it can be concluded that the particles are comprised of both lithium and lithium oxide. To be noted that the polycrystalline-featured rings are indexed to be lithium oxide while the isolated sharp single-crystalline diffraction spots are indexed to be lithium. Similar analysis for the sodium case is shown in Fig. 2j–r. Hence, these particles are composed of single-crystalline alkali metals and their polycrystalline oxide compounds.Fig. 2SAED analysis of alkali metal particles.a The selected lithium particles (the black circle represents the aperture area), b its SAED pattern and c rotational average spectrum. d Crystal structure of b.c.c. lithium, e its simulated polycrystalline diffraction rings and f rotational average spectrum. g Crystal structure of f.c.c. lithium oxide, h its simulated polycrystalline diffraction rings and i rotational average spectrum. j The selected sodium particles, k its SAED pattern and l rotational average spectrum. m Crystal structure of b.c.c. sodium, n its simulated polycrystalline diffraction rings and o rotational average spectrum. p Crystal structure of f.c.c. sodium oxide, q its simulated polycrystalline diffraction rings and r rotational average spectrum.Growth kinetics of lithium metalTo understand the formation sequence of alkali metals and their oxides, the growth of particles was traced at high spatiotemporal resolution18. Here, combining low dose AC-HRTEM imaging with in situ fast camera detection, an example of lithium-particle growth is demonstrated at atomic resolution with fast frame rate (40 ms per frame). It is confirmed that pure-phase lithium metal was formed and grew up at the beginning. Figure 3a shows an AC-HRTEM image of a lithium particle along [111] zone axis. The b.c.c. structure can be verified from the Fourier transformation pattern. The surface of two exposed <110> facets can be seen clearly. An in situ movie (Supplementary Movie 4 and snapshots shown in Fig. 3) indicates that the lithium particle grew along these two directions one after another. Figure 3a–c depicts the growth stage along \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[10\bar 1]$$\end{document}[101¯] direction, as indicated by red arrows. Along this direction, (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10\bar 1$$\end{document}101¯) facet moved forward while (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1\bar 10$$\end{document}11¯0) facet remained unchanged. Evolution of the (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10\bar 1$$\end{document}101¯) edge is highlighted (red single solid lines to dotted lines, and then to double solid lines), the layer-by-layer growth of lithium atoms can be observed. This stage lasted for about 600 ms (15 frames) and later on grew slowly for three seconds, then the particle changed growth direction to [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1\bar 10$$\end{document}11¯0], as shown in Fig. 3d–f. Along this direction, (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1\bar 10$$\end{document}11¯0) facet moved forward as indicated by red arrows. Compared with the (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10\bar 1$$\end{document}101¯) edge, growth along the [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1\bar 10$$\end{document}11¯0] direction is rather quick. Hence, atoms at the corner could not be replenished in a short time, and an incomplete (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\bar 1\bar 1$$\end{document}21¯1¯) facet was exposed as a transient state. As time went on, lithium atoms were replenished gradually, and the corner surface changed to even high index facets and the length of this corner shorten from 14 nm to 6 nm, resulting in a sharp corner eventually (also see Supplementary Movie 4). This stage lasted for about 400 ms. For alkali metals with b.c.c. structure, <110> face is the most densely packed face and has lowest surface energy, therefore the surface ended up with low index facets again. Millisecond temporal resolution enables the observation of fast structural evolution, confirming that the surface energy is one of the driving forces for the growth of alkali metals.Fig. 3Growth kinetics of lithium metal.a–c Growth along [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10\bar 1$$\end{document}101¯] direction. a A [111]-oriented lithium particle grown from lithium fluoride. Inset is the corresponding Fourier transformation pattern. Red single solid lines represent the edge of this particle. Image filtering was applied to enhance the signal to noise ratio. Dose-rate is ~1000 e Å−2 s−1. b Edge of the particle at this stage are tracked by red dotted lines. The middle area between solid lines and dotted lines represents the additive grown portion. c Edge of the particle at this stage are tracked by red double solid lines. The middle area between solid lines and dotted lines represents the additive grown portion. d–f Growth along [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1\bar 10$$\end{document}11¯0] direction. d Red single solid lines represent the edge of this particle. (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\bar 1\bar 1$$\end{document}21¯1¯) surface is exposed and the corresponding Fourier transformation spot is marked by red circle. e Edge of the particle at this stage are tracked by red dotted lines. f Edge of the particle at this stage are tracked by red double solid lines. g Growth length versus time. Arrows represent starts of new growing processes. h Growth rate versus time. Arrows mark the growing steps as in (g), respectively. i, j Images of two types of lithium whisker growth, Type I in (i) and Type II in (j), respectively. k, l Growth length (k) and growth rate (l) versus time.Figure 3g, h shows the growth length and growth rate (∆length/∆t) of the particle versus time, respectively. Particle length at t = 0 is regarded as the zero point. Arrows mark each start of a new growth period. As can be seen in Fig. 3g, in each growth period, the particle grows quickly at first and this rate slows down gradually, and therefore growth steps form. It is noted that the growing process is not continuous, and it may be possible that the delivery of new lithium source from alkali salts takes time for the start of the next growing process. After the change of growth direction, the growth along new direction is rather fast. As shown in Fig. 3h, growth rate at this moment (the blue arrow) is almost one order of magnitude quicker than the previous fastest growth rate (the leftmost red arrow). This indicates a sudden influx of a large number of lithium atoms or possible accumulation of a large number of diffused lithium atoms before the change of the growth direction. In terms of the different growth rate between [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10\bar 1$$\end{document}101¯] and [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1\bar 10$$\end{document}11¯0] direction, it is related with the anisotropic mass transfer along these two directions19. The rate of mass transfer could be determined by the decomposition of the matrix of the lithium salts. Anisotropic decomposition of the lithium salts may provide anisotropic diffusion flux of lithium source along different directions.Apart from the growth of lithium particles, we have also observed the growth of lithium dendritic whiskers. Investigation of the whisker growth here can be compared with the electrical current-driven growth of lithium dendrites in lithium-ion batteries20, which may provide better understanding of the dendrite growth. Figure 3i, j shows the growth of lithium whiskers from an as-synthesized lithium particle (defined as Type I, Supplementary Movie 5) and from the root lithium fluoride materials (defined as Type II, Supplementary Movie 6), respectively. Their corresponding growth length and smoothed growth rate versus time are shown in Fig. 3k, l, respectively. In these two examples, the whiskers grew straightly without deflection. It is found that the growth rate of Type I positively correlated with time (the correlation coefficient is 0.769 for the raw data and 0.993 for smoothed data), and the growth kinetics can be described as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L = \frac{1}{2}at^2$$\end{document}L=12at2Here L represents the length of the whisker, t represents the time, a represents the acceleration. Hence, this is an accelerated growth mode and a = 2.56 nm s−2 (0.6–5.6 s) for the case in Fig.3i. In contrast to Type I, the growth rate of Type II falls in the range of 60~100 nm s−1 and the length increases linearly with time (the correlation coefficient is 0.998). Therefore, it is a uniform growth mode and the kinetics follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L = vt$$\end{document}L=vtwhere v represents the speed. For the case shown in Fig. 3j, v = 85 nm s−1 (0–5.8 s).Uniform growth of Type II lithium whiskers suggests the decomposition reaction of lithium salts occurs uniformly, and the formation of lithium metal may be reaction-rate limited21. For the case of Type I, lithium whisker started to grow from an initially formed lithium particle, and the accelerated growth indicating a different mechanism with additional driving force. Interestingly, comparing with the situation of lithium whisker growth in the liquid electrolyte of lithium-ion batteries22, we found similar growth kinetics for whiskers grown from lithium particles, although the driving force (electron beam irradiation versus electrical current) and growth environment (vacuum versus liquid) are different. In addition, for the Type I growth demonstrated here, lithium source could not be replenished from surrounding environment as the case in liquid electrolyte, and the lithium particles should be the only lithium supplier. However, no obvious collapse or contrast change of the lithium particle can be observed as shown in Fig. 3i, indicating a rapid replenishment of lithium from other surrounding particles20. To summarize, 1D-growth could occur at the pinhole of the surface passivation layers of as-formed lithium particles (Type I whisker) or at some unique nucleation sites of the lithium salts where the diffusion is confined (Type II whisker). The different growth kinetics of two types of whisker could be related with different diffusion barriers in lithium metal and lithium salts23.In general, we propose that the growth process of both particles and whiskers could be a competition between radiation damage and diffusion flux. As the particle/whisker grows, radiation damage becomes more severe and the diffusion flux is depleted so that the growth rate slows down gradually. It seems that different environments (CO220, N220, vacuum, and liquid22) have little effect on the growth kinetics of lithium metal, whereas the morphology is related to the surface passivation layers, which is sensitive to the surrounding environment. The 1D- or 3D-growth mode is on one hand related to the confinement from surface passivation layers, and on the other hand related to the diffusion flux which is dependent of the strength of the driving force (electron beam irradiation or electrical current).Besides, the cross sections of these whiskers could be observed as deflection occurred occasionally as the whiskers became longer. One of the cases is shown in Supplementary Fig. 4 for a Type I whisker. The hexagonal cross section is again in consistent with the electrochemically deposited lithium whiskers7. Consequently, intrinsic features of lithium whiskers grown in vacuum shares many similarities with the situation in batteries, and therefore our method may provide an alternative way to study lithium metal-related structure problems as high spatiotemporal resolution is a major advantage over the other methods.During the whole in situ growing process, sharp crystalline surface facets, as well as single-crystalline feature of the particle, maintained throughout, and alkali metal oxides did not show up at an early stage. However, no matter for the decomposition of carbonates or fluorides, alkali metal oxides were found in all the final SAED analysis (Fig. 2 and Supplementary Fig. 5). This indicates that for the electron beam-induced decomposition of alkali carbonates/fluorides, pure alkali metals were formed originally. Oxidation would occur in a subsequent step, which will be detailed in the following.Oxidation of alkali metalsAfter the nucleation and fast growth process, alkali metal particles reached certain area/volume (Fig. 1) and the growth slowed down. Static imaging and spectroscopy were applied at this stage to verify the spatial distribution of the alkali metals and oxide compounds. Figure 4a shows an AC-HRTEM image (dose-rate ~1050 e Å−2 s−1) of a lithium particle along [110] direction. Single-crystalline feature of the particle can be seen at a glance. Figure 4b shows a close-up where the atom columns can be observed clearly. Its further magnified image in Fig. 4c agrees well with the simulated AC-HRTEM image (inset and also Supplementary Fig. 6) of lithium. Furthermore, Fig. 4d shows the elemental lithium mapping of a typical particle using energy-filtered TEM (EFTEM). Atomic resolution imaging together with compositional mapping shows that the fresh particle grows to be a single-crystalline lithium initially. Within the scope of Fig. 4a, only a little portion of the lattice fringes is presented in the form of Moiré fringes, as indicated by the blue square. The Moiré fringes arise from the overlap of Li2O with lithium (detailed analysis shown in Supplementary Fig. 7). We found that in most cases, Li2O distributed on the outer surface of the particles. Figure 4e shows several particles which have been placed in TEM vacuum chamber for tens of minutes. Its inset shows a partial close-up of the surface and the lattice of Li2O is observed. Fourier component analysis is utilized to isolate the different orientation of Li2O crystal grains, and the result is shown in Fig. 4f–h. Li2O with three different orientations were found and indicated by different colors, with their corresponding selected Fourier transformation spots shown inset. Figure 4h is the combination map with different orientations. The above AC-HRTEM analysis further proves that lithium is in the form of single crystal while surface Li2O is polycrystalline.Fig. 4AC-HRTEM of alkali metal particles.a A lithium particle grown along [110] direction. b AC-HRTEM of the particle, (c) partial enlarged view and simulated AC-HRTEM of lithium as inset. d EFTEM of a lithium particle. e Oxidized lithium particles and partial enlarged view as inset. f, g Lithium oxide with different orientations and their corresponding selected Fourier transformation spots as insets. h Superposition image of lithium oxide. i A sodium particle grown along [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$11\bar 2$$\end{document}112¯] direction. j AC-HRTEM of the particle, (k) partial enlarged view and simulated AC-HRTEM of sodium as inset. l EFTEM of a sodium particle. m Sodium oxide on the edge and its Fourier transformation spots inset. n, o Sodium oxide compounds, sodium, and their corresponding selected Fourier transformation spots inset, respectively. p Superposition image of different compounds. f–h, n–p These images are pseudo-colored.Similarly, Fig. 4i–p depicts the condition of sodium particles (dose-rate ~850 e Å−2 s−1). The single-crystalline feature of sodium particle can be observed, covered by surface oxide compounds. Besides the AC-HRTEM analysis, dark field TEM experiments were carried out and the results (Supplementary Fig. 8) agree with high magnification observations. Moreover, the thickness of lithium and sodium particles was measured taking advantage of the EFTEM technique. The average thickness is normally <70 nm (Supplementary Fig. 9). The difference in sodium particles is that the oxide layers (in Fig. 4m) are thinner than those of lithium particles (in Fig. 4e), as the AC-HRTEM observation of sodium particles was carried out after they have been placed in TEM vacuum chamber for only several minutes, shorter than the case of lithium particles. This indicates that the formation of oxide layer could be dependent of the time the particles exposed inside the vacuum chamber. Further comparison experiments confirm such deduction. As detailed in Supplementary Fig. 10, the thickness of oxide layers increased with the time exposed in the vacuum chamber. Meanwhile, in SAED experiments, the enhancement of the signal of Li2O polycrystalline rings versus time also supports this conclusion. To be noted that electron beam was blanked during the oxide layer thickening process (Supplementary Fig. 10), so that the possibility of beam-induced oxidation could be ruled out.Although alkali metal is easily to be oxidized, it is puzzling oxidation could occur in the TEM vacuum chamber (10−5 Pa). In the next step, the oxidation is investigated. For the alkali metal grown from the carbonate, one may doubt that the oxygen is originated from carbonate itself. However, the comparison experiments performed on lithium fluorides indicate that oxidation could occur even there was no oxygen source (Supplementary Fig. 5). This can be further proved by detailed energy dispersive spectroscopy (EDS) and electron energy loss spectroscopy (EELS) analysis (Supplementary Figs. 11–13). An in situ comparison EDS detection has revealed that there was almost no oxygen signal before the formation of lithium particles from LiF whereas relatively large amount of oxygen was detected at the end (Supplementary Fig. 11). Such difference rules out the oxygen source from the starting material as well as possible environmental oxygen adsorption on its surface. EDS and EELS mapping (Supplementary Figs. 12, 13) confirms our previous AC-HRTEM results (Fig. 4) that oxides were distributed on the surface of the particles. To further rule out the possible environmental oxygen adsorption on the carbon membrane of TEM grids, comparison experiments were carried out on pure gold and nickel-metal grids without any supporting membranes. It turned out that oxidation still occurred on the initially formed lithium particles (Supplementary Fig. 14). On this basis, we attribute the oxidation to the possible existence of trace oxygen in the high vacuum chamber, and interestingly alkali metals could react with the trace oxygen efficiently and result in oxygen segregation as oxides. Moreover, cooling experiments were also performed. Compared with the room temperature condition, alkali metal particles almost could not be generated under cryogenic condition at −178 °C (Supplementary Movies 7, 8). Lowering the temperature may increase the energy barrier for the alkali metal formation so that the formation of alkali metal particles cannot be observed.Ambiguities in lithium metal growth on garnet-type solid electrolytes of lithium-ion batteriesAs one of the important candidates of the next generation solid electrolyte, Li7La3Zr2O12 (LLZO) and its derivates have been widely investigated because of their relatively high ionic conductivity and good chemical stability against lithium metal anode24,25. It has been reported by different research groups that lithium metal could be grown from LLZO under the electron beam irradiation26,27. Here we employed our method on LLZO and our observations indicate lithium metal growth on LLZO could be an illusion.At low magnification, growth of lithium particles from the surface of LLZO could be observed, as shown in Fig. 5a (indicated by red arrows). Such phenomenon is similar with the observations in Fig. 1, and looks in agreement with previous reports using scanning electron microscopy (SEM)26,27. However, a detailed analysis on the thinner edge of the LLZO particle shows that the edge is lithium carbonate instead of LLZO. The SAED pattern in Fig. 5b is indexed to be lithium carbonate, corresponding to the selected area in Fig. 5a. Therefore, we conclude that lithium metal is grown from the lithium carbonate, which is the well-known surface contamination of LLZO. By comparison, we selected a clean LLZO particle and the phenomenon of lithium metal growth could not be observed (Fig. 5c). SAED on the edge of the particle suggests pure phase of LLZO, as shown in Fig. 5d. The only effect of electron beam bombardment is that the surface of LLZO turns into amorphous layer (outside the orange dotted line in Fig. 5c). Hence, compared with previous reports, our high-resolution observations indicate that lithium metal could not be grown on pure LLZO phase and the true origin may come from lithium carbonate contamination, which is easily formed on the garnet surface28. In contrast to low-resolution analysis methods such as using SEM, the combination of low dose-high-resolution imaging, electron diffraction in TEM as well as the EELS elemental determination is the key to comprehensive analysis of lithium growth phenomenon.Fig. 5Lithium growth from the surface contamination layer of LLZO.a Lithium particles (indicated by red arrows) grown from lithium carbonate contamination layer. Black circle represents the selected area for SAED in (b). b the SAED pattern is indexed to be Li2CO3. c No lithium particle growth can be observed on pure surface of LLZO upon electron irradiation. Orange dotted lines mark the amorphous layer caused by irradiation. d SAED of the region in (c) is indexed to be LLZO.Contact property of lithium metalThe research on lithium-lithium contact as well as lithium-lithium oxide contact is difficult to carry out in a conventional way. Another advantage of our method we demonstrate here is that the physical contact property of lithium metal and lithium oxide could be studied. Wettability of liquid lithium has been studied29 whereas the property of solid lithium contact is less known. On the other hand, contact between lithium metal and lithium oxide is of interest as lithium oxide is an important inorganic component of solid electrolyte interphase (SEI) on the surface of lithium dendrite in batteries7.The experiment setup is shown in Fig. 6a. Utilizing a scanning tunneling microscopy (STM) tip inside the microscope, lithium whisker could be produced by attaching the tip to the lithium particle and then pulling away. The black circle represents the area irradiated by electron beam for the generation of lithium. The in situ formation process of lithium whisker is shown in Supplementary Movie 9. Continuously stretching of the whisker would break it and fresh lithium tips were exposed (Fig. 6b and Supplementary Movie 10). Therefore, the contact property could be studied using these lithium tips. Firstly, Li–Li contact is demonstrated in Fig. 6c and Supplementary Movie 11. Pushing two Li tips to contact with each other right after the breakdown, the two tips can recombine together into one whisker again. On contrast, the Li2O–Li2O contact is demonstrated in Fig. 6d and Supplementary Movie 12. Waiting for several minutes after the whisker breakdown, the surface of Li tips was covered by oxide layers. When the two tips got in touch, they could not recombine. Further pushing of two tips caused deformation on both sides. Finally, the Li–Li2O contact is demonstrated in Fig. 6e and Supplementary Movie 13. When a fresh Li tip contacted with an oxidized tip, they could combine together. The corresponding schematic diagrams for all these cases are shown in Fig. 6f–j.Fig. 6Contact property of lithium metal and surface oxide layer.a Formation of lithium whisker with the help of an STM tip. The black circle represents the area where initial lithium particles were formed. b Breakdown of the lithium whisker, and two fresh lithium tips are obtained. c Li–Li contact. d Li2O-Li2O contact. e Li–Li2O contact. f–j Corresponding schematics of (a–e). k Li(110)-Li(110) interface model, l Li2O(001)-Li2O(001) interface model, m Li(001)-Li2O(001) with Li–Li contact at the interface, n Li(001)-Li2O(001) with Li–O contact at the interface. The interface regions are labeled by blue dotted boxes.To better understand the above contact properties, first-principle calculations were performed. The interfacial work of adhesion Wad between two surfaces is chosen as the criterion to evaluate the interaction between two surfaces. Figure 6k–n shows the relaxed interface structures of Li(110)-Li(110), Li2O(001)-Li2O(001), and two types of Li(001)-Li2O(001) interfaces, respectively. The interfacial work of adhesion was calculated to be −1.01 J m−2 for Li(110)-Li(110), and −1.21 J m−2 for Li(001)-Li2O(001) with Li–Li contact at the interface, and −6.43 J m−2 for Li(001)-Li2O(001) with Li–O contact at the interface. The negative values of Wad suggest the combination of the two surfaces is energetically favorable, in good agreement with the experimental results. However, Wad was calculated to be −7.92 J m−2 for Li2O(001)-Li2O(001), which is in contradiction with the experiment that oxide layers could not be combined together. It is speculated that the discrepancy is related to the polycrystalline feature of the lithium oxide layers. Random orientation of lithium oxides and grain boundaries may increase the interfacial work of adhesion so that the contact becomes unfavorable. The good contact between lithium metal and others could be closely related with the liquid-like fluidity of solid lithium as shown in these experiments. Our results on the basic contact property of lithium metal and surface passivation oxide layer provide support for better understanding of lithium dendrite growth and SEI related issues in batteries. For example, the phenomenon of “dead lithium” finds support from the bad contact property between surface passivation layers.Mechanism of the alkali metal formationFinally, combining with first-principle calculations, the formation mechanism of alkali metal particles is discussed. It is clear that the reaction is initiated by the electron beam, which deposits energy to the materials when it is incident to them. The interaction between high energy particles and materials30–33 can result in thermal heating30, knock-on damage31, radiolysis damage32, and charging effect32,33. In terms of thermal decomposition, in situ heating experiment of LiF (Supplementary Fig. 15) indicates that heating effect shouldn’t be the main driving force for the formation of alkali particles, as alkali metal particles did not show up just by heating. Furthermore, first-principle calculations (Supplementary Fig. 16) of the thermal decompositions for carbonates and fluorides under the ambient and vacuum conditions were performed, and pure alkali metals could not be the products thermodynamically, confirming that thermal decomposition should not be the predominant reaction mechanism. Hence, the formation of alkali metal is attributed to the radiolysis and charging effect. Irradiation of alkali salts can promote the formation of Frenkel defects in the crystal, and mobile state F* centers can recombine with surface terrace edge and initiate emission of alkali atoms21,34. These alkali atoms can aggregate on the surface of alkali salts and form alkali metals as a result. While the growth of alkali metal particles and whiskers could be the competition between knock-on damage and diffusion flux.DiscussionWe propose a simple and versatile method to form alkali metals in situ inside TEM directly. With this method, we have visualized the growth of alkali metals at atomic spatial resolution and millisecond temporal resolution. Oxidation of the alkali metals was observed, and the distribution and formation of surface oxide component was investigated. Throughout the text, this method shows its great compatibility in different experiments, and its applications in other in situ experiments are also predictable. As practical applications, we clarify the ambiguities in lithium metal growth on garnet-type solid electrolytes for lithium-metal batteries. On the other hand, we demonstrate a direct way to study the contact property of lithium metal as well as its surface passivation oxide layer.MethodsMaterialsFor the electron microscopy study, pure lithium carbonate (Li2CO3, 99.998%, from Aladdin), sodium carbonate (Na2CO3, 99.8%, from Tansoole), lithium fluoride (LiF, 99.99%, from Aladdin) were used.X-ray diffraction and electron microscopyPowder X-ray diffraction patterns were obtained using a Bruker AXS D8 Advance diffractometer with a Cu Kα source (λ
Cu Kα = 1.54 Å).TEM experiments were performed on JEOL JEM-2100 Plus (200 kV) and double aberration-corrected JEOL GrandArm (300 kV). To be specific, HRTEM images in Figs. 3–4, EFTEM images and EELS were obtained on JEOL GrandArm. The Gatan Oneview IS camera enabled fast in situ data acquisition. Other TEM results were obtained on JEOL JEM-2100 Plus. Cryogenic experiments were carried out with Gatan double tilt cooling holder (Model 636) which can sustain a low-temperature environment at −178 °C. In situ heating experiments were carried out with Protochips (Fusion 350) in situ heating holder. Lithium whisker contact experiments were carried out with an in situ STM tip holder (PicoFemto). AC-HRTEM images of lithium and sodium particles were collected under the negative spherical aberration (CS) imaging (NCSI) condition. Average background subtracted filtering was carried out based on the script from D. R. G. Mitchell and method by Kilaas35.Supplementary informationSupplementary InformationPeer Review FileMovieS1MovieS2MovieS3MovieS4MovieS5MovieS6MovieS7MovieS8MovieS9MovieS10MovieS11MovieS12MovieS13 | nature communications | [
"Article"
] | [
"Solid-state chemistry",
"Batteries",
"Transmission electron microscopy"
] | long history research alkali metals1 excitement energy conversion lithium metal ultimate choice anode battery highest theoretical capacity lowest electrochemical potential2,3 alkali metals less known structural information atomic-level reactive exist in air elemental form vacuum cryogenic transfer reliable alkali metal samples contaminations degradations electron beam sensitive alkali metals disables high-resolution transmission electron microscopy) imaging atomic-scale imaging of cryo-transferred lithium reported at low temperature7,8 room temperature imaging intercalating lithium into graphene sheets9 direct room temperature atomic resolution imaging of bare alkali metals never demonstrated direct in situ observation growth alkali metals at high spatiotemporal resolution not achieved missing link microstructure lithium metals hinders understanding development lithium-ion batteries work simple method to form alkali metals inside transmission electron microscope) alkali metals produced from alkali salts two types alkali salts carbonates fluorides chosen carbonates fluorides used industrial technical medical applications1 Li2CO3 first chemical in lithium production chain key component in lithium-air batteries11essential medicine manic depression bipolar Li2CO3 LiF starting materials ceramics glass cements1 research on stability decomposition lithium carbonates fluorides interesting electron beam-induced decomposition reaction found pure alkali metals products multistep reactions one-step reaction study structure structural kinetics alkali metals electron beam damage minimized atomic resolution obtained low dose aberration-corrected HRTEM Atomic resolution imaging lithium sodium at room temperature situ fast camera enables millisecond temporal resolution alkali metal growth paradigm investigate chemically reactive electron beam sensitive materials ambiguity about lithium metal growth solid electrolytes-metal batteries clarified physical contact property lithium metals surface passivation oxides revealed situ formation of alkali metals from alkali saltsFigure 1a configuration reaction triggered by electron beam rate controlled by intensity irradiation control dose-rate electron beam 10~1000 e Å−2 s−1 good choice focusing high energy electron beam alkali carbonate/fluoride particles alkali metal particles grow hundreds nanometersreaction rate controlled tuning beam intensity detector atomic resolution imaging of lithium sodium demonstrated at room temperature Details pure alkali carbonates/fluorides drop-casted onto TEM grids structural compositional characterizations of alkali salts shown in Supplementary Figs. 1–3-phase alkali salts starting materials three examples alkali metal formation growth process demonstrated Figure 1b–d show growth lithium from lithium carbonate sodium from carbonate lithium from fluoride sharp edges corners indicate crystalline feature particles particles expand along several growth directions white electron beam-induced formation of alkali metals from alkali salts common.Fig. 1In situ formation growth of alkali metal particles process alkali (Li Na) metal particle formed from materials under beam irradiation structure of alkali metals Atomic resolution images of lithium sodium metals at room temperature growth of Li from lithium carbonate Na from sodium carbonate Li from lithium fluoride confirm composition structure area electron diffraction (SAED) performed on lithium sodium particlesFigure 2a shows particles lithium carbonates black circle area diffraction Figure 2b c SAED pattern rotational average spectrum 2d g structures lithium lithium oxide polycrystalline diffraction patterns rotational average spectra Fig. 2e f h i particles lithium lithium oxide polycrystalline-featured lithium oxide single diffraction spots lithium analysis sodium case Fig. 2j–r particles single-crystalline alkali metals polycrystalline oxide compounds. 2SAED analysis alkali metal particles lithium particles black circle aperture SAED pattern rotational average spectrum Crystal structure b.c.c. lithium polycrystalline diffraction rings rotational average spectrum structure f lithium oxide diffraction rings spectrum sodium particles SAED pattern rotational average spectrum Crystal structure b.c.c. sodium polycrystalline diffraction rings rotational average spectrum Crystal structure f. sodium oxide diffraction rings rotational average spectrumGrowth kinetics lithium understand formation alkali metals oxides growth particles traced high spatiotemporal resolution18 combining low dose AC-HRTEM imaging fast camera detection lithium-particle growth demonstrated atomic resolution fast frame rate (40 ms per pure-phase lithium metal formed grew beginning Figure 3a shows AC-HRTEM image lithium particle [111] zone axis b.c. structure verified Fourier transformation pattern surface two exposed <110> facets seen in situ movie Movie 4 snapshots Fig. 3) indicates lithium particle grew two directions Figure 3a–c depicts growth stage direction indicated red arrows\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document}$10\bar\end{document}101 facet moved forward\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document}$1\bar\end{document}11 facet unchanged Evolution\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts{amsbsy{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$10\bar\end{document}101 edge highlighted single solid lines dotted lines double solid layer-by-layer growth of lithium atoms observedstage lasted 600 ms (15 frames grew slowly three seconds particle changed growth direction\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin}{-69pt}{document$$1\bar 10\end{document}11 ̄0 Fig. 3d–f\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}{document$1\bar 10{document}11 ̄0) facet moved forward red arrowsCompared with\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}-69pt}$10\bar{document}101 edge, growth along[12pt{amsmath{wasysym{mathrsfs{upgreek-69pt direction quick atoms at corner not replenished in short time incomplete\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amsbsy{mathrsfs{upgreek\oddsidemargin}{-69pt}{document$$2\bar 1\bar\end{document}21 ̄1 facet exposed as transient state time lithium atoms replenished gradually corner surface changed to high index facets length corner from 14 nm to 6 nm in sharp corner see Supplementary Movie 4) stage lasted for about 400 ms. For alkali metals with b.c.c.structure <110> face most densely packed lowest surface energy low index facets Millisecond temporal resolution enables fast structural evolution surface energy driving for growth alkali metals.Fig. 3Growth kinetics lithium metal.a–c Growth along\documentclass[12pt]{minimal}{amsmath{wasysym-69pt] direction. [111]-oriented lithium particle grown from lithium fluoride Fourier transformation pattern Red single solid lines represent edge particle Image filtering applied enhance signal to noise ratio Dose-rate ~1000 e Å−2 s−1. b Edge particle tracked by red dotted lines middle area additive grown portion. c Edge tracked red double solid lines middle area additive grown portion. d–f Growth along\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{upgreek-69pt$1}11 ̄0] direction. d Red single solid lines represent edge particle.\documentclass[12pt]{minimal\usepackage{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek\setlength\oddsidemargin-69pt}\begin{document}$$2\bar 1\bar\end{document}21 surface exposed Fourier transformation spot marked red circle Edge particle tracked red dotted lines f red double solid lines Growth length versus time Arrows represent starts new growing processes Growth rate versus time Arrows mark growing steps i j lithium whisker growth Type I Type II k l Growth length growth rate versus time.Figure 3g, h growth length rate particle versus time Particle length at t = 0 zero point Arrows mark start new growth period particle grows quickly slows down growth steps form growing process not continuous delivery new lithium source from alkali salts takes time After change growth direction growth new direction fast Fig. 3h growth rate blue arrow one quicker than previous fastest growth rate red indicates sudden influx large number lithium atoms accumulation diffused lithium atoms before change growth directiondifferent growth rate between\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts{amssymb{amsbsy{mathrsfs{upgreek and[12pt{minimal{amsmath{wasysym direction related with anisotropic mass transfer directions19 rate of mass transfer determined by decomposition of matrix lithium salts Anisotropic decomposition anisotropic diffusion flux of lithium source different directions growth lithium particles observed growth of lithium dendritic whiskers growth compared with electrical current-driven growth of lithium dendrites in lithium-ion batteries20 understanding dendrite growth Figure 3i, j shows growth of lithium whiskers from as-synthesized lithium particle Type I Movie 5) and root lithium fluoride materials Type II 6) growth length and smoothed growth rate versus time shown in Fig. 3k, l,two examples whiskers grew straightly without deflection growth rate Type I correlated with time correlation coefficient 0.769 raw data 0.993 smoothed data), growth kinetics described as:1\documentclass[12pt]{minimal}{amsmath{wasysym$$L = \frac{1}{2}at^2$${document}L L length whisker t time a acceleration accelerated growth mode a = 2.56 nm s−2 (0.6–5.6 s) for case Fig.3i contrast Type I growth rate Type II 60~100 nm s−1 length increases linearly with time correlation coefficient 0.998). uniform growth mode kinetics follows:2\documentclass[12pt]{minimal}{amsmath$$L = vt$${document}L=vtwhere v represents speed case in Fig.3j v = 85 nm s−1 (0–5.8 s).Uniform growth Type II lithium whiskers suggests decomposition lithium salts uniformly formation lithium metal Type I lithium whisker from formed lithium particle accelerated growth different mechanism additional driving force lithium whisker growth liquid electrolyte lithium-ion similar growth kinetics lithium particles driving force growth environment (vacuum liquid different Type I growth lithium source replenished environment liquid electrolyte lithium particles only lithium supplier no collapse contrast change lithium particle Fig. 3i rapid replenishment lithium from 1D-growth surface passivation layers as-formed lithium particles (Type I unique nucleation sites lithium salts diffusion confined (Type II different growth kinetics related diffusion barriers lithium metal lithium growth process particles whiskers competition between radiation damage diffusion flux particle/whisker grows radiation damage severe diffusion flux depleted growth rate slowsenvironments (CO220 N220 vacuum liquid22) effect growth kinetics lithium metal related to surface passivation layers sensitive to environment 1D- or 3D-growth mode related to confinement surface passivation layers diffusion flux driving force (electron beam irradiation electrical cross sections whiskers deflection longer in Fig. 4 Type I whisker hexagonal cross section consistent with electrochemically deposited lithium whiskers7 intrinsic features lithium whiskers grown in vacuum similarities with batteries method may alternative study lithium metal-related structure problems high spatiotemporal resolution advantage situ growing process sharp crystalline surface facets single-crystalline feature particle maintained alkali metal oxides show up early stage alkali metal oxides found in final SAED analysis (Fig. 2 Fig. 5) electron beam-induced decomposition pure alkali metals formed Oxidation subsequent step nucleation fast growth alkali metal particles reached area/volume (Fig. 1) growth slowed down Static imaging spectroscopy applied verify spatial distribution alkali metals oxide compoundsFigure 4a shows AC-HRTEM image-rate ~1050 Å−2 s−1) lithium particle [110] direction Single-crystalline feature Figure 4b close-up atom columns observed magnified image Fig. 4c agrees simulated AC-HRTEM image 6) lithium Fig. 4d shows elemental lithium mapping particle energy-filtered TEM Atomic resolution imaging mapping fresh particle grows single lithium Fig. 4a lattice fringes Moiré fringes blue square overlap Li2O lithium analysis Supplementary Fig. 7) Li2O distributed outer surface particles Figure 4e shows particles TEM vacuum chamber tens minutes partial close-up surface lattice Li2O Fourier component analysis orientation Li2O crystal grains result Fig. 4f–h Li2O three different orientations found colors Fourier transformation spots Figure 4h combination map different orientations AC-HRTEM analysis proves lithium single crystal surface Li2O polycrystalline.Fig. 4AC-HRTEM alkali metal particles lithium particle grown [110] direction AC-HRTEM partial enlarged view simulated AC-HRTEM lithiumEFTEM lithium particle Oxidized lithium particles partial enlarged view g Lithium oxide different orientations Fourier transformation spots Superposition image lithium oxide sodium particle grown[12pt{minimal{amsmath\oddsidemargin-69pt$11 direction AC-HRTEM particle partial enlarged view simulated AC-HRTEM sodium EFTEM sodium particle Sodium oxide edge Fourier transformation spots Sodium oxide compounds sodium Fourier transformation spots Superposition image different compounds f–h n–p images pseudo-colored Fig. 4i–p condition sodium particles (dose-rate ~850 e Å−2 s−1) single-crystalline feature sodium particle covered surface oxide compounds AC-HRTEM analysis dark field TEM experiments results agree high magnification observations thickness lithium sodium particles measured EFTEM technique average thickness <70 nm 9) difference sodium particles oxide layers Fig. 4m thinner lithium particlesAC-HRTEM observation of sodium particles after in TEM vacuum chamber several minutes shorter than lithium particles formation oxide layer of time particles exposed vacuum chamber comparison experiments confirm deduction thickness oxide layers increased with time exposed vacuum chamber SAED experiments enhancement signal of Li2O polycrystalline versus time supports conclusion electron beam blanked during oxide layer thickening beam-induced oxidation ruled out alkali metal oxidized oxidation in TEM vacuum chamber (10−5 oxidation investigated alkali metal from carbonate doubt oxygen from carbonate comparison experiments on lithium fluorides indicate oxidation could occur even no oxygen source proved by energy dispersive spectroscopy (EDS) electron energy loss spectroscopy (EELS) analysis EDS no oxygen signal before formation lithium particles from LiF large amount oxygen detected at end difference rules out oxygen source from starting material environmental oxygen adsorption EDS EELS mapping confirms AC-HRTEM results oxides distributed on surface particlesrule out environmental oxygen adsorption on carbon membrane TEM grids comparison experiments on gold nickel-metal grids without membranes oxidation occurred on initially formed lithium particles oxidation to trace oxygen in high vacuum chamber alkali metals react with oxygen result oxygen segregation as oxides cooling experiments performed alkali metal particles generated under cryogenic condition at −178 °C Lowering temperature energy barrier for alkali metal formation.Ambiguities in lithium metal growth on garnet-type solid electrolytes of lithium-ion Li7La3Zr2O12 (LLZO) derivates investigated high ionic conductivity chemical stability against lithium metal anode24 lithium metal grown from LLZO under electron beam irradiation26 employed method on LLZO observations indicate lithium metal growth low magnification growth lithium particles from surface LLZO observed in Fig. 5a similar with observations Fig. 1 with previous reports microscopy detailed analysis on thinner edge LLZO particle shows lithium carbonate instead of LLZO SAED pattern in Fig. 5b lithium carbonateconclude lithium metal grown from lithium carbonate surface contamination LLZO selected clean LLZO particle lithium metal growth observed (Fig. SAED edge suggests pure phase LLZO Fig. 5d electron beam bombardment surface LLZO turns amorphous layer orange line high-resolution observations indicate lithium metal grown on pure LLZO phase origin from lithium carbonate contamination garnet contrast low-resolution SEM low dose-high-resolution imaging electron diffraction TEM EELS elemental determination key analysis lithium growth.Fig. 5Lithium growth from surface contamination LLZO Lithium particles red grown from lithium carbonate contamination layer Black circle selected area for SAED SAED pattern indexed Li2CO3 No lithium particle growth on pure surface LLZO electron irradiation Orange lines mark amorphous layer caused irradiation SAED region (c) indexed LLZO.Contact property lithium research lithium-lithium contact oxide contact difficult advantage method physical contact property lithium oxide studied liquid lithium solid lithium contact less knowncontact between lithium metal lithium oxide important inorganic component solid electrolyte interphase) lithium dendrite experiment setup Fig. 6a scanning microscopy) tip lithium whisker produced attaching to lithium particle pulling away black circle area irradiated electron beam generation lithium formation lithium whisker Supplementary Movie 9. stretching whisker fresh lithium tips exposed (Fig. 6b Supplementary Movie contact property studied using lithium tips Li–Li contact in Fig. 6c Supplementary Movie 11. two Li tips breakdown recombine whisker Li2O–Li2O contact Fig. 6d Supplementary Movie 12. breakdown surface Li tips covered by oxide layers tips recombine pushing deformation Li–Li2O contact Fig. 6e Supplementary Movie 13. fresh Li tip with oxidized tip schematic diagrams Fig. 6f–j.Fig. 6Contact property lithium metal surface oxide layer Formation lithium whisker STM tip black circle initial lithium particles formed Breakdown lithium whisker two fresh lithium tips obtained Li–Li contact Li2O-Li2O contact Li–Li2O contactf–j schematics (a–e). Li(110) Li2O(001)-Li2O(001) Li(001)-Li2O(001) Li–Li Li(001)-Li2O(001 Li–O contact interface regions labeled blue boxes contact properties first-principle calculations performed interfacial work adhesion Wad surfaces criterion Figure 6k–n interface structures Li(110)-Li(110), Li2O(001)-Li2O Li(001)-Li2O(001) interfaces interfacial calculated −1.01 J m−2 Li(110)-Li(110), −1.21 J m−2 Li(001)-Li2O(001) Li–Li contact −6.43 J m−2 Li(001)-Li2O(001) Li–O contact negative values Wad suggest combination energetically favorable experimental results Wad −7.92 J m−2 Li2O(001)-Li2O(001), contradiction discrepancy polycrystalline feature lithium oxide layers lithium oxides increase interfacial work adhesion contact unfavorable good contact lithium metal liquid-like fluidity solid lithium results contact property lithium metal surface passivation oxide layer lithium dendrite growth SEI issues batteries“dead lithium” support from bad contact between surface passivation layers alkali metal formation mechanism of alkali metal particles discussed reaction initiated by electron beam deposits energy to materials interaction between high energy particles in thermal heating30 knock-on radiolysis charging situ heating experiment LiF heating main driving force for formation alkali particles by heating first-principle calculations thermal decompositions for carbonates fluorides under ambient vacuum conditions pure alkali metals thermal decomposition predominant reaction mechanism formation alkali metal attributed to radiolysis charging effect Irradiation of alkali salts Frenkel defects mobile state F* centers recombine with surface initiate emission alkali atoms21 alkali atoms on form alkali metals growth of alkali metal particles whiskers between knock-on damage diffusion flux simple method to form alkali metals in situ inside TEM visualized growth alkali metals at atomic spatial millisecond temporal resolution Oxidation metals observed distribution formation of surface oxide component investigatedmethod shows compatibility applications predictable clarify ambiguities lithium metal growth electrolytes batteries study contact property lithium metal surface passivation oxide layer electron microscopy lithium carbonate (Li2CO3 99.998% sodium carbonate (Na2CO3 99.8% lithium fluoride (LiF 99.99% Aladdin used-ray diffraction electron X-ray diffraction Bruker AXS D8 Advance diffractometer Cu Kα source 1.54 experiments JEOL JEM-2100 Plus (200 kV double aberration-corrected JEOL GrandArm (300 HRTEM EFTEM images EELS JEOL GrandArm Gatan Oneview IS camera fast data TEM results JEOL JEM-2100 Plus Cryogenic experiments Gatan double tilt cooling holder low −178 °C heating experiments Protochips 350 holder Lithium whisker contact experiments STM tip holder AC-HRTEM images lithium sodium particles collected negative spherical aberration imaging condition Average background subtracted filtering script D. R. G. Mitchell method Kilaas35Supplementary Review | 47.5 | 0.75518 |
10.1038/s41467-020-19625-9 | PMC7669885 | Yellow fever is absent from the Asia/Pacific region, despite presence of the mosquito vector. Here, the authors demonstrate that mosquitoes collected from field sites across the region are capable of transmitting yellow fever virus, indicating that vector competence is not a barrier to disease spread. | Historically endemic to Sub-Saharan Africa and South America, yellow fever is absent from the Asia-Pacific region. Yellow fever virus (YFV) is mainly transmitted by the anthropophilic Aedes mosquitoes whose distribution encompasses a large belt of tropical and sub tropical regions. Increasing exchanges between Africa and Asia have caused imported YFV incidents in non-endemic areas, which are threatening Asia with a new viral emergence. Here, using experimental infections of field-collected mosquitoes, we show that Asian-Pacific Aedes mosquitoes are competent vectors for YFV. We observe that Aedes aegypti populations from Singapore, Taiwan, Thailand, and New Caledonia are capable of transmitting YFV 14 days after oral infections, with a number of viral particles excreted from saliva reaching up to 23,000 viral particles. These findings represent the most comprehensive assessment of vector competence and show that Ae. aegypti mosquitoes from the Asia-Pacific region are highly competent to YFV, corroborating that vector populations are seemingly not a brake to the emergence of yellow fever in the region. | IntroductionIn 2016, the return from Angola of 11 yellow fever (YF)-infected workers to China posed the threat of a YF epidemic in Asia never before seen1. Increasing volumes of trade and travels between China and Africa increase the risk of disease introductions. Yellow fever virus (YFV), endemic to Africa and South America, has so far remained absent in Asia. The reasons explaining this absence (e.g., transmission barrier resulting from low compatibility between mosquito and virus genotypes2,3, limited duration and low viraemia in humans, absence of a sylvatic cycle4,5, competition with well-established flaviviruses as dengue and Japanese encephalitis viruses6) are still poorly explored, making the possibility of an epidemic unpredictable.Similar to other flaviviruses, the common symptoms of YF are fever, headache, muscle aches, nausea, and vomiting, however, the in-hospital case fatality rate (CFR) could dramatically reach 67%7,8, giving this disease a particular interest for public health. Traced back to around 3000 years ago, YF was mainly encountered in Africa where it was isolated in 1927 in Ghana2. YFV was transported via ships sailing from West Africa to the West Indies during the slave trade. Massive and recurrent transports of goods also brought competent vectors such as the mosquito Aedes aegypti contributing to initiate YFV transmission cycles in ships and later, on land at their destination. The in-depth understanding of YFV transmission cycle in the early 1900s9 permitted to implement successful vector control strategies since 191610 and to develop the YFV 17D vaccine in 193611. However, YFV still causes an estimated 51,000–380,000 annual severe cases, of which 19,000–180,000 are fatal in Africa12. Insecticide-resistance of mosquito populations, as well as a supply shortage, distribution, and uptake of YFV vaccines, are among the main causes of this current burden13.To transmit an arbovirus such as YFV, the mosquito should acquire the virus by ingesting viremic blood from an infected host, the virus enters into the midgut epithelial cells and replicates. After a few days of incubation, the virus should pass through the midgut basal lamina and disseminate into the hemocele, then it infects the salivary glands for transmission to the vertebrate host14. Parameters such as midgut infection, viral dissemination in hemocele, and transmission through infectious saliva are used to determine mosquito vector competence, which is an indicator of transmission risk15. In Africa and South America, YFV is primarily transmitted in a forest cycle between non-human primates (NHP) and zoophilic mosquitoes (Aedes in Africa and Haemagogus/Sabethes in South America). The urban cycle of YFV involves mainly the mosquito Ae. aegypti in both Africa and South America16.The mosquito vectors Ae. aegypti and Aedes albopictus, are present in 154 countries putting nearly half of the world population at risk of YFV transmission. Ecological disturbances induced by urban habitats contribute to the proliferation of Ae. aegypti, supplanting Ae. albopictus in urban areas in Asia17. Aedes spp. mosquitoes are vectors of chikungunya, dengue, and Zika viruses in East and South-East Asia, which serve as a suitable environment for YFV. Increasing exchanges between Asia and Africa has raised the number of passengers between Asia and YF-endemic countries18–20. Notable increase of travels between countries with different capacities to detect and control infectious diseases (e.g., growth of tourism in emerging countries) can facilitate the geographic spread of vector-borne diseases20,21.Of greater concern was the report of YFV laboratory-confirmed cases among Chinese travelers returning to Asia after a stay in Angola during the 2015–2016 YF outbreak3, threatening billions of immunologically naive populations in Asia living in close vicinity of Ae. aegypti and Ae. albopictus mosquitoes1. Africa receives a large number of Chinese workers who are usually unvaccinated against YFV, increasing the risk of importing YF in Asia22. The combination of repeated introductions of viraemic travelers and immunologically naive local population in an environment suitable to transmission accentuates the risk of YF emergence in Asia. Although the vector competence for YFV of mosquitoes in Africa, South America, and Caribbean regions, has been investigated23,24, only limited information for Asian-Pacific mosquitoes could be found to measure the possible risk of YFV transmission in this region25,26. Investigating the vector competence for YFV of mosquitoes in the Asia-Pacific region is essential to assess the potential threat of YFV transmission in a region where YF outbreaks have never been reported27. Here, we show the vector competence of 18 populations of Ae. aegypti and Ae. albopictus from the Asia-Pacific region. We demonstrate that (i) Ae. aegypti mosquitoes from the Asia-Pacific region are more susceptible to the West-African genotype of YFV than Ae. albopictus, (ii) mosquitoes from Singapore, Taiwan, Thailand, and New Caledonia are capable of transmitting YFV at 14 days post-infection, and (iii) Ae. aegypti mosquitoes excrete up to 23,000 viral particles in saliva, suggesting that YFV could be transmitted through the saliva of infected Ae. aegypti in laboratory conditions.ResultsAedes aegypti mosquitoes are highly competent to YFV infectionAedes aegypti populations from the Asia-Pacific region were used in experimental infections to evaluate different components of the vector competence at 14 and 21 days post-infection (dpi).At 14 dpi, infection rate (IR) ranged from 41.7% (CAMB, Cambodia) to 95.8% (TRUNG, Vietnam; CSP, Thailand; TAINAN, Taiwan; NOUMEA, New Caledonia) and were significantly different when comparing all 12 populations (Fisher’s exact test: P < 10−4), 10 Asian populations (P < 10−4) and the two populations from the Pacific region (P = 0.02) (Fig. 1a). Dissemination rate (DR) ranged from 42.8% (FENG, Taiwan) to 86.9% (CSP, Thailand; NOUMEA, New Caledonia), with some populations presenting higher DR (Fisher’s exact test: P = 0.06); the 10 Asian populations presented similar DR (P = 0.13) while the two populations from the Pacific region presented significantly different DR (P = 0.04) (Fig. 1b). Based on transmission rate (TR), seven among 12 populations did not excrete virus in saliva. For the other five populations (CSP, SING, ANNAN, FENG, NOUMEA), TR ranged from 12.5% (SING, Singapore) to 45% (CSP, Thailand), and was significantly different (Fisher’s exact test: P < 10−4) (Fig. 1c).Fig. 1Vector competence of 12 Aedes aegypti populations assessed 14 and 21 days after an infectious blood meal containing 107 ffu/mL of YFV (West-African genotype).Batches of 20–24 mosquitoes were examined in each population for viral infection (a, d), dissemination (b, e), and transmission (c, f) by estimating respectively the proportion of mosquitoes with infected bodies (1), head (2), and saliva (3). Titrations were performed on C6/36 cells in 96-well plates. Viral particles were detected by FFA using a primary anti-YFV antibody and a secondary fluorescent-conjugated antibody. Infection rate (IR) refers to the percentage of mosquitoes having an infected body among blood-fed mosquitoes. Dissemination rate (DR) is the percentage of mosquitoes with an infected head (containing viral particles having disseminated in the general cavity after crossing successfully the midgut) among mosquitoes with an infected body. Transmission rate (TR) corresponds to the percentage of mosquitoes with infectious saliva (viral particles having successively crossed the two anatomical barriers, midgut and salivary glands) among mosquitoes with infected head. Stars indicate statistical significance of comparisons by Fisher’s exact test (two-sided test; *P ≤ 0.05, ****P ≤ 0.0001). a ****P ≤ 0.0001, *P = 0.02; b *P0.042; c ****P ≤ 0.0001; d ****P ≤ 0.0001; e ****P ≤ 0.0001. ns (non-significant) indicates a lack of statistical significance (P > 0.05). In brackets are the numbers of mosquitoes tested. dpi days post-infection. Source data are provided in Supplementary Data 1 file.To test whether a longer incubation time of mosquitoes might improve the vector competence, we used the same protocol to assess IR, DR, and TR at 21 dpi. IR reached 100% in four populations (BLX, Laos; CSP, Thailand; SING, Singapore; NOUMEA, New Caledonia), and remained significantly different between the 12 populations (Fisher’s exact test: P < 10−4) (Fig. 1d). DR ranged from 47.6% (NANZI, Taiwan) to 95.8% (CSP, Thailand) and differed between populations (Fisher’s exact test: P < 10−4) (Fig. 1e). However, TR ranged from 10% (VIET, Vietnam) to 56.5% (CSP, Thailand), but no significant difference between the 12 populations was evidenced (Fisher’s exact test: P = 0.10) (Fig. 1f). Collectively, these results show that all Ae. aegypti mosquitoes examined in this study are competent vectors of YFV with 42% of populations (5/12) able to transmit at 14 dpi and all (12/12) at 21 dpi.Aedes albopictus mosquitoes are less competent to YFV than Ae. aegyptiTo examine whether Ae. albopictus native to Asia can sustain a local transmission of YFV, vector competence indices, IR, DR, and TR were calculated for six populations at 14 and 21 dpi.At 14 dpi, IR ranged from 4.2% (THAI, Thailand) to 62.5% (FOSHAN, China), and significantly differed between populations (Fisher’s exact test: P < 10−4) (Fig. 2a). DR ranged from 0% (THAI, Thailand) to 85.7% (LINGYA, Taiwan), but no significant difference was evidenced between populations (Fisher’s exact test: P = 0.41) (Fig. 2b). The transmission was only observed for FOSHAN (TR = 22.2%) (Fig. 2c).Fig. 2Vector competence of 6 Aedes albopictus populations assessed 14 and 21 days after an infectious blood meal containing 107 ffu/mL of YFV (West-African genotype).Batches of mosquitoes were examined in each population for viral infection (a, d), dissemination (b, e), and transmission (c, f) by estimating respectively the proportion of mosquitoes with infected bodies (1), head (2), and saliva (3). Infection rate (IR) refers to the percentage of mosquitoes having an infected body among blood-fed mosquitoes. Dissemination rate (DR) is the percentage of mosquitoes with an infected head (containing viral particles having disseminated in the general cavity after crossing successfully the midgut) among mosquitoes with an infected body. Transmission rate (TR) corresponds to the percentage of mosquitoes with infectious saliva (viral particles having successively crossed the two anatomical barriers, midgut and salivary glands) among mosquitoes with infected head. Stars indicate statistical significance of comparisons by Fisher’s exact test (two-sided test; *P ≤ 0.05, **P ≤ 0.01, ****P ≤ 0.0001). a ****P ≤ 0.0001; d **P = 0.003; e *P = 0.038. ns (non-significant) indicates a lack of statistical significance (P > 0.05). In brackets are the numbers of mosquitoes tested. dpi days post-infection. Source data are provided in Supplementary Data 1 file.At 21 dpi, IR ranged from 8.3% (THAI, Thailand) to 54.2% (FOSHAN, China) (Fisher’s exact test: P = 0.003) (Fig. 2d). DR ranged from 0% (XKH, Laos and THAI, Thailand) to 100% (LINGYA (Taiwan)) (Fisher’s exact test: P = 0.04) (Fig. 2e), and TR to 66.7% (LINGYA, Taiwan) (Fig. 2f). Four Ae. albopictus populations (YYG, Japan; XKH, Laos; THAI, Thailand; PMNI, Brazil) were not able to transmit at both 14 and 21 dpi. These results indicate that Ae. albopictus populations are less competent to disseminate and transmit YFV than Ae. aegypti (Supplementary Figs. 1 and 2).Higher loads of viral particles excreted in the saliva of Ae. aegypti than Ae. albopictusTo study whether Ae. aegypti delivered a higher load of viruses in saliva than Ae. albopictus, we collected individual mosquito saliva that was titrated. We observe that among the five populations able to transmit at 14 dpi, the number of viral particles varied from 101.6±1.5 (NOUMEA, New Caledonia) to 103 (FENG, Taiwan) (Fig. 3a). At 21 dpi, all 12 populations deliver viral particles in saliva ranging from 5 (VIET, Vietnam) to 103.7±4 (NANZI, Taiwan: min-max: 10–23,000) (Fig. 3b). Comparatively, Ae. albopictus mosquitoes were able to deliver 101.7±1.7 viral particles (FOSHAN, China) at 14 dpi (Fig. 3c) and 102.2±1.4 viral particles (LINGYA, Taiwan; min-max: 133–167) at 21 dpi (Fig. 3d).Fig. 3Viral loads measured in individual mosquito saliva at 14 and 21 days after an infectious blood meal with West-African YFV.a, b Saliva viral loads of Aedes aegpti at 14 and 21 dpi; c, d saliva viral loads of Aedes albopictus at 14 and 21 dpi. Saliva was collected for 30 min using the forced salivation technique by removing legs and wings and inserting a tip containing FBS in mosquito proboscis. Salivas were titrated on C3/36 cells and the numbers of viral particles are expressed in ffu/saliva. ns (non-significant) indicates the lack of statistical significance for comparisons using the Kruskal–Wallis test (two-sided test; P > 0.05). Bars indicate the mean. In brackets are the numbers of mosquitoes tested. dpi days post-infection. Source data are provided in Supplementary Data 1 file.Lower dissemination of YFV in Ae. aegypti from the Asia-Pacific region compared to African mosquitoesTo evaluate whether higher viral loads in the body and head of mosquitoes could increase the chance for virus transmission through saliva, viral particles in the body, head, and saliva were estimated only for mosquitoes capable of viral transmission (Supplementary Fig. 3). Viral loads in the body (Fig. 4a) and saliva (Fig. 4c) were not significantly different (Kruskal–Wallis: P > 0.05), while viral loads in the head were significantly higher (Fig. 4b) in mosquitoes from Africa (104.6±3.7) compared to Asia (103.9±3.9) and Pacific (103.7±3.6) regions (Kruskal–Wallis test: P > 0.05).Fig. 4Virals loads in body, head, and saliva of Aedes aegypti populations from Asia, Africa, and the Pacific region.Mosquito (a) body, (b) head, and (c) saliva were titrated on C6/36 cells and the number of viral particles was expressed in ffu/sample. Stars indicate statistical significance of comparisons by the Kruskal–Wallis test (two-sided test; **P ≤ 0.01). b **P = 0.0095. ns (non-significant) indicates a lack of statistical significance (P > 0.05). Bars indicate the mean. In brackets are the numbers of mosquitoes tested. Red dots: samples from Asia; green triangles: samples from Africa; blue squares: samples from the Pacific region. Source data are provided in Supplementary Data 2 file.Viral loads in the body were significantly correlated with viral loads in the head (ρ = 0.31, P = 0.012) (Supplementary Fig. 4a). However, no correlation was detected between viral loads in body and saliva (ρ = 0.22, P = 0.11; Supplementary Fig. 4b), or between virals loads in head and saliva (ρ = 0.04, P = 0.77; Supplementary Fig. 4c). To investigate the difference in terms of viral loads in body, head, and saliva between mosquitoes from different geographic origins, we used a linear regression model. To identify the main factor conditioning the correlation between viral loads in body, head, and saliva, we used a logistic regression model. The analysis corroborated that compared to mosquitoes from Africa used as the reference, Ae. aegypti from Asia presented a lower viral load in the head (the level is −0.73 log lower in mean in Asian than in African mosquitoes) as for mosquitoes from the Pacific region (−0.82 in mean) (Table 1, P = 0.01). When analyzing viral loads in body and saliva, no significant difference was found between mosquitoes from Africa, Asia, and the Pacific region (Table 1, respectively, P = 0.11 and P = 0.54).Table 1Univariate linear regression analyses for the body, head, and saliva in Aedes aegypti mosquitoes, 21 days after the infectious blood meal at a titer of 107 ffu/mL.ContinentCrude coefficient (95% CI)PKruskal–Wallis testBody Africa10.11 Asia−0.39 (−0.83; 0.06) Pacific+0.19 (−0.48; 0.85)Head Africa10.01 Asia−0.73 (−1.21; −0.26) Pacific−0.82 (−1.53; −0.12)Saliva Africa10.54 Asia−0.35 (−0.97; 0.27) Pacific−0.10 (−1.02; 0.82)Analyses were performed according to the continent where mosquitoes were collected. Source data are provided in Supplementary Data 2 file.In bold, significant values (P ≤ 0.05).Taken altogether, these results indicate that compared to mosquitoes from Africa, Ae. aegypti mosquitoes from the Asia-Pacific region hosted significantly lower viral particles in the head but presented similar viral loads in body and saliva, suggesting that only viral dissemination distinguishes Ae. aegypti mosquitoes from the three continents.Ability of different Aedes aegypti populations to transmit YFVTo determine the risk of mosquito-mediated YFV transmission at each location, we used transmission efficiencies (Supplementary Fig. 1) and probabilities of vector occurrence (data from Kraemer et al.28). When considering only Ae. aegypti from Asia, regions where CSP (Thailand), TRUNG (Vietnam) and NANZI (Taiwan) populations are located, presented a higher transmission risk of YFV (CSP: 54% [32.8–74.4%], TRUNG: 25% [9.8–48.7%], NANZI: 21% [7.1–42.2%]). In these regions, mosquito occurrence is predicted to be high and overall constant within a 5 km radius, allowing for competent vectors to place immunological naive populations (humans and natural reservoirs) at risk of YFV infection (Fig. 5).Fig. 5Risk of Aedes aegypti-mediated YFV transmission in Asia.a Original data from Kraemer et al. showing the probability of encountering Aedes aegypti in South-East Asia. The colors correspond to probabilities: lower (blue) or higher (red) than the median probability across the whole map (white). b Modeled vector occurrence (colored bars matching the values from the scale in a) along with mosquito transmission efficiency (gray bar) of Aedes aegypti populations tested in laboratory conditions shown in a. This map uses data published by Kraemer et al. and was generated with R v4.0.1 (package raster v3.1-5). Source data are provided in Supplementary Data 1 file28.DiscussionTo our knowledge, only two studies have been published on vector competence for YFV of mosquitoes from the Asia-Pacific region25,26. In our study, we examine 18 mosquito populations and find that Ae. aegypti populations from the Asia-Pacific region are more competent to transmit YFV than Ae. albopictus from the same geographical area. Compared to Ae. aegypti from YFV-endemic regions in Africa, mosquitoes from Singapore, Taiwan, Thailand, and New Caledonia presented the highest potential to transmit YFV; the risk of transmission to human populations is high. Based on these results, we cannot exclude the possibility of a YF epidemic occurring in the Asia-Pacific region where Ae. aegypti is well-established.A previous study using Asian Ae. aegypti populations showed that in laboratory conditions, Ae. aegypti from Laos (Bolikhamsai province) were able to transmit YFV at least 14 days after exposure to YFV S-79 strain25. Conversely, when infected with the American genotype 1 of YFV (strain 74018, from Brazil), Ae. aegypti from Cambodia (Phnom Penh) and Vietnam (Ho Chi Minh city) were found to be susceptible to YFV26 with however lower dissemination efficiencies than in our study. We find that Ae. aegypti populations from the Asia-Pacific region are highly competent to transmit a YFV of the West-African genotype, giving legitimacy to the evaluation of the risk of YF epidemics in this YF-free region. Originally from tropical rainforests in Africa where it circulates between non-human primates and zoophilic mosquitoes, YFV was introduced into the Americas during the slave trade from the 14th century, as was the YFV vector, Ae. aegypti2. The eradication of Ae. aegypti led to the success in controlling YF, but the relaxation of vector control in the 1970s permitted Ae. aegypti to recolonize the region29. This species then became responsible for urban dengue outbreaks30 but was excluded from the YFV cycle, mainly sylvatic in South America31. Thus, YFV is absent elsewhere in the world except in Africa and in America, until 2016 when 11 YFV-infected workers returning from Angola were reported in China, putting the YF risk back on the agenda3. In Asia, all the ingredients to fuel a sylvatic cycle are gathered as well as an urban cycle: 49 of the 52 countries are considered to be suitable for the proliferation of Ae. aegypti and/or Ae. albopictus32, offering the fertile ground for YF transmission in addition to dengue fever33,34; even though the YFV-susceptible non-human primates of South America are absent in Asia35, Macaca spp. monkeys widely distributed in Asia might have the role of a YFV reservoir36 alongside YFV-susceptible zoophilic mosquitoes37.Interestingly, we show that when infected with a West-African YFV, all Ae. aegypti populations examined in this study are able to transmit at day 21 post-infection. With higher rates of dissemination than transmission, our results indicate that the midgut has a less significant role as a physical barrier than the salivary glands. However, the number of viral particles excreted by these mosquitoes is similar to the viral loads estimated from African mosquitoes suggesting that once able to transmit, the Asian-Pacific Ae. aegypti mosquitoes are as efficient as mosquitoes from YFV-endemic regions in Africa. To note, we use a West-African YFV isolated in 1979 to infect African mosquitoes from Cameroon and Congo; YFV strains from Senegal show low rates of evolutionary change over time38 and Ae. aegypti from Cameroon, Congo, and Senegal belong to the ancestral form (namely Ae. aegypti formosus) and present relatively low levels of genetic differentiation39 which taken together, limits the bias in estimating vector competence. Laboratory-observed infection experiments show that the proportion of mosquitoes infected and able to transmit YFV was highest for Ae. aegypti from Thailand (>50%, Supplementary Fig. 1). Likewise, a recent modeling exercise (data extracted from28) suggests that Ae. aegypti can commonly be found throughout South-East Asia (Fig. 5a). These results suggest that Ae. aegypti from the Asia-Pacific region are competent to YFV and prone to trigger a YF outbreak, strengthening the conclusions drawn from metapopulation models to assess the probabilities of YFV spread based on international airline transportation40, or disease transmission models using infection data, vaccination coverage, and different environmental factors41,42. However, it is important to note that assessing the risk of YFV transmission based on vector competence data as was done in our study conducted in laboratory conditions, does not reflect alone the capacity of mosquitoes to act as a field vector. Some environmental factors might shorten mosquito lifespan and, therefore diminish the probability of infecting after the extrinsic incubation period. Moreover, the viral titers used in our experimental infections may differ from viremias encountered in patients, 4.98 (3.50–5.79) log10 copies/mL of YFV RNA in blood43. Finally, viral transmission in our study is determined by detecting viral particles in mosquito saliva collected using the forced salivation technique (see “Methods” section) which do not reflect the physiological dose of viral particles delivered by a mosquito during the bite. Moreover, vector capacity integrates biotic and abiotic factors in addition to vector competence, and therefore, varies in space and time across a region; it can be influenced by population density, vector feeding behavior, and vector lifespan44. Apart from making vaccination mandatory, preventing YF outbreaks in the region should rely on controlling Ae. aegypti populations, particularly in regions suffering from dengue, chikungunya, and Zika. Although it seems difficult considering the failure in preventing and controlling dengue using conventional insecticides, combining an early detection of imported cases, a vaccination mandatory for travelers returning from countries at risk, a plan for implementing mass vaccination campaigns and securing the vaccine stockpile (still produced in embryonated chicken eggs causing occasional issues of supply), and new promising vector control methods (e.g., Wolbachia strategy) would significantly improve the prevention of YF as of other arboviral diseases45.We believe that more work should be done to determine the evolution of viral populations after the escape from the midgut, in the mosquito general cavity where the virus disseminates in various peripheral organs and replicate in disparate tissues. Viral loads in mosquito heads are significantly lower in YFV-infected Ae. aegypti from the Asian-Pacific region suggesting a mechanism able to limit viral replication such as the mosquito immune responses, in particular, the RNA interference, the most important antiviral response against arboviruses46. This may refine the mutational spectrum over time, with implications for the diversity of viruses excreted from the mosquito salivary glands and, therefore, viruses injected into the vertebrate host47. Other flaviviruses are exclusively endemic to Asia such as the Japanese encephalitis virus (JEV)48. It is then legitimate to question if this resident virus might interfere with a non-resident virus, namely YFV49.Notably, even if the 17D vaccine has been available since the 1930s, concerns regarding the safety and supply of YFV vaccine make part of the world vulnerable to YF emergence since the manufacturing process of the YF vaccine cannot cover the need for an immediate mass vaccination campaign13, even though fractional-dose YF vaccination could be an alternative to a shortage of full-dose vaccine50. Altogether, our work brings critical data on mosquitoes that deepen our understanding of factors leading to the emergence of arboviruses in order to be better prepared when YF hits the Asia-Pacific region for decision makers51,52.MethodsEthics statementAnimals were housed in the Institut Pasteur animal facilities (Paris) accredited by the French Ministry of Agriculture for performing experiments on live rodents. Work on animals was performed in compliance with French and European regulations on care and protection of laboratory animals (EC Directive 2010/63, French Law 2013-118, February 6th, 2013). All experiments were approved by the Ethics Committee #89 and registered under the reference APAFIS (Autorisation de Projet utilisant des Animaux à des FIns Scientifiques) #6573-201606l412077987 v2.YFV strainYFV strain S-79 (accession number: MK060080) was isolated from a patient returning from Senegal in 1979, passaged twice on mice brains, and twice on C6/36 cells53. Virus stocks for mosquito infections were produced on C6/36 cells and stored at −80 °C until use.Mosquito populationsTwelve Ae. aegypti and six Ae. albopictus populations were analyzed (Table 2 and Fig. 6). Mosquito eggs were collected using ovitraps placed in each locality and shipped to the Institut Pasteur (Paris) for infections. After egg hatching, around 200 larvae were distributed per pan containing one liter of dechlorinated water and yeast tablets as food. Larvae were reared until the adult stage in controlled conditions54. OF1 mice for feeding mosquitoes were between 6-week and 2-month-old, maintained in an animal facility under standard conditions (23 °C and 14:10 light/dark cycle) at Institut Pasteur.Table 2Mosquito populations, countries, localities, and generation used.Mosquito speciesCountryPopulation nameLocalityGenerationaDate of collectionCollaborationsAedes aegyptiCambodiaCAMBPhnom Penh308. 2018Boyer S. (Institut Pasteur of Cambodia)VietnamVIETAn Giang108. 2018Huynh T. (Institut Pasteur of Ho Chi Minh City, Vietnam)TRUNGTrung Muoi310. 2018LaosBLXBolikhamxay309. 2019Marcombe S. (Institut Pasteur of Laos)ThailandCSPChiang Mai402. 2019Jupatanakul N. (National Center for Genetic Engineering and Biotechnology, Thailand)SingaporeSINGSingapore12019Pompon J. (National University of Singapore)TaiwanFENGKaohsiung104. 2019Chen C.H. (National Health Research Institute, Taiwan)NANZIKaohsiung104. 2019ANNANTainan104. 2019TAINANTainan104. 2019New CaledoniaNOUMEANouméa (quartier Normandie)22019Pocquet N. (Institut Pasteur of New Caledonia)French PolynesiaPAEATahitiLab colony1994Failloux A.B.Aedes albopictusChinaFOSHANGuangdong ProvinceLab colony1981Chen X.G. (Southern Medical University, Guangzhou, China)JapanYYGTokyoLab colony2014Sawabe K. (NIID)TaiwanLINGYALingya604. 2019Chen C.H. (National Health Research Institute, Taiwan)LaosXKHXieng Khouang309. 2019Marcombe S. (Institut Pasteur of Laos)ThailandTHAIChiang Mai702. 2019Jupatanakul N. (National Center for Genetic Engineering and Biotechnology, Thailand)BrazilPMNINova Iguaçu82015Lourenço-de-Oliveira R. (Instituto Oswaldo Cruz, Brazil)aGeneration refers to the generation of mosquitoes after field collection. Lab colony refers to a mosquito strain that has been adapted to laboratory conditions for more than 20 generations.Fig. 6Geographical distribution of the 18 mosquito sample locations (12 Aedes aegypti and 6 Aedes albopictus).Black dots: Aedes aegypti; red dots: Aedes albopictus. The map was built using the open-source map site “https://d-maps.com/conditions.php?lang=en/”. Each dot corresponds to a sampling location.Mosquito infectious blood mealBoxes of sixty 10-day-old female adults were transferred into biosafety level-3 (BSL-3) laboratory 24 h prior to infection. The blood meal was composed of 1.4 mL of rabbit erythrocytes supplemented with 10 mM adenosine triphosphate (ATP) as a phagostimulant, and 0.7 mL of viral stock to obtain a final titer of 107 ffu/mL. The infectious blood meal was placed in capsules of a Hemotek® blood-feeding system (Hemotek Ltd, Blackburn, UK) at 37 °C. The engorged mosquitoes were then kept at 28 °C in 80% humidity and fed with a 10% sucrose solution until processing at 14 and 21 days post-infection (dpi). The rabbits used for preparing infectious blood meals were between 3-month and 2-year-old and maintained in an animal facility under standard conditions (23 °C and 14:10 light/dark cycle) at Institut Pasteur.Preparation of mosquito samplesSaliva was collected after removing the wings and legs of mosquitoes and inserting the proboscis into a p20 tip filled with 5 µL of FBS (Fetal Bovine Serum)55. After 30 min, the saliva-containing FBS was expelled into 45 µL of L-15 medium and stored at −80 °C until analysis. To determine infection rate (IR) and dissemination rate (DR), bodies and heads were homogenized in 300 µL of L-15 medium supplemented with 2% of FBS. After centrifugation at 10,000 rpm for 10 min, supernatants were collected for virus detection. Moreover, to study if patterns of infection, dissemination, and transmission were different in Ae. aegypti populations from the Asia-Pacific region compared to mosquitoes from YFV-endemic regions in Africa, we included mosquito populations from Cameroon (Benoué, Douala, and Yaoundé) and Congo (Brazzaville) to our dataset; African Ae. aegypti analyzed were partly processed in the previous publication of Kamgang et al.54.Virus titrationSerially diluted samples were inoculated on C6/36 cells in 96-well plates; each well was inoculated with 50 µL of diluted samples for one hour at 28 °C and after removing the inoculum, cells were covered with 150 µL of carboxymethylcellulose (CMC) supplemented with L-15 medium. After incubation at 28 °C for 5 days, cells were fixed with 3.6% formaldehyde, washed and hybridized with YFV specific primary antibody (catalog number: NB100-64510, Novusbio, CO, USA), and revealed by using a fluorescent-conjugated secondary antibody (catalog number: A-11029, Life Technologies, CA, USA), with dilution factors 1:200 and 1:1000, respectively. Foci were counted under a fluorescent microscope and titers were expressed as ffu/sample.Risk of Ae. aegypti-mediated YFV transmissionThe work by Kraemer et al28. presents worldwide estimates of the occurrence of Ae. aegypti, i.e., the probability of encountering Ae. aegypti throughout at a resolution of 5 km × 5 km. We extracted these values at the sampling points where studied mosquito populations can be found as well as averaged these at each geographical point to illustrate possible heterogeneity in mosquito occurrence.Statistical analysesIR, DR, and TR were compared among populations using Fisher’s exact test. Virus titrations were compared among populations using Kruskal–Wallis non-parametric tests. Correlations between titration in bodies, heads, and saliva were estimated. Statistical analyses were performed using the Stata software (StataCorp LP, Texas, USA). P-values < 0.05 were considered statistically significant. If necessary, the significance level of each test was adjusted based on the number of tests run, according to the sequential method of Bonferroni56. The statistical details can be found in the figure legends and the effect of geographic origin was estimated using a linear regression model.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileDescriptions of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2Reporting Summary | nature communications | [
"Article"
] | [
"Viral transmission",
"Viral infection",
"Epidemiology",
"Experimental models of disease"
] | 2016, return from Angola of 11 yellow fever (YF-infected workers to China threat YF epidemic in Asia Increasing trade travels between China Africa increase risk disease Yellow fever virus endemic to Africa South America absent in Asia reasons transmission barrier low compatibility mosquito virus limited duration low viraemia sylvatic competition with dengue Japanese encephalitis poorly explored epidemic unpredictable symptoms YF are fever headache muscle aches nausea vomiting in-hospital fatality rate 67%7,8 interest public health Traced 3000 years ago YF encountered in Africa isolated in 1927 transported via ships West Africa to West Indies during slave trade transports vectors Aedes aegypti YFV transmission understanding of YFV transmission cycle vector control strategies since YFV 17D vaccine in YFV causes 51,000–380,000 annual severe cases 19,000–180,000 fatal in Insecticide-resistance mosquito supply shortage distribution uptake of YFV vaccines main causes current burden13YFV mosquito blood infected host midgut replicates After incubation virus midgut lamina into hemocele infects salivary glands for transmission vertebrate Parameters midgut infection viral dissemination transmission through saliva determine mosquito vector competence transmission Africa South America YFV transmitted forest cycle between non-human primates (NHP zoophilic mosquitoes urban cycle YFV involves mosquito Ae. aegypti in Africa South mosquito vectors Ae. aegypti Aedes albopictus present in 154 countries half world population risk YFV transmission Ecological disturbances urban contribute proliferation Ae. aegypti supplanting Ae. albopictus in Aedes spp. mosquitoes vectors of chikungunya dengue Zika viruses in East South-East Asia suitable environment for YFV Increasing exchanges between Asia Africa raised passengers between YF-endemic increase travels between countries growth geographic spread YFV cases among Chinese travelers returning Asia after Angola during 2015–2016 YF outbreak3 threatening immunologically naive populations in Asiaaegypti Ae. albopictus mosquitoes1 Africa receives Chinese workers unvaccinated against YFV risk importing YF repeated introductions viraemic travelers immunologically naive local population risk YF emergence Asia vector competence YFV Africa South America Caribbean investigated23 limited information for Asian-Pacific mosquitoes risk YFV transmission Investigating vector competence YFV Asia-Pacific essential threat YFV transmission YF outbreaks vector competence of 18 populations Ae. aegypti Ae. albopictus Asia-Pacific Ae aegypti mosquitoes more susceptible to West-African genotype YFV mosquitoes Singapore Taiwan Thailand New Caledonia YFV at 14 days post-infection Ae. aegypti mosquitoes excrete 23,000 viral particles in saliva YFV could through saliva infected aegypti mosquitoes to YFV populations Asia-Pacific used vector competence 14 21 days post-infection 14 dpi infection rate) ranged 41.7% to 95.8% (TRUNG Vietnam CSP Thailand TAINAN Taiwan NOUMEA New Caledonia different 12 populations < 10−4) 10 Asian populations two Pacific region = 0.02) (Fig. Dissemination rate 42.8% (FENG Taiwan to 86.9% (CSP Thailand NOUMEA some higher DR P = 10 Asian populations similar DR (P = 0.13) two Pacific different DR (P = 0.04) seven 12 populations excrete virus saliva five populations ANNAN FENG TR 12.5% to 45% significantly different P < 10−4) (Fig. 1c).Fig 1Vector competence 12 Aedes aegypti populations 14 21 days after infectious blood meal 107 ffu/mL YFV-African 20–24 mosquitoes examined viral infection dissemination transmission proportion infected saliva Titrations C6/36 cells 96-well plates Viral particles detected FFA primary anti-YFV antibody secondary fluorescent-conjugated antibody Infection rate percentage blood-fedDissemination rate (DR) percentage infected head viral particles cavity midgut Transmission rate (TR) infectious saliva (viral particles crossed midgut glands. Stars indicate statistical significance comparisons Fisher’s exact test (two *P ≤ 0.05, ****P ≤ 0.0001) ****P ≤ 0.0001 *P = 0.02 0.0001 0.0001 0.0001 ns (non-significant) lack statistical significance (P > 0.05) brackets numbers mosquitoes tested dpi days post-infection Source data Supplementary Data 1 file longer incubation time vector competence protocol IR DR TR at 21 dpi IR reached 100% in four populations (BLX different between 12 populations P < 10−4) DR ranged 47.6% to 95.8% (CSP differed between populations TR ranged 10% to 56.5% (CSP no significant difference between 12 populations (Fisher’s test: P = 0.10) results show Ae.aegypti mosquitoes vectors YFV 42% populations (5/12) transmit at 14 dpi (12/12) 21 dpi.Aedes albopictus mosquitoes less to YFV than Ae. aegyptiTo Ae. albopictus native Asia local transmission YFV vector competence indices IR DR TR calculated for six populations at 14 21 dpi 14 dpi IR ranged 4.2% (THAI Thailand to 62.5% (FOSHAN differed between populations DR 0% to 85.7% (LINGYA no significant difference between populations P = 0.41) transmission observed for FOSHAN (TR = 22.2%) (Fig. 2c).Fig 2Vector competence of 6 Aedes albopictus populations assessed 14 21 days after infectious blood meal 107 ffu/mL YFV (West-African genotype).Batches mosquitoes examined for viral infection dissemination transmission proportion infected bodies head saliva Infection rate (IR) percentage infected body Dissemination rate (DR infected head Transmission rate (TR) infectious saliva particlesStars indicate significance Fisher’s test (two-sided *P ≤ 0.05 **P ≤ 0.01 ****P ≤ 0.0001) ****P ≤ 0.0001 **P = 0.003 *P = 0.038 ns-significant lack statistical significance (P > 0.05) brackets numbers mosquitoes tested post-infection Source data Supplementary Data 1 21 dpi IR ranged 8.3% (THAI Thailand to 54.2% (FOSHAN China DR 0% Laos to 100% (LINGYA TR to 66.7% (LINGYA Taiwan Four Ae. albopictus populations (YYG Japan XKH Laos THAI Thailand PMNI Brazil transmit 14 21 dpi indicate Ae albopictus populations less disseminate transmit YFV Ae. aegypti.Higher loads viral particles saliva Ae. aegypti collected mosquito saliva titrated five populations 14 dpi viral particles varied 101.6±1.5 to 103 (FENG 21 dpi 12 populations viral particles saliva 5 (VIET to 103.7±4Ae. albopictus mosquitoes 101.7±1.7 particles (FOSHAN China 14 dpi 102.2±1.4 (LINGYA Taiwan 21 dpi (Fig 3Viral loads mosquito saliva 14 21 days after infectious blood meal West-African YFV Aedes aegpti 14 21 dpi Aedes albopictus 14 21 dpi Saliva collected 30 min forced salivation legs wings tip FBS mosquito proboscis Salivas titrated on C3/36 cells particles expressed in ffu/saliva (non-significant) lack statistical significance Kruskal–Wallis test P > 0.05) Bars indicate mean brackets numbers mosquitoes tested dpi days post-infection Source data Supplementary Data 1 file.Lower dissemination YFV in Ae. aegypti Asia-Pacific African higher viral loads body head virus transmission saliva viral particles estimated for Viral loads body saliva not different head higher in mosquitoes Africa (104.6±3.7) compared Asia (103.9±3.9) Pacific (103.7±3.6) P > 0.05)loads body head saliva Aedes aegypti populations Asia Africa Pacific region.Mosquito body head saliva titrated C6/36 cells viral particles expressed ffu/sample Stars indicate significance comparisons Kruskal–Wallis test-sided **P ≤ 0.01). **P = 0.0095-significant) lack statistical significance (P > 0.05) Bars indicate mean brackets numbers mosquitoes tested Red dots samples Asia green triangles Africa blue Pacific Source data Supplementary Data 2 file.Viral loads body correlated with head (ρ = 0.31 P = 0.012) no correlation body saliva (ρ = 0.22 P = 0.11 head = 0.04 P = 0.77 difference loads mosquitoes geographic origins used linear regression model identify main factor correlation logistic regression model analysis corroborated compared mosquitoes AfricaAsia lower viral load head −0.73 log lower Asian African Pacific (−0.82 P = viral loads body saliva no difference Africa Asia Pacific P = 0.11 P = 0.54) linear regression analyses body head saliva Aedes aegypti mosquitoes 21 days infectious blood meal 107 ffu/mL coefficient testBody Africa10 Asia−0.39 (−0.83; 0.06) Pacific+0.19 (−0.48; 0.85)Head Africa10.01 Asia−0.73 (−1.21; −0.26) Pacific−0.82 (−1.53 −0.12)Saliva Africa10.54 Asia−0.35 (−0.97; 0.27) Pacific−0.10 (−1.02; 0.82)Analyses continent Source data Supplementary Data 2 file values (P ≤ 0.05) aegypti Asia-Pacific lower viral particles head similar loads body saliva viral dissemination distinguishes three continents Aedes aegypti populations risk transmission transmission efficiencies probabilities vector occurrence Kraemer al.28)aegypti Asia CSP TRUNG NANZI (Taiwan higher transmission risk YFV (CSP 54% [32.8–74.4% TRUNG 25% [9.8–48.7% NANZI 21% [7.1–42.2% mosquito occurrence high constant within 5 km radius vectors naive populations risk YFV infection (Fig. 5) 5Risk Aedes aegypti YFV transmission Asia data Kraemer et al. probability Aedes aegypti South-East Asia colors probabilities lower higher median Modeled vector occurrence mosquito transmission efficiency Aedes aegypti populations laboratory map uses data Kraemer et al. generated R v4.0.1 Source data Supplementary Data 1 file28 two studies on vector competence YFV mosquitoes Asia-Pacific 18 mosquito populations Ae. aegypti populations Asia-Pacific more transmit YFV than Ae. albopictus mosquitoes Singapore Taiwan Thailand New Caledonia highest potential transmit YFV risk transmission human populations highresults exclude possibility YF epidemic Asia-Pacific Ae. aegypti-established previous study showed Ae Laos YFV 14 days after exposure YFV S-79 infected American genotype 1 YFV (strain 74018 Brazil), Ae aegypti from Cambodia (Phnom Penh Vietnam susceptible to YFV26 lower dissemination efficiencies Ae. aegypti populations Asia-Pacific transmit YFV West-African genotype risk YF epidemics YF-free region Originally from tropical rainforests Africa YFV introduced into Americas during slave trade 14th century YFV vector Ae. aegypti2. eradication of Ae. aegypti led success YF relaxation vector control 1970s permitted Ae. aegypti recolonize species responsible for urban dengue outbreaks30 excluded from YFV cycle South America31 YFV absent except Africa America until 2016 11 YFV-infected workers Angola reported in China YF risk back Asia sylvatic cycle urban cycle 49 of 52 countries suitable for proliferation Ae. aegypti Ae.albopictus32 fertile ground for YF transmission dengue fever33 YFV-susceptible non-human primates South America absent in Macaca spp. monkeys Asia might YFV reservoir36 YFV-susceptible zoophilic mosquitoes37 infected with West-African YFV Ae. aegypti populations transmit at day 21 post-infection. higher rates dissemination transmission midgut less barrier salivary glands viral particles excreted similar to African mosquitoes Asian-Pacific Ae. aegypti mosquitoes efficient as YFV-endemic regions Africa West-African YFV 1979 African mosquitoes Cameroon Congo YFV strains from Senegal low evolutionary change Ae. aegypti from Cameroon Congo Senegal ancestral form Ae. aegypti formosus low genetic differentiation39 limits bias vector competence mosquitoes infected YFV highest for Ae. aegypti from Thailand (>50% recent modeling Ae. aegypti found South-East Asiaaegypti Asia-Pacific to YFV prone trigger YF outbreak strengthening conclusions metapopulation models probabilities YFV spread international airline transportation40 disease transmission models infection data vaccination coverage environmental assessing risk YFV transmission vector competence data reflect capacity mosquitoes field vector environmental factors shorten mosquito lifespan diminish probability infecting viral titers experimental infections differ from viremias patients 4.98 (3.50–5.79) log10 copies/mL YFV RNA in viral transmission determined by detecting viral particles in mosquito saliva forced salivation technique reflect physiological dose viral particles mosquito bite vector capacity biotic abiotic factors varies space time region influenced by population density vector feeding behavior vector lifespan44 vaccination mandatory preventing YF outbreaks controlling Ae. aegypti populations regions dengue chikungunya Zika failure preventing controlling dengue conventional insecticides early detection imported cases vaccination mandatory for travelers countries risk plan mass vaccination campaigns securing vaccine stockpile new vector control methodsWolbachia strategy improve prevention YF other arboviral diseases45 more work determine evolution viral populations after escape midgut mosquito cavity virus disseminates peripheral organs tissues Viral loads mosquito heads lower YFV-infected Ae. aegypti Asian-Pacific viral replication mosquito immune responses RNA interference important antiviral response refine mutational spectrum implications diversity viruses excreted mosquito salivary glands injected vertebrate flaviviruses endemic Asia Japanese encephalitis virus (JEV question resident virus interfere with non-resident virus YFV49 17D vaccine available since 1930s concerns safety supply YFV vaccine vulnerable YF emergence cover immediate mass vaccination-dose YF vaccination alternative to shortage full-dose work brings data mosquitoes understanding factors emergence arboviruses YF Asia-Pacific region decision statementAnimals housed Institut Pasteur animal facilities (Paris) French Ministry Agriculture experiments rodents Work compliance French European regulations care protection laboratory animals (EC Directive 2010/63 French Law 2013-118 February 6th, 2013)experiments approved Ethics Committee #89 registered APAFIS #6573-201606l412077987 v2.YFV S-79 MK060080) isolated patient Senegal 1979 passaged twice mice brains C6/36 Virus stocks mosquito infections produced C6/36 cells stored −80 °C until use populationsTwelve Ae. aegypti six Ae. albopictus populations analyzed (Table 2 Fig. 6) Mosquito eggs collected ovitraps shipped Institut Pasteur (Paris) infections egg hatching 200 larvae distributed per pan dechlorinated water yeast tablets food Larvae reared until adult stage controlled mice feeding mosquitoes 6-week 2-month-old maintained animal facility standard conditions (23 °C 14:10 light/dark cycle Institut Pasteur.Table 2Mosquito populations countries localities generation speciesCountryPopulation aegyptiCambodiaCAMBPhnom 2018Huynh 2019Marcombe)ThailandCSPChiang 2019Jupatanakul Center Genetic Engineering Biotechnology University Singapore Health Research Institute2019ANNANTainan104 Normandie N. Pasteur New Caledonia colony1994Failloux A.B.Aedes albopictusChinaFOSHANGuangdong X.G. Medical University Guangzhou colony2014Sawabe K 2019Chen C.H. Health Research Institute Taiwan 2019Marcombe S. Pasteur Laos Mai702 2019Jupatanakul N. Center Genetic Engineering Biotechnology Iguaçu82015Lourenço-de-Oliveira R Oswaldo Cruz Brazil Lab colony mosquito strain adapted laboratory 20 generations distribution 18 mosquito sample locations (12 Aedes aegypti 6 Aedes albopictus).Black Aedes aegypti red Aedes albopictus-source dot sampling location infectious blood sixty 10-day-old female adults biosafety level-3 laboratory 24 h infection blood 1.4 mL rabbit erythrocytes 10 mM adenosine triphosphate) mL viral stock titer 107 ffu/mLinfectious blood meal in capsules Hemotek® blood-feeding system at 37 °C engorged mosquitoes kept at 28 °C 80% humidity fed 10% sucrose solution until processing 14 21 days post-infection rabbits 3-month 2-year-old conditions (23 °C 14:10 light/dark cycle Institut Pasteur mosquito samplesSaliva collected wings legs proboscis p20 tip 5 μL FBS After 30 min saliva FBS expelled into 45 μL L-15 medium stored at −80 °C until analysis infection dissemination bodies heads homogenized in 300 μL L-15 medium 2% FBS centrifugation 10,000 10 min supernatants collected for virus detection infection dissemination transmission Ae. aegypti populations Asia-Pacific YFV-endemic Africa included Cameroon Congo African Ae. aegypti processed diluted samples inoculated on C6/36 cells in 96-well plates 50 μL one hour 28 °C cells covered with 150 μL carboxymethylcellulose) L-15 mediumincubation 28 °C 5 days cells fixed 3.6% formaldehyde washed hybridized YFV primary antibody NB100-64510 Novusbio CO revealed fluorescent-conjugated secondary antibody A-11029 Life Technologies CA dilution factors 1:200 1:1000 Foci counted fluorescent microscope titers expressed ffu/sample.Risk Ae. aegypti YFV Kraemer et worldwide estimates occurrence Ae. aegypti probability 5 km × 5 km extracted values sampling points mosquito averaged heterogeneity mosquito occurrence.Statistical analysesIR DR TR compared Fisher’s exact test Virus titrations compared Kruskal–Wallis tests Correlations titration bodies heads saliva estimated analyses Stata software LP Texas P-values < 0.05 statistically significant significance level adjusted number tests run sequential method Bonferroni56 statistical details figure legends effect geographic origin estimated linear regression model Nature Research Reporting Summary.Supplementary Review Additional Supplementary FilesSupplementary Data | 49 | 0.548301 |
10.1038/s41467-020-16675-x | PMC7280211 | Narrowband organic photodetectors (OPDs) are attractive for emerging applications. Here, the authors report a simple strategy to produce filter-free narrowband OPDs with outstanding performances by manipulating exciton dissociation with a hierarchical device structure. | The high binding energy and low diffusion length of photogenerated Frenkel excitons have long been viewed as major drawbacks of organic semiconductors. Therefore, bulk heterojunction structure has been widely adopted to assist exciton dissociation in organic photon-electron conversion devices. Here, we demonstrate that these intrinsically “poor” properties of Frenkel excitons, in fact, offer great opportunities to achieve self-filtering narrowband organic photodetectors with the help of a hierarchical device structure to intentionally manipulate the dissociation of Frenkel excitons. With this strategy, filter-free narrowband organic photodetector centered at 860 nm with full-width-at-half-maximum of around 50 nm, peak external quantum efficiency around 65% and peak specific detectivity over 1013 Jones are obtained, which is one the best performed no-gain type narrowband organic photodetectors ever reported and comparable to commercialized silicon photodetectors. This novel device structure along with its design concept may help create low cost and reliable narrowband organic photodetectors for practical applications. | IntroductionNarrowband photodetectors, which only detect light within a specific wavelength of interest and have no response to light of other wavelengths (usually background or environmental radiation), have been widely used in imaging, chemical analysis, and have also shown great potential for application in emerging artificial intelligence networks, such as augmented/virtual reality, advanced driver assistance systems and full-weather robots1–3. Commercially available narrowband photodetectors are made of inorganic semiconductors (mostly silicon) and integrated with optical filters4. The use of filter creates additional optical interfaces, increases the device’s architectural complexity and limits the pixel density of the detector array5. Moreover, due to the small extinction coefficient of silicon, a thick silicon film (∼3 μm) is often required to sufficiently absorb light for operation, which results in an array with a greater likelihood of interpixel cross-talk. To overcome these defects, several strategies are emerging for narrowband photodetection without filter, including: (1) using absorbers with narrowband absorption6,7; (2) intentionally enhancing the absorption in a selected wavelength range by the plasmonic effect8–10; and (3) manipulating the internal quantum efficiency via charge collection narrowing (CCN)11–13. These pioneer works provided effective strategies of making filter-free narrowband photodetectors. Nevertheless, in order to suppress the response of the background light and realize the narrowband characteristic, most of these methods have to sacrifice the original sensitivity of the photodetectors and the resulting devices’ performances are still behind those of silicon photodetectors.Among all kinds of semiconductors, organic semiconductors have high extinction coefficients and can efficiently absorb light at thickness as little as a few hundred nanometers. The absorption spectrum of organic semiconductors can be easily tuned by adjusting their molecular structures14. In addition, because organic semiconductors are light-weight, flexible, and soluble in organic solvents, they are more suitable for wide applications that require inexpensive, flexible, and moldable photodetectors15–18. However, the typical high binding energy and short diffusion length of the excitons in organic semiconductors bring some challenges for their applications in photon–electron conversion devices. Unlike its inorganic counterparts, when an organic semiconductor absorbs a photon, a localized Frenkel exciton is created due to the weak intermolecular van der Waals interaction alongside with the low dielectric constant19,20. This exciton cannot spontaneously separate into free charge due to the strong Coulombic interaction21, and the short lifespan of exciton limits its diffusion length within tens of nanometers22. Therefore, many strategies, including molecular structure modification to improve dielectric constant23–25, and matching materials with different electron affinities (ionization potentials) to provide donor–acceptor (D/A) interfaces26,27, have been adopted to assist exciton dissociation before dissipation in organic photon–electron conversion devices.Herein, we find that these intrinsic “poor” properties of Frenkel excitons actually offer organic semiconductors great opportunities to achieve filter-free narrowband organic photodetectors (OPDs). We propose a simple strategy to produce narrowband OPDs by manipulating exciton dissociation (named as “exciton dissociation narrowing” [EDN]), with a hierarchical device structure where thick larger bandgap donor layers followed by a lower bandgap acceptor layer. During the operation, excitons generated by high-energy photons in donor front layers fail to separate into free charges due to the absence of D/A interfaces, thus dissipated. Only low-energy photons with a long penetration depth can reach the D/A interfaces and produce free charges for collection. Some other strategies based on thick active layer with bulk heterojunction (BHJ) structure have also been proposed to construct narrowband photodetectors, including no-gain type11 and gain type28,29. For example, the no-gain type narrowband detectors enabled by CCN concept modulate the external quantum efficiency (EQE) spectrum by manipulating the collection efficiency of free charges at the corresponding electrodes (photogenerated excitons firstly dissociate into free charges). The gain type detectors achieve photomultiplication via charge tunneling injection from external circuit under large applied voltage. In comparison, the novel device structure and its generic methodology proposed in this study is based on hierarchical active-layer structure and novel working mechanism of EDN. It was interestingly found that this novel methodology can efficiently suppress the response outside the detection window while retaining high sensitivity in the detection region, which enabled us to produce a series of simple structure visible-blind near-infrared (NIR) narrowband OPDs with outstanding performances.ResultsBasic device structure of self-filtering narrowband OPDFigure 1a shows the chemical structures of the materials30–33 used in this study. NT812 is a home designed high charge mobility (~10−2 cm2 V−1 s−1) donor polymer which shows ideal charge transport in organic solar cells even when the thickness reach to 1 μm30,34. The most crucial part of the proposed self-filtering (SF) narrowband OPD is the hierarchical structure of a thick larger bandgap donor layer followed by a lower bandgap acceptor layer. This hierarchical structure is usually difficult to be realized because the subsequent acceptor material will penetrate into the underlying donor layer network during solution processing35,36. To this end, a combination of methods was adopted. Firstly, to prevent the diffusion of the acceptor molecules (Y6, Fig. 1a), the donor polymer NT812 was crosslinked to be a robust film (with a thickness of 650 nm) by the azide crosslinker s-4PFA, which can crosslink semiconducting polymers with a sufficiently low concentration and has negligible influence on their crucial semiconductor properties37. Following that, a thin film of NT812 (100 nm) was deposited onto crosslinked NT812 to provide an interdiffusion region with following Y6 for efficient charge separation38. Then Y6 was deposited onto NT812 as the rear-layer to obtain a complete activelayer with a total thickness of 800 nm. Here, chloroform was selected as the solvent for rear-layer processing, in which Y6 exhibits good solubility, whereas the front-layer NT812 is nearly insoluble (as shown in Supplementary Fig. 1). This orthogonal solvent and its low boiling point further help the construction of the hierarchical structure.Fig. 1Hierarchical structure and EQE response of OPDs.a Chemical structures of Y6, NT812, s-4FPA and PFN-Br. b Time-of-flight secondary ion mass spectrometry (ToF-SIMS) depth profile through the prepared film. c Reconstructed three-dimensional negative ion images (100 μm × 100 μm area) of C74H69F4N8O2S5−, SN− and their composite from a depth profile of the prepared film. d Basic device structure of self-filtering (SF) narrowband organic photodetector (OPD). e External quantum efficiency (EQE) curves of the SF-narrowband OPD and reference device under −0.1 V bias.To verify the existence of this hierarchical structure, we carried out the time-of-flight secondary ion mass spectrometry (ToF-SIMS) depth profiling measurement39. The C74H69F4N8O2S5− (fragment structure is shown in Supplementary Fig. 2) and the SN− ions were selected to track the depth distribution of Y6 and NT812, respectively. Figure 1b shows the ToF-SIMS depth profile through the prepared film. As expected, Y6 signal shows strong intensity at the first 130 nm, and begins to fall after 130 nm and tends to be zero after 168 nm. Meanwhile, the intensity of SN− ion begins to rise after 130 nm and stabilize after 168 nm, which we speculate it is due to the higher thiadiazole content within NT812 leading to higher SN− ion yield. Figure 1c shows the reconstructed three-dimensional chemical images of the C74H69F4N8O2S5− and SN− ions, which exhibit evidently that Y6 molecules are concentrated in the upper layer of the whole film with a penetration depth of 168 nm. These results demonstrate that using the methods described above, we effectively restricted the diffusion of Y6 molecules into the underlying donor layer, guarantee a complete device structure of ITO/PEDOT:PSS/donor front layer NT812/acceptor rear layer Y6/PFN-Br/Ag (as illustrated in Fig. 1d).Figure 1e presents the EQE spectrum of the fabricated photodetector under −0.1 V bias, which exhibits a dominant narrow peak at 860 nm accompanied by a weak response at 520 nm. Figure 1e also gives the EQE spectrum of the reference device with a conventional structure (ITO/PEDOT:PSS/NT812:Y6 (150 nm)/PFN-Br/Ag). EQE curves of the SF-narrowband OPD and the reference device show that, as the corresponding EQE response in visible spectrum is effectively suppressed in SF-narrowband OPD, the peak EQE at 860 nm is not significantly reduced, maintaining more than 89% of the corresponding value of reference device. By comparison, the BHJ device11 with device struture of ITO/PEDOT:PSS/NT812:Y6 (800 nm, 1:4)/PFN-Br/Ag suffers tremendous EQE loss, only obtains a peak EQE around 10% (as demonstrated in Supplementary Fig. 3).Working mechanism of self-filtering narrowband OPDsTo understand the mechanism of EDN, the distribution of photogenerated excitons in the bulk of the front-layer material NT812 was analyzed. The continuity equation for neutral excitons can be described by Eq. (1):1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{\partial n(x,t)}}{{\partial t}} = g\alpha N_0\left( t \right)e^{ - \alpha x} - \frac{{n\left( {x,t} \right)}}{\tau } + D\frac{{\partial ^2n\left( {x,t} \right)}}{{\partial x^2}} - F\left( {x - x_{{\mathrm{int}}}} \right)n\left( {x,t} \right),$$\end{document}∂n(x,t)∂t=gαN0te−αx−nx,tτ+D∂2nx,t∂x2−Fx−xintnx,t,where n(x,t) is the time-dependent exciton density, x is the penetration depth of the incident light from the transparent electrode in the bulk of the front layer, g is the internal efficiency of photon-to-exciton, N0(t) is the number of incident photons, α is the absorption coefficient of the front layer material, τ is the exciton lifetime, D is the exciton diffusion coefficient, and F(x − xint) is the exciton dissociation rate at the D/A interface (xint). Equation (1) is solved with the stationary illumination ((∂n/∂t) = 0) and boundary conditions prosed by Stübinger et al.40:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n\left( x \right) = \frac{{gN_0}}{D} \, \, \frac{{\alpha L^2}}{{1 - \left( {\alpha L} \right)^2}}\left( {e^{ - \alpha x} - e^{ - \left( {x/L} \right)}} \right),$$\end{document}nx=gN0DαL21−αL2e−αx−e−x/L,where L is the one-dimensional diffusion length, which is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L = \sqrt {D\tau } $$\end{document}L=Dτ. Equation (2) offers the exciton density distribution in the bulk of the donor front layer. Furthermore, the exciton diffusion coefficient D of the front-layer material NT812 is calculated by using the Monte Carlo simulation method41. (The specific simulation details are recorded in Supplementary Note 1). Figure 2a shows the experimentally measured time-resolved photoluminescence of the pristine and blend film with a PC61BM volume fraction of 0.05% for NT812 (open circles), and the photoluminescence decay of blend film, which was modeled with the Monte Carlo simulation, is also depicted as solid lines. The model fits the experimental data very well and yields the initial diffusion coefficient D of 19.5 × 104 cm2 s−1; the diffusion length L was then calculated as 13.3 nm. Combined with the value of α measured from the ultraviolet–visible (UV–vis) light absorption spectrum (Supplementary Fig. 4), the exciton density distribution curves of incident light of various wavelengths can be eventually obtained (Fig. 2b). From the distribution of excitons in the donor front layer, the working mechanism of the SF-narrowband OPD can be summarized as follows (as illustrated in Fig. 2c).Fig. 2Working mechanism of SF-narrowband photodetectors.a Experimentally measured time-resolved photoluminescence of pristine (green open circles) and blend film (red open circles). Photoluminescence decay modeled with the Monte Carlo simulation of blend film (blue solid lines). b Exciton density in the bulk of NT812 versus penetration depth of incident light with different wavelength from the transparent ITO electrode. c Diagrammatic illustration of the working mechanism of self-filtering (SF) narrowband photodetectors.For incident light within the absorption range of the front donor material (i.e., high-energy photons in the spectral range of 380–850 nm for NT812), the corresponding photogenerated excitons are concentrated mainly on the front side of the film (e.g., within the first 200 nm layer, as shown in Fig. 2b), where the distance between these excitons and the D/A interfaces is far beyond the exciton diffusion length. During their limited lifetime, most excitons outside the depletion region cannot diffuse to the D/A interface to dissociate into free charges; therefore, the excitons generated by this spectral range of incident photons are bound in the bulk of the front donor material and eventually relax back to the ground state42,43. Moreover, according to the Beer–Lambert law44, the intensity of incident light decreases exponentially with the penetration depth in the front layer. Thus, most of the high-energy incident photons were absorbed by the thick front layer and cannot reach the acceptor rear layer (as shown by transmittance spectrum in Supplementary Fig. 5). As a result, neither the donor nor the acceptor material can contribute to the EQE in this spectral range of incident light. (as is typically represented by 720 nm incident light in Fig. 2b, c).For incident light with a longer wavelength (longer than the absorption onset of the front donor material; i.e., λ > 850 nm for NT812), the corresponding low-energy photons can penetrate the entire front layer and be harvested by the rear acceptor material with tiny loss. The excitons generated in the acceptor material then efficiently dissociate and separate into free charges at the D/A interfaces and are collected by the corresponding electrodes, so the corresponding EQE response is obtained at the long wavelength spectrum, which we call the acceptor generation region, which is the main contribution to the narrow response peak. (As that of 860 nm incident light in Fig. 2c).However, organic optoelectronic materials generally do not exhibit uniform absorption coefficients over a wide spectral range, photons within a specific spectral region (e.g., 520 and 810 nm for NT812; see Supplementary Fig. 4) can still penetrate deeply into the bulk of the donor front layer and create a small number of excitons within the depletion region (as shown in Fig. 2b). This portion of the incident photons can also partially reach the acceptor rear layer and create excitons for dissociation due to the low absorption coefficient of the front donor material (see Supplementary Fig. 5). Thus, the weak EQE response that corresponds to these incident spectral ranges can be attributed to both the donor and acceptor layer, which we call the dual-generation region.Self-filtering narrowband OPD with double donor layersTo further suppress the EQE response within the dual-generation region (spectra range of 500–600 nm), poly(3-hexylthiophene) (P3HT)45, which provides a cascade highest occupied molecular orbital (HOMO) level with NT812 (see Fig. 3a), was selected to replace PEDOT:PSS as the SF hole transport layer (SF-HTL). P3HT was also crosslinked by s-4FPA to prevent it from being washed away by the following solution, as verified by the cross-section scanning electron microscope image (Fig. 3b). The double donor layers that containing NT812 and P3HT delivered a complete device structure of ITO/SF-HTL P3HT/donor front layer NT812/acceptor rear layer Y6/PFN-Br/Ag.Fig. 3Device performances of SF-narrowband OPD with double donor layers.a Energy level diagram of materials used to construct self-filtering (SF) narrowband organic photodetector (OPD) with double donor layers. b Cross-section scanning electron microscope image of the double donor layer structure, the scale bar identifies a length of 1 μm. c Normalized ultraviolet-visible (UV–Vis) absorption spectra of P3HT, NT812, and Y6. d External quantum efficiency (EQE) curves (at 165 Hz) of the SF-narrowband OPD with double donor layers under different voltage biases as indicated. e Current density–voltage (J–V) curves in dark and under illumination. f Specific detectivity spectra obtained from dark current density (Jd).As shown in the UV–vis light absorption spectra (Fig. 3c), the absorption peak of P3HT is around 520 nm, which just covers the dip area of the NT812 absorption spectrum. As a result, the EQE response between 500 and 600 nm was successfully wiped out, while the narrowband EQE peak at 860 nm was not affected (as shown in Fig. 3d). Here, we define two parameters, out-of-band suppression factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S = \frac{{nor.R_{{\mathrm{ref}}}(\lambda )}}{{nor.R_{{\mathrm{SF}}}(\lambda )}}$$\end{document}S=nor.Rref(λ)nor.RSF(λ) and in-band transmission factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T = \frac{{R_{{\mathrm{SF}}}(\lambda _0)}}{{R_{{\mathrm{ref}}}(\lambda _0)}}$$\end{document}T=RSF(λ0)Rref(λ0), to characterize the self-filtering property of the narrowband photodetector, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$nor.R_{{\mathrm{ref}}}(\lambda )$$\end{document}nor.Rref(λ) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$nor.R_{{\mathrm{SF}}}(\lambda )$$\end{document}nor.RSF(λ)are the normalized responsivity of the reference device and SF-narrowband OPD, respectively. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{{\mathrm{ref}}}(\lambda _0)$$\end{document}Rref(λ0) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_{{\mathrm{SF}}}(\lambda _0)$$\end{document}RSF(λ0) are the responsivity of the reference device and SF-narrowband OPD at the narrow respond peak, respectively. Results are shown in Supplementary Fig. 6. The maximum out-of-band suppression factor at visible region is over 170, while the in-band transmission factor at 860 nm is 89.2%. Moreover, the spectral selectivity of our SF-narrowband OPD is based on the manipulation of excitons rather than free charge carriers after exciton dissociation11. Due to the high binding energy and the electrical neutrality of Frenkel exciton, the effect of applied electric field on the exciton diffusion is negligible46. Consequently, the EQE spectra of SF-narrowband OPDs tends to saturate at higher applied voltages and retain spectral selectivity even under −10 V bias (Supplementary Fig. 7), which is so significant to maintain the consistency of narrowband characteristics in practical applications.In addition to the narrow spectral response under illumination, the dark current density (Jd) is an important feature that has a pronounced impact on specific detectivity (D*). As a fundamental source of electronic noise, the dark current in organic semiconductor materials is derived mainly from charge injection from the metal contacts under an applied external bias47. This injection phenomenon becomes more serious when the bandgap of the semiconductor material becomes smaller due to the reduced gaps (hence the reduced injection barrier) between the material energy levels and the electrode Fermi levels, which is one of the main factors that limits the application of narrow-bandgap organic semiconductor materials in NIR photodetectors. Figure 3e shows the current density–voltage (J–V) characteristics of the two OPDs. Compared with the single donor layer device, the Jd of the double donor layer detector is nearly three orders of magnitude lower under reverse bias, which should be ascribed to the excellent electron blocking ability48 of wide-bandgap P3HT (Fig. 3a). Meanwhile, the cascade HOMO levels of the donor layers are favorable for the transport of photogenerated holes49, and the hole mobility of SF-HTL P3HT (150 nm)/donor front layer NT812 (750 nm) was 1.02 × 10−3 cm2 V−1 s−1 as demonstrated in Supplementary Note 2 and Supplementary Fig. 8, so the desired EQE peak can be maintained50,51. With the incorporation of a SF-HTL P3HT, the obtained photodetectors demonstrated a peak specific detectivity D* of 1.2 × 1013 Jones at 860 nm under −0.1 V bias (as shown in Fig. 3f) calculated from the expression of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D^ \ast = R/\sqrt {2qJ_{\mathrm{d}}} $$\end{document}D*=R/2qJd, where R is the responsivity (see Supplementary Fig. 9), and q is the elementary charge of the electron (When the thermal noise is taken into consideration, the calculated D* slightly decreases to 9.5 × 1012 Jones at 860 nm, see the Supplementary Note 3). The specific detectivity curve on a logarithmic scale in Supplementary Fig. 10 shows that the specific detectivity at the detection peak is two orders of magnitude higher than that outside the detection window. The above calculation method assumes that the total noise of the photodetector is dominated by the shot noise in Jd. When considered other noise, such as thermal and 1/f noise, and measured the actual noise spectrum directly, a lower detectivity could be obtained according to the equation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D^ \ast = R \times \sqrt A /S_{\mathrm{n}}$$\end{document}D*=R×A/Sn, where A is the device area and Sn is the noise spectral density, yielding a peak specific detectivity D* of 2.4 × 1012 Jones at 860 nm under a bias of −0.1 V and a frequency of 165 Hz (see Supplementary Fig. 11). The EQE values under various illumination intensity is also a critical performance parameter for photodiodes, and the result for the double donor layer device is shown in Supplementary Fig. 12. These critical parameters, including pretty high peak EQE value, responsivity and D* value, make our device perform much better than previously reported no-gain type narrowband OPDs, and even comparable to commercialized silicon photodetectors (see Supplementary Fig. 13).Universality of self-filtering narrowband OPD structureThe generic device structure proposed in this study is also applicable to other common organic semiconductor materials. The position and full-width-at-half-maximum (FWHM) of the response peaks can both be tuned by simply adjusting the combination of the donor and the acceptor materials in this study. Generally, when adopt donor and acceptor materials with red-shifted absorption onsets, the response peak will simultaneously red-shift, and when the absorption spectra of donor and acceptor materials overlap more with each other, the FWHM will be narrowed. Take the device structure based on P3HT as the SF-HTL (as shown in the Fig. 4a) for an example. The selected donor materials and acceptor materials are summarized in the Fig. 4b. As the normalized EQE spectra shown in the Fig. 4c, keeping the acceptor material Y6 unchanged, when the donor material NT812 was replaced by DT-PDPP2T-TT52, whose absorption onset is further redshifted than NT812 (Supplementary Fig. 14a), the FWHM of response peak was compressed from 72 to 43 nm, and the peak position was also red shifted from 860 to 910 nm. Furthermore, based on this donor front layer, when the acceptor material was changed from Y6 to IEICO-4F53, whose absorption onset is further redshifted than Y6 (Supplementary Fig. 14b), the response peak was further red shifted to 940 nm with a FWHM of 66 nm. And as shown in Fig. 4d, both photodetectors exhibit extremely low dark current density, delivering a peak specific detectivity D* about 1013 Jones in their respective detection windows (as shown in Fig. 4e, f, corresponding responsivity are shown in Supplementary Fig. 15), demonstrating the good universality of this EDN approach.Fig. 4Universality of the SF-narrowband OPD.a Typical device structure of self-filtering (SF) narrowband organic photodetector (OPD) with double donor layers. b Chemical structures of the selected donor materials, acceptor materials and SF hole transport layer (SF-HTL) materials. c Normalized external quantum efficiency (EQE) spectra of SF-narrowband OPDs fabricated by matching different materials, all devices were measured at −0.1 V bias. d Dark current density of SF-narrowband OPDs based on DT-PDPP2T-TT/Y6 and DT-PDPP2T-TT/IEICO-4F. e Specific detectivity spectra of OPDs based on DT-PDPP2T-TT/Y6 and f DT-PDPP2T-TT/IEICO-4F, respectively.DiscussionIn conclusion, we successfully constructed narrowband OPDs by addressing the disadvantages of a large binding energy and the low diffusion length of photogenerated Frenkel excitons in organic semiconductors. By calculating the diffusion length and hence the distribution of the photogenerated excitons in the donor layer, we could intentionally manipulate the dissociation efficiency of excitons generated by various wavelengths of incident light via a hierarchical device structure. The SF-narrowband OPDs were thus achieved via the concept of EDN. This strategy avoids the undesirable sensitivity degradation accompanied with the thick BHJ. Compared with conventional thin junction device under the same operating voltage, as the response in the visible range is completely suppressed, the peak EQE in the narrow NIR detection window still retains more than 89% of the corresponding value. Moreover, the utilizing of Frenkel exciton endows the resulting detector with electrically stable spectral selectivity, thus the device can still maintain narrow response even at −10 V bias, achieving a peak EQE value around 65%. In addition, the multilayer structure improved the charge injection barrier and effectively suppressed dark current, leading to higher detectivity. As consequence, a series of visible-blind NIR narrowband OPDs with FWHM of around 50 nm, peak detectivity over 1013 Jones were demonstrated. This novel device structure along with its generic design concept may endow organic semiconductors with the ability to create filter-free narrowband OPDs for practical applications, examples include active imaging for face identification in smart phones, for augmented/virtual reality goggles and for full-weather robots. The narrow wavelength selectivity at 860, 910, and 940 nm and the corresponding performance parameters demonstrated in this work provide an “ideal” photodetector for artificial intelligent applications. It should be noted that, different from the response peak which is highly dependent on the absorption spectra of donor and acceptor, the EQE and D* are affected by the photon–electron conversion efficiency of donor and acceptor combination. Therefore, narrowband OPDs with desired response peak, high EQE and D* can be expected by selecting existing donor and acceptor materials, or designing new materials (needs to co-work with material scientist).MethodsMaterialsP3HT, DT-PDPP2T-TT, IEICO-4F, Y6, and all reagents were purchased from commercial sources (1-Material and Solarmer Materials Inc.) and used without further purification. The polymer NT812 and the crosslinker s-4FPA were synthesized in-house following established procedures in the refs. 30,37.Device fabricationITO-coated glass was used as the substrate. Before device fabrication, the substrates were thoroughly cleaned by sequentially sonication with acetone, isopropanol, detergent, de-ionized water, isopropanol, and subsequently dried in a baking oven over night. After that, the substrates were treated by oxygen plasma for 4 min. For single donor layer device, PEDOT:PSS (P VP Al 4083) of 30 nm was first spun onto the substrates and then annealed at 150 °C on a hot plate for 20 min in air to remove the residual water. NT812 mixed with 10 wt% s-4FPA was dissolved in CB solvent, the blend of NT812 and s-4FPA was spun onto the PEDOT:PSS, annealed at 90 °C for 5 min in the glove box with nitrogen atmosphere, photo-exposed to deep-ultraviolet light (DUV, 254 nm wavelength), a crosslinked film of about 650 nm was then obtained. NT812 was dissolved in CB solution, and then spun onto the crosslinked film to obtain a thin film of about 100 nm. Y6 was dissolved in CF solvent, and then spun onto the donor front layer to obtain a complete activelayer with a total thickness of 800 nm. After spin-coating of 8 nm PFN-Br as cathode interface, a 100 nm Ag layer was sequentially deposited by thermal evaporation through a shadow mask in a vacuum chamber at a pressure of 4 × 10−7 torr. For double donor layer devices based on NT812/Y6, P3HT mixed with 12 wt% s-4FPA was dissolved in CB solvent, the blend of P3HT and s-4FPA was spun onto the ITO substrates, annealed at 90 °C for 5 min in the glove box with nitrogen atmosphere, photo-exposed to deep-ultraviolet light (DUV, 254 nm wavelength), and then the film was developed with CF to obtain the SF-HTL with a thickness of 150 nm. The following fabrication steps were the same as in single donor layer devices. For double donor layer devices based on DT-PDPP2T-TT/Y6, DT-PDPP2T-T mixed with 10 wt% s-4FPA was dissolved in CF solvent, the blend of DT-PDPP2T-T and s-4FPA was spun onto the SF-HTL P3HT (200 nm), photo-exposed to deep-ultraviolet light (DUV, 254 nm wavelength), a crosslinked DT-PDPP2T-T film of about 1400 nm was then obtained. DT-PDPP2T-T was dissolved in CF solution, and then spun onto the crosslinked film to obtain a thin film of about 100 nm. Y6 was dissolved in CF solvent, and then spun onto the donor front layer to obtain a complete activelayer with a total thickness of 1800 nm. For double donor layer device based on DT-PDPP2T-TT/IEICO-4F, the fabrication methods of 200 nm SF-HTL P3HT/1400 nm crosslinked DT-PDPP2T-T/150 nm DT-PDPP2T-T were the same as DT-PDPP2T-TT/Y6-based device, IEICO-4F was dissolved in CF, then spun onto the donor front layer to obtain a complete activelayer with a total thickness of 2000 nm. The following fabrication steps were the same as in single donor layer devices. The device active area was 0.0516 cm2.Device characterizationThe thickness of the thin films was determined by a Dektak 150 surface profiler. The EQE spectrum was measured on a commercial measurement system DSR100UV-B (Zolix Instruments Co., Ltd.) equipped with DC module. The light intensity at each wavelength was calibrated using a standard single crystal Si photovoltaic cell before the testing. The light frequency was 165 Hz. The light intensity was 2.4 μw cm−2 at 860 nm, 2.8 μw cm−2 at 910 nm and 3.1 μw cm−2 at 940 nm, respectively. The dark current density–voltage characteristics of the devices were recorded on a Keithley 2450 source-meter in an electrically and optically shielded box. The noise spectral density characteristics of the devices were recorded by a semiconductor parameter analyzer (Platform Design Automation, Inc. FS380 Pro). The frequency dependent measurements were carried out using a light-emitting diode (860 nm) modulated by the function generator (Aim-TTi TG120) as the excitation source. Square waves with different frequencies were applied. The photocurrent response of the photodiode was recorded using a digital storage oscilloscope (Tektronix TDS3052B).ToF-SIMS depth profilingA ToF-SIMS 5–100 (ION-TOF GmbH, Germany) instrument was used to acquire depth profiles from the prepared film. The instrument was equipped with a Bi/Mn liquid metal ion gun (LMIG) and an argon gas cluster ion gun, which were operated in the dual beam mode. For depth profiling, a 2.5 keV Ar-Cluster beam was used to sputter through the film at an area of 300 μm × 300 μm in 5 s intervals. A 25 keV Bi3+ analysis beam was used to analyze the central area between sputtering pulses over a 100 μm × 100 μm area inside the crater. The negative polarity data were used for sample analysis. The three-dimensional secondary ion images were reconstructed from the depth profile data.Time-resolved photoluminescence (TRPL)The TRPL measurements were carried out by time-correlated single photon counting (TCSPC) technique: Acton SP2150i spectrometer (Princeton instrument) equipped with a photomultiplier (PMA 182-N-M) were used to collect the photons, an event timer (HydraHarp-400 TCSPC) with 2 ps time resolution was used to measure the fluorescence decays, a 670 nm pulse laser was generated by the 80 MHz femtosecond laser with full width at half maximum about 120 fs, and the decay time fitting procedure was realized by the software FluoFit applying deconvolution with the instrument response function (IRF).UV–Vis absorption and transmittance spectraUV–Vis absorption spectra were acquired on a Shimadzu UV3600 spectrophotometer. Transmittance spectra were measured on an HP 8453E spectrophotometer.Scanning electron microscopeThe scanning electron microscope images were obtained with a Hitachi S-4800 FESEM.Supplementary information
Supplementary Information
Peer Review File | nature communications | [
"Article"
] | [
"Sensors and biosensors",
"Optical sensors"
] | IntroductionNarrowband photodetectors detect light specific wavelength no response to other wavelengths background used in imaging chemical analysis potential for artificial intelligence networks augmented/virtual reality advanced driver assistance systems full-weather Commercially narrowband photodetectors of inorganic semiconductors (mostly silicon integrated with optical filters4 filter creates optical interfaces increases architectural complexity limits pixel density small extinction coefficient silicon thick silicon film (∼3 μm required to absorb light interpixel cross-talk overcome defects strategies for narrowband photodetection without filter using absorbers with narrowband absorption6,7 enhancing absorption by plasmonic manipulating internal quantum efficiency via charge collection narrowing strategies filter-free narrowband photodetectors suppress response background light narrowband characteristic sacrifice original sensitivity performances behind silicon photodetectors organic semiconductors high extinction coefficients absorb light at few hundred nanometers absorption spectrum tuned by adjusting molecular structures14organic semiconductors light-weight flexible soluble in organic solvents suitable for inexpensive moldable high energy short diffusion length excitons semiconductors bring challenges photon–electron conversion devices organic semiconductor absorbs photon localized Frenkel exciton created weak intermolecular van der Waals interaction low dielectric constant19 separate into free charge strong Coulombic short lifespan limits diffusion length tens nanometers22 strategies molecular structure modification dielectric matching materials with different electron affinities adopted assist exciton dissociation before dissipation in properties of Frenkel excitons offer opportunities filter-free narrowband organic photodetectors propose strategy produce narrowband OPDs manipulating exciton dissociation narrowing” hierarchical device structure bandgap donor layers lower bandgap acceptor layer excitons high-energy photons fail separate into free charges D/A interfaces low-energy photons long penetration depth reach D/A interfaces produce free charges for collectionstrategies thick active layer bulk heterojunction structure proposed narrowband photodetectors no-gain type11 gain type28,29 no-gain narrowband detectors modulate external quantum efficiency spectrum collection efficiency free charges electrodes gain type detectors achieve photomultiplication via charge tunneling injection external circuit large voltage novel device structure methodology study based hierarchical active-layer structure novel working mechanism EDN methodology response outside detection window high in detection region simple structure visible-blind near-infrared) narrowband OPDs with outstanding performances device structure self-filtering narrowband OPDFigure 1a shows chemical structures used NT812 high charge mobility donor polymer shows ideal charge transport in organic solar cells 1 μm30 crucial proposed self-filtering narrowband OPD hierarchical structure thick larger bandgap donor layer lower bandgap acceptor layer structure difficult acceptor material into donor layer during solution processing35 combination methods adopted prevent diffusion acceptor moleculesdonor polymer NT812 crosslinked robust film 650 nm azide crosslinker s-4PFA low negligible influence semiconductor thin film NT812 (100 nm deposited crosslinked NT812 interdiffusion region Y6 charge Y6 deposited NT812 rear-layer activelayer thickness 800 nm chloroform solvent rear-layer processing Y6 good solubility front-layer NT812 nearly insoluble Supplementary Fig. 1) orthogonal solvent low boiling point hierarchical structure. 1Hierarchical structure EQE response OPDs Chemical structures Y6 NT812 s-4FPA PFN-Br Time-flight spectrometry profile prepared film Reconstructed three-dimensional negative ion images C74H69F4N8O2S5− SN− composite depth profile film device structure self-filtering narrowband organic photodetector External quantum efficiency (EQE) curves SF-narrowband OPD reference device under −0.1 V bias hierarchical structure depth profiling C74H69F4N8O2S5− (fragment structure Supplementary Fig.SN− ions selected track depth distribution Y6 NT812 Figure 1b ToF-SIMS depth profile film Y6 signal strong intensity first 130 nm after zero after 168 intensity SN− ion after 130 after 168 due higher thiadiazole content NT812 higher SN− ion yield Figure 1c chemical images C74H69F4N8O2S5− SN− ions Y6 molecules concentrated upper layer film penetration depth 168 nm restricted diffusion Y6 donor layer complete device structure ITO/PEDOT:PSS/donor front layer NT812/acceptor rear layer Y6/PFN-Br/Ag Fig. 1d).Figure 1e EQE spectrum photodetector under −0.1 V bias narrow peak at 860 nm weak response at 520 nm Figure EQE spectrum reference device conventional structure (ITO/PEDOT:PSS/NT812:Y6 (150 nm)/PFN-Br/Ag). EQE curves OPD show response suppressed SF OPD peak EQE at 860 nm not reduced 89% value reference deviceBHJ ITO/PEDOT:PSS/NT812:Y6 (800 nm, 1:4)/PFN-Br/Ag suffers EQE loss peak EQE 10% Supplementary Fig. 3) self-filtering narrowband distribution photogenerated excitons front-layer material NT812 analyzed continuity equation neutral excitons Eq.[12pt{amsmath\oddsidemargin-69pt\partial n(x,t t}} = g\alpha N x\left + D\partial ^2n\left {x x^2}} F\left {x - x{document}∂n(x,t)∂t=gαN0te−αx−nx,tτ+D∂2nx,t∂x2−Fx−xintnx,t n(x time-dependent exciton density x penetration depth incident light transparent electrode g internal efficiency photon-to-exciton N0(t number incident photons α absorption coefficient front layer material τ exciton lifetime D exciton diffusion coefficient F(x − xint) exciton dissociation rate D/A interface Equation (1) solved stationary illumination ((∂n/∂t) = 0) conditions Stübinger et al.40:2[12pt]{minimal{amsmath\oddsidemargin-69pt}{document}\left( x \right) =\frac{{gN_0}}{D}\alpha L^2}}{{1 - \left\alpha L \right^2}}\alpha x} - \left( {x/L} \right)}}\end{document}nx=gN0DαL21−αL2e−αx−e−x/L L one-dimensional diffusion length defined[12pt]{minimal}{amsmath\oddsidemargin{-69pt}{document}$$L = \sqrt {D\tau\end{document}L=Dτ Equation (2) exciton density distribution donor front layer exciton diffusion coefficient D front-layer material NT812 calculated Monte Carlo simulation method41 specific simulation details Supplementary Note 1)Figure 2a shows time-resolved photoluminescence pristine blend film PC61BM volume fraction 0.05% NT812 photoluminescence decay blend film modeled Monte Carlo simulation solid lines model fits data yields initial diffusion coefficient D 19.5 × 104 cm2 s−1 diffusion length L calculated 13.3 nm value α light absorption spectrum exciton density distribution curves incident light wavelengths obtained (Fig. 2b). front working mechanism SF-narrowband OPD summarized Fig. 2c).Fig. 2Working mechanism SF-narrowband photodetectors-resolved photoluminescence pristine blend film decay Monte Carlo simulation blend film Exciton density NT812 penetration incident light different wavelength transparent ITO electrode illustration mechanism self-filtering (SF) narrowband photodetectors incident light absorption range donor material high-energy photons 380–850 nm NT812) photogenerated excitons concentrated front side film first 200 nm layer Fig. distance excitons D/A interfaces beyond exciton diffusion lengthlimited lifetime excitons outside depletion region D/A interface into free charges bound in front donor material relax to ground Beer–Lambert intensity incident light decreases with penetration depth front layer high-energy photons absorbed by front layer reach acceptor rear layer Fig. 5) neither donor nor acceptor material contribute to EQE in represented by 720 nm incident light in Fig. 2b, longer wavelength > 850 nm for NT812) low-energy photons penetrate front layer harvested by rear acceptor material with loss excitons acceptor material dissociate separate into free charges at D/A interfaces collected by electrodes EQE response obtained at long wavelength spectrum acceptor generation region main contribution to narrow response peak 860 nm incident light in Fig. organic optoelectronic materials uniform absorption coefficients over spectral range photons within specific spectral region 520 and 810 nm for NT812 can penetrate into donor front layer create small number excitons within depletion region Fig.incident photons reach acceptor rear layer create dissociation low absorption coefficient front donor material Fig. 5) weak EQE response donor acceptor layer dual-generation region.Self-filtering narrowband OPD double donor suppress EQE response 500–600 poly(3-hexylthiophene) (P3HT)45 NT812 PEDOT:PSS SF hole transport layer P3HT crosslinked by s-4FPA double donor layers NT812 P3HT device structure ITO/SF-HTL P3HT/donor front layer NT812/acceptor rear layer Y6/PFN-Br/Ag.Fig. 3Device performances SF-narrowband OPD double donor layers Energy level diagram materials-filtering double donor layers Cross-section scanning electron microscope image double donor layer structure scale bar length 1 μm Normalized ultraviolet absorption spectra of P3HT, NT812 Y6 External quantum efficiency (EQE) curves 165 Hz SF-narrowband OPD double layers under voltage biases Current density–voltage (J–V) curves dark under illuminationdetectivity spectra from dark current density UV–vis light absorption spectra (Fig. absorption peak P3HT around 520 nm covers dip area NT812 absorption spectrum EQE response between 500 and 600 nm wiped out narrowband EQE peak at 860 nm not affected Fig. 3d). define two parameters out-of-band suppression factor[12pt{minimal{amsmath\oddsidemargin-69pt}{document} =\frac{{nor.R{ref{document}S=nor.Rref(λ)nor.RSF(λ) in-band transmission factor[12pt]{minimal{amsmath}{wasysym}{mathrsfs{upgreek}\oddsidemargin{-69pt}{document}$$T =\frac{{R\mathrm{SF}}}\lambda{ref{document}T=RSF(λ0)Rref(λ0) self-filtering property narrowband photodetector[12pt]{minimal}{amsmath}{wasysym{upgreek}\oddsidemargin}{-69pt}{document}$nor.R_{{\mathrm{ref}}}\lambda\end{document}nor.Rref(λ)[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$nor.R_{{\mathrm{SF}}}\lambda\end{document}nor.RSF(λ normalized responsivity reference device SF-narrowband OPD[12pt]{minimal{amsmath{wasysym\oddsidemargin-69pt}}$$R\mathrm{ref}}}{document}Rref(λ0)[12pt]{minimal}{amsmath{upgreek-69pt}\mathrm{SF}}} _0){document}RSF(λ0) responsivity reference device SF-narrowband OPD at narrow peak Results in Supplementary Fig. 6. maximum out-of-band suppression factor at visible region over 170 in-band transmission factor at 860 nm 89.2% spectral selectivity SF-narrowband OPD based on manipulation excitons free charge carriers exciton high binding energy electrical neutrality Frenkel exciton effect applied electric field on exciton diffusion negligible46 EQE spectra SF-narrowband OPDs saturate at higher voltages retain spectral selectivity under −10 V bias Fig. 7) significant consistency narrowband characteristicsnarrow spectral response illumination dark current density (Jd) detectivity electronic noise dark current organic semiconductor materials from charge injection metal contacts external injection serious bandgap smaller reduced gaps between material energy levels electrode Fermi levels application narrow-bandgap semiconductor materials in NIR photodetectors Figure 3e current density–voltage (J–V) characteristics two OPDs Jd double donor layer detector three orders lower under reverse bias excellent electron blocking wide-bandgap P3HT cascade HOMO levels donor layers favorable for transport photogenerated hole mobility of SF-HTL P3HT (150 nm)/donor front layer NT812 (750 nm) 1.02 × 10−3 cm2 V−1 s−1 Note 2 Fig 8 desired EQE peak SF-HTL P3HT photodetectors peak detectivity D* 1.2 × 1013 Jones at 860 nm under −0.1 V bias Fig3f calculated\documentclass[12pt]{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}$D = R/\sqrt {2qJ_\mathrm{d}}}\end{document}D*=R/2qJd R responsivity Supplementary Fig. 9) q elementary charge electron thermal noise calculated D* decreases 9.5 × 1012 Jones at 860 nm Supplementary Note 3) detectivity curve Supplementary Fig. 10 shows detection peak two orders magnitude higher outside detection window assumes total noise photodetector dominated shot noise Jd.considered noise thermal 1/f noise measured spectrum lower detectivity obtained equation\documentclass[12pt{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}{document$D = R A /S\mathrm}D*=R×A/Sn A device area Sn noise spectral density peak detectivity D* 2.4 × 1012 Jones at 860 nm bias −0.1 V frequency 165 Hz Supplementary Fig. 11). EQE values illumination intensity critical performance parameter for photodiodes result double donor layer device shown Supplementary Fig. 12. parameters high peak EQE value responsivity D* value device perform better than no-gain type narrowband OPDs comparable to silicon photodetectors Supplementary Fig. 13).Universality self-filtering narrowband OPD device structure applicable to organic semiconductor materials position response peaks tuned adjusting combination donor acceptor materials red-shifted absorption onsets response peak red-shift absorption spectra overlap FWHM narroweddevice structure P3HT SF-HTL Fig. 4a donor acceptor materials Fig. 4b normalized EQE spectra Fig. 4c acceptor material Y6 NT812 replaced by DT-PDPP2T-TT52 FWHM response peak compressed from 72 to 43 nm peak position shifted from 860 to 910 nm acceptor material changed from Y6 to IEICO-4F53 response peak shifted to 940 nm FWHM 66 nm Fig. 4d photodetectors low dark current density peak detectivity 1013 Jones detection windows Fig. 4e Fig. universality EDN approach.Fig. 4Universality SF-narrowband OPD device structure self-filtering (SF) narrowband photodetector double donor layers Chemical structures of donor materials acceptor materials SF transport layer materials Normalized external quantum efficiency (EQE) spectra of SF-narrowband OPDs measured at −0.1 V bias Dark current density of SF OPDs DT-PDPP2T-TT/Y6 DT-PDPP2T-TT/IEICO-4Fdetectivity spectra OPDs based DT-PDPP2T-TT/Y6 DT-PDPP2T-TT/IEICO-4F constructed narrowband OPDs large binding energy low diffusion length photogenerated Frenkel excitons in organic semiconductors calculating diffusion length distribution dissociation efficiency hierarchical device structure SF-narrowband OPDs achieved via EDN avoids degradation thick BHJ thin junction device peak EQE narrow NIR detection window retains 89% value Frenkel exciton endows detector electrically stable spectral selectivity narrow response at −10 V bias peak EQE value around 65% multilayer structure improved charge injection barrier suppressed dark current higher detectivity visible-blind NIR narrowband OPDs with FWHM 50 nm peak detectivity over 1013 Jones demonstrated novel device structure design concept organic semiconductors create filter-free narrowband OPDs for applications face identification augmented/virtual reality goggles full-weather robots narrow wavelength selectivity at 860, 910 940 nm performance parameters provide “ideal” photodetector for artificial intelligent applications.response peak dependent on absorption spectra donor acceptor EQE D* affected by photon–electron conversion efficiency donor acceptor narrowband OPDs response high EQE D* expected selecting existing donor acceptor materials or designing new materials co-work scientist).MethodsMaterialsP3HT-PDPP2T-TT IEICO-4F Y6 reagents purchased from commercial sources (1-Material Solarmer Materials without purification polymer NT812 crosslinker s-4FPA synthesized in-house-coated glass substrate substrates cleaned acetone isopropanol de-ionized water dried baking oven treated oxygen plasma 4 min single donor layer device PEDOT:PSS 30 nm spun substrates annealed at 150 °C 20 min NT812 mixed with 10 wt% s-4FPA dissolved in CB solvent spun onto PEDOT:PSS annealed 90 °C 5 min photo-exposed to deep-ultraviolet light crosslinked film 650 nm obtained NT812 dissolved in CB solution spun onto crosslinked film thin film 100 nmY6 dissolved in CF solvent spun donor front layer activelayer thickness 800 nm spin-coating 8 nm PFN-Br 100 nm Ag layer deposited thermal evaporation vacuum chamber pressure 4 × 10−7 torr double donor layer devices NT812/Y6 P3HT 12 wt% s-4FPA dissolved CB solvent spun ITO substrates annealed 90 °C 5 min photo-exposed deep-ultraviolet light film developed with CF SF-HTL thickness 150 nm fabrication steps same single donor layer double donor layer devices DT-PDPP2T-TT/Y6-T 10 wt% s-4FPA dissolved CF solvent spun SF-HTL P3HT (200 photo-exposed-ultraviolet light crosslinked DT-PDPP2T-T film 1400 nm dissolved in CF solution spun crosslinked film thin film 100 nm Y6 dissolved CF solvent spun donor front layer complete activelayer thickness 1800 nmdouble donor layer device DT-PDPP2T-TT/IEICO-4F fabrication methods 200 nm SF-HTL P3HT/1400 nm DT-PDPP2T-T/150 nm same IEICO-4F dissolved CF spun donor front layer activelayer thickness 2000 nm fabrication steps same single donor layer active area 0.0516 thickness thin films determined Dektak 150 surface profiler EQE spectrum measured DSR100UV-B (Zolix Instruments DC module light intensity wavelength calibrated standard single crystal Si photovoltaic cell light frequency 165 Hz intensity 2.4 μw cm−2 860 nm μw 910 nm 3.1 μw 940 nm dark current density–voltage characteristics recorded Keithley 2450 source-meter noise spectral density recorded semiconductor parameter analyzer frequency measurements light-emitting diode (860 nm function generator (Aim-TTi TG120) Square waves different frequencies photocurrent response recorded digital storage oscilloscope TDS3052B).ToF-SIMS depth-SIMS 5–100 instrument depth profiles filminstrument equipped Bi/Mn liquid metal ion gun argon gas cluster ion gun dual beam mode depth profiling 2.5 keV Ar-Cluster beam film 300 μm × 300 μm 5 s intervals 25 keV Bi3+ analysis beam area sputtering pulses 100 μm × crater negative polarity data sample analysis three-dimensional secondary ion images reconstructed from profile data-resolved photoluminescence (TRPL measurements time-correlated single photon counting technique Acton SP2150i spectrometer photomultiplier (PMA 182-N-M) photons event timer (HydraHarp-400 TCSPC) 2 ps time resolution fluorescence decays 670 nm pulse laser 80 MHz femtosecond laser decay time fitting software FluoFit absorption transmittance Shimadzu UV3600 spectrophotometer Transmittance HP 8453E spectrophotometer.Scanning electron Hitachi S-4800 FESEM.Supplementary information | 49.4 | 0.638811 |
10.1038/s41467-020-20878-7 | PMC7838200 | Selective hydrogenation of 5-(hydroxymethyl)furfural (HMF) to 5-Methylfurfural using H2 as reductant is very attractive, but remains challenging. Here, the authors report that isolated single atomic catalysts can catalyze the reaction efficiently with selectivity >99% at complete conversion of HMF. | 5-Methylfurfural (MF) is a very useful chemical. Selective hydrogenation of biomass platform molecule 5-(hydroxymethyl)furfural (HMF) to MF using H2 as the reducing agent is very attractive, but challenging because hydrogenation of C=O bond in HMF is more favourable than C–OH both kinetically and thermodynamically, and this route has not been realized. In this work, we prepare isolated single atomic catalysts (SACs) Pt1/Nb2O5-Ov, Pd1/Nb2O5-Ov, and Au1/Nb2O5-Ov, in which single metal atoms are supported on oxygen defective Nb2O5 (Nb2O5-Ov). It is discovered that the SACs can efficiently catalyze the hydrogenation of HMF to MF using H2 as the reducing agent with MF selectivity of >99% at complete conversion, while the selectivities of the metal nanocatalysts supported on Nb2O5 are very poor. A combination of experimental and density function theory (DFT) studies show that the unique features of the SACs for the reaction result from the cooperation of the Nb and Pt sites near the interface in the Pt1/Nb2O5-Ov. The Pt atoms are responsible for the activation of H2 and the Nb sites activate C-OH in the reaction. This work opens the way for producing MF by direct hydrogenation of biomass-derived HMF using H2 as the reductant. | IntroductionSelective hydrogenation is a critical class of reactions, and selectivity is an essential parameter in a chemical reaction. There are often different kinds of unsaturated groups in a compound such as C=C, C=O, C≡C, C=N, aromatic ring, and -NO2. Exploration of the methods to reduce one or some functional groups selectively while retaining others not only can synthesize many value-added chemicals but also can broaden chemical knowledge.5-(Hydroxymethyl)furfural (HMF) is an important biomass platform compound and it can be produced from cellulose that covers about 40% of lignocellulosic biomass1,2. The reduction of HMF is a very promising route to produce high-value chemicals. It has been reported that various products such as 2,5-bis(hydroxymethyl)furan (DHMF), 2,5-dimethylfuran (DMF), 2,5-dimethyltetrahydrofuran (DMTHF), 2,5-hexanediole (HDO) could be produced by the selective hydrogenation of HMF3–10. 5-Methylfurfural (MF) is a very useful chemical, which can be used as a food additive and commonly used synthetic intermediate11–14. However, the selective hydrogenation of the C–OH group in HMF to 5-methylfurfural (MF) has not been realized using H2 as the reducing agent, since in general the C=O bond is easier reduced than the C–OH bond in the hydrogenation of HMF from both kinetic and thermodynamic aspects15. Kinetically, hydrogenation of C=O bond in HMF is more favorable than the C–OH. Thermodynamically, MF has a strong tendency to be further hydrogenated to DMF. Ankur Bordoloi et al. studied the three reducible bonds in HMF (C=O, C=C, and C–OH) and found that C=O is the most prone to hydrogenation because of the strongest electrophilicity of aldehyde carbon. Furthermore, the activation energy required by the hydrogenation of C=O is much lower than that required by the dehydroxylation of C–OH16.Synthesis of MF from HMF was reported by indirect routes, which involved in the transformation of HMF to halogen compounds, such as 5-chloromethylfurfural (CMF) or 5-iodomethylfurfural (IMF), and MF was then prepared by the hydrodehalogenation of CMF or IMF (Fig. 1)17–19. The selective hydrogenolysis of HMF to MF was also reported using HCOOH as the reducing agent and the process proceeded via an esterification procedure. Unfortunately, the author found that no MF was detected when H2 was used20. Obviously, it is very attractive to prepare MF directly from the selective hydrogenation of HMF using H2 as the reducing agent because the process is simple, and the only byproduct is H2O.Fig. 1The synthetic routes to produce MF from HMF.The previous methods to synthesize MF with acid. This work to directly synthesize MF without any additives.Design of ideal catalysts is the key to achieve the selective hydrogenation of HMF to MF. In recent years, isolated single atomic catalysts (SACs) have been used in many reactions. For example, Pt1/N-C SACs showed high chemo- and regioselectivity towards terminal alkynes in hydrogenation21. An atomically dispersed copper (Cu) catalyst supported on defective nanodiamond graphene exhibited excellent catalytic performance for the selective conversion of acetylene to ethylene22. The Pt1/α-MoC catalyst showed promising activity and a strong chemospecificity towards the hydrogenation of nitro groups23. SACs have also been used to catalyze other important reactions, such as selective hydrogenation of nitroarenes, alkenes and carbonyl compounds, the catalytic transformation of methane, aqueous reforming of methanol, hydroformylation of olefins, olefin metathesis, and oxygen reduction, in which they showed excellent performance24–31.Nb2O5 has been used in the activation and cleavage of Caliphatic–O and Caromatic–O bonds in both lignin and its model compounds32–34. In this work, we fabricated Pt1/Nb2O5-Ov, Pd1/Nb2O5-Ov, and Au1/Nb2O5-Ov catalysts, in which single atomic metal sites were supported on oxygen defective Nb2O5 (Nb2O5-Ov). Very interestingly, it was discovered that they could efficiently catalyze selective hydrogenation of HMF to MF with >99% selectivity at complete conversion. The density function theory (DFT) calculations and experimental results indicated that the unusual feature of the catalysts for the reaction resulted from the excellent cooperation of the Pt and Nb sites. The Pt sites were responsible for the activation of H2 and the Nb sites near Pt sites selectively activated the C–OH group in HMF, and thus very high selectivity was achieved.ResultsStructural characterizationsThe Nb2O5 with oxygen-vacancy defects (Nb2O5-Ov) were prepared by thermal treating Nb2O5 in reducing atmosphere for 4 h at 500 °C (reducing gas contains 10 vol.% hydrogen and 90 vol.% argon). Defects of oxygen vacancies were detected by electron-paramagnetic resonance (EPR) measurement (Supplementary Fig. 1). A signal of oxygen vacancy at a g value of 2.003 was observed for Nb2O5-Ov and Pt1/Nb2O5-Ov, while no detectable EPR signal was observed for the intrinsic Nb2O5. In addition, the oxygen vacancy concentration of Pt1/Nb2O5-Ov was the highest. The transmission electron microscopy (TEM) image of the prepared Pt1/Nb2O5-Ov is shown in Fig. 2a. No Pt cluster or particle can be observed on the surface of Nb2O5-Ov, while the mapping images suggest the uniform distributions of Nb, O, and Pt over the entire architecture (Fig. 2b). The aberration-correction high-angle annular dark-field scanning transmission electron microscopy (AC–HAADF–STEM) image exhibited some marked bright points on the surface of Nb2O5-Ov, indicating the atomically dispersed Pt on the defective support (Fig. 2c). The content of Pt in Pt1/Nb2O5-Ov was 0.039 wt%, which was analyzed by inductively coupled plasma (ICP) mass spectrometry. Fourier-transformed k2-weighted extended X-ray absorption fine structure (EXAFS) in R space was performed to elucidate the coordination environments of Pt atoms anchored on Nb2O5-Ov. It was shown that there was only one notable peak at 1.5 Å from the Pt–O contribution, and no peak at 2.5 Å from the Pt-Pt contribution, confirming the sole presence of dispersed single Pt atoms in Pt1/Nb2O5-Ov (Fig. 2d). The least-squares EXAFS fitting curves of Pt1/Nb2O5-Ov and Pt foil and PtO2 are shown in Supplementary Fig. 2, and the corresponding structure parameters are listed in Supplementary Table 1.Fig. 2The structural characterization of Pt1/Nb2O5-Ov.a TEM images of Pt1/Nb2O5-Ov. b EDS mapping images of Pt1/Nb2O5-Ov. c AC–HAADF–STEM image of Pt1/Nb2O5-Ov. The atomically dispersed Pt atoms are highlighted by red circles. d The k2-weighted Fourier transform spectra of Pt1/Nb2O5-Ov and Pt foil.Pt species in Pt1/Nb2O5-Ov were partially positively charged as evidenced by X-ray absorption near-edge structure (XANES) of Pt1/Nb2O5-Ov sample between Pt foil and PtO2 (Fig. 3a). X-ray photoelectron spectroscopy (XPS) was used to study the valence states of Nb and O. As shown in Fig. 3b, Nb 3d5/2 binding energy of intrinsic Nb2O5 was 207.3 eV, which shifted to 207.1 eV for Pt1/Nb2O5-Ov. O 1s binding energy of intrinsic Nb2O5 was 530.3 eV, which shifted to 530.1 eV for Pt1/Nb2O5-Ov (Fig. 3c). The Pt 4f peak for the Pt1/Nb2O5-Ov sample can be deconvoluted into two peaks (Supplementary Fig. 3). The peaks at 76.1 and 72.9 eV correspond to the Ptδ+4f5/2 and Ptδ+4f7/2, respectively35–37. These results imply that there is a strong electron interaction between Nb2O5-Ov and Pt, and the electron transfers from Pt to the Nb2O5-Ov support.Fig. 3The electron state characterization of Pt1/Nb2O5-Ov.a The XANES spectra at the Pt L3-edge. b The XPS spectra of Nb 3d and c O1s of Pt1/Nb2O5-Ov and Nb2O5.Catalytic performance for HMF hydrogenationThe performance of Pt1/Nb2O5-Ov catalytic systems was studied for the selective hydrogenation of HMF and the results are shown in Table 1. The reaction could not occur in the absence of any catalyst (Table 1, entry 1). Very interestingly, Pt1/Nb2O5-Ov could efficiently catalyze the selective hydrogenation of HMF to MF with >99% selectivity at complete conversion of HMF in 4 h. The turnover frequency of producing MF could reach 1875 h−1 (Table 1, entry 2). The effects of different reaction temperatures and H2 pressures on the performance of Pt1/Nb2O5-Ov catalysts were studied. When the temperature changes from 160 °C to 120 °C, H2 pressure changes from 4.0 MPa to 1.0 MPa, the selectivities to MF are always >99% (Table 1, entry 3–6). These results indicate that the selectivity of MF was independent of H2 pressure and reaction temperature. The catalytic performances of Pt-based catalysts for the conversion of HMF are summarized in Supplementary Table 2.Table 1Selective hydrogenation of HMF over Pt1/Nb2O5-Ov catalysts[a].EntryCatalystConversion (%)Selectivityc (%)TOF/hb12341––––––02Pt1/Nb2O5-Ov>99>99–No–18753dPt1/Nb2O5-Ov92>99–No–17424ePt1/Nb2O5-Ov86>99–No–16285fPt1/Nb2O5-Ov71>99–No–13446gPt1/Nb2O5-Ov35>99–No–656aReaction conditions: HMF (0.3 mmol), catalyst (20 mg), solvent (THF 2 mL), reaction temperature (160 °C), H2 pressure (4.0 MPa), reaction time (4 h), stirring speed (600 rpm).bTOF = moles of product × moles of metal−1 × h−1.cThe selectivity was calculated based on the product detected by gas chromatography.dReaction temperature (160 °C), H2 pressure (2.0 MPa).eReaction temperature (160 °C), H2 pressure (1.0 MPa).fReaction temperature (140 °C), H2 pressure (4.0 MPa).gReaction temperature (120 °C), H2 pressure (4.0 MPa).The quantum chemical calculations demonstrated that HMF could be hydrogenated to produce MF and DMF with Gibbs free energy (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}ΔG) of −1.37 and −2.93 eV, respectively (Fig. 4a and Supplementary Table 3), indicating that MF is a thermodynamically unstable compound in the reaction. Even if MF is produced from the hydrogenation of HMF, it has the thermodynamic potential to be further hydrogenated to DMF. The variation of conversion and selectivity with reaction time over Pt1/Nb2O5-Ov was studied and the results are shown in Fig. 4b. It was found that the selectivity of MF was independent of the conversion of HMF and could reach >99% of all the reaction times. The conversion of HMF approached 100% at 4 h. Moreover, even the reaction time was prolonged to 6 h, the selectivity of MF was still >99%.Fig. 4Factors affected the selective hydrogenation of HMF to MF.a Gibbs free energy of HMF hydrogenation to MF and DMF from DFT calculation. C, H, and O atoms are represented as silver, white, and red balls. b The effect of reaction time over Pt1/Nb2O5-Ov.DiscussionTo unveil the reason for the high MF selectivity of Pt1/Nb2O5-Ov in selective hydrogenation of HMF, the selective chemisorption of C–OH and C=O on the surface of Pt1/Nb2O5-Ov was investigated by FT-IR spectroscopy using methanol and n-propanal as model molecules. It was found that methanol dissociates into methoxy species on Pt1/Nb2O5-Ov, and the bands of the methoxy species observed at 1218 cm−1 and 1155 cm−1 can be assigned to the ν(C–O) bands of the on-top and bridged sites, respectively (Supplementary Fig. 4). This is the direct evidence of chemical adsorption C–OH by Pt1/Nb2O5-Ov. However, Pt1/Nb2O5-Ov could not adsorb C=O groups chemically. The FT-IR spectroscopy experimental results indicate that C–OH in HMF is selectively activated over Pt1/Nb2O5-Ov. To further prove this, MF was also used as the substrate to check the performance of Pt1/Nb2O5-Ov. MF could not be converted over the Pt1/Nb2O5-Ov (Supplementary Table 4), indicating that it is not active for the hydrogenation of C=O, which is the main reason for the high selectivity of MF. The heterogeneous nature of Pt1/Nb2O5-Ov was evaluated by removing the catalyst after the reaction was conducted for 1 h, and then the reaction was continued for 5 h without the solid catalyst. The product yield was not further increased without Pt1/Nb2O5-Ov (Supplementary Fig. 5a), indicating no leaching of active species to the reaction mixture. The reusability of the catalyst was tested, and the results are shown in Supplementary Fig. 5b, the yield decreased slightly from run 1 to run 4 due to inevitable loss of catalyst in the recovery process. After the 4th run, 17 mg of Pt1/Nb2O5-Ov (85% original amount) was recovered. From run 5 to run 7, 0.25 mmol of the reactant was added, which was 85% of the first 4 runs. The yield of the product in the 5th run was nearly the same as that of the first run. The yields of runs 6 and 7 were slightly reduced because the slight loss of the catalyst in the recovery process. These results suggest that the intrinsic activity of the catalyst did not decrease during the recycle process. The used catalysts were characterized by HRTEM/elemental mapping and XPS. As shown in Supplementary Fig. 6, no Pt clusters or particles were observed on the surface of the used catalysts. The mapping images indicate the uniform distributions of Nb, O, and Pt over the used catalysts. The binding energy of Ptδ+4f5/2 (76.3 eV) and Ptδ+4f7/2 (73.0 eV) for used catalyst was almost the same as that of original Pt1/Nb2O5-Ov (Supplementary Fig. 7), suggesting that the electron state did not change during the reaction process. Moreover, the content of Pt in used Pt1/Nb2O5-Ov was 0.039 wt%, indicating no Pt leaching. These results show the excellent stability of Pt1/Nb2O5-Ov in the reaction systems.DFT calculations were performed to study the adsorption of HMF on Nb2O5 and Pt1/Nb2O5-Ov surfaces. All the calculations were conducted using the Vienna Ab initio Simulation Package (VASP), and the calculation details were provided in the experimental section. The three-layer p(3 × 3) surface slab for the Nb2O5(001) surfaces and Pt atoms modified Nb2O5 surfaces with (Pt1/Nb2O5-Ov) or without (Pt1/Nb2O5) oxygen vacancies were built (Supplementary Fig. 8). The distances of the Nb3-Pt is 3.363 Å in Pt1/Nb2O5-Ov (Supplementary Fig. 8c), which is consistent with the results observed in the experiment (Supplementary Table 1), indicating that the model we constructed is reasonable. It can be found that oxygen vacancies are easily formed on Pt1/Nb2O5, where the corresponding oxygen vacancy formation energy is exothermic 0.07 eV (Supplementary Fig. 9), indicating the Pt1/Nb2O5-Ov surface is stable.The adsorption energy of HMF adsorbed on various Nb2O5 surfaces by C=O adsorption mode or C–OH adsorption mode is first calculated (Fig. 5). The results show that the adsorption energy of HMF by C–OH adsorption mode is higher than the adsorption energy of HMF by C=O adsorption mode on the same sites of Nb2O5 or Pt1/Nb2O5-Ov surfaces, suggesting the preferential adsorption of C–OH. The adsorption energy of HMF by C–OH adsorption mode on Nb2 site of Pt1/Nb2O5-Ov surfaces was −1.14 eV, which is much higher than the adsorption energy of HMF by C=O adsorption mode (−0.78 eV). Furthermore, Pt and oxygen vacancies promoted the adsorption of HMF as evidenced that the adsorption energy of HMF by C–OH or C=O mode is the highest on Pt1/Nb2O5-Ov surfaces. In addition, we also calculated the H2 adsorption on Pt1/Nb2O5-Ov surfaces (Supplementary Fig. 10). The adsorption energy is exothermic 0.51 eV, which is weaker than HMF adsorption.Fig. 5Calculated adsorption energy and structures of HMF.(side view, Inset: top view). a, b Nb2O5 surface, which marked with blue background. c–f Pt1/Nb2O5-Ov surface, which marked with green background. The adsorption modes for C=O (upper row) and the C–OH (bottom row) in HMF. Silver: C, white: H, red: O, light blue: Nb, blue: Pt.It has been reported that the dissociation of H2 occurs via a heterolytic pathway over SACs, because all metal atoms are individually dispersed and no metal-metal pairs available for homolytic dissociation of H238. The reaction pathway over Pt1/Nb2O5-Ov for the selective hydrogenation of HMF was also calculated by DFT (Fig. 6). First, the C–OH of HMF adsorbed on Nb sites, and then H2 adsorbed on Pt was readily split into two H species. One of the H species moved to nearby oxygen on Pt1/Nb2O5-Ov to yield O–H species, leaving the other H on Pt formed Pt–H species. This step was calculated to be exothermic by 1.55 eV and exhibited a barrier of 0.45 eV (from c to e). It is generally accepted that the activity of Pt–H species is higher than that of O–H species37. The Pt–H species attacked the C site of C–OH in HMF accompanied by the breakage of C–O, leading to the formation of MF and -OH adsorbed on the Nb site. This step is exothermic 0.48 eV with an energy barrier of 0.90 eV (from e to g). Finally, H2O was formed with endothermic 0.16 eV and the entire reaction cycle was completed on Pt1/Nb2O5-Ov surfaces. Therefore, the ability of producing active H species and the adsorption modes of HMF determined the activity and selectivity of HMF hydrogenation reaction on Pt1/Nb2O5-Ov surfaces.Fig. 6Reaction pathway based on DFT calculations.Calculated energy profile of a HMF hydrogenation on Pt1/Nb2O5-Ov surfaces, as well as b–i are the key structures of HMF hydrogenation. Silver: C, white: H, red: O, light blue: Nb, blue: Pt, green: H (Hydrogen).Combining DFT calculation and experimental results, we can conclude that the Pt and Nb sites near the interface in the Pt1/Nb2O5-Ov cooperated very well to promote the selective hydrogenation of HMF to MF. The Pt atom was responsible for the activation of H2 and the Nb sites activated C–OH. Following this mechanism, it can be deduced that the selectivity of MF should be independent of the characteristics of metal for SACs. To verify this, Pd1/Nb2O5-Ov and Au1/Nb2O5-Ov were also prepared using the same method with Pt1/Nb2O5-Ov (the characterization results are shown in Supplementary Figs. 11 and 12) and their catalytic performances were checked (Fig. 7). Both Pd1/Nb2O5-Ov and Au1/Nb2O5-Ov showed >99% selectivity of MF, although the activity of Au1/Nb2O5-Ov was lower, showing that the nature of the metal species in the SACs only affected the activity but not the selectivity, which further supports the argument that the metal atom and the Nb sites activated the H2 and C–OH respectively in the reaction. For comparison, we also prepared the supported Pt, Pd, and Au nanocatalysts Ptn/Nb2O5, Pdn/Nb2O5, and Aun/Nb2O5, and used to catalyze the reaction. All the nanocatalysts showed very low selectivity to MF because the multiple active sites provided by nanoparticles could activate H2 and different chemical bonds in HMF. The results also further showed that the SACs had unique features for the selective hydrogenation of HMF to MF.Fig. 7Catalytic performance of M1/Nb2O5-Ov and Mn/Nb2O5.Reaction results for the catalyst of a Pt1/Nb2O5-Ov, Pd1/Nb2O5-Ov, and Au1/Nb2O5-Ov, reaction time 4 h; b Ptn/Nb2O5, Pdn/Nb2O5, and Aun/Nb2O5, reaction time 1 h; reaction condition: substrate (0.3 mmol), catalyst (20 mg), THF (2.0 mL), H2 (4.0 MPa), 160 °C. The products were detected by gas chromatography. The black arrows points to the conversion of HMF.The selectivity of different substrates with OH and CHO over Pt1/Nb2O5-Ov were investigated and the results are given in Supplementary Table 5. It was found that 98% yield of 5-methyl-2-thiophenecarboxaldehyde could be achieved for the selective hydrogenolysis of 5-(hydroxymethyl)thiophene-2-carbaldehyde (Supplementary Table 5, entry 1). The reactivity of furfural and furfuryl alcohol were checked, and it was found that Pt1/Nb2O5-Ov was active for the hydrogenolysis of the furfuryl alcohol to 2-methylfuran while inactive for the hydrogenation of furfural to furfuryl alcohol (Supplementary Table 5, entries 2 and 3). The similar results were obtained for the reactivity of 5-methylfurfural and (5-methyl-2-furyl)methanol (Supplementary Table 5, entries 4 and 5). The conversion of tetrahydrofurfuryl alcohol was very low over the catalyst (Supplementary Table 5, entry 6). The reactivity of benzaldehyde was higher than benzyl alcohol. In all, 24% conversion of benzyl alcohol and 55% conversion of benzaldehyde were obtained under the same reaction conditions (Supplementary Table 5, entries 7 and 8). However, the yield of p-methyl benzaldehyde could reach 23% for the selective hydrogenolysis of 4-(hydroxymethyl)benzaldehyde (Supplementary Table 5, entry 9). The catalyst could also catalyze the hydrogenolysis of aliphatic alcohols. The conversion of glycerol was 11% and the conversion of 1,2,6-hexanetriol was 26% (Supplementary Table 5, entries 10 and 11). Unfortunately, Pt1/Nb2O5-Ov could not catalyze the selective hydrogenation of aliphatic compounds with OH and CHO to the corresponding aldehyde (Supplementary Table 5, entries 12 and 13).In summary, SACs (Pt1/Nb2O5-Ov, Pd1/Nb2O5-Ov, and Au1/Nb2O5-Ov) can efficiently catalyze selective hydrogenation of HMF to MF. The selectivity to MF can be as high as >99% at complete conversion. The unusual and unique property of the SACs for the reaction originates from the excellent cooperation of the Nb and Pt sites near the interface in the catalysts. The Pt atom sites can only activate the H2, and the Nb sites solely activate C–OH in the reaction, while none of the Nb and Pt sites can activate the C=O group. Thus, the selectivity of MF is exceptionally high. Moreover, the catalysts can be reused at least 7 times without a decrease in the selectivity. We believe that the SACs and the reaction route have a high potential of application in producing MF from biomass-derived HMF by hydrogenation because of the apparent advantages, such as excellent selectivity, and high efficiency and stability, and the reaction process is very straightforward and uses biomass-derived feedstocks. We also believe that the protocol to combine the active sites of metal and support in SACs can also be used to catalyze some other selective hydrogenation reactions in biomass transformation.MethodsChemicals and materialsNiobium(V) oxalate hydrate (99%, alfa), cetyl-trimethyl-ammonium bromide (CTAB, 99%, Sigma-Aldrich), and H2PtCl6·6H2O (>99.0%) was purchased from Sinopharm Chemical Reagent Co. Ltd. Tetrahydrofuran, 5-hydroxymethylfurfural, 2,5-dimethylfuran, 2,5-bis(hydroxymethyl)furan, 2,5-bis(hydroxymethyl)tetrahydrofuran, 5-methylfurfural, 5-methylfurfurylalcohol, 5-methyltetrahydrofuran-2-methanol, 2-hexanol, 2,5-dimethyltetrahydrofuran and n-decane were purchased from J&K, which were all analytical grade. H2 (>99.99%) and Ar (>99.99%) were supplied by Beijing Analytical Instrument Company. Ultrapure water (resistivity ≥18 MΩ cm) was used in the experiments. All chemicals were used without further purification.Synthesis of Nb2O5 with oxygen-rich vacancyNb2O5 with oxygen-rich vacancy defects (Nb2O5-Ov) were prepared by the sol-gel method using CTAB as the surfactant. In all, 0.538 g of Niobium(V) oxalate hydrate was dissolved in 20 mL of ultrapure water, and a clear solution was obtained after stirring 2 h at room temperature. In all, 20 mL of CTAB surfactant solution (0.2 M) was added drop wise into the above prepared solution, followed stirring for 2 h at room temperature. Then they were transferred into a dried Teflon autoclave with a capacity of 100 mL at ambient temperature, followed by hydrothermal treatment at 180 °C for 10 h. After being cooled to room temperature, the white precipitate was separated by high-speed centrifugation, washed with ethanol for three times and distilled water for three times, and then dried in an oven at 70 °C for 6 h. The grinded powder was calcined in a tube furnace (Anhui Kemi Machinery Technology Co., Ltd.; Model TFV-1200-50-I-220) at 500 °C for 4 h (reducing gas contains 10 vol.% hydrogen and 90 vol.% argon) to obtain defective Nb2O5 supports with surface oxygen vacancies.Synthesis of M1/Nb2O5-OvPt1/Nb2O5-Ov SAC was synthesized according to the facile adsorption method39. In total, 100 mg of Nb2O5-Ov powders were suspended in 10 mL water. After stirring for 30 min, appropriate amount of H2PtCl6 solution was added dropwise into the Nb2O5-Ov dispersion under stirring. After stirring for 4 h and followed by aging for 4 h, the solution was centrifuged and washed with deionized water for several times, and then dried at 60 °C for 12 h. The synthesized catalyst was denoted as Pt1/Nb2O5-Ov. Nb2O5-Ov supported Pd SAC and Au SAC were prepared by the same method as that for Pt1/Nb2O5-Ov except that H2PdCl4 and HAuCl4 were used to substitute for H2PtCl6, and the catalyst were denoted as Pd1/Nb2O5-Ov and Au1/Nb2O5-Ov.Synthesis of Mn/Nb2O5200 mg of Nb2O5 powders were suspended in 20 mL water. After stirring for 30 min, appropriate amount of H2PtCl6 solution was added dropwise into Nb2O5 dispersion under stirring. Then aqueous solution containing fresh NaBH4 was added drop-wise with a continuous magnetic stirring under argon atmosphere, the reaction solution kept on stirring for 2 h to complete the reduction reaction. The obtained granules were collected by centrifuging and washing with ultrapure water for three times (3 × 30 mL) and ethanol twice (2 × 30 mL) and dried in a vacuum oven at 60 °C for 12 h. Nb2O5 supported Pd nanoparticles and Au nanoparticles were prepared by the same method.CharacterizationThe XRD experiment was performed on Rigaku D/max 2500 with nickel filtered Cu-Kα (λ = 0.154 nm) operated at 40 kV and 20 mA. X-band electron-paramagnetic resonance (EPR) measurement was performed at room temperature using a Bruker EMXplus-9.5/12 EPR spectrometer, with the sample mass of 50 mg. The TEM images of the catalysts were obtained using a JEOL-2100F electron microscope operated at 120 kV. Aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (AC–HAADF–STEM) and element energy dispersive spectroscopy (EDS) mapping images were conducted on a JEOL JEM–ARM200F equipment. The XPS spectra were collected on an ESCA Lab 220i-XL electron spectrometer (VG Scientific) using 300 W Al Kα radiation with a hemispherical energy analyser. The contents of different elements in Pt1/Nb2O5-Ov and Ptn/Nb2O5 catalysts were analyzed by ICP-AES (PROFILE. SPEC, Leeman). The binding energies were calibrated with the C1s level of adventitious carbon at 284.8 eV as the internal standard reference. The XAFS spectra of Pt L3 edge (11564 eV) were collected at 1W1B beamline of Beijing Synchrotron Radiation Facility (BSRF). The beam was tuned by the Si (111) double-crystal monochromators. The energies were calibrated according to the absorption edge of pure Pt foil. The XAFS data were recorded under fluorescence mode by Lytle detector. All collected spectra were processed and analyzed using Athena and Artemis program within the IFEFFIT package. For the XANES analysis, the experimental absorption coefficients as the function of energies were processed by background subtraction and normalization procedure, and reported as “normalized intensity”. Pt foil and PtO2 were used as the reference samples. For the extended X-ray absorption fine structure analysis, Fourier transformed data in R space were analyzed by applying metallic Pt model for the Pt-Pt shell. The passive electron factors, S02, were determined by fitting the experimental data of Pt foil and fixing the Pt-Pt coordination number (CN) to be 12, and then fixed for further analysis of the measured samples. The parameters describing the electronic properties (e.g., correction to the photoelectron energy origin, E0) and local structure environment including coordination number (CN), bond distance (R), and Debye Waller factor (σ2) around the absorbing atoms were allowed to vary during the fitting process. The fitted ranges for k and R spaces (k2 weighted) were k = 2.8-10.0 Å−1 and R = 1.6-3.2 Å. FT-IR spectra of the catalysts with the aldehyde groups and hydroxyl groups absorbed on the catalysts were recorded with a TENSOR 27 spectrometer. The sample (20 mg) of Pt1/Nb2O5-Ov catalyst was dispersed in n-propanal or methanol solution of THF at 160 °C and stirred for 12 h. The suspension was centrifuged and washed using THF to remove the physical absorbed species, and then dried at 100 °C for 12 h. The obtained samples were blended with KBr for IR characterization.Hydrogenation reactionThe reaction was carried out in a Teflon-lined stainless steel reactor of 16 mL with a magnetic stirrer. In a typical experiment, suitable amount of reactant, catalyst and solvent were loaded into the reactor. The reactor was sealed and purged with hydrogen to remove the air at room temperature. Then the reactor was placed in a furnace at desired temperature and H2 was charged to desired pressure. The stirrer was started with a stirring speed of 600 rpm. After the reaction, the reactor was placed in ice water, the gas was released and a known amount of internal standard (n-decane) was added into the reactor. The liquid reaction mixture in the reactor was transferred into a centrifuge tube. The reactor was washed using THF, which was combined the reaction mixture. The catalyst was separated by centrifugation. The quantitative analysis of the liquid products was conducted using a GC (Agilent 6820) equipped with a flame ionization detector (FID) and a HP-5MS capillary column (0.25 mm in diameter, 30 m in length). Identification of the products and reactant was done using a GC–MS (Agilent 7890B 5977 A, HP-5MS capillary column (0.25 mm in diameter, 30 m in length)) as well as by comparing the retention time with n-decane which was used as the internal standard in GC traces. The conversion of 5-hydroxymethylfurfural and the selectivity of the products were calculated based on the GC data.Computational detailsSpin-unrestricted density functional theory (DFT) calculations were performed using the Gaussian 16 program package40. The free energy of each compound was calculated at the B3LYP/6-311 + G*41–43 level, and the entropy were calculated in this work are for a standard state of 413.15 K and 1 atm. The computed Gibbs free energy (G) was obtained by Eq. (1):1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G = E - {\mathrm{TS}} + {\mathrm{ln}}\frac{P}{{P^{\it{0}}}}$$\end{document}G=E−TS+lnPP0All spin-polarized DFT calculations in this work were carried out using the Vienna Ab-initio Simulation Package (VASP)44. The projector augmented wave (PAW) method45 and the Perdew-Burke-Ernzerhof (PBE)46 functional utilizing the generalized gradient approximation (GGA)47 were applied throughout the calculations. The Brillouin zone integration was performed using 2 × 2 × 1 K-point mesh was used for all these models. The top two layers of all slabs were allowed to fully relax, while the bottom single layer was kept fixed to mimic the bulk region. The kinetic energy cut-off was set as 450 eV, the structure optimization force threshold was 0.03 eV/Å, and the self-consistent calculations applied a convergence energy threshold of 10−6 eV. We used a large vacuum height of 15 Å to eliminate the interaction between neighboring slabs. The transition states (TS) of surface reactions were located using a constrained optimization scheme and were verified when (i) all forces on atoms vanish and (ii) the total energy is a maximum along the reaction coordination but a minimum with respect to the rest of the degrees of freedom48–50. Vibrational analysis was carried out to ensure that the transition states have only one imaginary frequency along the reaction coordinate (Supplementary Table 6). The adsorption energy of species X on the surface, Eads(X), was calculated with2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{ads}}}\left( {\mathrm{X}} \right) = E_{{\mathrm{X}}/{\mathrm{slab}}}-E_{{\mathrm{slab}}}-E_{\mathrm{X}}$$\end{document}EadsX=EX/slab−Eslab−EXwhere EX/slab is the calculated total energy of the adsorption system, while Eslab and EX are calculated energies of the clean surface and the gas-phase molecule X, respectively. Obviously, a negative Eads(X) value indicates an exothermic adsorption process. The oxygen vacancy formation energy (EOv) was calculated according to3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{Ov}}} = E_{{\mathrm{slab}} - {\mathrm{vac}}} + 1/2E_{{\mathrm{O2}}}-E_{{\mathrm{slab}}}$$\end{document}EOv=Eslab−vac+1/2EO2−Eslabwhere Eslab-vac is the total energy of the surface with one oxygen vacancy, and EO2 is the energy of a gas-phase O2 molecule.Supplementary informationSupplementary InformationDescription of Additional Supplementary FilesSupplementary Data 1 | nature communications | [
"Article"
] | [
"Heterogeneous catalysis",
"Sustainability",
"Density functional theory"
] | hydrogenation critical essential unsaturated groups compound C=C C=O C≡C C=N aromatic ring -NO2. methods reduce functional groups selectively others synthesize value-added chemicals broaden chemical knowledge.5-(Hydroxymethyl)furfural (HMF) important biomass platform compound produced from cellulose 40% lignocellulosic biomass1,2 reduction HMF promising high chemicals products 2,5-bis(hydroxymethyl)furan 2,5-dimethylfuran 2,5-dimethyltetrahydrofuran 2,5-hexanediole (HDO) produced selective hydrogenation 5-Methylfurfural (MF) useful chemical food additive synthetic selective hydrogenation C–OH group HMF 5-methylfurfural not realized using H2 agent C=O bond easier reduced than C–OH HMF Kinetically hydrogenation C=O bond HMF favorable C–OH Thermodynamically MF hydrogenated to DMF Ankur Bordoloi studied three reducible bonds HMF (C=O C=C C–OH C=O prone to hydrogenation strongest electrophilicity aldehyde carbon activation energy required hydrogenation C=O lower than dehydroxylation C–OH16Synthesis MF from HMF indirect routes transformation HMF to halogen compounds 5-chloromethylfurfural MF prepared hydrodehalogenation CMF IMF (Fig. selective hydrogenolysis HMF to MF HCOOH esterification procedure no MF detected H2 attractive prepare MF from hydrogenation HMF H2 simple byproduct H2O. synthetic routes MF HMF previous methods synthesize MF acid MF without additives ideal catalysts key selective hydrogenation HMF MF isolated single atomic catalysts (SACs) used reactions Pt1/N-C SACs high chemo regioselectivity terminal alkynes atomically dispersed copper (Cu) catalyst defective nanodiamond graphene excellent catalytic performance conversion acetylene to Pt1/α-MoC catalyst promising activity strong chemospecificity hydrogenation nitro SACs used catalyze reactions selective hydrogenation nitroarenes alkenes carbonyl compounds transformation methane aqueous reforming methanol hydroformylation olefins olefin metathesis oxygen reduction excellentNb2O5 used activation cleavage Caliphatic–O Caromatic–O bonds lignin model fabricated Pt1/Nb2O5-Ov Pd1/Nb2O5-Ov Au1/Nb2O5-Ov catalysts single atomic metal sites supported oxygen defective Nb2O5 catalyze selective hydrogenation HMF to MF >99% selectivity conversion calculations results cooperation Pt Nb sites Pt sites activation H2 Nb sites activated C–OH group HMF high selectivity achieved Nb2O5 with oxygen-vacancy defects) prepared thermal treating atmosphere 4 h at 500 °C 10% hydrogen 90% Defects oxygen detected by-paramagnetic resonance (EPR) measurement signal oxygen vacancy value 2.003 observed for Nb2O5-Ov Pt1/Nb2O5-Ov no EPR signal intrinsic Nb2O5 oxygen vacancy concentration Pt1/Nb2O5-Ov highest microscopy) image prepared Pt1/Nb2O5-Ov Fig. 2a.No Pt cluster particle Nb2O5-Ov mapping images suggest uniform distributions Nb O Pt (Fig. image bright points Nb2O5-Ov atomically dispersed Pt defective support (Fig. content Pt in Pt1/Nb2O5-Ov 0.039 wt%, analyzed plasma mass spectrometry Fourier-transformed k2-weighted X-ray absorption) Pt atoms Nb2O5-Ov one peak 1.5 Å Pt–O no peak 2.5 Å Pt-Pt contribution dispersed Pt atoms Pt1/Nb2O5-Ov (Fig. 2d). least-squares EXAFS fitting curves Pt1/Nb2O5-Ov Pt foil PtO2 Supplementary Fig. 2 structure parameters Supplementary Table 1.Fig. structural characterization Pt1/Nb2O5-Ov TEM images EDS mapping images AC–HAADF–STEM image atomically dispersed Pt atoms highlighted red circles k2-weighted Fourier transform spectra Pt1/Nb2O5-Ov Pt foilPt species Pt1/Nb2O5-Ov partially positively charged-ray PtO2 X-ray spectroscopy valence states Nb O Nb 3d5/2 binding energy Nb2O5 207.3 eV shifted 207.1 eV Pt1/Nb2O5-Ov O 1s binding energy Nb2O5 530.3 eV 530.1 eV Pt1/Nb2O5-Ov Pt 4f peak two peaks peaks 76.1 72.9 eV Ptδ+4f5/2 Ptδ+4f7/2 strong electron interaction between Nb2O5-Ov Pt electron transfers from Pt to Nb2O5-Ov. electron state characterization Pt1/Nb2O5-Ov XANES spectra Pt L3-edge XPS spectra Nb 3d c O1s Pt1/Nb2O5-Ov Nb2O5 performance HMF Pt1/Nb2O5-Ov catalytic systems studied selective hydrogenation HMF results Table 1. reaction catalyst Pt1/Nb2O5-Ov catalyze hydrogenation HMF MF >99% selectivity conversion in 4 h.turnover frequency MF 1875 h−1 (Table 1 2) effects reaction temperatures H2 pressures Pt1/Nb2O5-Ov catalysts studied temperature 160 to 120 °C H2 pressure 4.0 to 1.0 MPa selectivities MF >99% (Table 1 3–6) selectivity independent H2 pressure temperature catalytic performances Pt-based catalysts conversion HMF summarized Supplementary Table 1Selective hydrogenation HMF Pt1/Nb2O5-Ov catalysts conditions HMF (0.3 catalyst (20 solvent (THF 2 temperature (160 H2 pressure (4.0 time (4 stirring speed (600 moles product × metal−1 × h−1 selectivity calculated product detected gas chromatographytemperature (160 °C), H2 pressure (2.0 pressure (1.0 (140 °C), H2 pressure (4.0 (120 °C), H2 pressure (4.0 quantum chemical calculations HMF hydrogenated MF DMF Gibbs free energy −1.37 −2.93 eV (Fig. 4a Supplementary Table 3) MF thermodynamically unstable compound hydrogenation HMF potential hydrogenated to DMF variation of conversion selectivity with reaction time over Pt1/Nb2O5-Ov studied results Fig. 4b selectivity MF independent of conversion HMF >99% conversion HMF 100% at 4 h reaction time prolonged to 6 h selectivity MF >99%.Fig. 4Factors affected hydrogenation HMF to MF Gibbs free energy HMF hydrogenation to MF DMF calculation C H O atoms represented silver white red balls effect of reaction time over Pt1/Nb2O5-Ovhigh MF selectivity Pt1/Nb2O5-Ov hydrogenation HMF chemisorption C–OH C=O investigated by FT-IR spectroscopy methanol n-propanal methanol dissociates into methoxy species on bands methoxy species at 1218 cm−1 1155 cm−1 assigned to ν(C–O) bands on-top sites Fig. 4) evidence chemical adsorption C–OH by Pt1/Nb2O5-Ov adsorb C=O groups FT-IR spectroscopy results indicate C–OH in HMF selectively activated over Pt1/Nb2O5-Ov MF used substrate performance Pt1/Nb2O5-Ov MF converted over 4) not active for hydrogenation C=O main reason high selectivity MF heterogeneous nature evaluated removing catalyst after reaction 1 h continued 5 h without catalyst product yield not increased without Pt1/Nb2O5-Ov no leaching active species reaction mixture reusability catalyst tested yield decreased slightly from run 1 to run 4 due to loss catalyst recovery4th run 17 mg Pt1/Nb2O5-Ov (85% original recovered run 5 to 7 0.25 mmol reactant added 85% first 4 runs yield 5th run same first run yields runs 6 7 slightly reduced loss catalyst recovery intrinsic activity catalyst decrease recycle used catalysts characterized HRTEM/elemental mapping XPS no Pt clusters particles surface mapping images uniform distributions Nb, O Pt catalysts energy Ptδ+4f5/2 (76.3 eV) Ptδ+4f7/2 (73.0 eV) same original Pt1/Nb2O5-Ov electron state change reaction content Pt Pt1/Nb2O5-Ov 0.039 wt%, no Pt leaching results show excellent stability Pt1/Nb2O5-Ov reaction systems.DFT calculations adsorption HMF on Nb2O5 Pt1/Nb2O5-Ov surfaces Vienna Ab initio Simulation Package details experimental section three-layer p(3 × 3) surface slab Nb2O5(001) surfaces Pt atoms Nb2O5 surfaces without oxygen vacancies builtdistances Nb3-Pt 3.363 Å Pt1/Nb2O5-Ov Fig. consistent results experiment Table 1) model reasonable oxygen vacancies formed Pt1/Nb2O5 energy exothermic 0.07 eV Fig. 9) Pt1/Nb2O5-Ov surface stable adsorption energy HMF Nb2O5 surfaces C=O C–OH calculated (Fig. 5) results C–OH higher C=O same sites Nb2O5 Pt1/Nb2O5-Ov surfaces preferential adsorption C–OH adsorption energy HMF C–OH mode Nb2 site Pt1/Nb2O5-Ov surfaces −1.14 eV higher C=O mode (−0.78 eV). Pt oxygen vacancies adsorption HMF C–OH=O mode highest on Pt1/Nb2O5-Ov surfaces calculated H2 adsorption Pt1/Nb2O5-Ov surfaces Fig. 10). adsorption energy exothermic 0.51 eV weaker than HMF adsorption.Fig. 5Calculated adsorption energy structures HMF Nb2O5 surface marked blue backgroundPt1/Nb2O5-Ov surface green background adsorption modes C=O C–OH HMF Silver C white H red O light blue Nb blue Pt dissociation H2 heterolytic pathway SACs metal atoms dispersed no metal-metal pairs homolytic dissociation reaction pathway Pt1/Nb2O5-Ov selective hydrogenation HMF calculated DFT. 6) C–OH HMF adsorbed Nb H2 Pt split two H species moved oxygen Pt1/Nb2O5-Ov O–H other Pt Pt–H exothermic 1.55 eV barrier 0.45 eV activity Pt–H species higher O–H Pt–H attacked C site C–OH HMF breakage C–O formation MF -OH Nb site exothermic 0.48 eV barrier 0.90 eV H2O formed endothermic 0.16 eV reaction cycle completed Pt1/Nb2O5-Ov surfaces H species adsorption modes HMF determined activity selectivity HMF hydrogenation reaction Pt1/Nb2O5-Ov surfaces. 6Reaction pathway DFT calculationsenergy profile HMF hydrogenation Pt1/Nb2O5-Ov surfaces b–i key structures Silver C white H red O light blue: Nb blue Pt green: H DFT calculation experimental results Pt Nb sites interface Pt1/Nb2O5-Ov selective hydrogenation HMF to MF Pt atom H2 Nb sites activated C–OH selectivity MF independent characteristics metal SACs Pd1/Nb2O5-Ov Au1/Nb2O5-Ov prepared same method Supplementary Figs. 11 12 catalytic performances checked (Fig. 7) Au1/Nb2O5-Ov showed >99% selectivity MF activity Au1/Nb2O5-Ov lower metal species affected activity not selectivity supports argument metal atom Nb sites activated H2 C–OH prepared Pt, Pd Au nanocatalysts Ptn/Nb2O5 Pdn/Nb2O5 Aun/Nb2O5 catalyze reaction nanocatalysts showed low selectivity to MF multiple active sites activate H2 different chemical bonds HMF SACs unique features selective hydrogenation HMF to MF.Fig.performance M1/Nb2O5-Ov Mn/Nb2O5 results Pt1/Nb2O5-Ov Pd1/Nb2O5-Ov Au1/Nb2O5-Ov time 4 h Ptn/Nb2O5 Pdn/Nb2O5 Aun/Nb2O5 time 1 h substrate (0.3 catalyst (20 mg), THF (2.0 H2 (4.0 160 °C products detected gas chromatography black conversion HMF selectivity substrates OH CHO over Pt1/Nb2O5-Ov investigated results Supplementary Table 5. 98% yield 5-methyl-2-thiophenecarboxaldehyde hydrogenolysis reactivity furfural furfuryl alcohol checked Pt1/Nb2O5-Ov active hydrogenolysis furfuryl alcohol 2-methylfuran inactive hydrogenation alcohol similar results 5-methylfurfural (5-methyl-2-furyl)methanol conversion tetrahydrofurfuryl alcohol low catalyst reactivity benzaldehyde higher benzyl alcohol24% conversion benzyl alcohol 55% benzaldehyde reaction conditions Table 5 entries 7 8). yield p-methyl benzaldehyde 23% selective hydrogenolysis 4-(hydroxymethyl)benzaldehyde Table 5 entry 9) catalyst catalyze hydrogenolysis aliphatic alcohols conversion glycerol 11% 1,2,6-hexanetriol 26% Table 5 entries 10 11). Pt1/Nb2O5-Ov catalyze selective hydrogenation aliphatic compounds OH CHO aldehyde Table 5 entries 12 SACs (Pt1/Nb2O5-Ov Pd1/Nb2O5-Ov Au1/Nb2O5-Ov) catalyze hydrogenation HMF MF selectivity MF >99% complete conversion SACs cooperation Nb Pt sites Pt activate H2 Nb activate C–OH Nb Pt activate C=O group selectivity MF high catalysts reused 7 times without decrease selectivity SACs reaction route high potential producing MF biomass-derived HMF hydrogenation excellent selectivity high efficiency stability reaction process straightforward uses biomass-derived feedstocksprotocol sites metal SACs catalyze selective hydrogenation reactions biomass transformation materialsNiobium(V) oxalate hydrate (99% cetyl-trimethyl-ammonium bromide H2PtCl6·6H2O>99.0% purchased Sinopharm Chemical Reagent Tetrahydrofuran 5-hydroxymethylfurfural 2,5-dimethylfuran 5-methylfurfural 5-methylfurfurylalcohol 5-methyltetrahydrofuran-2-methanol 2-hexanol 2,5-dimethyltetrahydrofuran n-decane purchased J&K grade H2>99.99% Ar>99.99% supplied Beijing Analytical Instrument Company Ultrapure water ≥18 MΩ used chemicals purification Nb2O5 oxygen-rich prepared sol-gel method CTAB surfactant 0.538 g Niobium(V oxalate hydrate dissolved 20 mL ultrapure water clear solution stirring 2 h 20 mL CTAB surfactant solution added solution 2 h transferred dried Teflon autoclave 100 mL hydrothermal treatment 180 °C 10 hcooled to room temperature white precipitate separated centrifugation washed ethanol distilled dried oven 70 °C 6 h grinded powder calcined tube furnace Kemi Machinery 500 °C 4 h 10% hydrogen 90% argon defective Nb2O5 supports surface oxygen vacancies.Synthesis M1/Nb2O5-OvPt1/Nb2O5-Ov SAC synthesized facile adsorption 100 mg Nb2O5-Ov powders suspended in 10 mL water stirring 30 min H2PtCl6 solution added dispersion 4 h solution centrifuged washed deionized water dried 60 °C 12 h synthesized catalyst Pt1/Nb2O5-Ov Pd SAC Au SAC prepared Pt1/Nb2O5-Ov H2PdCl4 HAuCl4 H2PtCl6 catalyst denoted Pd1/Nb2O5-Ov Au1/Nb2O5-Ov.Synthesis Mn/Nb2O5200 mg Nb2O5 powders suspended 20 mL water 30 min H2PtCl6 solution added Nb2O5 dispersion aqueous solution NaBH4 added drop-wise stirring 2 h reduction reaction.granules ultrapure water 30 ethanol dried vacuum oven 60 °C 12 h Nb2O5 Pd Au nanoparticles prepared XRD experiment Rigaku D/max 2500 nickel filtered Cu-Kα 0.154 nm) 40 kV 20 mA X-band room temperature Bruker EMXplus-9.5/12 EPR spectrometer sample mass 50 mg TEM images JEOL-2100F electron microscope 120 kV Aberration-corrected dispersive spectroscopy JEOL JEM–ARM200F XPS spectra ESCA Lab 220i-XL electron spectrometer 300 W Al Kα radiation hemispherical energy analyser Pt1/Nb2O5-Ov Ptn/Nb2O5 catalysts analyzed ICP-AES binding energies calibrated C1s level adventitious carbon 284.8 eV XAFS spectra Pt L3 edge (11564 eV collected 1W1B beamline Beijing Synchrotron Radiation Facility tuned (111) double-crystal monochromators energies calibrated absorption edge Pt foil XAFS data recorded fluorescence mode Lytle detectorcollected spectra processed analyzed Athena Artemis IFEFFIT package XANES analysis absorption coefficients processed subtraction normalization reported “normalized intensity”. Pt foil PtO2 reference samples extended X-ray absorption structure analysis Fourier transformed data R space analyzed metallic Pt model Pt-Pt shell passive electron factors determined fitting data Pt foil Pt-Pt coordination number 12 analysis parameters electronic properties photoelectron energy origin local structure coordination number bond distance Debye Waller factor atoms vary during fitting fitted ranges k R spaces k Å−1 R = 1.6-3.2 Å FT-IR spectra catalysts aldehyde hydroxyl groups recorded TENSOR 27 spectrometer sample (20 mg) Pt1/Nb2O5-Ov catalyst dispersed in n-propanal methanol solution THF 160 °C stirred 12 h suspension centrifuged washed THF dried 100 °C 12 h samples blended with KBr for IR characterization.Hydrogenation Teflon-lined stainless steel reactor 16 mL magnetic stirrer reactant catalyst solvent loaded sealed purged with hydrogenreactor placed in furnace at temperature H2 charged to pressure stirrer started 600 rpm After reaction placed in ice water gas released internal standard (n-decane) added liquid reaction mixture transferred into centrifuge tube reactor washed using THF catalyst separated by centrifugation quantitative analysis liquid products using GC (Agilent 6820) flame ionization detector HP-5MS capillary column Identification of products reactant using GC–MS retention time with n-decane conversion of 5-hydroxymethylfurfural selectivity of products calculated based GC data detailsSpin-unrestricted density functional theory (DFT) calculations using Gaussian 16 program free energy of each compound calculated at B3LYP/6-311 + G*41–43 level entropy for standard state 413.15 K 1 atm. computed Gibbs free energy (G) obtained by Eq.\documentclass[12pt{minimal\usepackage{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}{document$G = E -\mathrm{TS}} +\mathrm{ln}}{P}G=E−TS+lnPP0All spin-polarized DFT calculations Vienna Ab-initio Simulation Package (VASP)44 projector augmented wave) method45 Perdew-Burke-Ernzerhof (PBE)46 generalized gradient approximation (GGA)47 applied Brillouin zone integration 2 × 2 × 1 K-point mesh top two layers relax bottom layer fixed bulk region kinetic energy cut-off 450 eV structure optimization force threshold 0.03 eV/Å convergence energy threshold 10−6 eV large vacuum height 15 Å interaction slabs transition states) of surface reactions constrained optimization scheme verified when forces atoms vanish total energy maximum reaction coordination minimum degrees freedom48–50 Vibrational analysis transition states one imaginary frequency reaction coordinate (Supplementary Table 6)adsorption energy species X surface Eads calculated\documentclass[12pt{minimal\usepackage{amsmath}{wasysym{upgreek\oddsidemargin-69pt}{document}\mathrm{ads}}}\left\mathrm{X}} = E\mathrm\end{document}EadsX=EX/slab−Eslab−EXwhere EX/slab calculated total energy adsorption system Eslab EX energies clean surface gas-phase molecule X negative Eads(X) value indicates exothermic adsorption processoxygen vacancy energy (EOv calculated[12pt{minimal{amsmath{wasysym\oddsidemargin-69pt}{document}{Ov = E{slab}} -{vac + 1/2E{O2-E{slab{document}EOv=Eslab−vac+1/2EO2−Eslabwhere Eslab-vac total energy surface one oxygen vacancy EO2 energy gas-phase O2 molecule.Supplementary Additional Supplementary Data 1 | 51.2 | 0.688261 |
10.1038/s41467-020-18501-w | PMC7493924 | Caenorhabditis elegans responds to mitochondrial stress by activating the mitochondrial unfolded protein response (UPRmt). Here the authors show that HDA-1, the C. elegans ortholog of mammalian histone deacetylase (HDAC), coordinates with the genome organizer DVE-1 to activate UPRmt and modulate mitochondrial homeostasis. | The ability to detect, respond and adapt to mitochondrial stress ensures the development and survival of organisms. Caenorhabditis elegans responds to mitochondrial stress by activating the mitochondrial unfolded protein response (UPRmt) to buffer the mitochondrial folding environment, rewire the metabolic state, and promote innate immunity and lifespan extension. Here we show that HDA-1, the C. elegans ortholog of mammalian histone deacetylase (HDAC) is required for mitochondrial stress-mediated activation of UPRmt. HDA-1 interacts and coordinates with the genome organizer DVE-1 to induce the transcription of a broad spectrum of UPRmt, innate immune response and metabolic reprogramming genes. In rhesus monkey and human tissues, HDAC1/2 transcript levels correlate with the expression of UPRmt genes. Knocking down or pharmacological inhibition of HDAC1/2 disrupts the activation of the UPRmt and the mitochondrial network in mammalian cells. Our results underscore an evolutionarily conserved mechanism of HDAC1/2 in modulating mitochondrial homeostasis and regulating longevity. | IntroductionMitochondria play essential and pervasive roles in biology. Originating from endosymbiosis of proteobacteria, these organelles not only provide host cells with ATP generated through oxidative phosphorylation, but also participate in numerous biological processes from the regulation of calcium homeostasis to innate immune responses and programmed cell death1. The normal function of mitochondria is challenged by intrinsic stimuli and by extrinsic pathogens and xenobiotics2. Mitochondrial dysfunction has been extensively linked with aging3,4 and numerous diseases such as neurodegenerative disorders5. Given the importance of mitochondria, cells employ quality control mechanisms to actively surveil mitochondrial function and initiate protective programs upon mitochondrial damage.One such quality control mechanism is the mitochondrial unfolded protein response (UPRmt), which relays mitochondrial stress signals to the transcription of nuclear-encoded genes that protect mitochondria, such as chaperones and proteases6–9. In C. elegans, the transcriptional response is mainly governed by two transcription factors, ATFS-1 and DVE-17,9–11. In addition, growing evidence has also suggested that chromatin modifiers act in conjunction with these two transcription factors to modulate UPRmt activation12–14.Eukaryotic chromosomes are packed into three-dimensional higher-order structures. Chromatin modifier proteins induce changes in chromatin architecture and thereby control accessibility of DNA sequences to the transcription machinery. For example, post-translational modifications, such as acetylation, of histones H3 and H4 have been reported to dictate the active or inactive chromatin state and ultimately affect gene expression15,16. Histone acetylation states are tightly regulated by two chromatin modifiers with opposing function, histone acetyltransferase (HAT) and histone deacetylase (HDAC)16, allowing for the precise control of gene expression. Given that UPRmt requires the activation of a broad spectrum of genes to counteract mitochondrial stress and reset the cellular metabolic state, the biological importance of chromatin modifier proteins in UPRmt warrants further exploration.Here, we report that histone deacetylase HDA-1 is a key regulator of UPRmt in C. elegans. Mechanistically, HDA-1 acts in concert with DVE-1 to promote the transcription of genes involved in the mitochondrial stress response, innate immune response, and metabolism. Moreover, expression profiles in tissues from rhesus monkey and human indicate that transcript levels of primate HDAC1/2 strongly correlate with the expression of UPRmt genes. Collectively, our results highlight the conserved and crucial function of HDAC1/2 in regulating the mitochondrial stress response and its beneficial outcomes.Resultshda-1 is required for UPRmt activationTo understand the molecular mechanisms that govern activation of UPRmt, we carried out a genome-wide RNA interference (RNAi) screen by employing C. elegans UPRmt reporter strains such as hsp-6p::gfp2. Specifically, we fed age-synchronized wild-type worms expressing GFP fluorescent reporters with dsRNA-expressing bacteria and tested the animals for their abilities to activate GFP expression upon mitochondrial perturbation. hda-1, one of the HDAC genes, was recovered from this screen2. RNAi-mediated knockdown of hda-1 impaired the activation of the UPRmt reporter hsp-6p::gfp that is induced by RNAi of the nuclear-encoded mitochondrial gene atp-2 (ATP synthase F1 subunit beta) or cco-1 (cytochrome c oxidase subunit 5B) (Fig. 1a, b). Deficiency of hda-1 also suppressed the elevation of endogenous transcript levels of hsp-6 under mitochondrial stress (Fig. 1c). In addition, lack of hda-1 impaired the activation of another UPRmt reporter, hsp-60p::gfp, during mitochondrial perturbation (Fig. 1d). To test if hda-1 is specifically required for UPRmt activation, or if it actually mediates a more general stress response, we knocked down hda-1 in the endoplasmic reticulum (ER) stress reporter strain hsp-4p::gfp or the heat shock stress reporter strain hsp-16.2p::gfp, and challenged the animals with either the ER inhibitor tunicamycin or heat shock treatment. Deficiency of hda-1 did not affect the activation of ER stress or heat shock stress response (Fig. 1e, f; Supplementary Fig. 1a, b). Collectively, these results indicate that hda-1 plays a specific role in modulating the mitochondrial stress response.Fig. 1hda-1 is required for UPRmt activation.a Representative fluorescence images of hsp-6p::gfp worms. Worms on control (ctrl) or hda-1 RNAi were untreated or treated with atp-2 RNAi or cco-1 RNAi. Scale bar, 200 µm. b Immunoblotting of GFP levels in hsp-6p::gfp worms. Tubulin serves as a loading control. c qRT-PCR measures endogenous transcript levels of hsp-6 in worms (n = 3 independent experiments, n ≈ 1000 worms per sample). d Representative fluorescence images of hsp-60p::gfp worms. Scale bar, 200 µm. e Representative fluorescence images of hsp-4p::gfp worms. Worms were fed with control or hda-1 RNAi and untreated or treated with tunicamycin. Scale bar, 200 µm. f Representative fluorescence images of hsp-16.2p::gfp worms. Worms on control or hda-1 RNAi were untreated or treated with heat shock. Scale bar, 200 µm. g Heat map showing the expression pattern of 283 genes whose expression was upregulated under mitochondrial stress. Genes with an adjusted p value < 0.05 calculated by Cuffdiff were selected as differentially expressed genes. The heat map is scaled by row and colored according to the z-score. Genes with a higher expression level than the mean are colored red; genes with a lower expression level than the mean are colored blue. h Gene ontology (GO) term analysis of the 283 hda-1-dependent genes, whose expression levels were upregulated during UPRmt. Results in (c) are shown as mean + SD, and P values were calculated by one-way ANOVA and Tukey’s multiple comparisons test (***P < 0.001). In (h), P values are DAVID modified Fisher exact P values, one-sided. Source data are provided as a Source Data file.HDAC plays an evolutionarily conserved role in removing acetyl moieties from core histones. Based on their structures and functions, HDACs have been grouped into three classes: class I and II HDACs share sequence homology in their catalytic domain17,18, while class III HDACs, as exemplified by SIR2 (silent information regulator 2), have a unique feature and require NAD+ as a cofactor for catalysis19–23. C. elegans class I HDACs include hda-1, hda-2, and hda-3, and class II HDACs include hda-4, hda-5, hda-6, hda-10, and hda-1124,25. Considering the high homology between class I and II HDACs, we sought to test if any other HDAC in C. elegans plays a similar role to hda-1 in mediating the mitochondrial stress response. We knocked down individual class I and II HDACs in C. elegans, and tested the animals’ ability to induce UPRmt upon mitochondrial perturbation. Interestingly, knockdown of other HDACs failed to suppress atp-2 RNAi-induced UPRmt (Supplementary Fig. 1c, d), suggesting a specific role of hda-1 in mediating UPRmt.Next, we generated transgenic strains expressing HDA-1::GFP fusion protein driven by the hda-1 promoter. Notably, we found that hda-1 expression was elevated in the nuclei of C. elegans intestinal cells under cco-1 or atp-2 RNAi treatment (Supplementary Fig. 1e, f). In addition, overexpression of HDA-1 slightly upregulated the basal level of UPRmt and further elevated the induction of UPRmt under mitochondrial stress conditions (Supplementary Fig. 1g, h).To further validate the function of hda-1 in UPRmt, we collected total RNAs from wild-type animals or hda-1-deficient animals in the presence or absence of mitochondrial perturbation, and performed RNA sequencing (RNA-seq) analysis to characterize hda-1-dependent genes during the mitochondrial stress response. 805 genes were significantly upregulated in wild-type animals under atp-2 RNAi treatment. Among them, 283 genes were significantly less induced in hda-1-deficient animals (Fig. 1g; Supplementary Data 1). Gene ontology (GO) functional enrichment analysis revealed that these hda-1-dependent genes are enriched in GO Biological Processes such as Metabolic process, Response to stress, Immune response and Cellular detoxification (Fig. 1h and Supplementary Data 2). This further supports a crucial role of hda-1 in modulating the mitochondrial stress response.HDA-1 interacts with DVE-1 to regulate UPRmtTo explore the molecular mechanism by which HDA-1 regulates the mitochondrial stress response, we immunoprecipitated HDA-1 from C. elegans expressing HDA-1::GFP using anti-GFP antibody and performed mass spectrometry analysis to search for HDA-1-interacting proteins. Notably, we found that HDA-1 interacts with the homeodomain-containing transcription factor DVE-1, a well-known component of the C. elegans UPRmt pathway7, under control RNAi or atp-2 RNAi treatment that perturbs mitochondrial function. Immunoprecipitation experiments on lysates from a transgenic C. elegans strain expressing both FLAG-tagged HDA-1 and GFP-tagged DVE-1 further validated the interaction between these two proteins (Fig. 2a).Fig. 2HDA-1 interacts with DVE-1 to regulate UPRmt.a Immunoprecipitation followed by immunoblotting reveals that HDA-1 interacts with DVE-1 in worms. b Representative fluorescence images of hda-1p::hda-1::gfp worms. Worms were fed with control, dve-1, atfs-1, or ubl-5 RNAi. Images were taken when animals reached young adult stage. Red arrows indicate the posterior region of the intestine where hda-1p::hda-1::gfp is induced or suppressed. Scale bar, 200 µm. c Representative fluorescence images of dve-1p::dve-1::gfp worms, without (control) or with over-expression of hda-1p::hda-1::flag. Scale bar, 200 µm. d, e Representative fluorescence images of hda-1p::hda-1::mCherry; dve-1p::dve-1::gfp worms. Worms were fed with cco-1 RNAi from L1 stage and imaged on day 1 of adulthood. d Fluorescence images show the bottom part of the intestine. Scale bar, 50 µm. e Fluorescence images reveal the nucleus of an intestinal cell. Scale bar, 5 µm. f Venn diagram showing overlap between HDA-1 (blue) and DVE-1 (purple) ChlP peaks in worms fed on control or atp-2 RNAi. g Nucleotide binding motifs of HDA-1 and DVE-1 in the presence or absence of mitochondrial stress. The P values of the motifs are also shown. h Analysis of HDA-1 ChIP peak signals in control worms or dve-1 RNAi worms in the presence or absence of atp-2 RNAi-induced mitochondrial stress. i Analysis of DVE-1 ChIP peak signals in control worms or hda-1 RNAi worms in the presence or absence of atp-2 RNAi-induced mitochondrial stress. j Venn diagram comparing genes upregulated in response to mitochondrial perturbation that are dependent on hda-1 (blue) or dve-1 (purple). k GO term analysis of UPRmt-upregulated genes that are dependent on both hda-1 and dve-1. P values are DAVID modified Fisher exact P values, one-sided. Source data are provided as a Source Data file.To further characterize the relationship between HDA-1 and DVE-1, we examined the expression profile of HDA-1 in hda-1p::hda-1::gfp animals. Interestingly, we noticed that the protein level of HDA-1 was dramatically reduced by dve-1 RNAi (Fig. 2b and Supplementary Fig. 2a). The reduction of HDA-1 protein level under dve-1-deficient conditions seems to be mediated by ubiquitin-mediated degradation, because RNAi knockdown of both ubq-1 and ubq-2, the only two ubiquitin-encoding genes in C. elegans26, abolished the reduction of HDA-1 upon dve-1 RNAi treatment (Supplementary Fig. 2b). Conversely, knockdown of HDA-1 also reduced the DVE-1 protein level (Supplementary Fig. 2c), while overexpression of HDA-1 elevated the protein level of DVE-1 (Fig. 2c). Furthermore, qPCR analysis of hda-1 or dve-1 transcript levels in the presence or absence of the other partner revealed that the transcription of hda-1 or dve-1 is not affected by the loss of its partner protein (Supplementary Fig. 2d). Taken together, these results suggest that HDA-1 and DVE-1 may interact and stabilize each other.Like the human protein SATB1, with which it shares high homology27, DVE-1 has been reported to show a ‘cage-like’ distribution that surrounds heterochromatin12. Consistent with the interaction between HDA-1 and DVE-1, HDA-1 showed a similar nuclear distribution pattern in the intestinal cells of C. elegans, and formed dense structures that colocalized with DVE-1 to surround chromatin regions stained with DAPI (Fig. 2d, e). To further investigate if HDA-1 and DVE-1 regulate the same group of genes, we carried out chromatin immunoprecipitation followed by sequencing (ChIP-seq) to identify candidate genes regulated by HDA-1- or DVE-1 in the presence or absence of mitochondrial stress. 6794 HDA-1-enriched peaks were uncovered by ChIP-seq analysis under normal conditions. Among these peaks, 60.5% (4,109 out of 6,794) were also recovered in DVE-1 ChIP-seq (Fig. 2f), an overlap significantly higher than the expectation (Monte Carlo
P value < 0.0001). More importantly, after mitochondrial perturbation, the overlap between HDA-1 and DVE-1 occupancies significantly increased to 86.1% (5711 out of 6634) (Fig. 2f). Intrigued by the large overlap between HDA-1 and DVE-1 binding loci, we determined the consensus binding motifs of HDA-1 and DVE-1 in the presence or absence of mitochondrial stress (Fig. 2g). Interestingly, HDA-1 and DVE-1 shared more binding motifs upon mitochondrial perturbation, which is consistent with the increased overlap of binding peaks under stress condition (Fig. 2f, g). RNA-seq analysis revealed that 284 and 374 genes were coregulated by HDA-1 and DVE-1 under normal conditions or mitochondrial stress (Supplementary Fig. 2e). We further analyzed the ChIP-seq results and found that the binding of HDA-1 to its target loci may depend on DVE-1, as knockdown DVE-1 suppresses the upregulation of HDA-1 peaks (Fig. 2h and Supplementary Data 3). Conversely, knockdown of HDA-1 did not affect the upregulation of DVE-1 peaks (Fig. 2i and Supplementary Data 3). Consistent with this result, it has been reported that SATB1, the DVE-1 homolog in higher eukaryotes, provides a docking site to recruit the histone deacetylase HDAC1 onto SATB1 target sequences28.Further analysis of the RNA-seq results indicated that 283 and 218 genes induced under mitochondrial stress conditions in C. elegans were dependent on hda-1 and dve-1, respectively (Fig. 2j). Among the 283 stress-activated hda-1-dependent genes, 169 genes (59.7%) contain HDA-1 ChIP peaks within ±500 bp of the transcription start site (Supplementary Data 1). In addition, among the 218 stress-activated dve-1-dependent genes, 175 genes (80.3%) contain DVE-1 ChIP peaks within ±500 bp of the transcription start site (Supplementary Data 1). More importantly, 139 genes required both hda-1 and dve-1 for induction, accounting for 49.1% of hda-1-dependent genes and 63.8% of dve-1-dependent genes. These 139 genes that depend on both hda-1 and dve-1 were enriched for the GO terms Immune response, Response to stress, Metabolic process and Protein folding (Fig. 2k). We then employed qPCR to validate some of the key genes involved in HDA-1- and DVE-1-dependent regulation. We selected genes associated with some of the most strongly enriched GO terms (Fig. 2k). Knockdown of hda-1 or dve-1 significantly suppressed the induction of cdr-4, pgp-1, nhr-115, M04C3.2 (GO term: Immune response), dnj-10, djr-1.2 (GO term: Stress response), cyp-33C8, ipla-3, hmgs-1, tars-1 (GO term: Single-organism metabolic process) and cua-1 (GO term: Single-organism localization) upon mitochondrial stress (Supplementary Fig. 3a). Moreover, a hypomorphic allele of hda-1 also suppressed the induction of stress response genes (Supplementary Fig. 3b). Taken together, these results suggested a role of HDA-1, in coordination with DVE-1, to activate UPRmt and UPRmt-induced innate immune response and metabolic reprogramming.HDA-1 is required for UPRmt-mediated innate immunityMitochondrial function is greatly challenged by pathogens and xenobiotics that metazoans encounter in their natural habitats2. As a surveillance mechanism that defends against natural infection, UPRmt not only initiates mitochondrial protective responses to induce the expression of mitochondrial chaperones and proteases, but also activates innate immune responses. Our finding that HDA-1 regulates genes related to the GO terms Immune response and Metabolic process led us to further examine its critical function in animal fitness during mitochondrial stress. We first employed irg-1p::gfp transgenic animals, a reporter strain for pathogen-infected response and an indicator for the induction of innate immune response29. The irg-1p::gfp reporter worms were challenged with a Pseudomonas aeruginosa strain isolated from natural habitats harboring wild C. elegans populations, which has been shown to perturb mitochondrial function and induce expression of the UPRmt reporter hsp-6p::gfp2. In wild-type animals, irg-1p::gfp expression was induced upon Pseudomonas infection, whereas deficiency of hda-1 significantly suppressed the induction of irg-1, to a similar extent as deficiency of dve-1 (Fig. 3a; Supplementary Fig. 4a, b). Knockdown of hda-1 or dve-1 also suppressed the induction of several other immune response genes upon P. aeruginosa infection (Fig. 3b). In addition, deficiency of hda-1 or dve-1 reduced the survival rate of worms and promoted the accumulation of P. aeruginosa in C. elegans when they were exposed to Pseudomonas (Fig. 3c-e and Supplementary Fig. 4c, d)30. Conversely, overexpression of HDA-1 or DVE-1 promoted animal survival and reduced the accumulation of P. aeruginosa upon pathogen infection (Fig. 3f and Supplementary 4 f, e).Fig. 3HDA-1 is required for UPRmt-mediated innate immunity.a Representative fluorescence images of irg-1p::gfp worms. Worms on control, hda-1 or dve-1 RNAi were untreated or exposed to P. aeruginosa. Scale bar, 200 µm. b qRT-PCR measurement of the endogenous mRNA levels of immune response genes in wild-type worms raised on control RNAi, hda-1 RNAi or dve-1 RNAi and treated with P. aeruginosa. (n = 3 independent experiments, n ≈ 1000 worms per sample). c Survival curves of control, hda-1, or dve-1 RNAi worms in the P. aeruginosa slow-killing assay. n = 50 worms for each condition. d Representative fluorescence images showing accumulation of P. aeruginosa (GFP) in intestines of worms fed on control, hda-1 or dve-1 RNAi. Scale bar, 200 µm. e CFU (colony forming units) were quantified for experiments in (d). n = 30 worms for each sample. f Survival curves of wild-type, hda-1p::hda-1::gfp, or dve-1p::dve-1::gfp worms in the P. aeruginosa slow-killing assay. n = 50 worms for each sample. g Representative fluorescence images of irg-1p::gfp worms raised on control, hda-1 or dve-1 RNAi and treated with Rhodococcus. Scale bar, 200 µm. h Survival curves of control, hda-1, or dve-1 RNAi worms in the Rhodococcus killing assay. n = 55, 60, 59 worms respectively for each sample. Results in (b) are shown as mean + SD and results in (e) are shown as mean ± SD, P values were calculated by one-way ANOVA and Tukey’s multiple comparisons test (*P < 0.05; **P < 0.01; ***P < 0.001). Source data are provided as a Source Data file.Next, we asked if the function of HDA-1 can be partially mediated by its target genes. We tested whether hmgs-1, one of the 139 genes coregulated by HDA-1 and DVE-1, can modulate the innate immune response upon P. aeruginosa infection. hmgs-1 encodes an HMG-CoA synthase, which has been shown to be required for the induction of UPRmt 2. Knocking down hmgs-1 by RNAi reduced the survival rate of worms and promoted the accumulation of P. aeruginosa when worms were exposed to this pathogen (Supplementary Fig. 4g, h), similar to the effect observed under hda-1 RNAi. However, overexpression of hmgs-1 could not rescue the loss of immunity caused by hda-1 knockdown (Supplementary Fig. 4i, j), which suggests that expression of this target gene alone is not sufficient to rescue phenotypes caused by hda-1 deficiency.Moreover, hda-1 or dve-1 RNAi also suppressed the activation of the immune response and reduced the survival rate when the animals were challenged with another mitochondrial insult, a Rhodococcus strain isolated from the natural habitat of C. elegans (Fig. 3g, h; Supplementary Fig. 4k). Collectively, these results indicate that hda-1 is required for UPRmt-activated innate immune response.HDA-1 affects animal aging and age-related pathologyGenes important for mitochondrial function have been recovered from a genome-wide RNAi screen in C. elegans for determining worm lifespans4. Moreover, perturbations of mitochondrial electron transport in various species have been shown to extend their lifespans3,31–33. Since HDA-1 plays a critical role in mitochondrial stress response, we suspected that it may also play a role in mitochondrial stress-induced lifespan extension. Therefore, we examined the lifespans of wild-type or hda-1-deficient worms in the presence or absence of mitochondrial perturbation. Interestingly, hda-1 RNAi greatly reduced the lifespan extension induced by feeding worms with atp-2 RNAi to inhibit mitochondrial function (Fig. 4a). We noticed that hda-1 RNAi, in the absence of mitochondrial perturbation, shortened worm lifespan (Fig. 4a). However, a previous study reported that hda-1 RNAi does not affect C. elegans lifespan34. We speculated that there may be two reasons for this discrepancy: (1) different IPTG induction times were used for dsRNA expression, which may affect hda-1 RNAi efficiency (Supplementary Fig. 5a); (2) the previous study used liquid culture to carry out lifespan analysis, whereas we used solid agar. To further dissect the function of hda-1 in lifespan regulation, we measured the lifespan of worms carrying the hda-1(e1795) hypomorphic allele and found that this hypomorphic mutation also shortened worm lifespan (Supplementary Fig. 5b). It should be noted that we also observed a shortened lifespan when eat-2(ad1116) mutant animals were fed with hda-1 RNAi (Supplementary Fig. 5c). eat-2(ad1116) mutants had reduced pharyngeal pumping, therefore mimicking the caloric restriction effect and promoting lifespan extension35. It has been reported that eat-2 animals have decreased mitochondrial potentials36 and activate the ZIP-2 pathway, which contributes to the improvement of mitochondrial integrity37. Therefore, it will be interesting to understand if hda-1 regulates lifespan through maintenance of mitochondrial integrity, or through a more general mechanism.Fig. 4HDA-1 affects animal aging and age-dependent accumulation of protein aggregates.a Lifespan analysis of worms raised on the indicated RNAis. n = 130 worms per condition. b Mobility analysis of unc-54p::Q35::yfp worms on control, hda-1 or dve-1 RNAi (n = 2 independent experiments, 30 worms per condition). c Mobility analysis of unc-54p::Q35::yfp and unc-54p::Q35::yfp; hda-1p::hda-1::flag worms on day 8 of adulthood (n = 2 independent experiments, 30 worms per condition). d Representative fluorescence images of unc-54p::Q35::yfp worms fed with control or hda-1 RNAi. The images were taken on day 5 of adulthood. Scale bar, 200 µm. e Quantification of Q35 aggregates in (d). The numbers of Q35 aggregates in the body muscles (excluding the head and tail) of each worm were counted. n = 12 worms per condition. f Representative fluorescence images of unc-54p::Q35::yfp, unc-54p::Q35::yfp; hda-1p::hda-1::flag or unc-54p::Q35::yfp; dve-1p::dve-1::gfp worms on day 5 of adulthood. Scale bar, 200 µm. g Quantification of Q35 aggregates in (f). The numbers of Q35 aggregates in the body muscles (excluding the head and tail) of each worm were counted. n = 20 worms per sample. Results in (e, g) are shown as mean ± SD. In (e), the P value was calculated by two-tailed Student’s t test (*P < 0.05). In (g), P values were calculated by one-way ANOVA and Tukey’s multiple comparisons test (***P < 0.001). Source data are provided as a Source Data file.A progressive decline of responses toward proteostatic stress occurs with aging, leading to the toxic accumulation of protein aggregates, such as proteins containing polyglutamine (polyQ) repeats38. In line with the role of HDA-1 in mediating aging responses, knockdown of hda-1 enhanced polyQ toxicity and greatly impaired animal movement (Fig. 4b). Conversely, overexpression of HDA-1 suppressed polyQ toxicity and improved animal movement (Fig. 4c). In addition, knockdown of hda-1 increased the number of polyQ aggregates in C. elegans muscle cells (Fig. 4d, e), whereas overexpression of hda-1 or dve-1 reduced the accumulation of polyQ aggregates in aged animals (Fig. 4f, g). Taken together, these results indicate that HDA-1 plays an important role in mediating the beneficial impact of UPRmt to promote lifespan extension in C. elegans. The data also suggest the therapeutic potential of activating HDAC signaling to manage certain age-related diseases, such as neurodegeneration.HDA-1 regulates mitochondrial stress response in mammalsThe beneficial effect of HDA-1 in modulating C. elegans mitochondrial homeostasis encouraged us to examine its physiological relevance in higher eukaryotes. HDAC1 forms homo- or heterodimers together with HDAC2 in various transcription regulator complexes such as NuRD and CoREST to repress or activate gene expression39,40. Interestingly, we found that the expression levels of HDAC1/2 strongly correlate with SATB2 (mammalian ortholog of DVE-1), the mitochondrial chaperones HSPA9 and HSPD1, the mitochondrial proteases LONP1 and YME1L1, asparagine synthetase ASNS, and mitochondrial import inner membrane translocase TIMM17A in various human and rhesus monkey tissues (Fig. 5a; Supplementary Fig. 6). Furthermore, consistent with the interaction between HDA-1 and DVE-1 in C. elegans, HDAC1 also interacted with SATB2 in mammalian cells (Fig. 5b). Taken together, these results suggest that HDAC1/2 may play an evolutionarily conserved role in regulating mitochondrial homeostasis.Fig. 5Transcript levels of human HDAC1/2 strongly correlate with the expression of UPRmt genes.a Pearson’s correlation of HDAC1, HDAC2, SATB2 and UPRmt mRNA levels in human heart, colon, kidney, liver, lung, muscle, and testis tissues. Red circles indicate positive correlation and blue circles indicate negative correlation. The size of the circle corresponds to the correlation coefficient. b Immunoprecipitation followed by immunoblotting shows that HDAC1 interacts with SATB2 in HEK293T cells. Source data are provided as a Source Data file.To validate the function of HDAC1/2 in mediating UPRmt during mitochondrial stress, we first treated HEK293T cells with sodium butyrate (NaBt), a chemical inhibitor of HDAC (Supplementary Fig. 7a). NaBt administration suppressed the induction of LONP1, HSPA9, YME1L1, and HSPD1 in cells treated with antimycin A, a well-known inhibitor of mitochondrial electron transport chain complex III (Fig. 6a; Supplementary Fig. 7b). However, it is possible that chemical inhibitors may indiscriminately target several HDACs, and the effect may not be specifically due to HDAC1/2 inhibition. To directly validate the role of HDAC1/2 in modulating mitochondrial homeostasis and UPRmt activation, we used siRNA to knock down HDAC1/2 or SATB2 in HeLa cells and stained mitochondria with MitoTracker. Compared to wild-type cells treated with antimycin A, disrupted mitochondrial morphology and higher mitochondrial fusion were observed in HDAC1/2- or SATB2-deficient cells challenged with antimycin A (Fig. 6b and Supplementary Fig. 7c). In addition, the induction of human UPRmt genes such as HSPD1, LONP1, ASNS, and YME1L1 was significantly suppressed in HDAC1/2 or SATB2 knockdown cells challenged with mitochondrial insult (Fig. 6c and Supplementary Fig. 7d). Collectively, these results confirmed an evolutionarily conserved role of HDAC1/2 in mitochondrial surveillance and UPRmt activation.Fig. 6The functions of HDAC1/2 and SATB2 in UPRmt activation are conserved in mammals.a qRT-PCR measures mRNA levels of LONP1 and HSPA9 in HEK293T cells cultured under indicated conditions. b Representative fluorescence images of HeLa cells cultured under the indicated conditions. Scale bar, 10 µm. c qRT-PCR measures mRNA levels of UPRmt genes in HEK293T cells cultured under the indicated conditions. (a, c) n = 3 independent experiments. Results in (a, c) are shown as mean + SD. P values were calculated by two-way ANOVA and Tukey’s multiple comparisons test (**P < 0.01; ***P < 0.001). Source data are provided as a Source Data file.DiscussionDuring mitochondrial stress, a mitochondrion-to-nucleus communication named UPRmt is activated to initiate the transcriptional induction of genes involved in stress and immune responses, as well as those involved in metabolic reprogramming. Epigenetic regulation has emerged as another layer of regulation to relay mitochondrial stress signals to the expression of stress response genes12–14,41. Here we report that histone deacetylase HDA-1 coordinates with DVE-1 to activate UPRmt in C. elegans. In addition, we provide evidence to show that its mammalian homologs HDAC1/2 play a conserved role to act in conjunction with SATB2 (mammalian ortholog of DVE-1) to mediate mitochondrial homeostasis.Three-dimensional organization of chromatin architecture is important for regulating gene expression42–44. Interestingly, it has been well documented that SATB1 serves as a genome organizer to provide a landing platform for chromatin-remodeling enzymes such as HDAC27,28,45. Changes of chromatin organization such as the formation of chromatin loops will bring distant coregulated genes into close proximity and affect the accessibility of genomic loci. For instance, SATB1 has been reported to promote chromatin loop formation in T cells and activate cytokine gene expression27. Studies of HDAC1 have revealed that it can form chromatin-remodeling complexes such as NuRD and CoREST to regulate gene expression39,40. It will be important to examine the dynamic changes of chromatin architecture and characterize the function of HDA-1 and DVE-1 for rapid transcriptional induction in C. elegans upon mitochondrial perturbation. It will also be crucial to understand in the future how mitochondrial stress signals are relayed to HDA-1 activity.Activation of UPRmt not only upregulates the transcription of mitochondrion-specific chaperones and proteases to buffer the mitochondrial folding environment, but also initiates the transcriptional program to activate or suppress the expression of genes in different metabolic processes. Therefore, removal of the acetyl groups on histones through HDA-1 may indirectly adjust cellular metabolism to compensate for mitochondrial dysfunction by feeding additional acetyl-coA into metabolic reactions. Similarly, it has been shown that glucose or serum deprivation can reduce the levels of acetylated histones H3 and H4 in mammalian cells46. It will be of interest in the future to examine if global or specific acetylated histone marks are affected during mitochondrial perturbation.Moderate mitochondrial stress can initiate a beneficial hormetic stress defense to promote innate immunity and lifespan extension2,4,38,47,48, which suggests an intimate link between stress adaptation and the aging process49. Interestingly, we found that silencing of hda-1 suppressed mitochondrial stress-induced immune response and longevity. More importantly, overexpression of hda-1 delayed age-related accumulation of protein aggregates and polyglutamine toxicity in C. elegans. In addition, systematic correlation analysis in rhesus monkey and human strongly points to a conserved function of HDAC1/2 in regulating mitochondrial hormesis in primates. Therefore, future genetic or pharmacological treatments to target this pathway may provide therapeutic potential for the treatment of age-related diseases.MethodsStrains and cultureSJ4100(zcIs13[hsp-6::gfp]), SJ4058(zcIs9[hsp-60p::gfp]), CL2070 (dvIs70[hsp-16.2::gfp]), SJ4005(zcIs4[hsp-4p::gfp]), CB5535(hda-1(e1795)), SJ4197(zcIs39[dve-1p::dve-1::gfp]), AU133(agIs17[irg-1p::gfp]), AM140(rmIs132 [unc-54p::Q35::yfp]), SS104(glp-4(bn2)), and N2 wild-type C. elegans were obtained from the Caenorhabditis Genetics Center.The following strains were generated in our lab: YSL1(liuls1[hda-1p::hda-1::flag; odr-1p::dsRed]), YSL2(liuls2[hda-1p::hda-1::gfp; odr-1p::dsRed]), YSL3(liuls1[hda-1p::hda-1::flag; odr-1p::dsRed]; zcIs39[dve-1p::dve-1::gfp]), YSL4(zcIs39[dve-1p::dve-1::gfp]; liuEx1[hda-1p::hda-1::mCherry; odr-1p::dsRed]), YSL5(liuls1[hda-1p::hda-1::flag; odr-1p::dsRed]; zcIs13[hsp-6p::gfp]), YSL6(glp-4(bn2); liuls2[hda-1p::hda-1::gfp; odr-1p::dsRed]), YSL7(glp-4(bn2); zcIs39[dve-1p::dve-1::gfp]), YSL8(glp-4(bn2); rmIs132[unc-54p::Q35::yfp]), YSL9(glp-4(bn2); liuls1[hda-1p::hda-1::flag; odr-1p::dsRed]; rmIs132 [unc-54p::Q35::yfp]), YSL10(glp-4(bn2); zcIs39[dve-1p::dve-1::gfp]; rmIs132 [unc-54p::Q35::yfp]), and YSL11(glp-4(bn2); liuEx2[gly-19p::hmgs-1::flag; mec-7p::rfp]).HEK293T cells and HeLa cells were obtained from ATCC. Cells were cultured in DMEM medium supplemented with 10% (v/v) fetal bovine serum at 37 °C. SATB2 plasmid was constructed by PCR amplification of cDNA from total HEK293T RNA and ligated into the pCDNA3.3 vector. HDAC1 plasmid was a gift from Prof. Jiemin Wong. Plasmids were transfected into cells by Lipofectamine 2000 Reagent.Primer sequences for plasmid construction are provided in Supplementary Table 1.RNA interferenceMost RNAi clones were obtained from the Ahringer library, whereas hda-3 and hda-5 RNAi clones were generated by PCR amplification of worm cDNA and ligated into L4440 vector. The clones were then transformed into HT115 competent cells. RNAi clones were grown in LB at 37 °C overnight. 20X concentrated bacterial culture solution was seeded on to worm plates with 1.2 mg/ml IPTG. Dried plates were kept at room temperature overnight to allow IPTG induction of double-stranded RNA (dsRNA) expression. Synchronized L1 worms were raised on RNAi plates at 20 °C. For double RNAi experiments, mixed bacterial culture solutions were seeded onto RNAi plates, or the second RNAi bacterial culture solution was seeded 30 h after worms were cultured on the first RNAi clones.For siRNA knockdown in mammalian cells, 50 pmol of siRNA was transfected into a 6-well plate by Lipofectamine RNAi MAX Reagent.Induction of UPRmtFor induction of UPRmt by RNAi, synchronized L1 worms raised on atp-2 or cco-1 RNAi were imaged when they reached adulthood. For double RNAi experiments, 20X concentrated atp-2 or cco-1 RNAi bacterial culture (pre-induced by 0.2 mg/ml IPTG) was seeded 30 h after worms were cultured on the first RNAi. After another 30 h, worms were imaged.For induction of UPRmt by antimycin A in cells, HEK293T cells with 80% confluency were treated with 20 µg/ml antimycin A (Sigma #A8674) for 21 h.Induction of UPRER and HSRFor induction of UPRER by tunicamycin, synchronized L1 worms were raised on 6 cm RNAi plates for 43 h. 300 µl M9 buffer containing 18 µg tunicamycin was then spread across the entire surface of the plate. Induction of UPRER was examined after 12 h.To induce the heat shock response (HSR), synchronized L1 worms were raised on 6 cm RNAi plates for 34 h, cultured at 37 °C for 1 h and then transferred to 20 °C for another 21 h before examining the induction of HSR.Induction of innate immune responseFor induction of the innate immune response by pathogens, synchronized L1 worms were raised for 24 h on worm plates before exposure to P. aeruginosa or Rhodococcus. Worms were imaged on day 2 of adulthood.Sodium butyrate treatmentCells were treated with 5 mM sodium butyrate (Sigma #303410) for 24 h. For sodium butyrate and Antimycin A double treatment, Antimycin A was added into the cell culture medium 3 h after sodium butyrate treatment.MicroscopyWorms were picked into 100 mM NaN3 droplets on 2% agarose pads and imaged by a Zeiss Imager M2 microscope. Nuclear localization of worms was imaged by a Zeiss LSM880 microscope. Mitochondrial morphology of HeLa cells was imaged by a PerkinElmer Operetta CLS™. Comparable images were captured with the same exposure time and magnification. GFP fluorescence was quantified by ImageJ. For imaging mitochondria, HeLa cells were treated with 50 µg/ml antimycin A (Sigma #A8674) for 4 h and then stained with MitoTracker (Invitrogen #M7512).RNA isolation and real-time PCRSynchronized late L4 worms or HEK293T cells were collected and resuspended with trizol reagent (Cwbiol #cw0580A). Worm samples were frozen and homogenized three times in liquid nitrogen. Total RNA was isolated by chloroform extraction, precipitated with isopropanol, then washed with 75% (v/v) ethanol. cDNA was synthesized with reverse transcription kits (Transgen #AT311). Quantitative PCR was carried out using SYBR Green PCR Master Mix (Bio-Rad #1725121). For quantification, C. elegans transcripts were normalized to rpl-32, and transcripts from HEK293T cells were normalized to ACTB. Primer sequences for quantitative RT-PCR are provided in Supplementary Table 2.ImmunoblottingWorms or cells were resuspended with SDS loading buffer (100 mM Tris-HCl pH 6.8, 4% SDS, 20% glycerol, 10% β-mercaptoethanol, 0.004% bromophenol blue) and boiled at 95 °C for 10 min. Samples were separated by SDS-PAGE and transferred onto a PVDF membrane (Bio-Rad). After blockeding with 5% milk-TBST, the membrane was probed with the designated primary antibodies and secondary antibodies. Primary antibodies used in this study: anti-GFP (Sungene #KM8009, 1:1000 dilution), anti-tubulin (Abcam #ab6161, 1:1000 dilution), anti-Myc (CST #2276, 1:1000 dilution), anti-FLAG (Sigma #F7425, 1:2000 dilution) and anti-Acetylated-Lysine (CST #9441, 1:1000 dilution). Membranes were developed with the enhanced chemiluminescence method (Thermo) and visualized using the Tanon 5200 chemical luminescence imaging system.ImmunoprecipitationFor worm samples, 30,000 young adult worms with the indicated genotype were washed off by M9 buffer and resuspended in 3 ml lysis buffer (50 mM tris-HCl pH 8.0, 137 mM NaCl, 1% Triton X-100, 1 mM EDTA, 10% glycerol, proteinase inhibitor). Samples were homogenized with a glass homogenizer and sonicated. For HEK293T cells, cells in one 10 cm dish with 90% confluence were washed with 1X PBS buffer, then resuspended in 1 ml lysis buffer and sat on ice for 30 min. The worm or cell lysate was centrifuged at 20,000 × g for 15 min. The supernatant was then transferred into a new tube and rotated at 4 °C overnight in the presence of the designated antibody anti-GFP antibody (Abcam #ab290, 1 µl per sample), anti-FLAG magnetic beads (Sigma #M8823, 40 µl per sample), anti-Myc magnetic beads (Bimake #B26302, 20 µl per sample). For anti-GFP immunoprecipitation, protein G beads (Invitrogen #10004D, 40 µl per sample) were subsequently added to each sample and rotated at 4 °C for additional 2 h. After binding, the beads were washed three times with lysis buffer and boiled in 50 µl 2X SDS Laemmli buffer (4% SDS, 20% glycerol, 10% 2-mercaptoethanol, 0.02% bromophenol blue, 0.125 M Tri-HCL, pH 6.8) at 95 °C for 10 min.Lifespan measurementSynchronized L1 worms were raised on fresh plates seeded with the indicated bacteria. After reaching adulthood, worms were transferred to fresh plates every 2 days and monitored for survival. A worm that did not respond to three gentle touches on the head and displayed no pharyngeal pumping was considered dead. Those that died due to internal hatching, ruptured vulvae, or crawling off the agar were removed.P. aeruginosa slow-killing assayP. aeruginosa was cultured overnight in LB containing 50 µg/ml kanamycin at 37 °C. To prepare P. aeruginosa plates, 10 µl P. aeruginosa was seeded onto 3.5 cm slow-killing agar plates50 and spread slightly to a small circle using a sterile L spreader. Plates were air-dried, incubated at 37 °C for 24 h and then incubated at room temperature for another 24 h. Synchronized L1 glp-4(bn2) worms were first raised on RNAi plates at 20 °C for 49 h, then 50 worms were randomly picked onto P. aeruginosa plates and cultured at 25 °C. To score survival rate of N2 wild-type worms, 5-fluoro-2′-deoxyuridine (FUdR) (100 µg/ml) was added in the P. aeruginosa plates. Living worms were counted every 24 h. Those crawling off the agar were removed.Rhodococcus survival assayRhodococcus was cultured overnight in LB at 25 °C. A 30X concentrated solution of bacteria was seeded on to 6 cm slow-killing agar plates. Synchronized L1 glp-4(bn2) worms were first raised on RNAi plates at 20 °C for 49 h, then about 50 worms were randomly picked onto dried Rhodococcus plates and cultured at 25 °C. Living worms were counted every 24 h. Those crawling off the agar were removed.P. aeruginosa intestinal accumulation assayThe Pseudomonas aeruginosa strain PA01 with pMF230 (Addgene_62546), which expresses GFP, was provided by Dr. Huanqin Dai. To prepare PA01(GFP) plates, overnight cultures of PA01(GFP) were spread across the entire surface of the slow-killing plates. Plates were air-dried, incubated at 37 °C for 24 h and then incubated at room temperature for 24 h. Synchronized worms fed with RNAi at 20 °C for 53 h were washed twice in M9 buffer and transferred onto PA01(GFP) plates. Worms were imaged after another 45 h.Quantification of PA01(GFP) colony-forming units (CFU)Worms were exposed to PA01(GFP) in the same conditions as in the P. aeruginosa intestinal accumulation assay. After a defined period, worms were washed three times in M9 buffer and rotated for 1 h in M9 buffer containing kanamycin (1 mg/ml). After three additional washes with M9 buffer, 30 worms were picked into 100 µl M9 buffer in a 1.5 ml tube and grinded with a motorized pestle. Lysates were serially diluted, and 5 µl lysate solution was plated on LB plates containing carbenicillin (50 µg/ml). After overnight incubation at 37 °C, colonies with GFP fluorescence were counted.Mobility assayOn day 8 of adulthood, worms with Q35 polyglutamine expression were touched two times on the head and tail. Animals which could move and change their physical position were counted. Worms were randomly selected and tested under each condition.ChIP and ChIP-seq data analysisFor each ChIP sample, 240,000 worms with HDA-1-GFP or DVE-1-GFP overexpression were harvested. After cross-linking with 2% formaldehyde for 40 min and washing 3 times by cold PBS, worm pellets were resuspended in FA buffer (50 mM HEPES/KOH pH 7.5, 1 mM EDTA pH 8.0, 1% Triton-X-100, 0.1% sodium deoxycholate) with 150 mM NaCl and proteinase inhibitor, then homogenized, and sonicated. Next, the samples were centrifuged at 20,000 × g for 20 min at 4 °C. After dilution with FA buffer (containing 150 mM NaCl), 1% (V/V) supernatant was kept as the Input, and the rest was incubated with GFP-Trap agarose (ChromoTek #gta-20, 50 µl per sample) at 4 °C for 20 h. Then the GFP-Trap agarose was washed twice with 150 mM NaCl FA buffer, once with 1 M NaCl FA buffer, twice with 500 mM NaCl FA buffer, once with LiCl buffer (10 mM Tris-HCl pH 8.0, 1 mM EDTA pH 8.0, 1% NP40, 1% sodium deoxycholate, 250 mM LiCl) and three times with TE buffer. The bound fraction was eluted from the agarose by TE buffer (containing 1% SDS, 250 mM NaCl) and the elute was decross-linked together with the Input overnight at 65 °C. The samples were then digested by protease K for 2 h at 55 °C51. DNA was extracted with a ChIP DNA Clean & Concentrator kit (Zymo Research #D5201) and DNA libraries were prepared using an NEB kit (NEB # E7370). For ChIP-seq data analysis, adaptor sequences were filtered out of reads using an in-house script, then Trim Galore was used to do the quality control with the following parameters: -q 20 --fastqc --illumina --stringency 6 -e 0 --length 50 --trim-n --paired. Processed reads were mapped to the C. elegans genome (ce11) using BWA (version 0.7.13-r1126)52 with default parameters. Only uniquely mapping reads that were properly paired with mapping quality ≥20 and mismatches <8 were retained. PCR duplicates were removed by Picard (version 2.17.6) (http://broadinstitute.github.io/picard). MACS2 (version 2.1.1)53 was used to obtain peak regions with the following parameters: -g ce -f BAMPE --broad --SPMR. ChIP-seq signals were calculated using deepTools (version 2.4.2)54 by normalizing the read coverage to 10 million reads then subtracting input coverage from the ChIP samples to get the final ChIP-seq signals. Overlaps between peaks were determined by the BEDTools (version 2.27.1)55 intersect command. Motif analysis was performed by HOMER (version 4.10-0)56 with the following parameters: ce11 -gc -size given. Bwtool57 was used for peak signal comparison between groups.mRNA-seq sample preparation and data analysisFor worm samples, about 1000 synchronized L1 worms were fed with control, hda-1 RNAi or dve-1 RNAi for 24 h, followed by atp-2 RNAi for 48 h. Worms were washed and resuspended with trizol reagent for total RNA isolation.Reads were aligned to the C. elegans genome (ce11) using TopHat2 (version 2.1.1)58 with the following parameters: --read-mismatches 6 --read-edit-dist 6 --min-anchor 8 --segment-length 26. Uniquely mapping reads were used for further analysis. Gene expression levels and differentially expressed genes were generated by Cufflinks (version 2.2.1)59. GO term enrichment analysis was performed by DAVID60,61 using GOTERM_BP_2 category.Expression correlation analysesCorrelation analysis of the expression levels of HDAC1, HDAC2, SATB2 and UPRmt genes in different human tissues was performed using data from GTEx (GTEx v7 All Tissues RNA-Seq, https://www.gtexportal.org/home/datasets). Correlation analysis of the expression levels of HDAC1, HDAC2, SATB2, and UPRmt genes in different rhesus monkey tissues was performed using data from RhesusBase62 (http://rhesusbase.cbi.pku.edu.cn/download/download.jsp). In each tissue, the Pearson’s correlation for each pair of genes was calculated using the gene expression levels among all samples in human or rhesus macaque stated above.Statistics and reproducibilityStatistical analyses were performed with GraphPad Prism 5.0. Results were expressed as mean ± SD or mean + SD. ANOVA and Tukey’s multiple comparisons test or Student’s t test was used to calculate the P values. For details of the particular statistical analyses employed, precise P values and statistical significance for all graphs, see figure legends. Experiments yielding quantitative data for statistical analysis were performed independently at least twice, all with similar results. Micrographs and immunoblotting images shown in the figures are representative of three independent experiments, all with similar results.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2Supplementary Data 3Reporting Summary | nature communications | [
"Article"
] | [
"Cell biology",
"Molecular biology"
] | IntroductionMitochondria play essential roles in biology Originating from endosymbiosis proteobacteria provide cells ATP participate in biological processes calcium homeostasis immune responses cell death1 normal function mitochondria challenged by stimuli pathogens xenobiotics2. Mitochondrial dysfunction linked with diseases cells employ quality control mechanisms surveil function initiate protective programs damage mechanism mitochondrial unfolded protein response (UPRmt), relays stress signals to transcription of genes In C. elegans transcriptional response governed by factors ATFS-1 DVE-17 chromatin modifiers factors modulate UPRmt.Eukaryotic chromosomes three-dimensional higher-order structures Chromatin modifier proteins induce changes chromatin architecture control accessibility DNA sequences to transcription machinery post-translational modifications acetylation of histones H3 H4 dictate chromatin state affect gene expression15 Histone acetylation states regulated by chromatin modifiers histone acetyltransferase) deacetylase control gene expressionUPRmt requires activation genes mitochondrial stress cellular metabolic state biological importance chromatin modifier proteins warrants exploration histone deacetylase HDA-1 key regulator UPRmt in C. elegans HDA-1 acts with DVE-1 transcription genes mitochondrial stress response immune response metabolism expression profiles rhesus monkey human transcript levels primate HDAC1/2 correlate with expression UPRmt genes results highlight function HDAC1/2 regulating mitochondrial stress response beneficial outcomes required for UPRmt activationTo genome-wide RNA interference) screen C. elegans UPRmt reporter strains hsp-6p fed age-synchronized wild-type worms expressing GFP fluorescent reporters with dsRNA-expressing bacteria tested for GFP expression mitochondrial perturbation hda-1 recovered knockdown of hda-1 impaired activation UPRmt reporter hsp-6p::gfp atp-2 Deficiency hda-1 elevation endogenous transcript levels hsp-6 under mitochondrial stresslack hda-1 impaired activation UPRmt reporter hsp-60p::gfp during mitochondrial perturbation (Fig. 1d). test hda-1 required for UPRmt activation knocked down hda-1 in stress hsp-4p::gfp heat shock hsp-16.2p::gfp challenged with ER inhibitor tunicamycin or heat shock treatment Deficiency hda-1 affect activation ER stress heat shock response (Fig. 1e f results indicate hda-1 mitochondrial stress response 1hda-1 required for UPRmt activation images hsp-6p::gfp worms untreated treated with atp-2 or cco-1 RNAi Immunoblotting of GFP levels in hsp-6p::gfp worms control qRT-PCR measures transcript levels hsp-6 in worms 3 experiments 1000 worms hsp-60p::gfp worms hsp-4p::gfp worms fed control hda-1 RNAi untreated treated tunicamycin hsp-16.2p::gfp worms untreated treated with heat shockHeat map expression 283 genes upregulated under mitochondrial stress Genes adjusted p value < 0.05 Cuffdiff selected differentially expressed heat map scaled by colored z-score higher expression red lower blue Gene ontology analysis 283 hda-1-dependent genes expression upregulated during UPRmt Results mean + SD P values calculated one-way ANOVA Tukey’s multiple comparisons test (***P < 0.001) P values DAVID modified Fisher exact P values one-sided Source data file.HDAC removing acetyl moieties from core histones HDACs grouped three classes class I II share sequence homology catalytic class III unique feature require NAD+ cofactor C. elegans class I HDACs include hda-1 hda-2 hda-3 class II hda-4 high homology class I II HDACs test other HDAC C. elegans similar role hda-1 mitochondrial stress response knocked down class I II HDACs tested induce UPRmt mitochondrial perturbation knockdown HDACs suppress atp-2 RNAi-induced UPRmtrole hda-1 mediating UPRmt generated transgenic strains expressing HDA-1::GFP fusion protein hda-1 promoter hda-1 expression elevated in C. elegans cells under cco-1 atp-2 RNAi treatment overexpression HDA-1 upregulated level UPRmt elevated induction under mitochondrial stress conditions 1g hda-1 collected RNAs from wild-type hda-1-deficient animals performed RNA sequencing analysis hda-1-dependent genes stress response 805 genes upregulated in wild-type animals under atp-2 RNAi treatment 283 genes less induced in hda-1-deficient animals enrichment analysis hda-1-dependent genes enriched in Biological Processes Metabolic Response stress Immune response Cellular detoxification supports role hda-1 mitochondrial stress response.HDA-1 interacts with DVE-1 UPRmtTo immunoprecipitated HDA-1 from C. elegans expressing performed mass spectrometry analysis for HDA-1-interacting proteins HDA-1 interacts with homeodomain-containing transcription factor DVE-1 component C.elegans UPRmt pathway7 treatment mitochondrial function Immunoprecipitation experiments lysates transgenic C. elegans strain FLAG HDA-1 GFP DVE-1 validated interaction proteins (Fig. interacts DVE-1 UPRmt Immunoprecipitation immunoblotting HDA-1 interacts DVE-1 worms fluorescence images hda-1p::hda-1::gfp worms fed control dve-1 atfs-1 ubl-5 RNAi young adult stage Red arrows indicate posterior region intestine hda-1p:::gfp induced suppressed images dve-1p::dve-1::gfp worms without over-expression hda-1p:::flag images hda-1p dve-1p::dve-1::gfp worms fed RNAi L1 stage imaged day 1 adulthood images bottom intestine nucleus intestinal cell diagram overlap HDA-1 DVE-1 peaks worms fed control atp-2 RNAi Nucleotide binding motifs HDA-1 DVE-1 mitochondrial stress P values motifs shownAnalysis HDA-1 ChIP peak signals control worms dve-1 RNAi worms atp-2 RNAi-induced mitochondrial stress Analysis DVE-1 ChIP peak signals worms hda-1 RNAi worms atp-2 RNAi stress Venn diagram comparing genes upregulated dependent hda-1 dve-1 GO term analysis UPRmt-upregulated genes dependent hda-1 dve-1 P values DAVID modified Fisher exact P values one-sided Source data file relationship HDA-1 DVE-1 examined expression profile hda-1p::hda-1::gfp animals protein level HDA-1 reduced dve-1 RNAi reduction HDA-1 protein level dve-1-deficient conditions mediated ubiquitin-mediated degradation RNAi knockdown ubq-1 ubq-2 genes abolished reduction HDA-1 dve-1 RNAi treatment knockdown HDA-1 reduced DVE-1 protein level overexpression HDA-1 elevated protein level DVE-1 qPCR analysis hda-1 dve-1 transcript levels not affected loss partner protein results suggest HDA-1 DVE-1 interact stabilizehuman protein SATB1 DVE-1 ‘cage distribution heterochromatin12 HDA-1 DVE-1 similar distribution intestinal cells C. elegans structures with DVE-1 chromatin regions stained DAPI (Fig. 2d HDA-1 DVE-1 regulate genes chromatin immunoprecipitation sequencing-seq) genes regulated HDA-1 DVE-1 mitochondrial stress 6794 HDA-1-enriched peaks uncovered ChIP-seq analysis normal conditions 60.5% (4,109 out of 6,794) recovered in DVE-1 ChIP-seq overlap higher expectation P value < 0.0001) after mitochondrial perturbation overlap HDA-1 DVE-1 increased to 86.1% (5711 out of 6634) overlap determined consensus binding motifs HDA-1 DVE-1 mitochondrial stress (Fig. HDA-1 DVE-1 shared more binding motifs mitochondrial perturbation increased overlap stress. 2f RNA-seq analysis 284 374 genes coregulated by HDA-1 DVE-1 under normal stressanalyzed ChIP-seq results binding HDA-1 to target loci on DVE-1 knockdown DVE-1 suppresses upregulation HDA-1 peaks knockdown HDA-1 affect upregulation DVE-1 peaks SATB1, DVE-1 homolog provides docking site recruit histone deacetylase HDAC1 onto SATB1 target sequences28 RNA-seq results 283 218 genes induced under stress in C. elegans dependent on hda-1 dve-1 169 genes (59.7%) contain HDA-1 ChIP peaks within ±500 bp 218 dve-1-dependent 175 genes (80.3%) contain DVE-1 ChIP peaks within ±500 bp 139 genes required both hda-1 and dve-1 49.1% of hda-1-dependent 63.8% dve-1-dependent genes enriched for terms Immune response Response to stress Metabolic process Protein folding employed qPCR validate key genes HDA-1- DVE-1-dependent regulation selected genes enriched GO termsKnockdown of hda-1 dve-1 suppressed induction cdr-4 pgp-1 nhr-115 M04C3.2 dnj-10 djr-1.2 Stress cyp-33C8 ipla-3 hmgs-1 tars-1 cua-1 upon mitochondrial stress hypomorphic allele hda-1 suppressed stress response genes results suggested HDA-1 with DVE-1 UPRmt-induced innate immune response metabolic reprogramming.HDA-1 required for UPRmt-mediated innate immunityMitochondrial function challenged by pathogens xenobiotics UPRmt initiates mitochondrial protective responses chaperones proteases activates innate immune responses HDA-1 regulates genes Immune response Metabolic process led function in animal fitness during mitochondrial stress irg-1p::gfp transgenic animals reporter strain for pathogen-infected response indicator induction innate immune reporter worms challenged with Pseudomonas aeruginosa strainmitochondrial function expression UPRmt reporter hsp-6p::gfp2. wild-type animals irg-1p::gfp expression induced Pseudomonas infection deficiency hda-1 induction irg-1 dve-1 (Fig. 3a Knockdown hda-1 dve-1 induction immune response genes P. aeruginosa infection deficiency hda-1 dve-1 reduced survival accumulation P. aeruginosa. elegans. 3c overexpression HDA-1 DVE-1 promoted survival reduced accumulation P. aeruginosa infection (Fig. 3f 3HDA-1 required UPRmt-mediated innate immunity images irg-1p::gfp worms control hda-1 dve-1 RNAi untreated exposed P. aeruginosa qRT-PCR measurement mRNA levels immune response genes wild-type worms control hda-1 treated P. aeruginosa 3 experiments 1000 worms Survival curves control hda-1 dve-1 RNAi worms P. aeruginosa slow-killing assay 50 worms images accumulation P.aeruginosa intestines worms control hda-1 dve-1 RNAi Scale bar 200 μm CFU units quantified experiments n 30 worms each sample Survival curves wild-type worms P. aeruginosa slow-killing assay n 50 worms sample fluorescence images irg-1p::gfp worms control hda-1 dve-1 treated Rhodococcus Scale bar 200 μm Survival curves control hda-1 dve-1 RNAi worms Rhodococcus killing assay n 55 60 59 worms each sample Results (b) mean + SD (e) mean ± SD P values calculated one-way ANOVA Tukey’s multiple comparisons test (*P < 0.05 **P < 0.01 ***P < 0.001) Source data function HDA-1 mediated target genes tested hmgs-1 immune response P. aeruginosa infection encodes HMG-CoA synthase UPRmt 2. Knocking reduced survival accumulation P. aeruginosa hda-1 RNAi overexpression hmgs-1 rescue loss immunity hda-1 knockdownsuggests expression target gene not sufficient hda-1 deficiency hda-1 or dve-1 RNAi suppressed immune response reduced survival mitochondrial insult Rhodococcus strain C. elegans (Fig. 3g h Fig results indicate hda-1 required for UPRmt-activated immune response.HDA-1 affects animal aging age-related pathologyGenes mitochondrial function recovered from C. elegans worm lifespans4 perturbations mitochondrial electron transport extend lifespans3 HDA-1 mitochondrial stress response suspected stress-induced lifespan extension examined lifespans wild-type hda-1-deficient worms mitochondrial perturbation hda-1 RNAi reduced lifespan extension feeding worms with atp-2 RNAi (Fig. hda-1 RNAi perturbation shortened worm lifespan previous study reported hda-1 RNAi affect C. elegans lifespan34 speculated reasons discrepancy different IPTG induction times expression affect hda-1 RNAi efficiency previous study used liquid culture used solid agarhda-1 lifespan regulation measured lifespan worms carrying hda-1(e1795) hypomorphic allele mutation shortened lifespan observed shortened lifespan eat-2(ad1116) mutant animals fed hda-1 RNAi mutants reduced pharyngeal pumping caloric restriction lifespan eat-2 animals decreased mitochondrial activate ZIP-2 pathway improvement mitochondrial integrity37 understand hda-1 regulates lifespan mitochondrial integrity mechanism 4HDA-1 affects animal aging age-dependent accumulation protein Lifespan analysis worms raised RNAis = 130 worms per condition Mobility analysis unc-54p::Q35::yfp worms control hda-1 dve-1 RNAi 30 worms per Mobility analysis unc-54p::Q35::yfp worms day 8 adulthood 30 worms per fluorescence images unc-54p::Q35::yfp worms fed control hda-1 RNAi day 5 adulthood 200 μm Quantification Q35 aggregates body muscles worm counted n = 12 worms per conditionfluorescence images unc-54p::Q35::yfp hda-1p::flag-54p:yfp dve-1p::dve-1::gfp worms day 5 adulthood. Scale bar 200 μm Quantification Q35 aggregates (f). body muscles each worm counted n = 20 worms per sample Results (e, g) mean ± SD. P value calculated two-tailed Student’s t test (*P < 0.05) (g), P values calculated one-way ANOVA Tukey’s multiple comparisons test (***P < 0.001) Source data file decline responses proteostatic stress aging toxic accumulation protein aggregates polyglutamine knockdown hda-1 polyQ toxicity impaired animal movement overexpression suppressed polyQ toxicity improved movement knockdown hda-1 increased polyQ aggregates C. elegans muscle cells overexpression reduced accumulation polyQ results indicate HDA-1 impact UPRmt lifespan extension C. elegans suggest therapeutic potential activating HDAC signaling age-related diseases neurodegenerationHDA-1 regulates mitochondrial stress C. elegans mitochondrial homeostasis relevance eukaryotes HDAC1 forms with HDAC2 transcription regulator complexes NuRD CoREST gene expression levels HDAC1/2 correlate with SATB2 DVE-1) mitochondrial chaperones HSPA9 HSPD1 proteases LONP1 YME1L1 asparagine synthetase ASNS membrane translocase TIMM17A in human rhesus monkey tissues (Fig. 5a HDAC1 interacted with SATB2 in mammalian cells (Fig. HDAC1/2 mitochondrial homeostasis. HDAC1/2 correlate with expression UPRmt genes correlation HDAC1 HDAC2 SATB2 UPRmt mRNA levels in human heart colon kidney lung muscle testis tissues Red circles positive correlation blue circles negative correlation size correlation coefficient Immunoprecipitation HDAC1 interacts with SATB2 in HEK293T cells Source data HDAC1/2 UPRmt stress treated HEK293T cells with sodium butyrate inhibitor HDACNaBt LONP1 HSPA9 YME1L1 HSPD1 cells antimycin A inhibitor mitochondrial electron transport chain complex III 6a inhibitors target HDACs not due HDAC1/2 inhibition role HDAC1/2 mitochondrial homeostasis UPRmt activation used siRNA knock HDAC1/2 SATB2 HeLa cells MitoTracker cells antimycin disrupted mitochondrial higher mitochondrial fusion in HDAC1/2- SATB2-deficient cells (Fig. 6b induction human UPRmt genes HSPD1 LONP1 ASNS YME1L1 suppressed in HDAC1/2 SATB2 cells insult (Fig 6c 7d). results confirmed conserved role HDAC1/2 mitochondrial surveillance UPRmt activation.Fig. functions HDAC1/2 SATB2 UPRmt activation conserved in mammals qRT-PCR measures mRNA levels LONP1 HSPA9 HEK293T cells fluorescence images HeLa cells qRT-PCR measures levels UPRmt genes HEK293T cells 3 experiments Results mean + SD.P values calculated by two-way ANOVA Tukey’s multiple comparisons test (**P < 0.01 ***P < 0.001) Source data provided file mitochondrial stress-to-nucleus communication UPRmt activated transcriptional induction genes stress immune responses metabolic reprogramming Epigenetic regulation stress signals stress response genes12–14 histone deacetylase HDA-1 coordinates with DVE-1 UPRmt in C. elegans mammalian homologs HDAC1/2 with SATB2 DVE-1) mediate mitochondrial homeostasis-dimensional chromatin gene SATB1 organizer landing platform for chromatin-remodeling enzymes HDAC27 Changes chromatin organization loops distant genes affect accessibility genomic loci SATB1 chromatin loop formation T cells cytokine gene expression27 HDAC1 form chromatin-remodeling complexes NuRD CoREST regulate gene expression39 to examine dynamic changes chromatin architecture function HDA-1 DVE-1 for rapid transcriptional induction in C. elegans mitochondrial perturbation understand mitochondrial stress signals to HDA-1 activityUPRmt upregulates transcription mitochondrion chaperones proteases initiates transcriptional program expression genes metabolic processes removal acetyl groups histones HDA-1 metabolism mitochondrial dysfunction acetyl-coA glucose serum deprivation acetylated histones H3 H4 in mammalian cells46 examine global acetylated histone marks affected during mitochondrial perturbation.Moderate mitochondrial stress hormetic stress defense immunity lifespan link between stress adaptation aging silencing hda-1 suppressed stress immune response longevity overexpression hda-1 delayed age-related accumulation protein aggregates polyglutamine toxicity in C. elegans correlation analysis in rhesus monkey human points conserved function HDAC1/2 regulating mitochondrial hormesis primates future genetic treatments pathway may provide therapeutic potential age-related diseases.MethodsStrains cultureSJ4100 SJ4058 CL2070SJ4005[hsp-4p CB5535 SJ4197[dve-1p AU133[irg-1p:gfp AM140-54p SS104 N2 wild-type C. elegans Caenorhabditis Genetics Centerstrains generated lab YSL1 odr-1p:dsRed YSL2:gfp odr-1p:dsRed YSL3(liuls1 odr-1p:dsRed zcIs39:gfp YSL4 liuEx1[hda-1p odr-1p:dsRed YSL5(liuls1 odr-1p:dsRed zcIs13[hsp-6p:gfp YSL6(glp-4(bn2) liuls2[hda-1p:gfp odr-1p:dsRed YSL7:gfp YSL8(glp-4-54p:yfp YSL9(glp-4 odr-1p:dsRed:yfp YSL10(glp-4[dve-1p:gfp YSL11(glp-4(bn2) liuEx2[gly-19p mec-7p:rfp]).HEK293T cells HeLa cells obtained ATCCCells cultured DMEM medium 10% fetal bovine serum 37 °C SATB2 plasmid constructed PCR HEK293T RNA ligated pCDNA3.3 vector HDAC1 plasmid gift from Prof. Jiemin Wong transfected Lipofectamine 2000 Reagent plasmid construction Supplementary Table 1.RNA clones Ahringer library hda-3 generated PCR worm cDNA ligated L4440 vector clones transformed into HT115 cells grown in LB 37 °C overnight 20X bacterial culture solution seeded worm plates 1.2 mg/ml IPTG Dried plates kept room temperature overnight Synchronized L1 worms raised on RNAi plates 20 °C double RNAi experiments mixed bacterial culture 30 h after siRNA knockdown 50 pmol siRNA transfected into 6-well plate Lipofectamine RNAi MAX Reagent synchronized L1 worms atp-2 cco-1 RNAi imaged adulthood double RNAi 20X 0.2 mg/ml IPTG seeded 30 h after 30 h worms imagedUPRmt antimycin HEK293T cells 80% confluency treated 20 μg/ml antimycin (Sigma #A8674) 21 h UPRER L1 worms raised 6 cm RNAi plates 43 h 300 μl M9 buffer 18 μg tunicamycin spread examined 12 h heat shock response worms raised 6 cm RNAi 34 h cultured 37 °C 1 h transferred 20 °C 21 h innate immune L1 worms raised 24 h exposure P. aeruginosa Rhodococcus imaged day 2 adulthood butyrate 5 mM sodium butyrate (Sigma #303410) 24 h Antimycin added culture medium 3 h picked 100 mM NaN3 droplets 2% agarose pads imaged Zeiss Imager M2 microscope Nuclear localization Zeiss LSM880 microscope Mitochondrial imaged PerkinElmer Operetta CLSTM Comparable images GFP fluorescence quantified ImageJ cells treated 50 μg/ml antimycin A (Sigma #A8674) 4 h stained MitoTracker (Invitrogen #M7512)RNA isolation real-time late L4 worms HEK293T cells collected resuspended trizol reagent samples frozen homogenized liquid nitrogen RNA isolated chloroform extraction precipitated isopropanol washed 75% ethanol cDNA synthesized reverse transcription kits Quantitative PCR SYBR Green PCR Master Mix C. elegans transcripts normalized rpl-32 HEK293T cells ACTB sequences RT-PCR Supplementary Table 2.ImmunoblottingWorms cells resuspended SDS loading buffer (100 mM Tris-HCl pH 4% SDS 20% glycerol 10% β-mercaptoethanol bromophenol blue boiled 95 °C 10 min Samples separated SDS-PAGE transferred PVDF membrane 5% milk-TBST membrane probed primary secondary antibodies antibodies anti-GFP-tubulin-Myc-FLAG anti-Acetylated-Lysine Membranes developed enhanced chemiluminescence method visualized Tanon 5200 luminescence imaging system30,000 worms washed M9 buffer resuspended 3 ml lysis buffer (50 mM tris-HCl pH 8.0 137 mM NaCl 1% Triton X-100 1 mM EDTA 10% glycerol proteinase homogenized homogenizer sonicated HEK293T cells 10 cm dish 90% confluence washed 1X PBS buffer resuspended 1 ml lysis buffer ice 30 min centrifuged × g 15 min supernatant transferred tube rotated 4 °C overnight anti-GFP antibody 1 μl anti-FLAG beads anti-Myc anti-GFP immunoprecipitation protein G beads #10004D 40 added rotated 4 °C 2 h washed three times lysis buffer boiled 50 μl 2X SDS Laemmli buffer (4% SDS 20% glycerol 10% 2-mercaptoethanol 0.02% bromophenol blue 0.125 M Tri-HCL pH 6.8) 95 °C 10 min.Lifespan L1 worms raised plates bacteria transferred plates 2 days monitored survival three touches no pharyngeal pumping dead hatching removedaeruginosa slow-killing assayP aeruginosa cultured overnight LB 50 μg/ml kanamycin 37 °C 10 μl seeded 3.5 cm slow-killing agar spread circle spreader Plates air-dried incubated 37 °C 24 h room temperature 24 h L1 glp-4(bn2) worms raised plates 20 °C 49 h 50 plates cultured 25 °C 5-fluoro-2′-deoxyuridine) (100 μg/ml added Living worms counted 24 h removed.Rhodococcus survival cultured overnight LB 25 °C 30X concentrated solution bacteria seeded 6 cm slow-killing agar plates L1 glp-4 worms raised plates 20 °C 49 h 50 worms plates cultured 25 °C worms counted 24 h removed. aeruginosa intestinal accumulation Pseudomonas aeruginosa strain PA01 pMF230 (Addgene GFP provided Dr. Huanqin Dai overnight cultures spread Plates air-dried incubated 37 °C 24 h room temperature 24 hworms fed RNAi 20 °C 53 h washed transferred PA01(GFP plates imaged 45 h PA01(GFP units exposed P. aeruginosa washed three rotated 1 h kanamycin (1 mg 30 worms picked 100 μl buffer 1.5 grinded pestle Lysates diluted 5 μl lysate solution plated LB plates carbenicillin (50 μg overnight incubation 37 °C colonies GFP fluorescence counted day 8 worms Q35 polyglutamine touched two times counted randomly selected tested-seq 240,000 worms HDA-1-GFP DVE-1-GFP overexpression harvested cross 2% formaldehyde 40 min washing 3 times pellets resuspended FA buffer (50 mM HEPES/KOH 7.5 1 mM EDTA 8.0 1% Triton-X-100 0.1% sodium deoxycholate 150 mM NaCl proteinase inhibitor homogenized sonicated samples centrifuged 20,000 × g 20 min 4 °Cdilution FA buffer 150 mM 1% supernatant Input incubated GFP-Trap agarose 50 μl per sample 4 °C 20 h agarose washed 150 mM NaCl 1 M NaCl 500 mM NaCl LiCl buffer (10 mM Tris-HCl 1 mM EDTA pH 8.0 1% NP40 1% sodium deoxycholate 250 mM LiCl three TE buffer fraction eluted agarose by TE buffer 1% SDS 250 mM NaCl elute decross-linked Input overnight 65 °C samples digested by protease K 2 h 55 DNA extracted ChIP DNA Clean & Concentrator kit DNA libraries prepared NEB kit ChIP-seq data analysis adaptor sequences filtered Trim Galore quality control parameters 6 --length Processed reads mapped C. elegans genome BWA 0.7.13 default parameters mapping reads quality ≥20 mismatches <8 retained PCR duplicates removed by Picard (version 2.17.6) MACS2 (version 2.1.1 peak regionsChIP-seq signals calculated deepTools 2.4 normalizing coverage 10 million reads subtracting input final signals Overlaps peaks determined BEDTools 2.27.1 intersect command Motif analysis HOMER 4.10-0 parameters ce11 -gc -size Bwtool57 peak signal comparison-seq sample preparation 1000 synchronized L1 worms fed control hda-1 dve-1 RNAi 24 h atp-2 RNAi 48 h Worms washed resuspended trizol reagent RNA.Reads aligned C. elegans genome TopHat2 2.1.1 parameters --read-mismatches 6-edit-dist 6-anchor 8-length 26. mapping reads analysis Gene expression levels differentially expressed genes generated Cufflinks 2.2.1)59 term enrichment analysis DAVID60 GOTERM_BP_2 category.Expression correlation expression HDAC1 HDAC2 SATB2 UPRmt genes human tissues data GTEx v7 Tissues RNA-Seq rhesus monkey tissues RhesusBase62tissue Pearson’s correlation genes calculated using gene expression levels samples human rhesus macaque.Statistics reproducibilityStatistical analyses performed with GraphPad Prism 5.0 Results expressed mean ± SD or + SD ANOVA Tukey’s comparisons test Student’s t test P values details statistical analyses P values statistical significance see figure legends Experiments analysis performed twice similar results Micrographs immunoblotting images three experiments similar results.Reporting research design Nature Research Reporting Summary.Supplementary Review Additional Supplementary FilesSupplementary Data Summary | 49.7 | 0.910312 |
10.1038/s41467-020-18786-x | PMC7532528 | Early identification of COVID-19 patients at risk of progression may facilitate more individually aligned treatment plans. Here the authors develop an online nomogram incorporating CT severity score and clinical characteristics for early predicting the disease progression risk among COVID-19 pneumonia patients. | The outbreak of coronavirus disease 2019 (COVID-19) has rapidly spread to become a worldwide emergency. Early identification of patients at risk of progression may facilitate more individually aligned treatment plans and optimized utilization of medical resource. Here we conducted a multicenter retrospective study involving patients with moderate COVID-19 pneumonia to investigate the utility of chest computed tomography (CT) and clinical characteristics to risk-stratify the patients. Our results show that CT severity score is associated with inflammatory levels and that older age, higher neutrophil-to-lymphocyte ratio (NLR), and CT severity score on admission are independent risk factors for short-term progression. The nomogram based on these risk factors shows good calibration and discrimination in the derivation and validation cohorts. These findings have implications for predicting the progression risk of COVID-19 pneumonia patients at the time of admission. CT examination may help risk-stratification and guide the timing of admission. | IntroductionThe outbreak of coronavirus disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has rapidly spread to become a worldwide pandemic. Most COVID-19 patients have a mild clinical course, while a proportion of patients demonstrated rapid deterioration (particularly within 7–14 days) from the onset of symptoms into severe illness with or without acute respiratory distress syndrome (ARDS)1,2. There is no specific anti-coronavirus treatment for severe patients at present, and whether remdesivir is associated with significant clinical benefits for severe COVID-19 still requires further confirmation3,4. These patients have poor survival and often require intensive medical resource utilization, and their case fatality rate is about 20 times higher than that of nonsevere patients5,6. Thus, early identification of patients at risk of severe complications of COVID-19 is of clinical importance. Several studies reported that the prevalence of severe COVID-19 ranged from 15.7 to 26.1% among patients admitted to hospital and these cases were often associated with abnormal chest computed tomography (CT) findings and clinical laboratory data6–8. Guan et al. indicated that severe COVID-19 patients were more likely to show ground-glass opacity (GGO), local or bilateral patchy shadowing, and interstitial abnormalities on CT8. This likely reflects the clinical progression of disease but also offers an opportunity to investigate the clinical utility of chest CT as a predictive tool to risk-stratify the patients. Furthermore, the predictive value of chest CT for the prognosis of COVID-19 patients is warranted to assist the effective treatment and control of disease spread. Previous study suggested that higher CT lung scores correlated with poor prognosis in patients with Middle East Respiratory Syndrome (MERS)9. Chest CT has been proposed as an ancillary approach for screening individuals with suspected COVID-19 pneumonia during the epidemic period and monitoring treatment response according to the dynamic radiological changes10–12. Therefore, we retrospectively enrolled patients with moderate COVID-19 pneumonia on admission from multiple hospitals and observed for at least 14 days to explore the early CT and clinical risk factors for progression to severe COVID-19 pneumonia, and constructed a nomogram based on the independent factors. Meanwhile, we also compared the clinical and CT characteristics in patients with different exposure history or period from symptom onset to admission to provide a deep understanding of the relationship among CT findings, epidemiological features, and inflammation.Here, we show a nomogram incorporating age, neutrophil-to-lymphocyte ratio (NLR), and CT severity score on admission with good performance in the prediction of short-term disease progression in hospitalized patients with moderate COVID-19 pneumonia, which have implications for early predicting the progression risk and guiding individually aligned treatment plans among COVID-19 pneumonia patients.ResultsPatient characteristicsA total of 298 patients with COVID-19 were identified according to the inclusion criteria, of which 51 patients were excluded for having: (1) negative CT findings or severe/critical COVID-19 on admission (n = 39); (2) age younger than 18 years old (n = 12). Finally, 247 patients were included in our study, consisting of 141 patients in the derivation cohort and 106 patients in the validation cohort (Fig. 1). The clinical characteristics of included patients in the derivation and validation cohorts are presented in Table 1. Among them, 72 (51.1%) and 54 (50.9%) patients were male, with a median age of 44 years and 46 years in the derivation and validation cohorts, respectively. There were no significant differences in the clinical characteristics between the two cohorts (Supplementary Data 1). During the hospitalization, 15/141 (10.6%) and 10/106 (9.4%) patients progressed to severe pneumonia in the derivation and validation cohorts, and 6/15 (40.0%) and 5/10 (50.0%) severe cases further deteriorated critical illness. Besides, 245 (99.2%) patients of the included patients had recovered and discharged at the time of analysis.Fig. 1Study workflow.The flow diagram shows the study population enrollment and observation period.Table 1Clinical characteristics of patients with COVID-19 pneumonia in the derivation and validation cohorts.VariablesDerivation (n = 141)Validation (n = 106)P valueAge (years)44 (34–55)46 (35–56)0.695Male gender72 (51.1%)54 (50.9%)0.985Exposure history in Wuhan within 2 weeks0.075Yes76 (53.9%)45 (42.5%)No65 (46.1%)61 (57.6%)Smoking history7 (5.0%)7 (6.6%)0.581ComorbiditiesAny33 (23.4%)21 (19.8%)0.499Diabetes8 (5.7%)6 (5.7%)0.996Hypertension21 (14.9%)10 (9.4%)0.200Cardiovascular disease3 (2.1%)2 (1.9%)0.894COPD4 (2.8%)3 (2.8%)0.997Cerebrovascular disease1 (0.7%)0 (0)0.385Hepatitis B infection4 (2.8%)2 (1.9%)0.703Laboratory findingsLymphocyte count (×109/L)1.1 (0.8–1.5)1.1 (0.9–1.6)0.350NLR2.6 (1.9–3.7)2.7 (1.7–3.7)0.769Aspartate aminotransferase (U/L)24.0 (19.9–30.7)25.4 (19.9–34.0)0.272Albumin (g/L)37.1 (34.8–40.1)37.9 (34.9–40.3)0.348Lactic dehydrogenase (U/L)175.9 (138.9–221.9)183.4 (145.1–247.1)0.154C-reactive protein (mg/L)17.4 (7.4–38.2)16.9 (2.6–39.9)0.191Radiological findingsBilateral involvement123 (87.2%)94 (88.7%)0.731CT severity score6 (4–10)7 (4–10)0.149Clinical outcomesSevere pneumonia15 (10.6%)10 (9.4%)0.756Requiring mechanical ventilation6 (4.3%)5 (4.7%)0.862ICU admission4 (10.6%)4 (3.8%)0.728Death1 (0.7%)1(0.9%)0.839Data are presented as median (IQR) or n (%). Differences between groups are analyzed using Student’s t-test or Mann–Whitney U-test for continuous variables and Chi-square test or Fisher’s exact test for categorical variables. Two-sided P-values are reported.COPD chronic obstructive pulmonary disease, COVID-19 coronavirus disease 2019, CT computed tomography, ICU intensive care unit, IQR interquartile range, NLR neutrophil-to-lymphocyte ratio.Comparison between stable and progressive patientsThe clinical and CT characteristics of patients who progressed to severe COVID-19 (progressive group) and patients who did not (stable group) in the derivation cohort are presented in Table 2. Compared with those in the stable group, patients in the progressive group were significantly older (median age, 58 vs. 41 years old, P = 0.001) and more likely to have underlying hypertension (P = 0.004). But no significant difference was found in gender, exposure history, smoking history, and other co-morbidities including diabetes, chronic obstructive pulmonary disease (COPD), cardiovascular disease, cerebrovascular disease, and chronic hepatitis B infection between the two groups. The main clinical symptoms between the two groups were not statistically different, while slightly more patients manifested anorexia (P = 0.088), diarrhea (P = 0.065), and shortness of breath (P = 0.088) in the progressive group. There was no significant difference in percutaneous oxygen saturation on admission between the two groups (P = 0.110). Patients in the progressive group had lower baseline lymphocyte count and albumin, and higher NLR, aspartate aminotransferase, lactic dehydrogenase, and C-reactive protein (all P < 0.05). The major CT features of COVID-19 pneumonia patients were bilateral, peripheral or mixed distributed GGO, consolidation, and GGO with consolidation (Fig. 2a–f). Patients in the progressive group had more lobes and segments involved, with a higher proportion of crazy-paving sign and higher CT severity score (Supplementary Fig. 1a) compared with those in the stable group (all P < 0.05). However, no significant difference was found in hospital length of stay (P = 0.398) and duration of viral shedding after illness onset (P = 0.087) between the two groups.Table 2Clinical and CT characteristics between the stable and progressive patients in the derivation cohort.VariablesStable (n = 126)Progressive (n = 15)P valueAge (years)41 (33–52)58 (44–66)0.001Male (gender)65 (51.6%)7 (46.7%)0.719Exposure history in Wuhan within 2 weeks0.616Yes67 (53.2%)9 (60.0%)No59 (46.8%)6 (40.0%)Smoking history7 (5.6%)0 (0)0.349ComorbiditiesAny26 (20.6%)7 (46.7%)0.024Diabetes6 (4.8%)2 (13.3%)0.175Hypertension15 (11.9%)6 (40.0%)0.004Cardiovascular disease2 (1.6%)1 (6.7%)0.288COPD2 (1.6%)2 (13.3%)0.056Cerebrovascular disease1 (0.8%)0 (0)0.729Hepatitis B infection4 (3.2%)0 (0)0.484Signs and symptomsFever92 (73.0%)13 (86.7%)0.252Cough66 (52.4%)8 (53.3%)0.944Sputum production14 (11.1%)2 (13.3%)0.798Fatigue or myalgia28 (22.2%)3 (20.0%)0.844Anorexia3 (2.4%)2 (13.3%)0.088Diarrhea4 (3.2%)2 (13.3%)0.065Shortness of breath3 (2.4%)2 (13.3%)0.088Percutaneous oxygen saturation (%)97.5 (96.1–98.6)95.6 (94.5–98.5)0.110Laboratory findingsPlatelet count (×109/L)170.5 (137.8–224.0)148.0 (121.0–204.0)0.192White blood cell count (×109/L)4.4 (3.4–5.2)4.6 (3.3–5.7)0.683Neutrophil count (×109/L)2.8 (2.1–3.6)3.2 (2.4–4.4)0.237Lymphocyte count (×109/L)1.1 (0.9–1.5)0.7 (0.5–1.3)0.002NLR2.5 (1.8–3.4)4.8 (3.1–5.1)<0.001Alanine aminotransferase (U/L)20.0 (14.5–28.4)19.0 (14.2–31.4)0.794Aspartate aminotransferase (U/L)23.2 (19.6–28.9)30.4 (25.2–37.4)0.013Total bilirubin (μmol/L)11.3 (8.9–15.7)10.6 (8.8–13.8)0.339Albumin (g/L)37.5 (35.3–40.2)35.0 (31.7–37.3)0.008Creatinine (μmol/L)49.8 (40.1–60.3)52.3 (33.6–63.9)0.683Creatine kinase (U/L)69.6 (40.5–122.3)92.0 (57.5–386.0)0.120Lactic dehydrogenase (U/L)168.9 (134.9–217.3)197.4 (182.8–276.9)0.014C-reactive protein (mg/L)15.7 (6.8–36.8)36.9 (27.2–58.7)0.001CT featuresNumber of lobes involved0.013One lobe12 (9.5%)2 (13.3%)Two lobes25 (19.9%)0 (0)Three lobes15 (11.9%)0 (0)Four lobes28 (22.2%)1 (6.7%)Five lobes46 (36.5%)12 (80.0%)Number of segments involved9 (5–12)12 (10–15)0.007Bilateral involvement110 (87.3%)13 (86.7%)0.944Distribution pattern0.116Peripheral67 (53.2%)4 (26.7%)Central2 (1.6%)0 (0)Mixed57 (45.2%)11 (73.3%)GGO120 (88.9%)15 (100%)0.388Consolidation107 (84.9%)13 (86.7%)0.858GGO with consolidation100 (79.4%)13 (86.7%)0.503Crazy-paving34 (27.0%)8 (53.3%)0.035Air bronchogram71 (56.4%)11 (73.3%)0.207Discrete nodules10 (7.9%)1 (6.7%)0.862Lymphadenopathy5 (4.0%)1 (6.7%)0.625Pleural effusion4 (3.2%)0 (0)0.484CT severity score6 (4–9)10 (7–15)0.001Hospital length of stay (days)21 (16–28)22 (18–35)0.398Duration of viral shedding after illness onset (days)14 (10–24)18 (13–31)0.087Data are presented as median (IQR) or n (%). Differences between groups are analyzed using Student’s t-test or Mann–Whitney U-test for continuous variables and Chi-square test or Fisher’s exact test for categorical variables. Two-sided P values are reported.COPD chronic obstructive pulmonary disease, CT computed tomography, GGO ground-glass opacities, IQR interquartile range, NLR neutrophil-to-lymphocyte ratio.Fig. 2Representative chest CT images of patients with COVID-19 pneumonia.a Subpleural patchy areas of GGO with crazy-paving sign in the right middle lobe. b Multiple patchy areas of consolidation in the right middle lobe, left upper lobe, and bilateral lower lobes and air bronchogram in the right middle lobe. c Multiple patchy areas of organizing pneumonia in the right middle and lower lobes on the sagittal image with CT severity score of 9 for the right lung. d Bilateral and peripheral multiple patchy areas of GGO with reticular and intralobular septal thickening. e Multiple mixed distributed pure GGO, GGO with consolidation, and interlobular septal thickening in bilateral lungs. f Bilateral multiple patchy and thin areas of GGO in the posterior parts of the lungs.Logistic regression analysis and nomogram establishmentMultivariate logistic regression analysis showed that age (odds ratio [OR] and 95% confidence interval [CI], 1.06 [1.01–1.12]; P = 0.028), baseline NLR (OR and 95% CI, 1.74 [1.13–2.70]; P = 0.011), and CT severity score (OR with 95% CI 1.19 [1.01–1.41]; P = 0.043) were independent predictors for progression to severe COVID-19 pneumonia in the derivation cohort (Table 3). A nomogram incorporating these three predictors was then constructed (Fig. 3a). The calibration curve of the nomogram (Fig. 3b) and a nonsignificant Hosmer–Lemeshow test statistic (P = 0.378) showed good calibration in the derivation cohort. The favorable calibration of the nomogram was confirmed with the validation cohort (Fig. 3c), with a nonsignificant Hosmer–Lemeshow test statistic (P = 0.791). The area under the receiver operating characteristic curve (AUC) of the nomogram in the derivation and validation cohorts was 0.867 (95% CI, 0.770–0.963; Fig. 3d) and 0.898 (95% CI, 0.812–0.984; Fig. 3e), respectively, which revealed good discrimination. Therefore, our nomogram performed well in both the derivation and validation cohorts. The nomogram has been deployed in an online risk calculator, which is freely available at https://xy3yyfskfzc.shinyapps.io/DynNomapp2/. The decision curve analysis (DCA) for the nomogram in the derivation cohort is presented in Supplementary Fig. 2. The DCA indicated that when the threshold probability for a doctor or a patient was within a range from 0.02 to 0.92, the nomogram added more net benefit than the “treat all” or “treat none” strategies.Table 3Risk factors for progression to severe COVID-19 pneumonia in the derivation cohort.VariablesOR (95% CI)P valueOR (95% CI)P valueAge1.09 (1.04–1.14)0.0011.06 (1.01–1.12)0.028Hypertension4.93 (1.54–15.82)0.0070.676NLR2.13 (1.43–3.18)<0.0011.74 (1.13–2.70)0.012Aspartate aminotransferase1.04 (1.00–1.09)0.0700.682Albumin0.82 (0.71–0.96)0.0110.668Lactic dehydrogenase1.01 (1.00–1.02)0.0060.661C-reactive protein1.03 (1.01–1.06)0.0040.471Number of segments involved1.18 (1.04–1.35)0.0130.488Crazy-paving3.09 (1.04–9.18)0.0420.821CT severity score1.32 (1.14–1.54)0.0011.19 (1.01–1.41)0.043Univariate and multivariate logistic regression analyses are performed and the corresponding ORs are reported.CI confidence interval, CT computed tomography, COVID-19 coronavirus disease 2019, NLR neutrophil-to-lymphocyte ratio, OR odds ratio.Fig. 3Development and performance of nomogram.a A nomogram for the prediction of developing severe COVID-19 pneumonia. Calibration curves of the nomogram in the derivation (b) and validation (c) cohorts, respectively, which depict the calibration of the nomogram in terms of the agreement between the predicted risk of severe COVID-19 pneumonia and observed outcomes. The 45° blue line represents a perfect prediction, and the dotted red lines represent the predictive performance of the nomogram. The closer the dotted red line fit is to the ideal line, the better the predictive accuracy of the nomogram is. Plots show the ROC curves of the nomogram in the derivation (d) and validation (e) cohorts, respectively.Association of CT characteristics with inflammatory indexesIn the derivation cohort, correlation analysis showed that there were significantly positive associations between baseline CT severity score and inflammatory indexes (neutrophil count, lactic dehydrogenase, and C-reactive protein) on admission (Fig. 4a). Furthermore, for 134 patients who had available inflammatory indexes results on day 3 after admission in the derivation cohort, baseline CT severity score was not only positively associated with inflammatory indexes (white blood cell count, neutrophil count, lactic dehydrogenase, and C-reactive protein), but also negatively associated with lymphocyte count on day 3 after admission (Fig. 4b), which indicated the potential of CT severity score for early predicting lymphopenia.Fig. 4Correlation between CT characteristics and inflammatory indexes.Heatmaps depict the correlations between the baseline CT characteristics and inflammatory indexes (within the blue dotted box) on admission (a) and on day 3 after admission (b) showing the correlation coefficients r with P < 0.05 of all pairs.Comparison between patients with and without bacterial co-infectionBacterial coinfection may complicate the disease course, which needs to be treated using antibiotics. We further investigated the prevalence of bacterial coinfection in COVID-19 pneumonia patients and its association with clinical and CT characteristics in the derivation cohort (Supplementary Table 1). 17/141 (12.1%) developed bacterial coinfection during hospitalization. Patients who developed bacterial coinfection were significantly older (median age, 56 vs. 41 years old, P = 0.031), more likely to have underlying hypertension (P = 0.012), and less likely to show consolidation on CT at the time of admission (P = 0.012) while had comparable CT severity score (P = 0.817; Supplementary Fig. 1b) compared those without bacterial coinfection. Furthermore, patients with bacterial coinfection had significantly longer hospital length of stay and duration of viral shedding after illness onset (both P < 0.001).Comparison between other subgroupsThere was no significant difference in the clinical and CT characteristics between patients with and without exposure history in Wuhan within 2 weeks before illness onset in the derivation cohort (Supplementary Table 2), who had similar age, NLR level, and CT severity score (Supplementary Fig. 1c). In addition, the median period from symptom onset to admission in the derivation cohort was 4 days. For patients who were admitted more than 4 days from symptom onset, more lobes or segments involved, and higher CT severity score (Supplementary Fig. 1d) were found, while there was no significant difference in age and inflammatory indexes between patients with different period (≤4 days vs. >4 days) from symptom onset to admission in the derivation cohort (Table 4).Table 4Clinical and CT characteristics of patients with COVID-19 in the derivation cohort according to the period from symptom onset to admission.Variables≤4 days (n = 70)>4 days (n = 71)P valueAge (years)41 (31–53)46 (35–59)0.147Hypertension8 (11.4%)13 (18.3%)0.251Lymphocyte count (×109/L)1.1 (0.8–1.5)1.1 (0.9–1.4)0.944NLR2.7 (1.8–3.7)2.6 (1.9–3.9)0.867Aspartate aminotransferase (U/L)23.8 (19.8–30.4)24.0 (20.0–31.2)0.649Albumin (g/L)38.0 (35.3–40.6)36.4 (34.2–39.0)0.053Lactic dehydrogenase (U/L)173.3 (134.8–221.9)177.1 (141.9–232.1)0.598C-reactive protein (mg/L)18.1 (7.4–38.0)17.4 (6.8–38.7)0.918Number of lobes involved3 (2–5)4 (3–5)0.010Number of segments involved7 (3–12)10 (6–12)0.020Crazy-paving25 (35.7%)17 (23.9%)0.126CT severity score5 (3–10)7 (5–10)0.030Data are presented as median (IQR) or n (%). Differences between groups are analyzed using Student’s t-test or Mann–Whitney U-test for continuous variables and Chi-square test or Fisher’s exact test for categorical variables. Two-sided P values are reported.COVID-19 coronavirus disease 2019, CT computed tomography, IQR interquartile range, NLR neutrophil-to-lymphocyte ratio.DiscussionIn this study, we retrospectively assessed the clinical and CT characteristics of COVID-19 pneumonia patients from multiple hospitals and identified the baseline risk factors for clinical progression. Our results indicated that CT severity score was associated with inflammatory levels, and older age, higher NLR and CT severity score on admission were independent predictors for progression to severe COVID-19 pneumonia. The nomogram based on these risk factors showed good calibration and discrimination in the derivation and validation cohorts. In addition, patients who were admitted longer from symptom onset had more severe lung involvement.With the rapid increase of newly confirmed and severe cases, the management of severe patients become a challenging issue of the COVID-19 outbreak. Timely identification of patients with high risk to develop ARDS or multiorgan failure and risk stratification management might be helpful for more individually aligned treatment plans, optimized utilization of medical resource, and preventing further deterioration. In our cohorts, the prevalence of severe COVID-19 pneumonia was about 10%, which was lower than that in some large-scale reports6,8. This might be explained by the inclusion of only moderate patients on admission. Besides, the patients in our cohort were younger compared with those in Wuhan, which may be due to the fact that most of the patients with exposure history in Wuhan were young or middle-aged individuals working in Wuhan13. We found that progressive patients were more likely to be older and had underlying hypertension compared with stable patients. These data were in agreement with recent reports, which suggested that age and hypertension may be risk factors for progression in COVID-19 patients14–16.COVID-19 pneumonia patients who progressed in disease severity had lower baseline lymphocyte count, and higher NLR, lactic dehydrogenase, and C-reactive protein. SARSCoV-2 virus might act on lymphocytes as does SARSCoV, which induces a cytokine storm and triggers a series of immune responses17. Some studies suggested that the decrease of peripheral T lymphocyte count attributes to the inflammatory cytokine milieu and T cell recruitment to sites of infection, and reduced but hyperactivated or exhausted peripheral T cells were more frequently found in severe cases18,19. Lymphopenia has been confirmed as a potential factor associated with disease severity and mortality in COVID-1913. Thus, damage to lymphocytes and consequently immunologic abnormality might be an important factor leading to exacerbations of patients. The uncontrolled inflammatory response could also stimulate the production of neutrophils apart from speeding up the apoptosis of lymphocytes20. NLR, a simple biomarker to assess the systemic inflammatory status, is widely used for the prediction of prognosis of patients with pneumonia21,22. Increased NLR, resulting from decreased lymphocyte count and/or elevated neutrophil count, represents damaged lymphocyte function and/or increased inflammatory level and risk of bacterial infection. In addition, C-reactive protein is another serum maker produced by the liver in response to inflammation. Liu et al. reported that C-reactive protein might be predictive of disease severity in COVID-19 patients23. Thus, our results suggested that patients with higher inflammatory levels on admission had higher risk to develop severe COVID-19.To explore the predictive value of chest CT for progression, we compared the difference of CT characteristics in the stable and progressive patients and found that progressive patients had higher CT severity score. CT severity score is used to semiquantitatively estimate the pulmonary involvement, which is associated with both the number of involved lobes and extent of lesions24. In support of our findings, a previous report regarding MERS showed the predictive value of CT severity score for prognosis and short-term mortality9. Furthermore, a higher proportion of progressive patients showed crazy-paving sign which reflects interstitial thickening25. The binding of SARSCoV-2 spike protein to the receptor angiotensin-converting enzyme II (ACE2) contributes to the downregulation of ACE2, increased pulmonary capillary permeability, and diffuse alveolar damage26–28. In patients with SARS, mixed and predominant reticular patterns were also noted from the second week29,30. Therefore, we speculated that the involvement of the interstitial vascular endothelial cells results in interlobular and intralobular septal thickening, which may be associated with the disease severity.Our results further revealed that age, NLR, and CT severity score on admission were significant predictors for progression in moderate COVID-19 pneumonia patients. The predictive value of age and NLR has been reported in recent studies14,16,31. Previous study showed that the MuLBSTA score could early warn the mortality of viral pneumonia, which included lymphopenia and multilobe infiltration32. Our findings were consistent with theirs but more quantitatively in terms of imaging evaluation of lung involvement. Furthermore, we constructed a nomogram based on the multivariate logistic regression model to provide an easy-to-use tool for clinicians in the prediction of severe pneumonia in COVID-19 patients, which showed good performance in both the derivation cohort and external validation cohort. Recently, Liang et al. proposed a clinical risk score incorporating ten clinical variables to predict the occurrence of critical illness in hospitalized patients with COVID-1933. Their risk score included dichotomous chest X-ray abnormality instead of the severity of abnormality on CT. In contrast to their study, we adopted a quantitative CT severity score to accurately assess the degree of lung injury and aimed to early predict the in-hospital progression risk within 14 days in patients with moderate COVID-19 pneumonia on admission. Furthermore, the prediction model established in our study was simpler with only three easily accessible variables compared with theirs. Like SARS and MERS, some COVID-19 pneumonia patients progressed rapidly at about 7–14 days after onset likely due to the cytokine storm in the body as evidenced by increased plasma proinflammatory cytokines1,17,34. The results revealed the significant association between baseline CT severity score and inflammatory markers, particularly baseline CT severity score and lymphocyte count at day 3 after admission, which implied the potential value of chest CT on admission to estimate pulmonary inflammation or lung damage and to early predict lymphopenia.Patients with bacterial coinfection during hospitalization were older and more likely to have underlying hypertension than those without, which suggested that age and hypertension may be risk factors for concomitant bacterial infection. In addition, consolidation on baseline chest CT was less likely found in those with bacterial coinfection, which may be due to the weak antiviral immune response at the early stage of COVID-19 pneumonia in these older individuals with existing comorbidities35. The imaging findings may help physicians to identify the patients with higher risk of bacterial coinfection and those who need prophylactic antibiotic therapy to shorten hospital length of stay and duration of viral shedding. We also acknowledged that our findings were limited by the relatively small sample size, which should be interpreted with caution by clinicans and further confirmed by larger samples. Besides, patients who were admitted more than 4 days after symptom onset had higher CT severity scores, which likely attributed to the lung involvement progression as disease course extends36. CT examination may be important in guiding the rational timing of admission for the individual management of COVID-19 pneumonia patients.There were some limitations in our study. First, our study was retrospectively conducted, and the distribution of patients was imbalanced with only about 10% of cases developing severe pneumonia. Second, adjuvant treatment during hospitalization have not yet been analyzed, and multiple inflammatory cytokines were not available in this study. More comprehensive investigation of the relationship between CT characteristics and cytokine storm induced by COVID-19 needs to be performed. Third, CT is not used widely outside China for patients with COVID-19. Though the management guidelines in China recommend chest CT as a routine examination for COVID-19 pneumonia, the American College of Radiology advocates that CT should not be used as a first-line test to diagnose COVID-19 and should be used sparingly and reserved for hospitalized, symptomatic patients with specific clinical indications, which may limit the broad applicability of our findings37.In conclusion, our results indicated that older age, higher NLR, and CT severity score on admission were independent risk factors for clinical progression in moderate COVID-19 pneumonia patients, and the nomogram based on the three risk factors showed favorable predictive accuracy in the derivation and validation cohorts. Chest CT has the potential to early predict the risk of progression and reflect disease severity as well, which may also help guide the timing of admission for COVID-19 pneumonia patients.MethodsPatientsOur study was retrospectively conducted in compliance with the Health Insurance Portability and Accountability Act. We were authorized by the Hunan Provincial Health Commission in the collection of clinical and radiological data under anonymization according to the standard of care, which required the implementation to meet the criteria of the Helsinki declaration and follow all relevant regulations regarding the use of human study participants. The permission of the Institutional Review Board of Third Xiangya Hospital was obtained with waiver of informed consent. None of the patients in this study were reported in prior publications regarding their clinical characteristics.Health records were reviewed for patients who were treated at Third Xiangya Hospital, Changsha Public Health Treatment Center, Second People’s Hospital of Hunan, First Hospital of Yueyang, and Central Hospital of Shaoyang between January 17, 2020 and February 1, 2020. Patients were included in the study if they satisfied the following criteria: (1) confirmed COVID-19; (2) available chest CT scan on admission. The diagnosis of COVID-19 was established based on the World Health Organization interim guidance, and a confirmed case was defined as a positive result to high-throughput sequencing or real-time reverse transcription-polymerase chain reaction (RT-PCR) assay of SARS-CoV-2 for nasal and pharyngeal swab specimens. Patients were observed for at least 14 days from admission to determine whether they exacerbated to severe pneumonia or not. We divided the patients into two independent cohorts: patients treated at Third Xiangya Hospital, Changsha Public Health Treatment Center, and Second People’s Hospital of Hunan constituted the derivation cohort, whereas patients treated at First Hospital of Yueyang and Central Hospital of Shaoyang constituted the validation cohort.Clinical data collectionDemographic information (age, gender), exposure history, smoking history, comorbidities (including diabetes, hypertension, cardiovascular disease, COPD, cerebrovascular disease, and hepatitis B infection), clinical symptoms and signs (including percutaneous oxygen saturation), and laboratory data (platelet count, white blood cell count, neutrophil count, lymphocyte count, alanine aminotransferase, aspartate aminotransferase, total bilirubin, albumin, creatinine, creatine kinase, lactic dehydrogenase, and C-reactive protein) were obtained with data collection forms from electronic medical records. Exposure history was defined as exposure in Wuhan within 2 weeks before illness onset or exposure to local people with confirmed SARS-CoV-2 infection. The clinical classification of COVID-19 pneumonia is as follows: (1) moderate type, patients with fever, respiratory tract symptoms, and radiological evidence of confirmed pneumonia. (2) severe type, patients with one of the following: (a) respiratory distress (respiratory rate ≥ 30 beats/min); (b) hypoxia (oxygen saturation ≤ 93% in the resting state); (c) hypoxemia (arterial blood oxygen partial pressure/oxygen concentration ≤ 300 mmHg). (3) critical type, patients with one of the following: (a) respiratory failure requiring mechanical ventilation; (b) shock; (c) intensive care unit (ICU) admission is required for combined other organs failure. The primary endpoint of this study was the development of severe COVID-19 pneumonia by February 15, 2020, and other clinical outcomes including bacterial co-infection during hospitalization, requiring mechanical ventilation, ICU admission, discharge, and death were also recorded. Bacterial coinfection was diagnosed when patients showed clinical symptoms, signs, or radiological evidence of nosocomial pneumonia or bacteremia and a positive bacterial culture test was obtained from lower respiratory tract specimens or blood samples after admission1. Hospital length of stay was calculated by subtracting day of admission from day of discharge. Duration of viral shedding after illness onset was considered as the number of days from symptom onset to persistent negative results on respiratory tract viral RT-PCR testing. All samples from the same patient were tested until two consecutive samples showed negative results, with the first negative result defining the duration of shedding. Several co-authors (L.L., W.Z., X.M., M.Y., T.L., and X.L.) only contributed to the collection of original data. The researchers who collected original data or extracted results of index tests were not involved in the final data summary and analysis.CT examination and image analysisAll patients underwent chest CT examinations on admission. The images were reconstructed to 1.0-mm thickness for the transverse scans. Sagittal and coronal reconstructions with a 3.0-mm thickness were performed. All CT images were reviewed independently by two radiologists, each with over 10 years of experience in chest imaging. A third experienced radiologist was consulted if there was a disagreement in interpreting imaging results. The imaging features including GGO, consolidation, crazy-paving, and air bronchogram were recorded30,38. GGO was defined as hazy increased lung attenuation with preservation of bronchial and vascular margins, whereas consolidation was defined as opacification with obscuration of margins of vessels and airway walls. Crazy-paving refers to the appearance of ground-glass opacity with superimposed interlobular septal thickening and intralobular septal thickening. The lesion distribution pattern, lobe and segment involvement were also assessed. The CT findings in the outer one third of the lung were defined as peripheral, and those in the inner two thirds of the lung were defined as central. Besides, the presence of discrete nodules, lymphadenopathy, and pleural effusion were recorded. Each of the five lung lobes was reviewed for opacification and consolidation. The lesions extent within each lung lobe was semiquantitatively evaluated by scoring from 0 to 5 based on the degree of involvement: score 0, none involvement; score 1, ≤5% involvement; score 2, 6–25% involvement; score 3, 26–50% involvement; score 4, 51–75% involvement; score 5, >75% involvement. The total score was calculated by summing up scores of all five lobes to provide a CT severity score ranging from 0 to 2511.Statistical analysisContinuous variables are presented as the median and interquartile range, and categorical variables are presented as frequency and percentage. Differences between groups were analyzed using Student’s t-test or Mann–Whitney U-test for continuous variables according to the normal distribution and Chi-square test or Fisher’s exact test for categorical variables. Multivariate logistic regression with forward stepwise selection based on likelihood ratio was used to identify the risk factors for the development of severe COVID-19 in the derivation cohort, and a nomogram was then constructed by incorporating the independent factors. The calibration of the nomogram was assessed with a calibration curve and the Hosmer–Lemeshow test was performed to assess the goodness-of-fit. The discrimination performance of the nomogram was quantified using AUC. External validation of the nomogram was performed with the validation cohort. DCA was performed by calculating the net benefits of the nomogram for a range of threshold probabilities. The Spearman rank correlation was used in the correlation analysis. All clinical and imaging data items were entered into a worksheet in Microsoft Office Excel 2019. All statistical analyses were performed using SPSS statistics software (version 22.0, IBM Inc., Chicago, IL, USA) and R statistical software (version 3.6.1). A two-sided P value of less than 0.05 considered to be statistically significant.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Data 1Reporting Summary | nature communications | [
"Article"
] | [
"Infectious diseases",
"Medical imaging",
"Risk factors"
] | outbreak coronavirus disease 2019 (COVID-19), severe acute respiratory syndrome coronavirus 2 (SARS spread worldwide pandemic Most COVID-19 patients mild clinical course rapid deterioration 7–14 days into severe illness no specific anti-coronavirus treatment for severe patients remdesivir benefits for severe COVID-19 requires patients poor survival require intensive medical resource utilization fatality rate 20 times higher nonsevere early patients risk severe complications COVID-19 importance studies prevalence severe COVID-19 15.7 to 26.1% associated with abnormal chest computed tomography (CT) findings laboratory Guan severe COVID-19 patients likely show ground-glass opacity patchy shadowing interstitial abnormalities on progression opportunity investigate utility chest CT predictive tool predictive value chest CT for prognosis COVID-19 treatment control disease spread higher CT lung scores poor prognosis in Middle East Respiratory Syndrome (MERS)9 Chest CT proposed ancillary approach for screening suspected COVID-19 pneumonia epidemic monitoring treatment responseenrolled patients moderate COVID-19 pneumonia multiple hospitals observed 14 days early CT clinical risk factors for progression to severe constructed nomogram factors compared clinical CT characteristics in patients different exposure history relationship among CT findings epidemiological features nomogram age neutrophil-to-lymphocyte ratio CT severity score good performance short disease progression in hospitalized patients moderate COVID-19 pneumonia predicting progression risk guiding treatment plans patients COVID-19 identified 51 excluded for negative CT findings or severe/critical COVID-19 admission 39); age younger than 18 (n = 12). 247 patients included study 141 derivation cohort 106 validation cohort (Fig. 1) clinical characteristics in Table 1. 72 (51.1%) 54 (50.9%) patients male median age 44 years 46 years no significant differences clinical characteristics between cohorts hospitalization 15/141 (10.6%) 10/106 (9.4%) patients progressed to severe pneumonia 6/15 (40.0%) 5/10 (50.0%) severe cases deteriorated critical illness 245 (99.2%) patients recovered discharged at. 1Study workflowflow diagram population enrollment observation period 1Clinical characteristics COVID-19 pneumonia derivation validation cohorts 141)Validation 106 valueAge (34–55)46 (35–56)0.695Male gender72 (51.1%)54 (50.9%)0.985Exposure history Wuhan 2 weeks0.075Yes76 (53.9%)45 (42.5%)No65 (46.1%)61 (57.6%)Smoking history7 (23.4%)21 (19.8%)0.499Diabetes8 (5.7%)6 (5.7%)0.996Hypertension21 (14.9% (9.4%)0.200Cardiovascular disease3)0.894COPD4 B)0.703Laboratory findingsLymphocyte count)0.769Aspartate aminotransferase dehydrogenase-reactive protein)17.4)0.191Radiological findingsBilateral involvement123 (87.2%)94 (88.7%)0.731CT severity score6 pneumonia15 (10.6% (9.4% mechanical ventilation6 (4.3%)5 (4.7%)0.862ICU admission4 (10.6%(0.7%)1(0.9%)0.839Data median n (%) Differences groups analyzed Student’s t-test Mann–Whitney U-test variables Chi-square test Fisher’s exact test variables Two-sided P-values.COPD obstructive pulmonary disease COVID-19 coronavirus disease 2019 CT computed tomography intensive care unit IQR interquartile range NLR neutrophil-to ratio.Comparison stable progressive clinical CT characteristics severe COVID-19 Table 2. older 58 41 P = 0.001) likely underlying hypertension (P = 0.004) no difference gender exposure history smoking history co-morbidities diabetes cardiovascular cerebrovascular disease hepatitis B infection clinical symptoms anorexia diarrhea shortness of breath no difference percutaneous oxygen saturation admission = 0.110) lower baseline lymphocyte count albumin higher NLR aspartate aminotransferase lactic dehydrogenase C-reactive protein P < 0.05) major CT features COVID-19 pneumonia bilateral peripheral mixed distributed GGO consolidation GGO consolidation (Fig.progressive group more lobes segments higher crazy-paving sign CT severity score stable P < 0.05) no difference hospital length = 0.398) viral shedding 0.087) 2Clinical CT characteristics stable progressive patients cohort 126)Progressive 15 (33–52)58 (44–66)0.001Male (51.6%)7 (46.7%)0.719Exposure history 2 weeks0.616Yes67 (53.2%)9 (60.0%)No59 (46.8%)6 (40.0%)Smoking (20.6% (46.7%)0.024Diabetes6 (4)0.175Hypertension15 (11.9%)0.004Cardiovascular (1)0.056Cerebrovascular (0)0.729Hepatitis B (3 (73.0%)13 (86.7%)0.252Cough66 (52.4%)8 (53.3%)0.944Sputum (11)0.798Fatigue myalgia28 (22 (20.0%)0.844Anorexia3)0.088Percutaneous oxygen saturation (96.1–98.6)95.6 (94.5–98.5)0.110Laboratory findingsPlatelet count/L)170.5 (137.8–224.0(121.0–204.0 blood cell)0.683Neutrophil count)0.237Lymphocyte count (0.9–1.5 (1.8–3.4)4.8<0.001Alanine aminotransferase (14.5–28.4)19.4)0.794Aspartate aminotransferase (19.6–28.9)30.4 (25.2–37 bilirubin)0.339Albumin)37.5 (35.3–40.2)35.0 (31.7–37.3)0.008Creatinine.8 (40.1–60.3)52.3 (33.6–63.9)0.683Creatine kinase (40.5–122.3)92.0 (57.5–386.0)0.120Lactic dehydrogenase (134.9–217.3.4 (182.8–276.9)0.014C-reactive protein)15.7 (6.8–36.8)36.9 (27.2–58.7)0.001CT segments (10–15)0.007Bilateral involvement110 (87.3%)13 (86.7%)0.944Distribution pattern0.116Peripheral67 (53.2%)4 (26.7%)Central2 (45.2%)11 (73.3%)GGO120)0.388Consolidation107 (84.9%)13(86.7%)0.503Crazy-paving34 (27.0% (53.3%)0.035Air bronchogram71 (56.4% (73.3%)0.207Discrete nodules10 (6.7%)0.862Lymphadenopathy5 (6.7%)0.625Pleural effusion4 (3 severity score6 (4–9)10 (7–15)0.001Hospital length stay (16–28)22 viral shedding (10–24)18 (13–31)0.087Data median n Differences analyzed Student’s t-test Mann–Whitney U-test Chi-square test Fisher’s exact test Two-sided P values obstructive pulmonary disease CT tomography GGO-glass opacities IQR interquartile range NLR neutrophil-to-lymphocyte ratio. chest CT images COVID-19 pneumonia Subpleural GGO crazy-paving right consolidation left air bronchogram pneumonia CT severity score 9 right lung Bilateral peripheral GGO intralobular septal thickening mixed GGO consolidation interlobular septal thickening lungs patchy GGO posterior lungs regression analysis nomogram age ratio 95% confidence interval 1.06.01–1P = 0.028) baseline NLR 95% CI 1.74 [1.13–2.70 = 0.011) CT severity score 95% CI 1.19 [1.01–1.41] P = 0.043) predictors progression severe COVID-19 pneumonia derivation cohort nomogram predictors constructed (Fig. calibration curve nonsignificant Hosmer–Lemeshow test statistic (P = 0.378) good calibration derivation cohort favorable calibration confirmed validation cohort nonsignificant Hosmer–Lemeshow test statistic (P = area receiver operating characteristic curve) 0.867 0.898 good discrimination nomogram performed well validation cohorts deployed online risk calculator decision curve analysis) Supplementary Fig. 2. threshold probability 0.02 to 0.92 nomogram added more net benefit “treat all” “treat none” strategies.Table 3Risk factors progression severe COVID-19 pneumonia derivation cohort.VariablesOR (1.04–1.14)0.0011.06 (1.01–1.12)0.028Hypertension4.93 (1.54–15.82)0.0070.676NLR2.13 (1.18).74 (1.13–2.70)0.012Aspartate aminotransferase1.04 (1.00–1.09)0.0700.682Albumin0.82 (0.71–0.96)0.0110.668Lactic dehydrogenase1.01 (1.00–1.02)0.0060.661C-reactive protein1.03 (1.01–1.06)0.0040 segments.18 (1.04–1.35)0.0130.488Crazy-paving3.09 (1.04–9.18)0.0420.821CT severity score1.32 (1.14–1.54)0.0011.19 (1.01–1.41)0.043Univariate multivariate logistic regression analyses ORs confidence interval CT tomography COVID-19 coronavirus disease 2019 NLR neutrophil-to-lymphocyte ratio odds ratio. 3Development performance nomogram severe COVID-19 pneumonia Calibration curves cohorts 45° blue line perfect prediction dotted red lines predictive performance closer accuracy curves CT characteristics inflammatory positive associations baseline CT severity score inflammatory indexes lactic dehydrogenase C-reactive protein admission134 patients inflammatory indexes day 3 after admission baseline CT severity score positively associated with inflammatory indexes blood cell count neutrophil count lactic dehydrogenase C-reactive negatively associated lymphocyte count 3 potential CT severity score predicting lymphopenia 4Correlation CT characteristics inflammatory indexes correlations baseline CT characteristics inflammatory indexes admission day 3 correlation coefficients P < 0.05.Comparison patients with without bacterial co-infectionBacterial coinfection disease course antibiotics investigated prevalence coinfection in COVID-19 pneumonia patients association clinical CT characteristics cohort 17/141 (12.1%) developed bacterial coinfection during hospitalization Patients older 56 41 P = 0.031) likely underlying hypertension less consolidation on CT admission comparable CT severity score (P = patients with bacterial coinfection longer hospital length of stay duration viral shedding after illness onset no significant difference in clinical CT characteristics between patients with without exposure history Wuhan 2 weeks before illness onset cohort similar age NLR level CT severity scoremedian period to admission 4 days 4 more lobes higher CT severity score no difference age inflammatory indexes CT characteristics COVID-19>4)41 (31–53)46 (35–59)0.147Hypertension8 (11.4%)13 (18.3%)0.251Lymphocyte count)1.1 (0.8–1.5)0.944NLR2.7 (1.8–3.7)2.6 (1.9–3.9)0.867Aspartate aminotransferase)23.8 (19.8–30.4)24.0 (20.0–31.2)0.649Albumin (g/L)38.0.3–40.6)36.4 (34.2–39.0)0.053Lactic dehydrogenase)173.3.1)0.598C-reactive protein (mg/L)18.1 (7.4–38.0)17.4 (6.8–38 lobes (2–5)4 segments (3–12)10 (6–12)0.020Crazy-paving25 (35.7%)17 (23.9%)0.126CT severity score5 (3–10)7 (5–10)0.030Data median (%) Differences analyzed Student’s t-test Mann–Whitney U-test Chi-square test Fisher’s exact test Two-sided P valuesCOVID-19 coronavirus disease 2019 CT computed tomography IQR interquartile range NLR neutrophil-to-lymphocyte ratio study assessed clinical CT characteristics COVID-19 pneumonia patients hospitals identified baseline risk factors progression results CT severity score inflammatory levels older age higher NLR CT severity score admission predictors progression severe COVID-19 pneumonia nomogram risk factors showed good calibration discrimination derivation validation cohorts patients admitted longer symptom onset had more severe lung involvement rapid increase confirmed severe cases management severe patients COVID-19 Timely patients high risk risk stratification management aligned treatment plans optimized utilization medical resource preventing deterioration prevalence severe COVID-19 pneumonia 10% lower large-scale inclusion moderate patients on admission patients cohort younger Wuhan exposure history young middle-aged progressive patients older underlying hypertension patients reports age hypertension risk factors progression COVID-19 pneumonia patients severity lower baseline lymphocyte count higher NLR lactic dehydrogenase C-reactive protein SARSCoV-2 virus lymphocytes induces cytokine storm triggers immune responses17studies decrease peripheral T lymphocyte count inflammatory cytokine milieu cell recruitment infection reduced hyperactivated exhausted cells frequently found in severe cases18 Lymphopenia potential factor disease severity mortality in COVID-1913 damage lymphocytes immunologic abnormality exacerbations uncontrolled inflammatory response stimulate production neutrophils apoptosis NLR systemic inflammatory status used prognosis pneumonia21 Increased NLR lymphocyte count elevated neutrophil count represents damaged lymphocyte function increased inflammatory level risk bacterial infection C-reactive protein serum maker liver inflammation Liu reported predictive disease severity in COVID-19 results suggested higher inflammatory levels higher risk severe COVID-19 CT compared CT characteristics stable progressive patients progressive patients had higher CT severity score severity score pulmonary involvement associated number involved lobes extent lesions24. previous report showed predictive value CT severity score for prognosis mortality9 higher proportion progressive patients showed crazy-paving sign interstitial thickening25SARSCoV-2 protein to angiotensin-converting enzyme II contributes downregulation ACE2 increased pulmonary capillary permeability diffuse alveolar SARS mixed reticular patterns noted second speculated involvement interstitial vascular endothelial cells intralobular septal thickening associated disease severity results revealed age NLR CT severity score admission significant predictors progression in moderate COVID-19 pneumonia patients predictive value age NLR reported recent study MuLBSTA score mortality viral pneumonia lymphopenia multilobe infiltration32 Our findings consistent with imaging evaluation lung involvement constructed nomogram multivariate logistic regression model prediction severe pneumonia COVID-19 patients good performance derivation cohort validation cohort Liang et al. proposed clinical risk score ten variables critical illness COVID included dichotomous chest X-ray abnormality severity adopted quantitative CT severity score assess lung injury predict in-hospital progression risk within 14 days moderate COVID-19 pneumonia prediction model simpler three accessible variablesLike SARS MERS COVID-19 pneumonia progressed rapidly 7–14 days after onset due to cytokine storm increased plasma proinflammatory cytokines1 results revealed association between baseline CT severity score inflammatory markers lymphocyte count day 3 after admission potential value chest CT estimate pulmonary inflammation lung damage predict lymphopenia with bacterial coinfection older likely underlying hypertension risk factors for bacterial infection consolidation on baseline chest CT less likely bacterial coinfection weak antiviral immune response imaging findings may help identify patients higher risk bacterial coinfection prophylactic antibiotic shorten hospital length viral shedding findings limited by small sample size confirmed by larger samples patients admitted more than 4 days after symptom onset had higher CT severity scores lung involvement progression CT examination important timing admission management COVID-19 pneumonia limitations in study retrospectively conducted distribution imbalanced 10% cases developing severe pneumonia adjuvant treatment during hospitalization not analyzed multiple inflammatory cytokines available comprehensive investigation relationship between CT characteristics cytokine storm COVID-19 needsCT not used outside China for COVID-19 guidelines recommend chest CT American College of Radiology advocates first-line sparingly for hospitalized symptomatic patients specific clinical indications applicability results indicated older age higher NLR CT severity score independent risk factors for progression in moderate COVID-19 pneumonia showed favorable predictive accuracy in cohorts Chest CT risk progression disease severity guide timing admission for COVID-19 study conducted Health Insurance Portability and Accountability Act authorized by Hunan Provincial Health Commission clinical radiological data under anonymization Helsinki declaration regulations permission Institutional Review Board of Third Xiangya Hospital with waiver of informed consent patients reported in prior publications clinical records reviewed for patients treated at Third Xiangya Hospital Changsha Public Health Treatment Center Second People’s Hospital of Hunan First Hospital of Yueyang Central Hospital of Shaoyang between January 17, 2020 February 1, 2020. Patients included study if confirmed COVID-19 available chest CT scan on admissionCOVID-19 established World Health Organization guidance confirmed case positive result assay SARS-CoV-2 nasal pharyngeal swab specimens Patients observed 14 days from admission severe pneumonia divided patients cohorts Third Xiangya Hospital Changsha Public Health Treatment Center Second People’s Hospital Hunan derivation cohort First Hospital Yueyang Central Hospital Shaoyang validation cohort.Clinical data collectionDemographic information exposure history smoking history comorbidities diabetes clinical symptoms signs percutaneous oxygen laboratory data (platelet white blood cell neutrophil count lymphocyte alanine bilirubin albumin creatinine kinase lactic dehydrogenase C-reactive protein obtained Exposure history Wuhan 2 weeks before illness local people confirmed SARS-CoV-2 infection clinical classification COVID-19 pneumonia moderate type fever respiratory tract symptoms radiological evidence confirmed pneumoniasevere type respiratory distress rate ≥ 30 beats hypoxia (oxygen saturation ≤ 93% resting hypoxemia (arterial blood oxygen ≤ 300 mmHg). critical type respiratory failure ventilation shock intensive care unit (ICU admission combined other organs failure endpoint severe COVID-19 pneumonia by February 15, 2020 bacterial co-infection ventilation admission discharge death recorded Bacterial coinfection diagnosed symptoms signs radiological evidence nosocomial pneumonia bacteremia positive bacterial culture test lower respiratory tract specimens blood samples Hospital length stay calculated day admission from day discharge Duration viral shedding days symptom onset to negative results respiratory tract viral RT-PCR testing samples tested until two negative results first negative result duration shedding co-authors contributed original data not involved final data summary analysis patients underwent chest CT examinations admission images reconstructed to 1.0-mm thickness transverse scans Sagittal coronal reconstructions 3.0-mm thickness CT images reviewed by two radiologists over 10 years experience chest imagingradiologist consulted disagreement interpreting imaging results imaging features GGO consolidation crazy-paving air bronchogram recorded30 GGO lung attenuation preservation bronchial vascular margins consolidation opacification obscuration margins vessels airway walls Crazy-paving ground-glass opacity interlobular septal thickening intralobular septal thickening lesion distribution lobe segment involvement assessed CT findings outer one third lung peripheral inner two thirds central discrete nodules lymphadenopathy pleural effusion recorded five lung lobes reviewed for opacification consolidation lesions extent evaluated 0 to 5 involvement score 0 none 1 2 6–25% 3 26–50% 4 51–75% 5 >75% total score calculated five lobes CT severity score 0 to 2511.Statistical analysisContinuous variables median interquartile range categorical variables frequency percentage Differences groups analyzed Student’s t-test Mann–Whitney U-test Chi-square test Fisher’s exact test categorical variables Multivariate logistic regression forward stepwise selection risk factors severe COVID-19 derivation cohort nomogram constructed factorscalibration nomogram assessed curve Hosmer–Lemeshow test goodness-of-fit discrimination performance quantified AUC External validation validation cohort DCA net benefits threshold probabilities Spearman rank correlation used clinical imaging data entered worksheet Microsoft Office Excel 2019. statistical analyses SPSS statistics software (version 22.0 IBM Chicago R statistical software 3.6.1). two-sided P value less than 0.05 statistically significant.Reporting information research design Nature Research Reporting Summary.Supplementary Review Additional Supplementary FilesSupplementary Data 1Reporting Summary | 50 | 0.576875 |
10.1038/s41467-020-16867-5 | PMC7299982 | Based on b-value mapping, the author proposes the high likelihood of future rupture in the area of the 2019 Ridgecrest earthquakes. | Monitoring the Earth’s stress state plays a role in our understanding of an earthquake’s mechanism and in the distribution of hazards. Crustal deformation due to the July 2019 earthquake sequence in Ridgecrest (California) that culminated in a preceding quake of magnitude (M) 6.4 and a subsequent M7.1 quake caused stress perturbation in a nearby region, but implications of future seismicity are still uncertain. Here, the occurrence of small earthquakes is compared to larger ones, using b-values, showing that the rupture initiation from an area of low b-values, indicative of high stress, was common to both M6.4 and M7.1 quakes. The post-M7.1-quake sequence reveals that another low-b-value zone, which avoided its ruptured area, fell into an area near the Garlock fault that hosted past large earthquakes. If this area were more stressed, there would be a high-likelihood of further activation of seismicity that might influence the Garlock fault. | IntroductionThe 2019 Ridgecrest earthquakes, which occurred near the town of Ridgecrest, California, included a magnitude (M) 7.1 quake that struck on 5 July 2019 (UTC) as well as active foreshocks and aftershocks1 (Fig. 1a). A M6.4 event preceded the M7.1 quake 34 h later. The M7.1 quake ruptured the Earth’s surface and involved a right-lateral strike slip along a NW-SE trending fault. A predominant mechanism of the M6.4 quake was a left-lateral strike-slip fault motion along a NE-SW trending fault that conjugated with the fault of the M7.1 quake. The broad context of the Ridgecrest earthquakes is that they occurred under the current tectonic stress that created the Eastern California Shear Zone (ECSZ), a seismically active region east of the southern segment of the San Andreas Fault2. The M6.4 quake was followed by more than 1000 perceivable events until the M7.1 quake. A post-M7.1-quake sequence is still active. Over the first 8 months since the M7.1 quake, about 30,000 events with M ≥ 1 occurred, including more than 90 events with M ≥ 4.Fig. 1Ridgecrest earthquake sequence and b-values.a Map of earthquakes in the Ridgecrest region. The cross-section in Fig. 2 extends from P1 to P2 and from P3 to P4 with a width of 8 km. Inset shows the study region (black rectangle). Thin lines indicate major mapped faults. Los Angeles, Santa Barbara, Ridgecrest, and Coso geothermal field are indicated as LA, SB, RC, and Coso GF, respectively. b Map of b-values obtained from seismicity (M ≥ 1) at a depth of 7-13 km before the M6.4 quake. Inset: the b-value map at a depth of 0–7 km. c Plot of b as a function of time before the M6.4 quake for seismicity (depth of 7–13 km) falling in the circle with a radius of r = 10 km (red) and 12 km (green), centered at the M6.4 epicenter. Moving windows cover 100 events. Also included is the magnitude-time dependence.Crustal deformation due to the occurrence of large earthquakes causes stress perturbation in nearby regions. From the viewpoint of the physics of earthquakes, the probability of a subsequent large earthquake depends on the stress conditions set up by the previous events and long-term tectonic state3. Given the tectonic stress of the ECSZ, an investigation into the spatio-temporal state of stress along and near the faults coseismically ruptured by the M7.1 and M6.4 quakes can play a crucial role in understanding the distribution of post-seismic hazards after these quakes. Coulomb stress models were used to explain that the site of the M6.4 quake was stressed by the great 1872 Owens Valley (M~7.6), the 1992 Landers (M7.3), and the 1999 Hector Mine (M7.1) quakes, and that the M6.4 earthquake loaded the site where the M7.1 shock nucleated4–6. However, physics-based approaches employing Coulomb stress transfer have so far not been successful in forecasting upcoming large earthquakes any better than statistical models7. This is partly due to the fact that the locations of potential faults, essential inputs to the calculation of change in Coulomb stress, are unknown8.An alternative statistics-based approach is used to infer changes in the stress state, focusing on the fact that the Ridgecrest earthquakes occurred within the seismically active ECSZ, with data of good enough quality and sufficient abundance, collected by the SCSN (Southern California Seismic Network)9 (for details on the earthquake dataset, see Methods). This approach uses a statistical model based on seismicity: the b-value of the Gutenberg–Richter (GR) law10, given as log10N = a−bM, where N is the cumulative number of earthquakes with a magnitude larger than or equal to M, a characterizes seismic activity or earthquake productivity of a region, and the constant b is used to describe the relative occurrence of large and small events (i.e., a high b-value indicates a larger proportion of small earthquakes, and vice versa). The b-value is sensitive to differential stress, and its inverse dependence on differential stress has been confirmed many times in both laboratory and field studies11–15 (for details on the b-value estimation, see “Methods”).Here, earthquake triggering and characteristics of seismicity before, during, and after the Ridgecrest earthquakes are investigated. In particular, focus is placed on determining maps of b-values for different time periods, showing how the nucleation area for both the M6.4 and M7.1 quakes had low b-values before these events occurred, and mid-to-high b-values thereafter. The b-value map also correlates well with the slip distribution of the M7.1 quake. In addition, the local and time-dependent variations in b-values of the Ridgecrest earthquakes are linked with estimates of changes to Coulomb stress. The main conclusions of this study are that the b-values provide insight into the state of stress in the fault zone, which is likely closely related to the nucleation and evolution of earthquakes in the sequence. This combined approach of b-value and stress-change analyses to the post-M7.1-quake seismicity shows an area that is currently being stressed. Monitoring the spatio-temporal distribution of b, together with other seismological and geodetic observations, will contribute to an appreciation of the seismic hazard in the ECSZ.ResultsTemporal variation associated with stress changesDifferent periods were considered to find time-dependent signals that are consistent with stress increase and release. Two periods that separate data before and in between the two large quakes were selected: first, before the M6.4 quake; and second, 34 h after the M6.4 quake up to the M7.1 quake. For inference on the distribution in seismic hazards, another period that is about eight months after the rupture of the M7.1 quake until 23 March 2020 will be discussed later.Pre-M6.4-quake sequenceA map view (Fig. 1b) based on seismicity before the M6.4 quake with a depth range of 7–13 km shows a zone of low b-values (b ~ 0.6) around the future hypocenter of depth 10.7 km (for details on the mapping procedure, see Methods and Supplementary Figs. 1 and 2). Shallow seismicity (depth of 0–7 km) shows no clear zone of such low b-values near the future epicenter (inset of Fig. 1b). The low-b-value zone was seen, even when the M4.0 quake and its following events that occurred during the last 30 minutes before the M6.4 quake near the eventual hypocenter were excluded from the mapping (Supplementary Fig. 2d). For earthquakes around the M6.4 epicenter (Fig. 1c), the b-values were mostly above 1 until 2010. Since 2010, the b-values have shown a gradual decrease over time, to values near 0.7. The final values are remarkably similar to those immediately before the entire fracture, as was obtained in a previous laboratory experiment11.The M6.4 quake ruptured conjugate faults: the 6-km-long northwest-trending fault first slipped, followed by a slip in the ~15-km-long southwest-trending fault1,2. The initial portion of the M6.4 quake terminated about 4 km from the eventual M7.1 hypocenter. This 4-km gap was progressively filled by a series of moderate-sized earthquakes in the 34 hours after the M6.4 quake, which suggests that this portion of the fault acted as a barrier through which the M6.4 rupture was unable to propagate1. This was confirmed by the cross-sectional views (Fig. 2a, b) for the pre-M6.4-quake period (see Methods and Supplementary Figs. 3–5 for the mapping procedure). Low b-values (b < 0.9: purple to blue) were seen near the M6.4 hypocenter, while high b-values (b > 1: yellow to orange) were seen near the M7.1 hypocenter. This was interpreted as an indication of a weakly stressed area into which the M6.4 rupture was not allowed to propagate.Fig. 2Cross-sectional views of b-values.a
b-values for seismicity (M ≥ 1) before the M6.4 quake along the fault ruptured by the M7.1 quake. Stars shows the M7.1 and M6.4 hypocenters. b Same as a for the cross-section along the fault ruptured by the M6.4 quake. c, d Same as a and b for seismicity before the M7.1 quake. b-values were calculated for the period indicated by c, d in the inset of c from the first event after the M5.4 quake at a relative time of −0.672 days to the last event before the M7.1 quake. The use of seismicity soon after the M6.4 quake was avoided to remove the effect of strong temporal variability in b. Inset: plot of M and completeness magnitude (Mc) as a function of time relative to the M7.1 quake (see Supplementary Fig. 7). e Same as (a) for seismicity after the M7.1 quake. Events during the period from immediately after the M7.1 quake to 25 August 2019 (or 2019.65 decimal years) were not used to calculate b-values for the same reason as (c, d). f Slip distribution of the M7.1 quake23, and events (M ≥ 3) that occurred in the first 12 h. Symbol size is proportional to magnitude. g Top panel: frequency-magnitude distribution of earthquakes falling within a cylindrical volume with a 5‐km radius, centered at the location of the M6.4 hypocenter in (a and c): a with a = 2.44, b = 0.59 ± 0.17, and Mc = 1.5 and c with a = 3.14, b = 1.12 ± 0.33, and Mc = 1.5 (for details on b-value estimation, see Methods). Bottom panel: same as the top one for the location of the M7.1 hypocenter: a with a = 4.03, b = 1.01 ± 0.07, and Mc = 1.6, and c with a = 2.86, b = 0.66 ± 0.22, and Mc = 1.5. Values of logPb ≤ -1.3 indicate a significant difference in b48. h Plot of b as a function of time after the M7.1 quake for seismicity within the rectangle in (e). Plotting procedure is the same as that for Fig. 1c.Pre-M7.1-quake sequenceThe distribution of b-values (Fig. 2c, d) based on seismicity during a period before the M7.1 quake, indicated by the bidirectional arrow in the inset of Fig. 2c, shows a zone of low b-values near the eventual M7.1 hypocenter. A comparison with the pre-M6.4-quake period in Fig. 2a, b shows that an increase in b at the M6.4 hypocenter and a decrease in b at the M7.1 hypocenter are significant (Fig. 2g). The result indicates that the M6.4 rupture relaxed stress near the M6.4 hypocenter, which had been highly stressed before the M6.4 quake, but that it transferred stress to the nearby region of the M7.1 hypocenter, which had acted as a barrier before the M6.4 quake. The result was the erosion of this barrier by seismicity.To confirm that this erosion triggered the M7.1 quake, Coulomb stress transfer was calculated16,17 (for details on the fault models and the stress-change calculation, see Methods), revealing that a region around the hypocenter of the M7.1 quake became about 2 bars closer to failure by the M6.4 quake and its subsequent seismicity (Fig. 3a). To show this map, faults of the M6.4 quake and the relatively large events until immediately before the M7.1 quake were assumed as source faults (Supplementary Fig. 6). For the M6.4 quake, only the southwest-trending fault was assumed. This is because a large slip of the M6.4 quake occurred along the southwest-trending fault rather than along the conjugate northwest-trending fault. The former fault (~15 km long) is much longer than the latter (6 km long). A comparison with a case that only considered the M6.4 quake as a source fault (inset of Fig. 3a) shows that the large changes in Coulomb stress near regions of the M7.1 hypocenter were very likely due to the M6.4 quake as well as its subsequent earthquakes18. Even if the conjugate faults of the M6.4 quake were assumed as source faults, stress in the region near the M7.1 hypocenter increased5.Fig. 3Coulomb stress changes.a Stress changes resolved on the M7.1 quake fault as a result of the M6.4 quake and the following M ≥ 4.5 events. Star indicates the M7.1 hypocenter. Top inset: changes in Coulomb stress as a result of only the M6.4 earthquake, showing that the increase in stress near the region of the M7.1 hypocenter was as high as 1 bar (orange). Bottom inset: source faults projected on the Earth’s surface. The M6.4-quake fault is indicated by a segment because it is assumed to be a vertical plane. Rectangles indicate fault planes of M ≥ 4.5 events. For details on fault models, see Methods. b Changes in stress at a depth of 10 km as a result of the M6.4 and M7.1 quakes. Green segments indicate source faults (M6.4 and M7.1 quakes). Left panel: changes in stress resolved on M6.4-quake-type left-lateral faults (black line with a half-arrow pair). Right panel: changes in stress resolved on M7.1-quake-type right-lateral faults. The rectangle indicates an area of low b-values shown by the rectangle in Fig. 2e that displays the cross-section extending from P1 to P2. See Supplementary Figs. 11 and 12 for other depths.Additional insight into changes in the stress state was provided by temporal behavior of the sequence following the M6.4 quake. Relatively large events occurred early in the post-M6.4-quake sequence (grey stem plot in the inset of Fig. 2c), and the mean magnitude of these events evolved into small values over time. This behavior is well modeled by the Omori-Utsu (OU) power-law aftershock decay19, given as λ ~ t−p, where t is the time since the occurrence of a mainshock; λ is the number of aftershocks per unit time at t with a magnitude greater than or equal to a cutoff magnitude; and p is a constant (for details on the OU law, see Methods). p = 1 is a good approximation20, but spatio-temporal changes in p are observable. M ≥ 3 events were used, taking homogeneity of seismicity recordings into consideration (see Methods and Supplementary Fig. 7 for homogeneity of seismicity recordings). Modeling these events showed that p was smaller for the northern area, including the M7.1 hypocenter, than for the southern area (Fig. 4), revealing that decay in seismicity was slower in the former area than in the latter one (see also Supplementary Fig. 8). This result is interpreted as an indication of a slower decrease in stress in the northern area than in the southern area, according to fictional theory21. This supports the result of a b-value map before the M7.1 quake (Fig. 2c, d) that showed lower b-values (indicative of higher stress) in areas near the M7.1 hypocenter than in areas near the M6.4 hypocenter.Fig. 4Fitting of the OU law.a Plot of p-value of the OU law as a function of the length of the analyzed period since the M6.4 quake, based on seismicity (M ≥ 3) during the period between M6.4 and M7.1 quakes along the fault such that the M7.1 ruptured in the entire area (grey), in the northern area (North of 35.72°N) (blue), and in the southern area (South of 35.72°N) (red). The maximum-likelihood fit was used to determine a p-value. Uncertainties in p were computed by bootstrapping. Open circles for the northern area show p-values obtained based on N ≤ 20 earthquakes. For the periods ≤ 0.5 days, no p-value was obtained for the southern area, because the solution did not converge due to not enough data analyzed. Vertical lines indicating the periods of 1.404, 0.732, and 0 days correspond to the periods ending at the time of the M7.1, M5.4, and M6.4, quakes, respectively. b Number λ (day−1) of seismicity (M ≥ 3) as a function of time from the M6.4 quake for the analyzed period of 1.404 days in the entire area (grey).Low b-values near the M7.1 hypocenter (Fig. 2c, d), together with a temporal decay in seismicity (Fig. 4 and Supplementary Fig. 8), closely match another observation of increased Coulomb stresses near the M7.1 hypocenter (Fig. 3a). The sequence of stress jumps caused by the M6.4 quake and its subsequent events resulted in an increase of roughly 2 bars. This value is not surprising and is comparable to that obtained in previous studies2,5.To support the observation that the events preceding the M7.1 quake very likely played a role in triggering the eventual M7.1 event, an independent analysis from the above stress-related analyses was conducted. This was achieved by investigating if any sign indicative of the M7.1 quake could be found in the spatial organization in seismicity after the M6.4 quake (for details on spatial organization, see Methods). According to a previous study22, the spatial concentration of smaller magnitude events (retrospectively named foreshocks) near the eventual event (retrospectively named mainshock) was a common feature of large earthquakes in southern California. To examine whether this was observed for the M7.1 quake, the quantity ϕ = R−1/Rb−1 was selected22, where R−1 represents the inverse distance from position x to an event that occurred before a given time, averaged over the last n events before this given time, and Rb−1 is the same as R−1 but the average is taken over the second-to-last n events. ϕ>1 indicates a concentration of seismicity before the given time in an area surrounding x, and ϕ < 1 indicates the dispersion of seismicity. A cross-sectional view (Fig. 5) of ϕ-values with n = 25 (a typical value for southern California22) at the time immediately before the M7.1 quake shows a region of seismic concentration (ϕ~1.5) near the hypocenter of this quake. Similar to the p-value analysis, M≥3 events were used for the ϕ-value calculation. Near the future M7.1 hypocenter, there was a gradual increase in ϕ to values above 1, while in other regions, ϕ-values showed low values or a decreasing trend to values of ϕ ~ 1 or below 1. These results depend weakly on n for n = 15-35 (Supplementary Figs. 9 and 10), as was observed in a previous study22. Our results show that the spatial organization of the pre-M7.1-quake sequence in a region near the eventual hypocenter was similar to that observed for previous southern California earthquakes22, but it was dissimilar in other regions. This probably reflects the erosion by active seismicity toward a region near the M7.1 hypocenter. Thus, the spatial clustering before the M7.1 quake was a foreshock-type one indicative of a future mainshock, supporting the observation based on the above stress-related analyses.Fig. 5Seismicity concentration/dispersion.a ϕ-values with n = 25 at the time immediately before the M7.1 quake, based on seismicity (M ≥ 3) during the period between M6.4 and M7.1 quakes along the fault ruptured by the M7.1 quake. Note that seismicity used includes from the first event after the M6.4 quake to the last event before the M7.1 quake. Stars indicate the M6.4 and M7.1 hypocenters. b Same as a for the cross-section along the fault ruptured by the M6.4 quake. (c, d) Plot of ϕ as a function of relative time to the M7.1 quake at the locations indicated by arrows. Uncertainties in ϕ, used to draw error bars in c and d, were computed by bootstrapping (see also Supplementary Fig. 10). The occurrence time of the M6.4 quake, relative to that of the M7.1 quake, is indicated by a grey vertical line at −1.404 days. The relative time of the M7.1 quake is 0 days (grey vertical line), as is obvious. See Supplementary Fig. 9 for other n-values.Post-M7.1-quake sequenceThe M7.1 quake nucleated about 10 km to the northwest of the M6.4 event and its rupture propagated bilaterally, where most slips occurred near the M7.1 hypocenter23. The pre-M6.4-quake b-values (Fig. 2a) were compared with the slip distribution of the M7.1 quake23 (Fig. 2f), showing peak-slip values of 4–5 m around the M7.1 hypocenter (relative distance of −2 to 5 km with depth of −5 to 0 km). It was found that this peak-slip area did not overlap with high b-values (b > 1.1: indicative of low stress), a feature that is common to many other earthquakes24–26. The influence of structural heterogeneity on the spatial distribution of b-values was also noted in such a way that rupture propagation of the M7.1 quake to the northwest terminated at an area near the Coso geothermal production field27 with b > 1.1 (red). The high-temperature area around this field may have contributed to termination of the rupture and high b-values (b > 1.1: red). Similar behavior was observed for the 2016 Kumamoto earthquakes28.To show that coseismic rupture, which caused stress perturbation along the fault of the M7.1 events, played a role in the distribution of post-seismic hazards, the slip distribution of the M7.1 quake23 (Fig. 2f) and the b-value distribution based on post-M7.1-quake seismicity (Fig. 2e) were compared. An area of low b-values (b < 0.9: indicative of high stress), colored in blue to purple, within the rectangle shown in Fig. 2e, does not overlap with volumes of high slip (≥3 m: orange to red) but with volumes that remained unruptured (low slip in Fig. 2f), suggesting that the rupture of this quake released a pronounced amount of overall stress. Note that the rectangle that includes this low-b-value area is located not on the Garlock fault but near it. For events falling within this rectangle, the b-values show a decrease over time, to values around 0.8. The values are not as low as those immediately before the M6.4 and M7.1 quakes (Fig. 1b, c, and 2a–c), but contribute the most recent values in a decreasing trend of the b-value. Similar to laboratory observations of low and decreasing b-values that could previously be detected as a fault of a few centimeters in length that approached failure12,26, this was found for natural earthquakes with faults tens of kilometers in size.DiscussionGiven the current tectonic stress that drives the ECSZ2, it is likely possible to consider that a future activated fault is the one conjugating with the M7.1 rupture, as seen by the M6.4 and M7.1 quake couplet. We calculated changes in Coulomb stress resolved on the M6.4-quake-type left-lateral faults at a depth of 10 km (Fig. 3b), where the source faults are the right-lateral rupture of the M7.1 quake and the left-lateral rupture of the M6.4 quake (for details on the fault models, see Methods). This depth of 10 km was chosen because it is a typical depth of the rectangle with the low-b-value zone in Fig. 2e. The changes in stress pull most of the nearby left-lateral-type faults further from failure (blue lobes, namely stress shadows29) and push others of the same type closer to it (red lobes). We expect strong stress (red) at the region indicated by the rectangle in the left panel of Fig. 3b. Another possibility of future activation is rupture extension to the southeast: namely, the one along the fault of the M7.1 quake. We calculated changes in stress resolved on the M7.1-quake-type right-lateral faults, revealing that faults in the zone of low b-values are again in an area with stress changes to promote failure (red lobes) in the right panel of Fig. 3b. The same stress-change calculations were conducted for different depths (Supplementary Figs. 11 and 12). The result is not induced by a bias of choice of depth: stress patterns for a depth of 8–12 km covering the rectangle shown in Fig. 2e are similar to each other.If the zone of currently low b-values (Fig. 2e) were more stressed (decrease in b-value), seismic activity in this zone would be further enhanced with possibility of future ruptures propagating either along a M6.4-quake-type left-lateral fault or along a M7.1-quake-type right-lateral fault (Fig. 3b and Supplementary Figs. 11 and 12). If so, the influence of a likely future rupture on the Garlock fault would be inevitable. Although this fault has historically been seismically quiescent, it has hosted numerous large earthquakes over several thousand years30, and the last major earthquake occurred about 400 to 500 years ago31. Moreover, geodetic measurements1,18,23 showed that measurable surface creep was triggered by the Ridgecrest sequence, while no measurable creep was shown before the start of this sequence32. The timing of the precursory signal observed in Fig. 2h remains unexplained: the low-b-value patch may continue or subside without the occurrence of a large earthquake. It is not yet possible to make conclusions about the quantitative predictive power of b-value mapping. Thus, together with seismological and geodetic observations, it would be worthwhile to monitor the spatio-temporal distribution of b-values around the southeast rupture terminus of the M7.1 quake, which contributes to seismic hazard in the ECSZ.A question regarding the finding that the Garlock fault may be at risk of rupture due to the existence of a low b-value patch is that the estimate of risk is not quantitative, in the sense of a probability computation. One approach to quantitative evaluation of present level of risk is to apply some type of nowcasting method33 to the Ridgecrest sequence. While we have not examined it in details, previous studies have shown promise in its applications to seismically active regions33–36, and on a worldwide basis37–39. Our future work will be directed at answering this question.MethodsEarthquake datasetThe earthquake catalog produced by the SCSN (http://service.scedc.caltech.edu/eq-catalogs/date_mag_loc.php)9 was used in this study. The SCSN has been in operation for more than 87 years, since 1932, and has recorded and located earthquakes. Station density and technological sophistication have both increased steadily since 1932 leading to increased catalog precision over time.More than 105 earthquakes with M≥1 since 1980 at depths shallower than 20 km within the study region shown in Fig. 1a were processed. In view of the network updates, the completeness levels (Mc~1.5 ± 0.5) were obtained in and around the study region since the 1980s9. The levels are confirmed in the inset of Fig. 2c and Supplementary Figs. 2a and 7.For the location of the M7.1 quake, the hypocenter obtained from the relocated catalog from SCSN (https://scedc.caltech.edu/research-tools/alt-2011-dd-hauksson-yang-shearer.html)40,41 was used because the depth of this quake in the relocated catalog (1.9 km) is much shallower than its depth in the standard (non-relocated) catalog (8.0 km). This difference in depth was considered to be non-negligible because the target depth range in cross-sectional views, such as in Fig. 2, is mainly from 0 to 15 km. The relocated catalog was also used for the hypocenter of the M6.4 quake, although the difference in depth is small (10.7 km for the routinely generated catalog and 11.8 km for the relocated one).b-value estimationSpatial temporal changes in b are known to reflect a state of stress in the Earth’s crust13,15,42 and to be influenced by asperities and frictional properties24,43 and by an interface locking along subduction zones26,44,45. The results of laboratory experiments indicate a systematic decrease in the b-value approaching the time of the entire fracture11,12,14. To estimate b-values homogeneously over space and time, we employed the EMR (Entire-Magnitude-Range) technique46, which also simultaneously calculates the a-value of the GR law and the completeness magnitude Mc, above which all events are considered to be detected by a seismic network (a brief explanation of the EMR technique is provided in the next paragraph). The software package ZMAP47 was used to facilitate computing and mapping b-values based on the EMR method, as described below. EMR applies the maximum-likelihood method when computing the b-value to events with a magnitude above Mc. A b-value was always computed for the corresponding sample only if at least 20 events yielded a good fit to the GR law. Figure 2g shows a good fit of the GR law to observations in the present cases. The top panel of Fig. 2g shows the frequency-magnitude distribution of earthquakes falling within a cylindrical volume with a 5-km radius, centered at the location of the M6.4 hypocenter in Fig. 2a, c. The bottom panel shows the same as the top one for the location of the M7.1 hypocenter. The significant differences between b-values were computed with the Utsu test48 as the probability Pb that the b-values were not different. Values of logPb ≤ −1.3 indicate a significant difference. For both cases in Fig. 2g, the difference in b is significant. This observation is further supported by noting that the absolute difference in b is larger than the sum of uncertainties of the b-values for each of the hypocenters, where the uncertainties in b, as described in the legend of Fig. 2, were computed by bootstrapping. Uncertainty in b is quantified by the standard deviation of the b-values of the bootstrap samples.The EMR technique46 was initially proposed by Ogata and Katsura49,50, who combined the GR law with a detection rate function. Statistical modeling was performed separately for completely detected and incompletely detected parts of the frequency-magnitude distribution. The b- and a-values in the GR law are computed based on earthquakes above a certain magnitude (Mcc). For earthquakes whose magnitudes are smaller than Mcc, it has been hypothesized that the detection rate depends on their magnitudes in such a way that large earthquakes are almost entirely detected while smaller ones are detected at lower rates. Several studies46,49,50 assumed that the detection rate was expressed by the cumulative function of the Normal distribution. Earthquakes with magnitudes greater than or equal to Mcc are assumed to be detected with a detection rate of one. To evaluate the fitness of the model to data, the log-likelihood is computed by changing the value of Mcc. The best fitting model is that which maximizes the log-likelihood.The code of the EMR method is freely available together with the seismicity analysis software package ZMAP (http://www.seismo.ethz.ch/en/research-and-teaching/products-software/software/ZMAP/)47, which is written in Mathworks’ commercial software language Matlab® (http://www.mathworks.com). No knowledge of the Matlab language is needed since ZMAP is GUI-driven, although the ZMAP code is open. ZMAP combines many standard seismological tools. Evaluating spatial variations in seismicity is one of the primary research objectives of ZMAP. By creating a dense spatial grid and sampling overlapping volumes of circular shape, ZMAP users can map b-values calculated by the EMR46. Throughout this study, a grid spacing of 0.01 × 0.01° for map views (Fig. 1b and Supplementary Fig. 2) and 0.5 × 0.5 km for cross-sectional views (Fig. 2 and Supplementary Figs. 4 and 5) was used with a sampling radius r = 5 km, except for Supplementary Figs. 1 and 3 created for different radii r to identify the best representatives among them, as described below.Mapping procedureThe optimal sampling volume (Fig. 1b) was searched by mapping b-values with a wide range of radii r and the largest radius that provided the most detailed resolution of the b-value heterogeneity (inhomogeneity) was selected. The observation of a nearly identical pattern of b-values when sampled with radii of r = 5 km and 7 km suggests that using r < 5 km only reduces coverage (Supplementary Fig. 1). Sampling with r ≥ 9 km results in smoothed b-values and obscures any b-value contrast. Thus, the appropriate radius of the volumes is about 5 or 7 km, because sampling with smaller radii reduces coverage while sampling with larger radii obscures anomalies and contrasts. In making b-value maps throughout this study, earthquakes within a radius of r = 5 km (Fig. 1b and Supplementary Fig. 2c), a small radius between the appropriate radii, were sampled. The EMR technique46 also calculates Mc and a simultaneously, thus the maps of Mc and a, which were created when the b-value map in Fig. 1b and Supplementary Fig. 2c was obtained, are shown in Supplementary Fig. 2a, b. A similar search was conducted for cross-sectional views (Supplementary Fig. 3) and a decision was made to sample earthquakes within a radius of r = 5 km (Fig. 2), the same radius used for the map view in Fig. 1b and Supplementary Fig. 2c.A b-value analysis is critically dependent on a robust estimate of completeness of the processed earthquake data. In particular, underestimates in Mc lead to systematic underestimates in b-values. Attention was always paid to Mc when assessing Mc locally at each node. On the other hand, it is of interest to understand how Mc factors into the conclusion, so an additional test was conducted. This was achieved by creating b-value cross-sections and timeseries for an increased value of every local Mc by 0.2 and 0.5 magnitude units44 (Supplementary Figs. 4 and 5). The spatial and temporal pattern in b generally appears to remain stable with the Mc correction. However, due to a reduced plotted area, it was not possible to judge whether the predictive information in the b-value is contained in the very smallest earthquakes when using small values for the Mc correction or in the intermediate magnitude events when using large values. Future research will be to tackle this problem, using a seismicity catalog similar to that including highly abundant earthquakes, derived from template matching1.Fault modelsThe fault model of the M7.1 quake, which was used in this study, is based on the work by Xu et al.23, one of the currently available fault models of this quake. This is a finite-fault model (also called variable slip) with numerous small patches of slip, each having information on the location of a rectangular patch, strike, dip, and rake. Details of this fault model were obtained via the website created by the same authors (https://topex.ucsd.edu/SV_7.1/index.html)23 and by accessing the Earthquake Source Model Database (SRCMOD), an online database of finite-fault rupture models of past earthquakes (http://equake-rc.info/srcmod/)51. The model represents a northwesterly striking fault with right-lateral planes, showing one main rupture with two sub-parallel strands near the M7.1 hypocenter, to match the Earth’s surface deformation imaged by the Interferometric Synthetic Aperture Radar (InSAR) and others. Note that data obtained by using these imaging tools show exquisite details in the near field of earthquake rupture, in contrast to using teleseismic imaging ones. Considering the location of the hypocenter, the rupture radiation was bilateral. A slip distribution for the main rupture with peak-slip values of 4-5 m is given in Fig. 2f. Average strike, dip, and rake over patches for the main-rupture fault are 136°, 90°, and 176°, respectively, where the standard Aki & Richards sign conventions for fault geometry and slip are used52. The model involves predominantly strike-slip faulting.Similarly, the finite-fault model of the M6.4 quake, again proposed by Xu et al.23, was used. This fault was southwesterly striking with a left-lateral plane. The strike and dip of the plane are 226° and 90°, respectively, and average rake over patches is 6°, revealing a predominant strike-slip fault. A slip distribution with peak-slip values of ~3 m is given at https://topex.ucsd.edu/SV_7.1/index.html23 and http://equake-rc.info/srcmod/51. A previous observation1 shows that the M6.4 quake ruptured conjugate faults that were northwest- and southwest-trending, but the southwest-trending fault was assumed for the M6.4 quake, as described in the main text. Then, the model23 was assumed to be applicable for this fault.A moment tensor solution based on seismograms recorded by the SCSN (http://service.scedc.caltech.edu/eq-catalogs/date_mag_loc.php)9 was used to define M≥4.5 events during the period between the M6.4 and M7.1 quakes as source faults of Coulomb stress changes resolved on the fault of the M7.1 quake (Fig. 3a and Supplementary Fig. 6). To ensure quality of the mechanisms, solutions for events with a reduction in variance >80%, and generated by using at least three stations, were used. Solutions for two M5-class (M5.0 and M5.4) and five M4.5-class events met this criterion. The moment tensor catalog contains two nodal planes. A plane whose strike better matched the lineation in seismicity was chosen.Stress-change calculationStatic stress changes caused by the displacement of a fault (source fault) were calculated (Fig. 3 and Supplementary Figs. 6, 11, 12). Displacements in the elastic half space were used to calculate the 3D strain field; this was multiplied by elastic stiffness to derive the stress changes. A typical value for Poisson’s ratio (PR = 0.25), Young’s modulus (E = 8×105 bar), and friction coefficient (μ=0.4) was used, resulting in a shear modulus of G = E/[2(1+PR)] = 3.3×105 bar. The shear and normal components of the stress change were resolved on specified receiver fault planes. A receiver fault consists of planes, each characterized by specified strike, dip, and rake, on which the stresses imparted by the source faults were resolved. The Coulomb failure criterion, in which failure is hypothesized to be promoted (inhibited) when the Coulomb failure is positive (negative), was used. Coulomb16,17, the graphic-rich deformation and stress-change software for earthquakes, tectonics, and volcanoes, was used to calculate how earthquakes promote or inhibit failure on nearby faults (https://earthquake.usgs.gov/research/software/coulomb/).For the fault ruptured by the M7.1 quake (Fig. 3b), the finite-fault model proposed by Xu et al.23 was used (for details, see the section titled Fault models). The fault model of the M6.4 quake, again proposed by the same authors23, was used. These fault models were combined to create source faults in the Coulomb stress-change calculation in Fig. 3b and Supplementary Figs. 11 and 12. When slip on all patches consisting of a finite-fault model is set to zero, the model with zero slip can be used for a receiver fault. In creating Fig. 3a and Supplementary Fig. 6, which show stress changes resolved on the M7.1-quake fault, slip on all patches consisting of this fault was set to zero.To calculate Coulomb stress changes, shown in Fig. 3a and Supplementary Fig. 6, imparted by the M6.4 quake and the following M ≥ 4.5 events up to the M7.1 quake, the M≥4.5 events needed to be modeled. The SCSN moment tensor data, as described in the section titled Fault models, were used. To make realistically scaled source faults from the moment tensor information, a subroutine program in Coulomb was used16,17. The faults of the M≥4.5 events built by using this program were projected onto the Earth’s surface, and these are shown in the inset of Fig. 3a and Supplementary Fig. 6i, where the M6.4-quake fault is also included. An alternative perspective view to show the faults of the M≥4.5 events and the M6.4-quake is given in Supplementary Fig. 6.Homogeneity of seismicity recordingsSmall events in clusters such as swarms, aftershocks, and foreshocks are often missed in the earthquake catalog, as they are masked by the coda of large events and overlap with each other on seismograms. According to previous cases46,53, Mc depends on time t. In creating the inset of Fig. 2c and Supplementary Fig. 7a, a moving window approach was used, whereby the window covered 100 events. Mc-values from temporal analysis were plotted at the end of the moving window that they represent. Mc decreased with t from nearly 3 and reached a constant value at around 1.6. Relatively large events occurred early in the sequence, and the mean magnitude of these events evolved to small values over time (grey). The time-dependent decrease in Mc is consistent with the data. The use of M ≥ 3 events secures the homogeneity of recording for temporal analysis of seismicity after the M6.4 quake.We compared the EMR method with two frequently-used techniques (Supplementary Fig. 7): Maximum-curvature (MAXC) method and Goodness-of-fit (GOF) method54. The MAXC method defines Mc as the magnitude bin with the highest number of events (red). GOF defines Mc as the lowest magnitude for which the GOF is 90% or larger (green), where the GOF is based on the residual indicating the deviance of the fitted GR law model from the observed data54. Both techniques underestimated Mc (Supplementary Fig. 7a), that is, the EMR method gave the most conservative estimation of Mc, which justifies the use of this method. The same was conducted as Supplementary Fig. 7a for seismicity from 1 January 1980 to immediately before the M6.4 quake along the fault ruptured by the M7.3 quake (from P1 to P2), obtaining the same feature (Supplementary Fig. 7b).OU lawAn exact expression of the OU law19 is given as λ = k(c + t)−p, where t, λ, p are the same as those defined in the main text, and c and k are constants. Similar to the GR case, the maximum-likelihood fit was used to determine the parameters for this law. Uncertainties in p were computed by bootstrapping, where the standard deviation of the p-values of the bootstrap samples quantifies the corresponding uncertainty in p. The codes of the OU-fitting method are available together with the seismicity analysis software package ZMAP47. Similar to the observation that p = 1 is generally a good approximation20, a theory employing the rate- and state-dependent friction law21 also assumes p = 1 as a standard form. Variability in p is possible: p > 1 and p < 1 for special cases with rapidly and slowly decreasing stress, respectively.The inset of Fig. 4 shows a good fit of the OU law to activity with M ≥ 3 for the entire area along the M7.1 rupture (from P1 to P2) during the period between the M6.4 and M7.1 quakes (t = 0–1.404 days: 34 h). Note that setting of a minimum magnitude at M = 3 ensures the homogeneity of recording after the M6.4 quake (for details, see the section titled Homogeneity of seismicity recordings, inset of Fig. 2c, and Supplementary Fig. 7a).Seismicity along the M7.1 rupture was divided into two areas (Fig. 4): northern area (north of 35.72°N, blue data) and southern area (south of 35.72°N, red data), where the northern area includes the M7.1 hypocenter. Several lengths of the analyzed period were also considered. The difference in p between these areas was insignificant for periods shorter than 1 day, beyond which p-values for the southern area were significantly larger than for the northern one. p-values for the northern area were smaller than the typical decay with p~120 if unreliable estimates (open circles) were not taken into consideration. p-values for the southern area showed p > 1 if the longest period (1.404 days) was considered. The result in Fig. 4 was not induced by model selection bias: using the ETAS (Epidemic-type Aftershock Sequence) model55,56, similar results were observed (Supplementary Fig. 8), as described below. A model for seismicity rate resulting from stressing history21 explains the result in Fig. 4 and Supplementary Fig. 8: although the overall trend in stress decreased with time, low p-values (p < 1) in the northern area were due to slower decreasing stress than in the southern area.A more sophisticated model such as the ETAS model55 can be used, instead of the OU model. It was examined whether the ETAS model provided similar results to those obtained by the OU model, using the package SASeis200656 for facilitating statistical analysis of seismicity (https://www.ism.ac.jp/editsec/csm/index_j.html). This package includes a program that can be used to fit the ETAS model to earthquake (aftershock) data. The package also provides modules for plotting figures. Supplementary Figs. 8c and 8d showing a good fit of the ETAS model to observations in the present cases, were created by using one of the modules. For fitting the ETAS model, two time intervals were considered. One interval is called the target interval for which the ETAS model parameters are computed. The seismicity in this period may be affected by earthquakes which occurred before this period due to the long-lived nature of aftershock activity. To consider this effect, the other time interval that is precursory to the target interval (called precursory interval) was chosen and aftershock activities following earthquakes in this period were considered in the computation. The interval between the dashed lines in Supplementary Figs. 8c and 8d is the target interval (0.001 to 1.404 days) for which the ETAS model parameters were computed, and the precursory interval is 0-0.001 days. The correlation between p and the length of the analyzed period was independent of the choice of model for estimating p-values (Fig. 4 and Supplementary Fig. 8a). This suggests that the OU-based approach (Fig. 4) is sufficient to capture the essential aspects of the relaxation process that followed the M6.4 earthquake. Supplementary Fig. 8b shows a case that uses slightly different target intervals and precursory intervals from those in Supplementary Fig. 8a, obtaining a similar result.Spatial organizationA previous study22 using the SCSN earthquake catalog showed that an observation on seismic concentration and dispersion defined by ϕ-values was used to discriminate spatial clustering due to retrospectively named foreshocks from the one induced by aftershocks, and was implemented in an alarm-based model to forecast M > 6 earthquakes. Moreover, the probability of a daily occurrence presented an isolated peak due to a concentration of seismicity closely located in time and space to the epicenter of 5 out of 6 M > 6 earthquakes. A comparison with the present study shows that seismicity after the M6.4 quake displayed a foreshock-type sequence (ϕ>1) in a region near the eventual M7.1 hypocenter and an aftershock-type sequence in other regions.The previous study22 that introduced ϕ=R−1/Rb−1 to seismicity analysis did not take its uncertainty into consideration. A new uncertainty assessment based on a bootstrap method was thus developed. This approach is a Monte Carlo style simulation based on catalogs with permuted distances from position x to events. For each of the locations x where ϕ needs to be calculated, two catalogs are required: one consisting of n events to be used for calculating R−1 and the other consisting of n events for Rb−1 calculation. For the former catalog, events are drawn n times from its population of n events, allowing any event to be selected more than once. From these events, R−1 is computed, as defined in the main text. The same applies to Rb−1 to finally compute ϕ. This process was repeated 300 times, and then errors were estimated as the standard deviation of the ϕ-values, σϕ.σϕ with different values for n onto two cross-sections (from P1 to P2, and from P3 to P4) for the time immediately before the M7.1 quake was mapped, as shown in Supplementary Fig. 10. σϕ-values were also used to draw an error bar at each data point in the ϕ-value timeseries (Fig. 5c, d and Supplementary Fig. 10). General features are that high σϕ-values (σϕ>0.2) fall in regions with high ϕ-values and that the σϕ-value increases with a decrease of n-value. Using the newly developed uncertainty assessment, the significance of the results was quantified. Taking its uncertainty (σϕ) into consideration, ϕ-values in a region near the M7.1 hypocenter can be larger than 1 (seismic concentration) at the time immediately before the M7.1 quake, while ϕ-values in other regions are ϕ ~ 1 or ϕ < 1. Cases for n = 15-35 show that the high ϕ-value anomaly near the M7.1 hypocenter is stable and significant. A case with n = 10 shows an unstable and insignificant result due to small populations, resulting in high σϕ-values.Supplementary information
Peer Review File
Supplementary Information | nature communications | [
"Article"
] | [
"Natural hazards",
"Seismology"
] | 2019 Ridgecrest earthquakes Ridgecrest California 7.1 quake 5 July 2019 active foreshocks aftershocks1 M6.4 event preceded M7.1 quake 34 h M7.1 quake ruptured surface right-lateral strike slip NW-SE fault M6.4 quake left-lateral strike-slip NE-SW fault conjugated M7.1 quake earthquakes under tectonic stress Eastern California Shear Zone seismically active east San Andreas M6.4 quake followed 1000 events until M7.1 quake post-M7.1-quake sequence active 8 months since M7.1 30,000 events M ≥ 1 90 M ≥ 4.Fig. 1Ridgecrest earthquake sequence b-values Map earthquakes Ridgecrest region cross-section P1 to P2 P3 to P4 width 8 km study region lines major faults Los Angeles Santa Barbara Ridgecrest Coso geothermal field indicated LA SB RC Coso GF Map b-values seismicity ≥ 1) 7-13 km before M6.4 quake 0–7 km b function time before M6.quake seismicity 7–13 km circle radius r = 10 km (red) 12 km centered M6.4 epicenter Moving windows cover 100 events magnitude-time dependence.Crustal deformation large earthquakes causes stress perturbation nearby regions physics earthquakes probability subsequent large earthquake depends stress conditions previous events long-term tectonic state3 tectonic stress ECSZ investigation spatio-temporal state stress faults ruptured M7.1 M6.4 quakes distribution post-seismic hazards Coulomb stress models site M6.4 quake stressed 1872 Owens Valley (M 1992 Landers 1999 Hector Mine (M7.1) quakes M6.4 earthquake loaded site M7.1 shock physics-based approaches Coulomb stress transfer not successful forecasting large earthquakes statistical locations potential faults Coulomb stress unknown8alternative statistics approach stress state Ridgecrest earthquakes occurred seismically active ECSZ data good quality collected by SCSN (Southern California Seismic Network approach uses statistical model seismicity b-value of Gutenberg–Richter (GR) law10 log10N = a−bM N cumulative earthquakes magnitude larger than M a characterizes seismic activity region constant b relative occurrence large small events high b-value indicates larger small earthquakes b-value sensitive to differential stress inverse confirmed in laboratory field b-value estimation earthquake triggering characteristics seismicity before during after Ridgecrest earthquakes investigated focus determining maps b-values for time periods nucleation area M6.4 M7.1 quakes low b-values before mid-to-high b-values thereafter b-value map correlates with slip distribution M7.1 quake local time-dependent variations in b-values Ridgecrest earthquakes linked with Coulomb stress conclusions b-values provide insight into state stress fault zone related to nucleation evolution of earthquakescombined approach b-value stress-change analyses post-M7.1-quake seismicity shows area stressed Monitoring spatio-temporal distribution b seismological geodetic observations contribute seismic hazard ECSZ variation stress changesDifferent periods time-dependent signals consistent stress increase release Two periods data before quakes selected first before M6.4 quake second 34 h after M6.4 M7.1 quake inference distribution seismic hazards period eight months after M7.1 quake until 23 March 2020 discussed later-M6.4-quake map view (Fig. 1b before M6.4 quake 7–13 km shows zone low b-values (b ~ 0.6) around future hypocenter 10.7 km Shallow seismicity 0–7 km shows no zone low b-values near future epicenter low-b-value zone seen M4.0 quake events 30 minutes before M6.4 excluded from mapping earthquakes around M6.4 epicenter b-values above 1 until 2010. Since 2010, b-values decrease near 0.7 final values similar to before fracture previous M6.quake ruptured faults 6-km-long northwest-trending fault slipped followed ~15-km-long southwest-trending initial M6.4 quake terminated 4 km from M7.1 hypocenter 4-km gap filled by moderate-sized earthquakes 34 hours after fault barrier M6.4 rupture confirmed by cross-sectional views (Fig. 2a b pre-M6.4-quake period 3–5 Low b-values < purple to blue near M6.4 hypocenter high b-values > yellow to orange near M7.1 hypocenter weakly stressed area M6.4 rupture.Fig. 2Cross-sectional views of b-values seismicity (M ≥ 1) before M6.4 quake along fault ruptured M7.1 M7.1 M6.4 hypocenters ruptured seismicity before M7.1 quake b-values calculated period from first event after M5.4 quake −0.672 days to last event before M7.1 quake use seismicity after M6.4 quake avoided remove strong temporal variability plot M completeness magnitude (Mc) function time relative to M7.1 quake Supplementary Fig. 7)seismicity after M7.1 quake Events to 25 August 2019 not used b-values Slip distribution M7.1 events (M ≥ 3) first 12 h Symbol size proportional magnitude Top panel frequency-magnitude distribution earthquakes 5‐km radius M6.4 hypocenter a a = 2.44 b = 0.59 ± 0.17 Mc = 1.5 c a = 3.14 b 1.12 ± 0.33 Mc = 1.5 details b-value estimation Bottom panel M7.1 hypocenter a a = 4.03 b = 1.01 ± 0.07 Mc = 1.6 c a = 2.86 b = 0.66 ± 0.22 Mc = 1.5 logPb ≤ -1.3 significant difference b48 Plot b function time after M7.1 quake seismicity rectangle same Fig. 1c.Pre-M7.1-quake distribution b-values (Fig. 2c seismicity before M7.1 quake low b-values near M7.1 hypocenter comparison pre-M6.4-quake increase b M6.4 hypocenter decrease M7.1 hypocenter significant M6.4 rupture relaxed stress near M6.4 hypocenter stressed beforequake transferred stress to M7.1 hypocenter barrier before M6.4 quake result erosion barrier by seismicity erosion triggered M7.1 quake Coulomb stress transfer calculated16 region around hypocenter M7.1 quake 2 bars closer to failure by M6.4 quake subsequent seismicity (Fig. 3a). faults of M6.4 quake large events before M7.1 quake assumed as source faults M6.4 quake southwest-trending fault assumed large slip occurred former fault (~15 km long longer than (6 km comparison with M6.4 quake large changes in Coulomb stress near M7.1 hypocenter likely due to M6.4 quake subsequent earthquakes18 faults M6.4 quake stress in near M7.1 hypocenter increased5.Fig. 3Coulomb stress changes Stress resolved on M7.1 quake fault M6.4 quake following M ≥ 4.5 events Star indicates M7.1 hypocenter changes in Coulomb stress M6.4 earthquake increase stress near M7.1 hypocenter high as 1 bar source faults projected on Earth’s surface M6.4-quake fault indicated by segment vertical plane.Rectangles indicate fault planes M ≥ 4.5 events see Methods. Changes stress depth 10 km M6.4 M7.1 quakes Green segments indicate source faults (M6.4 M7.1 Left panel changes stress M6.4-quake-type left-lateral faults Right panel M7.1-quake-type right-lateral faults rectangle indicates low b-values Fig. 2e P1 to P2. See Supplementary Figs. 11 12 insight stress state temporal behavior sequence M6.4 quake large events occurred early post-M6.4-quake sequence (grey stem plot Fig. mean magnitude evolved small values over time modeled Omori-Utsu (OU) power-law aftershock decay19 λ ~ t−p t time mainshock λ number aftershocks per unit time t magnitude cutoff magnitude p constant Methods). p = 1 good spatio-temporal changes p observable M ≥ 3 events used homogeneity seismicity recordings Methods Fig. 7 Modeling showed p smaller northern area M7.1 southern area decay seismicity slower former Supplementary Figresult slower decrease stress northern southern fictional supports b-value map before M7.1 quake (Fig. 2c lower b-values higher stress near M7.1 hypocenter than M6.4 hypocenter.Fig. 4Fitting OU law Plot p-value function length analyzed period since M6.4 quake seismicity (M ≥ 3) between M6.4 M7.1 quakes northern southern maximum-likelihood fit p-value Uncertainties p computed by bootstrapping Open circles northern area show p-values N ≤ 20 earthquakes periods ≤ 0.5 days no p-value southern area not enough data Vertical lines periods 1.404 0 days correspond periods ending M7.1 M5.4 M6.4 quakes Number λ (day−1) seismicity (M ≥ 3) function time from M6.4 quake analyzed period 1.404 days entire area (grey).Low b-values near M7.1 hypocenter (Fig. 2c temporal decay in seismicity match increased Coulomb stresses near M7.1 hypocenter (Fig. 3a). sequence stress jumps M6.quake subsequent events increase 2 bars not surprising comparable previous events preceding M7.1 quake eventual M7.1 event independent analysis stress-related analyses conducted M7.1 quake spatial organization seismicity after M6.4 quake previous spatial concentration smaller magnitude events foreshocks near eventual event mainshock common large earthquakes southern California M7.1 quake quantity φ = R−1/Rb−1 R−1 inverse distance from position x to event before averaged over last n events Rb−1 same average second-to-last n events φ>1 indicates concentration seismicity before x φ < 1 indicates dispersion seismicity cross-sectional view (Fig. 5) of φ-values with n = 25 typical southern before M7.1 quake shows region seismic concentration (φ~1.5) near hypocenter M≥3 events used for φ-value calculation Near future M7.1 hypocenter gradual increase in φ to values above 1 other regions φ-values showed low decreasing trend to φ ~ 1 below 1. results depend on n for n = 15-35 Figs. 9 observed previousresults show spatial organization pre-M7.1-quake sequence near hypocenter similar previous southern California dissimilar other regions reflects erosion active seismicity near M7.1 hypocenter spatial clustering before M7.1 quake foreshock-type future mainshock supporting stress analyses.Fig. 5Seismicity concentration/dispersion φ-values n = 25 before M7.1 quake seismicity (M ≥ 3) between M6.4 M7.1 quakes fault ruptured seismicity first event M6.4 to last event before M7.1 Stars indicate M6.4 M7.1 hypocenters cross-section fault ruptured M6.4 Plot φ function relative time M7.1 quake locations Uncertainties φ computed by bootstrapping Fig occurrence time M6.4 quake M7.1 grey vertical line −1.404 days relative time M7.1 quake 0 days See Supplementary Fig. 9 n-values-M7.1-quake M7.1 quake nucleated 10 km northwest M6.4 rupture propagated bilaterally most slips near M7.1 pre-M6.4-quake b-values compared with slip distribution M7.1peak-slip values 4–5 m around M7.1 hypocenter distance −2 to 5 km −5 to 0 km). peak-slip area overlap with high b-values (b > 1.1 low common to other earthquakes24–26 influence structural heterogeneity on distribution b-values rupture propagation M7.1 quake terminated near Coso geothermal production field27 with b > 1.1 high-temperature area contributed to termination rupture high b-values > 1.1 Similar behavior 2016 Kumamoto earthquakes28 coseismic rupture stress perturbation M7.1 post-seismic hazards slip distribution M7.1 quake23 2f b-value distribution post-M7.1-quake seismicity (Fig. 2e) compared area of low b-values (b < 0.9 high blue to purple overlap with volumes high slip (≥3 m volumes unruptured (low rupture released stress rectangle low-b-value area not on Garlock fault near it b-values over time to around 0.8. values not as low as before M6.4 and M7.1 quakes contribute in decreasing trend b-valuelaboratory observations low decreasing b-values detected as fault few centimeters approached found for natural earthquakes with faults tens of kilometers current tectonic stress ECSZ2 likely future activated fault conjugating with M7.1 rupture M6.4 M7.1 quake couplet calculated changes Coulomb stress on M6.4-quake-type left-lateral faults at depth 10 km (Fig. 3b), source faults right-lateral rupture M7.1 left-lateral rupture M6.4 depth 10 km typical depth of rectangle low-b-value zone. 2e changes stress pull nearby left-lateral-type faults further from failure push others closer expect strong stress (red) at region rectangle left Fig. 3b possibility future activation rupture extension southeast along fault M7.1 quake calculated changes stress on M7.1-quake-type right-lateral faults faults in zone low b-values area stress changes promote failure lobes same stress-change calculations for different depths result not induced bias depth stress patterns for depth 8–12 km rectangle similar zone of low b-values (Figstressed (decrease b-value), seismic activity zone enhanced future ruptures propagating M6.4-quake-type left-lateral fault or M7.1-quake right-lateral fault (Fig. 3b Supplementary Figs. 11 and 12). influence likely future rupture on Garlock fault inevitable fault historically seismically quiescent hosted large earthquakes last major earthquake 400 to 500 years ago31 geodetic measurements1 showed surface creep triggered by Ridgecrest sequence no creep before start sequence32 timing precursory signal Fig. 2h unexplained low-b-value patch may continue or subside without large earthquake possible conclusions quantitative predictive power of b-value mapping monitor-temporal distribution of b-values around southeast rupture terminus of M7.1 quake contributes seismic hazard ECSZ Garlock fault risk rupture low b-value patch estimate risk not quantitative approach risk apply nowcasting method33 to Ridgecrest sequence previous studies shown promise applications seismically active regions33–36 worldwide future work answering.MethodsEarthquake datasetThe earthquake catalog SCSN.scedc.caltechedu/eq-catalogs/date_mag_loc.php used in study SCSN 87 years since 1932 recorded located earthquakes Station density technological sophistication increased since 1932 increased catalog precision 105 earthquakes M≥1 since 1980 depths shallower than 20 km study region Fig. 1a processed completeness levels (Mc~1.5 ± 0.5) obtained since 1980s9 levels confirmed in Fig. 2c Supplementary Figs. 2a 7.For location M7.1 quake hypocenter from relocated catalog SCSN used depth relocated catalog (1.9 km) shallower than standard (non catalog (8.0 km). difference depth non-negligible target depth range cross-sectional views 0 to 15 km relocated catalog used for hypocenter M6.4 quake difference depth small (10.7 km catalog 11.8 km relocated-value estimationSpatial temporal changes reflect stress Earth’s crust13 influenced by asperities frictional interface locking along subduction indicate decrease in b-value approaching fracture11estimate b-values over space time EMR (Entire-Magnitude-Range) technique46 calculates a-value GR law completeness magnitude Mc above events detected by seismic network explanation EMR software package ZMAP47 computing mapping b-values EMR method EMR applies maximum-likelihood method b-value events magnitude above Mc b-value computed if 20 events good fit to GR law Figure 2g shows good fit GR law observations cases top panel shows frequency-magnitude distribution of earthquakes cylindrical volume 5-km radius M6.4 hypocenter bottom panel same M7.1 hypocenter significant differences between b-values computed with Utsu test48 probability Pb Values logPb ≤ −1.3 indicate significant difference both cases difference in b significant absolute difference b larger than sum of uncertainties b-values hypocenters uncertainties computed bootstrapping Uncertainty b quantified by standard deviation of b-values bootstrap samples EMR technique46 proposed by Ogata Katsura49 combined GR law with detection rate function Statistical modeling performed separately for completely detected incompletely detected parts frequency-magnitude distributionb- a-values in GR law computed earthquakes above magnitude (Mcc). earthquakes smaller than Mcc hypothesized detection rate depends on magnitudes large earthquakes detected smaller lower studies46 assumed detection rate by cumulative function of Normal distribution Earthquakes magnitudes greater than or equal to Mcc assumed detected with detection rate one model log-likelihood computed by changing Mcc best fitting model maximizes log-likelihood code EMR method available seismicity analysis software ZMAP written in Mathworks’ Matlab® No knowledge Matlab language needed ZMAP GUI-driven code open combines standard seismological tools Evaluating spatial variations in seismicity primary research creating dense spatial grid sampling overlapping volumes circular shape ZMAP users map b-values calculated by EMR46 grid spacing of 0.01 × 0.01° for map views 0.5 × 0.5 km for cross-sectional views used sampling radius r = 5 km 1 different optimal sampling volume (Fig1b searched mapping b-values wide radii r largest radius detailed resolution-value heterogeneity selected identical pattern b-values radii r = 5 km and 7 km suggests r < 5 km reduces coverage Fig. 1) r ≥ 9 km smoothed b-values obscures contrast appropriate radius 5 or 7 km smaller radii reduces coverage larger obscures anomalies contrasts earthquakes within radius r = 5 km sampled EMR technique46 calculates Mc and a simultaneously maps of Mc and a created shown in Fig. 2a b similar search for cross-sectional views 3) decision sample earthquakes within radius r = 5 km (Fig. 2) same b-value analysis dependent on robust estimate completeness processed earthquake data underestimates in Mc lead to underestimates b-values Attention paid to Mc assessing understand Mc factors conclusion additional test conducted achieved creating b-value cross-sections timeseries for increased value local Mc by 0.2 and 0.5 magnitude units44 (Supplementary Figs. 4 and 5) spatial temporal pattern in b stable with Mc correctionreduced plotted area possible judge predictive information b-value smallest earthquakes or intermediate magnitude events Future research tackle problem seismicity catalog abundant earthquakes template.Fault model M7.1 quake based on work Xu et al.23 finite-fault model variable slip) small patches of slip information location rectangular patch strike dip rake Details obtained via website authors (https://topex.ucsd/SV_7.1/index Earthquake Source Model Database finite-fault rupture models-rc model represents northwesterly striking fault right-lateral planes one main rupture two sub-parallel strands near M7.1 hypocenter Earth’s surface deformation Interferometric Synthetic Aperture Radar (InSAR) data imaging tools show details near field earthquake rupture teleseismic imaging rupture radiation bilateral slip distribution main rupture peak-slip values 4-5 m in Fig. 2f Average strike, dip rake patches main-rupture fault are 136°, 90° 176° standard Aki & Richards sign conventions for fault geometry slip model involves predominantly strike-slip faultingfinite-fault model M6.4 quake proposed by Xu et al used fault southwesterly striking left-lateral plane strike dip 226° 90° rake 6° predominant strike-slip fault slip distribution peak-slip ~3 m at https://topex.ucsd/SV_7.1/index http://equake-rc/srcmod/51 M6.4 quake ruptured faults northwest southwest-trending southwest-trending fault assumed for M6.4 quake model23 applicable for fault moment tensor solution seismograms SCSN M≥4.5 events M6.4 M7.1 quakes source faults Coulomb stress changes M7.1 quake (Fig. 3a Fig 6) solutions events reduction variance >80% three stations used Solutions two M5-class five M4.5-class events met moment tensor catalog contains two nodal planes strike matched lineation seismicity chosen.Stress-change calculationStatic stress changes displacement fault calculated (Fig. 3 6 11 Displacements elastic half space 3D strain field multiplied by elastic stiffness stress changes Poisson’s ratio (PR = 0.Young’s modulus (E = 8×105 friction coefficient (μ=0.4) used shear modulus G = E/[2(1+PR)] = 3.3×105 bar shear normal components stress change resolved on receiver fault planes receiver fault planes strike dip rake stresses source faults resolved Coulomb failure criterion failure used Coulomb16,17 deformation stress-change software for earthquakes tectonics volcanoes failure on nearby faults.usgs.gov fault ruptured by M7.1 quake (Fig. 3b), finite-fault model Xu et al.23 used fault model M6.4 quake used models source faults in Coulomb stress-change calculation in Fig. 3b Supplementary Figs. 11 12. slip on patches finite-fault model set to zero used for receiver fault Fig. 3a Supplementary Fig. 6 stress changes M7.1-quake fault slip on all patches set to zero calculate Coulomb stress changes M6.4 quake M ≥ 4.5 events M7.1 quake M≥4.5 events modeled SCSN moment tensor data usedscaled source faults from moment tensor information subroutine program Coulomb faults M≥4.5 events projected Earth’s surface shown in Fig. 3a Supplementary Fig. 6i M6.4-quake fault included alternative perspective view faults M≥4.5 M6.4-quake Supplementary Fig. 6.Homogeneity seismicity recordingsSmall events clusters swarms aftershocks foreshocks missed in earthquake catalog masked by large events overlap seismograms Mc depends on time t Fig. 2c Supplementary Fig. 7a moving window approach used covered 100 events Mc-values plotted end moving window Mc decreased with t from 3 constant around 1.6 large events occurred early sequence mean magnitude evolved to small values over time time-dependent decrease in Mc consistent with data use M 3 events secures homogeneity for temporal analysis seismicity after M6.4 quake compared EMR method with techniques Fig. 7) Maximum-curvature (MAXC) Goodness-of-fit (GOF) MAXC defines Mc magnitude highest number events GOF lowest magnitude 90% deviance law model Both techniques underestimated Mc EMR method conservative estimation Mc justifies use methodconducted Supplementary Fig. 7a seismicity 1 January 1980 M6.4 quake fault ruptured M7.3 quake P1 to P2) same feature Fig. 7b).OU expression OU law19 λ = k(c + t)−p t λ p same text c k constants maximum-likelihood fit parameters Uncertainties p computed bootstrapping standard deviation p-values quantifies uncertainty codes OU-fitting method available seismicity analysis software package ZMAP47 p = 1 good approximation20 rate- state-dependent friction law21 assumes p = 1 Variability p possible p > 1 p < 1 special cases decreasing stress Fig. 4 shows good fit OU law activity M ≥ 3 area M7.1 rupture P1 to P2) between M6.4 M7.1 quakes (t = 0–1.404 days 34 minimum magnitude M = 3 ensures homogeneity recording after M6.4 quake Homogeneity seismicity recordings Supplementary Fig. 7a).Seismicity M7.1 rupture divided two areas (Fig. 4) northern southern includes M7.1 hypocenter lengths analyzed period considereddifference in p between areas insignificant for periods shorter than 1 day beyond p-values southern area larger northern p-values northern smaller than typical decay p~120 if unreliable estimates-values southern showed p > 1 if longest period (1.404 days) result Fig. 4 not induced by model selection bias ETAS similar results observed 8) model for seismicity rate stressing explains result trend stress decreased with time low p-values (p < 1) in northern area due to slower decreasing stress sophisticated ETAS model55 used instead of OU model examined ETAS model similar results to OU model using package SASeis200656 for analysis seismicity includes program ETAS model to earthquake data provides modules for plotting figures Supplementary Figs. 8c and 8d fit ETAS model observations created For fitting ETAS model two time intervals considered target interval ETAS model parameters computed seismicity may affected by earthquakes aftershock activityeffect time interval precursory target interval chosen aftershock activities earthquakes considered interval between dashed lines Supplementary Figs. 8c 8d target interval (0.001 to 1.404 days ETAS model parameters computed precursory interval 0-0.001 days correlation between p length analyzed period independent model p-values (Fig. 4 OU-based approach (Fig. 4) relaxation process M6.4 earthquake Fig. 8b different target precursory intervals similar result.Spatial previous study22 SCSN earthquake observation seismic concentration dispersion φ-values spatial clustering foreshocks aftershocks implemented forecast M > 6 earthquakes probability daily occurrence isolated peak concentration seismicity epicenter 5 out of 6 M > 6 earthquakes comparison present study seismicity after M6.4 quake foreshock-type sequence (φ>1) region near M7.1 hypocenter aftershock-type sequence other regions previous study22 introduced φ=R−1/Rb−1 seismicity analysis uncertainty new uncertainty assessment bootstrap method developed approach Monte Carlo style simulation catalogs permuted distances from position x to eventsx φ two catalogs required one n events R−1 other n Rb−1 former catalog events drawn n times from population n selected more than once R−1 computed same applies Rb−1 φ process repeated 300 times errors estimated as standard deviation φ-values n two cross-sections P1 to P2 P3 to P4) before M7.1 quake Supplementary Fig. 10. σφ-values error bar each data point φ-value timeseries (Fig. 5c d high σφ-values (σφ>0.2) in regions with high φ-values σφ-value increases with decrease n-value uncertainty assessment significance results quantified φ-values near M7.1 hypocenter larger than 1 before M7.1 quake other regions φ ~ 1 or φ < 1. Cases n = 15-35 show high φ-value anomaly near M7.1 hypocenter stable significant case n = 10 unstable insignificant result small populations high σφ-values.Supplementary information Peer Review File | 49.7 | 0.755274 |
10.1038/s41467-020-15304-x | PMC7090076 | Polyion complex vesicles have advantageous applications but have poorly tuneable permeability and nanostructure stability. Here, the authors develop polyion complex vesicles created using a stimuli-triggered transition from polymersomes and show high stability and switchable bilayer permaselectivity. | Compared to liposomes, polymersomes of block copolymers (BCPs) possess enhanced stability, along with compromised bilayer permeability. Though polyion complex vesicles (PICsomes) from oppositely charged block polyelectrolytes possess semipermeable bilayers, they are unstable towards physiologically relevant ionic strength and temperature; moreover, permselectivity tuning of PICsomes has remained a challenge. Starting from a single component diblock or triblock precursor, we solve this dilemma by stimuli-triggered chemical reactions within pre-organized BCP vesicles, actuating in situ polymersome-to-PICsome transition and achieving molecular size-selective cargo release at tunable rates. UV light and reductive milieu were utilized to trigger carboxyl decaging and generate ion pairs within hydrophobic polymersome bilayers containing tertiary amines. Contrary to conventional PICsomes, in situ generated ones are highly stable towards extreme pH range (pH 2-12), ionic strength (~3 M NaCl), and elevated temperature (70 °C) due to multivalent ion-pair interactions at high local concentration and cooperative hydrogen bonding interactions of pre-organized carbamate linkages. | IntroductionMimicking intricate structures and functions of complex biological systems such as cell membranes and viral capsids has inspired the creation of artificial nanostructures including liposomes and polymeric vesicles (polymersomes)1–6, both consisting of aqueous interiors enclosed by hydrophobic bilayers7–11. Compared to liposomes self-assembled from small molecule lipids, polymersomes of amphiphilic block copolymers (BCPs) possess much improved microstructural stability. They have been increasingly utilized to fabricate delivery of nanovehicles12–14, nanoreactors15–18, and artificial organelles3,19,20. However, effective permeation of active agents through/from aqueous lumens is prohibited by the hydrophobicity of thick bilayer membranes1,11. To solve this issue, several approaches such as membrane integration with channel proteins, copolymerization with responsive moieties, and bilayer post-modification have been proposed3,21–25. These approaches typically involve tedious procedures and need to introduce external additives, leading to unsatisfactory permselectivity and loss of structural integrity in certain cases. We previously proposed the concept of “traceless crosslinking” and achieved concurrent vesicle crosslinking and membrane permeabilization26–28. However, the nature of irreversible bilayer chemical crosslinking restricts their in vivo applications considering biodegradation and clearance issues; in addition, it still remains a challenge to achieve molecular size-selective release and permselectivity regulation features.In contrast to polymersomes with hydrophobic bilayers, polyion complex vesicles (PICsomes) constructed from two oppositely charged block polyelectrolytes possess semipermeable bilayers and improved permeability towards hydrophilic solutes29–31. However, PICsomes are intrinsically unstable under physiologically relevant temperature and ionic strength due to dynamic exchange with unit PICs, and their use as in vivo nanocarriers also need covalent stabilization via chemical crosslinking30,32–36. Though the stability issue has recently been partially solved by strategies, such as introducing guanidinium hydrogen bonding (HB) motif37 and longer alkyl spacer38 into the cationic block, and use of strong block polyelectrolyte39,40 or polyions with a dendritic topology41,42, PICsomes still encounter several major limitations. First, its fabrication process is not compatible with hydrophobic drugs and imaging agents29. Second, the formation of PICsomes involves two oppositely charged components with at least one of them being block polyelectrolytes29–31; the possibility of PICsome formation from a single component with oppositely charged comonomers arranged in a random, block, or alternating manner (e.g., block polyampholytes) remains to be explored43,44. Finally, the permselectivity modulation of PICsome bilayers has not been achieved yet, and macromolecular agents up to a molar mass of 10 kDa (e.g., dextran) could easily permeate though PICsome bilayers30. Thus, the loading capability and encapsulation stability of PICsomes towards hydrophilic small molecule drugs and functional agents need to be further improved.Although both polyion complex (PIC) micelles45–47 and PICsomes29–31 are intrinsically sensitive to high ionic strength, pH, and temperature40, previous literature reports also hint, in a retrospective view, that the stability of PICs or inter-polyelectrolyte complexes (IPECs) is highly dependent upon local concentrations and sequence arrangement of charged ion-pairs43,44,48–51. Sun et al.48 fabricated tough and viscoelastic polyamphoyte hydrogels via direct copolymerization of oppositely charged comonomers at >1.5 M total concentration in aqueous media containing 0.5 M NaCl. The resultant supramolecular hydrogels are stable towards high salt concentrations and elevated temperatures. For model peptides containing a well-positioned single pair of histidine and aspartate partially buried in an alanine-rich hydrophobic milieu, ion-pair interactions assisted by HB are not screened by high ionic strength up to ~1.0 M NaCl, contributing to helix stability51. Thus, rationally designed ion-pair interactions with preferred orientation and high density could provide Coulomb attraction instead of repulsion52.We then envisage that attractive ion-pair interactions could be exploited to stabilize nanostructures, such as PICsomes against physiologically relevant conditions if they are present at high local density and cooperatively assisted by other types of noncovalent interactions (e.g., HB, π–π, and hydrophobic). By taking advantage of the high local concentration of functional moieties (~1–2 mol/L) within hydrophobic bilayers of self-assembled polymersomes52, we herein propose a general strategy to in situ generate ion-pair interactions and trigger transformation from polymersomes to PICsomes (Fig. 1).Fig. 1Stimuli-triggered polymersome-to-PICsome transition and concurrent permeability regulation.For polymersomes self-assembled from amphiphilic diblock and triblock copolymers containing tertiary amine and caged carboxyl moieties in the hydrophobic block, external stimuli including UV light and reductive milieu trigger the decaging of carboxyl functionalities, which transfer protons onto neighboring tertiary amine moieties and render ion-pair interactions. Note that in situ generated PICsomes are highly stable towards extreme pH range, high ionic strength, and elevated temperature due to cooperative ion-pair interactions at high local concentration within pre-organized vesicle bilayers and synergistic contributions from hydrogen bonding interactions of carbamate side linkages. The polymersome-to-PICsome transition is accompanied with the transformation of hydrophobic bilayers into semipermeable membranes and switching of vesicle bilayer permeability. Moreover, sequence structure of the bilayer forming block, A-(B-co-C) vs. A-B-C, could be further utilized to regulate the permselectivity of resultant semipermeable PICsome bilayers.Polymersomes were self-assembled from amphiphilic BCPs containing tertiary amine (DPA or DEA) and caged carboxyl comonomers (NCMA or DCMA) in the stimuli-responsive block, being initially hydrophobic (Fig. 2). We intentionally integrate carbamate linkage into both tertiary amine-containing comonomers and caged carboxyl comonomers to strengthen intra- and inter-chain HB interactions within vesicle bilayers26–28,52, which provide a pre-organized milieu for cooperative noncovalent interactions with directional orientation and high local concentration. Upon triggered cleavage and generation of carboxyl moieties within bilayers, proton transfer from carboxyl to tertiary amine in situ generates ionized carboxyl and protonated amine moieties. Ion-pair interactions are strengthened by carbamate-relevant HB interactions and the hydrophobic bilayer-forming scaffold, contributing cooperatively to the high stability of in situ formed PICsomes towards extreme pH, high ionic strength, and elevated temperature.Fig. 2Chemical structures of four types of amphiphilic block copolymers (BCPs) used in this study.The strategy of in situ-triggered polymersome-to-PICsome transition bridges two main types of vesicles assembled from BCPs, starting from a single component block polyampholyte precursor. This feature differs from that of conventional PICsomes, which involves two oppositely charged polyion components. We further demonstrate that both UV light and reductive milieu could trigger polymersome-to-PICsome transition accompanied with the switching of vesicle bilayer permeability, manifesting the generality of the proposed strategy. Moreover, the permselectivity of resultant PICsomes could be finely tuned by the sequence arrangement of polyions (random vs. block type), and molecular size-selective release from PICsome interiors could be successfully achieved (Fig. 1). Note that this feature has not been previously realized for conventional PICsomes.ResultsSynthesis and self-assembly of amphiphilic BCPsAiming to fulfill the above design rationale, two series of carbamate-containing monomers possessing caged carboxyl and tertiary amine functionalities were designed (Supplementary Fig. 1). 2-Nitrobenzyl ester photo-caged carboxyl monomer (NCMA) and disulfide-caged carboxyl monomer (DCMA) generate carboxyl moieties upon actuation of UV irradiation and reductive milieu, respectively26–28. For the second series, pH-responsive DPA and DEA monomers containing both carbamate linkage and tertiary amine moieties were synthesized (Supplementary Fig. 1c)53. These four types of monomers (NCMA, DCMA, DPA, and DEA) and relevant synthetic intermediates were well-characterized by 1H and 13C NMR analysis (Supplementary Figs. 1 and 6–10). Next, reversible addition-fragmentation chain transfer (RAFT) polymerizations using PEG45-based macroRAFT agent afforded a series of amphiphilic BCPs with varying comonomer sequences and compositions, including PEO45-b-P(NCMA0.55-co-DPA0.45)29, PEO45-b-P(NCMA0.49-co-DEA0.51)32, and PEO45-b-P(DCMA0.45-co-DPA0.55)33 diblock copolymers, and PEO45-b-PNCMA17-b-PDPA21 triblock copolymer (Fig. 2, and Supplementary Figs. 2 and 3). These BCPs were characterized by 1H NMR and GPC analyses (Supplementary Figs. 11–14) and their structural parameters are summarized in Supplementary Table 1.For the hydrophobic block in these diblock and triblock copolymers, an almost equal ratio of caged carboxyl and tertiary amine comonomers was chosen. Upon decaging, carboxyl and tertiary amine moieties will be at roughly equivalent molar ratio, thus facilitating cooperative ion-pair interactions within vesicle bilayers (Fig. 1)14. Additionally, two types dye-labeled amphiphilic BCPs, PEO45-b-P(NCMA0.55-co-DPA0.45)29-Nile red and PEO45-b-P(NCMA0.55-co-DPA0.45)29-naphthalimide were also synthesized to fabricate vesicles conjugated with microenvironmental polarity-sensitive and pH-sensitive fluorescent probes (Supplementary Fig. 4). Furthermore, two types of control amphiphilic BCPs, PEO45-b-PNCMA30 and PEO45-b-PPA26, were also synthesized (Supplementary Figs. 5 and 15). For as-synthesized diblock and triblock copolymers, tertiary amine comonomer units are located within hydrophobic microenvironment containing caged carboxyl moieties, which considerably suppresses the apparent pKa for tertiary amines. For example, the amine pKa was determined to be ~5.7 for PEO45-b-P(NCMA0.55-co-DPA0.45)29, in contrast to the apparent pKa of ~6.4 for PDPA homopolymer (Supplementary Fig. 16)53.BCP self-assembly was triggered by slow addition of water into the polymer solution in acetone. Transmission electron microscopy (TEM) observation revealed the presence of typical vesicular nanostructures for all four types of BCPs (Fig. 3a and Supplementary Fig. 17). Dynamic laser light scattering (DLS) analysis revealed that resultant polymersomes possess intensity-average hydrodynamic diameters, 〈Dh〉, in the range of 530–640 nm and relative low polydispersities (μ2/Γ2 ~ 0.1), which are in general agreement with TEM results (Supplementary Table 1).Fig. 3TEM images of PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersomes and corresponding PICsomes.Original polymersomes a and PICsomes obtained after 10 min UV irradiation in neutral aqueous media b; c, e Polymersomes after being subjected to pH 4.0 and pH 12.0, exhibiting microstructural destruction in acidic media; d, f PICsomes in aqueous media at pH 2.0 and 12.0.Light-triggered polymersome-to-PICsome transitionNext, we investigated UV light-triggered evolution of PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersomes in aqueous media. As photo-labile NCMA and DPA comonomers are in a random sequence within the hydrophobic block, they should be in close contact with each other within polymersome bilayers. Upon UV irradiation, the process of photocleavage of 2-nitrobenzyl units and generation of carboxyl moieties were monitored by time-dependent UV/Vis absorption spectra (Supplementary Fig. 18). The photocleavage occurred quickly within the initial ∼5 min and then leveled off at ~10 min under 365 nm LED light irradiation. UV-triggered carboxyl decaging process was further examined with 1H NMR (Supplementary Fig. 19), demonstrating a photocleavage extent of >98% upon UV irradiation for 10 min. We also attempted to characterize UV-triggered chemical structural changes by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) technique, but no reliable signals could be recorded for both non-irradiated and UV-irradiated polymersome dispersions upon lyophilization (Supplementary Fig. 20).Apparently, the vesicular dispersion initially exhibited a bluish tinge, but changed to grayish yellow after UV irradiation (insets in Supplementary Fig. 21). Furthermore, 1H NMR spectra of the vesicle dispersion in D2O before and after UV irradiation showed only signals of well-solvated PEO coronas in both cases (Supplementary Fig. 21), indicating the presence of colloidal aggregates after UV-triggered carboxyl decaging. Moreover, direct TEM observations after UV irradiation revealed the presence of intact and robust vesicular nanostructures (Fig. 3b).Time-dependent DLS measurements were conducted to track the evolution of scattered light intensities and intensity-average hydrodynamic diameter, 〈Dh〉, upon UV irradiation (Fig. 4a). Scattered light intensities of PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersome dispersion exhibited an initial ∼15% decrease within ∼5 min UV irradiation and then reached a plateau, this was accompanied with a slight increase of 〈Dh〉 from 550 to 570 nm upon 10 min UV irradiation. In addition, the Dh distribution remained almost unchanged upon UV irradiation (Fig. 4b). Moreover, if we replace the N,N-diisopropylamine motif in the above BCP with N,N-diethylamine, the resultant PEO45-b-P(NCMA0.49-co-DEA0.51)32 BCP (Fig. 2 and Supplementary Table 1) should possess higher apparent pKa for tertiary amines (Supplementary Fig. 16). Again, both DLS and TEM characterization results confirmed UV light-triggered polymersome-to-PICsome transition, indicating the generality of the proposed strategy (Supplementary Fig. 22). To exclude possible UV-triggered photoreaction and/or decomposition of tertiary amine moieties in BCPs, we further examined the photostability of both N,N-diisopropylethylamine and triethylamine against UV irradiation, revealing essentially no discernible chemical structural changes (Supplementary Figs. 23 and 24).Fig. 4DLS characterization of polymersome-to-PICsome transition upon UV irradiation.a Irradiation duration-dependent evolution of scattered light intensities and 〈Dh〉 recorded for PEO45-b-P(NCMA0.55-co-DPA0.45)29 vesicles in neutral aqueous media. b Intensity-average hydrodynamic diameter, Dh, distributions before and after UV irradiation.The above results indicated that vesicles possess microstructural integrity after UV irradiation although light-triggered cleavage of 2-nitrobenzyl ester generated hydrophilic carboxyl moieties. Upon carboxyl generation, proton transfer from carboxyl to tertiary amine moieties will occur. This is reasonable considering that most of tertiary amines are initially in the unprotonated state (pKa ~ 5.7, Supplementary Fig. 16), and carboxyl functionalities of the control BCP, PEO45-b-PPA26, possess an apparent pKa of ~6.5 (Supplementary Fig. 25). Most importantly, the mutual presence of neighboring amine and carboxyl functionalities, and the generated hydrophilic milieu will considerably increase and decrease pKa values of amine and carboxyl moieties, respectively, thus facilitating proton transfer and ion-pair formation. Indeed, it is well-accepted that protonated amines tend to increase the acidity of neighboring carboxylic acids by stabilizing the conjugate base (carboxylate) via electrostatic interactions. We tentatively propose that upon UV decaging, vesicular nanostructures are stabilized by newly generated ion-pair interactions at high local concentrations and synergistically enhanced by side chain HB interactions (Fig. 1 and Supplementary Fig. 26). Note that before UV irradiation, original polymersomes are mainly stabilized by hydrophobic interactions; whereas UV-irradiation in situ generates extensive ion-pair interactions within vesicle bilayers, thus corresponding to light-actuated polymersome-to-PICsome transition. For resultant PICsomes, potentiometric titration experiments revealed that ion-pairs form within the pH range of 4.6–8.8; whereas below pH 4.6 and above pH 8.8, ionized carboxylates and protonated amines did not exist within vesicle bilayers, respectively (Supplementary Fig. 27).To probe the extent of proton transfer and ion-pair formation, we fabricated polymersomes conjugated with pH-sensitive fluorescent probe via the co-assembly of PEO45-b-P(NCMA0.55-co-DPA0.45)29 and PEO45-b-P(NCMA0.55-co-DPA0.45)29-naphthalimide BCPs (8:2 molar ratio) (Fig. 5a). Note that in the latter, the tertiary amine of naphthalimide probe possesses an apparent pKa comparable to those of PDPA homopolymer (Supplementary Fig. 16)54,55. As shown in Supplementary Fig. 28, for naphthalimide-conjugated polymersome dispersion in aqueous media, decreasing solution pH from 7 to 4 led to emission enhancement due to the loss of PET quenching for protonated tertiary amines; whereas in the pH range of 7–9, fluorescence emission exhibited only a slight decrease, which is in agreement with the apparent amine pKa (~5.7) for PEO45-b-P(NCMA0.55-co-DPA0.45)29 (Supplementary Fig. 16). For PICsomes obtained via UV irradiation, pH-dependent transition range of emission intensities shifted to higher pH, and the degree of amine protonation was determined to be ~51.2% at pH 7 (Supplementary Fig. 29).Fig. 5In situ proton transfer from newly generated carboxyl to amine species and formation of ion-pairs.a Schematics of the fabrication of naphthalimide-labeled polymersomes co-assembled from PEO45-b-P(NCMA0.55-co-DPA0.45)29 and PEO45-b-P(NCMA0.55-co-DPA0.45)29-naphthalimide (8:2 wt/wt); initially, naphthalimide emission is partially quenched by neighboring tertiary amine moieties via PET mechanism; upon decaging of carboxyl moieties and protonation of amine moieties, naphthalimide emission is prominently enhanced. b Evolution of fluorescence emission spectra during polymersome-to-PICsome transition upon UV irradiation. c Irradiation time-dependent evolution of the protonation fraction of tertiary amine moieties within vesicle bilayers during polymersome-to-PICsome transition. All data were obtained at a polymer concentration of 0.1 g/L at [NCMA] ~ 0.12 mM and [DPA] ~ 0.10 mM in Britton–Robinson buffer (pH 7.0, 12 mM; 25 °C).During UV light-triggered polymersome-to-PICsome transition under Britton–Robinson buffer media (pH 7.0, 12 mM; 25 °C), fluorescence emission intensities increased rapidly within the first 3–4 min (Fig. 5), which agrees with decaging kinetics shown in Supplementary Figs. 18 and 19. We further confirmed that upon UV light irradiation, naphthalimide dye exhibited negligible photobleaching (Supplementary Fig. 30). The above results clearly confirmed in situ proton transfer from newly generated carboxyl to amine species and formation of ion-pairs (Fig. 1). Moreover, on the basis of results shown in Supplementary Figs. 28 and 29 for dispersions of polymersomes and PICsomes before and after UV irradiation, respectively, the extent of amine protonation increased from ~7.8% to ~51.2% during polymersome-to-PICsome transition (Fig. 5c). This indicated that ~50% carboxyl and amine moieties formed ion-pairs within resultant PICsomes (Fig. 1).For the control BCP, PEO45-b-PNCMA30, its polymersome dispersion in neutral aqueous media exhibited an apparent pH decrease from ~7.4 to ~6.4 during UV irradiation (Supplementary Fig. 31). This agrees with the apparent pKa of ~6.5 for PEO45-b-PPA26 (Supplementary Fig. 25), the chemical structure of which is the same as decaged PEO45-b-PNCMA30 (Supplementary Fig. 5). On the other hand, for PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersome dispersion, UV irradiation only resulted in slight pH decrease from ~7.4 to ~7.2 (Supplementary Fig. 31), further confirming internal proton transfer and ion-pair formation, i.e., light-triggered polymersome-to-PICsome transition.For decaged PEO45-b-P(NCMA0.55-co-DPA0.45)29 directly generated via UV irradiation in DMSO solution, its self-assembly was actuated by using slow water addition method. As shown in Supplementary Fig. 32, we could only observe the formation of PIC micelles of non-uniform size distribution. This is in distinct contrast to UV-triggered PICsome formation (Fig. 3b), revealing the importance of pre-organization within precursor polymersome bilayers before triggered PICsome formation.PICsome nanostructures stabilized by multiple interactionsThe microstructural stability of both polymersomes and PICsomes is a prerequisite towards their potential applications in complex biological milieu1–5,29–31. We then examined the stability of polymersomes and resultant PICsomes towards pH, temperature and high ionic strength. As shown in Fig. 6a, original polymersomes (without UV irradiation) exhibited microstructural integrity over the pH range of 5–12; whereas below pH 5, vesicle disintegration occurred due to protonation of tertiary amine comonomers within bilayers (Supplementary Fig. 26, left column).Fig. 6Characterization of polymersomes and corresponding PICsomes.pH-dependent scattered light intensities a and intensity-average Dh distributions b recorded for aqueous dispersions of PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersomes and PICsomes in the pH range of 2–12. c Irradiation duration-dependent evolution of zeta potentials during polymersome-to-PICsome transition in neutral aqueous media. d Variation of zeta potentials for aqueous dispersions of polymersomes and PICsomes in the pH range of 2–12. All data were obtained at a polymer concentration of 0.1 g/L with [NCMA] ~ 0.12 mM and [DPA] ~ 0.10 mM in Britton–Robinson buffer (pH 2–12, 12 mM; 25 °C). The error bars indicate standard deviation (n = 3).As for PICsomes obtained via UV irradiation, its structural integrity was maintained over the pH range of 2–12, as confirmed by pH-dependent Dh distributions (Fig. 6b). Meanwhile, TEM observations confirmed that original polymersomes remained to be stable in neutral and alkali media, but disassembled into irregular aggregates under acidic media (pH < 4.0) (Fig. 3c, e). However, after UV irradiation, the resultant PICsomes exhibited highly microstructural stability in the pH range of 2–12 (Fig. 3d, f).Microstructural stability of PICsomes was further corroborated by zeta potential measurements during the process of polymersome-to-PICsome transition and against varying pH conditions. In neutral aqueous media, zeta potential values remained almost constant to be ca. +6 mV during UV irradiation (Fig. 6c). In the pH range of 7–12, zeta potentials of original polymersomes are in the range of ±5 mV. It slightly increased to ~+8 mV at pH 4, and no detectable data could be obtained when further decreasing pH to ~2–3 due to microstructural disassembly. As for PICsomes, the pH-dependent variation of zeta potentials in the pH range of 4–10 is similar to that of polymersomes. However, in the more acidic and alkaline media (pH 2 and 12), zeta potential values exhibited significant changes (+32 mV at pH 2 and −15 mV at pH 12), which demonstrated the structural integrity of PICsomes. The results are in good agreement with TEM and DLS characterization results (Figs. 3 and 6a).The stability of PICsomes under extreme pH conditions is quite unexpected, considering that electrostatic interactions of ion-pairs will be considerably weakened due to carboxyl protonation and amine deprotonation in strongly acidic and alkaline media, respectively (Supplementary Figs. 26–29). We tentatively ascribe the observed pH stability to the following two possible reasons48,50,52,56: (i) HB interactions between carboxyl moieties in the protonated state, together with cooperative HB and π–π interactions between benzyl carbamate side linkages, and HB interactions between carboxyl and carbamate moieties, could explain the stability at pH 2; (ii) concerning the stability under alkaline media, HB and π–π interactions between benzyl carbamate side linkages should be mainly responsible, in addition to the hydrophobic nature of non-protonated amine comonomer units (Fig. 7a).Fig. 7PICsome nanostructures stabilized by multiple and cooperative noncovalent interactions.a Microstructural stability of PICsomes contributing from multivalent ion-pair, hydrogen bonding, and π–π interactions at high local concentration due to the pre-organized nature of vesicle bilayers. b–d Intensity-average Dh distributions recorded for the aqueous dispersion of PICsomes (fabricated from PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersomes via UV irradiation for 10 min) at varying temperatures, ionic strengths, and combinations of them. All data were obtained at a polymer concentration of 0.1 g/L with [NCMA] ~ 0.12 mM and [DPA] ~ 0.10 mM in neutral aqueous media.Encouraged by the above pH stability of PICsomes, we further investigated the stability of PICsomes towards high ionic strengths and elevated temperatures. After obtaining PICsomes in situ via UV irradiation, different amount of NaCl salt was added, with the final concentration in range of 0–3.0 M. DLS results revealed that as-prepared PICsomes are very stable up to an NaCl concentration of 3.0 M, and Dh distributions remain almost the same with the variation of NaCl concentrations (Fig. 7b). In addition, PICsomes in neutral aqueous media are also stable towards elevated temperatures (25–70 °C); again, Dh distributions exhibit negligible changes in the temperature range investigated (Fig. 7c).Next, we challenge PICsomes with a combination of salt and elevated temperatures. In the presence of 2.0 M NaCl, PICsomes are stable up to ~50 °C. Further heating to even higher temperatures leads to the evolution of Dh distributions towards lower size ranges. At 70 °C, PICsomes disassemble into unimers and no reliable DLS signals could be detected (Fig. 7d and Supplementary Fig. 33). This is quite reasonable considering that ionic interactions and HB interactions will be largely suppressed by high salt concentrations and elevated temperatures. The disintegration of PICsomes at 70 °C in the presence of 2.0 M NaCl also confirmed that they are stabilized by cooperative noncovalent interactions, instead of chemical crosslinking (Fig. 1). Note that this is also in agreement with 1H NMR characterization data recorded in DMSO-d6 for lyophilized polymersome dispersion after being subjected to UV irradiation (Supplementary Fig. 19)48,50. It is interesting to note that original polymersomes (without UV irradiation) are very stable towards both high temperature (70 °C) and high ionic strength (2.0 M), and a combination of them (Supplementary Fig. 34), which should also be ascribed to cooperative carbamate-relevant HB interactions and the hydrophobic nature of vesicle bilayers (i.e., incompatible with NaCl salt).Effects of block sequences on permselectivity regulationAs demonstrated above, robust PICsome nanostructures could be in situ fabricated from PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersomes in aqueous media via UV-triggered carboxyl decaging. The bilayer forming block is a random copolymer of DPA and NCMA. We speculated that if DPA and NCMA comonomers are arranged in the block instead of random sequence, UV decaging of NCMA will afford triblock polyampholytes consisting of two oppositely charged blocks43,44, which is more comparable to conventional PICs29–31,45–47. PEO45-b-PNCMA17-b-PDPA21 with chemical compositions comparable to that of PEO45-b-P(NCMA0.55-co-DPA0.45)29 BCP was then synthesized (Fig. 2).According to similar procedures used for the diblock copolymer, PEO45-b-PNCMA17-b-PDPA21 triblock copolymer also self-assembled into polymersomes with quite uniform size distribution, as determined by TEM (Figs. 1 and 8a). Accordingly, UV irradiation of the triblock polymersome dispersion also generated carboxyl and ion-pairs within bilayers, leading to polymersome-to-PICsome transition. As shown in Fig. 8b, the vesicular nanostructure was well maintained, and the overall dimension of resultant PICsomes is larger than that of original polymersomes. DLS measurements indicated that the 〈Dh〉 of vesicles exhibited a rapid increase within the initial ∼3–4 min UV light irradiation; the increase of 〈Dh〉 from 540 to 770 nm during polymersome-to-PICsomes transition (Fig. 8c), which is quite prominent compared to that of PEO45-b-P(NCMA0.55-co-DPA0.45)29 vesicles (Fig. 4a; from 550 to 570 nm upon UV irradiation). It is worthy of noting that during the whole transition process, the vesicle size distribution remained to be quite uniform (μ2/Γ2 ∼ 0.10; Fig. 8d). To verify the importance of pre-organization within bilayers of precursor polymersomes before triggered PICsome formation, PEO45-b-PNCMA17-b-PDPA21 was subjected to direct decaging upon UV irradiation in DMSO, followed by self-assembly in aqueous media. We could only observe the formation of micellar nanoparticles instead of polymeric vesicles (Supplementary Fig. 35).Fig. 8Light-triggered microstructural evolution of triblock copolymer vesicles.TEM images of PEO45-b-PNCMA17-b-PDPA21 polymersomes a and corresponding PICsomes b. c Irradiation duration-dependent evolution of scattered light intensities and 〈Dh〉 recorded during polymersome-to-PICsome transition for PEO45-b-PNCMA17-b-PDPA21 vesicles in neutral aqueous media. d Evolution of intensity-average Dh distributions upon UV irradiation. All scale bars are 500 nm. All data were obtained at a polymer concentration of 0.1 g/L in neutral aqueous media.Drastically different extents of swelling during UV-actuated polymersome-to-PICsome transition for PEO45-b-P(NCMA0.55-co-DPA0.45)29 diblock and PEO45-b-PNCMA17-b-PDPA21 triblock copolymers reflected the effects of comonomer sequences upon formation of ion-pairs (Figs. 1 and 2). For diblock vesicles, newly generated carboxyl functionalities are in close proximity with tertiary amine moieties due to the random copolymerization nature. Thus, the local concentration of ion-pair interactions will be very high. In contrast, for triblock vesicles, initially generated carboxyl functionalities are more apart from amine moieties, leading to frustrated formation of ion-pairs and bilayer swelling due to partial ionization and protonation of carboxyl/amine residues (Fig. 8c). Only at later stages of UV irradiation, PIC formation between oppositely charged blocks (i.e., decaged NCMA block and DPA block) will occur, prohibiting further bilayer swelling. This explains that vesicle sizes remain almost constant after ~4 min UV irradiation.UV-triggered polymersome-to-PICsome transition should be accompanied with the transformation of bilayer polarity from being hydrophobic to hydrophilic due to the loss of hydrophobic 2-nitrobenzyl ester residues and generation of ion-pairs. We further examined this issue by using polymersomes covalently conjugated with a microenvironmental polarity-sensitive fluorescent probe, i.e., Nile red. As shown in Supplementary Fig. 36, for polymersomes co-assembled from PEO45-b-P(NCMA0.55-co-DPA0.45)29 and PEO45-b-P(NCMA0.55-co-DPA0.45)29-Nile red (8:2, molar ratio) in aqueous media, the initial Nile red emission at ~610 nm is quite strong, indicating the hydrophobic nature of polymersome bilayers. Upon UV irradiation, considerable decrease of emission intensities is clearly evident, accompanied with slight red shift of emission maxima. This confirms the generation of hydrophilic PICsome bilayers upon UV irradiation (Fig. 1). For diblock and triblock copolymer vesicle discussed above, different extents of swelling during polymersome-to-PICsome transition also hint varying mesh sizes and bilayer permeability for resultant PICsomes.To probe changes in bilayer permeability, we loaded several types of water-soluble anticancer drugs in the aqueous interior of polymersomes and examined release profiles upon UV light-actuated polymersome-to-PICsome transition (Fig. 9a). For gemcitabine hydrochloride (299.7 Da) loaded within polymersomes of PEO45-b-P(NCMA0.55-co-DPA0.45)29 without UV irradiation, ∼25% cumulative release was observed over 32 h (Fig. 9b). Upon UV irradiation for 2 and 5 min, sustained release of gemcitabine hydrochloride was achieved, with cumulative release extents being ~85% and ~97% after 32 h, respectively. However, doxorubicin hydrochloride (Dox·HCl) with larger molar mass (580.0 Da), only <5% drug could be released over 32 h from original polymersomes, and PICsomes upon UV irradiation (Fig. 9c). As both gemcitabine hydrochloride (299.7 Da) and Dox·HCl bear positive charges, the above discrepancy of release profiles should be ascribed to different molecular sizes, suggesting the excellent permselectivity of resultant PICsomes. This is in stark contrast to conventional PICsomes, from which the release of dextran with molar mass up to ~10 kDa could still be released30.Fig. 9Light-regulated polymersome-to-PICsome transition for molecular size-selective drug release.a, d Schematics of drug-loaded vesicles of PEO45-b-P(NCMA0.55-co-DPA0.45)29 diblock copolymer a and PEO45-b-PNCMA17-b-PDPA21 triblock copolymer d, and triggered release from corresponding PICsomes. b, c, e, f Release profiles of encapsulated b, e gemcitabine hydrochloride and c, f Dox·HCl from the aqueous lumen of PEO45-b-P(NCMA0.55-co-DPA0.45)29 diblock copolymer b, c and PEO45-b-PNCMA17-b-PDPA21 triblock copolymer e, f vesicles before and after UV irradiation. All data were obtained at a polymer concentration of 0.1 g/L in PB buffer (pH 7.4, 10 mM, 37 °C).We further checked permeabilities of neutral and negatively charged drugs through PICsome bilayers. As shown in Supplementary Fig. 37, neutral 2′-deoxy-5-fluorouridine (5-Fu; 246.2 Da) and negatively charged coumarin-343 (model drug; 285.3 Da) could be on-demand released from vesicles of PEO45-b-P(NCMA0.55-co-DPA0.45)29 via triggered polymersome-to-PICsome transition. On the other hand, the release of negative charged calcein with higher molar mass (622.6 Da) from both polymersomes and PICsomes are completely prohibited. The above results indicated that the polymersome-to-PICsome transition is accompanied with permeability switching of vesicle bilayers and molecular size-selective release of encapsulated drugs could be successfully achieved. This could be ascribed to the presence of paired ionic interactions at high local concentration within PICsome bilayers, providing accurate mesh size control by physical crosslinking through a combination of electrostatic ion-pair interactions and side chain HB interactions.For polymersomes of PEO45-b-PNCMA17-b-PDPA21 triblock copolymer, the hydrophobic-to-hydrophilic transition of bilayers also occurs during light-triggered polymersome-to-PICsome transition, as confirmed by fluorescence measurements of physically encapsulated Nile red probe (Supplementary Fig. 38). In contrast to PEO45-b-P(NCMA0.55-co-DPA0.45)29 vesicles, PICsomes of the triblock copolymer exhibited on-demand sustained release for gemcitabine hydrochloride, Dox·HCl, and calcein (Fig. 9d–f and Supplementary Fig. 39); note that PICsomes of the former exhibited almost no release of Dox·HCl and calcein (Fig. 9d and Supplementary Fig. 37). It is also intriguing to note that release rates of gemcitabine hydrochloride, Dox·HCl, and calcein from triblock PICsomes (5 min UV irradiation) decreased in the order of increasing molar mass and number of charges (∼100% cumulative release of gemcitabine hydrochloride over 16 h; ∼84% Dox·HCl release over 32 h, and ∼49% calcein release over 32 h, respectively). The dramatically enhanced permeability of PICsomes of the triblock copolymer compared to that of diblock copolymer could be safely ascribed to the significant swelling (from 540 to 770 nm; ~43% increase in 〈Dh〉) during light-triggered polymersome-to-PICsome transition; whereas diblock vesicles only exhibited an 〈Dh〉 increase ∼3.6% (from 550 to 570 nm). The block sequence-dependent permselectivity of in situ fabricated PICsomes from precursor polymersomes augurs well for their practical applications as both drug nanocarriers and nanoreactors. In addition, UV irradiation duration could serve as another dimension to modulate bilayer permeability and selectivity, with longer UV irradiation affording accelerated release of encapsulated functional agents (Fig. 1). To the best of our knowledge, the permselectivity modulation by both block copolymer sequences and magnitude of external stimuli has not been achieved before.Reduction-triggered polymersome-to-PICsome transitionThe previous sections established that light-triggered polymersome-to-PICsome transition affords ultrastable vesicles with excellent bilayer permselectivity towards a series of drug molecules of varying molar mass and number of charges. We further generalized the design by actuating polymersome-to-PICsome transition using reductive milieu trigger (Fig. 1). Note that the redox gradient across cell membranes is universal for all natural organisms14,57,58. For PEO45-b-P(DCMA0.45-co-DPA0.55)33 diblock copolymer containing disulfide-caged carboxyl comonomers (DCMA; Fig. 2), its self-assembly in aqueous media again afforded polymersomes of ~600 nm in diameter, with quite uniform size distribution (Fig. 10a).Fig. 10Reduction-triggered polymersome-to-PICsome transition and permselectivity regulation.a, b TEM images recorded for reductive milieu-responsive PEO45-b-P(DCMA0.45-co-PDPA0.55)33 vesicles a before and b after treating with 10 mM GSH in neutral aqueous media. c, d Incubation time-dependent evolution of c scattered light intensities and d intensity-average Dh distributions upon treating with 10 mM GSH. e, f Release profiles of e 5-Fu and f Dox·HCl from aqueous interiors of vesicles before and after treating with GSH at varying concentrations. All scale bars represent 1 μm. All data were obtained at a polymer concentration of 0.1 g/L in PB buffer (pH 7.4, 10 mM, 37 °C).Reduction-triggered evolution of vesicular microstructures was then explored by TEM and DLS measurements (Fig. 10b–d). Upon GSH addition, disulfide cleavage is accompanied with spontaneous 1,6-rearrangement of benzyl moieties14,57–59, generating carboxyl functionalities and leading to ion-pair formation (Fig. 1). The reduction-triggered decaging process was further examined with 1H NMR, exhibiting a decaging extent of >99% upon treating with 10 mM GSH for 24 h (Supplementary Fig. 40). Although we could observe the apparent decrease of scattered light intensities, 〈Dh〉 distributions remained almost unchanged upon treating with GSH; the vesicular integrity was further confirmed by TEM observations (Fig. 10b). Thus, although the reductive trigger and light trigger act on different time scales (hours vs. minutes), both could actuate polymersome-to-PICsome transition, with the formation of stable PICsomes with hydrophilic bilayers.Finally, we examined the permselectivity of in situ-fabricated PICsomes of PEO45-b-P(DCMA0.45-co-DPA0.55) upon actuating with reductive trigger (i.e., GSH). As shown in Fig. 10e, encapsulated model anticancer drug (5-Fu) exhibited retarded release from both original polymersomes and those treated with ∼2 μM GSH (comparable to extracellular and blood circulation milieu), with cumulative release extents being ∼14% and ∼19% over 32 h, respectively. On the other hand, upon treating with 5 and 10 mM GSH (comparable to cytosolic milieu), up to ∼73% and ∼94% 5-Fu release was achieved over 32 h incubation duration. However, in the case of loaded Dox·HCl, prohibited release (<8% cumulative release over 42 h) was observed for both original polymersomes and GSH-treated ones (Fig. 10f). This feature is quite similar to that of PEO45-b-P(NCMA0.55-co-DPA0.45)29 vesicles upon light-triggered polymersome-to-PICsome transition. Thus, irrespective of the types of external triggers (UV light or GSH), the permeability switching and bilayer permselectivity is mainly determined by the sequence of bilayer-formation block (e.g., random vs. block).DiscussionPolymersomes and PICsomes, with distinct bilayer compositions and polarity, represent two main categories of vesicular nanostructures self-assembled from BCPs. However, polymersomes possess retarded bilayer permeability, whereas PICsomes with semipermeable membrane are unstable towards physiologically relevant salt and temperature. We proposed a general strategy to solve this challenge by external stimuli-triggered polymersome-to-PICsome transition, starting from a single component functionalized BCP precursor. In situ-fabricated PICsomes via UV light trigger are ultrastable towards extreme pH (2–12), elevated temperature (up to 70 °C), and high ionic strength (3 M NaCl), due to cooperative ion-pair interactions at high local concentration and synergistic carbamate-relevant HB interactions within vesicle bilayers. The polymersome-to-PICsome transition is accompanied with prominent hydrophobic-to-hydrophilic permeability switching. As compared to that of conventional PICsomes fabricated from two oppositely charged block polyelectrolytes, the excellent permselectivity demonstrated by in situ-fabricated PICsome bilayers represents another important feature, which could be further regulated by comonomer sequences of the bilayer-forming block (random vs. block) and the magnitude of external stimuli. The proposed strategy of triggered polymersome-to-PICsome transition combines advantages of two main types of BCP vesicles and solves the stability issue of PICsomes without recourse to chemical crosslinking. In addition, reductive milieu could also be utilized to trigger polymersome-to-PICsome transition, auguring well for the generality of the proposed in situ transformation strategy.MethodsSample synthesisSynthetic routes employed for the preparation of 2-nitrobenzyl ester-photocaged carboxyl monomer (NCMA), DCMA, and two types of tertiary amine-containing monomer with carbamate linkages, DPA and DEA, are shown in Supplementary Fig. 1. Schematics of the synthesis of UV light-responsive PEO45-b-P(NCMAx-co-DPA1−x)n diblock copolymers and PEO45-b-PNCMAm-b-PDPAn triblock copolymers, and disulfide-caged PEO45-b-P(DCMAx-co-DPA1−x)n diblock copolymer are shown in Supplementary Figs. 2 and 3, respectively. Schematics of the synthesis of dye-functionalized amphiphilic diblock copolymers, PEO45-b-P(NCMAx-co-DPA1−x)n-Nile red and PEO45-b-P(NCMAx-co-DPA1−x)n-naphthalimide, and two types of control diblock copolymers without tertiary amine moieties, PEO45-b-PNCMA30 and PEO45-b-PPA26, are shown in Supplementary Figs. 4 and 5, respectively. Detailed procedures of sample synthesis and structural characterization data are described in the Supplementary Information.Self-assembly of amphiphilic BCPsIn a typical self-assembling procedure, 2 mg amphiphilic block copolymer was dissolved in 1 mL acetone, stirred and thermostated at 25 °C in a water bath. Next, 9 mL deionized water was slowly added over 9 h via a springe pump. The organic solvent was then removed by dialysis (MWCO 3.5 kDa) against deionized water for 8 h and the external dialysate was replaced with fresh deionized water at an ~2 h interval.Fabrication of dye-labeled vesiclesDye-labeled vesicles were fabricated via co-assembly of label-free amphiphilic diblock copolymers with dye-functionalized amphiphilic BCPs including P(NCMA0.55-co-DPA0.45)29-naphthalimide and PEO45-b-P(NCMA0.55-co-DPA0.45)29-Nile red. In a typical procedure employed for the fabrication of vesicles labeled with pH-sensitive naphthalimide-based probes, PEO45-b-P(NCMA0.55-co-DPA0.45)29 (1.6 mg) and P(NCMA0.55-co-DPA0.45)29-naphthalimide (0.4 mg) were dissolved in 1 mL acetone, stirred and thermostated at 25 °C in a water bath. Then, 9 mL water was slowly added within 9 h via a springe pump. The organic solvent was then removed by dialysis (MWCO 3.5 kDa) against deionized water for 8 h and the external dialysate was replaced with fresh deionized water at an ~2 h interval.Fabrication of drug/model drug-encapsulated vesiclesFor the physical encapsulation of hydrophobic Nile red into hydrophobic bilayers of self-assembled polymersomes, the amphiphilic block copolymer and Nile red were dissolved in acetone at final concentrations of 2.0 and 0.01 g/L, respectively. The solution mixture was then subjected to similar self-assembling procedures described above.For the encapsulation of hydrophilic drug and model drug molecules (e.g., anticancer drug 2′-deoxy-5-fluorouridin, 5-Fu) into the hydrophilic lumen of self-assembled polymersomes, 2.0 mg amphiphilic diblock copolymer was dissolved in 1 mL acetone, stirred and maintained at 25 °C in a water bath. Then, 5-Fu (16 mg, 125 μmol) dissolved in 9 mL water was slowly added within 9 h. The organic solvent was then removed by dialysis (MWCO 3.5 kDa) against deionized water for 8 h and the external dialysate was replaced with fresh deionized water at ~2 h interval. According to similar procedures, other water-soluble anticancer drugs and model drugs including gemcitabine hydrochloride, doxorubicin hydrochloride (Dox∙HCl), coumarin 343, and calcein were also encapsulated into the aqueous lumen of polymersomes, and vesicular self-assembly was actuated by slowing adding aqueous solution of corresponding drugs at the same molar concentration. The photostability assay results of drugs and model drugs against UV light irradiation are shown in Supplementary Fig. 41. Drug loading efficiency, loading content, and loaded drug concentration were quantified by fluorescence (Dox∙HCl, coumarin 343, and calcein) and UV–Vis absorbance (5-Fu and gemcitabine hydrochloride), respectively. Relevant results are summarized in Supplementary Table 2.Supplementary information
Supplementary Information
Peer Review File | nature communications | [
"Article"
] | [
"Drug delivery",
"Polymers",
"Self-assembly"
] | structures functions biological systems cell membranes viral capsids inspired artificial nanostructures liposomes polymeric vesicles aqueous interiors hydrophobic bilayers7–11 polymersomes amphiphilic block copolymers improved microstructural stability utilized nanoreactors15–18 artificial organelles3 permeation active agents aqueous lumens prohibited by hydrophobicity thick bilayer membranes1 approaches membrane integration with proteins copolymerization with moieties bilayer post-modification proposed3 involve tedious procedures external additives unsatisfactory permselectivity loss structural integrity proposed “traceless crosslinking” achieved vesicle crosslinking membrane irreversible bilayer chemical crosslinking restricts in vivo applications biodegradation clearance issues challenge achieve molecular size-selective release permselectivity regulation polyion complex vesicles (PICsomes) from oppositely charged polyelectrolytes possess semipermeable bilayers improved permeability towards hydrophilic unstable under temperature ionic strength exchange PICs need stabilization via chemical crosslinking30stability issue solved by introducing guanidinium hydrogen bonding alkyl spacer38 into cationic block strong block polyelectrolyte39 polyions with dendritic topology41 PICsomes encounter limitations fabrication process not compatible with hydrophobic drugs imaging agents29 formation involves two oppositely charged components block formation from single component permselectivity modulation of bilayers not achieved macromolecular agents 10 could permeate though bilayers30 loading capability encapsulation stability of PICsomes hydrophilic molecule drugs functional agents need polyion complex micelles45–47 PICsomes29–31 sensitive to high ionic strength pH temperature40 stability upon local concentrations sequence arrangement of charged ion-pairs43 Sun et al fabricated tough viscoelastic polyamphoyte hydrogels via copolymerization of oppositely charged comonomers at >1.5 M total concentration in aqueous media 0.5 M NaCl hydrogels stable towards high salt concentrations elevated temperaturesmodel peptides histidine aspartate alanine-rich hydrophobic milieu ion-pair interactions HB screened high ionic strength ~1.0 M NaCl helix ion-pair interactions high density provide Coulomb attraction ion-pair interactions stabilize nanostructures PICsomes against conditions high density assisted by noncovalent interactions HB high concentration functional moieties~1–2 mol/L hydrophobic bilayers strategy generate ion-pair interactions trigger transformation polymersomes to PICsomes.-triggered polymersome-to-PICsome transition permeability regulation polymersomes amphiphilic diblock triblock copolymers tertiary amine carboxyl moieties hydrophobic stimuli UV light milieu trigger decaging carboxyl functionalities transfer protons tertiary amine moieties render ion-pair interactions PICsomes stable towards extreme pH high ionic strength elevated temperature due ion-pair interactions high concentration hydrogen bonding interactions carbamate polymersome-to-PICsome transition transformation hydrophobic bilayers into semipermeable membranes vesicle bilayer permeabilitysequence structure bilayer block A--C-B-C permselectivity semipermeable PICsome bilayers.Polymersomes self-assembled from amphiphilic BCPs tertiary amine caged carboxyl comonomers stimuli-responsive block initially hydrophobic (Fig. 2) carbamate linkage tertiary amine carboxyl intra-chain HB interactions vesicle noncovalent interactions directional orientation high local concentration cleavage proton transfer carboxyl tertiary amine generates ionized carboxyl protonated amine moieties Ion interactions strengthened carbamate-relevant HB interactions hydrophobic bilayer-forming scaffold high stability PICsomes extreme pH high ionic strength elevated temperature 2Chemical structures four amphiphilic block copolymers (BCPs-triggered polymersome-to-PICsome transition vesicles single component block polyampholyte precursor differs conventional PICsomes charged polyion components UV light reductive milieu trigger polymersome-to-PICsome transition vesicle bilayer permeability proposed strategypermselectivity PICsomes tuned sequence arrangement polyions molecular size-selective release achieved (Fig. 1) feature not realized conventional PICsomes self-assembly amphiphilic two carbamate-containing monomers caged carboxyl tertiary amine functionalities designed 2-Nitrobenzyl ester photo-caged disulfide-caged carboxyl (DCMA generate carboxyl moieties UV irradiation reductive milieu second pH-responsive DPA DEA monomers carbamate linkage tertiary amine moieties synthesized monomers synthetic intermediates by 1H 13C NMR analysis 1 reversible-fragmentation) polymerizations PEG45-based macroRAFT agent amphiphilic BCPs varying comonomer sequences compositions PEO45-b-P(NCMA0.55-co-DPA0.45)29-P-co-DEA0.51)32-P(DCMA0.45-co-DPA0.55)33 diblock copolymers PEO45-b-PNCMA17-b-PDPA21 triblock copolymer BCPs characterized by 1H NMR GPC analyses11–14) structural parameters summarized Supplementary Table hydrophobic diblock triblock copolymers equal ratio caged carboxyl tertiary amine comonomers decaging carboxyl tertiary amine equivalent molar ion-pair interactions vesicle bilayers (Fig. dye-labeled amphiphilic BCPs PEO45-b-Nile red-naphthalimide synthesized vesicles polarity-sensitive pH-sensitive fluorescent probes Fig 4) control amphiphilic BCPs PEO45-b-PNCMA30 PEO45-b-PPA26 synthesized Figs. 5 as-synthesized diblock triblock copolymers tertiary amine comonomer units hydrophobic microenvironment carboxyl suppresses pKa tertiary amines amine pKa ~5.7 PEO45-b-P ~6.4 PDPA homopolymer 16 self-assembly triggered slow addition water polymer solution acetone microscopy) typical vesicular nanostructures four types BCPs. 3aDynamic laser analysis polymersomes-average hydrodynamic diameters 530–640 nm low polydispersities with TEM results 3TEM images PEO45-b-P-DPA0.45)29 polymersomes PICsomes polymersomes after 10 min UV irradiation neutral aqueous media pH 4.0 12.0 destruction acidic media PICsomes aqueous media at pH 2.0 12.0-triggered polymersome-to-PICsome investigated UV light-triggered evolution of polymersomes aqueous media photo-labile NCMA DPA comonomers random hydrophobic block close contact polymersome bilayers UV irradiation photocleavage 2-nitrobenzyl generation carboxyl moieties monitored-dependent UV/Vis absorption spectra photocleavage leveled off at ~10 min under 365 nm LED irradiation UV-triggered carboxyl decaging examined with 1H NMR photocleavage >98% upon UV irradiation 10 minattempted UV-triggered chemical changes laser/ionization spectrometry no reliable signals recorded non UV-irradiated polymersome dispersions upon lyophilization vesicular dispersion bluish changed to grayish yellow after UV irradiation 1H NMR spectra before after UV irradiation showed well-solvated PEO coronas colloidal aggregates after UV carboxyl decaging TEM observations after UV irradiation revealed intact vesicular nanostructures-dependent DLS measurements conducted light intensities diameter UV irradiation intensities PEO45-b-P(NCMA0.55-co-DPA0.45)29 dispersion initial ∼15% decrease within ∼5 min UV irradiation reached plateau slight increase 〈Dh〉 from 550 to 570 nm upon 10 min UV irradiation Dh distribution unchanged UV irradiation N,N-diisopropylamine with N,N-diethylamine resultant PEO45-b-P(NCMA0.49-co-DEA0.51)32 BCP higher pKa for tertiary aminesDLS TEM confirmed UV polymersome-to-PICsome transition proposed strategy Fig. photoreaction amine examined photostability N-diisopropylethylamine triethylamine against UV irradiation no chemical structural changes Figs. 23 24).Fig. 4DLS polymersome-to-PICsome transition UV irradiation Irradiation duration-dependent evolution light intensities 〈Dh〉 PEO45-b-P-DPA0.45)29 vesicles neutral aqueous media Intensity-average hydrodynamic diameter distributions before after UV irradiation vesicles microstructural integrity after UV irradiation-triggered cleavage 2-nitrobenzyl ester generated hydrophilic carboxyl moieties proton transfer to tertiary amine tertiary amines unprotonated (pKa ~ 5.7 carboxyl control BCP PEO45-b-PPA26 pKa ~6.5 Fig presence neighboring amine carboxyl functionalities hydrophilic milieu increase pKa values amine carboxyl proton transfer ion-pair formation protonated amines increase acidity carboxylic acids conjugate base)propose UV decaging vesicular nanostructures stabilized by ion-pair interactions high enhanced side chain HB interactions (Fig. 1 Fig before UV irradiation polymersomes stabilized hydrophobic interactions UV-irradiation generates ion-pair interactions vesicle bilayers polymersome-to-PICsome transition PICsomes potentiometric titration experiments ion-pairs form pH below 4.6 8.8 ionized carboxylates protonated amines exist Fig. ion-pair formation fabricated polymersomes conjugated with pH-sensitive fluorescent probe co-assembly PEO45-b-P-naphthalimide BCPs (8:2 ratio (Fig. 5a). tertiary amine naphthalimide probe pKa comparable PDPA homopolymer 16 28, naphthalimide-conjugated polymersome dispersion aqueous decreasing solution pH 7 to 4 emission enhancement loss PET quenching protonated tertiary amines pH 7–9 fluorescence emission slight decrease amine pKa (~5.7) PEO45-b-PPICsomes UV irradiation pH-dependent emission intensities shifted higher pH amine protonation ~51.2% at pH 7 Fig proton transfer carboxyl to amine formation ion-pairs fabrication naphthalimide-labeled polymersomes PEO45-b-naphthalimide (8:2 naphthalimide emission quenched by tertiary amine moieties decaging protonation naphthalimide emission enhanced Evolution fluorescence emission spectra polymersome-to-PICsome transition UV irradiation Irradiation time-dependent evolution protonation tertiary amine moieties vesicle data polymer concentration 0.1 g/L [NCMA] ~ 0.12 mM [DPA] ~ 0.10 mM Britton–Robinson buffer (pH 12 25 UV polymersome-to-PICsome transition fluorescence emission intensities increased first 3–4 min (Fig. 5) decaging kinetics Figs 18 19. UV irradiation naphthalimide dye negligible photobleaching Fig results confirmed proton transfer carboxyl amine formation ion-pairsresults Figs. 28 29 PICsomes before after UV irradiation amine protonation increased ~7.8% to ~51.2% during polymersome-to-PICsome transition ~50% carboxyl amine moieties formed ion-pairs PICsomes control BCP PEO45-b-PNCMA30 polymersome dispersion pH decrease ~7.4 to ~6.4 during UV irradiation agrees with pKa ~6.5 PEO45-b-PPA26 structure same as decaged PEO45-b-PNCMA30 PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersome dispersion UV irradiation slight pH decrease ~7.4 to ~7.2 internal proton transfer ion-pair formation light-triggered polymersome-to-PICsome transition decaged PEO45-b-P UV irradiation self-assembly slow water addition method PIC micelles non-uniform size distribution contrast to UV-triggered PICsome formation pre-organization polymersome bilayers before PICsome formation.PICsome nanostructures stabilized by microstructural stability polymersomes PICsomes applications complex biologicalexamined stability PICsomes pH temperature ionic strength Fig. 6a original polymersomes irradiation integrity pH 5–12 below pH 5 vesicle disintegration protonation tertiary amine comonomers Fig. 26, polymersomes PICsomes.pH-dependent light intensities Dh distributions dispersions PEO45-b-P-DPA0.45)29 polymersomes 2–12 Irradiation duration-dependent zeta potentials-PICsome transition neutral media Variation zeta potentials 2–12 data polymer concentration 0.1 g/L [NCMA] 0.12 mM [DPA] ~ 0.10 mM Britton–Robinson buffer (pH 2–12 12 mM 25 error bars standard deviation PICsomes UV irradiation structural integrity maintained pH 2–12 confirmed pH-dependent Dh distributions (Fig. TEM observations original polymersomes stable neutral alkali media disassembled under acidic media (pH < 4.0) after UV irradiation PICsomes stability pH 2–12stability PICsomes corroborated by zeta potential measurements-to-PICsome transition varying pH conditions neutral aqueous media zeta potential values constant +6 mV during UV irradiation (Fig pH 7–12 zeta potentials polymersomes ±5 mV increased to ~+8 mV at pH 4 no data pH to ~2–3 disassembly pH-dependent variation zeta potentials 4–10 similar polymersomes in acidic alkaline media (pH 2 zeta potential values changes (+32 mV at pH 2 −15 mV at pH structural integrity results with TEM DLS results (Figs. 3 stability PICsomes under extreme pH conditions unexpected electrostatic interactions ion weakened due to carboxyl protonation amine deprotonation in acidic alkaline media pH stability to two HB interactions between carboxyl moieties π–π interactions explain stability at pH 2 stability alkaline media HB π–π interactions benzyl carbamate responsible hydrophobic nature non-protonated amine comonomer units (Fig. 7a).Fignanostructures stabilized noncovalent interactions Microstructural stability multivalent ion-pair hydrogen bonding π–π interactions high concentration vesicle bilayers Intensity-average Dh distributions aqueous dispersion PICsomes PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersomes UV irradiation varying temperatures ionic strengths combinations data polymer concentration 0.1 g/L [NCMA] ~ 0.12 mM [DPA] ~ 0.10 mM neutral aqueous media investigated stability high ionic strengths elevated temperatures NaCl salt added final concentration 0–3.0 M PICsomes stable up to NaCl concentration 3.0 M Dh distributions same NaCl concentrations PICsomes neutral aqueous media stable elevated temperatures (25–70 Dh distributions negligible changes PICsomes with salt elevated temperatures 2.0 M NaCl stable up to ~50 °C heating Dh distributions lower size ranges At 70 °C PICsomes disassemble into unimers no DLS signals ionic interactions HB interactions suppressed by high salt concentrations elevated temperaturesdisintegration of PICsomes at 70 °C 2.0 M NaCl stabilized by cooperative noncovalent interactions chemical crosslinking (Fig. 1) with 1H NMR data DMSO-d6 lyophilized polymersome dispersion after UV irradiation original polymersomes UV stable towards high temperature (70 high ionic strength (2.0 cooperative carbamate-relevant HB interactions hydrophobic nature vesicle bilayers incompatible with NaCl salt).Effects block sequences on permselectivity robust PICsome nanostructures fabricated from PEO45-b-P(NCMA0.55-co-DPA0.45)29 polymersomes aqueous media via UV-triggered carboxyl decaging bilayer forming block random copolymer of DPA NCMA UV decaging triblock polyampholytes two oppositely charged comparable to conventional PEO45-b-PNCMA17-b-PDPA21 compositions comparable synthesized (Fig. 2) PEO45-b-PNCMA17-b-PDPA21 triblock copolymer self-assembled into polymersomes uniform size distributionUV irradiation triblock generated carboxyl ion-pairs-to-PICsome transition Fig. 8b vesicular nanostructure maintained dimension PICsomes larger than original DLS measurements 〈Dh〉 vesicles rapid min UV irradiation 540 to 770 nm during-to-PICsomes transition (Fig. PEO45-b-P(NCMA0.55-co-DPA0.45)29 vesicles. 4a 550 570 vesicle size distribution uniform (μ2/Γ2 ∼ 0.10 Fig. before PICsome formation PEO45-b-PNCMA17-b-PDPA21 decaging UV irradiation self in aqueous media formation micellar nanoparticles instead polymeric vesicles Fig. 35).Fig. 8Light-triggered microstructural evolution triblock copolymer vesicles images PEO45-b-PNCMA17-b-PDPA21 polymersomes PICsomes b Irradiation duration-dependent evolution light intensities 〈Dh〉-to-PICsome transition PEO45-b-PNCMA17-PDPA21 vesicles neutral aqueous media Evolution intensity-average Dh distributions UV irradiation scale bars 500 nm.data obtained polymer concentration 0.1 g/L neutral aqueous media different swelling UV polymersome-to-PICsome transition PEO45-b-P(NCMA0.55-co-DPA0.45)29 PEO45-b-PNCMA17-b-PDPA21 triblock copolymers comonomer sequences formation ion-pairs. 1 2) diblock vesicles generated carboxyl functionalities amine moieties local concentration ion-pair high triblock vesicles carboxyl apart amine moieties frustrated formation bilayer swelling partial ionization protonation residues later stages UV irradiation formation between charged blocks bilayer swelling vesicle sizes constant after ~4 min UV irradiation polymersome-to-PICsome transition transformation bilayer polarity hydrophobic to hydrophilic loss hydrophobic 2-nitrobenzyl ester residues generation ion-pairs examined polymersomes conjugated polarity-sensitive fluorescent probe Nile redpolymersomes from PEO45-b-P-Nile red (8:2 aqueous media initial Nile red emission ~610 nm strong hydrophobic bilayers UV irradiation decrease emission intensities slight red shift confirms generation hydrophilic bilayers UV irradiation (Fig. 1) diblock triblock copolymer vesicle swelling polymersome-to-PICsome transition varying mesh sizes bilayer permeability PICsomes loaded water-soluble anticancer drugs aqueous polymersomes examined release profiles UV transition gemcitabine hydrochloride (299.7 Da) PEO45-b-P without UV irradiation ∼25% cumulative release over 32 h UV irradiation 2 5 min sustained release cumulative release ~85% ~97% after 32 h doxorubicin hydrochloride) larger molar mass (580.0 <5% released over 32 h polymersomes UV irradiationgemcitabine hydrochloride (299.7 Dox·HCl positive charges discrepancy release profiles different molecular sizes permselectivity PICsomes contrast to conventional PICsomes dextran molar mass ~10 kDa. 9Light-regulated polymersome-to-PICsome transition for molecular size-selective drug release Schematics drug-loaded vesicles PEO45-b-P-PNCMA17-PDPA21 triblock copolymer triggered release PICsomes Release profiles of gemcitabine hydrochloride Dox·HCl from vesicles before after UV irradiation data obtained at polymer concentration 0.1 g/L in PB buffer (pH 7.4, 10 mM, 37 checked permeabilities of neutral negatively charged drugs through PICsome bilayers neutral 2′-deoxy-5-fluorouridine 246.2) negatively charged coumarin-343.3 Da) released from vesicles via polymersome-to-PICsome transition release negative charged calcein higher molar mass (622.6 Da) from PICsomes prohibitedresults polymersome-to-PICsome transition permeability switching vesicle bilayers size-selective release of encapsulated drugs paired ionic interactions PICsome bilayers mesh size control electrostatic ion side chain HB interactions polymersomes PEO45-b-PNCMA17-b-PDPA21 triblock copolymer hydrophobic-to-hydrophilic transition occurs during light confirmed by fluorescence measurements Nile red probe PEO45-b-P(NCMA0.55-co-DPA0.45)29 vesicles PICsomes-demand release for gemcitabine hydrochloride Dox·HCl calcein (Fig. 9d–f PICsomes almost no release of Dox·HCl calceinrelease rates gemcitabine hydrochloride Dox·HCl calcein from triblock PICsomes (5 min UV irradiation decreased increasing molar mass charges (∼100% release gemcitabine hydrochloride over 16 h ∼84% Dox·HCl 32 ∼49% calcein 32 h enhanced permeability PICsomes triblock copolymer swelling 540 to 770 nm ~43% increase 〈Dh〉 during light-triggered polymersome-to-PICsome transition diblock vesicles 〈Dh〉 increase ∼3.6% 550 to 570 block sequence-dependent permselectivity of PICsomes drug nanocarriers nanoreactors UV irradiation duration bilayer permeability selectivity accelerated release functional agents permselectivity modulation by block copolymer sequences magnitude external stimuli achieved.Reduction-triggered polymersome-to-PICsome ultrastable vesicles excellent bilayer permselectivity towards drug molecules varying molar mass generalized design transition reductive milieu trigger redox gradient across cell membranes universal forPEO45-b-P-DPA0.55)33 copolymer disulfide-caged carboxyl comonomers self aqueous polymersomes ~600 nm uniform size distribution.-triggered polymersome-to-PICsome transition permselectivity regulation TEM images PEO45 vesicles treating 10 mM GSH aqueous Incubation time evolution scattered light intensities intensity-average Dh distributions 10 mM GSH Release profiles 5-Fu Dox·HCl varying concentrations scale bars 1 μm data polymer concentration 0.1 g/L PB buffer (pH 7.4 10 mM 37 °C).Reduction-triggered evolution vesicular microstructures explored TEM DLS measurements (Fig. GSH addition disulfide cleavage 1,6-rearrangement benzyl carboxyl functionalities ion-pair formation (Fig. 1) reduction-triggered decaging process 1H NMR extent >99% treating 10 mM GSH 24 h light intensities 〈Dh〉 distributions unchanged GSH vesicular integrity confirmed TEM observationsreductive light trigger actuate polymersome-to-PICsome transition stable PICsomes hydrophilic bilayers examined permselectivity of situ-fabricated PICsomes PEO45-b-P(DCMA0.45-co-DPA0.55) reductive trigger GSH). encapsulated model anticancer drug (5-Fu) retarded release from original polymersomes treated with ∼2 μM GSH cumulative release ∼14% ∼19% over 32 h 5 10 mM GSH ∼73% and ∼94% 5-Fu release over 32 h loaded Dox·HCl prohibited release (<8% over 42 h observed for original GSH-treated ones feature similar to PEO45-b-P vesicles light-triggered polymersome-to-PICsome transition permeability switching bilayer permselectivity determined by sequence bilayer-formation block PICsomes distinct bilayer compositions polarity represent categories vesicular nanostructures polymersomes retarded bilayer permeability PICsomes with semipermeable membrane unstable towards salt temperatureproposed strategy external stimuli-triggered polymersome-to-PICsome transition from single component BCP precursor situ-fabricated PICsomes via UV light ultrastable towards extreme pH elevated temperature high ionic strength (3 due cooperative ion-pair interactions carbamate-relevant HB interactions vesicle bilayers transition hydrophobic-to-hydrophilic permeability switching conventional PICsomes excellent permselectivity bilayers regulated by comonomer sequences bilayer-forming block magnitude external stimuli proposed strategy polymersome-to-PICsome transition combines advantages BCP vesicles solves stability issue without chemical crosslinking reductive milieu trigger-PICsome transition transformation strategy.MethodsSample synthesisSynthetic routes preparation 2-nitrobenzyl ester-photocaged carboxyl monomer DCMA tertiary amine-containing monomer with carbamate linkages DPA DEA shown in Supplementary Fig.synthesis UV-responsive PEO45-b-P(NCMAx-co-DPA1−x-PDPAn triblock copolymers disulfide-caged PEO45-b-P-co-DPA1−x diblock copolymer Figs. 2 3 dye-functionalized amphiphilic diblock copolymers PEO45-b-P(NCMAx-co-DPA1−x-Nile red-naphthalimide control diblock copolymers amine PEO45-b-PNCMA30 PEO45-b-PPA26 Figs. 4 5 procedures synthesis structural characterization Supplementary Information-assembly amphiphilic 2 mg amphiphilic copolymer dissolved 1 mL acetone thermostated 25 °C 9 mL deionized water added 9 h solvent removed dialysis 8 h replaced fresh deionized water h dye-labeled co-assembly label-free amphiphilic diblock copolymers dye-functionalized amphiphilic BCPs P(NCMA0.55-co-DPA0.45)29-naphthalimide PEO45-b-P-Nile redprocedure vesicles pH-sensitive naphthalimide PEO45-b-P-DPA0.45)29 (1.6 mg-naphthalimide (0.4 mg) dissolved 1 mL acetone thermostated 25 °C 9 mL water added 9 h pump solvent removed dialysis deionized water 8 h replaced fresh deionized water ~2 h drug drug-encapsulated hydrophobic Nile red amphiphilic block copolymer Nile red dissolved acetone concentrations 2.0 0.01 g/L mixture self-assembling procedures encapsulation hydrophilic drug molecules anticancer drug 2′-deoxy-5-fluorouridin 2.0 mg amphiphilic diblock copolymer dissolved 1 mL acetone maintained 25 °C 5-Fu (16 mg, 125 μmol) dissolved 9 mL water added 9 h organic solvent removed dialysis 3.5 deionized water 8 h dialysate replaced fresh deionized water ~2 hprocedures water-soluble anticancer drugs gemcitabine doxorubicin coumarin 343 calcein encapsulated aqueous lumen polymersomes vesicular self-assembly actuated aqueous solution concentration photostability assay results UV light irradiation Supplementary Fig. 41 Drug loading efficiency content drug concentration quantified fluorescence coumarin calcein UV–Vis absorbance gemcitabine results summarized Supplementary Table 2.Supplementary information Review | 50.3 | 0.78357 |
10.1038/s41467-020-19107-y | PMC7568531 | The abnormally low concentration of xenon compared to other noble gases in Earth’s atmosphere remains debated, as the identification of mantle minerals that can capture and stabilize xenon is challenging. Here, the authors propose that xenon iron oxides could be potential Xe hosts in Earth’s lower mantle. | An enduring geological mystery concerns the missing xenon problem, referring to the abnormally low concentration of xenon compared to other noble gases in Earth’s atmosphere. Identifying mantle minerals that can capture and stabilize xenon has been a great challenge in materials physics and xenon chemistry. Here, using an advanced crystal structure search algorithm in conjunction with first-principles calculations we find reactions of xenon with recently discovered iron peroxide FeO2, forming robust xenon-iron oxides Xe2FeO2 and XeFe3O6 with significant Xe-O bonding in a wide range of pressure-temperature conditions corresponding to vast regions in Earth’s lower mantle. Calculated mass density and sound velocities validate Xe-Fe oxides as viable lower-mantle constituents. Meanwhile, Fe oxides do not react with Kr, Ar and Ne. It means that if Xe exists in the lower mantle at the same pressures as FeO2, xenon-iron oxides are predicted as potential Xe hosts in Earth’s lower mantle and could provide the repository for the atmosphere’s missing Xe. These findings establish robust materials basis, formation mechanism, and geological viability of these Xe-Fe oxides, which advance fundamental knowledge for understanding xenon chemistry and physics mechanisms for the possible deep-Earth Xe reservoir. | IntroductionThe chemical reaction of inert xenon (Xe), a quintessential full-shell element, was earliest predicted by Pauling in 1933 and the first xenon compound was experimentally produced in 19621. Then, more xenon compounds were experimentally synthesized at ambient pressure, containing some most electronegative atoms like fluorine2–5 or oxygen6–9. Subsequently, scientists found that pressure can effectively improve the chemical reactivity of Xe10–17. At moderate pressures, solid xenon forms weakly bonded compounds with other species, e.g., with H2O10 and O211,12 at 1 and 3 GPa, respectively. Strikingly, several novel Xe compounds with unusual stoichiometries are found to be thermodynamically stable at high pressures, e.g., Xe oxides13,14, Xe nitrides15, xenon-hydrogen16, and Xe–Mg compounds17.At ultra-high-pressure conditions, the high volatility, relative chemical inertia, and abundant isotopes of xenon make it a valuable tracer in the study of evolutionary dynamics and history of Earth. However, 99% of Earth’s primordial Xe is mysteriously missing as characterized by its very low abundance compared to other noble gases in Earth’s atmosphere18, which is known as the missing Xe problem19. Early hypotheses proposed that Xe might have escaped from the atmosphere after ionization20–22, or that it might be stored in the interior of Earth23–29. Attempts to incorporate Xe into ices, clathrates and sediments in the Earth’s crust were not successful27–29. Laboratory experiments have succeeded in trapping Xe in quartz30,31 and observing predicted stable xenon oxides13,14,32; but these results cannot explain the missing Xe mystery, because xenon oxides are unstable in equilibrium with metallic iron in lower mantle while xenon silicates decompose spontaneously at mantle pressures13. Reactions of Xe with iron and nickel in Earth’s core were predicted as a viable scenario33 and the predicted compounds were synthesized under core pressure and temperature conditions32,34. However, it remains highly intriguing and challenging to explore possible capture and stabilization of Xe by suitable minerals in Earth’s mantle, which is of special significance because it was estimated that the loss of atmospheric Xe occurred about 100 million years from accretion, corresponding to the time of mantle differentiation event30.Extensive past searches were unable to find chemical reactions of Xe with known mantle minerals. Recently discovered FeO2 synthesized at lower mantle conditions35 and stablized above 74 GPa in theoretical calculation36, offer an intriguing new possibility. This newly identified iron peroxide is able to react with helium to form a rare helium-bearing compound that explains deep-Earth primordial helium deposits revealed by geochemical evidences37. This finding raises exciting prospects that FeO2 may be able to react with Xe (actually P–T stability range of FeO2 has not been completely established in experiments) at mantle conditions, thereby forming compounds capable of trapping Xe in Earth’s interior. In this work, we have explored possible reactions of Xe with FeO2 in contrast with known mantle constituents FeO, SiO2, MgO, CaO, and Al2O3. We find that FeO2 has unique ability to react with Xe and form robust Xe-Fe oxides Xe2FeO2 and XeFe3O6 with surprisingly strong Xe–O bonding, while other mantle oxides do not react with Xe. We have further examined mass density and sound velocities of these Xe-Fe oxides, and the results support their viability in vast lower mantle region. These findings establish robust materials basis, formation mechanism, and geological viability of these Xe-Fe oxides, which advance fundamental knowledge for understanding xenon chemistry and physics mechanisms for the possible deep-Earth Xe reservoir.ResultsCrystal structuresWe take the crystal phases identified by the structure search process at various FeO2: Xe ratios and compute their enthalpy to determine the most viable structure at each composition, and based on the obtained results we construct the convex hull, as shown in Fig. 1a, which indicates stable products from reactions of FeO2 and Xe. This exercise has led to the discovery of two Xe-Fe oxides, Xe2FeO2 and XeFe3O6, that are stable against decomposition at 150 GPa and 200 GPa. The pressure–volume terms, associated with packing efficiency, make the major contribution to guarantee the thermodynamical stability of Xe2FeO2 and XeFe3O6 with formation enthalpies lying on the convex hull (Supplementary Fig. 1). These two Xe-Fe oxides are both still stable relative to all possible binary phases or pure simple substances of Xe-Fe-O2, which can be seen in Supplementary Information (Supplementary Figs. 2 and 3). The details Phonon dispersions calculated at 150 GPa (Fig. 1b, c) show that these compounds are dynamically stable and, as will be shown below, thermal effects further stabilize both oxides over a wider range of pressure at elevated temperatures at deep-Earth conditions. Here we first present a full analysis of structural and bonding characters of these two oxides at 150 GPa as a representative case study. The compound Xe2FeO2 is crystalized in a monoclinic structure with P21/c symmetry (Fig. 1d); its structural motif consists of stacked layers of corner-sharing octahedron with each Fe atom surrounded by six O atoms and the Fe atom is centered in a slightly distorted octahedron containing Fe–O bond lengths in a narrow range of 1.79–1.82 Å at 150 GPa. Each Xe atom in this structure has a coordination number of 3, bonding at the corners of FeO6 octahedra with the Xe–O bond lengths in the range of 2.40–2.42 Å at 150 GPa, which are similar to those found in Xe2O3 (~2.50 Å) and Xe2O5 (~2.37 Å) at the same pressure38. Meanwhile, XeFe3O6 is stabilized in a triclinic structure with P-1 symmetry, containing two formula units per cell (Fig. 1e); its corner-sharing FeO6 octahedra host Fe–O bonds with lengths of 1.73–1.81 Å at 150 GPa, forming a tubular structure, and each Xe atom has a coordination number of 6, located in the Fe-O tube with the nearest Xe-O distance of 2.08 Å, resulting in the vibrational mode of the lowest-frequency branch at F as shown in Fig. 1c. Further vibrational analyses are shown in Supplementary Fig. 4 and structural details of both Xe-Fe oxides at 150 GPa are listed in Supplementary Information (Supplementary Table 1).Fig. 1Energetic stability and structures of Xe-Fe oxides.a The ground-state convex hull with solid lines for the FeO2-Xe system constructed from calculated formation enthalpy (ΔH) data, identifying two stable Xe-Fe oxides, Xe2FeO2 and XeFe3O6, at 150 GPa and 200 GPa. The solid and open symbols represent the stable structures lying on the convex hull and the unstable structures above the convex hull, respectively. b Phonon dispersions of Xe2FeO2 at 150 GPa. c Phonon dispersions of Xe3FeO6 at 150 GPa. d The structure of Xe2FeO2 with a polyhedral and an enlarged view. e The structure of Xe3FeO6 with a polyhedral and an enlarged view. Xe, Fe, and O atoms are, respectively, shown in purple, gray, and red spheres.DiscussionChemical BondingTo assess bonding characters in the two Xe-Fe oxides, we have calculated their electronic density of states (DOS) at 150 GPa. The results shown in Fig. 2a, c reveal metallic nature of these compounds; crucially, in both cases the DOS in the vicinity of the Fermi level contain significant contributions from the Fe 3d, Xe 5p and O 2p states, indicating considerable Fe–O and Xe–O bonding interactions. We further calculated projected crystal orbital Hamiltonian population (pCOHP) that evaluates weighted population of wavefunctions on two atomic orbitals of a pair of selected atoms39. The results in Fig. 2b, d reveal characteristic Fe–O and Xe–O covalent bonding as indicated by the prominent features of strong low-energy bonding states together with some occupied antibonding states near the Fermi level in each case. It is noted that the occupied bonding states in Xe2FeO2 occur deeper below the Fermi level compared to those in XeFe3O6, indicating higher stability of Xe2FeO2. Moreover, integrated COHP (ICOHP) provides an estimate of the overall bonding strength39. Calculated ICOHP values for the Fe–O and Xe–O bonds at 150 GPa are -1.45 eV/pair and −0.24 eV/pair in XeFe3O6 and −1.01 eV/pair and −0.12 eV/pair in Xe2FeO2, respectively. These results show considerable Xe–O bonding compared to the strong Fe–O bonding, in sharp contrast to recently discovered He-FeO2 compound where He atoms show little direct bonding37 but instead serve as a Coulomb shield in stabilizing the structure40. We also examined charge distribution in Xe2FeO2 and XeFe3O6 by a Bader charge analysis41, and the results reveal a considerable amount of highly unusual charge transfer from Fe and Xe to O atoms. At 150 GPa, the Bader partial charges in Xe2FeO2 are +0.30, +1.40, −1.00 for Xe, Fe, and O, respectively; meanwhile, Bader partial charges in XeFe3O6 are +1.35, +1.35, −0.90 for Xe, Fe, and O, respectively, at the same pressure. As a result, Xe atoms in XeFe3O6 can donate more electrons than in Xe2FeO2 and Xe atoms can display different valence states in FeO2–Xe compounds. These significant charge transfers once again indicate strong bonding formation involving Xe, which is rare among noble gases atoms.Fig. 2Electronic properties of the two Xe-Fe oxides at 150 GPa.a Projected density of states (PDOS) of Fe-d, O-p, and Xe-p orbitals in Xe2FeO2. b Projected crystal orbital Hamiltonian Population (-pCOHP) of the newly predicted Xe2FeO2 compound. The values of -pCOHP >0 signify bonding states and the values of -pCOHP <0 signify antibonding states. c PDOS of Fe-d, O-p, and Xe-p orbitals in XeFe3O6. d -pCOHP of the newly predicted XeFe3O6 compound. The Fermi energy is set to zero of the energy.Thermal effectThermal effects play a crucial role in material stability at pertinent geophysical conditions, where temperatures reach 2000–4500 K. Here, we evaluate Gibbs free energy of Xe2FeO2 and XeFe3O6 by calculating lattice contributions to the entropic term using the quasiharmonic approximation to account for volume dependence of phonon frequencies due to temperature induced lattice expansion. In Fig. 3a we present relevant energetic terms affecting structure stability. It is seen that internal energy U values of the two Xe-Fe oxides are higher than those of their separate constituents, namely Xe and FeO2, but the PV terms are decisively favorable and dominant, producing the lower enthalpy H for the formation of both oxides. The temperature effect (i.e., thermal vibration of atomic positions) are favorable to reduce Gibbs free energy G of the Xe-Fe oxides even more relative to their separate constituents, further stabilizing the resulting crystal structures. Consequently, the threshold pressure above which these oxides are stable reduces considerably at increasing temperatures, thereby significantly expanding their stability field as will be seen in the phase diagram presented below.Fig. 3Calculated energetic terms and the mean square deviations.a Various energetic term (ΔE) for the two Xe-Fe oxides, enthalpy (H), pressure–volume (PV) terms, internal energy (U), and Gibbs free energy (G) at 150 GPa, 2000 K. b Mean square deviations (MSD) of Xe, O, and Fe atomic positions in Xe2FeO2 at 150 GPa, 3000 K. c MSD of Xe, O, and Fe atomic positions in XeFe3O6 at 150 GPa, 3000 K.Phase diagramFor a full assessment of temperature effects, we have performed extensive energetic and ab initio molecular dynamics (AIMD) simulations to evaluate structural stability and construct pressure–temperature (P–T) phase diagram for the Xe-Fe oxide system. We present in Fig. 3b, c the mean square deviations (MSD) of atomic positions in the Xe-Fe oxides at typical high P–T conditions of 150 GPa and 3000 K, and the results show that the Fe, O, and Xe atoms all remain near their lattice sites, indicating stability of the crystal structure. Similar AIMD simulations were performed systematically to probe each phase and determine the boundary where temperature-driven instability sets as indicated by deviating MSD from equilibrium positions; meanwhile Gibbs free energies were computed and compared to determine the boundaries between different solid phases in the P–T space. The resulting phase diagram (Fig. 4) spans a wide P–T range covering the lower mantle and higher P–T regions.Fig. 4The Pressure–temperature (P–T) phase diagram of the Xe-FeO2 system.The dotted and dashed lines indicate phase boundaries and temperature-driven phase instability, which are determined by Gibbs free energy and ab initio molecular dynamics (AIMD) calculations, respectively. The square symbols show phase transition points based on relative Gibbs free energy, and the solid and open triangle symbols represent stable and temperature-driven unstable solid phases, respectively, determined by AIMD simulations. The P–T regions of stable Xe2FeO2 are covered by slash lines and the P–T regions of stable XeFe3O6 are filled by back-slash. The pressure boundary between the lower mantle and the core is shown at the top of the figure. Also shown is the geotherm of the Earth’s interior65.We now analyze the stability fields of the predicted Xe-Fe oxides under the (P, T) conditions conforming to geological constraints dictated by the geotherm that is also shown in Fig. 4. It is seen that Xe2FeO2 is stable in the pressure range 110-120 GPa and temperatures around 2500 K inside the geotherm corresponding to the deep lower mantle region; meanwhile, both Xe2FeO2 and XeFe3O6 are stable between pressures 120–136 GPa and temperatures 2500–3600 K inside the geotherm corresponding to the lowest mantle to core-mantle boundary (CMB); finally, as pressure and temperature rise further, Xe2FeO2 becomes the sole stable phase.The above results suggest stable Xe-Fe oxides under the (P, T) conditions in vast deep-Earth regions. It is, however, necessary to assess the viability of the predicted Xe-Fe oxides in geological environments by examining their key material characteristics. To this end, we have run AIMD simulations to determine crystal structures at selected (P, T) conditions and used an AIMD-based strain-stress method42,43 to calculate the elastic tensors, from which elastic-wave velocities were determined by solving the Christoffel equation det |Tik-δikρV2 | = 0, where δik is the Kronecker delta function, V is one of the seismic velocities, and Tik is the Christoffel stiffness44.Mass density and sound velocitiesWe examine mass density and mean compressional (P-wave) and shear (S-wave) sound velocities, VP and VS, respectively, at two representative (P, T) points: (120 GPa, 2500 K) for lower mantle, and (135 GPa, 3500 K) for CMB, and compare with geological data. We first examine XeFe3O6, whose stability field compared to the geotherm indicates its stability in the lower mantle and CMB regions, but higher temperatures destabilize this compound. The calculated densities of XeFe3O6 are 8.86 and 9.06 g/cm3 at the selected lower mantle and CMB (P, T) points, respectively, which lie within or close to the range of 4.95–9.90 g/cm3 from the core rigidity zone (CRZ) model and the range of 5.57–8.91 g/cm3 from the ultralow velocity zone (ULVZ) model45. The calculated VP (VS) are 8.94(4.03) km/s and 9.20(4.26) km/s, respectively, which lie within the range of 8.20–10.70 km/s (1.00–5.00 km/s) for the ULVZs45. All these results indicate that XeFe3O6 is a viable constituent at the lower mantle and CMB (P, T) conditions.The calculated densities of Xe2FeO2 are 9.78 and 9.87 g/cm3 at the lower mantle and CMB points, respectively, which lie outside the range of 5.57–8.91 g/cm3 from the ULVZ model and at the top of the range of 4.95–9.90 g/cm3 from the CRZ model45. The calculated VP (VS) are 7.83(4.21) km/s and 7.94(4.02) km/s at the lower mantle and CMB points, respectively. While these VS values lie within the range of 1.00–5.00 km/s from the CRZ model or the range 2.91–6.17 km/s from the ULVZ model45, the VP values are out of the range of 10.97–13.03 km/s from the ULVZ model or the range of 8.20–10.70 km/s from the CRZ model45. These results render Xe2FeO2 a marginal lower mantle or CMB constituent at best.Reaction of noble gases and deep-Earth constituentsFinally, we highlight several significant aspects on the special role of FeO2 in trapping Xe in deep Earth. First, we have systematically examined possible reaction of Xe with major deep-Earth constituents FeO, SiO2, MgO, CaO, and Al2O3, and the resulting convex-hull data (Supplementary Fig. 5) show highly unfavorable energetics in all the cases, offering an underlying cause for unsuccessful past attempts to find Xe-bearing minerals in Earth’s mantle. Second, we have examined possible reactions of other noble gases Ne, Ar, and Kr with FeO2, and the results (Supplementary Fig. 6) indicate no tendency toward forming any stable noble-gas-Fe oxides up to 200 GPa. These results provide the possibility that Xe could be the sole inert element for reacting with deep-Earth constituents under mantle conditions. Moreover, while He-bearing compound FeO2He is found stable at CMB conditions, there is little direct bonding between He and Fe or O atoms in the compound37. Compared to Kr, Ar, Ne, and He, Xe has the lowest ionization energy and electronegativity, and consequently Xe is the easiest noble-gas atom to open up its outermost closed shell and form direct bonding as found in Xe2FeO2 and XeFe3O6.In summary, we have identified two Xe-Fe oxides, Xe2FeO2 and XeFe3O6, that are the first viable Xe-bearing compound at Earth’s lower mantle conditions. These new compounds are predicted by extensive crystal structure search in conjunction with ab initio energetic calculations and molecular dynamics simulations. Mass densities and compressional and shear sound velocities calculated at deep-Earth conditions are compatible with pertinent ULVZ and PREM data, thus confirming viability of Xe2FeO2 in geological environments. These results provide compelling evidence for a distinct deep-Earth Xe reservoir beyond previously proposed Xe-Fe and Xe-Ni intermetallic compounds in Earth’s inner core, thereby greatly expanding the range and scope of Xe-bearing compounds in deep Earth. The Xe-Fe oxides may enrich the understanding of prominent geophysical and geochemical processes, such as seismic anomalies near the CMB and possibly new chemical reactions inside Earth’s lower mantle.MethodsStructural predictionsOur structure search is based on a global optimization of free-energy surfaces using the CALYPSO methodology46,47, which has been successfully employed in predicting a large variety of crystal structures48–52. Evolutionary variable-cell calculations were performed at 120, 150, and 200 GPa with 1, 2, 3, and 4 formula units (f. u.) per cell. Most searches converge in 50 generations with about 2500 structures generated.Ab initio calculationsFirst-principles total-energy and electronic property calculations were carried out using the density functional theory with the Perdew–Burke–Ernzerhof exchange-correlation functional in the generalized gradient approximation (GGA)53,54 as implemented in the VASP code55, adopting frozen-core all-electron projector-augmented wave method56 with 3s2 3p6 3d7 4s1, 2s2 2p4, and 4d10 5s2 5p6 treated as valence electrons for Fe, O, and Xe, respectively. Correlation effects among the Fe 3d electrons were treated in the GGA + U approach57,58, adopting the recently proposed on-site Coulomb interaction U = 5.0 eV and a Hund’s coupling J = 0.8 Ev36,59–61. The spin-polarized and magnetic states were considered in obtaining the total-energy of the compounds containing iron. Zero-point energy was included in all reported calculations. A cutoff energy of 1200 eV for the plane-wave expansion and fine Monkhorst-Pack k meshes62 were chosen to ensure enthalpy convergence of better than 1 meV/atom.Phonon calculationsTo determine the dynamical stability, we performed phonon calculations by the direct supercell method using the Hellmann-Feynman theorem, as implemented in Phonopy code63. The harmonic interatomic force constants are calculated by 3 × 3 × 3 and 3 × 2 × 2 supercells for P21/c-Xe2FeO2 and P−1-XeFe3O6, respectively. Forces were calculated for atomic displacements of 0.01 Å, with a convergence threshold of 1 × 10−5 eV/Å.Van der Waals interactionTo examine the contribution of vdW interaction to the lattice energy, we have calculated the enthalpy of formation of Xe2FeO2 at high pressures using the vdW-DF2 density functional64. Our results show that the enthalpy of formation is less sensitive to the contribution of vdW correction at high-pressure conditions for Xe-FeO2 compounds, e.g., about 0.6 meV/atom for P21/c Xe2FeO2 at 135 GPa, thus vdW interaction is not considered in the calculations of lattice energy.Supplementary informationSupplementary InformationPeer Review File | nature communications | [
"Article"
] | [
"Geochemistry",
"Geochemistry",
"Theoretical chemistry"
] | chemical reaction of inert xenon predicted by Pauling 1933 first compound produced 19621 more compounds synthesized at ambient pressure electronegative atoms like pressure chemical reactivity moderate pressures solid xenon forms weakly bonded compounds with species H2O10 O211 at 1 3 GPa novel Xe compounds with unusual stoichiometries thermodynamically stable at high pressures Xe oxides13 nitrides15 xenon-hydrogen16 Xe–Mg compounds17 ultra-high-pressure high volatility chemical inertia abundant isotopes of xenon valuable tracer evolutionary dynamics history Earth 99% of Earth’s primordial Xe missing low abundance missing Xe problem19 hypotheses Xe escaped after or stored in interior Attempts to incorporate Xe into ices clathrates sediments not Laboratory experiments Xe in quartz30 stable xenon oxides13 explain missing Xe mystery xenon oxides unstable with metallic iron xenon silicates decompose at mantle pressures13Reactions Xe with iron nickel in Earth’s core predicted compounds synthesized under core pressure temperature challenging explore capture stabilization Xe by minerals mantle loss atmospheric Xe 100 million years from accretion mantle differentiation past searches chemical reactions Xe with mantle minerals FeO2 synthesized at lower mantle stablized above 74 GPa new possibility iron peroxide with helium rare helium-bearing compound deep-Earth primordial helium deposits prospects FeO2 react with Xe stability not at mantle conditions forming compounds trapping Xe interior explored reactions Xe with FeO2 mantle constituents FeO SiO2 MgO CaO Al2O3 FeO2 with Xe Xe-Fe oxides Xe2FeO2 XeFe3O6 strong Xe–O bonding other mantle oxides react Xe examined mass density sound velocities Xe-Fe oxides support viability in lower mantle region findings establish materials basis formation mechanism geological viability Xe-Fe oxides advance knowledge understanding xenon chemistry deep-Earth Xe reservoirResultsCrystal crystal phases FeO2: Xe ratios compute enthalpy viable structure construct convex hull Fig. 1a stable products reactions FeO2 Xe two Xe-Fe oxides Xe2FeO2 XeFe3O6 stable against decomposition at 150 200 GPa pressure–volume terms thermodynamical stability of Xe2FeO2 XeFe3O6 enthalpies convex 1) Xe-Fe oxides stable binary phases substances Xe-Fe-O2 Supplementary Information Figs. 2 3) Phonon dispersions at 150 GPa (Fig. 1b c show compounds dynamically stable thermal effects stabilize oxides pressure elevated temperatures deep-Earth conditions full analysis structural bonding characters oxides at 150 GPa compound Xe2FeO2 crystalized monoclinic structure P21/c symmetry (Fig. structural motif stacked layers corner-sharing octahedron Fe atom surrounded six O atoms Fe atom centered distorted octahedron Fe–O bond lengths 1.79–1.82 Å at 150 GPaXe atom coordination 3 bonding corners FeO6 octahedra Xe–O bond lengths 2.40–2.42 Å at 150 GPa similar Xe2O3 Xe2O5 XeFe3O6 triclinic structure P-1 symmetry two formula units per cell (Fig. corner FeO6 octahedra host Fe–O bonds 1.73–1.81 Å at 150 GPa tubular structure Xe atom coordination 6 Fe-O tube nearest Xe-O distance 2.08 Å vibrational mode lowest-frequency branch F Fig. 1c vibrational analyses Supplementary Fig. 4 structural details Xe-Fe oxides at 150 GPa Supplementary Information. 1Energetic stability structures Xe-Fe oxides-state convex hull solid lines FeO2-Xe system stable Xe-Fe oxides Xe2FeO2 XeFe3O6 at 150 200 GPa solid symbols stable structures unstable Phonon dispersions Xe2FeO2 150 GPa Xe3FeO6 150 Xe2FeO2 polyhedral enlarged view Xe3FeO6 Xe, Fe O atoms purple gray red spheresXe-Fe oxides calculated electronic density states at 150 GPa results Fig. 2a c reveal metallic nature DOS Fermi level contributions Fe 3d Xe 5p O 2p states Fe–O Xe–O bonding interactions calculated orbital Hamiltonian population) orbitals results Fig. 2b, d reveal Fe–O Xe–O covalent bonding strong low-energy bonding states occupied antibonding states near Fermi level occupied bonding states Xe2FeO2 deeper below Fermi level XeFe3O6 higher stability Xe2FeO2. bonding ICOHP values Fe–O Xe–O bonds at 150 GPa -1.45 eV/pair −0.24 eV/pair in XeFe3O6 −1.01 eV −0.12 eV/pair Xe2FeO2 results show Xe–O bonding Fe–O bonding contrast He-FeO2 compound He atoms Coulomb shield examined charge distribution in Xe2FeO2 XeFe3O6 unusual charge transfer from Fe Xe to O atoms At 150 GPa Bader partial charges Xe2FeO2 +0.30, +1.−1.00 Xe Fe O Bader partial charges XeFe3O6 +1.35 −0.90 Xe Fe O same pressure Xe atoms XeFe3O6 donate more electrons Xe2FeO2 different valence states FeO2–Xe compounds charge transfers indicate strong bonding Xe rare among noble gases.Fig. 2Electronic properties Xe-Fe oxides at 150 GPa Fe-d O-p Xe-p orbitals Xe2FeO2. crystal orbital Population-pCOHP) Xe2FeO2 compound -pCOHP >0 bonding <0 antibonding states PDOS Fe-d O-p Xe-p orbitals XeFe3O6 -pCOHP XeFe3O6 Fermi energy zero material stability geophysical temperatures 2000–4500 K Gibbs free energy Xe2FeO2 XeFe3O6 lattice contributions entropic term quasiharmonic approximation volume temperature lattice expansion Fig. 3a energetic terms structure stability internal energy U values Xe-Fe oxides higher than Xe FeO2 PV terms favorable lower enthalpy H formation oxides temperature effectthermal vibration atomic reduce Gibbs free energy G Xe-Fe oxides stabilizing crystal structures threshold pressure oxides reduces increasing temperatures expanding stability field phase diagram.Fig. energetic terms mean square deviations energetic term Xe-Fe oxides enthalpy pressure–volume terms internal energy Gibbs free energy) at 150 GPa, 2000 K Mean deviations) Xe O Fe atomic positions Xe2FeO2 at 150 GPa, 3000 K MSD Xe O Fe XeFe3O6 at 150 GPa, 3000 K.Phase temperature effects performed energetic molecular dynamics) simulations structural stability pressure–temperature phase diagram Xe-Fe oxide system Fig. 3b mean square deviations) atomic positions Xe-Fe oxides at high P–T conditions 150 GPa 3000 K Fe O Xe atoms remain near lattice sites stability crystal structure AIMD simulations performed temperature-driven instability Gibbs free energies computed compared solid phases phase diagram (Fig. 4) spans P–T range higher.Fig. Pressure–temperature phase diagram Xe-FeO2 systemdotted dashed lines indicate phase boundaries temperature instability determined by Gibbs energy initio molecular dynamics calculations symbols show phase transition points solid triangle symbols represent unstable phases AIMD simulations P–T regions stable Xe2FeO2 covered by slash lines XeFe3O6 filled by back-slash pressure boundary between lower mantle core top figure geotherm Earth’s analyze stability fields Xe-Fe oxides under (P, T) conditions geological constraints Fig. 4. Xe2FeO2 stable pressure 110-120 GPa temperatures around 2500 K geotherm mantle Xe2FeO2 XeFe3O6 stable pressures 120–136 GPa temperatures 2500–3600 K pressure temperature rise Xe2FeO2 sole stable phase results suggest stable Xe-Fe oxides under (P, T conditions in deep-Earth regions necessary to assess viability Xe-Fe oxides material characteristicsrun AIMD simulations crystal structures T conditions used strain-stress elastic tensors elastic-wave velocities determined Christoffel equation|Tik-δikρV2 = 0 δik Kronecker delta function V seismic Tik Christoffel density sound mass density compressional shear sound velocities points (120 GPa, 2500 K lower mantle (135 GPa, 3500 K CMB geological data XeFe3O6 stability stability lower mantle CMB higher temperatures destabilize calculated densities XeFe3O6 8.86 9.06 g/cm3 lower mantle CMB points 4.95–9.90 g/cm3 core rigidity zone model 5.57–8.91 g/cm3 ultralow velocity zone calculated (VS) 8.94(4.03) km/s 9.20(4.26) km/s 8.20–10.70 km/s (1.00–5.00 km/s) indicate XeFe3O6 viable constituent lower mantle CMB conditions calculated densities Xe2FeO2 9.78 9.87 g/cm3 at lower mantle CMB points outside range 5.57–8.91 g/cm3 ULVZ model top range 4.95–9.90 g/cm3 CRZ model45 calculated VP (VS) 7.83(4.21) km/s 7.94(4.02) km/s lower mantle CMB points VS values km/s CRZ 2.91–6.17 km/s ULVZ VP values 10.97–13.03 km/s ULVZ 8.20–10.70 km/s CRZ render Xe2FeO2 marginal lower mantle CMB constituent.Reaction noble gases deep-Earth FeO2 trapping Xe deep Earth examined reaction Xe with deep-Earth constituents FeO SiO2 MgO CaO Al2O3-hull data show unfavorable energetics unsuccessful attempts Xe-bearing minerals mantle examined reactions noble gases Ne Ar Kr with FeO2 results no tendency stable noble-gas-Fe oxides up to 200 GPa Xe sole inert element reacting deep-Earth constituents mantle conditions He-bearing compound FeO2He stable at CMB conditions little direct bonding between He Fe O atoms Xe lowest ionization energy electronegativity easiest noble-gas atom open form direct bonding Xe2FeO2 XeFe3O6.identified two Xe-Fe oxides Xe2FeO2 XeFe3O6 first viable Xe-bearing compound lower mantle conditions compounds predicted by crystal structure search initio energetic calculations molecular dynamics simulations Mass densities compressional shear sound velocities deep-Earth compatible with ULVZ PREM data confirming viability Xe2FeO2 results evidence deep-Earth Xe reservoir beyond Xe-Fe Xe-Ni expanding range Xe-bearing compounds Xe-Fe oxides enrich understanding geophysical geochemical processes seismic anomalies new chemical reactions lower mantle structure search based global optimization free-energy surfaces CALYPSO methodology46 crystal variable-cell calculations performed at 120 150 200 GPa 1 2 3 4 formula units per cell searches converge in 50 generations 2500 structures generated initio total-energy electronic property calculations density functional theory Perdew–Burke–Ernzerhof exchange-correlation gradient approximation frozen-core all-electron projector-augmented wave valence electrons for Fe, O XeFe 3d electrons treated GGA + U Coulomb interaction U 5.0 eV Hund’s coupling J 0.8 spin-polarized magnetic states considered total-energy compounds iron Zero-point energy included calculations cutoff energy 1200 eV plane-wave expansion Monkhorst-Pack enthalpy convergence 1 meV/atom calculations direct supercell method Hellmann-Feynman theorem Phonopy harmonic interatomic force constants calculated 3 × × 2 × 2 supercells P21/c-Xe2FeO2 P−1-XeFe3O6 Forces atomic displacements Å convergence threshold 1 × 10−5 eV/Å der Waals calculated enthalpy formation Xe2FeO2 high pressures vdW-DF2 density enthalpy less sensitive vdW correction high-pressure Xe-FeO2 compounds 0.6 meV/atom P21/c Xe2FeO2 135 GPa vdW interaction not considered calculations lattice energy.Supplementary Review File | 51.2 | 0.383785 |
10.1038/s41467-020-14638-w | PMC7021672 | The ATM kinase is a key regulator of the DNA damage response to double-strand breaks (DSBs) and its homozygous loss in patients predisposes to lymphoid malignancies. Here, the authors develop a Tdp2−/−
Atm−/− double-deficient mouse model to uncover topoisomerase II-induced DSBs as significant drivers of the genomic rearrangements that underpin these tumours. | The ATM kinase is a master regulator of the DNA damage response to double-strand breaks (DSBs) and a well-established tumour suppressor whose loss is the cause of the neurodegenerative and cancer-prone syndrome Ataxia-Telangiectasia (A-T). A-T patients and Atm−/− mouse models are particularly predisposed to develop lymphoid cancers derived from deficient repair of RAG-induced DSBs during V(D)J recombination. Here, we unexpectedly find that specifically disturbing the repair of DSBs produced by DNA topoisomerase II (TOP2) by genetically removing the highly specialised repair enzyme TDP2 increases the incidence of thymic tumours in Atm−/− mice. Furthermore, we find that TOP2 strongly colocalizes with RAG, both genome-wide and at V(D)J recombination sites, resulting in an increased endogenous chromosomal fragility of these regions. Thus, our findings demonstrate a strong causal relationship between endogenous TOP2-induced DSBs and cancer development, confirming these lesions as major drivers of ATM-deficient lymphoid malignancies, and potentially other conditions and cancer types. | IntroductionAtaxia Telangiectasia (A-T) is the paradigm of human inherited disease related to the DNA damage response (DDR)1,2. A-T is caused by loss-of-function mutations in the ATM gene, and is associated with multi-systemic features, affecting brain, gonads and the immune system; the main and most debilitating symptom of the disease being progressive early-onset cerebellar ataxia. Spontaneous chromosomal instability and profound hypersensitivity to agents that induce DNA double-strand breaks (DSBs) distinguish this syndrome from other spinocerebellar ataxias3, leading to the idea of DSBs underlying A-T symptomatology, although there is still an open debate regarding the molecular trigger of A-T, and the specific contribution of DSBs to the different aspects of the disease.The connection of DSBs with immunodeficiency and cancer predisposition in A-T is, however, well established and properly recapitulated in the Atm−/− mouse4. During lymphocyte development, the variable regions of T-cell receptor (TCR) and Immunoglobulin (Ig) loci, in T and B cells respectively, are randomly rearranged in a unique combination of the multiple possible V, D (in the case of TCRβ and δ and IgH) and J coding segments present in the germline5. This shuffling occurs through the generation of pairs of DSBs by the RAG1-RAG2 recombinase at recombination sequence signals (RSS), and their subsequent repair by the non-homologous end-joining (NHEJ) machinery. ATM, although not completely required, facilitates V(D)J recombination, providing a molecular explanation for the relatively mild immunodeficiency characteristic of A-T patients and Atm−/− mice. The main role of ATM in V(D)J recombination is to stabilize RAG post-cleavage complexes, thus protecting DNA ends and promoting their correct use for ligation6,7, but other functions that are not fully understood become relevant in conditions in which NHEJ is compromised8–10.Defects in V(D)J recombination are not only responsible for the immunological problems in A-T, but also underlie its characteristic cancer predisposition. Indeed, roughly one third of patients develop cancer, mainly lymphoma or lymphocytic leukemia, which in mice manifests as very aggressive thymic neoplasias, with characteristics of T-cell acute lymphoblastic leukemia (T-ALL)4. A-T and Atm−/− T-cell malignancies are frequently linked to genome rearrangements involving the TCR loci, strongly supporting the contribution of aberrant V(D)J recombination. Most prevalent in A-T are translocations or inversions (14;14) involving the TCRα/δ locus. Interestingly, this event is also frequent in mice as a t(14;12) translocation, providing thus a molecular link between ATM-deficient T-cell malignancies in human and mouse11–14. Furthermore, T-cell malignancies in Atm−/− mice and human share other characteristics such as chromosome 15 duplication, Notch1 amplification and Pten deletion13,14. In the currently accepted model, deficient V(D)J recombination and checkpoint defects caused by ATM loss lead to persistent DSBs that engage in oncogenic rearrangements15. Aberrant V(D)J recombination, however, is unlikely to represent the single driver of oncogenic translocations, as evidenced by additional V(D)J-unrelated regions of instability, and the persistent cancer predisposition observed upon RAG deficiency16,17. These results strongly suggest additional sources of DSBs as relevant contributors to the ATM-deficient oncogenic translocations responsible for T-cell cancer predisposition.In this sense, aberrant action of DNA topoisomerase II (TOP2) can constitute an important source of chromosomal breakage18. The physiological function of TOP2 is to solve topological problems arising from DNA metabolism19. To do so, it cleaves both strands of DNA to gate the passage of another DNA segment, via the formation of a catalytic intermediate, the cleavage complex (TOP2cc), in which the enzyme remains covalently linked to 5′ termini of the break. Although normally very transient, as the DNA gate is rapidly resealed after DNA passage, TOP2ccs can be stabilized and result in irreversible TOP2-induced DSBs upon conflict with cellular processes and processing of the structure by the proteasome20. Interestingly, this does not only occur accidentally as a consequence of errors in TOP2 catalytic cycles, but underlies the clinical efficacy of a heterogeneous group of chemotherapeutic agents collectively known as TOP2 “poisons”, of which etoposide is a widely characterised paradigmatic example21. Despite substantial efficacy, treatment with TOP2 poisons has been traditionally linked to the development of secondary haematological malignancies characterized by specific translocations22. In this regard, transcription and loop extrusion, a process that organizes the genome in functional loop domains by threading chromatin through the ring-shaped cohesin complex23–25, have been proposed as the main source of DSBs and chromosomal translocations induced by etoposide treatment26–29. However, the incidence and impact of endogenous TOP2-mediated lesions in the absence of treatment with TOP2 poisons and their relationship with cancer development remain to be established. Concerning this, compounds present in the diet and the environment and some forms of pre-existing DNA damage, such as nicks or abasic sites, can also poison TOP2 activity18.Remarkably, TOP2-induced DSBs are characterized by harbouring peptide blocks derived from the enzyme that remain covalently bound to 5′ termini through a phosphotyrosyl bond. TDP2, with its 5′-tyrosyl-DNA-phosphodiesterese activity30,31, is the only mammalian enzyme able to directly unblock TOP2-induced DSBs for their repair32. Interestingly, we have previously reported that ATM is specifically involved in facilitating repair of TOP2-induced DSBs in a TDP2-independent fashion, probably by promoting alternative nucleolytic pathways for the unspecific removal of the peptide adducts33. TDP2 and ATM, therefore, define two independent pathways for the repair of TOP2-induced DSBs, the combined absence of which has dramatic consequences for the cellular response to these lesions.In the present study, we generate and characterise Atm−/−
Tdp2−/− double-deficient mice to address the occurrence and relevance of TOP2-mediated DSBs in vivo, and, most importantly, their possible contribution to the different aspects of A-T symptomatology. We find that TDP2 loss further aggravates the predisposition of Atm−/− mice to thymic malignancies, and show a strong colocalization between RAG and TOP2 at regions that endogenously accumulate DSBs and TOP2ccs in Atm−/−
Tdp2−/− thymocytes. These results strongly suggest the co-occurrence and misrepair of RAG- and TOP2-mediated DSBs as a major driving force of thymic malignancies linked to ATM deficiency.ResultsTdp2−/−Atm−/− and Atm−/− mice are phenotypically similarConsidering the strong sensitivity and repair defect of Tdp2−/−
Atm−/− double-deficient cells to TOP2-induced DSBs33, we decided to address the physiological impact of these lesions in the context of the entire organism. We reasoned that, if TOP2 damage is mediating deleterious effects in the Atm−/− mouse, these would be further aggravated by Tdp2 loss. Mice from Tdp2 and Atm double heterozygote crosses were born at the expected Mendelian proportions and not displaying gross abnormalities (Supplementary Fig. 1a). Since Atm loss causes marked growth retardation4, we measured body weight of male mice at 4 weeks of age (Supplementary Fig.1b). As expected, Atm−/− mice showed a more than 20% reduction in size compared to wild-type animals. Deletion of Tdp2, however, did not have an impact on its own, as previously reported32, neither did it aggravate the growth retardation of Atm−/− mice. Furthermore, neither Tdp2−/−
Atm−/− nor, as previously reported4, Atm−/− mice displayed clear symptoms of neurological deficiencies or an ataxic behaviour, which we confirmed by analysing cerebellar integrity and Purkinje cellular density in 8-week old mice (Supplementary Fig. 1c). Finally, Tdp2−/− mutation did not substantially modify the already strong disruption of spermatogenesis found in Atm−/− mice (Supplementary Fig. 1d).Primary Tdp2−/−Atm−/− mouse embryonic fibroblasts (MEFs), however, showed a marked reduction in proliferation compared to single-mutant and wild-type cells (Supplementary Fig. 1e), and spontaneously accumulated DSBs in unchallenged growth conditions (Fig. 1a), with 60% of Tdp2−/−Atm−/− cells harbouring more than one 53BP1 focus compared to 25% in wild-type. Overall, these results suggest an accumulation of endogenous DSBs in the absence of TDP2 and ATM which, at least under certain growth conditions, can disrupt cellular fitness. The in vivo incidence of these lesions, however, does not seem sufficiently high to seriously compromise tissue homeostasis during development and early stages of life.Fig. 1Tdp2−/−Atm−/− causes the spontaneous accumulation of DSBs and etoposide hypersensitivity in mice.a Endogenous DSB occurrence in Tdp2+/+
Atm+/+, Tdp2−/−
Atm+/+, Tdp2+/+
Atm−/− and Tdp2−/−
Atm−/− primary MEFs measured by 53BP1 foci formation. The percentage of cells with more than one focus (left) and representative images (right) of 53BP1 foci (green) and DAPI staining (blue) are shown for each genotype. MEFs isolated from three different embryos were analysed per condition. Mean ± SEM and statistical significance by one-way ANOVA with Bonferroni post-test is shown (F = 5,402). b Body weight of 8-week old mice cohorts intraperitoneally injected with a single dose of etoposide (25 mg/kg). Average ± SEM of the percentage of initial body weight from five mice and statistical significance by Two-way ANOVA with Bonferroni post-test is shown (F = 29,11). Tdp2+ indicates both Tdp2+/+ and Tdp2+/− genotypes. Representative images of Haematoxylin-Eosin stained jejunum slices obtained 6 days after etoposide exposure are shown.Tdp2−/−Atm−/− mice are hypersensitive to TOP2-induced DSBsGiven the results above, we decided to challenge Tdp2−/−
Atm−/− mice with an increased load of TOP2-induced lesions. To do so, 8-week old adult mice of the relevant genotypes were subjected to a single intraperitoneal injection of the TOP2-poison etoposide (25 mg/kg). As can be seen in Fig. 1b, Tdp2−/−
Atm−/− double-knockout mice showed severe hypersensitivity to etoposide exposure, suffering a progressive weight loss that reached, on average, a 15% decrease after 6 days. Histopathological analysis revealed marked villous atrophy in the small intestine mucosa, which is a known etoposide target, as the likely cause of the drastic weight loss in Tdp2−/−
Atm−/− (Fig. 1b). In contrast, at this low etoposide concentration, wild-type and Atm−/− animals did not respond negatively to the treatment, while Tdp2−/− did so only mildly. This synergistic hypersensitivity of Tdp2−/−
Atm−/− mice to etoposide suggests a protective role of ATM against the adverse effects of TOP2 lesions in vivo, but only when TDP2 is absent, which recapitulates cellular observations in the context of the entire organism33.TDP2 loss increases thymic cancer predisposition of Atm−/−We decided to analyse Tdp2−/−
Atm−/− mice over a longer period, to allow an accumulation of a higher load of TOP2-mediated damage and/or the development of potential pathologies. For this, weight and general health status were monitored weekly in a minimum of 20 mice for each of the relevant genotypes, until reaching the experimental endpoint of 2 years (730 days) (Fig. 2a). As previously reported, we observed that Atm−/− mice showed a reduced life-span with an important number of animals succumbing to thymic neoplasia after the first few months of life (Fig. 2b, Supplementary Fig. 2a). Furthermore, a molecular characterization of the cellular content in these tumours showed a majority (70%) of double positive CD4+ CD8+ T-cell tumours (Fig. 2c, Supplementary Fig. 2b), consistent with the reported T-ALL predisposition.Fig. 2TDP2 deficiency increases thymic cancer predisposition of Atm−/− but not of Tpr53−/− mice.a, b Kaplan–Meier survival curve (a) and cumulative occurrence (b) of thymic tumours during the 1st year of life of at least 20 mice with the indicated Tdp2 and Atm genotypes. n = 21 (Tdp2+/+
Atm+/+), n = 23 (Tdp2−/−
Atm+/+), n = 23 (Tdp2+/+
Atm−/−), n = 21 (Tdp2−/−
Atm−/−). Tdp2+ indicates both Tdp2+/+ and Tdp2+/− genotypes. Statistical significance by two-sided Wilcoxon test is indicated. c Percentage of tumours composed by immature CD4+ CD8+ double positive, or CD4+/CD8+ single positive lymphocytes in the indicated genotypes. Thymocytes were selected by size and complexity prior classification by high or low CD4 and CD8 markers (see Supplementary Fig. 1a for gating strategy). d Median life-span of Tdp2+/+
Atm−/− and Tdp2−/−
Atm−/− mice affected by a thymic tumour (n = 11 and n = 16 for Tdp2+/+
Atm−/− and Tdp2−/−
Atm−/− mice respectively). e Kaplan–Meier survival curve and f cumulative incidence of thymic tumours of the indicated Tdp2 and Trp53 genotypes (n = 16 for Tdp2+/+ and n = 22 for Tdp2−/−).Single Tdp2−/− mice had a lifespan comparable to wild-type (Fig. 2a), and none of the animals developed thymic lymphoma within the course of the experiment (2 years). Strikingly however, in Atm−/− background, both the reduced overall life-span and the incidence of thymic tumours were largely aggravated by Tdp2 inactivation (Fig. 2a, b). Thus, Tdp2−/−Atm−/− mice showed a median survival of 140 days, contrasting with 307 days for the Atm−/− single mutant (Fig. 2a), and the probability of developing a thymic tumour during the 1st year of life increased from a 43% in Atm−/− to a 72% in the Tdp2−/−Atm−/− double mutant (Fig. 2b). Importantly, lifespan of Tdp2−/−Atm−/− and Atm−/− mice that did develop a thymic tumour was not significantly different (Fig. 2d), suggesting a direct effect on the incidence and not on the latency or aggressiveness of T-cell malignancies. In this sense, tumours observed in Tdp2−/−Atm−/−mice were also predominantly formed by double positive CD4+ CD8+ T-cell precursors (80%) (Fig. 2c). Double-knockout mice, however, did not show differential distribution in the populations of double negative (CD4−CD8−), double positive (CD4+CD8+) or single positives (CD4+ or CD8+) thymocytes compared to Atm−/− mice (Supplementary Fig. 2c), suggesting that the increased incidence of T-cell malignancy caused by TDP2 loss is unlikely to reflect further disruption of V(D)J recombination. In the same way, pre-B-cell, pro-B-cell and immature B-cell populations did not differ between single Atm−/− and double Tdp2−/−Atm−/− knock-out mice (Supplementary Fig. 2d), further supporting the idea that V(D)J recombination is not substantially altered by TDP2 loss.The checkpoint and proapoptotic functions of ATM, which are strongly signalled through the p53-dependent pathway, are known to be determinant for Atm−/− cancer predisposition15. Consistent with this, p53-deficient Trp53−/− mice are also highly susceptible to the development of double positive CD4+CD8+ thymic tumours34. In order to discern between the contribution of repair and checkpoint/apoptotic functions of ATM in the increased incidence of thymic tumours in Tdp2−/−Atm−/−, we analysed life span and tumour incidence in Tdp2−/−Trp53−/− animals, in which the repair component is not substantially affected. In contrast to what was observed in Atm−/− background (Fig. 2a, b), loss of functional TDP2 did not significantly decrease lifespan (Fig. 2e), nor increase the high incidence of cancer (Supplementary Fig. 2e), and thymic lymphoma in particular (Fig. 2f), observed in Trp53−/− animals. This suggests that Tdp2−/−Atm−/− tumours likely reflect direct functions of ATM in the repair of TOP2-induced DSBs, and not only in their signalling for checkpoint and apoptosis. Overall, the results are consistent with a model in which misrepair of TOP2-induced DSBs can strongly contribute to the development of thymic tumours characteristic of ATM deficiency.Tdp2−/−Atm−/− and Atm−/− malignancies are molecularly similarIn order to gain insights into the molecular events responsible for thymic tumour development in Tdp2−/−Atm−/− mice, we analysed and compared copy-number variation by comparative genomic hybridization (CGH) in six Tdp2−/−Atm−/− and three Atm−/− thymic tumours (Fig. 3a, Supplementary Fig. 3 and Supplementary Table 1). In Atm−/− animals (Fig. 3a, b), we observed the previously reported features of genomic instability13, with amplification upstream of the Tcra/d locus in chromosome 14 (2 out of 3 mice) and variable hemizygous loss of a Bcl11b-containing telomeric region of chromosome 12 (all 3 mice), which are indicative of the frequent t(14;12) translocation and subsequent breakage-fusion-bridge cycles. In addition, one mouse displayed instability at the Tcrb locus, with both gain and loss of DNA sequence. Two of the three animals also presented the characteristic duplication of chromosome 15. Also, two animals presented deletions in chromosome 19 covering the Pten tumour suppressor, although only in one of the cases this corresponded to the previously reported homozygous loss13. Finally, we did not observe amplification of Notch1 in chromosome 2, which is another common, although less frequent, feature in Atm−/− T-cell malignancies.Fig. 3Tdp2−/−Atm−/− and Atm−/− thymic malignancies display similar genome rearrangements.a Merged CGH analysis of Tdp2+/+Atm−/− (three mice, top) and Tdp2−/−Atm−/− (six mice, bottom) thymic lymphomas. DNA from each tumour sample was hybridized and analysed using kidney DNA from the same mouse as a control. Average amplification (red) or deletion (blue) score (Log2 tumour/kidney ratio) is shown. Significant copy number variations are defined by −0.66>Log2 tumour/kidney>0.66 (dashed lines). The location of relevant loci in Atm−/− thymic tumours is indicated. b Table summarising the number of mice of each genotype displaying copy number variation at the indicated locus of interest. Amp(Tcra/d): amplification upstream of Tcra/d; Del(12): deletion of the telomeric region of chromosome 12 covering Bcl11b; Del(Tcrb): deletion at the Tcrb locus; Amp(15): trisomy of chromosome 15; Del(Pten): deletion at the Pten locus; Amp(Notch1): duplication at the Notch1 locus.Surprisingly, we observed very similar features in Tdp2−/−Atm−/− mice (Fig. 3a, b). Four out of six tumours presented Tcra/d linked amplification, and all six showed chromosome 12 deletion with Bcl11b loss. Instability at the Tcrb locus was also observed in two of the six cases. Trisomy of chromosome 15 was also observed in four of the six animals, while only one of the six cases carried Notch1 amplification. All these events were coincident with the results obtained in Atm−/− animals (see above) and previous reports13. Finally, as observed for Atm−/−, one Tdp2−/−Atm−/− animal displayed hemizygous Pten loss as part of a larger deletion. The variable length of the copy-number variation events at common regions, together with additional sites of instability that are unique to individual tumours (Supplementary Fig. 3), are indicative of the stochastic nature of the oncogenic events. In summary, we can conclude that loss of TDP2 increases the incidence of the same oncogenic rearrangements that drive thymic tumour in Atm−/− animals, suggesting a strong contribution of TOP2-mediated lesions to ATM-deficient T-cell cancer predisposition.TOP2B is enriched at sites of DSB accumulationIn order to further establish aberrant TOP2 activity as a direct contributor to Atm−/− oncogenic T-cell genome rearrangements, we decided to map binding of topoisomerase IIß (TOP2B), which is the main source of TOP2 activity in G1 and non-cycling cells. ChIPseq analysis in freshly isolated thymocytes from wild-type mice revealed a pattern of TOP2B distribution consistent with that previously found in other cell types and tissues35–37 (Fig. 4a), with a clear enrichment at promoter, enhancer and insulator regions (34%, 19% and 13% of TOP2B peaks, respectively). In fact, regardless of the type of functional element analysed, and as previously reported36, we observed a strong correlation between TOP2B and the core cohesin subunit RAD21, with 81% of TOP2B peaks overlapping with those of cohesin (Fig. 4b), and detectable cohesin signal in virtually all TOP2B peaks, regardless of the presence of RNA polymerase II (POLR2A) or the insulating architectural protein CTCF (Fig. 4c). Fig. 4d shows the Cxcr4 locus merely as an illustrative example of TOP2B and cohesin co-localization within promoter (left) and insulator (right) regions. These results are consistent with the proposed connection between TOP2B function and 3D-genome organisation36,37.Fig. 4TOP2B colocalizes with cohesin at promoter, enhancer and insulator regions in thymocytes.a Genome-wide distribution of 1003 TOP2B ChIPseq called peaks at different genomic features (promoters, enhancers, insulators and others) in wild-type primary thymocytes. Chromatin states were defined using ENCODE ccREs registry: H3K4me3, TSS and low H3K27ac for promoters, H3K27ac and low H3K4me3 for enhancers and high CTCF and no presence of the previous histone marks for insulator-like regions.TOP2B peak distribution compared to the entire genome (top), and global signal profile compared to IgG control (bottom) at each genomic feature is shown. TOP2B peaks were defined as common peaks between TOP2B ChIPseq experiments using two different antibodies. b Overlap of TOP2B and RAD21 peaks in wild-type thymocytes. c Genome browser view of TOP2B, RAD21, CTCF, POLR2A, H3K4me3 (ENCODE) and H3K27ac (ENCODE) signal tracks at a representative genomic region encompassing an active promoter (left, positive for POLR2A, H3K4me3 and H3K27ac) and an insulator (right, negative for POLR2A, H3K4me3 and H3K27ac, and positive for CTCF and RAD21). Sites highlighted in grey show sites of major TOP2B signal accumulation over the region. d Heatmaps of TOP2B (this study), POLR2A (ENCODE), RAD2147 and CTCF (ENCODE) ChIPseq signals at TOP2B peaks (± 5 kb). K-means clusters based on the different signals represented are shown divided by black lines and named as C1 and C2.We decided to specifically study TOP2B occupancy at particular sites related to Atm−/− oncogenic translocations, integrating published maps of endogenous DSB accumulation determined by ENDseq in Atm−/− thymocytes 38. When concentrated on the Bcl11b, Notch1, and Pten loci (Fig. 5a–c), we observed clear TOP2B binding at all of these regions with strong accumulation at the promoters and towards the 3’ end of the genes, with a pattern following that of cohesin and RNA-polymerase II (POLR2A). This TOP2B accumulation was not thymocyte-specific, as a similar pattern was observed in MEFs (shown for Bcl11b, Supplementary Fig. 4a). In order to directly identify regions of DSB accumulation in Atm−/− thymocytes, we selected ENDseq peaks specific for Atm−/− thymocytes (highlighted in yellow; Fig. 5a–c). Interestingly, when these sites were compared with the distribution of TOP2B, overlap was highly variable with examples of both unique and common regions; compare region 1 and region 2 in Pten (Fig. 5c). TOP2B is therefore present at Atm−/− unstable regions, but it is difficult to establish a direct association with DSB accumulation, and may be a mere consequence of its widespread distribution at regulatory regions. Furthermore, the difficulty to assign original translocation breakpoints with sufficient precision strongly limits our capacity to directly link TOP2B binding to oncogenic DSB occurrence at these specific locations.Fig. 5TOP2B colocalizes with endogenous DSB accumulation in thymocytes.a–c Genome browser view of TOP2B, POLR2A, RAD21, and ENDseq (DSBs) signal tracks in wild-type and Atm−/− primary thymocytes, as indicated, at Bcl11b (a), Notch1 (b) and Pten (c) loci. Atm−/−-specific ENDseq peaks, defined as those not called in wild-type thymocytes, are highlighted in yellow. Two regions of interest in Pten are indicated as region 1 and region 2. d Global profile of ENDseq signal at TOP2B peaks in wild-type, Atm−/− and Rag2−/− primary thymocytes. Control input signal is also shown. e Global profile of TOP2B signal at ENDseq peaks in wild-type primary thymocytes. Set of ENDseq peaks were determined as the merge between those detected in any of the analysed genetic conditions.Based on this, we decided to compare the distribution of TOP2B and ENDseq signal at a genome-wide level. We observed a clear accumulation of DSBs at TOP2B peaks (Fig. 5d), supportive of a link between TOP2B function and spontaneous chromosome fragility, in the same line as what has been previously reported for etoposide treatments37. Conversely, and further supportive of this TOP2B-DSB association, ENDseq peaks displayed an accumulation of TOP2B signal (Fig. 5e). Most prominent peaks of ENDseq signal in thymocytes are known to be associated with RSS sites, disappearing in Rag2−/− and strongly accumulating in Atm−/− cells38, consistent with the high incidence and precision of RAG cleavage and the reported roles of ATM in facilitating V(D)J recombination. At TOP2B sites, however, DSBs were noticeably increased in both Rag2−/− and Atm−/− thymocytes (Fig. 5d), demonstrating that they are unrelated to RAG cleavage, and suggesting that these lesions can accumulate upon ATM deficiency. In summary, there is a significant genome-wide correlation between sites of TOP2B function and occurrence of RAG-independent endogenous DSBs in thymocytes, suggesting aberrant TOP2B activity as an additional relevant source of chromosome breakage that could contribute to characteristic Atm−/− oncogenic translocations.TOP2B and cohesin colocalize with the RAG endonucleaseBased on the results above, we decided to check how RAG and TOP2B contribute to DSB accumulation in more detail. In order to do so, we analysed TOP2B, RAD21 and ENDseq signal in previously reported peaks of RAG1 and RAG212 (Fig. 6a–c). RAG2 has a wide distribution with a strong presence at active promoters displaying the H3K4me3 histone mark, while the presence of RAG1 is more restricted, only co-localizing with a fraction of RAG2 peaks39,40. We observed a striking genome-wide co-localization of TOP2B and RAD21 with RAG2, which was irrespective of the strong presence of RAG1 or not (Fig. 6a). Thus, clear TOP2B accumulation was observed in peaks containing both RAG1 and RAG2 but also in those containing mostly RAG2 (Fig. 6a, b). In addition, 42% of TOP2B peaks colocalized with RAG (merge of RAG1 and RAG2 peaks) (Fig. 6b), consistent with their mutual enrichment at H3K4me3-positive promoter regions. In any case, although some ENDseq signal was observed at these regions, the accumulation of DSBs was not clear or sufficiently localized (Fig. 6a). Interestingly, when directly compared, spontaneous DSB accumulation was more robust at TOP2B and cohesin than at RAG1 or RAG2 peaks, both in wild-type and Atm−/− thymocytes (Fig. 6d). Altogether, these results uncover a surprising genome-wide connection between RAG and TOP2B-cohesin, and suggest that, when analysed globally, TOP2B activity can constitute an important source of DNA breakage in Atm−/− thymocytes, quantitatively more relevant than DSBs directly induced by RAG.Fig. 6TOP2B and cohesin colocalize with RAG genome-wide.a Heatmaps of RAG240, RAG140, TOP2B, RAD21 and ENDseq (DSBs) signal at RAG1 and RAG2 peaks (merged) in wild-type and Atm−/− primary thymocytes, as indicated. K-means clusters based on RAG1 and RAG2 signal are shown as C1 and C2. b Overlap of TOP2B and RAG peaks. Peaks of the RAG complex were defined as a merge of RAG1 and RAG2 proteins. c Global profile of TOP2B signal at RAG1 or RAG2 peaks. Control IgG signal is also shown. d Global profile of ENDseq signal at either RAG (merge of RAG1 and RAG2 peaks) or TOP2B-RAD21 (overlap between TOP2B and RAD21) peaks in wild-type, Atm−/− and Rag2−/− primary thymocytes.Based on this genome-wide colocalization between TOP2B and RAG, we decided to check the distribution of TOP2B specifically at sites actively undergoing V(D)J recombination in thymocytes (Fig. 7a–c; Supplementary Fig. 5a, b); this includes the Tcra/d, Tcrb, Tcrg and IgH15 loci. Strikingly, sites of strong and localized TOP2B accumulation were observed at all of these sites, and were particularly enriched at the V(D)J-initiating J segments. As expected, TOP2B coincided with the position of the cohesin subunit RAD21, and, in line with the genome-wide observations, also with the strongest peaks of RAG1 and RAG2. Furthermore, the TOP2B-cohesin sites concurred with a very strong ENDseq signal that was further enhanced in Atm−/− thymocytes. Interestingly however, DSB accumulation in these regions was completely dependent on RAG, and did not fully coincide with TOP2B-cohesin peaks, but was displaced a few kbs from TOP2B sites (Supplementary Fig. 6). This demonstrates that direct RAG cleavage, rather than aberrant TOP2B activity, is the main source of DSBs at Tcr and Igh loci in thymocytes, but suggests a strong association between TOP2B-cohesin function and sites undergoing active V(D)J recombination. In support of this, TOP2B accumulation in these regions was specific to thymocytes when compared to peaks called in MEFs (Supplementary Fig. 4b, c). ChIPseq of TOP2A also evidenced that TOP2 enrichment was preferential for the TOP2B isoform, both at V(D)J recombination sites and other genes related with the observed oncogenic translocations (Supplementary Fig. 4). Furthermore, when ENDseq signal at Tcrb was analysed in detail, there were two peaks that became apparent when the predominant signal derived from RAG cleavage was lost in Rag2−/− thymocytes (Fig. 7c; ENDseq Rag2−/− autoscaled). This relocation of ENDseq signal from major sites of RAG cleavage to additional less prominent sites may be responsible for the apparent increase in DSBs observed at TOP2B peaks in Rag2−/− background (Figs. 5a and 6d). In any case, similarly to what occurred on a genome-wide scale and at other particular locations (Fig. 5), these regions of DSB accumulation were perfectly coincident with the two major sites of TOP2B and RAD21 binding, further supporting an association between TOP2B function and RAG-independent DNA breakage. Nevertheless, despite a similar localized accumulation of TOP2B-cohesin, RAG-independent ENDseq signal was below detection levels at other Tcr and Igh loci (Fig. 7a, b; Supplementary Fig. 5), indicating some degree of variability in the incidence and/or detection of TOP2B-mediated DSBs at these regions.Fig. 7TOP2B is enriched at V(D)J-active regions colocalizing with RAG and endogenous DSB accumulation.a–c Genome browser view of TOP2B, RAD21, RAG1, RAG2 and ENDseq signal tracks in wild-type, Atm−/− and Rag2−/− primary thymocytes, as indicated, at IgH (a), Tcra/d (b) and Tcrb (c) loci. TOP2B enriched regions are highlighted in grey. Regions of V, D or J segments are indicated. ENDseq signal in Rag2−/− thymocytes is additionally shown with a smaller scale to appreciate sites of minor DSB accumulation (yellow, auto). d, e Endogenous accumulation of TOP2Bccs and TOP2-mediated DSBs, as measured with ICE-IP, at Tcrb (d) and Pten (e) loci in Tdp2+/+
Atm+/+, Tdp2−/−
Atm+/+, Tdp2+/+
Atm−/− and Tdp2−/−
Atm−/− freshly isolated thymocytes from three independent mice (n = 3). Regions 1 and 2 for Tcrb and Pten are indicated in Fig. 7c and 5c, respectively. Mean ± SEM and statistical significance by one-way ANOVA (non-parametric) with Dunn’s post-test is shown (F = 9 for region 1 of Pten and F = 5,8 and F = 7 for region 1 and region 2 of Tcrb).Tdp2−/−Atm−/− thymocytes accumulate endogenous TOP2 DSBsTo directly confirm aberrant TOP2B activity as a relevant source of DSB accumulation in Tdp2−/−
Atm−/− thymocytes, we used a technique recently developed in the laboratory, ICE-IP, in which TOP2 covalently attached to DNA is immunoprecipitated, and the abundance of the genomic region of interest is measured by qPCR (see the “Methods” section and Supplementary Fig. 7 for a detailed description). This allows locus-specific detection of TOP2ccs and TOP2-induced DSBs in which TOP2 has not been fully degraded or removed. In order to be able to detect endogenous lesions, we included a step of nested amplification. We first concentrated on two regions of Tcrb in which RAG-independent ENDseq peaks were detected (region 1 and region 2 in Fig. 7c). Interestingly, TOP2B ICE-IP in freshly isolated thymocytes showed a very clear enrichment of covalently attached TOP2B, exclusively, or at least only significantly, in Tdp2−/−Atm−/− thymocytes, with a >10-fold increase compared to wild-type animals (Fig. 7d). We then tested two regions of Pten with Atm−/−-specific ENDseq signal, one that coincided with a strong TOP2B signal and one that did not (region 1 and region 2 in Fig. 5c, respectively). As expected, covalently attached TOP2B accumulated to significant levels only in region 1 of Pten in Tdp2−/−Atm−/− thymocytes, and neither in region 2, nor in wild-type or single-mutant cells. These results strongly support the idea that TDP2 and ATM independently operate to prevent the endogenous accumulation of TOP2B-blocked DSBs, both globally at particular regions throughout the genome, and specifically at sites undergoing V(D)J recombination.Finally, in order to reinforce the idea that the rearrangements found in thymic lymphomas could be related to 3D-genome folding and the functions of TOP2B in this process, available Hi–C maps on wild-type thymocytes 41 were integrated with hotspots of Atm−/−-linked genomic instability. As a matter of fact, V(D)J recombination sites and other unstable sites were clearly associated with regions of strong long-range interactions (in the order of megabases), apparent as borders of squares in the contact density map (Supplementary Fig. 8), and that could be associated to topologically associating domains (TADs) borders. Interestingly, these regions have been shown to be particularly prone to breakage and instability in general 42,43, and to the accumulation of etoposide-induced DSBs in particular37, which has led to the idea of TOP2 activity at these functional elements of genome organization as a potential major driver of oncogenic translocations28,29.DiscussionIn this study, we show that the absence of TDP2 function significantly aggravates the predisposition of Atm−/− mice to develop thymic malignancies (Fig. 2). Furthermore, TOP2B genome-wide binding correlates with sites of endogenous DSB accumulation (Fig. 5), and endogenous TOP2B-blocked lesions indeed accumulate in Tdp2−/−Atm−/− thymocytes (Fig. 7). These observations strongly support TOP2B-mediated DSBs as potential drivers of tumour development, at least in the particular setting of A-T-linked cancer predisposition, establishing a causal connection between misrepair of endogenous TOP2 lesions and tumorigenesis. In this sense, the implications of these results reach beyond our understanding of A-T cancer predisposition, since ATM is a well-established tumour suppressor, very frequently mutated in in many tumour types, and lymphoid malignancies in particular44.Interestingly, thymic tumours in Tdp2−/−
Atm−/− mice display identical characteristics to those of Atm−/− animals, with a very aggressive form of T-ALL associated to recurrent clonal genome rearrangements enriched at loci undergoing V(D)J recombination (Figs. 2–3; Supplementary Fig. 2–3 and Supplementary Table 1). This is, in principle, difficult to reconcile with the incidence of thymic malignancies being increased by loss of TDP2, an enzyme that is highly specialized in TOP2-mediated DSBs, and therefore completely unnecessary for the repair of hairpin DNA ends produced by RAG. Nevertheless, one must bear in mind that translocations require two DSBs, so that TOP2-mediated lesions can contribute to the oncogenic events by providing partners for rearrangements that also engage Tcr or Ig loci, being the co-occurrence of TOP2- and RAG-mediated DNA breaks, and their stabilization and misrepair upon ATM loss, and not RAG-induced breaks per se, what drives ATM-deficient thymic lymphoma predisposition. This would occur in two different scenarios (Fig. 8).Fig. 8Model to explain aberrant TOP2 activity as a driver of ATM-deficient thymic malignancies.TOP2 activity associated with genome organization and transcription throughout the genome can accidentally result in DSBs (top). Additionally, DSBs can arise as a consequence of TOP2 activity derived from genomic reorganization associated to V(D)J recombination, and in particular, the mechanism of RAG scanning to find its RSS targets (bottom). These DSBs concur with RAG-mediated DSBs, increasing the probability of oncogenic translocations to occur. Efficient repair of TOP2-induced DSBs mediated by either TDP2 or ATM-dependent pathways, together with the checkpoint and apoptotic functions of ATM, strongly limit the oncogenic potential of these lesions.First, we consider TOP2-induced DSBs appearing independently of V(D)J recombination (Fig. 8, top). In this situation, aberrant TOP2 activity, which accidentally occurs throughout the genome as a consequence of genome organization or the removal of transcription-associated supercoiling27,37, results in the formation of DSBs that are normally repaired very efficiently by the action of either TDP2- or ATM-dependent repair pathways33. Furthermore, the checkpoint functions of ATM trigger cell cycle arrest and/or apoptosis if these breaks remain unresolved. A combined failure to properly repair and signal TOP2-mediated DSBs, which can occur accidentally, but is greatly enhanced in the Tdp2−/−
Atm−/− mutant, leads to the persistence and propagation of these lesions, as has been shown to occur with RAG-induced DSBs upon single ATM deficiency15. This substantially increases the probability of at least two DSBs concurring in time during thymocyte development, providing ideal conditions for the generation of chromosomal translocations that can drive oncogenesis. As a matter of fact, irradiation of Atm−/− B-cells has been shown to increase translocations involving the Igh locus15, strongly supporting the idea that additional sources of DSBs can enhance translocation frequency by providing partners for rearrangements with loci undergoing V(D)J recombination. Nevertheless, it is clear that, as previously suggested, off-target RAG cleavage can also constitute a source of additional DSBs during thymocyte development40. In this context, the striking genome-wide correlation between TOP2B and RAG binding (Fig. 6) complicates to unambiguously establish their individual contribution to endogenous DNA breakage. The fact of ENDseq signal being quantitatively more relevant at TOP2B than at RAG2 peaks, however, strongly vouches for aberrant TOP2 activity as a major source of endogenous DSB occurrence genome wide.Second, we consider a scenario in which TOP2-induced DSBs appear associated to V(D)J recombination (Fig. 8, bottom). The chromatin movements and changes in the three-dimensional conformation that necessarily accompany the process of V(D)J recombination result in topological problems that lead to particularly elevated levels of TOP2 activity at these regions, and therefore an increased probability of DSBs to occur. These TOP2-mediated DSBs concur both in time and space with RAG cleavage, constituting a particularly challenging and genome-threatening situation. This possibility is particularly appealing in the context of the recent RAG-scanning models, which invoke cohesin-mediated loop extrusion to explain how RAG finds and pairs RSSs23. Once RAG is loaded to form a V(D)J recombination centre, chromatin threading through the cohesin ring would allow RAG to linearly scan the extruded region and find a second RSS for pairing and cleavage. As has been proposed for general loop extrusion, TOP2 would facilitate RAG scanning by removing topological problems that are encountered by the extruding complex, but, as mentioned above, could result in the formation of DSBs that compromise the integrity of these regions37. Thus, the strong thymocyte-specific co-localization of TOP2B-cohesin with RAG at sites actively undergoing V(D)J recombination does not only provide us with a potential additional source of instability at these sites, but can actually be taken as an important support for a loop-extrusion model of RAG scanning.In summary, we provide a causal link between endogenously occurring TOP2-induced DSBs and cancer development, putting forward these lesions as important contributors to T-cell malignancies related to ATM loss, and opening the door for their potential involvement in other conditions and cancer types. Furthermore, we present evidence for the involvement of TOP2B in resolving topological problems derived from the chromatin movements and changes in genome conformation required for V(D)J recombination, providing an additional source of chromosome fragility at these regions beyond direct RAG cleavage, and supporting a model of loop extrusion-mediated RAG scanning for target RSS pairing.MethodsAnimal maintenanceAll animal procedures were performed in accordance with European Union legislation and with the approval of the Ethical Committee for Animal Experimentation of CABIMER and validation by the Regional Government of Andalusia (Consejería de Agricultura, Ganadería, Pesca y Desarrollo Sostenible de la Junta de Andalucía). Double heterozygotes Tdp2+/−Atm+/− and Tdp2+/−Trp53+/− were obtained by the crossing Tdp2+/−32 with Atm+/−4 or Trp53+/−45. The colonies were maintained by crossing double-heterozygotes and littermates were used for experiments. Animals were housed in isolated cages with controlled ventilation trough HEPA-filters, kept under standard housing conditions (21 ± 1 °C with a photoperiod of 12:12 h), and manipulated in flow cabins. Sterile food pellets and water were available ad libitum. Mice were genotyped using Phire Animal Tissue Direct PCR Kit (Thermo) following manufacturer instructions.In vivo etoposide sensitivityAt 8 weeks of age, mice underwent intraperitoneal injection with 3 µl/g of body weight of either DMSO (vehicle control) or etoposide at 25 mg/ml in DMSO for a final dose of 25 mg/kg. Weight and general health status were monitored daily from the day of injection (inclusive). Six days post-treatment mice were sacrificed by cervical dislocation and dissected for histopathological analysis. For this, organs were fixed in 4% paraformaldehyde for 2 days, embedded in paraffin, cut in 6 µm slices by microtome, stained with Haematoxylin-Eosin and visualized under the microscope.Lifespan analysisMinimum 20 mice per experimental condition were included in the analysis. Weight and general health status were monitored weekly. Animals were sacrificed by cervical dislocation and dissected for histopathological analysis in the case of showing a 20% loss of the maximum weight, the presence of a detectable tumour or signs of evident pain. Thymic lymphoma was macroscopically identified and confirmed by histopathological analysis. Overall survival and cumulative occurrence of thymic tumours were determined by Kaplan–Meier curves and statistically analysed with Wilcoxon tests.Histological analysisCerebella were fixed in 4% paraformaldehyde, embedded in paraffin, and sagittal 50 µm slices were obtained by vibratome. Immunohistochemistry was performed using anti-calbindin primary antibody (CB-38a, Swant) and Biotin-SP AffiniPure Goat Anti-Rabbit secondary antibody (111-065-003, Jackson). Signal was developed using VECTASTAIN Elite ABC HRP Kit (Pk-6100, Vector Laboratories) and DAB (D4418-50SET, Sigma).For histological analysis of testicles, organs were fixed in 4% paraformaldehyde, embedded in paraffin, cut in 6 µm slices by microtome and stained with Haematoxylin-Eosin.Cellular index assayPrimary MEFs were isolated from littermate embryos at day 13 p.c. and cultured at 37 °C, 5% CO2 and 3% O2 in Dubelcco's Modified Eagle's Medium (DMEM) supplemented with penicillin, streptomycin, 15% FCS and non-essential aminoacids. MEFs were plated in E-plates 16 in duplicates in a concentration of 4000 cells per well and analysed by xCELLigence® RTCA DP (ACEA Bioscences). The instrument measures cellular proliferation by changes in impedance-based signals in real time. Three independent sets of primary MEFs cells were analysed.Lymphocyte analysisHealthy thymus (4-weeks old) or thymic lymphomas were extracted from mice, disaggregated to single cells in EDTA-buffer (100 mM NaCl, 1 mM KH2PO4, 3 mM KCl, 10 mM Na2HPO4, 1 mM EDTA) and immunostained with anti-CD4-FITC and anti-CD8-APC. Samples were fixed 10 min in 4% paraformaldehyde for storage at 4 °C, and subsequently analysed using a BD FACSCalibur Flow Cytometer (BD Biosciences). Data was processed and analysed using FlowJo (v9, FlowJo, LLC).FACS analysisB-cell populations were isolated from 5-week old mice femur passing 2 ml of ice-cold EDTA buffer (100 mM NaCl, 1 mM KH2PO4, 3 mM KCl, 10 mM Na2HPO4, 1 mM EDTA) through the femur bone marrow using a 25-G needle syringe. Cells were centrifuged 10 seg at 10,000 rpm in a bench centrifuge and resuspended in 500 µl of PBS-5% FBS (blocking, 30 min). After blocking, 1:100 dilution of conjugated antibodies was added to the samples and incubated on ice for another 30 min. Cells were then washed twice with PBS-5%FBS (centrifuge 10 seg, 10000 rpm) and fixed adding 200 µl of 4%PFA-PBS drop by drop to cells resuspended in 20 µl of PBS. After 10 min of incubation, cells were washed with PBS and analysed by FACScalibur flow cytometer (Becton Dickinson). Antibodies: B220-APC (17–0452) and CD43-FITC (11–0431). CD43+ B220+ cells were considered Pro-B-cells, CD43- B220low Pre-B-cells and CD43- and B220high immature B-cells. Data was processed and analysed using FlowJo (v9, FlowJo, LLC).CGH analysisGenomic DNA was purified from thymic tumours and kidney from Tdp2+/+Atm−/− or Tdp2−/−Atm−/− mice using DNeasy Blood and Tissue Kit (Qiagen) following manufacturer recommendations. One microgram of genomic DNA from thymic lymphoma was profiled against 1 µg of matched normal kidney DNA from the same mouse. Processing, labelling and hybridization to Mouse Genome CGH Microarray 2 × 105 K (G4425B-014699, Agilent) was performed at the CBIMER Genomics Facility. Data was processed using the package snapCGH within R (version 3.3.1). Background correction was applied to fluorescence ratios of scanned images using the method ‘minimum’. To compare hybridizations, normalization between arrays by the method ‘scale’ was performed. Finally, data was processed and ordered by the function ‘processCGH’, and fluorescence ratios were plotted using the R function ‘plot’ and the package Gviz46. Thresholds for copy number alterations were set at log2 = ±0.6 for trisomy or hemizygous deletion.ImmunofluorescenceDSBs in primary MEFs were measured by 53BP1 foci. MEFs were grown on coverslips for 1 day in DMEM supplemented with penicillin, streptomycin, 15% FCS and non-essential aminoacids (37 °C, 5% CO2 and 3% O2) and then fixed for 10 min in ice-cold methanol at −20 °C. Cells were permeabilized for 2 min in PBS−0,2% Triton X-100 and blocked for 30 min in PBS−5% BSA. Coverslips were then incubated with the primary antibody (SantaCruz – sc22760) diluted 1:1000 in PBS−1% BSA for 1 h. After three washes of PBS−0,1% Tween 20, cells were incubated 30 min with the AlexaFluor-conjugated secondary antibody diluted 1:1000 in PBS−1% BSA (Jackson ImmunoResearch − 111-545-144) and washed three times with PBS−0,1% Tween 20. Finally, cells were counterstained with DAPI (Sigma) and mounted in Vectashield (Vector Labs). 53bp1 foci were counted manually (double-blind) in 40 cells per condition using ZEISS ApoTome microscope.ICE-IP (in vivo complex of enzymes immunoprecipitation)Cells from thymus medulla were isolated from 5-week old mice and placed into RPMI 20% FBS media for 3 h in order to collect only thymocytes in suspension. Cells were then collected by centrifugation at 300 × g 5 min and resuspended into 1% (w/v) N-Lauroylsarcosine sodium salt (Sigma-Aldrich, L7414) in TE buffer with complete protease inhibitor cocktail (Roche) for a denaturizing lysis of cells. Lysates were then homogenized using 25G syringe and subjected to DNA precipitation using a CsCl (Applichem-Panreac, A1098) density gradient. DNA was precipitated only with covalently bound proteins adding CsCl to a final concentration of 0,67 g/ml and centrifuging at 57,000 rpm for 20 h at 25 °C using 3.3 ml 13 × 33 polyallomer Optiseal tubes (Beckman Coulter) in a TLN100 rotor (Beckman Coulter).Forty micrograms of non-crosslinked DNA were digested for 6 h with HindIII-HF (NEB, R3104) and NdeI (NEB, R0111) restriction enzymes at 1 U/μl each one. Samples were then diluted five times in IP buffer (0,1% SDS, 1% TX-100, 2 mM EDTA, 20 mM TrisHCl pH8, 150 mM NaCl) and incubated o/n at 4 °C with 2 μg of TOP2B antibody (NOVUS, NB100-40842) and then with 25 μL of pre-blocked (1 mg/ml BSA) Dynabeads protein A and Dynabeads protein G (ThermoFisher). Beads were then sequentially washed with IP buffer, IP buffer containing 500 mM NaCl and LiCl buffer (0.25 M LiCl, 1% NP40, 1% NaDoc, 20 mM TrisHCl pH8 and 1 mM EDTA). DNA was eluted by 30 min incubation at 30 °C in 100 μL elution buffer (1% SDS, 100 mM NaHCO3) and then treated with 10 μg of Proteinase K (ThermoFisher) for 2 h at 37 °C prior to purification using Sera-Mag Select beads (GE Healthcare). Sites of interest were amplified using GO-TAQ polymerase and purified again using Sera-Mag Select beads before qPCR with nested primers. In order to avoid PCR saturation, the number of cycles (between 10 and 20) in the first PCR reaction were determined for each pair of primers to results in a range of 25–30 Cqs in the subsequent qPCR. Results for each region of interest were normalized against input and an intergenic control region which does not show significant TOP2B binding.Primers for PCR: Tcrb-Region1 (FW: GGAGACCCAGAACAGAGCAG, RV: ATAAATAGGGCTGGGGATGG), Tcrb-Region2 (FW: CACCTGCCATAGCTCCATCT, RV: CGGTGATAGCTAGAGGCTGAG), Pten-Region 1(FW: ACTGGCAAGCCAAGCTTAAA, RV: GTGGTTGGTTTCCTGCAGTT), Pten-Region 2 (FW: AACTCCGCTGTGAATTTTGG, RV: CTTTGGGAGGACATGCTAGG) and Control Region (FW: AGGAGAGAATGGAGACAAGAGC, RV: GGTCTCTATCACTGTTCTCATTGG).Nested primers for qPCR: Tcrb-Region1 (FW: GGAGACCCAGAACAGAGCAG, RV: ATAAATAGGGCTGGGGATGG), Tcrb-Region2 (FW: AGCTCCATCTCCAGGAGTCA, RV: TGAGGTAGAAAGGGCTGCAT), Pten-Region1 (FW: CCCCTCCCCACTTCTATTGT, RV: GTGGTTGGTTTCCTGCAGTT), Pten-Region2 (FW: CCGCTGTGAATTTTGGCTAT, RV: GCCTGTGAAACAGTGCTCAA) and Control Region (FW: TCACTGTTCTCATTGGTTGC, RV: CTCAGGAGTGTCAGGGAAGG).ChIPseqTo prepare chromatin, mouse thymus was extracted from 4–6-week old mice, washed twice with cold PBS and mechanically disaggregated into single cells in 1ml EDTA buffer (100 mM NaCl, 1 mM KH2PO4, 3 mM KCl, 10 mM Na2HPO4, 1 mM EDTA). Thymocytes were then washed twice and resuspended again in EDTA buffer. Cells were fixed in a final concentration of 1% formaldehyde and incubated at 37 °C for 10 min. Fixation was quenched by adding glycine to a final concentration of 125 mM. Cells were washed twice with cold PBS in the presence of complete protease inhibitor cocktail (Roche) and PMSF. Cell pellet was lysed in two steps using 0.5% NP-40 buffer for nucleus isolation and SDS 1% lysis buffer for nuclear lysis. Sonication was performed using Bioruptor (Diagenode, UCD-200) at high intensity and two cycles of 10 min (30” sonication, 30” pause) and chromatin was clarified by centrifugation (17,000 × g, 10 min, 4 °C).For IP, 50 μg chromatin and 4 μg antibody (anti-TOP2B NOVUS-NB100–40842, anti-TOP2B SantaCruz-sc13059, anti-TOP2A Abcam-EP1102Y and rabbit IgG SIGMA-I8140 as control) were incubated o/n in IP buffer at 4 °C, and then with 25 μL of pre-blocked (1 mg/ml BSA) Dynabeads protein A and Dynabeads protein G (ThermoFisher). Beads were then sequentially washed with IP buffer, IP buffer containing 500 mM NaCl and LiCl buffer. ChIPmentation was carried out essentially as previously described using Tagment DNA Enzyme provided by the Proteomic Service of CABD (Centro Andaluz de Biología del Desarrollo). DNA was eluted by incubation at 50 °C in 100 μL elution buffer (1% SDS, 100 mM NaHCO3), cross-linking reverted by incubation with 200 mM NaCl and 10 μg of Proteinase K (ThermoFisher) o/n at 65 °C and DNA purified using Qiagen PCR Purification columns. Libraries were amplified for N-1 cycles (being N the optimum Cq determined by qPCR reaction) using NEBNext High-Fidelity Polymerase (M0541, New England Biolabs), purified using Sera-Mag Select Beads (GE Healthcare) and sequenced using Illumina NextSeq 500 and single-end configuration.NGS analysisTags were aligned to mouse genome (mm9) using Bowtie (version 1.2.0). Peaks were called using HOMER and filtered by a fold change enrichment over control (IgG ChIPseq) of 10-fold and standard quality parameters for “factor-style” peak calling. When comparing peak colocalization, TOP2B called peaks were extended 1 kb in each direction. For genome tracks, we used bamCoverage (deepTools) to convert aligned reads to signal tracks (bigwig) using RPKM normalization. Visualization was done by UCSC browser and profiles and heatmaps were generated by Seqplots applying k-means algorithm for clustering. Two experiments with two different antibodies against TOP2B were performed in the study and one for TOP2A isoform. In order to increase the confidence of TOP2B peaks, TOP2B peak dataset was based on both experiments and defined as the common peaks called in the two replicates. For genome browser tracks, TOP2B ChIPseq performed using NOVUS Biologicals antibody (NB100-40842) was used.Genome annotation was obtained from UCSC classification. V, D, J segment annotations were obtained from Ji et al.39. All comparisons are performed with datasets derived from thymocytes except RAD21 data, which is derived specifically from CD4/CD8 DP thymocytes. Since DP is the most abundant (>90%) population of thymocytes, we consider this dataset valid for comparison with experiments performed in total thymocytes. RAG1 and RAG2 ChIPseq in thymocytes were obtained from Teng et al. 2015 (SRA: PRJNA285688)40, thymocyte ENDSeq datasets from Canela et al. 2016 (SRA: PRJNA326246)38, TOP2B ChIPseq in MEFs from Canela et al. (SRA: PRJNA387544)37 and RAD21 ChIPseq in CD4+CD8+ DP thymocytes from Loguercio et al. (SRA: PRJNA432324)47. POLR2A, H3K4me3, H3K27ac and CTCF thymocyte datasets were obtained from ENCODE (ENCSR000CEA, ENCSR000CCJ, ENCSR000CCH and ENCSR000CDZ respectively). All the raw reads from the mentioned available ChIPseq and ENDseq experiments were processed following the pipeline described above.Hi–C analysisHi–C data was obtained from Falk et al.41 (GEO accession number: GSE111032) and available contact matrixes displayed using HiGlass web-tool (http://higlass.io).Software and algorithmsUCSC Genome Browser – Kent et al. 2002. https://genome.ucsc.eduHiGlass - Kerpedjiev et al. 2018. https://higlass.io/Bowtie 1.2.0. – Langmead et al. 2009. https://sourceforge.net/projects/bowtie-bio/files/bowtieMACS – Zhang et al. 2008. https://pypi.python.org/pypi/MACSHOMER – Heinz et al. 2010. https://homer.ucsd.edu/homerSAMTOOLS-1.1 – Li et al. 2009. https://github.com/samtools/samtoolsdeepTools-2.4.1. – Ramirez et al. 2016. https://deeptools.readthedocs.ioR – R Development Core Team, 2008. https://www.r-projects.orgSeqPlots – Stempor et al. 2016. https://github.com/Przemol/seqplotsGraphPad Prism v6. for statistical analysis. https://graphpad.comFastQ Toolkit - http://hannonlab.cshl.edu/fastx_toolkit/Trimmomatic - Bolger et al. 2014. 10.1093/bioinformatics/btu170BD CellQuest pro and FlowJo v9. for FACS analysis.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary | nature communications | [
"Article"
] | [
"Cancer",
"Lymphoma",
"Genomic instability",
"DNA damage response",
"Double-strand DNA breaks"
] | IntroductionAtaxia Telangiectasia (A-T) human inherited disease related to DNA damage response (DDR caused by loss-function mutations ATM gene associated with multi-systemic features affecting brain gonads immune system main symptom progressive early-onset cerebellar ataxia chromosomal instability hypersensitivity to DNA double-strand breaks (DSBs distinguish other spinocerebellar DSBs underlying A-T symptomatology open debate molecular trigger-T contribution DSBs connection of DSBs with immunodeficiency cancer predisposition in-T established in Atm−/− mouse4 During lymphocyte development T-cell receptor (TCR) Immunoglobulin (Ig) loci randomly rearranged in V D J coding segments germline5 shuffling occurs through generation of DSBs by RAG1-RAG2 repair by non end-joining (NHEJ) machinery ATM facilitates V(D)J recombination molecular explanation for mild immunodeficiency A-T patients Atm−/− mice main role ATM in RAG post-cleavage complexes DNA ends use ligation6,7 other functions relevant in NHEJDefects in V(D)J recombination immunological problems A-T cancer predisposition one third patients develop cancer lymphoma leukemia aggressive thymic neoplasias T-cell acute lymphoblastic leukemia A-T Atm−/− T-cell malignancies linked to genome rearrangements TCR loci aberrant V(D)J recombination prevalent A-T TCRα/δ locus frequent in mice translocation molecular link between ATM-deficient T-cell malignancies human T-cell malignancies/− mice share chromosome 15 duplication Notch1 amplification Pten deficient V(D)J recombination defects ATM loss lead to persistent DSBs oncogenic Aberrant V(D)J recombination unlikely single driver oncogenic translocations V(D)J-unrelated regions of instability persistent cancer predisposition RAG deficiency16 suggest additional DSBs ATM-deficient oncogenic translocations T-cell cancer predisposition aberrant action DNA topoisomerase II (TOP2) chromosomal TOP2 topological problems DNAcleaves strands DNA gate passage another DNA segment catalytic intermediate cleavage complex (TOP2cc), enzyme linked to 5′ termini break transient DNA gate resealed after passage TOP2ccs result in irreversible TOP2-induced DSBs conflict with cellular processes processing proteasome20 errors in TOP2 catalytic cycles underlies clinical efficacy of chemotherapeutic agents TOP2 “poisons”, etoposide example21 efficacy treatment with TOP2 poisons linked to secondary haematological malignancies transcription loop extrusion genome chromatin through cohesin main source of DSBs chromosomal translocations by etoposide treatment26–29 incidence impact of endogenous TOP2-mediated lesions treatment TOP2 poisons relationship cancer development established compounds in diet environment pre-existing DNA damage nicks abasic sites can poison TOP2 activity18 TOP2-induced DSBs peptide blocks enzyme bound to 5′ termini through phosphotyrosyl bond TDP2 5′-tyrosyl-DNA-phosphodiesterese activity30 only mammalian enzyme unblock TOP2-induced DSBs for repair32ATM repair TOP2-induced DSBs promoting alternative nucleolytic pathways removal peptide TDP2 ATM define pathways repair TOP2-induced DSBs combined absence consequences cellular response present study Atm−/− Tdp2−/− double-deficient mice occurrence relevance TOP2-mediated DSBs vivo contribution A-T symptomatology TDP2 loss aggravates predisposition Atm−/− mice thymic malignancies colocalization between RAG TOP2 regions Atm−/− Tdp2−/− thymocytes results suggest co-occurrence misrepair RAG- TOP2-mediated DSBs thymic malignancies linked ATM deficiency/−Atm−/− Atm−/− mice phenotypically repair Tdp2− Atm−/− double-deficient cells to TOP2-induced DSBs33 address physiological impact lesions entire organism TOP2 damage deleterious effects Atm−/− mouse aggravated by Tdp2 loss Mice from Tdp2 Atm crosses born at expected Mendelian proportions gross abnormalities Atm loss causes growth measured body weight male mice at 4 weeks ageAtm−/− mice showed 20% reduction size compared wild-type Deletion of Tdp2 aggravate growth retardation neither Tdp2−/− Atm−/− mice displayed neurological deficiencies ataxic behaviour confirmed cerebellar integrity Purkinje cellular density in 8-week old mice Tdp2−/− mutation modify disruption spermatogenesis Tdp2−/−Atm−/− mouse embryonic fibroblasts showed reduction proliferation single-mutant wild-type cells accumulated DSBs in unchallenged growth conditions 60% Tdp2− cells harbouring more than one 53BP1 focus 25% wild-type results suggest accumulation of endogenous DSBs TDP2 ATM disrupt cellular fitness in vivo incidence lesions high compromise tissue homeostasis development early stages life 1Tdp2−/−Atm−/− causes accumulation DSBs etoposide hypersensitivity in miceEndogenous DSB occurrence in Tdp2+ MEFs measured by 53BP1 foci formation percentage cells focus images 53BP1 foci DAPI staining each genotype MEFs from three embryos analysed Mean ± SEM statistical significance one-way ANOVA (F = 5,402) Body weight 8-week old mice injected single dose etoposide (25 mg/kg). Average SEM initial body weight statistical significance ANOVA (F = 29,11) Tdp2+ indicates Tdp2+/+ Tdp2+/− genotypes images Haematoxylin-Eosin stained jejunum slices 6 days after etoposide exposure.Tdp2−/− mice hypersensitive to TOP2-induced Tdp2−/− mice increased load TOP2-induced lesions 8-week old mice single injection TOP2-poison etoposide (25 mg/kg). Tdp2− mice severe hypersensitivity etoposide progressive weight loss 15% decrease after 6 daysHistopathological analysis revealed villous atrophy small intestine mucosa etoposide target likely cause weight loss in Tdp2−/− Atm−/− (Fig. low etoposide concentration wild-type Atm−/− animals respond negatively treatment Tdp2−/− mildly hypersensitivity Tdp2−/− Atm−/− mice to etoposide suggests protective role ATM against effects TOP2 lesions when TDP2 absent loss increases thymic cancer predisposition Atm− Tdp2−/− Atm−/− mice longer period TOP2-mediated damage weight health status monitored weekly 20 mice 2 years (730 days) (Fig. Atm−/− mice showed reduced life-span animals succumbing thymic neoplasia after first months 2b molecular characterization showed majority (70%) double positive CD4+ CD8+ T-cell tumours. 2c 2TDP2 deficiency increases thymic cancer predisposition Atm−/− not Tpr53−/− miceKaplan–Meier survival curve occurrence thymic tumours 1st year 20 mice Tdp2 Atm genotypes 21 (Tdp2+/ 23 (Tdp2−/− Tdp2+ indicates Tdp2+/+ Tdp2+/− genotypes significance two Wilcoxon test Percentage tumours CD4+ CD8+ double positive/CD8+ single positive lymphocytes genotypes Thymocytes selected size complexity high low CD4 CD8 markers Fig. 1a Median life-span Tdp2+/+ Atm−/− Tdp2−/− Atm−/− mice thymic tumour (n = 11 16 Tdp2+ Kaplan–Meier survival curve incidence thymic tumours Tdp2 Trp53 genotypes = 16 Tdp2+/+ 22 Tdp2− Tdp2−/− mice lifespan comparable wild-type none developed thymic lymphoma experiment (2 Atm−/− background reduced life-span incidence thymic tumours aggravated Tdp2 inactivationTdp2−/−Atm−/− mice median survival 140 days 307 days/− single mutant probability thymic tumour 1st year increased 43% Atm−/− to 72% Tdp2− double mutant lifespan Tdp2−/−Atm− mice thymic tumour not different direct effect incidence not latency aggressiveness T-cell malignancies tumours Tdp2−/−Atm−/−mice predominantly formed by double positive CD4+ CD8+ T-cell precursors (80%) Double-knockout mice differential distribution double negative positive single positives thymocytes Atm−/− mice increased incidence T-cell malignancy TDP2 loss unlikely disruption V(D)J recombination pre-B-cell pro-B-cell B-cell populations differ between single Atm−/− double Tdp2−/−Atm−/− knock-out mice V(D)J recombination not altered by TDP2 loss checkpoint proapoptotic functions ATM p53-dependent for Atm−/− cancer predisposition15p53-deficient Trp53−/− mice susceptible to double positive CD4+CD8+ thymic tumours34 contribution repair checkpoint/apoptotic functions ATM increased incidence thymic tumours Tdp2−/−Atm−/− analysed life span tumour incidence in Tdp2−/−Trp53−/− animals repair component not affected loss of functional TDP2 decrease lifespan increase incidence cancer thymic lymphoma/− Tdp2−/−Atm−/− tumours reflect functions ATM repair TOP2-induced DSBs not results consistent with model misrepair of TOP2-induced DSBs thymic tumours ATM deficiency.Tdp2−/−Atm−/− Atm−/− malignancies molecularly thymic tumour development analysed compared copy-number variation by comparative genomic hybridization) in six Tdp2−/−Atm−/− three Atm−/− thymic tumours (Fig. 3a Atm−/− animalsobserved genomic instability13 amplification upstream Tcra/d locus chromosome 14 (2 3 mice variable hemizygous loss Bcl11b-containing telomeric region chromosome 12 3 frequent t(14;12) translocation breakage-fusion-bridge cycles one mouse instability Tcrb locus gain loss DNA sequence Two three animals duplication chromosome 15. two deletions chromosome 19 Pten tumour suppressor one corresponded homozygous amplification Notch1 chromosome 2 common Atm−/− T-cell malignancies. 3Tdp2−/−Atm−/− thymic malignancies similar genome rearrangements Merged CGH analysis Tdp2+/ thymic lymphomas DNA hybridized analysed kidney DNA mouse Average amplification deletion score tumour/kidney ratio copy number variations −0.66>Log2 tumour/kidney>0.66 location relevant loci Atm−/− thymic tumours indicated Table mice genotype displaying copy number variation locus(Tcra/d): amplification upstream deletion telomeric region chromosome 12 covering Bcl11b Del(Tcrb): deletion Tcrb locus(15) trisomy chromosome 15 Del(Pten): deletion Pten locus(Notch1) duplication Notch1 locus similar features Tdp2−/−Atm−/− mice (Fig. 3a Four six tumours Tcra/d amplification chromosome 12 deletion Bcl11b loss Instability Tcrb locus two cases Trisomy chromosome 15 four one Notch1 amplification events coincident with results Atm−/− animals previous one Tdp2− animal hemizygous Pten loss larger deletion variable length copy-number variation regions additional sites instability unique tumours indicative stochastic oncogenic events loss TDP2 increases oncogenic rearrangements Atm−/− animals contribution TOP2-mediated lesions ATM-deficient T-cell cancer predisposition.TOP2B enriched sites DSB aberrant TOP2 activity Atm−/− oncogenic T-cell genome rearrangements topoisomerase IIß main source TOP2 activity G1 non-cycling cellsChIPseq analysis thymocytes revealed TOP2B distribution consistent cell types. enrichment at promoter enhancer insulator regions (34% 19% 13% TOP2B peaks correlation between TOP2B cohesin subunit RAD21 81% TOP2B peaks overlapping with cohesin. cohesin all TOP2B peaks polymerase II) protein CTCF Fig. 4d shows Cxcr4 locus TOP2B cohesin co-localization promoter insulator regions consistent with connection TOP2B function 3D-genome 4TOP2B with cohesin at promoter enhancer insulator regions thymocytes Genome-wide distribution of 1003 TOP2B ChIPseq peaks genomic features in wild-type thymocytes Chromatin states defined ENCODE H3K4me3 TSS low H3K27ac for promoters H3K27ac H3K4me3 enhancers high CTCF no marks insulator-like regions.TOP2B peak distribution genome global signal profile IgG control genomic feature peaks common between TOP2B ChIPseq experiments antibodies Overlap TOP2B RAD21 peaks in wild-type thymocytesGenome browser view TOP2B RAD21 CTCF POLR2A H3K4me3 H3K27ac signal tracks genomic region active promoter positive POLR2A H3K4me3 H3K27ac insulator negative POLR2A H3K4me3 H3K27ac positive CTCF RAD21) Sites grey major TOP2B signal accumulation Heatmaps TOP2B POLR2A RAD2147 CTCF ChIPseq signals TOP2B peaks K-means clusters divided by black lines C1 TOP2B occupancy Atm−/− oncogenic translocations integrating endogenous DSB accumulation Atm−/− thymocytes Bcl11b Notch1 Pten loci TOP2B binding strong accumulation promoters 3’ end genes pattern cohesin RNA-polymerase II TOP2B accumulation not thymocyte-specific similar pattern in MEFs identify regions DSB accumulation Atm−/− thymocytes selected ENDseq peaks Atm−/− thymocytes (highlighted yellow sites compared with distribution TOP2B overlap variable unique common regions compare region 1 region 2 in PtenTOP2B present at Atm−/− unstable regions difficult association with DSB accumulation distribution regulatory regions difficulty assign original translocation breakpoints limits link TOP2B to oncogenic DSB occurrence 5TOP2B colocalizes with endogenous DSB accumulation in thymocytes Genome browser view TOP2B POLR2A RAD21 ENDseq signal tracks in wild-type Atm−/− primary thymocytes Bcl11b Notch1 Pten loci Atm−/−-specific ENDseq peaks highlighted in yellow regions interest in Pten region 1 2. Global profile ENDseq signal at TOP2B peaks in wild-type Atm−/− Rag2−/− primary thymocytes Control signal profile TOP2B signal ENDseq peaks wild-type primary thymocytes ENDseq peaks genetic conditions distribution TOP2B ENDseq signal genome-wide level observed accumulation of DSBs at TOP2B peaks (Fig. link between TOP2B function spontaneous chromosome fragility ENDseq peaks accumulation TOP2B signalpeaks ENDseq signal in thymocytes associated with RSS sites disappearing in Rag2−/− accumulating in Atm−/− consistent with high incidence RAG cleavage ATM V(D)J recombination TOP2B sites DSBs increased in Rag2−/− Atm−/− thymocytes (Fig. unrelated to RAG cleavage lesions accumulate upon ATM deficiency significant genome-wide correlation between TOP2B function RAG-independent endogenous DSBs thymocytes suggesting aberrant TOP2B activity chromosome breakage Atm−/− oncogenic translocations.TOP2B cohesin colocalize with RAG check RAG TOP2B DSB accumulation analysed TOP2B RAD21 ENDseq signal in peaks RAG1 RAG212 (Fig. RAG2 strong presence at active promoters RAG1 restricted co-localizing with RAG2 observed genome-wide co-localization of TOP2B RAD21 with RAG2 irrespective of RAG1 (Fig. TOP2B accumulation observed in peaks RAG1 RAG2 RAG2 42% of TOP2B peaks colocalized with RAGenrichment H3K4me3-positive promoter regions ENDseq signal observed accumulation DSBs not (Fig. DSB accumulation robust at TOP2B cohesin RAG1 RAG2 peaks wild-type Atm−/− thymocytes (Fig. results uncover genome-wide connection between RAG TOP2B-cohesin TOP2B activity source DNA breakage Atm−/− thymocytes relevant than DSBs induced RAG. 6TOP2B cohesin colocalize with RAG genome-wide RAG240 RAG140 TOP2B RAD21 ENDseq RAG1 RAG2 peaks wild-type Atm−/− primary thymocytes K-means clusters RAG1 RAG2 signal C1 C2. Overlap TOP2B RAG peaks RAG merge RAG1 RAG2 proteins profile TOP2B signal RAG1 RAG2 peaks Control IgG signal profile signal RAG TOP2B-RAD21 peaks wild-type Atm−/−/− primary thymocytes TOP2B RAG check distribution TOP2B sites undergoing V(D)J recombination thymocytes (Fig. 7a–cincludes Tcra/d Tcrb Tcrg IgH15 loci strong TOP2B accumulation observed particularly enriched at V(D)J-initiating J segments TOP2B coincided with cohesin subunit RAD21 peaks RAG1 RAG2. TOP2B-cohesin sites with strong signal enhanced in Atm−/− thymocytes DSB accumulation dependent on RAG coincide with TOP2B-cohesin peaks displaced from TOP2B sites Fig. 6) RAG cleavage main source of DSBs at Tcr Igh loci thymocytes association between TOP2B-cohesin function active V(D)J recombination TOP2B accumulation specific to thymocytes compared to peaks MEFs Fig 4b TOP2A TOP2 enrichment preferential for TOP2B isoform V(D)J recombination sites oncogenic translocations Fig. 4) signal at Tcrb two peaks apparent when signal RAG cleavage lost in Rag2−/− thymocytesENDseq Rag2−/− relocation ENDseq signal RAG increase DSBs TOP2B peaks Rag2−/− background (Figs. 5a regions DSB accumulation coincident with TOP2B RAD21 association TOP2B function RAG-independent DNA breakage TOP2B RAG-independent ENDseq signal below detection levels Tcr Igh loci (Fig. 7a b 5) variability incidence detection TOP2B-mediated DSBs 7TOP2B enriched at V(D)J-active regions RAG endogenous DSB accumulation Genome browser view TOP2B RAD21 RAG1 RAG2 signal tracks wild-type Atm−/− Rag2−/− primary thymocytes IgH Tcra/d Tcrb (c) loci TOP2B enriched regions highlighted grey Regions V, D J segments indicatedENDseq signal Rag2−/− thymocytes shown smaller scale minor DSB accumulation Endogenous accumulation TOP2Bccs TOP2-mediated DSBs measured ICE-IP Tcrb Pten loci Tdp2+/− isolated thymocytes three mice (n = 3) Regions 1 2 Tcrb Pten Fig. 7c 5c Mean ± SEM statistical significance one-way ANOVA Dunn’s post-test (F = 9 region 1 Pten F = F = 7 region 1 2 Tcrb).Tdp2−/−Atm−/− thymocytes accumulate endogenous TOP2 confirm aberrant TOP2B activity DSB accumulation Tdp2− thymocytes ICE-IP TOP2 attached DNA immunoprecipitated abundance genomic region measured qPCR Fig. 7 allows locus-specific detection TOP2ccs TOP2-induced DSBs TOP2 not degraded removed endogenous lesions nested amplification concentrated two regions Tcrb RAG-independent ENDseq peaks detected (region 1 2 FigTOP2B ICE-IP thymocytes showed enrichment attached TOP2B in Tdp2−/−Atm−/− thymocytes >10-fold increase compared to wild-type animals (Fig. tested two regions of Pten with Atm−/−-specific ENDseq signal with strong TOP2B signal (region 1 2 Fig. 5c TOP2B accumulated in region 1 Tdp2−/−Atm−/− thymocytes neither region 2 wild-type single-mutant cells results support TDP2 and ATM prevent accumulation of TOP2B-blocked DSBs sites V(D)J recombination rearrangements in thymic lymphomas related to 3D-genome folding functions TOP2B Hi–C maps on wild-type thymocytes integrated with Atm−/−-linked genomic instability V(D)J recombination sites unstable sites associated with strong long-range interactions borders in contact density map 8) topologically associating domains bordersregions prone to breakage instability accumulation etoposide-induced DSBs TOP2 activity potential driver oncogenic translocations28 of TDP2 function aggravates predisposition Atm−/− mice thymic malignancies TOP2B correlates with endogenous DSB accumulation TOP2B-blocked lesions accumulate in Tdp2−/−Atm−/− thymocytes support TOP2B-mediated DSBs potential drivers tumour development-T-linked cancer causal connection between misrepair endogenous TOP2 lesions tumorigenesis implications beyond ATM tumour suppressor frequently mutated in many tumour types lymphoid malignancies thymic tumours in Tdp2−/− Atm−/− mice display identical characteristics to Atm−/− animals aggressive form T-ALL associated to recurrent clonal genome rearrangements V(D)J recombination (Figs. 2–3 difficult to reconcile with incidence thymic malignancies increased by loss of TDP2 in TOP2-mediated DSBs unnecessary for repair of hairpin DNA ends RAGtranslocations require two DSBs TOP2-mediated lesions contribute to oncogenic events for rearrangements Tcr Ig loci co-occurrence of TOP2- RAG-mediated DNA breaks stabilization misrepair upon ATM loss not RAG-induced breaks drives ATM-deficient thymic lymphoma predisposition two scenarios (Fig. aberrant TOP2 activity ATM-deficient thymic malignancies activity genome organization result in DSBs DSBs TOP2 activity genomic reorganization V(D)J recombination RAG scanning DSBs concur with RAG-mediated DSBs probability oncogenic translocations Efficient repair of TOP2-induced DSBs TDP2 or ATM-dependent pathways checkpoint apoptotic functions ATM limit oncogenic potential TOP2-induced DSBs independently of V(D)J recombination aberrant TOP2 activity removal transcription results in formation DSBs repaired by TDP2- or ATM-dependent repair checkpoint functions ATM trigger cell cycle arrest apoptosis if breaks unresolvedfailure to repair signal TOP2-mediated DSBs enhanced in Tdp2−/− Atm−/− mutant leads to persistence propagation lesions RAG-induced DSBs ATM deficiency15 increases probability of two DSBs concurring during thymocyte development conditions for chromosomal translocations oncogenesis irradiation of Atm−/− B-cells translocations Igh locus15 additional DSBs enhance translocation frequency rearrangements V(D)J recombination off-target RAG cleavage can source of additional DSBs during thymocyte genome-wide correlation between TOP2B RAG (Fig. 6) complicates contribution to endogenous DNA breakage ENDseq signal more relevant at TOP2B than RAG2 peaks for aberrant TOP2 activity major source of endogenous DSB occurrence scenario TOP2-induced DSBs associated to V(D)J recombination (Fig. 8 chromatin movements changes conformation result topological problems to elevated levels TOP2 activity increased probability of DSBs TOP2-mediated DSBs concur with RAG cleavage challenging genome-threatening situationpossibility appealing recent RAG-scanning models invoke cohesin-mediated loop extrusion RAG RAG loaded V(D)J recombination centre chromatin threading through cohesin ring RAG scan extruded region find second RSS for pairing cleavage TOP2 RAG scanning topological problems could result DSBs integrity strong thymocyte co-localization of TOP2B-cohesin with RAG at sites undergoing V(D)J recombination potential additional source instability support for loop-extrusion model RAG scanning causal link between TOP2-induced DSBs cancer development lesions contributors to T-cell malignancies ATM loss potential involvement other conditions cancer types evidence for involvement TOP2B resolving topological problems chromatin movements changes genome conformation for V(D)J recombination additional source chromosome fragility beyond RAG cleavage loop extrusion-mediated RAG scanning for target RSS pairingMethodsAnimal procedures European Union legislation approval Ethical Committee Animal Experimentation CABIMER validation Regional Government Andalusia Agricultura heterozygotes Tdp2 obtained crossing Tdp2+/−32 Atm+/−4 Trp53+/−45 colonies maintained crossing-heterozygotes littermates experiments housed isolated cages controlled ventilation HEPA-filters standard housing conditions (21 ± 1 °C photoperiod 12:12 manipulated flow cabins Sterile food pellets water Mice genotyped Phire Animal Tissue Direct PCR Kit) vivo etoposide 8 weeks mice intraperitoneal injection 3 μl/g weight DMSO etoposide 25 mg/ml final dose 25 mg/kg Weight health status monitored daily Six days post mice sacrificed cervical dislocation dissected histopathological analysis organs fixed 4% paraformaldehyde 2 days paraffin cut 6 μm slices stained Haematoxylin-Eosin visualized under microscope 20 mice per condition Weight health status monitored weekly sacrificed dislocation dissected 20% loss weight detectable tumour painThymic lymphoma identified confirmed histopathological analysis survival occurrence Kaplan–Meier curves analysed Wilcoxon tests analysisCerebella fixed 4% paraformaldehyde paraffin sagittal 50 μm slices vibratome Immunohistochemistry anti-calbindin antibody Biotin-SP AffiniPure Goat Anti-Rabbit antibody Signal VECTASTAIN Elite ABC HRP Kit DAB testicles organs fixed 4% paraformaldehyde paraffin 6 μm slices stained Haematoxylin-Eosin index assayPrimary MEFs isolated embryos day 13 cultured 37 °C 5% CO2 3% O2's Medium supplemented penicillin streptomycin 15% FCS non aminoacids MEFs plated E-plates 4000 cells analysed xCELLigence® RTCA proliferation Three sets MEFs cells analysed analysisHealthy thymus lymphomas extracted disaggregated cells EDTA-buffer NaCl KH2PO4 3 KCl 10 Na2HPO4 EDTA immunostained anti-CD4-FITC anti-CD8-APCSamples fixed 10 min 4% paraformaldehyde 4 °C analysed BD FACSCalibur Flow Cytometer processed analysed FlowJo analysisB-cell populations isolated 5-week mice 2 ml ice-cold EDTA buffer (100 NaCl KH2PO4 3 KCl Na2HPO4 EDTA femur bone marrow 25-G needle syringe Cells centrifuged 10 seg 10,000 resuspended 500 μl PBS-5% FBS 30 1:100 dilution conjugated antibodies incubated ice 30 min Cells washed PBS-5%FBS 200 μl 4%PFA-PBS 20 μl PBS 10 min incubation cells washed PBS analysed FACScalibur flow cytometer Antibodies B220-APC (17–0452) CD43-FITC (11–0431) CD43+ B220+ cells Pro-B-cells Pre-B-cells immature B-cells processed analysed FlowJo analysisGenomic DNA purified thymic tumours kidney mice Blood Tissue Kit microgram DNA thymic lymphoma profiled 1 μg normal kidney DNAProcessing labelling hybridization Mouse Genome CGH Microarray 2 105 K (G4425B-014699 Agilent CBIMER Genomics Facility Data processed snapCGH R correction fluorescence ratios normalization arrays data processed ordered fluorescence ratios plotted R Gviz46 Thresholds copy alterations log2 trisomy hemizygous deletion MEFs measured 53BP1 foci grown coverslips 1 day DMEM penicillin streptomycin 15% FCS non aminoacids °C 5% CO2 3% O2) fixed 10 min-cold methanol −20 °C Cells permeabilized 2 min PBS−0,2% Triton X-100 blocked 30 min PBS−5% BSA incubated primary antibody sc22760 1:1000 PBS−1% BSA 1 h three washes incubated 30 min AlexaFluor secondary antibody 1:1000 washed three PBS−0,1% Tween 20. counterstained DAPI Vectashield 53bp1 foci counted manually-blind 40 cells condition ZEISS ApoTome microscopeimmunoprecipitation thymus medulla isolated 5-week mice RPMI 20% FBS media 3 h thymocytes collected centrifugation 300 × g 5 min resuspended 1% N-Lauroylsarcosine sodium salt buffer protease inhibitor cocktail denaturizing lysis Lysates homogenized 25G syringe DNA precipitation CsCl DNA precipitated proteins CsCl 0,67 g/ml 57,000 rpm 20 h 25 °C 3.3 ml 13 × 33 Optiseal tubes TLN100 rotor micrograms non DNA digested 6 h HindIII-HF NdeI enzymes 1 U/μl Samples diluted five times IP buffer SDS 1% TX-100 2 mM EDTA 20 mM TrisHCl pH8 150 mM NaCl incubated 4 °C 2 μg TOP2B antibody 25 μL pre-blocked Dynabeads protein A G Beads washed IP buffer 500 mM NaCl LiCl bufferM LiCl 1% NP40 1% NaDoc 20 mM TrisHCl 1 mM DNA eluted 30 min incubation 30 °C 100 μL buffer SDS mM NaHCO3) treated 10 μg Proteinase K 2 h 37 °C purification Sera-Mag Select beads amplified GO-TAQ polymerase purified Sera-Mag beads qPCR nested primers PCR saturation cycles 10 20 first PCR 25–30 Cqs qPCR normalized input intergenic control region TOP2B bindingPrimers PCR Tcrb-Region1 CACCTGCCATAGCTCCATCT Pten-Region 1 ACTGGCAAGCCAAGCTTAAA 2 AACTCCGCTGTGAATTTTGG CTTTGGGAGGACATGCTAGG Control Region AGGAGAGAATGGAGACAAGAGC primers qPCR Tcrb-Region1 ATAAATAGGGCTGGGGATGG),-Region2 AGCTCCATCTCCAGGAGTCA Pten-Region1 CCCCTCCCCACTTCTATTGT-Region2 CCGCTGTGAATTTTGGCTAT GCCTGTGAAACAGTGCTCAA Control Region TCACTGTTCTCATTGGTTGC chromatin mouse thymus extracted 4–6-week old mice washed cold PBS disaggregated cells 1ml EDTA buffer (100 mM NaCl 1 KH2PO4 3 KCl 10 Na2HPO4 Thymocytes washed twice resuspended EDTA buffer Cells fixed 1% formaldehyde incubated 37 °C 10 min glycine 125 mMCells washed cold PBS protease inhibitor cocktail PMSF Cell pellet lysed 0.5% NP-40 buffer nucleus SDS 1% lysis buffer nuclear lysis Sonication Bioruptor high intensity two cycles 10 min chromatin clarified centrifugation (17,000 4 50 μg chromatin 4 μg antibody-TOP2B NOVUS-TOP2A-EP1102Y rabbit IgG SIGMA-I8140 incubated IP buffer 4 °C 25 μL pre-blocked Dynabeads protein A protein G Beads washed IP buffer 500 mM NaCl LiCl buffer ChIPmentation Tagment DNA Enzyme Proteomic Service DNA eluted incubation 50 °C 100 μL elution buffer SDS 100 mM NaHCO3) 200 mM NaCl 10 μg Proteinase K 65 °C purified Qiagen PCR Purification columns Libraries amplified N-1 cycles NEBNext High-Fidelity Polymerase purified Sera-Mag Select Beads sequenced Illumina NextSeq 500 single-end configurationanalysisTags aligned mouse genome Bowtie Peaks called HOMER filtered change enrichment ChIPseq 10-fold standard quality parameters peak calling TOP2B peaks extended 1 kb each direction used bamCoverage aligned reads signal tracks RPKM normalization Visualization UCSC browser profiles heatmaps generated Seqplots k-means algorithm clustering Two experiments antibodies against TOP2B one TOP2A isoform confidence TOP2B peaks peak dataset based experiments defined common peaks two replicates genome browser tracks TOP2B ChIPseq Biologicals antibody (NB100-40842) annotation from UCSC classification V D J segment annotations from Ji et al comparisons with datasets thymocytes except RAD21 data CD4/CD8 DP thymocytes abundant>90% dataset valid comparison experiments thymocytes RAG1 RAG2 ChIPseq thymocytes from Teng et al. 2015 thymocyte ENDSeq datasets Canela et al 2016 TOP2B ChIPseq MEFs RAD21 ChIPseq in CD4+CD8+ DP thymocytes LoguercioPOLR2A H3K4me3 H3K27ac CTCF thymocyte datasets from ENCODE (ENCSR000CEA raw reads ChIPseq ENDseq experiments processed pipeline.Hi–C data from Falk et al.41 (GEO accession GSE111032) contact matrixes HiGlass web-tool.io).Software algorithmsUCSC Genome Browser Kent 2002..ucsc Kerpedjiev 2018./Bowtie 1.2.0. Langmead 2009. Zhang 2008. Heinz 2010..ucsd.edu-1.1 Li 2009.-2.4.1. Ramirez 2016. Development Core Team 2008..orgSeqPlots Stempor 2016./seqplotsGraphPad Prism v6. statistical analysis Toolkit.cshl/Trimmomatic Bolger al. 2014. 10.1093/bioinformatics/btu170BD CellQuest pro FlowJo v9. FACS analysis research design Nature Research Reporting Summary.Supplementary information Peer Review File Reporting Summary | 51 | 1.23919 |
10.1038/s41467-020-20225-w | PMC7749178 | Glioblastoma is divided into four subtypes based on molecular profiling at the methylome and transcriptome level. Here the authors perform an integrative analysis of these subtypes resulting in the identification of SOX10 whose loss induces a mesenchymal phenotype and promotes tumour progression. | Glioblastoma frequently exhibits therapy-associated subtype transitions to mesenchymal phenotypes with adverse prognosis. Here, we perform multi-omic profiling of 60 glioblastoma primary tumours and use orthogonal analysis of chromatin and RNA-derived gene regulatory networks to identify 38 subtype master regulators, whose cell population-specific activities we further map in published single-cell RNA sequencing data. These analyses identify the oligodendrocyte precursor marker and chromatin modifier SOX10 as a master regulator in RTK I-subtype tumours. In vitro functional studies demonstrate that SOX10 loss causes a subtype switch analogous to the proneural–mesenchymal transition observed in patients at the transcriptomic, epigenetic and phenotypic levels. SOX10 repression in an in vivo syngeneic graft glioblastoma mouse model results in increased tumour invasion, immune cell infiltration and significantly reduced survival, reminiscent of progressive human glioblastoma. These results identify SOX10 as a bona fide master regulator of the RTK I subtype, with both tumour cell-intrinsic and microenvironmental effects. | IntroductionGlioblastoma is a highly malignant brain cancer with a particularly poor prognosis despite aggressive treatment comprising surgical resection and radiochemotherapy with temozolomide1. Recent large-scale genomics studies2–5 have identified four mRNA expression/DNA-methylation subtypes of glioblastoma and their hallmark genetic lesions: (1) IDH, characterised by glioma CpG island methylation phenotype (G-CIMP) hypermethylation due to mutations in the isocitrate dehydrogenase (IDH) 1 or 2 genes3,6,7; (2) MES (mesenchymal), associated with NF1 aberrations and increased tumour infiltration by tumour-associated macrophages/microglia5; (3) RTK I (receptor tyrosine kinase I), in which tumours commonly have PDGFRA gene amplifications; and (4) RTK II, exhibiting the classical EGFR gene amplification8. The MES, RTK I and RTK II subtypes correspond to the mesenchymal, proneural and classical RNA expression subtypes4, which were recently refined based on the analysis of exclusively IDH wildtype glioblastoma5. Transitions between these subtypes have been observed during the treatment of patients9,10 and may lead to worse prognosis5,9. It remains unclear whether these transitions are due to tumour cell plasticity or expansion of pre-existing resistant subpopulations.Cancer Master Regulators (MRs) are proteins that define and regulate tumour cellular states11. It has been proposed that the systematic identification and characterisation of cancer MRs will provide a better understanding of basic cancer biology and potential therapeutic vulnerabilities12. While recent large-scale studies of transcriptomes and promoter-biased epigenomes have provided valuable insights into glioblastoma heterogeneity and cellular state transitions2,5,9,13–18, a comprehensive, genome-wide survey of the epigenetic landscape of primary glioblastoma subtypes using multidimensional data from the same patient tumour samples is, as yet, not available. Consequently, glioblastoma subtype MRs and their interactions with and effects on GB epigenetics remain largely unknown.Here, we present an integrated epigenetic analysis of the four subtypes of adult glioblastoma. We performed methylome, transcriptome and epigenome profiling on a cohort of 60 untreated patient tumours and show that enhancers vary across subtypes. We identified 10 consensus subtype MRs based on our analysis of these matched tumour epigenetic and gene expression data. Repression of the RTK I MR SOX10 in human glioblastoma cell lines caused a subtype transition to a mesenchymal cellular state via the remodelling of active enhancers. We further show, using a recently described immunocompetent syngeneic mouse model that SOX10 loss leads to a dramatic decrease in survival, increased tumour invasion and immune cell infiltration. These results show that GB subtype transitions can have striking effects on clinically relevant tumour phenotypes and, as such, require further investigation.ResultsPrimary glioblastoma epigenome profilingWe selected 60 adult glioblastoma primary tumours and 4 normal brain samples (Supplementary Data 1, 2) for DNA methylome (methylation microarrays and whole genome bisulphite sequencing (WGBS)) and transcriptome (strand-specific, rRNA-depleted, total RNA sequencing (ssRNA-seq); mean 2 × 108 reads) profiling (see Supplementary Data 3–5 for quality control data). The key resources used in this study are listed in Supplementary Table 1. Tumours were subtyped using a methylation microarray classifier (Supplementary Fig. 1a and Supplementary Data 6)3; the four subtypes (IDH: 12; MES: 19; RTK I: 12; RTK II: 17) and both major genotypes, IDH wildtype, present in about 90% of primary tumours (MES, RTK I, RTK II; 48), and IDH mutated (IDH; 12) were well represented. Subtyping based on WGBS data could clearly identify the IDH and RTK II groups, while MES and RTK I were less distinct. Consistent with this, methylation at gene-based features was generally variable across subtypes; however, the MES and RTK I subgroups could not be differentiated based on TSS or CGI methylation (Supplementary Fig. 1b, c) suggesting that these regions feature comparable methylation, and non-CGI and intergenic regions appear to be similar in the MES and RTK II subtypes (Supplementary Fig. 1c). For a subset of 20 tumours, we also profiled the H3K27ac, H3K4me1, H3K4me3, H3K36me3, H3K27me3 and H3K9me3 histone modifications by chromatin immunoprecipitation and sequencing (ChIP-seq) (Fig. 1a). The mutation status of IDH1 and IDH2 was determined by pyrosequencing. All IDH samples had IDH1 R132H mutations and G-CIMP6 (Supplementary Fig. 1), while remaining tumours were IDH wildtype (IDHwt). These CIMP-, IDHwt subtypes (MES, RTK I and RTK II) exhibited the classic glioblastoma copy number alterations (CNAs) consisting of gain of chromosome 7, loss of chromosome 10 and focal CDKN2A/B deletion. Amplifications of EGFR and gain of chromosomes 19 and 20 were strongly prevalent in RTK II, and PDGFRA, CDK4 and MDM2 or MDM4 amplifications were more frequent in RTK I tumours (Fig. 1b).Fig. 1A comprehensive dataset of glioblastoma subtype epigenomics.a Study design. Based on extensive molecular profiling of 60 glioblastoma tumours, epigenetic glioblastoma subtypes were characterised, subtype Master Regulators were derived based on epigenome and transcriptome data, and functionally validated in glioblastoma cell lines and a syngeneic mouse model. b Characteristics of the 60 glioblastomas used in this study, including age, gender, methylation subtype, IDH1 and IDH2 mutation status and copy number aberrations. c Genome-wide differences in DNA methylation between glioblastoma subtypes and control brain tissue. Mean methylation (WGBS beta-values) for 100 kbp genome bins were determined for each subtype, and the difference to the control brain average. d Mean subtype DNA methylation (WGBS beta) at genomic features. e Mean subtype DNA methylation (WGBS beta) for each state of the 18-state ChromHMM model. f MYT1 RNA-seq expression (log2 TPM + 1) in subtypes, visualised as Tukey boxplots. Boxes correspond to the 25th, 50th/median and 75th percentiles; whiskers denote 1.5× the IQR from the median. Points mark outliers beyond 1.5× IQR. NBr, n = 4; IDH, n = 12; MES, n = 19; RTK I, n = 12; RTK II, n = 17 samples. g Integrative view of the glioblastoma epigenomic landscape. For MYT1, an oligodendrocyte marker gene, per-subtype methylation (top, beta-values) and ChromHMM annotation (bottom) are displayed. Control non-neoplastic, control brain ChromHMM annotations from the Roadmap Epigenome are included as a reference. DNA hypomethylation (top) in the IDH and RTK I subtypes correlates with active TSS (E01–04, box 2) and enhancer (E07–E011, boxes 1 and 3) ChromHMM states (bottom).We next examined methylation differences between glioblastoma subtypes and normal brain. Large regions of the genome were hypomethylated in tumours relative to control, non-neoplastic normal brain tissue (Fig. 1c). Relative to the other tumour subtypes, RTK I tumours showed global hypomethylation. Similarly, IDH tumours were globally hypermethylated, showing that the G-CIMP phenotype extends beyond CpG islands and manifests across all genomic features. Tumour hypomethylation relative to normal tissue was most pronounced in intergenic regions, while CpG islands were relatively hypermethylated (Fig. 1d). Overall, these results agree with the current understanding of methylation changes in cancers.We then used the 18-state Roadmap Epigenome ChromHMM model19 with our tumour histone mark ChIP-seq data to annotate each sample’s genome. We defined consensus subtype ChromHMM states and calculated their mean subtype methylation (Fig. 1e). We found that active TSS states (E01–E04) have relatively low methylation, while transcription (E05-E06), repressive (E12–13, E16–E17) and non-functional (E18) states have relatively higher methylation. Generally, the IDH subtype is the most hypermethylated, while RTK I is the most hypomethylated. Interestingly, the bivalent TSS and enhancer states (E14–E15) showed the broadest ranges, with a striking degree of tumour-specific hypermethylation, suggesting that differences in the methylation of tumour subtype and normal tissue may be more prevalent in genomic loci of defined function. These effects are frequently subtype-specific, as illustrated by myelin transcription factor 1 (MYT1), a regulator of oligodendrocyte differentiation. MYT1 is overexpressed in IDH and RTK I glioblastoma relative to normal brain, and shows corresponding hypomethylation and active chromatin states in the gene promoter and in known enhancer regions (Fig. 1f, g).Active enhancers are highly variable across glioblastoma subtypesWe next took advantage of our unbiased WGBS data to call methylation features, i.e. differential methylation valleys (DMVs), partially methylated domains (PMDs) and lowly methylated regions (LMRs), which are enriched in promoter, heterochromatin and enhancer states, respectively20–22. We found that more than 60% of PMDs and DMVs were shared across all subtypes, and fewer than 17% were present in only one subtype each. Conversely, 37% of LMRs were specific for one subtype, suggesting that variable DNA methylation at LMRs is a substantial contributor to differences between subtypes (Fig. 2a). Consistent with this hypothesis, Uniform Manifold Approximation and Projection (UMAP) of LMR WGBS data clearly separated the subtypes, no matter whether all samples or only IDHwt samples were analysed (Fig. 2b, c).Fig. 2Active enhancer-LMR regions show highly variable methylation across glioblastoma subtypes.a Barplots showing the extent of subtype sharing of DNA methylation valleys (DMVs, left), lowly methylated regions (LMRs, middle) and partially methylated domains (PMDs, right). b Uniform manifold approximation and projection (UMAP) plot of the study glioblastoma samples based on the analysis of DNA methylation in LMRs. c UMAP plot of the glioblastoma samples based on the analysis of DNA methylation at enhancers. d Summary statistics for subtype DMVs, LMRs and PMDs, showing (left-right): mean DNA methylation (WGBS beta; boxes correspond to the 25th, 50th/median and 75th percentiles; whiskers denote 1.5× the IQR from the median); number of regions; ChromHMM state annotation; and ChromHMM state enrichment (see Methods for details of enrichment statistic calculation) for each subtype’s feature set. IDH, n = 12; MES, n = 19; RTK I, n = 12; RTK II, n = 17 samples.To learn more about the function of PMDs, LMRs and DMVs in glioblastoma, we used our ChromHMM model to annotate these features for each subtype (Fig. 2d). As expected, DMVs were strongly enriched in TSS states (E01–E03, E14) and PMDs contained mostly quiescent and repressive states (E16–E18). Chromatin states found in LMRs were more diverse, with the largest proportion (23%) being enhancer states (E07–E11 and E15). In the LMRs that are found only in one subtype, the enrichment of enhancer states was even more pronounced with 36%, indicating the importance of enhancers for defining subtype identities.Focusing on active enhancers (E9–E10), we noted that 64% of tumour active-enhancer regions were unique to tumours and not shared in normal brain19. Of these, 59% tumour-specific active enhancers were unique to a single subtype, while only about 6% were shared by all subtypes. These results suggest that GB subtypes and their differing gene expression programmes are, at least in part, the result of subtype-specific enhancer activity.Core regulatory circuitry analysis identifies subtype Master RegulatorsEnhancer activity is mediated by transcription factor proteins, including MRs. We therefore set out to identify the subtype MRs that are active in this heterogeneous enhancer landscape. Superenhancers (SEs) are a class of genomic loci that regulate cell identity genes, including MRs23. We performed subtype SE calling on our H3K27ac glioblastoma profiles (Fig. 3a). Some SEs exhibited subtype-specific enrichment; for example, RTK II tumours have an intronic SE in EGFR that is associated with higher H3K27ac signal and EGFR expression in this subtype. SEs with higher subtype H3K27ac signal correlated with target gene up-regulation in that subtype, suggesting that SEs regulate genes that are important for subtype identity in glioblastoma (Fig. 3b). This interpretation receives additional support from our observation that subtype LMRs are present in 65% of subtype SEs, on average.Fig. 3Core regulatory circuitry analysis identifies primary glioblastoma subtype Master Regulators.a Superenhancer (SE) identification using glioblastoma subtypes’ H3K27ac profiles. Left: Hockey stick plots showing enhancers (x-axis) ranked by their H3K27ac intensity (SES-normalised values, y-axis) are shown for the four subtypes. Selected SEs are labelled with their target genes. Centre: Exemplary subtype SEs, with mean subtype H3K27ac profiles for CALCRL (IDH), TGFBI (MES), GPR17 (RTK I) and EGFR (RTK II). Subtype SEs are depicted as coloured bars below each H3K27ac profile. Right: RNA-seq gene expression (log2 TPM + 1) for the indicated genes, by subtype, visualised as Tukey boxplots. IDH, n = 12; MES, n = 19; RTK I, n = 12; RTK II, n = 17 samples. Boxes correspond to the 25th, 50th/median and 75th percentiles; whiskers denote 1.5× the IQR from the median. Points mark outliers beyond 1.5× IQR. b Tukey boxplots showing the gene expression log FC (limma) for target genes of each subtype SEs (defined by ANOVA on H3K27ac signal, minimum log fold change 1, Benjamini–Hochberg adjusted P-value = 0.1), comparing expression in that subtype to the average of the other 3. Mean log FC is indicated by the white diamond; n indicates the number of target genes; Boxes correspond to the 25th, 50th/median and 75th percentiles; whiskers denote 1.5× the IQR from the median. Points mark outliers beyond 1.5× IQR. ***Two-tailed t-test, P-value <2.2 × 10−16. c Selected gene signature enrichment results for the target genes of each subtype’s SEs. The size of each circle corresponds to the ratio of SE target genes in that gene signature, while the colour represents the adjusted P-value. d Overview of subtype Master Regulator (MR) identification. Firstly, 56 MRs were predicted with CRCmapper on the tumour H3K27ac profiles (n = 20). We extended this MR activity inference to the full tumour cohort (n = 60 samples) using VIPER to predict these 56 MRs’ activity within a gene regulatory network inferred in the TCGA cohort (n = 525 samples). In total, 38 subtype MRs were identified. e Heatmap showing the mean subtype activity for each MR (n = 38).We annotated each subtype’s SE’s target genes with mSigDB genesets and found significant enrichment for known glioblastoma subtype signatures4,5 (Fig. 3c). Furthermore, genes upregulated in neural crest stem cells were enriched in IDH, MES and RTK I, while these genes were mostly downregulated in RTK II. Oligodendrocyte markers were enriched in RTK I, while astrocyte markers were enriched in RTK II, agreeing with recent reports that glioblastomas co-opt SE landscapes used in normal CNS development15.Core regulatory circuit (CRC) analysis identified a set of 56 candidate MRs across the four subtypes. We extended this analysis to the full cohort using a gene expression-based approach (Fig. 3d). To do so, we inferred a glioblastoma gene regulatory network using the TCGA gene expression microarray cohort (n = 525)2. We then used VIPER analysis24 to infer the activity of the 38 candidate CRC MRs that appear in this network. MRs were assigned to a subtype based on the average maximum VIPER NES (IDH: 9; MES: 16; RTK I: 4; RTK II: 9; Fig. 3e and Supplementary Data 7), including the previously reported MES MRs CEBPA and STAT313.Master Regulator activity in cell typesGlioblastoma consist of mixtures of cells of different tumour subtypes, as well as normal cell populations such as tumour-associated macrophages, which are especially prevalent in MES5,16,18,25,26. We addressed this potential confounding of MR predictions made using our bulk tissue data, by extending our analysis to published single-cell RNA-seq profiles of IDHwt glioblastoma25.Following cell quality control and filtering, we performed a pseudotime analysis27. Cells formed a 5-state trajectory, consisting of four terminal branches and one intermediate state (Fig. 4a). It is important to note that we did not distinguish between tumour and normal cells in this pseudotime analysis, and therefore tumour and normal cells with similar gene expression programmes will group together. Therefore, we annotated each state by scoring each cell for a set of normal brain cell signatures28 and assigning cells a glioblastoma subtype based on VIPER-calculated MR activity (Fig. 4b). Based on these results and examination of individual marker genes, each terminal branch was assigned the following identity: state 1 represents RTK I tumour cells and normal oligodendrocytes; state 3 are RTK II tumour cells and normal astrocytes; state 4 is a mixture of normal cell types; and state 5 consists of MES tumour cells and macrophages.Fig. 4Validation of core regulatory circuitry Master Regulator predictions using glioblastoma single-cell RNA-seq.a Pseudotime trajectory inferred with monocle using QC-filtered single cells from the Darmanis (2017)25 glioblastoma dataset. At top, the cells are coloured by the inferred pseudotime state. Below, cells are coloured by their source sample in the original study. b Assignment of glioblastoma subtype and normal cell identities to pseudotime states. Normal brain cell type signature scores (McKenzie et al., 2018)28; median scores are indicated by the white diamond; left), and subtypes assigned based on the VIPER-inferred activity of CIMP- CRC MRs (n = 25 analysable MRs) in the TCGA gene regulatory network were calculated for each cell. c Visualisation of relative MR activity (SREBF1, ubiquitous, top left; CEBPA, MES, top right; SOX10, RTK I, bottom left; NR3C1, RTK II, bottom right) on the pseudotime trajectory. The scaled VIPER NES was calculated for each MR using each cell’s expression profile and the RTN-derived regulons. d Heatmap visualisation of VIPER NES for the n = 28 CIMP- subtype CRC MRs, split by state. Each column corresponds to a single cell and is annotated with the source tumour sample. Within each state, samples were clustered by their activity profiles. The dashed white line delineates two subpopulations of cells in state 5 with differing MR activity. e tSNE projection of the MES tumour and TAM/microglial single cells in pseudotime state 5 (n = 2112) using the MES CRC MR (n = 13) activity matrix. Each cell is coloured by its relative expression of the macrophage markers AIF1 (left) and PTPRC (right). f As in (e), but coloured by the scaled MR activity (VIPER NES) of FLI1 (left) and STAT3 (right).We next visualised relative MR activity scores on the pseudotime trajectory, to confirm that the subtype predicted in our CRC analysis is consistent at this single-cell level (Fig. 4c). While some MRs exhibited ubiquitous activity (SREBF1), we observed that subtype MRs generally have higher activity in the cells of that state (CEBPA in MES, SOX10 in RTK I, NR3C1 in RTK II), as predicted.We clustered cells based on MR activity within each state (Fig. 4d), and observed that four MES MRs (CEBPA, FLI1, MAFB, MITF) separate two subpopulations within state 5. Interestingly, even within the cells with high activity of these four factors, further MR heterogeneity is observable (e.g. RXRA). This suggested us that MR activity might discriminate cell types.Therefore, we further examined the 2112 cells in state 5. A t-SNE projection of the cells’ MES MR (n = 13) VIPER activity matrix splits this state into two sub-clusters. Visualisation of microglial/macrophage marker gene expression (AIF1, PTPRC; Fig. 4e) identified the major population as these infiltrating immune cells, as expected. This suggests that the remaining population are MES glioblastoma cells. We found that the MRs CEBPA, FLI1, MAFB, MITF showed higher activities in this immune population, whereas STAT3, which was previously reported to induce mesenchymal transformation in NSCs13, appears to be active in both populations (Fig. 4f). Therefore, our analysis of scRNA-seq data were consistent with the analyses of bulk tissue data, in that it identified three tumour branches to which the respective subtype MRs could be assigned. Furthermore, for the MES subtype, which is known to be enriched with infiltrating immune cells, we were able to use scRNA-seq data to compare MR activity in tumour cells and tumour-associated macrophages.Loss of SOX10 results in an RTK I-to-MES transitionTo independently validate our CRC analysis, we analysed gene expression-based GB regulatory networks using the Reconstruction of Transcriptional Networks and Analysis of Master Regulators (RTN) package29. Along with the previously used TCGA network (Fig. 3d; Supplementary Data 8), we used a network inferred using an additional 569 microarray samples (Supplementary Data 9) to cross-validate our predictions. We defined glioblastoma subtype RNA expression signatures (Supplementary Fig. 2 and Supplementary Data 10), and used these to identify 117 subtype MRs that were statistically (two-sided t-test) significantly active in the same subtype, with the same direction of activity, in both networks (Supplementary Fig. 3 and Supplementary Data 7, 11).Overlap of these RNA-based predictions with chromatin-based CRC MRs (Fig. 3d) gave a consensus list of 10 MRs (Fig. 5a), from which we selected SOX10 for functional validation. SOX10 is an oligodendroglial lineage transcription factor30, and regulates a distinct epigenetic state that is linked to chromatin remodelling and therapy resistance in melanoma, which, like glioblastoma, originates from the neural-crest31,32. This evidence of its involvement in clinically relevant state transitions makes SOX10 a particularly interesting candidate to study epigenetic control and remodelling of subtype gene regulation in glioblastoma. In addition, SOX10 has been shown to affect cell fate decisions in neural lineage development in mice, and was described to antagonise the function of the transcription factor NFIA in driving astrocytic differentiation in normal development and later, overexpression in mouse tumour models33,34.Fig. 5SOX10 is a Master Regulator of the RTK I subtype.a Consensus Master Regulators. b Correlation of DNA methylation and SOX10 expression within the SOX10 gene body. Boxes in Tukey plots correspond to the 25th, 50th/median and 75th percentiles; whiskers denote 1.5× the IQR from the median. Points mark outliers beyond 1.5× IQR.IDH, n = 12; MES, n = 19; RTK I, n = 12; RTK II, n = 17 samples. c Epigenome landscape of SOX10 in glioblastoma. Per-sample methylation (WGBS beta, top) and subtype mean H3K27ac intensity (SES-normalised, bottom) are shown. d GSEA plots showing enrichment of proneural and mesenchymal gene signatures in control and SOX10 KD LN229 cells. Upper row: limma subtype signatures of tumour cell-specific gene expression; lower row: tumour cell-specific signatures of Wang et al. (2017)5. GSEA-calculated statistics for gene set enrichment are shown. P-values (all < 0.001) and FDR values were computed empirically using a permutation test (n = 1000 permutations) based on the enrichment score. e EnrichedHeatmap visualisation of genome regions with differential chromosome accessibility in LN229 control and SOX10 KD cells, as identified by ATAC-seq analysis. SES-normalised signals of SOX10 ChIP-seq, ATAC-seq and BRD4 ChIP-seq are displayed. Signal intensity is shown in the blue–red heatmaps, where each row shows a single ATAC peak, as indicated by the vertical dashed lines, and 1 kbp further 5′ and 3′. The lineplots at the top of each heatmap display the mean signal intensity across all the regions in that category (control: green; SOX10 KD: blue). f Volcano plot of de novo motif finding with HOMER from the differentially bound ATAC-seq peaks in LN229 cells. The significantly enriched motifs are labelled. g ChromHMM annotations of LN229 ATAC-seq peaks. Active TSS (E01–E04) and Enhancer (E07–E11) states in the NT (left) and SOX10 KD conditions (right) are shown. h Western blot of SOX10 and BRD4 co-immunoprecipitation in the cell line LN229 (two independent experiments). i Factor co-occupancy at SOX10 peaks in LN229. The SES-normalised signal for peak regions and 1 kbp up and downstream for SOX10, BRD4, H3K27ac, H3K4me1 and H3K4me3 were separately scaled. j, k Changes in SOX10 and BRD4 binding and ATAC-measured chromatin accessibility at the RTK I subtype genes SOX10 (j) and ERBB3 (k). SES-normalised ChIP-seq and ATAC-seq signal is shown in the NT and SOX10 KD conditions in the LN229 (top) and ZH487 (bottom) cell lines. The boxes indicate regulatory regions where SOX10, BRD4 and ATAC-seq signal change in a co-ordinated manner.SOX10 is over-expressed in RTK I tumours (Fig. 5b), correlating with genic hypomethylation and increased H3K27ac signal (Fig. 5c). We screened glioblastoma cell lines and selected two lines exhibiting high SOX10 expression and promoter hypomethylation, characteristic of RTK I patient samples for use as in vitro models: ZH487, a primary glioblastoma cell line that we established from an RTK I patient sample, and the conventional primary glioblastoma cell line LN229. Furthermore, SOX10 ChIP-seq identified a large number of binding sites that are shared by the two cell lines, showing that these are appropriate models of SOX10 activity (Supplementary Fig. 4).Suppression of SOX10 expression (Supplementary Fig. 5a) leads to extensive changes in RNA expression that we analysed by gene set enrichment analysis using our subtype-specific gene signatures, from which we selected only the tumour-specific genes (Supplementary Data 12) using a published gene list5. This analysis showed that SOX10 suppression resulted in RTK I-to-MES transition in the LN229 and ZH487 cell lines. We confirmed this finding applying the proneural (PN), mesenchymal (MES) and classical subtype gene signatures that selectively account for tumour cell-intrinsic effects (Fig. 5d and Supplementary Fig. 5b). Consistent with this finding, VIPER-inferred activity levels of the RTK I (n = 3, excluding SOX10) and MES (n = 15) MRs correlated with the control and SOX10 KD conditions (Supplementary Fig. 5c). The observed RTK I-MES transcriptomic transition was accompanied by increased cell invasion in trans-well invasion assay and organotypic ex vivo brain slice assays35 (Supplementary Fig. 5d, e), suggesting that an RTK I-to-MES transition had indeed occurred.SOX10 repression remodels the glioblastoma enhancer landscapeSOX10 has been implicated as a chromatin modifier36–38, suggesting us that the effects of SOX10 loss in this RTK I-to-MES transition may be mediated via chromatin changes. This was supported by ATAC-seq analyses showing that chromatin accessibility significantly (two-sided t-test) decreased at RTK I MR loci including SOX10, SOX8 and ERBB3, and increased at MES MR loci such as RUNX213, FOSL2 and SERPINE1 following SOX10 suppression (Supplementary Data 13, 14). Consistently, SOX10 binding sites identified in ChIP-seq data were preferentially located in genomic regions with increased accessibility in the control rather than in SOX10 KD cells (89% vs. 49%) (Fig. 5e and Supplementary Fig. 6a). Differential motif enrichment analysis of these ATAC-seq regions found enrichment of SOX motifs (SOX9, SOX10, SOX3, SOX4, SOX15, SOX2, SOX17) in control, and predicted MES MR motifs (Fosl1/2, Jun-AP1, RUNX, TEAD) in SOX10 KD cells (Fig. 5f and Supplementary Fig. 6b).ChromHMM annotation of ATAC-seq peaks revealed chromatin-accessibility changes to preferentially affect enhancer, but not TSS, states (Fig. 5g), indicating the importance of enhancers for subtype identity and agreeing with our analysis of subtype LMRs (Fig. 2). To verify that SOX10 binds to active enhancers in RTK I cells, we analysed the occupancy of Bromodomain containing 4 (BRD4) protein, a marker of active enhancers39. Mapping of BRD4 binding to the ATAC-seq regions showed redistribution of BRD4 following SOX10-mediated changes in chromatin accessibility, confirming the remodelling of the active enhancer landscape (Fig. 5e). JQ1 inhibition of BRD4 binding was sufficient to block up-regulation of the MES MR RUNX2 following SOX10 KD, suggesting that the RTK I-to-MES transition is dependent on enhancer dynamics (Supplementary Fig. 6d–f). Co-immunoprecipitation (Co-IP) confirmed that SOX10 and BRD4 physically interact (Fig. 5h), and SOX10 binding sites showed strong BRD4 binding and histone modifications typical of active enhancers (Fig. 5i and Supplementary Fig. 6c), suggesting that SOX10 recruits this co-factor to RTK I active enhancers. Consistent with this hypothesis, we observed loss of BRD4 binding and chromatin accessibility at regulatory regions of RTK1 genes following SOX10 repression (Fig. 5j, k).In summary, these results suggest that SOX10 maintains the RTK I cellular state via direct regulation of RTK I genes. Loss of SOX10 results in chromatin accessibility changes, enhancer remodelling and the release of BRD4 from RTK I enhancers. At this point it remains unclear, which factors are recruiting BRD4 to MES enhancers, leading to the manifestation of the MES cellular state.SOX10 repression results in mesenchymal phenotype in vivoThe mesenchymal subtype of glioblastoma has been associated with increased tumour cell invasion and immune cell infiltration in patients5,9. We therefore turned to a recently established immunocompetent syngeneic graft mouse model of glioblastoma with the genetic background of neural stem-cell specific Pten/Tp53 double knockout40,41 to investigate the role of SOX10 in these phenotypes. Repression of SOX10 resulted in faster in vivo tumour growth and a highly significant decrease in median survival time of engrafted mice (NT: 104 days, n = 10; SOX10 KD: 63 days, n = 9; P < 0.001, Fig. 6a–c).Fig. 6Loss of SOX10 induces a mesenchymal phenotype in vivo.a Representative MRI images of mouse brains bearing control (left) and SOX10 KD (right) tumours, taken 57 days post cell injection. b Median tumour volumes (µl) measured using MRI 57 days post-injection. NT: 1.85 µl, n = 10 animals; SOX10 KD: 50.3 µl, n = 9 animals. One-sided t-test, P = 8.65e-5. Boxes in Tukey plots correspond to the 25th, 50th/median and 75th percentiles; whiskers denote 1.5× the IQR from the median. Points mark outliers beyond 1.5× IQR. c Kaplan-Meier survival analysis. NT: median 104 days, n = 10 animals; SOX10 KD: median 63 days, n = 9 animals. Two-sided log-rank test, P = 4.89e-5. d H&E stainings of NT (left) and SOX10 KD (right) tumours were performed for two animals per group. Scale bars indicate 200 and 100 µm in gross and detail views, respectively. e Staining of tumour margins in 2 control and 2 SOX10 KD tumours with DAPI and antibodies against GFP, which is expressed only in tumour cells. Scale bars correspond to 100 µm. f RNA expression of myeloid and microglia marker genes in n = 3 (shNT) and n = 5 (shSOX10) tumours. Boxes in Tukey plots correspond to the 25th, 50th/median and 75th percentiles; whiskers denote 1.5× the IQR from the median. Points mark outliers beyond 1.5× IQR. g Immunohistochemistry staining of Aif1 (Iba1) at the tumour margin (top) and in the tumour bulk (bottom). Red boxes indicate the areas shown in close-up. Scale bar: 100 µm; two animals per group; 5 fields examined in each animal. h Quantification of Aif1 (Iba1) staining. Aif1-positive areas were computed for 10 fields per sample and 3 samples per condition. Mean ± standard deviation are shown; Two-sided t-test; P = 0.025.H&E staining of whole-brain sections showed that control tumours exhibit a fairly defined area of the tumour bulk and a marginal zone of tumour cells invading the surrounding tissue while in the case of SOX10 KD cells tumour boundaries appear more disrupted (Fig. 6d). This impression was supported by staining of GFP, which is expressed only in the tumour cells, showing better-defined tumour margins in control than in knockdown tumours (Fig. 6e). These findings suggest possible increased invasion of the surrounding normal tissue by tumour cells after SOX10 repression, consistent with our in vitro trans-well invasion and organotypic ex vivo brain slice assays (Supplementary Fig. 5d–f).RNA profiling of microenvironment-related genes revealed increased expression of markers for TAMs and resident microglia (Aif1, Itgam, Cd68 and Cx3cl1), and macrophage M1/M2 polarisation (Cd80 vs. Cd163) in knockdown tumours (Fig. 6f). In addition, immunohistochemistry staining identified increased numbers of Aif1 (Iba1) positive cells in SOX10 KD compared to control tumours (Fig. 6g). At the tumour margins, Aif1-positive cells showed a microglia-like morphology while they appeared more roundish in the tumour bulk (Fig. 6g, top vs. bottom row). In agreement with the RNA expression data, quantification of Aif1 staining showed an increase in tumour-associated macrophage infiltration in the SOX10 KD tumours (Aif1 positive area: NT: 4.57%; SOX10 KD: 17.11%; 10 fields of view in 3 tumours in each condition; P < 0.001) (Fig. 6h). In summary, these results show that SOX10 repression causes a phenotypic switch to a mesenchymal state in vivo, resulting in increased immune-cell infiltration and significantly decreased survival time.Finally, we returned to our primary tumour data to find evidence of SOX10-associated RTK I-to-MES transition in the RTK I and MES patient samples. Clustering of the 5000 most variable microarray probes in these 31 tumours identified 2 subtype clusters showing a gradient of methylation (Fig. 7a). We found that SOX10 expression is higher in RTK I than in MES tumours, and observed the same trend in proneural and MES5 gene expression (Fig. 7b–d). In agreement with results in vivo, MES tumours also exhibited higher myeloid marker gene expression than in RTK I (Fig. 7e). The differential H3K27ac enrichment of RTK I and MES SEs, the expression of SE-defined subtype identity genes and MR activity of the RTK I and MES MRs identified in the CRC analysis also correlated with this methylation gradient (Fig. 7f–h). In addition, the RTK I and MES tumour patients of our study cohort showed significantly different patients’ survival, and low SOX10 expression in MES-subtype glioblastoma of the TCGA cohort significantly correlated with adverse prognosis (Fig. 7i, j). These data add further support to the concept of a gradient of SOX10-dependent molecular and phenotypic characteristics in human glioblastoma.Fig. 7Genetic and epigenetic patterns of RTK I-to-MES transition in primary glioblastoma tissues.a Clustering of the 5000 most variable microarray probes in RTK I and MES tumours (n = 31) from our cohort identified 3 clusters consisting of RTK I tumours, MES tumours with intermediate genotypes and MES tumours with typical MES genotypes. b SOX10 RNA-seq (log2 TPM + 1) expression. c Wang PN and MES subtype signature ssGSEA score. d Relative expression of Wang PN and MES subtype marker genes. e Relative expression of myeloid cell marker genes. f RTK I and MES SEs differential H3K27ac enrichment (t-statistic) R. g Expression score of each subtype-specific SE’s target genes (MES: n = 422; RTK I: n = 279). h MR activity (VIPER NES) of the RTK I (n = 4) and MES CRC MRs (n = 12). i Kaplan-Meier survival curves for the RTK I and MES subgroups, considering overall (left panel; P = 0.030) and progression-free survival (right panel; P = 0.060). j Kaplan-Meier survival curves for MES tumours (n = 132) of the TCGA glioblastoma cohort stratified by average SOX10 expression. Overall survival (left panel; P = 0.036); progression-free survival, (right panel; P = 0.009). Group cut-off: average expression; statistical significance was determined by two-sided log-rank test.DiscussionIn this study, using chromatin and transcriptome gene regulatory network analyses, we show that glioblastoma subtypes have distinct enhancer landscapes and Master Regulator (MR) repertoires. In patient-derived and adherent cell line models, we found that SOX10 is an RTK I MR, and that its repression results in a transcriptomic and phenotypic RTK I-to-MES transition via remodelling of the enhancer landscape. Finally, we demonstrate that repression of SOX10 in an advanced syngeneic mouse model has a major effect on in vivo phenotypes, including altered growth patterns, increased immune cell content and a significant decrease in survival.Past methylome studies of glioblastoma have used microarray2,3 or RRBS9 data that sample a relatively small proportion of the genome. In contrast, our study generated WGBS data from a large number of glioblastoma primary tumours, which allowed us to analyse CpG methylation genome-wide, and describe subtype methylation differences that could not previously be appreciated. We found that the G-CIMP hypermethylation characteristic of IDH glioblastoma is not limited to CpG islands, but affects all genomic features and functional chromatin states. Furthermore, our analysis highlighted the importance of LMR and enhancer methylation for differentiating glioblastoma subtypes.We next identified the MRs that operate within these enhancer landscapes using complementary analyses of chromatin and transcriptome data generated from primary patient samples, providing the most comprehensive analysis to date. Previously, smaller studies had already provided evidence for the importance of enhancers and bivalent promoters for subtype identity42,43; however, these studies used RNA transcription-based subtype classification schemes that did not differentiate between IDHwt and IDH-mutated proneural-subtype tumours, or included the meanwhile abandoned concept of a “Neural” subtype of glioblastoma2,4. These classification schemes are only partially consistent with current concepts of glioblastoma subtyping1–3,5–7,44,45 and their findings, therefore, difficult to relate to ours.We validated our subtype MR predictions using published scRNA-seq data from IDHwt tumours25. Using this approach, we could separate cell states corresponding to normal cells and the three IDHwt glioblastoma subtypes. Recent single-cell studies of glioblastoma have proposed a variety of cell state models5,16,18,25,26. However, xenotransplantation studies have demonstrated that even a single glioblastoma cell can regenerate the cellular heterogeneity of the parent tumour16. Our analysis implicates enhancers and subtype MRs as key contributors to this cell state plasticity.Our complementary analyses identified 10 high-confidence, consensus subtype MRs. We selected the RTK I MR candidate SOX10 for functional characterisation. SOX10 is a member of the developmentally important SRY-related HMG-box containing (SOX) family of TFs46. In the CNS, SOX10 is an oligodendrocytic marker gene, and its activity is an example of tumour cells co-opting a developmental pathway to escape the terminal cell differentiation state34,47. In melanocytes, SOX10 binds promoters and distal elements48 and can recruit chromatin-modifying co-factors such as SMARCA438 and Chd736.We found that SOX10 maintains the RTK I transcriptomic state via regulation of subtype genes and its loss caused a shift to a mesenchymal phenotype. This transition is dependent on enhancer remodelling, as demonstrated by blockade of BRD4 activity using JQ1 inhibition, a phenomenon that was first described in recurrent glioblastomas10. Interestingly, recent data suggest that the reverse, MES-to-RTK I transition might be forced by depletion of the AP1 transcription factor FOSL1 in an NF1-mutant model49, supportive of a high level of plasticity of subtype identities. Other work has shown that RTK I-to-MES transition can be induced by therapy and correlates with resistance development50–54. Recent data indicate that MES glioblastoma has the worst prognosis of all subtypes5,9,18,42. Our in vivo results suggest that loss of SOX10, leading to changes in myeloid cells, may be the underlying cause of this decrease in survival, an interpretation consistent with the correlation of low SOX10 expression and adverse survival that we observed for human MES-type glioblastoma patients.Notably, loss of SOX10 has also been linked to adverse outcomes in other neural crest-derived tumours. For example, its loss in melanoma leads to transcriptome rewiring and drug resistance32. It is striking that neural crest cells undergo an analogous mesenchymal transition during normal development, suggesting that even in highly divergent tissues, hijacking of common developmental pathways by cancer can occur46,55.Neural lineage development has been shown to be regulated by sequentially interacting SOX transcription factors56, including SOX2, a reported marker of the proneural glioblastoma subtype and regulator of cell plasticity and astrocytic differentiation57,58. In our tumour cohort, SOX2 RNA-expression was about 2× lower in MES compared to other subtypes, but still remained at a very high level, about 3× higher than in normal brain. Furthermore, SOX2 showed strong H3K27ac activation in all glioblastoma subtypes (Fig. 3a), and its RNA expression did not significantly change in the SOX10 knockdown models. We therefore have no evidence on RNA-level to support a subtype-specific role of SOX2 or its relevance in the context of RTK I-to-MES transition induced by the loss of SOX10.In extensive studies of murine neural developmental pathways and their potential role in human gliomagenesis, SOX9-dependent activation of the transcription factor nuclear factor I-A (NFIA) was shown to antagonise SOX10 function and result in astrocytic differentiation of the tumour cells33,34. Although we identified SOX9 as an RTK II MR (Fig. 5a) and observed consistent upregulation of SOX9 in our SOX10 knockdown models, loss of SOX10 for unknown reasons did not result in the upregulation of NFIA RNA expression. This possibly provides an explanation to why we observed RTK I-to-MES rather than RTK I-to-RTK II transition after SOX10 suppression (Fig. 5d and Supplementary Fig. 5b).Our subtype MR analysis identifies many candidates beyond SOX10 that are known to play roles in CNS development, suggesting a model in which MR activity maintains and interacts with genetic and epigenetic factors to define a glioblastoma cell state. Evidence of plasticity, such as genetic, epigenetic and regulatory features of RTK I-to-MES transition, can be readily visualised in the multiple layers of data that we generated from primary tumours. The role of tumour cell-extrinsic components such as the microenvironment will add further complexity to this tumour cell-intrinsic plasticity, as suggested by our in vivo experiments. In this context, elucidating the mechanisms that promote myeloid-cell invasion and immune suppression will be of particular clinical relevance for the development of immunotherapy approaches. These findings are mirrored by a recent publication showing that knockdown of SOX10 is sufficient to convert a cell from melanocyte-like to mesenchymal-like in melanoma, and demonstrating that microenvironmental cues likely play a critical role in regulating melanoma cell state59. If, as is increasingly plausible, subtype plasticity contributes to therapy failure in glioblastoma, drug combinations simultaneously targeting both tumour cell growth and epigenetic plasticity may block the escape of cancer cells to a therapy-resistant state and thus lead to improved patient outcomes.MethodsPrimary tissue samplesSnap-frozen primary glioblastoma tumour samples and clinical data were collected at the time of primary diagnosis between 1994 and 2011 at the Burdenko Neurosurgery Institute (Moscow, Russia). Informed consent was obtained from all patients. Use of the material and clinical data for this study was approved by the ethics board at the Burdenko Neurosurgery Institute (Moscow, Russia). The patient cohort consisted of 32 males and 28 females with an average age of 52.5 ± 11 years (mean and standard deviation). Patients with IDH-subtype tumours tended to be younger than patients with tumours of the MES, RTK I and RTK II subtypes (42.6 ± 9.5 vs. 55.0 ± 9.9 years; mean and standard deviation). IDH1/2 mutation status were determined using either pyrosequencing or Sanger sequencing60,61. Samples of post-mortem normal brain were purchased from Biocat (Heidelberg, Germany).Cell lines and cell culture detailsThe human glioblastoma cell line LN229 (p53 mut, PTEN wt, p16 del; established from a white, 60-year old female in 1979) was obtained from ATCC (Cat#CRL-2611) and cultured in DMEM supplemented with 10% FCS, Penicillin/Streptomycin and glutamine. ZH487 patient-derived glioblastoma cells were established at the University of Zurich Hospital. ZH487 cells were cultured in Neurobasal medium (Cat#12348017, NBM) supplemented with 2% B27 (Cat#12587010, retinoic acid-free, Invitrogen), EGF (20 ng/ml, AF-100-15, Peprotech), FGF (20 ng/ml, Cat#100-18B, Peprotech) and glutamine (0.5 mM). HEK293T cells were used for lentivirus production and maintained as monolayer cultures in antibiotic-free DMEM supplemented with 10% FCS. All cells were cultured under 10% CO2 at 37 °C with humidity. Cell line identities were verified by the Multiplex human Cell line Authentication Test (MCA), and cells were tested for mycoplasma contamination with the Multiplex cell test (both Multiplexion GmbH, Friedrichshafen, Germany).In vivo syngeneic mouse modelAnimal experiments performed for this study comply with all relevant ethical regulations and were approved by the Regierungspräsidium Karlsruhe, Germany (reference no. G-156-15). The primary mouse glioblastoma cell line with Pten/Tp53 double knockout was established by the lab of Prof. Peter Angel in the German Cancer Research Center (DKFZ)40,41. mGB1 cells were characterised as the Proneural/RTK I subtype based on RNA-seq profiling with high SOX10 expression. mGB1 cells were culture at 37 °C in DMEM/F12 medium supplement with N2 supplement, EGF (20 ng/ml), FGF (20 ng/ml), Penicillin/Streptomycin and glutamine. SOX10 knockdown was carried out using lentivirus transduction with the shSOX10 (TRCN0000244290, Supplementary Table 3). All cells were Puromycin selected and SOX10 knockdown level was RT-PCR validated before injection; 200k cells (shNT and shSOX10, in 1 µl volume) were intracranially injected into adult C57/B6 mice (6 weeks female) brain under anaesthesia with Isoflurane. MRI scanning was performed in the DKFZ MRI core facility on 57 days post-injection.Whole genome bisulphite sequencing (WGBS)Primary GB whole-genome bisulphite library preparation was carried out as described previously20. Briefly, 5 μg of genomic DNA was sheared using a Covaris device. After adaptor ligation, DNA fragments with insert lengths of 200–250 bp were isolated using an E-Gel electrophoresis system (Life Technologies) and bisulphite converted overnight using the EZ DNA Methylation kit (Zymo Research). The fragments were PCR amplified using the FastStart High Fidelity PCR kit (Roche) for 6–8 cycles. Library aliquots were then purified and size selected with AMPure beads (New England BioLabs) and quality controlled with a Bioanalyzer (Agilent). Each library was sequenced using 2 lanes on an Illumina HiSeq 2000 in the DKFZ Genomics and Proteomics core facility.Whole genome sequencing (WGS)DNA (500 ng per sample) for whole genome sequencing were submitted to the DKFZ Genomics and Proteomics core facility and the library preparations were carried out using the standard protocol from Illumina. Each library was sequenced using 1 lane on an Illumina HiSeq X Ten.RNA sequencingPrimary GB RNA-seq libraries were prepared as described using methods to preserve strand specificity and deplete rRNA. Sequencing was carried out on the HiSeq 2000 platform with 1 lane per sample. All samples profiled by WGBS sequencing, WGS, 450k/Epic methylation microarray and RNA sequencing were genotyped in silico to exclude sample swaps.RNA-seq of GB cell lines (LN229 and ZH487, Control vs. shOX10) were performed using the polyA-selected RNA-seq libraries preparation protocol with the TruSeq stranded RNAseq Illumina kit by the DKFZ genomics & Proteomics core facility. Libraries were multiplexed-sequenced using 1 lane on a HiSeq 2000 v4, generating 50 bp single-end reads. RNA-seq libraries preparation of mouse tumour samples (shNT, n = 3; shSOX10, n = 5) were also used in the above mentioned PolyA protocol. Libraries were multiplexed-sequenced using 2 lanes on a HiSeq 2000 v4, generating 50 bp single-end reads.ChIP-seqThen, 10 µg each of H3K27Ac (Cat#AM39133, Active Motif), H3K4me1 (Cat#AM38297, Active Motif), H3K4me3 (Cat#AM39159, Active Motif), H3K9me3 (Cat#AM39161, Active Motif), H3K27me3 (Cat#07-449, Millipore) and H3K36me3 (Cat# AM61101, Active Motif) were used for ChIP library preparation of GB patient samples, which was performed at Active Motif according to proprietary methods. Libraries were multiplexed so that all libraries for each individual IP were sequenced on 1–4 lanes using the Illumina HiSeq 2000 platform. For the LN229 histone mark experiments, LN229 cells (LN229-shSOX10 without (Control) or with doxycycline induction of SOX10 knockdown) were cross-linked with 1% methanol-free formaldehyde for 10 min. After quenching with glycine, cells were washed three times with PBS and the cell pellet was treated with 4 U MNase per 1 × 106 cells for 15 min. MNase was stopped with 10× Covaris buffer and the chromatin was sheared for an additional 15 min with the LE220 Covaris device. The soluble chromatin was then recovered, quantified, and 2 µg chromatin was used in each immunoprecipitation (IP) with 2 µl each of each antibody (as above). Following the IP and washes with Covaris buffer, Li-buffer and TE, chromatin was digested with proteinase K and purified with AMPure beads. The purified DNA was cloned into illumina sequencing libraries with the NEBNext Ultra library preparation kit (NEB) according to standard protocols. For SOX10 and BRD4 ChIP-seq experiments, cells were cross-linked with 1% methanol-free formaldehyde for 15 min and quenched with 0.125 M glycine. Chromatin was isolated by adding lysis buffer and Dounce homogenisation. Collected chromatin was sheared via sonication to an average length of 300–500 bp. Input genomic DNA was prepared from collected chromatin by treatment with RNase, proteinase K and de-crosslinking under heat, and then isolated by ethanol precipitation. Pellets were re-suspended and DNA quantified on a NanoDrop spectrophotometer. Estimated total chromatin yield was calculated based on this amount; 30 µg of chromatin was pre-cleared with protein A agarose beads (Invitrogen), and DNA precipitated using 4 µg of antibody against SOX10 or BRD4. This DNA was isolated from the beads by washing followed by SDS buffer elution, RNase/proteinase K treatment and de-crosslinking under heat (65 °C overnight incubation). DNA was then purified using phenol-chloroform extraction and ethanol precipitation. Sequencing libraries were prepared and input DNA via standard protocols (enzymatic end-polishing, dA-addition and adaptor ligation) on an Apollo 342 NGS Library Prep system (Wafergen Biosystems/Takara). Prepared libraries were then sequenced (50 bp, single-end) on an Illumina HiSeq 2000.ATAC-seqATAC-seq was performed in biological duplicates, as previously described62. Briefly, viable frozen cells were incubated with Tn5 in 0.1% Igepal CA-630 (37 °C, 30’). Transposition was stopped with EDTA and DNA purified using AMPure beads. After DNA purification, barcodes were added using PCR and DNA re-purified on AMPure beads. These prepared libraries were then sequenced (50 bp, single-end) on an Illumina HiSeq 2000.Gene expression microarray profiling of cell linesDNase-treated total RNA (500 ng) was prepared for gene expression profiling on Affymetrix HG-U133-Plus2 and Illumina HumanHT-12 v4 Expression BeadChip microarrays at the DKFZ Genomics & Proteomics Core Facility. The GSEA results from Fig. 5d and Supplementary Fig. 5b were generated with the microarray data from the Affymetrix Human U133Plus 2.0 platform. For LN229 cells, the control group included the non-treated control and non-targeting sgRNA control, and for the SOX10 knockdown group, three guild RNA targeting SOX10 were used. For ZH487 cells, non-treated control and shNT was used as control with biological replicates and three shSOX10 shRNA were used to achieve SOX10 repression also with biological replicates.Methylation microarray data processing and CNV callingHere, 450k and EPIC DNA methylation array data were processed and analysed as previously described44 using the minfi (1.24.0)63 and conumee (1.3.0)64 Bioconductor packages. In brief, > 500 ng of DNA from snap-frozen samples was used as input material. minfi was used to extract raw signal intensities from IDAT files, and both colour channels corrected for background and dye-bias. Beta values were calculated using an offset of 100. CNVs were called using the standard conumee procedure, using two sets of 50 control samples with balanced CN profiles. Copy number aberrations were called from the conumee-processed values using the following numerical thresholds: for x < −1, as a deletion; −1 < x < −0.2 as a loss; −0.2 < x < 0.2 as no copy number change; 0.2 < x < 1 as a gain; x > 1 as an amplification.Subtype classification of patient samples using methylation microarraysFrom the previous GB classification,3 8000 probes (Supplementary Data 6) were used to cluster the methylation microarray data. We used only probes that appear on both 450k and EPIC microarrays (n = 7386). The methylation beta value matrix was used to calculate the sample pairwise Euclidean distance. This distance matrix was then used to hierarchically cluster samples using the ‘ward.D’ method.Clustering of MES and RTK I patient sample methylation microarraysThe MES (n = 19) and RTK I (n = 12) glioblastomas were clustered together (Fig. 7) to identify substructure in these subtypes. The 5000 most variable probes (by SD) were selected and the samples and probes hierarchically clustered using the Euclidean distance and the ‘ward.D’ method.RNA-seq processing and expression quantificationReads were aligned to the appropriate reference genome (hg19/mm10) with the Gencode reference transcriptome (v19/M2) using STAR (v2.3.0e). Read counts for each gene were quantified as the total number of reads mapping to exons using htseq-count (0.6.0) for human or featureCounts (Subread v1.5.3) for mouse samples. Gene expression values for each sample were quantified using the transcripts per million (TPM) metric.Tumour RNAseq and limma subtype gene signature analysisRaw read counts per gene were pre-filtered, retaining those genes with > 10 reads in > 6 samples for further analysis. Normalisation factors for the counts were calculated using the calcNormFactors function in ‘edgeR’ (3.20.1). voomWithQualityWeights from ‘limma’ (3.34.4) was used to transform the raw counts. limma differential expression analysis was used to compare the gene expression of each GBM subtype versus the other 3 (example contrast: IDH / ((MES + RTK I + RTK II)/3)). Genes were defined as significant for a subtype if they passed a BH adjusted P-value threshold of 0.001. Signatures from isolated mouse normal brain cell populations63 were used to compare subtypes to normal cell populations, and to GB subtype signatures. The enrichment of each gene set was tested in our samples using the ‘ssgsea’ method of the ‘GSVA’ R package (1.26.0), using TPM expression values. ESTIMATE65 (1.0.13) was used with default settings to determine immune and stromal cell content. Genes specifically up-regulated in a subtype (log FC > 0, adj. P-value < 0.05) were functionally annotated with Gene Ontology terms (“org.Hs.eg.db”, 3.5.0) using the enrichGO function from the R package “clusterProfiler” (3.6.0). MR activity in network A was inferred using the ‘R’ package VIPER (1.14.0; see below for further details).WGBS processingFor each sample, reads were mapped to the human genome (hg19) with bwa-mem (0.7.8) with a customised WGBS pipeline20. CpGs overlapping variable sites with a minor allele frequency higher than 0.25 were removed. Low coverage CpGs with 2 or fewer reads in more than 50% of the cohort were also removed from the analysis. The mean methylation of the two cytosines in a CpG dinucleotide (one C on forward strand and the other on reverse strand) was calculated by weighting their CpG coverage, i.e. m = (m1*c1 + m2*c2)/(c1 + c2), where m1 and m2 are the number of methylated CpGs of the two neighbouring cytosines and c1 and c2 are the corresponding CpG coverage. Similarly the mean coverage for the CpG dinucleotide is calculated by weighting the coverage itself: c = (c1*c1 + c2*c2)/(c1 + c2). Finally, the bsseq R package (1.10.0) was applied to smooth the methylation data and impute the missing methylation values with default parameters.Methylation feature (DMVs, LMRs, PMDs) analysisThe segmentation of the methylation features, partially methylated domains (PMDs), lowly methylated regions (LMRs) and DNA methylation valleys (DMVs) were performed by a customised pipeline. Further details can be found in Supplementary Methods. Briefly, chromosomes were split into blocks based on inter-CpG distance, and the mean and standard deviations of methylation of each block used to classify segments into low, intermediate and high methylation separately for each sample. DMVs and LMRs were then defined based on the characteristics of each block and its neighbours. PMDs were called using MethylSeekR (1.14.0)66 with 10 kbp minimum width. Subtype consensus regions (DMVs, LMRs, PMDs) were defined as segments with a cross-sample coverage of at least 4. Neighbouring LMRs were merged if the inter-LMR distance was less than 1 kb. Enrichments of other genomic features (e.g. CGIs, ChromHMM annotations) were computed by calculating the Jaccard Coefficient for the base-pair length of the two feature sets. This statistic was compared to a background of 1000 CpG content-matched regions to calculate z-scores and P-values for the enrichment. Classification of samples based on methylation patterns (enhancer and LMR methylation) was done using a projection of the methylation matrix using uniform manifold approximation and projection (UMAP)67.Subtype sharing methylation featuresFor each type of methylation features (DMVs, LMRs, PMDs), consensus methylation features in each subtype were first determined by selecting the genomic regions that occurred in more than 50% of samples in that subtype. The extent of subtype sharing methylation features was calculated as the fraction of the total width of regions that occur in 1, 2, 3 or 4 subtypes.Chromatin state enrichmentFor one set of methylation features and one set of genomic regions with a certain chromatin state, the Jaccard coefficient was used as the measurement of the overlap of two sets of regions, which was calculated as total base pairs of the intersected regions divided by the total base pairs of union of the two sets of regions. The significance of the Jaccard coefficient was calculated by permuting the methylation features restricted in a specific genomic background where the average CpG content was similar as in the methylation features. The selection of background regions was applied as follows: The methylation features were first split into small windows where the window size denoted as w was calculated as the 25th quantile of all widths of the methylation features. w was additionally rounded to the thousand digit. The window size is set to 10 kb if it is larger than 10 kb, and it is set to 1 kb if it is smaller than 1 kb. To the fact that small windows might cause bias for the calculation of CpG content due to the sparsity of CpG distribution, windows with a width less than w/4 were removed. For all the windows after filtering, the CpG content was defined as number of CpG sites per 1 kb window and denoted as p. To find proper background regions in the genome, the genome was split by windows with width w and CpG content was calculated for each window. Background windows with CpG content between the 5th percentile and the 95th percentile of P were finally selected as background regions. The methylated regions were randomly permuted within the background regions by using bedtools (v2.27.1) for 1000 times. In each permutation, the Jaccard coefficient was calculated. Finally, the z-score calculated as (s-μ)/σ was used as the measurement of the enrichment, where s is the Jaccard coefficient for the two sets of regions, μ and σ are the mean and standard deviation of the Jaccard coefficient in the random permutations.ATAC-seq and ChIP-seq processingATAC-seq and ChIP-seq datasets were processed using a custom pipeline implemented in Snakemake (v. 3.13.3). Briefly, reads were trimmed using the Trimgalore tool (https://github.com/FelixKrueger/TrimGalore) and aligned using Bowtie268 (v. 2.3.4.3) with standard parameters. Duplicates and multi-mapping reads were removed using the samtools package and the XS flag in the bam files. For the ChIP-seq data, input control (tumours: WGS; LN229: H3 ChIP-seq) and corresponding IP datasets were scaled using the SES method and converted into a bigwig track using the bamCompare tool of the deepTools2 suite69. For the ATAC-seq data, genome-wide coverage was calculated. Peaks were called using the callpeak mode in MACS2 (v. 2.1.1.20160309) (https://github.com/taoliu/MACS) for broad and narrow peaks. In addition, SICER70 was used to call peaks using the gap 600 and window 200 parameters. Various QC parameters (FRiP, PCR bottleneck coefficient, cross-strand correlation) were determined according to the ENCODE guidelines71. In addition, visual QC was performed using the signal profile at TSS of annotated genes and the fingerprint method from the deepTools2 suite.Chromatin segmentation with ChromHMMChromatin segmentation was defined using the ChromHMM (v. 1.19) tool. ChIP-seq (H3K27ac, H3K27me3, H3K36me3, H3K4me1, H3K4me3, H3K9me3) and corresponding input data were binarized using ChromHMM’s “BinarizeBam” command. The Roadmap Epigenome 18-state model19 was used to segment the genome of each sample. For the tumours, the consensus state for a subtype was defined as the state with the highest frequency in a given segment, and a minimum frequency of 50%.Superenhancer analysisThe union of H3K27ac peaks for each subtype’s samples were used as input regions for the ROSE2 superenhancer analysis pipeline, stitching together regions within 12.5 kbp of each other. Sample H3K27ac signal was calculated using ‘bigWigAverageOverBed’ (v2), and enhancers were ranked by the subtype average enrichment. SEs were defined using the default parameters for ROSE2. Subtype SEs were defined by combining all four subtype SE lists and then performing ANOVAs on the H3K27ac signal intensities, with a minimum log fold change of 1 and a Benjamini–Hochberg adjusted P-value threshold of 0.1. For comparisons of MES and RTK I SE activity, two statistics were calculated. Firstly, from the H3K27ac signal intensities, a t-statistic based on the relative distributions of signal in the two subtypes’ SEs was computed. Secondly, subtype-specific SE lists for MES (n = 422) and RTK I (n = 279) were defined as those SEs that have no overlap with any other subtype’s SE. A “subtype-specific SE gene score” was calculated using the mean expression of these targets. The comparison between SEs and subtype LMRs were done by overlapping SEs and LMRs requiring a 50% overlap of the LMR to count it as an overlap.Core regulatory circuit analysisCore regulatory circuits were determined using a modified version of the CRCmapper tool72. Instead of assigning the SE to the closest gene as implemented in the original CRCmapper tool, we computed the Spearman correlation of the H3K27ac signal on the SE over all samples, with the gene expression of all genes located within 500 kb around the SE across the same set of samples, provided the SE and the gene are located in the same topological associated domain (TAD). We assigned the highest correlated gene (within the range and within the same TAD as the SE) as the SE target gene. The rest of the CRCmapper procedure remains as implemented in the original tool: briefly, sets of autoregulatory TFs are identified by selecting TFs that are target genes of SE (where the target gene is determined as described above), under the condition that these SEs contain binding motifs for the corresponding TF. Then, cliques of autoregulatory TFs are identified in which the SEs contain binding motifs for all other TFs in the clique.Gene regulatory network inference with RTNTwo cohorts of glioblastoma gene expression microarray data were collected: A (from TCGA; n = 525 samples profiled on the Affymetrix HT HG-U133A)2 and B (samples with the pathological diagnosis “glioblastoma” in the metadata from the following five studies: E-MTAB-3073, GSE4290, GSE7696, GSE16011 and GSE43378; n = 569 samples profiled on the Affymetrix HG-U133A Plus2)73–77. The raw data were read into R and normalised using the ‘gcrma’ package (2.50.0). Study-associated batch effects were removed from cohort B using the ComBat function in ‘sva’ (3.26.0), specifying the study ID as the ‘batch’ option. ‘RTN’ (2.3.4) was used for the following steps of the analysis. Firstly, as the expression of any single gene can be measured by multiple microarray probes, the probe with the highest coefficient of variation in the expression matrix was kept for analysis. Regulatory relationships (‘edges’) between n = 1333 TFs (classes ‘a’, ‘b’ and ‘other’ as defined in69) and target genes were inferred using the ARACNe algorithm78. The direction of TF-target gene regulation (positive or negative) was inferred using Pearson correlation. TF-target edge P-values were calculated by permuting the Mutual Information matrix 1000 times, retaining edges with a BH-adjusted P-value < 0.01. The network was bootstrapped 100 times and TF-target edges found in 95% of the bootstrap samples retained. Finally, indirect TF-target edges were removed using the Data Processing Inequality (DPI) filter with a tolerance of 0.Identification of subtype Master Regulators with RTNThe limma subtype gene expression signatures were used to pre-filter potential subtype MRs using the tna.mra (BH adjusted P-value < 0.05) function in ‘RTN’. TFs passing the MRA pre-filtering step were then tested in a 1-tail GSEA using the tna.gsea1 function, using the limma-voom calculated log fold change as the GSEA phenotype. TFs regulating fewer than 15 genes were removed. Significant TFs with a BH-adjusted P-value < 0.01 (tested in 10,000 permutations) were retained. Subtype MRs were then identified using the 2-tailed GSEA test as implemented in the tna.gsea2 function, using the limma-voom calculated log fold change as the GSEA phenotype and the TF regulons inferred in the transcriptional network as the gene sets. TFs with a BH adjusted P-value < 0.01 (calculated using 10,000 permutations) were called as significantly active in that subtype. Common network MRs (n = 117) were defined as those passing this significance threshold within the same subtype, with the same direction of activity as measured by 2-tail GSEA differential Enrichment Score (dES), in both networks. The 2-tail GSEA dES in the two networks were calculated for each subtype signature for each of the consensus MRs. The results were visualised with the ‘ComplexHeatmap’ package. Finally, the two networks (A and B) were compared as follows: 2-tail GSEA dESs for each subtype signature were calculated for all TFs in both transcriptional networks (n = 512). The correlation of dESs between the two networks was calculated using Spearman’s rank correlation and visualised using the ‘ggplot2’ package (2.2.1).Single-cell RNA-seq analysisThe counts matrix from a published dataset (GSE84465) was analysed using the ‘monocle’ (2.10.1) package in R (v3.5.1). Briefly, the most variably expressed genes with a minimum mean expression of 0.1 were used to reduce dimensionality in a tSNE (first 4 principal components in 3 dimensions, regressing out the number of genes detected and the patient of origin), and cells clustered (ρ = 40, δ = 20). This cell clustering was used to identify genes detected in at least 10% of cells and differentially expressed between clusters, and the top 1000 ranked by statistical significance were used as the ordering genes for pseudotime analysis (again regressing out the number of genes detected and the source tumour). Normalised expression values (variance-stabilising transformed, VST) were used in downstream analyses. Normal brain cell population signatures (astrocytes, endothelia, microglia, neurons, oligodendrocytes) from McKenzie et al.28 were used to score cells. Briefly, for each gene in a signature, a background was defined consisting of the 100 genes with the smallest absolute expression difference. This matched background was subtracted from the expression value for each signature gene, and the sum of all of these background-corrected signature gene expression values was defined as the score. The Bioconductor package ‘viper’ (v1.14.0)24 was used to infer single-cell MR activity. The VST-normalised expression matrix was pre-filtered to remove genes with expression SD in the bottom quartile. MR regulons from network A were used to calculate the VIPER NES, and visualised in tSNE and pseudotime plots. MR activity profiles of VIPER NESs were visualised in heatmaps using ‘ComplexHeatmap’ (1.18.1), using Euclidean distance and the average clustering method. Subtype MR scores were defined by transforming each MR’s NES into z-scores, and then calculating the mean for all MRs of that subtype; the predicted subtype for a cell was the subtype with the maximum mean z-score. Cells in pseudotime State 5 (MES, microglia) were separated using tSNE (‘Rtsne’ 0.13) on the MES MR activity matrix inferred with network A. Gene expression (VST-normalised) and MR activity (network A) were visualised as for the pseudotime analysis. The RTK II MR ZBTB7C was not included in this analysis since its regulon was too small in RTN network A, which was used in this context.Cell line expression microarray data processing and analysisRaw Affymetrix microarray data were read into R and normalised using the ‘gcrma’ R package (2.50.0). Probes without an annotated gene were removed from the analysis, and the batch effect removed using ComBat from the ‘sva’ package for the ZH487 samples. Separately for each cell line model, samples were combined into two groups: control (untreated and non-targeting controls) or knockdown. Standard ‘limma’ differential expression analysis between the control and knockdown groups was performed, and genes with an adjusted P-value of < 0.05 were defined as significantly dysregulated. MR activity was inferred using each sample’s processed expression profile as input for VIPER (v1.14.0) analysis with network A. limma-calculated log FC profiles from LN229 and ZH487 cell lines with and without SOX10 repression were analysed by ‘GSEA’ (v3.0). Glioblastoma signature gene sets (Verhaak_glioblastoma: Proneural, Neural, Mesenchymal and Classical) were downloaded from the GSEA website (http://www.broadinstitute.org/gsea/). Results were visualised using Volcano plots. The activity of the MES and RTK I CRC MRs for these samples was calculated using VIPER and the RTN-inferred TCGA network A (n = 525 samples), taking the average of replicates for each condition; conditions’ MR activity profiles were clustered using the Euclidean distance and the ‘ward.D2’ method. Significance analysis of TF activities upon SOX KD (Fig. S5C) was done using a two-sided t-test to compare WT and SOX10 KD TF activities for each TF.ChIP-seq and ATAC-seq analysisMACS2 peaks calls were used as the input set of regions for this analysis in R (v3.4.3). For ATAC-seq, a consensus peakset was defined from the two biological replicates for each condition by taking the merged peaks generated by the findOverlapsOfPeaks function from the R/Bioconductor package ChIPpeakAnno (3.12.4). Signal intensity was calculated using ‘bigWigAverageOverBed’, using the SES-normalised.bigWig file for the factor of interest and the bed file of regions of interest, and visualised using the ‘EnrichedHeatmap’ package (1.9.2). Separately for the two cell lines, differential ATAC peaks were identified using the R/Bioconductor package ‘DiffBind’ (2.6.6) using an FDR threshold of 0.05. Peaks were functionally annotated based on the largest state overlap in a comparison to LN229 cell line ChromHMM states. States were collapsed to the following summary states: E01-E04 were defined as TSS; E07-E11 were defined as Enh.ATAC-seq motif findingHOMER (v4.9.1) de novo motif finding was used with the default settings, apart from defining the background to be the union of ATAC peaks in both conditions (control + shSOX10) for that cell line.Visualisation of genomic dataThe circlize package (0.4.6)78 was used for the circular visualisation of genome-wide methylation differences and chromatin states transitions. The ComplexHeatmap package (1.19.1)79 is used for visualisation of heatmaps and complex summary plots. The EnrichedHeatmap package (1.9.2)80 is used for visualisation of epigenetic signals at genomic regions. Genome browser views were visualised using the WashU Epigenome Browser or Gviz (1.22.3)81. The epik package (https://github.com/jokergoo/epik) is used for general integrative visualisation and analysis.SOX10 knockdown systemsCRISPRi SOX10 knockdown LN229 cells were used for gene expression microarray experiments; 20 nt sgRNA sequences were designed using the CRISPR web design tool (http://crispr.mit.edu), targeting a genomic window of −50 to +200 bp relative to the transcription start site as defined by the NCBI RefSeq database. sgRNA oligonucleotides were cloned into a 5′ BstXI-BlpI 3′ digested backbone of a pU6-sgRNA EF1Alpha-puro-T2A-BFP expression plasmid by adding additional sequences to obtain compatible sticky ends (see Supplementary Table 3 for oligonucleotide sequences). Stable dCas9-expressing LN229 cells were transduced with lentivirus containing gRNAs targeting SOX10, or lentivirus containing negative guide RNA using an MOI of 2.5. Positive transduced cells were selected with puromycin (1 µg/ml) for 48 h. SOX10 knockdown was evaluated 4 days after viral transduction. Inducible SOX10 knockdown cells were used in RNA-seq, ChIP-seq (SOX10, BRD4 and 6 histone marks) and ATAC-seq experiments. Inducible SOX10 knockdown cells were established by infecting LN229 cells with pLKO-Tet-On non-targeting (nt) shRNA and pLKO-Tet-On SOX10 shRNA (TRCN0000018988) lentiviral particles and puromycin selection (1 µg/ml) for 7 days. shRNA expression was induced by adding 1 µg/ml doxycycline to the medium for at least 7 days. Cells were cultured in DMEM containing 1 g/l glucose (D5921, Sigma) supplemented with 10% tetracycline-free fetal bovine serum (Clontech), 1% penicillin and streptomycin (P/S) mix and glutamine (0.5 mM). SOX10 knockdown in ZH487 cells were carried out with constitutive shRNA lentivirus infection system (shSOX10-1, 2 and 3) (see Supplementary Table 3 for shRNA sequences).RNA isolation, cDNA synthesis and qRT-PCRTotal RNA was isolated using the RNeasy Micro kit (Qiagen) according to the manufacturer’s protocol; 1000 ng was reverse transcribed using random hexamer primers and QuantiTect Rev. Transcription Kit (Qiagen) according to manufacturer’s instructions. Each cDNA sample was analysed in technical triplicate with the Applied Biosystems Prism 7900HT Fast Real-Time PCR System and Absolute SYBR Green ROX Mix (ABgene). The relative amount of specific mRNA was normalised to levels of ARF1 and DCTN2 mRNA. Expression levels were calculated according to the ΔCt method. Primer sequences are given in Supplementary Table 2.ChIP-qPCRFor the LN229 BRD4 ChIP-PCR experiments, 10 million cells per condition (doxycycline inducible system, control vs. SOX10-KD, DMSO treated vs. JQ1 (500 nM, 6 h)) were cross-linked with 1% methanol-free formaldehyde for 10 min and quenched with 0.125 M glycine. Chromatin was isolated by adding cell lysis buffer (50 mM HEPES pH 7.9,140 mM NaCl,1 mM EDTA,10% glycerol, 0.5% NP-40, 0.25%Triton-100) with protease inhibitor cocktail (Roche, Cat#11836170001). Collected chromatin was sheared via sonication in low SDS shearing buffer (0.1% SDS; 1 mM EDTA;10 mM Tris, pH 8.1) to an average length of 300–500 bp with Covaris S2 system under the conditions indicated (Covaris MicroTube; duty cycle 5%; intensity 4; cycle per burst 200, sonification time 5 min). Input genomic DNA was prepared from collected chromatin by treatment with RNase, proteinase K and de-crosslinking under heat, and then purified with the QIAquick PCR purification kit (Cat#28106, Qiagen). Sonicated chromatin DNA buffer was diluted by adding 10% Triton X-100 and 5 M NaCl (final concentrations of IP buffer: 0.1% SDS; 1 mM EDTA; 10 mM Tris, pH 8.1; 1% Triton-100; 150 mM NaCl). The chromatin was pre-cleared with magnetic protein A/G beads (Cat #CS204457, Millipore), and the samples subjected to immunoprecipitation with 10 µg antibody against BRD4 with overnight incubation in the 4 °C cold room with rotation. The next day, 20 µl fully resuspended Magnetic Protein A/G Beads was added to each sample and incubated in the cold room with rotation for 2 h. Chromatin DNA was purified from the beads by sequential washing with Low Salt Wash Buffer (0.1% SDS; 1% Triton-100; 2 mM EDTA; 20 mM Hepes-KOH, pH 7.9;150 mM NaCl), High Salt Wash Buffer (0.1%SDS; 1% Trtiton X-100; 2 mM EDTA; 20 mM Hepes-KOH, pH7.9; 500 mM NaCl), LiCl Wash Buffer (100 mM Tris-HCl, pH 7.5; 0.5 M LiCl; 1%NP-40; 1%Sodium Deoxycholate) and TE buffer (10 mM Tris-HCl, pH 8.0; 0.1 mM EDTA). Then the magnetic beads containing DNA was eluted with the Elution buffer (10 mM Tris-HCl, pH 8.0; 0.1 mM EDTA; 1% SDS) and de-crosslinking with RNase/proteinase K treatment under heat (65 °C) for 2 h. DNA was then purified with QIAquick PCR purification column. The purified DNA was ready for qRT-PCR analysis. The ChIP primers used are listed in Supplementary Table 2. The relative binding of the investigated proteins to the gene of interest was calculated from qPCR data by calculating the percentage of recovery from the ChIP to the initial input.Western blottingCells were washed with PBS and were lysed in modified RIPA lysis buffer (0.5% SDS) supplemented with protease inhibitors cocktails and phosphatase inhibitor phosSTOP. The cells were sheared and clear supernatant was collected for protein concentration measurement with the BCA assay. The protein lysates were diluted to 0.5 µg/µl with NuPAGE™ LDS Sample Buffer and reducing agent and boiled at 95 °C for 5 min. A total of 5 µg of protein samples were loaded and resolved on 4–12% Bis-Tris Protein gels according to manufacturer’s instructions. After SDS-PAGE, the proteins were then transferred onto transfer buffer pre-wetted PVDF membrane. Membranes were blocked in 5% skimmed milk or BSA in TBS-T at room temperature for 1 h with gentle shaking. The membranes were incubated with anti-SOX10 (Cat#sc-17342, Santa Cruz, 1:1000 dilution), anti-BRD4 (Cat#A301-985A100, Bethyl Lab., 1:2000 dilution) or anti-alpha-Tubulin (Cat#T9026, Sigma, 1:5000 dilution) in 5% skimmed milk TBS-T at 4 °C overnight with gentle shaking. The membranes were washed with TBS-T for 10 min and repeated three times. Corresponding horseradish peroxidase (HRP) conjugated secondary antibodies (Anti-mouse IgG, Cell Signalling Technology, Cat#7076, 1: 5000; anti-rabbit IgG, Cell Signalling Technology Cat#7074, 1:5000; anti-goat IgG, Santa Cruz sc-2354, 1:5000) were added and incubated for 1 h with gentle agitation at room temperature. The membranes were washed again with TBS-T three times for a total of 30 min before addition of ECL reagents or ECL plus reagents. Signals were subsequently detected by light-sensitive film. Alpha-tubulin was used as loading control. Uncropped images of the western blots are provided in Supplementary Fig. 7.Co-immunoprecipitationHere, 10 million LN229 cells per condition were harvested and washed with PBS. The cells were lysed with Pierce IP lysis buffer (Cat#87787, Life Technologies) supplemented with protease inhibitor cocktail (Cat#11836170001, Roche). Samples were incubated on ice for 30 min with intermittent vertexing. After centrifugation, the supernatant was collected and pre-cleared with Protein G Dynabeads (Cat#10003D, Life Technologies). Protein concentration was determined and Input sample was collected (500 µg each); 10 µg SOX10 antibody (Cat#155279, Abcam) per condition was used to immunoprecipitate protein. Protein G beads were added to the lysate and subjected to overnight incubation, with rotation, in a cold room. The supernatant was removed, isolated beads were washed with TBST three times and resuspended in 20 µl 2× loading buffer (Cat#NP0007, Abcam). Samples were incubated at 70 °C for 10 min and the supernatant was collected via magnetic separation. A further 10 µl 2× loading buffer was added to the beads, and the beads were incubated at 95 °C for 5 min. The supernatant was collected and combined with the previous eluate. This combined eluate was then resolved using SDS-PAGE and the proteins visualised via Western Blot.In vitro invasion assayHere, 200,000 ZH487 cells were seeded into each well of a 6-well plate. Cells were transduced with shRNA targeting SOX10 and NT control, expanded for 4 days, then collected, counted and seeded (50,000 per well) for 36 h in the Neurobasal growth medium (without B27/EGF/FGF) in Biocoat™ Matrigel invasion chambers (8 µm pores, Cat# 354480, BD Bioscience, Bedford, MA). Invasion was then induced by incubation with full growth medium supplemented with B27/EGF/FGF in the lower chamber. Non-invading cells were removed and the remaining cells fixed and stained with haematoxylin. Images were taken with a light microscope (Zeiss, Germany) at 100× magnification.Ex vivo brain slice invasion assayThe assay was performed as described35. Briefly, a 6–8 week old C57Bl/6 N mouse was euthanized, the brain was isolated and the cerebellum removed with a scalpel. The brain was cut in 350 μm thick coronal slices with a vibratome (Leica VT1200 S). The slices were cultivated on top of a filter (Cat#PICM03050, Millipore) in a 6-well plate with a medium composed of: MEM (Cat# M2279, Sigma), 25% heat-inactivated horse serum (Cat# 26050070, Life Technologies), 25 mM HEPES (Cat#H0887-100 ml, Sigma), 1 mM L-glutamine (Cat#G7513, Sigma), 5 mg/ml glucose (Sigma Cat#G8769), 100 U/ml penicillin/streptomycin (Cat#P4333, Sigma). Control (shNT) and inducible SOX10-KD LN229 cells were treated with Dox for at least 7 days. LN229, control and SOX10-KD, glioma cells cultivated in medium (DMEM, Cat# D5671, Sigma; 10% FBS, Cat#F7524, Sigma; 2 mM L-glutamine) were trypsinized and counted. 1 × 106 cells/ml PBS were incubated with 5 μl lipophilic dye DiD (1 mg/ml in DMSO, Cat#60014, Biotium) for 30 min at 37 °C. After two washing steps 500 cells/well for control LN229 and 1200 cells/well for LN229 SOX10-KD were seeded in a flat-bottom 96-well plate coated with 50 μl low melt agarose (Cat# M3049.0010, Genaxxon; 1% in PBS). After 3 days spheroids were collected and manually implanted in the brain slices using a blunt Hamilton syringe (701 N; 10 μl; 26 s/51/3) and a binocular microscope. Three days after implantation the brain slices were fixed with 4% PFA and cleared according to the SeeDB protocol82.Immunohistochemistry of mouse tumoursFor immunohistochemistry experiments, paraffin embedded (4 μm) tumour tissues were first subjected to deparaffinization and citric acid-based antigen retrieval was performed following standard protocols. Sections were either stained with hematoxylin and eosin (H&E) or subjected to immunohistochemistry Iba1 (Cat#019-19741, Wako, 1:2000). Immunohistochemistry images were obtained with light microscope (Zeiss, Germany) with 20× and 100x objectives. Iba1 staining quantification was performed with ImageJ (NIH) using random areas from the tumour core region. For immunofluorescence staining, sections were stained for GFP (Cat#13970, abcam, 1:500 and anti-Chicken IgY (H + L) secondary antibody, Alexa Fluor 488, Invitrogen A-11039, 1:1000) and immunofluorescence images of GFP were captured using Leica TCS SP8 confocal microscope with 20× objective.Quantification and statistical analysisStatistical analysis was performed using R and Students t-test. Kaplan-Meier analysis was performed to estimate the survival time of different GBM subgroups and a log rank test was used to test for differences of more than one survival curve. Details of the statistical tests applied are stated in the figure legends and the main text. In all Tukey-style boxplots, the box corresponds to the 25th, 50th/median and 75th percentiles and the whiskers denote 1.5× the IQR from the median. Outliers beyond 1.5× IQR are shown as points. Other boxplots indicate mean values ± standard deviation.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationDescription of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2Supplementary Data 3Supplementary Data 4Supplementary Data 5Supplementary Data 6Supplementary Data 7Supplementary Data 8Supplementary Data 9Supplementary Data 10Supplementary Data 11Supplementary Data 12Supplementary Data 13Supplementary Data 14Reporting Summary | nature communications | [
"Article"
] | [
"Cancer genomics",
"CNS cancer"
] | malignant brain cancer poor prognosis despite aggressive treatment surgical resection radiochemotherapy genomics identified four mRNA expression/DNA-methylation subtypes glioblastoma genetic lesions IDH glioma CpG island methylation phenotype hypermethylation mutations isocitrate dehydrogenase (IDH) 1 2 MES NF1 aberrations increased tumour infiltration macrophages RTK I PDGFRA gene amplifications (4) RTK II classical EGFR gene MES RTK I RTK II subtypes correspond mesenchymal proneural classical RNA expression subtypes4 refined IDH wildtype glioblastoma5 Transitions between subtypes observed during treatment worse unclear transitions due to tumour cell plasticity or expansion resistant subpopulations Master Regulators (MRs) tumour cellular proposed systematic characterisation cancer MRs understanding cancer biology therapeutic vulnerabilities12 studies transcriptomes epigenomes insights glioblastoma heterogeneity cellular state comprehensive genome-wide survey epigenetic landscape primary glioblastoma subtypes not availableglioblastoma subtype MRs interactions with effects GB epigenetics unknown present integrated epigenetic analysis of four subtypes adult glioblastoma performed methylome transcriptome epigenome profiling on 60 untreated patient tumours enhancers vary across subtypes identified 10 consensus subtype MRs Repression of RTK I MR SOX10 in human glioblastoma cell lines caused subtype transition to mesenchymal cellular state enhancers SOX10 loss leads to decrease survival increased tumour invasion immune cell infiltration GB subtype transitions on tumour phenotypes require further investigation glioblastoma epigenome selected 60 adult glioblastoma primary tumours 4 normal brain samples for DNA methylome transcriptome profiling key resources in Supplementary Table 1. Tumours subtyped using methylation microarray classifier four subtypes (IDH: 12 MES: 19 RTK I: 12 RTK II: 17) genotypes IDH wildtype in 90% of primary tumours IDH mutated representedSubtyping WGBS data IDH RTK II groups MES RTK I less distinct methylation gene-based features variable across subtypes MES RTK I subgroups TSS CGI methylation comparable methylation non-CGI intergenic regions similar MES RTK II subtypes 20 tumours profiled H3K27ac histone modifications chromatin immunoprecipitation sequencing mutation status IDH1 IDH2 determined pyrosequencing IDH samples had IDH1 R132H mutations G-CIMP6 remaining tumours IDH wildtype- IDHwt subtypes RTK I II exhibited glioblastoma copy number alterations gain chromosome 7 loss chromosome 10 CDKN2A/B deletion Amplifications EGFR chromosomes 19 20 prevalent in RTK II PDGFRA CDK4 MDM2 MDM4 amplifications frequent RTK I tumours. 1A dataset glioblastoma subtype epigenomicsmolecular profiling 60 glioblastoma tumours epigenetic subtypes characterised Master Regulators derived epigenome validated cell lines syngeneic mouse model Characteristics 60 glioblastomas age gender methylation subtype IDH1 IDH2 mutation status copy number aberrations Genome-wide differences DNA methylation subtypes control brain tissue Mean methylation beta-values 100 kbp genome bins difference control brain average DNA methylation genomic features 18-state ChromHMM model MYT1-seq expression subtypes Tukey boxplots Boxes 25th 50th 75th percentiles whiskers 1.5× IQR median Points outliers beyond 1.5× IQR NBr 4 IDH 12 MES 19 RTK 12 II 17 samples glioblastoma epigenomic landscape MYT1 oligodendrocyte marker gene per-subtype methylation beta-values ChromHMM annotation Control-neoplastic ChromHMM annotations Roadmap Epigenome DNA hypomethylation IDH RTK I subtypes correlates active TSS ChromHMM states methylation differences between glioblastoma subtypes normal brainregions genome hypomethylated in non-neoplastic brain tissue (Fig. RTK I tumours global hypomethylation IDH tumours globally hypermethylated G-CIMP phenotype genomic features hypomethylation intergenic regions CpG islands hypermethylated (Fig. results agree understanding methylation changes cancers used 18-state Roadmap Epigenome ChromHMM tumour histone mark ChIP-seq data genome defined consensus subtype ChromHMM states calculated mean subtype methylation (Fig. active TSS states (E01–E04) low methylation transcription repressive non-functional states higher methylation IDH subtype most hypermethylated RTK I most hypomethylated bivalent TSS enhancer states (E14–E15) broadest ranges tumour-specific hypermethylation differences methylation genomic loci defined function effects subtype-specific illustrated myelin transcription factor 1 regulator oligodendrocyte differentiation overexpressed in IDH RTK I glioblastoma hypomethylation active chromatin states gene promoter enhancer regionsenhancers variable across glioblastoma WGBS data methylation features differential methylation valleys partially methylated domains lowly methylated regions promoter heterochromatin enhancer states 60% PMDs DMVs shared across subtypes fewer than 17% one subtype 37% LMRs specific one subtype variable DNA methylation LMRs differences subtypes (Fig. Uniform Manifold Approximation Projection) LMR WGBS data separated subtypes (Fig. 2b c).Fig. 2Active enhancer-LMR regions show variable methylation across glioblastoma subtypes Barplots subtype sharing DNA methylation valleys lowly methylated regions partially methylated domains Uniform manifold approximation projection (UMAP plot glioblastoma samples DNA methylation LMRs UMAP plot glioblastoma samples DNA methylation enhancersstatistics DMVs LMRs PMDs mean DNA methylation beta boxes 25th 50th 75th percentiles whiskers 1.5× IQR regions ChromHMM state annotation enrichment IDH 12 MES 19 RTK I 12 RTK II 17 samples PMDs LMRs DMVs glioblastoma ChromHMM model features subtype (Fig. DMVs enriched TSS states (E01–E03 E14) PMDs quiescent repressive states (E16–E18) Chromatin states LMRs diverse largest proportion (23%) enhancer states (E07–E11 E15) LMRs one subtype enrichment enhancer states 36% importance enhancers subtype identities enhancers (E9–E10) 64% tumour active-enhancer regions unique tumours not shared normal 59% tumour-specific enhancers unique single subtype 6% shared all subtypes GB subtypes differing gene expression programmes subtype-specific enhancer activity regulatory circuitry analysis subtype Master RegulatorsEnhancer activity mediated transcription factor proteins MRs identify subtype MRs heterogeneous enhancer landscapeSuperenhancers cell identity genes including MRs23 performed SE calling H3K27ac glioblastoma profiles SEs subtype-specific enrichment RTK II tumours SE EGFR higher H3K27ac signal expression higher H3K27ac signal correlated target gene up-regulation regulate genes important identity glioblastoma subtype LMRs in 65% SEs regulatory circuitry analysis glioblastoma subtype Master Regulators Superenhancer) identification glioblastoma H3K27ac profiles Hockey stick plots enhancers ranked H3K27ac intensity four subtypes Selected SEs labelled target genes subtype SEs H3K27ac profiles CALCRL TGFBI GPR17 EGFR SEs coloured bars below H3K27ac profile-seq gene expression TPM + 1) genes subtype Tukey boxplots IDH 12 MES 19 RTK I 12 RTK II 17 samples Boxes 25th 50th 75th percentiles whiskers 1.5× IQR median Points mark outliers beyond 1.5× IQRboxplots gene expression target genes subtype ANOVA H3K27ac signal minimum fold change 1 Benjamini–Hochberg adjusted P-value comparing average 3. Mean log FC white diamond n target genes Boxes 25th 50th 75th percentiles whiskers 1.5× IQR median Points mark outliers beyond 1.5× IQR-tailed t-test P-value <2.2 × 10−16 gene signature enrichment results target genes size circle ratio target genes colour adjusted P-value subtype Master Regulator) 56 MRs predicted CRCmapper tumour H3K27ac profiles extended MR activity inference full tumour cohort 60 VIPER 56 network TCGA cohort 525 38 subtype MRs identified Heatmap mean subtype activity each MR annotated target genes mSigDB genesets enrichment glioblastoma subtype (Fig. genes upregulated neural crest stem cells enriched IDH MES RTK I downregulated RTK II Oligodendrocyte markers enriched RTK I astrocyte markers RTK II glioblastomas co-opt SE landscapes normal CNSregulatory circuit analysis identified 56 candidate MRs four subtypes extended analysis to full cohort gene expression-based approach (Fig. inferred glioblastoma gene regulatory network TCGA gene expression microarray cohort (n = 525 used VIPER activity 38 candidate CRC MRs MRs assigned subtype average maximum VIPER (IDH 9 MES 16 RTK I 4 II 9 Fig 3e 7) including MES MRs CEBPA STAT313 activity cell typesGlioblastoma mixtures tumour subtypes normal cell populations prevalent in addressed MR predictions analysis to single-cell RNA-seq profiles IDHwt glioblastoma25 performed pseudotime Cells formed 5-state trajectory four terminal branches one intermediate state (Fig. 4a). distinguish tumour normal cells cells similar gene expression programmes group together annotated state scoring cell normal brain cell cells glioblastoma subtype based VIPER-calculated MR activity (Figresults terminal branch assigned identity state 1 RTK I cells normal oligodendrocytes 3 RTK II astrocytes 4 mixture normal cell types 5 MES tumour cells macrophages.Fig. 4Validation core regulatory circuitry predictions glioblastoma single-cell RNA-seq Pseudotime trajectory inferred monocle QC-filtered single cells Darmanis glioblastoma dataset cells coloured inferred pseudotime state source sample original study Assignment glioblastoma subtype normal cell identities pseudotime states Normal brain cell type signature scores subtypes assigned VIPER-inferred activity CIMP- CRC MRs gene calculated each cell Visualisation relative MR activity (SREBF1 CEBPA SOX10 NR3C1 pseudotime trajectory scaled VIPER calculated each MR expression profile RTN-derived regulons Heatmap VIPER NES = 28 CIMP- CRC MRs split by state column single cell annotated source tumour sample samples clustered by activity profiles dashed white line subpopulations state 5 differing MR activitytSNE projection MES tumour TAM/microglial cells pseudotime state 5 (n = 2112) MES CRC MR = 13) activity matrix cell coloured macrophage markers AIF1 PTPRC scaled MR activity FLI1 STAT3 visualised relative MR activity scores pseudotime trajectory subtype CRC analysis single-cell level (Fig. some MRs ubiquitous activity subtype MRs higher activity (CEBPA SOX10 RTK NR3C1 clustered cells MR activity state (Fig. four MES MRs (CEBPA FLI1 MAFB MITF separate two subpopulations state 5. high activity MR heterogeneity MR activity discriminate cell types examined 2112 cells state 5. t-SNE projection MES MR (n = 13) VIPER activity matrix splits state two sub-clusters Visualisation microglial/macrophage marker gene expression (AIF1 PTPRC major population infiltrating immune cells remaining population MES glioblastoma cells MRs CEBPA FLI1 MAFB MITF higher activities population STAT3 mesenchymal transformation active both populationsanalysis of scRNA-seq data consistent with bulk tissue data identified three tumour branches subtype MRs MES subtype enriched with infiltrating immune cells scRNA-seq data MR activity in tumour cells macrophages.Loss of SOX10 results in RTK I-to-MES analysis analysed gene expression-based GB regulatory networks using Reconstruction of Transcriptional Networks Analysis of Master Regulators (RTN) network used network inferred additional 569 microarray samples cross-validate predictions defined glioblastoma subtype RNA expression signatures 117 subtype MRs active in same subtype same direction activity both networks 11).Overlap RNA-based predictions with chromatin-based CRC MRs gave consensus list of 10 MRs selected SOX10 for functional validation SOX10 oligodendroglial lineage transcription regulates epigenetic state linked to chromatin remodelling resistance in melanoma makes SOX10 interesting candidate to study epigenetic control remodelling subtype gene regulation in glioblastomaSOX10 cell fate neural lineage development mice transcription factor NFIA astrocytic differentiation overexpression mouse tumour. 5SOX10 Master Regulator RTK I subtype Correlation DNA methylation SOX10 expression Boxes Tukey plots 25th 50th 75th percentiles whiskers 1.5× IQR median Points outliers beyond 1.5× IQR.IDH 12 MES 19 RTK I 12 RTK II 17 samples Epigenome landscape SOX10 glioblastoma Per-sample methylation mean H3K27ac intensity GSEA plots enrichment proneural mesenchymal gene signatures control SOX10 KD LN229 cells tumour cell-specific Wang et al. GSEA-calculated statistics gene set enrichment P-values FDR values computed permutation test enrichment score EnrichedHeatmap genome regions differential chromosome accessibility LN229 control SOX10 KD cells ATAC-seq analysis SES-normalised signals SOX10 ChIP-seq ATAC-seq BRD4 ChIP-seq Signal intensity blue–red heatmaps single ATAC peak vertical lines 1 kbp 5′ 3′lineplots heatmap display signal intensity regions green SOX10 KD Volcano plot motif HOMER ATAC-seq peaks LN229 cells enriched motifs labelled ChromHMM annotations LN229 ATAC-seq peaks Active TSS (E01–E04) Enhancer (E07–E11) states NT SOX10 KD conditions Western blot SOX10 BRD4 co-immunoprecipitation cell LN229 Factor co-occupancy SOX10 peaks LN229 SES-normalised signal peak regions 1 kbp downstream SOX10 BRD4 H3K27ac H3K4me1 scaled Changes SOX10 BRD4 binding ATAC chromatin accessibility RTK I subtype genes SOX10 ERBB3 SES-normalised ChIP-seq ATAC-seq signal NT SOX10 KD conditions LN229 ZH487 cell lines boxes indicate regulatory regions SOX10 BRD4 ATAC-seq signal.SOX10 over-expressed RTK I tumours genic hypomethylation increased H3K27ac signal glioblastoma cell lines selected two lines high SOX10 expression hypomethylation ZH487 LN229SOX10 ChIP-seq identified sites shared two cell lines models SOX10 activity Fig. 4).Suppression SOX10 expression leads changes RNA expression analysed gene set enrichment analysis selected tumour-specific genes 12 SOX10 suppression RTK I-to-MES transition in LN229 ZH487 cell lines confirmed applying proneural mesenchymal classical subtype gene signatures tumour cell effects (Fig. 5d Fig 5b). VIPER-inferred activity levels RTK I 3 MES 15) MRs correlated with control SOX10 KD conditions 5c). RTK I-MES transition increased cell invasion in trans assay ex vivo brain slice assays35 5d RTK I-to-MES transition occurred.SOX10 repression remodels glioblastoma enhancer chromatin effects SOX10 loss RTK I-to-MES transition mediated via chromatin changesATAC-seq analyses chromatin accessibility decreased at RTK I MR loci SOX10 SOX8 ERBB3 increased at MES MR loci RUNX213 FOSL2 following SOX10 suppression SOX10 binding sites in regions increased accessibility control SOX10 KD cells (89% vs 49%) (Fig. 5e Differential motif enrichment analysis SOX motifs in control predicted MES MR motifs RUNX in SOX10 KD cells (Fig. 5f 6b).ChromHMM chromatin-accessibility changes enhancer TSS states importance of enhancers for subtype identity analysis LMRs SOX10 binds to enhancers in RTK I cells analysed occupancy Bromodomain (BRD4) protein marker Mapping BRD4 binding-seq regions redistribution following SOX10 changes chromatin accessibility remodelling active enhancer landscape (Fig. inhibition BRD4 binding up-regulation MES MR RUNX2 following SOX10 RTK I-to-MES transition dependent on enhancer dynamics6d–f). Co-immunoprecipitation confirmed SOX10 BRD4 interact (Fig. SOX10 binding sites showed BRD4 binding histone modifications enhancers (Fig. 5i SOX10 recruits co-factor to RTK I enhancers observed loss BRD4 binding chromatin accessibility at regions RTK1 genes following SOX10 repression (Fig. 5j results suggest SOX10 maintains RTK I cellular state genes Loss SOX10 chromatin accessibility changes enhancer remodelling release BRD4 from enhancers unclear factors recruiting BRD4 to MES enhancers.SOX10 repression mesenchymal phenotype in glioblastoma increased tumour cell invasion immune cell infiltration immunocompetent syngeneic graft mouse model of glioblastoma role SOX10 Repression SOX10 faster tumour growth median survival time mice 104 days 63 days Fig. 6Loss SOX10 induces mesenchymal phenotype MRI images of mouse brains SOX10 tumours 57 days post injection Median tumour volumes (μl 57 days post-injectionNT 1.85 μl 10 SOX10 KD 50.3 μl 9 One-sided t-test P 8.65e-5 plots 25th 50th 75th percentiles whiskers 1.5× IQR Points outliers beyond Kaplan-Meier survival analysis NT 104 days 10 SOX10 63 days 9 Two log-rank test P 4.89e-5 H&E stainings NT SOX10 tumours two Scale bars 200 100 μm Staining tumour margins 2 control 2 SOX10 KD tumours DAPI antibodies GFP Scale bars 100 μm RNA expression myeloid microglia genes 3 5 tumours 25th 50th 75th percentiles whiskers 1.5× IQR Points outliers beyond 1.5× IQR Immunohistochemistry staining Aif1 tumour margin bulk Red boxes close-up Scale bar 100 μm two animals group 5 fields Quantification Aif1 staining areas 10 fields 3 Mean standard deviation Two-sided t-test P = 0.025.H&E staining-brain showed control tumours defined marginal invading tissue SOX10 KD cells boundaries disrupted. supported staining GFP expressed tumour cells better-defined tumour margins control knockdown tumours suggest increased invasion normal tissue cells after SOX10 repression vitro trans-well invasion ex vivo brain slice assays 5d–f).RNA profiling revealed increased expression markers TAMs resident microglia (Aif1 Itgam Cd68 Cx3cl1) macrophage M1/M2 polarisation knockdown tumours immunohistochemistry staining identified increased Aif1 positive cells SOX10 KD control tumours tumour margins Aif1-positive cells microglia-like roundish tumour bulk Aif1 staining increase tumour macrophage infiltration SOX10 KD tumours (Aif1 positive area NT 4.57% SOX10 KD 17.11% 10 fields 3 tumours P < 0.001) SOX10 repression causes switch mesenchymal state increased immune-cell infiltration decreased survival timereturned primary tumour data SOX10-associated RTK I-to-MES transition samples Clustering 5000 microarray probes 31 tumours identified 2 subtype clusters gradient methylation (Fig. SOX10 expression higher in RTK I MES tumours same trend proneural MES5 gene expression MES tumours higher myeloid marker gene expression I differential H3K27ac enrichment RTK I MES SEs expression SE-defined subtype identity genes MR activity correlated methylation gradient RTK I MES tumour patients different survival low SOX10 expression MES-subtype glioblastoma correlated adverse prognosis (Fig 7i data support gradient SOX10-dependent molecular characteristics in human glioblastoma. 7Genetic epigenetic patterns RTK I-to-MES transition in primary glioblastoma tissues Clustering 5000 probes RTK I MES tumours 31 identified 3 clusters RTK I SOX10 RNA-seq expression Wang PN MES subtype signature ssGSEA score Relative expression Wang PN MES subtype marker genes myeloid cell marker genesRTK I MES SEs differential H3K27ac enrichment-statistic Expression score target genes 422 RTK I 279) MR activity RTK I 4) MES CRC MRs Kaplan-Meier survival curves RTK I MES subgroups overall P 0.030) progression-free survival 0.060)-Meier survival curves MES tumours = 132 TCGA glioblastoma cohort SOX10 expression Overall survival P = 0.036) progression-free survival P = 0.009) Group cut-off average expression statistical significance two-sided log-rank test study glioblastoma subtypes distinct enhancer landscapes Master Regulator (MR) repertoires SOX10 RTK I MR repression RTK I-to-MES transition remodelling enhancer landscape repression SOX10 syngeneic mouse model in vivo phenotypes altered growth patterns increased immune cell content decrease survival studies glioblastoma RRBS9 data genome study generated WGBS data large glioblastoma primary tumours CpG methylation genome-wide subtype methylation differencesG-CIMP hypermethylation IDH glioblastoma limited CpG islands affects genomic features functional chromatin states analysis highlighted importance LMR enhancer methylation differentiating glioblastoma subtypes identified MRs enhancer landscapes using chromatin transcriptome data comprehensive analysis smaller studies importance enhancers promoters subtype used RNA transcription-based subtype classification schemes differentiate IDHwt IDH-mutated proneural-subtype tumours “Neural” subtype partially consistent with concepts glioblastoma difficult relate validated subtype MR predictions using scRNA-seq data IDHwt cell states normal cells three IDHwt glioblastoma subtypes single-cell studies glioblastoma cell state xenotransplantation studies single glioblastoma cell regenerate cellular heterogeneity analysis implicates enhancers subtype MRs key cell state plasticity analyses identified 10 high-confidence consensus subtype MRs selected RTK I MR candidate SOX10 for functional characterisation member developmentally important SRY-related HMG-box (SOX familyCNS SOX10 oligodendrocytic marker gene tumour cells pathway escape terminal cell differentiation melanocytes SOX10 binds promoters distal chromatin-modifying co-factors SMARCA438 Chd736 SOX10 maintains RTK I transcriptomic state subtype genes loss shift to mesenchymal phenotype transition dependent on enhancer remodelling blockade BRD4 activity JQ1 inhibition recurrent glioblastomas10 reverse MES-to-RTK I transition forced by depletion AP1 transcription factor FOSL1 NF1-mutant RTK I-to-MES transition induced by correlates with resistance MES glioblastoma worst prognosis in vivo results loss of SOX10 changes myeloid cells decrease survival correlation low SOX10 expression adverse survival MES-type glioblastoma loss SOX10 linked to adverse outcomes in neural crest-derived tumours loss in melanoma leads to transcriptome rewiring drug neural crest cells undergo analogous mesenchymal transition during normal development hijacking developmental pathways by cancerNeural lineage development regulated by SOX transcription including SOX2 marker of proneural glioblastoma subtype regulator cell plasticity astrocytic differentiation57 In tumour cohort SOX2 RNA-expression 2× lower in MES remained high 3× higher than normal brain SOX2 showed H3K27ac activation in all glioblastoma subtypes. RNA expression change in SOX10 knockdown models no evidence subtype-specific role SOX2 RTK I-to-MES transition loss of SOX10 SOX9-dependent activation SOX10 function in astrocytic differentiation tumour identified SOX9 as RTK II MR observed upregulation SOX9 in SOX10 knockdown models loss of SOX10 in upregulation NFIA RNA expression explanation RTK I-to-MES II transition after SOX10 suppression subtype MR analysis identifies candidates beyond SOX10 in CNS development suggesting MR activity interacts with genetic epigenetic factors glioblastoma cell state Evidence of plasticity genetic regulatory features RTK I-to-MES transition visualised in layers data from primary tumourstumour cell components microenvironment complexity plasticity suggested vivo experiments elucidating mechanisms myeloid-cell invasion immune suppression for immunotherapy approaches findings mirrored publication knockdown SOX10 cell melanocyte to mesenchymal-like in melanoma microenvironmental cues melanoma cell subtype plasticity failure in glioblastoma drug combinations targeting tumour cell growth epigenetic plasticity block escape cancer cells therapy-resistant state improved outcomes tissue samplesSnap-frozen primary glioblastoma tumour samples clinical data collected diagnosis between 1994 2011 Burdenko Neurosurgery Institute Informed consent obtained from patients approved by ethics board patient cohort 32 males 28 females average age 52.5 ± 11 years Patients IDH-subtype tumours younger than MES RTK I RTK II subtypes (42.6 9.5 vs 55.0 9.9 years IDH1/2 mutation status determined using pyrosequencing or Sanger sequencing60Samples post-mortem normal brain purchased Biocat (Heidelberg lines culture human glioblastoma cell line LN229 (p53 p16 established 60-year female 1979 obtained ATCC cultured DMEM 10% FCS Penicillin/Streptomycin glutamine ZH487 patient-derived glioblastoma cells established University Zurich Hospital cultured Neurobasal medium#12348017 2% B27 EGF FGF glutamine HEK293T cells lentivirus production monolayer cultures antibiotic-free DMEM 10% FCS cells cultured 10% CO2 37 °C humidity Cell line identities verified Multiplex Cell line Authentication Test tested mycoplasma contamination Multiplex cell test vivo syngeneic mouse modelAnimal experiments ethical regulations approved Regierungspräsidium Karlsruhe primary mouse glioblastoma cell line Pten/Tp53 double knockout established Prof. Peter Angel German Cancer Research Center mGB1 cells Proneural/RTK I subtype high SOX10 expression 37 °C DMEM/F12 medium N2 supplement EGF FGF Penicillin/Streptomycin glutamineSOX10 knockdown lentivirus transduction shSOX10 (TRCN0000244290 Table 3) cells Puromycin selected SOX10 RT-PCR validated injection 200k cells shSOX10 1 μl volume intracranially injected adult C57/B6 mice (6 weeks female Isoflurane MRI scanning DKFZ MRI 57 days post genome bisulphite sequencing bisulphite preparation 5 μg DNA sheared Covaris device DNA fragments lengths 200–250 bp isolated E-Gel electrophoresis bisulphite converted EZ DNA Methylation kit fragments PCR amplified FastStart High Fidelity PCR 6–8 cycles Library aliquots purified size selected AMPure beads quality controlled Bioanalyzer library sequenced 2 lanes Illumina HiSeq 2000 DKFZ Genomics Proteomics (500 ng per sample submitted DKFZ Genomics Proteomics library preparations standard protocol Illumina 1 lane Illumina HiSeq.RNA sequencingPrimary RNA-seq libraries strand specificity rRNA HiSeq 2000 1 lane per samplesamples profiled by WGBS sequencing WGS 450k/Epic methylation microarray RNA sequencing genotyped in silico exclude swaps.RNA-seq GB cell lines (LN229 ZH487 Control vs. shOX10) performed polyA-selected RNA-seq libraries TruSeq RNAseq Illumina kit DKFZ genomics Proteomics Libraries multiplexed-sequenced 1 lane HiSeq 2000 v4 50 bp single-end reads-seq mouse tumour samples (shNT n = 3; shSOX10 n = 5) used PolyA protocol multiplexed 2 lanes HiSeq 2000 v4 50 bp reads 10 μg H3K27Ac for ChIP library preparation GB patient samples Active Motif Libraries multiplexed sequenced 1–4 lanes Illumina HiSeq 2000 platform LN229 experiments cells-shSOX10 cross-linked with 1% methanol-free formaldehyde 10 min. cells washed three times with PBS cell pellet treated 4 U MNase per 1 × 106 cells 15 min.MNase stopped Covaris buffer chromatin sheared 15 min LE220 Covaris soluble chromatin recovered quantified 2 μg immunoprecipitation 2 μl antibody chromatin digested proteinase K purified AMPure beads purified DNA cloned sequencing libraries NEBNext SOX10 BRD4 ChIP-seq experiments cells cross-linked 1% methanol-free formaldehyde 15 min quenched 0.125 M glycine Chromatin isolated lysis buffer Dounce homogenisation chromatin sheared sonication 300–500 bp Input genomic DNA RNase proteinase K de-crosslinking isolated ethanol precipitation Pellets re-suspended DNA quantified NanoDrop spectrophotometer chromatin yield calculated 30 μg chromatin pre-cleared protein A agarose beads precipitated 4 μg antibody SOX10 BRD4 isolated washing SDS buffer elution RNase/proteinase K treatment de-crosslinking purified-chloroform extraction ethanol precipitationSequencing libraries prepared-polishing dA-addition ligation Apollo 342 NGS Library Prep system Biosystems libraries sequenced (50 bp-end Illumina HiSeq 2000.ATAC biological duplicates viable frozen cells incubated Tn5 0.1% Igepal CA-630 (37 °C Transposition stopped EDTA DNA purified AMPure beads barcodes added PCR re-purified beads libraries sequenced Illumina HiSeq 2000.Gene expression microarray profiling-treated RNA (500 ng) prepared profiling Affymetrix HG-U133-Plus2 Illumina HumanHT-12 v4 Expression BeadChip microarrays DKFZ Genomics Proteomics Core Facility GSEA results Fig. 5d Fig. 5b generated Affymetrix Human U133Plus 2.0 platform LN229 cells control non non-targeting three RNA ZH487 cells non-treated control shNT three shSOX10 repression microarray data processing DNA methylation processed analysed minfi (1.24.0 conumee (1.3.0)64 Bioconductor packages > 500 ng DNA snap-frozen samples input materialminfi signal intensities IDAT colour channels corrected background dye-bias Beta values calculated offset 100 CNVs called standard conumee procedure two sets 50 control samples balanced CN profiles Copy number aberrations called conumee-processed values thresholds x < −1 deletion −1 < < −0.2 loss −0.2 < < 0.2 no change 0.2 < < 1 gain > 1 amplification.Subtype classification patient samples methylation 8000 probes methylation microarray data 450k EPIC microarrays 7386) methylation beta value matrix sample pairwise Euclidean distance samples ‘ward.D’ method.Clustering MES RTK I methylation MES 19 RTK I 12) glioblastomas clustered. 7) substructure 5000 variable probes selected clustered Euclidean distance ‘ward.D’ method.RNA-seq processing expression quantificationReads aligned reference genome/mm10) Gencode reference transcriptome (v19/M2) STAR Read counts gene quantified total exons htseq-count (0.6.0) human mouse samples Gene expression values quantified transcripts per million (TPM)Tumour RNAseq limma gene signature counts gene pre-filtered genes > 10 reads > 6 samples Normalisation factors calculated calcNormFactors ‘edgeR’ (3.20.1). voomWithQualityWeights (3.34.4) counts limma differential expression analysis gene expression GBM subtype versus 3 Genes significant passed BH adjusted P-value threshold 0.001 Signatures mouse normal brain cell subtype signatures enrichment gene tested ‘ssgsea’ method R package (1.26.0) TPM expression values ESTIMATE65 (1.0.13) immune stromal cell content Genes up-regulated FC > 0 P-value < 0.05) annotated with Gene Ontology terms 3.5 enrichGO function R package “clusterProfiler” (3.6.0) MR activity network A inferred package VIPER (1.14.0 reads mapped human genome bwa-mem (0.7.8) WGBS CpGs overlapping variable sites minor allele frequency higher 0.25 removed Low coverage CpGs with 2 or fewer reads 50% cohort removedmean methylation two cytosines CpG dinucleotide calculated weighting CpG coverage m = (m1*c1 + m2*c2)/(c1 + c2) m1 m2 methylated CpGs c1 c2 CpG coverage mean coverage CpG dinucleotide calculated weighting coverage c = (c1*c1 + c2*c2)/(c1 + c2) bsseq R package (1.10.0) smooth methylation data missing methylation values parameters.Methylation feature (DMVs LMRs PMDs segmentation methylation features lowly methylated regions DNA methylation valleys performed customised pipeline chromosomes split into blocks inter-CpG distance mean standard deviations methylation low intermediate high methylation DMVs LMRs defined characteristics block PMDs called MethylSeekR (1.14.0)66 10 kbp minimum Subtype consensus regions defined cross-sample coverage 4. Neighbouring LMRs merged if inter-LMR distance less than 1 kb. Enrichments genomic features computed Jaccard Coefficient base-pair length setsstatistic compared to 1000 CpG content-matched regions calculate z-scores P-values for enrichment Classification samples methylation patterns using projection methylation matrix uniform manifold approximation projection.Subtype sharing methylation featuresFor consensus methylation features determined by selecting genomic regions in more than 50% samples extent of subtype sharing methylation features calculated as of total width of regions in 1 2 3 4 subtypes.Chromatin state enrichmentFor Jaccard coefficient overlap of two sets regions calculated as total base pairs divided by base pairs union significance Jaccard coefficient calculated by permuting methylation features in specific genomic background where average CpG content similar selection of background regions methylation features split into small windows window size w calculated as 25th quantile of widths methylation features w rounded to thousand digit window size set to 10 kb if larger 1 kb if smaller 1 small windows cause bias windows width less than w/4 removed filtering CpG content defined as number sites per 1 kb window denoted as p.background regions genome split by windows CpG content calculated windows CpG content between 5th 95th percentile selected background regions methylated regions randomly permuted bedtools (v2.27.1) 1000 times Jaccard coefficient calculated z-score calculated as (s-μ)/σ enrichment s Jaccard coefficient μ σ mean standard deviation Jaccard coefficient permutations.ATAC-seq ChIP-seq datasets processed custom pipeline Snakemake (v. 3.13.3). reads trimmed Trimgalore tool aligned Bowtie268 (v. 2.3.4.3) standard parameters Duplicates multi-mapping reads removed samtools package XS flag files ChIP-seq data control IP datasets scaled SES method converted into bigwig track bamCompare tool deepTools2 ATAC-seq data genome-wide coverage calculated Peaks called callpeak mode MACS2 (v. 2.1.1.20160309) SICER70 used peaks gap 600 window 200 parameters QC parameters (FRiP PCR bottleneck coefficient cross-strand correlation determined ENCODE guidelines71visual QC performed signal profile TSS genes fingerprint method deepTools2.Chromatin segmentation defined ChromHMM (v. 1.19) tool ChIP-seq (H3K27ac input data binarized ChromHMM’s “BinarizeBam” command Roadmap Epigenome 18-state model19 genome consensus state highest frequency minimum frequency 50%.Superenhancer union H3K27ac peaks input regions ROSE2 superenhancer analysis pipeline regions within 12.5 kbp H3K27ac signal calculated ‘bigWigAverageOverBed’ enhancers ranked by subtype average enrichment SEs defined default parameters ROSE2. SEs combining four subtype SE lists ANOVAs H3K27ac signal intensities minimum log fold change 1 Benjamini–Hochberg adjusted P-value threshold 0.1 comparisons MES RTK I SE activity two statistics calculated H3K27ac signal intensities t-statistic distributions computed subtype-specific SE lists for MES = 422) RTK I 279) defined SEs no overlap “subtype-specific SE gene score” calculated mean expression targetscomparison between SEs LMRs overlapping requiring 50% overlap.Core regulatory circuit determined using modified CRCmapper assigning SE to closest gene computed Spearman correlation of H3K27ac signal on SE over all samples expression genes within 500 kb around SE gene in same topological domain assigned highest correlated gene as SE target gene CRCmapper procedure remains autoregulatory TFs identified selecting target genes SE SEs contain binding motifs for corresponding TF cliques of autoregulatory TFs identified SEs binding motifs for all other TFs.Gene regulatory network inference with RTNTwo cohorts of glioblastoma gene expression microarray data collected A (from TCGA 525 samples Affymetrix HG-U133A B studies E-MTAB-3073 GSE4290 GSE7696 GSE16011 GSE43378 569 samples raw data read into R normalised using ‘gcrma’ package (2.50.0). Study-associated batch effects removed from cohort B using ComBat function in ‘sva’ (3.26.0) specifying study ID as ‘batch’ option. ‘RTN’ (23.4 used for steps analysis expression gene measured by microarray probes probe highest coefficient variation kept for analysis Regulatory relationships between = 1333 TFs target genes inferred using ARACNe direction TF-target gene regulation or inferred using Pearson correlation TF-target edge P-values calculated permuting Mutual Information matrix 1000 times retaining edges BH-adjusted P-value < 0.01. network bootstrapped 100 times TF-target edges in 95% samples retained indirect TF-target edges removed Data Processing Inequality (DPI) filter 0.Identification of subtype Master Regulators with limma subtype gene expression signatures pre potential subtype MRs TFs passing tested in 1-tail GSEA TFs regulating fewer than 15 genes removed Significant TFs with BH-adjusted P-value < 0.01 10,000 retained Subtype MRs identified using 2-tailed GSEA test.gsea2 limma-voom fold change TFs with BH adjusted P-value < 0.01 10,000 permutations significantly active in subtypenetwork MRs (n = 117) defined passing significance threshold same subtype direction activity measured 2-tail GSEA differential Enrichment Score both networks 2-tail GSEA dES calculated each subtype signature consensus MRs results visualised ‘ComplexHeatmap’ package two networks (A B compared 2-tail GSEA dESs each subtype calculated all TFs networks (n = 512). correlation calculated Spearman’s rank correlation visualised ‘ggplot2’ package (2.2.1).Single-cell RNA-seq counts matrix dataset (GSE84465) analysed ‘monocle’ (2.10.1) package (v3.5.1) most variably expressed genes minimum mean expression 0.1 reduce tSNE cells clustered (ρ = 40 = clustering genes 10% cells differentially expressed clusters top 1000 ranked statistical significance ordering genes pseudotime analysis Normalised expression values (variance-stabilising transformed) used downstream analyses Normal brain cell population signatures McKenzie et al.28 score cells each gene background defined 100 genes smallest expression differencematched background subtracted from each signature gene sum defined as score Bioconductor package ‘viper’ (v1.14.0)24 single-cell MR activity VST-normalised expression matrix pre-filtered genes SD bottom quartile MR regulons network A calculate VIPER NES visualised in tSNE pseudotime plots MR activity profiles VIPER visualised in heatmaps ‘ComplexHeatmap’ Euclidean distance average clustering method Subtype MR scores defined transforming NES into z-scores mean predicted subtype maximum mean z-score Cells in pseudotime State 5 separated using tSNE 0.13) MES MR activity matrix Gene expression MR activity visualised pseudotime analysis RTK II MR ZBTB7C not included regulon small in RTN network A line expression microarray data processing Affymetrix microarray data read into R normalised ‘gcrma’ R package (2.50.0). Probes without annotated gene removed batch effect removed ZH487 samples samples combined into groups control knockdown differential expression analysis genes adjusted P-value < 0.05 significantly dysregulatedMR activity inferred expression profile VIPER (v1.14.0 analysis-calculated FC profiles LN229 ZH487 cell lines SOX10 repression analysed (v3.0). Glioblastoma signature gene sets downloaded GSEA website Results visualised Volcano plots activity MES RTK CRC MRs calculated VIPER RTN-inferred TCGA network A (n = 525 average replicates condition MR activity profiles clustered Euclidean distance ‘ward.D2’ method analysis TF activities SOX KD (Fig S5C two-sided t-test WT SOX10 KD TF activities.ChIP-seq ATAC-seq analysisMACS2 peaks input (v3.4.3) ATAC-seq consensus peakset defined two replicates merged peaks findOverlapsOfPeaks R/Bioconductor Signal intensity calculated ‘bigWigAverageOverBed’ SES-normalised.bigWig file visualised ‘EnrichedHeatmap’ package (1.9.2). differential ATAC peaks identified R/Bioconductor package ‘DiffBind’ (2.6.6 FDR threshold 0.05. Peaks annotated largest state overlap LN229 cell line ChromHMM statesStates collapsed E01-E04 TSS E07-E11 Enh.ATAC-seq motif findingHOMER (v4.9.1) motif finding default settings background union ATAC peaks genomic circlize package (0.4.6)78 methylation differences chromatin states transitions ComplexHeatmap (1.19.1)79 plots EnrichedHeatmap (1.9.2)80 epigenetic signals genomic regions Genome browser views WashU Epigenome Browser Gviz (1.22.3)81 epik package integrative visualisation analysis.SOX10 knockdown LN229 cells gene expression microarray experiments 20 sgRNA sequences CRISPR targeting genomic window −50 to +200 bp transcription start site NCBI sgRNA oligonucleotides cloned 5′ BstXI-BlpI 3′ pU6-sgRNA EF1Alpha-puro-T2A-BFP expression plasmid additional sequences ends Table 3 dCas9-expressing LN229 cells transduced lentivirus SOX10 RNA MOI 2.5 Positive transduced cells selected puromycin (1 μg/ml) 48 h SOX10 knockdown evaluated 4 days after transductionSOX10 cells RNA ChIP-seq ATAC-seq experiments LN229 cells pLKO-Tet-On shRNA SOX10 shRNA lentiviral particles puromycin (1 μg/ml 7 days expression induced 1 μg/ml doxycycline 7 days Cells cultured DMEM 1 g/l glucose 10% tetracycline-free fetal bovine serum 1% penicillin streptomycin glutamine (0.5 SOX10 knockdown ZH487 cells shRNA lentivirus infection system 2 3) Supplementary Table 3 cDNA synthesis qRT-PCRTotal RNA isolated RNeasy Micro kit 1000 ng reverse transcribed hexamer primers QuantiTect Rev. Transcription Kit cDNA sample analysed triplicate Applied Biosystems Prism 7900HT Fast Real-Time PCR System SYBR Green ROX Mix mRNA normalised ARF1 DCTN2 mRNA Expression levels calculated ΔCt method sequences Supplementary Table 2.ChIP LN229 BRD4 ChIP-PCR experiments 10 million cells per conditionnM 6 cross-linked 1% methanol formaldehyde 10 min quenched 0.125 glycine Chromatin isolated cell lysis buffer (50 mM HEPES mM NaCl mM EDTA glycerol 0.5% 0.25%Triton-100 protease inhibitor cocktail chromatin sheared low SDS buffer (0.1% SDS 1 mM EDTA Tris pH 8.1 300–500 bp Covaris S2 system 5% 4 5 genomic DNA RNase proteinase K de-crosslinking purified QIAquick PCR purification kit diluted 10% Triton X-100 5 M NaCl 0.1% SDS 1 mM EDTA 10 mM Tris pH 8.1 1% Triton-100 150 mM chromatin pre-cleared magnetic protein A/G beads immunoprecipitation 10 μg antibody BRD4 overnight incubation 4 °C 20 μl resuspended Magnetic Protein A/G Beads incubated 2 h DNA purified Low Salt Wash Buffer.1% SDS 1% Triton-100 2 mM EDTA High Buffer1%SDS Trtiton X-100 2 mM EDTA 20 mM Hepes-KOH pH7.9 500 mM LiCl Wash Buffer (100 mM Tris-HCl pH 7.5 0.5 M LiCl 1%NP-40 1%Sodium Deoxycholate TE buffer (10 mM Tris-HCl pH 8.0 0.1 mM magnetic beads DNA eluted buffer (10-HCl 8.0 0.1 EDTA 1% SDS RNase/proteinase K °C 2 h DNA purified QIAquick PCR ready qRT-PCR analysis ChIP primers Supplementary Table 2. binding proteins gene calculated recovery blottingCells washed lysed RIPA lysis buffer (0.5% SDS protease inhibitors sheared supernatant collected protein concentration protein lysates diluted 0.5 μg/μl NuPAGETM LDS Sample Buffer reducing agent boiled 95 °C 5 min 5 μg protein samples 4–12% Bis-Tris Protein gels proteins transferred PVDF membrane blocked 5% skimmed milk BSA 1 h incubated anti-SOX10 anti-BRD41:2000 anti-alpha-Tubulin#T9026 1:5000 dilution 5% milk 4 °C overnight shaking membranes washed TBS-T 10 min repeated three times horseradish peroxidase antibodies-mouse IgG anti-rabbit anti-goat IgG added incubated 1 h room temperature membranes washed TBS-T three times 30 min ECL Signals detected light-sensitive film Alpha-tubulin loading control images western blots Supplementary Fig. 7.Co 10 million LN229 cells condition harvested washed PBS lysed Pierce IP lysis buffer#87787 protease inhibitor cocktail#11836170001 Samples incubated 30 min intermittent vertexing supernatant collected pre-cleared Protein G Dynabeads#10003D Protein concentration determined Input sample collected (500 μg 10 μg SOX10 antibody#155279 per condition immunoprecipitate protein Protein G beads added lysate overnight incubation cold roomsupernatant removed beads washed TBST resuspended 20 μl 2× buffer Samples incubated 70 °C 10 min supernatant collected separation 10 μl buffer added incubated 95 °C 5 min supernatant collected combined eluate resolved SDS-PAGE proteins visualised Western Blot vitro invasion 200,000 cells seeded 6-well plate transduced shRNA SOX10 NT control expanded 4 days collected counted seeded 36 h Neurobasal growth medium B27/EGF BiocoatTM Matrigel invasion chambers (8 Bioscience Invasion incubation full growth medium B27/EGF/FGF Non cells removed remaining fixed stained haematoxylin Images light microscope 100× magnification vivo brain slice invasion 6–8 week old C57Bl/6 N mouse euthanized brain isolated cerebellum removed scalpelbrain cut in 350 μm slices vibratome (Leica VT1200 slices cultivated filter#PICM03050 6-well plate MEM M2279 25% heat-inactivated horse serum 25 mM HEPES 1 mM L-glutamine 5 mg/ml glucose 100 U/ml penicillin/streptomycin Control SOX10-KD LN229 cells treated Dox 7 days LN229 SOX10-KD glioma cells medium D5671 10% FBS 2 mM L-glutamine trypsinized counted 106 cells/ml PBS incubated 5 μl lipophilic dye DiD 30 min 37 °C two 500 cells/well control LN229 1200 cells/well LN229 SOX10-KD seeded-bottom 96-well plate 50 μl low melt agarose 3 days spheroids collected implanted brain slices Hamilton syringe binocular microscope Three days after implantation slices fixed 4% PFA cleared SeeDB.Immunohistochemistry mouse paraffin embedded (4 μm) tumour tissues deparaffinization citric acid antigen retrievalSections stained hematoxylin eosin subjected immunohistochemistry Iba1 (Cat#019-19741 Wako 1:2000). Immunohistochemistry images obtained light microscope (Zeiss Germany 20× 100x objectives Iba1 staining quantification ImageJ (NIH areas tumour core immunofluorescence staining sections stained GFP (Cat#13970 1:500 anti-Chicken IgY (H L) secondary antibody Alexa Fluor Invitrogen A-11039 1:1000 immunofluorescence images GFP captured Leica TCS SP8 confocal microscope 20× objective.Quantification statistical R Students t-test Kaplan-Meier analysis survival time GBM subgroups log rank test differences survival curve Details statistical tests figure legends main text Tukey-style boxplots box 25th 50th 75th percentiles whiskers 1.5× IQR median Outliers beyond 1.5× IQR points boxplots mean values standard deviation Nature Research Reporting Summary.Supplementary | 49.6 | 1.600644 |
10.1038/s41467-020-16477-1 | PMC7244735 | The synthesis of vinylboronates and alkylboronates often suffers from step-tedious and poorly stereoselective procedures. Here, the authors report a bench-stable redox-active reagent for the radical difunctionalization of alkenes and alkynes affording fluorine-containing vinylboronates and alkylboronates. | Vinylboronates and alkylboronates are key components in variegated transformations in all facets of chemical science. The synthesis of vinylboronates and alkylboronates suffers from step-tedious and poor stereoselective procedures. We have developed a regulated radical difunctionalization strategy for the construction of fluorine-containing vinylboronates and alkylboronates with an integrated redox-active reagent IMDN-SO2RF. This bench-stable imidazolium sulfonate cationic salt offers a scalable and operational protocol for the fluoroalkylation-borylation of unsaturated hydrocarbons in a high regio- and stereoselective manner. The products can be further transformed into valuable fluorinated building blocks. | IntroductionDifunctionalization of alkenes and alkynes has been widely explored for rapid diversification of double/triple bonds1–6. Traditional transition metal-catalyzed difunctionalization methods have been well-developed to control the regioselectivity and stereoselectivity7,8. Along these lines, cascade radical addition of unsaturated hydrocarbons in the absence of metallo-intermediate has been realized through careful manipulation of the radical reactivities9–11. A single process to achieve radical difunctionalization with extensive functionality tolerance, especially fluorine-containing moieties, is of great value in altering the physical and biological properties of the unsaturated hydrocarbons4,12–18. Studer and co-workers19 have reported a radical 1,2-trifluoromethylboration of unactivated alkenes using gaseous CF3I. Meanwhile, the direct 1,2-fluoroalkylboration of alkynes has also been explored20–22. The consequential vicinial vinylboronates, which can be readily transformed to a myriad of fluorine-containing building blocks, have been seldom realized. The only two existing approaches of trifluoromethylated vinylboronates were derived from fluorinated alkynes23 and oxiranes24. The inaccessibility of those pre-functionalized precursors and operationally tedious procedures prohibited the stepwise fluorination-borylation strategies from practical use (Fig. 1a). Thus, the development of regioselective installation of fluoroalkyl and boronated functionalities to unactivated hydrocarbons in the absence of transition-metal catalysts represents great challenge.Fig. 1Origin of the reaction design.a Early reports on borylation of CF3-containing substrates. b Lewis base-activated radical borylation. c The design for dual functional redox-active reagent. d Imidazolium sulfonate-derived bench-stable fluoroalkylating reagent. e Mechanistic insight of sequential radical trifluoromethylborylation.Due to the complexity of radical process incorporating C–C and C–B bonds formation, several issues need to be addressed, including the reactivities of carbon and boron-centered radicals, the regioselectivity of the radical additions to unsymmetrical alkynes, and stereoselectivity towards E/Z vinylboronates. Typically, a radical borylation process25–32 often employs Lewis basic solvents/mediators such as DMAc33,34, DMF19,35, phthalimide36, and pyridine37,38 for the activation of diboron reagents via homolytic cleavage of B–B bonds (Fig. 1b). However, the exogenous Lewis base-activated diboron species inevitably deplete CF3 radicals that generated promptly from the trifluromethylating reagents, unable to engage in the desired trifluoromethylborylation sequence (for DFT calculation details, see Supplementary Fig. 18). Inspired by recent radical-induced difunctionalization strategies39,40, we intend to design an integrated reagent that progressively releases CF3 radical for alkene/alkyne addition, and the endogenous Lewis basic residue subsequently activates the B–B bond for further borylation (Fig. 1c).Trifluoromethanesulfinate-derived fluorinating reagents have been devised and adopted for direct functionalization of alkenes, alkynes, and arenes41. In contrast, the highly hydroscopic and corrosive trifluoromethanesulfonic anhydride (Tf2O) as a trifluoromethyl source is rarely explored. For its strong electrophilic nature, triflic anhydride is commonly used as an alcohol and amine protecting agent42–44. Qing and co-workers45 have described a triflated pyridine intermediate (Tf−Py+·OTf−) that generated in situ for trifluoromethylated arenes and alkynes. However, the strong electron-withdrawing triflate-derived pyridinium complex is preferably dissociated and unattainable in solid or liquid phase. We speculated that a more basic N-heterocycle such as imidazole could harness the highly reactive Tf2O to assemble a bench-stable redox-active reagent. The positive charge of the resulting imidazolium trifluoromethanesulfonate can be delocalized on both nitrogen. Through the cleavage of the stabilized N–S bond (BDE ≈ 70 kcal mol−1)46, this cationic complex undergoes SET process to generate CF3SO2 radical. Meanwhile, as a Lewis base, the imidazole residue can further activate the diboron reagents towards homolytic cleavage of the diboron reagent20,34–39. Herein, we have synthesized a dual functional reagent IMDN-SO2CF3
1a–1g, a scalable and air-stable crystalline salt for a sequential radical fluoroalkylation-borylation of unsaturated hydrocarbons (Fig. 1d). First, under the irradiation, Ir(III)* can reduce the cationic reagent 1 to a neutral radical I and releases CF3 radical, SO2, and imidazole. Then the addition of ·CF3 to the alkyne regioselectively furnishes vinylic radical II. Subsequent addition of vinyl radical II to B2cat2 affords a Z-vinyl diboron radical III. The control of stereoselectivity is governed by steric repulsion between the trifluoromethyl group and the boronates. The following activation of diboron by the Lewis basic imidazole forms a highly reactive B–N heteroleptic intermediate IV, which leads to the carboborylation product 3 and imidazole-stabilized boryl radical V. Finally, photo-oxidation of V followed by coupling with −OTf affords boryl imidazolium salt VI and regenerates Ir(III) (Fig. 1e). These proposed intermediates and selectivities are supported by DFT calculations (see Supplementary Figs. 16–19) This photoinduced cascade radical difunctionalization offers a concise and applicable protocol for constructing highly regio- and stereoselective fluorine-substituted vinylboronates and vicinal fluoroalkyl boronates.ResultsReaction optimizationTo validate the above hypothesis, we selected phenylacetylene (2a) as pilot substrate to test the trifluoromethylborylation reaction (Table 1). After extensive screening of conditions (see Supplementary Tables 1 and 2), we found that when using 2.5 equivalents of IMDN-SO2CF3 (1a) (E1/2red = −1.385 V vs SCE), 2 mol% of fac-Ir(ppy)3 (E1/2IV/III* = −1.73 V vs SCE)47, 2.5 equivalents of B2cat2 in a mixed solvent of MeCN and EtOAc (1:3 v/v) at room temperature under the irradiation of 30 W blue LEDs, the vinylboronate product 3a could be obtained in 82% yield (determined by 19F NMR) with over 20:1 Z/E ratio. Different imidazolium sulfonate reagents 1b–1h were then examined (Table 1). The yield of 3a descended when the benzoimidazolium reagents 1b and 1c were used (entries 2 and 3). The counterion was found to be important for this transformation, as evidenced by the low yield (53%) obtained when using BF4− salt (1d, entry 4). The reaction proceeded with 2-phenylimidazole reagents 1e and 1f in 76% and 60% yield, respectively (entries 5–6). The dimethylated reagent 1g resulted in a lower conversion (entry 7). The electroneutral reagent 1h failed to produce the desired product under irradiation, which may due to the low reduction potential (E1/2red = −1.808 V vs SCE) (entry 8). This result validated the precedential presumption that the cationic reagent can serve as a better electron acceptor to furnish N-centered neutral radicals48–51. Other diboron reagents, such as bis(pinacolato)diboron (B2pin2) and bis(neopentylglycolato)-diboron (B2neop2), did not provide the corresponding borylated products (entries 9–10). Addition of excess bases such as imidazole and pyridine resulted in much lower yields (entries 11–12).Table 1Optimization of the reaction conditions.EntryVariation from the conditionsYield of 3aa (%)Z:E of 3ab1None82 (65)c>20:121b instead of 1a67>20:131c instead of 1a69>20:141d instead of 1a53>20:151e instead of 1a76>20:161f instead of 1a60>20:171g instead of 1a32>20:181h instead of 1a0—9B2pin2 instead of B2cat20—10B2neop2 instead of B2cat20—112.0 equiv of pyridine58>20:1122.0 equiv of 1-methylimidazole47>20:1aYield determined by 19F NMR spectroscopy using trifluoromethoxybenzene as an internal standard. b The Z/E ratio was determined by 19F NMR. c Isolated yield.Substrate scope with respect to the alkynesUsing 2 mol% of fac-Ir(ppy)3, IMDN-SO2CF3 (1a) (2.5 equiv), and B2cat2 (2.5 equiv) at ambient temperature, a range of alkyens underwent fluoroalkylation-borylation with good efficiency. As shown in Fig. 2, the reaction can be performed at a gram scale to give 3a in 61% yield and high stereoselectivity. Aromatic alkynes with electro-donating or electro-withdrawing substituents afford the desired products 3b–3j in good to excellent yields (60–93%) with high regio- and stereoselectivity (Z:E > 20:1).Fig. 2Substrate scope of the alkynes.aCrude yields determined by 19F NMR spectroscopy using benzotrifluoride or trifluoromethoxybenzene as an internal standard. bValues in parentheses are of isolated yields. cThe E/Z ratio was determined by 19F NMR. dThe E/Z ratio was determined by 1H NMR. eCrude yields determined by 1H NMR spectroscopy using dibromomethane as an internal standard. f2.0 mmol of 2p was used.Functionalities including halides (3b, 3m, 3n), nitrile (3d), ester (3g), and boronate (3j) are tolerated. Naphthyl- and thienyl-substituted alkynes also readily transformed into the Z-products 3h and 3i in good yields. The reaction could also be applied to alkynyl deuterium to produce the (Z)-selective deuterated vinylboronates 3k–3n in 52–72% yields. An attempt of more challenging internal alkyne substrate resulted in the tetrasubstituted olefin in high regio- and stereoselectivity (3o, 45%). For further investigation of the reaction scope, different fluoroalkylating reagents 1i–1l have been synthesized and applied to the standard cabonborylation conditions. Perfluoro-butyl (1i), hexyl (1j), and octanyl (1k) reagents could furnish the corresponding products 3q–3s in good yields (81–87%). Using a perhalogenated ether-derived sulfonate (1l), the vinylboronate 3t was formed in high yield. To demonstrate the scalability of such radical carboborylation protocol, the reaction was carried out on 10 mmol scale to afford 3a in 61% yield with equally high Z/E ratio. Under the standard reaction conditions, alkyl-substituted alkynes could not transform to the desired products. DFT calculations illustrate that the energy barrier of CF3 radical addition to aliphatic alkynes is higher than that to aromatic alkynes. Furthermore, a competing pathway of CF3 radical addition to B2cat2 leads to other trifluoromethylated products. Therefore, an excess amount of alkyl alkyne substrate is needed to facilitate the main reaction pathway. By using four equivalents of the alkyne, the borylated product 3p can be obtained in 26% yield. For internal aliphatic alkynes, the computed barrier with the CF3 radical is much higher than that for the reaction of B2Cat2 with the CF3 radical. Therefore, no desired product is obtained using internal aliphatic alkynes as substrate.Substrate scope with respect to the olefinsThe α-fluoroalkylated boronates are also useful fluorine-containing synthons for further elaboration. By slight variation of the standard reaction conditions (see Supplementary Tables 3–10), we have extended this carbonborylation protocol to a range of unactivated alkenes (Fig. 3). Using IMDN-SO2CF3 (1e), alkenes bearing ester and amide functionalities underwent radical 1,2-carbonboration to afford trifluoromethylated boronates (5a–5h) in good yields. Heteroaryl (5i–5j), sulfonyl (5k), and oxygenated alkyl groups (5l–5n) at various positions of the alkenes were also found effective. Cyclic alkenes could also transform into the desired products 5q and 5y. Noteworthy, biorelevant molecules, such as boldenone, lanosterol, (+)-α-tocopherol, and estrone-derived terminal alkenes afforded β-trifluoromethylboronates (5r–5u) in good yields. Additionally, fluoroalkyl radicals including ·C4F9 (1i), ·C6F13 (1j), ·C8F17 (1k), and ·CF2CF2O(CF2)7CF2Cl (1l) were successfully stitched to unactivated olefins to afford fluoroalkylborylated products in moderate yields (5v–5dd). The reaction with styrene failed to afford the desired product due to inert reactivity of benzylic radical.Fig. 3Substrate scope of the olefins.aCrude yields determined by 19F NMR. bValues in parentheses are isolated yields. cThe diastereomeric ratio determined by 1H NMR.Synthetic applicationsThe synthetic utility of the method was demonstrated in a number of transformations of the highly functionalized alkylboronates and alkenylboronates20,52,53 (Fig. 4). Oxidation of β-CF3 boronate 5a afforded hydroxylated product 6 in 62% yield. Silver-catalyzed radical deboronofluorination of 5a in aqueous solution provided the alkyl fluoride 7 in 76% yield. Vinylation, oxidative coupling, and homologation of 5p afforded functionalized products 8–10 in good yields. Halogenation of vinylboronic ester 3a resulted in the formation of β-CF3-vinyl bromide 11(53%). Palladium-catalyzed Suzuki–Miyaura cross-coupling of 3a with (hetero)aryl iodides afforded the corresponding trisubstituted alkenes 12 (90%) and 15 (88%). Olefination and alkynylation using vinyl bromide or alkynyl bromide also proceeded smoothly to generate 13 and 14 in 73% and 96% yields, respectively. The coupling of 3a with bioactive estrone-derived triflate produced the corresponding product 16 with high stereoselectivity.Fig. 4Further transformations.aH2O2 (30%), NaOH (3 M), THF, 0 °C to rt. bSelectfluor, AgNO3, TFA, H3PO4, DCM/H2O, 50 °C. cVinylmagnesium bromide, I2, THF, −78 °C to 0 °C. dn-BuLi, NBS, thiophene, THF, −78 °C. en-BuLi, dibromomethane, THF, −78 °C to rt. fCuBr2, MeOH, 80 °C. gPd(PPh3)4 (5 mol%), Cs2CO3, toluene, H2O, 80 °C.DiscussionIn summary, we have described an air-stable redox-active reagent IMDN-SO2RF
1 with high reactivity and scalability. A key design feature of this dual functional imidazolium sulfonate reagent is the cationic nature that favors the progressive formation of fluoroalkyl radicals by SET reduction under photocatalytic conditions. Meanwhile, the in situ-generated Lewis basic imidazole residue promotes the B–B bond cleavage. The integrated reagent is applicable to regulate the reaction sequence of carbon and boron-centered radicals to access various fluorine-bearing vinylboronates and alkylboronates with high stereo- and regioselectivities. Further study of this reagent is underway in our laboratory.MethodsGeneral procedure for the synthesis of imidazolium salts 1To a one-necked 1000 mL flask equipped with a magnetic stirrer, the corresponding imidazole (100 mmol), Et3N (150 mmol), and 600 mL DCM were added. The flask was then cooled in an ice bath, and 130 mmol (36.8 g) (CF3SO2)2O was bubbled into the flask slowly. The mixture was stirred at room temperature for 2 h and evaporated in vacuo, quenched with water, and extracted with ethyl acetate (300 mL × 3). The combined organic layers were dried over Na2SO4, filtered, and concentrated. The product was purified by flash column chromatography on silica gel with n-pentane/ethyl acetate as eluent to give the imidazolyl sulfonamide. Under argon, to a solution of the imidazolyl sulfonamide in dried DCM (400 mL) was added dropwise MeOTf (or Me3OBF4) (130 mmol) at 0 °C. Then, the mixture was stirred at room temperature for 12 h. (If EtOTf is used, the reaction is refluxed for 24 h.) After that, the mixture was concentrated under rotary evaporation to give a white solid (or a viscous liquid) crude product, to which Et2O (300 mL) was added. With vigorous stirring, a solid precipitate was formed and washed with Et2O (200 mL × 3) and dried in vacuo to yield the imidazolium salt 1 as a white solid.General procedure for the synthesis of vinylboronates 3Under argon, to a solution of 1 (0.50 mmol, 2.5 equiv), B2Cat2 (0.5 mmol, 2.5 equiv) and fac-Ir(ppy)3 (2 mol%) in MeCN:EtOAc (1:3) (3 mL) was added corresponding alkynes 2 (0.2 mmol) at room temperature. After that, the tube was exposed to 30 W blue LEDs at room temperature about 30 h until the reaction was completed as monitored by TLC or GC-MS analysis. A solution of pinacol (236 mg, 2 mmol) in MeCN (1.0 mL) was added dropwise to the mixture at 0 °C. After 1 h, saturated ammonium chloride solution (15 mL) was added and the aqueous layer was extracted with hexane (3 × 15 mL). The combined organic layers were dried over Na2SO4, filtered, and concentrated. The product was purified by flash column chromatography on silica gel with n-pentane/ethyl acetate as eluent to give the vinylboronates 3.General procedure for the synthesis of alkylboronates 5Under argon, to a solution of 1 (0.50 mmol, 2.5 equiv), B2Cat2 (0.6 mmol, 3.0 equiv) and fac-Ir(ppy)3 (2 mol%) in 1:1 MeCN/acetone (0.2 mL) was added Et3B (0.6 mmol, 3.0 equiv, 1 mol/L in THF) and corresponding alkenes 4 (0.2 mmol) at room temperature. After that, the tube was exposed to 30 W blue LEDs at room temperature for 30 h until the reaction was completed as monitored by TLC or GC-MS analysis. A solution of pinacol (142 mg, 1.2 mmol) in Et3N (1.1 mL) was added to the mixture. After 1 h, the reaction mixture was evaporated in vacuo. The product was purified by flash column chromatography on silica gel with n-pentane/ethyl acetate as eluent to give the alkylboronates 5.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Data 1 | nature communications | [
"Article"
] | [
"Reaction mechanisms",
"Synthetic chemistry methodology",
"Photochemistry"
] | alkenes alkynes explored diversification metal-catalyzed difunctionalization methods regioselectivity stereoselectivity7,8 radical addition unsaturated hydrocarbons metallo-intermediate realized manipulation radical single process radical difunctionalization functionality tolerance fluorine-containing physical biological properties hydrocarbons4 Studer co reported radical 1,2-trifluoromethylboration unactivated alkenes gaseous CF3I direct 1,2-fluoroalkylboration alkynes vinylboronates fluorine-containing building blocks seldom realized two approaches trifluoromethylated vinylboronates fluorinated alkynes23 oxiranes24. inaccessibility tedious procedures fluorination-borylation strategies development regioselective installation fluoroalkyl boronated functionalities unactivated hydrocarbons transition-metal catalysts challenge reaction design reports borylation CF3-containing substrates base-activated radical borylation design dual functional redox-active reagent Imidazolium sulfonate-derived-stable fluoroalkylating reagent Mechanistic insight sequential radical trifluoromethylborylationcomplexity radical process C–C bonds issues reactivities carbon boron-centered radicals regioselectivity unsymmetrical alkynes stereoselectivity towards E/Z vinylboronates radical borylation employs Lewis solvents phthalimide36 pyridine37 activation diboron reagents B–B bonds. exogenous Lewis-activated diboron species deplete CF3 radicals trifluromethylating reagents trifluoromethylborylation sequence Supplementary Fig. design integrated reagent releases CF3 radical for alkene/alkyne addition endogenous Lewis residue activates B–B bond borylation (Fig. 1c).Trifluoromethanesulfinate-derived fluorinating reagents functionalization alkenes alkynes hydroscopic corrosive trifluoromethanesulfonic anhydride (Tf2O) trifluoromethyl source rarely explored used alcohol amine protecting Qing co described triflated pyridine intermediate (Tf−Py+·OTf−) for trifluoromethylated arenes alkynes electron-withdrawing triflate-derived pyridinium complex dissociated unattainable in solid liquid.speculated N-heterocycle imidazole Tf2O bench-stable redox-active reagent positive charge imidazolium trifluoromethanesulfonate nitrogen N–S bond complex CF3SO2 radical imidazole residue diboron reagents homolytic cleavage synthesized dual functional reagent IMDN-SO2CF3 scalable air-stable crystalline salt sequential radical fluoroalkylation-borylation unsaturated hydrocarbons. irradiation Ir(III) reagent 1 neutral radical I releases CF3 radical SO2 imidazole addition ·CF3 alkyne furnishes vinylic radical II addition vinyl radical II B2cat2 Z-vinyl diboron radical III stereoselectivity steric repulsion trifluoromethyl group boronates activation diboron imidazole forms reactive B–N heteroleptic intermediate IV carboborylation product 3 imidazole-stabilized boryl radical V photo-oxidation coupling −OTf boryl imidazolium salt VI regenerates Ir(III) intermediates selectivities supported DFT calculations16–19 photoinduced cascade radical difunctionalization protocol constructing regio- stereoselective fluorine-substituted vinylboronates vicinal fluoroalkyl boronates selected phenylacetylene (2a) pilot substrate trifluoromethylborylation reaction screening using 2.5 equivalents IMDN-SO2CF3 −1.385 V 2 mol% fac-Ir(ppy)3 −1.73 2.5 equivalents B2cat2 mixed solvent MeCN EtOAc (1:3) room temperature 30 W blue LEDs vinylboronate product 3a 82% yield NMR over 20:1 Z/E ratio imidazolium sulfonate reagents 1b–1h examined yield 3a descended benzoimidazolium reagents 1b 1c used counterion important low yield (53%) salt (1d reaction proceeded 2-phenylimidazole reagents 1e 1f 76% 60% yield dimethylated reagent 1g lower conversion electroneutral reagent 1h failed produce desired product irradiation low reduction potential (E1/2red −1.808 V vs SCE cationic reagent better electron acceptor N-centered neutraldiboron reagents bis(pinacolato)diboron (B2pin2) bis(neopentylglycolato)-diboron (B2neop2) provide borylated products (entries 9–10) excess bases imidazole pyridine lower yields (entries 11–12).Table 1Optimization reaction conditions conditionsYield 3aa:E>20:121b:131c:141d:151e:161f:171g>20:181h pyridine58>20:1122.0 1-methylimidazole47>20:1aYield determined 19F NMR spectroscopy trifluoromethoxybenzene Z/E ratio 19F NMR Isolated yield 2 mol% fac-Ir(ppy)3 IMDN-SO2CF3 B2cat2 ambient temperature alkyens fluoroalkylation-borylation efficiency. 2 reaction scale 3a 61% yield high stereoselectivity Aromatic alkynes electro-donating-withdrawing substituents afford products 3b–3j good excellent yields (60–93%) high regio- stereoselectivity (Z:E > 20:1).Fig. 2Substrate scope alkynesyields determined 19F NMR spectroscopy benzotrifluoride trifluoromethoxybenzene standard isolated yields E/Z ratio 19F NMR 1H NMR 1H NMR dibromomethane mmol 2p used halides (3b nitrile ester boronate (3j) tolerated Naphthyl thienyl-substituted alkynes transformed into Z-products 3h 3i reaction alkynyl deuterium (Z)-selective deuterated vinylboronates 3k–3n 52–72% yields substrate tetrasubstituted olefin high regio- stereoselectivity (3o 45%). fluoroalkylating reagents 1i–1l synthesized standard cabonborylation conditions Perfluoro-butyl hexyl octanyl (1k reagents products 3q–3s good yields (81–87%) perhalogenated ether-derived sulfonate vinylboronate 3t formed high yield reaction 10 mmol scale 3a 61% yield high Z/E ratio standard conditions alkyl-substituted alkynes transform desired products energy barrier CF3 radical addition aliphatic alkynes higher aromatic alkynescompeting CF3 radical B2cat2 leads trifluoromethylated products excess alkyl alkyne substrate needed reaction four equivalents alkyne borylated product 3p 26% yield internal aliphatic alkynes barrier CF3 radical higher B2Cat2 CF3 no desired product alkynes substrate α-fluoroalkylated boronates useful fluorine-containing synthons elaboration extended carbonborylation protocol unactivated alkenes (Fig IMDN-SO2CF3 alkenes ester amide 1,2-carbonboration trifluoromethylated boronates (5a–5h) good yields Heteroaryl (5i–5j), sulfonyl oxygenated alkyl groups (5l–5n) effective alkenes transform products 5q 5y biorelevant molecules boldenone lanosterol-α-tocopherol estrone-derived terminal alkenes afforded β-trifluoromethylboronates (5r–5u) good yieldsfluoroalkyl radicals ·C4F9(CF2)7CF2Cl unactivated olefins fluoroalkylborylated products moderate yields reaction styrene inert reactivity benzylic radical. 3Substrate olefins yields 19F NMR isolated yields diastereomeric ratio 1H NMR.Synthetic transformations alkylboronates alkenylboronates20 Oxidation β-CF3 boronate 5a hydroxylated product 6 62% yield Silver-catalyzed deboronofluorination 5a alkyl fluoride 7 76% yield Vinylation oxidative coupling homologation 5p products 8–10 good yields Halogenation vinylboronic ester 3a β-CF3-vinyl bromide 11(53%). Palladium-coupling 3a)aryl iodides trisubstituted alkenes 12 (90%) 15 (88%) Olefination alkynylation vinyl 13 14 73% 96% yields coupling 3a with bioactive estrone triflate product 16 high stereoselectivity.Fig. transformations (30%), NaOH (3 THFAgNO3 TFA H3PO4 DCM/H2O 50 °C bromide THF −78 °C 0-BuLi NBS thiophene THF −78 °C-BuLi dibromomethane THF −78 °C MeOH 80 °C(PPh3)4 Cs2CO3 toluene H2O 80 °C air-stable redox-active reagent IMDN-SO2RF 1 high reactivity scalability imidazolium sulfonate nature formation fluoroalkyl radicals photocatalytic Lewis imidazole residue promotes B–B bond cleavage reagent reaction carbon boron radicals fluorine vinylboronates alkylboronates study synthesis imidazolium salts 1000 mL flask imidazole (100 Et3N (150 600 mL DCM added cooled ice bath 130 mmol (36.8 g) (CF3SO2)2O bubbled mixture stirred room temperature 2 h evaporated vacuo quenched water extracted ethyl acetate (300 mL organic layers dried Na2SO4 filtered concentratedpurified chromatography silica gel n-pentane/ethyl acetate imidazolyl sulfonamide argon DCM (400 mL added MeOTf (130 mmol 0 °C mixture stirred room temperature 12 h EtOTf refluxed 24 h concentrated rotary evaporation white Et2O (300 mL added solid precipitate formed washed Et2O (200 mL × 3) dried vacuo imidazolium salt 1 white solid synthesis vinylboronates argon solution 1 (0.50 B2Cat2 (0.5 mmol fac-Ir)3 (2 MeCN:EtOAc (1 (3 mL added alkynes 2 (0.2 mmol) room temperature tube exposed 30 W blue LEDs 30 h reaction solution pinacol (236 mg 2 mmol MeCN (1.0 mL added 0 °C 1 h saturated ammonium chloride solution (15 mL added aqueous layer extracted hexane (3 × 15 mL). organic layers dried Na2SO4 filtered concentrated purified chromatography silica gel n-pentane/ethyl acetate vinylboronates synthesis alkylboronates argon solution 1mmol 2.5 B2Cat2 mmol 3.0 fac-Ir)3 (2 1:1 MeCN/acetone mL Et3B (0.6 mmol 3.0 equiv 1 mol/L THF alkenes (0.2 mmol room temperature tube exposed 30 W blue LEDs 30 h reaction TLC-MS pinacol (142 mg 1.2 mmol Et3N (1.1 mL 1 h evaporated vacuo purified flash column chromatography silica gel n-pentane/ethyl acetate eluent alkylboronates information | 53 | 0.380456 |
10.1038/s41467-021-21055-0 | PMC7865077 | Despite many studies on the oxygen-evolution perovskite-type electrocatalysts thus far, significant constraints prevent precise evaluations of catalytic activity. Here authors examine over 50 heteroepitaxial LaNiO3 and LaCoO3 (001) thin films to assess the intrinsic catalytic behaviors. | Although numerous studies on oxide catalysts for an efficient oxygen evolution reaction have been carried out to compare their catalytic performance and suggest new compositions, two significant constraints have been overlooked. One is the difference in electronic conduction behavior between catalysts (metallic versus insulating) and the other is the strong crystallographic surface orientation dependence of the catalysis in a crystal. Consequently, unless a comprehensive comparison of the oxygen-evolution catalytic activity between samples is made on a crystallographically identical surface with sufficient electron conduction, misleading interpretations on the catalytic performance and mechanism may be unavoidable. To overcome these limitations, we utilize both metallic (001) LaNiO3 epitaxial thin films together with metal dopants and semiconducting (001) LaCoO3 epitaxial thin films supported with a conductive interlayer. We identify that Fe, Cr, and Al are beneficial to enhance the catalysis in LaNiO3 although their perovskite counterparts, LaFeO3, LaCrO3, and LaAlO3, with a large bandgap are inactive. Furthermore, semiconducting LaCoO3 is found to have more than one order higher activity than metallic LaNiO3, in contrast to previous reports. Showing the importance of facilitating electron conduction, our work highlights the impact of the near-Fermi-level d-orbital states on the oxygen-evolution catalysis performance in perovskite oxides. | IntroductionAs many devices for electrochemical energy storage and conversion operate at room temperature, control of electrochemical redox reactions is of great significance to reduce the activation barriers and thereby boost the overall storage and conversion efficiencies. Among different types of redox reactions, the oxygen evolution reaction (OER) is an indispensable kinetic process taking place at the anode during water splitting in electrolyzers and at the cathode during charging in metal–air rechargeable batteries1–3. In particular, multiple transfers of electrons and protons during the OER are considered to result in a much large activation barrier4, compared with that of the hydrogen evolution reaction. The utilization of efficient OER electrocatalysts is thus imperative to significantly reduce the overpotential of the anodic reaction in water electrolysis for hydrogen production.In addition to traditional Ir- and Ru-based oxides5–8, many other oxide materials have been suggested as promising OER catalysts over the last decade9,10, encompassing complex perovskite oxides11–16, layered (oxy)hydroxides17–20, and spinel oxides21,22. Furthermore, as several notable descriptors have been successfully developed with the aid of ab initio density functional theory (DFT) calculations4,23, they enable us to understand the overall origin of activity variations and sometimes even to theoretically predict the catalytic properties when new catalysts are reported. As known well, charge transfer between adsorbates (O* and OH*) and a metal (M) on the catalyst surface will always be involved in the OER4. In this regard, electronic structure-associated descriptors recently have been proposed24–33 in addition to the early descriptor based on the bond strength between M and the adsorbates since the work by Bockris and Otagawa11,12. The filling of the eg-level electrons of the metal cations24, the position of the O p-band center25, the degree of covalency between transition metals and oxygen26,30, and the charge-transfer energy13,29 are noticeable examples of the important descriptors on the basis of the electronic structure of oxides.Although these recent descriptors provide reasonable insight and further prediction in determining the relative OER activity between metal oxides, serious constrains should be noted when the experimentally obtained activities are compared with the theoretical descriptors. First, if a catalyst is electronically insulating, a great deal of applied overpotential may be dissipated as Ohmic resistance, resulting in much lower OER current even though the catalyst may be intrinsically very active. Therefore, a direct comparison of the OER activity cannot be made when one catalyst is metallic and the other is insulating or semiconducting with a bandgap. Second, as demonstrated in several oxides including SrRuO334,35, the OER activity of crystalline catalysts is considerably orientation-dependent, showing fairly different values of the OER current density on each crystalline facet. This indicates that the precise difference of the OER activity between catalysts is difficult to identify unless an identical crystallographic surface is measured. Third, most theoretical studies regarding the variation of M−O(H)* bond strength have dealt with merely one specific facet4,23, despite that the experimentally measured OER properties come from numerous random facets of polycrystalline catalysts. Moreover, many DFT calculations to obtain the density of states (DOS) of M d and O p orbitals have been carried out by using the bulk supercells rather than the supercells containing the surface.To overcome these limitations in this work, we utilize heteroepitaxial (001)cubic thin films of LaNiO3 and LaCoO3 perovskite oxides instead of polycrystalline particles36–39. In one series of experiments, metallic LaNiO3 thin films doped with a small amount of seven different trivalent metal dopants, Fe, Co, Cr, Mn, Sc, Al, and In are used. As these dopants are all 3+, neither cation vacancies nor oxygen vacancies are created during thin-film fabrication. As a result, we can straightforwardly focus on the correlation between the OER activity and the constitution from the transition-metal d-orbital states by doping the identical (001) metallic surface without interference by the charged point defects and the crystallographic orientation difference. One of the significant findings in this study is that Fe, Cr, and Al, the counter La-perovskites of which (LaFeO3, LaCrO3, and LaAlO3) are known to be OER inactive11,12,14, make a notable contribution to the OER activity when doped in metallic LaNiO3 films. The DFT calculations also consistently demonstrate the appearance of a large density of the d-orbital states near the Fermi level in cases where the OER activity is enhanced. In addition, when the low conductivity limitation is eliminated in semiconducting LaCoO3 films by introducing a metallic interlayer in the other series of experiments, the OER current density of the LaCoO3 film is observed to be one order higher than that of the LaNiO3 film, in stark contrast to previous results11,12,14,40, showing the best OER activity among LaMO3-type perovskite oxides. In addition to clarifying the importance of the electronic conductivity of OER catalysts, our study dealing with more than 50 heteroepitaxial thin films of different compositions highlights the impact of the near-Fermi-level d-orbital states on the OER performance in perovskite oxides.ResultsDFT calculations for DOSPrior to directly measuring the OER current density of each thin-film sample, we carried out DFT calculation to acquire the DOS information at the (001) LaNiO3 surface with dopants. A series of DOS plots of Ni 3d, O 2p, and dopant metal p and d orbitals shown in Figs. 1 and 2 were obtained from the first-row octahedra at the surface without La. A readily recognizable feature in Fig. 1 is that a significantly high density of the 3d states of three dopants, Fe, Co, and Cr (white circles in the schematic illustrations of the supercells), is commonly identified between −2 and 0 eV below the Fermi level, as denoted by a black arrow in each of the three plots. In contrast, the O 2p DOS in each plot does not substantially vary with doping when compared with that of the pristine (001) surface.Fig. 1DOS at the LaNiO3 (001)cubic surface with various dopants (1).The first set of DOS plots for O 2p, Ni 3d, and metal 2p or 3d states is provided for each case. A black arrow together with a gray shadow in each plot indicates the noticeably high density of the 3d states near the Fermi level. It is noteworthy that a rise of the 3d states of Ni-1 is induced by Al doping, although Al does not provide a significant contribution to the total DOS between −2 and 0 eV.Fig. 2DOS at the LaNiO3 (001)cubic surface with various dopants (2).The second set of DOS plots for O 2p, Ni 3d, and metal 3d states is provided for each case. In contrast to the first set for Fe, Co, Cr, and Al doping, the dopants, Sc, Mn, and In, have neither a substantial influence on the total DOS nor induce any Ni 3d-state variation near the Fermi level.Another intriguing aspect can be found in the Al-doped surface. As indicated by a black arrow in the DOS of “Ni-1,” a noticeable increase of Ni 3d DOS between −1 and 0 eV below the Fermi level is identified in the plot, demonstrating a doping-induced increment of the Ni 3d states, whereas no major contribution of the Al 2p states independently to the total DOS is verified between −2 and 0 eV. This variation of the Ni 3d DOS induced by doping near the Fermi level is also observed in the Fe-doped surface (see Supplementary Fig. 1 for the DOS plots of the Fe-doped surface). By contrast, as shown in Fig. 2, neither significant density of the d-orbital states from the other three dopants, Sc, Mn, and In, below the Fermi level, nor notable variation in the Ni 3d states was obtained for the (001) surface. Based on the series of DOS plots provided in Figs. 1 and 2, two distinct types of trivalent dopants can be categorized (Fe, Co, Cr, and Al vs. Sc, Mn, and In), depending on how significantly the d-orbital states contribute to the total DOS near the Fermi level.Initial OER activities of doped LaNiO3 filmsTo measure and compare the initial OER current values during the first anodic cycle from the identical crystallographic surface of doped LaNiO3, we fabricated heteroepitaxial (001) LaNiO3 thin films deposited on SrTiO3 single-crystal substrates by using a sol–gel method36,41. The heteroepitaxy nature of the films was first verified by X-ray diffraction selectively showing the (00 l) Brag reflections (see Supplementary Fig. 2 for the detailed X-ray patterns). Based on this X-ray diffractometry, both high crystallinity and heteroepitaxy of the films appear to be preserved up to 20% doping. As demonstrated in a series of composition maps acquired by energy-dispersive X-ray spectroscopy (EDS) along with high-angle annular dark-field (HAADF) scanning transmission electron microscopy (STEM) images in Fig. 3a, each of the dopants was also verified to be homogeneously distributed in a film with 25–30 nm thickness. In addition, we verified the trivalent state of dopants by using X-ray photoemission spectroscopy (XPS) and electron energy-loss spectroscopy (EELS) (see Supplementary Figs. 3 and 4 for details)Fig. 3Fabrication of LaNiO3 heteroepitaxial (001)cubic thin films with dopants.a A series of EDS maps is provided to verify the homogeneous distribution of the dopants in the films. The doping level was adjusted to be 10%. b HAADF images and their enlargements for the surface region are shown to directly confirm both the heteroepitaxy and the clean (001) surface of the films on SrTiO3 substrates.As the Cr-Kα1 peak (5.42 keV) in EDS overlaps with the La-Lβ2 peak (5.38 keV), the EDS analysis cannot be applied to La-containing materials when a compositional analysis of Cr is necessary (see Supplementary Fig. 5 for more details on the artifact of Cr detection in EDS). We thus used EELS in STEM instead to investigate the Cr distribution. In contrast to the successful fabrication of doped LaNiO3 epitaxial thin films for six dopants, homogeneous incorporation of Cr was found to be very difficult, as excessively rapid gelation of the Cr precursor was unavoidable during the sol–gel process. Consequently, as shown in Supplementary Fig. 6, serious inhomogeneity of Cr in the film and the formation of secondary phases on the film surface were frequently observed in Cr-added films during the STEM and EELS observation. Atomic-column-resolved cross-sectional HAADF-STEM images provided in Fig. 3b directly confirm the high crystallinity of each film with the perovskite structure and a clean and flat (001) surface without secondary impurity phases for the six dopants except the Cr case. Supplementary Fig. 7 shows the (001) surface of pristine LaNiO3 films without impurity or amorphous phases as well.The optimum doping concentration in each film sample for the OER activity measurement was adjusted to be comparatively small, 5%, as the metallic conduction behavior of the matrix LaNiO3 should be maintained so as to have sufficiently high electronic conductivity on an order of 103 S/cm. Figure 4a presents the OER current densities of the (001) surfaces during the forward sweep of the first cycle as a function of the anodic potential vs. reversible hydrogen electrode (RHE). The OER activity of our pristine (001) LaNiO3 films in this work (~75 μA/cm2) is exactly comparable with the activity (30–90 μA/cm2) of (001) LaNiO3 films reported in previous studies42–44 (see Supplementary Fig. 8 for details). As specifically represented in the bar graph in Fig. 4b, the catalytic activity of Fe-, Co-, Cr-, and Al-doped (001) films shows more than twofold enhancement, compared with that of the pristine film. This improvement by a small amount of doping is thus a significant indication that Fe, Cr, and Al in addition to Co can be beneficial for the OER catalysis if sufficiently high electronic conductance is supported. Although 6 out of 11 Cr-doped thin-film samples exhibit remarkable OER activity enhancement, no improvement is observed in the other five samples (see Supplementary Fig. 9 for details). We therefore denote this variation by using a different color in the bar graph for Cr along with black lines in Fig. 4b to indicate the maximum and minimum values. This wide range of variation in the catalytic activity among Cr-doped samples appears to be attributable to the serious inhomogeneity of Cr, as shown in Supplementary Fig. 6. Nevertheless, the observation of enhanced activity by Cr demonstrates that Cr is a catalytically active dopant in LaNiO3.Fig. 4OER activity comparison of the doped LaNiO3 (001)cubic surfaces.a This plot shows the variation of the OER current density of the films with 5% doping as a function of applied potential vs. RHE. b The bar graph compares the difference between the OER current densities of the doped (001) surfaces measured at 1.63 V vs. RHE. For an easier comparison, a black vertical line is added on the graph, showing the OER current density of the pristine (001) surface. As denoted, Fe, Co, Cr, and Al have a notably beneficial effect on the OER activity, while Sc, Mn, and In make little contribution.In contrast to these four dopants, Sc, Mn, and In were identified to have little influence on the OER catalysis, revealing nearly the same OER current density with that of the pristine film, as shown in Fig. 4b. Entire sets of the OER current values and the Tafel slopes with different doping levels (up to 20%) for each of the dopants are provided in Supplementary Fig. 10 and Supplementary Table 1. As all the thin-film samples were fabricated in an identical manner, the double-layer (DL) capacitance representing the electrochemically active surface area9,45,46 does not vary between the samples regardless of the doping species (see Supplementary Fig. 11 and Supplementary Table 1 for the DL capacitance values). Furthermore, as demonstrated in Supplementary Fig. 12, the nearly invariable surface topologies obtained by atomic force microscopy (AFM) and the merely 15% variation of the DL capacitances after OER cycling confirm that the notable enhancement of the OER activities by addition of Fe, Co, Cr, and Al shown in Fig. 4 stems from the doping effect rather than surface morphology variation. An XPS analysis also verifies no serious dopant dissolution during the anodic cycling (see Supplementary Fig. 13 for the XPS results).A combination of the series of DOS plots shown in Figs. 1 and 2, and the OER activity results in Fig. 4 reasonably suggests a noticeable correlation between the d-orbital states and the OER electrocatalysis. As indicated by black arrows in Fig. 1, the substantially high density of 3d states between −2 and 0 eV below the Fermi level is a common feature of Fe, Co, and Cr dopants for the significantly improved OER catalytic activity. Even though there is no contribution of Al to the DOS neat the Fermi level, it is noted that Al doping induces a rise of the 3d DOS of neighboring Ni, as also denoted by a black arrow in Fig. 1. In particular, the results of OER enhancements by Cr and Al are notable new findings because their counterpart perovskites, LaCrO3 and LaAlO3, are known to be inactive to the OER catalysis.OER activities of La(Ni,Co)O3 solid-solution filmsTo further examine the effect of doping and subsequent d-orbital states on the OER activity, we carried out another set of experiments using La(Ni,Co)O3 solid-solution thin films. As LaCoO3 has the same crystal structure (trigonal, space group: R\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar 3$$\end{document}3¯m) and very similar lattice parameters with LaNiO3, full-range solid solutions between LaCoO3 and LaNiO3 are achievable. However, it is noted that LaCoO3 is not metallic but semiconducting with a comparatively small bandgap (0.5–0.6 eV)47. Therefore, two different sample configurations were adopted for comparison. As shown in the STEM images and EDS maps in Fig. 5a, b, one configuration is a simple La(Ni,Co)O3 thin film directly deposited on a SrTiO3 substrate (Fig. 5a) and the other contains a conductive perovskite interlayer between the film and the substrate (Fig. 5b) to provide sufficient contact for electron conduction to the entire bottom surface of the film (see the schematic illustration in Fig. 5g). Metallic LaNiO3 was used for the conductive interlayer. The preservation of the epitaxy among the substrate, the interlayer, and the film was directly confirmed by atomic-scale STEM observation as shown in Fig. 5c. A whole cross-sectional HAADF image showing the entire configuration with the epitaxial interlayer at an atomic level is provided in Supplementary Fig. 14.Fig. 5La(Ni,Co)O3 (001) solid-solution thin films.Two different configurations of thin-film samples with and without a conductive interlayer were prepared. a, b Two sets of EDS maps along with a pair of HAADF- and BF-STEM images exemplify LaCoO3 films (a) directly deposited on a SrTiO3 substrate and (b) with a LaNiO3 conductive interlayer on a substrate. c Atomic-scale HAADF images verify the heteroepitaxy among the film, the interlayer, and the substrate, showing the preservation of their perovskite structure. d This set of atomic-scale EDS maps shows homogeneous distribution of Ni and Co in a La(Ni0.4Co0.6)O3 film. e The current–potential variations of La(Ni,Co)O3 solid-solution thin films with and without a conductive interlayer are plotted. f This bar graph represents the OER current densities measured at 1.63 V vs. RHE from the plots in e for comparison. In addition to consistent increment of the OER current density with Co addition up to 80%, a huge difference of the OER activity between LaCoO3 films with and without a conductive interlayer is noted. g A schematic illustration indicates that the conductive interlayer provides a viable conduction path at the entire bottom surface of the film. h The sheet resistance was also measured in a series of La(Ni,Co)O3 solid-solution films. The conductive interlayer is verified to play a crucial role to efficiently reduce the sheet resistance of Co-rich films including the pristine LaCoO3 film.The solid-solution behavior between LaCoO3 and LaNiO3 was verified by observing the consistent shift of the (002) Brag reflection in thin-film samples during the X-ray diffractometry (see Supplementary Fig. 15 for a series of X-ray diffraction patterns of La(Ni1−xCox)O3 (x = 0–1) films deposited on SrTiO3 substrates). Atomic-scale EDS maps48–50 in Fig. 5d specifically exemplify the homogeneous distribution of Ni and Co in a La(Ni0.4Co0.6)O3 film, directly proving the solid-solution mixture. When we measured the OER current density of a series of the La(Ni,Co)O3 solid-solution (001) films, an important finding could be readily acquired. As shown in Fig. 5e, f, a systematic enhancement of the OER activity with increasing Co substitution is identified for the samples with a metallic interlayer, demonstrating more than 10-fold higher OER current density of the (001) LaCoO3 surface (1.276 mA/cm2geo) than that of the (001) LaNiO3 surface (0.075 mA/cm2geo). It is worth noting that this remarkably large activity difference between LaCoO3 and LaNiO3 has not been reported thus far and previous studies rather have shown that LaNiO3 has the highest OER activity among many La-based perovskite catalysts11,12,14 (see Supplementary Fig. 16 for the comparison of the previously reported OER current densities of perovskite oxides). The considerably low OER current density of the (001) LaCoO3 surface without the conductive interlayer, 0.05 mA/cm2geo, which is even lower than that of the (001) LaNiO3 surface, straightforwardly reveals that exceptional OER activity of Co can be seriously interrupted if sufficient electron conduction to LaCoO3 is prevented. Indeed, the plot of sheet resistance of films in Fig. 5h consistently shows a reduction of resistance by four orders of magnitude in the LaCoO3 thin-film sample containing the interlayer. All the information regarding the OER current densities, the Tafel slopes, and the DL capacitance of La(Ni1−xCox)O3 films with and without the conductive interlayer is provided in Supplementary Tables 2 and 3. In addition, the X-ray reciprocal space maps (RSMs)51 of the LaCoO3 thin film with the interlayer (the total thickness of >50 nm) demonstrate no substantial epitaxial strain (see Supplementary Fig. 17 for the RSM). As a consequence, the remarkably high OER activity observed in the LaCoO3 film with a LaNiO3 interlayer does not relate to the epitaxial strain.OER activities of doped LaCoO3 filmsOur work shows that the intrinsic OER catalytic capability of LaCoO3 is exceptionally high, based on the results shown in Figs. 4 and 5 along with the prominently high density of Co 3d states near the Fermi level (see Supplementary Fig. 18 for the DOS of O 2p and Co 3d at the (001)cubic surface in LaCoO3). In particular, if the conditions of a sufficient conduction path and nanosize catalyst morphology are satisfied to prevent serious Ohmic losses caused by the nonmetallic behavior of LaCoO3, the additional contribution of the 3d states from a small amount of dopants is anticipated to be marginal in contrast to the LaNiO3 cases. For the third set of experiments, we thus prepared (001) doped LaCoO3 thin-film samples with a conductive interlayer in an identical manner as in La(Ni,Co)O3 solid-solution (001) films. Figure 6a shows a series of EDS maps visualizing the homogeneous distribution of dopants in (001) LaCoO3 films with a metallic LaNiO3 interlayer on SrTiO3 substrates (see Supplementary Fig. 19 for the X-ray diffraction patterns of the films to prove the epitaxial growth). As compared in Fig. 6b, c, the dopant addition does not make a substantial contribution to the OER activity in LaCoO3, showing rather a detrimental influence in most cases of dopants, except for Fe. The electrochemical data including the OER current densities of doped LaCoO3 films with a conductive interlayer up to a 20% doping level are provided in Supplementary Table 4 in addition to Supplementary Fig. 20.Fig. 6Fabrication of LaCoO3 heteroepitaxial (001)cubic thin films with dopants.a A LaNiO3 metallic interlayer was utilized in all the thin films, as shown in the Ni map. This whole series of EDS maps also verifies the homogeneous distribution of the dopants in the films. The doping level was adjusted to be 10%. b This plot shows the variation of the OER current density of the films with 5% doping as a function of applied potential vs. RHE. No substantial enhancement of the OER current by doing is identified. c The bar graph indicates the OER current densities of the doped (001) surfaces measured at 1.63 V vs. RHE.Durability of high-activity LaCoO3In addition to the distinct initial OER activities between LaCoO3 and LaNiO3, another property that should be mentioned is the durability of high catalytic activity. Based on previous reports on several oxides, it appears to be generally accepted that the stability of OER catalysts (or facets of a crystal) shows an inversely proportional correlation with the catalytic activity34,52. In this regard, we examined how long the high activity of LaCoO3 films could be preserved. Figure 7a presents the OER current densities of a pristine (001) LaCoO3 film with a conductive interlayer as a function of the anodic cycling number. It is noteworthy that three different potential ranges were utilized during cycling. As can be found in the first plot, its high activity (1 mA/cm2) does not substantially deteriorate even after 100 cycles in a range of 1.2–1.7 V vs. RHE, whereas a metallic (001) LaNiO3 film shows a gradual and substantial decrease of the OER activity with cycling (see Supplementary Fig. 21a for comparison). In contrast, when a much higher overpotential is applied up to 1.9 V (the third plot in Fig. 7a), significant degradation of the OER current density is observable within 20 cycles.Fig. 7Durability of OER activity in LaCoO3 thin films.a A set of three plots for the current-density variation with anodic cycling is provided. If the potential range is adjusted so as not to exceed 1.7 V vs. RHE, the high OER current density of LaCoO3 is preserved even after 100 anodic cycles, demonstrating notable durability of the high activity. b A series of STEM images show the surface variations with different potential ranges during cycling. In addition to the image of a pristine film, each images was acquired after the 20th cycle, as indicated in red shadows in a. The formation of an amorphous layer is indicated by a pair of yellow lines in each image. As compared in the EDS spectra, this amorphization is induced by Co dissolution at high overpotential.Atomic-column-resolved STEM images in Fig. 7b directly reveal that a Co-deficient amorphous layer is induced by anodic cycling on the LaCoO3 film surface (see the EDS spectra), demonstrating the remarkable dependence of the amorphization on the potential range. As already observed in a previous report31 on LaNiO3 films showing Ni dissolution (also see Supplementary Fig. 21b, c for the Ni-deficient composition of a surface amorphous layer), Co cations at high overpotential appear to dissolve into an alkaline electrolyte during cycling as well, resulting in the formation of a Co-deficient amorphous phase (see Supplementary Fig. 22) and subsequent reduction of the OER activity with cycling in the range of 1.2–1.9 V. Although larger overpotential is required (e.g., 1.8 V) in (001) LaNiO3 films in order to obtain a high OER current density (1 mA/cm2) (see Supplementary Fig. 21a), much lower overpotential (1.7 V) is sufficient to preserve the same OER activity in (001) LaCoO3 films. Breaking the conventional wisdom regarding oxide-based OER catalysts, this notable advantage of Co in terms of durability thus opens various possibilities for new design of catalyst materials achieving both high activity and long-time stability.DiscussionThree sets of experiments in the present study provide several significant implications regarding the oxide-based catalysts for the OER. First, the electronic conduction through the catalysts appears to be much more impactful than what has been usually considered in overall OER catalysis. The one order-of-magnitude variation of the OER current density in the LaCoO3 (001) surface with and without the conductive interlayer in Fig. 5f directly emphasizes the importance of electronic conduction, including the electronic conductivity of catalysts, in OER electrocatalysis. To precisely evaluate the real OER activity of a nonmetallic catalyst, such as LaCoO3 in this work, two resistive factors should be taken into account. One is the Ohmic loss (iROhmic) caused by the nonmetallic behavior of a catalyst and the other is the contact resistance (iRcont) induced between the catalyst and the electrode. Unless these two resistive components are efficiently eliminated, a large amount of applied potential during the OER would be seriously dissipated without directly contributing to the OER. We thus used nanoscale thin films to reduce iROhmic and utilized a conductive interlayer deposited on the entire substrate to suppress iRcont. Under this sample configuration, we could successfully evaluate the notably high OER activity of LaCoO3 without interference of the two resistive components.Second, Cr, the oxide form of which is known as one of the inactive transition-metal oxides for the OER, can be effective in addition to Fe, if the total composition of catalysts is optimized so that the high electrical conduction of charge is attained during the OER (note that the experimentally obtained bandgaps of pristine LaCrO3 and LaFeO3 are very large, ~3.4 eV and ~2.3 eV, respectively)53,54. Indeed, recent reports dealing with Cr doping and solid solutions in various materials to enhance the OER catalysis55–58 agree very well with our observations and DFT calculation for the Cr 3d DOS. Third, from the Al doping results in Figs. 1 and 4, catalytically inactive metal ions with no electrons in the d orbitals can make a viable contribution to the OER catalysis. As their substitution may induce the local distortion of neighboring [NiO6] octahedra, the Ni d-orbital levels become nondegenerate, resulting in the DOS variation31,41, as schematically illustrated in Supplementary Fig. 23a. Consequently, the density of neighboring Ni 3d states can change even by doping of catalytically inactive dopants. Indeed, we verified significantly different bond lengths in a neighboring [NiO6] octahedron of doped Al and subsequent distortion of the octahedron during the DFT calculation (see Supplementary Fig. 23b). In this respect, when a mixing approach with multiple components is utilized, a balanced consideration of the electronic structure on both the dopants and the matrix cation appears necessary for better understanding of catalysis enhancement.Finally, the charge transfer from transition metals to adsorbates via oxygen (e.g., M3+−O(OH*)− → M4+−OO*2−) during the OER on LaMO3 will be much easier and more likely to occur if a larger number of density of M 3d states are placed near the Fermi level rather than far below it. Consequently, Fe, Co, and Cr having substantial 3d states near the Fermi level appears to be considerably effective for the OER catalysis. However, please note that LaMO3 (M = Fe, Cr, Mn, In, Sc) is a typical insulating oxide with a large bandgap (>2 eV), whereas LaCoO3 is semiconducting with a narrow bandgap (~0.5 eV). As a consequence, when Fe, Mn, In, and Sc were doped in metallic LaNiO3 films, we could observe a serious reduction of the electronic conductivity (see Supplementary Fig. 24). Therefore, in the case of Fe doping, it appears that there is a trade-off between the benefit of Fe doping and the disadvantage from the detrimental conductivity in terms of the concentration, demonstrating that 5% is the optimum concentration.We have investigated the intrinsic contribution of the 3d-orbital states of various metal dopants to the OER activity on the LaMO3-type perovskite surface. For a consistent and precise comparison, a crystallographically identical (001) surface was measured with a small amount of the dopants in metallic LaNiO3 and semiconducting LaCoO3 heteroepitaxial thin films with a conductive interlayer to exclude electronic conduction constraints. Consistent with the DOS at the (001) surface acquired by the DFT calculations, Co, Fe, Cr, and even Al were directly identified to be effective elements to enhance the OER catalysis in LaNiO3. In particular, the OER activity of Co was proved to be one order higher than Ni in the (001) LaMO3-type perovskite surface together with notably better stability, in contrast to the conventional wisdom. Our study emphasizes that the 3d-orbital states near the Fermi level have a notable influence on the OER catalysis and also how seriously misleading the intrinsic OER activity of transition metals in oxides can be if the catalysts are not metallic.MethodsThin-film fabrication and X-ray analysisHeteroepitaxial LaNiO3 and LaCoO3 thin films were prepared by using a sol–gel process. Starting materials for the preparation of precursor solutions were La(NO3)3 ∙ 6H2O (99.999%, Sigma Aldrich), Ni(CH3COO)2 ∙ 4H2O (99.998%, Sigma Aldrich), and Co(CH3COO)2 ∙ 4H2O (99.999%, Alfa Aesar). For doping experiments, Al(NO3)3 ∙ 9H2O (99.997%, Sigma Aldrich), Sc(NO3)3 ∙ 4H2O (99.9%, Sigma Aldrich), Mn(CH3COO)2 ∙ 4H2O (99.99%, Sigma Aldrich), Fe(NO3)3 ∙ 9H2O (99.95%, Sigma Aldrich), Cr(NO3)3 ∙ 9H2O (99.99%, Sigma Aldrich), and In(NO3)3 ∙ 3H2O (99.99%, Sigma Aldrich) were used as dopants. Each of the starting materials were first dissolved in 2-methoxyethanol (99.9%, Sigma Aldrich) and refluxed at 80 °C for 1 h under a constant stirring condition to prepare homogeneous precursor solutions with 0.2 M. Each of the precursor solutions was deposited on (001) SrTiO3 single-crystal substrates by a spin-coating method at 5000 r.p.m. for 10 s. The wet films were dried at 150 °C for 10 min on a hot plate, subsequently heat-treated at 400 °C for 10 min for pyrolysis, and finally annealed at 800 °C for 1 h in air for crystallization. A 5%-Fe-doped LaNiO3 epitaxial film was deposited on each SrTiO3 single-crystal substrate in an identical manner as a conductive metallic interlayer to facilitate the electron conduction to the film. The epitaxy of the grown films was confirmed by X-ray diffractometry (X’Pert-PRO MRD, PANalytical) with Cu-Kα radiation and direct STEM observation. The X-ray RSMs for the (103) reflections from the substrates and the films were acquired by using the same diffractometer.STEM, EDS, EELS, XPS, and AFMSamples for all the STEM analyses were fabricated by focused ion-beam system (Helios Nanolab 450 F1, Thermo Fisher Scientific). Protective amorphous carbon and thin Pt layers were deposited on the epitaxial films before ion-beam milling. To minimize the sidewall damage and sufficiently thin the specimens for electron transparency, final milling was conducted at a voltage of ~2 kV. HADDF and BF-STEM images were taken with a transmission electron microscope (Titan cubed G2 60–300, Thermo Fisher Scientific) at 300 kV with a spherical aberration corrector (CEOS GmbH). The optimum size of the electron probe was ~1 Å with a convergence semiangle of 24 mrad. Chemical mapping with EDS was carried out in the Titan cubed G2 at 300 kV along with four integrated silicon-drift EDS detectors (ChemiSTEM™ technology) at a collection solid angle of 0.7 srad. La-Lα (4.6 keV), Ni-Kα (7.5 keV), Co-Kα (6.9 keV), Fe-Kα (6.4 keV), Al-Kα (1.5 keV), Mn-Kα (5.9 keV), Sc-Kα (4.1 keV), In-Lα (3.3 keV), Sr-Lα (1.8 keV), and Ti-Kα (4.5 keV) lines were selected during elemental mapping. The probe current was adjusted to be 50−100 pA with a scanning time of <300 s. The EDS maps were low-pass filtered using Bruker ESPRIT software after the reduction of background noise for better visualization. As the Cr-Kα1 (5.42 keV) and La-Lβ2 (5.38 keV) lines seriously overlap with each other, EELS analysis was performed for Cr mapping with a Gatan Image Filter (GIF Quantum 965, Gatan, Inc.). Electron energy-loss spectra for the Cr-L edges were acquired for spectrum imaging with a dispersion of 0.25 eV per channel and a collection aperture of 5 mm in diameter. The valence state of dopants was investigated using an X-ray photoelectron spectroscope (K-Alpha XPS, Thermo Scientific) with monochromatic Al-Kα radiation (1486.7 eV) and flood gun emission of 150 μA. The Ni 2p, Co 2p, Cr 2p, Al 2p, Sc 2p, Mn 2p, and In 3d peaks were compared with those from the reference crystals of a DyScO3 single crystal, Cr2O3 polycrystals, In2O3 polycrystals, a LaAlO3 single crystal, and LaMnO3 polycrystals. No significant chemical shift or broadening of the peaks was found, verifying the trivalent state of the dopants. In addition, because the photoemission Fe 2p1/2 and 2p3/2 peaks seriously overlap with the Ni Auger peaks in XPS, an EELS analysis was performed by using LaFeO3 polycrystals as a reference to examine the valence state of Fe. The surface topology of each film was examined by an atomic force microscope (Cypher VRS, Oxford Instruments) with a Si probe (tip radius of curvature: ~7 nm, AC160TS-R3, Oxford Instruments) in non-contact tapping mode.DFT calculationsAb initio DFT calculations for DOS of oxygen 2p and metal p and d orbitals at the (001) surface of doped LaNiO3 were carried out using the spin-polarized local density approximation (LDA) functional for exchange correlation, along with the ultrasoft pseudopotentials for ionic cores, as implemented in the CASTEP code (Biovia Inc.). A sufficiently long (001)-surface slab along with a 10-Å vacuum layer was constructed as an optimum supercell for each calculation to make the relaxation layer of each slab more than 10 Å in thickness. To account for the electron localization around the transition-metal ions, the LDA + U method with the Hubbard U parameter (4.0 eV for Ni 3d, Co 3d, and Fe 3d states; 2.0 eV for Mn 3d states; 3.5 eV for Cr 3d states) was employed59–62. Low-spin (t2g6)(eg1) for d7 Ni3+, intermediate-spin (t2g5)(eg1) for d6 Co3+, high-spin (t2g3)(eg2) for d5 Fe3+, high-spin (t2g3)(eg1) for d4 Mn3+, and high-spin (t2g3)(eg0) for d3 Cr3+ configurations were assumed, respectively61. The plane-wave basis set for the kinetic energy cutoff was 500 eV. Relaxation of the internal coordinates for each atom was performed using the Broyden–Fletcher–Goldfarb–Shanno algorithm with convergence tolerances of 0.1 eV/Å for the maximum ionic force, 5 × 10−5 eV/atom for the total energy, and 0.005 Å for the maximum ionic displacement.Electrochemical and electrical tests and measurementsAll electrochemical reactions and measurements were conducted with a potentiostat (Biologic SP-300) in a 0.1 M KOH aqueous solution (pH 12.9) prepared by using the Milli-Q water (18.2 MΩ·cm) and KOH pellets (Sigma Aldrich, 99.99%) to achieve sufficiently high purity. A Pt counter electrode and a saturated Ag/AgCl reference electrode were used. The measured potential values vs. the Ag/AgCl reference electrode were converted into the RHE scale by using the following equation at 25 °C,1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{RHE}}} = E_{{\mathrm{Ag}}/{\mathrm{AgCl}}} + 0.059 \cdot {\mathrm{pH}} + E^\circ _{{\mathrm{Ag}}/{\mathrm{AgCl}}}$$\end{document}ERHE=EAg/AgCl+0.059⋅pH+EAg/AgCl∘where ERHE is the converted potential vs. RHE, EAg/AgCl is the measured potential against the Ag/AgCl reference electrode, and E°Ag/AgCl is the standard potential of Ag/AgCl (KCl, 3 M) at 25 °C, i.e., 0.21 V. For thin-film samples, the substrate and the connecting copper wire were completely covered with chemically inert insulating epoxy resin (Henkel Loctite® EA 9460) after application of silver paint (Cans Elcoat P-100) on the back side of a thin-film sample so as to expose the film surface only. All electrolyte solutions were presaturated by bubbling O2 for 30 min under constant O2 bubbling. Cyclic potential was applied to the samples at a rate of 10 mV/s, sweeping from 1.2 to 1.75 V vs. RHE to record the first-cycle OER current density in each case. During the durability tests of thin films, the current densities were measured in various cyclic potential ranges up to 1.9 V. Samples were taken out of the electrolyte solution in order to remove O2 gas bubbles adsorbed on the film surface every five anodic cycles during the durability tests. Electrochemical impedance spectroscopy was also carried out in the same potentiostat in a frequency range from 0.1 Hz to 1 MHz with an amplitude of 10 mV for iR correction of the uncompensated series resistance. DL capacitance was measured to examine the variation of the electrochemical active surface area between thin-film samples. The DL capacitance measurements were carried out in a non-faradaic potential range of 1.03–1.17 V vs. RHE by recording the current values as a function of scan rate. To precisely present the OER current values on the basis of the DL capacitance, the converted plots showing the OER current normalized by the DL capacitance are provided in Supplementary Fig. 25. A two-point probe method along with Pt electrodes was utilized to measure the sheet resistance of thin-film samples. The thickness of each thin-film sample was directly measured by STEM observation.Supplementary informationSupplementary Information | nature communications | [
"Article"
] | [
"Catalysis",
"Electron transfer",
"Materials for energy and catalysis",
"Surfaces, interfaces and thin films"
] | devices for electrochemical energy storage conversion operate at room temperature control redox reactions activation barriers storage conversion efficiencies oxygen evolution reaction (OER) indispensable at anode during water splitting in electrolyzers cathode charging in metal–air rechargeable multiple transfers of electrons protons during OER in large activation barrier4 hydrogen evolution reaction utilization efficient OER electrocatalysts reduce overpotential anodic reaction in water electrolysis for hydrogen production Ir- Ru-based other oxide materials suggested promising OER catalysts complex perovskite layered (oxy)hydroxides17–20 spinel oxides21 descriptors developed with initio density functional theory (DFT) calculations4 enable understand origin of activity variations predict catalytic properties new catalysts charge transfer between adsorbates metal (M) on catalyst surface involved in OER4 electronic structure-associated descriptors proposed24–33 bond strength between M adsorbates filling of eg-level electrons of metal cations24 position O p-band center25 covalency between transition metals oxygen26 charge-transfer energy13 important descriptors electronic structure of oxidesrecent descriptors provide insight OER activity between metal oxides constrains when activities with theoretical descriptors if catalyst insulating overpotential dissipated as Ohmic resistance lower OER current active direct comparison OER when one catalyst metallic other insulating or semiconducting with OER activity of crystalline catalysts-dependent different values OER current on each facet precise difference OER activity between catalysts difficult identify unless identical crystallographic surface measured theoretical studies variation M−O(H)* bond strength dealt with one specific facet4 experimentally measured OER properties from numerous facets of polycrystalline catalysts DFT calculations of M d O orbitals bulk supercells surface limitations heteroepitaxial)cubic thin films of LaNiO3 LaCoO3 perovskite oxides instead of polycrystalline metallic LaNiO3 thin films doped with seven trivalent metal dopants Fe Co Cr Mn Sc Al In used dopants 3+ cation oxygen vacancies created during thin-film fabricationfocus correlation between OER activity transition-metal d-orbital states doping identical (001) metallic surface without interference charged point defects crystallographic difference Fe, Cr Al counter La-perovskites (LaFeO3 OER inactive11 contribution OER activity when doped in metallic LaNiO3 films DFT calculations demonstrate large density d-orbital states near Fermi level OER activity enhanced low conductivity limitation eliminated in semiconducting LaCoO3 films metallic interlayer OER current density LaCoO3 film one order higher than LaNiO3 film best OER activity among LaMO3-type perovskite oxides study 50 heteroepitaxial thin films highlights impact near-Fermi-level d-orbital states on OER performance in perovskite oxides calculations for OER current density DFT calculation DOS information (001) LaNiO3 surface with dopants DOS plots of Ni 3d, O 2p dopant metal p d orbitals Figs. 1 2 obtained from first-row octahedra at surface without Lahigh density 3d states dopants Fe, Co Cr identified between −2 0 eV below Fermi level denoted by black arrow in plots O 2p DOS vary with doping pristine (001) surface.Fig. 1DOS LaNiO3 (001)cubic surface dopants first DOS plots for O 2p Ni 3d metal 2p states black gray shadow indicates high density 3d states near Fermi level of 3d states Ni-1 induced by Al doping Al significant contribution to total DOS between −2 0 eV.Fig. 2DOS LaNiO3 (001)cubic surface dopants second set DOS plots for O 2p Ni 3d metal 3d states dopants Sc, Mn In total DOS induce Ni 3d-state variation near Fermi level Al-doped surface black arrow “Ni-1 increase of Ni 3d DOS between −1 0 eV below Fermi level doping-induced increment Ni 3d states no major contribution Al 2p states to total DOS between −2 and 0 eV variation Ni 3d DOS by doping near Fermi level observed in Fe-doped surface Supplementary Fig. 1 DOS plots Fig.neither significant density d-orbital states dopants Sc Mn In below Fermi level variation Ni 3d states obtained (001) surface DOS plots Figs. 1 2 two trivalent dopants categorized (Fe, Co Cr Al vs. Sc Mn depending d-orbital states DOS near Fermi level.Initial OER activities doped LaNiO3 OER values first anodic cycle doped LaNiO3 fabricated heteroepitaxial (001) LaNiO3 thin films on SrTiO3 single-crystal substrates sol–gel heteroepitaxy nature verified by X-ray diffraction Brag reflections Fig. 2 high crystallinity heteroepitaxy films preserved up to 20% doping composition maps energy-dispersive X-ray spectroscopy high-field Fig 3a dopants homogeneously distributed in film 25–30 nm thickness verified trivalent state dopants X-ray photoemission spectroscopy electron energy-loss spectroscopy) Figs 3 4. 3Fabrication LaNiO3 heteroepitaxial (001)cubic thin films with dopants EDS maps verify homogeneous distribution dopantsdoping level adjusted 10% HAADF images enlargements confirm heteroepitaxy clean (001) surface films SrTiO3 substrates Cr-Kα1 peak (5.42 keV EDS overlaps with La-Lβ2 peak (5.38 keV), EDS analysis La-containing materials compositional analysis Cr Fig. 5 used EELS Cr distribution doped LaNiO3 epitaxial films six dopants homogeneous incorporation Cr difficult rapid gelation Cr precursor Fig 6 inhomogeneity Cr film formation secondary phases surface observed in Cr-added films STEM EELS Atomic-column-resolved cross-sectional HAADF-STEM images Fig. 3b confirm high crystallinity film perovskite structure clean flat (001) surface without secondary impurity phases for six dopants except Cr Fig. 7 shows (001) surface pristine LaNiO3 films without impurity amorphous phases optimum doping concentration OER activity 5% metallic conduction LaNiO3 high electronic conductivity 103 S/cm. Figure 4a presents OER current densities (001) surfaces forward sweep first cycle anodic potential.hydrogen electrode OER activity pristine (001) LaNiO3 films (~75 μA/cm2) comparable with (30–90 μA/cm2) (001) LaNiO3 previous Supplementary Fig. 8 catalytic activity Fe- Co Cr- Al-doped (001) films twofold enhancement pristine film improvement doping Fe Cr Al beneficial for OER catalysis if high electronic conductance 6 out of 11 Cr-doped thin-film samples OER activity enhancement no improvement other five samples Supplementary Fig. 9 variation different color bar graph for Cr black lines Fig. 4b maximum minimum values variation catalytic activity Cr-doped samples inhomogeneity Cr Supplementary Fig. 6. enhanced activity Cr demonstrates Cr active dopant in LaNiO3.Fig. 4OER activity comparison doped LaNiO3 (001 surfaces plot shows variation OER current density films 5% doping potential vs RHE bar graph OER current densities doped (001) surfaces V vs RHE black vertical line added OER current density pristine (001) surface Fe Co Cr Al OER activity Sc Mn In little contributionfour dopants Sc Mn In little influence OER catalysis same OER current density with pristine film in Fig. 4b OER current values Tafel slopes different doping levels in Supplementary Fig. 10 Table 1. thin-film samples fabricated double-layer (DL) capacitance vary between samples doping species Supplementary Fig. 11 Table 1 DL capacitance Supplementary Fig. 12 invariable surface topologies by atomic force microscopy 15% variation DL capacitances after OER cycling confirm enhancement OER activities Fe Co Cr Al Fig. 4 doping effect surface morphology variation XPS analysis no serious dopant dissolution during anodic cycling Supplementary Fig. 13 DOS plots Figs. 1 2 OER activity results Fig. 4 correlation between d-orbital states OER electrocatalysis high density of 3d states between −2 0 eV below Fermi level common feature Fe, Co Cr dopants improved OER catalytic activity no contribution Al to DOS Fermi level Al doping induces rise 3d DOS of neighboring Niresults OER enhancements by Cr and Al new counterpart perovskites LaCrO3 and LaAlO3 inactive to OER catalysis.OER activities of La(Ni,Co)O3 solid-solution filmsTo examine effect of doping d-orbital states on OER activity experiments using La(Ni,Co)O3 solid-solution thin films LaCoO3 has same crystal structure (trigonal space group similar lattice parameters with LaNiO3 full-range solid solutions between LaCoO3 LaNiO3 achievable LaCoO3 not metallic but semiconducting with small bandgap (0.5–0.6 eV)47 two sample configurations adopted for comparison one configuration simple La(Ni,Co)O3 thin film deposited on SrTiO3 substrate other conductive perovskite interlayer (Fig. 5b for electron conduction Fig Metallic LaNiO3 used for conductive interlayer preservation of epitaxy among substrate interlayer film confirmed by atomic-scale STEM observation in Fig. 5c.cross-sectional HAADF image epitaxial interlayer atomic Supplementary Fig. 5La(Ni,Co)O3 (001) solid-solution thin films.Two configurations samples with without conductive interlayer EDS maps HAADF- BF-STEM images exemplify LaCoO3 films deposited SrTiO3 substrate LaNiO3 conductive interlayer Atomic-scale HAADF images verify heteroepitaxy film interlayer substrate preservation perovskite structure EDS maps homogeneous distribution Ni and Co in La(Ni0.4Co0.6)O3 film current–potential variations of La(Ni,Co)O3 films with without conductive interlayer plotted bar represents OER current densities 1.63 V vs. RHE addition difference OER activity between LaCoO3 films with without conductive interlayer conductive interlayer provides conduction path bottom surface sheet resistance measured in La(Ni,Co)O3 solid-solution films conductive interlayer sheet resistance Co-rich films solid-solution behavior between LaCoO3 LaNiO3 verified consistent shift (002) Brag reflection in thin-film samples during X-ray diffractometryX-ray diffraction patterns La(Ni1−xCox)O3 films SrTiO3 Atomic-scale EDS Fig. 5d homogeneous distribution Ni Co La(Ni0.4Co0.6)O3 film solid-solution mixture OER current density La(Ni,Co)O3 solid-solution (001) films finding Fig. 5e enhancement OER activity increasing Co substitution samples metallic interlayer 10-fold higher OER current density (001) LaCoO3 surface (1.276 mA/cm2geo) (001) LaNiO3 surface (0.075 mA/cm2geo). large difference between LaCoO3 LaNiO3 reported LaNiO3 highest OER activity La-based perovskite Supplementary Fig. 16 comparison OER low OER current density (001) LaCoO3 surface without conductive interlayer 0.05 mA/cm2geo lower (001) LaNiO3 surface reveals OER activity Co interrupted if electron conduction to LaCoO3 prevented plot sheet resistance Fig. 5h reduction resistance four orders LaCoO3 thin-film sample interlayerinformation OER densities Tafel slopes DL capacitance of La(Ni1−xCox)O3 films with without interlayer in Supplementary Tables 2 and 3. X-ray maps LaCoO3 thin film with interlayer >50 nm demonstrate no epitaxial strain Supplementary Fig. 17 high OER activity in LaCoO3 film with LaNiO3 interlayer epitaxial strain.OER activities doped LaCoO3 intrinsic OER catalytic capability of LaCoO3 high results Figs. 4 5 high density of Co 3d states near Fermi level Supplementary Fig. 18 if conduction path nanosize catalyst morphology additional contribution 3d states from dopants marginal LaNiO3 cases third experiments prepared (001) doped LaCoO3 thin-film samples with conductive interlayer La solid films Figure 6a EDS maps distribution dopants in (001) LaCoO3 films with metallic LaNiO3 interlayer on SrTiO3 substrates Supplementary Fig. 19 X-ray diffraction patterns epitaxial compared in Fig.dopant addition OER activity in LaCoO3 detrimental influence except Fe electrochemical data OER current densities of doped LaCoO3 films up to 20% doping level Supplementary Table 4 Fig. 20.Fig 6Fabrication LaCoO3 heteroepitaxial (001 thin films with dopants LaNiO3 metallic interlayer in films EDS maps homogeneous distribution dopants doping level adjusted 10% plot shows variation OER current density films with 5% doping applied potential vs. RHE No substantial enhancement OER current bar graph indicates OER current densities of doped (001) surfaces at 1.63 V vs. RHE.Durability of high-activity LaCoO3In distinct initial OER activities LaNiO3 durability of high catalytic activity stability OER catalysts with catalytic examined high activity LaCoO3 films Figure 7a OER current densities of pristine (001) LaCoO3 film with conductive interlayer anodic cycling number three potential ranges utilized during cycling high activity (1 mA/cm2) deteriorate after 100 cycles 1.2–1.7 V vs.RHE metallic (001) LaNiO3 film shows decrease OER activity with cycling Supplementary Fig. 21a higher overpotential applied up to 1.9 V third plot Fig. degradation OER current density within 20 cycles.Fig. 7Durability OER activity in LaCoO3 thin films three plots for current-density variation with anodic cycling If potential range adjusted not exceed 1.7 V vs. RHE high OER current density LaCoO3 preserved after 100 anodic cycles notable durability STEM images show surface variations with potential ranges during film acquired after 20th cycle formation amorphous layer indicated by yellow lines amorphization induced by Co dissolution at high overpotential STEM images Fig. 7b reveal Co-deficient amorphous layer induced by anodic cycling on LaCoO3 film surface dependence amorphization on potential range Co cations at high overpotential dissolve into alkaline electrolyte during cycling Co-deficient amorphous phase 22 reduction OER activity with cycling 1.2–1.9 V larger overpotential required1.8 V) in (001) LaNiO3 films high OER current density (1 mA/cm2) Fig. lower overpotential (1.7 V) OER activity in (001) LaCoO3 films oxide-based OER catalysts advantage of Co durability opens possibilities for new design catalyst materials high activity long-time stability experiments implications oxide-based catalysts for OER electronic conduction through catalysts more impactful OER catalysis one order-of-magnitude variation of OER current density in LaCoO3 (001) surface with without conductive interlayer Fig. 5f emphasizes importance of electronic conduction OER electrocatalysis evaluate OER activity of nonmetallic catalyst LaCoO3 two resistive factors Ohmic loss) nonmetallic behavior contact resistance) catalyst electrode Unless eliminated applied potential OER dissipated without used nanoscale thin films to reduce iROhmic conductive interlayer substrate to suppress iRcont high OER activity of LaCoO3 without interference resistive componentsCr oxide inactive transition-metal for OER effective to Fe if composition catalysts optimized high electrical conduction during OER bandgaps of LaCrO3 and LaFeO3 large ~3.4 eV ~2.3 eV reports Cr doping solid solutions OER agree with observations DFT calculation for Cr 3d DOS Al doping results Figs. 1 4 catalytically inactive metal with no electrons in d orbitals OER catalysis substitution distortion [NiO6] octahedra Ni d-orbital levels nondegenerate DOS Supplementary Fig. 23a density of Ni 3d states change by doping inactive dopants verified different bond lengths in [NiO6] octahedron of doped Al distortion during DFT calculation mixing approach multiple components balanced consideration of electronic structure on dopants matrix cation necessary for catalysis enhancement charge transfer from transition metals to adsorbates via oxygen M3+−O(OH*)− → M4+−OO*2− during OER on LaMO3 easier if M 3d states near Fermi levelFe, Co Cr 3d states near Fermi level effective for OER catalysis LaMO3 (M = Fe Cr Mn In Sc) insulating oxide large bandgap (>2 eV), LaCoO3 semiconducting narrow bandgap (~0.5 eV). Fe Mn In Sc doped in metallic LaNiO3 films reduction electronic conductivity (see Supplementary Fig. 24). Fe doping trade-off between benefit disadvantage detrimental conductivity 5% optimum concentration investigated contribution 3d-orbital states metal dopants to OER activity on LaMO3-type perovskite surface identical (001) surface measured with dopants in metallic LaNiO3 semiconducting LaCoO3 heteroepitaxial thin films Co Fe Cr Al effective OER catalysis in LaNiO3 OER activity of Co one order higher than Ni in (001) LaMO3-type perovskite surface better stability study emphasizes 3d-orbital states near Fermi level OER catalysis misleading OER activity of transition metals oxides if catalysts not metallic-film fabrication X-ray LaNiO3 LaCoO3 films prepared sol–gel process Starting materials La(NO3)3 ∙ 6H2O (99.999% Ni(CH3COO)2 ∙ 4H2O (99.998% Co(CH3COO)2 ∙ 4H2O (99.999% doping experiments Al(NO3)3 ∙ 9H2O (99.997% Sc(NO3)3 ∙ 4H2O Mn(CH3COO)2 ∙ 4H2O Fe(NO3)3 ∙ 9H2O Cr(NO3)3 ∙ 9H2O In(NO3)3 ∙ 3H2O (99.99% used dopants dissolved in 2-methoxyethanol refluxed 80 °C 1 h precursor solutions deposited on SrTiO3 single-crystal substrates spin-coating 5000 r.m. 10 s wet films dried 150 °C heat-treated 400 °C 10 annealed 800 °C 1 h crystallization 5%-Fe-doped LaNiO3 epitaxial film deposited each SrTiO3 substrate interlayer conductionepitaxy grown films confirmed by X-ray diffractometry-PRO Cu-Kα radiation STEM observation X-ray RSMs reflections substrates films acquired same diffractometer.STEM EDS EELS XPS AFMSamples fabricated ion-beam system (Helios Nanolab 450 F1 Thermo Fisher Protective amorphous carbon thin Pt layers deposited epitaxial films before milling damage thin specimens electron transparency final milling ~2 kV HADDF BF-STEM images taken transmission electron microscope 300 kV spherical aberration corrector optimum size electron probe ~1 Å convergence semiangle 24 mrad Chemical mapping EDS Titan cubed G2 300 kV four silicon-drift EDS detectors collection solid angle 0.7 srad La-Lα Ni-Kα Co-Kα Fe-Kα Al-Kα Mn-Kα Sc-Kα In-Lα Sr-Lα Ti-Kα (4.5 lines selected mapping probe current adjusted 50−100 pA scanning time <300 s EDS maps low-pass filtered Bruker ESPRIT software noise Cr-Kα1 (5 La-Lβ2keV lines overlap EELS analysis Cr mapping Gatan Image Filter Electron energy-loss spectra Cr-L edges acquired spectrum imaging dispersion 0.25 eV per channel collection aperture 5 mm valence state dopants investigated X-ray photoelectron spectroscope (K-Alpha XPS monochromatic Al-Kα radiation (1486.7 eV flood gun emission 150 μA Ni Co Cr Al Sc 2p Mn 2p In 3d peaks compared with DyScO3 Cr2O3 In2O3 LaAlO3 LaMnO3 No significant chemical shift broadening peaks verifying trivalent state dopants photoemission Fe 2p1/2 2p3/2 peaks overlap Ni Auger peaks XPS EELS analysis LaFeO3 polycrystals valence state Fe surface topology film examined atomic force microscope VRS Si probe ~7 nm non-contact.DFT oxygen 2p metal p d orbitals surface doped LaNiO3 spin-polarized local density approximation) exchange correlation ultrasoft pseudopotentials ionic cores CASTEP codelong (001)-surface slab 10-Å vacuum layer constructed optimum supercell relaxation layer 10 Å electron localization transition-metal ions LDA + U method Hubbard U parameter (4.0 eV Ni Co Fe 2.0 eV Mn 3.5 eV Cr 3d Low-spin d7 Ni3+ intermediate Co3+ high Fe3+ d4 Mn3+ d3 Cr3+ assumed plane-wave basis kinetic energy cutoff 500 eV Relaxation internal coordinates Broyden–Fletcher–Goldfarb–Shanno algorithm convergence tolerances 0.1 eV/Å maximum ionic force 5 × 10−5 eV/atom total energy 0.005 Å maximum ionic displacement.Electrochemical tests reactions measurements potentiostat (Biologic SP-300) 0.1 M KOH aqueous solution (pH 12.9) Milli-Q water (18.2 KOH pellets Pt counter electrode saturated Ag/AgCl reference electrode measured potential valuesAg/AgCl electrode converted into RHE scale equation at 25 °C{amsmath-69pt{RHE ={AgCl + 0.059 + E}ERHE=EAg/AgCl+0.059⋅pH+EAg ERHE converted potential vs RHE EAg/AgCl measured potential against Ag/AgCl electrode E°Ag/AgCl standard potential of Ag/AgCl) at 25 °C 0.21 V thin-film samples substrate connecting copper wire covered with chemically inert insulating epoxy resin (Henkel Loctite® EA 9460) silver paint (Cans Elcoat P-100) electrolyte solutions presaturated by bubbling O2 for 30 min Cyclic potential applied samples 10 mV/s from 1.2 to 1.75 V vs. RHE first-cycle OER current density durability tests current densities measured in potential ranges up to 1.9 VSamples taken electrolyte solution O2 gas bubbles film surface five cycles durability tests Electrochemical impedance spectroscopy same potentiostat 0.1 Hz to 1 MHz amplitude 10 mV iR correction uncompensated series resistance DL capacitance measured variation electrochemical active surface area between thin-film samples measurements non-faradaic potential range 1.03–1.17 V vs. RHE current values scan rate converted plots normalized DL capacitance Supplementary Fig. 25. two-point probe method Pt electrodes sheet resistance thin-film samples thickness thin-film sample measured by STEM observation.Supplementary | 51.1 | 0.653465 |
10.1038/s41467-020-14589-2 | PMC7028998 | It is unclear whether bird migration patterns are restricted to interglacial periods or are maintained during glacial maxima. Somveille et al. apply a global migration simulation model to climate reconstruction to show that the prevalence of this phenomenon has likely been largely maintained up to 50,000 years ago. | Migration is a widespread response of birds to seasonally varying climates. As seasonality is particularly pronounced during interglacial periods, this raises the question of the significance of bird migration during past periods with different patterns of seasonality. Here, we apply a mechanistic model to climate reconstructions to simulate the past 50,000 years of bird migration worldwide, a period encompassing the transition between the last glacial period and the current interglacial. Our results indicate that bird migration was also a prevalent phenomenon during the last ice age, almost as much as today, suggesting that it has been continually important throughout the glacial cycles of recent Earth history. We find however regional variations, with increasing migratory activity in the Americas, which is not mirrored in the Old World. These results highlight the strong flexibility of the global bird migration system and offer a baseline in the context of on-going anthropogenic climate change. | IntroductionBird migration, which dramatically rearranges avian assemblages worldwide in direct response to seasonality1–5, is a labile trait6–9. Previous research suggests that the original machinery of migration (physiological, behavioural, genetic)10 evolved deep in the avian lineage8,11 and its expression can change as a function of environmental conditions12–17. Accordingly, previous phylogenetic analyses found generally high (even if varying) rates of transition between sedentary and migratory behaviours and vice versa18–20. Analyses of current bird migration patterns have also shown that the seasonal ranges of migratory species, as well as the composition of avian communities in terms of migrant and resident species, are well explained by current climatic factors5,21–24, suggesting that the global distribution of birds is approximately at equilibrium with current climate23. Consequently, on-going climate change is already affecting migration routes14 and the prevalence of migrant species in avian assemblages25. Over time, climate change might, therefore, contribute to significant changes in global migration patterns, potentially leading to important net gains or losses of migratory behaviour in the avifauna.Previous authors have thus hypothesised that the important variations in the Earth’s climate during the glacial cycling of the Pleistocene had a major role in shaping current migratory pathways7,26–29. According to this hypothesis, the shifts and expansions of seasonal breeding ranges from glacial refugia into interglacial temperate regions, and away from non-breeding grounds, could have triggered a migratory behaviour in many species and shaped the global migration patterns observed today (e.g., by increasing migration distances). If this is the case, the importance of bird migration as an ecological phenomenon may be restricted to the warm interglacial periods (like the one we are in today), which are characterised by extensive areas featuring temperate climate conditions30.Understanding how the importance, prevalence and magnitude of migration across the avifauna vary throughout glacial cycles has relevance not just for understanding migration as a behavioural phenomenon, but also for gauging the past seasonal dynamics and the functional roles of birds in communities and ecosystems. However, addressing this problem directly is particularly difficult because the migratory behaviour is not recorded in the fossil record (i.e., fossils may indicate where species were present, but not if they migrated).Previous species-specific studies have used analysis of genetic data and ecological niche modelling to investigate how the last glacial cycle has affected the evolution of migration, generally finding a strong effect27–30. However, recent simulation analyses of global bird migration23,24 have suggested that, in addition to species-specific climatic niches, the seasonal redistribution of species is shaped by inter-specific competition for access to limited resources, for example associated with mutual interference31,32, increasing search time33 and territorial defence34. Investigating the past dynamics of bird migration can thus benefit from reconstructions at the scale of the avifauna, which go above and beyond independent, species-by-species models. A mechanistic model of the global seasonal distribution of birds—the Seasonally Explicit Distributions Simulator (SEDS)—has been recently developed to simulate spatial diversity patterns reflecting an equilibrium between the distribution of the global avifauna and climate23. This model relies upon the concept of energy efficiency (i.e., optimisation of energy budgets) to simulate the seasonal distributions of species, which can be sedentary or migratory. With the availability of climate reconstructions, this model provides a unique opportunity for simulating global bird migration in the past in order to investigate the response of the migratory avifauna to glacial cycles.Here, we develop a new version of the SEDS model that integrates annual energy budgets explicitly (see details in Methods, and a complete description and discussion of the original model in ref. 23). This framework is based on modelling the balance between costs and benefits of migration, with energy as a common currency, assuming a local carrying capacity to species richness based on primary productivity. The model simulates bird species’ seasonal distributions—i.e., breeding and non-breeding ranges (coincident in resident species; different in migratory species)—that progressively saturate a virtual world with similar geography and seasonal distribution of energy supply as the real world. These virtual bird species distribute in a way that maximises energetic fitness, i.e., they maximise the amount of energy allocated to reproduction and survival by optimising the balance between energy assimilation and energetic costs associated with migration distance (both being a function of where the species’ seasonal ranges are located), while taking into account the distribution of the other species, thus considering inter-specific competition for access to energy supply (Fig. 1, see Methods).Fig. 1Model description.The SEDS model was applied to each time slice of a climate reconstruction from present to 50,000 before present (BP); i.e., every 1000 years from present to 22,000 BP and every 2000 years before 22,000 BP. For a given time slice (example of 12,000 BP in this figure), the simulation starts (T0) with a virtual world empty of bird species (R = 0). At this point, the energy available to birds is equal to the energy supply (EA = ES), estimated from the net primary productivity (NPP) obtained from the climate reconstruction. In each simulation step (Ti, sub-steps 1 to 4) a new virtual species is added to the virtual world. Its geographical distribution (combination of a breeding range and a non-breeding range) is selected among 1,040,000 candidate distributions, which are pairs of range options. Each range option is seeded randomly across the world and grown until a fixed size using a stochastic algorithm constrained by climatic conditions (see Methods; sub-step 1). The candidate distribution with the highest energetic fitness (i.e., maximum value of energy used for production, Eprod) is selected (sub-steps 2 and 3). Eprod is computed as the energy assimilated by the species (Eassim), which is a linear function of the energy available within the geographical ranges (βEA, type I functional response), minus energetic costs, which is equal to 1 (the basal energy used for survival) plus a linear function of the migration distance (α = 6.45e−5, see Methods). This way, the model optimises the balance between the energy assimilated through access to energy supplies and the energy used for travelling to determine the species’ seasonal geographical distributions. As this new species is added to the virtual world (R = n + 1; n indicating the total number of species at the start of a simulation step), the energy available EA is further depleted in the corresponding breeding and non-breeding ranges (sub-step 4). The simulation ends (Tend) when, after nmax species are simulated, the maximum value for Eprod is negative, meaning that any new virtual species would not have access to enough energy to compensate for the energetic costs associated with survival and therefore cannot exist.We apply the new version of the SEDS model to past climate data in order to simulate a reconstruction of the global seasonal distribution of birds over the past 50,000 years. This period encompasses the transition between the last glacial period and the current interglacial period, thus allowing us to investigate the effect of major climatic changes on the spatial patterns and importance of bird migration worldwide. Our results indicate that the prevalence of avian migration has remained largely stable across the globe over the past 50,000 years, albeit with noticeable geographical variations, which suggests that this phenomenon has been continually important throughout the glacial cycles of the Quaternary and that its origin might be more ancient.ResultsPredicting the current global seasonal distribution of birdsThe SEDS model simulates the distribution of the global avifauna in two seasons (capturing summer in the Northern Hemisphere, and summer the Southern Hemisphere) using simple rules reflecting a few key mechanisms that are derived from first principles of ecology and energetics, with only a single free parameter that could not be estimated directly from the literature (see Methods). Despite its simplicity, this model performs well in simulating the patterns associated with the current global seasonal distribution of birds, i.e., richness in breeding migrants, non-breeding migrants and residents (Supplementary Fig. 1). The model predicts particularly well the fact that breeding migrants concentrate around 50°N, with a strong asymmetry between the northern and southern hemisphere (correlation between empirical and simulated pattern = 0.795; Supplementary Fig. 1). It also correctly predicts the empirical observation that during their non-breeding season migratory birds largely concentrate in the southern part of the northern hemisphere (correlation between empirical and simulated pattern = 0.611; Supplementary Fig. 1). The pattern of resident bird diversity, with a peak in the tropics, is also well captured (correlation between empirical and simulated pattern = 0.637; Supplementary Fig. 1), even if the model underestimates the magnitude of this peak (leading to an underestimation of empirical total species richness). The model’s good performance at simulating current breeding and non-breeding patterns of the global migratory avifauna supports an important role of energy efficiency (i.e., optimising the interplay between energy assimilation, which is affected by inter-specific competition for access to resources, and the energetic cost of travelling) in driving bird migration.Predicting the past global seasonal distribution of birdsA mechanistic, simulation-based model with good explanatory power is particularly useful for making predictions into environmental conditions different from those in which the model was calibrated. Assuming that the apparent current equilibrium between climate and the distribution of the global avifauna equally applied to the past, we therefore used the SEDS modelling framework to simulate the global seasonal distribution of migratory birds through time (Fig. 1). We used a climate reconstruction covering the past 50,000 years (with 1000-year intervals between present and 22,000 years ago and 2000-year intervals earlier; Supplementary Movie 1) combined with a global vegetation model to obtain estimates of energy supply at regular intervals over that period (see details in Methods). When applied to environmental conditions over the past 50,000 years, our model predicts breeding distributions of migratory bird species progressively closer to the Equator, up to the Last Glacial Maximum (LGM, ~20,000 years ago), particularly noticeable in North America and the Western Palaearctic (Fig. 2). In particular, avian assemblages north of ~50°N are predicted to have been significantly poorer in breeding migrants than they are today, particularly prior to 10,000 BP (Fig. 2). We also predict that the geographical distribution of non-breeding migratory birds were concentrated closer to the Equator than at the present, although this effect is less noticeable than for the breeding distributions (Fig. 2).Fig. 2Contrast between past and present simulated patterns of migratory bird diversity.The global patterns were computed as the predicted richness in the past, i.e., predicted number of species per hexagon: 10,000 years before present (a, b); 20,000 BP (c, d); and 50,000 BP (e, f), minus the predicted richness in the present. BP: before present. Red areas had more species than today, blue areas fewer.Our model predicts variations in the proportion of bird species that are migratory during the last glacial cycle. In the Americas, this proportion would have been ~20% smaller at the LGM than today (Fig. 3), corresponding to species that were resident during the last ice age and started migrating seasonally since then. A somewhat opposite trend is predicted to have occurred in the Old World, with the proportion of migrants similar to today or even higher during more ancient time periods (Fig. 3). A similar asymmetry is predicted in terms of migration distances. In the Americas, the model predicts that migration distance in the LGM was on average ~500 km shorter than today (Fig. 3), or conversely that today birds travel on average ~40% longer distances than they did at the end of the last ice age. In the Old World, however, the average distance travelled by migratory species is predicted to have slightly oscillated but remained on average largely stable over the last 50,000 years (Fig. 3).Fig. 3Predicted proportion of migrants and average migration distance over the last 50,000 years.a Evolution of the total proportion of simulated bird species that are migrants across the world relative to the present value. b Evolution of the average distance between breeding and non-breeding grounds for migrant species, computed as the great circle distance between the centroids of the seasonal ranges. These simulated time series are shown for the Americas (in black) and the Old World (in grey). 1 kyr = 1000 years.Our model has no ‘memory’, in the sense that the global seasonal distribution of birds for any year is simulated independently of other time points. Despite this, it predicts a stable overall number of simulated bird species over the last 50,000 years (Supplementary Fig. 2) boosting confidence in the model predictions, because it is indeed not expected that the number of avian species changed much over that period.DiscussionOverall, our findings suggest that throughout the last 50,000 years, spanning the last glacial maximum (~20,000 years ago), bird migration remained an important global phenomenon, refuting the hypothesis that this is mainly a phenomenon of interglacial periods during which the planet features large areas with temperate climate conditions29,30. This contrasts with previous results by Zink and Gardner30, who predicted from species-specific climate niche models for 56 North American migratory species that most species were sedentary during the last glacial maximum and that glaciations are major ‘migratory switches’. Our results, which are based on simulations at the avifauna-scale, point instead to an origin of migration over a much longer time scale than the glacial cycles of recent Earth history18–20,35–37.Our simulations also indicate that the magnitude of the avian response in terms of migratory behaviour to past global change was likely variable across the world. North America is the region of the world that is predicted to have seen the greatest changes in bird migration since the last ice age, alongside the retreat of the large Laurentide ice sheet. In this region, we predict a southwards compression of bird migration as we go back in time (Fig. 4), particularly of breeding ranges (Fig. 2), with a predicted shift of the transition zone between southern avian assemblages that are net senders of breeding migrants and northern assemblages that are net receivers of breeding migrants from ~35°N today to <30°N at the LGM (Fig. 4). This corresponds to a significant increase in the average migration distances covered in this region since the LGM (~40% increase) and in the proportion of species that are migratory (~25% increase; Fig. 3). In the Old World, the western Palaearctic is also predicted to have experienced relatively important changes alongside the retreat of the ice, with seasonal grounds of migrants at the LGM being closer to the equator than today (Fig. 2). This is somewhat less pronounced than in North America, which is in line with the relatively smaller extent of the Eurasian ice sheets, and we predict that it had little effect on the proportion of species that are migratory or in the average distance they travel (the former even appearing to be somewhat higher in the past than today; Fig. 3 and Supplementary Fig. 2).Fig. 4Fifty-thousand years of predicted seasonal difference in richness due to migration, across the northern Western Hemisphere.The maps a–h show the seasonal difference in avian richness, computed as the richness in breeding migrants minus the richness in non-breeding migrants, for eight time slices between the present and 50,000 years ago. BP: before present. Red colours are regions with avian higher richness during the breeding season; blue colours are regions with higher richness during the non-breeding season.In our simulations, patterns in the seasonal redistribution of the world’s avifauna emerge from the optimisation of energy budgets as birds use migration as a strategy to maximise energetic fitness. The number of current resident species, however, is substantially underestimated, particularly around the equator (Supplementary Fig. 1), which results in a too high fraction of migrants in relation to resident species across the world (Fig. 3 and Supplementary Fig. 2). A significant amount of the variation in the seasonal distribution of the migratory avifauna also remains unexplained by the model (Supplementary Fig. 1). This suggests that additional mechanisms need to be included in future versions of the model in order to better explain the empirical diversity patterns. This may include more realistic approaches to modelling migratory costs than the shortest distance between seasonal grounds, by taking into account geographical features (e.g., seas, mountains, distribution of stopovers)3,38–40, wind patterns41, and predation risk42. Moreover, the model currently treats all species as equivalent and equally abundant locally, but differences in species’ evolutionary history and ecology could also be important for explaining global empirical patterns. For example, we underestimate the number of species across high-latitude northern temperate regions over winter, yet several bird lineages (e.g., parids, corvids, woodpeckers, finches) have evolved adaptations other than migration to increase survival in these seasonally energy-depleted environments43. Furthermore, the current version of the model does not perform as well at predicting the proportion of all species that are migrants as the original SEDS model23, suggesting that the model of net primary productivity we applied here (one that can be projected back in the past, see Methods), might not be as adequate at estimating energy supply as the remote-sensing data used in the original SEDS model.Given that shifts in the spatial distribution of biodiversity are the combined result of individual species’ responses44, a further development would be to apply our global model in tandem with single species models, the former being used to model the background community and the latter for making species-specific predictions on the evolution of their distributions and migration since the last glacial period. This would allow investigating for example if rapid transitions between being sedentary and being migratory and vice versa occurred since the last ice age, something that cannot be captured by our global model alone. Estimating the rate of species-specific gains and losses of migratory behaviour due to glacial cycling could in turn inform phylogenetic analyses over evolutionary time (e.g., refs. 18–20,37) and thus bring new insights into the origin and evolution of migration.Our results suggest that bird migration systems across continents have not responded the same to past climate change. The differences in the past waxing and waning of the migratory phenomenon between continents can potentially explain patterns observed today, such as differences in migration strategies between avifaunas. For example, the need for communication calls during migratory flights might be higher in the New World45 to compensate for the fact that species’ migratory behaviours have been particularly variable over time.The rapid anthropogenic climate change that Earth is currently experiencing is likely to have a strong impact on the distribution and movement of species and biodiversity. While non-mobile species will likely have to locally adapt to change, highly mobile species might be able to move and track changing environmental conditions. In this context, the magnitude and flexibility of the response of bird migration to global change highlighted by our results offers a baseline for predicting how migratory birds will respond to future climate change.MethodsClimate reconstructionMonthly climate data (temperature, precipitation and cloud cover) covering the past 50,000 years were obtained from the HadCM3 global circulation model46. These data are at 1000-year intervals between present and 22,000 years before present (BP) and at 2000-year intervals between 22,000 and 50,000 BP. The original simulation data on a 2.5° × 3.5° were bias corrected and downscaled using the delta method47, which builds a difference map between simulated and observed data (in our case, high-resolution present-day estimates from ref. 48) and uses it across time intervals. This approach has been shown to be the most robust solution to debias paleoclimatic reconstructions for the late Pleistocene49. We first downscaled our paleoclimatic variables to a 1/6 degree grid, and subsequently remap them onto a global grid of equal-area, equal-shape hexagons (internode spacing of ~153 km). With the HadCM3 simulations, we used the global ice sheets reconstruction data set ICE-6G version 1.250. Hexagons covered by ice sheets were considered not habitable for all birds, regardless of climatic conditions. Seasonal climate values for temperature and precipitation were obtained by averaging the monthly climate values over the northern winter (November to February included) and the northern summer (May to August included). Visualisations of the climate reconstructions are presented in Supplementary Movie 1.Model overviewWe developed a new version of the Seasonally Explicit Distributions Simulator (SEDS) model, which was originally described and discussed in ref. 23. Here, we integrate energy budgets more explicitly and we reduce the number of free parameters. The SEDS model is based on modelling the balance between costs and benefits, with energy as a common currency. It is built on three main components: (1) a set of virtual species’ range options, (2) the estimated energy requirements associated with key biological processes, and (3) the estimated spatial and seasonal variation of the energy supply available to birds in the environment. Integrating these three components, the model is applied through a sequence of simulation steps whereby a virtual world—with the same geography and seasonality as the real world, mapped onto the global hexagonal grid described above—is filled with virtual species until it becomes saturated. In this model, virtual bird species are functionally equivalent (i.e., we ignore differences in traits and ecology) and are represented by a combination of a breeding range and a non-breeding range that can be either congruent (resident species) or different (migratory species).At the start of a simulation, the virtual world is empty of bird species, and each simulation step consists of adding a virtual species into it. The geographical distribution of each virtual species (i.e., combination of breeding and non-breeding ranges) is selected among candidate distributions as the one with the maximal energetic fitness (Fig. 1). The newly simulated virtual species then consumes energy within its geographical distribution equivalent to its corresponding energetic cost, effectively depleting the energy available in all the hexagons across its geographical distribution. We stop simulating species when the virtual world is saturated with simulated species. Each simulation was performed separately on the Western Hemisphere (WH < 30°W) and Eastern Hemisphere (EH > 30°W).The model is mechanistic in the sense given by Connolly et al.51: ‘a characterisation of the state of a system [here, the global seasonal distribution of birds] as explicit functions of component parts [species’ geographical ranges optimising energy budgets] and their associated actions and interactions [inter-specific competition for access to energy supply]’.Virtual species range optionsFor each time slice separately, we generated 1000 contiguous geographical ranges in our virtual world (400 in the WH, 600 in the EH, reflecting differences in area) to serve as options from which the distributions of virtual species were simulated (Fig. 1). These range options all had a size of 131 hexagons in the western hemisphere and 180 hexagons in the eastern hemisphere, which correspond to respective median values in a global data set of avian species range maps52. Ranges were generated using a method adapted from the spreading dye algorithm53,54 through a climate-driven approach of range expansion that has been shown to accurately capture the empirical distribution of bird ranges’ shape55. Each range was seeded from a single hexagon, randomly selected among all hexagons each with a probability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_h = 1/\left( {1 + S_h} \right)$$\end{document}Ph=1∕1+Sh (Eq. (1)), with Sh denoting the number of species already simulated and occurring in hexagon h. The probability of selecting a given hexagon as a seed hexagon was thus a function of the local richness in virtual species in order to avoid simulating range options that are too clustered spatially. From the selected seed hexagon, we then allowed a stochastic spread into adjacent unoccupied hexagons, constrained by climatic conditions, until the virtual range reached a fixed size. For each range, an initial climatic optimum was obtained from the position of the seed hexagon in a climatic space defined by a mean annual temperature (z-standardised) and a mean annual precipitation (z-standardised after being log-transformed). We then selected two neighbours of the seed hexagon, with the probability of selection being higher for neighbours closer to the climatic optimum (that is, lower Euclidian distance d in the climatic space between itself and the climate optimum), calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{{\mathrm{select}}} = 2(d + 1)^{ - 30}$$\end{document}Pselect=2(d+1)−30 (Eq. (2)), divided by the sum of these values across all neighbours (hence decaying exponentially with increasing climatic distance d). We then repeated this procedure, each time redefining the climatic optimum as the average climatic condition across the already selected (that is, occupied) hexagons and selecting 25% (rounded to the larger integer) of the set of unoccupied neighbours of the occupied hexagons (summing the probabilities of the ones being neighbours of more than one occupied hexagon), until the desired range size was reached (131 hexagons, ~3.05 million km2, in the WH, 180 hexagons, ~4.20 million km2, in the EH). Visualisation of the geographical distribution of range options through time is presented in Supplementary Movie 2.Energy supplyIn each hexagon, the energy supply ES is the total amount of available resources that can be used to fuel bird species’ metabolism. In ref. 23, it was modelled as proportional to the Normalised Difference Vegetation Index (NDVI), but NDVI is a remote-sensing measure of greenness of the land that cannot directly be reconstructed in the past. Therefore, here, we modelled energy supply as proportional to net primary productivity (NPP). Monthly NPP for each time slice was estimated with the Biome4 global vegetation model56 using monthly temperature, precipitation and cloud cover reconstructions (see above) as inputs (Supplementary Movie 1).We assumed that in any given hexagon j, the carrying capacity for birds is proportional to NPP, such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{S}}_j} = \mu \ast log_{10}\left( {{\mathrm{NPP}}_j + 1} \right)$$\end{document}ESj=μ*log10NPPj+1 (Eq. (3)). Negative NPP values were set to zero. The parameter μ was used for adjusting the energy supply (that is, acting as carrying capacity) for the model to generate a realistic total number of virtual species. In this study, however, we were not interested in replicating precisely the total number of bird species occurring in the world, but rather in investigating how the spatial patterns associated with the global seasonal distribution of birds changed in the past. We thus used a fixed value of μ = 65 for all of our simulations. This value for μ was chosen, after looking at the range of values for NPP, to obtain simulated species richness that are in the same order of magnitude as the empirical data. We conducted a sensitivity analysis to make sure that this value for μ was not crucial for the results (see section on sensitivity analysis below). The energy available in any given hexagon \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{A}}_j}$$\end{document}EAj in a given simulation step is equal to the energy supply minus the energy already consumed by species simulated during previous simulation steps and occurring in hexagon j. The energy available to a species at a given season (breeding or non-breeding) was computed as the mean of the energy available in all the hexagons across the seasonal range.Energetic costsThe energetic requirements associated with a species’ seasonal range were modelled as a function of two terms: the basal energy use for existence, BEU, which was set to be 1 (arbitrary) unit of energy use, and the additional cost associated with migration (mC), which was converted into these same (arbitrary) units. The cost of migration corresponds to the energetic cost of, each year, travelling between the breeding and non-breeding ranges. We assumed that mC increases linearly with distance travelled (thus, mC = 0 for resident species), and migration happens instantaneously at the end of each season (its cost added to the corresponding season to reflect the previous investment in fat reserves). For each season, mC was computed as a function of the great circle distance, dm, between the centroids of the breeding and non-breeding geographical ranges (average distance travelled by individual birds of the species assuming that they migrate using the shortest route). Thus, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\mathrm{C}} = \alpha \ast d_{\mathrm{m}}$$\end{document}mC=α*dm (Eq. (4)), with α being a parameter determining the energy required for a bird to travel a unit of distance.The parameter α can be estimated directly from the literature on flight physiology57 as equals to flight power (FW, in J/s) divided by flight speed (FS, in m/s). To rescale the cost of migration in terms of the arbitrary units of energy use, we compared the energy used for the migratory journey to the basal metabolic rate (which approximates minimum levels of energy expenditure for existence) over a whole season (BMRS, in J), such that: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha = \frac{{F_{\mathrm{W}}}}{{F_{\mathrm{S}}\;{\mathrm{BMR}}_{\mathrm{S}}}}$$\end{document}α=FWFSBMRS (Eq. (5)). Detailed comparative studies found that FW and FS scale with body mass (M, in g) such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{\mathrm{W}} = 0.257M^{0.763}$$\end{document}FW=0.257M0.763 (Eq. (6)) (estimated using data from ref. 58 on the cost of forward flapping flight for 31 avian species, excluding seabirds) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{\mathrm{S}} = 6.4773M^{0.13}$$\end{document}FS=6.4773M0.13 (Eq. (7)) (estimated by ref. 59 measuring the cruising speed of 138 species of migratory birds in flapping flight), respectively. We used the allometric relationship for the basal metabolic rate (BMR, in mlO2/h) described for 211 avian species in ref. 60 as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{BMR}} = 6.7141M^{0.6452}$$\end{document}BMR=6.7141M0.6452 (Eq. (8)), which we then converted to J/s using the conversion factor 1J/s = 172mlO2/h, and multiplied by the number of seconds in 6 months (i.e., ~15,724,800) to obtain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{BMR}}_{\mathrm{S}} = 6.15{\mathrm{e}}^5M^{0.6452}$$\end{document}BMRS=6.15e5M0.6452 (Eq. (9)). The resulting estimation for α was therefore approximately independent of body mass, with the cost of migration equal to: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{\mathrm{c}} = 6.45{\mathrm{e}}^{ - 5}d_{\mathrm{m}}$$\end{document}mc=6.45e−5dm (Eq. (10)), with dm the travel distance in kilometres. This corresponds to an energetic cost for migration of ~0.065 or ~6.5% of the yearly basal energy use for existence if the species travels an average of 1000 km between its breeding and non-breeding geographical ranges.Maximising energetic fitnessAs a model development in relation to ref. 23, here we modelled fitness explicitly, using an energetic definition (e.g., ref. 61). Birds assimilate biochemical energy initially converted mostly from solar radiation energy via photosynthesis. This assimilated energy (Eassim) fuels two main metabolic processes: respiration, which powers the work of living, and production, which generates new biomass. Using energy as the common currency, it translates into two components of fitness: energy used for survival (Esurv) and energy used for production (Eprod). The following relationship can be derived from this definition:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{prod}}} = E_{{\mathrm{assim}}} - E_{{\mathrm{surv}}}$$\end{document}Eprod=Eassim−EsurvWe assume that, to maximise fitness, birds maximise Eprod on an annual basis. For each virtual species to be simulated, we therefore looked for the candidate distribution (i.e., combination of virtual seasonal range options) with the highest associated value for year-round Eprod. To do so, for each candidate distribution, we estimated annual Eassim and Esurv. To estimate Eassim during a given season, we assumed a type I functional response of birds to the energy supply available in the environment (EA) as:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{assim}}} = \beta E_{\mathrm{A}}$$\end{document}Eassim=βEAwhere β is a parameter governing this linear relationship.To estimate Esurv during a given season, we computed the sum of the basal energy use for existence (BEU), which was set to 1 arbitrary unit, and the energetic cost of migrating between the seasonal distributions (see details in the section on energetic costs above), as:13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{surv}}} = {\mathrm{BE}}_{\mathrm{U}} + \alpha d_{\mathrm{m}} = 1 + 6.45{\mathrm{e}}^{ - 5}d_{\mathrm{m}}$$\end{document}Esurv=BEU+αdm=1+6.45e−5dmThe year-round amount of energy allocated to production for a given candidate distribution was therefore obtained using the following formula:14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{prod}}} = \beta \left( {E_{\mathrm{a}}^{({\mathrm{NS}})} + E_{\mathrm{a}}^{({\mathrm{NW}})}} \right) - 2\left( {{\mathrm{BE}}_{\mathrm{U}} + \alpha d_{\mathrm{m}}} \right)$$\end{document}Eprod=βEa(NS)+Ea(NW)−2BEU+αdmwhere NS indicates northern summer, and NW indicates northern winter.The geographical distribution (i.e., combination of breeding and non-breeding virtual ranges) of the virtual species to be simulated was selected as the candidate distribution with the highest value for Eprod (i.e., the highest energetic fitness). This candidate distribution was thus the one with an optimal balance between the energy assimilated through access to energy supplies and the energy used for travelling. The breeding season was set to be the season with the highest Eprod, essentially assuming that species maximise the amount of energy directly allocated to reproduction, and the geographical range associated with this season was therefore assigned as the breeding distribution of the species while the other range was assigned as the non-breeding distribution of the species. The newly simulated virtual species consumes energy within its seasonal geographical ranges equivalent to the corresponding Eassim, effectively depleting the energy available in all the hexagons across its geographical distribution. We stopped simulating species when the Eassim value for the selected distribution was below BEU, meaning that the species could not assimilate enough energy to fuel the basal energy use for survival. This indicates that the virtual world is saturated with simulated species so that no new species can be added to it and survive.Global patterns in the seasonal distribution of birdsThe SEDS model outputs virtual species seasonal distributions across the world, from which global diversity patterns can be mapped. We generated the following three basic spatial patterns that captured the global seasonal distribution of terrestrial birds: ‘richness in breeding migrants’, the number of species present in each hexagon only during their breeding season; ‘richness in non-breeding migrants’, the number of species present in each hexagon only during their non-breeding season; and ‘richness in residents’, the number of bird species present in each hexagon year-round. In parallel, we have also quantified these patterns using empirical data on bird distribution: spatial polygons representing the global distribution of 9783 non-marine bird species, obtained from BirdLife International and NatureServe52. The data and their treatment for generating these global richness patterns are described in detail in ref. 23.Parameters scanThe improved SEDS model used in this study has only one free parameter that could not be estimated directly from the literature: β, which determines the type I functional response between energy available in the environment and energy assimilation. We explored the following range of values for this parameter:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta} \in \left\{ {0.003, \ldots ,0.035} \right\}$$\end{document}β∈0.003,…,0.035 with a step of 0.001.Values below 0.003 resulted in not even one virtual species being able to survive as the energy it assimilated was already below its energy requirement for survival. Also we bounded the range of values to an upper limit of 0.035 because above this value the energy assimilation for the first simulated species (i.e., with maximum possible energy available) became highly unrealistic: >17 times the basal energy used for survival.To assess the quality of the model outputs for each simulation given a β parameter value, we computed the correlation coefficient between empirical and simulated patterns for the global seasonal distribution of birds by summing correlation coefficients for the three patterns described above: richness in breeding migrants, richness in non-breeding migrants and richness in residents.Low values of β lead to poor model performance, i.e., low correlation between the simulated global patterns and the empirical ones (Supplementary Fig. 1), as well as a very low number of species generated. For β > 0.01 the performance of the model plateaus above a mean correlation of 0.6 between empirical and simulated patterns, indicating fairly good performance of the model (Supplementary Fig. 3). However, for β > 0.015 the correlation tends to slightly decreases as β increases. The total species richness generated also peaks between 0.01 > β > 0.015, even though it does not go above 4000. This is less than half the actual number of bird species in the world. For every value of β investigated, the model also predicts a total proportion of migrants in the global avifauna that is much higher than the real one ( > 45% vs. 15%, respectively). We selected β = 0.012 as our best-fit (i.e., most realistic) value to be used for back-casting the global seasonal distribution of birds. This value gives the best compromise between maximising the match to empirical patterns, generating the maximum number of species and minimising the total proportion of migrants (Supplementary Fig. 3).Sensitivity analysisWe explored the robustness of the results obtained for the best-fit model by running the model with varying values for α, the parameter associated with the cost of migration previously estimated directly from the literature, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \in \left\{ {0.00001, \ldots ,0.0005} \right\}$$\end{document}α∈0.00001,…,0.0005, and μ, the parameter for rescaling NPP into energy supply, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu \in \left\{ {45, \ldots ,150} \right\}$$\end{document}μ∈45,…,150. Each time, we ran the model keeping all other parameters fixed and investigated the model performance at re-producing the empirical patterns associated with the global seasonal distribution of birds.We also explored the sensitivity of the model to the way we simulated geographical range options. We investigated the model performance using different values of range size (i.e., number of hexagons \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\in \left\{ {100, \ldots ,200} \right\}$$\end{document}∈100,…,200 for range options in the western hemisphere, and number of hexagons \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\in \left\{ {150, \ldots ,250} \right\}$$\end{document}∈150,…,250 for range options in the eastern hemisphere), as well as varying the strength of the climatic constraints when growing range options, i.e., varying the probability of selecting neighbours calculated as 2(d + 1)−x by investigating values for x
\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\in \left\{ {1, \ldots ,45} \right\}$$\end{document}∈1,…,45.The model performances were very robust to variations in μ, the parameter for rescaling NPP into energy supply for birds (Supplementary Fig. 4), as well as to how range options were simulated (Supplementary Figs. 5 and 6). Varying μ did not affect much the ability of the model to predict the current global seasonal distribution of birds (i.e., correlations between simulated and empirical patterns always >0.6), and did not affect the total proportion of migrants simulated (Supplementary Fig. 4). However, when μ increases then the total number of species generated increases almost linearly, which is not surprising since μ determines the total amount of energy supply, and thus the carrying capacity of the virtual world. Varying the size of the geographical range options simulated, as well as the strength of the climatic constraints determining their shape, did not affect the model performance (i.e., correlations between simulated and empirical patterns always >0.6) nor the total proportion of migrants simulated (Supplementary Figs. 5 and 6). Increasing the size of the simulated range options led to a nearly linear decrease in the total number of simulated species as less species can be fit in the virtual world for a given total carrying capacity. In contrast to the other sensitivity analyses, variation in α, the parameter determining the cost of migration, did affect significantly model performances, with relatively high values (α > 0.001), leading to the model poorly capturing the current global seasonal distribution of birds and model performance also decreasing with decreasing α below α = 0.0005 (Supplementary Fig. 7). The direct estimation of α from the literature (α = 0.000645, see section on energetic costs above) is within the peak of model performance (i.e., 0.0004 > α > 0.0008), which boosts the realism of the model. This peak of model performance also corresponds to a dip in the total number of simulated species, although the latter does not vary much with varying α (Supplementary Fig. 7). The total proportion of migrants decreases almost linearly with increasing α, which is not surprising as this corresponds to an increase in the cost of migration.We also reconstructed the global seasonal distribution of birds over the last 50,000 years for several combinations of parameter values other than the best-fit model. In addition to three alternative values for β, we investigated the outputs for three alternative values for α and three alternative values for μ, selected as performing relatively well for the present. The predicted evolution of global migration over the last 50,000 years is fairly robust to variation in values for parameters β, α and μ. The total number of species simulated remains stable over the last 50,000 years, with a slight decrease between 10,000 BP and 20,000 BP, for every combinations of parameter values investigated (Supplementary Figs. 8–16). The total proportion of migrants among simulated species also remains relatively stable over the last 50,000 years, with a slight decrease often observed in the Americas between 10,000 BP and 20,000 BP, for every combinations of parameter values investigated (Supplementary Figs. 8–16). The model exhibits a very similar pattern through time for the average migration distance among simulated migratory species for every combinations of parameter values investigated, being very stable in the Old World while decreasing by ~200–700 km between the present and the LGM in the Americas (Supplementary Figs. 8–16), except for when the cost of migration is very high (α = 0.005; Supplementary Fig. 13) as migratory species already travel very short distances to avoid the high energetic costs.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Information
Supplementary Movie 1
Supplementary Movie 2 | nature communications | [
"Article"
] | [
"Animal migration",
"Biogeography",
"Ecological modelling",
"Palaeoecology"
] | IntroductionBird migration rearranges avian assemblages to seasonality1–5 is labile trait6–9 research suggests original machinery of migration evolved in avian lineage8 can change environmental conditions12–17 previous phylogenetic analyses found high rates transition between sedentary and migratory behaviours Analyses bird migration patterns shown seasonal ranges of migratory species composition of avian communities explained by climatic factors5 global distribution birds at equilibrium with climate23 climate change affecting migration routes14 prevalence of migrant species climate change might contribute to changes in global migration patterns leading to gains or losses of migratory behaviour avifauna authors hypothesised variations in climate during glacial cycling Pleistocene migratory pathways7 shifts of seasonal breeding ranges from glacial into interglacial temperate regions could triggered migratory behaviour shaped global migration patterns increasing migration distances). importance of bird migration may be restricted to warm interglacial periods temperate climateUnderstanding importance prevalence magnitude migration avifauna glacial cycles for understanding migration gauging past seasonal dynamics functional roles birds in communities ecosystems addressing problem difficult migratory behaviour not recorded in fossil record fossils indicate species not species-specific studies used genetic data ecological niche modelling last glacial cycle evolution migration strong recent simulation analyses global bird migration23 suggested seasonal redistribution shaped by inter-specific competition for access limited resources mutual interference31 increasing search time33 territorial defence34 Investigating past dynamics bird migration from reconstructions at scale avifauna species-by-species models mechanistic model global seasonal distribution Seasonally Explicit Distributions Simulator (SEDS developed simulate spatial diversity patterns equilibrium between avifauna climate23 model relies energy efficiency optimisation energy budgets simulate seasonal distributions species climate reconstructions model provides unique opportunity for simulating global bird migration response avifauna to glacial cycles develop new version SEDS model integrates annual energy budgets details in Methods original modelframework modelling balance costs benefits migration energy common currency assuming local carrying capacity to species richness based primary productivity model simulates bird species’ seasonal breeding non-breeding ranges resident species different migratory species virtual world with similar geography seasonal distribution energy supply virtual bird species distribute energetic fitness energy reproduction survival optimising balance energy assimilation energetic costs migration distance distribution other species considering inter-specific competition for access energy supply (Fig. 1 SEDS model applied to each time slice climate reconstruction present to before present 1000 years to 22,000 BP 2000 years before 22,000 BP time slice 12,000 simulation starts virtual world empty of bird species = energy available to birds equal to energy supply estimated from net primary productivity (NPP) climate reconstruction each simulation step new virtual species added geographical distribution breeding non-breeding range selected among 1,040,000 candidate distributions range range seeded randomly across grown until fixed size using stochastic algorithm constrained by climatic conditions distribution with highest energetic fitness maximum value energy production selectedEprod computed as energy assimilated by species linear function of energy within geographical ranges (βEA minus energetic costs equal to 1 basal energy for survival plus linear function migration distance (α = 6.45e−5 model optimises balance between energy assimilated energy for travelling seasonal geographical distributions new species added to virtual world (R = n + 1 energy depleted in breeding and non-breeding ranges (sub-step 4) simulation ends when after nmax species maximum value for Eprod negative new species enough energy for costs survival exist apply SEDS model to past climate data simulate reconstruction global seasonal distribution of birds past 50,000 years period transition between last glacial period and current interglacial period effect of climatic changes on spatial patterns bird migration results indicate prevalence of avian migration stable past 50,000 years with geographical variations important throughout glacial cycles origin might be ancientResultsPredicting current global seasonal distribution of birdsThe SEDS model simulates distribution global avifauna in two seasons summer Northern Hemisphere Southern Hemisphere using simple rules mechanisms from principles ecology energetics single free parameter not estimated literature model performs well patterns current global seasonal distribution birds breeding migrants non-breeding migrants residents model predicts breeding migrants concentrate around 50°N strong asymmetry between northern southern hemisphere (correlation = 0.795 predicts non-breeding season migratory birds concentrate in southern northern hemisphere (correlation = 0.611 pattern resident bird diversity peak in tropics captured (correlation = 0.637 model underestimates magnitude peak underestimation total species richness). model’s good performance simulating breeding non-breeding patterns migratory avifauna supports energy efficiency optimising interplay energy assimilation energetic cost of travelling in driving bird migration.Predicting past global seasonal distribution mechanistic simulation-based model good explanatory power useful for predictions environmental conditions differentequilibrium between climate distribution global avifauna applied past used SEDS modelling framework simulate global seasonal distribution of migratory birds used climate reconstruction past 50,000 years 1000-year intervals 22,000 2000-year intervals global vegetation model estimates energy supply past years model predicts breeding distributions migratory bird species closer to Equator Last Glacial Maximum ~20,000 years noticeable in North America Western Palaearctic avian assemblages north of ~50°N poorer in breeding migrants prior to 10,000 BP geographical distribution of non-breeding migratory birds concentrated closer to Equator less noticeable breeding distributions 2Contrast between past present simulated patterns of migratory bird diversity global patterns computed as predicted richness past species per hexagon 10,000 years before 20,000 BP 50,000 BP minus present Red areas more species blue areas fewer model predicts variations in proportion bird species migratory during last glacial cycle In Americas proportion ~20% smaller at LGM species resident during last ice age migrating seasonallyopposite trend in Old World proportion migrants similar to today or higher during ancient periods (Fig. 3) similar asymmetry migration distances Americas model predicts migration distance LGM average ~500 km shorter today today birds travel ~40% longer distances than end last ice age Old World average distance migratory species slightly oscillated stable last 50,000 years.Fig. 3Predicted proportion of migrants average migration distance last 50,000 years Evolution proportion bird species migrants average distance between breeding and non-breeding grounds for migrant species simulated time series shown for Americas black Old World grey). 1 kyr = 1000 years model no ‘memory’ global seasonal distribution of birds simulated independently predicts stable number of simulated bird species over last 50,000 years boosting confidence model predictions not number avian species changed much findings suggest last 50,000 years last glacial maximum~20,000 years bird migration remained important global phenomenon refuting hypothesis mainly interglacial periods temperate climatecontrasts with results Zink and Gardner30 predicted climate for 56 North American migratory species most sedentary during last glacial maximum glaciations major ‘migratory switches’ Our results based point to origin of migration over longer time than glacial cycles simulations indicate avian response migratory behaviour to past global change likely variable across North America predicted greatest changes in bird migration since last age retreat of large Laurentide ice sheet predict southwards compression of bird migration breeding ranges predicted shift of transition zone between southern assemblages and northern receivers from ~35°N today to <30°N at LGM corresponds to significant increase in average migration distances since LGM (~40% increase proportion of species migratory (~25% increase western Palaearctic experienced important changes retreat ice seasonal grounds of migrants at LGM closer to equator than less pronounced than North America smaller extent Eurasian ice sheets little effect on proportion of species migratory or average distance travel higher in past. 4Fifty-thousand years of predicted seasonal difference in richness due to migration northern Western Hemispheremaps a–h show seasonal difference in avian richness breeding minus non-breeding migrants for eight slices between present and 50,000 years ago. Red colours higher richness during breeding season blue higher richness non-breeding season patterns seasonal redistribution avifauna emerge from optimisation energy budgets birds use migration energetic fitness resident species underestimated particularly around equator results in high of migrants resident species variation in seasonal distribution migratory avifauna unexplained by model suggests additional mechanisms need in future versions explain diversity patterns may include realistic approaches modelling migratory costs geographical features wind patterns41 predation risk42 model treats all species as equivalent equally abundant locally differences in evolutionary history ecology could important for explaining global patterns underestimate number species across high-latitude northern temperate regions over winter several bird lineages evolved adaptations other than migration to increase survival in seasonally energy-depleted environments43current model proportion species migrants original SEDS model23 suggesting model of net primary productivity might not adequate estimating energy supply remote-sensing data original SEDS model shifts in spatial distribution biodiversity are result of species’ responses44 development apply global model with single species models background community species-specific predictions on evolution migration since last glacial period investigating if rapid transitions between sedentary and migratory since last ice age by global model Estimating species-specific gains losses of migratory behaviour due to glacial cycling could inform phylogenetic analyses over time bring insights into origin evolution of migration results suggest bird migration systems across continents not responded same to past climate change differences in between continents can explain patterns today differences in migration strategies need for communication calls during migratory flights might higher in New World45 for migratory behaviours variable rapid anthropogenic climate change likely on distribution movement species biodiversity non-mobile species adapt to change highly mobile species might track changing environmental conditions magnitude flexibility of response bird migration to global change offers baseline for predicting migratory birds future climate changeMethodsClimate reconstructionMonthly data (temperature precipitation cloud cover past 50,000 years from HadCM3 global circulation data 1000-year intervals between 22,000 years before 2000-year intervals between 22,000 50,000 BP original simulation data on 2.5° × 3.5° bias corrected downscaled using delta method47 difference between simulated observed data approach robust paleoclimatic reconstructions for late Pleistocene49 downscaled paleoclimatic variables to 1/6 degree grid global grid equal-area equal-shape hexagons spacing ~153 HadCM3 simulations used global ice sheets reconstruction data set ICE-6G version 1.250 Hexagons covered by ice sheets not habitable for all birds Seasonal climate values for temperature precipitation averaging monthly values northern winter northern summer Visualisations reconstructions in Supplementary Movie 1.Model new version Seasonally Explicit Distributions Simulator (SEDS) model energy budgets free parameters balance between costs benefits energy common currency built on three components virtual species’ range options estimated energy requirements biological processes spatial seasonal variation of energy supply birdsIntegrating three components model applied through simulation steps virtual same geography seasonality real mapped global hexagonal grid filled with virtual species until saturated virtual bird species functionally equivalent differences represented by breeding non-breeding range congruent or different start simulation virtual world empty of species each step virtual species geographical distribution of each virtual species selected as with maximal energetic fitness (Fig. 1) newly simulated virtual species consumes energy equivalent to energetic cost depleting energy in all hexagons stop simulating species when world saturated species Each simulation performed separately on Western Hemisphere < Eastern Hemisphere (EH > model mechanistic characterisation state of system global seasonal distribution birds as functions of component parts geographical ranges energy budgets actions interactions competition for access energy supply.Virtual species range optionsFor each generated 1000 contiguous geographical ranges in virtual world (400 in WH, 600 in EH options distributions virtual species simulated (Fig. 1)range options 131 western 180 eastern median values global avian species range maps52 Ranges generated spreading dye algorithm53 climate-driven approach range expansion distribution bird ranges’ shape55 range seeded from single hexagon randomly selected\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$P_h = 1/\left {1 + S_h}\right\end{document}Ph=1∕1+Sh (Eq. (1)), Sh number species simulated hexagon h probability selecting hexagon seed hexagon function local richness virtual species avoid simulating range options clustered selected seed hexagon allowed stochastic spread into adjacent unoccupied hexagons climatic conditions until virtual range reached fixed sizeeach range initial climatic optimum obtained from position seed hexagon in space defined by mean annual temperature precipitation selected two neighbours seed hexagon probability selection higher for closer to optimum lower Euclidian distance d calculated as[12pt{minimal}{amsmath{upgreek\oddsidemargin-69pt}{select = 2(d + 1)^{ - 30}(d+1)−30 divided by sum values across neighbours decaying with increasing climatic distance d). repeated procedure redefining climatic optimum as average condition across selected hexagons selecting 25% larger integer of unoccupied neighbours probabilities until desired range size reached (131 hexagons, ~3.05 million km2 in WH, 180 hexagons, ~4.20 million km2 in EH). Visualisation geographical distribution of range options through time in Supplementary Movie 2.Energy supplyIn each hexagon energy supply ES is total available resources bird species’ metabolismmodelled proportional to Normalised Difference Vegetation Index (NDVI), NDVI remote-sensing measure greenness land reconstructed past modelled energy supply proportional to net primary productivity (NPP). Monthly NPP each time slice estimated with Biome4 global vegetation model56 using monthly temperature precipitation cloud cover reconstructions (Supplementary Movie 1) assumed hexagon j carrying capacity for birds proportional to NPP\documentclass[12pt{minimal}\usepackage{amsmath{wasysym\oddsidemargin-69pt}{document}\mathrm_j = log{NPP}}_j + 1{document}ESj=μ*log10NPPj+1 Negative NPP values set to zero parameter μ used adjusting energy supply carrying capacity model generate realistic total number virtual species not interested replicating total number bird species investigating spatial patterns global seasonal distribution of birds past used fixed value μ = 65 for simulations chosen obtain simulated species richness same as empirical dataconducted sensitivity analysis μ not crucial for results energy in hexagon\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}{\oddsidemargin{-69pt}{document}$$E\mathrm{A}}_j}EAj simulation equal to energy supply minus energy consumed by species simulated previous steps in hexagon j energy available species season computed as mean of energy in hexagons across seasonal range.Energetic energetic requirements species’ seasonal range modelled function of basal energy use for existence BEU 1 additional cost migration (mC), converted into units cost of migration corresponds to energetic cost travelling between breeding non-breeding rangesassumed mC increases distance travelled = 0 for resident migration instantaneously end each season cost added season previous investment fat reserves). each season mC computed function of great circle distance dm between centroids breeding non-breeding ranges (average distance travelled birds species shortest\documentclass[12pt{minimal}\usepackage{amsmath}{wasysym{upgreek}\setlength{\oddsidemargin-69pt}\begin{document}\mathrm{C}} = \alpha\mathrm{m}}{document}mC=α*dm (Eq. (4)), α parameter energy required bird travel distanceparameter α estimated from flight physiology57 equals to flight power (FW J/s divided by flight speed (FS m/s). rescale cost migration compared energy for migratory journey to basal metabolic rate minimum energy expenditure over season (BMRS\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{mathrsfs{upgreek-69pt{document}\alpha = \mathrm{BMR}}{document}α=FWFSBMRS (Eq. (5)). comparative studies found FW and FS scale with body mass (M\documentclass[12pt]{minimal}{amsmath{upgreek\oddsidemargin-69pt}{document}_{\mathrm{W}} = 0.257M^{0.763}{document}FW=0.257M0.763 (Eq. (6)) (estimated using data from ref.cost forward flapping flight 31 avian species excluding seabirds\documentclass[12pt]{minimal{amsmath\oddsidemargin-69pt}{document\mathrm{S}} = 6.4773M^{0.13\end}FS=6.4773M0.13 (Eq. (7) (estimated ref. 59 cruising speed 138 species migratory birds used allometric relationship basal metabolic rate (BMR mlO2/h) 211 avian species ref. 60\documentclass[12pt]{minimal}{amsmath\oddsidemargin-69pt}{document\mathrm{BMR}} = 6.7141M^{0.6452}{document}BMR=6.7141M0.6452 (Eq. converted to J/s conversion factor 1J/s = 172mlO2/h multiplied seconds 6 months~15,724,800) obtain\documentclass[12pt]{minimal{amsmath{wasysym\oddsidemargin-69pt}\mathrm{BMR}} = 6.15{e}}^5M^{0.6452}BMRS=6.15e5M0.6452 (Eq. (9)). estimation α independent of body mass cost migration equal[12pt]{minimal}{amsmath{upgreek\oddsidemargin-69pt}\mathrm{c}} = 6.45{\mathrm{e}} 5}mc=6.45e−5dm (Eq. dm travel distance in kilometres energetic cost for migration ~0.065 or ~6.5% yearly basal energy use species travels average 1000 km between breeding non-breeding ranges.Maximising energetic model development ref. 23, modelled fitness energetic definition ref. 61). Birds assimilate biochemical energy solar radiation via photosynthesisassimilated energy (Eassim) fuels metabolic processes respiration production biomass energy translates into two components fitness survival (Esurv) production (Eprod). relationship derived from definition:11\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}_{{\mathrm{prod}}} = E{assim - E{surv\end{document}Eprod=Eassim−EsurvWe birds maximise Eprod each virtual species for distribution with highest value for year-round Eprod estimated annual Eassim and Esurv.estimate Eassim season assumed type I functional response birds energy supply\documentclass[12pt]{minimal}{amsmath}{wasysym{upgreek}\oddsidemargin-69pt}\mathrm{assim = \beta E_{A}}}Eassim=βEAwhere β parameter governing linear relationship estimate Esurv season computed basal energy use existence (BEU), set 1 arbitrary unit energetic cost migrating seasonal distributions details section energetic costs\documentclass[12pt]{minimal}{amsmath}{wasysym{upgreek\oddsidemargin-69pt}$E\mathrm{surv}}} =\mathrm{BE}}{U}} +\alpha d{m}} = 1 + 6.45{{e}}^{ - 5}d_{m}}{document}Esurv=BEU+αdm=1+6.45e−5dmThe year-round energy production for candidate distribution obtained using formula:14\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin{-69pt}\begin{document}_{{\mathrm{prod}}} = \beta \left( {E{\mathrm{a}}\mathrm{NS}})} +{NW}})}} - 2\left( {{\mathrm{BE}}_\mathrm{U}} + \alpha d_{\mathrm{m}}} \right\end{document}Eprod=βEa(NS)+Ea(NW)−2BEU+αdmwhere NS northern summer NW northern winter geographical distribution breeding non-breeding of virtual species selected as highest value for Eprod highest energetic optimal balance between energy assimilated energy travelling breeding season highest Eprod energy reproduction geographical range assigned as breeding distribution other range non-breeding distributionsimulated virtual species consumes energy within seasonal ranges equivalent to Eassim depleting energy in hexagons stopped simulating species when Eassim value below BEU species assimilate enough energy basal for survival virtual world saturated with simulated species no new species added.Global patterns in seasonal distribution of birdsThe SEDS model outputs virtual species seasonal distributions global diversity patterns mapped generated three spatial patterns global seasonal distribution of terrestrial birds ‘richness in breeding migrants’ non-breeding migrants’ in residents’ species year-round quantified patterns using empirical data on bird distribution spatial polygons global distribution of 9783 non-marine bird species from BirdLife International NatureServe52 data treatment patterns described in ref. 23.Parameters improved SEDS model has one free parameter not estimated: β determines type I functional response between energy and energy assimilationexplored values for parameter:\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\beta}\left {0.003, ,0.035}\right\end{document}β∈0.003,...,0.035 step of 0.001.Values below 0.003 virtual species energy below requirement survival bounded range values to upper limit 0.035 above energy assimilation for first simulated species unrealistic >17 times basal energy for survival assess quality model outputs β parameter value computed correlation coefficient between empirical simulated patterns for global seasonal distribution birds summing breeding non-breeding residents.Low values β lead to poor model performance low correlation between low number species generated For β > 0.01 performance plateaus above mean correlation 0.6 good performance β > 0.015 correlation decreases as β increases total species richness peaks between 0.01 > β > 0.015 not above 4000.less than half actual bird species world every value β investigated model predicts proportion migrants global avifauna higher than real > 45% vs. 15% selected β = 0.012 best-fit most realistic value for back-casting global seasonal distribution birds value compromise between maximising match to empirical patterns generating maximum species minimising total proportion migrants (Supplementary Fig. 3).Sensitivity explored robustness results best-fit model running with varying values for α parameter cost of migration\documentclass[12pt{minimal}{amsmath{wasysym-69pt}{document}$\alpha\left {0.00001, ,0.0005}\right\end{document}α∈0.00001,...,0.0005, μ parameter for rescaling NPP into energy supply[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin{-69pt}{document}$$\mu\left {45, ,150}\right\end{document}μ∈45,...,150.ran model parameters fixed investigated performance re-producing patterns global seasonal distribution birds explored sensitivity to geographical range options investigated performance using different values range size number of hexagons\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document} {100, ,200}}∈100,...,200 for range options western hemisphere number of hexagons\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}\oddsidemargin}{-69pt} {150, \ldots ,250}{document}∈150,...,250 for range options eastern hemisphere), varying strength climatic constraints when growing range optionsvarying probability selecting neighbours calculated as 2(d + 1)−x investigating values x\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\in\left {1, ,45}\right\end{document}∈1,...,45 model performances robust to variations μ rescaling NPP energy supply for birds range options simulated Varying μ affect model predict global seasonal distribution birds correlations affect total proportion migrants simulated μ increases total number species generated increases linearly μ determines energy supply carrying capacity virtual world Varying size geographical range options climatic constraints determining shape affect model performance correlations >0.6 total proportion of migrants simulated Increasing size simulated range options led linear decrease total number simulated species less species in virtual world for total carrying capacitycontrast analyses variation in α cost migration model performances high values (α > 0.001) model poorly capturing global seasonal distribution birds performance decreasing α below α = 0.0005 direct estimation of α (α = 0.000645 within peak of model performance 0.0004 > α > 0.0008) boosts realism model peak performance corresponds to dip in total number of simulated species with α total proportion of migrants decreases linearly with increasing α corresponds to increase cost migration reconstructed global seasonal distribution of birds last 50,000 years for values three for β investigated outputs for α μ performing well predicted evolution of global migration 50,000 years robust to variation in β, α μ total number of species simulated stable slight decrease between 10,000 BP and 20,000 BP for combinations values proportion of migrants among species stable slight decrease in Americas between 10,000 BP and 20,000 BP combinations model exhibits similar pattern for average migration distance among simulated migratory species for combinations values stable in Old World decreasing by ~200–700 km between present and LGM in Americas8–16) cost migration high (α = 0.005 Fig 13 migratory species travel short distances avoid costs.Supplementary information Peer Review File Movie 1 2 | 48.6 | 0.845402 |
10.1038/s41467-020-15842-4 | PMC7239903 | Thalamic head direction (HD) cells are necessary to establish spatial maps in the hippocampus. Here, the authors show that HD cells tuned to a particular direction are coupled to individual hippocampal ripple events during sleep, suggesting an influence of the replay of specific trajectories during sleep memory consolidation. | The anterior thalamus is a key relay of neuronal signals within the limbic system. During sleep, the occurrence of hippocampal sharp wave-ripples (SWRs), believed to mediate consolidation of explicit memories, is modulated by thalamocortical network activity, yet how information is routed around SWRs and how this communication depends on neuronal dynamics remains unclear. Here, by simultaneously recording ensembles of neurons in the anterior thalamus and local field potentials in the CA1 area of the hippocampus, we show that the head-direction (HD) cells of the anterodorsal nucleus are set in stable directions immediately before SWRs. This response contrasts with other thalamic cells that exhibit diverse couplings to the hippocampus related to their intrinsic dynamics but independent of their anatomical location. Thus, our data suggest a specific and homogeneous contribution of the HD signal to hippocampal activity and a diverse and cell-specific coupling of non-HD neurons. | IntroductionThalamocortical loops form the canonical circuits of the most complex cognitive functions1–4. The dorsal thalamus (henceforth referred to as the thalamus for simplicity) is organized in a large family of cytoarchitectonically defined nuclei1. Among them, the anterior thalamic nuclei (AT) are located at the central stage of the Papez circuit5 and play a key role in memory6,7, spatial navigation8–10, and arousal states11,12. However, the dynamics of AT neurons and their functional integration with the limbic system remain elusive.The role of AT in navigation is indicated by the presence of head-direction (HD) neurons, which discharge when the animal is facing a particular direction and are primarily located in the anterodorsal (AD) nucleus8,10. HD cells of the AD are necessary for the establishment of higher level spatial representations in the brain’s navigation system, for example the grid cells of the medial entorhinal cortex9.The relationship between neuronal activity and memory in the AT remains elusive. In contrast, this relationship has now been clearly established in the hippocampus. Non-Rapid Eye Movement (NREM) sleep is instrumental for memory consolidation13. During NREM sleep, the hippocampal activity patterns that form during exploration are reactivated14, thus recapitulating spatial trajectories in the environment15,16. These replay events are associated with SWRs in local field potential (LFP)17 and are necessary for memory18–20. However, it remains unclear whether this phenomenon is entirely generated within the hippocampus or is influenced by its inputs.During NREM sleep, the timing of SWRs is orchestrated by thalamocortical dynamics21–24, especially the thalamocortical slow oscillation that arises from the fluctuation of membrane potentials between UP and DOWN states (corresponding to active and silent states, respectively)25. The AD is a key hub for the propagation of the slow oscillation12, suggesting that HD cells play a role in coordinating neuronal activity within the limbic systen during NREM sleep.The contribution of AD HD cells to offline processing of spatial signals is further demonstrated by the fact that HD cells maintain a coherent firing activity such that a single direction can be decoded at any moment during sleep10,26). Additionally, AD HD cells lead their main cortical targets (in the post-subiculum), independently of brain states27. This suggests that during sleep AD neurons continue conveying a coherent HD signal to the navigation system, including the hippocampus, independent of the current heading of the animal28.The role of the other AT nuclei in coordinating activity in the limbic system and how this activity relates to neuronal dynamics is still unclear. One challenge when investigating thalamic function is that, while each nucleus is characterized by specific connectivity with other brain areas, exact input/output patterns can differ across neurons from the same anatomically defined nucleus1,29. Furthermore, the firing of thalamic neurons is often believed to be homogeneous (e.g. burst firing during NREM sleep25,30), yet variability in spike train dynamics has been previously reported31,32.By simultaneously recording thalamic neuronal activity and hippocampal LFP in freely moving and naturally sleeping mice, we first addressed the question of how AT neurons, and specifically thalamic HD cells, are coupled to SWRs. Then, we examined how the variability in this coupling correlates with intrinsic properties of AT neurons, as determined from spike train dynamics across brain states and timescales. We show that HD cells are specifically coupled to SWRs, increasing their gain and firing coherently for a particular direction during SWRs. Furthermore, HD cells of the AD nucleus (the majority of cells in this nucleus) form a very homogeneous population of neurons, both in terms of their intrinsic dynamics as well as hippocampal coupling. Non-HD cells offer a striking contrast: their modulation by hippocampal SWRs and intrinsic dynamics are highly diverse and differ even for proximal neurons of the same nucleus.ResultsThe HD signal is precisely coupled to SWRsWe performed recordings of the anterior thalamus (n = 2016 AT neurons) in freely moving mice in an open environment, while they foraged for food, and in their home cages during sleep periods that preceded and followed exploration. Neurons were classified as HD cells by measuring the modulation of firing rate with respect to the direction of the head of the animal in the horizontal plane (n = 161 HD cells; see Methods). The tuning curves of 24 simultaneously recorded HD cells are shown in Fig. 1a.Fig. 1Stability of HD decoding before SWRs.a–e HD decoding with ISOMAP around time of SWRs for one example session. a Tuning curves of HD neurons recorded simultaneously in the AD nucleus. b Simultaneous recording of CA1 pyramidal layer LFP (top) and neuronal activity in AD (same neurons as in a) around the time of SWRs during NREM sleep. Spikes from HD neurons are sorted according to their preferred direction during wakefulness (as in a). Red asterisks indicate SWRs. c Two-dimensional embeddings (using ISOMAP) of HD neuronal population activity during wakefulness and around the times of SWRs (colored and gray points, respectively). Curves display the population activity embeddings of the three examples shown in b (same red asterisks). d Distribution of distances from the embedding center (c) during wakefulness (blue) and NREM sleep (gray). e Distribution of average angular directions around the time of SWRs (±30 ms). f Average normalized radii (i.e. distance from embedding centers relative to shuffles) around times of SWRs (gray lines; n = 7 sessions, 3 animals) and session-wide average (black line). g Same as f for angular velocity (i.e. angular displacement between two consecutive activity bins). h Average density of spikes (±s.e.m.) for simultaneously recorded HD (black) and non-HD (gray) neurons, and rate of occurrence of SWRs (red line) during DOWN and UP states (normalized time).HD cells of the AT are believed to convey a coherent signal to the parahippocampal system during sleep10,27,28. We thus made the prediction that the HD system provides the hippocampus with a specific direction during SWRs in which place cell ensembles replay previously formed patterns associated with spatial trajectories15,16. To this end, we simultaneously recorded LFP in the pyramidal layer of the dorsal CA1.Around the time of SWRs only a subset of HD neurons fired, and these neurons demonstrated a similar preferred HD (Fig. 1b). To better visualize this effect, we first uncovered the topology of population firing using ISOMAP (see Methods)26. This non-linear dimensionality reduction method revealed the embedding of HD cell population vectors, and was computed without knowledge of the behavioral correlates (i.e. the HD tuning curves). During wakefulness, this topology had a ring structure with the relative angular position along the ring clearly mapping the animal’s HD (Fig. 1c). While population activity was restricted to the ring during wakefulness (the center of the ring corresponds to “forbidden states”), its intrinsic dimensionality increased during NREM sleep activity, as seen by the scattering of population vector embeddings, in particular at the center of the ring. This apparent violation of the intrinsic topology resulted from the modulation of network activity by the slow oscillation25,26.The population activity around the time of SWRs corresponds to trajectories within the projection space. In the three cases shown in Fig. 1b, activity started near the ring’s center (corresponding to putative DOWN states) and settled on fixed points in the outer part of the ring (corresponding to “allowed” states during wakefulness) at the time of SWRs.To quantify population states around SWRs and across sessions, the population vectors at each time bin were expressed in polar coordinates by taking the center of the ring as the reference point. We analyzed data sets that included more than 10 HD neurons recorded together and showed a clear ring topology during wakefulness (n = 7 sessions, 3 animals, see Supplementary Fig. 1). This topology was assessed by examining the distance of each point (i.e. a population vector projected on the manifold) to the center (i.e. mean) of the embeddings. During wakefulness, distances were all distributed within a fixed range from the center and, importantly, no point was close to the center (Fig. 1d and Supplementary Fig. 1). This indicates that the embeddings form an annulus-like shape around the center, as expected for a population of HD cells26. At time of SWRs, the HD population pointed to near-random directions (Fig. 1e and Supplementary Fig. 1), and the slight bias towards one direction certainly results from the non-homogeneous sampling of preferred HD in the population (due to the limited number of recorded neurons).For all sessions examined, the radius of the trajectories peaked around times of SWRs compared to baseline (see Methods) (Fig. 1f). Average radius was maximal before SWR peaks (−17 ms ± 6 ms; t = −2.82, p = 0.03; t-test). Was the angular position stable or was it drifting at these times? To address this question, we computed the relative angular velocity of the signal (i.e. angular displacement relative to a baseline). Angular velocity was minimal around SWRs (Fig. 1g), and, similarly to the radius, reached its minimum before SWR peaks (−18 ms ± 15 ms; t = −2.88, p = 0.02; t-test excluding SWRs occurring <100 ms after UP onset). Similar results were obtained upon using Bayesian decoding of angular position (i.e. a decoding based on HD tuning curves; see Supplementary Fig. 2)10,33.It is possible that the stabilization of the network into particular states results from the increase of HD population firing. However, these two aspects of HD cell population activity seem independent as HD cell population outside of SWRs showed fast sweeps at high population firing rates (i.e. on the outer ring of the population activity topology)26. Hence the increase in gain and the stabilization of the network into particular states is characteristic of epochs preceding SWRs.The coupling between HD neuronal population and SWRs could result from non-specific coordination of SWRs with slow oscillations22,23,34, in particular at the transition between DOWN and UP states. To test for this possibility, we first detected putative DOWN and UP states from ensemble activity (composed of HD and non-HD neurons) and duration of each epoch was normalized (Fig. 1h). We did not see any change in the probability of SWR occurrence around DOWN/UP transitions of AT activity, but instead saw a constant rate of SWR occurrence during UP states (Fig. 1h). If anything, SWRs were slightly less prevalent during early onset of UP states. In contrast, when the modulation of SWRs by the slow oscillation was examined with a cross-correlation analysis which does not correct for UP state duration, SWRs seemed more prominent at the beginning and end of the UP states (Fig. 1h, inset). However, this analysis primarily reflects the distribution of UP state duration and not a preferred phase within UP states. Finally, we asked whether the co-occurrence of SWRs and stable HD signal was biased by their relative position within UP states. SWRs were clustered according to the time elapsed since the UP state onset. No difference was observed in the time-lags of the radius peaks and minima of angular velocity relative to SWRs. The only exception was for SWRs occurring in the first 200 ms of UP state as, at those times, the system is certainly more governed by dynamics relative to the transition from the DOWN states (Supplementary Fig. 1b, c).In conclusion, SWRs were specifically preceded by high-activity and stable HD population states independently of the coordination between the thalamocortical slow oscillation and SWRs.Homogeneous firing of thalamic HD cells around SWRsThe observed coupling of the HD network to SWRs raised the question of whether or not this was specific to HD AT neurons. Non-HD AT neurons were recorded at increasing dorsoventral depth from session to session (Fig. 2a), thus spanning most AT nuclei, not only the most dorsal nuclei adjacent to AD. Taking advantage of multiple channels and geometry of silicon probes we were able to reconstruct the putative tracks of each shank across the AT (Fig. 2b). The position of the probes was calibrated along the dorso-ventral axis by the first appearance of spiking activity on at least one shank (indicating the dorsal border of the thalamus) and along the medio-lateral axis by matching the highest density of HD neurons with the AD nucleus8,10) (Fig. 2b and Supplementary Fig. 3).Fig. 2Homogeneous coupling of HD, but not non-HD neurons to SWRs.a Histology of an example mouse (Mouse 17, DAPI staining). Note the tracks of the eight-shank silicon probe and lesion sites at the end of the tracks. Contour of the AD nucleus is shown in red. b Schematic of the AT (coronal plane) with recording sites (black dots) for the same animal as in a. Red dots indicate where HD neurons were detected. c Top, spikes for successive SWRs. Middle, mean firing rates. Bottom, z-scored modulation (see text) for one HD neuron (red) and 2 non-HD neurons (gray) differently modulated by SWRs. d Top, average SWRs modulation (±s.e.m.) for HD (red) and non-HD neurons (black) for all sessions. Middle, bottom, SWR cross-correlograms for HD neurons (top) and all other AT neurons (bottom). Neurons were sorted according to the level of correlation with SWR at zero-time lag. e Examples of SWR cross-correlograms for HD (top) and non-HD neurons (middle) recorded during the same session but on different shanks (arrow in b). Bottom, Pearsonʼs correlation coefficients between SWR cross-correlograms of pairs of neurons recorded on the same shanks. Red dots indicate pairs of HD neurons. f Distribution of Pearson’s correlation coefficients between SWR cross-correlograms of pairs of HD (red) and non-HD (gray) neurons (similar to e) recorded on the same shanks, for all sessions. AD anterodorsal, AM anteromedial, AVd anteroventral, dorsomedial part, AVv anteroventral, ventrolateral part, IAD interanterodorsal, LD laterodorsal, MD mediodorsal, PV paraventricular, sm stria medullaris, VA Ventral anterior.As no particular behavioral correlates and no topology of the population activity could be assumed for non-HD neurons, we first analyzed neuronal activation around times of SWRs on a cell-by-cell basis. Similar to the potential caveat of HD cell population analysis (Fig. 1), a challenge when evaluating neuronal responses to SWRs with cross-correlation analysis is that SWRs are co-modulated with the slow oscillation21–25. To isolate the specific responses of neurons to SWRs, cross-correlograms were compared to a baseline expected in the condition of an absence of fast co-modulation between neuronal spikes and SWRs (within ±150 ms, corresponding to the typical duration of a sharp wave; see Methods). Cross-correlograms were then expressed in a number of standard deviations (z) from these null distributions.Three example neurons (including one HD neuron) showed different modulation by SWRs (Fig. 2c). The average SWR modulation of HD neurons shows a clear population response with a highly synchronized increase of firing before the SWRs (Fig. 2d), thus confirming the specific coupling of the HD signal to the SWRs (Fig. 1). On the contrary, non-HD cells showed various responses around the time of SWRs which resulted in a near-uniform average modulation. While this observation suggests that unlike HD cells, non-HD cells do not act as a synchronized population immediately before SWRs, the non-HD cell populations were recorded from different nuclei of the AT (Fig. 2b). It is thus possible that, locally, some nuclei show the same coupling to SWRs as the HD neurons of AD. To rule out this possibility, we computed pairwise correlations of neurons recorded on the same shanks, assuming that those neurons were most likely to come from the same nuclei. As shown in Fig. 2e, in a session containing HD and non-HD neurons, SWR modulation of non-HD neurons was more heterogeneously distributed than that of HD cells. This locally non-homogeneous response of non-HD cell population to SWRs compared to HD neurons was confirmed by computing the distribution of pairwise correlations within all shanks and across sessions (Fig. 2f; n = 7234 pairs, 58 sessions, t = −18.08, p = 10−71, t-test).It can thus be concluded that HD neurons (and, by extension, neurons of the AD nucleus) fire homogeneously around the time of SWRs, pointing in a particular direction. On the contrary, the activation of other AT neurons around the time of SWRs is highly variable, even within local networks (presumably within a nucleus).Modulation by the hippocampus is brain-state invariantWhat is the origin of the high variability in the coordination of AT neurons with SWRs? These interactions certainly depend on the input/output connections of each neuron. In the case of a hard-wired network, we hypothesized that the coordination of AT neurons with hippocampal activity should not depend on brain states. During wakefulness and REM sleep, the hippocampus is dominated by theta oscillations (6–9 Hz)35 which modulate neurons in the entire limbic system35, including in the AT36. We thus tested for a relationship between SWR modulation (during NREM sleep) and phase coupling to theta oscillation (during wakefulness and REM sleep) (Fig. 3a).Fig. 3Relationship between temporal coupling of AT neurons to SWRs and modulation by theta oscillations.a Example of a simultaneous recording of LFP in the CA1 pyr. layer (top) and neuronal activity in the AT (bottom) during REM sleep. Red ticks indicate spikes from HD neurons, sorted according to their preferred direction during wakefulness. b Top, spikes for successive theta cycles. Bottom, histograms of spike density within theta cycles for three example neurons during REM sleep. c Theta modulation versus SWR energy for all neurons (red points indicate HD neurons). Side panels indicate distribution of each group. “Theta mod” (blue distributions) indicates the group of neurons significantly modulated by theta (p < 0.01). d Density map of theta-modulated neurons (min = 0.016) for one mouse. Note the absence of theta modulation for AD. e
Z-scored SWR cross-correlograms for all theta-modulated AT neurons (equivalent to the group theta-mod in c). Neurons are sorted according to their modulation at SWR peaks. f First two jPC vectors (black and gray) of SWR cross-correlograms. g Projection of individual SWR cross-correlograms onto the first two jPCs. Point colors indicate 0-lag modulation (same color code as in e). Gray arrows indicate corresponding SWR-jPCA phase for four example neurons (green points, a-d). Theta phase distributions for the same four neurons are shown at each corner. h SWR-jPCA phase as a function of theta phase (r = 0.18, p < 0.001, circular correlation). The four example neurons are shown in green.The firing of AT neurons relative to REM theta was analyzed, as commonly done, in terms of preferred phase and modulation amplitude. Three example AT neurons, recorded simultaneously but on different shanks of the probe, showed different theta modulation profiles, both in amplitude and preferred phase (Fig. 3b). Overall, 38% of all AT neurons were significantly modulated by theta rhythm during REM sleep (n = 767/2016, p < 0.001). This proportion was lower during wakefulness (n = 333/2016, p < 0.001).How can theta modulation be compared to SWR responses? We first quantified the “SWR energy”, defined as the variance of the normalized cross-correlograms (as in Fig. 2). SWR energy and theta modulation amplitude were correlated for the ensemble of AT neurons (Fig. 3c; r = 0.81, p < 0.014, Pearson’s correlation). Those similarities in modulation strength by hippocampal dynamics across different brain states possibly reflect the strength of inputs to each AT neuron from the parahippocampal area, potentially through multiple synaptic pathways. If so, this should also be reflected in the temporal response of each neuron.Interestingly, HD neurons were overall not modulated by theta (only n = 21/161 HD neurons), thus showing once again that HD neurons of the AD nucleus form a homogeneous population of neurons and certainly share the same input/output connectivity profile. Furthermore, the anatomical density of theta-modulated neurons revealed a clear segmentation: they can be found anywhere in AT except in the putative location of AD (Fig. 3d, see Supplementary Fig. 2 for the three other mice).Firing of AT neurons relative to SWRs cannot, overall, be trivially described (e.g. excited or inhibited). Rather, they show a wide range of temporal profiles (Fig. 3e for theta-modulated neurons). While modulation by oscillations is commonly characterized in terms of phase preference, such description for modulation by SWRs is lacking. To capture the dynamics of AT neurons around time of SWRs, we used jPCA, a method that captures the rotational dynamics of a neuronal population during non-periodic behavior37). The projection onto the jPC subspace describes the various temporal responses of a population of neurons in a two-dimensional trajectory during a pseudo-cycle. Specifically, we determined the jPC basis from the ensemble of z-scored cross-correlograms relative to SWRs (see Methods) and projected each cross-correlogram on the first two jPC components (Fig. 3f). In the two-dimensional jPC projection space, each neuron can be attributed a “phase” that corresponds to the angle from the positive direction on the first jPC axis.The resulting phases were, for four example neurons, in good agreement with their preferred phase to theta oscillations (Fig. 3g, note that arrows point towards similar directions for theta and SWR phases). In the population of AT neurons that were significantly modulated by theta, theta phases were correlated with SWR phases (Fig. 3h; r = 0.18, p = 2.3 × 10−7, circular correlation). Hence, AT neurons were similarly modulated both in amplitude and in time by hippocampal population dynamics in all brain states, revealing an invariant property at the circuit level.A link between firing dynamics and hippocampal modulationAlthough the relationship between the modulation of individual AT neurons by hippocampal dynamics and their detailed connectivity profile is intractable in vivo, does this modulation depend on other intrinsic neuronal characteristics available from extracellular recordings? Spike train dynamics, which are well captured by auto-correlation functions (or “auto-correlograms”), reflect the complex interaction between morphological, input, and membrane properties. Individual HD neurons exhibit quantitatively different auto-correlograms during wake, REM sleep and NREM sleep. Yet, as shown for two example HD neurons, both responses to SWRs and brain state-specific auto-correlograms are similar between HD neurons (only their firing rates, and thus absolute levels of auto-correlation, are different; Fig. 4a). Examination of three non-HD example neurons recorded simultaneously and on the same shank illustrates that nearby neurons can share common properties but can also be strikingly different (Fig. 4b, c). Specifically, a pair of simultaneously recorded neurons from a given shank had similar auto-correlograms and SWR cross-correlograms (Fig. 4b). On the same shank, a third nearby neuron (see anatomical distribution of waveforms) showed largely different spike train dynamics and opposite modulation by SWRs (Fig. 4c).Fig. 4Covariation between single-cell spike train auto-correlograms and SWR cross-correlograms.a SWRs cross-correlograms (top left) and auto-correlograms across brain states (bottom) for two example HD neurons. Circular plots (top right) indicate HD tuning curves. b, c Same as a for three non-HD neurons recorded simultaneously on the same shank. Top right, average waveforms on the eight recording sites of the shank, from dorsal (top) to ventral (bottom) positions. The neuron shown in gray is the same for both panels. Note that, although the cell bodies of the two neurons in c are close to each other (see waveforms), auto-correlograms and responses to SWRs are strikingly different. d Same as a–c but for three neurons recorded on different shanks and different sessions in the same animal. Their anatomical location is shown on the left-side map of the anterior thalamus. e PCA projections of first 20ms of auto-correlograms (top) and SWR cross-correlograms (bottom) for the three example neurons of b, c. f Correlation matrix between the two series of PCA weights (left) and for shuffled neuronal identities (right) for all neurons. ρ2: total correlation. g Distribution of total correlation ρ2 for actual (red line) and for random shuffling of cell identities (1000 times) between all cells irrespective of shank and recording days (black) and between cells recorded on the same shank on a given day (gray). h Distribution of total correlation ρ2 for actual (vertical lines) and shuffled data between SWRs and auto-correlograms from wake (red), REM sleep (light blue) and NREM sleep (dark blue). Dashed line shows correlation between SWRs and stacked auto-correlograms of all epochs (same as in g). i Same as h but for cell identities shuffled within shanks.The previous examples suggest a high variability in both spike train dynamics and responses to SWRs, even within a given nucleus (with AD being the sole exception). Could these two properties of neuronal firing be related to each other? First, we asked the question of whether neurons showing the same response to SWRs across sessions (and, thus, anatomical locations) had similar auto-correlograms. This was indeed the case for the three example neurons shown in Fig. 4d. To further test for a possible relationship between individual spike train dynamics and SWR responses, we directly quantified the correlations between the two measures. To this end, principal component analysis (PCA) was used to reduce the dimensionality of the features (i.e. the number of time bins) of the first 20 ms of auto-correlograms (from wake, REM and NREM epochs) and cross-correlograms with SWRs (±500 ms). The first ten components were considered (explaining 99% and 91% of the variance of auto-correlograms and SWR cross-correlograms, respectively) (Fig. 4e). The correlation matrix of these weights had visibly more structure than the correlation matrix of PC weights in which neuron identities were shuffled (Fig. 4f; the diagonal blocks of the full correlation matrix are not displayed because PCs are orthogonal to each other). In fact, the total correlation ρ2 of PC weights was significantly higher than shuffles (Fig. 4g; ρ2 = 0.31, p < 0.001).Shuffling neurons recorded simultaneously on a given shank led to a distribution of total correlation that was still significantly smaller than the actual total correlation (Fig. 4g), thus indicating that the relationship between spike train dynamics and responses to SWRs is specific to each neuron rather solely determined by anatomical location. However, the within-shank shuffled correlations were higher than when cell identities were shuffled irrespectively of recording days and shanks (t = −82; p = 10−10), suggesting some similarity in both auto-correlation and SWR response within each nucleus. The strong correlation at the individual cell level cannot be accounted for by differences in firing rates (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{rate}^{2}=0.3$$\end{document}ρrate2=0.3 vs. ρ2 = 0.31) nor in burstiness alone (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{burst}^{2}=0.3$$\end{document}ρburst2=0.3). Because SWR-associated firing could potentially bias the auto-correlograms during NREM sleep, we repeated the analysis by considering auto-correlograms of only one brain state at a time. The total correlation remained significantly higher than shuffled data (p < 0.001) when wake, REM sleep and NREM sleep auto-correlograms were used separately (Fig. 4h-i). Besides, the relationship between auto-correlograms and SWR cross-correlograms holds when HD neurons were excluded (p < 0.001, data not shown). These results constitute a demonstration that intrinsic dynamics of individual AT neurons are directly related to their participation in circuit-level activity.Spike train dynamics of non-HD AT neurons are heterogeneousThe link between spike train dynamics and coordination with SWRs, independent of the anatomical location of the neurons (except for HD cells), begs the question of whether there exist different sub-classes of neurons in the AT based solely on the distribution of spike train auto-correlograms. An example HD neuron (cell 1 in Fig. 5a) had a relatively short refractory period across brain states. It also showed a low level of burstiness, if any, during NREM sleep (identified as a peak in the first 8 ms of the auto-correlogram; for the sake of simplicity, burstiness is reported only during NREM sleep), in agreement with previous intracellular studies of AD neurons32. In contrast, example cell 2 in Fig. 5a showed a high level of burstiness, as expected for typical thalamic neurons during NREM sleep25,30, and had a longer refractory period than the example HD neuron (cell 1) during wakefulness and REM sleep. A third cell (cell 2) showed intermediate properties, i.e. a long refractory period but a total lack of burstiness. We thus hypothesized that neurons can be segregated based on their auto-correlograms.Fig. 5Variability and clustering of spike auto-correlograms in the AT.a Auto-correlograms of an HD (1), non-bursty (2) and bursty (3) neuron during wake, REM sleep and NREM sleep epochs. b Clustering and t-SNE embedding of auto-correlograms (each point represents the stacked auto-correlograms from wake, REM sleep and NREM sleep of a neuron). HD neurons are marked with a white dot. The three example neurons from c are circled. K-means clustering of the auto-correlograms result in two clusters (see text). Cluster #2 is color-coded relative to burst index during NREM sleep. c Left, 15 superimposed auto-correlograms of randomly selected cells from cluster #1 (normalized by the baseline between 50 and 100 ms), during REM (top) and NREM (bottom) sleep. Color (white to dark red) indicates average firing rates (from low to high). Note the high similarity of auto-correlograms during REM sleep, independent of basal firing rates, and the mild variability during NREM sleep. d Left, density of cluster #1 (i.e. mostly HD neurons) and, right, mean burst index of cluster #2 over the anatomical schematics, for the same example mouse. e Distribution of Pearson’s correlation coefficients between pairs of neurons recorded on the same shank showing the dynamical homogeneity of HD neurons compared to non-HD neurons. f Classification score of HD versus non-HD neurons (based on auto-correlograms) for increasing duration of auto-correlation. Score is relative to classifiers trained with shuffled data (0, chance level; 1, perfect classification).To visually determine the clustering of auto-correlograms, we used t-distributed stochastic neighbor embedding (t-SNE38,) to project the auto-correlograms from the three different brain states (i.e. three auto-correlograms per neuron) in a two-dimensional embedding. Instead of distinct groups, AT neurons were continuously distributed (along a gradient of burstiness, among other factors) with the exception of HD neurons that formed a separated “island” (Fig. 5b). To confirm this bimodal distribution, we applied the common clustering algorithm, K-means (with k = 2), directly to the space of auto-correlograms. The first resulting cluster contained most of the HD neurons (99/127, p < 0.001, binomial test). Within cluster #1, the shape of auto-correlograms was highly similar and independent of average firing rates (Fig. 5c). Moreover, the anatomical density of the neurons belonging to this particular cluster was concentrated around the putative location of the AD nucleus (Fig. 5d). These observations demonstrate that the dynamics of spike train emission, independent of excitability (i.e. firing rate), is sufficient to categorize HD neurons. There was not a direct match between cluster #2 and anatomy, as shown for example by the broad anatomical distribution of average burstiness. These observations were confirmed by the high level of correlation between auto-correlograms of neuronal HD pairs recorded on the same shanks. In contrast, pairs of non-HD neurons from the same shanks had less correlated auto-correlograms (Fig. 5e, t = 18.32, p = 10−73, t-test).How much information was sufficient to classify HD versus non-HD neurons? To address this question, we trained automatic classifiers with auto-correlograms to identify the distinction between HD and non-HD neurons. Specifically, we used gradient boosted trees (XGB,39), a robust and fast non-linear classifier, on a binary output (HD or non-HD) and we trained classifiers on auto-correlograms of various duration. Classification quality (or “score”) was evaluated as a percentage above chance level (0%: chance level; 100%: perfect clustering). HD cells were labeled 50% above chance with only the first 6 ms of spike train auto-correlogram in each of the three brain states (Fig. 5f). This classification on such a limited amount of information suggests that duration of the refractory period and level of burstiness are both unique to HD cells in the AT.Together, these results provide evidence that HD neurons, in addition to their homogeneous response to hippocampal SWRs and unlike non-HD neurons, share common and specific dynamical properties. This further suggests that AD neurons are a distinct class of neurons in the anterior thalamus while non-HD neurons have broadly distributed properties, irrespective of their nucleus of origin.Slow firing dynamics distinguish HD and non-HD neuronsAlthough neurons showed a large spectrum of fast timescale dynamics (i.e. faster than 100 ms), does it mean that they were similarly activated during wakefulness on slow (“behavioral”) timescales (i.e. on the order of seconds)40? These timescales correspond, for example, to the typical firing duration of an HD neuron as an animal rotates its head through the neuron’s HD receptive field. For AT HD cells, these timescales are similar during wakefulness and REM sleep10. AT neurons showed a wide range of slow dynamics, not only during wakefulness but also during REM sleep, as observed in the various decay times of their auto-correlograms (Fig. 6a). The characteristic timescale, τ, of auto-correlogram decay was well captured by an exponential fit40. These behavioral timescales were intrinsic markers of neurons. The average decay times during wake and REM sleep are significantly correlated across neurons (Fig. 6b; r = 0.434, p < 0.001, Pearson’s correlation). HD cells, again, showed marked differences compared with other AT neurons: their intrinsic slow timescales were the slowest (Fig. 6b, inset; t = −0.23, p = 10−18, t-test for Wake; t = −0.37, p = 10−41, t-test for REM). In conclusion, neurons of the AT exhibited intrinsic behavioral timescale dynamics, thus possibly reflecting involvement in cognitive functions requiring different integration times40,41, which is another property that is a sufficient feature to distinguish HD from non-HD neurons.Fig. 6Intrinsic slow dynamics during wakefulness and REM sleep in the AT and its relation to fast dynamics during NREM sleep.a Auto-correlograms of three neurons (HD and non-HD neurons) over long timescale (2.5 s) during wake. Dashed gray lines display exponential fit. b Decay time τ between wake and REM sleep. Red dots indicate HD neurons. Inset shows average (±s.e.m.) decay time for HD and non-HD neurons during wakefulness and REM sleep. c Decay time τ during REM sleep as a function of burst index during NREM sleep (r = −0.374, p < 0.001, Pearson’s correlation). HD cells and non-HD cells are indicated by red and dark dots, respectively.Spike train dynamics, at fast and slow timescales, must result from complex interactions between intrinsic properties of neurons and network states. It is noteworthy that during wakefulness and REM sleep, thalamic neurons show very rare bursts25. Interestingly, burst index (calculated during NREM sleep) was negatively correlated with slow dynamics during both wakefulness and REM sleep (Fig. 6c during REM sleep; during wakefulness: r = −0.24, p < 0.001, data not shown). This relationship is potentially due to intrinsic properties which manifest at timescales that are orders of magnitude apart, and suggest that spike train statistics at millisecond timescales recapitulate the firing properties of the neurons in processes taking place at behavioral timescales.DiscussionWe described how HD cells of the AT (almost exclusively from the AD nucleus) form a homogeneous and specific cell population that plays a key role in limbic processing: while HD cells obviously inform the brain’s navigation system of the animal’s current HD during wakefulness, they also point towards particular directions at times of hippocampal SWRs during NREM sleep. Thus, they possibly constrain the dynamics of the hippocampus to allow for the generation of coherent replayed trajectories along certain directions. On the contrary, other AT neurons were heterogeneous, even when recorded from the same anatomical location. However, we revealed a link between the dynamics of spike trains and responses to NREM sleep SWRs, thus revealing how intrinsic neuronal properties and circuit integration are certainly related in the AT. One limitation of our study is that we only analyzed NREM sleep SWRs. The behavioral task was not designed to maximize the number of awake SWRs, and further studies are needed to determine the relationship between HD and non-HD neurons of the AT and awake SWRs.Here, we used a non-linear dimensionality reduction technique, ISOMAP42, to decode population states independent of their behavioral correlates26. We showed how the HD cell population of the AD nucleus, which normally drifts at high angular speed during NREM sleep10,26, stabilized at fixed points and at high gain immediately before SWRs. It remains to be shown whether HD cells firing in the AD can in fact orchestrate hippocampal activity. However, this transient gain increase in the HD cell population preceding SWRs was independent of the relative timing within the UP states. This suggests that the AD nucleus can rapidly influence hippocampal dynamics, and not only at the timescale of the slow oscillation21,23,24,34. While the impact of AD firing on the timing of SWRs is indirect and through polysynaptic routes, the projection fields of AD neurons, including targets in the post-subiculum, retrosplenial cortex and medial entorhinal cortex, suggest a broad contribution of AD neurons to the activity of the limbic system, as evidenced by their central role in synchronizing the slow oscillation during NREM sleep12.Not only do AD neurons exert a widespread influence on the limbic system during NREM sleep, they also convey a coherent message. Indeed, the population activity of HD neurons of the AD nucleus is still highly organized during NREM sleep10,26 in such a way that a virtual HD can be reliably decoded at any time with the exception of DOWN states that correspond to a singularity in HD cell population dynamics. It is thus possible that spatially tuned neurons of the limbic system are also updated by the HD signal during NREM sleep. While the HD cell population shows fast “sweeps” during NREM sleep10,26, the stabilization of the HD signal during SWRs suggests that the navigation system is transiently constrained in a particular direction. While our study did not include neuronal recordings in downstream structures, one prediction is that replay events of the hippocampus15,16 correspond to linear bouts of possible trajectories. Following exploration of a two-dimensional environment, it was recently suggested that replayed trajectories follow random (and non necessarily linear) rather than previously experienced paths16. Whether the directions of the replayed trajectories, from start to end points, or the initial direction of the replayed trajectories correspond to the direction encoded by the HD network remains an open question. Another possibility is that the HD cells influence replay in the medial entorhinal cortex. Indeed, grid cells maintain their coordination during NREM sleep43,44 and replay previously formed trajectories45,46, possibly independent of the hippocampus46. Finally, the coherent HD signal provided by the AD nucleus may constitute a subcortical process of coordinated neuronal sequences during sleep that do not correspond to any experienced spatial trajectories16,47. Paired recordings of AT HD neurons and place or grid cells will be necessary to test these predictions.One outstanding question is whether or not the subcortical HD network itself is under the influence of an external signal related to SWR generation and content. Indeed, one view posits that the HD network, including the AD nucleus and upstream structures, randomly fluctuates during sleep by integrating noisy inputs26. A transient fluctuation in excitability would favor the occurrence of a SWR through polysynaptic pathways. Conversely, it is possible that such a state is controlled by cortical feedback: AD neurons, as well as their presynaptic neurons of the lateral mammillary nuclei, receive a feedback from the post-subiculum. However, post-subicular cells remain largely under the influence of the AD nucleus during NREM sleep27. At any rate, AD alone is certainly not the only structure influencing SWRs. For example, auditory cortex neurons fire prior to SWR and predict the content of hippocampal replays48.While the exact input and output connectivity pattern of each AT neuron is intractable in vivo, their coordination with hippocampal activity is an opportunity to determine their functional integration within the circuit. Unlike HD neurons, the co-modulation with SWRs was broadly distributed for non-HD neurons, largely independent of their anatomical location. One prediction is that with the exception of the AD nucleus, other cytoarchitecturally defined AT nuclei are characterized by a large diversity of connectivity patterns from neurons to neurons, as previous anatomical studies have suggested1,29.The flow of information in the limbic system, and in particular to and from the hippocampus, may vary across brain states, especially between wakefulness and NREM sleep49. AT neurons showed similar responses to SWRs (during NREM sleep) and theta (during wakefulness and REM), thus suggesting that they play a role in routing information independently of brain states. Whether this results from specific coupling with the hippocampus or from modulation by other limbic structures (e.g. the medial septum) remains to be answered.Surprisingly, spike train dynamics are another potential marker of the functional coupling of AT neurons with the hippocampus. In fact, we observed a strong relationship between spike train auto-correlograms across brain states and responses to SWRs. While our study could not determine the intracellular properties of AT neurons, spike train dynamics directly reflect – at least in part – these intrinsic properties50,51. Again, HD cells formed a highly homogeneous population of neurons that could be identified as separate from the non-HD cells. In contrast, non-HD cells showed a continuous spectrum of dynamics with graded burstiness and refractory periods, among other features. Furthermore, AT neurons show a wide range of slow (~1 s) intrinsic timescale dynamics. The intrinsic nature of these timescales is suggested by the similar time constant of their spike train auto-correlogram during wakefulness and REM sleep, two activated states of the brain. HD cells show markedly longer behavioral timescales than other AT neurons. Together, these results suggest that the cytoarchitectural definition of AT nuclei is an insufficient level of description with the exception of the AD nucleus in which neurons show homogeneous behavioral correlates, for functional coupling with the hippocampus and spike train dynamics. In support of this view, gene expression profile in the AD is unique in the thalamus, while other AT neurons are continuously distributed rather than clustered in cytoarchitecturally defined nuclei52.A growing body of evidence points towards an intimate link between gene expression, spontaneous activity and circuit development in thalamocortical pathways29,52–55. It is thus possible that spike train dynamics and coupling to network activity are two facets of common processes that are directly related to the acquired (but not necessarily pre-configured) role that each neuron plays in the circuit and associated with a specific molecular makeup29,52–57. Whether the variability in spike train dynamics reflects specific processes for signal transmission to downstream readers remains an open question.The AT is a key communication relay between structures of the limbic system7. Interactions between the thalamus, the hippocampus and more generally the limbic system during learning and around SWRs (nested in slow and spindle oscillations, two NREM-specific thalamocortical rhythms) are believed to be crucial for supporting hippocampus-dependent memory consolidation processes at play during NREM sleep12,13,17,21,22,24,28,58,59. Interestingly, while the HD cell population is stabilized at times of SWR occurence, it drifts at maximum speed during thalamocortical spindles26. Future work is necessary to characterize the processes controlling HD and non-HD cells of the AT, their contribution in determining hippocampal replay content and their role in routing hippocampal signals across the structures of the limbic system. The analytical framework of our study provides a foundation for investigating the relationship between intrinsic neuronal properties at various timescales and functional integration in brain networks in vivo. It is possible that such a relationship is an organizational principle of all thalamic neurons.MethodsSurgery and experimental designAll experiments were approved by the Institutional Animal Care and Use Committee of New York University Medical Center. Details of the surgeries were described previously10. Briefly, four male mice weighing ~30 g (3–6 months) were implanted under isoflurane anesthesia with silicon probes (Neuronexus) above the anterior thalamus (AP: −0.6 mm; ML: −0.5 to −1.9 mm; DV: 2.2 mm, with a 10–15° angle, the shanks pointing toward midline). Hippocampal wire bundles were implanted above CA1 (AP: −2.2 mm; −1 to −1.6 mm ML; 1 mm DV). The probes consisted of 8 shanks separated by 200-μm and each shank had eight recording sites (160 μm2 each site, 1–3-MΩ impedance) that were staggered to provide a two-dimensional arrangement (20-μm vertical separation). In all experiments, ground and reference screws or 100-μm diameter tungsten wires were implanted in the bone above the cerebellum.Electrophysiological data acquisitionThe animals were recorded over several days for successive epochs of pre-sleep (1–2 h), food foraging in a circular arena (sweetened cereals or regular food pellets) for 30 min, and post-sleep (1–2 h). Overall, sessions lasted on average 5 h (±0.94 s.d.). Electrophysiological signals were acquired continuously at 20 kHz on a 256-channel Amplipex system (Szeged; 16-bit resolution, analog multiplexing). The broadband signal was down sampled to 1.25 kHz and used as LFP.To track the position of the animals in the open maze and in their home cage during rest epochs, two small light-emitting diodes (LEDs; 5-cm separation), mounted above the headstage, were recorded by a digital video camera at 30 frames per second. The LED locations were detected online and resampled at 39 Hz by the acquisition system. Spike sorting was performed semi-automatically, using KlustaKwik (http://klustakwik.sourceforge.net/). This was followed by manual adjustment of the waveform clusters using the Klusters software.Sleep scoringAs previously described10, stages of sleep were identified semi-automatically based on the CA1 LFP spectrogram and on animal movements that were continuously tracked with the LEDs, as during behavior. Overall, sleep was defined as a long period of immobility. Within each sleep episode, NREM sleep was defined as periods with high delta (1–4 Hz) and spindle (10–15 Hz) activity. REM sleep was defined as periods with strong power in the theta (5–12 Hz) range and low delta.UP and DOWN state detectionUP and DOWN states were detected by computing the total firing rate of all simultaneously recorded neurons in bins of 10 ms, smoothed with a Gaussian kernel of 20 ms s.d. Epochs during which total firing rate was lower than 20% of the maximum firing rate were considered as DOWN epochs. Epochs shorter than 30 ms and longer than 500 ms were discarded. UP states were defined as the epochs between each DOWN states.SWR detectionTo detect SWRs, CA1 LFP was first bandpassed between 80 and 300 Hz with a Gaussian filter. The squared signal was then smoothed using a digital filter with a window length of 11 and z-scored using mean and standard deviation. The normalized squared signal was then thresholded between 3 and 7 standard deviations yielding a first set of candidate ripples. The set was then reduced by keeping only candidates with a duration between 25 ms and 350 ms (in some sessions with lower amplitude ripples, the thresholds were lowered to 2 and 5 s.d., respectively, and rippled detection was validated by visual inspection). Ripples closer than 30ms were merged and considered a unitary event.Head-direction classificationFor each session, the direction of the head of the animal in the horizontal plane was calculated by the relative orientation of a blue and red LED located on top of the head. Head-direction neurons were detected by computing tuning curves i.e. the ratio between the histogram of angular direction associated to each spike divided by the total time spent by the animal in each angular bins. Similar to Peyrache et al.10, a Rayleigh test was performed to test for the null hypothesis of uniformly distributed firing in all angular directions, and neurons were classified as HD cells if peak firing rates of the tuning curves was >1, a probability of non-uniform distribution <0.001 and a concentration parameter (i.e. inverse of the variance of the tuning curves) larger than 1.ISOMAP projection and decodingISOMAP projections (Fig. 1 and Supplementary Fig. 1) were performed for sessions containing at least 10 HD neurons (eight sessions in total from three different animals). First, spike trains were binned during wakefulness and around time of SWRs. As such dimensionality reduction techniques are computationally heavy, we included in the analysis only the first 15 min of exploration and a period of 1 s around SWRS (±500 ms). The duration of the bins was 400 ms and 30 ms (with a 50% overlap) during wakefulness and around SWR occurrence times, respectively (different bins were used to capture the different timescales of the HD signal during wakefulness and NREM sleep). To visually inspect the topology of HD cell population vectors (Fig. 1c and Supplementary Fig. 1a), we used a bin size of 200 ms for wakefulness and 100 ms (with a overlap of 75%) around SWRs. To quantify the change in radii and angular velocity across sessions, ISOMAP projections were compared to a baseline taken around randomly selected events during NREM sleep. Similar to Chaudhuri et al.26, we computed the square root of the rates to normalize for the variance in firing rates. Binned firing rates were smoothed with a Gaussian kernel of three bin standard deviation (independent of absolute bin duration).In each condition (wakefulness, SWR, and controls), binned firing rates were stacked together, yielding a rate matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R\in {{\mathbb{R}}}^{T\times N}$$\end{document}R∈RT×N, where N is the number of neurons and T is the total number of time points. This rate matrix was then projected to a two dimension plane \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I\in {{\mathbb{R}}}^{T\times 2}$$\end{document}I∈RT×2 using ISOMAP42. The number of neighbors was set to 200. To further reduce the computational requirement for long spike trains (i.e. typically more than 30000 time bins), we repeated the ISOMAP projection using a random subset of 150 SWRs and 150 random NREM sleep time bins for baseline quantification with the same 15 min of awake data, until all SWRs were analyzed. This ensures that at each iteration, ISOMAP gives a similar topology of embeddings and that neuronal “trajectories” around SWRs can be compared to each other for a given session.Angular direction and radius (i.e. ring size) was decoded during SWRs by computing the element-wise arc tangent and Euclidean norm of each time point. Angular velocity was then evaluated by computing the angular difference between two consecutive time points. The same procedure was performed for randomly selected events during NREM sleep that projected within the same ISOMAP embedding. Radius and angular velocity (as shown in Fig. 1d, e) are thus the ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({\bar{X}}_{{\mathrm{SWRs}}}-{\bar{X}}_{{\mathrm{random}}})/{\bar{X}}_{{\mathrm{random}}}$$\end{document}(X¯SWRs−X¯random)/X¯random with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{X}$$\end{document}X¯ denoting average values).Bayesian decodingTo validate the decoding in the embedding space (Fig. 1), we used Bayesian decoding to predict angular velocity during SWRs (Fig. 2). Let n = (n1, n2, . . . , nN) be the number of spikes fired by the HD neurons within a given time window (30 ms) and Φ be the animal’s head direction or, during sleep, the angle of the internal HD signal. The goal of this decoder is to compute the posterior probability P(Φ∣n), which can be achieved using Bayes’ rule of conditional probabilities:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(\Phi | {\bf{n}})=\frac{P({\bf{n}}| \Phi )P(\Phi )}{P({\bf{n}})}$$\end{document}P(Φ∣n)=P(n∣Φ)P(Φ)P(n)Assuming that (i) neuronal firing is independent from each other and (ii) spike counts follow Poisson distributions, the probability P(n∣Φ) is equal to33:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P({\bf{n}}| \Phi )=\mathop{\prod }\limits_{i = 1}^{N}P({n}_{i}| \Phi )=\mathop{\prod }\limits_{i = 1}^{N}\frac{{(\tau {f}_{i}(\Phi ))}^{{n}_{i}}}{{n}_{i}!}{\mathrm{ex}}{{\mathrm{p}}}^{-\tau {f}_{i}(\Phi )}$$\end{document}P(n∣Φ)=∏i=1NP(ni∣Φ)=∏i=1N(τfi(Φ))nini!exp−τfi(Φ)where τ is the bin size and fi(Φ) the average firing rate of cell i for the direction Φ during wakefulness (i.e. the HD tuning curve at angle Φ). The decoded direction is the value Φ associated with the highest posterior probability P(Φ∣n).Similar to ISOMAP decoding, angular velocity was evaluated as the ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({\bar{v}}_{{\mathrm{SWRs}}}-{\bar{v}}_{{\mathrm{random}}})/{\bar{v}}_{{\mathrm{random}}}$$\end{document}(v¯SWRs−v¯random)/v¯random with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bar{v}}_{SWRs}$$\end{document}v¯SWRs being the average angular velocity during SWRs and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bar{v}}_{{\mathrm{random}}}$$\end{document}v¯random, the average angular velocity for randomly selected events during non-REM sleep. We used bin size of 30 ms with 50% overlap and angular bins of 6 degrees.Map alignmentTo align the position of the electrodes with their putative anatomical location, we used the following procedure. First, after each session, the probe was lowered by 70–140 μm. Relative position was estimated from the first day of recording when the probe entered the thalamus (appearance of spiking activity on 1–2 shanks). This yielded a relative depth for each session while the horizontal distance between shanks was kept constant (200 μm). This two-dimensional grid of recording sites was then rotated by 15° clockwise corresponding to the angle of penetration of the silicon probe. Then, the relative position of recording sites was aligned to the anatomical map (from bregma −0.82 mm; Fig. 38 in Paxinos et al.60) by matching the relative configuration of the electrodes with the putative location of the AD nucleus (shanks on which large number of head-direction neurons were recorded)10. Note that the anatomical map from the atlas was enlarged by 10% (it is commonly accepted that slices used for anatomy are shrunk by approximately this amount). As shown in Figs. 2 and S2, the density of HD neurons matched well with the estimated anatomical position of the anterodorsal nucleus.SWRs cross-correlogramsThe SWR modulation was computed for each neuron by estimating cross-correlograms (average firing rates in 5 ms bins, ±500 ms from SWR peak time). Because neuronal discharge and SWRs are both co-modulated by the slow oscillation21–24, cross-correlograms were normalized relative to their expected values under the null hypothesis of no short timescale coupling between neuronal discharge and SWR occurrence. To this end, SWR cross-correlograms were convolved with Gaussian windows of 150 ms s.d. (a process similar to low-pass filtering the firing rates). The value of 150 ms was chosen as it corresponds to the upper bound duration of SWRs. From this distribution of low-pass filtered (“expected”) rates around the time of SWRs, we inferred an “expected”, standard deviation under the assumption of a Poisson process (i.e. the square root of the expected rate at a given time bin). For each time bin, we subtracted the expected rate from the observed rate, and then divided the difference by the expected standard deviation. The observed cross-correlograms were thus expressed in z-values from the expected distribution under the null hypothesis of no short timescale coupling. This method enables us to extract the specific and fast modulation of a neuron by SWRs, independent of the co-modulation of the SWRs with the thalamocortical slow oscillations22–25. In mathematical terms, we thus obtained a set of time series z(t) = [z1(t), . . . , zN(t)] where N is the number of neurons. Note that the z values are not a typical z-score, but correspond to deviation in z from the null hypothesis. Note that the mean and s.d. of the resulting z-transforms are not normalized. To describe the amplitude of SWR modulation, we introduced SWR energy, defined as the total power of the z-scored cross-correlograms: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {z}_{i}| =\sqrt{\mathop{\sum }\nolimits_{t = 1}^{T}{z}_{i}{(t)}^{2}}$$\end{document}∣zi∣=∑t=1Tzi(t)2.Theta modulationTheta modulation for each neuron was computed separately for epochs of wake and REM sleep. Using continuous wavelets transform, a phase was assigned to each theta cycle and histograms of spike counts between 0 and 2π were then computed for each neuron as shown in Fig. 3b. A Rayleigh test was then performed with the null hypothesis of uniformly distributed firing rates during a theta cycle.jPCATo describe SWR modulation in term of phases, we used the jPCA method37. First, PCA was used to reduce the dimensionality of the ensembles of cross-correlograms (K = 6 principal components were considered, as in Churchland et al.37). Overall, the goal of jPCA is to describe temporal response profiles, assuming that the data are governed by a linear dynamical system of the form: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{{\bf{x}}}=M{\bf{x}}$$\end{document}x˙=Mx with M ∈ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathbb{R}}}$$\end{document}RK,K (or M ∈ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathbb{R}}}$$\end{document}RN,N if no dimension reduction is applied first). The matrix M can be split into its symmetric and anti-symmetric matrix M = Msym + Manti, related to expansion/contraction and rotational dynamics, respectively. The latter matrix has pure imaginary eigenvalues, hence its association with rotational dynamics. To capture phase response in the rotational space, it is thus sufficient to find the best fit for the matrix Manti ∈ Γk×k. This can be done by minimizing the error made on predicting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{X}$$\end{document}X˙ from MX, which can be expressed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${M}_{{\mathrm{anti}}}^{* }={\mathrm{argmi}}{{\mathrm{n}}}_{M\in \Gamma }| | \dot{X}-MX| {| }_{F}$$\end{document}Manti*=argminM∈Γ∣∣X˙−MX∣∣F, where F denotes the Froebenius norm (see Churchland et al.37 for additional details on the methods). Finally, jPCA entails the decomposition of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${M}_{{\mathrm{anti}}}^{* }$$\end{document}Manti* into its eigenvectors made (by definition) of complex conjugate pairs. A suitable pair of projection vectors U = (u1, u2) ∈ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathbb{R}}}$$\end{document}RT,2 can be obtained by combining complex conjugate pairs (v1, v2) such that u1 = v1 + v2 and u2 = i(v1 − v2).The last step consists in projecting SWR cross-correlograms onto the first two jPCs (u1 and u2): y = ZTu ∈ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathbb{R}}}$$\end{document}RN,2. The SWR phase was defined as the angle from origin in this 2D space: ϕSWR = atan2(y2, y1).Auto-correlogramsFor each neuron, three auto-correlograms were computed separately for the epoch of wake, REM and NREM sleep respectively. The short auto-correlograms shown in Figs. 4 and 5 were computed using a bin size of 0.5 ms while the long auto-correlograms in Fig. 6 were computed using a bin size of 5 ms.Embedding of the auto-correlograms in a 2D map (Fig. 5b) was performed with the t-SNE algorithm38 by concatenating only the part corresponding to positive time lag (i.e. the right part of the auto-correlograms). Given \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2\,{\mathrm{ms}}\to 40\,{\mathrm{ms}}}^{{\mathrm{epoch}}}\in {{\mathbb{R}}}^{N}$$\end{document}A2ms→40msepoch∈RN the auto-correlogram vector of one epoch for one neuron, each point in the t-SNE projection is thus a mapping \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f:{X}_{i}\in {{\mathbb{R}}}^{3N}\to {Y}_{i}\in {{\mathbb{R}}}^{2}$$\end{document}f:Xi∈R3N→Yi∈R2 with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{i}=[{A}_{2\,{\mathrm{ms}}\to 40\,{\mathrm{ms}}}^{{\mathrm{wake}}}{A}_{2\,{\mathrm{ms}}\to 40\,{\mathrm{ms}}}^{{\mathrm{REM}}}{A}_{2\,{\mathrm{ms}}\to 40\,{\mathrm{ms}}}^{{\mathrm{NREM}}}]$$\end{document}Xi=[A2ms→40mswakeA2ms→40msREMA2ms→40msNREM] for each neuron i.In Fig. 5f, neurons were classified (HD versus non-HD) with gradient boosted trees using the XGBoost package39. Inputs given to the classifier is the same as the t-SNE algorithm (i.e. stacked auto-correlograms of wake, REM and NREM sleep episodes). Classifier was trained with 1000 iterations, using default parameters and a 10-fold cross-validation procedure. Chance levels were determined from a thousand classifications obtained with random shuffling of the neuron labels. For N neurons and C classes, classification score was defined as:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s=\frac{\mathop{\sum }\nolimits_{i\, = \,1}^{N}1({c}_{i}\,=\, {\hat{c}}_{i})-\mathop{\sum }\nolimits_{i\, = \,1}^{N}1({c}_{i} \, =\, {\hat{c}}_{i}^{s})}{N-\mathop{\sum }\nolimits_{i \,=\, 1}^{N}1({c}_{i}\, =\,{\hat{c}}_{i}^{s})}$$\end{document}s=∑i=1N1(ci=c^i)−∑i=1N1(ci=c^is)N−∑i=1N1(ci=c^is)with 1(x) the indicator function, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{c}}_{i}$$\end{document}c^i the class predicted for the ith neuron and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{c}}_{i}^{s}$$\end{document}c^is the labels predicted with a classifier trained on shuffled data. A score of 0 indicates chance level (compared to shuffles) and a score of 1 indicates perfect classification.Burst indexFor each epoch, bursts were defined as groups of spikes with interspike intervals (ISI) between 2 ms and 9 ms. Burst index61 was then computed as the ratio between the observed count N2 ms<ISI<9 ms and the count expected from a homogeneous process with the same average firing rate: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bar{N}}_{{\tau }_{1}{<}ISI{<}\,{\tau }_{2}}=T(exp(-{\tau }_{1}\times r)-exp(-{\tau }_{2}\times r))$$\end{document}N¯τ1<ISI<τ2=T(exp(−τ1×r)−exp(−τ2×r)) with T the duration of the epoch, r the mean firing rate during this epoch, τ1 = 2 ms and τ1 = 9 ms.Auto-correlograms and SWR cross-correlograms correlationTo compute the correlation between SWR cross-correlograms and auto-correlograms (Figs. 4e–g), auto-correlograms were first stacked across brain states (as done for the t-SNE projection, see above). We used PCA to reduce the dimensionality of each matrix (of either SWRs cross-correlograms or stacked auto-correlograms) and we kept the first 10 components accounting for respectively 91% and 99% of the variance. We computed the 20-by-20 correlation matrix C for the 10 SWR PCs and 10 auto-correlogram PCs. The matrix is diagonal for the upper-left and bottom right 10-by-10 blocks as PCs from a given dataset are perpendicular to each other. We thus show only the off-diagonal matrix in Fig. 4f. We then defined the total correlation as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{{\mathrm{total}}}^{2}=1-| C|$$\end{document}ρtotal2=1−∣C∣ with ∣C∣ being the determinant of the correlation matrix (if only one PC was considered for auto-correlograms and SWR cross-correlograms, then total correlation would be the linear correlation between the two PCs). From a geometrical perspective, this measure captures the fraction of state-space volume occupied by both measures. The null correlation was determined by shuffling the identity of the neurons in both datasets (1000 times).We controlled for correlation between neurons belonging to the same nucleus by shuffling neurons within groups recorded on the same shank and session (1000 times). The correlation between SWR cross-correlograms and auto-correlograms was further tested for possible correlation by common underlying factors (firing rates or burstiness). The correlation rij between the ith auto-correlogram PC and the jth SWR PC was replaced by the partial correlation (correlation knowing factor f): \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${r}_{ij| f}=\frac{{r}_{ij}\,-\,{r}_{if}{r}_{jf}}{\sqrt{1\,-\,{r}_{if}^{2}}\sqrt{1\,-\,{r}_{jf}^{2}}}$$\end{document}rij∣f=rij−rifrjf1−rif21−rjf2 where rkf (k = i or j) is the correlation between PC weights and the external factor f (firing rates or burstiness).Quantification and statistical analysisAnalyses were done using customized code written in MATLAB (MathWorks) and Python (3.5) with the following libraries: numpy, scipy and scikit-learn.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary Information
Supplementary Information
Peer Review File
Reporting Summary | nature communications | [
"Article"
] | [
"Non-REM sleep",
"Replay",
"Hippocampus",
"Spatial memory",
"Neural circuits"
] | IntroductionThalamocortical loops form circuits complex cognitive functions1–4 dorsal thalamus organized in nuclei1 anterior thalamic nuclei (AT) at central stage Papez circuit5 key in memory6,7 spatial arousal states11 dynamics AT neurons integration with limbic system elusive role AT in navigation indicated by head-direction (HD) neurons discharge when direction in anterodorsal (AD) nucleus8 HD cells necessary for higher level spatial representations navigation system relationship between neuronal activity memory in AT elusive relationship established in hippocampus Non-Rapid Eye Movement (NREM) sleep instrumental for memory consolidation13 hippocampal activity patterns reactivated14 spatial trajectories replay events associated with SWRs in local field potential necessary for unclear generated hippocampus or influenced by inputs NREM sleep timing of SWRs orchestrated by thalamocortical dynamics21–24 thalamocortical slow oscillation from fluctuation membrane potentials between UP DOWN statesAD key hub for slow oscillation12 HD cells neuronal activity limbic systen during NREM sleep contribution AD HD cells to processing spatial signals maintain coherent firing activity single direction decoded during sleep10 AD HD cells lead main cortical targets independently of brain states27 during sleep AD neurons coherent HD signal to navigation system hippocampus independent of current heading role other AT nuclei coordinating activity limbic system neuronal dynamics unclear challenge investigating thalamic function nucleus input/output patterns differ across neurons firing thalamic neurons homogeneous NREM variability in spike train dynamics reported31 recording thalamic neuronal activity hippocampal LFP in freely moving sleeping mice addressed AT neurons thalamic HD cells coupled to SWRs examined variability coupling with intrinsic properties AT neurons spike train dynamics across brain states HD cells specifically coupled to SWRs increasing gain firing coherently for direction during SWRs HD cells AD nucleus form homogeneous population neurons intrinsic dynamics hippocampal couplingNon-HD cells contrast modulation by hippocampal SWRs dynamics diverse differ proximal neurons nucleus HD signal coupled to SWRsWe recordings anterior thalamus (n = 2016 neurons in moving mice open environment home cages sleep Neurons classified as HD cells modulation firing rate direction head horizontal plane (n = 161 HD cells tuning curves of 24 recorded HD cells in Fig. 1a HD decoding before SWRs HD decoding with ISOMAP around SWRs session Tuning curves of HD neurons recorded AD nucleus Simultaneous recording of CA1 pyramidal layer LFP neuronal activity in AD around SWRs during NREM sleep Spikes HD neurons sorted preferred direction during wakefulness Red asterisks indicate SWRs Two-dimensional embeddings HD neuronal population activity during wakefulness around SWRs Curves Distribution distances from embedding center during wakefulness NREM sleep average angular directions around SWRs Average normalized radii from embedding centers around SWRs 7 sessions 3 session-wide average angular velocityangular displacement between activity bins). Average density of spikes for recorded HD non-HD (gray neurons rate occurrence SWRs during DOWN and UP states cells convey signal to parahippocampal system during sleep10 HD system provides hippocampus specific direction during SWRs cell ensembles replay patterns spatial recorded LFP in pyramidal layer dorsal CA1 SWRs subset HD neurons fired preferred HD (Fig. 1b). uncovered topology of population firing using ISOMAP revealed HD cell population vectors computed without behavioral correlates HD tuning During wakefulness topology ring structure angular position mapping animal’s HD (Fig. 1c). population activity restricted to ring during wakefulness “forbidden increased during NREM sleep scattering of population vector embeddings center ring violation topology resulted from modulation network activity by slow oscillation25 population activity around SWRs corresponds to trajectories within projection space three cases Fig.activity started near center putative DOWN states settled on fixed points outer part “allowed” states wakefulness at SWRs population states around SWRs sessions population vectors at each time bin expressed in polar coordinates center ring reference point analyzed data sets 10 HD neurons clear ring topology during wakefulness (n = 7 sessions, 3 animals Fig 1) topology assessed distance each point to center embeddings During wakefulness distances distributed within fixed range from center no point close to center embeddings form annulus-like shape around center for HD cells26 SWRs HD population pointed to near-random directions slight bias towards one direction from non-homogeneous sampling of preferred HD limited number recorded all sessions radius trajectories peaked around SWRs baseline Average radius maximal before SWR peaks (−17 ms ± 6 ms; t = −2.82, p = 0.03 t angular position stable or drifting? computed relative angular velocity signal velocity minimal around SWRsreached minimum before SWR peaks (−18 ms ± 15 ms t = −2.88, p = 0.02-test excluding SWRs <100 ms after UP onset). Similar results obtained using Bayesian decoding angular position HD curves stabilization network into states results from increase HD population firing independent SWRs showed fast sweeps at high firing rates increase in gain stabilization network characteristic of epochs preceding SWRs coupling between HD neuronal population SWRs could result from coordination SWRs with slow oscillations22 at transition between DOWN UP states detected putative DOWN and UP states from activity duration epoch normalized (Fig. change in probability SWR occurrence around DOWN/UP transitions constant rate SWR occurrence during UP states SWRs less prevalent during early onset UP states modulation SWRs by slow oscillation SWRs more prominent at beginning end of UP states analysis reflects distribution UP state duration not preferred phase co-occurrence of SWRs stable HD signal biased by position within UP states SWRs clustered according to time since UP state onsetdifference in time-lags radius peaks minima angular velocity relative to SWRs exception for SWRs first 200 ms UP state governed by dynamics transition DOWN states Fig. 1b SWRs preceded by high-activity stable HD population states coordination thalamocortical slow oscillation SWRs.Homogeneous firing HD cells around coupling HD network to SWRs question specific HD AT neurons Non-HD AT neurons recorded increasing dorsoventral depth (Fig. spanning most AT nuclei multiple channels silicon probes tracks each shank across AT (Fig. position probes calibrated dorso-ventral axis spiking activity shank medio-lateral axis matching highest density HD neurons with AD nucleus8 (Fig. 2b 3). 2Homogeneous coupling HD not non-HD neurons to SWRs Histology mouse 17, eight-shank silicon probe lesion sites end tracks Contour AD nucleus shown in red Schematic AT (coronal plane) recording sites animal Red dots indicate HD neurons detected Top spikes SWRs Middle mean firing ratesz-scored modulation one HD 2 non-HD neurons modulated SWRs average SWRs modulation HD non-HD neurons sessions SWR cross-correlograms HD AT neurons Neurons sorted correlation SWR zero-time lag SWR cross-correlograms HD non-HD neurons same session different shanks Pearsonʼs correlation coefficients SWR-correlograms same shanks Red dots indicate HD neurons Distribution Pearson’s correlation coefficients SWR-correlograms HD non-HD neurons same shanks all sessions AD anterodorsal AM anteromedial AVd anteroventral dorsomedial AVv anteroventral ventrolateral IAD interanterodorsal LD laterodorsal MD mediodorsal PV paraventricular medullaris Ventral anterior no behavioral correlates topology population non-HD neurons analyzed neuronal activation SWRs cell-by-cell challenge evaluating neuronal responses SWRs cross-correlation analysis SWRs co-modulated slow oscillation21–25responses neurons to SWRs cross-correlograms compared to baseline fast co-modulation neuronal spikes SWRs ±150 ms typical duration sharp wave Cross-correlograms expressed in standard deviations (z) from null distributions.Three neurons one HD different modulation by SWRs (Fig. 2c). average SWR modulation HD neurons response synchronized increase firing before SWRs (Fig. confirming coupling HD signal SWRs 1) non-HD cells various responses SWRs near-uniform average modulation non-HD cells before SWRs recorded different nuclei possible some nuclei show same coupling to SWRs HD neurons rule out computed pairwise correlations neurons same shanks same nuclei Fig. 2e session HD non-HD neurons SWR modulation non-HD neurons heterogeneously distributed HD non-homogeneous response non-HD cell population to SWRs HD confirmed distribution pairwise correlations shanks across sessions (Fig. 2f; n = 7234 pairs, 58 sessions, t = −18.p = 10−71 t HD neurons AD nucleus fire homogeneously SWRs activation AT neurons SWRs variable local networks hippocampus brain-state origin high variability coordination AT neurons with SWRs depend input/output connections hard-wired network coordination AT hippocampal activity brain states wakefulness REM sleep hippocampus dominated by theta oscillations (6–9 Hz modulate neurons limbic including tested relationship between SWR modulation NREM sleep phase coupling theta oscillation wakefulness REM sleep (Fig. 3a).Fig. 3Relationship coupling AT neurons SWRs modulation theta oscillations recording LFP CA1 pyr. layer neuronal activity AT during REM sleep Red ticks spikes HD neurons spikes theta cycles spike density theta cycles neurons REM sleep Theta modulation versus SWR energy all neurons points indicate HD distribution “Theta mod” indicates neurons significantly modulated by theta (p < 0.01). Density map of theta-modulated neurons (min = 0.016) for one mouse absence theta modulation for AD.Z-scored SWR cross-correlograms theta-modulated AT neurons group theta-mod sorted modulation SWR peaks First two jPC vectors gray SWR cross-correlograms Projection SWR cross-correlograms jPCs Point colors 0-lag modulation Gray arrows SWR-jPCA phase four neurons points a Theta phase distributions corner SWR-jPCA phase function theta phase (r = 0.18 p < 0.001 circular four neurons green firing AT neurons REM theta analyzed preferred phase modulation amplitude Three AT neurons different theta modulation profiles (Fig. 38% AT neurons modulated theta REM sleep (n = 767/2016 p < 0.001) lower wakefulness (n = 333/2016 p < 0.001) theta modulation compared SWR responses quantified “SWR variance normalized cross-correlograms Fig. SWR energy theta modulation amplitude correlated AT neurons (Fig. 3c r = 0.81 p < 0.014 Pearson’s similarities modulation strength reflect strength inputs AT neuron parahippocampal temporal responseHD neurons not modulated by theta n = 21/161 neurons AD nucleus form homogeneous population share same input/output connectivity profile anatomical density of theta-modulated neurons segmentation found anywhere in AT except AD (Fig. 3d Supplementary Fig. 2 AT neurons relative to SWRs show wide range of temporal profiles (Fig. 3e for theta-modulated neurons). modulation by oscillations phase preference description for by SWRs lacking dynamics AT neurons around SWRs used jPCA rotational dynamics non-periodic projection jPC subspace describes temporal responses two-cycle determined jPC basis from z-scored cross-correlograms relative to SWRs projected on first two jPC components (Fig. 3f). each neuron attributed “phase” to angle from positive direction first jPC axis resulting phases with preferred phase theta oscillations (Fig. 3g for theta SWR AT neurons modulated by theta theta phases correlated with SWR phases (Fig. 3h r = 0.p = 2.3 × 10−7 circular AT neurons modulated by hippocampal dynamics in brain states invariant property at circuit level link between firing dynamics hippocampal relationship modulation AT neurons hippocampal dynamics connectivity profile vivo on neuronal characteristics extracellular recordings? Spike train dynamics captured by auto-correlation functions reflect complex interaction between input membrane properties HD neurons different auto-correlograms during wake REM NREM sleep HD neurons responses to SWRs brain state-specific auto-correlograms similar firing rates levels auto-correlation different Fig three non-HD neurons shank common properties different 4b pair neurons similar auto-correlograms SWR cross-correlograms third neuron showed different spike train dynamics opposite modulation SWRs 4Covariation between single-cell spike train auto-correlograms SWR cross-correlograms cross-correlograms auto-correlograms brain states for two HD neurons Circular plots indicate HD tuning curves Same for three non-HD neurons shankTop right average waveforms eight recording sites shank dorsal to ventral positions neuron gray same both panels cell bodies two neurons c close auto-correlograms responses SWRs different. Same as a–c three neurons different shanks sessions same animal anatomical location left-side map anterior. PCA projections first 20ms auto-correlograms SWR cross-correlograms three neurons b, c Correlation matrix PCA weights shuffled neuronal identities all neurons ρ2 total correlation. Distribution correlation ρ2 actual random shuffling cell identities between all cells shank recording days cells same shank day Distribution total correlation ρ2 actual shuffled data between SWRs auto-correlograms wake sleep NREM sleep Dashed line correlation between SWRs stacked auto-correlograms all epochs Same h cell identities shuffled within shanks examples suggest high variability spike train dynamics responses SWRs given nucleus AD sole exception). properties neuronal firing related? neurons same response SWRs across sessions locations similar auto-correlograms three neurons Fig. 4d.relationship spike train dynamics SWR responses quantified correlations principal component analysis (PCA) time first 20 ms auto-correlograms cross-correlograms SWRs (±500 first ten components considered 99% 91% variance auto-correlograms SWR cross-correlograms (Fig. 4e). correlation matrix weights more structure than PC weights shuffled (Fig. 4f diagonal blocks not PCs orthogonal total correlation ρ2 of PC weights higher than shuffles (Fig. 4g; ρ2 = 0.31, p < 0.001).Shuffling neurons led distribution total correlation smaller than actual total correlation (Fig. relationship spike train dynamics responses SWRs specific each neuron determined anatomical location within-shank shuffled correlations higher than cell identities shuffled (t = −82; p = 10−10) suggesting similarity auto-correlation SWR response within nucleusstrong correlation at cell level by differences firing rates (\documentclass[12pt{minimal}{amsmath{wasysym-69pt{rate=0.3}ρrate2=0.3 vs ρ2 = 0.31) nor in burstiness alone[12pt{minimal-69pt{burst}=0.3}ρburst2=0.3). SWR-associated firing could bias auto-correlograms during NREM sleep repeated analysis auto-correlograms of one brain state total correlation higher than shuffled data (p < 0.001) when wake REM NREM sleep used separately (Fig. 4h-i). relationship between auto-correlograms and SWR cross-correlograms holds when HD neurons excluded (p < 0.001 data not shown). results dynamics of AT neurons related to participation circuit-level activitySpike train dynamics non-HD AT neurons heterogeneousThe link between coordination with SWRs independent anatomical location question sub-classes neurons AT based distribution spike train auto-correlograms HD neuron (cell 1 Fig. 5a short refractory period across brain states low burstiness during NREM sleep peak first 8 ms agreement with previous studies AD cell 2 Fig. 5a high burstiness during NREM longer refractory period during wakefulness REM sleep third cell (cell 2) intermediate properties long refractory period lack of burstiness neurons segregated based on auto-correlograms.Fig. 5Variability clustering of spike auto-correlograms in AT Auto-correlograms HD (1) non-bursty (2) bursty (3) neuron during wake REM NREM sleep Clustering t-SNE embedding auto-correlograms HD neurons marked white dot three example neurons from c circled K-means clustering in two clusters Cluster #2 color-coded relative to burst index during NREM sleep15 superimposed auto-correlograms cells cluster #1 50 100 REM NREM sleep Color to dark red indicates average firing rates low to high similarity auto-correlograms REM independent firing mild variability NREM density cluster #1 HD neurons mean burst index cluster #2 mouse Distribution Pearson’s correlation coefficients between neurons homogeneity HD neurons non-HD neurons Classification score HD versus non-HD neurons auto-correlograms increasing duration auto-correlation Score relative classifiers shuffled data 1 perfect clustering auto-correlograms used t-distributed stochastic neighbor embedding auto-correlograms three brain states per neuron two-dimensional neurons continuously distributed gradient burstiness exception HD neurons separated “island” (Fig. 5b). distribution applied clustering algorithm K-means = 2) to space auto-correlograms first cluster contained most HD neurons (99/127, p < 0.001 cluster #1 shape auto-correlograms similar independent of average firing rates (Figanatomical density neurons cluster concentrated around AD nucleus (Fig. 5d). observations dynamics spike train emission independent of excitability to categorize HD neurons direct match between cluster #2 anatomy distribution average burstiness confirmed by high correlation between auto-correlograms neuronal HD pairs same shanks non-HD neurons less correlated auto-correlograms (Fig. 5e, t = 18.32, p = 10−73 t classify HD versus non-HD neurons? trained classifiers with auto-correlograms identify HD non-HD neurons used gradient boosted trees binary output trained classifiers on auto-correlograms duration Classification quality evaluated as percentage above chance level 100% HD cells labeled 50% above chance with first 6 ms of spike train auto-correlogram in brain states (Fig. 5f). classification limited information suggests duration refractory period level burstiness unique to HD cells results HD neurons share common dynamical properties AD neurons distinct in anterior thalamus non-HD neurons have broadly distributed properties irrespective nucleus originSlow firing dynamics distinguish HD non-HD neurons fast dynamics 100 activated during wakefulness slow timescales timescales correspond to typical firing duration HD neuron HD cells timescales similar during wakefulness REM AT neurons showed slow dynamics wakefulness sleep decay times auto-correlograms (Fig. characteristic timescale τ auto-correlogram decay captured by exponential behavioral timescales intrinsic markers neurons average decay times during wake REM sleep correlated across neurons (Fig. 6b r = 0.434 p < 0.001 HD cells showed differences neurons slow timescales slowest (Fig. 6b t = −0.23 10−18 −0.37 10−41 neurons AT exhibited behavioral timescale dynamics reflecting cognitive functions different integration distinguish HD non-HD neurons. 6Intrinsic slow dynamics during wakefulness REM sleep AT relation to fast dynamics NREM sleep Auto-correlograms three neurons (2.5 s during wake lines display exponential fit Decay time τ between wake REM sleep Red dots indicate HD neuronsshows average decay time HD non-HD neurons during wakefulness REM sleep Decay time REM function of burst index NREM sleep (r = −0.374 p < 0.001 Pearson’s HD non-HD cells indicated by red dark dots train dynamics fast slow from complex interactions between properties neurons network states during wakefulness REM sleep neurons show rare bursts25 burst index NREM sleep negatively correlated with slow dynamics wakefulness REM sleep (Fig. 6c REM sleep wakefulness r = −0.24 p < 0.001 data not relationship due to intrinsic properties timescales orders magnitude apart spike train statistics recapitulate firing properties HD cells nucleus form homogeneous population key in limbic processing inform navigation system wakefulness point towards directions hippocampal SWRs during NREM sleep constrain dynamics hippocampus other AT neurons heterogeneous revealed link between dynamics spike trains responses to NREM sleep SWRs intrinsic neuronal properties circuit integration related in AT only analyzed NREM sleep SWRsbehavioral task not maximize awake SWRs further studies needed relationship between HD non-HD neurons awake SWRs used non-linear reduction technique ISOMAP42 decode population states independent behavioral correlates26 HD cell population AD nucleus drifts high speed NREM stabilized at fixed points high gain before SWRs HD cells firing AD orchestrate hippocampal activity transient gain increase HD cell population preceding SWRs independent of timing UP states suggests AD nucleus influence hippocampal dynamics impact AD firing on timing SWRs indirect polysynaptic routes projection fields AD neurons suggest broad contribution AD to activity limbic system role synchronizing slow oscillation during NREM AD neurons influence limbic system NREM sleep convey coherent message population activity HD neurons AD nucleus organized during NREM virtual HD decoded any exception DOWN states HD cell population dynamics possible spatially tuned neurons limbic system updated by HD signal during NREM sleep HD cell population shows fast “sweeps” during NREM stabilization HD signal during SWRs suggests navigation system constrained directionstudy include neuronal recordings downstream structures prediction replay events hippocampus15 correspond to linear trajectories two-dimensional environment suggested replayed trajectories follow random experienced paths16 replayed initial correspond to HD network open question possibility HD cells influence replay medial entorhinal cortex cells maintain coordination during NREM replay trajectories45 independent of hippocampus46 coherent HD signal AD nucleus may subcortical process coordinated neuronal sequences during sleep correspond to experienced spatial trajectories16 Paired recordings of AT HD neurons grid cells necessary to test predictions question subcortical HD network influence external signal SWR generation content HD network randomly fluctuates during sleep integrating noisy inputs26 transient fluctuation excitability SWR through polysynaptic pathways state controlled by cortical feedback AD neurons presynaptic neurons receive feedback from post-subiculum post-subicular cells under influence AD nucleus during NREM sleep27 AD not only structure influencing SWRs auditory cortex neurons fire prior to SWR predict hippocampal replays48input output connectivity AT neuron intractable in vivo coordination with hippocampal activity functional integration Unlike HD neurons co-modulation with SWRs distributed for non-HD neurons independent of anatomical location prediction exception AD nucleus other AT nuclei diversity connectivity patterns previous studies flow information limbic system to hippocampus vary across brain states wakefulness NREM AT neurons similar responses to SWRs NREM sleep theta wakefulness suggesting routing information independently brain states results from coupling with hippocampus or modulation by other limbic structures spike train dynamics potential marker of functional coupling AT neurons with hippocampus observed strong relationship between spike train auto-correlograms responses to SWRs study determine intracellular properties AT neurons spike train dynamics reflect intrinsic HD cells homogeneous population neurons non-HD cells non-HD cells showed continuous spectrum dynamics graded burstiness refractory periods AT neurons show slow (~1 intrinsic timescale dynamics similar time constant spike train auto-correlogram during wakefulness REM sleepHD cells show longer behavioral timescales AT neurons results suggest cytoarchitectural definition AT nuclei insufficient exception AD nucleus show homogeneous behavioral correlates functional coupling with hippocampus spike train dynamics gene expression profile in AD unique in thalamus AT neurons continuously distributed evidence points link between gene expression spontaneous activity circuit development in thalamocortical pathways29 spike train dynamics coupling network activity related to acquired role neuron circuit molecular variability in spike train dynamics reflects signal transmission downstream open question AT key communication relay between limbic Interactions between thalamus hippocampus limbic system during learning SWRs crucial hippocampus memory consolidation NREM HD cell population stabilized at SWR occurence drifts maximum speed during thalamocortical spindles26 Future work necessary to characterize processes controlling HD non-HD cells AT hippocampal replay content routing hippocampal signals across limbic system framework study foundation for investigating relationship between intrinsic neuronal properties timescales functional integration in brain networks in vivorelationship organizational principle thalamic neurons experiments approved Institutional Animal Care Use Committee New York University Medical Center surgeries four male mice ~30 g (3–6 months implanted isoflurane anesthesia silicon probes above anterior thalamus −0.6 mm ML −0.5 to −1.9 mm DV 2.2 mm 10–15° angle shanks Hippocampal wire bundles implanted above CA1 −2.2 mm −1 to −1.6 mm ML 1 mm probes 8 shanks separated 200-μm eight recording sites (160 μm2 each 1–3-MΩ impedance staggered two-dimensional arrangement (20-μm vertical ground reference screws 100-μm tungsten wires implanted bone above cerebellum.Electrophysiological data animals recorded pre-sleep food foraging post-sleep sessions 5 h (±0.94 s Electrophysiological signals acquired 20 kHz 256-channel Amplipex system broadband signal sampled 1.25 kHz LFP cage two light-emitting diodes recorded digital video camera 30 frames per second LED locations detected resampled 39 HzSpike sorting performed semi-automatically using KlustaKwik followed manual adjustment waveform clusters Klusters software.Sleep stages sleep identified-automatically CA1 LFP spectrogram animal movements tracked sleep defined long period immobility NREM sleep high delta (1–4 Hz spindle (10–15 Hz) activity REM sleep strong power theta (5–12 Hz low delta DOWN state detected total firing rate neurons 10 ms smoothed Gaussian kernel 20 ms lower than 20% maximum DOWN epochs shorter than 30 ms longer than 500 ms discarded UP states epochs between DOWN states.SWR CA1 LFP bandpassed between 80 and 300 Hz Gaussian filter signal smoothed digital filter 11 z-scored mean standard deviation normalized signal thresholded between 3 7 standard deviations first set candidate ripples set reduced candidates duration between 25 ms and 350 ms thresholds lowered to 2 5 s.d. rippled detection validated by visual Ripples closer than 30ms merged unitary eventHead-direction classificationFor each session direction head animal calculated by orientation blue red LED top head Head-direction neurons detected by tuning curves ratio between angular direction each spike divided total time animal Peyrache et al.10 Rayleigh test performed null hypothesis uniformly distributed firing angular directions neurons classified as HD cells if peak firing rates >1 probability non-uniform distribution <0.001 concentration parameter inverse variance larger than 1.ISOMAP projection decodingISOMAP projections (Fig. 1 performed for sessions 10 HD neurons (eight sessions three spike trains binned during wakefulness SWRs heavy included first 15 min exploration period 1 s around SWRS (±500 duration bins 400 ms 30 ms 50% overlap during wakefulness SWR bins HD signal NREM topology HD cell population vectors used bin size 200 ms for wakefulness 100 ms overlap 75% around SWRs change radii angular velocity across sessions ISOMAP projections compared to baseline events NREM sleep Chaudhuri et al.26 computed root rates for variance in firing ratesBinned firing rates smoothed with Gaussian kernel three bin standard deviation (independent of bin duration).In each condition (wakefulness SWR controls), rates stacked together yielding rate matrix\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek-69pt{T\times N}}R∈RT×N N number neurons T total number time points rate matrix projected to two dimension plane[12pt]{minimal}{amsmath{upgreek-69pt}\times 2}I∈RT×2 using ISOMAP42 number of neighbors set to 200 reduce computational requirement for long spike trains repeated ISOMAP projection using random 150 SWRs 150 random NREM sleep time bins for baseline quantification same 15 min awake data until all SWRs analyzed ensures ISOMAP similar topology of embeddings neuronal “trajectories” around SWRs compared sessionAngular direction radius ring size decoded during SWRs computing element-wise arc tangent Euclidean norm each time point Angular velocity evaluated angular difference between two time points same procedure performed for randomly selected events during NREM sleep same ISOMAP embedding Radius angular velocity in Fig. 1d, e ratio\documentclass[12pt]{minimal{amsmath{wasysym{upgreek-69pt{document}\bar{X}}{SWRs{random{document(X ̄SWRs−X ̄random\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document\bar{X\end{document}X denoting average values).Bayesian decodingTo decoding used Bayesian decoding predict angular velocity during SWRs (Fig. 2) n = (n1, n2, number spikes fired HD neurons time window (30 ms) Φ animal’s head direction angle internal HD signaldecoder compute posterior probability P(Φ∣n), achieved Bayes’ rule conditional probabilities:1[12pt{amsmath-69pt$$P(\Phi \bf{n}}={P({\bf{n}} \Phi )P(}\bf{n}})}}P(Φ∣n)=P(n∣Φ)P(Φ)P(n neuronal firing independent spike counts follow Poisson distributions probability P(n∣Φ) equal to33:2[12pt]{minimal}{amsmath}-69pt}$$P({\bf{n}} \Phi )=\mathop{\prod }\limits{i = 1}^{N}P({n}_{i} \Phi )=\mathop{ 1}^{N}\frac{{(\tau {f}_{i}(\Phi ))}_{i\mathrm{ex}}{{{p}}}{-\tau {f}_{i}(\Phi )\end(n∣Φ(τfi(Φ(Φ τ bin size fi(Φ average firing rate cell i direction Φ wakefulnessHD tuning curve angle Φ). decoded direction value Φ highest posterior probability P(Φ∣n).Similar ISOMAP decoding angular velocity evaluated as ratio\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}{mathrsfs{upgreek}\oddsidemargin-69pt}{document}\mathrm{SWRs{random\end{document} ̄SWRs−v ̄random\documentclass[12pt]{minimal}{amsmath}{wasysym}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document\bar{v}}{SWRs\end{document average angular velocity during SWRs[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document\bar{v}}\mathrm{random}}}\end{documentaverage angular velocity events non-REM sleep used bin size 30 ms 50% overlap angular bins 6 degrees alignmentTo position electrodes putative anatomical location used procedure after each session probe lowered 70–140 μm Relative position estimated first day recording thalamus spiking activity 1–2 yielded relative session horizontal distance between shanks constant (200 μm). two-dimensional recording sites rotated 15° clockwise angle penetration silicon probe relative position aligned to anatomical map −0.82 mm matching configuration electrodes location AD nucleus head-direction neurons anatomical map enlarged 10% Figs. 2 S2 density HD neurons matched estimated anatomical position anterodorsal nucleus.SWRs cross-correlogramsThe SWR modulation computed neuron estimating cross-correlograms (average firing rates 5 ms bins ±500 ms SWR peak neuronal discharge SWRs co-modulated slow cross-correlograms normalized expected values null hypothesis no short timescale coupling SWR SWR cross-correlograms convolved with Gaussian windows 150 ms s.dprocess similar low-pass filtering firing 150 ms chosen corresponds upper bound duration SWRs distribution low-pass filtered rates SWRs inferred standard deviation Poisson process root expected rate time each subtracted expected rate from observed rate divided difference by standard deviation observed cross-correlograms expressed in z-values from expected distribution null hypothesis no short timescale coupling method enables specific fast modulation neuron by SWRs independent co-modulation thalamocortical slow obtained time series z(t) = [z1(t), zN(t)] N number neurons z values not typical z-score correspond to deviation from null hypothesis mean s.d. z-transforms not normalizedSWR modulation introduced SWR energy total power z-scored cross-correlograms[12pt{minimal\usepackage{amsmath{wasysym{upgreek\oddsidemargin{-69pt}{document} {z}{i\sqrt\mathop\sum\nolimits{t = 1}^{T}{z}{i(t^{2}}{document=∑t=1Tzi(t)2.Theta each neuron computed separately wake REM sleep continuous wavelets transform phase assigned each theta cycle spike counts 0 2π computed each neuron Fig. 3b Rayleigh test performed null hypothesis uniformly distributed firing rates theta cycle SWR modulation used jPCA method37 reduce cross-correlograms = 6 principal components considered Churchland et al.37)goal of jPCA temporal response profiles data governed by linear dynamical system\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs}{upgreek}\oddsidemargin}{-69pt}\begin{document}\bf{x}}}=M{{document}x=Mx with M[12pt{amsmath{upgreek}\oddsidemargin}{-69pt}{document\mathbb{R\end{document}RK,K (or M\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document{R\end{document}RN,N if no dimension reduction applied matrix M split into symmetric anti-symmetric matrix M = Msym + Manti related to expansion/contraction rotational dynamics latter matrix has imaginary eigenvalues association with rotational dynamicscapture phase response rotational space sufficient find best fit matrix Manti ∈ Γk×k minimizing error predicting\documentclass[12pt]{minimal}{amsmath{wasysym\oddsidemargin}{-69pt}\begin{document}\dot{X\end{document}X from MX expressed\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek}\oddsidemargin{-69pt}{document}$${M}\mathrm{anti}}}{*\mathrm{argmi}}{{{n}}}{M\in \Gamma }\dot{X}-MX|{F}\end{document}Manti*=argminM∈Γ∣∣X ̇−MX∣∣F F Froebenius norm Churchland et al.37 additional details jPCA decomposition\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin{-69pt}\begin{document}\mathrm{anti{document eigenvectors complex conjugate pairs suitable pair projection vectors U = (u1, u2)\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek\oddsidemargin}{-69pt}\mathbb{R\end{document}RT,2 obtained combining complex conjugate pairs (v1, v2) u1 = v1 + v2 u2 = i(v1 − v2)last step projecting SWR cross-correlograms first two jPCs (u1 u2) y = ZTu[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document{document SWR phase defined angle from origin 2D space φSWR = atan2(y2, y1).Auto-correlogramsFor each neuron three computed wake REM NREM sleep short Figs. 4 5 bin size 0.5 ms long Fig. 6 bin size 5 ms.Embedding auto-correlograms in 2D map (Fig. 5b) performed t-SNE algorithm38 concatenating part positive time lagright part auto-correlograms).\documentclass[12pt]{minimal}{amsmath}{wasysym}}}}}\oddsidemargin}{-69pt}{document}{A}_{2\\mathrm{ms}}\to 40\{ms{epoch}}}\mathbb{R}}}^{N}\end{document}A2ms→40msepoch∈RN auto-correlogram vector one epoch one neuron each point t-SNE projection mapping[12pt]{minimal}{amsmath}{wasysym}}\oddsidemargin}{-69pt}{document}:{X}_{i}\mathbb{R}}}^{3N} {Y}_{i}{R}}}^{2}\end{document}f:Xi∈R3N→Yi∈R2[12pt]{minimal}{amsmath}{wasysym}}{upgreek}{\oddsidemargin}{-69pt}{document}${X}_{i}={A{2\\mathrm 40{wake 40{NREM[A2ms→40mswakeA2ms→40msREMA2ms→40msNREM each neuron Fig. 5f neurons classified (HD versus non-HD gradient boosted trees XGBoost package39 Inputs classifier same t-SNE algorithm stacked auto-correlograms wake NREM sleep Classifier trained 1000 iterations default parameters 10-fold cross-validation procedure Chance levels determined thousand classifications random shuffling neuron labelsN neurons C classes classification score defined[12pt]{minimal}{amsmath{wasysym\oddsidemargin}{-69pt}{document}\frac{\mathop{\sum\nolimits,1}^{N}1({c}_{i}{c}},1}^{N}1({c}_{i}\hat{c}}_{i{s}-\mathop{\sum\nolimits 1}^{N}1({c}_{i}\hat{c}}_{i}{s}\end{document}s=∑i=1N1(ci=c^i)−∑i=1N1(ci=c^is)N−∑i=1N1(ci=c^is 1(x) indicator function[12pt]{minimal}{amsmath}{wasysym{upgreek}\setlength{\oddsidemargin}{-69pt}{document}$\hat{c}}\end{document}c class predicted ith neuron[12pt]{minimal}\usepackage{amsmath}wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}\begin{document labels predicted with classifier on shuffled data score 0 chance level 1 perfect classification.Burst bursts groups spikes interspike intervals between 2 ms and 9 ms Burst index61 computed ratio between observed count N2 ms<ISI<9 ms count expected homogeneous process same average firing rate\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts{mathrsfs{upgreek-69pt}{document}{N}}\tau{1\tau{2}}=T(exp(-\tau r{2}\times r\end{document ̄τ1<ISI<τ2=T(exp(−τ1×r)−exp(−τ2×r)) T duration epoch r mean firing rate τ1 = 2 ms τ1 = 9 msAuto-correlograms SWR cross-correlograms correlationTo compute correlation between SWR cross-correlograms auto-correlograms (Figs. 4e–g), stacked across brain states t-SNE projection used PCA reduce dimensionality each matrix kept first 10 components for 91% 99% variance computed 20-by-20 correlation matrix C for 10 SWR PCs 10 auto-correlogram PCs matrix diagonal for upper-left bottom right 10-by-10 blocks PCs perpendicular off-diagonal matrix in Fig. 4f defined total correlation as \documentclass[12pt{minimal}{amsmath-69pt{total ∣C∣ determinant of correlation matrix (if one PC for auto-correlograms SWR cross-correlograms total correlation linear correlation between two PCs). measure captures fraction state-space volume occupied by both measuresnull correlation determined shuffling identity neurons in datasets (1000 controlled correlation between neurons same nucleus shuffling groups same shank session (1000 times). correlation between SWR cross-correlograms auto-correlograms tested for correlation factors (firing rates burstiness). correlation between ith auto-correlogram PC jth SWR PC replaced by partial correlation factor\documentclass[12pt{minimal\usepackage{amsmath\oddsidemargin-69pt}{document}=rij−rifrjf1−rif21−rjf2 rkf (k = i or j correlation between PC weights external factor f (firing rates burstiness).Quantification statistical analysisAnalyses using customized code MATLAB Python (3.5) libraries numpy scipy scikit-learn.Reporting information research design Nature Research Reporting Summary.Supplementary Information Peer Review File Reporting Summary | 47.6 | 1.327674 |
10.1038/s41467-020-17931-w | PMC7426427 | Columns are the functional and morphological unit of the brain, but how neurons assemble into this structure was unclear. Here, the authors show that Dscam gene rewires neurons that derive from the same stem cell to establish columns through the process of lineage-dependent repulsion. | The brain is organized morphologically and functionally into a columnar structure. According to the radial unit hypothesis, neurons from the same lineage form a radial unit that contributes to column formation. However, the molecular mechanisms that link neuronal lineage and column formation remain elusive. Here, we show that neurons from the same lineage project to different columns under control of Down syndrome cell adhesion molecule (Dscam) in the fly brain. Dscam1 is temporally expressed in newly born neuroblasts and is inherited by their daughter neurons. The transient transcription of Dscam1 in neuroblasts enables the expression of the same Dscam1 splice isoform within cells of the same lineage, causing lineage-dependent repulsion. In the absence of Dscam1 function, neurons from the same lineage project to the same column. When the splice diversity of Dscam1 is reduced, column formation is significantly compromised. Thus, Dscam1 controls column formation through lineage-dependent repulsion. | IntroductionColumns are the higher-order morphological and functional units of the brain. A group of neurons gather to form individual columnar units, which are then precisely arranged to establish the brain. Several types of columnar units have been described in the cerebral cortex: cortical columns are groups of cells that share similar response selectivity, and microcolumns are cell type-specific clusters of neurons1–3, which are found in nearly all examined cortical regions.The radial unit hypothesis was proposed to explain the mechanism of column formation in the mammalian cerebral cortex. According to this hypothesis, columns are formed by clonally related neurons that are produced from a common progenitor cell4. The neurons of an individual radial unit are suggested to form a column that shares a similar response selectivity. However, sister neurons actually undergo lateral dispersion during development, become sparsely distributed and are mixed with neurons derived from other progenitors, calling into question the organization of columns simply via the clonally related neurons5,6. Thus, developmental mechanisms of columnar unit formation and significance of the neuronal lineage remain elusive.Like the mammalian brain, the fly visual system shows columnar organizations such as ommatidia in the retina, cartridges in the lamina, and columns in the medulla7–10. Photoreceptor neurons R1–8 form an ommatidium in the retina, while R1–6 neurons and lamina neurons L1–5 form a cartridge in the lamina. The medulla is the largest component of the fly visual system and each medulla column contains as many as 100 neurons. Medulla neurons make connections and form columnar units in the medulla neuropil8. Note that medulla neurons, whose cell bodies are situated in the medulla cortex, project their neurites toward the medulla neuropil (Fig. 1a). The axons and dendrites of neurons form repetitive columnar units in the medulla neuropil.Fig. 1Linage dependent repulsion in the larval medulla.a, b Schematics of the developing Drosophila medulla in L3 larval stage. Lateral (a) and dorsal (b) views. The dotted line in (a) indicates the plane showing the columns in the medulla layer. The dotted lines in (b) indicate the planes showing the NB and neuron layers. b, c Schematics of the proneural wave and temporal transcription factors. d, e Neurons of the same lineage are visualized by elav-Gal4 MARCM clones (GFP, white). Dpn (red) and Ncad (blue) visualize the NBs and neuropil, respectively. n = 26 in (d). n = 21 in (e). f, g The medulla structures visualized by Ncad (blue). f M0 and medulla layers in a lateral view. g The medulla columns in an anterior view. h–k Neurons of the same lineage are visualized by drf-Gal4 twin-spot MARCM clones (GFP in white, RFP in magenta). Ncad (blue). Sister neurons reroute in M0 layer, and innervate different columns in the medulla layer. Arrows and arrowheads indicate arborizations in the medulla layer and cell bodies, respectively. l A histogram showing the distance between pairs of neurons on the surface of the medulla layer (n = 110). 0 μm indicates fused indistinguichable pairs. m A box plot of the distance between distinguishable 110 pairs of neurons in 38 clones found in 38 independent brain samples. The cases of 0 μm distance are not included. Center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range. Median and average are 5.78 and 8.50 μm, respectively. Source data are provided as a Source Data file. n, o Projections of medulla neurons in early L3 larval stage visualized by elav-Gal4 MARCM clones (GFP in white). Ncad (blue). n 0–32 h L3 larva only showing M0 layer (n = 20). Neurites are already defasciculated within M0 layer. o 32–48 h L3 larva showing M0 and medulla layers (n = 49). Neurites are defasciculated in M0 layers and innervate the medulla layer. d–f, h–k, n, o Lateral views showing the neuron layer in (b). g Anterior view showing the columns along the dotted line in (a). Scale bars indicate 20 μm in (d, f, n, o) and 5 μm in (e, g, h–k).In our previous study, we demonstrated that R7, R8, and Mi1 are the core neurons that are concentrically arranged in the larval medulla according to N-cadherin-dependent differential adhesion7. Compared with the retina and lamina, developmental sequence of the medulla is more similar to that in the cerebral cortex. A single neuroblast (NB; a neural stem-like cell), produces a group of radially oriented and clonally related neurons that is analogous to the radial unit, a group of neurons that is produced from a single neural progenitor cell and migrates along the same radial fiber in the developing cerebral cortex (Fig. 1a)11,12. The development of ommatidia and lamina cartridges does not accompany NB and lineage-dependent development.In a previous study, we demonstrated that the wave of differentiation known as the proneural wave sweeps across the sheet of neuroepithelial cells (NEs) in the developing larval optic lobe. NEs are sequentially differentiated into medulla NBs behind the proneural wave (Fig. 1b)13–15. Thus, NBs are arranged according to their birth order and sequentially produce different types of neurons inside the brain. We and other groups have demonstrated that temporal transcription factors, such as Homothorax (Hth), Klumpfuss (Klu), and Eyeless (Ey), which are sequentially expressed in NBs, specify birth order-dependent production of neurons (Fig. 1c). Each NB and its daughter neurons form a radially arranged cluster of neurons during larval stage (Fig. 1a). The birth order of medulla neurons correlates with the concentric gene expression found in the medulla cortex. For example, Hth-positive NBs produce neurons that inherit Hth expression, which in turn induces Brain-spefic-homeobox (Bsh) expression. As a result, newly differentiated NBs produce Hth/Bsh double-positive Mi1 neurons that are located in the inner most concentric domain in the medulla cortex (Fig. 1a)16,17. Similarly, Klu- and Ey-positive NBs produce Runt- and Drifter (Drf)-positive neurons, respectively, forming the concentric domains outside of the Hth/Bsh domain11,16,17. Similar temporal patterning of neural stem cells and neurogenesis are found in the developing cerebral cortex12.During larval stage, medulla neurons of the same lineage are radially arranged, forming a radial cluster. However, these neurons are tangentially dispersed in pupal stage. As a result, the neurons of the same lineage are no longer clustered anymore beyond 24 h after puparium formation (APF)11,18,19. This phenomenon is similar to rather sparse distribution of the sister neurons found in the mature cerebral cortex5,6. Thus, columns are not simply formed according to the radial unit of the same neuronal lineage in the fly medulla and mammalian cerebral cortex.In this study, we demonstrate that sister neurons of a given lineage project to different columns in the larval medulla, suggesting that the neurons of the same lineage repel each other, which we refer to as lineage-dependent repulsion. As a potential candidate molecule that could regulate this process, we focus on Down syndrome cell adhesion molecule (Dscam), a gene significantly contributing to the phenotypes observed in Down syndrome20,21.Drosophila Dscam1 gene has three alternative exons encoding Ig2, Ig3, and Ig7 domains containing 12, 48 and 33 different splice variants, respectively. In total, Dscam1 encodes as many as 19,008 different ectodomains22. Homophilic binding of Dscam1 only occurs between identical isoforms that match at all three variable Ig domains and produces a repulsive signal23. Thus, neurons expressing the same Dscam1 isoforms show a repellent interaction. In addition, splicing of Dscam1 in each cell is probabilistic24. The vast diversity of Dscam1 isoforms is necessary for correct development of neural circuits25. Expression of the same Dscam1 isoform in a single cell causes self-avoidance, which is important for correct dendritic wiring26.We hypothesize that Dscam1 may be temporally expressed in NBs and is inherited by neurons of the same lineage to regulate the lineage-dependent repulsion. Indeed, we show that Dscam1 is temporally expressed in NBs under the control of Hth, a temporal transcription factor. Expression of Dscam1 in a radial unit is essential for lineage-dependent repulsion. Our findings suggest a function of Dscam1 in lineage-dependent repulsion, which provides a link between temporal patterning, neuronal lineage and column formation.ResultsLineage-dependent repulsion in the developing medullaNBs are located in the outermost region of the larval medulla primordium and produce a group of neurons toward the inner area of the medulla cortex with a radial orientation, as visualized by GFP expressed under the control of elav-Gal4 using the MARCM technique in order to label neurons of the same lineage (Fig. 1a, d). Daughter neurons of the same NB are linearly arranged in the larval brain forming a radial unit until the onset of tangential dispersion between 12 and 24 h APF11. By closely focusing on their neurites, we found that neurons of the same lineage widely project their axons encompassing multiple columns (Fig. 1e). During the late 3rd larval stage (L3), the developing neuropil, as visualized with the Ncad antibody, contains two distinct layers (Fig. 1e, f). The medulla layer contributes to adult medulla layers M1–M10. The other layer, located outside the medulla layer, is a temporal layer that disappears during the pupal stage11. We refer to this temporal structure as M0 layer (Fig. 1a, e, f). The medulla columns can be observed within the medulla layer in a frontal view (Fig. 1a, g). Note that the distance between neighboring columns is ~5 µm, while neurites of a radial unit extend as far as 50 µm distance and more (Fig. 1e, g; n = 8).The axons change their directions within M0 layer and eventually project to the medulla layer (Fig. 1e). To clearly distinguish each of the axons, we used twin-spot MARCM technique under the control of drf-Gal4, which visualizes a smaller number of neurons (Fig. 1h–k)27. In many cases, the axons change their direction within M0 layer, and project to the medulla layer or to the other brain region such as lobula through the medulla layer, which is reminiscent of the projection patterns in Tm-type neurons11,28.Usually, sister neurons that derive from the same NB do not form projections to the same columns. Instead, they are often rerouted in M0 layer and form projections to different columns within different regions of the medulla layer (Fig. 1h–j). Within the medulla cortex, axons of the same radial unit are bundled and project together toward M0 layer. However, they are defasciculated within M0 layer and project to different columns in the medulla layer.We quantified the distance between sister neurons that derive from the same radial unit by focusing on brain samples that contain small number of isolated clones (Fig. 1l, m). When the axons were fused, we regarded the distance as 0 μm. Otherwise, the distance between axon shafts on the surface of the medulla layer was measured (Fig. 1l). According to Ncad staining, the distance between columns is about 5 μm in larval medulla (Fig. 1g). Among 110 pairs of neurons, 11 and 37 pairs were 0 μm and 2–5 μm distant, respectively, and 62 pairs projected to the medula layer showing distance >5 μm (Fig. 1l). Note that 2–5 μm distance does not necessarily mean that they are projecting to the same column, because they may still project to the adjacent domains of the neighboring columns.Distribution of the distance between axon pairs is plotted in Fig. 1m. Median and average of the distance are 5.78 and 8.50 μm, respectively. The behavior of the axons suggests that the sister neurons repel each other very often. Since this repulsion occurs between neurons of the same lineage, we refer to this process as lineage-dependent repulsion.When we focused on the anterior part of the developing medulla, neurons of the same lineage often projected to distinct parts of the medulla neuropil (n = 27/32; Fig. 1h–j). However, in the posterior part of the medulla, the terminals of the sister neurons were often indistinguishable (n = 8/24; Fig. 1k). Among 11 pairs of fused axons, 9 were located in the posterior part of the medulla (Fig. 1l). Thus, lineage-dependent repulsion may be less prominent in the posterior part of the medulla. In the followings sections, we only focus on the anterior part of the medulla.To examine when lineage-dependent repulsion takes place early in development, we examined 0–32 h and 32–48 h L3 larval brains (Fig. 1n, o). In 0–32 h L3 brains, there was only one Ncad-positive layer, which is most likely M0 layer, because all axons project through M0 layer. The medulla layer is then found in 32–48 h L3 brains. Importantly, the axons of the same lineage are already defasciculated within M0 layer in the 0–32 h L3 stage. Thus, lineage-dependent repulsion takes place even before the formation of the medulla layer (Fig. 1n).To enable lineage-dependent repulsion, daughter neurons that derive from the same NB must remember the identity of their common mother NB and repeal each other according to their lineage. Dscam1 potentially exhibits nearly 20,000 splice variants (Fig. 2a). Identical Dscam1 isoforms bind with each other and provides a repulsive signal (Fig. 2b). Self-avoidance of dendritic processes is controlled by the same Dscam1 isoform expressed in the same neuron26. A similar mechanism may regulate lineage-dependent repulsion in the medulla column. However, in this case, repulsion must occur between a group of neurons that derive from the same NB. Since splicing diversity of Dscam1 is thought to be stochastically selected24, we assume that each NB temporally expresses a single Dscam1 variant, which is inherited by its daughter neurons. Therefore, the daughter neurons that are produced by the same NB likely express the same Dscam1 variant and repel each other, leading to projection to different medulla columns (Fig. 2c). In contrast, neurons of different lineages expressing different variants do not repel each other and are able to project to the same column.Fig. 2Detecting transcription of Dscam1 in NB and neurons using in situ RT-PCR.a Schematic of Dscam1 gene structure and alternative splicing indicating the primers used for in situ RT-PCR. b Homophilic binding of the identical Dscam1 isoform causes repulsion. c Schematic representation of lineage depend repulsion between neurons that derive from the same NB expressing the same Dscam1 isoform. d Control PCR (green), Lsc (blue) and Dpn (magenta) on the surface of the brain in a lateral view showing the NB layer, n = 22 (see Fig. 1b). e Quantification of signal intensity in the dotted box in (d). f Control PCR (green), Lsc (blue) and Dpn (magenta) in a dorsal view showing the decrease of mRNA signal in older NBs, n = 16 (same orientation as Fig. 1b). g, h, j Lateral views showing the neuron layer (see Fig. 1b). g Control PCR (green) and Dpn (magenta), n = 47. h Intron PCR (green) and Dpn (magenta), n = 13. i Quantification of signal intensity in the boxes in (g, h). Background signal was subtracted for Dpn. jDscam1 9.1 PCR (green) and Dpn (magenta) in a lateral view showing the neuron layer, n = 10 (see Fig. 1b). k Quantification of signal intensity in the boxes in (j). Each box contains one NB. Intensities in the dotted boxes are plotted with dotted lines. Scale bars indicate 20 μm. Source data are provided as a Source Data file (e, i, k).Transcription of Dscam1 essentially occurs in NEs and NBsTo test this hypothesis, transcription pattern of Dscam1 in NBs and neurons was examined. To detect low levels of mRNA, reverse-transcribed cDNA was PCR-amplified to perform in situ RT-PCR for Dscam1 mRNA (Fig. 2a; “Methods”). We designed control primers that amplify a fragment containing exons 8–10, which is shared by all Dscam1 isoforms (Fig. 2a).During larval development, NEs sequentially become NBs in a medial-to-lateral orientation on the surface of the developing medulla behind the proneural wave (Fig. 1b). Lsc is transiently expressed in a narrow band of 1–2 NE cells at the wavefront15, whereas Dpn is strongly expressed in all NBs. Strong mRNA signals were found in NEs and NBs following the wavefront of the proneural wave, as indicated by Lsc expression and gradually decreased in the older NBs (Fig. 2d–f), suggesting that Dscam1 is temporally transcribed in the newborn NBs. The signals decreased as the neurons became older in the inner part of the brain (Fig. 2g, i). The observation that strong Dscam1 mRNA signals form a circle encompassing the entire larval brain hemisphere (Fig. 2g, Supplementary Fig. 1a) indicates that they are temporally transcribed in all of the newborn medulla NBs and inherited to their daughter neurons.To confirm whether Dscam1 is newly transcribed in the medulla neurons, or not, we detected pre-mRNA of Dscam1 by using a primer set that amplifies the intron between the exons 9.33 and 10 (Fig. 2a). As we expected, the intron PCR signals were highly restricted to NBs on the surface of the brain hemisphere (Fig. 2h, i). The signals in neurons were hardly detectable compared with control RT-PCR results (Fig. 2i). These results suggest that Dscam1 mRNA is essentially transcribed in NEs and newborn NBs immediately behind the proneural wave, and is inherited to the daughter neurons.To confirm the validity of our in situ RT-PCR technique, we examined Dscam1 mRNA signal in clones homozygous for Dscam20, a null mutant of Dscam1 (Supplementary Fig. 1b). Compared with control cells, signals for Dscam1 protein and Dscam1 mRNA as visualized by in situ RT-PCR were abolished, suggesting that in situ RT-PCR specifically detects Dscam1 mRNA.We also visualized Ncad mRNA by using in situ RT-PCR (Supplementary Fig. 1c–e). Consistent with Ncad protein expression in medulla neurons, we observed strong Ncad mRNA signal in the inner region of the medulla cortex. Relatively uniform Ncad mRNA signal throughout the medulla strongly suggests that the sharp decrease of Dscam1 mRNA signal inside the brain indeed recapitulates Dscam1 expression (Fig. 2g–i).Neurons of the same lineage express similar Dscam1 isoformsTo test the hypothesis that the same Dscam1 isoform is inherited by the daughter neurons of a NB, we performed in situ RT-PCR for a single variant of exons 4, 6, and 9 (Fig. 2a and Supplementary Fig. 1f–k). According to the results of the previous study, one splice variant is stochastically chosen from the alternative exon 424. If the same phenomenon occurs for exons 6 and 9, a limited number of medulla NBs should express the same exon variant, which will then be inherited by their daughter neurons. Indeed, we observed a cluster of NBs and their daughter neurons expressing the same variant of exons 4, 6 and 9 (Fig. 2j, k and Supplementary Fig. 1f–n). The location of 9.1-positive NB cluster was not uniform, but variable in each brain sample. In many cases, a brain contained one or two domains that express a particular exon variant. We repeated the same experiment for 22 splice variants from exon 4 (4 variants), 6 (10 variants) and 9 (8 variants; “Methods”). At least ten samples were observed for each variant, and we obtained essentially the same results as quantified in Supplementary Fig. 1l–n, suggesting that an alternative splice variant is stochastically chosen.If the stochastic choise solely occurs in NBs, a salt-and-pepper-like pattern should appear. Indeed, closer look at their expression patterns occasionally reveals a lack of in situ RT-PCR signals in an expression domain (Supplementary Fig. 1i). However, this may be due to cell cycle dependent changes in mRNA distribution. Since Dscam1 expression is initiated in NEs (Fig. 2d–f), which quickly divide symmetrically, a group of NBs is supposed to share the same exon variant. Or, there might be unknown mechanisms that provide a bias on the choise of alternative exons.We assume that alternative splicing of all of the alternative exons (exons 4, 6, and 9) is independently and stochastically determined according to the previous study24. If so, a cluster of NBs expressing the same variant of exon 9 most likely contains NBs expressing different variants of the other exons and can presumably be further subdivided by the selection of exons 4 and 6. Thus, one or very small number of NB lineages could share exactly the same splice variants.On the other hand, we also expressed a single isoform of Dscam1 in neurons by generating clones of cells containing a single Dscam1 isoform (dscam3.31.8; Supplementary Fig. 2a–e). The neurons expressing the single isoform showed a normal radial arrangement, and their neurites were normally defasciculated in M0 layer projecting to the wide area of the medulla neuropil, as found in wild-type control clones (Supplementary Fig. 2e).In wild-type condition, we assume that exon 9.8 is stochastically chosen upon alternative splicing. In contrast, the single-isoform mutant, dscam3.31.8, lacks all variants of exon 9 except for 9.8. We next asked what happens to the expression pattern of exon 9.8 in this single variant mutant background. Surprisingly, mRNA for exon 9.8 was uniformly detected in the medulla NBs forming a circle encompassing the entire larval brain hemisphere (Supplementary Fig. 2f, g), suggesting that exon 9.8 is always chosen in the absence of the other variants of exon 9. Consistently, mRNA for exons 9.1 and 9.4 were not detected in the same mutant background (Supplementary Fig. 2h, i). These findings support our hypothesis that neurons of same lineage express similar Dscam1 isoforms.Dscam1 protein is stabilized in medulla neuronsNext, we examined the expression pattern of the Dscam1 protein in NBs and neurons. We found that Dscam1 protein is weakly expressed in Lsc-positive NEs and 1–2 rows of Dpn-positive NBs behind the proneural wavefront and decreases in older NBs (Fig. 3a, b, e), suggesting that Dscam1 protein is temporally expressed accompanying NB differentiation, but is rapidly downregulated in older NBs.Fig. 3Expression pattern of Dscam1 protein.a, b Expression pattern of Dscam1 protein (white) in NEs and NBs in a lateral view showing the NB layer (see Fig. 1b). (a) Lsc (blue). (b) Dpn (blue). c, d Expression pattern of Dscam1 protein (white or magenta) in neurons in a lateral view showing the neuron layer. c Dpn (blue). d Arrows indicate Dscam1 signals along the axons of GFP expressing neurons (Ay-Gal4 UAS-GFP; green). e, f Quantification of Dscam1 signal intensity in the boxes in (a) and (c), respectively. g Dscam1 (white) and Ncad (blue) visualize the columnar structures in the medulla layer in an anterior view (dotted line in Fig. 1a). Scale bars indicate 20 μm. Source data are provided as a Source Data file (e, f).In contrast to mRNA distribution, Dscam1 protein is strongly accumulated along the neural fibers that are radially oriented in the medulla cortex, which colocalize with neurites projecting from the radial cluster of neurons (Fig. 3c, d, f). Since Dscam1 signals in NBs and neurons are eliminated in Dscam1 null mutant clones (dscam120; Supplementary Fig. 1b), the above signals indeed reflect the expression patterns of Dscam1. Thus, we assume that Dscam1 is predominantly transcribed in newborn NBs behind the proneural wave, while Dscam1 mRNA inherited by their daughter neurons is rapidly degraded. On the other hand, Dscam1 protein, which may be translated in NBs and neurons, is stabilized in neurons and localizes to the neurites (Fig. 3d).The temporal restriction of Dscam1 transcription may be essential for lineage-dependent repulsion. During alternative splicing, spliceosome machinery assembles at the splice sites forming a complex that leads to the selection of a single splice variant29. If the duration of transcription is restricted, a small number of splice variants will be selected. As a result, a NB will produce a single or very small number of splice isoforms, which are shared among its daughter neurons. When a group of neurons expresses the identical Dscam1 isoform, the recognition between Dscam1 proteins causes mutual repulsion, leading to lineage-dependent repulsion (Fig. 2b, c).The strong Dscam1 signals found in M0 layer (Fig. 3c, d) are consistent with the idea that Dscam1 regulates the spreading of neurites within M0 layer in the larval medulla (Fig. 1e, h–j). The columnar distribution pattern of Dscam1 in the medulla layer, which overlaps with the columnar distribution of Ncad, also suggests its essential role in column formation (Fig. 3g).Hth regulates temporal Dscam1 expression in NBsSimilar to Dscam1, expression of Hth and Ey in NBs is inherited by the daughter neurons in the larval medulla11,16,17. Hth is the first temporal transcription factor expressed in NEs and NBs. Hth activates expression of Bsh and Ncad in the early born medulla neurons, which differentiate to a single type of medulla neuron, Mi1 (Fig. 1c)11,30.We compared the expression patterns of Dscam1, Hth, Bsh, and Ncad, and found that Dscam1 and Hth are coexpressed in NEs and newborn NBs (Fig. 4a). In contrast, Bsh and Ncad are specifically expressed in neurons and not in NEs/NBs11,30. Strong Dscam1 signals were detected in neurons located in the inner area of the developing medulla (Fig. 3c, d). Similarly, a transcriptional regulator, Engrailed, regulates the expression of a guidance receptor, Frazzled, in NBs to control axon guidance during Drosophila embryonic development31.Fig. 4Hth regulates the expression of Dscam1.Lateral views of L3 larval brains showing the NB (a, b, d) and neuron layers (c, e; see Fig. 1b). Dscam1 (white in a–c, magenta in d, e). a Dscam1 expression overlaps with Hth (green) in NEs, and with Hth and Dpn (magenta) in NBs. b, c Background level of Dscam1 signal in hth mutant clones visualized by the absence of GFP (green; arrows) compared with the control cells (arrowheads). Pixel intensity of Dscam1 signal was uniformly enhanced throughout the image (b). d, e Ectopic Dscam1 upregulation in clones expressing hth visualized by GFP (green; arrows) under the control of Ay-Gal4. Ncad (blue). The areas indicated by the dotted boxes are enlarged in the right panels. Scale bars indicate 20 μm. f–i Quantification of signal intensity in the boxes in (b–e). Intensities in the dotted boxes are plotted with dotted lines. f, g Dscam1 signal is reduced in GFP-negative hth mutant clones (b, c). h, i Dscam1 signal is enhanced in GFP-positive hth expressing clones (d, e). Background signal was subtracted for Dscam1 (f, h). Experiment was independently repeated at least three times with similar results (a–e). Source data are provided as a Source Data file (e, g, h, i).Since Hth is a temporal transcription factor whose expression in NEs and NBs overlaps with that of Dscam1 and activates the expression of Bsh and Ncad in neurons, it may also regulate the expression of Dscam1. To test this possibility, we generated hth mutant clones. As expected, Dscam1 expression in the medulla NEs and NBs was autonomously eliminated in hth mutant clones (17/43; Fig. 4b, f). The Dscam1 signals in medulla neurons were reduced (Fig. 4c, g). The residual Dscam1 signals may be due to nonspecific background of Dscam1 antibody; because similar background signals were also detectable in Dscam1 null mutant clones (Supplementary Fig. 3b). The strong Dscam1 signals along the neurites were completely eliminated in hth mutant clones (10/41; Figs. 3c, d and 4c, g).Note that the hth mutant used in this study (hthP2) is the most commonly used allele, but is hypomorphic (Fly Base). The incomplete loss of Dscam1 expression may be due to its hypomorphic nature. Or, there might be additional unknown factors that act partially redundantly with hth.To test whether hth expression is sufficient to induce Dscam1 expression, we generated clones ectopically expressing hth (Fig. 4d, e). We found that ectopic hth expression in NBs effectively upregulated the expression of Dscam1 (11/21; Fig. 4d, h). In addition, Dscam1 expression was upregulated in neurons and localized along the neurites upon ectopic hth expression (21/47; Fig. 4e, i). The upregulation of Ncad found in hth expressing clones suggest that the ectopic hth expression causes premature neuron differentiation, which may indirectly upregulate Dscam1 (Fig. 4e). However, Dpn-positive NBs also show upregulation of Dscam1 expression on the surface of the brain (Supplementary Fig. 4). Taken together, these results indicate that Hth acts as a trigger of Dscam1 expression in NEs and NBs.While Hth is widely expressed in NEs, Dscam1 expression is found in a part of NEs (Figs. 2d–f and 3a). Thus, Hth expression in NEs may not be sufficient to induce Dscam1 expression. The other unknown factor might be necessary to cause the full induction of Dscam1 expression.Loss of Dscam1 causes the loss of neural defasciculationDscam1 protein is expressed in NBs and their daughter neurons. As we demonstrated, neurites of sister neurons that derive from the same NB are rerouted within M0 layer and project to distinct columns (Fig. 1e, h–j). We hypothesize that the sister neurons express the same or similar Dscam1 isoforms, causing repulsion between neurons of the same lineage. To test this possibility, we compared projection patterns of neurites of the radial units in control and Dscam1 null mutant clones (dscam20; Fig. 5a–d).Fig. 5Loss of Dscam1 impairs lineage-dependent repulsion.Neurons of the same lineage are visualized by elav-Gal4 MARCM clones (GFP in white). Ncad (blue) visualizes the neuropil structure and columns. Projection patterns of neurons of the same lineage in control (a) and Dscam1 mutant clones (b). Lateral views of L3 larval brains showing the neuron layer (see Fig. 1b). Wide spread tangential projections found in M0 layer in control clones (a) are suppressed in Dscam1 clones (b; arrowheads). The neurons innervate the medulla layer following M0 layer. c Quantification of the distance between neurites of the same lineage. Control: n = 8. Dscam1 mutant: n = 10, Center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range. Average projection distances are 49.2 and 17.45 μm, respectively. SD are 14.8 and 8.09, respectively (two-sided t test, P = 0.00043). d, e Column morphology of 48 h APF pupal medulla showing the M1–2 layers. Control (d) and Dscam1 mutant clones (e). Regular column morphology is disrupted in and around Dscam1 mutant clones (arrows). f Classification of column morphology. In contrast to normal columns, abnormal columns are classified into irregular, fused and unclear columns. g Quantification of column morphology. Control: 87 columns from 3 brains. Dscam1 mutant clones: 136 columns from 3 brains (Fisher exact test, normal P = 8.4 × 10−47, unclear P = 7.23 × 10−19, irregular P = 7.18 × 10−12, fused P = 0.00046). h, j Tangential migration of neuronal cell bodies of the same lineage at 24 h APF. Dorsal views showing control (h) and Dscam1 mutant clones (j). i, k Spatial distribution of the cell bodies in the medulla cortex in (h, j) are quantified. The wide spread tangential distribution in control (h, i; SD = 42 μm, n = 3) is suppressed in Dscam1 mutant clones (j, k; SD = 24 μm, n = 3, two-sided t test, P = 0.0022). l Schematic representation of the optic lobe at 24 h APF. Scale bars indicate 20 μm in (a, b, d, e, h, j) and 5 μm in (f). Experiment was independently repeated at least three times with similar results (a, b, d, e). Source data are provided as a Source Data file (c, g, i, k).In control clones, neurites of a radial unit were defasciculated and projected to remote columns (distance = 50 μm, n = 8; Fig. 5a, c). Note that we measured the largest distance among multiple neurons, which is greater than the average distance between two neurons (Fig. 1l). In contrast, Dscam1 mutant neurites were bundled in M0 layer and projected to the same or nearby columns (distance = 20 μm, n = 10; Fig. 5b, c), suggesting that Dscam1 is responsible for lineage-dependent repulsion.To determine whether lineage-dependent repulsion is essential for column formation, we examined changes in columnar structure, as visualized with Ncad antibody at 48h APF, in the presence of control and Dscam1 null mutant clones (Fig. 5d, e). We classified column morphology into normal, unclear, irregular and fused (Fig. 5f; “Methods”), and quantified column morphology (Fig. 5g). Abnormal, unclear, irregular, and fused columns were significantly increased in the presence of mutant clones.The shape of individual columns and column arrangement were widely affected when the medulla contained Dscam1 mutant radial units (Fig. 5e). The nonautonomous columnar defects caused by Dscam1 mutant clones suggest that axons of the same lineage need to project to a wide range of columns under the control of Dscam1-dependent repulsion. Thus, Dscam1 is essential for lineage-dependent repulsion and subsequent column formation.Although the neurites of a radial unit repel each other by projecting to remote columns early in the larval stage (Fig. 1e, h–j), dispersion of their cell bodies occurs between 12 and 24 h APF11. To examine whether the dispersion occurs in Dscam1 mutant radial units, we compared distributions of cell bodies in control and Dscam1 mutant clones (dscam20; Fig. 5h–k). In control clones, the cell bodies of a radial unit were widely distributed throughout the medulla cortex showing tangential dispersion (n = 3, 20 neurons/brain, SD = 42 µm; Fig. 5h, i). However, when the radial unit lacked Dscam1 function, the cell bodies remained close to each other, forming a cluster at 24 h APF (n = 3, 20 neurons/brain, SD = 24 µm; Fig. 5j, k). Thus, the dispersion of cell bodies also depends on Dscam1.Loss of Dscam1 diversity leads to columnar defectsPrevious studies have demonstrated that diversity of Dscam1 isoforms is critical for neuronal wiring using Dscam1 mutant alleles in which the number of splice isoforms is reduced (Fig. 6a)32. We asked whether Dscam1 diversity is also crucial for column formation.Fig. 6Loss of Dscam1 isoform diversity leads to defects in column formation.a Schematic representation of Dscam1 gene in wild type and Dscam1 single-isoform mutant backgrounds. Column morphology of 48 h APF pupal medulla visualized by Ncad (blue) in the M1–2 (top) and M9–10 (bottom) layers in control (b) and Dscam1 single mutant backgrounds (c; Dscam123/Dscam13.31.8). Scale bars indicate 20 μm. Experiment was independently repeated at least three times with similar results (b, c). d Schematic representation of the optic lobe at 48 h APF. (e) Quantification of column morphology in the top layer. Control: 74 columns from 5 brains. Single-isoform mutant: 143 columns from 5 brains (Fisher exact text, normal P = 2.11 × 10−40, fused P = 5.4 × 10−17, irregular P = 0.33, unclear P = 0.21). Source data are provided as a Source Data file.By combining the single-isoform mutants Dscam3.31.8 and a null allele, Dscam23, we generated a mutant background that produces only one Dscam1 isoform (Fig. 6a). We compared column shape and column arrangement as visualized with Ncad antibody at 48 h APF (Fig. 6b, c). In control brains, regular arrangement of the donut-like columns was observed (Fig. 6b, d)7. In contrast, shape and arrangement of the columns were significantly disorganized in the Dscam1 single-isoform backgrounds (Fig. 6c). The defects in the top layer were quantified and statistically tested (Fig. 6e). Since the bottom layer was too disorganized and column shape was unidentifiable in the mutant, quantification of column morphology was not applicable in the bottom layer (n = 5).DiscussionIn this paper, we demonstrate that dscam1 is temporally transcribed in NEs and newborn NBs under the control of the temporal transcription factor, Hth. dscam1 mRNA and Dscam1 protein are then inherited by neurons in a lineage-dependent manner (Fig. 7). Among 20,000 splice isoforms of Dscam1, small number of isoforms are stochastically selected. Since Dscam1 is temporally expressed in NBs, neurons of the same lineage tend to share the same or similar splice isoforms, which causes homophilic binding and subsequent repulsion. As a result, neurons of the same lineage repel each other and contribute to different medulla columns (Fig. 7b).Fig. 7Inheritance of Dscam1 mRNA and Dscam1 protein, and lineage depend repulsion.Schematics of Dscam1 transcription in NEs and newer NBs under the control of Hth behind the proneural wave (a). Graded distribution of Dscam1 mRNA and Dscam1 protein in neurons of the same lineage that controls lineage-dependent repulsion within M0 layer and innervation to different columns (b).However, our results cannot explain the following issues. First, sister neurons occasionally project to very distant regions of the medulla. For example, the sister neurons can project to the ventral and dorsal halves of the medulla neuropil at the same time (Fig. 1i). Since repulsive action of Dscam1 is triggered by its direct homophilic interaction, the projections of sister neurons over such a long distance cannot be explained solely by Dscam1 function. As-yet-unknown chemorepulsive molecules or other guidance molecules that are expressed in distinct subdomains of the medulla might work together with Dscam1. For example, Slit and Netrin act as repulsive guidance molecules during larval optic lobe development33,34. Optix is expressed in the dorsal and ventral subdomains of the medulla19,35 and may regulate the region specific expression of downstream guidance molecules.Second, in the presence of the same Dscam1 isoform in a radial unit, the neurites are closely bundled within the medulla cortex prior to defasciculation (Fig. 1). A previous study showed that low expression levels of Dscam1 produce adhesive signal, while high expression levels provide repulsive signal23. Thus, the function of Dscam1 may be switched from adhesive to repulsive depending on its expression level, as demonstrated in Netrin signaling34,36. Consistent with this idea, Dscam1 is more strongly localized in M0 than its localization within the medulla cortex (Fig. 3c). Alternatively, an as-yet-unknown mechanism may be involved.The waves of differentiation are observed during development of a wide variety of visual systems in animals from flies to mammals9. In addition, the radial unit, a group of neurons generated by a common neural stem cell, is also found during column formation in the mammalian cerebral cortex4, and neurons of the same lineage are dispersed later in development, as found in the fly medulla5,6. Thus, the mechanism demonstrated in this study combining the wave of differentiation and the temporal expression of guidance molecule encoding column identification code might be an evolutionarily conserved strategy of column formation from fly to mammalian brains.In this study, we propose that splice diversity of Dscam1 regulates column formation through lineage-dependent repulsion. Similar but distinct mechanisms are found in the mammalian adaptive immune systems and olfactory systems. The immunoglobulin and T cell receptor (TCR) genes contain multiple gene segments that are stochastically selected and rearranged to generate variable molecules that recognize various antigens37. In contrast to the irreversible and permanent recombination processes found in the adaptive immune systems, the alternative splicing of Dscam1 is more flexible and reversible.In the mouse olfactory system, each olfactory sensory neuron expresses one olfactory receptor (OR) gene out of ~1000 OR genes38. A single OR gene is stochastically selected by a cis-acting regulatory element that controls multiple genes located at a genetic locus. Furthermore, a negative feedback mechanism inactivates expression of the other OR genes after one OR gene is selected. Thus, the expression of OR genes is rigorously regulated.Thus far, we only suggest that a single Dscam1 isoform is selected due to occupation of a single splice variant by spliceosome and temporally restricted transcription. If there is no specific mechanism that represses the expression of the other unselected isoforms, the process of lineage-dependent repulsion may not be very strictly controlled. Nevertheless, the mechanism that we propose in this study is very simple. We only need to assume the existence of multiple splice isoforms and temporally restricted transcription in stem cell-like progenitor cells. It will be interesting to determine whether similar mechanisms exist in other biological systems including column formation in mammalian brains.MethodsFly strainsFly strains were maintained on standard Drosophila medium at 25 °C. The following mutant and transgenic flies were used: dscam2039, dscam21, dscam2340, dscam3.31.8, dscam10.27.2532, hthp230, UAS-hth1-1241, elav-Gal4, hs-flp, FRT42D, FRT82B, ubiGFP, tub-Gal80, UASCD8GFP42, and Ay-Gal443.Clonal analysisNeurons of the same lineage were visualized by crossing hs-flp elav-Gal4 UAS-CD8GFP; tub-Gal80 FRT42D with FRT42D, dscam20FRT42D, dscam21FRT42D, dscam23FRT42D and dscam3.31.8FRT42D, and applying 34 °C 30 min heat shock (Figs. 1d, e, n, o, 5, and Supplementary Fig. 2a–e). Small number of neurons of the same lineage were visualized by crossing hs-flp; UAS-CD2RFP UAS-GFP-Mir FRT40A with UAS-CD8GFP UAS-CD2-Mir FRT40A; drf-Gal4, and applying 34 °C 30 min heat shock (Fig. 1h–k). Dscam1 null mutant clones were generated by crossing hs-flp; ubi-GFP FRT42D with dscam20FRT42D, and applying 37 °C 60 min heat shock (Fig. 3, Supplementary Figs. 1b and 3). hth mutant clones were generated by crossing hs-flp; ubiGFP FRT82B flies with hthP2FRT82B, and applying 37 °C 60 min heat shock (Fig. 4b, c). hth overexpression clones were generated by crossing hs-flp; Ay-Gal4 UAS-GFP strain with UAS-hth1-12, and applying 37 °C 60 min heat shock (Fig. 4d, e and Supplementary Fig. 4).In situ RT-PCRIn situ RT-PCR was performed as described below. Larval brains were dissected in fresh PBS and promptly transferred to ice cold 4%formaldehyde/PBS solution. The brains were transferred to ice cold 4% formaldehyde/PBS solution in a tube and fixed at 4 °C overnight. The formaldehyde solution was removed, and the brains were washed with ~800 μl of –20 °C methanol and fixed in methanol at −20 °C overnight.The brains were washed with ~800 μl of 100% ethanol twice and incubated in ~800 μl of 50% xylene/ethanol solution at room temperature for 30 min. The brains were washed with ~800 μl of 100% ethanol twice. 100% ethanol was gradually replaced with a series of 75, 50, and 25% ethanol/H2O solutions and H2O. The tube was cooled down on ice. H2O was replaced with ~800 μl of –20 °C 80% acetone/H2O solution. The tube was incubated for 10 min on ice. The acetone solution was replaced with H2O and the brains were washed with ~800 μl of PTw (0.1% Tween20 in PBS) twice. PTw was replaced with 4% formaldehyde/PBS solution and the brains were fixed at room temperature for 30 min. The brains were washed with ~800 μl of PTw three times and transferred to a PCR tube.Prior to reverse transcription, the brains were incubated at 65 °C for 5 min in a solution containing 0.34 mM 3′ reverse transcription primer (see below) and 1mM dNTP mixture in H2O (15 μl scale), and were cooled down on ice. Adding 6 μl of 5× PrimeScript Buffer, 32U RNase inhibitor, 200U PrimeScript Reverse Transcriptase (TaKaRa) and H2O to the solution (30ul scale), reverse transcription was performed by incubating at 30 °C for 10 min, at 42 °C for 30–60 min and 70 °C for 15 min. The tube was cooled down on ice and the brains were washed in H2O.PCR was performed in 50 μl of PCR solution containing 1× KOD Buffer, 0.2mM dNTP mixture, 2.5U KOD DNA polymerase (Thermo Fisher), 1 μM 5′ primer, 1 μM 3′ primer (see below) and 0.02 mM digoxigenin-11-dUTP (Sigma-Aldrich). Initial denaturation for 1 min at 95 °C, 20 cycles of 10 s denaturation at 98 °C, 5 s annealing at 55 °C and 50 s elongation at 68 °C, and final extension for 5min at 72 °C.The brains were washed in PBT (0.3% TritonX in PBS) and blocked in 5–10% normal serum/PBT solution at room temperature for 30–60 min. Primary antibody reaction was performed in a solution containing mouse anti-Dig antibody (1:200), other primary antibodies and 1% normal serum in PBT at 4 °C overnight. The brains were washed in PBT four times. Secondary antibody reaction was performed in a solution containing anti-mouse FITC secondary antibody (1:200), other secondary antibodies and 1% normal serum in PBT at 4 °C overnight. The brains were washed in PBT three times. PBT was replaced with PBS, and the brains were mounted in VECTASHIELD.PrimersReverse transcription of Dscam1 was performed by using the reverse transcription primer RT11 (GTGTTGGACCTTGACGTCTT). Control in situ PCR was performed by using the exons 8–10 primers Ctrl8_5 (GCTGATTATCGAGAATGTGGAA) and Ctrl10_3 (TTCTTCCATGTAACTTGGGGTTT).In situ PCR of Dscam1 exon 9 was performed by using the primers Ctrl11_3 (AACTCCGGTGGAAAGGATCT) and Ds9.1 (CCTTTGATTTCGGTGAGGAA), Ds9.2 (ACGAGTTGGACATGG), Ds9.3 (ACGAGCTGGATATGGTCTCG), Ds9.4 (ACATGGTGTCCGCCTATTGT), Ds9.5 (GCGATGTCCCAATTACCAT), Ds9.6 (TCGGCAGCGAAGTCTTTAAT), Ds9.7 (GCGGAGAAGTGGCTAGTGTC), and Ds9.8 (ATCCAAGCGTTTGACTTTGG).In situ PCR of Dscam1 pre-mRNA was performed by the intron primers Int1_5 (ACCGCATCAGAAAACCAATC) and Int1_3 (GTGCTGTGTGTGGATTTTGC), or Int2_5 (TCATGCTCCAACACCGAATA) and Int2_3 (CAGGGCGAATTGTTTACGTT).In situ PCR of exon 4 was performed by using the primers Exon5R (CTCTCCAGAGGGCAATACCA) and Ds4.1 (GAGGCGGATGTTAACAAGGA), Ds4.2 (ACACAAGGCATTTGTCATCC), Ds4.3 (CCTATGTAATACGCGGCAATG), and Ds4.4 (AATCGGAGGTCAACAACGAG).In situ PCR of exon 6 was performed by using the primers Exon7R (TCCTCGACTACTGCGTCCTT) and Ds6.4 (TACGCTCCTTTGTCCAGCTC), Ds6.8 (GCAGATCCAGAGCGGAACTA), Ds6.12 (TCGAACAATGGAGGTGTCTG), Ds6.16 (TTTCTCCATGCAATGTCCTG), Ds6.20 (TGCAGGATAAGTTTGGTGTGA), Ds6.24 (AAAGGACGGTTTCAGTCACG), Ds6.28 (AGGAAGTGGGACCCTGCTAT), Ds6.32 (TCCACCGCAATACTTTGTCC), Ds6.36 (CATCGAGGTGCAAAAGTCAA), and Ds6.40 (GTCGATTAAGGCCAGCTTTG).Reverse transcription of Ncad was performed by using the reverse transcription primer RT4R (GAATTGGGTCCATTGCTGTT). In situ PCR for Ncad was performed by using the primers Ncad_E2 (GTATCGAAGGCAATCCCACA) and Ncad_E3 (TTTGGAAATGTGCCATCCTT).HistochemistryImmunohistochemistry was performed as described below. Larval brains were dissected in PBS, and fixed in 4% formaldehyde/PBT solution at room temperature for 30–60 min. The brains were washed in PBT and blocked in 5–10% normal serum/PBT solution at room temperature for 30 min. Primary antibody reaction was performed in a solution containing primary antibodies and 1% normal serum in PBT at 4 °C overnight. The brains were washed in PBT. Secondary antibody reaction was performed in a solution containing secondary antibodies (1:200) and 1% normal serum in PBT at 4 °C overnight. The brains were washed in PBT and mounted in VECTASHIELD.Primary antibodies: rabbit anti-Hth (1:1000; Adi Salzberg, Israel Institute of Technology, Israel), rat anti-Dpn (1:100; 11D1CH11, abcam), guinea pig anti-Lsc (1:1200), rat anti-Ncad (1:20; DSHB), mouse anti-Dscam1 (1:200; S. Lawrence Zipursky, UCLA, USA), mouse anti-Dig (1:200; 21H8, abcam) and rabbit anti-GFP Alexa488 conjugated (1:1000; Invitrogen A21311) antibodies. Secondary antibodies: anti-mouse Cy3, anti-mouse FITC, anti-mouse Cy5, anti-guinea pig Cy5, anti-guinea pig FITC, anti-rat Cy5, anti-chicken Cy3 (Jackson ImmunoResearch Laboratories) antibodies.Confocal images were obtained by Zeiss LSM880 and processed using ZEN 2.3, ImageJ 1.52a and Adobe Photoshop CC 2019.Image processingDistance between pairs of neurons were measured by focusing on axons on the surface of the medulla layer visualized by Ncad staining using Straight Line and Measure tools of ImageJ according to the scale bar provided by ZEN (Fig. 1l). When the distance was >10 μm, Segmented Line tool was used to measure the distance along the surface of the medulla layer.Signal intensity was quantified within the indicated rectangle areas by ImageJ (Figs. 2d–k and 3a, c). In Supplementary Fig. 1l–n, a circle that encompass the Dpn-positive medulla NB area was drawn using Segmented Line tool starting from the posterior end of the brain in a counter clockwise manner. Signal intensity of mRNA was measured along the circle. Removing background signals by subtracting 100, the number of mRNA expressing domain and the relative size of each expression domain were quantified for each brain sample. 100% indicates that mRNA is expressed in all NBs on the surface of the brain as found in control in situ RT-PCR (see Supplementary Fig. 1a).Column morphology was classified into normal and abnormal columns (Fig. 5f). Normal columns show regular donut-like morphology. Abnormal columns were further clasified into unclear, irregular, and fused columns (Fig. 5f). Unclear columns do not show clear shape, while donut-like shape is distorted in irregular columns. When adjacent columns are connected, these columns are regarded as fused columns. According to this classification, column morphology was quantified in brains containing control and Dscam1 mutant clones (Fig. 5g).Spatial distribution of neuronal cell bodies was quantified by ImageJ as follows (Fig. 5h–k). Subtract Background and Threshold to reduce background noises below the threshold level. Fill holes, Convert to Mask and Watershed to separate neurons that are close to each other. Analyze Particles (size = 20-Infinity) to remove garbage and extract positions of cells in the X–Y coordinates44.Statistics and reproducibilityFor quantification and statistical analysis, distinct brain samples were measured and analyzed as indicated in the text. Image intensities were not artificially processed except as otherwise noted. When statistics were not applicable, experiments were independently repeated at least three times with similar results.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Developmental neurogenesis",
"Neural stem cells",
"Axon and dendritic guidance",
"Cell fate and cell lineage"
] | IntroductionColumns higher functional units brain neurons form columnar units brain types columnar units described in cerebral cortex cortical columns similar response selectivity microcolumns cell type-specific clusters found in cortical regions radial unit hypothesis proposed column formation in mammalian cerebral cortex columns formed by clonally related neurons from common progenitor neurons radial unit form column similar selectivity sister neurons undergo dispersion sparsely distributed mixed with progenitors organization columns developmental mechanisms columnar unit formation significance neuronal lineage elusive fly visual system shows columnar organizations ommatidia retina cartridges lamina columns in Photoreceptor neurons R1–8 form ommatidium retina R1–6 lamina neurons L1–5 form cartridge in lamina medulla largest component each column contains 100 neurons Medulla neurons form columnar units in medulla medulla neurons project neurites toward medulla neuropil (Fig. axons dendrites form repetitive columnar units in medulla neuropil 1Linage dependent repulsion in larval medullaSchematics Drosophila medulla L3 stage Lateral dorsal views dotted line columns medulla layer dotted lines NB neuron layers Schematics proneural wave temporal transcription factors Neurons lineage visualized elav-Gal4 MARCM clones Dpn Ncad visualize NBs neuropil n = 26 21 medulla structures Ncad M0 medulla layers lateral view medulla columns anterior view Neurons lineage visualized drf-Gal4 twin-spot MARCM clones Ncad Sister neurons reroute M0 layer innervate columns medulla layer Arrows arborizations medulla layer cell bodies histogram distance neurons medulla layer = 0 μm fused indistinguichable pairs box plot distance 110 pairs neurons 38 clones samples cases 0 μm distance not included Center line median box limits upper lower quartiles whiskers 1.5× interquartile range Median average 5.78 8.50 μm Source data file Projections medulla neurons early L3 larval stage elav-Gal4 MARCM clones Ncad L3 larva M0 layer = Neurites defasciculated M0 layer32–48 h L3 larva M0 medulla layers (n = 49). Neurites defasciculated M0 layers innervate medulla d–f h–k n o Lateral views neuron layer (b). Anterior view columns dotted line (a). Scale bars 20 μm (d f n o 5 μm (e g previous study R7 R8 Mi1 core neurons concentrically arranged medulla N-cadherin-dependent differential developmental sequence medulla similar cerebral cortex single neuroblast produces radially oriented clonally related neurons analogous radial unit migrates radial fiber cerebral cortex (Fig. 1a development ommatidia lamina cartridges accompany NB lineage-dependent development differentiation proneural wave sweeps neuroepithelial cells larval optic lobe NEs sequentially differentiated medulla NBs (Fig. 1b NBs arranged birth order produce different types neurons brain temporal transcription factors Homothorax Klumpfuss Eyeless expressed NBs specify birth order-dependent production neurons (Fig. NB daughter neurons form radially arranged cluster stage (Figbirth order medulla neurons correlates with concentric gene expression Hth-positive NBs produce Hth expression induces Brain-homeobox (Bsh) expression differentiated NBs produce Hth/Bsh double-positive Mi1 neurons inner concentric domain medulla cortex. 1a Klu- Ey-positive NBs produce Runt- Drifter (Drf)-positive neurons concentric domains outside Hth/Bsh Similar patterning neural stem cells neurogenesis in developing cerebral stage medulla neurons same lineage arranged cluster dispersed in pupal stage clustered beyond 24 h after puparium formation similar to sparse distribution sister neurons mature cerebral columns not formed radial unit neuronal lineage fly medulla mammalian cerebral cortex sister neurons lineage project to different columns medulla repel each other lineage-dependent repulsion potential candidate molecule on Down syndrome cell adhesion molecule (Dscam), contributing Down.Drosophila Dscam1 gene has three alternative exons Ig2 Ig3 Ig7 domains 12 48 33 splice variantsDscam1 encodes 19,008 ectodomains22 Homophilic binding occurs between identical isoforms Ig domains produces repulsive neurons expressing same isoforms show repellent interaction splicing Dscam1 in each cell diversity of isoforms necessary for development neural Expression same isoform in single cell causes self-avoidance important for dendritic wiring26 Dscam1 temporally expressed in NBs inherited by neurons same lineage lineage-dependent repulsion expressed Hth temporal transcription factor Expression radial unit essential for lineage-dependent repulsion findings suggest function Dscam1 in lineage-dependent repulsion link between temporal patterning neuronal lineage column formation-dependent repulsion in medullaNBs outermost larval medulla primordium produce neurons toward medulla cortex radial orientation (Fig. 1a Daughter neurons same NB linearly arranged in larval brain forming radial unit until tangential dispersion between 12 and 24 h APF11 neurons same lineage project axons multiple columnslate 3rd stage (L3) developing neuropil visualized Ncad antibody contains two layers (Fig. 1e medulla layer contributes to adult medulla layers M1–M10 other layer outside temporal disappears pupal structure M0 layer (Fig. 1a e medulla columns observed frontal view (Fig. 1a distance between neighboring columns ~5 μm neurites radial unit extend 50 μm (Fig. 1e axons change directions M0 layer project to medulla layer (Fig. distinguish used twin-spot MARCM technique drf-Gal4 visualizes smaller number neurons (Fig. 1h–k axons change direction M0 project to medulla layer other brain region projection patterns Tm-type neurons11 sister neurons same NB form projections same columns rerouted in M0 layer different columns (Fig. 1h–j). medulla cortex axons same radial unit project toward M0 layer defasciculated M0 project to different columns quantified distance between sister neurons same radial unit brain samples small clones (Fig. 1l axons fused distance 0 μmdistance between axon shafts medulla measured (Fig. 1l). Ncad staining distance columns 5 μm in medulla (Fig. 1g). 110 pairs neurons 11 37 0 μm 2–5 μm distant 62 projected medula >5 μm (Fig. 1l). 2–5 μm distance same column may project adjacent domains distance between axon pairs Fig. 1m. Median average distance 5.78 8.50 μm sister neurons repel each other repulsion lineage lineage-dependent repulsion anterior part medulla neurons same lineage projected to distinct parts medulla (n = 27/32; Fig posterior part terminals neurons indistinguishable (n = 8/24; Fig. 1k). 11 pairs fused axons 9 in posterior part medulla (Fig. lineage-dependent repulsion less prominent posterior medulla anterior part lineage-dependent repulsion early in development examined 0–32 h 32–48 h L3 brains (Fig. 1n 0–32 h L3 brains one Ncad-positive layer likely M0 layer axons project medulla layer in 32–48 h L3 brainsaxons same lineage defasciculated M0 layer 0–32 h L3 stage lineage-dependent repulsion before formation medulla layer (Fig. daughter neurons same NB remember mother NB repeal other according lineage Dscam1 exhibits splice variants (Fig. Identical Dscam1 isoforms bind repulsive signal (Fig. Self-avoidance controlled by same Dscam1 isoform similar mechanism lineage-dependent repulsion medulla column repulsion between neurons same NB splicing diversity Dscam1 stochastically each NB expresses single Dscam1 variant inherited by daughter neurons daughter neurons same NB express same Dscam1 variant repel each other projection different medulla columns (Fig. 2c). neurons different lineages different variants repel project same column.Fig. 2Detecting transcription Dscam1 NB neurons in situ RT-PCR Dscam1 gene structure alternative splicing Homophilic binding identical Dscam1 isoform causes repulsion lineage repulsion between neurons same NB Dscam1 isoform Control PCR Lsc) Dpn) surface brain NB layer n = 22 Fig.signal intensity dotted box (d). Control PCR Lsc Dpn dorsal view decrease mRNA signal older NBs n = 16 Fig. g h j Lateral views neuron layer Control PCR Dpn n = 47 Intron PCR Dpn n = 13. signal intensity boxes (g Background signal subtracted Dpn jDscam1 9.1 PCR Dpn lateral view neuron layer n = 10 signal intensity boxes (j). Each box one NB Intensities plotted lines Scale bars 20 μm Source data file (e k).Transcription Dscam1 occurs NEs transcription pattern NBs neurons examined low levels mRNA reverse-transcribed cDNA PCR-amplified RT-PCR Dscam1 mRNA. 2a control primers fragment exons 8–10 shared Dscam1 isoforms larval development NEs become NBs medial-to-lateral orientation medulla proneural wave (Fig. 1b). Lsc expressed 1–2 NE cells wavefront15 Dpn expressed all NBsmRNA signals in NEs NBs following proneural wave decreased in older NBs (Fig. Dscam1 temporally transcribed in newborn NBs signals decreased neurons older brain (Fig. 2g strong Dscam1 mRNA signals form circle brain hemisphere temporally transcribed in newborn medulla NBs inherited to daughter neurons detected pre-mRNA Dscam1 primer between exons 9.33 10 (Fig. 2a). intron PCR signals restricted to NBs brain hemisphere (Fig. 2h signals neurons hardly detectable with control RT-PCR results (Fig. suggest Dscam1 mRNA transcribed in NEs newborn NBs proneural wave inherited to daughter neurons RT-PCR examined Dscam1 mRNA signal in clones homozygous for Dscam20 null mutant Dscam1 Fig. 1b). signals for Dscam1 protein mRNA-PCR abolished situ RT-PCR detects Dscam1 mRNA visualized Ncad mRNA in situ RT-PCRNcad protein expression medulla neurons observed strong Ncad mRNA signal inner medulla cortex uniform Ncad mRNA signal suggests decrease Dscam1 mRNA signal recapitulates Dscam1 expression (Fig. 2g–i).Neurons same lineage express similar Dscam1 test Dscam1 isoform inherited daughter neurons NB performed situ RT-PCR exons 4 6 9 (Fig. 2a Fig 1f–k). one splice variant stochastically chosen from exon 424. same exons 6 9 limited medulla NBs express same exon variant inherited by daughter neurons observed cluster NBs daughter neurons same variant exons 4 6 9 (Fig. 2j k 1f–n). location 9.1-positive NB cluster variable each brain sample domains exon variant repeated experiment 22 splice variants exon 4 6 9 ten samples observed each variant same results Fig. 1l–n suggesting alternative splice variant stochastically chosen stochastic choise occurs in NBs salt-and-pepper-like pattern lack situ RT-PCR signals expression domain Fig 1i). cell cycle changes mRNA distributionDscam1 expression initiated in NEs (Fig. divide symmetrically group NBs share same exon variant unknown mechanisms bias on choise alternative exons alternative splicing of exons 4 6 9) independently determined previous cluster NBs same variant exon 9 contains different variants subdivided by selection exons 4 6. NB lineages could share same splice variants expressed single isoform Dscam1 in neurons generating clones (dscam3.31.8 2a–e). neurons normal radial arrangement neurites defasciculated in M0 layer medulla neuropil wild-type control clones wild-type exon 9.8 stochastically chosen upon alternative splicing single-isoform mutant dscam3.31.8 lacks variants exon 9 except 9.8. expression pattern exon 9.8 in single variant mutant background mRNA for exon 9.8 uniformly detected in medulla NBs circle larval brain hemisphere Fig 2f exon 9.8 chosen other variants mRNA for exons 9.1 9.4 not detected in same mutant backgroundfindings support hypothesis neurons same lineage express similar Dscam1 isoforms.Dscam1 protein stabilized medulla examined expression pattern NBs neurons weakly expressed in Lsc-positive NEs Dpn-positive NBs decreases older NBs (Fig. 3a b protein temporally expressed NB differentiation rapidly downregulated older NBs.Fig. 3Expression pattern Dscam1 protein pattern (white NEs NBs Fig Lsc Dpn (blue). d Expression pattern Dscam1 protein (white magenta neurons Arrows indicate Dscam1 signals axons GFP expressing neurons f Quantification Dscam1 signal intensity boxes (a) (c), Dscam1 (white) Ncad (blue) visualize columnar structures medulla layer Fig. Scale bars indicate 20 μm Source data provided file contrast mRNA distribution Dscam1 protein accumulated along neural fibers radially oriented medulla cortex neurites projecting (Fig. 3c d Dscam1 signals NBs neurons eliminated in Dscam1 null mutant clones signals reflect expression patterns Dscam1Dscam1 transcribed in newborn NBs mRNA inherited daughter neurons degraded Dscam1 protein translated NBs neurons stabilized neurons localizes neurites (Fig. temporal restriction Dscam1 transcription essential for lineage-dependent repulsion alternative splicing spliceosome machinery assembles splice sites selection single splice duration transcription restricted small splice variants selected NB single splice isoforms shared among daughter neurons group neurons identical Dscam1 isoform recognition causes mutual repulsion lineage-dependent repulsion (Fig. 2b strong Dscam1 signals in M0 layer. 3c regulates spreading neurites medulla. 1e columnar distribution Dscam1 medulla overlaps with Ncad suggests role in column formation (Fig. 3g).Hth regulates Dscam1 expression in Hth Ey inherited by daughter neurons larval Hth first temporal transcription factor expressed in NEs NBs activates expression Bsh Ncad early born medulla neurons single type medulla neuron Mi1 (Fig. 1ccompared expression Dscam1 Hth Bsh Ncad Hth coexpressed in NEs newborn NBs (Fig. Bsh Ncad expressed neurons not NEs Dscam1 signals detected neurons inner developing medulla (Fig. 3c transcriptional regulator Engrailed regulates expression guidance receptor Frazzled NBs guidance Drosophila embryonic. 4Hth regulates expression Dscam1 views L3 larval brains NB b d neuron layers e. Dscam1 magenta d expression overlaps Hth NEs Hth Dpn NBs Background level Dscam1 signal hth mutant clones GFP control cells Pixel intensity Dscam1 enhanced image Ectopic Dscam1 upregulation clones expressing hth GFP control Ay-Gal4 Ncad areas dotted boxes enlarged panels Scale bars indicate 20 μm Quantification signal intensity boxes (b–e). plotted dotted linesDscam1 signal reduced GFP-negative hth mutant clones (b Dscam1 enhanced GFP-positive hth clones (d Background signal subtracted for Dscam1 (f Experiment repeated three times similar results Source data file (e g h Hth temporal transcription factor overlaps Dscam1 activates Bsh Ncad regulate Dscam1 generated hth mutant clones Dscam1 expression medulla NEs NBs eliminated hth mutant clones (17 4b Dscam1 signals medulla neurons reduced (Fig 4c residual Dscam1 signals due background Dscam1 antibody similar signals Dscam1 null mutant clones strong Dscam1 signals eliminated in hth mutant clones (10/41 3c d hth mutant (hthP2) commonly used allele hypomorphic incomplete loss Dscam1 expression hypomorphic nature additional unknown factors test hth expression Dscam1 expression generated clones ectopically expressing hth 4d ectopic hth expression upregulated Dscam1 (11/21Dscam1 expression upregulated neurons localized ectopic hth expression Fig. 4e upregulation Ncad clones ectopic hth causes premature neuron differentiation upregulate Dscam1 Dpn-positive NBs show Dscam1 expression brain 4) Hth Dscam1 expression NEs NBs Hth expressed NEs Dscam1 expression NEs (Figs. 2d–f Hth expression not induce Dscam1 expression unknown factor full induction Dscam1 expression Dscam1 causes loss neural defasciculationDscam1 protein expressed NBs daughter neurons neurites sister neurons NB rerouted M0 layer project distinct columns (Fig. 1e sister neurons express Dscam1 isoforms repulsion between neurons lineage compared projection patterns neurites radial units control Dscam1 null mutant clones Fig. 5Loss Dscam1 impairs lineage-dependent repulsion same lineage visualized elav-Gal4 MARCM clones Ncad visualizes neuropil structure columnsProjection patterns neurons lineage control Dscam1 mutant clones views L3 larval brains Fig. tangential projections M0 layer clones suppressed Dscam1 clones neurons innervate medulla M0 distance neurites lineage Control 8. Dscam1 10 median limits 1.5× interquartile Average projection distances 49.2 17.45 μm SD 14.8 8.09 P = 0.00043) Column morphology 48 h APF pupal medulla M1–2 layers Control Dscam1 mutant clones column disrupted Dscam1 mutant clones Classification column abnormal columns irregular fused unclear column morphology Control 87 columns 3 brains Dscam1 mutant clones 136 columns 3 brains normal 10−47 unclear 7.23 10−19 irregular 7.18 10−12 fused 0.00046) Tangential migration neuronal cell bodies lineage 24 h APF Dorsal views control Dscam1 clones Spatial distribution cell bodies medulla cortextangential distribution control SD 42 μm n = 3) suppressed Dscam1 mutant clones (j SD 24 μm n = 3 two-sided test P = 0.0022) optic lobe 24 h APF Scale bars indicate 20 μm (a b d e h j 5 μm (f). repeated three times similar results Source data file (c g control clones neurites defasciculated projected remote columns (distance 50 μm n = 8 Fig 5a measured largest distance neurons greater than average distance Dscam1 mutant neurites bundled M0 layer projected columns (distance 20 μm n = 10 Dscam1 lineage-dependent repulsion formation examined changes columnar structure antibody 48h APF control Dscam1 mutant clones classified column normal unclear irregular fused quantified column morphology Abnormal columns increased mutant clones shape columns arrangement affected medulla Dscam1 mutant radial unitsnonautonomous columnar defects by Dscam1 mutant clones suggest axons project to columns under Dscam1-dependent repulsion Dscam1 essential for lineage-dependent repulsion column formation neurites radial unit repel to remote columns early stage dispersion cell occurs between 12 and 24 h APF11 Dscam1 mutant radial units compared distributions cell in control Dscam1 mutant clones control clones cell bodies distributed medulla cortex tangential dispersion unit lacked Dscam1 function cell bodies remained close forming cluster at 24 h APF 24 dispersion depends on Dscam1.Loss of Dscam1 diversity leads to columnar diversity of Dscam1 isoforms critical for neuronal wiring Dscam1 diversity crucial for column formation 6Loss of Dscam1 isoform diversity leads to defects in column formation representation of Dscam1 gene in wild type single-isoform mutant backgroundsColumn morphology 48 h APF pupal medulla visualized Ncad M1–2 M9–10 layers control Dscam1 mutant backgrounds Dscam123 Scale bars 20 μm repeated three times similar results Schematic representation optic lobe 48 h APF Quantification column top layer Control 74 columns 5 brains Single-isoform mutant 143 columns 5 brains normal P 2.11 × 10−40 fused P 5.4 × 10−17 irregular P 0.33 unclear P 0.21) Source data combining single-isoform mutants Dscam3.31.8 Dscam23 generated mutant background one Dscam1 isoform. compared column shape arrangement Ncad antibody 48 h APF control brains regular arrangement donut-like columns shape disorganized Dscam1 single-isoform backgrounds defects top layer quantified tested bottom layer disorganized column shape unidentifiable quantification not applicable bottom layer (n = 5) dscam1 temporally transcribed NEs newborn NBs factor mRNA protein inherited neurons lineage-dependent20,000 isoforms Dscam1 small number stochastically selected Dscam1 temporally expressed in NBs neurons same lineage share same isoforms causes homophilic binding repulsion neurons repel contribute different medulla columns (Fig. 7Inheritance Dscam1 mRNA protein lineage repulsion.Schematics Dscam1 transcription in NEs NBs control Hth proneural wave Graded distribution Dscam1 mRNA protein neurons lineage controls lineage-dependent repulsion M0 layer innervation different columns results explain issues sister neurons project distant regions medulla ventral dorsal halves medulla (Fig. repulsive action Dscam1 triggered homophilic interaction projections Dscam1 function chemorepulsive molecules guidance molecules distinct subdomains medulla work with Dscam1 Slit Netrin repulsive guidance molecules optic lobe Optix expressed dorsal ventral subdomains regulate region specific expression guidance molecules same Dscam1 isoform radial unit neurites bundled medulla cortex prior defasciculation (Fig. 1)previous study showed low Dscam1 produce adhesive high repulsive function Dscam1 may adhesive to repulsive depending expression level Netrin Dscam1 localized in M0 than medulla cortex (Fig. mechanism may involved waves differentiation observed during visual systems in flies to radial unit found during column formation mammalian cerebral neurons same lineage dispersed later in development fly medulla5,6 mechanism combining wave differentiation temporal expression guidance molecule column might be evolutionarily conserved strategy column formation fly to mammalian splice diversity Dscam1 regulates column formation through lineage-dependent repulsion Similar mechanisms in mammalian adaptive immune systems olfactory systems immunoglobulin T cell receptor) genes contain multiple gene segments selected generate variable molecules antigens37 irreversible recombination alternative splicing Dscam1 flexible reversible mouse olfactory system each neuron expresses one receptor gene out of ~1000 OR single OR gene selected by cis-acting regulatory element multiple genes negative feedback mechanism inactivates expression other OR genes after one selectedexpression OR genes regulated single Dscam1 isoform selected splice variant temporally restricted transcription no mechanism expression isoforms lineage-dependent repulsion controlled mechanism simple assume multiple splice isoforms temporally restricted transcription stem cell progenitor cells determine similar mechanisms other systems column formation mammalian brains maintained Drosophila medium 25 °C mutant transgenic flies used dscam2039.27.2532 hthp230 UAS-hth1-1241 elav-Gal4 hs-flp FRT42D FRT82B ubiGFP tub-Gal80 UASCD8GFP42 Ay-Gal443.Clonal analysisNeurons same lineage visualized crossing hs-flp elav-Gal4-CD8GFP tub-Gal80 FRT42D FRT42D 34 °C 30 min heat shock (Figs. 1d e n 5 Fig. neurons same lineage crossing hs-flp UAS-CD2RFP UAS-GFP-Mir FRT40A-CD8GFP drf-Gal4 34 °C 30 min heat shockDscam1 null mutant clones crossing hs-flp ubi-GFP FRT42D dscam20FRT42D 37 °C 60 min heat shock (Fig. 3 1b hth clones-flp ubiGFP FRT82B hthP2FRT82B 37 °C 60 min heat shock (Fig. 4b overexpression clones crossing hs-flp Ay-Gal4 UAS-GFP UAS-hth1-12 37 °C 60 min heat shock (Fig. 4d e Fig 4) situ RT Larval brains dissected PBS transferred ice cold 4%formaldehyde/PBS solution fixed 4 °C overnight washed μl –20 °C methanol overnight washed μl 100% ethanol incubated 50% xylene/ethanol 30 min replaced with 75 50 25% ethanol/H2O solutions H2O ice H2O replaced μl –20 °C 80% acetone/H2O solution incubated 10 min ice acetone replaced H2O brains washed ~800 μl PTw (0.1% Tween20 PBS) twicePTw replaced 4% formaldehyde/PBS brains room temperature 30 min washed μl PTw transferred PCR tube brains incubated 65 °C 5 min 0.34 mM 3′ transcription primer 1mM dNTP cooled 6 μl 5× PrimeScript Buffer 32U RNase inhibitor 200U PrimeScript Reverse Transcriptase H2O transcription 30 °C 10 42 °C 30–60 70 °C 15 min cooled washed H2O 50 μl PCR 1× KOD Buffer 0.2mM dNTP 2.5U KOD DNA polymerase 1 μM 5′ primer 3′ primer 0.02 mM digoxigenin-11-dUTP denaturation 1 min 95 °C 20 cycles 98 °C 5 s annealing 55 °C 50 s elongation 68 °C 5min 72 °C brains washed PBT blocked 5–10% normal serum/PBT 30–60 min Primary antibody reaction mouse anti-Dig antibody 1% normal serum PBT 4 °C overnight washed four timesantibody reaction solution anti-mouse FITC secondary antibody (1 antibodies 1% normal serum PBT 4 °C overnight brains washed PBT three times PBT replaced with PBS brains VECTASHIELD transcription Dscam1 RT11 in situ PCR exons 8–10 primers Ctrl8_5 (GCTGATTATCGAGAATGTGGAA Ctrl10_3 (TTCTTCCATGTAACTTGGGGTTT).In situ PCR Dscam1 exon 9 primers Ctrl11_3 Ds9.1 Ds9.2 Ds9.3 Ds9.4 Ds9.5 Ds9.6 Ds9.7 Ds9.(ATCCAAGCGTTTGACTTTGG).In situ PCR Dscam1 pre-mRNA primers Int1_5 (ACCGCATCAGAAAACCAATC Int1_3 Int2_5 (TCATGCTCCAACACCGAATA Int2_3 PCR exon 4 primers Exon5R (CTCTCCAGAGGGCAATACCA Ds4.1 Ds4.2 Ds4.3 Ds4.4 PCR exon 6 primers Exon7R (TCCTCGACTACTGCGTCCTT Ds6.4 (TACGCTCCTTTGTCCAGCTC), Ds6.8 Ds6.12 Ds6.16 Ds6.20 Ds6.24 Ds6.28 Ds6.32 Ds6.36 Ds6.40 (GTCGATTAAGGCCAGCTTTG).Reverse transcription Ncad primer RT4R (GAATTGGGTCCATTGCTGTT). situ PCR Ncad primers Ncad_E2 (GTATCGAAGGCAATCCCACA Ncad_E3 (TTTGGAAATGTGCCATCCTT).HistochemistryImmunohistochemistryLarval brains dissected PBS fixed 4% formaldehyde/PBT solution room temperature 30–60 min washed PBT blocked 5–10% normal serum/PBT solution 30 min Primary antibody reaction antibodies 1% normal serum PBT 4 °C overnight brains washed PBT Secondary antibody reaction secondary antibodies (1:200) 1% normal serum PBT 4 °C overnight brains washed PBT VECTASHIELD.Primary antibodies rabbit anti-Hth (1:1000 rat anti-Dpn (1 guinea pig anti-Lsc (1 rat anti-Ncad mouse anti-Dscam1 (1:200 mouse anti-Dig (1 rabbit anti-GFP Alexa488 (1:1000 Invitrogen antibodies Secondary antibodies anti-mouse Cy3 FITC Cy5-guinea pig-chicken Cy3 ImmunoResearch Laboratories.Confocal images Zeiss LSM880 processed ZEN 2.3 ImageJ 1.52a Adobe Photoshop CC between neurons measured axons medulla layer Straight Line Measure ImageJ distance >10 μm Segmented Line tool distanceSignal intensity quantified rectangle ImageJ (Figs. 2d–k 3a Supplementary Fig. 1l–n circle Dpn-positive medulla NB area drawn Segmented Line tool posterior end brain counter clockwise Signal intensity mRNA measured background signals subtracting 100 mRNA expressing domain size quantified brain sample 100% mRNA expressed all NBs brain RT-PCR Supplementary Fig. 1a).Column classified normal abnormal columns (Fig. Normal columns regular donut-like Abnormal columns unclear irregular fused Unclear columns donut-like shape distorted irregular columns columns connected fused columns column morphology quantified brains control Dscam1 mutant clones (Fig. 5g).Spatial distribution neuronal cell bodies quantified ImageJ (Fig. 5h–k). Subtract Background Threshold reduce background noises Fill holes Convert to Mask Watershed separate neurons Analyze Particles (size 20-Infinity remove garbage extract positions cells X–Y brain samples measured analyzed Image intensities not artificially processed experiments repeated three times similar resultsReporting research design Nature Research Reporting Summary article.Supplementary Review FileReporting Summary | 49.1 | 0.877237 |
10.1038/s41467-020-14627-z | PMC7029035 | Water confined in natural or synthetic hydrophobic nano-spaces behaves differently than in the bulk. Here the authors investigate water in hydrophobic synthetic 1D nanochannels revealing water clustering in tetramers and octamers and high proton conductivity, along with a continuous liquid to solid transition. | Water confined within one-dimensional (1D) hydrophobic nanochannels has attracted significant interest due to its unusual structure and dynamic properties. As a representative system, water-filled carbon nanotubes (CNTs) are generally studied, but direct observation of the crystal structure and proton transport is difficult for CNTs due to their poor crystallinity and high electron conduction. Here, we report the direct observation of a unique water-cluster structure and high proton conduction realized in a metal-organic nanotube, [Pt(dach)(bpy)Br]4(SO4)4·32H2O (dach: (1R, 2R)-(–)-1,2-diaminocyclohexane; bpy: 4,4’-bipyridine). In the crystalline state, a hydrogen-bonded ice nanotube composed of water tetramers and octamers is found within the hydrophobic nanochannel. Single-crystal impedance measurements along the channel direction reveal a high proton conduction of 10−2 Scm−1. Moreover, fast proton diffusion and continuous liquid-to-solid transition are confirmed using solid-state 1H-NMR measurements. Our study provides valuable insight into the structural and dynamical properties of confined water within 1D hydrophobic nanochannels. | IntroductionWater confined within a hydrophobic nanospace has been of particular interest for researchers due to the unusual structural and dynamical properties, which differ significantly from those of bulk water1,2. In a hydrophobic nanochannel with a small aperture size (<3−4 nm), confined water shows remarkable physical properties which arise from the reduced number of neighbouring water molecules, resulting in a restricted hydrogen-bond network2,3. To study the effect of confining water in a one-dimensional (1D) nanospace, water-filled carbon nanotubes (CNTs) have been an attractive research focus, with interesting water-cluster formation4,5, fast water transport6,7, and high proton conductivity observed8–10. Such studies on the dynamics of water confinement in hydrophobic nanochannels are especially important as biomimetics11,12 for understanding transport mechanisms in proton pumps13 or water transport proteins14. However, the poor crystallinity and high electrical conductivity of CNTs cause difficulties for detailed investigation of the structure and proton transport properties of confined water within the 1D hydrophobic nanochannels.In contrast to CNTs, structural fabrication of nanochannels using coordination chemistry is a promising approach due to the richness in structural degrees of freedom and the relative ease of structural characterisation through crystallographic techniques. Among coordination complexes, metal-organic frameworks (MOFs), which are infinite coordination networks of metal ions bridged through organic ligands, have attracted much interest due to their structural designability and uniform porosity15–17. Recently, we have developed a rational bottom-up synthesis of a metal-organic nanotube18 based on the oxidative polymerisation of a square-shaped platinum complex using elemental iodine. This nanotube is composed of four 1D halogen-bridged transition-metal chains (MX-chains) which are connected linearly bridging organic ligands to form a right square prism-shaped nanotubular structure (MX-tube)19.Herein, we present the direct observation of the proton dynamics of water confined in the hydrophobic channel of a synthetic nanotubular material20,21 (Fig. 1a). We synthesise the metal-organic nanotube [Pt(dach)(bpy)Br]4(SO4)4·32H2O (1, dach: (1R, 2R)-(-)-1,2-diaminocyclohexane; bpy: 4,4′-bipyridine), and elucidate unique water clustering composed of water tetramers and octamers within the hydrophobic nanochannel of 1 using single-crystal X-ray analyses. Single-crystal impedance spectroscopy measurements and solid-state 1H-NMR studies, both supported by theoretical calculations, are used to show high proton conductivity of 1 that originates from the confined water within the hydrophobic nanochannel. In addition, we find a continuous liquid-to-solid transition of the confined water, which is significantly different from that of bulk water.Fig. 1Proton-conducting hydrophobic nanochannel.a Schematic illustration of a water-filled hydrophobic nanochannel. b The four-legged tubular structure of 1 viewed along the a-axis (100 K). Counter anions and water molecules have been omitted for clarity. c The packing structure of 1 in the bc plane. Channel A and channel B are highlighted by light blue and light green circles, respectively. Water molecules are omitted for clarity. Platinum, bromine, sulfur, carbon and nitrogen are represented by orange, brown, yellow, grey and blue spheres, respectively.ResultsConfined water in the nanochannel of a metal-organic nanotubeSingle crystals of 1 were obtained by the oxidative polymerisation of a square-shaped complex, [Pt(dach)(bpy)]4(SO4)4, using elemental Br2 (Supplementary Figs. 1–7, see Supplementary Information for details). The crystal structure of 1 was determined by single-crystal X-ray diffraction (SCXRD) at 100 K (Fig. 1b, c, Supplementary Figs. 8–16, and Supplementary Table 1). The structure consists of symmetrically equivalent nanotubes, with square [Pt(dach)(bpy)]4 units bridged to each other along through bromide anions at the corners of the square, resulting in tubes that propagate parallel to one another along the a-axis. Overall, there is a net 8+ charge per square [Pt(dach)(bpy)Br]4 unit, which is charge-balanced by four sulfate anions that reside between the tubes. The tubes are spaced 14.1 Å (17.0 Å) apart along the b- and c-direction and are propped apart due to steric interactions between the bulky exterior dach ligands and sulfate counter anions. This spacing leaves two types of water-filled 1D channels in the crystal, a hydrophobic interior channel A and a hydrophilic exterior channel B (Fig. 1c). Within channel B, water molecules form a 1D hydrogen-bond network which includes terminal dach amino groups and sulfate anions (Supplementary Fig. 9). Within channel A on the other hand, water forms two unique clusters of alternating octamers and tetramers, having strong hydrogen-bond interactions within the cluster and weaker interactions between clusters (Fig. 2). These tetramers and octamers are similar to theoretically-predicted conformations of small water clusters22. There are three crystallographically-unique water molecules in the octamer and two in the tetramer, and medium strength hydrogen bonds hold each cluster together at 2.5–2.7 Å O–O distances. The clusters form weak hydrogen bonds between one another in the channel direction at distances of 3.0–3.4 Å. Water in the clusters are separated from the walls of the tube by >2.9 Å, indicating weak contact forces between the clusters and the inner surface of the nanochannel. This clustering arises from the hydrophobic nature of the channel, which is similar to that of CNTs or biological channels in ion transport proteins6,13,14. In 1, the alternating M-X nature of the tube gives rise to a corrugated interior which narrows near the bpy surfaces, which is in contrast to the smooth interior surface of a CNT (Supplementary Fig. 12). Variable-temperature SCXRD was used to understand the dynamics of the water clusters with increasing temperature; water-clusters were clearly observed below 200 K, but became slightly disordered above 250 K (Supplementary Fig. 13).Fig. 2Water-cluster formation within channel A (100 K).The results of SCXRD are displayed. a ORTEP drawing (30% probability ellipsoids) of the 12 unique oxygen atoms in the nanotube. b Tetramer and octamer-like water clusters. The red dotted line represents the hydrogen bonds. c 1D hydrogen-bonding network of water clusters. Light blue dotted line represents the hydrogen bonds between water clusters.High proton conductivityTo investigate any unique transport properties arising from the nanoconfinement of water, impedance spectroscopy measurements on a single crystal of 1 were performed along the channel direction (a-direction; Fig. 3b inset). We first confirmed that the dc conductivity is less than the lower limit of measurement, ruling out significant electrical conductivity in the tube. On the other hand, when measured with ac impedance spectroscopy, 1 gave a semi-circular response in the Nyquist plot, suggesting that the conductivity is protonic in nature. Upon increasing the relative humidity (RH) from 40 to 95%, the conductivity increased significantly by over three orders of magnitude (Fig. 3b). This strong humidity dependence clearly suggests that the included water molecules are essential for the high level of conductivity. At 55 °C and 95% RH, the conductivity of 1 reached 1.7 × 10−2 S cm−1 (Fig. 3c). This value is quite high and in the range of superprotonic conductivity, which is comparable to the commercial proton-exchange polymer membrane Nafion23. This high level of conductivity means that 1 is among the most proton-conductive MOFs despite the fact that 1 does not have any free functional acidic groups or free acids inside the pores.24,25 The proton conductivity measured on a pelletised powder sample (Supplementary Figs. 17–20) was about a hundredth lower than that measured with a single crystal at 95% RH, suggesting that the proton conduction is highly anisotropic and that the conducting pathway is in the channel direction. The conductivity showed Arrhenius behaviour with varying temperature, with an activation energy (Ea) of 0.22 eV (Fig. 3d). This low Ea is close to those of typical proton-conductive MOFs showing Grotthuss-type proton hopping, where a protonic charge defect diffuses through the hydrogen-bond network17,26,27. Moreover, the water-cluster structures of 1 were retained above 200 K (~250 K, Supplementary Fig. 13), indicating that high proton conductivity is derived not from mobile hydronium ions (H3O+) themselves but rather from proton shuttling through the hydrogen-bond network between interacting water clusters.Fig. 3Humidity dependence of proton conductivity of 1.a Water vapour sorption isotherms (298 K). The blue open and closed circles denote adsorption and desorption processes, respectively. b Humidity dependence of conductivity: Inset shows the experimental setup. c Nyquist plots of 1 at elevated temperature under a relative humidity of 95%. d Arrhenius plots of the conductivity under a relative humidity of 95%.Proton-conducting mechanismFor further investigation of the transport mechanism of 1, solid-state 1H-NMR measurements were performed (Supplementary Figs. 21–27). In the 1H-MAS NMR spectra, the peak of the crystallization water was observed as a single component, indicating the rapid exchange of protons of water between channel A and B (Supplementary Fig. 22). The self-diffusion coefficient of 1H in a water molecule can be estimated by the pulsed-field gradient (PFG) method28–30, by which the diffusion movement is detected as the attenuation of an observed echo signal. Because the self-diffusion coefficient of 1H in water has a linear relationship with the mobility of the proton carrier in proton conduction, PFG-NMR measurements provide useful insight into the mechanism of conduction. Figure 4 shows the results of PFG-NMR measurements at 25 °C and 95% RH. The peak centred at 4.54 ppm was assigned as crystallization water protons and was observed to decay with the increasing strength of the gradient pulse (g) (Fig. 4a and Supplementary Figs. 23–27). The echo attenuation follows different equations depending on the dimensionality of the diffusion path (isotropic, anisotropic, or unidirectional; see Supplementary Materials for detail)29–31, and in our case, the curve fitting was successful when considering the single component of an anisotropic diffusion model (Fig. 4b) expressed as,Fig. 4Intensity decay of the echo of the crystallization water of 1.a Echo decay with selected g values at 25 °C, 95% RH. b Fit to the attenuation of the normalized signal intensity (E/E0) as a function of g2 based on the two-component anisotropic diffusion model (see text). Here, the gradient pulse duration (δ) is 1 ms, and the diffusion time (Δ) is 20 ms. γ is the gyromagnetic ratio of 1H. Grey dotted line, red solid line and black dashed line are the best fits to the experimental data (red circle) using isotropic, anisotropic and unidirectional diffusion model, respectively.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E/E_0 = \exp \left( { - kD_ \bot } \right)\int_0^1 {{\mathrm{exp}}} ( - k[D_{||} - D_ \bot ]x^2){\mathrm{d}}x$$\end{document}E∕E0=exp−kD⊥∫01exp(−k[D∣∣−D⊥]x2)dxwhere k = γ2δ2γ2(Δ−δ/3), and D is the self-diffusion coefficient. The subscripts of D represent the diffusion parallel (D||) and perpendicular (D⊥) to the channel alignment direction. The fitting result gave self-diffusion coefficients of D|| = 2.9 × 10−11 m2s−1 and D⊥ = 1.6 × 10−12 m2s−1. The diffusion time (Δ) dependence of D and mean-square displacement of 1H also supports the anisotropic nature of the diffusion (Supplementary Figs. 24 and 25). The self-diffusion coefficient of D|| is comparable to that reported for Nafion (0.2–20 × 10–10 m2s–1; D of Nafion largely varies with water mass fraction)32. From the temperature dependence of self-diffusion coefficients, the activation energies are estimated to be 0.18 eV for D|| and 0.14 eV for D⊥ (Supplementary Fig. 26). Compared to the impedance result, smaller activation energies were likely determined because PFG-NMR observes shorter-range diffusion. It should be noted that the observed D|| contains the contributions from both channel A and B. The very smooth proton diffusivity in the hydrophobic nanochannel A is expected because of weak interactions among the confined waters and the inner wall of the tube6–10, resulting in the lowering of the potential barrier heights of the rotational and translational movements of the included water molecules. Conversely, moderate diffusivity in channel B would be expected due to the low dimensionality of the diffusion path and the large number of hydrogen bonds among the water molecules, sulfate anions, and amines of the tubular framework, which restricts water reorientation and diffusion29. In addition, our recent study on related nanotubular analogues having ultrasmall diameters revealed that the observed proton conductivities are much lower than that of 133. In these analogue compounds, no water molecule exists inside the hydrophobic nanotube, whereas the hydrophilic channel outside of the nanotube contains several water molecules. These results suggest that the existence of confined waters within hydrophobic channel A with smooth proton diffusivity plays an important role in the realization of the high proton conductivity of 1.As the conductivity (σ) can be expressed as σ = neμ (n: the number of carriers, e: charge, μ: mobility of carriers), the existence of mobile proton sources is very important for attaining high proton conduction. In our case, the amine sites of the dach ligand that are coordinated to the PtIV sites could be considered as possible proton sources. Compared to the pKa value of free amines (pKa = 25–30), the pKa of amines coordinated to metal ions are known to be much lower34. DFT calculations clearly support this assumption and indicate that the amine groups coordinated to PtIV cations in 1 have similar acidity to carbonic acid (Supplementary Figs. 28 and 29 and Supplementary Table 3).To obtain atomistic insights on the proton diffusivity in the hydrophobic nanochannel, we performed quantum-mechanical molecular dynamics (QM-MD) simulations (Supplementary Figs. 30–35, Supplementary Tables 4–6 and Supplementary Movies 1–3; see Supplementary Materials for details). The vehicular diffusion35 coefficients (Dv) estimated from the time dependence of mean square displacement indicate that water molecules in channel A move approximately 1.5 times faster than those in channel B (Supplementary Fig. 35a). According to the decomposition into parallel and perpendicular contributions with respect to channel alignment direction, the vehicular diffusion is not completely isotropic in both channel A and B (Supplementary Table 7). The Grotthuss diffusion26,27 coefficients (DG) were calculated based on the proton transfer rate, which was quantified by counting proton shuttling at each simulation step36,37. The proton transfer rate in channel A was more than two times higher that of channel B because attractive interaction with the sulfate anion in channel B hindered the proton shuttling (Supplementary Fig. 35b). The averaged value of DG obtained from channel A and B accounts for the larger portion in overall proton diffusion coefficient (Dp) given by the summation of DG and Dv38–40. The comparison of Dp among three water environments showed that the proton diffusivity in the hydrophobic nanochannel is in between the liquid and solid state (Supplementary Table 8).Liquid-to-solid phase transition of confined waterThe liquid-to-solid phase transition behaviour of confined water has been of interest because the nature of the freezing of water in quasi-one-dimensional systems is not simple and has not been fully understood4,5,41,42. The experimental and theoretical investigations of CNTs have indicated that the phase transition behaviour strongly depends on the pore diameter and pressure; the transition behaviour has been predicted to be discontinuous (first-order-like) for a pore diameter of >1.2 nm and continuous for a pore diameter of <1.2 nm at ambient pressure43. We investigated the liquid-to-solid phase transition behaviour of confined water in 1 using differential scanning calorimetry (DSC), solid-state 1H-NMR and impedance measurements at low temperature (Supplementary Figs. 36–39). The DSC measurements showed no obvious phase transition (ice transition or melting) of the confined water between 163 and 293 K (Supplementary Fig. 36), suggesting that the confined water shows a continuous liquid-to-solid transition in the channels. The temperature dependence of the solid-state 1H-NMR signal line width for the crystallisation water of 1 showed a gradual and continuous broadening as the temperature decreased (Supplementary Figs. 37 and 39). These results indicate that the confined water in 1 exhibits no clear liquid-to-solid phase transition, which is markedly different from the behaviour of the bulk water or water confined in micropores5,43. This is also supported by the proton conductivity showing a gradual increase with temperature, rather than an inflection indicative of a phase transition (Supplementary Fig. 38). From the combination of solid-state 1H-NMR and proton conductivity, the correlation between the proton conductivity and the continuous liquid-to-solid phase transition is clearly visualized (Fig. 5).Fig. 5Temperature dependence of 1H NMR spectra and proton conductivity.Red squares and blue circles indicate FWHM of 1H NMR spectra (Supplementary Fig. 37) and proton conductivity (pelletised sample, Supplementary Fig. 38), respectively.DiscussionWe have synthesised and characterised a metal-organic nanotube that has water-filled hydrophobic nanochannels in the crystalline state. The hydrophobic nanochannel of the nanotube has the square-shaped aperture with the size of ca. 1.0 nm (Fig. 1). The confined water in the hydrophobic nanochannel of 1 forms unique clustered structures, which was confirmed by SCXRD measurements. Single crystal impedance spectroscopy revealed that 1 exhibits very high proton conductivity along the 1D channel direction. The high conductivity of 1 is attributed to both the high proton diffusivity within the hydrophobic nanochannel combined with the acidity of the coordinated amine protons, both of which were supported by PFG-NMR and theoretical calculations. In addition, the continuous liquid-to-solid phase transition of the confined water in 1 was observed from DSC, solid-state 1H-NMR and impedance measurements, showing markedly different behaviour from bulk water. In 2001, Brewer et al.8, pointed out that the properties of confined water and proton transport through that water in a hydrophobic nanochannel are highly sensitive to the size of the channel. These properties of confined water originate from the reduced number of possible hydrogen-bond interaction in the nanochannel and negligible water–channel wall interaction2,3, which become similar to those of bulk water as the diameter of the nanochannel become larger. Recent experimental results using CNTs also validated these theoretical predictions.2,5,44 For example, Noy et al., experimentally demonstrated that 0.8-nm-diameter CNT porins show extraordinarily fast proton transport while 1.5-nm-diameter CNT porins show the proton transport rate comparable to bulk water44. Consistent with these previous works on CNTs with the aperture sizes of <1.5 nm, our nanotube experimentally demonstrated the unique clustering structures, high proton mobility, and a continuous liquid-to-solid phase transition of the confined water. The present results show strong experimental evidence of the predicted behaviour of confined water in hydrophobic nanochannels. Nanotube fabrication based on coordination chemistry can allow for systematic structural tuning of pore size, shape and surface properties. This tunability can thus enable us to systematically investigate the properties of confined water. We believe that these findings will provide valuable structural and dynamic insights about confined molecular species in nanotubular materials as well as in biological channels.MethodsSyntheses and characterisation of compoundsReagents and solvents were purchased from Wako Pure Chemical Industries, Ltd., TCI Co., Ltd., and Sigma-Aldrich Chemical Co. and used without further purification. Elemental analyses for all compounds were performed using Yanaco MT-5 and MT-6 CHN recorders at the Centre for Organic Elemental Microanalysis, Kyoto University.Synthesis of [Pt(dach)(bpy)Br]4(SO4)4·32H2O (1)To an aqueous solution containing equimolar amounts of the square-shaped unit [Pt(dach)(bpy)]48+ and its brominated square unit [Pt(dach)(bpy)Br2]48+, an excess amount of tetrabutylammonium sulfate [(C4H9)4N]2(SO4) were added, and then the mixture was stirred at rt. A red-purple precipitate of 1 was collected by filtration (60% yield from the starting material (Pt(dach)(NO3)2)). Single crystals were obtained by slow tetrahydrofuran diffusion to an aqueous solution of the powder sample. Elemental analysis (%) calcd for C64H88N16O16S4Br4Pt4·28H2O: C 25.04, H 4.73, N 7.30, S 4.18; found: C 24.73, H 4.35, N 7.26 S 4.25. (IR spectrum and TGA curve are shown in Supplementary Figs. 4 and 5, respectively. See Supplementary Discussion for the experimental detail).Adsorption/desorption isothermsThe sorption experiments for H2O (298 K) were carried out using a BELSORP-max (MicrotracBEL). The as-synthesized sample 1 was dried under high vacuum (<10−1 Pa) at 35 °C for 2 days to remove the crystallisation water before measurements. Isotherms are shown in Fig. 2a.Single-crystal X-ray crystallographyAn X-ray crystal structure analyses were carried out using a Bruker SMART APEX II CCD detector with graphite-monochromated Mo Kα radiation (λ = 0.71073 Å) at 100 K. Single crystals were prepared by the vapour diffusion method using a water solution of 1 and THF, as described above. Suitable single-crystals were selected in the mother liquid and quickly transferred to paraton oil to avoid undesired solvent loss. The single crystals were mounted on MicroMesh (MiTeGen). The structures were solved by direct methods (SIR92), expanded using Fourier techniques (DIRDIF99) and refined by full-matrix least-squares refinement on F2 (SHELXL-97) using the CrystalStructure software package. The refinement result for 1 is summarized in Supplementary Table 1.Impedance measurementsImpedance measurements were carried out with a Solartron SI 1260 Impedance/Gain-Phase Analyser and 1296 Dielectric Interface in the frequency range of 1–1 × 106 Hz. The relative humidity and temperature were controlled by an Espec Corp. SH-221 incubator. The measurements in the low temperature region (140–300 K) were taken using Oxford OptistatDN2 and Lakeshore 340 temperature controllers.Solid-state 1H NMR measurementsSolid-state 1H nuclear magnetic resonance (NMR) measurements of 1 were performed on an AVANCE II+ 400 NMR spectrometer (Bruker Biospin K. K.) with an UltraShieldTM 400 WB 9.4 T superconducting magnet. 1H magic angle spinning (MAS) spectra and 13C cross-polarization (CP) MAS spectra were measured with a rotor of 4 mm diameter. Pulse field gradient (PFG)-NMR measurements were performed with a Diff 50 diffusion probe (Bruker Biospin K. K.).Supplementary information
Supplementary Information
Description of Additional Supplementary Files
Supplementary Data 1
Supplementary Data 2
Supplementary Data 3
Supplementary Movie 1
Supplementary Movie 2
Supplementary Movie 3 | nature communications | [
"Article"
] | [
"Reaction kinetics and dynamics",
"Nanoscale materials"
] | confined hydrophobic nanospace unusual structural dynamical properties from bulk hydrophobic nanochannel small aperture size (<3−4 confined water shows physical properties from reduced neighbouring water molecules restricted hydrogen-bond confining water water-filled carbon nanotubes (CNTs) attractive research focus water-cluster fast water high proton conductivity studies on dynamics water confinement important understanding transport mechanisms proton pumps13 water transport proteins14 poor crystallinity high electrical conductivity of CNTs cause difficulties for structure proton transport properties confined water 1D hydrophobic nanochannels contrast CNTs structural fabrication of nanochannels using coordination chemistry promising structural degrees freedom ease structural characterisation metal-organic frameworks networks metal ions organic ligands interest structural designability uniform developed synthesis metal-organic nanotube18 oxidative polymerisation square-shaped platinum complex using elemental iodine nanotube four 1D halogen-bridged transition-metal chains (MX connected organic ligands prism-shaped nanotubular structurepresent proton dynamics water hydrophobic channel synthetic nanotubular (Fig. synthesise metal-organic nanotube [Pt(dach)(bpy)Br]4(SO4)4·32H2O-diaminocyclohexane elucidate water clustering tetramers octamers hydrophobic nanochannel single-crystal X-ray analyses Single-crystal impedance spectroscopy measurements solid-state 1H-NMR studies show high proton conductivity water continuous liquid-to-solid transition different bulk water. 1Proton-conducting hydrophobic nanochannel water-filled four-legged tubular structure 1 a-axis (100 Counter anions water molecules omitted packing structure plane Channel A B highlighted light blue green circles Water molecules omitted Platinum bromine sulfur carbon nitrogen represented orange brown yellow grey blue spheres.ResultsConfined water nanochannel metal-organic nanotubeSingle crystals 1 obtained oxidative polymerisation-shaped complex [Pt(dach)(bpy)]4(SO4)4 elemental Br2 (Supplementary Figs. 1–7 crystal structure determined single-crystal X-ray diffraction) 100 K1b Figs. 8–16 Table 1) structure equivalent nanotubes square [Pt(dach)(bpy)]4 units bridged bromide anions corners tubes parallel a-axis net 8+ charge per square [Pt)Br]4 unit charge-balanced by four sulfate anions between tubes tubes spaced 14.1 Å (17.0 Å) apart b- c-direction steric interactions dach ligands sulfate anions spacing two water-filled 1D channels hydrophobic interior channel A hydrophilic exterior channel B (Fig. channel B water molecules form 1D hydrogen-bond network terminal dach amino groups sulfate anions Fig. 9) channel A water forms clusters octamers tetramers strong hydrogen-bond interactions weaker (Fig. 2) similar small water three-unique water molecules in octamer two tetramer medium strength hydrogen bonds hold cluster together 2.5–2.7 Å O–O distances clusters weak hydrogen bonds 3.0–3.4 Å Water clusters separated from walls tube >2.9 Å weak contact forces clustering hydrophobic nature channel similar CNTs1 alternating M-X tube corrugated interior narrows near bpy surfaces contrast smooth CNT Fig 12). Variable-temperature SCXRD dynamics water clusters increasing temperature water-clusters observed below 200 K disordered above 250 K Fig 13).Fig. 2Water-cluster formation channel A (100 results SCXRD ORTEP drawing 12 oxygen atoms nanotube Tetramer octamer-like water clusters red dotted line hydrogen bonds 1D hydrogen-bonding network water clusters blue dotted line hydrogen bonds.High proton impedance spectroscopy measurements single crystal 1 channel direction dc conductivity less than lower limit ruling out electrical conductivity ac impedance spectroscopy 1 semi-circular response Nyquist plot conductivity protonic increasing relative humidity 40 to 95% conductivity increased three orders magnitude (Fig. 3b). humidity water molecules essential high conductivity 55 °C 95% RH conductivity 1 1.7 × 10−2 S cm−1 (Fig 3c). high superprotonic conductivity comparable commercial proton-exchange polymer membrane Nafion23.high conductivity 1 proton-conductive MOFs free functional acidic groups acids pores proton conductivity pelletised powder sample 17–20) hundredth lower than single crystal at 95% RH conduction anisotropic conducting pathway in channel direction conductivity showed Arrhenius behaviour varying temperature activation energy (Ea) 0.22 eV (Fig. low Ea close to typical proton-conductive MOFs Grotthuss-type proton hopping protonic charge defect diffuses through hydrogen-bond water-cluster structures 1 retained above 200 K high proton conductivity not from hydronium ions proton shuttling through hydrogen-bond network water clusters. 3Humidity dependence proton conductivity Water vapour sorption isotherms (298 blue circles adsorption desorption processes Humidity conductivity Nyquist plots elevated temperature under humidity 95% Arrhenius under humidity 95%.Proton-conducting solid-state 1H-NMR measurements performed 21–27) peak crystallization water single component rapid exchange of protons between channel A and Bself-diffusion coefficient 1H in water molecule estimated by pulsed-field gradient (PFG) diffusion movement detected as attenuation echo signal-diffusion coefficient mobility proton carrier conduction PFG-NMR measurements conduction Figure 4 results PFG-NMR measurements at 25 °C 95% RH peak 4.54 ppm crystallization water protons decay with strength gradient pulse (Fig. 4a echo attenuation follows equations diffusion path (isotropic anisotropic unidirectional curve fitting successful anisotropic diffusion model (Fig. 4b 4Intensity decay echo crystallization water values at 25 °C 95% RH attenuation normalized signal intensity function g2 two-component anisotropic diffusion model gradient pulse duration (δ 1 ms diffusion time 20 ms γ gyromagnetic ratio of 1H Grey dotted line red solid line black dashed line best fits to experimental data isotropic anisotropic unidirectional diffusion model[12pt{amsmath\oddsidemargin-69pt$E/E_0\exp kD_0^1\mathrm{exp[D D\mathrm{document}E∕E0=exp−kD⊥∫01exp(−k k = γ2δ2γ2(Δ−δ/3), D self-diffusion coefficient subscripts D represent diffusion parallel perpendicular channel alignment direction fitting result self-diffusion coefficients D|| = 2.9 × 10−11 m2s−1 D⊥ = 1.6 × 10−12 m2s−1 diffusion time (Δ) dependence D mean-square displacement 1H supports anisotropic diffusion Figs. 24 self-diffusion coefficient D|| comparable Nafion (0.2–20 × 10–10 m2s–1 varies with water mass fraction temperature activation energies 0.18 eV for D|| 0.14 eV D⊥ smaller activation energies PFG-NMR observes shorter-range diffusionobserved D|| contains contributions channel A and B smooth proton diffusivity in hydrophobic nanochannel A expected weak interactions among confined waters inner wall potential barrier heights rotational translational movements water molecules moderate diffusivity in channel B expected to low dimensionality diffusion path large hydrogen bonds among water molecules amines tubular restricts water reorientation diffusion29 recent study on nanotubular analogues observed proton conductivities lower than 133 no water molecule inside hydrophobic nanotube hydrophilic channel outside contains several water molecules suggest confined waters within channel A with smooth proton diffusivity high proton conductivity conductivity (σ) expressed as = neμ mobile proton sources important for high proton conduction amine sites of dach ligand coordinated to PtIV sites possible proton sources amines coordinated to metal ions lower34 DFT calculations support assumption amine groups coordinated to PtIV cations have similar acidity to carbonic acid (Supplementary Figs. 28 29 Table 3)insights proton diffusivity hydrophobic nanochannel performed quantum molecular dynamics) simulations Figs. 30–35 Tables 4–6 Movies 1–3 vehicular diffusion35 coefficients indicate water molecules channel A move 1.5 times faster than channel B Fig. 35a). diffusion not isotropic in channel A and B Table 7) Grotthuss diffusion26,27 coefficients calculated proton transfer rate quantified counting proton shuttling each simulation step36 proton transfer rate channel A two times higher channel B interaction sulfate anion hindered proton shuttling Fig. 35b). averaged value DG from channel A and B accounts larger portion overall proton diffusion coefficient (Dp) comparison Dp among water environments showed proton diffusivity in hydrophobic nanochannel between liquid solid state Table 8).Liquid-to-solid phase transition of confined freezing water quasi-one-dimensional systems not simpleexperimental investigations CNTs phase transition depends on pore diameter pressure predicted discontinuous for >1.2 nm continuous for <1.2 nm at ambient investigated liquid-to-solid phase transition of confined water in 1 using differential scanning calorimetry (DSC), solid-state 1H-NMR impedance measurements at low temperature DSC measurements showed no phase transition between 163 and 293 K shows continuous liquid-to-solid transition temperature dependence of solid-state 1H-NMR signal line width for crystallisation water showed gradual continuous broadening as temperature decreased results indicate water no clear liquid-to-solid phase transition different from bulk water micropores5 by proton conductivity gradual increase with temperature correlation between proton continuous liquid-to-solid phase transition visualized (Fig. 5) 5Temperature dependence of 1H NMR spectra proton conductivity squares blue circles indicate FWHM of 1H NMR spectra proton conductivity synthesised characterised metal-organic nanotube water-filled hydrophobic nanochannels in crystalline statehydrophobic nanochannel-shaped aperture 1.0 nm (Fig. 1) confined water forms unique clustered structures confirmed by SCXRD measurements Single crystal impedance spectroscopy 1 high proton conductivity 1D channel direction high conductivity attributed to high proton diffusivity acidity coordinated amine protons supported by PFG-NMR theoretical calculations continuous liquid-to-solid phase transition observed from DSC solid-state 1H-NMR impedance measurements different from bulk water 2001, Brewer et al properties confined water proton transport sensitive to size originate from reduced hydrogen-bond interaction negligible water–channel wall similar to bulk water diameter larger experimental results validated predictions Noy et al. 0.8-nm-diameter CNT porins show fast proton transport 1.5-nm-diameter CNT porins transport comparable to bulk nanotube demonstrated unique clustering structures high proton mobility continuous liquid-to-solid phase transition water results show evidence of predicted behaviour confined water in hydrophobic nanochannels Nanotube fabrication coordination chemistry structural tuning of pore size shape surface propertiestunability properties confined water findings provide structural dynamic insights molecular species nanotubular biological channels characterisation compoundsReagents solvents purchased Wako Pure Chemical TCI Sigma-Aldrich Chemical without purification Elemental analyses Yanaco MT-5 MT-6 CHN recorders Centre Organic Elemental Microanalysis Kyoto University [Pt(dach)(bpy)Br]4(SO4)4·32H2O aqueous solution [Pt]48++ tetrabutylammonium sulfate [(C4H9)4N]2(SO4) added stirred red-purple precipitate collected filtration (60% yield material crystals tetrahydrofuran diffusion aqueous Elemental analysis C64H88N16O16S4Br4Pt4·28H2O C 25.04 H 4.73 N 7.30 S 4.18 C 24.73 H 4.35 N 7.26 S 4.25 spectrum TGA curve Supplementary Figs. 4 5Supplementary Discussion experimental experiments H2O (298 K BELSORP-max sample 1 dried high vacuum<10−1 Pa 35 °C 2 days crystallisation water Isotherms Fig. 2a.Single-crystal X-ray analyses Bruker SMART APEX II CCD detector-monochromated Mo Kα radiation 0.71073 100 K crystals prepared vapour diffusion water solution 1 THF Suitable single-crystals selected transferred paraton oil solvent loss crystals mounted MicroMesh structures solved direct methods expanded Fourier techniques refined full-matrix least-squares refinement F2 CrystalStructure software refinement result summarized Supplementary Table 1.Impedance Solartron SI 1260 Impedance/Gain-Phase Analyser 1296 Dielectric Interface 1–1 × 106 Hz humidity temperature controlled Espec Corp. SH-221 incubator measurements low temperature region (140–300 K Oxford OptistatDN2 Lakeshore 340 temperature controllers-state 1H NMR AVANCE II+ 400 NMR spectrometer UltraShieldTM 400 WB 9.4 T superconducting magnet1H magic angle spinning 13C cross-polarization measured rotor 4 mm diameter Pulse field gradient-NMR measurements Diff 50 diffusion probe (Bruker Biospin K.).Supplementary information Additional Files Data 1 2 3 Movie | 49.4 | 0.407148 |
10.1038/s41467-021-21912-y | PMC7955110 | Amyotrophic lateral sclerosis (ALS) and frontotemporal lobar degeneration (FTLD) patients have brain deposits with amyloid-like aggregates from large C-terminal fragments of the transactive response DNA-binding protein of 43 kDa (TDP-43). Here, the authors present the cryo-EM structure of amyloid fibrils generated from the complete C-terminal TDP-43 low complexity domain and they discuss the effects of disease-causing mutations and phosphorylation of specific Ser residues. | Amyotrophic lateral sclerosis and several other neurodegenerative diseases are associated with brain deposits of amyloid-like aggregates formed by the C-terminal fragments of TDP-43 that contain the low complexity domain of the protein. Here, we report the cryo-EM structure of amyloid formed from the entire TDP-43 low complexity domain in vitro at pH 4. This structure reveals single protofilament fibrils containing a large (139-residue), tightly packed core. While the C-terminal part of this core region is largely planar and characterized by a small proportion of hydrophobic amino acids, the N-terminal region contains numerous hydrophobic residues and has a non-planar backbone conformation, resulting in rugged surfaces of fibril ends. The structural features found in these fibrils differ from those previously found for fibrils generated from short protein fragments. The present atomic model for TDP-43 LCD fibrils provides insight into potential structural perturbations caused by phosphorylation and disease-related mutations. | IntroductionProteinaceous brain inclusions containing the transactive response DNA-binding protein of 43 kDa (TDP-43) are a pathological hallmark of amyotrophic lateral sclerosis (ALS) and frontotemporal lobar degeneration (FTLD)1–3. Similar inclusions have also been found in several other neurodegenerative disorders, including Alzheimer’s disease, cerebral age-related TDP-43 with sclerosis, dementia with Lewy bodies, hippocampal sclerosis, Huntington’s disease, and chronic traumatic encephalopathy, among others2,3.TDP-43 consists of an N-terminal domain, two RNA recognition motifs, and a C-terminal low-complexity domain (LCD) that maps to residues ~267–414 and is rich in Gln/Asn and Gly residues3. Pathological inclusions in TDP-43 proteinopathies typically consist of C-terminal fragments of different sizes, the main component of which is the LCD4–6. The latter domain, which contains most disease-associated mutation sites2,3, is believed to drive the aggregation process7,8. Even though morphologically distinct TDP-43 aggregates have been observed in patient-derived histopathological samples1,9, many of them appear to have fibrillar structure and stain with amyloid-specific dyes10–12. Furthermore, recent reports indicate that these aggregates can self-propagate by the prion-like mechanism13–16. Consistent with these findings, polypeptides corresponding to TDP-43 LCD or its fragments have been shown to form amyloid fibrils in vitro3,8,17–19.In contrast to recent progress in high-resolution structural studies of other amyloids19–36, however, similar studies with TDP-43 are limited to fibrils formed by relatively short fragments of the protein19. To bridge this critical gap, here we report a near-atomic-resolution cryo-electron microscopic (cryo-EM) structure of fibrils formed in vitro by the entire LCD domain of TDP-43. Using our structural model for wild-type protein, we also assess the potential impact of disease-associated mutations and phosphorylation of individual Ser residues.ResultsFibril morphology as assessed by atomic force microscopy (AFM)To optimize sample preparation for cryo-EM studies, TDP-43 LCD fibrils were first analyzed by AFM. Consistent with the previous report19, fibrils formed at pH 7 were found to clump together, precluding detailed morphological analysis. This clumping tendency was reduced for fibrils prepared at pH 6, even though they were still not sufficiently dispersed to allow high-resolution cryo-EM studies. When analyzed by AFM, the later fibrils displayed two different morphologies, one twisted and one without a twist (Supplementary Fig. 1a). The twisted fibrils were left-handed, showing in AFM images a maximum height of 7.2 ± 0.3 nm and periodicity of 49.0 ± 6.0 nm.Fibrils formed under mildly acidic conditions (pH 4) were much better dispersed and more suitable for cryo-EM studies. Akin to pH 6 fibrils, these fibrils displayed two different morphologies, one twisted and one lacking such a twist (Supplementary Fig. 1b). In an attempt to select one preferential morphology, four rounds of sequential seeding reactions were performed by adding preformed, sonicated fibrils (10% w/w) to freshly prepared, non-aggregated protein under the same buffer conditions. Despite these efforts, akin to the first-round fibrils, the final sample used for cryo-EM studies also contained two distinct fibril morphologies (Supplementary Fig. 1c). The relative populations of the two fibril types did not change significantly during sequential seeding reactions, with the twisted species in the non-seeded reaction and last-round seeded reaction accounting for 51 ± 16 and 45 ± 10% of the entire fibril population, respectively. The morphology of pH 4 twisted fibrils was very similar to that of pH 6 counterparts: they were left-handed and characterized by the height maximum of 7.8 ± 0.9 nm and periodicity of 46.3 ± 2.8 nm.Cryo-EM structure of TDP-43 LCD fibrilsConsistent with AFM data, cryo-EM images of fibrils formed at pH 4 revealed two types of morphologies, one of them showing a helical twist and the second one lacking such a twist (Supplementary Fig. 2a, b). Helical reconstruction of twisted fibrils (all of which share the same fold) allowed us to determine a density map with a nominal resolution of 3.2 Å (Supplementary Table 1 and Supplementary Fig. 2c). Each twisted fibril consists of a single left-handed protofilament in which subunits stack along the fibril axis with a helical rise of 4.73 Å and a helical twist of −1.66° (Fig. 1a). A near-atomic-resolution structural model could be unambiguously built for the fibril core, which maps to residues 276–414 of TDP-43 (Fig. 1b and Supplementary Table 1).Fig. 1Cryo-EM structure of TDP-43 LCD fibrils.a Cryo-EM density map showing a left-handed helix with a half-pitch of 513 Å. Repeating densities representing β-strands along the z-axis with a 4.73-Å interval indicate a parallel in-register β-sheet architecture. b The atomic model and superimposed cryo-EM map of one cross-sectional layer of each fibril (top view). c Tilted cross-section of a TDP-43 LCD fibril. β-Strands in the top subunit are shown in blue, highly ordered turns in orange, and the less ordered region in yellow. d Schematic representation of one cross-sectional layer of the amyloid core, with β-strands shown as thicker arrows and the less ordered region (residues 295–299) marked as dotted lines. e Hydrophobicity of the fibril cross-section, with hydrophobicity levels colored according to Kyte–Doolittle54. A major hydrophobic core (green box) is made up of the 311–318, 336–341, and 381–383 segments.This core consists of 14 β-strands linked by relatively rigid turns and loops (Fig. 1c, d). Each protein subunit can be divided into two regions that differ with regard to an overall amino acid composition as well as the backbone geometry (Figs. 1d and 2a, b). The C-terminal region (residues 344–414, strands β9–β14) is characterized by a small proportion of hydrophobic amino acids. Relatively long β-strands pack mostly via polar interfaces and many residues within strands β9, β10, β11, and β13 are engaged in steric–zipper interactions (Figs. 1d and 2a). The stacking of subunits within this region is largely maintained by a network of intermolecular backbone H-bonds between β-strands, with additional stabilization through H-bonds between numerous stacked Gln and Asn side chains (Supplementary Fig. 3). This C-terminal region is largely planar, with most β-strands packed through interactions within the same subunit (Fig. 2b).Fig. 2Differences between the N- and C-terminal regions of the amyloid core of TDP-43 LCD fibrils.a Top view of one subunit within the fibril. Large (red) and smaller (orange) hydrophobic cores are present within the N-terminal part, stabilizing this region. The C-terminal region is stabilized largely by steric–zipper interactions involving side chains within strands β9, β10, β11, and β13 (green). b Side view of one subunit within the fibril (with the view angle indicated by the blue arrow in a). In contrast to a largely planar C-terminal region, the N-terminal region extends along the fibril axis over the distance of 22.4 Å. The lowest (Phe276) and the highest (Asn306) points are marked in red. c Side view of the 311–327 segment extending along the fibril axis. This segment in subunit i interacts with the 327–345 segment in subunits i + 1, i + 2, and i + 3. d, e Side view of interlayer interactions between F313 and A341 (d) and I318 and M336 (e). Layers of side chains are packed in a staggered fashion, resulting in a very compact hydrophobic core.By contrast, the N-terminal region of LCD (residues 276–343, strands β1–β8) is rich in hydrophobic amino acids and contains mostly short β-strands that are not involved in steric–zipper interactions. Instead, strands and turns pack against each other in all directions, allowing non-polar side chains to be tightly packed and buried, with a hydrophobic interface between the 311–318 and 336–341 segments (Fig. 1d, e). Side chains on the opposite side of the 336–341 segment form another hydrophobic interface with residues 381–383 within the LCD C-terminal region. Altogether, these three segments make up a large hydrophobic core. A second, smaller hydrophobic core involves residues within the Ala- and Met-rich 321–330 segment that pack against Phe283, Phe289, and Trp334 (Figs. 1e and 2a). To facilitate such a tight packing of hydrophobic residues, the backbone in this region of each subunit makes numerous turns within the x–y plane while also extending along the z-axis over a distance of ~22.4 Å (Fig. 2b). As a result, each subunit (i) not only interacts with the layer directly above (i + 1) and below (i − 1) but also with layers up to (i + 3) and (i − 3), as illustrated in Fig. 2c for the 311–327 segment. Furthermore, some of the interactions are between stacks of hydrophobic residues, which are arranged in a staggered fashion (Fig. 2d, e). Thus, even though only ~40% of residues are involved in intermolecular H-bonds within the cross-β motif, fibrils are likely further stabilized through the interlayer interactions between side chains.The non-planar backbone conformation of TDP-43 LCD subunits (which is not unusual among amyloids22,28,35) results in rugged surfaces of fibril ends with well-defined “ridges” and “grooves”. The ridges are composed mostly of water-exposed hydrophobic residues, with one end containing 15 such residues from the top three subunits [M311 (m, m − 1, m − 2), F313 (m, m − 1, m − 2), F316 (m, m − 1), I318 (m, m − 1), M322 (m), A324 (m), M337 (m), M339 (m), and M405 (m)], and the other end containing nine hydrophobic residues from the bottom three subunits [A341 (n, n + 1, n + 2), L340 (n, n + 1), M336 (n), A381 (n), I383 (n), and A388 (n)] (Fig. 3).Fig. 3Rugged surface of the top and bottom ends of TDP-43 LCD fibril.a, b Surface representation of fibril ends with solvent-accessible hydrophobic residues from different layers marked in different colors. c, d Ribbon representation of the structure at both fibril ends. Exposed hydrophobic residues from different layers are labeled using the same color scheme as in a, b. Fifteen hydrophobic side chains from three different layers [M311 (m, m − 1, m − 2), F313 (m, m − 1, m − 2), F316 (m, m − 1), I318 (m, m − 1), M322 (m), A324 (m), M337 (m), M339 (m), and M405 (m)] are exposed to water at the top end. Nine hydrophobic side chains [A341 (n, n + 1, n + 2), L340 (n, n + 1), M336 (n), A381 (n), I383 (n), and A388 (n)] are exposed at the bottom end.DiscussionRapidly accumulating cryo-EM data for different types of amyloid fibrils point to a large structural polymorphism of fibrils formed even by the same protein. In the present study, we determined a near-atomic-resolution structure of one polymorphic form of fibrils formed by the entire LCD of TDP-43, a protein associated with many neurodegenerative diseases. The second type of fibrils observed in our samples lacks any twist, precluding high-resolution structural characterization by cryo-EM. Even though the present structure has been determined for fibrils formed under mildly acidic conditions (pH 4), similar morphology of fibrils generated at pH 6 suggests that the latter fibrils may be structurally similarity also at the atomic level.A notable feature of the present structure of TDP-43 LCD fibrils is the size of the amyloid core that is comprised of 139 residues (out of 148 present in the LCD). To the best of our knowledge, the size of this core region is the largest among all amyloid fibrils structurally characterized to date37. Interestingly, in contrast to most other reported fibril structures that contain two or more protofilaments37, TDP-43 LCD fibrils consist of a single protofilament. This is, however, not an unprecedented feature, as single protofilament structures have been reported for fibrils formed by antibody light chains32,33, transthyretin34, FUS LCD20, and some polymorphic forms of tau24–26 and α-synuclein29.Another notable feature of TDP-43 LCD fibrils is a highly non-planar backbone conformation within the N-terminal region of the amyloid core. This results in rugged surfaces of fibril ends with many water-exposed hydrophobic residues. Such highly hydrophobic, rugged surfaces of fibril ends may be especially conducive to the recruitment of TDP-43 monomers and their templated conversion into the amyloid conformation. Distinct surfaces at opposite ends could also result in fibril polarity, with different elongation rates at each end. Furthermore, detailed structural characterization of these surfaces (Fig. 3) may provide a starting point for a rational design of TDP-43 amyloid inhibitors.A recent study reported cryo-EM structures of fibrils formed by two relatively short fragments of TDP-43 LCD19. The first fragment (residues 311–360) formed polymorphic amyloid structures with a common motif described as a dagger fold, in which residues from Phe313 to Ala341 form tight hydrophobic interaction. A sharp (~160°) kink at Gln327 defines the tip of the dagger (Supplementary Fig. 4a). The fold adopted by this segment in our structure of fibrils formed by the entire LCD is substantially different, with no sharp kink at Gln 327. Instead, there is a ~90° turn at this residue, followed by another ~90° turn at Gln331, such that the overall fold shows little resemblance to the dagger motif (Supplementary Fig. 4b). Fibrils formed by the second fragment (residues 286–331 with A315E mutation) consisted of four protofilaments that each contain another common motif characterized as R-shaped fold (Supplementary Fig. 4c). It was proposed that a similar fold (stabilized by hydrophobic interactions between Ala315, Ala297, and Phe313) would be adopted by this segment without the mutation19. Again, the fold within this region of fibrils formed by the entire LCD is quite different: Ala297, Phe313, and Ala315 are not in close contact, and the local conformation is dictated by interactions with other parts of the molecule (Supplementary Fig. 4d).Over 30 point mutations within the TDP-43 LCD are associated with ALS and FTLD2,3. The mechanisms by which these mutations facilitate disease are poorly understood. Mapping the pathogenic mutations on the structure of wild-type TDP-43 LCD fibrils revealed that ~50% of them are not compatible with this structure due to severe steric clashes within tightly packed segments of the protein, introduction of charges into the dehydrated fibril interior, or both (Fig. 4a). Thus, these mutations will likely result in substantially different fibril structures, and this may affect the disease phenotype.Fig. 4Disease-related point mutations and phosphorylation sites mapped on the structure of one subunit of fibrils formed from the wild-type, non-phosphorylated TDP-43 LCD.a Mutations that are compatible with this structure are labeled in blue. The remaining mutations (labeled in red) are not compatible with the structure determined for wild-type protein fibrils due to steric clashes within tightly packed segments of the protein (S292N, M311V, S332N, G348C/V, S379P, A382P/T, S393L), introduction of charges into the dehydrated fibril interior (N378D, N390D), or both (G335D, G348R, G376D, G384R). P363 mutation would likely interfere with the formation of a turn observed in the fibril structure for the wild-type protein. Given that the 295–299 segment is relatively flexible in the present structure, the compatibility of A315E substitution (labeled in purple) with this structure is difficult to assess. b Phosphorylation sites exposed on the surface and those buried inside the structure determined for wild-type TDP-43 LCD fibrils are labeled in blue and red, respectively. Phosphorylation of exposed S410 would require only very small structural rearrangement of the backbone of C-terminal residues.One of the features of ALS and FTLD pathology is phosphorylation of Ser residues within the C-terminal part of the TDP-43 LCD, with the consensus pathological phosphorylation sites at Ser403, Ser404, and Ser409/410, and additional disease-specific phosphorylation sites at Ser379 and Ser3693,38. The full phosphorylation landscape and population of molecules phosphorylated at specific sites are, however, unknown. Furthermore, non-phosphorylated protein seems to be also present in pathological inclusions, as indicated by staining with antibodies raised against non-phosphorylated C-terminal epitopes39. Interestingly, while two of the common phosphorylation sites (Ser404 and Ser410) are exposed on the fibril surface, others (Ser403, Ser409, Ser379, and Ser369) are buried in the interior of the fibril structure that we determined for the non-phosphorylated protein (Fig. 4b). Thus, phosphorylation of the latter residues will likely affect the structure of fibrillar aggregates, potentially resulting in a large, phosphorylation site-dependent structural polymorphism. This, in turn, could further influence the disease phenotype. The present structure for non-phosphorylated TDP-43 LCD fibrils provides a necessary foundation for future high-resolution structural studies with fibrils containing protein variants with different phosphorylation patterns.MethodsProtein expression and purificationThe plasmid for bacterial expression of TDP-43 LCD with an N-terminal His-tag was described previously8. The protein was expressed overnight using RosettaTM (DE3) pLysS competent cells (MilliporeSigma) after induction with 1 mM isopropyl β-d-1-thiogalactopyranoside. Cells were collected by centrifugation, lysed by sonication in Buffer A (20 mM Tris-HCl buffer, pH 8, containing 8 M urea, 500 mM NaCl, and 25 mM imidazole), and purified over Ni-charged nitrilotriacetic acid column using 4–5 column volume washes with Buffer A followed by elution with Buffer B (20 mM Tris-HCl buffer, pH 8, containing 8 M urea, 200 mM NaCl, and 250 mM imidazole). Protein was concentrated and purified by high-performance liquid chromatography using a C4 column with acetonitrile gradient in water containing 0.05% trifluoroacetic acid. Protein purity (>95%) was confirmed by gel electrophoresis. Pure protein was flash frozen and lyophilized for later use.Fibril formation and morphology analysis using AFMLyophilized protein was dissolved in Milli-Q purified H2O and passed through an Amicon Ultra centrifugal filter with 100 kDa molecular weight cut-off to remove preformed aggregates8. Fibrils were formed at a protein concentration of 30 μM in 20 mM MES buffer, pH 6.0, or 20 mM sodium acetate buffer, pH 4.0. The samples were placed on rotation (~6–8 rpm) at 37 °C. For AFM imaging, 2 μl of fibril suspension was diluted 10-fold in Milli-Q H2O, deposited on freshly cleaved mica substrate and incubated for 5 min. The surface was then washed four times with Milli-Q H2O and dried under N2. The images were obtained by NanoScope 9.1 using scan assist mode and a silicon probe (spring constant, 40 newtons/m) on a Bruker multimode atomic force microscope equipped with Nanoscope V controller. Image analysis was performed using Nanoscope Analysis 1.5. Fibrils generated from both reactions displayed two different morphologies on AFM images, one twisted and one lacking such a twist. The reported height maxima and periodicity of twisted fibrils are based on measurements using 20 randomly selected fibrils in each group. In an attempt to select one preferential morphology in pH 4 fibrils, 4 rounds of sequential seeding reactions were then performed by adding preformed, sonicated pH 4 fibrils (10% w/w) to freshly prepared, non-aggregated protein under the same buffer conditions. Despite these efforts, akin to the first-round fibrils, the final sample used for cryo-EM studies also contained two types of fibril morphologies (Supplementary Fig. 1). To assess the percentage of twisted fibrils, the total length of twisted and non-twisted fibrils in three randomly selected 2 μm × 2 μm AFM images was measured.Cryo-electron microscopyTwo hundred mesh lacey carbon grids (Ted Pella) were first coated with 0.1 mg/ml graphene oxide and then with 0.1% poly-lysine as described previously40,41. Three microliters of TDP-43 LCD fibril suspension (30 μM) was applied to the coated grid, blotted for 6 s, and plunge-frozen in liquid ethane using a Vitrobot Mark IV (ThermoFisher Scientific). Movies were collected on a Titan Krios G3i microscope (ThermoFisher Scientific) equipped with a BioQuantum K3 camera (Gatan, Inc.), with 0.414 Å/pixel at super resolution mode, 42 e−/Å2 total dose, and 60 total frames. A total of 6589 micrographs were automatically collected using SerialEM42 with 6 shots per position. Beam image shift was applied, and defocus range was between −0.8 and −1.5 μm.Data processingMovies were corrected for drifting and binned by a factor of 2 using MotionCor243. Contrast Transfer Functions (CTF) were estimated by Gctf44. All further processing was carried out using RELION 3.145–47. Fibrils with apparent helical twists were manually picked and a total of 294,168 segments were extracted using an overlap of 97% between neighboring segments and a box size of 512 pixels. Segments were first subjected to several rounds of reference-free two-dimensional (2D) classification using T = 8 and K = 100 to remove poorly defined classes, resulting in 65,075 segments contributing to clear 2D averages. These segments were then used for subsequent three-dimensional (3D) classification employing an initial model of a featureless cylinder generated by relion_helix_toolbox46. The initial helical rise (4.73 Å) was calculated from the 2D class layer line profile, and the initial helical twist (−1.64°) was calculated from the crossover distance. The handedness of the helix was determined by AFM images. The tilt of all segments were kept at 90° throughout the 3D processing. Two rounds of 3D classification were performed using K = 3 and T = 4, resulting in 11,026 segments that contributed to a high-resolution reconstruction. Additional rounds of 3D classification were performed on these segments using a single class and increasing T value (4, 8, 20, 40, and 100) with local optimization of helical twist and rise. In the last round of 3D classification, β-strands were well separated and large side chains could be resolved. The model and data were then used for high-resolution gold-standard 3D refinement. Iterative Bayesian polishing47 and CTF refinement48 were performed to further improve the resolution. The overall resolution was calculated to be 3.2 Å from Fourier shell correlations at 0.143 between two independently refined half-maps. Refined helical symmetry (twist = −1.66°, rise = 4.73 Å) was imposed on the post-processed map for further model building.Model buildingAn initial model was built in Coot49 using large side chains of 343QQNQ346 segment as a guide. Five chains were built at the central region of the density map, which covered all types of intermolecular contact within the map. The model was then subjected to iterative real-space refinements in PHENIX50. At later stages, segments favoring β-strand conformation were identified and the direction of backbone oxygen and nitrogen atoms were adjusted manually to facilitate hydrogen bonding in β-sheets. Such restraints were also implemented in the subsequent refinements. After real-space refinement, side-chain orientations were manually adjusted to ensure energy-favored geometry. The final model was validated using the comprehensive validation method in PHENIX50. A map containing five copies of subunits was extracted manually from the reconstructed map using the UCSF Chimera package51 to calculate Fourier shell correlations between the map and the atomic model.The density of the TDP-43 LCD segment 295–299 was weaker than that of the rest of the molecule, possibly due to the less-ordered structure of this segment. In this region, we displayed the map at a low contour level and built the model manually. After automatic refinement in PHENIX50, this region showed no significant clashes or Ramachandran outliers. It has been reported that such less-defined turns or loops of an amyloid core may adopt more than one conformation30. In our study, due to the relatively low number of particles, we were not able to classify and determine different structures in this less-defined region. Thus, the structure of this segment in our refined model may represent the average of multiple conformations.When assessing structural compatibility of disease-related mutations and phosphorylation at individual Ser residues with our structural model determined for wild-type, non-phosphorylated protein, the Reduce52 and Probe53 programs were used to test for severe steric clashes (>1 Å overlap) in the absence of the backbone movement.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Structural biology",
"Cryoelectron microscopy",
"Alzheimer's disease"
] | brain inclusions transactive response DNA-binding protein 43 kDa (TDP-43) hallmark amyotrophic lateral sclerosis (ALS) frontotemporal lobar degeneration Similar inclusions in neurodegenerative disorders Alzheimer’s disease age-related TDP-43 sclerosis dementia Lewy bodies hippocampal sclerosis Huntington’s disease chronic traumatic encephalopathy-43 N-terminal domain two RNA recognition motifs C-terminal low-complexity domain) residues ~267–414 rich Gln/Asn Gly residues3 Pathological inclusions TDP-43 proteinopathies C-terminal fragments different sizes main component LCD4–6 latter domain disease-associated mutation distinct TDP-43 aggregates observed histopathological fibrillar structure stain amyloid-specific aggregates self-propagate prion-like polypeptides TDP-43 LCD fragments form amyloid fibrils in vitro3 studies studies TDP-43 limited fibrils short fragments near-atomic-resolution cryo-electron microscopic) structure of fibrils formed vitro entire LCD domain TDP-43structural model wild-type protein assess potential impact disease mutations phosphorylation Ser residues morphology assessed by atomic force microscopy (AFM cryo-EM TDP-43 LCD fibrils analyzed by AFM fibrils at pH 7 clump together morphological analysis clumping reduced for fibrils at pH 6 not sufficiently dispersed high-resolution cryo-EM studies fibrils displayed two one twisted one without twist twisted fibrils left-handed maximum height 7.2 ± 0.3 nm periodicity 49.0 ± 6.0 nm.Fibrils mildly acidic conditions (pH 4) better dispersed suitable for cryo-EM studies two morphologies one twisted one twist select morphology four rounds of sequential seeding reactions preformed sonicated fibrils to prepared non-aggregated protein final sample contained two distinct fibril morphologies relative populations fibril types change during seeding reactions twisted species non for 51 ± 16 45 ± 10% of fibril populationmorphology pH 4 fibrils similar pH 6 counterparts left-handed height maximum 7.8 ± 0.9 nm periodicity 46.3 ± 2.8 nm.Cryo-EM structure TDP-43 LCD cryo-EM images pH 4 two morphologies helical twist lacking 2a Helical reconstruction density map resolution 3.2 Å fibril single left-handed protofilament subunits stack fibril axis helical rise 4.73 Å helical twist −1.66° (Fig near-atomic-resolution structural model fibril core maps residues 276–414 TDP-43 1b 1Cryo-EM structure TDP-43 LCD fibrils density map left-handed helix half-pitch 513 Å Repeating densities β-strands z-axis 4.73-Å interval parallel in-register β-sheet architecture atomic model cryo-EM map cross-sectional layer each fibril Tilted cross-section TDP-43 LCD fibril β-Strands top subunit blue highly ordered turns orange less ordered region yellowSchematic representation cross-sectional layer amyloid core β-strands arrows less ordered region (residues 295–299) dotted lines Hydrophobicity fibril cross-section levels colored Kyte–Doolittle54 major hydrophobic core 311–318 336–341 segments 14 β-strands rigid turns loops (Fig. 1c protein subunit divided two regions amino acid composition backbone geometry. 1d 2a C-terminal region (residues strands β9–β14) small hydrophobic amino acids long β-strands pack polar interfaces β9 β10 β11 β13 steric–zipper interactions. 1d stacking subunits maintained intermolecular backbone H-bonds β-strands stabilization H-bonds Gln Asn side chains C-terminal region planar β-strands packed subunit (Fig. 2b).Fig. 2Differences N- C-terminal regions amyloid core TDP-43 LCD fibrils view subunit Large smaller hydrophobic cores N-terminal stabilizing regionC-terminal region stabilized by steric–zipper interactions strands β9 β10 β11 β13 Side view subunit fibril blue arrow N-terminal region extends along fibril axis 22.4 Å lowest (Phe276) highest (Asn306) points marked in red view 311–327 segment fibril axis interacts with 327–345 segment i 1 2 interlayer interactions between F313 A341 I318 M336 chains compact hydrophobic core N-terminal region (residues 276–343 strands β1–β8) rich in hydrophobic amino acids short β-strands not involved in steric–zipper interactions strands pack against non-polar side chains hydrophobic interface between 311–318 336–341 segments (Fig. Side chains opposite side 336–341 form interface with residues 381–383 LCD C-terminal region three segments large hydrophobic core second smaller hydrophobic core residues Ala- Met-rich 321–330 segment against Phe283 Phe289 Trp3341e 2a). tight packing hydrophobic residues backbone subunit makes turns x–y plane z-axis ~22.4 Å (Fig. 2b). each subunit (i) interacts with layer above (i + 1) below (i − 1) layers up to (i + 3) (i − 3) illustrated Fig. 2c 311–327 segment interactions between stacks hydrophobic residues arranged staggered (Fig. 2d, e). ~40% residues involved intermolecular H-bonds cross-β motif fibrils stabilized through interlayer interactions side chains non-planar backbone conformation TDP-43 LCD subunits results in rugged surfaces fibril ends “ridges” “grooves”. water-exposed hydrophobic residues one end 15 residues from top three subunits [M311 F313 A324 M405 other end nine residues from bottom three subunits [A341 L340 M336 A381 I383 A388 (Fig. 3).Fig. 3Rugged surface top bottom ends TDP-43 LCD fibril.b Surface fibril ends with solvent-accessible hydrophobic residues layers marked colors c d Ribbon representation structure both fibril ends Exposed hydrophobic residues layers labeled same color scheme Fifteen hydrophobic side chains from three layers [M311 F313 F316 I318 M322 A324 M337 M339 M405 exposed to water top end Nine side chains [A341 L340 M336 A381 I383 A388 exposed bottom end cryo-EM data for amyloid fibrils point to large structural polymorphism same protein present study determined near-atomic-resolution structure of one polymorphic form fibrils formed by LCD TDP-43 protein associated with neurodegenerative diseases second type fibrils lacks twist precluding high-resolution structural characterization cryo-EM structure determined for mildly acidic conditions (pH 4) similar at pH 6 suggests similarity atomic level notable feature TDP-43 LCD fibrils size amyloid core 139 residues (out of 148 in LCD). size largest among all amyloid fibrils structurally characterizedcontrast to fibril structures two protofilaments37 TDP-43 LCD fibrils single protofilament not unprecedented single protofilament structures reported for by antibody light transthyretin34 FUS LCD20 tau24–26 α-synuclein29 TDP-43 LCD fibrils non backbone conformation N-terminal region amyloid core rugged surfaces water-exposed hydrophobic residues conducive to TDP-43 monomers conversion amyloid Distinct surfaces at opposite ends in fibril polarity different elongation rates detailed structural characterization (Fig. 3) starting point for design TDP-43 amyloid inhibitors study reported cryo-EM structures by two fragments TDP-43 LCD19 first fragment (residues 311–360) formed polymorphic amyloid structures dagger fold residues from Phe313 to Ala341 hydrophobic interaction sharp~160° kink at Gln327 defines tip dagger different no sharp kink at Gln 327 ~90° turn at residue another ~90° turn at Gln331 little resemblance to dagger motifFibrils second fragment (residues 286–331 with A315E mutation four protofilaments R-shaped fold Fig. 4c). proposed similar fold interactions Ala315 Ala297 Phe313) without fold LCD different Ala297 Phe313 Ala315 not close contact conformation dictated by interactions parts molecule Fig. 4d).Over 30 point mutations TDP-43 LCD associated with ALS FTLD2,3 mechanisms disease poorly understood Mapping pathogenic mutations wild-type TDP-43 LCD fibrils ~50% not compatible due steric clashes charges dehydrated fibril interior (Fig. 4a). mutations different fibril structures may affect disease phenotype.Fig. 4Disease-related point mutations phosphorylation sites structure subunit fibrils wild-type non-phosphorylated TDP-43 LCD Mutations compatible structure labeled in bluemutations red compatible with structure wild-type protein fibrils due to steric clashes segments (S292N M311V S332N G348C/V S379P A382P/T S393L), charges dehydrated fibril interior (N378D (G335D G348R G376D G384R). P363 mutation formation turn fibril structure 295–299 segment flexible compatibility A315E substitution purple difficult Phosphorylation sites exposed buried inside structure-type TDP-43 LCD fibrils labeled in blue and red Phosphorylation of S410 small structural rearrangement C-terminal residues ALS FTLD pathology phosphorylation Ser residues C-terminal TDP-43 LCD phosphorylation sites at Ser403 Ser404 Ser409/410 Ser379 Ser3693 full phosphorylation landscape population unknown non-phosphorylated protein present in pathological inclusions staining antibodies against non-phosphorylated C-terminal two phosphorylation sites (Ser404 Ser410) exposed on fibril surface others (Ser403 Ser409 Ser379 Ser369 buried in interior fibril structurephosphorylation structure fibrillar aggregates phosphorylation site-dependent structural polymorphism disease phenotype structure non-phosphorylated TDP-43 LCD fibrils foundation future high-resolution studies phosphorylation expression bacterial expression TDP-43 LCD N-terminal His-tag protein expressed overnight RosettaTM pLysS cells induction 1 mM isopropyl β-d-1-thiogalactopyranoside Cells collected centrifugation lysed sonication Buffer A-HCl buffer 8 urea 500 NaCl 25 mM purified Ni-charged nitrilotriacetic acid column 4–5 column washes elution Buffer B Protein concentrated purified high liquid chromatography 0.05% trifluoroacetic acid Protein purity (>95% confirmed gel electrophoresis Pure protein flash frozen lyophilized formation morphology analysis AFMLyophilized protein dissolved Milli-Q purified H2O passed Amicon Ultra centrifugal filter 100 cut-off Fibrils formed protein concentration 30 μM 20 mM MES buffer pH 6.0 20 mM sodium acetate buffer pH 4.0samples placed rotation rpm at 37 °C AFM imaging 2 μl fibril suspension diluted 10-fold in Milli-Q H2O deposited mica substrate incubated 5 min surface washed four times-Q H2O dried under N2. images obtained NanoScope 9.1 scan assist mode silicon probe 40 newtons/m Bruker multimode atomic force microscope Nanoscope V controller analysis Nanoscope Analysis 1.5 Fibrils two one twisted one lacking twist height maxima periodicity twisted fibrils based 20 selected fibrils select morphology 4 rounds seeding reactions preformed sonicated pH 4 fibrils to non-aggregated protein final sample contained two fibril morphologies total length non in three 2 μm × 2 μm AFM images measured.Cryo-electron microscopyTwo hundred mesh lacey carbon grids coated 0.1 mg/ml graphene oxide 0.1% poly-lysinemicroliters TDP-43 LCD fibril suspension (30 applied coated grid blotted 6 s plunge-frozen ethane Vitrobot Mark IV Movies collected Titan Krios G3i microscope BioQuantum K3 camera 0.414 Å/pixel super resolution 42 e−/Å2 dose 60 frames 6589 micrographs collected SerialEM42 6 shots per position Beam image shift defocus −0.8 −1.5 μm corrected drifting binned 2 MotionCor243 Contrast Transfer Functions estimated Gctf44 processing RELION 3.145–47 Fibrils helical twists picked 294,168 segments extracted overlap 97% box size 512 pixels two-dimensional (2D classification T = 8 K = 100 65,075 segments 2D averages three-dimensional (3D classification featureless cylinder relion_helix_toolbox46 helical (4.73 Å) calculated 2D class layer line profile helical twist (−1.64°) crossover distance handedness helix determined AFM images tilt 90° Two rounds 3D classification K = 3 T = 4 11,026 segments high-resolution reconstructionrounds 3D classification performed segments single class increasing T value (4 8 20 40 100) optimization helical twist rise last round β-strands separated large side chains resolved model data used for high-resolution gold-standard 3D refinement Bayesian CTF refinement48 resolution overall resolution calculated 3.2 Å Fourier shell correlations 0.143 between refined half-maps Refined helical symmetry (twist −1.66° 4.73 Å imposed post-processed map model building initial model built Coot49 side chains 343QQNQ346 segment Five chains built central region density map intermolecular contact model subjected real-space refinements PHENIX50 segments favoring β-strand conformation identified direction backbone oxygen nitrogen atoms adjusted hydrogen bonding restraints implemented subsequent refinements side-chain orientations adjusted energy-favored geometry final model validated validation PHENIX50 map five subunits extracted UCSF Chimera calculate Fourier shell correlations density TDP-43 LCD segment 295–299 weaker less-ordered structure displayed map low contour level built model manuallyrefinement PHENIX50 region showed no clashes Ramachandran outliers less-defined turns loops amyloid core may adopt more one study low number particles classify structures less-defined region structure segment refined model may represent average multiple conformations assessing structural compatibility disease-related mutations phosphorylation Ser residues structural model wild-type non-phosphorylated protein Reduce52 Probe53 programs severe steric clashes (>1 Å overlap backbone movement.Reporting Nature Research Reporting Summary.Supplementary Review FileReporting Summary | 49.7 | 0.595317 |
10.1038/s41467-020-20348-0 | PMC7801705 | Trafficking of IgA/commensal complex in the gut has been implicated in inflammatory bowel diseases such as Crohn’s disease, but molecular insights are still lacking. Here the authors show, using mouse model or human cells, that NOD2 mutation increases IgA transport, potentially by altering gut microfold cells from the gut, to impact gut inflammation. | Intestinal microfold cells are the primary pathway for translocation of secretory IgA (SIgA)-pathogen complexes to gut-associated lymphoid tissue. Uptake of SIgA/commensals complexes is important for priming adaptive immunity in the mucosa. This study aims to explore the effect of SIgA retrograde transport of immune complexes in Crohn’s disease (CD). Here we report a significant increase of SIgA transport in CD patients with NOD2-mutation compared to CD patients without NOD2 mutation and/or healthy individuals. NOD2 has an effect in the IgA transport through human and mouse M cells by downregulating Dectin-1 and Siglec-5 expression, two receptors involved in retrograde transport. These findings define a mechanism of NOD2-mediated regulation of mucosal responses to intestinal microbiota, which is involved in CD intestinal inflammation and dysbiosis. | IntroductionThe ability of the host immune system to discriminate between pathogens and commensals is essential to maintain mucosal homeostasis1,2. The critical importance of maintaining a mucosal homeostatic mechanism in the intestine is highlighted when functional or genetic deficiencies exist. An example of such failure in maintaining a finely balanced immune response is the development of chronic intestinal inflammation, such as Crohn’s disease (CD). CD is an idiopathic, chronic regional enteritis that most commonly affects the terminal ileum but has the potential to affect any part of the gastrointestinal tract from mouth to anus. CD is thought to occur as a result of a breakdown in self-recognition of commensal bacteria together with mucosal barrier dysfunction in individuals with a given genetic background3–5. The most strongly associated genetic risk factor for CD in Western populations remains NOD2, an intracellular pattern recognition receptor important in immune defense against intracellular microbes6–8. NOD2 is known to regulate the intestinal barrier function, limiting the transcellular permeability and bacterial translocation9,10. The CD-associated mutation in NOD2 (Leu1007fsinsC, Gly908Arg, and Arg702Trp)10, located within the LRR region of the protein, results in loss of NF-κB activation in response to muramyl dipeptide (MDP). However, the reasons why the inactivation of NOD2 can result in chronic colitis remain largely speculative.Secretory IgA (SIgA) is the most abundant immunoglobulin on mucosal surfaces of humans and many other mammals. SIgA can protect the intestinal epithelium by discriminating commensal bacteria from enteric pathogens11–16. Recognition of enteric pathogens by the intestinal immune system results in the production of high affinity, T-cell-dependent, pathogen-specific IgA, which is “transcytosed” into the intestinal lumen. SIgA exhibits also the striking feature to adhere to the apical membrane of M cells, promoting the uptake and delivery of antigens (Ags) to dendritic cell (DC) located in Peyer’s patches (PP). Under pathological conditions such as infection invading IgA opsonized micro-organisms, these immune complexes amplifies the production of proinflammatory cytokines such as TNF, IL-1β, and IL-23 by human CD103 + DCs17. This retrograde transport is called reverse transcytosis, and is mediated by epithelial M cells18–21. Both the Cα1 domain of SIgA2 and its associated Sialic acid (Sia) residue glycosylation are involved in IgA reverse transcytosis, as well as Dectin-1 and Siglec-5, identified as receptors for SIgA uptake on M-cells19. However, the regulation and pathway(s) whereby SIgA is retro transported across M cells still need to be elucidated.Increase of the intestinal permeability has for many years been recognized as a pathogenic factor in CD. An abundance of clinical, epidemiologic, and animal model studies have assessed the impact of various commensal and potentially pathogenic enteric bacteria that may trigger or exacerbate IBD22,23. In a population-based cohort study, an increased risk of IBD was demonstrated in individuals notified in laboratory registries with an episode of Salmonella gastroenteritis24. This finding promotes the concept that pathogens that cause acute intestinal inflammation may predispose individuals to later development of CD, perhaps by causing initial intestinal inflammation or alterations of the intestinal microbiota to promote the formation of colitogenic microbes. We hypothesized that the mucosal inflammation observed in CD patients could be due to an increasing transport of IgA-pathogen complexes from lumen to PP immune cells through M cells. Indeed, after reverse transcytosis, bacteria-IgA complexes are taken up by CD11c+ DCs, and can induce inflammatory responses18,19. Moreover, intestinal bacteria selected on the basis of high coating with IgA is associated with reduced gut microbial diversity in human25 and conferred dramatic susceptibility to colitis in germ-free mice12,26. The earliest observable CD lesions are reported to occur in the follicle-associated epithelium (FAE), where M cells are abundant27, where the PPs are more numerous, and where IgA2 predominates28. NOD2 mutations associated with CD primarily predispose to the development of lesions in the ileal compartment29, indicating that disease susceptibility is increased by altering signaling interactions between intestinal microbiota and the mucosal innate immunity. Hence, we next hypothesized that NOD2 regulates IgA retrograde transport might explain the dysregulation observed in patients with NOD2 mutations.Here we observe an increasing transport of IgA in human PP biopsies obtained from CD patients with or NOD2 mutations. We demonstrate that NOD2 has a regulatory effect in the IgA transport through human and mouse M cells by decreasing the Dectin-1 and Siglec-5 expression on M cells, already identified to act as co-receptors in this process. Our findings define a mechanism of NOD2-mediated regulation of innate immune responses to intestinal microbiota which is involved in the initiation and/or perpetuation of the mucosal inflammation observed in CD patients.ResultsNOD2 mutation increases SIgA retro transport in CD patientsWe initially examined whether the transport of SIgA was modified as a consequence of the disease, and second refined the analysis to samples collected from CD patients with or lacking NOD2 mutations. To address this question, we first determined NOD2 genotypes from PP biopsies in the terminal ileum from healthy (n = 8) or CD (n = 20) followed by quantification of SIgA-positive cell numbers. The mean fold-increase of SIgA-positive cells per PP was 2.8 in CD patients expressing NOD2 frameshift polymorphisms (n = 10) compared to healthy individuals (n = 8) and CD patients expressing wild type (WT) NOD2 (n = 10). A first colocalization (Fig. 1b) with anti-IgA Ab and anti-secretory component (SC) Ab confirm that IgA was not enriched from inside (interstitial fluids) but from the lumen (retrotranslocation of SIgA from the lumen to the PP). Colocalization IgA+/SC+ was observed in all tested biopsies (healthy n = 8; CD patients with NOD2 frameshift polymorphisms n = 10; CD patients with WT NOD2
n = 10). A second colocalization IgA2+/DC-SIGN+ (Fig. 1b) indicate that counted SIgA+ cells derived from the retrograde transport of SIgA through M cells followed by their DC uptake. Increase of SIgA retro transport is not observed in segments of the intestine lacking PPs (Fig. 1a, b). Weak differences in numbers of SIgA+ cells were measured in patients with NOD2 polymorphisms R702W (n = 4) or FS1007insC (n = 3) or R702W/G908R (n = 3) (Fig. 1a). The sum of the data demonstrate an increase in PP-associated SIgA in the subset of CD patients displaying mutated NOD2 gene solely.Fig. 1NOD2 mutation increases SIgA reverse transcytosis in CD patients.a The two first graph compares the frequency of IgA-positive cells between healthy donors (n = 8), CD WT NOD2 patients (n = 10) and CD patients expressing NOD2 polymorphisms (n = 10) in PP and in villi. The third graph shows the number of IgA positive PP cells in patients expressing different NOD2 polymorphisms. NOD2 polymorphisms (R702W, G908R, FS1007insC) were determined by qPCR and sequencing. The last graph shows the ratio PP/villi. Vertical bars show the mean value ± SEM. A nonparametric Mann–Whitney U-test was used (p values: **p < 0.01, ***p < 0.005). b Images obtained from patient biopsy samples taken from the terminal ileum. This experiment was repeated in all patients with similar results: healthy donors (n = 8), CD WT NOD2 patients (n = 10) and CD patients expressing NOD2 polymorphisms (n = 10). Biopsies were labeled with anti-human GFP-IgA2 and PE-GP2 to label M cells or PE-DC-SIGN or PE anti-human SC at room temperature for 2 h. The left panels show representative images from CD patients expressing NOD2 polymorphism, the middle panels, CD patients with WT NOD2, and the right panels depict images from healthy donors. Bars: 200 µm. On all pictures, dotted lines delineate the follicle–associated epithelium (FAE) separating the intestinal lumen and the lymphoid tissue (side view). Source data are provided as a Source Data file.NOD2 deficiency increases IgA retro transport in miceIn the absence of NOD2, PPs present a higher level of CD4+ T cells and M cells in the FAE and increased levels of Th1 (IFN-γ, TNF-α and IL-12) and Th2 (IL-4) cytokines. These immune alterations are associated with an increase of paracellular permeability and yeast/bacterial translocation30,31. Hence, we next hypothesized that at steady-state, the dysregulation observed in CD patients with NOD2 mutations may favor SIgA reverse transcytosis, and ultimately lead to excessive uptake of bacteria-SIgA complexes32.To this aim, we sought to determine whether the loss of NOD2 expression affected the transport of SIgA in a NOD2 knock-out (KO) mouse model. Transport of fluorescently labeled SIgA was compared between NOD2 KO mice and WT mice by examining the fate of the antibody molecule after administration into a ligated intestinal loop containing a PP. As observed in human gut biopsies obtained from patients with NOD2 polymorphisms, the mean fold-increase (2.3) of SIgA transport from the lumen into the PP in NOD2 KO mice (112.4 ± 14.5) was significantly different when compared to WT mice (49.4 ± 7.6). This significant increase in IgA transport, together with a significant decrease after MDP-PAM treatment demonstrates the implication of NOD2 in the process of transcytosis (Fig. 2a). Finally, we did not observe any retrotranscytosis of negative controls as BSA or an irrelevant mouse IgG using a ligated intestinal loop model suggesting that the effect is SIgA-dependant (Fig. 2a).Fig. 2NOD2 deficiency increases IgA retrograde transport in mice.a Tissue section showing a PP obtained from a ligated intestinal loop in WT, NOD2 KO or WT mice after MDP-PAM stimulation following exposure to SIgA-Cy3 for 60 min. Mouse IgG and the BSA were used as negative controls. The graph shows the number of SIgA-positive cells per PP. This experiment was repeated on 6 mice per group and an average of 3 PP per mouse was done. Vertical bars show the mean value±SEM. One-way ANOVA followed by Bonferroni post hoc test was used (p values: *p < 0.05; ***p < 0.005). b Fecal bacteria from WT or Nod2KO mice were stained with anti-IgA-FITC and the MFI were calculated on IgA-bacteria. This experiment was repeated on 4 mice per group. Vertical bars show the mean value ± SEM. A nonparametric Mann–Whitney U-test was used (ns: not significant). c Lysed bacteria were used as the target antigens in a western blot, using serum (S) or fecal (F) supernatant from WT or NOD2 KO mice as primary antibody and anti–IgA-HRP as the secondary antibody. This experiment was repeated on eight cohoused mice (WT (n = 4), NOD2 KO (n = 4)) with similar results. NOD2 KO and littermate WT mice were cohoused and immunized orally with p24-SIgA. d Levels of Ag-specific mucosal IgA and e levels of serum IgG in littermate WT or NOD2 KO mice were shown. Vertical bars show the mean value ± SEM. n = 5 biologically independent mice per group. One-way ANOVA followed by Bonferroni post hoc test was used (p values: *p < 0.05; **p < 0.01). f Cytokine concentrations in serum and faeces were determined in triplicate by Luminex-multiplex cytokine assay in littermate WT (n = 3) and NOD 2 KO mice (n = 3). Cytokine profiles are shown as radar charts; each axis displays the mean quantity (pg/ml) of each cytokine after immunization with p24-SIgA, p24 and PBS. A nonparametric Mann–Whitney U-test was used. P values have been calculated by comparing the p24-SIgA group with the p24 group; *p < 0.05. Source data are provided as a Source Data file.Oral administration of HIV p24 in association with SIgA to littermate WT mice leads to the generation of a robust immune response against the viral antigen. In this experimental setting, SIgA serves as a delivery system carrying an intact HIV p24 antigen into intestinal M cells in PPs20. We thus speculated that the increased retro-transport of SIgA observed in NOD2 KO mice would result in improved p24-specific reactivity after oral immunization as compared to littermate WT mice. First, we confirmed the same ability of IgA to bind to the same microbiota pattern in littermate or NOD2 KO mice, as previously described33 (Fig. 2c). It was also confirmed by quantification of IgA+ bacteria with flow cytometry (Fig. 2b). As oral immunization is well known to induce both systemic and mucosal responses, the antigen-specific response was measured in both serum and feces samples 1 week after the last immunization. Systemic p24-specific IgG and fecal IgA titers were significantly increased in NOD2 KO mice as compared to littermate WT mice (Fig. 2d, e). Oral delivery of an equivalent amount of p24 alone was not sufficient to induce a specific Ab responses in either KO or littermate WT animals (Fig. 2d, e). Because the local tissue-dependent nature of immune responses relies highly on the controlled production of cytokines, a panel of them was measured by Luminex in serum and feces obtained 24 h after the last immunization. In comparison with p24 alone, p24-SIgA was a potent inducer of all the tested cytokines in serum but also TNF-α, INF-γ and IL-17 measured in feces of NOD2 KO mice (Fig. 2f). Immunization with SIgA alone has no effect on the production of cytokines. Except for IL-10, no difference in cytokine secretion was observed between littermate WT and NOD2 KO mice. The increase of IL-10 in serum after immunization with p24-SIgA in the NOD2 KO mice may indicate regulation of the systemic inflammatory process following initial priming of the specific Th cell response. Taken together, these results indicate that transport of p24-SIgA complexes is dependent on functional NOD2 and results in the regulated passage of the hooked Ag, which is subsequently processed to trigger the onset of mucosal and systemic antibody and cytokine responses.SIgA-Salmonella amplify Salmonella-induced colitis in miceThe demonstration that SIgA reverse transcytosis in NOD2 KO mice favors the transport of a bound Ag, as reflected by the detection of specific immune responses, suggests that such a mechanism may be involved in the pathogenesis of chronic colitis. To address this hypothesis, KO mice displaying opposed effects on intestinal SIgA retrograde transport (i.e., NOD2 KO mice identified as promoting intestinal retrograde transport (this study), and Dectin-1 KO mice known to abrogate reverse transcytosis19) were used and compared with WT littermate animals. The disease condition was triggered upon administration of dextran sodium sulfate (DSS) in drinking water (positive control), or after oral delivery of Salmonella Typhimurium alone (Salmonella-induced colitis model) or in association with murine Salmonella-specific SIgASal4 (working hypothesis). In control experiments, oral administration of the Dectin-1 antagonist laminarin, known to suppress the development of DSS-colitis34, was used to further verify the impact of SIgA retrograde transport inhibition.In comparison with the PBS control group, oral administration of Salmonella alone in littermate WT mice resulted in a significant increase of inflammation severity, as determined by the measurement of the disease activity index (DAI) (Fig. 3a). Oral delivery of Salmonella-SIgASal4 complexes led to a weak, yet statistically significant further increase in DAI at day 5 (Fig. 3a). DSS-treated littermate WT mice showed the highest signs of disease. Blocking of Dectin-1, and thus reduction of SIgA reverse transcytosis by addition of laminarin in the drinking water, markedly decreased the severity of colitis. In support of our working hypothesis, NOD2 KO mice with increased SIgA reverse transcytosis and thus higher intestinal transport of Salmonella exhibited a DAI resembling that of DSS-treated animals (Fig. 3a), well above the index observed in the Salmonella alone condition. Such a colitis-preventing effect was impaired upon delivery of laminarin to mice (Fig. 3a). The “clinical” progression of colitis was also confirmed by quantifying neutrophil infiltration in the lamina propria (Fig. 3b), according to the Nancy histological score (Fig. 3c), by measuring weight loss (Supplementary Fig. 1a), serum IL-6, CRP, and Lipopolysaccharide (LPS) (Supplementary Fig. 1b). The high level of blood LPS in Salmonella-IgA treated mice reflects the IgA-based retrotranscytosis of Salmonella through the intestinal epithelium. This seems to demonstrate the inability of Sal4 IgA to neutralize Salmonella under our experimental conditions as previously described35. To confirm this point, we measured the aggregation of Salmonella Typhimurium after SIgASal4 binding35. We did not observe in vitro a significant increase in Salmonella aggregates after SIgA binding (Supplementary Fig. 1c). The absence of aggregates confirms the non-neutralizing character of Sal4 IgA in our experimental conditions. This could indicate that these aggregates are only capable of forming in vivo and do not alter the retrotranscytosis ability of IgA.Fig. 3SIgA-Salmonella amplify Salmonella-induced colitis in mice.Cohoused mice were challenged orally either with DSS, PBS, Salmonella Typhimurium or Salmonella Typhimurium bound with murine IgA. They were left untreated or treated with 5% laminarin in drinking water for 3 days prior to colitis challenge. a Disease activity index (DAI) score was undertaken daily to evaluate the clinical progression of colitis. The DAI was the combined score of weight loss compared to initial weight, stool consistency, and bleeding. Vertical bars show the mean value±SEM. n = 5 biologically independent mice per group. One-way ANOVA followed by Bonferroni post hoc test was used (p values: *p < 0.05; ***p < 0.005). b Sections from the colon of mice showing neutrophil infiltrates in the lamina propria. This experiment was repeated on all mice showing similar results. c Nancy histological score applied in each colon section for littermate and NOD2 KO mice. n = 5 biologically independent mice per group. One-way ANOVA followed by Bonferroni post hoc test was used (p values: **p < 0.01). Source data are provided as a Source Data file.SIgA-Salmonella treated mice without laminarin have been isolated, plotted and significantly highlighted by comparing littermate WT and Nod2KO mice (Supplementary Fig. 1d). The significative difference of DAI at day 5 after SIgA-Salmonella administration clearly confirms the role of Nod2 in SIgA-Salmonella transport. This statement was also observed using a Salmonella-GFP bound with SIgA administered into a ligated intestinal loop containing a PP in NOD2 KO mice and WT littermate mice (Supplementary Fig 1e). Results obtained by blocking Dectin-1 by the addition of laminarin in the drinking water were similar by using Dectin-1 KO mice (Supplementary Fig. 1f) and are consistent with the previous reports34. Thus, SIgA reverse transcytosis seems to play a role in colitis induction by a NOD2-mediated increased transport of pathogens.NOD2 modulates the Dectin-1 and Siglec-5 expressionIn the mouse intestine, selective IgA reverse transcytosis across the epithelium has been shown to depend on the expression of Dectin-1 and to a lesser extent Siglec-5 on M cells19. IgA reverse transcytosis can be recapitulated in the in vitro model of FAE comprising human polarized Caco-2 cells and M-like cells. Traditional transport (transcytosis) across polarized epithelial of serosal polymeric IgA into secretions requires polymeric Ig receptor and a plethora of evolutionary well-conserved intracellular proteins including EEA-1, Rab-5, Rab-9, Rab-11, Rab-17, and Rab-25. Because the regulation of vesicle trafficking along the retro-transcytotic pathway is unknown, we found it important to determine whether NOD2 may somehow contribute to fine-tune the expression of either one of the two receptors and/or the putative organelle-specific proteins.In order to address the nature of proteins possibly involved, we took advantage of the in vitro model of FAE. In the absence of M-like cell conversion, the plain Caco-2 cell polarized monolayer (mono-culture) allowed weak retrotranscytosis of the reporter IgA-luciferase fusion protein (IgA-Luc), in agreement with previous data19. This background level was not affected after transfection with siRNA targeting a selection of extra- and intracellular proteins, as indicated in Fig. 4a. In the presence of interspersed M-like cells in the polarized monolayer (co-culture), SIgA reverse transcytosis of IgA-Luc was increased by a factor of 4 (Fig. 4a, lanes control and random non targeting oligo). Among the candidate assayed, targeting of EEA-1, Rab-5, Rab-17, as well as SYK and TAK1 proteins involved in Dectin-1 signaling pathway led to a significant 3-fold decrease of SIgA-Luc transport, as compared to controls (Fig. 4a). This decrease compares with that resulting from the targeting of Dectin-1 and Siglec-5 used as positive controls. A significant twofold increase of IgA transport in cells treated with NOD2 siRNA, together with a significant threefold decrease after MDP-PAM treatment confirm the implication of NOD2 in the process of transcytosis, and indirectly validates the results obtained via the FAE model (Fig. 4a). Western Blot (Supplementary Fig. 2a) or quantitative RT-PCR was performed to correlate the decrease of mRNA level and its expected increase after MDP/PAM treatment, (Supplementary Fig. 2b). In additional control experiments, both the decrease in the level of targeted proteins and the monolayer integrity (TEER) were systematically monitored after siRNA transfection (Supplementary Fig. 2c). No significant modifications between controls, knockdown and stimulated cells were observed. Staining of tight junctions using ZO-1 detection confirmed the monolayer integrity after siRNA transfection (Supplementary Fig. 2d). Finally, the retro transport of cholera toxin (CT), which is known to use the Rab-5 pathway36, has been using as a positive apical-to-basal transport control to test the efficacy of siRNA knockdown on either monoculture or coculture conditions (Supplementary Fig. 2e). After Rab-5 siRNA knockdown, CT transport is significantly reduced in our inverted in vitro model, confirming the role of Rab-5 in CT transport and the efficacy of siRNA knockdown.Fig. 4IgA reverse transcytosis after protein inhibition or stimulation.After siRNA knockdown or muramyl dipeptide—Pam3Cys (MDP-PAM) stimulation, IgA2 conjugated with luciferase a transport were quantified in the inverted in vitro model of FAE. Vertical bars show the mean value ± SEM. n = 3 independent experiments. One-way ANOVA followed by Bonferroni post hoc test was used (p values: **p < 0.01; ***p < 0.005). b Immunoprecipitation made 30 or 60 min after IgA incubation. This experiment was repeated twice with similar results. c Immunofluorescence staining after IgA incubation (60 min), with FITC anti-human IgA and anti-Rabs mAbs or anti-EEA-1 mAbs followed by corresponding PE secondary Abs. Colocalization between IgA and endosomal proteins resulted in yellow dots present in EEA-1, Rab-5, and Rab-17 images. This experiment was repeated twice with similar results. Source data are provided as a Source Data file.To identify if EEA-1, Rab-5, Rab-17, Dectin-1, and Siglec-5 are key partners in IgA reverse transcytosis, IgA2 was added apically in the in vitro model of FAE for 30 or 60 min, and IgA2-associated proteins complexes were recovered after immunoprecipitation with protein M-agarose beads. Association between IgA2 and EEA-1, Rab-5, Rab-17, Dectin-1, and Siglec-5 was revealed by immunodetection with a battery of specific antibodies (Fig. 4b). Specific colocalizations IgA+/EEA-1+, IgA+/Rab-5+, and IgA+/Rab-17+ were also confirmed by immunofluorescence. No colocalization was observed with Rab-7, Rab-9, Rab-11, Rab-25, and pIgR with IgA (Fig. 4c).The role of NOD2 in the modulation of protein expression associated with SIgA reverse transcytosis was investigated next. The in vitro model containing M-like cells (co-culture) or not (mono-culture) was used as such, or following of NOD2 knockdown with siRNA. Proteins were recovered from whole cell lysates and their relative abundance was assessed by Western blot (Fig. 5a). The expression of Dectin-1 and Siglec-5 was significantly increased after NOD2 knockdown, whereas no relevant changes were observed upon analysis of the endosomal proteins tested. MDP-PAM treatment led to a decrease of the expression of the two IgA receptors (Fig. 5b). When examining specifically the expression of Dectin-1 and Siglec-5 in GP2+ M-like cells, as tested by flow cytometry, a consistent threefold increase of either receptor that occurred in NOD2 knockdown cells was confirmed as compared to unconverted cells with an enterocyte phenotype (Fig. 5c). In support of data in Fig. 5b, MDP-PAM treatment did not result in any significant changes in the expression of either receptor. Dectin-1 and/or Siglec-5 expression has not been detected on enterocytes. This observation has been confirmed in NOD2 KO mice where Dectin-1 expression is increased on PP M cells compared to WT mice (Fig. 5d). Finally, we investigate the involvement of NOD2 in other stages of the mucosal IgA pathway, such as production. Figure 5e clearly shows no significate difference in IgA concentration observed in serum or feces between NOD2KO mice and littermate WT mice. The sum of the data implicates NOD2 in M-like cells as the plausible player involved in Dectin-1 and Siglec-5-mediated facilitated retrograde transport of IgA across a tight epithelial mimic.Fig. 5NOD2 modulates the expression of Dectin-1 and Siglec-5 receptors.Western blot showing the expression of proteins in an in vitro model containing (co-culture) or not (mono-culture) M-like cells before “−” or after “+” NOD2 blocking of transcription with siRNA knockdown (a) or NOD2 stimulation with MDP-PAM treatment (b). These experiments were repeated twice with similar results. Flow cytometry was used to examine the role of NOD2 on Dectin-1 and Siglec-5 expression in M like cells in vitro (c) (n = 3 independent experiments; Vertical bars show the mean value ± SEM; One-way ANOVA followed by Bonferroni post hoc test was used, p values: ***p < 0.005) and in vivo (d) (n = 4 independent experiments; Vertical bars show the mean value ± SEM; A nonparametric Mann–Whitney U-test was used, p values: *p < 0.05). e Total IgA concentration was determined by ELISA in serum and faeces of NOD2KO mice and littermate WT mice. n = 4 biologically independent mice. One-way ANOVA followed by Bonferroni post hoc test was used. Vertical bars show the mean value ± SEM. kd knockdown, stim stimulation. Source data are provided as a Source Data file.DiscussionSo far, the mechanisms involved in the loss-of-function polymorphisms on downstream NOD2 signaling and the pathogenesis of CD remain largely unknown. A commonly recognized finding is that defects in NOD2 result in a constitutively weak inflammatory response that can lead to increased intestinal bacterial load and with time to chronic intestinal inflammation observed in CD8,37. Here we demonstrate that the mucosal inflammation observed in CD patients is could be due to increased transport through M cells of IgA-bacteria complexes from the lumen to immune cells present in the PP. First, we found that IgA reverse transcytosis was significantly increased through human (Fig. 1) and mouse (Fig. 2) M cells when NOD2 is mutated or absent. At a steady-state, NOD2 seems to down-regulate the IgA retrograde transcytosis through M cells. In CD or colitis mouse models, the absence of NOD2 could increase the transport of IgA-bacteria complexes inducing mucosal inflammation. It is well established that the polymorphism of the NOD2 gene is a major risk factor in CD. However, a molecular explanation of how such loss of function leads to increased susceptibility to CD remains unclear. Hedl et al.38 have shown that NOD2 signaling activates the mTOR pathway which induces the upregulation of anti-inflammatory mediators and simultaneously the downregulation of pro-inflammatory cytokines. These data support the idea that NOD2 modulates innate immune responses to intestinal microflora and thus suggest that the absence of such regulation leads to increased susceptibility to CD.The role of the IgA reverse transcytosis in the pathogenesis of chronic colitis was identified in Dectin-1 KO mice and littermate WT mice using the well-known Salmonella colitis model39,40. Dectin-1 signaling has been described to regulate intestinal inflammation by controlling commensal Lactobacillus-mediated colonic regulatory T cells34. It seems that the transport of IgA-bacteria complexes through M cells via Dectin-1 receptor19 induces intestinal inflammation, suggesting some degrees of correlation between the abundance of M cells and chronic intestinal inflammation. In this respect, Bennet et al.41 found that in both the dextran sodium sulfate and Citrobacter rodentium models of colitis, significantly increased numbers of PP M cells were induced at the peak of inflammation in the colonic epithelium. Using KO mice, we demonstrate that NOD2 acts as an immune regulator of IgA reverse transcytosis. Indeed, NOD2 KO mice administered with Salmonella-IgA complexes exhibit an increase of the inflammation’s severity as compared to littermate WT mice. This observation is supported by studies showing that NOD2 can modulate inflammation and mediate efficient clearance of bacteria from the mucosal tissue during Salmonella colitis42,43.We next investigated the molecular mechanisms which could explain how NOD2 modulates IgA retrograde transport. Using a specific siRNA knockdown approach in the in vitro model containing M-like cells, the involvement of NOD2 as a regulator of IgA transport was confirmed. The role of endosomal proteins already known to be involved in the IgA transport through enterocytes was also studied44. Our data reveal that IgA reverse transcytosis through M cells seems to be mediated by EEA-1, Rab-5, and Rab-17 endosomal proteins as it has been previously described for epithelial IgA transcytosis. The role of Dectin-1 and Siglec-5 in IgA reverse transcytosis was also confirmed19. NOD2 has also been shown to influence MHC cross-presentation29, autophagy induction, and resistance to intracellular bacterial infection45,46. Thus, while principally well-known for its acute signaling effects, NOD2 activation causes a variety of cellular changes in vivo that are also likely important for immune homeostasis. SYK and TAK1 proteins from the Dectin-1 signaling pathway are also involved in the blocking of IgA reverse transcytosis. The role of NOD2 interaction with TAK1 through its leucine-rich repeat (LRR) region to exert its inhibitory effect on TAK1-induced NF-κB activation has been published47. This suggests that NOD2 inhibits TAK1-induced NF-κB activation, which results in the downregulation of IgA transport. A significant decrease of Dectin-1 and Siglec-5 expression after NOD2 knockdown in M cells is observed in the current study, suggesting that NOD2 contributes to regulate the expression of these two receptors for SIgA (Fig. 6). Of interest in the more global context of intestinal diseases, NOD2 has been identified as a negative regulator of TLR248 and of TLR4 in necrotising enterocolitis49.Fig. 6The retrograde transport of IgA-pathogen complexes across M cells is increased in CD patients as compared to healthy individuals.This translocation of IgA is regulated by WT NOD2 upon decrease of Dectin-1 and Siglec-5 expression in M cells, ensuring proper homeostasis. This pathway is controlled by endosomal proteins such as EEA-1, Rab-5, and Rab-17. We would like to emphasize that this identified mechanism is likely one among others that is involved in the initiation and/or perpetuation of mucosal inflammation observed in CD patients.Our finding should be interpreted with caution, as the etiology of CD involves a combination of genetic, environmental, and microbial factors37. Hence, we find it fair to underline that the SIgA-dependent route identified in the study is likely to be one among other mechanisms that eventually initiates and/or perpetuates mucosal inflammation observed in CD. In any case, our data confirm and extend the knowledge that the NOD2 genotype status is currently the strongest genetic marker associated with a severe CD course. In Fig. 1, NOD2 mutations of the studied population (R702W, 1007 fs and R702W/G908R) were compared and no differences were observed in IgA-positive cells distribution in PP. These results are consistent with the study of Hugot et al., which provided strong evidence that the penetrance of the most at-risk genotypes is low. They found no clear relationship between mutation frequencies and the disease incidence in their studied populations and no significant deficit of double-dose mutation carriers among healthy controls50. However, another study predicts an 8% increase in the risk for complicated disease with the presence of a single NOD2 mutation, and a 41% increase with 2 mutations51. Nevertheless, these two studies confirm that CARD15/NOD2 acts in interaction with other unknown risk cofactors.A theory proposes that antibodies generated in response to microbial colonization of the intestine shape the microbiota composition to benefit the health of the host through a process called antibody-mediated immunoselection (AMIS)52. Immunoselection refers to a process of natural selection within a host organism that is mediated by the immune system to influence microbial fitness and hence microbial ecology and evolution. A significant fraction of commensal bacteria is heavily coated with IgAs26,53,54. However, such coating seems to be non-specific, as there is a significant overlap of bacterial species between IgA-coated and non-coated fractions. Under such conditions, the bacteria diversity was significantly reduced. Thus, it seems that reduced affinity maturation of IgAs is associated with reduced diversity and skewed microbiota and abundant coating of bacteria with natural IgAs. A recent paper goes further by saying that bacteria recognized by human SIgM were dually coated by SIgA and showed increased richness and diversity compared to IgA-only-coated or uncoated bacteria55. This bacterial selection mechanism could be a causal agent for CD development or other IgA-based diseases where IgA reverse transcytosis could play a role. Indeed, CD patients have several features in common with IgA nephropathy and celiac disease. Increased small intestinal mucosal permeability has been demonstrated in these pathologies56,57. In IgA nephropathy, this may allow the influx of food and bacterial Ags resulting in immune complex formation and deposition. In celiac disease, the alteration of para- and transcellular pathways has been proposed to explain the retrograde transport of intact peptides, and notably the apico-to-basal translocation of SIgA ensured by the transferrin receptor58.A better understanding of molecular mechanisms driving chronic gut inflammation has led over the past two decades to therapeutic strategies with major impacts for the current management of IBD59,60. However, despite spectacular successes, mainly attributable to the anti-TNF therapy, not all patients respond to the drugs and about one third of the responders relapse within a short period of time. Further works are therefore needed to identify therapeutic molecules in IBD. In the present study, we demonstrated that IgA retro-transport is involved in promoting inflammation in CD by acting on the transport of IgA-bacteria immune complexes in the PP through M cells. Therapeutic strategies aiming at blocking IgA reverse transcytosis during the acute phase of CD may be considered to design new efficient immunotherapeutic strategies.MethodsImmunolabeling of human PP ex vivoInformed and consenting CD patients or healthy individuals who had undergone lower endoscopy for routine diagnostic purposes with normal intestinal mucosa provided two biopsy samples from the terminal ileum. Biopsies were fixed for 2 h in 3% paraformaldehyde and included in OCT embedding solution, before being cryosectioned using a Leica cryostat model CM1950. Seven-micrometers sections were captured on Ultra+ Superfrost microscope slides and stained for M cells. Slides were washed in PBS to eliminate residual OCT embedding solution, and blocked with PBS containing 5% FBS for 30 min at room temperature. Immunolabeling was performed using a combination of GFP-IgA2 (Invivogen), anti-human PE-GP2 mAb (MBL), anti-human PE-DC-SIGN mAb (ThermoFisher scientific), and mouse anti-IgA secretory component Ab (Abcam) followed by goat anti-mouse PE-IgG (Abcam) diluted to 20 µg/ml for 2 h at room temperature. The slides were then washed in PBS, air-dried, and mounted with Fluoprep (Biomérieux). Slides were observed by immunofluorescence microscopy (Eclipse, Nikon).Oligonucleotide sequences for NOD2 polymorphism PCRPCR of NOD2 polymorphisms61 (R702W, G908R, FS1007insC) were performed on DNA extracted from human biopsies with a commercial extraction kit (QIAamp DNA mini kit, Qiagen, Hilden, Deutschland). The following primers were used to identify specific mutations: R702W: forward, 5′-GAA TTC CTT CAC ATC ACT TTC CAG T -3′ and reverse, 5′- GTC AAC TTG AGG TGC CCA ACA TT-3′; G908R: forward, 5′-CCC AGC TCC TCC CTC TTC-3′ and reverse, 5′-AAG TCT GTA ATG TAA AGC CAC-3′; FS1007insC: forward, 5′-CTG AGC CTT TGT TGA TGA GC-3′ and reverse, 5′- TCT TCA ACC ACA TCC CCA TT-3′. We purified PCR products with a PCR purification kit (Qiagen) before sequencing (Eurofins).MiceNOD2 KO mice were obtained from Gabriel Nunez (University of Michigan, USA). Dectin-1 KO mice were obtained from Gordon D. Brown (University of Aberdeen, UK). NOD2 KO mice62 and Dectin-1 KO mice63 have been described. Littermate mice were obtained from Nod2-heterozygous crosses. All mice were born and hosted at the PLEXAN (Platform for Experiments and Analysis, Faculty of Medicine, Université de Saint-Etienne, France) which is a conventional animal facility with infectious sector P2. All mice were co-housed in the same conditions (Temperatures of 20 °C with 50% humidity, 12 light/12 dark cycle, unlimited access to food and water), and were females between 2- and 4-months old. The experimental protocols have been approved by the French ministry of research, the local ethical committee (CEEA-Loire) and the Animal Welfare Committee of the PLEXAN (agreement no. 2017011315316714_v4).SIgA administration into ligated loopsFor ileal loop preparation, mice were starved overnight, anesthetized by intra-peritoneal injection of a mix of ketamine and xylazine (100 and 10 mg/kg animal weight, respectively) and kept warm at 37 °C throughout the surgical procedure. Hundred microliters of a 1 mg/ml solution of SIgA-Cy3 diluted in PBS or Salmonella(GFP)-SIgASal4 immune complexes (see a model of mouse colitis), BSA (Sigma), or mouse IgG (anti-human TAK1 mAb (R&D System) were administered into a 1.5-cm ileal loop containing a PP. Salmonella enterica subsp. enterica serovar Typhimurium GFP were obtained from ATCC (14028GFP™). Upon completion of the experiment, the mice were sacrificed by cervical dislocation and the piece of intestine was removed, extensively washed with PBS, fixed for 2 h in 3% paraformaldehyde, and included in optimal cutting tissue (OCT) embedding solution. Seven-micrometers sections (Leica cryostat model CM1950, Leica Microsystems) were captured on Ultra+ superfrost microscope slides (VWR International). Slides were observed by immunofluorescence microscopy (Eclipse, Nikon). The protocol followed the guidance of the regional Ethics Committee for animal testing, CREEA (Permit number: No. 69387487).Mice stimulated by MDP-PAM were administered with MDP-PAM in the ligated loop, two hours before SIgA-Cy3 administration.To measure mouse SIgA retrotranscytosis, a polymeric IgA Ab (clone IgAC5 specific to S. flexneri serotype 5a LPS64) was obtained as previously described65. Purified free human SC was produced in Chinese Hamster Ovary cells66. SIgA was obtained by mixing in PBS pIgA molecules with a twofold excess of human SC for 2 h at room temperature according to the conditions described previously67. Cy3-SIgA complexes were obtained by conjugation with indocarbocyanine (Cy3) using the FluoroLink mAb Cy3 labeling kit (Amersham Biosciences).Study of IgA/fecal microbiota interactionsThe ability of fecal IgA to bind gut microbiota has been measured by western blot33. Lysed bacteria were used as the target antigens in a western blot assay. Serum or fecal supernatant from WT or Nod2KO mice were used as primary antibody and anti-IgA-HRP as secondary antibody. This experiment was repeated on 4 mice per group. The quantity of IgA already bound with fecal bacteria were also quantified by flow cytometry. Fecal supernatant from WT and Nod2KO mice were stained with anti-IgA-FITC and the MFI were calculated on IgA-bacteria. This experiment was analysed using 4 mice per group.In vivo delivery of p24-SIgAHIV-1 p24 capsid protein from clade B strain (Px Therapeutics, France) was covalently associated with polymeric SIgAC5 using the Sulfo-KMUS heterobifunctional crosslinker (Thermo Scientific). Covalent complex formation was verified by Western blot with a polyclonal anti-HIV-1 serum and revealed with anti-human IgG HRP-conjugated secondary Ab (Amersham). Mouse oral immunizations were performed by orogastric intubation with polyethylene tubing under light anesthesia with isofluroan (Halocarbon Laboratories). The tubing was introduced at a fixed distance of 1.8 cm from the incisors. Immunizations consisted of three administrations of 100 ml at 1-week intervals. Littermate WT (n = 5), and Nod2 KO (n = 5) mice were immunized with 100 mg of HIVp24-SIgA, or SIgA alone, or HIVp24 alone per administration.Measurement of HIVp24-specific IgG and IgA AbsSerum and feces samples were recovered 1 week after the last immunization. Five fresh feces were collected from each animal. Feces were incubated with Halt Protease Inhibitor Cocktail (Thermo Scientific), centrifuged at 16,000 × g, and stored at −20 °C until use. Specific Abs against HIVp24 were measured using a quantitative ELISA. Maxisorp 96-well plates were coated with either 50 µl of HIVp24 Ag solution (5 µg/ml in sterile PBS) or 50 µl of a 1/3200 dilution of an equal mixture of anti-mouse Ig kappa and lambda light chain–specific mAbs (Serotec), and then incubated O/N at 4 °C. Murine IgG or IgA immunoglobulins (Igs) (Southern Biotech) were used as standards. Bound or captured Igs were detected by incubation with HRP-conjugated goat antimouse (IgG), while IgA was detected using biotinylated goat antimouse IgA (Southern Biotech) followed by streptavidin-HRP (Amersham). Results are given as the means of concentrations ±SEM.Cytokines and chemokinesThe evaluation of multiple cytokines/chemokines was performed with a Luminex 100 instrument (Luminex Corporation, Austin, TX, USA), in combination with the Bio-Plex mouse cytokine 23-plex panel and Bio-Plex mouse cytokine Th17 panel B 8-Plex Group III (Bio-Rad, Berkeley, CA, USA). Biological fluids were recovered 1 week after the last immunization. Cytokine and chemokine concentrations were determined as the mean of three replicates.Model of mouse colitisThe virulent streptomycin-resistant Salmonella enterica serovar Typhimurium strain SL1344 was cultured in LB (LB Broth, Sigma) supplemented with 90 µg/ml of streptomycin (LB-St). A day before infection, a SL1344 colony was cultured overnight at 37 °C, 100 rpm in 3 ml of LB-St. Mouse IgASal4, specific for Salmonella Typhimurium surface carbohydrates was produced as described previously and used to treat mice with SIgA-Salmonella complexes. In other experiments, colitis was also induced by giving 5% dextran sodium sulfate solution in drinking water ad libitum for 7 days68. Body weight and DAI were monitored each day.NOD2 KO (n = 5), Dectin-1 KO (n = 5), and littermate WT (n = 5) mice were given either Salmonella-SIgASal4 complexes or Salmonella alone (i.e., 106 CFU/mouse) in 100 µl PBS by orogastric intubation with polyethylene tubing under light anesthesia with isoflurane (Halocarbon Laboratories). Mice were not pre-treated with streptomycin as we would like to measure the role of SIgA reverse transcytosis in the context of a normal microbiota. Infectivity and dissemination of Salmonella SL1344 were fist tested at different doses. A dose of 1 × 106 CFU/mouse was used as it was not lethal but induce strong inflammation.Induction of colitis was compared in NOD2 KO (n = 5), Dectin-1 KO (n = 5), and littermate WT (n = 5) mice after adding 5% laminarin in drinking water for 3 days before colitis induction34.Assessment of colitis severityDuring the duration of the experiment, the DAI score was monitored daily to evaluate the clinical progression of colitis69. The DAI score combines read-outs including: weight loss compared to initial weight, stool consistency, and rectal bleeding. Scores were defined as follows: weight loss: 0 (no loss), 1 (1–5%), 2 (5–10%), 3 (10–20%), and 4 (>20%); stool consistency: 0 (normal), 2 (loose stool), and 4 (diarrhoea); and rectal bleeding: 0 (no blood), 2 (visual pellet bleeding), and 4 (gross bleeding, blood around anus). The experimental endpoint was reached when mice exhibited weight loss >20% of initial weight with dehydration and diarrhoea had to be euthanized by cervical dislocation following inhalation of isoflurane.To evaluate histological damages reflecting colitis severity, a small fragment (0.5 cm) of the colon was cut, embedded in OCT, and frozen in isopentane cooled with liquid nitrogen. Seven-micrometer sections were prepared as previously described and stained with hematoxylin/eosin using the published procedures70. Scoring of neutrophil infiltration was performed using the Nancy histological score which has been evaluated by a blinded pathologist.Mice were bled 5 days after treatment through retro-orbital plexus. The presence of IL-6, LPS, and CRP in sera was assessed by ELISA (Mouse IL-6 ELISA MAX, Biolegend, San Diego, CA; Mouse Lipopolysaccharide ELISA Kit, ELISAgenie, London, UK; Mouse C-Reactive Protein/CRP DuoSet ELISA, R&D system, USA).IgA-Salmonella agglutination assaySalmonella aggregates formed by IgASal4 has been quantified before oral challenge35. To test the in vitro agglutination of S. typhimurium by IgASal4 antibodies, 0.1 ml of hybridoma culture supernatants (1 µg) was added to 0.1 ml of an overnight culture of bacteria and incubated in round-bottom ELISA plates. Unrelated IgA hybridoma supernatant (anti-V. cholera71) or fresh culture medium was used as a control. Agglutination was measured by flow cytometry after 3 h at 23 °C.Cell cultureThe human intestinal cell line Caco-2 (clone 1) (obtained from Dr. Maria Rescigno, University of Milan-Bicocca, Milan, Italy)72 was cultured in Dulbecco’s modified Eagle’s medium (DMEM) (PAA) supplemented with 10% (v/v) fetal bovine serum (FBS, Thermo-Fisher), 1% (v/v) non-essential amino-acids (PAA), and 1% (v/v) penicillin-streptomycin (PAA), referred to as complete DMEM. The human Burkitt’s lymphoma cell line Raji B (American Type Culture Collection), was cultured in RPMI 1640 supplemented with 10% (v/v) FBS, 1% (v/v) non-essential amino-acids, 1% (v/v) L-glutamine and 1% (v/v) penicillin-streptomycin.Inverted in vitro model of the human FAE19,73Inverted Transwell polycarbonate inserts (12 wells, pore diameter of 3.0 μm, Corning) were coated with Matrigel™, a basement membrane matrix (BD Biosciences) prepared in pure DMEM to a final protein concentration of 100 μg/mL for 1 h at room temperature. The coating solution was removed and inverted inserts were washed with 300 μl of DMEM. Caco-2 cells (3 × 105), suspended in 300 μL of complete DMEM, were seeded on the lower insert side and cultured overnight. The inserts were then inverted and placed in a 12-well culture dish and kept for 9 days. Raji B cells (5 × 105) were resuspended in complete DMEM and then added to the basolateral compartment of the Caco-2 cells, and co-cultures were maintained for 5 days. Mono-cultures of Caco-2 cells, cultivated as above but without back-addition of Raji B cells, were used as controls. The establishment of polarized co- and mono-cultures was controlled by measurement of TEER using an EndohmTM tissue resistance chamber (Endohm-12, World Precision Instruments) connected to a Millicell-ERS Ohmmeter (Millipore). The mean TEER value of medium alone (9 Ω/cm2) was subtracted from each measurement. In random samples, the barrier integrity of the tight junctions was confirmed by zonula occludens-1 (ZO-1) immunolabeling74. For transcytosis analyses, inserts were inverted prior to incubation in a 6-well plate, and a piece of silicon tubing (14 (height) × 20 (diameter) mm, Labomoderne) serving as a medium reservoir was placed on the surface facing the basolateral pole of the cell monolayer.Gene inhibition by small interfering (si) RNACells in the inverted in vitro model of FAE were transfected at a final concentration of 5 nM with ON-TARGETplus SMARTpool siRNAs (Dharmacon) using Silentfect reagent (Bio-Rad) according to the procedure provided by the manufacturer. The reference numbers for gene targeting were as follows: Dectin-1: L-021476-00-0005; Siglec-5: L-019522-02-0005; EEA-1: L-004012-00-0005; pIgR: L-017729-00-0005; Rab-5: L-004009-00-0005; Rab-7: L-010388-00-0005; Rab-9: L-004177-00-0005; Rab-11: L-004726-00-0005; Rab-25: L-010366-00-0005; Syk: L-003176-00-0005; NOD2: L-003464-00-0005. TAK-1 was silenced by using sequence 6317S (Cell Signaling Technology).Treatment with MDP - Pam3Cys (MDP-PAM)In order to specifically activate NOD2, the inverted in vitro model of FAE was exposed to 1 μg/ml MDP and 1 μg/ml Pam3Cys-Ser-(Lys)4 hydrochloride (Invivogen) for 24 h.Immunoprecipitation and Western Blot10 μg of IgA2 (Invivogen) were added to the apical compartment of the in vitro model of FAE. And the cells were incubated for 30 or 60 min at 37 °C. After two washes with PBS, the cells were lysed with Mammalian Protein Extraction Reagent (Thermo Scientific). The lysate was cleared by centrifugation and the protein concentration was brought to 5 μg/ml. IgA2 (and bound proteins) were concentrated by immunoprecipitation with protein M-agarose beads (Invivogen). Post washes, elution was performed with 0.1 m glycine (pH 3.0), with immediate neutralization by 1 M Tris buffer (pH 8.0). The eluted material was subjected to SDS-PAGE followed by transfer onto a hybond ECL nitrocellulose membrane (GE Healthcare Life Science). Immunodetection of targeted proteins was performed with a selection of Abs/antisera including: Goat anti-human Dectin-1/CLEC 7A serum, anti-human CD170 (Siglec-5) mAb (mouse IgG1, Clone #194128), sheep anti-human EEA-1 serum, anti-human pIgR mAb (Mouse IgG3, Clone # 825724), anti-human SYK mAb (mouse IgG1, Clone # 720402), anti-human TAK1 mAb (mouse IgG1, Clone # 491840) were purchased from R&D System, and rabbit anti-human Rab-5 serum, anti-human Rab-7 mAb (mouse IgG2b, Clone # Rab-7-117), anti-human Rab-9 mAb (mouse IgG1, Clone # Mab9), rabbit anti-human Rab-11 serum, anti-human Rab-25 mAb (Rabbit IgG, Clone EPR18353) were obtained from Abcam. Appropriate HRP-coupled secondary Abs were used for detection with the “Clarity Western ECL Substrate” (Biorad).Immunofluorescence stainingTen micrograms of IgA2 (Invivogen) were added to the apical compartment of the in vitro model of FAE. The cells were incubated for 60 min at 37 °C. Inserts were washed in HBSS to eliminate residual medium, incubated in 4% paraformaldehyde for 30 min, permeabilized with 0.1% Triton X-100 (Sigma-Aldrich), and blocked with PBS containing 5% FBS for 15 min at room temperature. Immunolabeling was performed using a combination of same anti-rabs and anti-EEA-1 mAbs described in the previous section. Each reagent was diluted to 1/100 and incubated for 2 h at room temperature. Corresponding secondary antibodies labeled with a PE were incubated for 1 h at room temperature. After two washes, inserts were air-dried, mounted with Fluoprep (BioMerieux), and observed by Immunofluorescence microscopy (Eclipse Ti, Nikon).NOD2 RT-qPCR75Total RNA was extracted using TRIZOL (Invitrogen). Reverse transcription was performed using the PrimeScript RT reagent kit (TaKaRa Biotechnology, Dalian, PRC). The SYBR Premix Ex Taq™ II kit (TaKaRa Biotechnology) was used to amplify GAPDH and NOD2 gene products. The oligonucleotide primers used were as follows: NOD2: forward, 5′-CTG AAG AAT GCC CGC AAG GT-3′ and reverse, 5′-GTC TCT TGG AGC AGG CGG ATG-3′; GAPDH: forward, 5′-TGC ACC ACC AAC TGC TTA GC-3′ and reverse, 5′- GGC ATG GAC TGT GGT CAT GAG-3′. The double standard curve method was used to analyze the relative gene expression76.IgA RT experimentAfter 48 h of siRNA mediated gene knockdown or 24 h of MDP-PAM stimulation, 10 µg of Ab conjugated with luciferase (Luc) or colostrum IgA or 1 µg of CT from V. cholerae (sigma) were added to the apical side of the in vitro model of FAE at 37 °C for 90 min19. Basolateral solutions were then recovered and the number of retro-transcytosed Ab-Luc measured by luminometry using the Gaussia Luc Assay Kit (Biolux) according to the procedure provided by the manufacturer. Transported colostrum IgA and CT were detected by ELISA using respectively biotinylated goat anti-human IgA (Southern Biotech) and biotinylated rabbit anti-CT pAb (Invitrogen) followed by streptavidin-HRP (Amersham).Statistical analysisStatistical analyses were performed using the InStat version 2.01 from the GraphPad Software. A nonparametric Mann–Whitney U-test or one-way ANOVA followed by Bonferroni post hoc test was used where appropriate77,78. The limit of significance for p values was set at 0.05 (marked by * in the plot); ** indicates p values ≤ 0.01, and *** stands for p values ≤ 0.005. Statistically significant differences between groups are emphasized by bars connecting the relevant columns under comparison.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Antibodies",
"Chronic inflammation",
"Mucosal immunology",
"Crohn's disease"
] | host immune system between pathogens commensals essential mucosal homeostasis1,2 importance highlighted when functional or genetic deficiencies exist example failure immune response is chronic intestinal inflammation Crohn’s disease (CD). CD idiopathic chronic regional enteritis affects terminal ileum potential affect any part gastrointestinal tract to CD breakdown self-recognition of commensal bacteria mucosal barrier dysfunction in genetic genetic risk factor for CD NOD2, intracellular pattern recognition receptor immune defense NOD2 intestinal barrier function transcellular permeability bacterial translocation9 CD-associated mutation in NOD2 results in loss of NF-κB activation muramyl dipeptide inactivation of NOD2 chronic colitis speculative.Secretory IgA) abundant immunoglobulin on mucosal surfaces humans mammals intestinal epithelium commensal bacteria from enteric pathogens11–16 Recognition pathogens results in production of high affinity, T-cell-dependent pathogen-specific IgA “transcytosed” into intestinal lumenSIgA to apical membrane M cells uptake delivery antigens to dendritic cell (DC) Peyer’s patches pathological conditions infection IgA immune production proinflammatory cytokines TNF IL-1β IL-23 by human CD103 + DCs17 retrograde transport reverse transcytosis mediated by epithelial M cells18–21 Cα1 domain SIgA2 Sialic acid (Sia) residue glycosylation involved in IgA reverse transcytosis Dectin-1 Siglec-5 receptors for SIgA uptake M-cells19 regulation pathway SIgA across M cells need.Increase intestinal permeability pathogenic factor in CD clinical animal studies impact of pathogenic enteric bacteria IBD22 increased risk IBD in individuals Salmonella gastroenteritis24. pathogens acute intestinal predispose later development CD colitogenic microbes mucosal inflammation in CD patients due to increasing transport IgA-pathogen complexes from lumen to PP immune cells after reverse transcytosis bacteria-IgA complexes taken up by CD11c+ DCs induce inflammatory responses18intestinal bacteria high coating with IgA reduced gut microbial diversity susceptibility to colitis in germ-free mice12 earliest CD lesions in follicle-associated epithelium M cells abundant27 PPs numerous IgA2 predominates28 NOD2 mutations CD predispose lesions ileal compartment29 disease susceptibility signaling interactions intestinal microbiota mucosal innate immunity NOD2 regulates IgA retrograde transport dysregulation in patients NOD2 mutations increasing transport of IgA in human PP biopsies from CD patients with NOD2 mutations NOD2 IgA transport cells decreasing Dectin-1 Siglec-5 expression on M cells findings define NOD2-mediated regulation innate immune responses mucosal in CD patients mutation increases SIgA retro transport in CD examined transport SIgA modified disease refined analysis to samples CD patients with lacking NOD2 mutations determined NOD2 genotypes from PP biopsies terminal ileum SIgA-positive cell numbersmean fold-increase SIgA-positive cells per PP 2.8 in CD patients NOD2 polymorphisms 10 compared healthy 8) wild NOD2 first colocalization 1b anti-IgA Ab anti-secretory component Ab IgA not enriched lumen Colocalization IgA+/SC+ observed in all biopsies (healthy 8 CD NOD2 10 WT NOD2 second colocalization IgA2+/DC-SIGN+ SIgA+ cells from retrograde transport SIgA M cells DC uptake Increase SIgA transport not observed in intestine lacking PPs Weak differences SIgA+ cells in patients NOD2 polymorphisms R702W FS1007insC/G908R data increase PP-associated SIgA in CD patients mutated NOD2 gene 1NOD2 mutation increases SIgA reverse transcytosis in CD patients first graph compares frequency IgA-positive cells between healthy donors CD WT NOD2 patients patients polymorphisms third graph shows IgA positive PP cells in patients different NOD2 polymorphismsNOD2 polymorphisms (R702W G908R FS1007insC determined by qPCR sequencing ratio PP/villi bars mean value ± SEM nonparametric Mann–Whitney U-test used < 0.01 ***p < 0.005) Images biopsy samples terminal ileum repeated in healthy donors 8) CD WT NOD2 10 NOD2 polymorphisms Biopsies labeled anti-human GFP-IgA2 PE-GP2 PE-DC-SIGN anti-human SC room temperature 2 h left panels patients NOD2 polymorphism NOD2 healthy donors Bars 200 μm dotted lines follicle–associated epithelium intestinal lumen lymphoid tissue Source data.NOD2 deficiency increases IgA retro transport NOD2 higher CD4+ T cells M cells FAE increased Th1-γ TNF-α IL-12 Th2 (IL-4) cytokines immune alterations paracellular permeability yeast/bacterial dysregulation CD patients NOD2 mutations favor SIgA reverse transcytosis excessive uptake bacteria-SIgAloss NOD2 expression transport SIgA mouse model Transport SIgA compared NOD2 KO WT mice after administration ligated intestinal loop PP mean fold-increase (2.3) SIgA transport PP NOD2 KO mice (112.4 ± 14.5) different WT mice (49.4 ± 7.6). increase transport decrease MDP-PAM treatment implication NOD2 transcytosis (Fig retrotranscytosis negative controls BSA irrelevant mouse IgG effect SIgA-dependant 2NOD2 deficiency increases IgA transport Tissue section PP ligated intestinal loop WT NOD2 KO WT mice after MDP-PAM stimulation exposure SIgA-Cy3 60 min Mouse IgG BSA negative controls SIgA-positive cells per PP repeated 6 mice per group average 3 PP per mouse bars mean value±SEM One-way ANOVA Bonferroni post hoc test used (p values < 0.05; < 0.005) Fecal bacteria WT Nod2KO mice stained anti-IgA-FITC MFI calculated IgA-bacteria repeated 4 mice per group bars mean value ± SEMnonparametric Mann–Whitney U-test used not Lysed bacteria target antigens western blot serum fecal supernatant WT NOD2 KO mice primary anti–IgA-HRP secondary repeated on eight cohoused mice similar results immunized orally p24-SIgA Ag-specific mucosal IgA serum IgG NOD2 KO shown Vertical bars mean value ± SEM n = 5 independent mice per group One-way ANOVA Bonferroni post hoc test (p values < 0.05 < Cytokine concentrations serum faeces determined Luminex-multiplex cytokine assay WT NOD 2 KO Cytokine profiles radar charts mean quantity (pg/ml cytokine after immunization p24-SIgA p24 PBS nonparametric Mann–Whitney U-test P values calculated p24-SIgA group p24 group *p < 0.05. Source data file administration HIV p24 SIgA WT mice robust immune response SIgA delivery system HIV p24 antigen intestinal M cells increased retro-transport SIgA NOD2 KO mice improved p24-specific reactivity after immunizationconfirmed IgA bind microbiota littermate NOD2 KO mice (Fig. confirmed quantification IgA+ bacteria flow cytometry (Fig. oral immunization systemic mucosal responses antigen-specific response measured serum feces 1 week after last immunization Systemic p24-specific IgG fecal IgA titers increased NOD2 KO mice littermate WT mice (Fig. 2d Oral delivery p24 not responses KO littermate WT local tissue-dependent immune responses controlled production cytokines measured Luminex serum feces 24 h after last immunization p24-SIgA cytokines serum TNF-α INF-γ IL-17 feces NOD2 KO mice (Fig. Immunization SIgA no effect production cytokines no difference cytokine secretion between littermate WT NOD2 KO mice increase IL-10 serum after immunization p24-SIgA NOD2 KO mice regulation systemic inflammatory process Th cell response transport p24-SIgA complexes dependent functional NOD2 regulated passage hooked Ag mucosal systemic antibody cytokine responsesSIgA-Salmonella amplify-induced colitis in SIgA reverse transcytosis in NOD2 KO mice favors transport bound Ag immune responses suggests pathogenesis chronic colitis KO mice opposed effects intestinal SIgA transport Dectin-1 KO compared with WT littermate animals disease condition triggered administration dextran sodium sulfate (DSS) water after oral delivery Salmonella Typhimurium association with murine Salmonella-specific SIgASal4 control experiments oral administration Dectin-1 antagonist laminarin DSS-colitis34 impact SIgA retrograde transport inhibition control group oral administration Salmonella alone in WT mice severity Oral delivery Salmonella-SIgASal4 complexes increase in DAI at day 5 DSS-treated littermate WT mice showed highest signs disease Blocking Dectin-1 reduction of SIgA reverse transcytosis by addition laminarin drinking water decreased severity colitis NOD2 KO mice with increased SIgA reverse transcytosis higher intestinal transport Salmonella exhibited DAI resembling DSS-treated animalsabove index Salmonella alone condition colitis-preventing effect impaired delivery laminarin progression colitis confirmed neutrophil infiltration lamina propria Nancy histological score measuring weight loss serum IL-6 CRP Lipopolysaccharide (LPS) high level blood LPS in Salmonella-IgA treated mice reflects IgA retrotranscytosis Salmonella intestinal epithelium inability Sal4 IgA neutralize Salmonella measured aggregation Salmonella Typhimurium after SIgASal4 increase Salmonella aggregates after SIgA binding absence confirms non-neutralizing character Sal4 IgA aggregates in vivo alter retrotranscytosis IgA 3SIgA-Salmonella-induced colitis in mice mice challenged orally with DSS PBS Salmonella Typhimurium IgA left untreated with 5% laminarin water 3 days prior colitis challenge Disease activity index (DAI) score daily progression colitis weight loss initial weight stool consistency bleeding bars show mean value±SEM = 5 independent mice per groupOne-way ANOVA Bonferroni post test (p values < 0.05 < 0.005) Sections colon mice neutrophil infiltrates lamina propria repeated mice Nancy histological score colon littermate NOD2 KO mice n = 5 independent mice per group One-way ANOVA Bonferroni post test (p values < 0.01). Source data file.SIgA-Salmonella treated mice without laminarin isolated littermate WT Nod2KO mice difference DAI day 5 SIgA-Salmonella administration confirms role Nod2-Salmonella transport observed Salmonella-GFP SIgA intestinal NOD2 KO WT littermate mice Results blocking Dectin-1 laminarin similar Dectin-1 KO mice consistent previous SIgA reverse transcytosis colitis induction NOD2-mediated transport.NOD2 modulates Dectin-1 Siglec-5 selective IgA reverse transcytosis expression Dectin-1 Siglec-5 M IgA reverse transcytosis recapitulated in vitro model FAE human polarized Caco-2 cells M-like cellstransport (transcytosis across epithelial of serosal polymeric IgA into secretions requires polymeric Ig receptor intracellular proteins EEA-1 Rab-5 Rab-9 Rab-11 Rab-17 Rab-25 regulation vesicle trafficking unknown to determine NOD2 expression receptors organelle proteins in vitro model FAE M-like cell conversion Caco-2 cell polarized monolayer allowed weak retrotranscytosis of IgA-luciferase fusion protein not affected after transfection with siRNA targeting extra intracellular proteins Fig 4a interspersed M-like cells SIgA reverse transcytosis of IgA-Luc increased by factor 4 targeting of EEA-1 Rab-5 Rab-17 SYK TAK1 proteins Dectin-1 led 3-fold decrease of SIgA-Luc transport controls decrease compares with targeting Dectin-1 Siglec-5 positive controls twofold increase of IgA transport in cells treated with NOD2 siRNA threefold decrease after MDP-PAM treatment confirm implication NOD2 in transcytosis validates results FAE model Western BlotRT-PCR mRNA level increase after MDP/PAM treatment control experiments decrease targeted proteins monolayer integrity monitored after siRNA transfection No significant modifications controls stimulated cells Staining tight junctions ZO-1 confirmed monolayer integrity siRNA transfection retro transport cholera toxin Rab-5-basal transport control efficacy siRNA knockdown coculture After Rab-5 siRNA knockdown CT transport reduced vitro model confirming role Rab-5 CT transport efficacy 4IgA reverse transcytosis after protein inhibition stimulation siRNA knockdown stimulation IgA2 luciferase transport quantified inverted in vitro model mean value ± SEM 3 experiments One-way ANOVA Bonferroni post hoc test values < < 0.005) Immunoprecipitation 30 60 min after IgA incubation repeated twice similar results Immunofluorescence staining after IgA incubation (60 anti-human IgA anti-Rabs anti-EEA-1 mAbs Colocalization IgA endosomal proteins yellow dots EEA-1 Rab-5 Rab-17 images repeated twice similar resultsSource data file EEA-1 Rab-5 Rab-17 Dectin-1 Siglec-5 IgA reverse transcytosis IgA2 added in vitro model 30 60 min proteins recovered after immunoprecipitation M-agarose beads Association IgA2 EEA-1 Rab-5 Rab-17 Dectin-1 Siglec-5 revealed immunodetection antibodies (Fig. colocalizations IgA+/EEA-1+/Rab-5+/Rab-17+ confirmed immunofluorescence No colocalization Rab-7 Rab-9 Rab-11 Rab-25 pIgR IgA (Fig role NOD2 protein expression SIgA reverse transcytosis investigated in vitro model M-like cells used NOD2 knockdown Proteins recovered cell lysates abundance assessed Western blot (Fig. expression Dectin-1 Siglec-5 increased after NOD2 knockdown no changes endosomal proteins MDP-PAM treatment expression IgA receptors (Fig. expression Dectin-1 Siglec-5 GP2+ M-like cells threefold increase receptor NOD2 knockdown cells confirmed unconverted cells MDP-PAM treatment significant changes expression receptorDectin-1 Siglec-5 expression detected on enterocytes confirmed in NOD2 KO mice Dectin-1 expression increased on PP M cells WT mice (Fig. involvement NOD2 in mucosal IgA pathway production Figure 5e shows no difference in IgA concentration serum feces between NOD2KO WT mice implicates NOD2 in M-like cells Dectin-1 Siglec-5 transport IgA epithelial mimic. 5NOD2 modulates expression Dectin-1 Siglec-5 receptors expression proteins in vitro model-like cells NOD2 blocking transcription siRNA knockdown stimulation MDP-PAM treatment experiments repeated twice similar results Flow cytometry used role NOD2 on Dectin-1 Siglec-5 expression in M cells in vitro 3 experiments One-way ANOVA Bonferroni post hoc test < in vivo 4 experiments nonparametric Mann–Whitney U-test Total IgA concentration determined by ELISA in serum faeces NOD2KO littermate WT mice n = 4 mice One-way ANOVA Bonferroni post hoc test Source data provided filemechanisms polymorphisms NOD2 signaling pathogenesis of CD unknown defects in NOD2 weak inflammatory response to increased intestinal bacterial load chronic intestinal inflammation in CD8 mucosal inflammation in CD patients due to increased transport of IgA-bacteria complexes from lumen to immune cells IgA reverse transcytosis increased human mouse M cells when NOD2 mutated or absent steady-state NOD2 down IgA transcytosis In models absence NOD2 transport IgA-bacteria mucosal inflammation polymorphism NOD2 gene major risk factor in CD molecular explanation loss function susceptibility CD unclear Hedl NOD2 signaling activates mTOR pathway anti-inflammatory mediators pro-inflammatory cytokines NOD2 modulates immune responses absence regulation leads to increased susceptibility to CD role IgA reverse transcytosis in chronic colitis identified in Dectin-1 KO mice littermate WT mice Salmonella colitis Dectin-1 signaling intestinal inflammation Lactobacillus-mediated colonic regulatory T cells34transport IgA-bacteria complexes through M cells via Dectin-1 induces intestinal inflammation correlation cells chronic inflammation Bennet et al dextran sodium sulfate Citrobacter rodentium models colitis increased PP M cells induced peak inflammation colonic epithelium mice NOD2 immune regulator IgA reverse transcytosis NOD2 KO mice administered with Salmonella-IgA complexes severity littermate WT mice studies NOD2 inflammation clearance bacteria from mucosal tissue during Salmonella colitis42 investigated molecular mechanisms NOD2 IgA retrograde transport siRNA knockdown approach vitro model NOD2 IgA transport confirmed role endosomal proteins transport IgA reverse transcytosis cells mediated by EEA-1 Rab-5 Rab-17 endosomal proteins role of Dectin-1 Siglec-5 in IgA reverse transcytosis NOD2 MHC cross-presentation29 autophagy induction resistance to intracellular bacterial infection45 NOD2 activation causes cellular changes in vivo important for immune homeostasisSYK TAK1 proteins Dectin-1 pathway IgA reverse transcytosis NOD2 interaction with TAK1 inhibitory effect on TAK1-induced NF-κB activation NOD2 inhibits NF-κB activation IgA transport significant decrease of Dectin-1 Siglec-5 expression after NOD2 knockdown in M cells observed study NOD2 expression receptors for SIgA (Fig. 6) NOD2 negative regulator of TLR248 TLR4 in necrotising enterocolitis49 retrograde transport of IgA-pathogen complexes across M cells increased in CD patients healthy translocation IgA regulated by WT NOD2 upon decrease Dectin-1 Siglec-5 expression cells homeostasis pathway controlled by endosomal proteins EEA-1 Rab-5 Rab-17 mechanism likely in initiation perpetuation mucosal in CD patients caution etiology of CD involves genetic environmental microbial factors37 SIgA-dependent route likely initiates mucosal inflammation in CD data confirm NOD2 genotype status strongest genetic marker with severe CD course Fig.NOD2 mutations (R702W 1007 R702W/G908R compared no differences in IgA-positive cells distribution PP results consistent with study Hugot et al. penetrance at-risk genotypes low no relationship between mutation frequencies disease incidence no deficit double-dose mutation carriers healthy another study predicts 8% increase risk complicated disease single NOD2 mutation 41% increase 2 mutations51 studies confirm CARD15/NOD2 unknown risk cofactors theory antibodies microbial colonization shape microbiota antibody-mediated immunoselection (AMIS Immunoselection natural selection immune system microbial fitness ecology evolution commensal bacteria heavily coated with IgAs26 coating non-specific significant overlap bacterial species between IgA-coated non-coated fractions bacteria diversity reduced reduced affinity maturation IgAs with reduced diversity skewed microbiota abundant coating bacteria natural IgAs paper bacteria recognized by human SIgM dually coated by SIgA increased richness diversity IgA-only-coated uncoated bacterial selection could for CD development IgA-based diseases reverse transcytosisCD patients features with IgA nephropathy celiac disease Increased intestinal mucosal permeability IgA nephropathy influx of food bacterial Ags immune complex formation deposition celiac disease alteration para transcellular pathways retrograde transport of intact peptides apico-to-basal translocation of SIgA transferrin understanding molecular mechanisms chronic gut inflammation led to therapeutic strategies IBD59 successes anti-TNF therapy not all patients respond one third relapse Further works needed therapeutic molecules in IBD study IgA retro-transport inflammation in CD transport IgA-bacteria immune complexes M cells strategies IgA reverse transcytosis acute phase CD may.MethodsImmunolabeling of human PP ex CD patients two biopsy samples from terminal ileum Biopsies fixed 2 h in 3% paraformaldehyde OCT embedding solution cryosectioned Leica cryostat model CM1950 Seven-micrometers sections captured on Ultra+ Superfrost microscope slides stained for M cells Slides washed in PBS blocked with PBS 5% FBS for 30 minImmunolabeling GFP-IgA2 anti-human PE-GP2 mAb PE-DC-SIGN mAb mouse anti-IgA secretory component Ab (Abcam) goat anti-mouse PE-IgG (Abcam) diluted 20 μg/ml 2 h room temperature slides washed PBS air-dried with Fluoprep (Biomérieux). observed by immunofluorescence microscopy (Eclipse Nikon).Oligonucleotide sequences for NOD2 polymorphism polymorphisms61 (R702W G908R FS1007insC performed on DNA human biopsies commercial extraction kit (QIAamp DNA mini kit primers mutations R702W G908R FS1007insC purified PCR products with kit (Qiagen before sequencing (Eurofins).MiceNOD2 KO mice from Gabriel Nunez (University of Michigan Dectin-1 KO mice Gordon D. Brown (University of Aberdeen NOD2 KO mice62 Dectin-1 KO mice63 described Littermate mice from Nod2-heterozygous crossesmice born hosted PLEXAN Faculty Medicine Université Saint-Etienne France conventional animal facility infectious sector P2. mice co-housed conditions 20 °C 50% humidity 12 light/12 dark cycle access food females 2- 4-months old protocols approved French ministry research local ethical committee-Loire Animal Welfare Committee (agreement 2017011315316714.SIgA administration ligated mice starved overnight anesthetized ketamine xylazine (100 10 mg/kg weight kept warm 37 °C 1 mg/ml SIgA-Cy3 diluted PBS Salmonella(GFP)-SIgASal4 immune complexes BSA mouse IgG administered 1.5-cm ileal loop Salmonella enterica serovar Typhimurium GFP obtained from ATCC mice sacrificed cervical dislocation intestine removed washed with PBS fixed 2 h 3% paraformaldehyde included cutting tissue) solutionSeven-micrometers sections CM1950 captured Ultra+ superfrost microscope slides observed immunofluorescence microscopy protocol regional Ethics Committee testing CREEA 69387487).Mice stimulated MDP-PAM administered hours before SIgA-Cy3 administration SIgA retrotranscytosis polymeric IgA Ab S. serotype 5a LPS64) obtained Purified human SC Chinese Hamster Ovary SIgA obtained mixing PBS pIgA molecules human SC 2 h room temperature Cy3-SIgA complexes indocarbocyanine (Cy3) FluoroLink mAb Cy3 kit (Amersham IgA/fecal microbiota fecal IgA gut microbiota measured western Lysed bacteria target antigens Serum fecal supernatant primary antibody anti-IgA-HRP secondary antibody repeated 4 mice per group IgA bound fecal bacteria quantified flow cytometry Fecal supernatant stained anti-IgA-FITC MFI calculated IgA-bacteria 4 mice groupvivo-SIgAHIV-1 protein strain Therapeutics associated polymeric SIgAC5 Sulfo-KMUS crosslinker complex Western blot polyclonal anti-HIV-1 serum anti-human IgG HRP secondary Mouse oral immunizations orogastric intubation polyethylene tubing light anesthesia isofluroan tubing 1.8 cm incisors Immunizations three administrations 100 ml 1-week intervals Littermate Nod2 KO mice immunized 100 mg HIVp24-SIgA HIVp24 administration HIVp24-specific IgG IgA AbsSerum feces samples recovered 1 week immunization Five fresh feces Feces incubated Halt Protease Inhibitor Cocktail centrifuged 16,000 g stored −20 °C Abs HIVp24 measured ELISA Maxisorp 96-well plates coated 50 μl HIVp24 Ag solution 1/3200 anti-mouse Ig kappa mAbs incubated 4 °C Murine IgG IgA immunoglobulins (Southern Biotech standardsIgs detected incubation HRP-conjugated goat antimouse IgA detected biotinylated goat antimouse IgA streptavidin-HRP (Amersham). Results concentrations ±SEM.Cytokines evaluation multiple cytokines Luminex 100 instrument Austin Bio-Plex mouse cytokine 23-plex panel Th17 panel 8-Plex Group Berkeley CA Biological fluids recovered 1 week after immunization Cytokine chemokine concentrations mean three replicates mouse streptomycin-resistant Salmonella enterica Typhimurium strain SL1344 cultured LB 90 μg/ml streptomycin SL1344 colony cultured overnight 37 °C 100 rpm 3 ml LB-St Mouse IgASal4 Salmonella Typhimurium produced-Salmonella complexes colitis induced 5% dextran sodium sulfate solution 7 Body weight DAI monitored Dectin-1 littermate WT mice given Salmonella-SIgASal4 complexes or Salmonella alone106 CFU/mouse 100 μl PBS orogastric intubation polyethylene tubing light anesthesia isoflurane (Halocarbon Laboratories). not pre-treated streptomycin measure SIgA reverse transcytosis normal microbiota Infectivity dissemination Salmonella SL1344 tested different doses dose 1 × 106 CFU/mouse used not lethal strong colitis compared NOD2 KO 5) Dectin-1 KO 5) littermate WT 5) mice 5% laminarin drinking water 3 days before colitis colitis DAI score monitored daily clinical progression weight loss stool consistency rectal bleeding Scores weight loss 0 1 2 3 4 ( stool consistency 2 4 rectal bleeding 0 2 4 endpoint mice weight loss >20% initial weight dehydration diarrhoea euthanized cervical dislocation inhalation isoflurane histological damages colitis small fragment (0.5 cm) colon cut OCT frozen isopentane cooled liquid nitrogen Seven-micrometer sections prepared stained hematoxylin/eosinneutrophil infiltration Nancy histological score evaluated blinded pathologist bled 5 days after treatment retro-orbital plexus IL-6 LPS CRP sera assessed ELISA IL-6 ELISA MAX Diego Lipopolysaccharide ELISA Kit London C-Reactive Protein/CRP DuoSet ELISA-Salmonella agglutination assaySalmonella IgASal4 quantified oral vitro agglutination S. typhimurium IgASal4 0.1 ml hybridoma culture supernatants overnight culture incubated ELISA plates Unrelated IgA hybridoma supernatant-V control Agglutination measured flow cytometry after 3 h 23 °C human intestinal cell line Caco-2 Dr. Maria Rescigno University Milan-Bicocca cultured Dulbecco’s Eagle’s medium 10% fetal bovine serum non amino-acids penicillin-streptomycin human Burkitt’s lymphoma cell line B cultured RPMI 1640 10% FBS non-essential amino-acids L-glutamine penicillin-streptomycinInverted vitro model human FAE19 Transwell polycarbonate inserts (12 wells 3.0 μm coated MatrigelTM basement membrane matrix DMEM protein concentration 100 μg/mL 1 h room temperature coating removed inserts washed 300 μl DMEM Caco-2 cells (3 × 300 μL DMEM seeded lower insert side cultured overnight inserts inverted 12-well culture dish 9 days Raji B cells (5 × 105 resuspended DMEM added basolateral compartment Caco-2 cells co-cultures maintained 5 days Mono-cultures Caco-2 cells Raji B cells controls polarized co mono-cultures controlled measurement TEER EndohmTM tissue resistance chamber Millicell-ERS Ohmmeter mean TEER value (9 Ω/cm2) subtracted measurement barrier integrity junctions confirmed transcytosis analyses inserts inverted incubation 6-well plate silicon tubing (14 20 mm medium reservoir basolateral pole cell monolayerGene inhibition small interfering RNACells vitro model FAE transfected 5 nM ON-TARGETplus SMARTpool siRNAs Silentfect reagent reference numbers gene targeting Dectin-1 Siglec-5 EEA-1 pIgR Rab-5 Rab-7 Rab-9 Rab-11 Rab-25 L Syk L NOD2 L TAK-1 silenced sequence 6317S MDP - Pam3Cys activate NOD2 exposed 1 μg/ml MDP 1 Pam3Cys-Ser-(Lys)4 hydrochloride (Invivogen) 24 h.Immunoprecipitation Western Blot10 μg IgA2 (Invivogen) added apical compartment cells incubated 30 or 60 min at 37 °C cells lysed with Mammalian Protein Extraction Reagent lysate cleared centrifugation protein concentration 5 μg/ml IgA2 proteins concentrated immunoprecipitation protein M-agarose beads (Invivogen). elution 0.1 m glycine (pH neutralization 1 M Tris buffer (pH 8.0)eluted material subjected SDS-PAGE hybond ECL nitrocellulose membrane Healthcare Life Immunodetection proteins Goat anti-human Dectin-1/CLEC 7A serum anti-human CD170 mAb sheep anti-human EEA-1 serum pIgR mAb SYK mAb TAK1 mAb R&D System rabbit anti-human Rab-5 serum Rab-7 mAb Rab-9 mAb rabbit anti Rab-11 serum Rab-25 mAb Abcam HRP-coupled secondary Abs Western ECL Substrate” (Biorad).Immunofluorescence stainingTen micrograms IgA2 (Invivogen) added apical compartment in vitro model FAE cells incubated 60 min 37 °C Inserts washed HBSS incubated 4% paraformaldehyde 30 min permeabilized 0.1% Triton X-100 blocked PBS 5% FBS 15 min room temperature Immunolabeling anti-rabs anti-EEA-1 mAbs reagent diluted 1/100 incubated 2 h secondary antibodies labeled PE incubated 1 h room temperaturetwo washes inserts air-dried Fluoprep observed Immunofluorescence microscopy Nikon).NOD2 RT-qPCR75Total RNA extracted TRIZOL Reverse transcription PrimeScript RT reagent kit (TaKaRa Biotechnology SYBR Premix Ex TaqTM II kit GAPDH NOD2 gene products oligonucleotide primers NOD2 5′-CTG AAG GAPDH-TGC GC-3′ double standard curve method gene.IgA RT 48 h siRNA gene knockdown 24 h MDP-PAM stimulation 10 μg Ab luciferase (Luc colostrum IgA 1 μg CT V. cholerae added apical side in vitro model FAE 37 °C 90 Basolateral solutions recovered retro-transcytosed Ab-Luc measured luminometry Gaussia Luc Assay Kit Transported colostrum IgA CT detected ELISA biotinylated goat anti-human IgA biotinylated rabbit anti-CT pAb streptavidin-HRP (Amersham).Statistical InStat version 2.01 GraphPad Softwarenonparametric Mann–Whitney U-test one-way ANOVA Bonferroni post hoc test used limit significance p values 0.05 by * ** ≤ 0.01 *** p values ≤ 0.005 significant differences emphasized by bars connecting columns Nature Research Reporting Summary.Supplementary Review FileReporting Summary | 49.9 | 1.01666 |
10.1038/s41467-020-20071-w | PMC7718259 | Electrocatalytic processes are promising for automated and scalable synthesis of singlet oxygen, but they are energy- and chemical-intensive. Here the authors present a Janus electrocatalytic membrane that selectively produces singlet oxygen with low energy consumption and free of chemical precursors. | The importance of singlet oxygen (1O2) in the environmental and biomedical fields has motivated research for effective 1O2 production. Electrocatalytic processes hold great potential for highly-automated and scalable 1O2 synthesis, but they are energy- and chemical-intensive. Herein, we present a Janus electrocatalytic membrane realizing ultra-efficient 1O2 production (6.9 mmol per m3 of permeate) and very low energy consumption (13.3 Wh per m3 of permeate) via a fast, flow-through electro-filtration process without the addition of chemical precursors. We confirm that a superoxide-mediated chain reaction, initiated by electrocatalytic oxygen reduction on the cathodic membrane side and subsequently terminated by H2O2 oxidation on the anodic membrane side, is crucial for 1O2 generation. We further demonstrate that the high 1O2 production efficiency is mainly attributable to the enhanced mass and charge transfer imparted by nano- and micro-confinement effects within the porous membrane structure. Our findings highlight a new electro-filtration strategy and an innovative reactive membrane design for synthesizing 1O2 for a broad range of potential applications including environmental remediation. | IntroductionThe seminal work by H. Kautsky in 1938 revealed an exciting form of molecular oxygen, singlet oxygen (1O2), which later has been proven to be pivotal in the chemical and biomedical fields1–3. This unique nonradical derivative of oxygen, together with hydroxyl radicals (•OH), is considered the most potent among reactive oxygen species (ROS)4,5. Compared with the short-lived and unselective •OH radical, the meta-stable 1O2, possessing an unoccupied π* orbital, shows high selectivity toward electron-rich substances6,7, such as pharmaceuticals8, unsaturated biomolecules7,9, and microbial pathogens10. Consequently, 1O2 targeted reactions are applied in diverse fields, including water decontamination10, photodynamic cancer therapy11, and green organic synthesis12.Greater recognition of the paramount importance of 1O2 has motivated research for more effective 1O2 production. Current approaches for 1O2 production mainly include (i) photosensitization using elaborately designed photosensitizers (e.g., molecular dyes13 or quantum nanodots14) for visible or UVA light adsorption8,15 and (ii) enzymatic reactions (e.g., peroxidases16 and oxygenases17 catalysis) in biological systems, relying on rigorous pH and temperature conditions of the intracellular environment7,18. While both approaches face challenges for industrial scale-up, electrocatalysis (EC), a highly automated and facile technology, could offer more approachable 1O2 production pathways19–21. Specifically, EC enables the flexible generation of critical 1O2 precursors, such as hydrogen peroxide22,23 and hypochlorite20,24. In addition, as an alternative to photoexcitation or chemiexcitation, EC provides efficient electro-excitation to 1O2 precursors, such as cathodic activation of peroxymonosulfate21 and anodic excitation of ferrocene25.EC, however, is a much less explored process for 1O2 generation compared with photosensitization or enzymatic reactions. Present EC pathways rely on high dosages of precursors, which compromise the environmental sustainability of EC26,27. Further, for effective oxidation of target molecules by the in situ generated 1O2, EC currently requires long reaction times (range of hours) and consequently consumes high electric energy20,21. Therefore, it is critical to improving EC efficacy by eliminating the use of chemicals, significantly shortening residence time, and enhancing the Faradaic efficiency of the process.Porous membranes have emerged as ideal electrode substrates to enhance EC performance28,29. A variety of membrane-like structures have been proposed, including horizontally stacked lamellar channels30 and vertically aligned straight-through pores31. These porous substrates spatially confine reactants and electrified active sites at the micrometer- and nanometer-scale. Spatial confinement by membrane pores and channels allows for optimal utilization of electrons by reactants due to minimal diffusion distance and mass transfer enhancement30,32, thereby overcoming intensive chemical usage in conventional EC systems. In addition, studies also highlighted the importance of flow-through configuration of membrane electrodes, which renders electrons to flow parallel to the fluid flow direction, further enhancing the charge transfer rates compared to the conventional flow-by mode (i.e., when the flow direction and electric field direction are perpendiculars)33,34. To date, reported electrocatalytic membranes render only half-cell reaction (cathodic reduction or anodic oxidation) within a membrane substrate. In such cases, the electrode redox reactions are not fully utilized. Therefore, exploiting the electrocatalytic flow-through porous membrane configuration incorporating both cathodic reduction and anodic oxidation may offer an innovative strategy for 1O2 synthesis, featuring enhanced Faradaic and energy efficiency.Herein, we conceive a Janus electrocatalytic membrane realizing ultra-efficient 1O2 production without the addition of chemical precursors. The Janus membrane (Pd–Pt–CM) features double-sided electrochemically reactive surfaces formed by sputtering palladium (Pd) and platinum (Pt) nanoparticles on the respective sides of a ceramic membrane (CM) substrate. We confirmed the co-existence of 1O2 and its precursors, and investigated their spatiotemporal distribution throughout the Pd–Pt–CM during electrocatalytic filtration using the Pd- and Pt-sides as cathode and anode, respectively. We then evaluated the efficacy of the in situ generated 1O2 for water decontamination using sulfamethoxazole as a model organic molecule in the electrocatalytic filtration process. We further proposed a mechanism for the stepwise electrocatalytic production of 1O2 by the Pd–Pt–CM through a set of sequential ROS-mediated redox chain reactions within distinct membrane regions. Overall, our findings demonstrate a new paradigm and electrochemical material platform for 1O2 production by using a Janus electrocatalytic flow-through membrane.ResultsFabrication and properties of Pd–Pt–CMDue to their intrinsic electrical insulating properties and large thickness35–37, CM can serve as ideal substrates for incorporating both cathodic and anodic electrodes on separate sides of CM. This unique feature allows for sequential electro-redox reactions occurring in different inner-porous CM regions during electrocatalytic filtration. As illustrated in Fig. 1a, the Pd–Pt–CM was synthesized via confocal magnetron co-sputtering (details in “Methods”). Palladium (Pd) and platinum (Pt) nanoparticles were, respectively, sputtered on the feed- and permeate-side surfaces of the CM (300 kDa, thickness of 2.5 mm). Noble metal nanoparticles are chosen as they exhibit high electrical conductivity and render efficient EC reactions with lasting stability38,39.Fig. 1Fabrication and characterization of the Pd–Pt–CM.a Schematic depicting the fabrication of the Pd–Pt–CM by magnetron sputtering. b, c SEM cross-section images of the Pd–CM and Pt–CM surfaces, overlapped with EDS mapping of Zr (yellow), O (blue), Ti (green), and Pd (red) or Pt (purple) elements. The insets illustrate the SEM surface images of the respective functionalized membranes overlapped with EDS mapping of Pd (orange) or Pt (purple) element. d Nyquist plots of the Pd–CM and Pt–CM surfaces of the Pd–Pt–CM and corresponding CM substrates in 100 mM Na2SO4 solution at applied frequencies varied from 106–1 Hz. e, f XPS spectra of the Pd 3d and Pt 4f on the Pd–Pt–CM, respectively. g Grazing incidence XRD patterns of the Pd–CM and Pt–CM surfaces of the Pd–Pt–CM with their respective CM substrates. h Pore size distributions of the Pd–Pt–CM (orange circles) and the pristine CM (blue squares). The inset shows the water fluxes of the Pd–Pt–CM and the pristine CM as a function of transmembrane pressure (error bars represent standard deviation from triplicate experiments).Discernible color transformations for both the Pd–CM and Pt–CM surfaces (Supplementary Fig. 1) confirmed the respective metal deposition. Figure 1b, c depicts the morphologies of the Pd–Pt–CM by scanning electron microscopy (SEM) overlapped with the elemental mapping of X-ray EDS. Sputtered Pd and Pt nanoparticles penetrated into the membrane surfaces (detailed EDS mapping results are shown in Supplementary Figs. 2 and 3), exhibiting notable sputtered depths of approximately 60 μm (Fig. 1b, orange) and 90 μm (Fig. 1c, purple), respectively. Such deep inner-pore sputtering guarantees EC reactions within the confined porous membrane structure, instead of only the membrane surface. The larger sputtered depth of Pt–CM than that of Pd–CM, under the same sputtering thickness, is attributed to the distinct asymmetric porous architecture of the pristine CM (Supplementary Fig. 4). In addition, Pd and Pt nanoparticles were uniformly deposited on the respective surfaces of the Pd–Pt–CM as shown by the even distribution of the EDS signals of Pd (Fig. 1b inset, orange) and Pt (Fig. 1c inset, purple), suggesting unimpeded electron transfer over the membrane surfaces during EC reactions.Conductance properties of the Pd–Pt–CM were further quantified by electrochemical impedance spectroscopy (EIS). As illustrated in Fig. 1d, we observed a 6-fold and 40-fold decrease in the amplitudes of the semicircles for the Pd–CM and Pt–CM surfaces, respectively, compared to the corresponding sides of the pristine CM. This observation indicates that the sputtering of Pd and Pt nanoparticles notably reduces the charge transfer resistance and facilitates electron transfer of the functionalized membrane surfaces40. Further, the Pt–CM surface exhibits much lower charge transfer resistance than that of the Pd–CM surface, which could be attributed to the synergistic effect of the higher intrinsic conductivity of Pt and the larger sputtered depth.The elemental composition and the crystalline structure of the Pd–Pt–CM were investigated by X-ray photoelectron spectroscopy (XPS) and the grazing incidence X-ray diffraction (GI-XRD). XPS peaks of Pd 3d3/2 and Pd 3d5/2, centered at 340.2 and 335.1 eV (Fig. 1e), suggest that Pd0 is the dominant component with minor Pd2+ coexisting on the Pd–CM surface41, possibly due to surface oxidation. Similarly, the deconvoluted peaks of Pt 4f5/2 and Pt 4f7/2 at binding energies of 74.8 and 71.5 eV (Fig. 1f) indicate the predominance of Pt0 along with partially oxidized Pt2+ on the Pt–CM surface42 (XPS survey spectra is provided in Supplementary Fig. 5). The Pt2+ signal is likely ascribed to the oxidation of Pt0 on the near-surface of the anodic membrane region since XPS can only retrieve information <10 nm past the surface of the sample43. Notably, the bulk sputtered Pt underneath the anodic membrane surface is mainly composed of Pt0, the more thermodynamically stable Pt species. The XRD analysis reveals that the Pd–CM surface (Fig. 1g, orange) exhibits characteristic diffraction peaks of the (111) and (200) planes of Pd0 (JCPDS No. 46-1043) with a face-centered cubic (fcc) crystal structure at 39.2° and 45.5°, respectively44. The other peaks correspond to rutile (JCPDS No. 73-1232) and zirconium oxide (JCPDS No. 49-1642), indicative of the CM substrate. The presence of crystalline Pt for the Pt–CM surface (Fig. 1g, purple) is corroborated by the diffraction peaks at 39.7°, 46.5°, and 67.7°, consistent with (111), (200), and (220) planes of the fcc Pt0 (JCPDS No. 04-0802)45. The mean nanoparticle sizes of Pd and Pt are calculated as 20.1 and 22.8 nm, respectively, using the Scherrer equation46. Notably, the high dispersion of the nano-sized Pd and Pt is indicated by the broadening of the XRD diffraction peaks47, as compared with the intensive peaks of the pristine CM grains (Fig. 1g, gray).We further analyzed the influence of sputtering on membrane performance. The resembling pore size distributions (Fig. 1h) of the Pd–Pt–CM and the pristine CM suggest a negligible effect of sputtering on membrane structure. The inset in Fig. 1h also shows that the Pd–Pt–CM has comparable water flux as the pristine CM, indicating the negligible influence of sputtering on membrane permeability. We also observed reduced surface roughness of both functionalized surfaces of the Pd–Pt–CM comparing with those of the pristine CM (Supplementary Fig. 6), implying that the seeding of Pd and Pt nanoparticles smooths the surface morphology. In addition, the Pd–Pt–CM possesses hydrophilic surface properties as the pristine CM, as evidenced by the similar water contact angles of the Pd–Pt–CM (21°) and CM (19°) (Supplementary Fig. 7). The smooth surface and hydrophilic nature of the Pd–Pt–CM suggest a lessened fouling potential by organic molecules35,37 and thus more efficient electron transfer during filtration.Spatiotemporal distribution of 1O2 and other ROS in Pd–Pt–CMElectrocatalytic filtration with the Pd–Pt–CM was conducted in a customized cross-flow filtration device (Supplementary Fig. 8). As shown schematically in Fig. 2a, the Pd–Pt–CM was mounted in-between a feed chamber (Region B, orange) and a permeate chamber (Region C, blue). In our setup, the Pd–CM surface functioning as a cathode faced the feed chamber, while the Pt–CM surface serving as an anode faced the permeate. The feed solution was circulated within the feed reservoir (Region A, gray) at a cross-flow velocity of 0.8 L min–1 under an applied pressure of 0.1 bar.Fig. 2Spatiotemporal distribution of reactive oxygen species (ROS) generated in the Pd–Pt–CM.a Schematic illustrating the cross-flow filtration system containing a feed reservoir (Region A, gray) and an electrofiltration module with a 12 mL feed chamber (Region B, orange), a 12 mL permeate chamber (Region C, blue), and the Pd–Pt–CM. b EPR analysis for 1O2 in different regions of the filtration system using 2,2,6,6-tetramethylpiperidine (TEMP, 25 mM) as the trapping agent. EPR measurement of the TEMP-1O2 adduct in the permeate (Region C) of the filtration system using an N2-saturated feed solution is conducted for comparison. c
1O2 generation indicated by the degradation of furfuryl alcohol (FFA, 50 μM). d H2O2 detection measured by Amplex Red color reaction. e Detection of O2•− by using 2,3-Bis-(2-methoxy-4-nitro-5-sulfophenyl)-2H-tetrazolium-5-carboxanilide (XTT, 100 μM) as the probe. f
•OH measurement by applying terephthalate (TPA, 1 mM) which reacts with •OH to form hydroxyterephthalate (hTPA). Error bars represent standard deviations from triplicate experiments.We first detected singlet oxygen (1O2) generation by the Pd–Pt–CM electrofiltration with only Na2SO4 electrolyte as the feed solution. Electron paramagnetic resonance (EPR) spectra (Fig. 2b) showed that by applying the trapping agent 2,2,6,6-tetramethylpiperidine (TEMP), the typical triplet signal (1:1:1, a(N) = 16.9 G) of 2,2,6,6-tetramethyl-4-piperidinol-N-oxyl (TEMPO) in Region C (blue line) was observed, indicating the presence of 1O2 in the permeate8,48. Detailed EPR measurement is provided in the Supplementary methods; the EPR spectrum of the blank control solution, i.e., 100 mM Na2SO4 with 25 mM TEMP, is provided in Supplementary Fig. 9. The lack of TEMPO signals in Regions A and B (Fig. 2b, gray and orange lines) suggests that 1O2 was generated from the Pd–Pt–CM inner-pore structure during the electrocatalytic filtration. We observed a minimum applied voltage of 1.6 V for producing 1O2 (EPR spectra in Supplementary Fig. 10). Given that side reactions, i.e., water splitting for hydrogen and oxygen evolutions, are intensified at high voltage and thus reducing the Faradaic efficiency49,50, a minimum voltage of 1.6 V was selected for the subsequent experiments. In addition, 1O2 production was confirmed by furfuryl alcohol (FFA) probe51, which is highly selective to 1O2 with a rate constant of 1.2 × 108 M−1 s−1. Near 17% degradation of FFA (50 μM) was continuously achieved by the Pd–Pt–CM electrified at 1.6 V (Fig. 2c, Region C, blue triangles), suggesting a constant production of 1O2 within the membrane. The detected products of FFA agree well with the oxidation products of FFA by 1O2 oxidation32,52 (Supplementary Fig. 11; detailed detection methods provided in the Supplementary methods). In contrast, less than 3% removal of FFA was observed from the Pd–CM surface (Fig. 2c, Region B, orange circles), likely due to adsorption of FFA on the Pd cathode39 (<1% FFA removal by conventional non-electrocatalytic Pd–Pt–CM filtration is shown in Supplementary Fig. 12, and <4.5% FFA removal by N2-saturated Pd–Pt–CM electrofiltration is shown in Supplementary Fig. 13).Decreasing the oxygen level in the feed solution via nitrogen purging substantially reduced the triplet peak of TEMPO (Fig. 2b), suggesting oxygen to be the precursor for the electro-generated 1O2. Other ROS potentially produced in situ via the sequential electro-redox reaction during filtration, including hydrogen peroxide (H2O2), superoxide (O2•−), and hydroxyl radicals (•OH), could be involved as critical intermediates for 1O2 production32,53 (detailed ROS measurement provided in the Supplementary methods). Hydrogen peroxide (H2O2) was generated at the vicinity of the Pd–CM surface (Fig. 2d), which we attribute to the two-electron cathodic reduction of O250,54. H2O2 tended to accumulate above the Pd–CM surface, as evidenced by the higher H2O2 concentration in Region B (Fig. 2d, orange circles) than that in Region A (Fig. 2d, gray squares). A near twofold decrease of H2O2 in the permeate (Fig. 2d, blue triangles) indicates that H2O2 was further consumed within the Pd–Pt–CM inner-pore structure. In contrast, superoxide (O2•−) was more notably produced within the Pd–Pt–CM, as evidenced by the higher concentration in the permeate (Fig. 2e, blue triangles) as compared with the other regions (Fig. 2e, orange circles and gray squares). O2•− accumulation within the Pd–Pt–CM could be attributed to two pathways: (i) one-electron reduction of O2 via the Pd cathode50 and (ii) one-electron oxidation of H2O2 via the Pt anode54,55. In addition, free hydroxyl radicals (•OH) were not detected within the system (Fig. 2f); we attribute this observation to the intrinsic nature of Pt as an “active” anode, which exhibits low production of physisorbed Pt(•OH)56.Efficacy and energy efficiency of electrocatalytic 1O2 productionThe efficacy of the electrocatalytic Pd–Pt–CM for water decontamination was investigated. Sulfamethoxazole (SMX), an antibiotic molecule, was used as a model organic compound, which can be specifically attacked by 1O2 via the electron-rich aniline group57 (Fig. 3a). As shown in Fig. 3b, the Pd–Pt–CM (i.e., mode I) exhibited an ultra-efficient SMX removal (82.9%) during a single-pass electrofiltration (detailed SMX measurement provided in the Supplementary methods). The residence time within the membrane is ~23 s (Supplementary Eq. 3), which is two orders of magnitude lower than the reaction times in current flow-by EC configurations for water decontamination21,58. Notably, the uncompromised membrane permeability (Supplementary Fig. 14) indicates negligible fouling during the electrofiltration, further ensuring durable and unimpeded electron transfer within the membrane (as evidenced by the stable current curve in Supplementary Fig. 15). The electrocatalytic performance and stability of the Pd–Pt–CM were also well retained, as shown in Supplementary Fig. 16.Fig. 3Effectiveness and energy efficiency of 1O2 in situ generated by the Pd–Pt–CM.a Schematic illustrating the reaction between 1O2 and SMX. b SMX removal efficiency and electric energy consumption by Pd–Pt–CM under different electrocatalytic and conventional filtration modes. Modes I–III are for flow-through electrofiltration. For mode, I, the Pd–CM and Pt–CM surfaces of the Pd–Pt–CM serve as cathode and anode, respectively. Modes II and III apply each side of the Pd–Pt–CM as the cathode (mode II) or anode (mode III) and a Ti mesh plate as the counter electrode inserted in the feed chamber (mode II) or permeate chamber (mode III). Modes IV and V are electrocatalytic flow-by modes using the Pd- and Pt-functionalized surfaces as the cathode and anode, respectively. The reaction times of modes IV and V are the same as the filtration duration in mode I (i.e., 38 min). Mode VI and VII are conventional filtration modes using Pd–Pt–CM and CM without applying voltage. For each filtration mode, the feed solution contains 10 μM SMX and 100 mM Na2SO4. Error bars represent the standard deviation from triplicate experiments. In the schematic above each bar, the orange layer and purple layer represent Pd–CM and Pt–CM surfaces, respectively, and the blue arrows denote the direction of water flow. c ROS quenching by applying 10 mM TPA, 10 mM p-benzoquinone (pBQ), 2 mg L−1 catalase, and 10 mM sodium azide (NaN3) as quenchers for •OH, O2•−, H2O2, and 1O2, respectively. Measurements were performed in mode I with the respective quenching agent in the feed solution. Error bars represent the standard deviation from triplicate experiments. d CV curves of the Pd–CM surface in 100 mM Na2SO4 electrolyte at a scan rate of 0.1 V s−1. e CV curves of the Pt–CM surface in 100 mM Na2SO4 electrolyte at a scan rate of 0.1 V s−1.To validate the role of 1O2 and other ROS in the efficient SMX removal during the Pd–Pt–CM electrofiltration (i.e., mode I), we have conducted quenching tests using different ROS scavengers. It is well established that sodium azide (NaN3), p-benzoquinone (pBQ), catalase, and terephthalic acid (TPA) are specific scavengers for 1O2, O2•−, H2O2, and •OH, respectively7,59–61. The efficient quenching using NaN3 confirms the main contribution of 1O2 to SMX removal in mode I (Fig. 3c). The oxidation products of SMX were also examined and were found to be consistent with oxidation products of SMX by 1O2 oxidation57 (Supplementary Fig. 17). Considering that 1O2 reacts with SMX at a minimal molar ratio of 1:157, the minimum 1O2 yield by the Pd–Pt–CM can thus be calculated as 6.9 mmol per m3 of permeate, and the 1O2 yield rate is calculated as 0.3 μM s−1 accordingly (Supplementary Eqs. 4 and 5). In addition, given the considerably low reactivities of O2•− and H2O2 with SMX (Supplementary Fig. 18), the observed quenching by pBQ and catalase suggests that O2•− and H2O2 are critical intermediates to 1O2 production. The insignificant quenching by TPA indicates that free •OH plays a negligible role in SMX removal in mode I. In addition, N2-saturated filtration with the Pd–Pt–CM significantly suppressed SMX removal, in agreement with the previous EPR result (Fig. 2b), indicating oxygen to be the source of the in situ generated 1O2.We also investigated the contribution of direct electro-redox reactions on either side of the Pd–Pt–CM toward SMX removal via cyclic voltammetry (CV) measurements. At the cathodic Pd–CM surface (Fig. 3d), the distinctive CV peak (blue curve) and enhanced current density in the O2-saturated electrolyte compared with that in the N2-saturated electrolyte (black curve), corroborate the generation of H2O2 via oxygen reduction at the Pd–CM cathode. The addition of SMX to the Pd–CM cathode (Fig. 3d, orange curve) caused no obvious difference in CV curves, indicating inert interaction of SMX in direct electro-reduction. Similarly, for the anodic Pt–CM surface (Fig. 3e), the indiscernible difference in the CV curves following the addition of SMX suggests an insignificant influence of direct electro-oxidation towards SMX removal.The efficacy of the Janus electrocatalytic Pd–Pt–CM (i.e., mode I) was further compared with those of other EC configurations, i.e., single-sided electrocatalytic flow-through Pd–Pt–CM (mode II and III) and electrocatalytic flow-by modes (mode IV and V) (detailed mode schematics in Supplementary Fig. 19). As we observe in Fig. 3b, mode I exhibited an over the threefold and sevenfold increase of SMX removal rate compared with mode II and mode III, respectively, highlighting the paramount importance of the Janus configuration of the electrocatalytic Pd–Pt–CM in enhancing SMX removal compared to single-sided electrocatalytic membrane configurations. Compared to mode I, mode IV and V demonstrated notably lower SMX removal rates of <10% by the Pd–Pt–CM, indicating an order of magnitude lower efficiency of flow-by mode than that of flow-through mode. Without adding electricity, conventional membrane filtration in modes VI and VII showed negligible SMX removal. In addition, following a similar trend as the SMX removal efficiency, mode I showed enhanced current density compared with modes II–V (Supplementary Fig. 15), which corroborates the improved transfer of electrons within the confined Janus Pd–Pt–CM porous structures. Taken together, these results collectively emphasize the key role of the electrocatalytic flow-through configuration using both sides of the Pd–Pt–CM as electrodes (i.e., mode I) for highly efficient 1O2 production and enhanced electron transfer in a single-pass filtration.Electric energy consumption is another important consideration in the application of the electrocatalytic membrane. As observed in Fig. 3b (right axis, orange filled squares), mode I exhibited slightly higher energy consumption (detailed calculations in the Supplementary methods), i.e., 13.3 Wh m−3, than those in modes II–V (orange filled squares) mainly due to the higher current density. However, when taking the removal efficacy into account, the specific energy consumption of mode I (i.e., 6.33 Wh g−1 SMX) was significantly lower than those of modes II–V (Fig. 3b right axis, blue filled triangles), indicating an enhanced energy efficiency of the Janus electrocatalytic Pd–Pt–CM. We further compared the energy consumption of the Pd–Pt–CM (in mode I) with other reported electrocatalytic membranes for similar use of water decontamination (Supplementary Fig. 20). Remarkably, our Pd–Pt–CM (in mode I) had a substantially lower energy consumption than those reported in other studies.DiscussionBased on our findings discussed above, we propose the following mechanism for the in situ 1O2 production (Fig. 4). Due to the lack of TEMP-1O2 EPR fingerprint (Fig. 2b, blue lines) and the inhibited SMX removal (Fig. 3c, yellow diamond) in the absence of O2, we confirm that the presence of the ground state O2 is requisite for producing 1O2 by the Pd–Pt–CM. Differing from the general route of photosensitization, where sensitizers with the excited-triplet structures directly transfer energy from excitons to the ground state O2 for 1O2 generation8,14, the electrocatalytic Pd–Pt–CM does not serve as a medium for the direct activation from the inert O2 to 1O2. Alternatively, the Pd–Pt–CM allows indirect routes for 1O2 synthesis with the ROS H2O2 and O2•− as critical intermediates, as confirmed by the inhibited SMX removal by quenching these ROS (Fig. 3c, orange circle, and purple triangle).Fig. 4Proposed mechanism for 1O2 production by the electrocatalytic Pd–Pt–CM filtration.a Schematic describing the various regions in the Pd–Pt–CM. b Schematic of the cathodic Pd-functionalized (black and orange core–shell balls) membrane region, where H2O2 (green balls) and O2•− (yellow balls) are generated via oxygen reduction in close vicinity to the Pd-functionalized pores. c Anodic Pt-functionalized (black and purple core–shell balls) membrane region, where H2O2 consumption, O2•− accumulation, and 1O2 (red balls) generation occur concurrently. d A zoom-in close-to-bottom Pt-functionalized channel, where 1O2 is released throughout the permeate flow. e Proposed reaction pathway of singlet oxygen production. The orange and purple arrows represent the reactions occurring in the cathodic reduction region and anodic enhanced region, respectively. Dashed lines linked to the electric power supply denote the electron transfer.Figure 4a–d depict three functional regions in the porous Pd–Pt–CM undergoing a flow-through electrofiltration process (Fig. 4a) and the corresponding spatiotemporal distribution of ROS (Fig. 4b–d). H2O2 is first generated by an oxygen reduction reaction (ORR) in the cathodic Pd–CM region (Fig. 4b), followed by O2•− accumulation chiefly by the subsequent oxidation of H2O2 in the anodic Pt–CM region (Fig. 4c)50,54,55. Benefiting from the continuous production of H2O2 and O2•−, the Pd–Pt–CM triggers two possible pathways for 1O2 formation: (i) the Haber–Weiss reaction between O2•−/HO2• and H2O2 (Fig. 4e, route ①) and (ii) the recombination of O2•−/HO2• (Fig. 4e, route ②)32,53,62. EPR tests further confirm the vital role of O2•− and H2O2 in 1O2 formation (Supplementary Figs. 21 and 22). Although an •OH radical-mediated route can also be feasible for 1O2 formation32,53 (Fig. 4e, route ③, dash line), this pathway is negligible due to the inhibition of the Pt anode for free •OH generation, as Pt with its low oxygen evolution overpotential readily transforms the generated physisorbed Pt(•OH) into PtO superoxide56. This is evidenced by the absence of detected •OH signal during electrofiltration (Fig. 2f) and the marginal inhibition of TPA (•OH scavenger) for SMX removal (Fig. 3c, blue triangle).The formation of O2•− intermediates by the H2O2 oxidation at the anodic Pt–CM region, as compared with that by the ORR pathway at the cathodic Pd–CM region, is critical for 1O2 synthesis in the Pd–Pt–CM (Fig. 4e, thick lines). As the Pd cathode favors the formation of H2O2 and H2O via a serial two-plus-two-electron transfer in ORR, its generation of O2•− intermediates from the one-electron transfer is minor63,64. This proposed analysis is also verified by tuning the electrofiltration modes. By shifting from the use of only cathodic Pd–CM (Fig. 3b, mode II) to the use of sequential cathodic Pd–CM and anodic Pt–CM (Fig. 3b, mode I), a significant increase in the removal rate from 27.0 to 82.9% was observed, corroborating the key role of the anodic Pt–CM in contributing for O2•− formation. Additionally, switching the electro-redox sequence, i.e., applying Pd–CM as anode facing the feed solution and Pt–CM as cathode facing the permeate, fails to produce 1O2 (Supplementary Figure 23). These results collectively validate that the electro-redox sequence, which induces sequential generation of H2O2 and O2•−, is critical for the 1O2 formation in the Pd–Pt–CM.The distinct material structure of the Pd–Pt–CM, endowed by the confocal magnetron sputtering technique, offers additional mechanistic insights into the favorable 1O2 synthesis. One important structural feature achieved by the sputtering is the high dispersion of Pd and Pt nanoparticles (Fig. 1g), which prevents the formation of large agglomerates on the CM substrate65. This feature is also evidenced by the negligible changes in permeability and pore size distribution of the Pd–Pt–CM compared to the pristine CM (Fig. 1h). The high dispersion of Pd nanoparticles renders the favorable electrocatalytic formation of H2O2 over H2O in the ORR pathway, which is critical for inducing the subsequent ROS chain reactions (Fig. 4e, thick lines). Given the latter H2O product in ORR is formed by a desorption–readsorption of the H2O2 intermediate on a proximate Pd reactive site64,66, the increased interparticle distances of Pd nanoparticles imparted by sputtering prevent H2O2 from subsequent reduction by adjacent Pd agglomerates. Another important structural feature achieved by the Pd and Pt sputtering is the large sputtered depth in the near-surface regions of both sides of the Pd–Pt–CM (i.e., 60 and 90 μm for Pd–CM and Pt–CM regions from Fig. 1a, c). These sputtered depths are over an order of magnitude greater than the inner-porous reactive lengths of reported electrocatalytic membranes67,68. Taken together, the high dispersion of the electro-catalysts, as well as the elongated reactive regions, guarantee maximal exposure of the reactive sites in the Pd–Pt–CM. Furthermore, with the presence of the (100) facet of the Pt nanoparticles, the Pt anodic membrane region inherently favors the spin state of 1O2, thus likely facilitating 1O2 formation by promoting electron transfer between 1O2 precursors adsorbed on the Pt nanoparticles61,69,70 (for specific reaction steps see Supplementary Table 1).Enhanced mass transfer induced by the flow-through electrofiltration is another important feature to promote in situ 1O2 formation. The forced convection, together with the confinement of the solution within the membrane inner-pores, synergistically enhances the transfer of electrons and reactants at the membrane-solution interface33,71. This synergy also facilitates the intimate contact of reactants to maximally exploit their redox capacity for 1O2 synthesis72. In addition, we carried out an SMX removal test in a batch mode without the forced filtration using the same electrocatalytic Pd–Pt–CM configuration as mode I. As shown in Supplementary Fig. 24, an over tenfold decrease of SMX removal efficiency (7.4%) and current density (0.006 mA cm−2) compared with those of mode I was observed, thus corroborating the enhanced mass and charge transfer by convection within the confined porous membrane regions.In summary, we have developed a Janus electrocatalytic membrane, which leverages sequential reduction–oxidation reactions in distinct membrane regions for ultra-efficient 1O2 synthesis via a flow-through electrofiltration. This automated, sustainable and scalable approach of highly efficient 1O2 production and utilization has promising potential in diverse applications, including targeted water purification and in situ sensing in environmental remediation, green organic synthesis, and biomedical engineering. Moreover, we envision that the proposed Janus electrocatalytic membrane geometry, which is radically different from the general assembly of incorporating half-cell reaction with a membrane substrate, will pave the way for future smarter design strategies for high-performance and multifunctional electrocatalytic membranes.MethodsFabrication of Pd–Pt–CMThe Janus membrane (Pd–Pt–CM) was prepared via confocal magnetron co-sputtering (AJA International ATC 2200). Pd and Pt nanoparticles were sputtered on the feed- and permeate-side surfaces of a ceramic membrane (CM) substrate from respective targets with purity ≥99.95%. The sputtering targets were aligned in tetrahedral configuration with an angle and distance to the CM substrate set as 30° and 18 cm, respectively. A base pressure was retained at below 10−7 Pa prior to sputtering. Ultrapure argon gas was then employed for providing a working pressure of 0.3 Pa to eliminate possible contamination in the sputtering chamber. Subsequently, we placed the CM substrate on a silicon holder with its feed-side surface facing to the Pd target, followed by initiating Pd sputtering at a power of 30 W with a deposition rate of 2.1 nm min−1. The sputtering of Pt on the permeate-side surface of CM was then conducted at identical power with a deposition rate of 2.3 nm min−1. Deposition thicknesses for the Pd- and Pt-functionalized surfaces were controlled as ~30 nm by a quartz thickness gauge meter positioned in the center of the sputtering chamber.Characterization of Pd–Pt–CMSEM (SU8230, Hitachi) coupled with X-ray energy dispersive spectroscopy (EDS, XFlash 5060FQ, Bruker) was employed to investigate the morphology and elemental distribution of the Pd–Pt–CM and the pristine CM substrate. The roughness of the Pd–CM surface and Pt–CM surface was measured by atomic force microscopy (Dimension Fastscan, Bruker) and a Zygo Nexview 3D optical profiler (AMETEK), respectively. Grazing incidence X-ray diffraction (GI-XRD, Smart lab, Rigaku) was applied for evaluating the crystal structures of the Pd–Pt–CM and the CM substrate. X-ray photoelectron spectroscopy (XPS, VersaProbe II, PHI) was used to characterize the elemental composition of the functionalized membrane surfaces. Membrane pore size distribution was measured by AutoPore V—mercury intrusion porosimetry. The water contact angle was measured by the sessile drop method using a contact angle goniometer (OneAttension, Biolin Scientific).Electrofiltration procedureElectrofiltration experiments were conducted using a cross-flow membrane filtration system (Supplementary Fig. 8). The membrane system consists of a 12-mL feed chamber and a 12-mL permeate chamber. Pd–Pt–CM was inserted in-between the feed and permeate chambers; Pd–CM surface faced the feed chamber, while Pt–CM surface faced the permeate chamber. The Pd–CM and Pt–CM surfaces serving as cathode and anode, respectively, were connected to a DC supply (E3617A, HEWLETT PACKARD) via carbon tape. Porous titanium (Ti) mesh plate with an identical surface area as Pd–Pt–CM (17.34 cm2) was placed in each chamber and used as a counter electrode in specific flow-through modes (i.e., mode II and III in Fig. 3b). A constant voltage of 1.6 V was applied to the Pd–Pt–CM throughout the filtration experiments, and a multimeter (Fluke 87-V, Everett, WA) was used for current measurement.Experiments were carried out in flow-through and flow-by modes. For each mode, feed solution was circulated between the feed chamber (i.e., Region B in Fig. 2a) and feed reservoir (i.e., Region A in Fig. 2a) at a flow velocity of 0.8 L min−1 by a gear pump (Cole-Parmer Instrument Company, Vernon Hills, IL, USA). In flow-through mode, the feed solution flowed through the membrane at a constant transmembrane pressure of 0.1 bar, while in the flow-by mode, the feed solution flowed tangent to the membrane surface.Electrochemical measurementsElectrochemical performance of the Pd–Pt–CM was analyzed by an electrochemical workstation (CHI 660E, CH Instruments) using a typical three-electrode configuration electrochemical cell. We applied one membrane surface as the working electrode, a platinum wire as the counter electrode, and an Ag/AgCl electrode as the reference electrode, in 100 mM Na2SO4 solution. EIS for each membrane surface was conducted by applying frequencies varied from 106 to 1 Hz. CV curves of each membrane surface were collected at a scan rate of 0.1 V s−1 in either O2-saturated or N2-saturated Na2SO4 solution (100 mM). Pure nitrogen or oxygen gas was used to purge the electrolyte for 1 h prior to the tests.Supplementary information
Supplementary Information | nature communications | [
"Article"
] | [
"Pollution remediation",
"Electrocatalysis",
"Chemical engineering"
] | work H. Kautsky 1938 revealed molecular oxygen singlet oxygen (1O2) pivotal in chemical biomedical fields1–3 nonradical derivative oxygen hydroxyl radicals most potent among reactive oxygen species (ROS)4,5 short-lived •OH radical meta-stable 1O2 unoccupied π* orbital shows high selectivity toward electron-rich substances6,7 pharmaceuticals8 unsaturated biomolecules7,9 microbial pathogens10 1O2 targeted reactions applied in diverse fields water decontamination10 photodynamic cancer therapy11 green organic synthesis12 importance 1O2 motivated research for effective 1O2 production approaches 1O2 production include photosensitization enzymatic reactions biological systems pH temperature conditions intracellular approaches challenges electrocatalysis (EC), automated more approachable 1O2 production EC enables flexible generation of critical 1O2 precursors hydrogen peroxide22 hypochlorite20 alternative to EC provides efficient electro-excitation to 1O2 precursors.EC less explored process for 1O2 generation photosensitization enzymaticEC pathways rely high dosages precursors compromise environmental sustainability oxidation molecules 1O2 EC requires long reaction times consumes high electric critical EC efficacy eliminating chemicals shortening residence time enhancing Faradaic efficiency.Porous membranes ideal electrode substrates enhance EC performance28 structures proposed horizontally stacked lamellar channels30 vertically aligned straight-through pores31 substrates confine reactants electrified active sites micrometer nanometer-scale confinement allows optimal utilization electrons minimal diffusion distance mass transfer overcoming chemical usage conventional EC systems studies importance flow-through configuration membrane electrodes electrons flow parallel fluid flow direction charge transfer rates electrocatalytic membranes render half-cell reaction reduction anodic oxidation electrode redox reactions not utilized electrocatalytic flow-through porous membrane configuration cathodic reduction anodic oxidation innovative strategy 1O2 synthesis enhanced Faradaic energy efficiency conceive Janus electrocatalytic membrane ultra-efficient 1O2 production without chemical precursorsJanus membrane (Pd–Pt–CM features double-sided electrochemically reactive surfaces sputtering palladium platinum (Pt nanoparticles ceramic substrate confirmed co-existence 1O2 precursors investigated distribution during electrocatalytic filtration Pd- Pt-sides cathode anode evaluated efficacy generated 1O2 for water decontamination using sulfamethoxazole proposed mechanism stepwise electrocatalytic production 1O2 Pd–Pt–CM sequential ROS-mediated redox chain reactions membrane regions findings demonstrate new paradigm electrochemical material platform for 1O2 production Janus electrocatalytic flow membrane properties of electrical insulating properties large ideal cathodic anodic electrodes sequential electro-redox reactions inner-porous CM regions filtration Pd–Pt–CM synthesized via confocal magnetron co-sputtering Palladium platinum nanoparticles sputtered on feed- permeate-side surfaces CM thickness 2.5 Noble metal nanoparticles high electrical conductivity efficient EC reactions. 1Fabrication characterization Pd–Pt–CMSchematic fabrication Pd–Pt–CM magnetron sputtering SEM cross-section images Pd–CM Pt–CM surfaces EDS mapping Zr O Ti Pd Pt (purple elements insets illustrate SEM surface images functionalized membranes EDS mapping Pd Pt Nyquist plots Pd–CM Pt–CM surfaces CM substrates 100 mM Na2SO4 solution frequencies 106–1 Hz XPS spectra Pd 3d Pt 4f Pd–Pt–CM Grazing incidence XRD patterns Pd–CM Pt–CM surfaces CM substrates Pore size distributions Pd–Pt–CM circles pristine CM (blue water fluxes Pd–Pt–CM pristine CM transmembrane pressure bars standard deviation color transformations Pd–CM Pt–CM surfaces metal deposition Figure 1b c Pd–Pt–CM microscopy) elemental mapping X-ray EDS Sputtered Pd Pt nanoparticles penetrated membrane surfaces EDS mapping results Figs. 2 3) sputtered depths 60 μm 90 μmdeep inner-pore sputtering guarantees EC reactions within membrane structure larger sputtered depth of Pt–CM Pd–CM attributed to asymmetric porous architecture pristine CM Pd Pt nanoparticles uniformly deposited on surfaces Pd–Pt–CM even distribution EDS signals unimpeded electron transfer during EC reactions properties Pd–Pt–CM quantified by electrochemical impedance spectroscopy observed 6-fold 40-fold decrease in amplitudes semicircles for Pd–CM Pt–CM surfaces pristine CM sputtering of Pd Pt nanoparticles reduces charge transfer resistance facilitates electron transfer membrane Pt–CM surface lower charge transfer resistance than Pd–CM higher intrinsic conductivity Pt larger sputtered elemental composition crystalline structure of Pd–Pt–CM investigated by X-ray spectroscopy X-ray diffraction XPS peaks of Pd 3d3/2 Pd 3d5/2 at 340.2 335.1 eV suggest Pd0 dominant minor Pd2+ on Pd–CM possibly due to surface oxidationdeconvoluted peaks Pt 4f5/2 4f7/2 energies 74.8 71.5 eV (Fig. 1f indicate predominance Pt0 partially oxidized Pt2+ Pt–CM (XPS survey spectra Supplementary Fig. 5) Pt2+ signal oxidation Pt0 anodic membrane information <10 nm past bulk sputtered Pt anodic membrane mainly Pt0 thermodynamically stable Pt XRD analysis Pd–CM surface (Fig. 1g exhibits diffraction peaks (111) (200) planes Pd0 face-centered cubic crystal structure at 39.2° 45.5° other peaks correspond rutile zirconium oxide CM substrate presence crystalline Pt Pt–CM surface. 1g diffraction peaks at 39.7° 46.5° 67.7° consistent with (111) (220) planes Pt0 mean nanoparticle sizes Pd Pt calculated 20.1 22.8 nm high dispersion nano-sized Pd Pt broadening XRD diffraction peaks pristine CM grains analyzed influence sputtering on membrane performance pore size distributionsPd–Pt–CM pristine CM suggest negligible effect sputtering on membrane structure Fig. 1h Pd–Pt–CM comparable water flux pristine CM negligible influence sputtering on membrane permeability reduced surface roughness Pd–Pt–CM pristine CM seeding Pd Pt nanoparticles smooths surface Pd–Pt–CM hydrophilic surface properties pristine CM similar water contact angles (21°) (19°) 7) smooth surface hydrophilic nature Pd–Pt–CM suggest lessened fouling potential organic efficient electron transfer during filtration distribution of 1O2 ROS in Pd–Pt–CMElectrocatalytic filtration customized cross-flow filtration device Fig. Pd–Pt–CM mounted feed chamber permeate chamber Pd–CM surface cathode faced feed chamber Pt–CM surface anode permeate feed solution circulated reservoir cross-flow velocity 0.8 L min–1 under pressure 0.1 bar.Fig. distribution of reactive oxygen species (ROS) Pd–Pt–CMSchematic cross-flow filtration system feed reservoir A electrofiltration module 12 mL feed chamber B 12 mL permeate chamber C Pd–Pt–CM EPR analysis 1O2 regions 2,2,6,6-tetramethylpiperidine 25 mM trapping agent measurement TEMP-1O2 permeate C N2-saturated feed solution 1O2 generation degradation furfuryl alcohol 50 H2O2 detection Amplex Red reaction Detection O2•− 2,3-Bis-(2-methoxy-4-nitro-5-sulfophenyl)-2H-tetrazolium-5-carboxanilide 100 μM probe •OH measurement terephthalate 1 mM •OH hydroxyterephthalate Error bars standard deviations experiments detected singlet oxygen (1O2) generation Pd–Pt–CM electrofiltration Na2SO4 electrolyte feed spectra trapping 2,2,6,6-tetramethylpiperidine triplet signal (1:1:1 16.9 G 2,2-tetramethyl-4-piperidinol-N-oxyl Region C 1O2 Detailed EPR measurement Supplementary methods control solution100 mM Na2SO4 25 mM TEMP Supplementary Fig. 9. lack TEMPO signals Regions A B 1O2 generated Pd–Pt–CM inner-pore electrocatalytic filtration minimum voltage 1.6 V 1O2 Fig. side reactions intensified high voltage Faradaic minimum voltage 1.6 V experiments 1O2 production confirmed furfuryl alcohol (FFA) probe51 selective 1O2 rate 1.2 × 108 M−1 s−1 17% degradation FFA (50 μM) Pd–Pt–CM 1.6 V. 2c Region C constant production 1O2 membrane detected products FFA agree oxidation products FFA 1O2 Fig. 11 less than 3% removal FFA Pd–CM surface. B adsorption FFA Pd<1% FFA removal conventional non-electrocatalytic Pd–Pt–CM filtration Supplementary Fig. 12 <4.5% FFA removal N2-saturated Pd–Pt–CM electrofiltration Fig. 13).Decreasing oxygen level feed solution nitrogen purging reduced triplet peak TEMPOsuggesting oxygen precursor for electro-generated 1O2. ROS produced electro-redox reaction filtration hydrogen peroxide (H2O2) superoxide hydroxyl radicals critical intermediates for 1O2 production32 (detailed ROS measurement Supplementary Hydrogen peroxide (H2O2) generated Pd–CM surface (Fig. two-electron cathodic reduction O250 H2O2 above Pd–CM surface higher H2O2 concentration Region B Region A twofold decrease H2O2 in permeate indicates H2O2 consumed Pd–Pt–CM inner-pore structure superoxide (O2•−) produced within Pd–Pt–CM higher concentration in permeate 2e O2•− accumulation Pd–Pt–CM to two pathways one-electron reduction O2 via Pd cathode50 one-electron oxidation H2O2 via Pt anode54 free hydroxyl radicals (•OH) not detected system (Fig. Pt “active” anode low production physisorbed Pt(•OH)56Efficacy energy efficiency electrocatalytic 1O2 electrocatalytic Pd–Pt–CM water decontamination investigated Sulfamethoxazole antibiotic molecule model organic compound attacked 1O2 electron-rich aniline group57 (Fig. Pd–Pt–CM mode I ultra-efficient SMX removal (82.9%) single-pass electrofiltration residence time membrane ~23 s two lower reaction times current EC configurations water uncompromised membrane permeability 14 negligible fouling electrofiltration durable electron transfer stable current curve electrocatalytic performance stability Pd–Pt–CM retained Fig. 16.Fig. 3Effectiveness energy efficiency 1O2 Pd–Pt–CM Schematic reaction 1O2 SMX SMX removal efficiency electric energy consumption Pd–Pt–CM electrocatalytic conventional filtration modes Modes I–III flow-through electrofiltration mode I Pd–CM Pt–CM surfaces serve cathode anodeModes II III apply Pd–Pt–CM cathode anode Ti mesh plate counter electrode Modes IV V electrocatalytic flow-by modes Pd Pt surfaces cathode anode reaction times same filtration duration I 38 VI VII conventional filtration Pd–Pt–CM CM without voltage feed solution contains 10 μM SMX 100 mM Na2SO4 Error bars standard deviation triplicate experiments orange purple layer represent Pd–CM Pt–CM surfaces blue direction water flow ROS quenching 10 mM TPA 10 mM p-benzoquinone 2 mg L−1 catalase 10 mM sodium azide quenchers •OH O2•− H2O2 1O2 Measurements performed mode I agent feed solution Error bars standard deviation triplicate experiments curves Pd–CM surface 100 mM Na2SO4 electrolyte 0.1 V Pt–CM surface 100 mM Na2SO4 electrolyte validate 1O2 ROS SMX removal Pd–Pt–CM electrofiltration conducted quenching tests ROS scavengerssodium azide (NaN3) p-benzoquinone catalase terephthalic acid) scavengers 1O2 O2•− H2O2 •OH quenching NaN3 confirms contribution 1O2 to SMX removal mode I (Fig. oxidation products SMX examined consistent with 1O2 1O2 reacts with SMX ratio 1:157 minimum 1O2 yield Pd–Pt–CM mmol per m3 permeate 1O2 yield rate 0.3 μM s−1 4 low reactivities O2•− H2O2 with SMX quenching by pBQ catalase suggests O2•− H2O2 critical intermediates 1O2 production insignificant quenching by TPA •OH negligible role in SMX removal mode I N2-saturated filtration with Pd–Pt–CM suppressed SMX removal oxygen source generated investigated contribution electro-redox reactions Pd–Pt–CM SMX removal cyclic voltammetry measurements Pd–CM surfacedistinctive CV peak (blue curve enhanced current density O2-saturated electrolyte N2-saturated electrolyte corroborate generation H2O2 via oxygen reduction at Pd–CM cathode addition SMX Pd–CM cathode. 3d orange curve caused no difference CV curves inert interaction SMX direct electro-reduction anodic Pt–CM surface (Fig. difference CV curves addition SMX suggests insignificant influence direct electro-oxidation SMX removal efficacy Janus electrocatalytic Pd–Pt–CM mode I compared with other EC configurations single-sided electrocatalytic flow-through flow-by modes IV V Fig mode I threefold sevenfold increase SMX removal rate II III importance Janus configuration enhancing SMX removal IV V lower SMX removal rates <10% lower efficiency flow-by mode Without adding electricity conventional membrane filtration in modes VI VII showed negligible SMX removal mode I showed enhanced current density compared modes II–V improved transfer electrons within Janus Pd–Pt–CM porous.results emphasize role electrocatalytic flow-through configuration using both sides Pd–Pt–CM as electrodes mode I for efficient 1O2 production enhanced electron transfer in single-pass filtration.Electric energy consumption important consideration electrocatalytic membrane Fig. 3b mode I higher energy consumption 13.3 Wh m−3 than modes II–V due to higher current density specific energy consumption of mode I 6.33 Wh g−1 SMX) lower than modes II–V indicating enhanced energy efficiency Janus electrocatalytic Pd–Pt–CM compared energy consumption Pd–Pt–CM mode I with other electrocatalytic membranes for water decontamination Pd–Pt–CM) lower energy consumption than propose mechanism for in situ 1O2 production (Fig. 4) to lack of TEMP-1O2 EPR fingerprint inhibited SMX removal presence ground state O2 requisite for producing 1O2 by Pd–Pt–CMphotosensitization sensitizers transfer energy to O2 for 1O2 electrocatalytic Pd–Pt–CM for activation inert O2 to 1O2. allows indirect routes for 1O2 synthesis ROS H2O2 O2•− intermediates confirmed by SMX removal ROS.. mechanism for 1O2 production by electrocatalytic Pd–Pt–CM filtration Schematic regions in Pd–Pt–CM cathodic Pd-functionalized membrane region H2O2 O2•− generated via oxygen reduction Anodic Pt-functionalized membrane region H2O2 consumption O2•− accumulation 1O2 generation occur zoom-in close-to-bottom Pt-functionalized channel 1O2 released reaction pathway of singlet oxygen production orange purple represent reactions in cathodic reduction anodic enhanced region Dashed lines electric power supply denote electron transfer.Figure 4a–d three functional regions in Pd–Pt–CM flow-through electrofiltration spatiotemporal distribution of ROSH2O2 generated oxygen cathodic Pd–CM region followed O2•− accumulation oxidation H2O2 anodic Pt–CM region 4c production H2O2 O2•− Pd–Pt–CM triggers two pathways 1O2 formation Haber–Weiss reaction between O2•−/HO2• H2O2 1) recombination O2•−/HO2• EPR tests confirm role O2•− H2O2 in 1O2 formation 21 •OH radical-mediated route feasible for 1O2 3 negligible inhibition Pt anode •OH generation transforms Pt(•OH) into PtO •OH signal during electrofiltration. 2f marginal inhibition TPA (•OH scavenger) for SMX removal 3c formation O2•− intermediates H2O2 oxidation anodic Pt–CM region ORR cathodic Pd–CM critical for 1O2 synthesis Pd–Pt–CM 4e Pd cathode favors formation H2O2 H2O via two-plus-two-electron transfer ORR generation O2•− intermediates from one-electron transfer minor63proposed analysis verified by tuning electrofiltration modes shifting from cathodic Pd–CM to sequential cathodic Pd–CM anodic Pt–CM increase removal rate from 27.0 to 82.9% role anodic Pt–CM O2•− formation switching electro-redox sequence Pd–CM Pt–CM fails produce 1O2 validate electro-redox sequence generation H2O2 O2•− critical for 1O2 formation in Pd–Pt–CM distinct material structure Pd–Pt–CM confocal magnetron sputtering technique offers mechanistic insights 1O2 synthesis high dispersion of Pd Pt nanoparticles prevents large agglomerates on CM negligible changes in permeability pore size distribution Pd–Pt–CM compared to pristine CM high dispersion Pd favorable electrocatalytic formation of H2O2 over H2O ORR pathway critical for ROS chain reactionsthick latter H2O product ORR formed by desorption–readsorption H2O2 intermediate Pd reactive increased interparticle distances Pd nanoparticles sputtering prevent H2O2 reduction by Pd agglomerates Pd Pt sputtering large sputtered depth in near-surface regions Pd–Pt–CM 60 90 μm Fig. 1a greater than inner-porous reactive lengths electrocatalytic membranes67 high dispersion electro-catalysts elongated reactive regions guarantee maximal exposure reactive sites Pd–Pt–CM presence Pt nanoparticles anodic membrane favors spin state 1O2 1O2 formation electron transfer between 1O2 precursors Pt nanoparticles61 Supplementary Table 1).Enhanced mass transfer flow-through electrofiltration 1O2 formation forced convection confinement solution membrane enhances transfer electrons reactants at membrane-solution facilitates intimate contact reactants redox capacity for 1O2 synthesis72 carried SMX removal test batch mode without forced filtration same electrocatalytic Pd–Pt–CM configuration mode I Supplementary Fig.tenfold decrease SMX removal efficiency (7.4% current density (0.006 mA cm−2) mode I observed corroborating enhanced mass charge transfer convection membrane regions developed Janus electrocatalytic membrane leverages reduction–oxidation reactions-efficient 1O2 synthesis flow-through electrofiltration automated sustainable scalable approach potential applications water purification environmental remediation green organic synthesis biomedical engineering proposed Janus membrane geometry different future design strategies high-performance multifunctional electrocatalytic membranes Pd–Pt–CMThe Janus membrane prepared confocal magnetron co-sputtering ATC Pd Pt nanoparticles sputtered feed- permeate-side surfaces ceramic membrane substrate targets purity ≥99.95% sputtering targets aligned tetrahedral angle distance to CM substrate 30° 18 cm base pressure below 10−7 Pa sputtering Ultrapure argon gas employed working pressure 0.3 Pa eliminate contamination placed CM substrate on silicon holder facing Pd target Pd sputtering 30 W deposition rate 2.1 nm min−1sputtering Pt surface CM conducted identical power deposition rate 2.3 nm min−1 Deposition thicknesses Pd- Pt surfaces controlled ~30 nm by quartz thickness gauge meter center sputtering chamber.Characterization Pd–Pt–CMSEM X-ray energy dispersive spectroscopy elemental distribution Pd–Pt–CM CM substrate roughness Pd–CM Pt–CM measured atomic force microscopy Zygo Nexview 3D optical profiler Grazing incidence X-ray diffraction crystal structures Pd–Pt–CM CM substrate X-ray photoelectron spectroscopy elemental composition membrane surfaces Membrane pore size distribution measured AutoPore V—mercury intrusion porosimetry water contact angle measured sessile drop method contact angle goniometer cross-flow membrane filtration system 12-mL feed chamber 12-mL permeate chamber Pd–Pt–CM inserted-between feed permeate chambers Pd–CM feed Pt–CM permeate chamberPd–CM Pt–CM surfaces cathode anode connected to DC supply (E3617A HEWLETT via carbon tape Porous titanium mesh plate identical surface area Pd–Pt–CM (17.34 cm2) in each chamber counter electrode flow-through modes constant voltage 1.6 V Pd–Pt–CM filtration multimeter (Fluke 87-V for current measurement flow-through flow-by modes feed solution circulated between feed chamber feed reservoir velocity L min−1 by gear pump (Cole-Parmer Instrument Company Vernon Hills flow-through feed solution through membrane constant transmembrane pressure 0.1 bar flow-by mode tangent to membrane surface.Electrochemical performance Pd–Pt–CM analyzed electrochemical workstation 660E three-electrode applied membrane surface working platinum wire counter electrode Ag/AgCl electrode reference electrode in 100 mM Na2SO4 solution frequencies 106 to 1 Hz CV curves collected scan rate 0.1 V s−1 in or N2-saturated Na2SO4 solution Pure nitrogen oxygen gas electrolyte 1 h prior tests.Supplementary information | 50.5 | 0.630737 |
10.1038/s41467-020-15052-y | PMC7057995 | RNA polymerase I (Pol I) catalyses the transcription of ribosomal RNA precursors, and its transcription initiation mechanism differs from that of Pol II and Pol III. Here the authors present the cryo-EM structure of a trapped early intermediate stage of promoter-recruited Pol I, which reveals the interactions of the basal rDNA transcription machinery with the native promoter, and discuss the mechanistic implications. | Transcription of the ribosomal RNA precursor by RNA polymerase (Pol) I is a prerequisite for the biosynthesis of ribosomes in eukaryotes. Compared to Pols II and III, the mechanisms underlying promoter recognition, initiation complex formation and DNA melting by Pol I substantially diverge. Here, we report the high-resolution cryo-EM reconstruction of a Pol I early initiation intermediate assembled on a double-stranded promoter scaffold that prevents the establishment of downstream DNA contacts. Our analyses demonstrate how efficient promoter-backbone interaction is achieved by combined re-arrangements of flexible regions in the ‘core factor’ subunits Rrn7 and Rrn11. Furthermore, structure-function analysis illustrates how destabilization of the melted DNA region correlates with contraction of the polymerase cleft upon transcription activation, thereby combining promoter recruitment with DNA-melting. This suggests that molecular mechanisms and structural features of Pol I initiation have co-evolved to support the efficient melting, initial transcription and promoter clearance required for high-level rRNA synthesis. | IntroductionThe transcription of the ribosomal RNA (rRNA) precursor by RNA polymerase (Pol) I is a prerequisite for ribosome biosynthesis in all known eukaryotes1. As such, Pol I transcription is tightly regulated, mostly at the level of pre-initiation complex (PIC) formation2–6. Whereas Pol II and Pol III use related initiation mechanisms, the processes underlying Pol I promoter recognition, PIC formation and DNA melting substantially diverge7–9. In bakers’ yeast Saccharomyces cerevisiae, a basal initiation system required for Pol I activity consists of the promoter DNA core element (CE), specific initiation factor Rrn3 and heterotrimeric core factor (CF)10. CF binds a CE stretch between ~15 and 38 base pairs (bps) upstream of the transcription start site (TSS)11 and recruits Rrn3-stabilized Pol I that is primed for initiation12–14. DNA melting occurs at a position slightly upstream of the TSS between the Pol I ‘clamp core’ and ‘protrusion’ domains15–17. No additional factors are required to commence initial transcription and promoter escape. In a complete system, however, upstream activating factor (UAF) recognizes an upstream element (UE) and cooperates with the TATA-binding protein (TBP) to stabilize CF association with the promoter, increasing Pol I initiation rates by up to 40-fold in vitro18–21. Furthermore, the factor Net1 may reside at Pol I promoters and has been described to enhance initiation in vivo and in vitro22,23.During transcription initiation, Pols are recruited to their promoters by a set of general transcription factors, forming a ‘closed complex’ (CC). After melting of both DNA strands, an ‘open complex’ (OC) is established, transitioning into an ‘initially transcribing complex’ (ITC) with the beginning of RNA chain synthesis. In ITCs, a stable DNA/RNA hybrid is formed and the polymerase has initiated movement into the gene before establishment of a processive elongation complex (‘EC’; for a review of initiation phases compare refs. 24,25). Previous structural analyses of Pol I initiation complexes by us and others relied on an artificially stabilized, mismatched bubble scaffold assembled with an initially transcribed RNA sequence and a double-stranded DNA (dsDNA) sequence extending to up to 24 bps downstream of the TSS15–17. This experimental approach originates from the analysis of Pol II elongation complexes (ECs), preventing heterogenic sample conformations and making use of the tight DNA/RNA hybrid association with the polymerase26,27. In the case of the Pol I PIC a similar experimental strategy results in the visualization of late initiation intermediates. Consequently, an inconsistent occupancy of Rrn3 and divergent localization of the tandem-winged helix (twh) domain of Pol I subunit A49 and the C-terminal domain of subunit A12.2 have been observed15–17, leaving room for speculation with regard to the functional roles and temporal classification of the analyzed conformations during initiation7–9.Therefore, we aimed at analyzing Pol I initiation mechanisms at an early initiation stage, allowing the visualization of promoter recognition, Pol I recruitment and DNA melting in a scenario as close to the native situation as possible. For this purpose, we assembled a complete initiation complex on double-stranded (ds) promoter DNA and performed single-particle cryo-EM analysis. The dsDNA scaffold was truncated on its downstream edge at position +8 relative to the TSS, thus preventing a contact with the clamp core and jaw domains of the polymerase. Three-dimensional particle reconstruction, cryo-EM density refinement and structural modeling allow the placement of basal PIC components and a comparative PIC analysis of the three eukaryotic Pols. Furthermore, structure-guided analysis indicates how Pol-I-specific ribosomal DNA (rDNA) promoter melting may be achieved.ResultsComplex formation and cryo-EM analysisTo study promoter recognition and DNA melting, we formed a complete Pol I initiation complex in vitro. UAF was assembled on a dsDNA promoter scaffold ranging from position −155 to +8 relative to the TSS together with TBP, CF, and a fragment of the protein Net122,28,29 (Methods). Endogenously purified Pol I13,30,31 was pre-incubated with recombinant Rrn332 to reconstitute a complete early PIC that was stable throughout size exclusion chromatography (Supplementary Fig. 1a; Methods). Accordingly, Pol I could be recruited to a UAF/TPB/Net1/CF-bound promoter scaffold lacking sequence stretches required for forming extended downstream contacts with the jaw- and clamp-head domains of the polymerase. Single-particle cryo-EM data was collected on a Titan Krios equipped with Gatan K2 summit direct electron detector basically as described12,13. Following pre-processing, two-dimensional (2D)- and three-dimensional (3D)-classification in RELION33, a total of 122,099 particles were selected from 4,088 micrograph movies (Methods; Supplementary Fig. 1). A final cryo-EM reconstruction exhibits an overall resolution of 3.5 Å and shows a Pol I early intermediate PIC (eiPIC; Fig. 1 and Supplementary Fig. 1). The cryo-EM density clearly reveals secondary structure features for the entire particle and side chain orientations in most regions (Fig. 1c and Supplementary Fig. 2a–f). Despite protein–protein crosslinking, TBP, UAF, and Net1-CTR remain flexible, although apparently stabilizing CF similar to the co-activator ‘mediator’ in context of a Pol II PIC34.Fig. 1Cryo-EM reconstruction of a Pol I early intermediate PIC.a Overview of the Pol I eiPIC cryo-EM reconstruction at 3.5 Å resolution (unsharpened; transparent gray envelope) overlaid with the PDB model (colored ribbon) and DNA (space filing). The right panel shows transparent density (gray) for protein components and solid density for the DNA path (template strand in blue and non-template in light blue). PAD promoter-associated domain (of Rrn11); PIR polymerase interacting region (of CF). b Schematic representation of promoter dsDNA used for PIC assembly, densities observed in the eiPIC reconstruction are highlighted in blue and light blue for template strand and non-template strand, respectively. c Atomic model of the bridge helix in subunit A190 overlaid with sharpened eiPIC density (gray mesh) indicates residue orientations.An early intermediate PIC exhibits a well-defined architectureInitial assignment located template and non-template DNA strands, Pol I, CF subunits, and Rrn3, followed by manual model building and real-space refinement, resulting in a model of high quality (Methods, Table 1). Upstream DNA is well-ordered between CF-interacting regions and entry into the Pol I active center cleft. Following the canonical DNA-path further downstream, however, no density is visible around the active center itself, but ≥12 well-defined base-pairs can be placed on the downstream edge between bridge helix and the clamp-head/jaw domains, even though our scaffold should not extend this far. Most likely the conserved35 and highly charged region is bound by foreign DNA or the far upstream end of our scaffold. A similar effect was observed for patches of the nucleosome, after transcription by Pol II ‘peeled’ off supercoiled DNA36. Well in line, in vitro initiation assays previously showed a strong preference for Pol I to initiate from dsDNA ends of synthetic sequences15.Table 1Cryo-EM data collection, refinement, and validation statistics.eiPIC EMDB-10544 PDB 6TPSCF in eiPIC EMDB-10663Data collection and processing Magnification105,000105,000 Voltage (kV)300300 Electron exposure (e–/Å2)5656 Defocus range (μm)−1.5 to −3.1−1.5 to −3.1 Pixel size (Å)1.09 (0.545 superres)1.09 (0.545 superres) Symmetry imposedC1C1 Initial particle images (no.)311,557311,557 Final particle images (no.)122,099122,099 Map resolution (Å)3.543.91 FSC threshold0.1430.143 Map resolution range (Å)3.3 to 9.93.4 to 16.7Refinement Initial model used (PDB code)6TPS Model resolution (Å)3.5 FSC threshold0.143 Map sharpening B factor (Å2)−75Model composition Non-hydrogen atoms50,070 Protein residues6,109 Ligands8 (Zn and Mg)B factors (Å2) Protein65.6 Ligand102.9R.m.s. deviations Bond lengths (Å)0.009 Bond angles (°)0.985Validation MolProbity score1.85 Clashscore5.96 Poor rotamers (%)0.59Ramachandran plot Favored (%)91.10 Allowed (%)8.75 Disallowed (%)0.15Initiation factor Rrn3 is tightly bound to Pol I ‘stalk’ and ‘dock’ subdomains12,13 in all analyzed particles, agreeing with chromatin immuno precipitation (ChIP) and biochemical studies in yeast2,32,37 and mouse38,39 cells. CF is associated with the Pol I core via its polymerase interacting regions (PIR) similar to ITC conformations15–17. Excellent quality of the cryo-EM density allowed us, to rebuild the CF subunits Rrn6, Rrn7 and Rrn11, consolidating divergent assignments in the crystal structure15 (PDB 5O7X) and an ITC EM-based model17 (PDB 5W66). In contrast to inactive Pol I30,31,40, the ‘expander’ and ‘connector’ subdomains are flexible and the central bridge helix is refolded in the eiPIC (Fig. 1c) as expected from EC structures41,42. The C-terminal domain of subunit A12.2 shows only residual density in funnel domain of subunit A190 (Supplementary Fig. 2b), but is not localized on the A135 lobe as observed in a 12-subunit EC43. Our eiPIC reconstruction shows strong density for the A49/A34.5 dimerization and A34.5 C-terminal tail domains (Supplementary Fig. 2e), indicating that the heterodimer is constitutively attached. The twh and linker domains of subunit A49 are detached in the eiPIC, agreeing with a proposed role in promoter escape17.Core factor embraces the promoter DNAThe eiPIC density allows the construction of a CF model, which we found to resemble the overall ITC conformation. To define the structural changes that take place upon promoter recruitment, we compared the architecture of CF in free (PDB 5O7X) and promoter-engaged eiPIC conformation (Supplementary Fig. 3). This shows that CF module I and II retract from each other by up to 12 Å upon binding of the CE promoter sequence. This retraction leads to the exposure of positively charged residues that are now free to engage the phosphate backbone (Supplementary Fig. 3a–c). These DNA-binding regions lie within the Rrn11 promoter-associated domain (‘PAD’) and the cyclin domains of Rrn7. The same regions engage the DNA in ITCs15–17 and have been described in detail in late ITCs devoid of Rrn317. Remarkably, the Rrn7 residues involved in DNA-binding are not conserved within TFIIB or Brf1, which share a similarity in their overall fold44–46 and would clash with TBP15 in canonical TFIIB-TBP47 or Brf1-TBP48,49 complex.Comparison of free and promoter-engaged CF also shows that the Rrn7-specific helix α4a in the N-terminal cyclin domain shifts and is inserted into the minor groove of the CE promoter DNA, while loop α7-α8 in cyclin II becomes well-structured and contacts the major groove further upstream upon eiPIC formation (Fig. 2a). Thereby, the distal upstream DNA-path is modified towards the C-terminal domain of Rrn7 and the β-propeller-domain of Rrn6. Thus, promoter binding by Rrn7-specific regions on one face and by the TFIIB-unrelated CF subunit Rrn11 on the opposite face tightly squeeze the DNA. This may explain why the basal Pol I initiation system does not require TBP association opposite of the Rrn7 cyclins.Fig. 2Core Factor—promoter interactions in eiPIC.a Model of promoter-bound CF in the eiPIC. The same regions of Rrn7 and Rrn11 contribute to promoter phosphate backbone interactions compared to ITC reconstructions. b Electrophoretic mobility shift assay (EMSA) shows that wild-type CF interacts with double-stranded promoter DNA (0.25 pmol, 0.5 pmol, and 1 pmol CF added). Mutation of Rrn7 (Δα4a and Δloop α7-α8) does not impair promoter-DNA association. c In contrast to DNA binding, initiation efficiency of CF assembled with Rrn7 mutants Δα4a and Δloop α7-α8 is impaired (promoter-dependent in vitro transcription assay from a minimal scaffold).To address the importance of these residues, we constructed CF mutants with deletions in helix α4a and in loop α7-α8. Both can still associate with promoter DNA (Fig. 2b), but show defects in basal initiation in vitro (Fig. 2c). Engagement of these regions may therefore be important to induce a specific DNA conformation required for Pol I recruitment or promoter melting.The Pol I ‘sandwich’ region is important for PIC formationWe have previously described a Pol-I-specific proximal upstream promoter-binding region consisting of loop α11a-α12 (residues 452–456) and the loop β28-β28 (residues 815–818) in the protrusion and wall domains of Pol I subunit A135, respectively15. In the eiPIC, a positively charged loop (892–895, wall domain of subunit A135) re-orients towards the promoter DNA, contributing additional phosphate-backbone interactions (Supplementary Fig. 4) similar to other ITC/PIC structures16,17. These promoter interactions are all specific to Pol I, because the residues are not conserved in Pol II50 and III51. Furthermore, DNA is occluded from the corresponding region in Pol II and III PICs by the N-terminal cyclin domains of TFIIB34,52,53 and Brf1/Brf248,49,54,55, respectively. Fittingly, this Pol I region was previously named ‘sandwich’17.In the eiPIC, the sandwich region tightly holds the promoter in place between the wall and protrusion domains at the bottom of the cleft. sandwich elements contact both DNA strands, therefore rendering it specific for an un-melted duplex. Density for the DNA directly downstream of the sandwich is not observed, indicating a higher degree of flexibility. Consequently, the recruitment of the Pol-I-Rrn3 complex seems to mainly rely on (1) contacts between the promoter and the sandwich and (2) protein–protein contacts between CF and the Pol-I-Rrn3 complex. In contrast, further promoter contacts with the Pol I cleft or downstream elements and/or A49 appear not to be required for recruitment.TFIIB-related elements in Rrn7 adopt divergent positionsThe TFIIB-related ‘reader’ and ‘linker’ elements within Rrn744,46 are mostly ordered in the active center cleft of the eiPIC, with the exception of the residues 46–56 (B-reader homologous56). The protein backbone extends from the N-terminal zinc ribbon into the Pol I cleft, apparently trapping the well-ordered ‘lid’ subdomain of Pol I subunit A190 before forming two anti-parallel strands and exiting the Pol I upstream face on the side of the shelf module (Fig. 3a). The path of Rrn7 differs from a Pol I ITC17 and from TFIIB in complex with Pol II57 (Supplementary Fig. 5). During Pol II initiation, the TFIIB-reader-loop contacts the ‘rudder’ and the ‘fork loop I’ domains, while the TFIIB-linker binds the top of the rudder and forms a helix that interacts with the clamp core domain57. In the eiPIC, rudder and fork loop I apparently interact neither with each other nor with the TFIIB-reader-homologous regions of Rrn7. Instead, rudder and fork loop I are oriented towards the bridge helix and an Rrn7 helix that is similar to the TFIIB linker connects to CF module II.Fig. 3The N-terminal region of Rrn7 is partially ordered within the eiPIC.a Ribbon model of TFIIB-homologous regions in the N-terminus of Rrn7 (green) overlaid with sharpened eiPIC density (gray mesh). The lid domain of Pol I subunit A190 (dark gray ribbon) is trapped between well-ordered regions of Rrn7. Residues 46 to 56 of Rrn7 are partially flexible, hinting at a function during promoter melting. b Amino acid sequence of flexible Rrn7 region is shown in green with schematic representation of deletion mutants indicated by black bars. Basal in vitro initiation assay shows the effect of Rrn7 mutations within this loop: Deletion of the entire loop and its C-terminal part (51 to 56) show reduced initiation activity.In addition to a divergent path of Rrn7 compared to TFIIB, the residues contacting the template strand in a Pol I ITC17 and Pol II ITC57 are mostly flexible in the eiPIC, but not in Pol II CCs52 or in a Pol II-TFIIB57 complex. Furthermore, TFIIB reader-loop arginine residue 78, which is important for TSS selection by Pol II58, does not exist in Rrn7. This adds to overall sequence44,46 and architecture differences15 between Rrn7 and TFIIB.To clarify the importance of Rrn7 loop residues disordered in the eiPIC, we mutated the entire loop or smaller stretches and analyzed CF initiation activity in a basal assay (Fig. 3b). The loop-deletion Rrn7 mutant shows strongly reduced initiation efficiency, which can mainly be attributed to the residues 51–56, but not to residues 43–50. The Rrn7 version with loop-deletion still assembles well with Rrn6 and Rrn11 and is able to form a basal PIC in vitro (Supplementary Fig. 3d, e). Thus, the Rrn7-reader-loop is likely important for promoter melting.Pol I is primed for initiation at the eiPIC stageModeling of the active center based on our eiPIC density indicates, that aspartate 629 in subunit A190 (Asp483 in Pol II subunit Rpb1) has apparently changed its orientation with respect to the dimeric crystal structures30,31 (Fig. 4). Assuming its active orientation in the eiPIC, Asp629 now allows coordination of the catalytic magnesium ion (‘metal A’), together with Asp627 and Asp631 for which we observe a clear cryo-EM density peak (Fig. 4b). In addition, the hybrid-binding domain of subunit A135 re-arranges to form a one-turn helix in the eiPIC. This helix also resembles the active Pol I, II, and III EC conformations and its formation exposes histidine 1038 to the bottom of the cleft, which is now free to contact the hybrid upon initial transcription as observed in ITCs. Furthermore, the previously buried lysines 462 and 463 in subunit A190 become exposed in the eiPIC (Supplementary Fig. 2f), now resembling the active Pol-II-fold59 and contacting the first visible downstream DNA base pair. This may contribute to a high affinity for foreign DNA and to the Pol I preference for initiation from ends of dsDNA. With the described structural changes upon eiPIC formation, Pol I enters a conformation that is primed for initial transcription via a conserved mechanism60 in the presence of NTPs.Fig. 4Pol I is primed for initiation in the eiPIC.a Cleft contraction throughout activation stages. Pol I structural models were overlaid via their A135 subunits (protrusion subdomain in gray, space filling). Cleft contraction is indicated by colored clamp core helices (subunit A190). Monomeric Pol I and ITC stages are similar to Rrn3-bound- and EC-conformations, respectively (not shown for clarity). PDB models displayed: 4C2M (orange), 5G5L (green), eiPIC (magenta) and 5M3F (black). b Atomic model of the active center and hybrid-binding domains within Pol I subunits A190 and A135, respectively. Overlaid with sharpened eiPIC density (gray mesh). The metal A site is occupied and a one-turn-helix α30 is formed in A135, exposing positively charged residues. c Inactive Pol I (PDB 4C2M) region for comparison to b. d Active Pol II region for comparison to b and c.We also observe, that the Pol I cleft continues to contract downstream of the sandwich region, adapting an intermediate conformation between the Rrn3-bound and ITC/actively elongating states (Fig. 4a). This adds an additional intermediate to the set of Pol I structures9, but is in line with the suggestion, that cleft modulation is a major regulatory mechanism of Pol I transcription14,30,61. At the stage of DNA-melting during the transition from CC to OC states, dsDNA cannot be accommodated between clamp core and protrusion domains any longer15. Hence, simultaneous promoter loading and cleft contraction allosterically destabilize the upstream duplex at the position of the clamp core and may foster spontaneous melting at this position. Notably, the initially melted region shows the highest conservation among rDNA promoters identified thus far62. Thus, the eiPIC apparently represents a trapped CC-OC transition intermediate conformation, which is important for spontaneous DNA-melting to take place during promoter association of the polymerase.DiscussionWithin this work, we describe an early intermediate initiation complex. The structure enables the independent discussion of promoter recruitment and DNA-melting in a sequential manner. Apparently the polymerase is recruited to its dsDNA promoter but cannot complete the melting process due to a lack of fixated downstream DNA. We described the eiPIC reconstruction in the context of PIC formation and continue to update our model of Pol I recruitment and DNA-melting in light of these findings. Our interpretation is well in line with the idea that targeting of the initiation machinery to the rDNA promoter depends mostly on UAF, and TBP serves to position CF downstream of the UE, while interacting with the promoter using a divergent interface63. Recruitment of the Pol-I-Rrn3 complex then relies on a specific DNA architecture64, namely a bendability that allows interactions of the Rrn11 TPR domain with the Pol I protrusion15 and binding of a promoter element to the Pol I sandwich region (Supplementary Fig. 4). Since our assembly originally comprised UAF and TBP, and only a single reconstruction was obtained from 39% of all recorded particles, it is likely that we capture a physiologically relevant conformation, while factors were artificially positioned by DNA/RNA hybrid scaffolds simulating initial transcription in previous analyses15–17, even though RNA was lost in one case16.Within the eiPIC structure, re-arrangements between CF module I and II enable Rrn7 and Rrn11 to bind promoter DNA, mainly by phosphate backbone interactions of basic loops. This explains the (low) sequence specificity of DNA-binding by CF and thus the overall similar eiPIC architecture compared to ITCs and late PIC reconstructions. Likely, Rrn7-specific DNA-interacting loops contribute to DNA-conformational modulation (compare Fig. 2). We further confirm cleft contraction between the protrusion and clamp core domains and exposure of basic residues bottom of the cleft during DNA-melting by Pol I in the eiPIC.While our findings do not oppose the idea of an upstream ratchetting mechanism to open Pol I promoter DNA, we also see no evidence to support such a mechanism deduced from shifts in CF-positions observed in ITC reconstructions17.Instead, we propose a simplified melting-mechanism based on steric DNA-distortion and electrostatic single-strand trapping which, in this combination, is only possible in Pol I, but not in Pol II and III. Firstly, Pol I recruitment relies on DNA-duplex binding to the sandwiching region and DNA positioning within the expanded cleft of the Pol-I-Rrn3 complex (Fig. 4a and Supplementary Fig. 4). Sequence specificity is determined by proximal upstream bendability15,65 and distal upstream recognition by UAF, which is linked to the PIC via CF and TBP. Divergent TFIIB reader-loop elements within Rrn7 are placed in the Pol I cleft, may play a role in duplex-destabilization and bind the melted template strand similar to observations in ITCs17. In addition, allosteric duplex-destabilization resulting from a cleft contraction between the clamp and protrusion domains observed in the eiPIC likely contributes to melting (Fig. 4a). This contraction primes Pol I for initial transcription by re-ordering previously inactivated regions (Figs. 1 and 4 and Supplementary Fig. 2). Exposed basic residues can then contribute to stabilization of the initially melted template strand and ultimately the DNA/RNA hybrid at the bottom of the cleft. Furthermore, the non-template strand may be bound by the A49 linker (as observed in ref. 17), thereby preventing collapse of the early bubble similar to the σ-factor in bacterial Pol66,67. Only after initial transcription, the growing RNA chain can interact with Rrn7 and would finally clash with reader/linker elements, freeing the exit channel and expelling Rrn7 from the polymerase. This is probably concerted with the association of the flexible A49 twh domain at the back of the clamp core domain, leading to dissociation of CF and Rrn3 and preventing re-association, thereby fostering promoter escape.In Pol II and Pol III initiation complexes48,49,53,68, TFIIB/Brf1 cyclin domains occlude the sandwiching region and reader/linker domains diverge from Rrn7, preventing a similar mechanism. Arguing for a model of combined adaptations, a number of CF-mutations impaired in vitro initiation rates, but only large deletions completely abolished functionality15,45. Furthermore, a 12-subunit Pol I lacking A49/A34.5 is still able to initiate from its native promoter (although the lack of A49 linker-positioning strongly impaired the process)13,37, TBP is not necessary for basal transcription11,21 and single A49 mutations have only minor effects on Pol I function69. Thus, the overall functionality of the system is robust and highly adaptive to conditional variations. However, full initiation rates required for physiological growth depend on the combined action of all Pol-I-specific elements that have accumulated throughout evolutionary adaptation and are basically conserved throughout eukaryotic organisms10,70,71. These adaptations increase initial transcription to such efficiency, that formation of a stable closed complex under physiological conditions appears unlikely. While such a state may be transiently established, the instant cleft contraction and Rrn7-dependent duplex-destabilization by the combined action of Pol I and CF elements directly lead to melting and prime the polymerase for initial transcription and hybrid stabilization.During the final stages of revision of this work, a related study was published72. Sadian et al. provide an excellent description of CF-promoter contacts in detail and investigate the role of an acidic loop in Rrn3, based on higher resolution reconstructions. Compared to our results, interpretation relies on a minor subset of 0.7% or 0.5% of particles from two datasets indicating a transient nature of CCs. In our UAF/TBP-containing samples, however, 39% of initial particles contribute to the final reconstruction and divergent CF-positions are not observed. This may be due to a lack of available particles in our datasets, or due to stabilization of a more ‘native’ CF-orientation in the presence of UAF/TBP. Detailed structure-function analysis of UAF- and TBP-contributions are now instrumental to understanding the process of Pol I initiation in its entirety.MethodsProtein expression and purificationPartially purified endogenous Saccharomyces cerevisiae Pol I is a by-product of Pol III purification via a TAP-tag on subunit AC4049. The Pol-I-containing MonoQ fractions were a gift from A. Vannini and G. Abascal-Palacios. Fractions were pooled, diluted fivefold in buffer A (20 mM HEPES/KOH pH 7.8, 10% glycerol, 1 mM MgCl2, 10 µM ZnCl2, 5 mM dithiothreitol (DTT)) and loaded onto a MonoS HR 5/5 column (GE Healthcare). Separation was performed with a gradient from 10–37,5% buffer B (buffer A with 2 M KAc) with a 2 CV plateau at 17,5% B. Pol I eluted at 470 mM KAc73, peak fractions were pooled, flash frozen in liquid nitrogen and stored at −80 °C.Rrn332 was expressed in BL21(DE3) pRIL (Agilent) cells, by autoinduction in TB medium (1.2% tryptone; 2.4% yeast extract; 0.5% glycerol); 1/10 volume of a sterile solution containing 0.17 M KH2PO4 and 0.72 M K2HPO4 and 1/50 volume of a sterile solution containing 25% glycerol; 10% lactose and 1% glucose were added. A culture was grown at 37 °C to an OD600 of 0.6, after cooling the culture on ice, incubation was continued at 16 °C overnight. Cells were harvested (6000 g; 10 min), resuspended in lysis buffer (50 mM HEPES at pH 7.8, 200 mM NaCl, 3 mM DTT, 10% glycerol). A 3 ml Ni-NTA column (Qiagen) was equilibrated with lysis buffer, the supernatant loaded, and the column was washed with lysis buffer containing 25 mM imidazole. Elution was carried out in lysis buffer containing 150 mM imidazol. Next, Rrn3 was further purified by anion exchange chromatography (Mono Q 5/50, GE Healthcare). The column was equilibrated in MonoQ buffer 1 (50 mM HEPES at pH 7.8, 5 mM DTT, 10% glycerol), and proteins were eluted with a linear gradient of 20 column volumes from 100 mM to 1 M NaCl. After concentration (Amicon, 35 kDa cutoff), the sample was applied to a Superdex 200 increase 10/300 size exclusion column (GE Healthcare) equilibrated with buffer Rrn3-SEC (20 mM HEPES at pH 7.8, 300 mM NaCl, 5 mM DTT).CF subunits15 were co-expressed in E. coli BL21-CodonPlus(DE3)-RIL cells (Agilent) from two plasmids. A 4 l culture was grown in LB medium at 37 °C until OD600 reached 0.5–0.7. Cultures were cooled on ice for 20 min and expression was induced with 0.1 mM IPTG. Cells were grown at 18 °C overnight. Cells were harvested by centrifugation, washed with phosphate-buffered saline (PBS) at 4 °C, flash frozen in liquid nitrogen and stored at −80 °C. One pellet was suspended in buffer CF-A (20 mM imidazole, 350 mM NaCl, 10 mM MgCl2, 10% (v/v) glycerol, 20 mM HEPES pH 7.8, 1 mM DTT, 1x protease inhibitor). Cells were lysed by sonication using a Branson Digital Sonifier, the lysate was cleared by centrifugation and the supernatant was filtered with a 0.22 µm filter (Millipore) to remove cell debris. Cell lysate was then applied to a Ni-NTA column (5 ml, GE Healthcare) and bound CF washed with 5 CV of buffer CF-B (25 mM imidazole, 200 mM NaCl, 10 mM MgCl2, 10% (v/v) glycerol, 20 mM HEPES pH 7.8, 1 mM DTT) at 4 °C. The column was transferred to room temperature, washed with 2.5 CV of buffer CF-C (50 mM imidazole, 200 mM NaCl, 10 mM MgCl2, 10% (v/v) glycerol, 20 mM HEPES pH 7.8, 1 mM DTT, 5 mM ATP, 2 mg/ml denatured protein), incubated for 10 min, and washed again with 2.5 CV buffer CF-C. The column was transferred to 4 °C and washed with 5 CV buffer CF-D (50 mM imidazole, 200 mM NaCl, 10 mM MgCl2, 10% (v/v) glycerol, 20 mM HEPES pH 7.8, 1 mM DTT). Elution was performed with 5 CV of buffer CF-E (350 mM imidazole, 200 mM NaCl, 10 mM MgCl2, 10% (v/v) glycerol, 20 mM HEPES pH 7.8, 1 mM DTT). Protein was then loaded on a 5 ml heparin column (GE Healthcare) in buffer CF-F (200 mM NaCl, 1 mM MgCl2, 10% (v/v) glycerol, 20 mM HEPES pH 7.8, 1 mM DTT) and eluted with a gradient ranging from 0.2 to 2.0 M NaCl, including a plateau at 550 mM NaCl of 2 CVs. CF-containing fractions were concentrated using a 100 kDa cutoff centrifugal filter (Millipore). Size exclusion chromatography was carried out with a Superose 6 increase 10/300 column (GE Healthcare) in buffer CF-G (200 mM NaCl, 1 mM MgCl2, 5% (v/v) glycerol, 10 mM HEPES pH 7.8, 10 μM ZnCl2, 1 mM DTT). CF-containing fractions were concentrated using a 100 kDa cutoff centrifugal filter (Millipore) and directly used or flash frozen in liquid nitrogen for storage at −80 °C.S. cerevisae TBP was cloned into vector pET28b via NheI/Not I restriction sites (compare Supplementary Table 1). Recombinant His6-TBP protein was expressed in BL21(DE3) pRIL (Agilent) cells, by autoinduction in TB medium (1.2% tryptone; 2.4% yeast extract; 0.5% glycerol; 1/10 volume of a sterile solution containing 0.17 M KH2PO4 and 0.72 M K2HPO4 and 1/50 volume of a sterile solution containing 25% glycerol; 10% lactose and 1% glucose were added. A culture was grown at 37 ° C to an OD600 of 0.6, after cooling the culture on ice, incubation was continued at 16 ° C overnight. Cells were harvested (6000 g; 10 min), resuspended in lysis buffer (50 mM HEPES/KOH; 10% glycerol; 10 mM MgAc2; 200 mM KCl; 10 mM imidazole; 5 mM β-mercaptoethanole; 1 mM phenylmethylsulphonyl fluoride (PMSF); 2 mM benzamidine), and lysed by sonication (Branson Sonifier 250 macrotip, cooling in icewater). The cell extract was cleared twice (40,000 g for 40 min at 4 °C) and incubated with 1 ml equilibrated NiNTA Agarose (Qiagen) at 4 °C for 2 h on a rotating wheel. The resin was transferred to a polypropylene column (Bio-Rad), washed with wash buffer 1 (20 mM HEPES/KOH; 10% glycerol; 5 mM MgAc2; 1 M KCl; 20 mM imidazole; 5 mM b-mercaptoethanole), wash buffer 2 (as wash buffer 1 but with 0.2 M KCl) and eluted with elution buffer (20 mM HEPES/KOH; 10% glycerol; 5 mM MgAc2; 0,2 M KCl; 200 mM imidazole; and 5 mM b-mercaptoethanole). The sample was diluted with buffer C (20 mM HEPES/KOH pH 7.8, 10% glycerol, 1 mM MgCl2, 5 mM DTT) to ≈100 mM KCl and loaded onto a MonoS 5/50 GL column (GE Healthcare) and eluted with a linear gradient from 10–100% buffer D (buffer C + 1 M KCl). TBP containing fractions were pooled, concentrated and loaded onto a Superdex 75 10/300 Increase column (GE Helthcare) equilibrated in buffer E (20 mM HEPES/KOH pH 7.8, 10% glycerol, 200 mM KCl, 1 mM MgCl2, 5 mM DTT) peak fractions were pooled, flash frozen in liquid nitrogen and stored at −80 °C.Net1-CTR-TAP was expressed in baculovirus infected SF21 cells as recently published22,74–76. Specifically, 50 × 106 cells were resuspended in buffer F (20 mM HEPES/KOH pH 7.8, 10% glycerol, 200 mM KCl, 1 mM MgCl2, 0.2% NP40, 1 mM DTT), lysed by sonication (Branson Sonifier 250 macrotip, cooling in icewater) and cleared by centrifugation (40,000 × g for 40 min at 4 °C). The supernatant was incubated four hours with 1 ml IgG Sepharose 6 Fast Flow (GE Healthcare), washed with buffer F and incubated with TEV protease for two hours at 16 °C. Eluate fractions were collected, flash frozen in liquid nitrogen and stored at −80 °C. The MultiBac system was also used to generate a bacculo virus co-expressing Rrn5-HA, Rrn9-Flag, Uaf30-His7, Rrn10, Histones H3 and 4 in SF21 cells (Supplementary Table 1). In all, 1 × 109 cells were resuspended in lysis buffer (50 mM HEPES/KOH pH 7.8, 10% glycerol, 400 mM (NH4)2SO4, 10 mM MgCl2, 20 mM imidazole, 1 mM DTT) and lysed by sonication (Branson Sonifier 250, cooling in icewater). The cleared lysate (2 × 70,000 g, 45 min, 4 °C) was incubated with 2 ml Ni-NTA Agarose (Qiagen) for 2 h, beads were washed with buffer G (20 mM HEPES/KOH pH 7.8, 10% glycerol, 1 M KCl, 5 mM MgCl2, 20 mM imidazole, 1 mM DTT), buffer H (20 mM HEPES/KOH pH 7.8, 10% glycerol, 400 mM KCl, 2 mM MgCl2, 50 mM imidazole, 1 mM DTT) and eluted with buffer I (20 mM HEPES/KOH pH 7.8, 10% glycerol, 400 mM KCl, 2 mM MgCl2, 300 mM imidazole, 1 mM DTT). The eluate was diluted with buffer C, loaded onto a MonoS HR 5/5 column and eluted with a gradient to buffer D (see above). Eluate fractions were collected, concentrated, flash frozen in liquid nitrogen and stored at −80 °C. Raw SDS-PAGE gels are shown in Supplementary Fig. 6.Promoter-dependent in vitro transcriptionPromoter-dependent in vitro transcriptions were performed following our previously published protocols13,15 on core promoter scaffolds from position −38 to +24 relative to the TSS (Supplementary Table 1). Specifically, Promoter-dependent in vitro transcription reactions were performed as follows: A total of 50 ng template dsDNA template were used for each transcription reaction (25 μl reaction volume. CF was added to a final concentration of 20 nM pre-incubated Pol-I-Rrn3 complex was added to a final concentration of 4 nM. 20 mM HEPES/KOH pH 7.8 and 2 M KAc were added to adjust volume and salt concentration to the final reaction conditions of 150 mM KAc in 25 µl. Transcription was started by the addition of 5 μl 5x transcription buffer (100 mM HEPES/KOH pH 7.8, 50 mM MgCl2, 25 mM EGTA, 0.25 mM EDTA, 3 mM DTT, 1 mM ATP, 1 mM UTP, 1 mM CTP, 0.05 mM GTP complemented with 0.3 µl [α-32P]GTP (10 mCi/ml; Hartmann Analytic). The samples were incubated at 24 °C for 30 min. Next, 200 μl Proteinase K buffer (0.5 mg/ml Proteinase K in 0.3 M NaCl, 10 mM Tris/HCl pH 7.5, 5 mM EDTA and 0.6% SDS) was added to stop transcription. The samples were incubated at 56 °C for 15 min. Ethanol (700 μl) was added to allow precipitation of nucleic acids (30 min at −80 °C). The samples were centrifuged for 10 min at 12,000 × g, the supernatant was removed and the precipitate was washed with 150 µl 70% ethanol. After centrifugation, the supernatant was removed and the pellets were dried at 95 °C for 1 min. RNA in the pellet was dissolved in 12 μl 80% formamide, 0.1 M TRIS-Borate-EDTA (TBE), 0.02% bromophenol blue and 0.02% xylene cyanol. Samples were heated for 2 min under vigorous shaking at 95 °C and briefly centrifuged. After separation on a 20% polyacrylamide gel containing 8 M urea and 1x TBE. Radiolabelled transcripts are visualized using a PhosphoImager (GE Healthcare). Raw gels are shown in Supplementary Fig. 6.Electrophoretic mobility shift assays (EMSA)For EMSA experiments, a fluorescently labeled promoter fragment (−83 to +26 relative to the TSS) was annealed from oligonucleotides labeled with fluorescent dyes (NTS position −3 Atto647N and TS position −5 Cy3) as described below. 0.2 pmol dsDNA was incubated for 30 min without or with increasing amounts (0.25, 0.5, 1 pmol) of recombinant mutant CF in incubation buffer containing 20 mM HEPES/KOH pH 7.8, 10% glycerol, 200 mM KCl, 1 mM MgCl2, 0.1 mg/ml BSA, 1 mM DTT. The reaction was loaded on a pre-run 6% native acrylamide gel in 0.5x TBE buffer and imaged on a Typhoon FLA 9000 (GE Healthcare) imaging system. Raw gels are shown in Supplementary Fig. 6.Pol I PIC assemblyThe Pol I PIC was assembled on complementary rDNA promoter oligonucleotides AGCTTAAATTGAAGTTTTTCTCGGCGAGAAATACGTAGTTAAGGCAGAGCGACAGAGAGGGCAAAAGAAAATAAAAGTAAGATTTTAGTTTGTAATGGGAGGGGGGGTTTAGTCATGGAGTACAAGTGTGAGGAAAAGTAGTTGGGAGGTACTTCATGCGAAA (NTS), TTTCGCATGAAGTACCTCCCAACTACTTTTCCTCACACTTGTACTCCATGACTAAACCCCCCCTCCCATTACAAACTAAAATCTTACTTTTATTTTCTTTTGCCCTCTCTGTCGCTCTGCCTTAACTACGTATTTCTCGCCGAGAAAAACTTCAATTTAAGCT (TS) (Integrated DNA Technologies). Oligonucleotides were dissolved in TE buffer (10 mM Tris pH 8, 0.5 mM EDTA), mixed in equimolar amounts to a final concentration of 10 µM each, heated to 95 °C and slowly cooled down to 10 °C with a cooling rate of 1 °C/min.In all, 0.11 nmol promoter DNA was incubated with equimolar amounts of UAF, and threefold molecular excess of TBP and Net1-CTR. After 20 min incubation at 28 °C, 0.17 nmol CF was added and incubated for additional 20 min. 0.095 nmol Pol I, pre-incubated overnight with fivefold molar excess of Rrn3 on ice, was added and the sample was diluted with buffer G (20 mM HEPES/KOH pH 7.8, 2 mM MgCl2, 5 mM DTT) to final assembly conditions (20 mM HEPES/KOH pH 7.8, 50 mM KAc, 50 mM KCl, 2 mM MgCl2, 10 µM ZnCl2, 5 mM DTT; buffer H), incubated further 30 min and concentrated to 50 µl. The sample was crosslinked with 1 mM (bis(sulfosuccinimidyl) suberate) (BS3) for 30 min at 28 °C. The Crosslinking reaction was quenched with 100 mM NH4HCO3 final concentration for 15 min at 28 °C. The sample was loaded onto a Superose 6 PC 3.2/30 column (GE Healthcare) equilibrated with buffer H (20 mM HEPES/KOH pH 7.8, 50 mM KAc, 50 mM KCl, 2 mM MgCl2, 10 µM ZnCl2, 5 mM DTT) and collected in 60 µl fractions. A raw sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) gel of non-crosslinked sample is shown in Supplementary Fig. 6.Cryo-EM sample preparation and data acquisitionGrids were glow discharged in Argon/Oxygen plasma 90/10 (Fischione) for one minute. Four microliters of sample was applied to a grid (Quantifoi R 2/1 + 2 nm carbon, Quantifoil), incubated for 30 s, blotted 4 s with blot force ‘8’, at 100% humidity and 4 °C in a Vitrobot Mark IV (FEI) and plunged in liquid ethane.Images were collected on a Cs-corrected Titan Krios microscope (FEI), operated at 300 kV using the multi-shot feature of the serialEM software77 for automated data collection. Movie frames were acquired on a 4k × 4k Gatan K2 summit direct electron detector in super-resolution mode at a nominal magnification of 105,000, which yielded a pixel size of 0.545 Å. Forty movie frames were recorded at a dose of 1.4 electrons per Å2 per frame corresponding to a total dose of 56 e/Å2.Image processingMovie frames were aligned, dose-weighted, binned by a factor of 2 and averaged using MotionCor278. Contrast Transfer Function (CTF) parameters were estimated with the Gctf 79 program. The RELION 3-beta suite33 was used for the whole-image processing workflow unless stated otherwise. The dataset was divided into four subsets with ~1000 images each. In a first step the reference-free auto-picking procedure based on a Laplacian-of-Gaussian (LoG) filter was used to identify ~100,000 starting coordinates (per subset), which were used to extract particles with threefold binning in a 140 pixel box and the particles were grouped by reference-free 2D classification. Classes with contamination and damaged particles were discarded and the remaining particles were aligned on a reference generated from the PDB entry 5G5L low-pass filtered to 40 Å. Three-dimensional (3D) classes containing only Pol I and Rrn3, or damaged particles were discarded. The remaining 227,718 particles from the four subsets were merged, re-extracted without binning and refined against an initial model generated in RELION. CTF Refinement and Bayesian polishing was performed and the polished particles were refined and 2D and 3D classification without alignment were performed to remove misaligned particles and the remaining 168,532 particles were subjected to a second round of CTF refinement. A 3D classification without sampling and a CF-only mask revealed one class with partial CF occupancy and another with damaged particles that were both discarded. Refinement of the remaining 122,099 particles resulted in an early intermediate PIC reconstruction. For details, compare Supplementary Fig. 1. During post-processing in RELION, a B-factor of −75 Ų was determined and applied for map sharpening, resulting in an overall resolution of 3.5 Å. Focused refinements the with a Pol-I-Rrn3 mask (3.5 Å after post-processing) or a CF-DNA mask (3.9 Å after post-processing) were additionally carried out to assist subdomain conformation determination and aid CF chain tracing, respectively. Directional FSC were calculated as described80.Model buildingAt a resolution of 3.5 Å, we derive an atomic model of an early intermediate PIC. We first placed Pol I domains as described for PDB 5G5L12 originating from the crystal structure (PDB 4C2M30), an Rrn3 monomer (PDB 3TJ132), a CF monomer (PDB 5O7X15) and an ITC DNA (PDB 5W6617) in the unsharpened eiPIC map generated with RELION 3 (beta version)33. Using COOT81, we adjusted protein backbone traces consulting focused maps of CF or the Pol-I-Rrn3 complex and finally build side chain residues where appropriate. DNA-sequences were mutated to poly-A (-T, -G, -C). For the structure-based modeling of the TFIIB-related domains in the N-terminal region of Rrn7, the strong density for aromatic residue Phe70 was used as a marker. The final model was refined using the real-space refinement tool of the Phenix suite82 and evaluated using MolProbity83. Figures were prepared with UCSF Chimera84 or PyMOL (pymol.org).It should be noted that promoter-binding regions within CF are highly flexible and thus poorly ordered in DNA-free CF crystals15. While we refrained from building most of these regions in the crystal structure, the putatively assigned residue numbers within helix α2 of CF subunit Rrn11 were now adjusted in the eiPIC, similar to a de novo built model based on a cryo-EM reconstruction of an ITC17.An additional cryo-EM density stretch between the Rrn7 ribbon and the Pol I wall domain may potentially be attributed to a flexible loop in Rrn3 (249–323) or to a part of the Rrn6 C-terminal domain. Whereas the latter assignment would agree with a previously published crosslinking/mass spectrometry analysis16 and direct Rrn6-Rrn3 interaction studies85, it remains as speculative at this point.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Reporting Summary | nature communications | [
"Article"
] | [
"Biochemistry",
"Electron microscopy",
"Molecular modelling"
] | transcription ribosomal RNA precursor RNA polymerase I prerequisite for ribosome biosynthesis in eukaryotes1 Pol I transcription regulated pre-initiation complex Pol II III related initiation mechanisms Pol I promoter recognition PIC formation DNA melting In bakers’ yeast Saccharomyces cerevisiae basal initiation system for Pol I promoter DNA core initiation factor Rrn3 heterotrimeric core factor CF binds CE 38 base pairs upstream transcription start site recruits Rrn3-stabilized Pol I for DNA melting occurs upstream between Pol I No additional factors required transcription promoter escape upstream activating factor recognizes element cooperates with TATA-binding protein CF association Pol I initiation rates 40-fold in factor Net1 at Pol I promoters initiation in vivo transcription initiation Pols recruited to promoters by transcription factors ‘closed complex’ After melting DNA strands ‘open complex’ established into ‘initially transcribing chain synthesisITCs stable DNA/RNA hybrid formed polymerase movement gene before processive elongation complex 24,25) analyses Pol I initiation complexes stabilized mismatched bubble scaffold transcribed RNA sequence double-stranded DNA sequence 24 bps downstream approach Pol II elongation complexes heterogenic sample conformations tight DNA/RNA hybrid association polymerase26 Pol I PIC strategy late initiation intermediates inconsistent occupancy Rrn3 divergent localization tandem-winged helix domain Pol I subunit A49 C-terminal domain subunit A12.2 speculation functional roles temporal classification conformations aimed Pol I initiation mechanisms early initiation stage promoter recognition Pol I recruitment DNA melting close native assembled initiation complex double-stranded promoter DNA performed single-particle cryo-EM analysis dsDNA scaffold truncated downstream edge position +8 TSS preventing contact clamp core jaw domains polymerase Three-dimensional particle reconstruction cryo-EM density refinement structural modeling allow basal PIC components comparative PIC analysis three eukaryotic Polsstructure-guided analysis indicates Pol-I-specific ribosomal DNA promoter melting formation cryo-EM promoter recognition DNA melting formed Pol I initiation complex in vitro UAF assembled on dsDNA promoter scaffold position −155 to +8 relative TSS TBP CF fragment protein Net122 purified Pol I13 pre-incubated with Rrn332 early PIC stable size exclusion chromatography Pol I recruited to UAF/TPB/Net1/CF-bound promoter scaffold lacking sequence stretches extended downstream contacts-head domains polymerase Single-particle cryo-EM data collected Titan Krios Gatan K2 summit direct electron detector pre-processing RELION33 122,099 particles selected from 4,088 micrograph movies final cryo-EM reconstruction resolution 3.5 Å Pol I early intermediate PIC cryo-EM density reveals secondary structure features particle side chain orientations crosslinking TBP UAF Net1-CTR flexible stabilizing CF Pol II-EM reconstruction Pol I early intermediate PICOverview Pol I eiPIC cryo-EM reconstruction 3.5 Å resolution transparent gray envelope PDB model DNA right panel transparent density protein components solid density DNA path blue non-template light PAD promoter-associated domain Rrn11) PIR polymerase interacting region Schematic representation promoter dsDNA PIC assembly densities eiPIC highlighted blue light blue template non-template strand Atomic model helix subunit A190 sharpened eiPIC density mesh residue orientations early intermediate PIC-defined architectureInitial assignment template non-template DNA strands Pol I CF subunits Rrn3 manual model building real-space refinement model high quality Upstream DNA-ordered CF-interacting regions Pol I active center cleft no density active center ≥12 base-pairs downstream edge helix clamp-head/jaw domains highly charged region bound foreign DNA upstream end scaffold similar effect patches nucleosome transcription Pol II supercoiled in vitro initiation assays preference Pol I dsDNA ends synthetic 1Cryo-EM data collection refinement validation statisticsEMDB-10544 6TPSCF EMDB-10663Data collection processing Voltage)300300 Electron exposure Defocus range)−1.5 Pixel size (0.545 superres Symmetry Initial particle images,557 Final images Map resolution)3.543.91 FSC threshold0.1430.143 range 16.7Refinement Initial model)6TPS resolution threshold0.143 Map sharpening B factor composition Non atoms50,070 Protein residues6,109 Ligands8 (Zn Mg factors Protein65.6 Ligand102 deviations Bond lengths)0.009 angles)0.985Validation score1.85 Clashscore5.96 rotamers (%)0.59Ramachandran plot Favored.10 Allowed Disallowed (%)0.15Initiation factor Rrn3 bound Pol I subdomains12 chromatin immuno precipitation biochemical studies mouse38 cells CF associated Pol I core polymerase interacting regions ITC cryo-EM density rebuild CF subunits Rrn6 Rrn7 Rrn11 assignments crystal ITC EMinactive Pol I30,31,40 subdomains flexible central bridge helix refolded eiPIC. 1c EC C-terminal domain subunit A12.2 residual density funnel domain subunit A190 not localized A135 lobe 12-subunit EC43 eiPIC reconstruction shows strong density A49/A34.5 dimerization A34.5 C-terminal tail domains heterodimer constitutively attached twh linker domains subunit A49 detached eiPIC proposed role promoter escape17.Core factor embraces promoter eiPIC density allows construction CF model ITC conformation structural changes promoter recruitment compared architecture CF free promoter-engaged eiPIC conformation CF module I II retract 12 Å binding CE promoter sequence retraction positively charged residues engage phosphate backbone DNA-binding regions Rrn11 promoter-associated domain cyclin domains Rrn7 regions engage DNA in ITCs15–17 described late ITCs devoid Rrn317.Rrn7 residues DNA-binding conserved TFIIB Brf1 clash with TBP15 TFIIB-TBP47 Brf1-TBP48,49 complex free promoter-engaged CF Rrn7-specific helix α4a N-terminal cyclin domain shifts inserted into minor groove CE promoter DNA loop α7-α8 II contacts major groove upstream eiPIC formation distal upstream DNA-path modified towards C-terminal domain Rrn7 β-propeller-domain Rrn6 promoter binding Rrn7-specific regions TFIIB-unrelated CF subunit Rrn11 squeeze DNA basal Pol I initiation system require TBP association Rrn7 cyclins. 2Core Factor—promoter interactions eiPIC promoter-bound CF regions Rrn7 Rrn11 contribute promoter phosphate backbone interactions wild-type CF interacts with double-stranded promoter DNA Mutation Rrn7 (Δα4a α7-α8) impair promoter-DNA association initiation efficiency CF with Rrn7 mutants Δα4a α7-α8 impaired constructed CF mutants with deletions in helix α4a loop α7-α8associate with promoter DNA show defects in basal initiation vitro Engagement regions DNA conformation for Pol I recruitment promoter melting Pol I ‘sandwich’ region important for PIC Pol-I-specific proximal promoter-binding region loop α11a-α12 (residues 452–456 β28-β28 815–818) protrusion wall domains Pol I subunit A135 eiPIC positively charged loop (892–895 re-orients towards promoter DNA-backbone interactions 4) promoter interactions specific to Pol I residues not conserved in Pol II50 III51 DNA occluded from Pol II III PICs by N-terminal cyclin domains TFIIB34,52,53 Brf1/Brf248,49,54,55 Pol region named eiPIC sandwich region holds promoter between wall protrusion domains contact DNA strands specific for un-melted duplex Density DNA downstream observed higher flexibility recruitment Pol-I-Rrn3 complex contacts promoter sandwich protein–protein contacts between CF Pol-Rrn3 promoter contacts with Pol I cleft downstream elements A49 required for recruitmentTFIIB elements Rrn7 divergent elements Rrn744,46 ordered center cleft eiPIC exception residues 46–56 protein backbone extends N-terminal zinc ribbon into Pol I cleft subdomain Pol subunit A190 forming two anti-parallel strands exiting Pol shelf module path Rrn7 differs from Pol I ITC17 TFIIB Pol II57 Pol II initiation TFIIB-reader-loop contacts ‘rudder’ ‘fork loop domains TFIIB-linker binds rudder forms helix clamp core domain57 rudder fork loop I interact TFIIB-reader-homologous regions Rrn7 oriented towards helix Rrn7 helix connects CF module II N-terminal region Rrn7 partially ordered within eiPIC model TFIIB-homologous regions N-terminus Rrn7 sharpened eiPIC density lid domain Pol I subunit A190 trapped between regions Rrn7 Residues 46 to 56 Rrn7 partially flexible function promoter melting Amino acid sequence flexible Rrn7 region green deletion mutants black barsvitro initiation assay Rrn7 mutations Deletion loop C-terminal part (51 to 56) show reduced initiation activity divergent path Rrn7 TFIIB residues contacting template strand Pol I ITC17 Pol II ITC57 flexible in eiPIC not Pol II CCs52 Pol II-TFIIB57 complex TFIIB reader-loop arginine residue 78 important for TSS selection in Rrn7 adds between Rrn7 TFIIB Rrn7 residues eiPIC mutated entire loop analyzed CF initiation activity (Fig. loop-deletion Rrn7 mutant reduced initiation efficiency residues 51–56 not 43–50 Rrn7 version loop-deletion assembles with Rrn6 Rrn11 basal PIC in vitro Rrn7-reader-loop important for promoter melting I primed for initiation eiPIC aspartate 629 in subunit A190 (Asp483 Pol II subunit Rpb1) changed orientation dimeric crystal. Asp629 allows coordination catalytic magnesium ion with Asp627 Asp631 clear cryo-EM density peakhybrid-binding domain subunit A135 re-arranges one-turn helix eiPIC helix resembles active Pol I II III EC conformations exposes histidine 1038 bottom cleft contact hybrid initial transcription buried lysines 462 463 subunit A190 exposed eiPIC resembling active Pol-II contacting downstream DNA base pair high affinity foreign DNA Pol I preference initiation ends dsDNA structural changes eiPIC formation Pol I enters conformation primed for initial transcription 4Pol I primed initiation eiPIC Cleft contraction activation stages Pol I structural models overlaid A135 subunits colored clamp core helices Monomeric Pol I ITC stages similar Rrn3-bound- EC-conformations PDB models 4C2M 5G5L eiPIC 5M3F Atomic model active center hybrid-binding domains Pol I subunits A190 A135 sharpened eiPIC density metal A site occupied one-turn-helix α30 formed in A135 exposing positively charged residues Inactive Pol I region Active Pol II regionPol I cleft downstream sandwich region adapting intermediate conformation between Rrn3-bound ITC/actively elongating states (Fig. 4a). adds intermediate Pol I cleft modulation Pol I transcription14 DNA-melting transition CC to OC dsDNA between clamp core protrusion domains promoter loading cleft contraction destabilize upstream duplex foster spontaneous melting initially melted region highest conservation among rDNA promoters eiPIC represents trapped CC-OC transition intermediate conformation important for spontaneous DNA-melting during promoter association polymerase early intermediate initiation complex enables promoter recruitment DNA-melting polymerase recruited to dsDNA promoter complete melting lack of fixated downstream DNA eiPIC reconstruction context PIC formation model Pol I recruitment DNA-melting targeting initiation machinery to rDNA promoter depends on UAF TBP CF downstream UE interacting promoter divergent Recruitment Pol-I-Rrn3 complex relies on specific DNA architecture64 bendability interactions Rrn11 TPR domain with Pol I protrusion15 binding promoter element to Pol I sandwich region Fig. 4)assembly comprised UAF TBP single reconstruction from 39% particles likely physiologically relevant conformation factors positioned by DNA/RNA scaffolds initial transcription RNA lost one eiPIC re-arrangements CF module I II enable Rrn7 Rrn11 bind promoter DNA phosphate backbone interactions explains sequence specificity DNA-binding CF similar eiPIC Rrn7-specific DNA-interacting loops contribute DNA-conformational modulation Fig confirm cleft contraction protrusion clamp core domains exposure residues during DNA-melting Pol I findings oppose upstream ratchetting open Pol I promoter DNA no evidence CF-positions propose simplified melting-mechanism steric DNA-distortion electrostatic single-strand trapping possible in Pol I not Pol II III Pol I recruitment relies on DNA-duplex binding sandwiching region DNA positioning expanded cleft Pol-I-Rrn3 complex. Sequence specificity determined by proximal upstream distal upstream recognition UAF linked to PIC via CF TBPDivergent TFIIB reader-loop elements Rrn7 Pol I cleft duplex-destabilization bind melted template strand allosteric duplex-destabilization cleft contraction clamp protrusion domains contributes to melting (Fig. contraction primes Pol I for transcription inactivated regions (Figs. 1 4 2) residues stabilization melted template strand DNA/RNA hybrid cleft non-template strand bound by A49 linker collapse early bubble after transcription growing RNA chain with Rrn7 reader/linker elements freeing exit channel expelling Rrn7 from polymerase with association A49 twh domain clamp core domain dissociation CF Rrn3 re-association promoter escape Pol II Pol III initiation complexes48 TFIIB/Brf1 cyclin domains occlude sandwiching region reader/linker domains diverge from Rrn7 preventing similar mechanism CF-mutations vitro initiation rates large deletions abolished functionality1512-subunit Pol I lacking A49/A34.5 from native promoter lack A49 linker-positioning process TBP not necessary for basal transcription11 A49 mutations minor effects on Pol I function69 functionality system robust adaptive to conditional variations full initiation rates for growth depend on combined action Pol-I-specific elements conserved eukaryotic increase initial transcription stable closed complex unlikely cleft contraction Rrn7-dependent duplex-destabilization Pol I CF lead to melting polymerase for initial transcription hybrid stabilization related study published72 Sadian et al. CF-promoter contacts role acidic loop in Rrn3 interpretation relies on minor subset of 0.7% or 0.5% particles from datasets transient nature UAF/TBP-containing samples 39% initial particles contribute to final reconstruction divergent CF-positions not observed due to lack of particles or stabilization ‘native’ CF-orientation in UAF/TBP Detailed structure-function analysis of UAF- TBP-contributions instrumental to understanding Pol I initiationexpression Saccharomyces cerevisiae III MonoQ fractions Vannini Abascal-Palacios pooled diluted buffer A (20 mM HEPES 7.8 10% glycerol 1 mM MgCl2 10 ZnCl2 5 dithiothreitol loaded MonoS HR 5/5 column Separation 10–37,5% buffer B plateau 17,5% I eluted 470 mM KAc73 fractions pooled frozen nitrogen stored −80 °C expressed BL21 pRIL cells autoinduction TB medium (1.2% tryptone 2.4% yeast extract 0.5% 0.17 M KH2PO4 0.72 M K2HPO4 1/50 25% glycerol 10% lactose 1% glucose culture grown 37 °C OD600 0.6 incubation 16 °C Cells harvested resuspended lysis buffer (50 mM HEPES pH 7.8 200 mM NaCl 3 mM DTT 10% 3 ml Ni-NTA column equilibrated lysis buffer washed buffer 25 mM imidazole Elution buffer 150 mM imidazolRrn3 purified chromatography equilibrated buffer 1 (50 mM HEPES pH 7.8 5 mM DTT 10% proteins eluted 100 to 1 M NaCl Superdex 200 10/300 column equilibrated buffer Rrn3-SEC (20 mM HEPES pH 7.8 300 mM NaCl 5 mM subunits15 co-expressed E. coli BL21-CodonPlus-RIL cells 4 l culture grown LB medium 37 °C OD600 0.5–0.7 cooled ice 20 min expression induced 0.1 mM IPTG 18 °C overnight harvested washed frozen nitrogen stored −80 °C pellet suspended buffer CF-A (20 mM imidazole 350 mM NaCl 10 mM MgCl2 10% glycerol 20 mM HEPES 7.8 1 mM DTT protease Cells lysed cleared centrifugation filtered 0.22 μm filterlysate Ni-NTA column (5 GE Healthcare washed buffer-B (25 imidazole 200 NaCl 10 MgCl2 10% glycerol 20 HEPES pH 7.8 1 DTT 4 °C transferred room temperature washed 2.5 buffer CF-C (50 imidazole NaCl MgCl2 glycerol HEPES 7.8 1 DTT 5 ATP 2 mg/ml denatured incubated 10 min washed transferred 4 °C washed 5 buffer CF-D (50 imidazole 200 NaCl MgCl2 glycerol 20 HEPES pH 7.8 1 Elution 5 buffer CF-E imidazole 200 NaCl MgCl2 glycerol 7.8 1 Protein loaded 5 ml heparin column buffer CF-F NaCl MgCl2 10% glycerol HEPES pH 7.8 1 DTT eluted 0.2 2.0 M NaCl plateau 550 mM NaClCF fractions concentrated 100 kDa filter Size exclusion chromatography Superose 6 10/300 column Healthcare buffer CF-G (200 mM NaCl 1 mM MgCl2 5% glycerol 10 mM HEPES pH 7.8 10 ZnCl2 1 mM kDa frozen liquid nitrogen −80 °C cerevisae TBP cloned pET28b sites Recombinant His6-TBP protein expressed BL21(DE3) pRIL cells autoinduction TB medium (1.2% tryptone 2.4% yeast extract 0.5% glycerol 0.17 M KH2PO4 0.72 M K2HPO4 25% glycerol 10% lactose 1% glucose culture grown 37 ° C OD600 0.6 incubation 16 ° C Cells harvested g resuspended lysis buffer (50 mM HEPES/KOH 10% glycerol 10 MgAc2 200 KCl imidazole 5 β-mercaptoethanole 1 fluoride 2 lysed sonication cell extract cleared incubated 1 ml NiNTA Agarose 4 °C 2 hresin transferred polypropylene column washed buffer 1 10% glycerol 5 MgAc2 1 KCl imidazole b 2 0.2 M KCl eluted buffer (20 10% glycerol 5 MgAc2 0,2 M KCl 200 imidazole 5 b diluted buffer C MgCl2 to mM KCl loaded MonoS 5/50 GL column Healthcare eluted 10–100% buffer D TBP fractions pooled concentrated loaded Superdex 75 10/300 Increase column equilibrated buffer E (20 10% glycerol 200 mM KCl 1 mM MgCl2 5 mM DTT fractions pooled frozen liquid nitrogen stored −80 °C.Net1-CTR-TAP expressed baculovirus infected SF21 cells50 106 cells resuspended buffer F (20 HEPES pH 7.8 10% glycerol 200 mM KCl 1 MgCl2 0.2% NP40 lysed sonication cleared centrifugation 40 min 4 supernatant incubated hours 1 ml IgG Sepharose 6 washed buffer F incubated TEV protease two hours 16 °C Eluate collected frozen nitrogen stored −80 °C MultiBac system virus Rrn5-HA Rrn9-Flag-His7 SF21 cells109 cells resuspended lysis buffer (50 mM pH 10% glycerol 400 mM (NH4)2SO4 10 mM MgCl2 20 mM imidazole 1 mM DTT lysed sonication Sonifier 250 cleared lysate (2 × 70,000 g 45 min 4 °C incubated 2 ml Ni-NTA Agarose 2 h washed buffer G (20 glycerol 1 KCl 5 mM MgCl2 20 imidazole 1 H 400 KCl 2 MgCl2 50 imidazole 1 DTT eluted buffer I (20 400 KCl 2 MgCl2 300 mM imidazole 1 mM eluate diluted buffer C loaded MonoS HR 5/5 column eluted buffer D Eluate collected concentrated frozen liquid nitrogen stored −80 °C SDS-PAGE gels Supplementary Fig. 6.Promoter-dependent vitro performed core promoter scaffolds position −38 to +24 TSS 50 ng dsDNA template eachCF 20 nM Pol-I-Rrn3 4 nM 20 mM HEPES/KOH pH 7.8 2 M KAc 150 mM KAc 25 μl Transcription 5 μl buffer (100 mM HEPES/KOH pH 7.8 50 MgCl2 25 EGTA 0.25 EDTA 3 DTT 1 ATP UTP CTP 0.05 GTP 0.3 μl [α-32P]GTP samples incubated 24 °C 30 min 200 μl Proteinase K buffer mg/ml 0.3 M NaCl 10 mM Tris/HCl pH 7.5 5 mM EDTA 0.6% SDS incubated 56 °C 15 min Ethanol μl centrifuged 10 min 12,000 g washed 150 μl 70% ethanol pellets dried 95 °C 1 min dissolved 12 μl 80% formamide 0.1 M TRIS-Borate-EDTA 0.02% bromophenol blue 0.02% xylene cyanol heated 2 min 95 °C centrifuged 20% polyacrylamide gel 8 M urea 1x TBE Radiolabelled transcripts PhosphoImager gels Figmobility shift assays (EMSA labeled promoter fragment (−83 to +26 TSS annealed from oligonucleotides fluorescent dyes (NTS −3 Atto647N TS −5 Cy3) 0.2 pmol dsDNA incubated 30 min (0.25 0.5 1 pmol recombinant mutant CF buffer 20 mM HEPES/KOH pH 7.8 10% glycerol 200 mM KCl 1 mM MgCl2 0.1 mg/ml BSA 1 mM DTT reaction loaded-run 6% native acrylamide gel 0.5x TBE buffer imaged Typhoon FLA 9000 imaging system Raw gels Supplementary Fig. 6.Pol I PIC rDNA promoter oligonucleotides Oligonucleotides dissolved TE buffer (10 mM Tris pH 8 0.5 mM mixed 10 μM each heated to 95 °C cooled 10 °C 1 °C/min0.11 nmol DNA incubated UAF threefold TBP Net1-CTR 20 min incubation 28 0.17 nmol CF added incubated 20 min 0.095 nmol Pol I pre-incubated Rrn3 added diluted buffer G (20 mM HEPES/KOH pH 7.8 2 mM MgCl2 5 mM DTT KCl MgCl2 ZnCl2 incubated 30 min concentrated 50 μl crosslinked 1 mM (bis(sulfosuccinimidyl) suberate 30 min 28 °C quenched 100 mM NH4HCO3 15 min loaded Superose 6 PC 3.2/30 column equilibrated buffer H KCl 2 MgCl2 10 ZnCl2 5 DTT collected 60 μl fractions raw sodium dodecyl sulfate–polyacrylamide gel electrophoresis non-crosslinked sample Supplementary Fig. sample preparation discharged Argon/Oxygen plasma 90/10 one minutemicroliters sample applied (Quantifoi R 2/1 + 2 nm carbon incubated 30 s blotted 4 s force 100% humidity 4 °C Vitrobot Mark IV plunged liquid ethane collected Cs-corrected Titan Krios microscope 300 kV multi-shot feature serialEM Movie frames 4k × 4k Gatan K2 summit direct electron detector super-resolution magnification 105,000 pixel size 0.545 Å Forty frames recorded 1.4 electrons per Å2 total 56 e/Å2.Image frames aligned dose-weighted binned factor 2 averaged MotionCor278 Contrast Transfer Function) parameters estimated Gctf 79 program RELION 3-beta suite33 whole-image processing dataset divided four subsets ~1000 images each reference-free auto-picking-Gaussian filter ~100,000 starting coordinates particles threefold binning 140 pixel box grouped reference-free 2D classification contamination damaged particles discarded remaining particles aligned filtered 40 Å Three-dimensional (3D) classes damaged discarded remaining 227,718 particles merged re-extracted refined initial model RELIONCTF Refinement Bayesian polishing performed polished particles refined 2D 3D classification without alignment misaligned particles remaining 168,532 particles second round CTF refinement 3D classification sampling CF-only mask class partial CF occupancy damaged particles discarded Refinement remaining 122,099 particles early intermediate PIC reconstruction Supplementary Fig. 1. post-processing RELION B-factor −75 Å2 determined applied map sharpening resolution 3.5 Å refinements Pol-I-Rrn3 mask CF-DNA mask subdomain conformation determination CF chain tracing Directional FSC calculated resolution 3.5 Å atomic model early intermediate PIC placed Pol I domains PDB 5G5L12 Rrn3 monomer CF monomer ITC DNA unsharpened eiPIC map RELION 3 COOT81 adjusted protein backbone traces maps CF Pol-I-Rrn3 complex side chain residues DNA-sequences mutated to poly-A ( structure-based modeling TFIIB-related domains N-terminal region Rrn7 strong density aromatic residue Phe70 markerfinal model refined Phenix suite82 evaluated MolProbity83 Figures prepared UCSF Chimera84 PyMOL promoter-binding regions CF flexible poorly ordered in DNA-free CF crystals15 refrained from building regions crystal structure residue numbers helix α2 CF subunit Rrn11 adjusted in eiPIC model cryo-EM reconstruction ITC17 additional cryo-EM density stretch between Rrn7 ribbon Pol I wall domain attributed flexible loop Rrn3 (249–323) or Rrn6 C-terminal domain latter crosslinking/mass spectrometry Rrn6-Rrn3 interaction studies85 remains speculative information Nature Research Reporting Summary.Supplementary information | 47.2 | 0.788788 |
10.1038/s41467-021-21992-w | PMC7979801 | Favipiravir has broad-spectrum antiviral activity against a variety of RNA viruses. Here the authors investigate the safety, pharmacokinetics and anti-SARS-CoV-2 efficacy of different drug dosage in the a Syrian hamster model of infection and, combined with genetic analyses, they show that Favipiravir at high doses decrease viral infectivity while inducing the emergence of mutations in viral genomes, decreasing fitness. | Despite no or limited pre-clinical evidence, repurposed drugs are massively evaluated in clinical trials to palliate the lack of antiviral molecules against SARS-CoV-2. Here we use a Syrian hamster model to assess the antiviral efficacy of favipiravir, understand its mechanism of action and determine its pharmacokinetics. When treatment is initiated before or simultaneously to infection, favipiravir has a strong dose effect, leading to reduction of infectious titers in lungs and clinical alleviation of the disease. Antiviral effect of favipiravir correlates with incorporation of a large number of mutations into viral genomes and decrease of viral infectivity. Antiviral efficacy is achieved with plasma drug exposure comparable with those previously found during human clinical trials. Notably, the highest dose of favipiravir tested is associated with signs of toxicity in animals. Thereby, pharmacokinetic and tolerance studies are required to determine whether similar effects can be safely achieved in humans. | IntroductionIn March 2020, the World Health Organization declared coronavirus disease 2019 (COVID-19) a pandemic1. The COVID-19 outbreak was originally identified in Wuhan, China, in December 2019 and spread rapidly around the world within a few months. The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the causative agent of COVID-19, belongs to the Coronaviridae family and is closely related to the SARS-CoV, which emerged in China in 20022. After an incubation period of about 5 days, disease onset usually begins with an influenza-like syndrome associated with high virus replication in respiratory tracts3,4. In some patients, a late acute respiratory distress syndrome, associated with high levels of inflammatory proteins, occurs within one to two weeks3. As of 11 November 2020, more than 90 million cases of COVID-19 have resulted in more than 1,936,000 deaths5. In the face of this ongoing pandemic and its unprecedented repercussions, not only on human health but also on society, ecology and economy, there is an urgent need for effective infection prevention and control measures.Whilst host-directed and immune-based therapies could prove useful for the clinical management of critically ill patients, the availability of safe and effective antiviral molecules would represent an important step toward fighting the current pandemic. As conventional drug development is a slow process, repurposing of drugs already approved for any indication was extensively explored and led to the implementation of many clinical trials for the treatment of COVID-196. However, the development of effective antiviral drugs for the treatment of COVID-19, should, as much as possible, rely on robust pre-clinical in vivo data, not only on efficacy generated in vitro. Accordingly, rapid implementation of rodent and non-human primate animal models should help to assess more finely the potential safety and efficacy of drug candidates and to determine appropriated dose regimens in humans7,8.Favipiravir (6-fluoro-3-hydroxypyrazine-2-carboxamine) is an anti‐influenza drug approved in Japan that has shown broad-spectrum antiviral activity against a variety of other RNA viruses9–15. Favipiravir is a prodrug that is metabolized intracellularly into its active ribonucleoside 5′-triphosphate form that acts as a nucleotide analog to selectively inhibit RNA-dependent RNA polymerase and induce lethal mutagenesis16,17. Recently, several studies reported in vitro inhibitory activity of favipiravir against SARS-CoV-2 with 50% effective concentrations (EC50) ranging from 62 to > 500 µM (10 to > 78 µg/mL)18–20. Based on these results, more than 20 clinical trials on the management of COVID-19 by favipiravir are ongoing (https://clinicaltrials.gov/).In the present study, we evaluate the efficacy of favipiravir in vitro and using a Syrian hamster model (Mesocricetus auratus). Our results show that preventive or preemptive administration of high doses favipiravir induce significant reduction of infectious titers and histopathological damages in lungs and clinical alleviation of the disease. Analysis of genetic diversity of viral populations in lungs also confirms the mutagenic effect of favipiravir.ResultsIn vitro efficacy of favipiravirUsing VeroE6 cells and an antiviral assay based on reduction of cytopathic effect (CPE), we recorded EC50 and EC90 of 204 and 334 µM using a multiplicity of infection (MOI) of 0.001, 446, and > 500 µM with an MOI of 0.01 (Table 1 and Supplementary Fig. 1) in accordance with previous studies18–20. Infectious titer reductions (fold change in comparison with untreated cells) were ≥ 2 with 125 µM of favipiravir and ranged between 11 and 342 with 500 µM. Using Caco-2 cells, which do not exhibit CPE with SARS-CoV-2 BavPat1 strain, infectious titer reductions were around 5 with 125 µM of favipiravir and ranged between 144 and 7721 with 500 µM of the drug. 50% cytotoxic concentrations (CC50) in VeroE6 and Caco-2 cells were > 500 µM.Table 1In vitro efficacy of favipiravir.Cell lineMOIDrug effective concentrationaInfectious titer reductionbEC50EC90125 µM250 µM500 µMVero E60.001204 µM334 µM2.213.2341.90.01446 µM>500 µM2.05.710.9Caco-20.001nana5.6137.47720.80.01nana4.07.2144.0MOI multiplicity of infection, na not applicable.aEstimated from dose–response curves of antiviral activity (Supplementary Fig. 1).bCalculated using mean infectious titers without favipiravir (virus control).Infection of Syrian hamsters with SARS-CoV-2Following Chan et al., we implemented a hamster model to study the efficacy of antiviral compounds7. Firstly, we intranasally infected 4-week-old female Syrian hamsters with 106 TCID50 of virus. Groups of two animals were sacrificed 2, 3, 4, and 7 days post-infection (dpi). Viral replication was quantified in sacrificed animals by RT-qPCR in organs (lungs, brain, liver, small/large bowel, kidney, spleen, and heart) and plasma. Viral loads in lungs peaked at 2 dpi, remained elevated until 4 dpi and dramatically decreased at 7 dpi (Supplementary 2). Viral loads in plasma peaked at 3 dpi and viral replication was detected in the large bowel at 2 dpi (Supplementary Fig. 2 and Supplementary Data 1). No viral RNA was detected in almost all the other samples tested (Supplementary Data 1). Subsequently, we infected animals with two lower virus inocula (105 and 104 TCID50). Viral RNA was quantified in lungs, large bowel, and plasma from sacrificed animals 2, 3, 4, and 7 dpi (Supplementary Fig. 2 and Supplementary Data 1). Viral loads in lungs peaked at 2 and 3 dpi with inocula of 105 and 104 TCID50, respectively. Maximum viral loads in lungs of animals infected with each virus inoculum were comparable. Viral RNA yields in plasma and large bowel followed a similar trend but with more variability, with this two lower inocula. In addition, clinical monitoring of animals showed no marked symptoms of infection but normalized weights (i.e., % of initial weights) significantly lower from 3 dpi when compared to animals intranasally inoculated with sodium chloride 0.9% (Supplementary Fig. 2).In vivo efficacy of favipiravirTo assess the efficacy of favipiravir, hamsters received the drug, intraperitoneally, three times a day (TID). We used three doses of favipiravir: 18.75, 37.5, and 75 mg/day (corresponding to 340 ± 36, 670 ± 42 and 1390 ± 126 mg/kg/day, respectively).In a first set of experiments, treatment was initiated at the day of infection (preemptive antiviral therapy) and ended at 2 dpi. We infected groups of 6 animals intranasally with three virus inocula (106, 105, and 104 TCID50) and viral replication was measured in lungs and plasma at 3 dpi (Fig. 1a). Each virus inoculum was assessed in an independent experiment. When analysis of virus replication in clarified lung homogenates was based on infectious titers (as measured using TCID50 assay), the effect of favipiravir in reducing infectious titers was dose dependent, in particular when low virus inocula were used to infect animals (Fig. 1b). At each virus inoculum, mean infectious titers for groups of animals treated with 75 mg/day TID were significantly lower than those observed with untreated groups (p ≤ 0.0001): reduction of infectious titers ranged between 1.9 and 3.7 log10. For animals infected with 105 or 104 TCID50, significant infectious titer reductions of around 0.8 log10 were also observed with the dose of 37.5 mg/day TID (p ≤ 0.038). Drug 90 and 99% effective doses (ED90 and ED99) were estimated based on these results and ranged between 31–42 mg/day and 53–70 mg/day, respectively (Table 2). When analysis of virus replication in clarified lung homogenates were assessed on viral RNA yields (as measured using quantitative real-time RT-PCR assay), significant differences with groups of untreated animals, ranging between 0.7 and 2.5 log10, were observed only with the higher dose of favipiravir (p ≤ 0.012). Once again, this difference was more noticeable with lower virus inocula (Fig. 1c). Since we found higher reductions of infectious titers than those observed with viral RNA yields, we estimated the relative infectivity of viral particle (i.e., the ratio of the number of infectious particles over the number of viral RNA molecules). Decreased infectivity was observed in all treated groups of animals. These differences were always significant with the higher dose of favipiravir (p ≤ 0.031) and were significant with the dose of 37.5 mg/day TID for animals infected with 105 or 104 TCID50 of virus (p ≤ 0.041) (Fig. 1d). We then measured plasma viral loads using quantitative real-time RT-PCR assay and found, with the higher dose of favipiravir and the groups of animals infected with 106 or 104 TCID50, significant reductions of 2.1 and 2.62 log10, respectively (p ≤ 0.022) (Fig. 1e). Finally, signs of toxicity were observed with animal treated with the dose of 75 mg/day TID: normalized weights were significantly lower than those of untreated animals (Fig. 1f).Fig. 1Virological results with preemptive favipiravir therapy.a Experimental timeline. Groups of 6 hamsters were intranasally infected with 106, 105 or 104 TCID50 of virus. b Viral replication in lung based on infectious titers (measured using a TCID50 assay) expressed in TCID50/copy of ɣ-actine gene (n = 6 animals/group). c Viral replication in lung based on viral RNA yields (measured using an RT-qPCR assay) expressed in viral genome copies/copy of ɣ-actine gene (n = 6 animals/group). d Relative lung viral particle infectivities were calculated as follows: ratio of lung infectious titer over viral RNA yields (n = 6 animals/group). e Plasma viral loads (measured using an RT-qPCR assay) are expressed in viral genome copies/mL of plasma (the dotted line indicates the detection threshold of the assay) (n = 6 animals/group). f Clinical course of the disease (n = 6 animals/group). Normalized weight at day n was calculated as follows: % of initial weight of the animal at day n. Data represent mean ± SD (details in Supplementary Data 2). Two-sided statistical analysis were performed using Shapiro–Wilk normality test, Student t-test, Mann–Whitney test, Welch’s test, and two-way ANOVA with Post-hoc Dunnett’s multiple comparisons test (details in Supplementary Data 3 and 4). ****, ***, ** and * symbols indicate that the average value for the group is significantly lower than that of the untreated group with a p-value < 0.0001, ranging between 0.0001–0.001, 0.001–0.01, and 0.01–0.05, respectively. Source data are provided as a Source data file.Table 2Drug effective doses (ED) on reducing viral titers according to the level of viral inoculum.Virus inoculumED50mg/day (95%CIa)ED90mg/day (95%CIa)ED99mg/day (95%CIa)Preemptive therapy104 TCID5034 (30–37)42 (38–46)53 (48–58)105 TCID5026 (21–30)37 (31–44)56 (46–65)106 TCID5015 (10–20)31 (21–41)70 (48–93)Preventive therapy104 TCID5027 (25–29)35 (32–38)47 (44–51)Dose–response curves are presented in Supplementary Fig. 3.a95% confidence interval.In a second set of experiments, we assessed, over a period of 7 days, the impact of the preemptive therapy on the clinical course of the disease using weight as the primary criterion (Fig. 2a). Since signs of toxicity were noticed during the first set of experiments, we evaluated the toxicity of the three doses of favipiravir with groups of four non-infected animals treated during four days (Fig. 2b). Important toxicity was observed with the dose of 75 mg/day TID with, from the first day of treatment, normalized weights significantly lower than those of untreated animals (Supplementary Data 5). We also found a constant, but moderate, toxicity with the dose of 37.5 mg/day TID that was significant at day 4, 5, and 6 only. No toxicity was detected with the lower dose of favipiravir. To assess if the toxicity observed with the highest dose of favipiravir was exacerbated by the infection, we compared normalized weights of infected and non-infected animals treated with the dose of 75 mg/day TID. Regardless of the virus inoculum, no significant difference was observed at 1, 2, and 3 dpi (Supplementary Fig. 4). After this evaluation of favipiravir toxicity, we intranasally infected groups of 10 animals with two virus inocula (105 or 104 TCID50). Each virus inoculum was assessed in an independent experiment. Treatment with a dose of 37.5 mg/day TID was initiated on the day of infection (preemptive antiviral therapy) and ended at 3 dpi (Fig. 2a). With both virus inocula, treatment was associated with clinical alleviation of the disease (Fig. 2c, d). With the inoculum of 105 TCID50, mean weights of treated animals were significantly higher than those of untreated animals at 5 and 6 dpi (p ≤ 0.031). Similar observations were made with the inoculum of 104 TCID50 at 5, 6, and 7 dpi (p < 0.0001).Fig. 2Clinical follow-up of animals.a Experimental timeline. b Evaluation of the toxicity of the three doses of favipiravir (mg/day TID) with groups of four uninfected animals following the experimental timeline described in panel a but without infection. c, d Clinical follow-up with groups of 10 animals infected respectively with 105 and 104 TCID50 of virus and treated with a dose of favipiravir of 37.5 mg/day TID. Normalized weight at day n was calculated as follows: % of initial weight of the animal at day n. Data represent mean ± SD (details in Supplementary Data 2). Two-sided statistical analysis were performed using two-way ANOVA with Post-hoc Dunnett’s multiple comparisons test or Post-hoc Sidak’s multiple comparisons test (details in Supplementary Data 5). ****, ***, ** and * symbols indicate that the average value for the group is significantly lower than that of the untreated group with a p-value < 0.0001, ranging between 0.0001–0.001, 0.001–0.01, and 0.01–0.05, respectively Source data are provided as a Source data file.In a third set of experiments, treatment was started 1 day before infection (preventive antiviral therapy) and ended at 2 dpi. We intranasally infected groups of 6 animals with 104 TCID50 of virus and viral replication was measured in lungs and plasma at 3 dpi (Fig. 3a). Once again, an inverse relationship was observed between lung infectious titers and the dose of favipiravir (Fig. 3b). Mean infectious titers for groups of animals treated with 37.5 and 75 mg/day TID were significantly lower than those observed with untreated groups (p ≤ 0.002). Of note, undetectable infectious titers were found for all animals treated with the higher dose. Estimated ED90 and ED99 were 35 and 47 mg/day, respectively (Table 2). Significant reductions of viral RNA yields of 0.9 and 3.3 log10, were observed with animals treated with 37.5 and 75 mg/day TID, respectively (p ≤ 0.023) (Fig. 3c). Resulting infectivity of viral particle was decreased, with a significant reduction only for the higher dose of favipiravir (p = 0.005) (Fig. 3e). Finally, we found significantly reduced plasma viral loads with animals treated with 37.5 and 75 mg/day TID (p ≤ 0.005) (Fig. 3f). Once again, signs of toxicity were observed with animal treated with the dose of 75 mg/day TID: normalized weights were significantly lower than those of untreated animals (Fig. 3d).Fig. 3Virological results with preventive favipiravir therapy.a Experimental timeline. Groups of 6 hamsters were intranasally infected with 104 TCID50 of virus. b Viral replication in lungs based on infectious titers (measured using a TCID50 assay) expressed in TCID50/copy of ɣ-actine gene (n = 6 animals/group). c Viral replication in lungs based on viral RNA yields (measured using an RT-qPCR assay) expressed viral genome copies/copy of ɣ-actine gene (n = 6 animals/group). d Clinical course of the disease (n = 6 animals/group). Normalized weight at day n was calculated as follows: % of initial weight of the animal at day n. e Relative lung virus infectivities were calculated as follows: ratio of lung infectious titer over viral RNA yields (n = 6 animals/group). f Plasma viral loads (measured using an RT-qPCR assay) are expressed in viral genome copies/mL of plasma (the dotted line indicates the detection threshold of the assay) (n = 6 animals/group). Data represent mean ± SD (details in Supplementary Data 2). Statistical analysis were performed using Shapiro–Wilk normality test, Student t-test, Mann–Whitney test, One-sample t-test and two-way ANOVA with Post-hoc Dunnett’s multiple comparisons test (details in Supplementary Data 3 and 4). ****, ** and * symbols indicate that the average value for the group is significantly different from that of the untreated group with a p-value < 0.0001, ranging between 0.001–0.01 and 0.01–0.05, respectively. Source data are provided as a Source data file.In a last set of experiments, we assessed the impact of favipiravir treatment on lung pathological changes induced by SARS-CoV-2. Animals were intranasally infected with 104 TCID50 of virus. Treatment with two doses of favipiravir (37.5 and 75 mg/day TID) was initiated one day before infection (preventive antiviral therapy) or at day of infection (preemptive antiviral therapy) and ended at 3 dpi. For each therapeutic strategy and for each dose of favipiravir, a group of four animals was sacrificed at 3 and 5 dpi (Fig. 4a and c). As a control, we used four vehicle-treated groups of four animals (one at 3 dpi and one at 5 dpi for each therapeutic strategy). Based on the severity of inflammation, alveolar hemorrhagic necrosis and vessel lesions, a cumulative score from 0 to 10 was calculated and assigned to a grade of severity (0 = normal; 1 = mild; 2 = moderate; 3 = marked and 4 = severe; details in Supplementary Data 7). Overall, lungs of untreated animals displayed typical lesions of air-borne infection (i.e., broncho-interstitial pneumonia), with a progression between 3 dpi and 5 dpi that reflects the virus dissemination within the respiratory tree as previously demonstrated7,21. At 3 dpi, 7/8 untreated animals displayed mild pulmonary pathological changes (Fig. 4b and d) leading to difficulty to assess the efficacy of the treatment even if almost all mean cumulative scores of treated animals were significantly lower than those of untreated groups. In contrast, at 5 dpi all untreated animals displayed severe pulmonary impairments and we observed a dose-dependent effect of favipiravir (Fig. 4b and d). When using a preemptive antiviral strategy, all animals treated with 37.5 mg/day TID had marked histopathological damages in lungs and animals treated with 75 mg/day TID displayed mild or moderate histopathological damages (Supplementary Fig. 5). When using a preventive antiviral strategy, all animals treated with 37.5 mg/day TID had mild to marked damages in lung and animals treated with 75 mg/day TID displayed no or mild histopathological damages (Fig. 4e–h). At 5 dpi, significant cumulative score reductions were observed with both doses of favipiravir regardless the therapeutic strategy used (p = 0.0286, details in Supplementary Data 8).Fig. 4Lung histopathological changes with preemptive or preventive favipiravir therapy.Groups of four animals were intranasally infected with 104 TCID50 of virus and sacrificed at 3 and 5 dpi. Experimental timelines for preemptive (a) and preventive (c) favipiravir therapies. At day of sacrifice, lungs were collected, fixed, and embedded in paraffin. Tissue sections were stained with hematoxylin-eosin (H&E). Based on severity of inflammation, alveolar hemorrhagic necrosis, and vessel lesions, a cumulative score from 0 to 10 was calculated and assigned to a grade of severity (I, II, III, and IV). Scoring of pathological changes for preemptive (b) and preventive (d) favipiravir therapies (n = 4 animals/group) (details in Supplementary Data 7). Two-sided statistical analysis were performed using Shapiro–Wilk normality test, Student t-test, Mann–Whitney test, and two-way ANOVA with Post-hoc Dunnett’s multiple comparisons test (details in Supplementary Data 7 and 8). * Symbol indicates that the average value for the group is significantly different from that of the untreated group with a p-value ranging between 0.01 and 0.05. e Representative images of lung tissue (left lobe) (scale bar: 4 mm): multifocal and extensive areas of inflammation for untreated animal, multifocal but limited areas of inflammation for 37.5 mg/day treated animal and normal lung for 75 mg/day treated animal (n = 4 samples/group). f Representative images of bronchial inflammation (scale bar: 100 µ): severe peribronchiolar inflammation and bronchiole filled with neutrophilic exudates for untreated animal, mild peribronchiolar inflammation for 37.5 mg/day treated animal and normal bronchi for 75 mg/day treated animal (n = 4 samples/group). g Representative images of alveolar inflammation (scale bar: 100 µ): severe infiltration of alveolar walls, alveoli filled with neutrophils/macrophages for untreated animal, moderate infiltration of alveolar walls, some alveoli filled with neutrophils/macrophages for 37.5 mg/day treated animal and normal alveoli for 75 mg/day treated animal. h Representative images of vessel inflammation (scale bar: 50 µ): infiltration of vascular wall with neutrophils/cell debris and endothelial damage (arrow) for untreated animal, moderate endothelial leukocytic accumulation for 37.5 mg/day treated animal and normal vessel for 75 mg/day treated animal (n = 4 samples/group). Clinical courses of the disease are presented in Supplementary Fig. 6. Source data are provided as a Source data file.Favipiravir pharmacokinetics (PK) in a hamster modelWe first assessed the PK and lung distribution of favipiravir in a subgroup of uninfected animals. Groups of animals were treated respectively with a single dose of favipiravir administrated intraperitoneally: 6.25 mg, 12.5 mg, and 25 mg. In each dose group, we sacrificed three animals at specific time points post-treatment (0.5, 1, 5 or 8 h) for determination of favipiravir in plasma. Drug concentration in lung tissue was determined at 0.5 and 5 h post-treatment. Subsequently, we assessed the favipiravir concentration after multiple dose in animals intranasally infected with 105 TCID50 of virus. Groups of nine animals received the three doses evaluated for 3 days (Fig. 1): 18.75 mg/day, 37.5 mg/day or 75 mg/day TID and were sacrificed at 12-h after the last treatment dose. Favipiravir trough concentrations were quantified in plasma (n = 9) and lung tissue (n = 3).Results are presented in Table 3 and Supplementary Fig. 7. The single dose PK analysis showed that the maximum concentration of favipiravir was observed at 0.5 h at all doses, then plasma drug concentrations decreased exponentially to reach concentrations below 10 µg/ml at 12 h. Favipiravir PK exhibited a non-linear increase in concentration between the doses. After multiple doses, trough concentrations (12 h) of favipiravir also exhibited a non-linear increase between doses. The extrapolated 12 h post-treatment concentrations after a single dose were calculated in order to determine the accumulation ratio. Accumulation ratios were respectively 6, 16, and 21 at the three doses, confirming the non-proportional increase between doses. The average concentration after single dose administration over 0–12-h intervals was calculated and the respective values obtained were 10.1 µg/mL, 38.7 µg/mL, and 100.5 µg/mL for the three favipiravir doses.Table 3Plasma and lung concentrations of favipiravir after administration of a single dose or multiple dose of favipiravir.Time post-treatmentSingle doseMultiple dosea (Day 3)DosePlasma (µg/mL)Lung (µg/g)L/p ratioDosePlasma (µg/mL)Lung (µg/g)L/p ratio0.5 h25 mg372 ± 47.5216 ± 390.58 ± 0,0475 mg/day TID1 h279 ± 49.95 h135 ± 49.081.3 ± 240.62 ± 0.108 h5.77 ± 1.3412 h1.43b29.9 ± 9.8316.0 ± 4.870.44 ± 0,070.5 h12.5 mg166 ± 52.090.7 ± 12.70.58 ± 0.1437.5 mg/day TID1 h155 ± 20.65 h10.7 ± 5.163.84 ± 1.490.37 ± 0.0528 h1.94 ± 0.0612 h0.16b2.57 ± 1.221.36 ± 0.140.35 ± 0,030.5 h6.25 mg86.3 ± 4.1150.2 ± 16.40.58 ± 0.1718.75 mg/day TID1 h35.2 ± 27.85 h2.90 ± 0.251.09 ± 0.050.38 ± 0.058 h0.56 ± 0.1612 h0.05b0.31 ± 0.14Not detectedNANA not applicable.Data represent mean ± SD; three animals for each condition except at multiple dose (n = 9 for plasma; n = 3 for lung); details in Supplementary Data 9.aPK realized after 3 days of favipiravir administered three times a day, at the end of the dosing interval (trough concentrations).bExtrapolated C12h.Favipiravir lung concentrations were 1.6–2.7-fold lower than in plasma for both administration of single and multiple doses. After a single dose, the mean lung to plasma ratio ranged from 0.37 to 0.62 according to the time post-treatment and was similar between the three doses of favipiravir at 0.5 h. A high ratio 5 h post-treatment was observed at the highest dose (25 mg) with an increase by a factor 1.6–1.8 compared with the lower doses. After multiple doses, the lung penetration of favipiravir was confirmed with a mean lung to plasma ratio ranging from 0.35 to 0.44. Favipiravir was not detected in the lungs at the lowest dose (18.75 mg/day).Mutagenic effect of favipiravirTo understand which genomic modifications accompanied favipiravir treatment, direct complete genome sequencing of clarified lung homogenates from animals intranasally infected with 106 TCID50 of virus and treated with the two highest doses of drug (preemptive antiviral therapy; Fig. 1) was performed. Data were generated by next-generation sequencing from lung samples of four animals per group (untreated, 37.5 mg/day TID and 75 mg/day TID). The mean sequencing coverage for each sample ranged from 10,991 to 37,991 reads per genomic position and we subjected substitutions with a frequency ≥ 1% to further analysis. The genetic variability in virus stock was also analyzed: 14 nucleotide polymorphisms were detected of which 5 recorded a mutation frequency higher than 10% (Supplementary Data 10).In order to study the mutagenic effect of favipiravir, we used the consensus sequence from virus stock as reference and all the mutations simultaneously detected in a lung sample and in virus stock were not considered in the further analysis (1–4 mutations per sample, see Supplementary Data 10). Overall, no majority mutations were detected (mutation frequency > 50%), and almost all of the mutations occurred at a frequency lower than 10% (Fig. 5a). In addition, mutations were distributed throughout the whole genome (Fig. 5b).Fig. 5Mutagenic effect of favipiravir.a Viral genetic diversity in clarified lung homogenates. For each condition, four samples were analyzed. Each triangle represents a mutation (only substitutions with a frequency ≥ 1% were considered). b Patterns of mutation distribution on complete viral genome. Each variable nucleotide position was counted only once when found. The variability was represented using 75 nt sliding windows. For each condition, variable nucleotide positions were determined and represented using a 300 nt sliding window. c Mean number of mutations (n = 4 samples/group). Data represent mean ± SD. d Mutation characteristics (n = 4 samples/group). For each sample, the frequency of a given mutation was calculated as follows: number of this kind of mutation detected in the sample divided by the total number of mutations detected in this sample. Data represent mean ± SD (details in details in Supplementary Data 10 and 13). Two-sided statistical analysis were performed using Shapiro–Wilk normality test, Student t-test, Mann–Whitney test, and Welch’s test (details in Supplementary Data 11 and 12). ***, ** and * symbols indicate that the average value for the group is significantly lower than that of the untreated group with a p-value ranging between 0.0001–0.001, 0.001–0.01, and 0.01–0.05, respectively. e Association between lung infectious titers (measured using a TCID50 assay) and frequency of non synonymous, synonymous and G → A mutations. Each dot represent data from a given animal. Statistical analysis was performed using univariate linear regression. The error band (in gray) represent the 95% confidence interval of the regression line. Source data are provided as a Source data file.Results revealed a relationship between the number of mutations detected per sample and the dose of favipiravir (Fig. 5c): the mean number of mutations increased by a factor 2 and 4.8 with groups of animals treated with 37.5 and 75 mg/day, TID respectively. The difference is significant only with a dose of 37.5 mg/day TID (p = 0.029). This increase of the number of mutations is mainly the consequence of the occurrence of a large number of G → A substitutions and, to a lesser extent, C → U substitutions. Consequently, regardless of the dose of favipiravir, mean frequency of G → A substitutions was significantly increased by a factor of 4.2 (p ≤ 0.009). This rise of these transition mutations led to increased frequency of all transition mutations (significant only at dose of 37.5 mg/day TID; p = 0.037) and increased frequency of non-synonymous mutations (significant only at dose of 75 mg/day TID; p = 0.009) (Fig. 5d). We investigated whether or not effectiveness in treated animals was linked with the characteristics of the mutations detected on viral populations and found that infectious titers in lungs were negatively associated with frequency of non-synonymous and G → A mutations, and positively associated with frequency of synonymous mutations (p < 0.03; Fig. 5e). Finally, our experiments revealed some parallel evolution events; 32 substitutions in viral sub-populations were detected in two independent animals. Notably, 18 of these shared mutations were detected only with treated animals, 14 of them being non-synonymous (Supplementary Data 13). These mutations are located in nsp2, 3, 4, 5, 6, 14, N protein, Matrix, ORF 3a and 8. At this stage, one cannot conclude if these substitutions reflect the adaptation to the hamster model or are the result of the antiviral selection.DiscussionIn the current study, we used a hamster model to assess efficacy of the favipiravir against SARS-CoV-2. Following infection, viral RNA was mainly detected in lungs, blood, and, to a lesser extent, in the large bowel. Peak of viral replication was observed at 2–3 dpi, in line with recently reported investigations that involved 6–10-weeks-old hamsters7. Clinically, the main symptom was the lack of weight gain, observed from the first day of infection and followed by recovery at 7 dpi. Histopathological changes are comparable to those previously described7,21. Notably, our results revealed that all animals with marked or severe pulmonary impairments displayed vascular lesions (endothelitis, vasculitis) as previously described in humans22. Overall, this confirmed that the in vivo model, with younger animals (4 weeks-old), is suitable for preclinical evaluation of antiviral compounds against SARS-CoV-2.Using a preemptive strategy, we demonstrated that doses of favipiravir of around 700–1400 mg/kg/day TID reduced viral replication in the lungs of infected animals and allowed clinical alleviation of the disease (Figs. 1 and 2). In the most favorable situation, where high doses were used as a preventive therapy, favipiravir led to undetectable viral replication in lung and plasma. These results showed that the use of high doses of favipiravir could expand its in vivo spectrum against RNA viruses. Reduction of viral replication was greater when estimated on the basis of infectious titers than on total viral RNA as previously observed in non-human primates treated with Remdesivir and in hamsters treated with favipiravir23,24. Furthermore, the analysis of pulmonary histopathological changes revealed that favipiravir played a protective role by reducing the severity of the lesions. However, the effective doses of favipiravir were higher than those usually used in rodent models (≈100–400 mg/kg/day)10,12,25–28. This can be correlated with the high favipiravir EC50 found in vitro for SARS-CoV-2. Moreover, effective doses were associated with significant toxicity in our hamster model (Fig. 2). This observed toxicity reflected only the adverse effects of favipiravir and was not exacerbated during SARS-CoV-2 infection. Indeed, similar weights were measured among infected and non-infected animals treated with the highest dose of favipiravir at 1, 2, and 3 dpi.In the present study, reduction of viral replication was correlated with the dose of favipiravir administrated and inversely correlated with the virus inoculum. In a recent study, the efficacy of favipiravir intraperitoneally or orally administrated twice daily (loading dose of 900 and 1200 mg/kg/day followed by 600 and 1000 mg/kg/day, respectively) was assessed using a similar hamster model (6–10 weeks old) with high virus inocula (2 × 106 TCID50)24. Treatment with the highest dose of favipiravir resulted in a moderate decrease of viral RNA yields in lung tissue and the lowest dose induced an even smaller inhibitory effect. However, significant infectious titers reduction were observed in a dose-dependent manner in lungs. Both doses were also associated with regression of pulmonary histopathological impairments. Overall, these results are in accordance with ours at the medium and the high doses of favipiravir (around 670 and 1390 mg/kg/day TID). However, in this other study, no signs of toxicity were associated with favipiravir treatment regardless the dosing regimen. This discrepancy could be due to the difference between (i) the highest daily doses used (1000 mg/kg/day in regards to 1390 mg/kg/day in our study), (ii) the dosing regimens (BID instead of TID in our study), and/or (iii) the age of the hamsters at day of infection (6–10 weeks old in comparison to 4 weeks old in our study).With influenza viruses, favipiravir acts as a nucleotide analog since it is recognized as a purine nucleotide by the viral RNA-dependent RNA polymerase. It is metabolized intracellularly to its active form and incorporated into nascent viral RNA strands. This inhibits RNA strand extension and induces abnormal levels of mutation accumulation into the viral genome16,17. Recently, it was shown in vitro that favipiravir has a similar mechanism of action with SARS-CoV-2 through a combination of chain termination, reduced RNA synthesis and lethal mutagenesis20. Our genomic analysis confirmed the mutagenic effect of favipiravir in vivo24. Indeed, we found that favipiravir treatment induced appearance of a large number of G → A mutations into viral genomes (Fig. 5). This was associated to a decrease of viral infectivity probably because alteration of the genomic RNA disturb the replication capacity. Similar findings were described in vitro and in vivo with other RNA viruses9,16,29,30. Of note, we also observed a strong inverse association between infectious titers in lungs and the proportion of non-synonymous mutations detected in viral populations. Because random non-synonymous mutations are more deleterious than synonymous mutations31, this suggests that they were randomly distributed over the three positions of the codons and that no compensatory mechanism was triggered by the virus to eliminate them (i.e. negative selection). Finally, the inverse correlation between lung infections titers and the frequency of G → A substitutions showed that an increased proportion of these mutations beyond an error threshold might be expected to cause lethal mutagenesis.Genomic analyses revealed that 18 mutations detected in viral sub-populations were shared only with treated animals. Two of them were located in the nsp14 coding region involved in the proof-reading activity of the viral RNA polymerisation32,33. However, they were located in the N7 MTase domain involved in viral RNA capping34,35. By comparison, resistance mutations selected against Remdesivir in β-coronavirus murine hepatitis virus model were obtained in the RdRP (nsp12) coding sequence36. Further investigations are needed to assess the impact of these mutations on the antiviral effect of favipiravir.Favipiravir PK in our hamster model displayed a non-linear increase in plasma exposure between the doses as already reported in non-human primates37. The observed favipiravir concentration versus time profiles were in agreement with previous results of a PK study performed in 7–8-week-old hamsters orally treated with a single dose of 100 mg/kg of favipiravir38. The maximum plasma drug concentration occurred at 0.5 h after oral administration, earlier than in humans, and then decreased rapidly in agreement with its short half-life39. After repeated doses, plasma exposure confirmed non-linear PK over the entire range of doses, further emphasized by accumulation ratios. The important accumulation observed at the highest dose could explain in part the toxicity observed in hamsters at this dose. Favipiravir undergoes an important hepatic metabolism mainly by aldehyde oxidase producing an inactive M1 metabolite and inhibits aldehyde oxidase activity in a concentration- and time-dependent manner. These properties explain the self-inhibition of its own metabolism as observed in our study in which the highest dose of favipiravir led to a greater increase in favipiravir concentrations40.A good penetration of favipiravir in lungs was observed with lung/plasma ratios ranging from 35 to 44% after repeated doses, consistent with its physicochemical properties. Lung exposure was also in accordance with the previous studies38.The medium dose of favipiravir used in this study (670 mg/kg/day TID) is within the range of the estimated doses required to reduce by 90% (ED90) the level of infectious titers in lungs (ranging between 31 and 42 mg/day corresponding to 570–780 mg/kg/day) (Table 2) and displayed limited drug-associated toxicity (Fig. 2b). Animals infected with 105 and 104 TCID50 of virus, and treated following a preemptive strategy with this dose displayed significant reduction of infectious titers and histopathological scores in lungs and clinical alleviation of the disease (Figs. 1, 2, and 4). Animal treated following a preventive strategy with this dose also displayed significant reduction of viral replication and histopathological scores in lungs (Figs. 3 and 4). Regarding the accumulation ratio after repeated doses and the good penetration of favipiravir in lungs, effective concentrations can be expected in lungs, throughout the course of treatment using this dose of 670 mg/kg/day TID.How clinically realistic are these results? To address this question we compared the drug concentrations obtained in the hamster model with those obtained in patients. In 2016, a clinical trial evaluated the use of favipiravir in Ebola-infected patients41. The dose used in Ebola-infected patients was 6000 mg on day 0 followed by 1200 mg BID for 9 days. The median trough concentrations of favipiravir at day 2 and day 4 were 46.1 and 25.9 µg/mL, respectively. This is within the range observed here in hamsters treated with the highest dose (around 1400 mg/kg/day), with a mean trough concentration of 29.9 µg/mL. However, additional investigations are required to determine whether or not similar favipiravir plasma exposure in SARS-COV-2 infected patients are associated with antiviral activity. The major differences in PK between hamster and humans, and the toxicity observed at the highest doses in our animal model limits the extrapolation of our results. Therefore, whether safe dosing regimens in humans may achieve similar plasma exposure and recapitulate the profound effect on viral replication is unknown. Further, the intracellular concentration of the active metabolite was not determined and which parameter of the drug pharmacokinetics best drives the antiviral effect remains to be established.In summary, this study establishes that high doses of favipiravir are associated with antiviral activity against SARS-CoV-2 infection in a hamster model. The better antiviral efficacy was observed using a preventive strategy, suggesting that favipiravir could be more appropriate for a prophylactic use. Our results should be interpreted with caution because high doses of favipiravir were associated with signs of toxicity in our model. It is required to determine if a tolerable dosing regimen could generate similar exposure in non-human primates, associated with significant antiviral activity, before testing a high dose regimen in COVID-19 patients. Furthermore, subsequent studies should determine if an increased antiviral efficacy can be reached using favipiravir in association with other effective antiviral drugs, since this strategy may enable to reduce the dosing regimen of favipiravir. Finally, this work reinforces the need for rapid development of animal models to confirm in vivo efficacy of antiviral compounds and accordingly, to determine appropriate dose regimens in humans before starting clinical trials.MethodsCellsVeroE6 cells (ATCC CRL-1586) and Caco-2 cells (ATCC HTB-37) were grown at 37 °C with 5% CO2 in minimal essential medium (MEM) supplemented with 7.5% heat-inactivated fetal bovine serum (FBS), 1% penicillin/streptomycin and 1% non-essential amino acids (all from ThermoFisher Scientific).VirusAll experiments with infectious virus were conducted in biosafety level (BSL) 3 laboratory. SARS-CoV-2 strain BavPat1, supplied through European Virus Archive GLOBAL (https://www.european-virus-archive.com/), was provided by Christian Drosten (Berlin, Germany). Virus stocks were prepared by inoculating at MOI of 0.001 a 25 cm 2 culture flask of confluent VeroE6 cells with MEM medium supplemented with 2.5% FBS. The cell supernatant medium was replaced each 24 h and harvested at the peak of infection, supplemented with 25 mM HEPES (Sigma), aliquoted and stored at −80 °C.In vitro determination of EC50, EC90, CC50, and infectious titer reductionsOne day prior to infection, 5 × 104 VeroE6 cells were seeded in 96-well culture plates (5 × 104 cells/well in 100 µL of 2.5% FBS medium (assay medium). The next day, seven 2-fold serial dilutions of favipiravir (Courtesy of Toyama-Chemical; 0.61 µg/mL to 78.5 µg/mL, in triplicate) were added (25 µL/well, in assay medium). Eight virus control wells were supplemented with 25 µL of assay medium and eight cell controls were supplemented with 50 µL of assay medium. After 15 min, 25 µL of virus suspension, diluted in assay medium, was added to the wells at an MOI of 0.01 or 0.001 (except for cell controls). Three days after infection, cell supernatant media were collected to perform TCID50 assay (at concentration of 500, 250, and 125 µM), as described below, in order to calculate infectious titer reductions and cell viability was assessed using CellTiter-Blue reagent (Promega) following the manufacturer’s intstructions. Fluorescence (560/590 nm) was recorded with a Tecan Infinite 200Pro machine (Tecan). The 50 and 90% effective concentrations (EC50, EC90) were determined using logarithmic interpolation (% of inhibition were calculated as follows: (ODsample − ODvirus control)/(ODcell control − ODvirus control)). For the evaluation of CC50 (the concentration that induced 50% cytoxicity), the same culture conditions were set as for the determination of the EC50, without addition of the virus, then cell viability was measured using CellTiter Blue (Promega). CC50 was determined using logarithmic interpolation.In vivo experimentsApproval and authorizationIn vivo experiments were approved by the local ethical committee (C2EA—14) and the French ‘Ministère de l’Enseignement Supérieur, de la Recherche et de l’Innovation’ (APAFIS#23975) and performed in accordance with the French national guidelines and the European legislation covering the use of animals for scientific purposes. All experiments were conducted in BSL 3 laboratory.Animal handlingThree-week-old female Syrian hamsters were provided by Janvier Labs. Animals were maintained in ISOcage P - Bioexclusion System (Techniplast) with unlimited access to water/food and 14 h/10 h light/dark cycle. Animals were weighed and monitored daily for the duration of the study to detect the appearance of any clinical signs of illness/suffering. Virus inoculation was performed under general anesthesia (isoflurane). Organs and blood were collected after euthanasia (cervical dislocation) which was also realized under general anesthesia (isoflurane).Hamster InfectionAnesthetized animals (four-week-old) were intranasally infected with 50 µL containing 106, 105 or 104 TCID50 of virus in 0.9% sodium chloride solution. The mock group was intranasally inoculated with 50 µL of 0.9% sodium chloride solution.Favipiravir administrationHamster were intraperitoneally inoculated with different doses of favipiravir. Control group were intraperitoneally inoculated with a 0.9% sodium chloride solution.Organ collectionOrgans were first washed in 10 mL of 0.9% sodium chloride solution and then transferred to a 2 mL or 50 mL tube containing respectively 1 mL (small/large bowel pieces, kidney, spleen, and heart) or 10 mL (lungs, brain and liver) of 0.9% sodium chloride solution and 3 mm glass beads. They were crushed using a Tissue Lyser machine (Retsch MM400) for 5 min at 30 cycles/s and then centrifuged 5 min at 16,200 × g. Supernatant media were transferred to a 2 mL tube, centrifuged 10 min at 16,200 × g, and stored at −80 °C. One milliliter of blood was harvested in a 2 mL tube containing 100 µL of 0.5 M EDTA (ThermoFischer Scientific). Blood was centrifuged for 10 min at 16,200 × g and stored at −80 °C.Quantitative real-time RT-PCR (RT-qPCR) assaysTo avoid contamination, all experiments were conducted in a molecular biology laboratory that is specifically designed for clinical diagnosis using molecular techniques, and which includes separate laboratories dedicated to perform each step of the procedure. Prior to PCR amplification, RNA extraction was performed using the QIAamp 96 DNA kit, and the Qiacube HT kit and the Qiacube HT (both from Qiagen) following the manufacturer’s instructions. Shortly, 100 µl of organ clarified homogenates, spiked with 10 µL of internal control (bacteriophage MS2)42, were transferred into an S-block containing the recommended volumes of VXL, proteinase K, and RNA carrier. RT-qPCR (SARS-CoV-2 and MS2 viral genome detection) were performed with the Express one step RT-qPCR Universal kit (ThermoFisher Scientific) using 3.5 µL of RNA and 6.5 µL of RT-qPCR mix that contains 250 nmol of each primer and 75 nmol of probe. Amplification was performed with the QuantStudio 12K Flex Real-Time PCR System (ThermoFisher Scientific) using the following conditions: 50 °C for 10 min, 95 °C for 20 s, followed by 40 cycles of 95 °C for 3 s, 60 °C for 30 s. qPCR (ɣ-actine gene detection) was perfomed under the same condition as RT-qPCR with the following modifications: we used the Express one step qPCR Universal kit (ThermoFisher Scientific) and the 50 °C step of the amplification cycle was removed. Data were collected using the QuantStudio 12K Flex Software v1.2.3. Primers and probes sequences used to detect SARS-CoV-2, MS2 and ɣ-actine are described in Supplementary Table 1.Tissue-culture infectious dose 50 (TCID50) assayTo determine infectious titers, 96-well culture plates containing confluent VeroE6 cells were inoculated with 150 μL per well of serial dilutions of each sample (four-fold or ten-fold dilutions when analyzing lung clarified homogenates or cell supernatant media, respectively). Each dilution was performed in sextuplicate. Plates were incubated for 4 days and then read for the absence or presence of cytopathic effect in each well. Infectious titers were estimated using the method described by Reed & Muench43.Favipiravir pharmacokineticsAnimal handling, hamster infections, and favipiravir administrations were performed as described above. A piece of left lung was first washed in 10 mL of sodium chloride 0.9% solution, blotted with filter paper, weighed and then transferred to a 2 mL tube containing 1 mL of 0.9% sodium chloride solution and 3 mm glass beads. It was crushed using the Tissue Lyser machine (Retsch MM400) during 10 min at 30 cycles/s and then centrifuged 5 min at 16,200 × g. Supernatant media were transferred to 2 mL tubes, centrifuged 10 min at 16,200 × g and stored at −80 °C. One milliliter of blood was harvested in a 2 mL tube containing 100 µL of 0.5 M EDTA (ThermoFischer Scientific). Blood was centrifuged for 10 min at 16,200 × g and stored at −80 °C.Quantification of favipiravir in plasma and lung tissues was performed by a validated sensitive and selective validated high-performance liquid chromatography coupled with tandem mass spectrometry method (UPLC-TQD, Waters, USA) with a lower limit of quantification of 0.1 µg/mL. Precision and accuracy of the three quality control samples (QCs) were within 15% over the calibration range (0.5 µg/mL to 100 µg/mL) (Bekegnran et al., submitted). Favipiravir was extracted by a simple protein precipitation method, using acetonitrile for plasma and ice-cold acetonitrile for clarified lung homogenates. Briefly, 50 µL of samples matrix was added to 500 µL of acetonitrile solution containing the internal standard (favipiravir-13C,15N, Alsachim), then vortexed for 2 min followed by centrifugation for 10 min at 4 °C. The supernatant medium was evaporated and the dry residues were then transferred to 96-well plates and 50 µL was injected. To assess the selectivity and specificity of the method and matrix effect, blank plasma and tissues homogenates from 2 control animals (uninfected and untreated) were processed at each run. Moreover, the same control samples spiked with favipiravir concentration equivalent to the QCs (0.75, 50, and 80 µg/mL) were also processed and compared to the QCs samples. Data were collected using the MassLynx Mass Spectrometry Software 4.1.Noncompartemental analysis conducted using software Pkanalix2019R2 (www.lixoft.com). Areas under the plasma concentration time curve were computed using medians of favipiravir concentrations at 0.5, 1, 5, and 8 h, and extrapolated until T = 12 h. Ctrough were extrapolated at T = 12 h using lambda-z loglinear regression on the decreasing slope of concentrations.HistologyAnimal handling, hamster infections, and favipiravir administrations were performed as described above. Lungs were collected after intratracheal instillation of 4% (w/v) formaldehyde solution, fixed 72 h at room temperature with a 4% (w/v) formaldehyde solution and embedded in paraffin. Tissue sections of 3.5 µm, obtained following guidelines from the “global open RENI” (The standard reference for nomenclature and diagnostic criteria in toxicologic pathology; https://www.goreni.org/), were stained with hematoxylin-eosin (H&E) and blindly analyzed by a certified veterinary pathologist. Microscopic examination was done using a Nikon Eclipse E400 microscope. Different anatomic compartments were examined (see Supplementary Table 2): (1) for bronchial and alveolar walls, a score of 0 to 4 was assigned based on the severity of inflammation; (2) regarding alveoli, a score of 0 to 2 was assigned based on presence and severity of hemorrhagic necrosis; (3) regarding vessel lesions (endothelitis/vasculitis), absence or presence was scored 0 or 1 respectively. A cumulative score was then calculated and assigned to a grade of severity (see Supplementary Table 3).Sequence analysis of the full-length genome200 µL of lung clarified homogenate or infectious cell supernatant (virus stock) was inactivated with an equal volume of VXL lysis buffer (Qiagen) and viral RNA was extracted using an EZ1 Advanced XL robot with the EZ1 mini virus 2.0 kit (both from Qiagen) and linear acrylamide (ThermoFisher Scientific) in place of carrier RNA. cDNA was generated in a final volume of 40 µL using 14 µL of nucleic acid extract, random hexamer and the Protoscript II First Strand cDNA Synthesis Kit (New England Biolabs). A specific set of primers (Supplementary Table 4) was used to generate thirteen amplicons covering the entire genome with the Q5 High-Fidelity DNA polymerase (New England Biolabs). PCR mixes (final volume 25 µL) contained 2.5 µL of cDNA, 2 µL of each primer (10 µM), and 12.5 µL of Q5 High-Fidelity 2X Master Mix. Amplification was performed with the following conditions: 30 s at 98 °C, then 45 cycles of 15 s at 98 °C and 5 min at 65 °C. Size of PCR products was verified by gel electrophoresis. For each sample, an equimolar pool of all amplicons was prepared and purified using Monarch PCR & DNA Cleanup Kit (New England Biolabs). After DNA quantification using Qubit dsDNA HS Assay Kit and Qubit 2.0 fluorometer (ThermoFisher Scientific), amplicons were fragmented by sonication into fragments of around 200 bp long. Libraries were built by adding barcodes, for sample identification, and primers using AB Library Builder System (ThermoFisher Scientific). To pool equimolarly the barcoded samples a quantification step by real-time PCR using Ion Library TaqMan Quantitation Kit (ThermoFisher Scientific) was performed. Then, emulsion PCR from pools and loading on 530 chip was performed using the automated Ion Chef instrument (ThermoFisher Scientific). Sequencing was performed using the S5 Ion torrent technology v5.12 (ThermoFisher Scientific) following the manufacturer’s instructions. Consensus sequence was obtained after trimming of reads (reads with quality score < 0.99, and length < 100 pb were removed and the 30 first and 30 last nucleotides were removed from the reads). Mapping of the reads on a reference (determine following blast of De Novo contigs) was done using CLC genomics workbench software v.20 (Qiagen). A de novo contig was also produced to ensure that the consensus sequence was not affected by the reference sequence. Mutation frequency for each position was calculated as the number of reads with a mutation compared to the reference divided by the total number of reads at that site. Only substitutions with a frequency of at least 1% were taken into account for the analysis (Supplementary Data 10).ED50, ED90, and ED99 determinationWe conducted a nonlinear regression of infectious viral load against dose, using an Emax model, giving \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$VL = VL_0 \times \Big( 1 - \left.\left( {\frac{{D^\gamma }}{{D^\gamma + D_{50}^\gamma }}} \right) \right)$$\end{document}VL=VL0×1−DγDγ+D50γ with VL0 being infectious viral load of untreated animals. We estimated D50 the dose required to decrease viral load by 50%, using a coefficient γ to account for the high sigmoidicity of the relation between dose and titers. γ coefficient was chosen as the one maximizing likelihood of the model. We extrapolated the D90 and D99 using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_{90} = \root {\gamma } \of {{9 \times D_{50}^\gamma }}$$\end{document}D90=9×D50γγ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D_{99} = \root {\gamma } \of {{99 \times D_{50}^\gamma }}$$\end{document}D99=99×D50γγ, as well as their 95% confidence interval using the delta method.Graphical representations and statistical analysisGraphical representations and statistical analyses were performed with Graphpad Prism 7 (Graphpad software) except linear/nonlinear regressions and their corresponding graphical representations that were performed using R statistical software (http://www.R-project.org). Statistical details for each experiment are described in the figure legends and in corresponding Supplementary data. When relevant, two-sided statistical tests were always used. P-values lower than 0.05 were considered statistically significant. Experimental timelines were created on biorender.com.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary DataSupplementary Data 1Supplementary Data 2Supplementary Data 3Supplementary Data 4Supplementary Data 5Supplementary Data 6Supplementary Data 7Supplementary Data 8Supplementary Data 9Supplementary Data 10Supplementary Data 11Supplementary Data 12Supplementary Data 13Reporting Summary | nature communications | [
"Article"
] | [
"Therapeutics",
"Drug development",
"Experimental models of disease",
"Infection"
] | March 2020 World Health Organization declared coronavirus disease 2019 (COVID-19) pandemic1 COVID-19 outbreak identified Wuhan China December 2019 spread months severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), causative agent COVID-19 Coronaviridae family related to SARS-CoV emerged China 20022. incubation 5 days disease onset begins influenza-like syndrome high virus replication late acute respiratory distress syndrome high inflammatory proteins occurs one to two weeks3 11 November 2020 90 million cases COVID-19 resulted 1,936,000 deaths5 pandemic repercussions health society ecology economy urgent need for effective infection prevention control host-directed immune-based useful safe effective antiviral molecules important step pandemic conventional drug development slow repurposing drugs explored led clinical trials treatment COVID-196 development effective antiviral drugs rely on pre-clinical in vivo data not rapid implementation rodent non-human primate animal models assess potential safety efficacy drug determine appropriated dose regimensFavipiravir (6-fluoro-3-hydroxypyrazine-2-carboxamine anti‐influenza drug approved Japan broad-spectrum antiviral activity against RNA prodrug metabolized intracellularly into active ribonucleoside 5′-triphosphate form RNA polymerase lethal mutagenesis16 studies reported vitro inhibitory activity against SARS-CoV-2 50% effective concentrations 62 to > 500 μM (10 to > 78 μg/mL 20 clinical trials COVID-19 by favipiravir ongoing present study efficacy favipiravir in vitro Syrian hamster model preventive administration high doses reduction infectious titers histopathological damages clinical alleviation disease genetic diversity viral confirms mutagenic effect vitro efficacy recorded EC50 EC90 of 204 334 μM multiplicity of infection (MOI) of 0.001, 446, > 500 μM MOI of 0.01 Infectious titer reductions ≥ 2 with 125 μM favipiravir between 11 and 342 with 500 μMCaco-2 cells CPE SARS-CoV-2 BavPat1 strain infectious titer reductions 5 125 μM favipiravir 144 7721 500 μM 50% cytotoxic concentrations (CC50) VeroE6 Caco-2 cells > 500 μM.Table 1In vitro efficacy favipiravir.Cell lineMOIDrug titer μM250 μM500 μMVero μM334>500 multiplicity infection dose–response curves antiviral activity mean infectious titers without favipiravir Syrian hamsters SARS hamster model efficacy antiviral intranasally infected 4-week-old female Syrian hamsters 106 TCID50 virus Groups two animals sacrificed 2 3 4 7 days post-infection Viral replication quantified RT-qPCR organs (lungs brain bowel kidney spleen heart plasma Viral loads lungs peaked 2 dpi 4 dpi decreased at 7 dpi plasma 3 dpi replication large bowel 2 dpi No viral RNA other samplesinfected animals with two lower virus inocula (105 104 TCID50). Viral RNA quantified in lungs large bowel plasma 2 3 4 7 dpi Fig. 2 Viral loads lungs peaked at 2 3 dpi with inocula 105 104 TCID50 Maximum viral loads lungs comparable RNA yields plasma large bowel similar trend more variability lower inocula clinical monitoring no symptoms infection normalized weights lower from 3 dpi inoculated with sodium chloride 0.9% Fig 2) vivo efficacy favipiravirTo hamsters received intraperitoneally three times a day used three doses 18.75 37.5 75 mg/day 340 ± 36 ± 42 1390 ± 126 mg/kg/day treatment initiated day infection ended at 2 dpi infected groups 6 animals intranasally with three virus inocula (106 105 104 TCID50 viral replication measured in lungs plasma at 3 dpi (Fig. Each virus inoculum assessed experiment effect favipiravir dose dependenteach virus inoculum infectious titers 75 mg/day TID lower untreated (p ≤ 0.0001) ranged 1.9 and 3.7 log10 infected with 105 or 104 TCID50 titer reductions 0.8 log10 with 37.5 mg/day TID (p ≤ 0.038) Drug 90 99% effective doses estimated ranged 31–42 mg/day 53–70 mg/day (Table 2) virus replication lung homogenates viral RNA yields differences untreated 0.7 and 2.5 log10 observed higher dose favipiravir (p ≤ 0.012) difference noticeable with lower virus inocula reductions estimated relative infectivity of viral particle Decreased infectivity in all treated groups differences significant with higher dose favipiravir (p ≤ 0.031) 37.5 mg/day TID infected with 105 or 104 TCID50 (p ≤ 0.041) measured plasma viral loads RT-PCR higher dose favipiravir 106 or 104 TCID50 reductions of 2.1 and 2.62 log10 (p ≤ 0.022)signs observed animal treated 75 mg/day weights lower untreated animals (Fig. 1Virological results preemptive favipiravir therapy Experimental timeline 6 hamsters intranasally infected 106 105 104 TCID50 virus Viral replication lung infectious titers TCID50 assay TCID50/copy ɣ-actine gene 6 animals replication RNA yields RT-qPCR assay copies 6 Relative lung viral particle infectivities calculated ratio lung infectious titer viral RNA yields Plasma viral loads RT-qPCR assay viral genome copies/mL plasma 6 animals Clinical course disease 6 animals Normalized weight day n calculated % initial weight Data mean ± SD Supplementary Data 2) Two statistical analysis Shapiro–Wilk normality test Student t-test Mann–Whitney test Welch’s test two-way ANOVA Post-hoc Dunnett’s multiple comparisons test Supplementary Data 3 4) **** * symbols average value group lower untreated group p-value < 0.0001 0.0001–0.001 0.01–0.05 Source dataTable 2Drug doses viral titers inoculum.Virus inoculumED50mg/day)ED90mg/day/day)Preemptive therapy104 TCID5034 (30–37)42 (38–46)53 (48–58)105 TCID5026 (21–30)37 (31–44)56 (46–65)106 TCID5015 (10–20)31 (21–41)70 (48–93)Preventive therapy104 TCID5027 (25–29)35 (32–38)47 (44–51)Dose–response curves Supplementary Fig. confidence interval second experiments assessed 7 days impact preemptive therapy clinical course disease weight criterion (Fig. 2a). toxicity evaluated toxicity three doses favipiravir four non-infected animals. Important toxicity 75 mg/day TID normalized weights lower untreated moderate toxicity 37.5 mg/day TID significant day 4, 5 6 No toxicity lower dose favipiravir infection compared weights infected non-infected animals 75 mg/day no significant difference 1 2 3 dpi Fig. 4) intranasally infected groups 10 animals two virus inocula (105 or 104 TCID50). Each virus inoculum assessed independent experimentTreatment 37.5 mg/day TID initiated day infection ended 3 dpi (Fig. 2a). associated clinical alleviation (Fig. 2c inoculum 105 TCID50 mean weights treated animals higher untreated 5 6 dpi (p ≤ 0.031) Similar inoculum 104 TCID50 5 6 7 dpi (p < 0.0001).Fig. 2Clinical follow-up Experimental timeline three doses favipiravir four uninfected animals without infection Clinical follow-up 10 animals infected 105 104 TCID50 treated favipiravir 37.5 mg/day Normalized weight day n calculated % initial weight Data mean ± SD Supplementary Data 2) Two-sided statistical analysis two-way ANOVA Post-hoc Dunnett’s Sidak’s test Supplementary Data 5) **** *** ** * symbols average value group lower untreated group p-value < 0.0001 0.0001–0.001 0.01–0.05 Source data third treatment started 1 day before infection ended 2 dpi intranasally infected groups 6 animals 104 TCID50 viral replication measured lungs plasma at 3 dpi (Fig.inverse relationship between lung infectious titers dose favipiravir (Fig. infectious titers 37.5 75 mg/day TID lower untreated (p ≤ 0.002) undetectable infectious titers higher dose Estimated ED90 ED99 35 47 mg/day reductions viral RNA yields 0.9 3.3 log10 37.5 75 mg/day TID (p ≤ 0.023) (Fig. infectivity viral decreased reduction higher dose favipiravir (p = 0.005) (Fig. reduced plasma viral loads 37.5 75 mg/day TID (p ≤ 0.005) (Fig. signs of observed 75 mg/day TID normalized weights lower untreated (Fig. 3d).Fig. 3Virological results preventive favipiravir therapy Experimental timeline 6 hamsters intranasally infected with 104 TCID50 virus Viral replication infectious titers TCID50 assay/copy ɣ-actine gene 6 Viral replication RNA yields RT-qPCR assay Clinical course disease 6 Normalized weight at day n calculated % initial weightlung virus infectivities calculated ratio lung infectious titer over viral RNA yields = 6 animals Plasma viral loads RT-qPCR assay expressed in viral genome copies/mL dotted line detection threshold (n = 6 animals/group). Data represent mean ± SD Supplementary Data 2) Statistical analysis Shapiro–Wilk normality test Student t-test Mann–Whitney test One-sample t-test two-way ANOVA Post Dunnett’s multiple comparisons test Supplementary Data 3 4) **** ** * symbols average value group different from untreated group p-value < 0.0001 0.001–0.01 0.01–0.05 Source data impact favipiravir treatment lung pathological changes SARS-CoV-2 Animals intranasally infected 104 TCID50 virus Treatment two doses favipiravir (37.5 75 mg/day initiated before ended at 3 dpi four animals sacrificed at 3 5 dpi (Fig. 4a control four vehicle-treated groups severity inflammation alveolar hemorrhagic necrosis vessel lesions cumulative score 0 to 10 calculated severity Supplementary Data 7)lungs untreated animals displayed lesions air-borne infection broncho-interstitial progression between 3 5 dpi virus dissemination At 3 dpi 7/8 animals mild pulmonary changes difficulty efficacy treatment scores lower 5 dpi severe pulmonary impairments dose-dependent effect favipiravir preemptive antiviral strategy animals 37.5 mg/day TID marked histopathological damages 75 mg/day TID mild or moderate damages preventive antiviral strategy 37.5 mg/day TID mild marked damages 75 mg/day TID no or mild damages At 5 dpi significant cumulative score reductions observed both doses favipiravir (p = 0.0286. 4Lung histopathological changes with preemptive preventive favipiravir four animals intranasally infected with 104 TCID50 virus sacrificed at 3 5 dpi Experimental timelines preemptive favipiravir therapies day sacrifice lungs collected fixed embedded in paraffinTissue sections stained with hematoxylin-eosin severity inflammation alveolar hemorrhagic necrosis vessel lesions cumulative score 0 to 10 calculated assigned to grade severity (I II III Scoring pathological changes for preemptive preventive favipiravir therapies (n = 4 animals/group Supplementary Data 7) Two statistical analysis using Shapiro–Wilk normality test Student t-test Mann–Whitney test two-way ANOVA Post-hoc Dunnett’s multiple comparisons test Supplementary Data 7 8). average value group different from untreated group p-value 0.01 and 0.05. images lung tissue multifocal extensive inflammation untreated limited 37.5 mg treated normal 75 mg/day 4 samples/group). bronchial inflammation severe filled with neutrophilic exudates mild 37.5 mg normal 75 mg/day 4 samples/group). alveolar inflammation severe infiltration filled with neutrophils moderate some 37.5 mg normal 75 mgimages vessel inflammation infiltration vascular wall neutrophils debris endothelial damage untreated moderate endothelial leukocytic accumulation 37.5 mg normal vessel 75 mg/day 4 samples Clinical courses Supplementary Fig. 6. Source data.Favipiravir hamster assessed PK lung distribution favipiravir uninfected animals treated single dose favipiravir 6.25 mg 12.5 mg 25 mg sacrificed three animals post-treatment 1 5 8 h favipiravir plasma concentration lung tissue 0.5 5 h post-treatment assessed favipiravir concentration multiple dose animals infected 105 TCID50 virus nine animals received three doses 18.75 37.5 mg 75 mg/day sacrificed 12-h after last dose Favipiravir trough concentrations quantified plasma (n = 9) lung tissue (n = 3).Results Table 3 Supplementary Fig 7. dose maximum concentration 0.5 h concentrations decreased below 10 μg/ml at 12 h non-linear increase between doses multiple doses trough concentrations non-linear increase extrapolated 12 h post-treatment concentrations dose calculated accumulation ratioAccumulation ratios 6 16 21 three doses non-proportional increase average concentration single dose 0–12-h values 10.1 38.7 μg 100.5 μg/mL three favipiravir doses 3Plasma lung concentrations-treatmentSingle doseMultiple/mL mg372 47.5216 390.58 ± 0,0475 mg/day TID1 h279 49.95.081.3 240.62 0.108 h5.77 1.3412 9.8316.0 4.870.44 0,070.5 h12.5 52.090.7 12.70.58 0.1437.5 mg/day TID1 h155 h10.7.84 1.490.37 0.0528 1.221.36 0.140.35 0,030.5 4.1150.2 16.40.58 0.1718.75 mg/day TID1 h35.2 27.85 0.251.09 0.050.38 0.058 h0.56 0.1612 detectedNANA applicableData mean ± SD three animals each condition multiple dose = 9 plasma 3 Supplementary Data after 3 days favipiravir three times dosing interval lung concentrations 1.6–2.7-fold lower plasma single multiple doses single dose mean lung to plasma ratio 0.37 to 0.62 similar between three doses 0.5 h high ratio 5 h post highest dose (25 mg increase 1.6–1.8 lower doses multiple doses lung penetration mean lung to plasma ratio 0.35 to 0.44 not detected lungs lowest dose (18.75 mg/day).Mutagenic effect genomic modifications genome sequencing lung homogenates animals infected 106 TCID50 virus treated two highest doses Data next-generation sequencing lung samples four animals group 37.5 mg/day 75 mg/day mean sequencing coverage 10,991 to 37,991 reads per genomic position subjected substitutions frequency ≥ 1% analysisgenetic variability in virus stock analyzed 14 nucleotide polymorphisms detected 5 mutation frequency higher than 10% (Supplementary Data mutagenic effect favipiravir used consensus sequence virus stock mutations not considered (1–4 mutations per sample no majority mutations > almost lower than 10% (Fig. 5a). mutations distributed throughout genome (Fig. 5b).Fig. 5Mutagenic effect of favipiravir Viral genetic diversity in lung homogenates four samples analyzed triangle represents mutation substitutions frequency ≥ 1% mutation distribution viral genome variable nucleotide position counted once variability represented using 75 nt sliding windows nucleotide positions 300 nt sliding window Mean number of mutations 4 Mutation characteristics frequency calculated divided by total mean ± SD Supplementary Data 10 13). Two-sided statistical analysis using Shapiro–Wilk normality test Student t-test Mann–Whitney test Welch’s test Supplementary Data 11 12). *** ** * symbols average value group lower than untreated group p-value between 0.0001–0.001 0.001–0.01 0.01–0.05Association between lung infectious titers TCID50 assay frequency non synonymous synonymous G → A mutations data animal analysis univariate linear regression error band 95% confidence interval Source data relationship mutations sample dose favipiravir (Fig. 5c): mean mutations increased factor 2 4.8 with treated 37.5 75 mg/day TID difference significant 37.5 mg/day TID (p = 0.029) increase G → A substitutions C → U substitutions regardless dose favipiravir mean frequency G → A substitutions increased factor 4.2 (p ≤ 0.009) increased frequency 37.5 mg/day TID p = 0.037) non-synonymous mutations 75 mg/day TID p = 0.009) (Fig. 5d). investigated effectiveness animals characteristics mutations viral infectious titers negatively associated with non-synonymous G → A mutations positively synonymous mutations (p < 0.03 Fig. experiments parallel evolution events 32 substitutions in viral sub-populations detected in two independent animals18 shared mutations detected with treated animals 14 non-synonymous (Supplementary Data 13). mutations located in nsp2, 3 4 5 6 14 N protein Matrix ORF 3a 8. if substitutions reflect to hamster model or antiviral selection current study used hamster model efficacy favipiravir against SARS-CoV-2 infection viral RNA detected in lungs blood large bowel Peak viral replication at 2–3 dpi 6–10-weeks-old symptom lack of weight gain recovery at 7 dpi Histopathological changes comparable to animals with pulmonary impairments displayed vascular lesions in vivo model younger animals (4 weeks-old), suitable for preclinical evaluation antiviral compounds against SARS-CoV-2 doses favipiravir 700–1400 mg/kg/day reduced viral replication allowed clinical alleviation (Figs. 1 2) high doses favipiravir led to undetectable viral replication in lung plasma high doses favipiravir expand in vivo spectrum against RNA virusesviral replication greater estimated infectious titers than total viral RNA non-human primates Remdesivir hamsters favipiravir23 pulmonary changes favipiravir severity lesions effective doses favipiravir higher rodent models (≈100–400 mg/kg/day correlated with high favipiravir EC50 SARS-CoV-2 effective doses hamster model toxicity adverse effects favipiravir not exacerbated SARS-CoV-2 infection similar weights measured among infected non-infected animals highest dose favipiravir 1 2 3 dpi reduction viral replication correlated with dose favipiravir inversely virus inoculum study efficacy favipiravir intraperitoneally orally daily 900 1200 mg/kg/day 600 1000 mg/kg/day assessed hamster model (6–10 weeks high virus inocula highest dose favipiravir moderate decrease viral RNA yields lung tissue lowest dose smaller inhibitory effect infectious titers reduction dose-dependent Both doses associated regression pulmonary histopathological impairmentsresults accordance with at medium high doses favipiravir 670 1390 mg/kg/day study no signs with favipiravir treatment discrepancy could due to difference between highest daily doses (1000 mg/kg/day 1390 mg/kg dosing regimens age hamsters at infection (6–10 weeks old 4 weeks influenza viruses favipiravir acts as nucleotide analog purine nucleotide by viral metabolized intracellularly to active form incorporated into nascent viral RNA strands inhibits RNA strand extension induces mutation accumulation viral genome16 favipiravir similar action with SARS-CoV-2 chain termination reduced RNA synthesis lethal mutagenesis20 genomic analysis confirmed mutagenic effect favipiravir in favipiravir treatment induced G → A mutations into viral genomes viral infectivity Similar findings in vitro vivo with other RNA viruses9 strong inverse association between infectious titers in lungs non-synonymous mutations in viral populations random non-synonymous mutations more deleterious than synonymous suggests randomly distributed no compensatory mechanism triggered by virus eliminateinverse correlation between lung infections G → A substitutions increased mutations beyond error threshold cause lethal mutagenesis.Genomic analyses 18 mutations in viral sub-populations shared with treated animals Two in nsp14 coding region proof-reading viral RNA polymerisation32 in N7 MTase domain viral RNA capping34 resistance mutations against Remdesivir in β-coronavirus hepatitis virus in RdRP (nsp12) coding sequence36 investigations needed impact on antiviral effect favipiravir.Favipiravir PK hamster model non-linear increase in plasma exposure between doses favipiravir concentration time profiles agreement with results PK study 7–8-week-old hamsters single dose 100 mg/kg maximum plasma drug concentration 0.5 h after oral administration decreased rapidly short half After repeated doses plasma exposure confirmed non-linear PK over doses accumulation ratios accumulation at highest dose in hamsters Favipiravir hepatic metabolism inactive M1 metabolite inhibits aldehyde oxidase activity concentration- time-dependent self-inhibition metabolism highest dose to increasegood penetration favipiravir lungs observed lung/plasma ratios 35 to 44% after repeated doses consistent physicochemical properties Lung exposure accordance previous medium dose favipiravir (670 mg/kg/day within estimated doses reduce 90% infectious titers lungs 31 42 mg/day 570–780 mg/kg/day limited drug-associated Animals infected with 105 104 TCID50 virus treated preemptive reduction infectious titers histopathological scores clinical alleviation (Figs. 1 2 4) treated preventive strategy reduction viral replication histopathological scores 3 accumulation ratio after repeated doses good penetration effective concentrations expected lungs treatment mg/kg clinically realistic results compared drug concentrations hamster model with patients 2016, clinical trial favipiravir Ebola-infected dose 6000 mg day 0 1200 mg 9 days median trough concentrations favipiravir day 2 day 4 46.1 25.9 μg/mL within range hamsters highest dose 1400 mg/kg mean trough concentration 29.9 μg/mL additional investigations required favipiravir plasma exposure in SARS-COV-2 infected patients antiviral activitydifferences in PK between hamster humans toxicity at highest doses limits extrapolation safe dosing regimens humans plasma exposure effect viral replication unknown intracellular concentration active metabolite determined parameter drug drives antiviral effect study high doses favipiravir antiviral activity against SARS-CoV-2 in hamster model better antiviral efficacy preventive strategy favipiravir for prophylactic use high doses required determine tolerable dosing regimen similar exposure in non-human primates before testing high dose regimen in COVID-19 patients studies determine increased antiviral efficacy favipiravir other antiviral drugs dosing regimen need for rapid development animal models confirm efficacy antiviral determine appropriate dose regimens humans before clinical trials.MethodsCellsVeroE6 cells Caco-2 cells grown at 37 °C 5% CO2 medium 7.5% heat-inactivated fetal bovine serum 1% penicillin/streptomycin 1% non-essential amino acids from ThermoFisher experiments virus in biosafety level (BSL) 3 laboratory SARS-CoV-2 strain BavPat1 European Virus Archive GLOBALeuropean-virus-archive Christian Drosten stocks 0.001 25 cm 2 flask VeroE6 cells MEM medium 2.5% FBS cell supernatant medium replaced 24 h harvested infection supplemented 25 mM HEPES stored −80 °C vitro determination EC50 EC90 CC50 infectious titer 5 104 VeroE6 cells seeded 96-well culture plates 100 μL 2.5% FBS medium seven 2-fold dilutions favipiravir-Chemical 0.61 μg 78.5 μg added (25 μL/well Eight virus control wells supplemented 25 μL eight cell controls 50 μL 15 min 25 μL virus suspension added wells MOI 0.001 Three days after infection cell supernatant media collected TCID50 assay 500 250 125 infectious titer reductions cell viability assessed CellTiter-Blue reagentFluorescence (560/590 nm recorded Tecan Infinite 200Pro machine 50 90% concentrations (EC50 EC90) determined logarithmic interpolation% inhibition calculated (ODsample ODvirus control control CC50 50% same culture conditions EC50 virus cell viability measured CellTiter Blue CC50 determined interpolation vivo experimentsApproval approved local ethical committee French ‘Ministère l’Enseignement Supérieur Recherche l’Innovation’ French guidelines European legislation experiments BSL 3 laboratory handlingThree-week-old female Syrian hamsters provided Janvier Labs ISOcage P - Bioexclusion System unlimited access water/food 14 h/10 h light/dark cycle weighed monitored daily signs illness Virus inoculation general anesthesia Organs blood collected after euthanasia InfectionAnesthetized animals intranasally infected 50 μL 106 105 104 TCID50 virus 0.9% sodium chloride solution mock group intranasally inoculated 50 μL 0.9% sodium chloride solution.Favipiravir intraperitoneally inoculated different doses favipiravirControl group inoculated 0.9% sodium chloride collectionOrgans washed 10 mL 0.9% sodium chloride transferred 2 50 mL tube 1 mL bowel kidney spleen heart 10 mL (lungs brain 0.9% sodium chloride solution 3 mm glass beads crushed Tissue Lyser machine 5 min 30 cycles/s centrifuged 5 min 16,200 × g Supernatant media transferred 2 mL tube centrifuged 10 min 16,200 stored −80 °C milliliter blood harvested 2 mL tube 100 μL 0.5 M EDTA centrifuged 10 min 16,200 stored −80 °C real-time RT-PCR experiments molecular biology laboratory RNA extraction QIAamp 96 DNA kit Qiacube HT kit 100 μl organ homogenates 10 μL internal control (bacteriophage MS2)42 transferred S-block VXL proteinase K RNA carrier RT-qPCR-CoV-2 MS2 performed Express one step RT-qPCR Universal kit 3.5 μL RNA 6.5 μL RT-qPCR mix 250 nmol primer 75 nmol probeAmplification QuantStudio 12K Flex Real-Time PCR System conditions 50 °C 10 min 95 °C 20 s 40 cycles 95 °C 3 s 60 °C 30 s qPCR (ɣ-actine gene detection-qPCR modifications Express one step qPCR Universal kit 50 °C step removed Data collected QuantStudio 12K Flex Software v1.2.3. Primers SARS-CoV-2 MS2 ɣ-actine Supplementary Table 1.Tissue-culture infectious dose 50) 96-well culture plates VeroE6 cells inoculated 150 μL per well dilutions Plates incubated 4 days read cytopathic effect Infectious titers estimated Reed & Muench43.Favipiravir handling hamster infections favipiravir administrations left lung washed 10 mL sodium chloride 0.9% solution blotted filter paper weighed transferred 2 mL tube 3 mm glass beads crushed Tissue Lyser machine 10 min 30 cycles/s centrifuged 5 min 16,200 × g Supernatant media transferred 2 mL tubes centrifuged 10 min 16,200 stored −80 °Cmilliliter blood harvested 2 mL tube 100 μL 0.5 M EDTA (ThermoFischer centrifuged 10 min 16,200 g stored −80 °C favipiravir plasma lung tissues high-performance liquid chromatography tandem mass spectrometry limit 0.1 μg/mL Precision accuracy three samples 15% calibration range (0.5 μg/mL to 100 μg/mL Favipiravir extracted protein precipitation acetonitrile plasma-cold acetonitrile lung homogenates 50 μL matrix added 500 μL acetonitrile solution (favipiravir-13C,15N vortexed 2 min centrifugation 10 min 4 °C medium evaporated residues transferred 96-well plates 50 μL injected selectivity plasma homogenates 2 control animals processed control samples favipiravir (0.75 50 80 μg/mL compared Data collected MassLynx Mass Spectrometry Software.Noncompartemental analysis Pkanalix2019R2 plasma concentration time curve computed medians favipiravir concentrations 0.5 1 5 8 h extrapolated until T = 12 hCtrough extrapolated T = 12 h lambda-z loglinear regression decreasing concentrations.HistologyAnimal handling hamster infections favipiravir administrations performed Lungs collected after intratracheal instillation 4% formaldehyde solution fixed 72 h room temperature embedded in paraffin Tissue sections 3.5 μm RENI” stained hematoxylin-eosin) blindly analyzed by veterinary pathologist Microscopic examination Nikon Eclipse E400 microscope anatomic compartments examined bronchial alveolar walls score 0 to 4 severity inflammation alveoli score 0 to 2 hemorrhagic necrosis vessel lesions absence presence scored 0 or 1 cumulative score calculated grade severity analysis full-length genome200 μL lung clarified homogenate infectious cell supernatant inactivated VXL lysis buffer viral RNA extracted EZ1 Advanced XL robot EZ1 mini virus 2.0 kit linear acrylamide carrier RNAcDNA generated 40 μL 14 μL nucleic acid extract random hexamer Protoscript II First Strand cDNA Synthesis Kit (New England Biolabs). primers Table 4) amplicons Q5 High-Fidelity DNA polymerase PCR mixes 25 μL 2.5 μL cDNA 2 μL primer (10 12.5 μL Q5 High 2X Master Mix Amplification 30 s 98 °C 45 cycles 15 s 98 °C 5 min 65 °C PCR verified gel electrophoresis equimolar pool amplicons purified Monarch PCR DNA Cleanup Kit DNA quantification Qubit dsDNA HS Assay Kit 2.0 fluorometer amplicons fragmented sonication fragments 200 bp long Libraries built barcodes primers AB Library Builder System quantification real-time PCR Ion Library TaqMan Quantitation Kit emulsion PCR loading 530 chip Ion Chef instrument Sequencing S5 Ion torrent technology v5.12 Consensus sequence trimming reads quality score < 0.99 length < 100 pb removed 30 first 30 last nucleotides Mapping reads CLC genomics workbench software v.(Qiagen). de novo contig produced consensus sequence not affected by reference sequence Mutation frequency each position calculated as reads mutation compared reference divided by total reads site substitutions with frequency 1% analysis (Supplementary Data 10).ED50 ED90 ED99 conducted nonlinear regression of infectious viral load against dose Emax model\documentclass[12pt{amsmath-69pt}{document} = VL_0 1^\gamma + D\gamma{document}VL=VL0×1−DγDγ+D50γ VL0 infectious viral load of untreated animals estimated D50 dose to decrease viral load by 50% coefficient γ high sigmoidicity relation dose titers γ coefficient chosen maximizing likelihood modelextrapolated D90 D99 using\documentclass[12pt{minimal{amsmath{wasysym\oddsidemargin-69pt}$D_{90} =\root {\gamma {{9 \times D_{50}^\gamma}D90=9×D50γγ[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin{-69pt}$D_{99} =\root {\gamma } {{99 \times D_{50}^\gamma}D99=99×D50γγ 95% confidence interval using delta method.Graphical representations statistical analyses performed with Graphpad Prism 7 except linear/nonlinear regressions representations R statistical software-project Statistical details experiment in figure legends Supplementary data two-sided statistical tests used P-values lower than 0.05 statistically significant Experimental timelines created on biorender.com.Reporting summaryFurther information research design Nature Research Reporting SummarySupplementary informationSupplementary InformationPeer Review FileDescription Additional Supplementary DataSupplementary Data Data Data 9Supplementary Data 10Supplementary Data Data 13Reporting Summary | 48.5 | 0.948323 |
10.1038/s41467-020-19676-y | PMC7688929 | High-performance and low-cost indicators are important in food and cosmetics industry but market uptake is low due to several challenges such as toxicity, cost and unclear reading. Here, the authors report on optically-programmed, non-colorimetric indicators based on nanotextured organic non-wovens, encoded by controlling their cross-linking degree. | Spoiled perishable products, such as food and drugs exposed to inappropriate temperature, cause million illnesses every year. Risks range from intoxication due to pathogen-contaminated edibles, to suboptimal potency of temperature-sensitive vaccines. High-performance and low-cost indicators are needed, based on conformable materials whose properties change continuously and irreversibly depending on the experienced time-temperature profile. However, these systems can be limited by unclear reading, especially for colour-blind people, and are often difficult to be encoded with a tailored response to detect excess temperature over varying temporal profiles. Here we report on optically-programmed, non-colorimetric indicators based on nano-textured non-wovens encoded by their cross-linking degree. This combination allows a desired time-temperature response to be achieved, to address different perishable products. The devices operate by visual contrast with ambient light, which is explained by backscattering calculations for the complex fibrous material. Optical nanomaterials with photo-encoded thermal properties might establish new design rules for intelligent labels. | IntroductionPerishable goods, such as foods, cosmetics, and drugs, spoil over time spontaneously, and at an accelerated pace upon increasing temperature; as such, their appropriate handling and storage are critically important for public safety1,2. A crucial piece of information for citizens is given by the expiration date, or date of minimum durability of a product, defined as the interval of time within which the properties, quality, and specifications of a good under proper storage conditions remains effective and safe for human consumption. Also, the suitable storage temperature must be present on the package leaflet and on the item labels as required, for instance, by the European Union for medicinal products3. However, stability tests allowing such data to be determined are based on the use of controlled temperatures, which is generally different from the real life, during which undesired temperature variations can occur by an enormous variety of issues during transportation, distribution, and storage. The consequences cause several million illnesses every year, including intoxication from improperly stored food in which high temperature induced enhanced growth rates of microbial pathogens (e.g., verocytotoxigenic Escherichia coli, Listeria, Salmonella, and Campylobacter)4, scombroid poisoning from histamine-contaminated fish5, and diseases related to suboptimal potency of deteriorated vaccines6. While the control over the supply cold chain is strictly regulated for the manufacturers, wholesalers, distributors, and transporters at each level of good distribution, even inappropriate handling by consumers and domestic storage conditions must be taken into account as potential sources of risk.Time–temperature indicators (TTIs)7–9 are devices fabricated, for instance, in the form of labels or stickers that can largely mitigate these risks for the final user. They should be small and flexible to comply with items of any size and shape2, and preferably rely on a solid-state technology to avoid any possible leakage of fluid reactants. Electronics embedding radio-frequency identification tags and wireless antennas, whose resonant response evolves while they are conformed to spoiling food surfaces10, are elegant data loggers capable of transmitting quality information, but intelligent indicators providing a clear and user-friendly visual information of the overall time–temperature history of an item have the potential to induce a much better perception in consumers7, thus greatly improving general safety without affecting electronic waste. To this aim, an indicator must be based on a material showing a continuous and irreversible change of a physico-chemical property depending on the time–temperature profile undergone by the product with which it is coupled. This effect has been reported for a variety of colorimetric systems, including organic semiconductors9, plasmonic metal nanocrystals8,11, nanocomposites of chitosan and gold nanoparticles12, photochromic inks with temperature-dependent discoloration13, enzymatic indicators14, and microencapsulated microorganisms15. However, most of the TTIs have barely superficially penetrated the market. This is due to a number of reasons, such as potential unclear reading7, concern for the eventual diffusion of toxic reactants from the device, and high cost of the technology behind, which might account for 50–100% of the whole packaging cost16 and lead to increased product price. Also, colorimetric indicators can be unsuitable for color-blind people. Furthermore, indicator materials are generally highly difficult or impossible to be programmed, i.e. to be encoded with a tailored response of their components in order to detect excess temperature over varying temporal profiles, ranging from a few minutes up to several days to address the need of many different classes of perishable products (food, drugs, vaccines, etc.). While efforts towards programming have been focused on finely controlling the colloidal11 or polymer17 chemistry of involved components, the possibility of embedding response instructions by optical control might open a much more practical perspective for the extensive use of smart TTIs.Here we report on photo-programmed, non-colorimetric indicators based on nano-textured organic non-wovens. A two-level information is incorporated in the devices by controlling the degree of cross-linking in organic fibers by initial system encoding through calibrated photo-exposure doses (which allows a desired time–temperature profile response to be achieved) and embedding readable patterns that emerge along with temperature exposure in a continuous way (which enables the device to operate by visual contrast with ambient light, in a highly user-friendly way).ResultsElectrospinning and photo-programming of organic non-wovensThe photo-programming steps of the indicators are illustrated in the scheme of Fig. 1a. Uniform and smooth fibers are electrospun from SU-8 3025, which is a negative epoxy resist largely used in optical and electron-beam lithography, by a proper choice of the spinning parameters (Figs. S1 and S2 in Supplementary Information). Compared to other organic compounds, SU-8 exhibits a high refractive index (∼1.59 in the visible spectral range), which has motivated its past use in waveguides incorporating quantum dots18. In the framework of the present work, this property is relevant to provide surfaces coated by SU-8 non-wovens with remarkable diffuse reflectance as better explained below. As-spun organic fibers are not stable over time, with melting observed within 30 min even at room temperature as shown in Figs. S3 and S4. Therefore, in our process the non-woven is initially stabilized by a first photo-programming step, which is performed soon after spinning to promote SU-8 crosslinking by a calibrated UV exposure dose (<100 mJ/cm2) on all of the surface (top-right part of Fig. 1a). Then, a second photo-programming step is carried out that is spatially selective, i.e., a further UV exposure is performed on a part of the fabric through an optical mask (bottom-right part of Fig. 1a). The two consecutive illuminations are complementary in defining the properties and subsequent evolution of the programmed indicator, since the first one encodes the overall time–temperature behavior of the system through its thermal response, whereas the second provides further dose to design a pre-determined, non-colorimetric ultimate visual output.Fig. 1Photo-programming of thermal properties.a Schematic representation of the indicator photo-programming. SU-8 fibrous non-woven deposition and photo-programming (blue vertical rays) at calibrated optical dose and pattern activation. A warning sign will appear when the subsequently experienced time–temperature profile overcomes targeted programmed values (i.e., correct item storage conditions) by diffuse reflectance contrast (green rays). b Calculated average values of the cross-linking degree, Γ, at different exposure doses during photo-programming (left vertical scale). Data are averaged over at least three samples for each dose (error bars: SD). DSC inflection points (Tg, right vertical scale) for the non-woven fabrics following the first step of photo-programming at different UV doses are also displayed. The blue line is a linear fit to the experimental data (diamonds). Tg0 and Tg80 refer to samples photo-programmed (step 1) with UV doses: 0 and 80 mJ/cm2, respectively. c Optical transmission change (dots, ΔTr) at λ = 510 nm, vs. temperature, for a fabric with first step of photo-programming at 80 mJ/cm2. The dashed line is a guide for the eye. Inset: Normalized Trt0 (black line, left vertical scale) and Rt0 (blue line, right vertical scale) spectra. Trt0 and Rt0 are the optical transmission and the diffuse reflectance measured at t0 = 0, respectively. Rt0 was measured by means of an integrating sphere (see Methods). d Temporal variation of ΔTr (at λ = 510 nm, left vertical scale) for samples kept at Tamb (~ 20 °C, pink dots), 25 °C (green dots), 30 °C (red dots) and 35 °C (black dots). The corresponding optical reflectance change, ΔR (at 510 nm, empty dots, right scale), vs. time is also shown for samples at 35 °C. Dashed lines are guides for the eye.In photocurable thermoset resins complete solidification induced by light-activated cationic polymerization normally occurs during post-exposure baking well above the glass transition temperature, Tg (∼50 °C in SU-8 uncured films). Very few reactions are expected to take place below the glass transition temperature due to the largely limited molecular motions19. Instead, at variance with the common lithographic processes, no post-exposure baking is performed here to promote resin crosslinking in the very thin organic fibers. The degree of cross-linking of the material might be affected by various mechanisms, including the fast interface dynamics impacting on the cooperative motion of monomers20, local chain restrictions21, and photothermal effects generated by the absorption of light22,23. With the purpose of investigating this behavior in depth, we analyze SU-8 non-woven fabrics before and after exposure by Fourier-transform infrared (FTIR) spectroscopy and differential scanning calorimetry (DSC), evidencing a correlation of the material properties with the dose shown in Fig. 1b. The FTIR spectra (Fig. S5) show peaks corresponding to the C–O stretching modes of the epoxy groups (at 862 and 911 cm−1), whose intensity decreases upon SU-8 cross-linking, and peaks at 1508 and 1607 cm−1, corresponding to C–C stretch modes of the aromatic ring, whose intensities do not vary during the polymerization process24,25. The absorption values of the peaks at 911 and 1607 cm−1 can be used to estimate the degree of polymerization upon UV exposure (Fig. 1b), through the expression, Γ = 1−[(AE/AR)/(AE0/AR0)], where AE (AE0) and AR (AR0) denote the intensities of the peak at 911 cm−1 and of the reference peak at 1607 cm−1, before (AE0, AR0) and after (AE, AR) exposure, respectively26,27. Γ is found to increase up to a maximum around 10% upon increasing the exposure dose in the first photo-programming step (0–80 mJ/cm2), highlighting that a fine control is achievable for the cross-linking degree of the non-woven material and, consequently, for the device time–temperature response. For the as-spun non-woven, the glass transition temperature (Tg0 in Fig. 1b) is 13 °C (Fig. S6), i.e., below room temperature, which is in agreement with the low stability and melting behavior observed in unexposed fibers. The Tg value moves to higher temperatures (Fig. S6), increasing roughly linearly upon UV exposure at low dose, at a rate of about 0.1 °C×cm2/mJ (Fig. 1b). In addition, it is worth noticing that the UV doses used here in the first photo-programming step are well below those usually employed in lithographic processes on the resist compound (150–250 mJ/cm2)28, thus making the material still reactive to UV for the second step. Indeed, Γ is additionally doubled after the second photo-programming step (>200 mJ/cm2), highlighting remarkably increased cross-linking (Fig. 1b).Due to the varied cross-linking degree, we find that each programmed non-woven is very sensitive to a specific working temperature, Tw, in terms of its optical (light-scattering) properties. This aspect is exemplified in Fig. 1c, where we show how the optical transmittance, Tr, measured at an incident light wavelength (λ) of 510 nm for non-wovens exposed by 80 mJ/cm2 (Tw ≅ 38 °C), changes upon heating at different temperatures. Here, ΔTr is given by Tr − Trt0, where Tr is measured upon keeping the samples at each given temperature for 60 min, and Trt0 is the transmittance at t0 = 0, i.e., immediately after photo-programming. Data show an almost stable optical transmittance up to 30 °C (ΔTr ~ 0.6 %) followed by a rapid increase of the system transparency for temperatures in the interval 30–45 °C (ΔTr ~ 80%) and a new plateau condition (ΔTr ~ 90%) above 45 °C. This behavior is related to the rapid increase in the mobility and the formation of a viscous flow, following chain disentanglements in the fibrous material as promoted by heating. An analogous mechanism has been recently exploited with self-healing thermoplastic polyurethanes17. The corresponding transmittance and diffuse (hemispherical) reflectance (R) spectra for the as-spun (t = 0) material are displayed in the inset of Fig. 1c, highlighting a wavelength-independent (white) behavior in the range 400–700 nm, which makes the system optimal for use with ambient light (Fig. S7). Long-term experiments are additionally performed by maintaining the system at constant temperature such as, 20, 25, 30, and 35 °C up to 25 days (Fig. 1d). The material is stable at T ≤ 25 °C, with no significant change found in the transmittance, thus setting an upper limit for prior-use storage temperature of this specific system. Instead, a remarkable response in the long term starts to be found for T ≥ 25 °C, with ample changes corresponding to variations of few degrees, thus highlighting accurate temperature-sensitivity. At 35 °C, the transmittance of incident light for fabrics firstly programmed by 80 mJ/cm2 increases up to ΔTr ~ 30% (black dots in Fig. 1d), while the surface reflectance correspondingly decreases (open dots in Fig. 1d).Pattern activation: morphological and theoretical studiesAn exemplary device displaying a warning sign upon exposure to a temperature of 55 °C is photographed at various time instants in the left images in Fig. 2a–d (an exemplary transition is also shown in the Supplementary Movie), and the corresponding thermal pictures collected by an infrared thermo-camera are shown in the right images. The visual output of the device is clearly based on the contrast in reflectance, which arises upon heating, between regions that were or were not previously exposed during the second photo-programming step, respectively. Exposed regions remain whitish, with high reflectance values, whereas unexposed areas that did not complete photo-induced cross-linking activation gradually melt into an almost continuous film and drastically decrease their reflectance, thus unveiling the substrate used for deposition. As such, the TTI exploits only ambient light with no need for powering or other light sources, and can display any desired shape or pattern depending on the optical mask used in the second photo-programming. Interestingly, the thermal images in Fig. 2b–d highlight that regions in which fibers are still present (i.e., where the negative SU-8 compound underwent the second photo-programming step) maintain a local temperature appreciably lower (by ∼1 °C, Fig. S8) than regions where fibers melt and a higher optical transmission is achieved, due to the cooling efficiency of the organic non-woven, which is likely to promote high heat fluxes29. The best view of the transition undergone by the non-woven fabric upon increasing temperature is captured by scanning electron microscopy (SEM) imaging of the interface across two regions that are either exposed or not exposed, during the second photo-programming step. Optical micrographs of this interface collected at different time points while the TTI is kept at 55 °C are shown in Fig. 2e–h, the corresponding SEM pictures of the fibrous interfaces are in Fig. 2i–l, and views at higher magnification of the unexposed fibers while melting over time are in Fig. 2m–p. As shown in Fig. 2i–l, fibers exposed during the second photo-programming are instead stable over the investigated times, which explains the different optical properties from a microscopic point of view.Fig. 2Pattern activation and changing morphology.a Photographs (left) and thermal images (right) of fibrous samples after the second step of photo-programming. b–d Photographs and thermal images of the sample in a captured at different times during heating at 55 °C. The color scale in thermal images indicates the temperature of the sample. (Scale bars in (a–d): 5 mm). e–h Corresponding optical micrographs of a–d for a feature of the letter “H” in the sign “Hot” (red arrow in (b–d)). w-UV (w/o-UV) indicates the area of the sample exposed (unexposed) to UV light during the second step of photo-programming. Scale bars, 200 μm. i–l Corresponding SEM micrographs at the interface between the UV exposed (w-UV) and the unexposed areas (w/o-UV), collected at the various time points. Scale bars, 20 μm. m–p Higher magnification view of the areas shown in (i–l), that are not exposed to UV light during the second step of photo-programming. Scale bars, 10 μm.To analyze this behavior in depth, we model the light-scattering properties of the fibrous non-woven fabrics by the transition matrix (T-matrix) technique30–32, schematizing each polymer filament as a linear aggregate of particle clusters with spherical subunits of 700 nm diameter (corresponding to the fiber average diameter). To mimic the random morphology of the fibrous surface, we consider three layers of fibers distributed along the thickness direction (z axis in Fig. 3a, b). The model structure is schematized in Fig. 3a, b, where green, red, and blue clusters represent filaments in the first, second, and third layer, respectively. The backscattering intensity map at λ = 510 nm (given by the squared ratio of the backscattered and the incident field, |Es/E0|2) is obtained by considering a plane-wave illumination with unitary amplitude impinging perpendicularly to the structure (i.e., along the z direction), averaging over polarization, and it is calculated at ∼1 μm distance from the first layer of the structure (Fig. 3c). A critically important parameter that yields quantitative information on the amount of the backscattered light is the albedo value, A = Cscat/Cext, defined as the ratio of the scattering, Cscat, and extinction, Cext, light-scattering cross-sections. T-matrix calculations yield an albedo of about A510 = 0.73 at λ = 510 nm, which is one order of magnitude higher than the reflectivity of a continuous film of SU-8 calculated from the Fresnel coefficients at normal incidence (0.06). To assess if this behavior has general validity for ambient light, we determine the backscattering maps at wavelengths corresponding to the RGB standard, 450, 530, and 600 nm (Fig. S9). In Fig. 3d we show the backscattering intensity map obtained as the sum of the single-wavelength components, evidencing the highly intense scattering occurring by ambient illumination that leads to an averaged albedo, ARGB = 0.733. In addition, upon the fiber melting due to increasing temperature, we find that the albedo values dramatically decrease and approach those of SU-8 films. This investigation is based on a three-dimensional model taking into account a fibrous multi-layer, and it fully rationalizes the whitish appearance found for nanofiber non-wovens and the surfaces coated with them, as well as the reflectance contrast arising in our indicators during the operation.Fig. 3Light back-scattering from three-dimensional fiber non-wovens.a Scheme of the structure of nanofibers used for the calculations. To mimic the morphological structure of the fibrous non-woven, an area of 10 × 10 µm2 is considered in the x–y plane, and three layers of fibers are positioned along the axial (z) direction with total thickness 2.1 μm (green nanofibers are placed in the first layer, red ones in the second layer, and blue ones in the third layer). The average fiber diameter (700 nm) equals the value measured by SEM (Fig. S2, Supplementary Information). b Planar view (x–y) of cluster discretization used to calculate the light-scattering properties. Each polymer filament is schematized as a linear cluster with 700 nm spherical subunits. c Backscattering intensity map at λ = 510 nm, calculated as the ratio of the backscattered field (Es) and the incident one (E0). The map is obtained by considering a plane wave illumination with unitary amplitude impinging perpendicularly to the fibers (i.e., along the z axis), averaging over polarization, and it is calculated at about 1 μm distance from the first layer of the structure. d RGB intensity map obtained as the sum of the backscattering maps calculated at wavelengths corresponding to red (λ = 600 nm), green (530 nm,) and blue (450 nm) wavelengths.T-matrix calculations also help in rationalizing the photon transport properties of the non-wovens, which are characterized by two length scales33. The first one is the scattering mean free path, ls, which is the distance of free propagation between scattering events34. This is calculated from the average cluster subunit density (N = 0.51 μm−3 for our clusters) and the spherical subunit scattering cross-section (σscat = 1.6 μm2 at λ = 510 nm), leading to ls = 1/Nσscat = 1.2 μm. The second relevant length scale is associated with the transport mean free path, l*, which is the distance over which the direction of propagation of the photon is randomized. This quantity is calculated by considering the subunit anisotropy parameter, g = <cos θ>, where θ is the scattering angle, and the similarity relation33, l* = ls/(1−g). For our structures we obtain g = 0.65 and a typical transport length of l* = 3.4 μm (see Supplementary Information for details). Thus, visible light impinging on the fibrous material is efficiently extinct within few microns of propagation. In addition, estimating the albedo values for samples with thickness matching experimental values (20–30 μm) shows how the intensity of the backscattered light is efficiently maximized with the used geometry (Fig. S10), thus leading to reflectance contrast in the indicators.Time–temperature indicator devicesPhoto-programming through sequential exposures makes these devices highly versatile in terms of targeted time–temperature profiles, since calibrated doses tailored to the cross-linking degree and hence the structural relaxation of the non-wovens allow the optical response of the indicators to match the expiration intervals of very different perishable products (see Tables S1 and S2 in the Supplementary Information) at a given temperature or through a varied thermal history. Examples to illustrate the ample range of times that need to be monitored include food, for which a few minutes up to hours at unsuitable temperature are enough to induce pathogen growth and toxin formation, drugs used by people traveling across different climatic zones (∼hours), and vaccines such those based on toxoids or polysaccharides that are found to degrade significantly in a few days just above the room temperature (35–45 °C)6. All these cases could be well covered by the thermal response of properly photo-programmed TTIs as shown in Fig. 1b. Strategies to further increase the operational timescale would include enhancing the thermal stability by calibrating the amount of photoacid generator in electrospun blends or tuning the substrate reflectivity properties, or electrospinning nanocomposites with slower relaxation dynamics, namely with reduced local mobility of filler-interacting polymer chains or cross-linked clusters, as demonstrated for other organic matrices35.Exemplary visual sequences of two devices encoded by a first exposure step at 72 and 80 mJ/cm2, respectively, and then kept at 35 °C, are shown in Fig. 4a, b. Here, a “Do not use” sign appears while the reflectance contrast increases in timescales of minutes for the first devices and of hours for the second one. Higher or lower targeted temperature, as well as longer time response, can be achieved by controlling the cross-linking as shown in Fig. 1b. In addition, the devices can be made flexible and water-proof by directly depositing the fibers onto paper and encapsulating in commercial polyethylene terephthalate (PET) sheets (Figure S11), thus being conformable even to curved surfaces (Fig. 4c, inset). Yellow PET sheets can be also used for encapsulation, which allows the TTIs to operate outdoor under direct sun light as shown in Fig. 4c, d.Fig. 4Flexible time–temperature indicator on paper substrate during operation.a, b Photographs of two differently programmed indicators while working at 35 °C with different response timescales: a minutes (device first photo-programming dose = 72 mJ/cm2) and b hours (first photo-programming dose = 80 mJ/cm2), captured at different observation times. (Scale bar: 5 mm). c, d Operation during sunlight-induced warming. Exemplary goods (milk, drugs) are exposed to sunlight for 30 min. The temperature of each item, measured by a thermocouple, is displayed in the red box. The photographs also show the corresponding visual response of non-wovens, encapsulated by yellow plastic sheets to screen UV radiation and prevent undesired residual crosslinking (c, top-right inset, scale bars, 5 mm). In the top-right inset in (c), arrows highlight the flexible and robust multilayer interface at the device edge. Bottom-right insets: magnified view of the indicator on the drug blister, before (c) and after (d) sunlight warming.Furthermore, we point out that even changes in the device reflectance contrast that are too low to be appreciated by eye, could be easily detected by automatized time-lapse analysis of the Weber’s contrast (Cw):1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{\mathrm{w}} = \frac{{I_{\mathrm{b}} - I}}{{I_{\mathrm{b}}}}$$\end{document}Cw=Ib−IIbwhere I is the mean gray intensity of the darker features and Ib is the background gray intensity in the activated pattern, respectively36,37. The typical temporal evolution of Cw exhibits a relatively rapid increase at Tw, then approaching a saturation value (Fig. S12), similar to ΔTr data (black dots in Fig. 1d). A strategy for contrast enhancement is to vary the reflectivity properties of the deposition substrate. To this aim, TTIs were realized by using a glossy paper as substrate (Fig. S13a–d). A maximum value of Cw = 0.5 is found in this way (Fig. S13e, f) that is a factor 2.5 higher than the Cw measured for TTI devices with a similar characteristic timescale but made on opaque paper. Here, the dependence of the Cw on the thickness (h) of the non-woven can be rationalized considering the effective UV dose delivered to the regions that are exposed only during the first photo-programming step (corresponding to the dark regions of the patterns shown in Fig. S13a–d). For h < hM (hM = 30 µm, Fig. S13f), the UV intensity at the glossy paper substrate (IsUV) is higher than for a sample with thickness hM: IsUV(h < hM) > IsUV(h = hM). Due to the diffuse reflectivity of the glossy paper (∼5%), a non-woven with h < hM thus receives an effective higher UV dose with respect to the one with h = hM, leading to relatively higher SU-8 crosslinking and improved capability to retain the filamentary structure at the programmed working temperature. This in turn enhances the intensity of the light that is backscattered from these regions, and relatively decreases the pattern contrast. While the intensity at the glossy paper also decreases for thicker non-woven [IsUV(h = hM) > IsUV(h > hM)], for such samples an additional effect becomes more relevant. Indeed, in this case additional UV exposure of the regions that have undergone the first photo-programming step can occur during the second programming step, because the scattering of UV light through the non-woven can determine a lateral spread of the illuminated features. This effect becomes more relevant upon increasing h, leading to the data curve plot in Fig. S13f. Finally, we carry out long-term stability experiments on the devices to determine the most appropriate storage temperature prior use, finding that freezing conditions are suggested to assure a month-scale shelf-life for these indicators (Figs. S14 and S15). Further studies are currently underway to further extend storage limits.DiscussionCollectively, these results showed that the thermal response of materials for an intelligent indicator can be engineered by tuning the degree of cross-linking and translating to nano-textured non-wovens the working principle of negative resists commonly used in micro- and nano-lithographies. At working temperature, changes take place that alter the material architecture, and consequently, the optical behavior. Therefore, the obtained devices are very cheap, flexible, and light-weight, compatible with paper and common layer encapsulation, and conformable to monitored items with any size and shape. The operation simply consists of visual testing, with no need for electronics or moving/liquid component, and is explained by backscattering intensity from the complex fibrous material, here analyzed though three-dimensional calculations. Outperforming existing solutions in terms of programming capability and ease of use, nano-materials with photo-encoded thermal properties have established the design rule for intelligent labels for the supply chain of perishable products.MethodsPhoto-programmed nanofibersSU-8 3025 (Microchem, product Y3110720500L1GL, lot number 18030250) was diluted with acetone (Sigma) at different relative volumes (acetone/SU-8 volume) in the range 0–16% to form test solutions for electrospinning. Uniform and smooth fibers (Fig. S1) with an average diameter of 700 nm (Fig. S2) were obtained with 9% relative acetone/SU-8 volume and an applied electric field of 0.7 kV/cm, producing electrified jets from a 27-gauge stainless-steel needle. Electrospinning times between 7 and 13 min led to non-wovens with thickness between 20 and 30 μm, measured by profilometry (DektakXT, Bruker). Substrates used to successfully deposit the fibers were silicon wafers, aluminum foils, indium tin oxide-coated glass, and paper. The first step of photo-programming was carried out by UV exposure (λmax ∼ 365 nm) from cylindrical lamps at a distance of 10 cm from the sample, with doses in the range 0.3–80 mJ/cm2. The second, spatially selective step of photo-programming (pattern activation, 250, 305 mJ/cm2) was carried out by a MJB4 (SUSS MicroTec) mask aligner system or by a low-cost bromograph. The doses of UV exposure were measured by using a calibrated power meter (mod. 843-R, Newport) equipped with a UV-visible detector (mod. 818-UV/DB, Newport). More specifically, the detector was positioned in place of the sample and the UV intensity measured as a function of time with a temporal resolution of 0.5 s. The effective UV dose impinging on the samples was then calculated by integrating the measured temporal curve of the UV intensity in a time interval corresponding to the various exposure times used in the photo-programming steps. Packaging of flexible indicators was performed by embedding the non-woven within two plastic foils with a thickness of 75 μm, through a commercial cold laminating machine (Speedy A4, Buffetti).Thermal, optical, and morphological propertiesDSC was performed by a Q200 differential scanning calorimeter from TA Instruments-Waters LLC, equipped with a RCS90 cooling system and a nitrogen gas purge set at a flow rate of 50 mL/min. Standard sample pans in aluminum were used and sealed before measurements. Non-woven samples with mass in the range 2.5–4.3 mg were heated from −60 to 250 °C at 10 °C/min. IR spectroscopy was performed with a FTIR spectrophotometer (Spectrum 100, Perkin-Elmer Inc.) in transmission mode, measuring at least three different samples for each dose. Thermograms were analyzed by TA Universal analysis. The optical transmission of non-wovens was measured by an UV/VIS spectrophotomer Lambda 950 (Pelkin Elmer). The FEG-SEM Merlin from Zeiss at acceleration voltages of 1–18 kV was used to image fibers deposited on silicon substrates. The diffuse reflectance was measured by using a broadband lamp38 as light source and an integration sphere (Labsphere) coupled to a spectrometer through an optical fiber (mod. Flame, Ocean Optics). The sample was positioned in place of the access port of the sphere, which is collinear with the one used for the incident light beam. The incident light beam was directed onto the sample along the direction perpendicular to the sample surface, and only the hemispherically diffused light was detected by the sphere and spectrometer. The data of light backscattered by the sample (along the direction of the incident beam) at a solid angle of 1.1 × 10−3 sr was not collected. As a reference sample, we used a blank port coated with the same high diffusive layer of the internal surface of the integration sphere. An infrared camera (FLIR A655sc with a macroscopic lens) was used to monitor the temperature of the sample during heating at 55 °C for 10 min.ModelingTo calculate the maps of the backscattered intensity and the albedo, A, of the model non-woven, light scattering theory in the T-matrix formalism was exploited30–32 as generalized in the cluster model39,40 (see Supplementary Information). Organic filaments are modeled as a one-dimensional cluster of spherical particles with diameter equal to the average nanofiber transverse size. The structure of the network is constructed in three dimensions to mimic a 10 × 10 µm2 area of the nanofiber mats with a thickness of 2.1 µm. The backscattering intensity maps are calculated as the ratio |Es/E0|2, where Es and E0 are the amplitude of the backscattered and incident fields, respectively, considering a plane wave illumination and averaging over polarization states.Weber’s contrastThe temporal evolution of Cw was evaluated on devices prepared by using the following procedure: (i) fiber deposition on a paper substrate, (ii) first step of photo-programming (72–80 mJ/cm2), (iii) second step of photo-programming (305 mJ/cm2), and (iv) device encapsulation by cold lamination between PET sheets. Images of the TTIs maintained at a constant temperature of 35 °C were measured at several time points using either a smartphone photo-camera or an imaging system based on a CMOS camera (DCC1645C, Thorlabs) coupled with a long working-distance optical system (MVL7000, Thorlabs). For the calculation of Cw the images of the sample were converted to 8-bit and the mean gray intensities of the exposed and unexposed areas were estimated by using the Fiji-ImageJ software (U.S. National Institutes of Health, Bethesda, Fig. S12 and S13).Supplementary informationSupplementary InformationDescription of Additional Supplementary FilesSupplementary Movie 1 | nature communications | [
"Article"
] | [
"Polymers",
"Environmental, health and safety issues",
"Polymers"
] | IntroductionPerishable goods foods cosmetics drugs spoil accelerated increasing temperature appropriate handling storage important for public safety1,2 crucial information for citizens by expiration date minimum durability of product interval time properties quality specifications under proper storage conditions effective safe for human consumption suitable storage temperature must present on package leaflet item labels required by European Union for medicinal products3 stability tests based on controlled temperatures different from real life undesired temperature variations occur during transportation distribution storage consequences cause million illnesses intoxication from improperly stored food growth scombroid poisoning from histamine-contaminated fish5 diseases related to suboptimal potency of deteriorated vaccines6 control over supply cold chain regulated for manufacturers wholesalers distributors transporters inappropriate handling domestic storage conditions as potential sources of risk.Time–temperature indicators (TTIs are devices mitigate risks for user should be small flexible comply with size rely on solid-state technology to avoid leakage of fluid reactantsElectronics embedding radio-frequency identification tags wireless antennas response evolves spoiling food surfaces10 are data loggers transmitting quality information intelligent indicators providing clear visual information time–temperature history item potential induce better perception safety without affecting electronic waste indicator must based on material showing continuous change physico-chemical property depending on time–temperature profile product effect reported for colorimetric systems organic semiconductors9 plasmonic metal nanocrystals8 nanocomposites of chitosan gold nanoparticles12 photochromic inks with temperature-dependent discoloration13 enzymatic indicators14 microencapsulated microorganisms15 TTIs barely penetrated market due to unclear concern diffusion toxic reactants high cost technology for 50–100% packaging increased product price colorimetric indicators unsuitable for color-blind people indicator materials difficult or to be programmed detect excess temperature profiles perishable products efforts programming on controlling colloidal11 polymer17 chemistry of components embedding response instructions by optical control might practical perspective for use smart TTIs report on photo-programmed non-colorimetric indicators based on nano-textured organic non-wovenstwo-level information in devices controlling cross-linking in organic fibers encoding calibrated photo-exposure doses desired time–temperature profile response readable patterns temperature exposure device visual contrast with ambient light user way).ResultsElectrospinning photo-programming of organic non-wovensThe photo-programming steps illustrated in Fig. 1a Uniform fibers electrospun from SU-8 3025 negative epoxy resist used in optical electron-beam lithography choice spinning parameters (Figs. S1 S2 SU-8 high refractive index (∼1.59 motivated use in waveguides quantum dots18 relevant surfaces coated SU-8 non-wovens diffuse reflectance As-spun organic fibers not stable over time melting within 30 min at room temperature in Figs. S3 S4 non-woven stabilized by first photo-programming step after spinning SU-8 crosslinking calibrated UV exposure dose (<100 mJ/cm2) surface second photo-programming step spatially selective further UV exposure on fabric through optical masktwo illuminations properties evolution programmed indicator first encodes time–temperature behavior second provides dose non-colorimetric visual output.Fig. 1Photo-programming thermal properties Schematic representation indicator photo-programming SU-8 fibrous non-woven deposition photo-programming rays calibrated optical dose pattern activation warning sign time–temperature profile overcomes programmed values diffuse reflectance contrast average values cross-linking degree Γ different exposure doses Data averaged over three samples each dose bars DSC inflection points non-woven fabrics first step photo-programming UV doses displayed blue line linear fit experimental data Tg0 Tg80 samples photo-programmed UV doses 0 80 mJ/cm2 Optical transmission change ΔTr) at λ = 510 nm vs temperature fabric first step photo-programming 80 mJ/cm2. dashed line guide eye Normalized Trt0 Rt0 spectra optical transmission diffuse reflectance measured at t0 = 0Rt0 measured integrating sphere Temporal variation ΔTr λ = 510 nm samples at 20 °C 25 °C 30 °C 35 °C (black optical reflectance change ΔR 510 nm time shown samples at 35 °C Dashed lines guides eye photocurable thermoset resins solidification light-activated cationic polymerization occurs post-exposure baking above glass transition temperature Tg (∼50 °C SU-8 uncured few reactions below glass transition temperature limited molecular no post-exposure baking resin crosslinking thin organic fibers cross-linking affected by fast interface dynamics local chain photothermal effects SU-8 non-woven fabrics before after exposure Fourier-transform infrared (FTIR) spectroscopy differential scanning calorimetry correlation material properties dose Fig. 1b FTIR spectra (Fig. S5) show peaks C–O stretching modes epoxy groups 862 911 cm−1) intensity decreases SU-8 cross-linking peaks 1508 1607 cm−1 C–C stretch modes aromatic ring vary during polymerizationabsorption values peaks at 911 1607 cm−1 polymerization UV exposure (Fig. Γ = 1−[(AE/AR)/(AE0/AR0)] AE AR intensities peak 911 1607 cm−1 before after exposure Γ 10% exposure dose first photo-programming step (0–80 mJ/cm2) fine control achievable cross-linking degree non-woven material device time–temperature response as-spun non-woven glass transition temperature (Tg0 Fig. 1b) 13 °C (Fig. S6) below room temperature low stability melting behavior unexposed fibers Tg value moves to higher temperatures linearly UV exposure low dose 0.1 °C×cm2/mJ UV doses first photo-programming step below lithographic processes (150–250 mJ/cm2)28 material reactive UV second step Γ doubled after second photo-programming step (>200 mJ/cm2) increased cross-linking varied cross-linking degree each programmed non-woven sensitive to specific working temperature Tw optical (light-scattering) properties exemplified in Fig.optical transmittance Tr 510 nm for non-wovens exposed 80 mJ/cm2 38 changes heating temperatures ΔTr Tr − Trt0 Tr measured samples temperature 60 min Trt0 transmittance at t0 = 0 after photo-programming stable transmittance up to 30 °C (ΔTr ~ %) rapid increase system transparency 30–45 °C (ΔTr ~ 80%) plateau (ΔTr ~ 90%) above 45 °C related to mobility viscous flow disentanglements fibrous material heating analogous mechanism with self-healing thermoplastic polyurethanes17 transmittance reflectance spectra as-spun (t = 0) material displayed in Fig. 1c wavelength-independent behavior 400–700 nm system optimal for ambient light Long-term experiments system at constant temperature 20 25 30 35 °C 25 days material stable at T ≤ 25 °C no significant change transmittance upper limit-use storage temperature remarkable response for T ≥ 25 °C changes variations few degrees temperature-sensitivity At 35 °C transmittance fabrics programmed by 80 mJ/cm2 increases to ΔTr ~ 30%1d), surface reflectance decreases (open dots Fig. 1d).Pattern activation theoretical device displaying warning sign exposure to temperature 55 °C photographed left images Fig. 2a–d transition Supplementary thermal pictures infrared thermo-camera right images visual output based on contrast in reflectance heating between regions exposed second photo-programming step Exposed regions remain whitish high reflectance unexposed areas melt into continuous film decrease reflectance unveiling substrate for deposition TTI exploits ambient light no need powering sources display desired shape pattern depending optical mask second photo-programming thermal images Fig. 2b–d highlight regions fibers present negative SU-8 compound maintain local temperature lower (by ∼1 °C Fig than higher optical transmission due to cooling efficiency organic non-woven high heat view transition non-woven fabric increasing temperature captured by scanning electron microscopy (SEM) imaging interface across two regions exposed second photo-programming step Optical micrographs interface TTI at 55 °C shown in Fig. 2e–h SEM pictures of fibrous interfaces in Fig.2i–l views higher magnification unexposed fibers melting Fig. 2m–p Fig. 2i–l fibers exposed second photo-programming stable times explains different optical properties.Fig. 2Pattern activation changing Photographs thermal images fibrous samples after second step photo-programming b–d Photographs thermal images heating 55 °C color scale indicates temperature (Scale bars (a–d): 5 mm). e–h optical micrographs a–d letter “H” sign “Hot” arrow w-UV) indicates area exposed UV second step Scale bars 200 μm i–l SEM micrographs interface UV exposed unexposed areas Scale bars 20 μm m–p Higher magnification areas not exposed UV second step Scale bars 10 μm analyze behavior model light-scattering properties fibrous non-woven fabrics transition matrix polymer filament linear aggregate particle clusters spherical subunits 700 nm diameter fiber average random morphology fibrous surface three layers fibers thickness direction (z axis Fig. 3a, b). model structure schematized Fig.3a b green red blue clusters represent filaments first second third layer backscattering intensity map at λ = 510 nm ratio backscattered incident field |Es/E0|2) obtained plane-wave illumination amplitude perpendicularly structure averaging over polarization calculated ∼1 μm distance from first layer (Fig. 3c). important parameter albedo value A = Cscat/Cext ratio scattering Cscat extinction Cext light-scattering cross-sections T-matrix calculations yield albedo A510 = 0.73 at λ = 510 nm one higher than reflectivity continuous film SU-8 Fresnel normal incidence (0.06) ambient light backscattering maps at wavelengths RGB standard 450 530 600 nm (Fig. S9). Fig. 3d backscattering intensity map sum single-wavelength components intense scattering ambient illumination averaged albedo ARGB = fiber melting temperature albedo values decrease approach SU-8 films investigation based three-dimensional model fibrous multi-layer rationalizes whitish appearance nanofiber non-wovens reflectance contrast indicators operation.Fig.back-scattering three-dimensional fiber non-wovens Scheme structure nanofibers calculations structure fibrous non-woven area 10 × 10 μm2 x–y plane three layers axial (z) direction thickness 2.1 μm (green first red second blue third average fiber diameter (700 nm) equals SEM (Fig. S2 Planar view cluster discretization light-scattering properties polymer filament linear cluster 700 nm spherical subunits Backscattering intensity map at λ = 510 nm calculated ratio backscattered field incident one plane wave illumination unitary amplitude perpendicularly fibers polarization calculated 1 μm distance from first layer RGB intensity map sum backscattering maps red green blue wavelengths-matrix calculations photon transport properties non-wovens two length first scattering mean free path ls distance free propagation between scattering calculated average cluster subunit density (N = 0.51 μm−3 spherical subunit scattering cross-section (σscat = 1.6 μm2 at λ = 510 ls = 1/Nσscat = 1.2 μm.second length scale with transport mean free path l* distance direction propagation photon randomized calculated subunit anisotropy parameter g = <cos θ> θ scattering angle similarity relation33 l* = ls/(1−g). structures g = 0.65 typical transport length l* = 3.4 μm Supplementary Information visible light on fibrous material extinct within microns of propagation estimating albedo values for samples thickness experimental values (20–30 μm intensity backscattered light maximized with geometry (Fig. S10) leading reflectance contrast in indicators.Time–temperature indicator devicesPhoto-programming sequential exposures makes devices versatile targeted time–temperature profiles calibrated doses cross-linking degree structural relaxation non-wovens allow optical response indicators match expiration intervals of different perishable products Tables S1 S2 at temperature varied thermal history Examples include food pathogen growth toxin formation drugs vaccines degrade in days above room temperature cases covered by thermal response of photo-programmed TTIs in Fig. 1b.Strategies increase operational timescale thermal stability calibrating photoacid generator electrospun blends tuning substrate reflectivity electrospinning nanocomposites slower relaxation dynamics reduced local mobility filler-interacting polymer chains cross-linked clusters visual sequences two devices encoded first exposure at 72 80 mJ/cm2 kept at 35 °C Fig. 4a b “Do not use” sign appears reflectance contrast increases minutes first hours second Higher targeted temperature longer time response controlling cross-linking Fig. 1b devices flexible water-proof depositing fibers paper encapsulating polyethylene terephthalate (PET sheets conformable curved surfaces Yellow PET sheets encapsulation TTIs operate outdoor under direct sun light Fig. 4c d. 4Flexible time–temperature indicator paper substrate operation Photographs two programmed indicators 35 °C different response timescales minutes 72 hours 80 mJ times Operation during sunlight-induced warming goods (milk drugs) exposed sunlight 30 min temperature each measured displayed in red boxphotographs show visual response of non-wovens yellow plastic sheets screen UV radiation prevent crosslinking (c top-right inset scale bars 5 mm). arrows highlight flexible multilayer interface at device edge Bottom magnified view indicator on drug blister before after (d sunlight warming changes in device reflectance contrast low detected by automatized time-lapse analysis Weber’s contrast\documentclass[12pt{minimal{amsmath-69pt\mathrm{w}} =}Cw=Ib−IIbwhere I mean gray intensity darker features background gray intensity activated pattern,37 temporal evolution of Cw rapid increase at Tw approaching saturation value (Fig. S12) similar to ΔTr data (black Fig. 1d). strategy for contrast enhancement vary reflectivity properties deposition substrate glossy paper as substrate (Fig. S13a–d). maximum value Cw = 0.5 (Fig.S13e f) factor 2.5 higher than Cw TTI devices opaque paper Cw on thickness non-woven effective UV dose regions exposed during first photo-programming step dark regions patterns Fig. S13a–d). h < hM = 30 μm Fig. S13f), UV intensity glossy paper substrate higher sample thickness hM IsUV(h < hM) > IsUV(h = hM). diffuse reflectivity glossy paper non-woven with h < hM receives higher UV dose h = hM higher SU-8 crosslinking improved retain filamentary structure programmed working temperature enhances intensity light decreases pattern contrast intensity at glossy paper decreases for thicker non-woven [IsUV(h = hM > > hM additional effect relevant additional UV exposure regions first photo-programming step during second programming step scattering UV light lateral spread illuminated features effect relevant increasing h data curve plot Fig. S13f. long-term stability experiments appropriate storage temperature freezing conditions month-scale shelf-life (Figs. S14 S15) studies extend storage limits.results showed thermal response for intelligent indicator engineered tuning cross-linking translating nano-textured non-wovens negative resists micro-lithographies working temperature changes alter material architecture optical behavior devices cheap flexible light-weight compatible with paper encapsulation conformable to monitored items size shape operation visual testing no electronics/liquid explained by backscattering intensity from fibrous material three-dimensional calculations programming nano-materials photo-encoded thermal properties established design rule for intelligent labels perishable products-programmed nanofibersSU-8 3025 diluted with acetone) volumes 0–16% test solutions electrospinning Uniform smooth fibers diameter 700 nm obtained with acetone/SU-8 volume electric field 0.7 kV/cm electrified jets from 27-gauge stainless-steel needle Electrospinning times 7 13 min to non-wovens thickness 20 30 μm measured by profilometry Substrates fibers silicon wafers aluminum foils indium tin oxide-coated glass paperfirst step photo-programming UV exposure 365 nm from cylindrical lamps 10 cm sample doses 0.3–80 mJ/cm2. second selective step (pattern activation 250, 305 mJ/cm2) MJB4 mask aligner system or low-cost bromograph doses UV measured calibrated power meter 843-R UV-visible detector 818-UV/DB detector positioned sample UV intensity measured function time temporal resolution 0.5 s effective UV dose calculated integrating temporal curve UV intensity time interval exposure times Packaging flexible indicators non-woven plastic foils thickness 75 μm commercial cold laminating machine optical propertiesDSC Q200 differential scanning calorimeter TA Instruments-Waters RCS90 cooling system nitrogen gas purge flow rate 50 mL/min Standard sample pans aluminum sealed before measurements Non-woven samples 2.5–4.3 mg heated −60 to 250 °C 10 °C/min IR spectroscopy FTIR spectrophotometer three samples each dose Thermograms analyzed by TA Universal analysisoptical transmission of non-wovens measured by UV/VIS spectrophotomer Lambda 950 FEG-SEM Merlin Zeiss voltages 1–18 kV image fibers on silicon substrates diffuse reflectance measured broadband lamp38 integration sphere) spectrometer optical fiber sample positioned access port collinear with incident light beam light beam directed sample perpendicular surface hemispherically diffused light detected by sphere spectrometer data light backscattered at solid angle 1.1 × 10−3 sr not collected reference sample used blank port coated high diffusive layer surface integration sphere infrared camera (FLIR A655sc temperature during heating at 55 °C 10 min maps backscattered intensity albedo non-woven light scattering theory T-matrix formalism cluster model39 Organic filaments modeled one-dimensional cluster spherical particles diameter equal average nanofiber transverse size structure network three dimensions 10 × 10 μm2 area nanofiber mats thickness 2.1 μm backscattering intensity maps calculated as ratio |Es/E0|2 Es amplitude backscattered incident fields plane wave illumination polarization statesWeber’s temporal evolution Cw evaluated devices fiber deposition paper substrate first photo-programming (72–80 mJ/cm2) second (305 mJ/cm2) device encapsulation cold lamination PET sheets Images TTIs temperature 35 °C measured smartphone photo-camera CMOS camera (DCC1645C Thorlabs long working-distance optical system (MVL7000 calculation Cw images converted 8-bit mean gray intensities exposed unexposed estimated Fiji-ImageJ software (U. National Institutes of Health Bethesda Fig. S12 S13).Supplementary | 48.2 | 0.624037 |
10.1038/s41467-020-20523-3 | PMC7814032 | Long-term potentiation at hippocampal CA1 synapses can be due to increasing the number and/or single-channel conductance of AMPA receptors. The authors show that PKA and CaMKII are necessary and together sufficient to increase single channel conductance, via insertion of calcium-permeable AMPA receptors. | Long-term potentiation (LTP) at hippocampal CA1 synapses can be expressed by an increase either in the number (N) of AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptors or in their single channel conductance (γ). Here, we have established how these distinct synaptic processes contribute to the expression of LTP in hippocampal slices obtained from young adult rodents. LTP induced by compressed theta burst stimulation (TBS), with a 10 s inter-episode interval, involves purely an increase in N (LTPN). In contrast, either a spaced TBS, with a 10 min inter-episode interval, or a single TBS, delivered when PKA is activated, results in LTP that is associated with a transient increase in γ (LTPγ), caused by the insertion of calcium-permeable (CP)-AMPA receptors. Activation of CaMKII is necessary and sufficient for LTPN whilst PKA is additionally required for LTPγ. Thus, two mechanistically distinct forms of LTP co-exist at these synapses. | IntroductionLong-term potentiation (LTP) of synaptic function is considered the major process underlying learning and memory1 where it is involved in synaptic engram formation2,3, yet the underlying cellular mechanisms remain incompletely understood. The best-characterized form of LTP occurs at the Schaffer collateral-commissural pathway (SCCP) in the hippocampus, where it is triggered by synaptic activation of NMDA (N-methyl-D-aspartate) receptors4 and is expressed as a persistent increase in AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptor-mediated synaptic transmission5. This modification is primarily due to a functional modulation of AMPA receptors (AMPARs), which may involve a change in the number of active channels (N) (termed LTPN) and/or their single-channel conductance (γ) properties (termed LTPγ) (e.g.6–9). Whilst there is considerable evidence that LTPN is triggered by activation of Ca2+/calmodulin-dependent kinase II (CaMKII)10,11 and involves exocytosis and lateral diffusion of AMPARs12,13, the mechanisms underlying LTPγ are largely unknown. The two most likely molecular mechanisms involve (i) CaMKII-mediated phosphorylation of Ser831 of GluA1, which can result in an increase in the time AMPARs dwell in higher conductance states14–16 or (ii) the insertion of calcium-permeable AMPA receptors (CP-AMPARs), which have a higher γ than their calcium-impermeable (CI) counterparts17,18.In the present study, we have tested the hypothesis that LTPγ is due to the insertion of CP-AMPARs in young adult rodents using two theta burst stimulation (TBS) induction protocols that differed only in the timing between episodes, and applied peak-scaled non-stationary fluctuation analysis (NSFA)19–21 to estimate γ before and after the induction of LTP6,15,22–25. We found that the compressed TBS protocol (cTBS – inter-episode interval of 10 s) resulted exclusively in LTPN, for which CaMKII was both necessary and sufficient. In contrast, a spaced TBS protocol (sTBS – inter-episode interval of 10 min) resulted in a transient increase in γ, lasting ~15 min, which was due to the insertion of CP-AMPARs and required both CaMKII and PKA. Insertion of CP-AMPARs mediates both the initial expression of LTPγ, by enhancing the net synaptic unitary conductance, and helps trigger the processes that lead to a persistent increase in synaptic efficacy that outlasts the increase in γ. Since the PKA-dependent form of LTP also requires de novo protein synthesis and has stimulation features similar to spaced behavioural learning, LTPγ is likely to underlie the formation of synaptic engrams and long-term memory.ResultsAn increase in γ is specifically triggered by a sTBS protocolSimultaneous field excitatory postsynaptic potential (fEPSP) recordings from stratum radiatum and somatic whole-cell recordings were obtained in response to baseline stimulation of two independent SCCP inputs (Fig. 1a). TBS was delivered to one input (test), while the second input served as a control for stability and heterosynaptic effects (Fig. 1c, d). Synaptic potentiation was quantified and γ was estimated using NSFA (Fig. 1e, f), as described previously6. To optimize the estimates of γ we used minimal stimulation and restricted our measurements to the first 20–30 min following TBS, since γ estimates are extremely sensitive to small fluctuations in series resistance20. Thus, our study focused on the induction and initial expression mechanisms of LTP.Fig. 1LTP and non-stationary fluctuation analysis (NSFA) methodology.a Schematic of a hippocampal brain slice for LTP experiments, and the positioning of recording (R1, R2) and stimulating (S1, S2) electrodes. The CA3 region was cut (dashed line) to reduce neuronal excitability. Representative field and whole cell responses (fEPSP and EPSC), simultaneously obtained from CA1 neurons. Five consecutive responses were averaged and the stimulus artifacts were blanked for clarity. b Induction protocols for weak, compressed and spaced TBS (wTBS, cTBS and sTBS) are graphically summarized. c Representative fEPSP recordings for LTP evoked by a single episode of TBS (weak TBS, blue arrow). d Simultaneously obtained EPSC recordings. e Upper traces are two sets of representative waveforms for individual sweeps (thin lines), superimposed with the scaled mean of 57 EPSCs (thick lines). Lower traces are the subtraction of the scaled mean from the representative individual EPSCs. f Corresponding current–variance relationship to estimate the unitary conductance (γ). Fluctuation of the individual decays was plotted against the mean EPSC. Solid line is a parabolic fit with 95% confidence intervals (shaded). Dotted line, the background average variance.In the first series of experiments we delivered three episodes of TBS, with each episode comprising 5 shocks at 100 Hz delivered 5 times at 5 Hz (i.e. 75 stimuli in total; see Fig. 1b schematic); in interleaved experiments we either delivered these three episodes as a cTBS (10 s inter-episode interval) or as a sTBS (10 min inter-episode interval). We referred to the resultant potentiation as cLTP (Fig. 2a–i) and sLTP (Fig. 2j–r), respectively. In response to cTBS there was a substantial cLTP (Fig. 2a), with EPSC amplitudes increasing to 212 ± 11% of baseline, averaged over the first 10 min after induction (Fig. 2b). For 22 neurons from 15 rats (n = 22/15), we obtained γ estimates in 10 min epochs and found it to be unaltered throughout (Fig. 2c–g). The γ values were 5.1 ± 0.3 pS, (baseline), 5.3 ± 0.4 pS (first 10 min epoch post cTBS; LTP10’; t21 = 1.23, p = 0.2327, vs baseline, paired Student’s t test) and 5.2 ± 0.4 pS (second 10 min epoch post cTBS; LTP20’; t21 = 0.33, p = 0.7452; Fig. 2d). The control input was also stable throughout (4.9 ± 0.4 pS, 4.5 ± 0.3 pS and 4.8 ± 0.3 pS at the corresponding time-points; Fig. 2d). The lack of change in γ was also clearly evident in the plots from individual experiments for control (Fig. 2e) and test inputs (Fig. 2f) and in the cumulative distribution plots (Fig. 2g). The lack of change in γ was observed over a wide range of cLTP magnitudes (Fig. 2h).Fig. 2Increased AMPA receptor unitary conductance (γ) during sLTP, but not cLTP.a A representative LTP experiment with sample traces for baseline and post TBS – the mean of selected records for analysis, superimposed with peak-scaled individual traces (10 successive sweeps, thin lines; baseline = grey, LTP = black). Scaled trace is from the baseline normalized to the LTP. Scale bars: 20 pA and 10 ms. Two inputs were stimulated alternately and cTBS (3 x TBS with an inter-episode interval of 10 s; blue arrows) delivered to one input (filled symbols) with the second input (open symbols) serving as a control (Con). b Levels of cTBS-induced LTP (cLTP) for control and test inputs, quantified during the 10 min epoch after the induction (mean ± SEM, n = 22 neurons from 15 animals; t21 = 8.545, p < 0.0001, two-sided paired Student’s t test). c Corresponding current–variance relationship of the EPSCs for the test input. The unitary channel conductance (γ) of AMPA receptors was estimated during baseline (grey) and after the induction of LTP (LTP10’; black). d Grouped comparison of control and test input γ estimates for baseline and the initial 10 min epoch (LTP10’) and the subsequent 10 min epoch (LTP20’). n = 22 neurons from 15 animals. e, f Summary plot for the γ at baseline (left) and LTP10’ (right) for control (e) and test (f) inputs. Individual values from each neuron are connected by lines. Circles indicate mean values. g Cumulative distribution of the same data set for LTP10’. Dotted lines indicate the mean values for each input. h, i Analysis of the relationships of γ with LTP (p = 0.6517, F(1,20) = 0.2101, F-test) (h) and EPSC decay time (p = 0.9521, F(1,20) = 0.0037, F-test) (i). Linear regression with 95% confidence intervals (shaded) for the amount of cLTP and the corresponding level of γ. j–r Equivalent analysis for the LTP induced by sTBS (3 x TBS at inter-episode interval of 10 min; see arrows). The whole-cell recordings were obtained after the second TBS. This was necessary due to the lability of LTP washout. k Levels of sTBS-induced LTP (sLTP) for control and test inputs, quantified during the 10 min epoch after the induction (n = 23 neurons from 17 animals; t22 = 5.238, p < 0.0001, two-sided paired Student’s t test). m–o Statistical analysis between control and test pathways (t22 = 3.220, p = 0.0039 for baseline and t22 = 6.123, p < 0.0001 for LTP10’, two-sided paired Student’s t test) (m) and within pathway analysis for control (t22 = 1.065, p = 0.2986, two-sided paired Student’s t test) (n) and test (t22 = 3.753, p = 0.0011, two-sided paired Student’s t test) (o) pathway reveals a time- and pathway-dependent increase in γ. Note that higher conductance was observed in the test input (o) compared to the control (n) under the “baseline” state, suggesting that the first + second TBS were sufficient to increase γ. The third TBS triggered a small but discernible further increase in γ. q, r Analysis of the relationships of γ with LTP (p = 0.0225, F(1, 21) = 6.066, F-test) (q) and decay time of EPSCs (p = 0.0712, F(1,21) = 3.612, F-test) (r). Data are presented as mean ± SEM. Source data are provided as a Source Data file.In response to sTBS the results were strikingly different. For this set of experiments, whole-cell recordings were obtained shortly after delivery of the second TBS episode and the effects of the third TBS were evaluated (Fig. 2j). This method was necessary because of the rapid wash-out of LTP with low access whole-cell recordings. In response to the third TBS there was a substantial additional LTP, with EPSC amplitudes increasing to 177 ± 9% of baseline, averaged over the first 10 min after induction (Fig. 2k). The estimate of γ upon break in was significantly higher (6.9 ± 0.4 pS) compared to the control input (4.9 ± 0.4 pS; Fig. 2n–o; t22 = 3.22, p = 0.0039, paired Student’s t test) and this was further increased in response to the third episode of TBS to 8.4 ± 0.4 pS (LTP10’; t22 = 3.75, p = 0.0011, Fig. 2l, m, o, p; n = 23/17). However, when we quantified γ at 10–20 min after the last TBS, the value (5.5 ± 0.3 pS) was no longer significantly different from the control input (LTP20’; t22 = 2.01, p = 0.0570, paired Student’s t test; Fig. 2m). In contrast to the test input, sTBS did not result in a significant γ change in the control input (4.9 ± 0.4 pS, 5.4 ± 0.4 pS and 4.6 ± 0.3 pS at the corresponding time points; Fig. 2m, n). Thus, the increase in γ is specifically related to sLTP. Furthermore, this increase in γ positively correlated with the magnitude of sLTP (Fig. 2q).Since sLTP, but not cLTP, is associated with the insertion of CP-AMPARs26,27 these results suggest that CP-AMPARs may account for the increase in γ. CP-AMPARs have slightly faster decay kinetics (τdecay) than CI-AMPARs25,28, which can be detected using single exponential fits to EPSC decays. We found that cLTP was not associated with an alteration in τdecay (Fig. 2i, Supplementary Table 1; t21 = 0.66, p = 0.5146, paired Student’s t test), whereas sLTP was associated with a highly significant decrease in τdecay (Supplementary Table 1; p = 0.0051, t22 = 3.11, paired Student’s t test). A regression analysis showed a trend for the τdecay to be inversely related with the increase in γ (Fig. 2r; p = 0.0712, F(1,21) = 3.61). Therefore, the kinetic analysis provides additional support for the notion that insertion of CP-AMPARs occurs during the induction of LTP in response to a sTBS.The role of PKA in LTPγIt is established that elevating cAMP by, for example, use of the phosphodiesterase 4 inhibitor rolipram, enables a weak stimulus to generate an enhanced PKA-dependent form of LTP29. Previously, we found that in the presence of rolipram a weak TBS, comprising one episode of TBS, generated an LTP that is largely dependent on the insertion of CP-AMPARs26. Here we used this same method as an independent means to investigate whether insertion of CP-AMPARs are responsible for the increase in γ. Since only one TBS is required to induce the PKA-dependent form of LTP in the presence of rolipram we could make γ measurements before and after the full induction of LTP. As illustrated in Fig. 3a, b, a single episode of TBS (wTBS; comprising 25 stimuli), when delivered in the presence of rolipram (1 µM), generated a robust LTP (234 ± 14% of baseline for test vs. 121 ± 6% for control input). We found that this LTP was also associated with a transient increase in γ (baseline = 4.9 ± 0.4 pS, LTP10’ = 8.0 ± 0.6 pS; t20 = 5.90, p < 0.0001) that returned to baseline by the second 10 min epoch (LTP20’ = 5.4 ± 0.3 pS; t20 = 1.39, p = 0.1810, paired Student’s t test) following the wTBS (n = 21/15; Fig. 3c, d, f, g). This potentiation required the wTBS since the control input was largely unaffected (5.1 ± 0.3 pS, 5.4 ± 0.5 pS and 4.8 ± 0.3 pS at the corresponding time points; Fig. 3d, e) and since the baseline γ values in the presence of rolipram were not significantly different to the baseline γ values in its absence (Fig. 3d–f; Supplementary Table 1). As was the case with the sLTP, the size of the change in γ correlated with the magnitude of LTP (p = 0.0024, F(1,19) = 12.27; Fig. 3h). Additionally, there was an associated reduction in τdecay (p = 0.0007, t20 = 3.99, paired Student’s t test; Supplementary Table 1) that also negatively correlated with the increased γ (p = 0.0199, F(1,19) = 6.46; Fig. 3i). These results further support the idea that insertion of CP-AMPARs mediates LTPγ.Fig. 3Increased AMPA receptor unitary conductance (γ) during LTP in the presence of rolipram.a–g Equivalent experiments to those illustrated in Fig. 2 for the LTP induced by a wTBS (a single episode of TBS) in the presence of rolipram (1 µM; n = 21 neurons from 15 animals). Scale bars: 10 pA and 10 ms. b Quantification for the levels of LTP for control and test inputs during the 10 min epoch after the induction (mean ± SEM, t20 = 7.860, p < 0.0001, two-sided paired Student’s t test). d–f Statistical analysis between control and test pathways for LTP10’ (t20 = 5.901, p < 0.0001, two-sided paired Student’s t test) (d) and within pathway analysis for control (t20 = 0.4416, p = 0.6635, two-sided Student’s t test) (e) and test (t20 = 6.059, p < 0.0001, two-sided paired Student’s t test) (f) inputs. h, i Analysis of the relationships of γ with LTP (p = 0.0024, F(1,19) = 12.27, F-test) (h) and decay time of EPSCs (p = 0.0199, F(1,19) = 6.462, F-test) (i). Data are presented as mean ± SEM. Source data are provided as a Source Data file.To more specifically test the requirement of PKA for driving alterations in γ, we included the catalytic subunit of PKA (PKA Cα; 300 U/mL) in the patch solution (Fig. 4). This treatment had little effect on the control input that did not receive any wTBS (Fig. 4a), suggesting that PKA alone has minimal effect on synaptic transmission. However, as was the case with rolipram, the wTBS in the presence of PKA Cα generated a robust potentiation (Fig. 4a) that was associated with an increase in γ (Fig. 4b, c). The levels quantified during baseline and 10 min post TBS (LTP10’) were 5.2 ± 0.5 pS and 7.8 ± 0.8 pS (t16 = 5.80, p < 0.0001, paired Student’s t test; n = 17/13; Fig. 4b). Once again, the increase in γ was only transient, since estimates of γ made between 10 and 20 min following the wTBS (i.e. LTP20’) were not significantly different from baseline (5.3 ± 0.5 pS; t16 = 0.37, p = 0.7163, paired Student’s t test; Fig. 4b).Fig. 4wTBS with PKA Cα transiently increases γ via CP-AMPAR insertion.a–b A wTBS in the presence of intracellular PKA Cα (300 U/mL) transiently increased γ (n = 17 neurons from 13 animals, mean ± SEM, F(1.931,27.04) = 52.89 for EPSCs and F(1.863,26.08) = 12.59 for γ, one-way repeated measures ANOVA followed by Bonferroni’s multiple comparisons test; *p < 0.05, **p < 0.01 and ***p < 0.001 vs. baseline). EPSCs (a) and γ (b) were analyzed in 10-min bins. A single episode of TBS (at time marked by an arrow) was delivered to one input (filled symbols) with the second input (open symbols) serving as a control; base = baseline. c A representative current–variance plot for PKA Cα plus wTBS for baseline, the first 10 min (LTP10’) and the last 10 min of LTP (LTP30’). Sample traces were obtained from baseline and LTP10’. Scale bars: 10 pA and 10 ms. d–f Equivalent experiments in the presence of IEM-1460 (IEM, 30 µM; n = 16 neurons from 13 animals, F(1.095,13.14) = 25.66 for EPSCs and F(2.184,26.21) = 0.2547 for γ, one-way repeated measures ANOVA followed by Bonferroni’s multiple comparisons test; *p < 0.05, **p < 0.01 and ***p < 0.001 vs. baseline). g, h Quantification of the levels of LTP (t31 = 3.006, p = 0.0052, two-sided unpaired Student’s t test) (g) and (t31 = 3.544, p = 0.0013, two-sided unpaired Student’s t test) γ (h) measured during the 10 min after wTBS with cumulative distributions (right). n = 17 neurons from 13 animals (PKA Cα + wTBS) and 16 neurons from 13 animals (PKA Cα + wTBS + IEM). i, j Analysis of the relationships between γ and LTP for PKA Cα + wTBS (p = 0.0021, F(1,15) = 13.72, F-test) (i) and PKA Cα + wTBS + IEM (p = 0.9090, F(1,14) = 0.0136, F-test) (j). k, l Analysis of the relationships between γ and EPSC decay time for PKA Cα + wTBS (p = 0.0117, F(1,15) = 8.243, F-test) (k) and PKA Cα + wTBS + IEM (p = 0.3931, F(1,14) = 0.7764, F-test) (l). Data are presented as mean ± SEM. Source data are provided as a Source Data file.To establish whether the increase in γ was indeed due to the insertion of CP-AMPARs we used IEM-1460 (IEM, 30 µM). Previously, we showed that IEM inhibited LTP triggered by a sTBS without affecting LTP triggered by a cTBS26,27. Since these two induction protocols activate NMDARs to a similar extent, the effects of IEM is unlikely to be due to a direct action on NMDARs. To establish whether this is indeed the case, we examined the effects of IEM on NMDAR-mediated EPSCs evoked by single pulses and during TBS. IEM had no effect whatsoever on NMDAR-mediated synaptic transmission (Supplementary Fig. 1).In the presence of bath applied IEM and PKA Cα in the patch pipette, the level of LTP triggered by the wTBS was significantly less than in its absence (202 ± 16% vs. 276 ± 19% of baseline, 10 min after wTBS; t31 = 3.01, p = 0.0052, unpaired Student’s t test; Fig. 4d, g), consistent with a component of LTP being generated by the insertion of CP-AMPARs when PKA is activated26,30–32. IEM completely prevented the transient increase in γ (baseline vs. LTP10’; 4.3 ± 0.5 pS vs. 4.3 ± 0.5 pS; t15 = 0.08, p = 0.9338, paired Student’s t test; Fig. 4e, h; n = 16/13; also see Supplementary Table 1). There was a strong correlation between the increase in γ with both the magnitude of LTP (Fig. 4i; p = 0.0021, F(1,15) = 13.72) and the decrease in τdecay (Fig. 4k; p = 0.0117, F(1,15) = 8.24) when wTBS was delivered in the presence of PKA Cα, but there were no such correlations when IEM was also present (Fig. 4j, l).In conclusion, we find that activation of PKA, that occurs during (i) a sTBS, (ii) a wTBS in the presence of rolipram or (iii) a wTBS in the presence of the catalytic subunit of PKA, results in the transient insertion of CP-AMPARs and that these receptors are responsible for the increase in γ during the initial expression phase of LTP.The role of CaMKII in LTPγCaMKII has been demonstrated to be both necessary and sufficient for the induction of LTP10,11. Consistent with this notion, when tested using a CaMKII selective antagonist, KN-62 (10 µM), we found that both cLTP and sLTP were substantially reduced (Fig. 5a, b). The levels of potentiation of 108 ± 8% (after 90 min of cTBS, n = 4 slices from individual animals; Fig. 5a) and 105 ± 7% (after 120 min of sTBS, n = 5 slices; Fig. 5b), respectively, were significantly less than that in the corresponding interleaved untreated control groups that potentiated 155 ± 5% (p = 0.0010, t9 = 4.79; unpaired Student’s t test; n = 7 slices) and 159 ± 3% (p = 0.0001, t13 = 6.87; unpaired Student’s t test; n = 10 slices), but were not significantly different from their respective control inputs (t3 = 1.50, p = 0.2315 and t4 = 0.66; p = 0.5426; paired Student’s t test).Fig. 5CaMKII does not affect γ.a, b CaMKII dependence of both forms of LTP (mean ± SEM). a cLTP, measured using fEPSP recordings, was inhibited by the CaMKII inhibitor, KN-62 (10 µM; n = 4 animals; green). b sLTP showed a similar sensitivity to KN-62 (n = 5 animals). Interleaved control experiments (n = 7 and 10 animals; black) are superimposed. t9 = 4.786 (p = 0.0010; cTBS) and t13 = 6.865, (p < 0.0001; sTBS) by two-sided unpaired Student’s t test. The sample traces were obtained at the time indicated by the numbers. Scale bars: 0.2 mV and 10 ms. c A representative whole-cell recording with the inclusion of activated CaMKII (250 U/mL) in the internal solution. The sample traces are averages of selected records for analysis, superimposed with individual scaled traces (10 successive sweeps, thin lines) after 5 and 15 min of whole-cell recording. Scale bars: 10 pA and 10 ms. d Pooled results (mean ± SEM, 5-min bins) for the effects on EPSC (%) by activated CaMKII (n = 15 neurons from 12 animals, F(1.378, 19.30) = 10.52, p = 0.0021, one-way repeated measures ANOVA) and interleaved control, heat-inactivated, CaMKII (n = 14 neurons from 11 animals, F(2.585, 33.61) = 1.096, one-way repeated measures ANOVA). Bonferroni’s post hoc multiple comparisons test; *p < 0.05, **p < 0.01 and ***p < 0.001 vs. the initial 5 min of whole-cell recording. e Rise times (20–80%, τrise) and decay time constants (τdecay) were plotted for the EPSCs used in the NSFA analysis for the neuron illustrated in c. f Current–variance relationships for this neuron used to estimate γ over 5 min epochs starting at 0 and 10 min after commencing whole-cell recording. g Time course for the estimates of γ for active vs. inactive CaMKII. One-way repeated measures ANOVA; F(3.837,53.72) = 0.2498 (n = 15 neurons from 12 animals, active CaMKII), F(3.706, 48.18) = 0.0795 (n = 14 neurons from 11 animals, inactive CaMKII). Bonferroni’s post hoc multiple comparisons test; *p < 0.05, **p < 0.01 and ***p < 0.001 vs. the initial 5 min of whole-cell recording. h, i Quantification for the levels of LTP (t27 = 3.125, p = 0.0040, two-sided unpaired Student’s t test) (h) and γ (t27 = 0.3539, p = 0.7262, two-sided unpaired Student’s t test) (i) measured over a 5 min epoch, commencing 10 min after starting whole-cell recording. j, k Analysis of the relationships between γ and LTP (p = 0.2265, F(1,13) = 1.611, F-test) (j) and EPSC decay time (p = 0.2813, F(1,13) = 1.264, F-test) (k) for the active CaMKII experiments. Data are presented as mean ± SEM. Source data are provided as a Source Data file.It has been suggested that the role of CaMKII in LTP involves an increase in γ14,15. To further examine the role of CaMKII in LTP we interleaved experiments where we applied either active or inactive (heat inactivated) CaMKII (250 U/mL) via the patch pipette and delivered baseline (low frequency) stimulation to monitor basal synaptic transmission. Consistent with previous reports33,34, activated CaMKII, but not inactive CaMKII, was sufficient to potentiate synaptic transmission (Fig. 5c, d, h). However, this potentiation was not associated with an increase in γ (Fig. 5f, g, i) or a change in rise and decay kinetics (Fig. 5e; see also Supplementary Table 1). The respective γ values for baseline (i.e. first 5 min of recording) and 10–15 min of whole-cell recording were 4.7 ± 0.6 pS and 4.3 ± 0.5 pS (t14 = 0.74, p = 0.4740, paired Student’s t test; n = 15/12; Fig. 5g). There was no correlation between γ change and either the magnitude of LTP (Fig. 5j; p = 0.2265, F(1,13) = 1.61) or τdecay (Fig. 5k; p = 0.2813, F(1,13) = 1.26). We can conclude, therefore, that CaMKII alone can generate substantial potentiation that does not involve any alteration in γ.Activation of CaMKII and PKA are both necessary and sufficient for LTPγSince neither PKA alone nor CaMKII alone affected γ, we explored whether the combination of the two kinases may be sufficient for the effect. We, therefore, patch loaded PKA Cα (300 U/mL) with either the active or inactive forms of CaMKII (250 U/mL). In interleaved experiments, we found that PKA Cα + active CaMKII produced a robust potentiation of synaptic responses, specifically 178 ± 10% of baseline when quantified 15 min after whole-cell (Fig. 6a, b, f). In this case, the effect was also associated with an increase in γ (Fig. 6d, e, g). The levels of conductance for the baseline and potentiation (calculated between 10 to 15 min of recording) were 4.6 ± 0.4 pS and 6.5 ± 0.4 pS, respectively (t17 = 5.38, p = 0.0002, paired Student’s t test; n = 18/15). Again, this effect was only transient, as the γ returned to baseline levels within 20–30 min of whole-cell recording (Fig. 6e). In contrast, inactive CaMKII plus PKA Cα, had no significant effect on synaptic transmission (112 ± 9%; Fig. 6b, f) or on γ (4.2 ± 0.4 pS vs. 4.2 ± 0.4 pS, n = 16/14; Fig. 6e, g). These results suggest that (i) both CaMKII and PKA are required for and (ii) their combined activity is sufficient for LTPγ at these synapses.Fig. 6CaMKII plus PKA Cα results in a transient synaptic insertion of CP-AMPARs and increase in γ.a–k Equivalent experiments as described in Fig. 5c–k but with the inclusion of activated CaMKII (250 U/mL) plus the catalytic subunit of PKA (PKA Cα, 300 U/mL) in the internal solution. Scale bars: 10 pA and 10 ms. b Pooled results (mean ± SEM, 5-min bins) for the effects on EPSC (%) by active CaMKII + PKA Cα (n = 18 neurons from 15 animals, F(2.560,43.52) = 38.34, one-way repeated measures ANOVA), CaMKII + PKA Cα + IEM-1460 (IEM, 30 µM; n = 20 neurons from 15 animals, F(2.064,39.22) = 4.079) and heat-inactivated CaMKII + PKA Cα (n = 16 neurons from 14 animals, F(2.682,40.23) = 1.301). Bonferroni’s post hoc multiple comparisons test; *p < 0.05, **p < 0.01 and ***p < 0.001 vs. the initial 5 min of whole-cell recording. e Time course for the estimates of γ. One-way repeated measures ANOVA followed by Bonferroni’s multiple comparisons test (*p < 0.05, **p < 0.01 and ***p < 0.001 vs. the initial 5 min of whole-cell recording); F(3.219, 54.72) = 9.927 (CaMKII + PKA Cα), F(4.067,77.28) = 0.5990 (CaMKII + PKA Cα + IEM) and F(3.587,53.80) = 0.1366 (heat-inactivated CaMKII + PKA Cα). Quantification for the levels of LTP (F(2,51) = 14.19) (f) and γ (F(2, 51) = 9.210) (g). One-way ANOVA followed by Bonferroni’s multiple comparisons test; *p < 0.05, **p < 0.01 and ***p < 0.001. h, i Analysis of the relationships between γ and LTP (p = 0.0618, F(1,16) = 4.073, F-test) (h) and EPSC decay time (p = 0.0340, F(1,16) = 5.440, F-test) (i) for CaMKII + PKA Cα. j, k Analysis of the relationships between γ and LTP (p = 0.4775, F(1,18) = 0.5263, F-test) (j) and EPSC decay time (p = 0.3760, F(1, 18) = 0.8241, F-test) (k) for CaMKII + PKA Cα + IEM. Data are presented as mean ± SEM. Source data are provided as a Source Data file.In additional interleaved experiments, the sensitivity to IEM was tested on the potentiation produced by CaMKII plus PKA Cα. Consistent with the involvement of CP-AMPARs, there was a reduced level of potentiation (Fig. 6b, f) and no change in γ in the presence of IEM (Fig. 6e, g). The respective amounts, quantified after 10–15 min of whole-cell recording, were 128 ± 9% of baseline (p = 0.0003 vs. CaMKII + PKA Cα, one-way ANOVA with Bonferroni’s correction) and 4.1 ± 0.4 pS (p = 0.0009 vs. CaMKII + PKA Cα, one-way ANOVA with Bonferroni’s correction; n = 20/15).The correlations between the change in γ with the level of LTP and decay times for these experiments are summarized in Fig. 6h–k. There was a substantial correlation between the increase in γ and the level of potentiation (Fig. 6h) and the decrease in τdecay (Fig. 6i) with CaMKII plus PKA Cα but no such correlation was found in the presence of IEM (Fig. 6j, k).The proportion of synaptically incorporated CP-AMPARs during LTPγTogether, the previous experiments provide multiple lines of evidence that LTPγ is due to the insertion of CP-AMPARs into synapses that contain CI-AMPARs. In order to determine the relative proportions of each it was necessary to measure γ for synapses containing either just CI-AMPARs or just CP-AMPARs, under our recording conditions. To achieve this, we used lentivirus-driven CRISPR/Cas9 expression to delete GluA2 in a fraction of neurons in vivo, allowing a direct comparison between a knock-out (KO) and a wild-type (WT) neuron within each adult brain slice (Fig. 7a). When compared with uninfected neighbouring neurons, the KO cells showed a reduced AMPAR synaptic transmission (Fig. 7b) and an inwardly rectifying current–voltage relationship (Fig. 7c, d). The level of γ in KO neurons was significantly higher at 17.3 ± 1.2 pS (n = 16) compared to 4.6 ± 0.4 pS for WT neurons (n = 17; t31 = 10.09, p < 0.0001, unpaired Student’s t test, data from 12 animals; Fig. 7e–h). This increase in γ in these neurons was associated with a decreased τdecay from 6.8 ± 0.4 ms in WT to 5.3 ± 0.4 ms in KO neurons (t31 = 2.85, p = 0.0026, unpaired Student’s t test, Fig. 7i). Assuming that these EPSCs were comprised of 100 and 0 % CP-AMPARs, respectively, then the increase in γ that we observed during sLTP can be explained by CP-AMPARs comprising ~30% of the synaptic current during the first 10 min following LTP induction.Fig. 7CP-AMPAR characterization in CRISPR_Gria2 knock-out neurons.a Schematic of dual whole-cell recordings for the CRISPR_Gria2 knock-out and neighbouring uninfected (Uninf.) neurons. Sparse expression following stereotactic lentivirus injection detected by co-expressed EGFP (green); blue, DAPI staining; Pyr, stratum pyramidale; Rad, stratum radiatum. Scale bar = 30 microns. b Scatterplot shows amplitudes of AMPAR EPSCs for each pair recorded simultaneously (open circles) and the mean ± SEM (filled circle; n = 18 pairs from 12 animals). c, d Quantification of the rectification index for pharmacologically isolated AMPAR-mediated EPSCs and the corresponding current–voltage relationship (mean ± SEM, n = 18 pairs from 12 animals, t17 = 17.38, p < 0.0001, two-sided paired Student’s t test). Scale bars: 100 pA and 10 ms. e–g Representative traces to measure the γ for the control and CRISPR_Gria2 knock-out neurons. Individual traces (thin lines) superimposed with the average. Scale bars: 30 pA and 10 ms. The lower panels are corresponding colour-coded images of all sweeps used in the NSFA (g). h, i Quantification of γ (t31 = 10.09, p < 0.0001, two-sided unpaired Student’s t test) and decay time (t31 = 3.273, p = 0.0026, two-sided unpaired Student’s t test) constants (n = 17 and 16 for Uninf. vs. CRISPR_Gria2 knock-out neurons from 12 animals). Data are presented as mean ± SEM. Source data are provided as a Source Data file.DiscussionNMDA receptor-dependent LTP has been extensively studied as the primary mechanisms utilized are crucial for the formation of long-term memories. Despite many molecules being discovered and different aspects of their regulation being uncovered, there are crucial gaps in our knowledge. One relates to the fact that long-term memory requires de novo protein synthesis yet most of our mechanistic understanding of LTP has been obtained from the study of a protein synthesis-independent form of LTP. A second pertains to the fact that much of this understanding has been derived from the study of juvenile animals, where technical issues have permitted more in-depth analysis, whereas most studies of learning and memory are conducted in adult animals. In the present study, we have addressed these issues by studying LTP at CA1 synapses in young adult rodents and have compared induction protocols that are known to activate the protein synthesis-independent (cTBS) and protein synthesis-dependent (sTBS, rolipram + wTBS) forms35,36. Using a cTBS protocol, LTP involved the insertion of additional CI-AMPARs, for which activation of CaMKII is both necessary and sufficient. Using a sTBS there was an additional LTP component that involved the transient insertion of CP-AMPARs, for which activation of CaMKII and PKA are both necessary and, in combination, sufficient. The insertion of CP-AMPARs increases AMPA receptor γ and this underlies the initial expression of this form of LTP, which we have termed LTPγ. The insertion of CP-AMPARs is transient and is replaced by a persistent increase in the number of CI-AMPARs.Two distinct postsynaptic forms of LTP at CA1 synapsesThe division of NMDA receptor-dependent LTP into multiple components was made on the basis of sensitivity to various pharmacological agents and substantiated by genetic studies36. In particular, when a single train (tetanus or TBS) is employed, the resultant LTP may be independent of both PKA activation and de novo protein synthesis; this is commonly referred to as LTP1 or E-LTP35,37. In contrast, when multiple trains are delivered, with an interval in the order of minutes, then there is often the generation of an additional PKA and de novo protein synthesis-dependent component of LTP, which is commonly referred to as LTP2 or L-LTP27,36,38,39. LTP2 is generally assumed to underlie long-term memory formation, that also requires de novo protein synthesis.NMDA receptor-dependent LTP has also been divided into two distinct postsynaptic mechanisms of expression, one involving an increase in the number of AMPARs without a change in γ (LTPN) and the other involving an increase in γ (LTPγ)6. Here, one of our goals was to determine whether these separate expression mechanisms specifically relate to LTP1 and LTP2. We found that LTP1 never involved an alteration in γ whereas LTP2 invariably did. The increase in γ was transient, lasting between 10 and 20 min and could be fully explained by the insertion of CP-AMPARs. In terms of signalling cascades, we found that activation of CaMKII was both necessary and sufficient for LTP1 whereas both CaMKII and PKA were required, and in combination were sufficient, for LTP2 (see model in Fig. 8). Our findings do not conflict with a large body of literature regarding alterations in AMPARs underlying LTP at these synapses and the roles of both CaMKII and PKA (e.g.26,30,40–42).Fig. 8Schematic outlining the induction of two mechanistically distinct forms of LTP.1 Under baseline conditions synaptic transmission is mediated by GluA2-containing, calcium-impermeable (CI)-AMPARs, two shown for simplicity. 2 The first theta-burst stimulation (TBS) activates NMDA receptors (NMDARs) and this drives more CI-AMPARs into the synapse by lateral diffusion from a peri-synaptic pool, via a process that involves CaMKII. PKA is also activated (via adenyl cyclase, not shown) and this induces the process of inserting GluA2-lacking calcium-permeable (CP)-AMPARs into peri-synaptic sites on the plasma membrane. 3 LTP is expressed by the increase in number of CI-AMPARs (LTPN) but synapses also become primed for LTPγ by the availability of peri-synaptic CP-AMPARs. 4 Within ~1 h, the peri-synaptic CP-AMPARs are removed and, presumably, degraded. 6 If a second TBS is delivered whilst the synapses are still primed (5) then NMDAR activation drives the peri-synaptically located CP-AMPARs into the synapse, via a CaMKII-dependent process. This might involve an exchange of CP-AMPARs for CI-AMPARs, which are removed from the synapse via a mechanism triggered by PICK1. 7 These CP-AMPARs increase synaptic strength due to their higher single channel conductance (LTPγ). However, their dwell time in the synapses is quite short (~15 min) before they are removed. If synapses remain active, such as by basal stimulation, activation of the transiently available, synaptic CP-AMPARs triggers protein synthesis and the insertion of more CI-AMPARs (8), which can extend the expression of LTP for long periods.The roles of CP-AMPARs and alterations in γ in LTP have been controversial6,24,43–45. However, these controversies can now be reconciled on the basis of the type of LTP under investigation. Under the conditions of our study, we could readily switch between forms of LTP that do not (LTP1) or do (LTP2) involve a CP-AMPAR component by simply altering the timing between TBS episodes. We saw similar effects when we compared whole-cell recordings, using minimal stimulation, with field potential recordings, which provide an average measure of synaptic transmission across a wide range of release probabilities P(r). Therefore, we do not expect that our observations are dependent on the P(r) of the synapse under investigation. However, the extent to which CP-AMPARs are involved in synaptic plasticity is likely to involve additional factors, such as the developmental stage of the animal, the level of stress experienced prior to euthanasia and the precise experimental conditions used, including the stimulus parameters employed31,32,43,44,46,47.We can conclude that LTP1 equates to LTPN and LTP2 with LTPγ. It is important to note, however, that although a compressed induction protocol (cTBS) will ordinarily result in just LTP1/LTPN, a spaced protocol will comprise a mixture of LTP1 and LTP2 (LTPγ), since the initial train will induce LTP1 upon which subsequent trains will add LTP2 under our experimental conditions. The relative proportion of these two components will depend on a variety of conditions, including the interval between the trains, with ~10 min being optimal for the induction of LTP226.On the mechanism of LTPγThe increase in γ is regulated by the c-terminal tail of GluA148 and could result from a CaMKII-dependent phosphorylation of Ser831 of GluA1 to directly modulate their multiple conductance states14,16 and/or by the insertion of CP-AMPARs25, since these have a higher single channel conductance than CI-AMPARs17,18. Our findings have demonstrated that LTPγ can be explained exclusively by the latter mechanism, since all changes in γ were eliminated by IEM. Furthermore, we found that activation of PKA plus CaMKII increased γ whereas CaMKII alone did not, despite leading to a substantial potentiation. The failure of CaMKII alone to increase γ, which is contrary to some previous studies14,15, could be explained on the basis of the native AMPAR configuration since γ alterations are affected by the subunit combination and accessary protein composition of AMPA receptors16,18. It is worth noting, however, that whilst activation of CaMKII alone was not sufficient to induce LTPγ its activation was necessary. It is possible, therefore, that phosphorylation of Ser831 of GluA1 is a necessary step for LTPγ. Such a mechanism would involve dual phosphorylation of GluA1 on Ser831 and Ser845, which is known to occur49. One scenario is that GluA1 is firstly phosphorylated on Ser845 to drive CP-AMPARs to peri-synaptic sites, for which considerable evidence already exists30,41,42,49–51. From here, they are next phosphorylated on Ser831 to drive them into the synapse. In this model, both phosphorylation steps are required for synaptic γ to increase because they are regulating different trafficking steps on route to the synapse. In which case, CaMKII should not be able to increase γ further in neurons lacking GluA2 because CP-AMPARs are already synaptically expressed. Future work could address this and other aspects of the temporal sequence and consequences of PKA and CaMKII-dependent phosphorylation of GluA1 for LTP.Our data are compatible with an exchange of a subset of CI-AMPARs for CP-AMPARs. The latter could be explained by a mechanism involving the Ca2+ sensor PICK152,53, which has been shown to bind and internalize GluA2-containing AMPARs to enable the insertion of CP-AMPARs during LTP54. The next step involves the replacement of the newly inserted CP-AMPARs with CI-AMPARs, a process that requires baseline (low frequency) synaptic activation26,43 and probably involves Ca2+ permeation through the CP-AMPARs themselves55. The rapid replacement of CP-AMPARs with CI-AMPARs was originally described at excitatory synapses onto cerebellar stellate neurons from P18-P20 rats56. At this synapse, high-frequency stimulation (tetanus) induces CP-AMPARs to be replaced with the equivalent number of CI forms resulting in a reduction in the synaptic current by a third, reflecting lower γ of the latter form. We observed an initial reduction in EPSC amplitude following the triggering of LTP, which might be explained, in part, by a one-to-one exchange of CP-AMPARs for CI-AMPARs. Additionally, the transient expression of CP-AMPARs could trigger an increase in the number of AMPAR slots at synapses that enables an increase in the number of CI-AMPARs above and beyond what can occur during LTP1.Since CP-AMPARs increase synaptic conductance why does there need to be an exchange for a greater number of CI-AMPARs to maintain the enhanced synaptic response? One possibility is that the expression of CP-AMPARs at these synapses needs to be restricted in time due to potential excitotoxicity57. Therefore, they can only provide a transient mechanism of expression whilst triggering the more persistent switch resulting in a larger number of CI-AMPARs.Developmental regulation of the expression mechanisms of LTPThere is strong evidence that the expression mechanisms of LTP are developmentally regulated. The co-existence of two mechanisms involving the insertion of CI-AMPARs and CP-AMPARs can account for the LTP at P146 and in young adults, as observed herein. However, at around P7, LTP is associated with a decrease in γ22, which is most likely explained by the replacement of CP-AMPARs with a larger number of CI-AMPARs. Early in development there is also the initial insertion of CP-AMPARs into synapses55 that appear to lack AMPARs altogether; so-called “silent” synapses58,59. A potential scenario is as follows: first synapses acquire CP-AMPARs, next these are replaced by more CI-AMPARs. Thereafter LTP can increase the number of these CI-AMPARs via two mechanisms, one of which involves the transient insertion of CP-AMPARs and one that does not.There have been far fewer studies regarding the mechanisms of synaptic plasticity in tissues obtained from adult animals compared to juvenile animals, although most learning and memory studies are conducted in adult animals. This is a concern when attempting to relate mechanisms of synaptic plasticity to learning and memory. Our present study, conducted exclusively in tissue from young adult animals, shows that two distinct forms of synaptic plasticity can be readily induced simply by altering the patterns of activation. Our result that a cTBS protocol induces LTP that does not involve an alteration in γ is consistent with another study in adult animals24. Our finding that a sTBS induces an additional component of LTP that involves an increase in γ is the first evidence that such a process occurs beyond early developmental stages.Functional significance of two forms of LTPThis raises the question as to why there are two distinct mechanisms to increase the synaptic complement of CI-AMPARs. Previous work has shown that the insertion of CP-AMPARs is specifically associated with the PKA and protein synthesis component of LTP27. It is reasonable to assume, therefore, that the transient insertion of CP-AMPARs is part of the machinery that triggers de novo protein synthesis and the consequential morphological changes (spine enlargement and/or new spine formation). In contrast, in the absence of de novo protein synthesis, the increase in synaptic CI-AMPAR number can support increased synaptic efficacy. Although both processes can increase synaptic strength lasting many hours in vitro, it seems probably that only the protein synthesis-dependent form triggers synaptic changes that underpin long-lasting memories (lasting from days to lifetimes). Indeed, it has been shown that spaced training with access to reward enhances the persistence of memory, and treatment with rolipram after training enhances memory retention60. It seems likely that PKA triggers protein synthesis by phosphorylating GluA1 on S845 to promote the insertion of CP-AMPARs and by phosphorylating other regulatory targets and that together these regulate gene expression. The requirement for PKA to trigger the protein synthesis-dependent form of LTP also provides the opportunity for extensive neuromodulation. Neurotransmitters, such as noradrenaline and dopamine, and stress hormones, such as corticosterone, may, via the insertion of CP-AMPARs, augment protein synthesis-dependent LTP to enhance and/or prolong the persistence of the associated memory (e.g., 30,31,42,60–62).Concluding remarksWe have identified the molecular basis of two independent forms of LTP that co-exist at hippocampal synapses in young adult animals, the occurrence of which is controlled by the patterns of synaptic activation during induction. The existence of these two distinct LTP mechanisms goes a long way in explaining many of the controversies that have plagued the field. LTP1 can be induced by a cTBS and involves the insertion of CI-AMPARs, and for this to occur activation of CaMKII is both necessary and sufficient. A sTBS, however, triggers both LTP1 and LTP2. This latter form of LTP involves the transient insertion of CP-AMPARs and this requires activation of PKA in addition to CaMKII.MethodsHippocampal slice preparationTransverse hippocampal slices (400 μm) were prepared from male Sprague-Dawley rats (1–3 months of age). Animals were anesthetized with isoflurane and euthanised by decapitation in accordance with UK Animals (Scientific Procedures) Act of 1986. The brain was then removed and placed in ice-chilled slicing solution that contained (mM): 124 NaCl, 3 KCl, 26 NaHCO3, 1.25 NaH2PO4, 10 MgSO4, 10 D-glucose and 1 CaCl2, saturated with 95% O2 and 5% CO2. The hippocampi were rapidly isolated from the brain and sliced using a vibratome (Microslicer) while maintained in the slicing solution. The CA3 region was removed to suppress the upstream neuronal excitability, and the slices were transferred to an incubation chamber that contained the recording solution (artificial cerebrospinal fluid, ACSF; mM): 124 NaCl, 3 KCl, 26 NaHCO3, 1.25 NaH2PO4, 2 MgSO4, 10 D-glucose and 2 CaCl2 (carbonated with 95% O2 and 5% CO2). Slices were allowed to recover at 32–34 °C for 30 min, and then maintained at 26–28 °C for a minimum of 1 h before recordings were made.Field excitatory postsynaptic potential (fEPSP) recordingsThe extracellular electrophysiology was performed in both interface and submerged type chambers maintained at 32 °C, and continuously perfused at 2–4 mL/min with oxygenated ACSF. The slope of evoked fEPSPs (V/s) was measured in the CA1 region of hippocampal slices and bipolar stimulating electrodes were used at a constant voltage intensity (0.1 ms pulse width) throughout the experiments. Signals were amplified using Axopatch 1D (Molecular Devices) and digitized with BNC-2110 (National Instruments) A/D board at a sampling rate of 20 kHz. Recordings were monitored and analyzed using WinLTP v2.363. Each specific experiment was conducted on a single slice from an animal, so the n-value reflects both the number of slices and animals used.Two independent Schaffer collateral-commissural pathways (SCCPs) were stimulated alternately to obtain the evoked synaptic responses, each at a constant baseline frequency of between 0.033 and 0.1 Hz. Following a stable baseline period of at least 20 min, LTP was induced using theta-burst stimulation (TBS) delivered at the same basal stimulus intensity. An episode of TBS comprises 5 bursts at 5 Hz, with each burst composed of 5 pulses at 100 Hz. For LTP induced by compressed TBS (cTBS), three TBS episodes were delivered with an inter-episode interval (IEI) of 10 s. For spaced TBS (sTBS), the same number of episodes were given with an IEI of 10 min (see Fig. 1b). Representative sample traces are an average of 5 consecutive responses, collected from typical experiments (stimulus artefacts were blanked for clarity).Whole-cell patch clamp recordingWhole-cell recording was made with ACSF that contained 50 µM picrotoxin (Abcam) and 20 µM (+)-bicuculline (Hello Bio) to prevent GABAA receptor mediated contribution. CA1 pyramidal cells were visualized with IR-DIC optics (Zeiss). The whole-cell solution comprised (mM): 8 NaCl, 130 CsMeSO3, 10 HEPES, 0.5 EGTA, 4 Mg-ATP, 0.3 Na3-GTP, 5 QX-314 and 0.1 spermine. The pH was adjusted to 7.2–7.3 with CsOH and osmolarity was set to 285–290 mOsm/L. The peak amplitude of evoked EPSCs (pA) was monitored and analyzed using WinLTP v2.363. Two independent SCCPs were stimulated alternately, each at a baseline frequency of 0.1–0.5 Hz. Borosilicate glass pipettes were fire-polished with a final resistance of 2–4 MΩ. Access resistance (RA) was estimated by fitting whole-cell capacitance current with a double exponential, and experiments were only accepted for analysis if RA varied by <15%. RA values were 8.8 ± 0.3 MΩ; range from 6.2 to 12.8 MΩ. Signals were amplified using an Axopatch 200B (Molecular Devices), filtered at 2–5 kHz, and digitized at 20 kHz using a BNC-2110 (National Instruments) A/D board.Cells were voltage-clamped at −70 mV throughout unless otherwise indicated. LTP was induced using TBS delivered at basal stimulus intensity while in current-clamp mode, and was triggered within 10 min of whole-cell to prevent the dialysis effect. In some experiments, the PKA catalytic subunit (PKA Cα, 300 U/mL) and/or CaMKII (250 U/mL) were included in the internal solution. CaMKII was activated (1× NEBuffer for Protein Kinases; 50 mM Tris-HCl, 10 mM MgCl2, 0.1 mM EDTA, 2 mM DTT and 0.01% Brij 35; 200 µM ATP, 1.2 µM calmodulin and 2 mM CaCl2; incubated for 10 min at 30 °C) or heat-inactivated (65 °C for 20 min) as described in the suppliers’ manual (New England Biolabs). It is a Ca2+/calmodulin-dependent, truncated monomer (1–325 amino acid residues) of the α subunit, isolated from Spodoptera frugiperda (Sf9) cells infected with recombinant baculovirus carrying the truncated rat CaMKII (New England Biolabs).To ensure recording stability, extracellular field EPSPs were simultaneously monitored as described previously26. Peak amplitude (pA) and initial slope (V/s) of EPSCs and fEPSPs were measured, and displayed on-line, using WinLTP v2.363. Whole-cell recordings were initiated following collection of at least 10 min of stable baseline assessed by extracellular recordings.Peak-scaled, non-stationary fluctuation analysis (NSFA)The unitary conductance (γ) of AMPA receptors was estimated using NSFA according to ref. 6 (see also19–21). Whole-cell responses were carefully selected for analysis using WinWCP v5.1 (University of Strathclyde, Glasgow) and Mini Analysis v6.0 (Synaptosoft) software on the basis of the following criteria: first, precise alignment of traces on the rise phase; second, no contamination by spontaneous or polysynaptic currents; third, complete decay from the peak EPSCs. The traces were analyzed and the variance of the decay was plotted as a function of the amplitude at that time point. The x-axis was divided into 50-bins of equal current decrement from the peak. The single channel conductance was estimated by fitting the plot to a second polynomial equation, σ2 = iI - I2N + b1, where σ2 is the variance, I is the mean current, N is the number of channels activated, i is the single channel current and b1 is the background noise. In the conductance conversion (i.e. γ = i/V), the driving force (V) is the difference between the holding (−70 mV) and reversal potential (assumed to be 0 mV).The kinetics of the mean EPSC from each neuron was estimated in Clampfit v10.1 (Molecular Devices) by measuring 20–80% rise time (τrise) and the time constant for the decay (τdecay). Representative sample traces are the averages of all of the traces that were selected for analysis, superimposed with individual peak-scaled traces (10 successive sweeps), unless otherwise stated. Stimulus intensity was set to obtain a sporadic observation of transmission failures but high enough to obtain a reliable estimate of γ.Plasmid constructs and lentivirus productionThe following oligonucleotide sequences were used to generate single guide RNA (sgRNA) for GluA2 knockout: forward (5′ to 3′) CACC G ctaacagcatacagataggt; reverse (5′ to 3′) AAAC acctatctgtatgctgttag C64. These were annealed and ligated into the lentivirus backbone developed by the Zhang lab65. The construct was modified and used with the CaMKIIα promoter for Cas9-P2A-EGFP expression.Lentivirus was produced by transfecting Lenti-X 293 T cells (Takara Bio) with pMD2.G, psPAX2 and lentiCRISPR65. The 293 T cells were maintained in serum-free UltraCULTURE media (supplemented with 4 mM L-glutamine, 2 mM GlutaMAX-I, 0.1 mM MEM non-essential amino acids, 1 mM sodium pyruvate, 1× penicillin/streptomycin). Three days after transfection, the supernatant was filter sterilized (0.45 µm pore membrane, Millipore) and ultracentrifuged at 110,000 × g (Beckman Coulter) with an additional sucrose filtration. The lentivirus pellet was resuspended in Dulbecco’s PBS and kept at −80 °C.In vivo stereotactic injections and dual whole-cell recordingsThe surgical procedure was performed under sterile conditions in accordance with the Institutional Animal Care and Use Committee of Seoul National University. Male C57BL/6 mice (2–3 months of age) were anesthetized by intraperitoneal injection of a ketamine (130 mg/kg body weight) and xylazine (10 mg/kg) mixture. The anesthetized mice were immobilized on a stereotactic apparatus and the lentiviral medium (0.5 µL per each at a flow rate of 0.1 µL/min; 5 × 109 TU/ml) was bilaterally injected at CA1 area using a microinjection syringe (Hamilton). The coordinates used were −1.7 mm posterior, ±1.2 mm lateral to bregma and −1.5 mm below the skull surface.Following 4–6 weeks of expression, the hippocampal slices were prepared and whole-cell recordings were made as described above. EGFP-positive and neighbouring uninfected neurons were identified by epifluorescence microscopy and compared by dual whole-cell recordings. Rectification index was measured as described in ref. 26. AMPAR currents were isolated using a mixture of D-AP5 (100 µM) and L-689,560 (5 µM). The index was calculated by taking the responses from −70, 0 and +40 mV of holding voltages. Following the recordings, brain slices were PFA-fixed, stained with DAPI, and imaged on a confocal microscope (Leica SP8).CompoundsDrugs were prepared as frozen stock solutions (stored below −20 °C). Compounds were as follows: N,N,H,-Trimethyl-5-[(tricyclo[3.3.1.13,7]dec-1-ylmethyl)amino]-1-pentanaminium bromide hydrobromide (IEM-1460; Hello Bio); 4-(3-(cyclopentyloxy)-4-methoxyphenyl)pyrrolidin-2-one (rolipram; Abcam); 4-[(2 S)-2-[(5-isoquinolinylsulfonyl)methylamino]-3-oxo-3-(4-phenyl-1-piperazinyl)propyl] phenyl isoquinolinesulfonic acid ester (KN-62; Tocris and Hello Bio); D-AP5 (Hello Bio); L-689,560 (Tocris); a catalytic subunit of protein kinase A (PKA Cα, New England Biolabs); Ca2+/calmodulin-dependent protein kinase II (CaMKII, New England Biolabs).Statistical analysisAll treatment groups were interleaved with control experiments. Data are presented as mean ± SEM (standard error of the mean). Responses were normalized to the baseline prior to LTP induction unless otherwise stated. Statistical significance was assessed using (two-tailed) paired or unpaired Student’s t tests or one-way ANOVA as appropriate using Graphpad Prism 8. Adjustments were made for multiple comparisons using Bonferroni’s correction. The level of significance is denoted on the figures as follows: *p < 0.05, **p < 0.01 and ***p < 0.001.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Electrophysiology",
"Long-term potentiation"
] | -term potentiation (LTP) synaptic function major process underlying learning involved synaptic engram cellular mechanisms incompletely understood form LTP occurs Schaffer collateral-commissural pathway) hippocampus triggered by synaptic activation NMDA (N-methyl-D-aspartate receptors4 expressed persistent increase in AMPA receptor-mediated synaptic modification due to functional modulation AMPA receptors change active channels LTPN single-channel conductance (γ properties LTPγ evidence LTPN triggered by activation Ca2+/calmodulin-dependent kinase II involves exocytosis lateral diffusion of AMPARs12 mechanisms underlying LTPγ unknown two likely mechanisms CaMKII-mediated phosphorylation of Ser831 GluA1 increase time AMPARs dwell higher conductance of calcium-permeable AMPA receptors higher γpresent study tested hypothesis LTPγ due insertion CP-AMPARs young adult rodents two theta burst stimulation induction protocols timing applied peak-scaled non-stationary fluctuation analysis (NSFA estimate γ before after induction LTP6 compressed TBS protocol resulted LTPN CaMKII necessary sufficient spaced TBS protocol resulted transient increase γ ~15 min due CP-AMPARs required CaMKII PKA Insertion CP-AMPARs mediates initial expression LTPγ synaptic unitary conductance persistent increase synaptic efficacy increase γ PKA-dependent form LTP requires de novo protein synthesis stimulation features similar spaced behavioural learning LTPγ likely formation synaptic engrams long-term memory increase γ triggered by sTBS field excitatory postsynaptic potential) recordings stratum radiatum whole-cell obtained baseline stimulation two SCCP inputs (Fig TBS delivered one input second input control for stability heterosynaptic effects Synaptic potentiation quantified γ estimated using NSFAestimates γ used minimal stimulation restricted measurements first 20–30 min TBS γ estimates sensitive fluctuations series study focused induction initial expression mechanisms LTP. 1LTP non-stationary fluctuation analysis methodology Schematic hippocampal brain slice LTP experiments positioning recording stimulating electrodes CA3 region cut reduce neuronal excitability field cell responses (fEPSP obtained CA1 neurons Five consecutive responses averaged stimulus artifacts blanked Induction protocols weak compressed spaced TBS summarized fEPSP recordings LTP evoked single episode TBS obtained EPSC recordings Upper traces representative waveforms sweeps scaled mean 57 EPSCs Lower subtraction scaled mean EPSCs current–variance relationship unitary conductance Fluctuation decays plotted against mean EPSC parabolic fit 95% confidence intervals background average variance first series delivered three episodes TBS 5 shocks 100 Hz 5 times 5 Hz 75 stimuli totalexperiments delivered three episodes cTBS (10 s or sTBS (10 min potentiation cLTP (Fig. 2a–i sLTP (Fig. response cTBS substantial cLTP (Fig. EPSC amplitudes 212 ± 11% baseline first 10 min induction (Fig 22 neurons 15 rats obtained γ estimates 10 min epochs unaltered (Fig. 2c–g). γ values 5.1 ± 0.3 pS 5.3 ± 0.4 pS post cTBS 5.2 ± 0.4 pS post cTBS 0.7452 control input stable (4.9 ± 0.4 pS 4.5 ± 0.3 pS 4.8 ± 0.3 pS lack of change in γ evident plots experiments control test inputs cumulative distribution plots lack change γ range cLTP magnitudes (Fig. 2h).Fig 2Increased AMPA receptor unitary conductance (γ) during sLTP not cLTPrepresentative LTP experiment sample traces baseline post TBS mean records peak-scaled traces (10 sweeps thin lines baseline grey LTP Scaled trace baseline normalized LTP Scale bars 20 pA 10 ms Two inputs stimulated cTBS (3 inter interval 10 s blue arrows one input second symbols control Levels cTBS-induced LTP control test inputs quantified 10 min epoch induction ± SEM n 22 neurons 15 animals t21 = p < 0.0001 two-sided paired current–variance relationship EPSCs test input unitary channel conductance (γ) AMPA receptors estimated baseline after induction LTP comparison control test input γ estimates baseline initial 10 epoch subsequent n = 22 neurons 15 animals Summary plot γ baseline LTP10’ control test inputs values connected lines Circles mean values Cumulative distribution data LTP10’ Dotted lines mean values input relationships γ LTP (p = 0.6517 F(1 = 0.2101 EPSC decay time (p = 0.9521 Linear regression 95% confidence intervals cLTP level γanalysis LTP induced by sTBS (3 x TBS 10 min whole-cell recordings obtained after second TBS LTP washout sTBS-induced LTP control test inputs quantified during 10 min epoch after induction 23 neurons 17 animals t22 = 5.238 p < 0.0001 t analysis control test pathways (t22 = 3.220 p = baseline = 6.123 p < 0.0001 analysis control (t22 = 1.065 p = 0.2986 test = 3.753 p = 0.0011 time- pathway-dependent increase in γ higher conductance in test input (o) control first second TBS increase γ third TBS triggered increase γ relationships γ with LTP (p = 0.0225 F(1 = 6.066 decay time of EPSCs (p = 0.0712 F(1 = 3.612 Data presented mean ± SEM Source data file response to sTBS results different whole-cell recordings obtained after second TBS effects third TBS evaluated (Fig necessary rapid wash-out LTP low access whole-cell recordingsthird TBS additional LTP EPSC amplitudes 177 ± 9% baseline first 10 min (Fig. estimate γ break in higher (6.9 ± 0.4 pS control input (4.9 ± 0.4 pS t22 = 3.22 p = 0.0039 increased third episode 8.4 ± 0.4 pS (LTP10’ t22 = 3.75 p = 0.0011 quantified γ 10–20 min after last TBS (5.5 ± 0.3 pS) control input (LTP20’ t22 = 2.01 p = 0.0570 sTBS significant γ change control input (4.9 ± 0.4 pS 5.4 ± 0.4 pS 4.6 ± 0.3 pS Fig. 2m increase γ related sLTP correlated magnitude sLTP (Fig. sLTP not cLTP associated CP-AMPARs26 CP-AMPARs account increase γ faster decay kinetics CI-AMPARs25 detected EPSC cLTP not associated alteration τdecay (Fig. 2i Table 1 t21 = 0.p = 0.5146 Student’s t sLTP decrease τdecay (Supplementary Table 1 p = 0.0051 t22 = 3.11 t regression analysis τdecay inversely related increase γ (Fig. 2r p = 0.0712 F(1,21) = 3.61). kinetic analysis insertion CP-AMPARs during induction LTP sTBS role PKA elevating cAMP 4 inhibitor rolipram weak stimulus enhanced PKA-dependent rolipram weak TBS LTP dependent insertion CP method insertion CP-AMPARs increase γ one TBS required PKA-dependent LTP rolipram γ measurements before after induction LTP Fig. 3a single episode TBS rolipram (1 generated robust LTP (234 ± 14% baseline 121 ± 6% control LTP transient increase γ (baseline = 4.9 ± 0.4 pS LTP10’ = 8.0 ± 0.6 pS t20 = 5.90 p < 0.0001) returned baseline second 10 min epoch (LTP20’ = 5.4 ± 0.3 pS t20 = 1.39 p = 0.1810 test following wTBS (n = 21/15d f potentiation required wTBS control input unaffected (5.1 ± 0.3 pS 5.4 ± 0.5 pS 4.8 ± 0.3 pS Fig. 3d baseline γ values rolipram not different absence (Fig. 3d–f 1) change γ correlated magnitude LTP (p = 0.0024 F(1,19) = 12.27 Fig. reduction τdecay (p = 0.0007 t20 = 3.99 negatively correlated increased γ (p = 0.0199 F(1,19) = 6.46 results support CP-AMPARs mediates LTPγ. 3Increased AMPA receptor conductance LTP rolipram Equivalent experiments Fig. 2 LTP induced wTBS rolipram (1 21 neurons 15 Scale bars 10 pA 10 ms levels LTP control test inputs 10 min epoch after induction (mean ± SEM t20 = p < 0.0001 Statistical analysis control test pathways LTP10’ (t20 = 5.901 p < 0.0001 analysis control (t20 = 0.4416 p = 0.6635 test = 6.059 <Analysis relationships γ LTP (p = 0.0024 F(1,19) = 12.27-test decay time EPSCs (p = 0.0199 F(1,19) = 6.462 Data mean ± SEM Source data file PKA alterations γ included catalytic PKA Cα 300 U/mL) patch solution (Fig. 4) treatment little effect control input wTBS PKA minimal effect synaptic transmission wTBS PKA Cα generated potentiation increase γ levels baseline 10 min post TBS 5.2 ± 0.5 pS 7.8 ± 0.8 pS (t16 = 5.80 p < 0.0001 increase γ transient estimates 10 20 min wTBS not different baseline (5.3 ± 0.5 pS t16 = 0.37 p = 4wTBS PKA Cα increases γ CP-AMPAR insertion wTBS intracellular PKA Cα (300 U/mL increased γ (n = 17 neurons 13 animals mean F(1.931 = 52.89 EPSCs F(1.86312.59 γ ANOVA Bonferroni’s comparisons test *p < 0.05 **p < 0.01 ***p < 0.001 EPSCs γ analyzed 10-min bins episode TBS one input second input control base current–variance plot PKA Cα wTBS baseline first 10 last 10 min Sample traces baseline LTP10’ Scale bars 10 pA 10 ms experiments IEM-1460 30 μM 16 neurons 13 animals F(1.095,13 25.66 EPSCs F(2.184,26.21) 0.2547 γ ANOVA comparisons test *p < 0.05 **p < 0.01 ***p < 0.001h LTP (t31 = 3.006 p = 0.0052 (t31 = 3.544 p = 0.0013 γ measured 10 min after wTBS cumulative distributions 17 neurons 13 (PKA Cα + wTBS 16 + wTBS + IEM). relationships γ LTP PKA Cα + wTBS (p = 0.0021 F(1,15) = 13.72 PKA Cα + wTBS + IEM (p = 0.9090 F(1,14) = 0.0136 relationships γ EPSC decay time PKA Cα + wTBS (p = 0.0117 F(1,15) = 8.243 PKA Cα + wTBS + IEM (p = 0.3931 F(1,14) = 0.7764-test Data mean ± SEM Source data file increase γ CP-AMPARs used IEM-1460 IEM inhibited LTP sTBS LTP cTBS26 activate NMDARs IEM unlikely NMDARs examined effects NMDAR-mediated EPSCs TBS no effect NMDAR-mediated synaptic transmission 1)IEM PKA Cα patch pipette LTP triggered wTBS less absence (202 ± 16% vs. 276 ± 19% baseline 10 min after wTBS t31 = 3.01, p = 0.0052 Fig. 4d, consistent LTP generated insertion CP-AMPARs PKA activated26 IEM prevented increase γ (baseline 4.3 ± 0.5 pS t15 = 0.08 p = Fig. 4e, h n = 16/13 strong correlation increase γ magnitude LTP (Fig. 4i p = 0.0021, F(1,15) = 13.72) decrease τdecay (Fig. 4k p = 0.0117 F(1,15) = 8.24) wTBS delivered PKA Cα no correlations IEM present (Fig. 4j, activation PKA sTBS wTBS rolipram catalytic subunit PKA results CP-AMPARs receptors responsible increase γ initial expression phase LTP role CaMKII necessary sufficient induction LTP10 tested CaMKII selective antagonist KN-62 (10 μM), cLTP sLTP reduced (Fig.levels potentiation 108 ± 8% (after 90 min cTBS 4 105 ± 7% (after 120 min sTBS 5 slices less control groups 155 ± 5% (p = 0.0010 t9 = 4.79 7 159 ± 3% (p = 0.0001 = 6.87 10 not different control inputs (t3 = 1.50 p = 0.2315 t4 = 0.66 0.5426 5CaMKII affect CaMKII dependence LTP cLTP inhibited CaMKII KN-62 (10 4 animals sLTP similar KN-62 (n = 5 control experiments = 7 10 animals superimposed t9 = 4.786 (p = 0.0010 t13 = 6.865 < 0.0001 two Student’s t test sample traces obtained time numbers Scale bars 0.2 mV 10 ms whole-cell recording activated CaMKII (250 U/mL) solution sample traces averages records superimposed scaled traces after 5 15 min-cell recording Scale bars 10 pA 10 msresults 5-min effects EPSC activated CaMKII 15 neurons 12 animals F(1.378 = 10.52 p heat-inactivated CaMKII 14 neurons 11 animals(2.585 33.61) = 1.096 Bonferroni’s post comparisons test *p < 0.05 **p < ***p < 0.001 initial 5 min whole-cell recording Rise times (20–80% decay time constants plotted EPSCs NSFA analysis neuron Current–variance relationships γ 5 min epochs Time course estimates γ active inactive CaMKII F(3.837,53.72) = 0.2498 15 neurons 12 F(3.706, 48.18) = 0.0795 14 11 inactive Bonferroni’s comparisons test *p < 0.05 **p < 0.01 ***p < 0.001 initial 5 min-cell recording levels LTP (t27 = 3.125 p = γ (t27 = 0.3539 p = 0.7262 measured 5 min epoch 10 min whole-cell recordingrelationships γ LTP (p = 0.2265 F(1,13) = 1.611-test EPSC decay time (p = 0.2813 F(1,13) = 1.264 active CaMKII experiments Data mean ± SEM Source data file suggested role CaMKII LTP increase interleaved experiments applied active or inactive CaMKII (250 U/mL) patch pipette baseline (low frequency stimulation basal synaptic transmission activated CaMKII not inactive potentiate synaptic transmission (Fig. 5c potentiation not associated increase γ change rise decay kinetics γ values baseline 10–15 min whole-cell recording 4.7 ± 0.6 pS 4.3 ± 0.5 pS (t14 = 0.74, p = n = 15/12 no correlation between γ change magnitude LTP p = 0.2265 F(1,13) = 1.61) τdecay (Fig 5k p = 0.2813 F(1,13) = 1.26) CaMKII generate substantial potentiation alteration γ.Activation CaMKII PKA necessary sufficient for neither PKA CaMKII affected γ combination two kinasespatch loaded PKA Cα (300 U/mL with active inactive CaMKII (250 U/mL). PKA Cα + CaMKII potentiation synaptic responses 178 ± 10% baseline 15 min after whole-cell (Fig. 6a increase γ (Fig. 6d levels conductance baseline potentiation 4.6 ± 0.4 pS 6.5 ± 0.4 pS (t17 = 5.38 p = 0.0002 n = effect transient γ returned baseline within 20–30 min whole-cell inactive CaMKII plus PKA Cα no effect synaptic transmission (112 ± 9% γ (4.2 ± 0.4 pS results suggest CaMKII and PKA required activity sufficient for LTPγ. 6CaMKII plus PKA Cα transient synaptic insertion CP-AMPARs increase γ Equivalent experiments Fig. 5c–k activated CaMKII (250 U/mL) PKA Cα 300 U/mL) solution Scale bars 10 pA 10 ms results effects EPSC (%) by active CaMKII + PKA Cα (n = 18 15 = 38.measures CaMKII PKA Cα IEM-1460 30 μM 20 neurons 15 animals F(2.064,39.22) = 4.079) heat-inactivated CaMKII PKA Cα 16 neurons 14 animals F(2.682,40.23) = 1.301) Bonferroni’s comparisons test *p < 0.05 **p < 0.01 0.001 5 min estimates γ comparisons < 0.05 < < 0.001 F(3.219, 54.72) = (CaMKII PKA F(4.067,77.28) = 0.5990 (CaMKII PKA IEM F(3.587,53.80) = 0.1366-inactivated CaMKII PKA levels LTP (F(2,51) = 14.19) γ 9.210-way ANOVA comparisons test *p < 0.05 **p < 0.01 ***p < 0.001 relationships γ LTP 0.0618 4.073 EPSC decay time 0.0340 CaMKII PKA Cα relationships γ LTP 0.4775 F(1 0.5263 EPSC decay time 0.3760 0.8241 CaMKII PKA Cα IEM Data mean ± SEM Source dataexperiments to IEM tested on potentiation CaMKII plus PKA Cα CP-AMPARs reduced potentiation (Fig. 6b f no change in γ IEM (Fig. 6e g). amounts after 10–15 min recording 128 ± 9% baseline (p = 0.0003 vs PKA 4.1 ± 0.4 pS (p = 0.0009 vs n = correlations between γ LTP decay times summarized in Fig. 6h–k substantial correlation between increase γ potentiation decrease τdecay with CaMKII plus PKA Cα no correlation IEM (Fig. 6j proportion of synaptically incorporated CP-AMPARs during experiments evidence LTPγ due to insertion CP-AMPARs synapses proportions measure γ for synapses CI-AMPARs or CP-AMPARs recording conditions used lentivirus-driven CRISPR/Cas9 expression delete GluA2 neurons comparison between knock-out (KO) wild-type (WT) neuron brain (Fig. 7a). uninfected KO cells showed reduced AMPAR synaptic transmission (Fig. 7b rectifying current–voltage relationshipγ KO neurons higher 17.3 ± 1.2 pS = 16 4.6 ± 0.4 pS WT neurons (n = 17 t31 = 10.09 p < 0.0001 12 animals Fig increase γ decreased τdecay 6.8 ± 0.4 ms WT to 5.3 ± 0.4 ms KO (t31 = 2.85 p = 0.0026 100 0 % CP-AMPARs increase γ sLTP CP-AMPARs ~30% synaptic current first 10 min LTP induction. 7CP-AMPAR characterization CRISPR_Gria2 knock-out neurons dual-cell recordings CRISPR_Gria2 knock-out uninfected neurons expression lentivirus injection co-expressed EGFP Scale bar 30 microns amplitudes AMPAR EPSCs mean ± SEM 18 pairs 12 rectification index isolated AMPAR-mediated EPSCs current–voltage relationship (mean ± SEM = 18 pairs 12 t17 = 17.38 p < 0.0001 t Scale bars 100 pA 10 ms traces γ control CRISPR_Gria2 knock-out neurons traces averageScale bars 30 pA 10 ms. lower panels colour-coded images sweeps in NSFA (g). Quantification of γ (t31 = 10.09 p < 0.0001 decay time (t31 = 3.273 p = 0.0026) constants (n = 17 16 Uninf. vs. CRISPR_Gria2 knock-out neurons from 12 animals). Data presented as mean ± SEM Source data Source Data file.DiscussionNMDA receptor-dependent LTP extensively studied crucial for formation long-term memories molecules discovered regulation gaps knowledge long-term memory requires de protein synthesis understanding LTP from protein synthesis-independent form LTP understanding derived from study juvenile animals issues analysis studies learning memory in adult animals present study addressed issues LTP at CA1 synapses in young adult rodents compared induction protocols protein synthesis-independent cTBS protocol LTP involved insertion additional CI-AMPARs activation CaMKII necessary sufficient sTBS additional LTP component transient insertion CP-AMPARs activation CaMKII necessary sufficientinsertion CP-AMPARs increases AMPA receptor γ underlies initial expression LTP LTPγ insertion transient replaced by increase CI-AMPARs.Two postsynaptic forms LTP at CA1 synapsesThe division NMDA receptor-dependent LTP into multiple components sensitivity to pharmacological agents substantiated by genetic studies36 single train (tetanus) resultant LTP independent of PKA activation de novo protein synthesis referred LTP1 or E-LTP35 multiple trains delivered additional PKA de novo protein synthesis-dependent component LTP LTP2 or L-LTP27 LTP2 long-term memory formation requires de novo protein synthesis.NMDA receptor-dependent LTP divided into two postsynaptic mechanisms expression one increase AMPARs without change γ other increase γ (LTPγ determine expression mechanisms to LTP1 and LTP2. LTP1 involved alteration in γ LTP2 did increase in γ transient 10 20 min explained by insertion CP-AMPARs activation CaMKII necessary sufficient for LTP1 CaMKII and PKA required sufficient for LTP2findings conflict with literature alterations AMPARs LTP at synapses roles CaMKII PKA. 8Schematic induction of two distinct forms LTP baseline conditions synaptic transmission mediated by GluA2-containing calcium-impermeable (CI-AMPARs first theta-burst stimulation activates receptors drives more CI-AMPARs into synapse diffusion PKA activated induces inserting GluA2-lacking calcium-permeable-AMPARs into peri-synaptic sites plasma membrane LTP expressed by increase CI-AMPARs synapses primed for LTPγ by peri-synaptic CP-AMPARs Within ~1 h peri-synaptic CP-AMPARs removed degraded second TBS delivered synapses primed NMDAR activation drives CP-AMPARs into synapse CaMKII-dependent process exchange CP-AMPARs for CI-AMPARs removed PICK1 CP-AMPARs increase synaptic strength higher single channel conductance dwell time in synapses short (~15 min) before removedsynapses active basal stimulation activation synaptic CP-AMPARs triggers protein synthesis insertion more CI-AMPARs extend expression LTP long periods roles of CP-AMPARs alterations in γ in LTP controversial6,24,43–45 controversies reconciled type LTP under investigation study switch between forms LTP or CP-AMPAR altering timing between TBS episodes similar effects compared whole-cell recordings minimal stimulation with field potential recordings average synaptic transmission release probabilities observations dependent on P(r) synapse under investigation CP-AMPARs synaptic plasticity likely additional factors developmental stage animal stress prior euthanasia experimental conditions stimulus parameters employed31,32,43,44,46,47 LTP1 equates to LTPN LTP2 with LTPγ compressed induction protocol) in LTP1/LTPN spaced protocol mixture of LTP1 and LTP2 initial train LTP1 subsequent trains add LTP2 conditions relative proportion of components on conditions interval between trains ~10 min optimal for induction of LTP226.mechanism LTPγThe increase γ regulated by c-terminal tail GluA148 from CaMKII-dependent phosphorylation of Ser831 GluA1 insertion CP-AMPARs25 higher single channel conductance than CI-AMPARs17 findings LTPγ explained by latter mechanism changes in γ eliminated by IEM activation of PKA plus CaMKII increased γ CaMKII alone not potentiation failure CaMKII to increase γ explained native AMPAR configuration γ alterations affected by subunit combination protein composition AMPA receptors16 activation CaMKII not sufficient LTPγ necessary phosphorylation of Ser831 GluA1 necessary for LTPγ mechanism dual phosphorylation of GluA1 on Ser831 Ser845 scenario GluA1 phosphorylated on Ser845 drive CP-AMPARs to peri-synaptic sites next phosphorylated on Ser831 synapse both phosphorylation steps required for synaptic γ increase trafficking steps CaMKII increase γ in neurons lacking GluA2 CP-AMPARs synaptically expressed Future work address temporal sequence consequences of PKA CaMKII-dependent phosphorylation of GluA1 for LTP.data compatible with exchange CI-AMPARs for CP-AMPARs explained by mechanism Ca2+ sensor PICK152 GluA2-containing AMPARs CP-AMPARs during LTP54 next step replacement inserted CP-AMPARs with CI-AMPARs requires baseline frequency synaptic activation26 involves Ca2+ permeation through CP-AMPARs rapid replacement CP-AMPARs with CI-AMPARs described at excitatory synapses cerebellar stellate neurons P18-P20 rats56 high-frequency stimulation induces CP-AMPARs with CI forms reduction synaptic current by third lower γ observed initial reduction in EPSC amplitude following triggering LTP explained by exchange CP-AMPARs for CI-AMPARs transient expression CP-AMPARs could trigger increase AMPAR slots CI-AMPARs CP-AMPARs increase synaptic conductance exchange for greater number CI-AMPARs enhanced synaptic response? expression CP-AMPARs due to potential excitotoxicity57 provide transient mechanism expression persistent switch larger number CI-AMPARs.Developmental regulation expression mechanisms evidence expression mechanisms LTP developmentally regulatedco-existence mechanisms insertion CI-AMPARs CP-AMPARs for LTP at P146 young adults at P7 LTP with decrease in γ22 likely explained by replacement CP-AMPARs with CI-AMPARs Early in development initial insertion of CP-AMPARs into synapses55 AMPARs “silent” synapses58,59 potential scenario synapses acquire CP-AMPARs replaced by more CI-AMPARs LTP increase CI-AMPARs via two mechanisms transient insertion CP-AMPARs one fewer studies synaptic plasticity in adult animals juvenile most learning memory studies concern learning memory present study young adult shows two forms of synaptic plasticity induced by altering patterns of activation cTBS protocol induces LTP alteration in γ consistent with study adult sTBS induces additional LTP increase in γ first evidence process beyond early developmental stages.Functional significance of two forms of two distinct mechanisms to increase synaptic complement CI-AMPARs Previous work insertion of CP-AMPARs associated with PKA protein synthesis LTP27assume transient insertion CP-AMPARs triggers novo protein synthesis morphological changes (spine enlargement new spine absence protein synthesis increase in synaptic CI-AMPAR number increased synaptic efficacy both processes increase synaptic strength protein synthesis-dependent form triggers synaptic changes long-lasting memories spaced training reward enhances persistence memory treatment with rolipram enhances memory retention60 PKA triggers protein synthesis phosphorylating GluA1 on S845-AMPARs other regulatory targets gene expression requirement for PKA trigger protein synthesis-dependent form LTP provides opportunity neuromodulation Neurotransmitters noradrenaline dopamine stress hormones corticosterone may augment protein synthesis-dependent LTP enhance prolong persistence memory identified molecular basis of two independent forms of LTP co-exist at hippocampal synapses in young adult animals occurrence controlled by synaptic activation during induction two distinct LTP mechanisms controversies LTP1 induced by cTBS involves insertion CI-AMPARs activation of CaMKII necessary sufficientsTBS triggers LTP1 LTP2. CP-AMPARs activation PKA CaMKII slice hippocampal slices (400 μm Sprague-Dawley rats (1–3 months anesthetized isoflurane euthanised decapitation UK Animals Procedures Act 1986 brain removed-chilled slicing solution 124 NaCl 3 KCl 26 NaHCO3 1.25 NaH2PO4 10 MgSO4 D-glucose 1 CaCl2 95% O2 5% CO2. hippocampi isolated sliced vibratome CA3 region removed slices transferred incubation chamber recording solution 124 NaCl 3 KCl 26 NaHCO3 1.25 NaH2PO4 2 MgSO4 10 D-glucose 2 CaCl2 O2 5% CO2) Slices recover 32–34 °C 30 min maintained 26–28 °C 1 h before recordings potential extracellular electrophysiology interface submerged chambers 32 °C perfused 2–4 mL/min oxygenated ACSFslope evoked fEPSPs measured CA1 region hippocampal slices bipolar stimulating electrodes used constant voltage intensity (0.1 ms pulse Signals amplified Axopatch 1D digitized BNC-2110 A/D board rate 20 kHz Recordings monitored analyzed WinLTP v2.363 experiment single slice animal n-value reflects slices animals Schaffer collateral-commissural pathways (SCCPs stimulated alternately synaptic responses constant baseline frequency 0.033 0.1 Hz stable baseline 20 min LTP induced theta-burst stimulation (TBS same basal stimulus intensity episode 5 bursts 5 Hz 5 pulses 100 Hz LTP compressed TBS three episodes inter interval 10 s spaced TBS same episodes IEI 10 min Fig 5 consecutive responses experiments artefacts blanked-cell patch clamp ACSF 50 μM picrotoxin 20 μM (+)-bicuculline) prevent GABAA receptor contributionCA1 pyramidal cells visualized IR-DIC optics-cell solution 8 NaCl 130 CsMeSO3 10 HEPES 0.5 EGTA 4 Mg-ATP 0.3 Na3-GTP 5 QX-314 0.1 spermine pH 7.2–7.3 CsOH osmolarity 285–290 mOsm/L peak amplitude EPSCs monitored WinLTP v2.363 Two SCCPs stimulated 0.1–0.5 Hz Borosilicate glass pipettes fire-polished resistance 2–4 MΩ estimated capacitance current double exponential accepted <15% values ± 0.3 MΩ 6.2 to 12.8 MΩ Signals amplified Axopatch 200B filtered 2–5 kHz digitized 20 kHz BNC-2110 voltage-clamped −70 mV LTP induced TBS stimulus triggered 10 min dialysis PKA catalytic subunit CaMKII (250 included solutionCaMKII activated NEBuffer Protein Kinases 50 mM Tris-HCl 10 mM MgCl2 0.1 mM EDTA 2 mM DTT 0.01% Brij 35 200 μM ATP 1.2 μM calmodulin 2 mM CaCl2 incubated 10 min 30 °C heat-inactivated (65 °C 20 min manual England Ca2+/calmodulin-dependent truncated monomer (1–325 amino acid residues α subunit Spodoptera frugiperda cells infected recombinant baculovirus extracellular field EPSPs monitored Peak amplitude initial slope/s fEPSPs measured displayed on-line WinLTP v2.363 Whole-cell recordings initiated 10 min stable baseline-scaled fluctuation analysis unitary conductance (γ) AMPA receptors estimated Whole-cell responses selected WinWCP v5.1 Mini Analysis v6.0 precise traces no contamination currents complete decay peak EPSCs traces analyzed variance decay plotted amplitude x-axis divided 50-bins current decrement peaksingle channel conductance estimated plot second polynomial equation σ2 = iI - I2N + b1 σ2 variance I mean current N channels activated single channel current b1 background noise conductance conversion γ = i driving force (V) difference holding (−70 mV reversal potential 0 kinetics mean EPSC neuron estimated Clampfit v10.1 20–80% time constant decay Representative sample traces averages superimposed peak-scaled traces (10 Stimulus intensity set sporadic transmission failures high reliable estimate γ.Plasmid constructs lentivirus oligonucleotide sequences single guide RNA GluA2 knockout forward reverse annealed ligated into lentivirus backbone Zhang lab65 construct modified CaMKIIα promoter for Cas9-P2A-EGFP expression.Lentivirus produced transfecting Lenti-X 293 T cells) with pMD2.G, psPAX2 lentiCRISPR65.293 T cells serum-free UltraCULTURE media 4 mM L-glutamine 2 mM GlutaMAX-I 0.1 mM amino acids 1 mM sodium pyruvate penicillin transfection supernatant sterilized ultracentrifuged 110,000 × g sucrose filtration lentivirus pellet resuspended Dulbecco’s PBS −80 °C injections dual whole-cell surgical procedure sterile conditions Committee Seoul National University C57BL/6 mice (2–3 months anesthetized ketamine xylazine (10 mg/kg immobilized stereotactic apparatus lentiviral medium (0.5 μL 0.1 μL/min injected CA1 area microinjection syringe coordinates −1.7 mm posterior mm lateral −1.5 mm skull surface 4–6 weeks hippocampal slices prepared whole-cell recordings EGFP-positive uninfected neurons identified epifluorescence microscopy compared dual whole-cell recordings Rectification index measured AMPAR currents isolated D-AP5 L-689,560 (5 index calculated −70 0 +40 mV voltagesrecordings brain slices PFA-fixed stained DAPI imaged confocal microscope (Leica SP8) prepared frozen solutions below −20 N-Trimethyl-5--ylmethyl)amino-pentanaminium bromide hydrobromide 4-(3-(cyclopentyloxy)-4-methoxyphenyl)pyrrolidin-2-one 4---isoquinolinylsulfonyl)methylamino]-3-oxo-3-(4-phenyl-1-piperazinyl)propyl isoquinolinesulfonic acid ester D-AP5 L-689,560 catalytic subunit protein kinase A Ca2+/calmodulin-dependent protein kinase II treatment groups interleaved control experiments Data mean ± SEM Responses normalized baseline LTP induction Statistical significance assessed Student’s t tests one-way ANOVA Graphpad Prism 8. Adjustments multiple comparisons Bonferroni’s correction level significance figures *p < 0.05 **p < 0.01 ***p < 0.001 Nature Research Reporting Summary | 47.7 | 1.006786 |
10.1038/s41467-020-19291-x | PMC7599336 | The 17q23 amplicon is associated with poor outcome in ER+ breast cancers, but the causal genes responsible endocrine resistance in this region are unclear. In this study, the authors demonstrate that PRR11 located at 17q23, is critical for conferring endocrine resistance through activation of PI3K signalling and therefore propose PI3K inhibition as a treatment for PRR11-amplified breast cancers. | The 17q23 amplicon is associated with poor outcome in ER+ breast cancers, but the causal genes to endocrine resistance in this amplicon are unclear. Here, we interrogate transcriptome data from primary breast tumors and find that among genes in 17q23, PRR11 is a key gene associated with a poor response to therapeutic estrogen suppression. PRR11 promotes estrogen-independent proliferation and confers endocrine resistance in ER+ breast cancers. Mechanistically, the proline-rich motif-mediated interaction of PRR11 with the p85α regulatory subunit of PI3K suppresses p85 homodimerization, thus enhancing insulin-stimulated binding of p110-p85α heterodimers to IRS1 and activation of PI3K. PRR11-amplified breast cancer cells rely on PIK3CA and are highly sensitive to PI3K inhibitors, suggesting that PRR11 amplification confers PI3K dependence. Finally, genetic and pharmacological inhibition of PI3K suppresses PRR11-mediated, estrogen-independent growth. These data suggest ER+/PRR11-amplified breast cancers as a novel subgroup of tumors that may benefit from treatment with PI3K inhibitors and antiestrogens. | IntroductionApproximately 80% of breast cancers are estrogen receptor (ER)-positive and depend on estrogen for growth1. Therapies for ER+ breast cancer inhibit ER signaling by directly antagonizing ER (i.e., fulvestrant) or by abolishing estrogen production (i.e., aromatase inhibitors). Adjuvant anti-ER therapies significantly reduce the risk of recurrence in patients with ER+ breast cancer2. However, approximately 20% of patients treated with adjuvant endocrine therapy eventually relapse with metastatic disease3. To date, several mechanisms of de novo and acquired resistance to endocrine therapy have been reported4. Due to advances in large-scale tumor DNA sequencing, several somatic alterations that promote endocrine resistance have been discovered. Mutations in the ligand-binding domain of ESR1, the gene encoding ERα, confer resistance to estrogen suppression via ligand-independent ERα transcriptional activity5. Amplification of growth factor receptors such as ERBB2 and FGFR1 has also been associated with endocrine therapy resistance6,7. Enrichment of CCND1 amplification in luminal B tumors also suggests a potential causal role with a drug-resistant phenotype1. More recently, Razavi and colleagues reported that mutations in components of the mitogen-activated protein kinase (MAPK) pathway and the ER transcriptional program, found in approximately 20% of ER+ breast cancers, are associated with shorter response to antiestrogen therapy8. Preclinical and clinical studies have suggested a critical role for hyperactivation of the phosphoinositide 3-kinase (PI3K)/AKT pathway in endocrine resistance9–12. In line with this causal role, the PI3Kα inhibitor alpelisib in combination with the ER antagonist fulvestrant was clearly superior than fulvestrant alone in patients with advanced ER+/PIK3CA mutant breast cancer13, leading to the approval of alpelisib + fulvestrant in this subgroup of ER+ breast cancers.We recently reported genomic profiling of ER+ breast tumors after short-term treatment with the aromatase inhibitor (AI), letrozole14. In this study, the 11q13.3, 8p11.23, and 17q21-23 amplicons significantly correlated with high levels of the proliferation marker Ki67 upon drug-induced estrogen suppression. FGFR1 and CCND1 amplification, in 8p11-12 and 11q13, respectively, were associated with resistance to letrozole as defined by maintenance of a high Ki67 score on treatment. Although the 17q23 amplicon has been associated with highly proliferative luminal B tumors and high risk of recurrence in ER+ breast cancers15,16, a specific gene or genes in this region that would be causal to endocrine resistance have not been uncovered. In a recent study, we performed whole transcriptome analysis on RNA extracted from 58 ER+ breast cancers from patients treated with prolonged neoadjuvant letrozole17. In this cohort, we identified PRR11 (Proline rich 11), a protein-coding gene located in chromosome 17q22-23, to be overexpressed in tumors resistant to estrogen suppression compared to letrozole-sensitive tumors. PRR11 has been implicated in poor outcome of various cancer types18–20, but the molecular basis for this association is unclear. We hypothesized that PRR11 amplification in the 17q23 amplicon promotes endocrine resistance in ER+ breast cancer. We show herein that high PRR11 is causally associated with estrogen-independent growth of ER+ breast cancer cells. This action involved a PR (proline rich) domain-dependent interaction of PRR11 with the p85 regulatory subunit of PI3K which reduces homodimerization of p85 and, in turn, is permissive of ligand-induced association of p110α with insulin receptor substrate 1 (IRS1) and activation of PI3K. Ectopic expression of PRR11 failed to promote estrogen-independent growth when p110α was knocked down, and PRR11-overexpressing cells were highly sensitive to PI3K inhibitors, suggesting that PRR11 amplification generates dependence on PI3K signaling, particularly in the setting of estrogen deprivation. Taken together, these data suggest a combination of PI3K and ERα targeted therapies is a rational approach against ER+ breast cancers with PRR11 amplification.ResultsPRR11 is associated with poor outcome of ER+ breast cancersIn order to identify genes associated with poor outcome of ER+ berast cancers treated with antiestrogens, we had performed whole transcriptome analysis on RNA extracted from 58 ER+ breast cancers from patients treated with long-term letrozole for a median of 7.2 months (Supplementary Table 1; cohort of Guerrero-Zotano et al.17). PRR11 mRNA was significantly upregulated in resistant tumors. In this study, resistance to estrogen suppression was defined by a preoperative endocrine prognostic index (PEPI) ≥ 4 and/or evidence of cancer relapse after a median follow-up of 5 years [log2 fold change > 1 and false discovery rates (FDR) < 0.05; Fig. 1a]. RNA-seq analysis of the treated tumors showed that among 51 genes in 17q23, PRR11 was the only gene significantly overexpressed in resistant vs. sensitive cancers (Log2 fold change = 1.15, FDR = 0.004, p = 1.02E-05; Fig. 1b). Patients with PRR11-high cancers displayed an increased risk of relapse in the same cohort [hazard ratio (HR) = 3.753; 95% confidence interval (CI), 1.045–13.47; Fig. 1c]. In the Kaplan–Meier Plotter database21, high PRR11 mRNA levels were also associated with a shorter relapse-free survival (RFS) of ER+/HER2− breast cancers treated with endocrine therapy (HR = 3.85; 95% CI, 1.95–7.59; Fig. 1d), but this association was not present in patients with HER2+ or triple-negative breast cancer (TNBC; Supplementary Fig. 1a). We further interrogated the association of PRR11 expression with response to estrogen suppression in two other clinical studies of ER+ breast cancers treated with a neoadjuvant aromatase inhibitor (Supplementary Table 1; cohort of Giltnane et al.14; cohort of Miller et al.22). In these studies, maintenance of a high Ki67 index on treatment was used as a surrogate of resistance to estrogen suppression. In these two cohorts, we also found a statistically significant correlation between PRR11 mRNA and on-treatment high Ki67 levels (Fig. 1e, f). To correlate PRR11 expression with response to estrogen suppression, PRR11 protein levels were evaluated by IHC in tumor sections from a previously reported cohort of 175 ER+ breast cancers treated with letrozole before surgery (NCT0065197614). Based on the exponential curve of PRR11 protein levels, tumors were classified as PRR11 ≤ 1% (negative), 1–15% (positive) and >15% (high; Supplementary Fig. 1b). PRR11 protein levels were statistically higher in tumors with poor response to letrozole, as defined by on-treatment high Ki67 levels14 (Fig. 1g, h).Fig. 1PRR11 is associated with poor clinical outcome of ER+ breast cancers treated with antiestrogens.a Volcano plot of genes differentially expressed in non-responding tumors compared to responding tumors. Log2 fold change (FC) and false discovery rates (FDR) were calculated using DeSeq2 package. b Log2 FC of 17q23 locus genes in ER+/HER2− breast cancers treated with long-term letrozole (n = 51). Genes in 17q23 locus were selected based on the Atlas of Genetics and Cytogenetics25. Red bars indicate genes with FDR < 0.05. c Recurrence-free survival in ER+ breast cancers, treated with long-term neoadjuvant letrozole, with low or high PRR11 mRNA levels. A PRR11 FPKM cut-off (3.93) obtained from the human protein atlas was used to divide PRR11-high (n = 43) and -low (n = 15) tumors (https://www.proteinatlas.org/ENSG00000068489-PRR11/pathology/breast+cancer). The Mantel–Cox model was used to calculate the hazard ratio (HR) and p value. d Relapse-free survival of ER+/HER2− breast cancers, treated with endocrine therapy, with low (n = 104) or high (n = 98) PRR11 mRNA levels by the auto select best cutoff in Kaplan–Meier Plotter. HR and p were adopted from the Kaplan–Meier Plotter (http://kmplot.com/analysis/). e, f Correlation between the on-treatment percent of Ki67+ tumor cells and PRR11 mRNA level in breast tumors from the cohort of Giltnane et al. and Miller et al. (Pearson correlation). g Representative PRR11 immunohistochemistry (IHC) images of primary ER+ breast tumors. h PRR11 levels were plotted as a function of response to estrogen suppression with letrozole in trial NCT0065197614. Data represent the mean ± SD (n = 91, 25, 39 for drug sensitive, intermediate and resistant group, respectively; two-tailed unpaired t-tests). Source data are provided as a Source data file.PRR11 has been reported as an estrogen-responsive gene23. Thus, we examined the possibility that high PRR11 mRNA remained high in resistant tumors because ER is not sufficiently suppressed by letrozole treatment. In another cohort of 18 ER+ breast tumors treated with an aromatase inhibitor (cohort of Miller et al.22), PRR11 mRNA was not downregulated in post-treatment compared to pre-treatment tumors (Supplementary Fig. 1c), implying PRR11 is not regulated by ligand-induced ER in vivo. Moreover, PRR11 mRNA levels did not correlate with estrogen response gene set signatures in three cohorts of tumors treated with an aromatase inhibitor (cohorts of Guerrero-Zotano et al., Giltnane et al., and Miller et al.; Supplementary Fig. 1d). Consistent with these correlations, exogenous estrogen did not increase PRR11 mRNA levels in MCF7 and HCC1428 cells (Supplementary Fig. 1e).PRR11 is a key gene associated with endocrine resistanceThe cytogenetic band of PRR11 is designated 17q22 or 17q23.2 (Ensembl or HGNC, respectively), located at the terminal region of 17q22 close to the 17q23 region (Supplementary Fig. 2a). The rate of PRR11 amplification is 15.9% in the Metastatic Breast Cancer (MBC) project, but 9.5% and 9.4% in METABRIC and The Cancer Genome Atlas (TCGA), respectively (Fig. 2a). These last two cohorts comprise primary breast tumors, thus suggesting a higher rate of PRR11 amplification in metastatic (in the MBC project) compared to ER+ primary breast tumors. Of note, most metastatic ER+ breast cancers have undergone adjuvant treatment with endocrine therapy, so metastatic ER+ tumors often harbor somatic alterations associated with endocrine resistance. In ER+ tumors in the MBC project dataset, PRR11 amplification was mutually exclusive with mutations in ESR1 and NF1, both established mechanisms of antiestrogen resistance (Fig. 2b). In the METABRIC cohort, PRR11 copy number gain or amplification predicted shorter disease-free survival of patients with ER+/HER2− breast cancer treated with antiestrogens (HR = 1.408; 95% CI, 1.029–1.926; Fig. 2c). Also in METABRIC, PRR11 copy number alterations (CNAs) were correlated with PRR11 mRNA expression (Supplementary Fig. 2b). In a panel of 56 breast cancer cell lines in the CCLE dataset24, high PRR11 copy number was significantly correlated with PRR11 mRNA (Supplementary Fig. 2c). Moreover, we observed that PRR11-amplified breast cancer cell lines express higher levels of PRR11 protein compared to non-amplified cell lines (n = 11; Supplementary Fig. 2d). We finally verified PRR11 amplification by fluorescence in situ hybridization (FISH) in a letrozole-resistant primary breast cancer (Supplementary Fig. 2e).Fig. 2PRR11 is a key gene in 17q23 associated with endocrine resistance.a Frequency of PRR11 amplification in MBC project, TCGA or METABRIC ER+ breast cancers. b Oncoprint of putative endocrine-resistant drivers in metastatic ER+ tumors from MBC project dataset. ER+ tumors classified as ‘METASTATIC DISEASE PRESENT’ were interrogated (n = 44). c Plot of disease-free survival of METABRIC ER+/HER2− breast cancer patients treated with anti-hormone therapy as a function of PRR11 copy number [gain/amplification (n = 163) vs. diploid/deletion (n = 834)]. d Venn diagram of genes in 17q23 that significantly correlated with on-treatment Ki67 levels (Pearson r > 0.4; p < 0.05). Source data are provided as a Source data file.The gene(s) in the 17q23 amplicon that are causal to endocrine resistance have not been identified. Using the Atlas of Genetics and Cytogenetics25, we found 90 genes located in the 17q23 amplicon (Supplementary Table 2). We next examined the association of these genes with on-treatment Ki67 levels in three clinical studies of ER+ breast cancers treated with a neoadjuvant aromatase inhibitor (Supplementary Table 1; cohort of Guerrero-Zotano17; cohort of Giltnane14; cohort of Miller22). This analysis revealed that four of these 90 genes in 17q23 (PRR11, BRIP1, SMARCD2, and TACO1) correlated statistically with on-treatment Ki67 levels (Pearson r > 0.4, p < 0.05; Fig. 2d; Supplementary Tables 3–5). Across these three studies, PRR11 was the only one of these four genes that exhibited a significant correlation with a high Ki67 score. Finally, among 67 genes in 17q23, high PRR11 mRNA levels also correlated with a shorter RFS in patients with ER+/HER2− breast cancer treated with endocrine therapy in the Kaplan–Meier Plotter database (Supplementary Fig. 2f). Collectively, these data suggest that PRR11 may be a key gene in 17q23 associated with endocrine resistance.Several genes in the 17q23 amplicon have been speculated to be associated with poor outcome in breast cancer15,26. To determine whether PRR11 is an essential gene in 17q23-amplified breast cancer cells, we interrogated genome-scale RNAi screening data of MCF7 cells that harbor 17q23 amplification27 in Project Achilles dataset (v2.4.3)28. shRNAs targeting 47 genes in 17q23 were screened in Project Achilles dataset. Of these 47 genes, PRR11 displayed the lowest Analytic Technique for Assessment of RNAi by Similarity (ATARiS)29 score, thus implying a high dependency of 17q23-amplified breast cancer cells on PRR11 (Supplementary Fig. 2g).PRR11 overexpression confers resistance to antiestrogensTo further prioritize genes in 17q23 associated with resistance to estrogen suppression, we transduced PRR11, BRIP1, SMARCD2, and TACO1 into MDA-MB-134VI and MDA-MB-175VII cells, which do not harbor 17q23 amplification (Fig. 3a). PRR11 was the only gene that promoted growth of both cell lines under conditions of estrogen deprivation (Fig. 3b). Next, we employed MCF7 LTED (long-term estrogen deprived) and HCC1428 LTED cells9. MCF7 and HCC1428 wild type cells harbor PRR11 high copy number24. PRR11 ablation by siRNA abolished estrogen-independent growth of MCF7 LTED and HCC1428 LTED cells (Fig. 3c). The inhibitory effect of PRR11 ablation was rescued by re-expression of PRR11 (Fig. 3d, e). To assess the role of PRR11 in vivo, we generated xenografts of MCF7 cells stably expressing a doxycycline-inducible PRR11 shRNA in ovariectomized athymic mice. Treatment with doxycycline markedly reduced PRR11 protein levels and arrested growth of established MCF7 xenografts expressing PRR11 shRNA but not a control shRNA (Fig. 3f and Supplementary Fig. 3a).Fig. 3PRR11 overexpression confers resistance to antiestrogens.a Lysates of MDA-MB-134VI and MDA-MB-175VII cells that had been transduced with pLX304-PRR11, -BRIP1, -TACO1, and -SMARCD2 were subjected to immunoblot analysis. b Low density monolayers of MDA-MB-134VI and MDA-MB-175VII pLX304-PRR11, -BRIP1, -TACO1, and -SMARCD2 cells were grown in estrogen-deprived condition. After 2 weeks, cell monolayers were stained with crystal violet and cell viability quantified as described in Methods. Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). c MCF7 LTED and HCC1428 LTED cells were transfected with PRR11 siRNAs. Low density monolayers of cells were treated ± 1 nM estrogen (E2) for 10 days. Cell monolayers were stained with crystal violet. Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). d, e MCF7 LTED and HCC1428 LTED cells were transduced with shRNA targeting the 3′ UTR of PRR11 and then, re-transduced with pLX304-GFP or pLX304-PRR11. Cell lysates were subjected to immunoblot analysis (d). Upper and lower arrows indicate exogenous and endogenous PRR11, respectively. Low density monolayers of cells shown in d were grown in absence of E2 for 10 days (e). Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). f MCF7 cells stably expressing doxycycline-inducible-PRR11 shRNA and control shRNA were injected s.c. in the dorsum of athymic ovariectomized mice supplemented with a 14-day release 17β-estradiol pellet. After 4 weeks, mice were randomized to treatment with 10 mg/kg doxycycline (i.p) for 4 weeks. Each data point represents the mean tumor volume in mm3 ± SD; n of mice per group are shown in parenthesis (two-tailed unpaired t-tests). g Fulvestrant sensitivity of ER+ breast cancer cell lines (n = 11, PRISM Repurposing 19Q3 dataset). Y-axis, drug sensitivity, represents relative barcode abundance following fulvestrant treatment (Pearson correlation). h, i Fulvestrant GR50 were calculated using the GR metrics calculator56 (http://www.grcalculator.org/grcalculator/). Cell numbers on days 0 and 6 were used as the input data. MDA-MB-134VI and MDA-MB-175VII cells were stably transduced with pLX302-LacZ (control) or pLX302-PRR11 (h). MCF7 FulvR cells were transfected with control or PRR11 siRNA (i). Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). Source data are provided as a Source data file.In the PRISM repurposing 19Q3 dataset30, PRR11-amplified ER+ breast cancer cell lines displayed a lower sensitivity to the ER antagonist fulvestrant compared to cells without PRR11 amplification (Fig. 3g). Transduction of PRR11 into MDA-MB-134VI and MDA-MB175VII cells attenuated growth inhibition by fulvestrant (Fig. 3h and Supplementary Fig. 3b). PRR11 knockdown re-sensitized fulvestrant-resistant (FulvR) MCF7 cells to fulvestrant (Fig. 3i and Supplementary Fig. 3c). A similar result was observed in tamoxifen-resistant (TamR) MCF7 and TamR HCC1428 cells (Supplementary Fig. 3d). These data suggest PRR11 promotes resistance to antiestrogens and its downregulation enhances the action of ER-targeted therapies against ER+ breast cancer cells.PRR11 promotes proliferation of ER+ breast cancer cellsWe next analyzed RNA-seq data in the cohort of Guerrero-Zotano et al.17 using 125 previously reported breast cancer-related gene expression signatures31. Nine proliferation-associated signatures were significantly enriched in tumors with high PRR11 mRNA expression (FDR < 0.01; Fig. 4a). Hallmark gene sets associated with proliferation, including “E2F_TARGETS” and “G2M_CHECKPOINT” were significantly enriched in ER+/HER2− tumors with PRR11 gain or amplification in METABRIC and in ER+ tumors with PRR11 amplification in TCGA (Supplementary Fig. 4a, b).Fig. 4PRR11 overexpression enhances cancer cell proliferation.a Single sample gene set analysis was performed using a set of 125 previously reported breast cancer-related signatures. Gene sets that were differentially enriched between PRR11 high vs. low tumors (FDR < 0.01). b Complementary DNA (cDNA) of MCF7 LTED cells transfected with control or PRR11 siRNA was tested in a 84-cell cycle gene PCR array. Expression of 6 genes in the array was reduced by PRR11 siRNA transfection (FC < 0.5). Each data point represents the average of duplicate experiments. c MCF7 LTED and HCC1428 LTED cells transfected with control or PRR11 siRNA for 48 h were stained with propidium iodide and analyzed by flow cytometry. Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). d MCF7 LTED and HCC1428 LTED cells transduced with shRNA targeting the 3′ UTR of PRR11 and either pLX304-GFP or pLX304-PRR11 were stained with propidium iodide and analyzed by flow cytometry. Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). e MDA-MB-134VI and MDA-MB-175VII cells transduced with pLX302-LacZ or pLX302-PRR11 were grown in estrogen (E2)-deprived condition for 4 days. Cells were stained with propidium iodide and analyzed by flow cytometry. Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). f HCC1428 LTED, MCF7 LTED, MCF7 TamR, and MCF7 FulvR cells were transfected with control or PRR11 siRNA for 48 h. Cell lysates were subjected to immunoblot analysis. g MDA-MB-134VI and MDA-MB-175VII cells transduced with pLX302-LacZ or pLX302-PRR11 were grown ± 1 nM E2 for 4 days. Cell lysates were subjected to immunoblot analysis. Source data are provided as a Source data file.To inquire further into the mode of action of PRR11 in cell proliferation, we determined the expression of 84 cell cycle-associated genes using a PCR array. Six genes were reduced upon transfection of PRR11 siRNA into MCF7 LTED cells (SKP2, CDKN1A, CCNB2, CCNA2, CKS2, and CCNB1; FC < 0.5; Fig. 4b). With the exception of SKP2 and CDKN1A, expression of each of these genes was significantly elevated in ER+ breast cancers with PRR11 copy number gain/amplification in the METABRIC and TCGA dataset (Supplementary Fig. 4c, d). PRR11 ablation arrested cell cycle progression and inhibited proliferation in MCF7 LTED, MCF7 TamR, MCF7 FulvR, and HCC1428 LTED cells (Fig. 4c and Supplementary Fig. 4e) and these were rescued by re-expression of PRR11 (Fig. 4d). Conversely, PRR11 overexpression resulted in a marked increase in cells in S phase in estrogen-deprived MDA-MB-134VI and MDA-MB-175VIII cells (Fig. 4e). Consistent with a cytostatic effect, PRR11 knockdown reduced RB phosphorylation levels (Fig. 4f). Even in absence of estrogen, ectopic expression of PRR11 was capable to induce RB phosphorylation (Fig. 4g). Finally, ER transcriptional activity measured with an estrogen response element (ERE)-luciferase reporter was not affected by siRNA-mediated PRR11 ablation in parental and LTED MCF7 and HCC1428 cells (Supplementary Fig. 4f), suggesting that PRR11 confers resistance to estrogen suppression via growth-promoting signaling pathways independent of ER. PRR11 ablation did not suppress proliferation in PRR11-overexpressing triple negative and HER2+ breast cancer cells (Supplementary Fig. 4g, h).PRR11 reduces p85 homodimers and enhances PI3K activationGene set signature analysis shown in Fig. 4a also revealed that PIK3CA and IGF1 signaling pathways, both reported to be associated with endocrine resistance9,32, were significantly enriched in PRR11-high tumors. In line with this association, PRR11 knockdown resulted in a reduction in phosphorylated AKT (p-AKT) and p110α protein in LTED, TamR, and FulvR cells (Fig. 5a). AKT inactivates GSK3β through the phosphorylation, which in turn, promotes the stabilization of cyclin D1 protein33,34. Indeed, PRR11 knockdown resulted in an inactivation of GSK3β (Fig. 5a). PIK3CA mRNA levels were not affected by PRR11 ablation, suggesting that PIK3CA transcription is not regulated by PRR11 (Supplementary Fig. 5a). A similar reduction in p-AKT by PRR11 ablation were observed in MCF7 cells whose PIK3CA E545K allele had been corrected to a wild type sequence by somatic cell gene targeting, implying that the effect of PRR11 in PI3K activity is not limited to cells with PIK3CA mutations (Supplementary Fig. 5b–d). Conversely, overexpression of PRR11 resulted in an increase of p-AKT and p-GSK3β levels in MDA-MB-175VII and MDA-MB-134VI cells (Fig. 5b). To determine whether PRR11 affects phosphatidylinositol-3,4,5-trisphosphate (PIP3) formation, we utilized live cell imaging with a GFP-based biosensor fused with the PH domain of AKT. Detection of this biosensor at the plasma membrane is a surrogate of PI3K-induced PIP3 formation. Indeed, PRR11 silencing by siRNA resulted in a decrease of GFP-biosensor signals at the plasma membrane and membrane ruffling in MCF7 LTED cells (Supplementary Movies 1 and 2; control siRNA and siRNA targeting PRR11, respectively).Fig. 5PRR11 overexpression reduces p85 homodimers and enhances ligand-induced PI3K activation.a Lysates of MCF7 LTED, FulvR, TamR, and HCC1428 LTED cells transfected with PRR11 or control siRNA for 48 h were subjected to immunoblot analysis. b Lysates of MDA-MB-175VII and MDA-MB-134VI cells transduced with pLX302-LacZ or pLX302-PRR11 were subjected to immunoblot analysis. c MCF7 parental, MCF7 LTED, HCC1428 parental, and HCC1428 LTED cell lysates were immunoprecipitated with PRR11 or IgG antibodies. Immune complexes were then subjected to immunoblot analysis. d MCF7 LTED cells and HEK293 cells transduced with pLenti7.3-PIK3R1-Flag and pLX302-PRR11-V5 were subjected to proximity ligation assay (PLA) with PRR11, p85α, V5 and Flag antibodies. e Schema of p85 monomers associating via a SH3-PR domain intermolecular interaction, potentially disrupted by PRR11 overexpression. f, g HEK293 cells were co-transduced with pLenti7.3-PIK3R1-Flag, -PIK3R1-HA, and pLX302-PRR11-V5 (0, 0.25, 0.5, 1 µg; f). MCF7 LTED cells that had been stably transduced with pIND-PIK3R1-HA and pLenti6.3-PIK3R1-Flag were transfected with PRR11 siRNA (0, 1, 5, 25 pM) for 48 h in presence of 2 µg/mL doxycycline (g). Cell lysates were precipitated with HA or Flag antibodies and then subjected to immunoblot analysis. h, i MCF7 LTED cells transfected with control or PRR11 siRNA (h) and MDA-MB-175VII cells transduced with pLX302-LacZ or pLX302-PRR11 (i) were serum-starved for 24 h and then treated with 100 nM insulin for 10 min. Cell lysates were prepared and immunoprecipitated with a p110α antibody or IgG. Antibody pulldowns were then subjected to immunoblot analysis. j MDA-MB-175VII cells transduced with pLX302-LacZ or pLX302-PRR11 were serum-starved for 24 h and then treated with 100 nM insulin for 10 min. MCF7 LTED and HCC1428 LTED cells transfected with control or PRR11 siRNA were treated with insulin in the same way. Cell lysates were subjected to immunoblot analysis. k HEK293 cells were co-transduced with pLenti7.3-PIK3R1-Flag and either pLX302-PRR11 wild type (WT) or pLX302-PRR11 ∆PR, a mutant lacking the PR motif. Lysates were prepared and immunoprecipitated with a Flag antibody; immune complexes were then subjected to immunoblot analysis. Source data are provided as a Source data file.Proline-rich (PR) motifs bind to src homology 3 (SH3) domains, which are critical for the assembly of signaling complexes involved in aberrant cell proliferation35. Of potential relevance to PRR11, the p85 regulatory subunit of PI3K contains a SH3 domain and stabilizes the p110α catalytic subunit of PI3K36. Thus, to explore if PRR11 would interact with the PI3K pathway, we examined whether PRR11 associates with p85 via its PR motif. In MCF7 and HCC1428 cells, PRR11 was physically associated with p85α, as measured by both co-immunoprecipitation followed by immunoblot analysis and proximity ligation assay (PLA; Fig. 5c, d). This association between PRR11 and p85α was confirmed in HEK293 cells transfected with exogenous PRR11 and Flag-tagged p85α (Fig. 5d and Supplementary Fig. 6a).The intermolecular interaction between the SH3 and PR domains in p85 monomers mediates p85 homodimerization37,38. The p85 homodimer contains four SH2 domains and, as a consequence, outcompetes p85/p110 heterodimers for binding to phosphorylated Tyr residues in insulin receptor substrate 1 (IRS1), thus inhibiting insulin/IGF-stimulated PI3K activity39. Therefore, we hypothesized that the association of surplus PRR11 with p85α would impair the formation of p85 homodimers (Fig. 5e). This would potentially enhance both the association of heterodimeric p85/p110 with IRS1 at the plasma membrane and PI3K activation stimulated by insulin/IGFs. To test this hypothesis, we co-transduced PIK3R1 (p85α) tagged with either Flag or human influenza hemagglutinin (HA) into HEK293 cells. Ectopic expression of PRR11 impaired the association between p85α-Flag and p85α-HA as measured by HA and Flag precipitation followed by Flag and HA immunoblot, respectively (Fig. 5f), and by PLA (Supplementary Fig. 6b), suggesting PRR11 inhibits p85 homodimerization. In MCF7 LTED cells stably expressing both PIK3R1-Flag and doxycycline-inducible PIK3R1-HA, knockdown of PRR11 resulted in an increase of the p85 homodimers (Fig. 5g and Supplementary Fig. 6c). PRR11 silencing also reduced insulin-mediated association of p110α with IRS1 and insulin/IGF-stimulated p-AKT in MCF7 LTED and HCC1428 LTED cells (Fig. 5h, j and Supplementary Fig. 6d). Conversely, PRR11 overexpression enhanced the p110α-IRS1 association and p-AKT in MDA-MB-175VII cells (Fig. 5i, j).To determine whether the PR domain in PRR11 is crucial to the interaction of PRR11 with p85, we generated PRR11 mutant (PRR11-ΔPR) lacking the PR motif as revealed by Motif Scan, an in silico motif prediction analysis (Supplementary Fig. 7a, b). We co-transfected Flag-tagged PIK3R1 with either V5-PRR11-WT or V5-PRR11-ΔPR into HEK293 cells. Immunoprecipitation with a Flag antibody revealed that deletion of the PR motif in PRR11 reduced its interaction with p85α compared to PRR11 WT (Fig. 5k), suggesting that the PR motif of PRR11 mediates the association with p85α. Moreover, PRR11-ΔPR did not induce p-AKT levels to the same degree as PRR11 wild type (WT) in MCF7 LTED and HEK293 cells stably transduced with PRR11 3’ UTR shRNA (Supplementary Fig. 7c). Together, these data suggest that the PR motif of PRR11 promotes association with p85α, leading to reduced p85 homodimerization and enhanced PI3K activation.PRR11 amplification is associated with PI3K activationWe next computed signature scores of gene sets associated with the insulin/PI3K pathway in ER+/HER2− breast tumors in METABRIC. Signature scores of these gene sets were significantly higher in tumors harboring PRR11 gain or amplification compared to those with PRR11 deletion or diploid tumors (Fig. 6a, b). Similarly, hallmark gene sets associated with the PI3K/AKT pathway, such as “PI3K_AKT_MTOR_SIGNALING” and “MTORC1_SIGNALING”, were enriched in METABRIC ER+/HER2− breast tumors with PRR11 gain or amplification compared to PRR11 deleted/diploid tumors (Fig. 6c). We then computed the connectivity map (CMap) scores40 with the list of genes significantly upregulated in ER+/HER2− breast cancers with PRR11 gain or amplification vs. diploid/deletion in METABRIC, and in ER+ breast cancers with PRR11 amplification vs. no amplification in TCGA (Top 150 genes, FDR < 0.05). PI3K inhibitors and AKT signaling loss-of-function (LOF) were found as perturbation classes with < −95 connectivity score (tau) in MCF7 cells (Fig. 6d), representing an opposite connectivity between perturbation and gene set. This suggests that genes overexpressed in PRR11 amplified ER+ breast cancers can be downregulated by perturbations that inhibit PI3K/AKT. Of note, CDK4/6 inhibitors exhibited similar opposite connectivity, suggesting these agents may also be effective against PRR11-amplified cancers.Fig. 6PRR11 amplification is associated with hyperactivation of the PI3K pathway in ER+ breast cancers.a, b Signature score of the PI3K gene set in ER+/HER2−/PIK3CA wild type breast cancers in METABRIC plotted as a function of PRR11 copy number (a: n = 503 and 135 for deletion/diploid and gain/amplification group, respectively). Signature score of IGF1 gene set in ER+/HER2- breast cancers in METABRIC plotted as a function of PRR11 copy number (b: n = 1058 and 199 for deletion/diploid and gain/amplification group, respectively). Data represent the mean ± SD (two-tailed unpaired t-tests). c GSEA of mRNA expression data from ER+/HER2- tumors in METABRIC (PRR11 gain/amplification vs deletion/diploid); analyses show enrichment in PI3K_AKT_MTOR_SIG and MTORC1_SIG signatures. d Connectivity scores (tau) were computed using the connectivity map (CMap) with genes significantly upregulated in ER+ (TCGA) breast cancers harboring PRR11 amplification vs. no amplification and ER+/HER2− (METABRIC) breast cancers with PRR11 gain/amplification vs. PRR11 deletion/diploid. Connectivity score of 44 perturbation classes out of 171 are highlighted (tau < –95). e
PIK3CA mutation frequency in ER+/HER2− (METABRIC:
n = 1398) and ER+ (TCGA: n = 594) breast cancers plotted as a function of PRR11 copy number alterations vs. no alterations (two-tailed Fisher’s exact test). Source data are provided as a Source data file.Finally, in the METABRIC and TCGA datasets, PRR11 amplification and PIK3CA mutations were mutually exclusive of each other in ER+ breast cancers. This mutual exclusivity also supports the notion that PRR11 amplification is functionally linked to aberrant activation of PI3K (Fig. 6e).PRR11-amplified breast cancer cells rely on the PI3K pathwayThe data shown so far suggest that a potential oncogenic role of PRR11 depends on activation of PI3K and, as such, PI3K inhibitors would be effective against PRR11-overexpressing breast cancer cells. To explore this, we first utilized MCF10A cells that require EGF and insulin to propagate. PIK3CA knockdown significantly inhibited the proliferation of MCF10A cells grown in media containing EGF/insulin or insulin alone (Supplementary Fig. 8a, b). PIK3CA knockdown abolished growth promoted by PRR11 overexpression, suggesting that PRR11-mediated cell growth requires PIK3CA. Consistent with these data, the PIK3CA dependence score of 57 breast cancer cell lines significantly correlated with PRR11 copy number in DEMETER2, a combined large-scale RNAi screening dataset41 (Fig. 7a). In the PRISM repurposing 19Q3 dataset30, PRR11 copy number of 27 breast cancer cell lines significantly correlated with sensitivity to the PI3K inhibitors pictilisib and taselisib (Fig. 7b). Furthermore, breast cancer cell lines with PRR11 amplification displayed significantly higher sensitivity to the PI3K inhibitor pictilisib compared to cell lines without PRR11 amplification in the LINCS MGH/Sanger dataset of Drug/Cell-line Browser (DCB42; Fig. 7c). It was previously shown that ectopic expression of mutant PIK3CA sensitizes these cells to PI3K inhibitors43. Likewise, PRR11 overexpression sensitized MCF10A and MDA-MB-134VI cells to PI3K inhibitors (Fig. 7d, e and Supplementary Fig. 8c, d). Together, these suggest that PRR11-overexpressing cells rely on PIK3CA and, as a result, are highly sensitive to PI3K inhibitors.Fig. 7PRR11-amplified breast cancer cells are dependent on the PI3K pathway.a
PIK3CA dependency scores of 57 breast cancer cell lines were plotted against PRR11 copy number (DEMETER2 V5 dataset; Pearson correlation). b Sensitivity of 27 breast cancer cell lines to pictilisib and taselisib was plotted against PRR11 copy number (PRISM Repurposing 19Q3 dataset; Pearson correlation). Y-axis shows the log 2 cell fraction as per the relative barcode abundance following drug treatment. c Pictilisib sensitivity score of 26 breast cancer cell lines in the LINCS MGH/Sanger dataset (n = 22 and 4 for non-amplification and amplification group, respectively). Data represent the mean ± SD (two-tailed unpaired t-tests). d, e Alpelisib (d) and taselisib (e) GR50 of MDA-MB-134VI and MDA-MB-175VII cells transduced with pLX302-LacZ or pLX302-PRR11 were calculated using the GR metrics calculator. Cell numbers counted at day 0 and day 6 were used as the input data. Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). f Low density monolayers of MDA-MB-134VI pLX302-LacZ and -PRR11 cells were grown in estrogen (E2)-free medium. After 14 days, cell monolayers were stained with crystal violet. Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). g, h Low density monolayers of MDA-MB-175VII and MDA-MB-134VI cells transduced with pLX302-PRR11 or pLX302-LacZ were treated ± 1 µM alpelisib (g) or ± 1 µM taselisib (h) in estrogen-free medium. After 14 days, cell monolayers were stained with crystal violet. Data represent the mean ± SD of three replicates (two-tailed unpaired t-tests). Source data are provided as a Source data file.We finally tested whether genetic or pharmacological inhibition of PI3K would overcome PRR11-mediated resistance to estrogen suppression. PIK3CA knockdown with siRNA significantly abrogated estrogen-independent growth promoted by PRR11 overexpression in MDA-MB-134VI cells (Fig. 7f and Supplementary Fig. 8e). In addition, treatment with the PI3K inhibitors alpelisib and taselisib abolished estrogen-independent growth of MDA-MB-175VII and MDA-MB-134VI cells stably transduced with PRR11 (Fig. 7g, h). Consistently, PI3K inhibitors led to complete blockade of cyclin D1 protein levels that were induced by PRR11 overexpression (Supplementary Fig. 8f). These data suggest that PRR11-mediated escape from estrogen suppression can be blocked by PI3K inhibition.DiscussionThe 17q23 locus is amplified in ≈7% of breast cancers and has been suggested as a molecular subgroup through a clustering analysis of joint copy number and gene expression15. This cluster (IntClust1), which is predominantly composed of highly proliferative ER+ luminal B breast cancers, exhibits a poor prognosis and high genomic instability. A recent report showed that IntClust1 is one of four clusters associated with a high risk of distant relapse of ER+ breast cancers in METABRIC16. Initial studies aimed at identifying potential oncogenes in 17q23 focused on genes in this locus that are both amplified and overexpressed. RPS6KB1 and TBX2 were first proposed as putative candidates following extensive mapping of the amplicon in breast tumors and breast cancer cell lines44. Subsequent comprehensive analysis of copy number and gene expression predicted MUL, APPBP2, and TRAP240 as potential oncogenes27. However, functional studies of these alterations have been incomplete. More recently, WIP1 (PPM1D) and MIR21, also located in 17q23, were shown to cause resistance to anti-HER2 therapy45. However, WIP1 and MIR21 were not associated with a worse outcome of patients with ER+ breast cancer treated with endocrine therapy in the Kaplan–Meier Plotter dataset (data not shown). In this report, our comprehensive analysis with survival data of patients with ER+ breast cancer suggests that PRR11 is strongly associated with breast cancer progression.Aberrant activation of PI3K/AKT/β-catenin has been suggested as a mechanism by which PRR11 promotes cell proliferation in ovarian and hepatocellular carcinoma19,46, but the molecular basis for a potential role of PRR11 in cancer virulence is unclear. We show herein that the PR domain of PRR11 physically binds the SH3 domain of p85, thus inhibiting p85 homodimer formation. Homodimers or monomers of p85 form a sequestration complex with IRS1, thus competing with p85-p110 dimers for binding to IRS1 at the plasma membrane and attenuating insulin/IGF stimulated PI3K activity39. Hence, an excess of PRR11 would favor p85-p110 dimers and be permissive of p110 binding to IRS1, retention of p110 at the plasma membrane, and enhanced PIP3 formation. Homodimers of p85 can also inhibit PI3K signaling via direct association and stabilization of PTEN37, but we did not observe that knockdown of PRR11 alters PTEN protein levels (Fig. 5a).We and others have identified hyperactivation of the PI3K pathway as a mechanism for ER+ breast cancers to bypass hormone dependence9,47. In this study, we found PRR11 overexpression was strongly associated with resistance to estrogen suppression in primary ER+ postmenopausal tumors treated with letrozole. Other analyses shown herein also support a functional connection between PRR11 and the PI3K pathway. For instance, there was a significant correlation between PI3K/IGF1 gene set signature scores and high PRR11 mRNA levels in clinical cohorts of ER+ breast tumors treated with neoadjuvant letrozole. Further, PRR11 amplifications were mutually exclusive of PIK3CA mutations in genomic breast cancer databases. Next, overexpression of PRR11 in breast epithelial and breast cancer cells stimulated growth and this effect was abolished by RNA interference of p110α and by treatment with PI3K inhibitors. Finally, PRR11-amplified breast cancer cell lines exhibited higher sensitivity to PI3K inhibitors compared to cells that do not harbor PRR11 amplification. Treatment of patients with advanced ER+ breast cancer with PI3Kα inhibitors in combination with antiestrogens has significantly improved progression-free survival13,48. In these studies, PIK3CA mutations in tumors predicted clinical benefit from the PI3K inhibitor. However, some patients in these trials with wild type PIK3CA also benefitted clinically, suggesting that other alterations resulting in PI3K pathway dependence also respond to PI3K inhibitors and, as such, should be explored as biomarkers for enrollment of patients into trials with this class of drugs. We posit PRR11 amplification may also serve as a predictive biomarker of sensitivity to PI3Kα inhibitors, particularly in PIK3CA-wild type tumors.In summary, we identified PRR11, a gene in the 17q23 amplicon, as a potential driver of antiestrogen resistance in ER+ breast cancer. Integrative analyses, including clinical data from patients with ER+ breast cancer treated with an aromatase inhibitor strongly implicated a role for PRR11 in endocrine resistance. PRR11 blocks p85 homodimerization and sensitizes to ligand-induced PI3K activation, suggesting that PRR11 amplification confers resistance to estrogen deprivation through hyperactivation of the PI3K pathway. Finally, we propose that, in conjunction with endocrine therapy, PI3K may be an actionable target in ER+ breast cancers harboring PRR11 amplification.MethodsCell linesThe MCF7 (ATCC® HTB-22), HCC1428 (ATCC® CRL-2327), MDA-MB-175-VII (ATCC® HTB-25) and MDA-MB-134VI (ATCC® HTB-23), BT-474 (ATCC® HTB-20) human breast cancer cells, HEK293 (ATCC® CRL-1573) human embryonic kidney cells and MCF10A (ATCC® CRL-10317) breast epithelial cells were purchased from ATCC in 2018 or 2019. The 293FT (R70007) cells were purchased from Invitrogen in 2016. HCC38 and MDA-MB-231 human breast cancer cells were kindly provided by Dr. Jennifer A. Pietenpol at Vanderbilt University Medical Center. Cell lines were authenticated by the short-tandem repeat (STR) method and tested for Mycoplasma contamination. MCF7, MDA-MB-175-VII, MDA-MB-134VI, BT-474, MDA-MB-231, HEK293, and 293FT cells were maintained in DMEM/10% fetal bovine serum (FBS)/1% Antibiotic-Antimycotic (AA). HCC38 and HCC1428 cells were maintained in RPMI 1640/10% FBS/1% AA. MCF10A cells were maintained in DMEM/F12 supplemented with 5% horse serum, 20 ng/mL EGF, 10 µg/mL insulin, 0.5 µg/mL hydrocortisone, 0.1 µg/mL cholera toxin, and 1% AA. Long-term estrogen-deprived (LTED) cell lines have been described previously9. To generate fulvestrant-resistant MCF7 cells, cells were cultured in the presence of increasing concentrations of fulvestrant starting at 50 nM. Cells were deemed resistant when they grew as parental cells in 1 µM fulvestrant. To generate tamoxifen-resistant MCF7 and HCC1428 cells, cells were cultured in the presence of increasing concentrations of tamoxifen starting at 500 nM. Cells were deemed resistant when they grew as parental cells in 2 µM tamoxifen. For experiments outlined here, resistant cells were removed from each drug for at least 24 h prior to treatment. MCF7 cells (29C-1) whose PIK3CA E545K allele has been corrected to a wild type sequence by somatic cell gene targeting were kindly provided by Dr. Ben Ho Park at Vanderbilt University Medical Center.Xenograft studiesAll mice were maintained according to the guidelines of the Care and Use of Laboratory Animals published by the US National Institutes of Health and the Institutional Animal Care and Use Committee. All procedures were approved by the Institutional Ethics Review Committee of the University of Texas Southwestern Medical Center. Eight-week old female ovariectomized athymic mice (Hsd:Athymic Nude-Foxn1nu, Envigo) were implanted with a 14-day release 17β-estradiol pellet (0.17 mg, Innovative Research of America). The following day, 107 MCF7 cells stably expressing a doxycycline-inducible control or PRR11 shRNA were suspended in IMEM and growth factor reduced Matrigel (BD Biosciences) at a 1:1 ratio and then injected subcutaneously into the right flank of each mouse. Approximately 4 weeks later, mice bearing tumors measuring ≥250 mm3 were randomized to treatment with vehicle (0.9% NaCl) or doxycycline (10 mg/kg/daily, by intraperitoneal injection). Tumor diameters were measured with calipers weekly and tumor volume was calculated with the formula: volume = width2 × length/2. After 4 weeks, tumors were harvested and homogenized using TissueLyser II (Qiagen) for subsequent immunoblot analysis.Gene set signature analysesSingle-sample gene set enrichment for 125 previously published breast cancer-related gene expression signatures were computed as described previously17; signatures with false discovery rate (FDR) < 0.01 were considered as differentially activated pathways between PRR11 high and low tumors. Gene set enrichment analysis (GSEA) was conducted with the javaGSEA interface downloaded from Broad Institute (http://software.broadinstitute.org/gsea/index.jsp). The h.all.v6.2.ymbols.gmt [Hallmarks] was used as gene sets database49.Immunohistochemistry (IHC)Formalin-fixed paraffin-embedded (FFPE) 4-µm tumor sections were deparaffinized. Antigen retrieval was performed with a citrate buffer (pH 6) in a decloaking chamber (Biocare). Endogen peroxidase was blocked with 3% H2O2 and protein block (Agilent). Tumor sections were next incubated with a PRR11 antibody (Novus, NBP1-83784; dilution 1:200) overnight at 4 °C. Envision (Agilent) was used for visualization with DAB as the chromogen (Agilent); hematoxylin was applied as the counterstain. Whole sections were digitally acquired using an AxioScan Z1 slide scanner (Carl Zeiss) at ×20. Automated semi-quantitative scoring was performed using QuPath software50. Color deconvolution stains were set form a representative area. A cell segmentation was determined on hematoxylin OD. An object classification was trained to differentiate tumor from stroma. Percentage of PRR11+ cells was calculated with the cell detection algorithm according to the cytoplasm DAB OD mean. Each selected region was visually assessed for correct performance of the quantification algorithm excluding areas of non-invasive tumor.Fluorescence in situ hybridizationFour-μm tissue sections were mounted on charged slides and hybridized overnight with a PRR11 FISH probe (EMPIRE GENOMICS) and a centromere 17 control probe (EMPIRE GENOMICS); FISH was performed as described previously14. Twenty to sixty tumor cell nuclei from random areas were individually evaluated with the 100× oil immersion objective by counting green PRR11 and orange centromere 17 (CEN17) signals. A PRR11/CEN17 ratio ≥2.0 was considered as PRR11 amplification.siRNA transfectionCells seeded in 6-well plates or 60-mm dishes were transfected with 20 pmole or 40 pmole of siRNAs, respectively, using Lipofectamine RNAiMAX reagent (Invitrogen) as per the manufacturer’s instructions. Control siRNA (4390843), PRR11 siRNAs (siRNA#1: 4392420-S31473, siRNA#2: 4392420-S31475), and PIK3CA siRNA (4390824-S10520) were purchased from ThermoFisher Scientific.PlasmidsHuman PRR11 open reading frame (ORF) in pENRT221 from The UltimateTM ORF Lite human cDNA collection (Life Technologies) was cloned into pLX302 or pINDUCER20 (pIND) using GatewayTM LR ClonaseTM II Enzyme Mix (ThermoFisher). To generate pLX302-PRR11-ΔPR, the PR motif was predicted using Motif Scan (https://myhits.isb-sib.ch/cgi-bin/motif_scan) and then removed from the pLX302-PRR11-WT construct using Q5-Site-Directed Mutagenesis Kit (NEW ENGLAND BioLabs). pLX304-PRR11-ΔPR was generated from pLX302-PRR11-ΔPR through BP/LR cloning. SMART doxycycline-inducible PRR11 shRNA and SMART PRR11 shRNA targeting 3’UTR was purchased from Dharmacon (V3SH11252 and V3SH11240, respectively). pLenti7.3-PIK3R1-Flag and pLenti7.3-PIK3R1-HA were kindly provided by Dr. Gordon Mills at Oregon Health Sciences University. To generate pLenti6.3-PIK3R1-Flag and pIND-PIK3R1-HA, PIK3R1-Flag and PIK3R1-HA were cloned into pDORNTM221 using GatewayTM BP ClonaseTM II Enzyme mix (ThermoFisher). Next, pDONR221-PIK3R1-Flag and pDONR221-PIK3R1-HA were cloned into pLenti6.3/V5-DSETTM GatewayTM vector (ThermoFisher) and pINDUCER20, respectively, using GatewayTM LR ClonaseTM II Enzyme Mix. pLX304-PRR11(HsCD00444919), -BRIP1 (HsCD00440250), -SMARCD2 (HsCD00440288) and -TACO1 (HsCD00442382) were purchased from DNASU.Lentiviral transductionTo generate stably transduced cell lines, 1 µg of the plasmids were co-transfected with 0.75 µg of psPAX2 (2nd generation lentiviral packaging plasmid) and 0.5 µg of pMD2.G (VSV-G envelope expressing plasmid) into 293FT cells using Lipofectamine2000 (ThermoFisher). Cell medium was changed to fresh medium 24 h post-transfection, and cells were collected 48 h later. Lentiviral particles of sgRNA editing PRR11 were purchased from Dharmacon (VSGH11937-247492131). Virus-containing medium was applied to target cells with 8 µg/mL polybrene. Puromycin were used for selection of cells transduced with SMART doxycycline-inducible PRR11 shRNA, SMART PRR11 shRNA targeting 3’ UTR, pLX302-PRR11, pLX302-LacZ, control sgRNA and PRR11 sgRNA. Blasticidin were used for selection of cells transduced with pLX304-GFP, -PRR11, -BRIP1, -SMARCD2, -TACO1 or pLenti6.3-PIK3R1-Flag. MCF7 LTED cells transduced with pIND-PIK3R1-HA were selected with G418 sulfate.Cell proliferation assaysAfter 24 h from transfection with control siRNA or PRR11 siRNAs, HCC1428, and MCF7 (LTED, TamR, and FulvR) cells were seeded in 6-well plates. Cells were trypsinized and then counted every 3 days for 6 days. Number of cells were counted with Z2 coulter counter analyzer (Beckman coulter).RT-qPCRRNA was extracted from cells using Maxwell RSC simplyRNA Cells Kit (Promega) according to the manufacturer’s protocol. cDNA was synthesized using the iSCRIPT cDNA synthesis Kit (Bio-Rad) and then subjected to PCR with PowerUpTM SYBRTM Green Master Mix (ThermoFisher), PIK3CA (Qiagen), or GAPDH (Qiagen) primers using QuantStudio3 Real-Time PCR System (ThermoFisher).Dual luciferase assayCells (1 × 104/well) were seeded in 96-well plates in triplicate. Next day, cells were transfected with pGLB-MERE and pCMV-Renilla, each with either control or PRR11 siRNA. Twenty-four h post-transfection, cells were switched to estrogen-free medium (IMEM/5% charcoal stripped FBS) for 24 h, followed by treatment with 1 nM estradiol for 24 h. Renilla and firefly luciferase activities were measured using Dual-Luciferase® Reporter Assay System according to manufacturer’s protocol (Promega).Clonogenic assaysMCF7 and HCC1428 (LTED, TamR, and FulvR) cells were seed in 12-well plate and then transfected with control siRNA or PRR11 siRNA. Cells were grown ± 1 nM estradiol, ± 1 µM tamoxifen or ± 1 µM fulvestrant for 10 days. MDA-MB-175VII cells transduced with pLX304-PRR11, -BRIP1, -SMARCD2 and -TACO1 were grown in absence of estradiol for 14 days. MDA-MB-175VII and MDA-MB-134VI cells transduced with pLX302-PRR11 or -LacZ were grown in absence of estradiol ± 1 µM taselisib or ± 1 µM alpelisib for 14 days. MCF10A pLX302-PRR11 or -LacZ cells were grown ± 1 µM taselisib or ± 1 µM alpelisib for 10 days. Fresh media containing drugs were replaced every 4 days. Cells were fixed with cold-iced methanol for 10 min at −20 °C and then stained with 0.5% crystal violet solution for 10 min at room temperature. Stained monolayers were imaged using Gelcount mammalian cell colony counter (Oxford Optronix). For quantification, cell monolayers were dissolved with 10% acetic acid for 15 min at room temperature. Supernatants were transferred into 96-well plates for measuring absorbance at 560 nm using GloMax Discover microplate reader (Promega).Immunoblot analysisCells were lysed with RIPA buffer (ThermoFisher) containing protease inhibitors (Protease Inhibitor Cocktail, Roche) and phosphatase inhibitors (PhosSTOP, Roche), scraped and then incubated on ice for 30 min. Supernatants were collected after centrifugation at 14,000 rpm for 10 min. Protein concentration in cell lysates was measured with the Pierce BCA Protein Assay kit (ThermoFisher). Thirty µg of protein were subjected to SDS-PAGE followed by transfer to nitrocellulose membrane for immunoblot analysis. Information of antibodies used in this study is included as Supplementary Table 6.Cell cycle analysisAfter 48 h from transfection with control siRNA or PRR11 siRNAs, cells were trypsinized and then fixed with 70% ethanol for 3 h at −20 °C. Cells were washed x3 with PBS and then resuspended in PBS containing 100 µg/mL RNase A and 40 µg/mL propidium iodide for 10 min at room temperature. Stained cells were analyzed with BD LSRFORTESSA (BD Biosciences) followed by profiling of cell cycle phases with the Dean-Jett-Fox model of FlowJo (ver.10). Gating strategy is shown in Supplementary Fig. 4i.PCR arrayRNA was extracted from MCF7 LTED cells transfected with control siRNA or PRR11 siRNA for 48 h using Maxwell® RSC simplyRNA Cells Kit (Promega). Complementary DNA (cDNA) was synthesized using RT2 First Strand Kit according to the manufacturer’s protocol and then subjected to real-time PCR using the RT2 ProfilerTM PCR Array Human Cell Cycle (Qiagen) in duplicate.Live cell imagingA plasmid encoding the GFP-based AKT-PH biosensor was designed as described in a previous report51. Briefly, the PH domain of AKT1 (residues 2-149) was fused with the COOH terminus of enhanced GFP. MCF7 LTED cells were transduced with pEGFP-AKT-PH and then sorted for GFP expression by flow cytometry. For three-dimensional imaging, we used a modified variant of the light-sheet microscope52. Briefly, this microscope illuminates the specimen from the epi-direction with an obliquely launched light-sheet. The beam is rapidly scanned, and subsequent fluorescence descanned, using a mirror galvanometer. For primary, secondary, and tertiary objectives, we used a high-NA silicone immersion objective (Nikon ×100 NA 1.35, 0.28–0.31 mm working distance), a high-NA air immersion objective (Nikon ×40 NA 0.95, 0.25–0.16 mm working distance), and a bespoke glass-tipped objective (AMS-AGY v1.0, NA 1.0, 0 mm working distance. https://andrewgyork.github.io/high_na_single_objective_lightsheet/), respectively. Images were acquired for each plane on a high-speed scientific CMOS camera (Hamamatsu Flash 4.0) using custom LabView-based software, sheared in the frequency-domain with Python, and rotated with an affine transform using either MATLAB or IMOD. No deconvolution was performed on the data shown. For each condition, 8 positions in a 35 mm dish were chosen at random, and each position encompassed multiple cells. Movies were acquired at a volumetric imaging rate of 0.1 Hz for 10 min.ImmunoprecipitationAll immunoprecipitation experiments were performed using the DynabeadsTM Protein G Immunoprecipitation Kit (ThermoFisher) according to the manufacturer’s protocol. Briefly, 10 µg of antibodies were pre-incubated with 1.5 mg Dynabeads for 10 min at room temperature and then incubated overnight at 4 °C with 1 mg of cell lysate. Precipitates were eluted from the magnetic bead-antibody-antigen complex using the elution buffer containing NuPAGE LDS sample buffer (ThermoFisher) and NuPAGE sample reducing agent (ThermoFisher) for 10 min at 70 °C. Eluted samples and 20 µg of input lysates were subjected to immunoblot analysis.Proximity ligation assayMCF7 LTED (5 × 104/well) cells were seeded in 8-well chamber slides (Lab-Tek, 177445) in triplicate. HEK293 cells (5 × 104) were seeded 24 h post-transfection with pLX302-PRR11-V5, pLenti7.3-PIK3R1-Flag, and pLenti7.3-PIK3R1-HA. PRR11 (Cat# LS-B15222, LS Bio). p85 (Cat# 05-212, Millipore), V5 (Cat# 13202, Cell Signaling Technology), HA (Cat# 3724, Cell Signaling Technology) and Flag (Cat# 8146, Cell Signaling Technology) antibodies were used for this assay. PLA was performed with Duolink In Situ Red Starter Kit Mouse/Rabbit (Sigma) according to the manufacturer’s protocol and then imaged with a DMi8 inverted microscope (Leica). The number of PLA foci was quantified by Duolink Image Tool software; 5 images per sample were analyzed.Statistics and reproducibilityPearson r correlation, hazard ratio and t-tests (Nonparametric tests) were performed with GraphPad Prism version 8 or Microsoft Excel 2016. Data were represented as mean ± SD. All experiments were conducted at least three times. A p value < 0.05 was considered statistically significant. A false discovery rate (FDR) was computed using the Benjamini–Hochberg procedure. R version 3.5.2 and R studio version 1.1.463 were used.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Movie 1Supplementary Movie 2Reporting Summary | nature communications | [
"Article"
] | [
"Breast cancer",
"Cancer therapeutic resistance"
] | 80% breast cancers estrogen receptor-positive depend on estrogen Therapies ER+ breast cancer inhibit ER signaling ER or abolishing estrogen production aromatase Adjuvant anti-ER therapies reduce risk recurrence ER+ 20% patients relapse metastatic mechanisms de novo acquired resistance to endocrine tumor DNA sequencing somatic alterations endocrine resistance discovered Mutations in ligand-binding ESR1 confer resistance to estrogen suppression Amplification growth factor receptors ERBB2 FGFR1 associated endocrine therapy CCND1 amplification in B tumors suggests causal role drug-resistant phenotype1 mutations in mitogen-activated protein kinase (MAPK) pathway ER transcriptional program in 20% ER+ breast cancers shorter response antiestrogen therapy8 studies hyperactivation phosphoinositide 3-kinase (PI3K)/AKT pathway endocrine PI3Kα inhibitor alpelisib with ER antagonist fulvestrant superior than fulvestrant in advanced ER+/PIK3CA mutant breast approval alpelisib + fulvestrantreported genomic profiling ER+ breast tumors after treatment 11q13.3 8p11.23 17q21-23 amplicons correlated with high proliferation marker Ki67 drug-induced estrogen suppression FGFR1 CCND1 amplification 8p11-12 11q13 associated with resistance letrozole high Ki67 score 17q23 amplicon associated with proliferative luminal B tumors high risk recurrence specific gene causal endocrine resistance uncovered study transcriptome analysis 58 ER+ breast cancers identified PRR11 protein-coding gene chromosome 17q22-23 overexpressed in tumors resistant estrogen suppression letrozole-sensitive PRR11 implicated poor outcome cancer molecular basis unclear hypothesized PRR11 amplification 17q23 amplicon promotes endocrine resistance ER+ breast cancer high PRR11 associated with estrogen-independent growth ER+ cancer cells interaction PRR11 p85 regulatory subunit PI3K reduces homodimerization p85 association p110α insulin receptor substrate 1 activation PI3KEctopic PRR11 estrogen-independent growth p110α PRR11-overexpressing cells sensitive PI3K inhibitors PRR11 amplification dependence PI3K signaling estrogen deprivation data suggest PI3K ERα ER+ breast cancers PRR11 amplification poor outcome ER+ breast antiestrogens transcriptome analysis RNA 58 ER+ breast cancers long letrozole 7.2 months PRR11 mRNA upregulated in resistant tumors resistance estrogen suppression defined preoperative endocrine prognostic index (PEPI) ≥ 4 cancer relapse after follow-up 5 years [log2 fold change > 1 false discovery rates) < 0.05 RNA-seq analysis 51 genes PRR11 overexpressed in resistant sensitive cancers (Log2 fold change 1.15 FDR = 0.004 p = PRR11-high cancers increased risk relapse [hazard ratio) = 3.753 95% 1.045–13.47 high PRR11 mRNA levels shorter relapse-free survival ER+/HER2− breast cancers treated endocrine therapy = 3.85 95% CI 1.95–7.59association not present in HER2+ triple-negative breast cancer interrogated association PRR11 expression response estrogen suppression in two studies ER+ breast cancers neoadjuvant aromatase inhibitor Giltnane Miller et al.22) high Ki67 index resistance estrogen suppression significant correlation between PRR11 mRNA-treatment high Ki67 levels (Fig. 1e PRR11 protein levels evaluated in tumor sections 175 ER+ breast cancers treated letrozole before surgery PRR11 tumors classified PRR11 ≤ 1% 1–15% (positive) >15% (high PRR11 levels higher in tumors poor response letrozole high Ki67 (Fig. 1g 1PRR11 poor clinical outcome ER+ breast cancers treated antiestrogens plot genes non-responding tumors change false discovery rates calculated DeSeq2 package 17q23 locus genes in ER+/HER2− breast cancers treated long-term letrozole (n = 51). Genes selected Atlas of Genetics Cytogenetics25Red bars indicate genes FDR < 0.05. Recurrence-free survival ER+ breast cancers treated letrozole low high PRR11 mRNA levels PRR11 FPKM cut-off (3.93) human protein atlas PRR11-high 43 -low 15 tumors Mantel–Cox model hazard ratio p value Relapse-free survival ER+/HER2− breast cancers treated endocrine low 104 or high = 98) PRR11 mRNA levels Kaplan–Meier Plotter HR p Kaplan–Meier Plotter Correlation on-treatment percent Ki67+ tumor cells PRR11 mRNA level tumors Giltnane et al. Miller et. PRR11 immunohistochemistry images ER+ breast tumors PRR11 levels response estrogen suppression letrozole trial NCT0065197614 Data mean ± SD (n = 91 25 39 sensitive intermediate resistant group Source data.PRR11 estrogen-responsive high PRR11 mRNA resistant tumors suppressed letrozole treatment 18 ER+ breast tumors aromatase inhibitor PRR11 mRNA not downregulated post-treatmentPRR11 not regulated by ligand-induced ER PRR11 mRNA levels correlate with estrogen response gene signatures in three cohorts tumors treated with aromatase inhibitor exogenous estrogen increase PRR11 mRNA levels in MCF7 HCC1428 cells 1e).PRR11 key gene endocrine cytogenetic band 17q22 or 17q23.2 at terminal region 17q22 close to 17q23 region PRR11 amplification 15.9% in Metastatic Breast Cancer (MBC) project 9.5% and 9.4% in METABRIC The Cancer Genome Atlas cohorts higher rate PRR11 amplification in metastatic ER+ metastatic ER+ breast cancers adjuvant treatment endocrine therapy alterations endocrine resistance PRR11 amplification mutually exclusive with mutations in ESR1 and NF1 antiestrogen resistance METABRIC cohort PRR11 copy number gain predicted shorter disease-free survival of patients ER+/HER2− breast cancer treated with antiestrogens PRR11 copy number alterations correlated with PRR11 mRNA expression56 breast cancer cell lines CCLE high PRR11 copy number correlated with PRR11 mRNA PRR11-amplified lines express higher levels PRR11 protein non-amplified 11 verified PRR11 amplification by fluorescence in situ hybridization letrozole-resistant breast cancer 2PRR11 key gene 17q23 endocrine resistance Frequency PRR11 amplification MBC project TCGA METABRIC ER+ breast cancers Oncoprint endocrine-resistant drivers metastatic ER+ tumors (n = 44). disease-free survival METABRIC ER+/HER2− breast cancer patients treated anti-hormone PRR11 copy number/amplification 163)/deletion 834) Venn diagram genes 17q23 on-treatment Ki67 levels (Pearson r > 0.4 p < 0.05) Source data gene 17q23 amplicon causal to endocrine resistance identified found 90 genes 17q23 amplicon 2) examined association genes on-treatment Ki67 levels three studies ER+ breast cancers treated neoadjuvant aromatase inhibitoranalysis revealed four 90 genes in 17q23 (PRR11 BRIP1 SMARCD2 TACO1) correlated with on-treatment Ki67 levels (Pearson r > 0.4 p < 0.05 Fig. 2d Tables 3–5) PRR11 only correlation with high Ki67 score high PRR11 mRNA levels correlated with shorter RFS in patients ER+/HER2− breast cancer endocrine Kaplan–Meier Plotter database Fig 2f). data suggest PRR11 key gene in 17q23 endocrine resistance genes 17q23 poor outcome breast PRR11 interrogated genome RNAi screening data MCF7 cells Project Achilles dataset 47 genes 17q23 screened PRR11 displayed lowest score high dependency cancer cells on PRR11 Fig 2g).PRR11 overexpression confers resistance to antiestrogensTo transduced PRR11 BRIP1 SMARCD2 TACO1 into MDA-MB-134VI MDA-MB-175VII cells 17q23 amplification (Fig. 3a). PRR11 promoted growth under estrogen deprivationemployed MCF7 LTED HCC1428 LTED PRR11 high copy PRR11 ablation abolished estrogen-independent growth. inhibitory effect rescued by re-expression PRR11 (Fig. 3d generated xenografts MCF7 cells doxycycline-inducible PRR11 in ovariectomized athymic mice Treatment doxycycline reduced PRR11 protein levels arrested growth MCF7 xenografts (Fig. 3f 3PRR11 overexpression resistance antiestrogens Lysates MDA-MB-134VI MDA-MB-175VII cells transduced with-PRR11 -BRIP1 -TACO1 -SMARCD2 immunoblot analysis Low density monolayers grown in estrogen-deprived condition After 2 weeks cell monolayers stained with crystal violet cell viability quantified Data replicates MCF7 HCC1428 LTED cells transfected with PRR11 siRNAs monolayers treated ± 1 nM estrogen for 10 days monolayers stained with crystal violetData mean ± SD three replicates t MCF7 HCC1428 cells transduced shRNA UTR PRR11 re-transduced pLX304-GFP-PRR11 Cell lysates immunoblot analysis Upper lower arrows indicate exogenous endogenous PRR11 Low density monolayers cells grown E2 10 days mean ± SD three replicates MCF7 cells doxycycline-inducible-PRR11 control injected dorsum athymic ovariectomized mice 14-day 17β-estradiol pellet After 4 weeks randomized 10 mg/kg doxycycline 4 weeks data point mean tumor volume mm3 ± SD mice group parenthesis Fulvestrant sensitivity ER+ breast cancer cell lines = 11 Y-axis sensitivity barcode abundance treatment Fulvestrant GR50 calculated GR metrics Cell numbers days 0 6 input dataMDA-MB-134VI-175VII cells transduced pLX302-LacZ pLX302-PRR11 MCF7 FulvR cells transfected control PRR11 siRNA Data mean ± SD three replicates t Source data PRISM 19Q3 PRR11-amplified ER+ breast cancer cell lines lower fulvestrant without PRR11 Transduction PRR11-MB-134VI-MB175VII cells attenuated growth inhibition fulvestrant PRR11 knockdown re-sensitized fulvestrant-resistant MCF7 cells fulvestrant similar result tamoxifen-resistant MCF7 HCC1428 cells PRR11 promotes resistance antiestrogens enhances ER-targeted therapies against breast cancer cells promotes proliferation ER+ breast cancer analyzed RNA-seq data Guerrero-Zotano et al 125 breast cancer-related gene expression Nine proliferation-associated signatures enriched in tumors high PRR11 mRNA expression gene sets “E2F_TARGETS” “G2M_CHECKPOINT” enriched in ER+/HER2− tumors PRR11 ER+. 4PRR11 overexpression enhances cancer proliferation Single sample gene set analysis 125 breast cancer signatures sets enriched PRR11 high low tumors (FDR < DNA MCF7 LTED cells transfected tested 84-cell gene PCR array Expression 6 genes reduced PRR11 transfection (FC < average duplicate experiments MCF7 LTED HCC1428 LTED cells transfected PRR11 48 h stained propidium iodide analyzed flow cytometry Data mean ± SD three replicates MCF7 HCC1428 cells transduced PRR11 pLX304-GFP-PRR11 stained propidium iodide analyzed cytometry SD MDA-MB-134VI MDA-MB-175VII cells transduced pLX302-PRR11 grown estrogen-deprived 4 days stained propidium iodide analyzed flow cytometry mean ± SD HCC1428 LTED MCF7 LTED TamR FulvR cells transfected PRR11 siRNA 48 h lysates immunoblot analysisMDA-MB-134VI-MB-175VII cells transduced with pLX302-LacZ grown ± 1 nM E2 4 days lysates subjected immunoblot analysis Source data PRR11 determined expression 84 cell cycle genes PCR Six genes reduced transfection PRR11 siRNA into MCF7 LTED cells (SKP2 CDKN1A CCNB2 CCNA2 CKS2 CCNB1 < 0.5 SKP2 CDKN1A expression elevated in ER+ breast cancers PRR11 gain METABRIC dataset PRR11 ablation arrested cell cycle progression inhibited proliferation in MCF7 LTED TamR FulvR HCC1428 LTED cells rescued by re-expression PRR11 PRR11 overexpression cells S phase estrogen-deprived MDA-MB-134VI MDA-MB-175VIII cells PRR11 reduced RB phosphorylation levels ectopic expression PRR11 RB phosphorylation ER transcriptional activity not affected by PRR11 ablation in parental LTED MCF7 HCC1428 cellsPRR11 resistance estrogen suppression growth ER ablation suppress proliferation in-overexpressing triple negative HER2+ breast cancer cells Fig. 4g reduces p85 homodimers enhances PI3K activationGene analysis 4a PIK3CA IGF1 signaling pathways associated endocrine enriched in PRR11-high tumors PRR11 knockdown AKT p110α protein in LTED TamR FulvR cells. AKT inactivates GSK3β promotes stabilization cyclin D1 PRR11 knockdown GSK3β PIK3CA mRNA levels not affected PRR11 ablation transcription not regulated by PRR11 similar reduction in p-AKT in MCF7 cells PIK3CA E545K allele corrected wild effect PI3K activity not limited PIK3CA mutations overexpression PRR11 p-AKT p-GSK3β levels in MDA-MB-175VII MDA-MB-134VI cells (Fig. PRR11 phosphatidylinositol-3,4,5-trisphosphate (PIP3) formation live cell imaging GFP-based biosensor domain AKTbiosensor plasma membrane PI3K PIP3 formation PRR11 silencing GFP-biosensor signals ruffling MCF7 LTED cells overexpression reduces p85 homodimers enhances PI3K activation Lysates MCF7 LTED FulvR TamR HCC1428 LTED cells transfected PRR11 48 h immunoblot analysis Lysates MDA-MB-175VII-MB-134VI cells transduced pLX302-LacZ pLX302-PRR11 MCF7 LTED HCC1428 LTED lysates immunoprecipitated PRR11 IgG antibodies immunoblot analysis MCF7 LTED HEK293 cells transduced-PIK3R1-Flag pLX302-PRR11-V5 ligation assay PRR11 p85α V5 Flag antibodies p85 monomers SH3-PR interaction disrupted PRR11 overexpressionHEK293 cells co-transduced with pLenti7.3-PIK3R1-Flag -PIK3R1-HA pLX302-PRR11-V5 μg MCF7 LTED cells transduced-PIK3R1-Flag transfected PRR11 siRNA 48 h 2 μg/mL doxycycline lysates precipitated HA Flag antibodies immunoblot analysis MCF7 cells MDA-MB-175VII cells pLX302-LacZ pLX302-PRR11 serum-starved 24 h treated 100 nM insulin 10 min lysates prepared immunoprecipitated p110α antibody IgG Antibody pulldowns immunoblot analysis MDA-MB-175VII cells serum-starved 24 h treated 100 nM insulin 10 min MCF7 HCC1428 LTED cells transfected PRR11 treated lysates immunoblot analysis HEK293 cells co-transduced with pLenti7.3-PIK3R1-Flag pLX302-PRR11 ∆PR Lysates prepared immunoprecipitated Flag antibody immune complexes immunoblot analysis Source dataProline motifs bind src 3 (SH3) domains critical signaling aberrant cell PRR11 p85 regulatory subunit PI3K contains SH3 domain stabilizes p110α subunit PI3K examined p85 PR motif MCF7 HCC1428 cells PRR11 associated with p85α measured co-immunoprecipitation immunoblot analysis proximity ligation assay association confirmed in HEK293 cells transfected PRR11 Flag-tagged p85α interaction SH3 PR domains p85 mediates p85 p85 homodimer contains four SH2 domains outcompetes p85/p110 heterodimers phosphorylated Tyr residues insulin receptor substrate 1 inhibiting insulin/IGF-stimulated PI3K hypothesized association surplus PRR11 with p85α formation p85 homodimers association heterodimeric p85/p110 with IRS1 PI3K activation insulin co-transduced PIK3R1 (p85α) tagged Flag human influenza hemagglutinin) into HEK293 cellsEctopic expression PRR11 association p85α-Flag-HA measured precipitation immunoblot PLA PRR11 inhibits p85 homodimerization MCF7 LTED cells PIK3R1-Flag PIK3R1-HA knockdown PRR11 p85 homodimers (Fig. 5g PRR11 silencing reduced insulin association p110α IRS1 insulin/IGF-stimulated p-AKT MCF7 HCC1428 LTED cells. 5h PRR11 overexpression enhanced p110α-IRS1 association p-AKT MDA-MB-175VII cells (Fig. 5i PR PRR11 interaction p85 generated PRR11 mutant-ΔPR) lacking PR motif 7a co-transfected Flag-tagged PIK3R1 V5-PRR11-WT V5-PRR11-ΔPR HEK293 cells Immunoprecipitation Flag antibody deletion PR motif PRR11 reduced interaction p85α (Fig. mediates association p85α PRR11-ΔPR induce p-AKT levels PRR11 wild type MCF7 LTED HEK293 cells PRR11 3’ UTR shRNAdata suggest PRR11 promotes association p85α reduced p85 homodimerization enhanced PI3K activation amplification PI3K computed signature scores gene sets insulin/PI3K pathway ER+/HER2− breast tumors scores higher in tumors PRR11 gain amplification PRR11 deletion diploid. hallmark gene sets PI3K/AKT pathway_AKT_MTOR_SIGNALING” enriched in ER+/HER2− tumors PRR11 gain amplification PRR11 deleted/diploid computed connectivity map) genes upregulated ER+/HER2− cancers PRR11 gain amplification ER+ no TCGA PI3K inhibitors AKT signaling loss-of-function) perturbation classes < −95 connectivity score) in MCF7 cells (Fig. opposite connectivity genes overexpressed PRR11 amplified ER+ breast cancers downregulated by perturbations PI3K/AKT CDK4/6 inhibitors opposite connectivity effective against PRR11-amplified cancers 6PRR11 amplification hyperactivation PI3K pathway in ER+ breast cancersSignature score PI3K gene ER+/HER2−/PIK3CA breast cancers PRR11 copy number n = 503 135 deletion/diploid gain/amplification group score IGF1 gene ER+/HER2- breast cancers PRR11 copy number n = 1058 199 deletion/diploid gain/amplification Data mean ± SD-tailed t-tests). mRNA expression data ER+/HER2- tumors (PRR11 gain/amplification deletion enrichment PI3K_AKT_MTOR_SIG MTORC1_SIG signatures Connectivity scores genes upregulated ER+) breast cancers PRR11 ER+/HER2− Connectivity score 44 perturbation classes out 171 highlighted (tau < –95) PIK3CA mutation frequency ER+/HER2− 1398) ER+ (TCGA n = 594) breast cancers function PRR11 copy number alterations no alterations Source data METABRIC TCGA datasets PRR11 amplification PIK3CA mutations mutually exclusive ER+ breast cancers PRR11 amplification linked activation PI3K-amplified breast cancer cells rely PI3K oncogenic role PRR11 depends activation PI3K PI3K inhibitors effective against PRR11-overexpressing cells utilized MCF10A cells EGF insulin PIK3CA knockdown inhibited proliferation MCF10A cells EGF/insulin PIK3CA knockdown abolished growth PRR11 overexpression PRR11-mediated cell growth requires PIK3CA PIK3CA dependence score 57 breast cancer cell lines correlated with PRR11 copy number DEMETER2 PRISM 19Q3 PRR11 copy number 27 cancer cell lines correlated with PI3K inhibitors pictilisib taselisib lines with PRR11 amplification higher to PI3K inhibitor pictilisib LINCS/Sanger dataset ectopic expression PIK3CA sensitizes cells PI3K PRR11 overexpression sensitized MCF10A MDA-MB-134VI cells to PI3K inhibitors PRR11-overexpressing cells rely on PIK3CA sensitive to PI3K inhibitors-amplified breast cancer cells dependent on PI3K pathwayPIK3CA dependency scores 57 cell lines plotted PRR11 V5 dataset Sensitivity 27 lines pictilisib taselisib PRR11 Repurposing 19Q3 dataset Y-axis shows log 2 cell fraction barcode abundance drug treatment Pictilisib sensitivity score 26 cell lines LINCS MGH/Sanger dataset (n = 22 4 non-amplification amplification group Data mean ± SD-tailed t-tests). Alpelisib taselisib GR50 MDA-MB-134VI-MB-175VII cells pLX302-LacZ calculated GR metrics calculator Cell numbers day 0 day 6 input Data mean ± SD three replicates Low density monolayers cells grown estrogen-free medium 14 days stained crystal violet mean ± SD three replicates Low density monolayers MDA-MB-175VII-134VI cells treated ± 1 μM alpelisib 1 μM taselisib estrogen-free medium 14 days cell monolayers stained crystal violet Data mean ± SD three replicates Source datatested genetic inhibition PI3K PRR11 resistance estrogen suppression PIK3CA knockdown siRNA abrogated estrogen-independent growth PRR11 overexpression in MDA-MB-134VI cells (Fig. 7f Fig 8e). treatment PI3K inhibitors alpelisib taselisib abolished estrogen-independent growth MDA-MB-175VII-MB-134VI cells PRR11 (Fig. 7g PI3K inhibitors D1 protein levels PRR11 overexpression Fig 8f). data suggest PRR11 escape estrogen suppression blocked by PI3K inhibition 17q23 locus amplified in breast cancers suggested molecular subgroup clustering analysis cluster proliferative ER+ B breast cancers poor prognosis high genomic instability high risk distant relapse ER+ breast cancers studies potential oncogenes 17q23 genes amplified overexpressed RPS6KB1 TBX2 proposed candidates analysis gene expression predicted MUL APPBP2 TRAP240 potential oncogenes27 functional studies incomplete WIP1 (PPM1D) MIR21 in 17q23 cause resistance to anti-HER2 therapy45WIP1 MIR21 not associated worse outcome ER+ breast cancer endocrine therapy Kaplan–Meier Plotter dataset analysis suggests PRR11 associated breast cancer progression activation PI3K/AKT/β-catenin PRR11 promotes cell proliferation ovarian hepatocellular carcinoma19 molecular basis role PRR11 cancer virulence unclear PRR11 binds SH3 domain p85 p85 homodimer formation Homodimers p85 form sequestration complex IRS1 p85-p110 dimers binding attenuating insulin/IGF PI3K excess PRR11 p85-p110 dimers p110 binding IRS1 retention p110 PIP3 formation Homodimers p85 inhibit PI3K signaling PTEN37 knockdown PRR11 alters PTEN protein levels hyperactivation PI3K pathway ER+ breast cancers bypass hormone PRR11 overexpression associated resistance estrogen suppression in primary ER+ postmenopausal tumors treated letrozole analyses support connection PRR11 PI3K pathway significant correlation PI3K/IGF1 gene set signature scores high PRR11 mRNA levels ER+ breast tumors treated neoadjuvant letrozolePRR11 amplifications exclusive PIK3CA mutations breast cancer databases overexpression PRR11 breast epithelial cancer cells stimulated growth abolished by RNA interference p110α treatment PI3K inhibitors PRR11-amplified breast cancer cell lines higher to PI3K inhibitors Treatment advanced ER+ breast cancer PI3Kα inhibitors antiestrogens improved PIK3CA mutations predicted benefit PI3K inhibitor some patients wild type PIK3CA benefitted alterations PI3K pathway respond to PI3K inhibitors biomarkers for enrollment PRR11 amplification biomarker PI3Kα inhibitors PIK3CA-wild type tumors PRR11 gene 17q23 amplicon potential driver antiestrogen resistance in ER+ breast cancer analyses data role PRR11 endocrine resistance PRR11 blocks p85 homodimerization sensitizes PI3K activation amplification confers resistance estrogen deprivation hyperactivation PI3K pathway therapy PI3K actionable target in ER+ breast cancers PRR11 amplificationMCF7 HCC1428 MDA-MB-175-VII-134VI BT-474 breast cancer HEK293 embryonic kidney MCF10A epithelial cells purchased ATCC 2018 2019. 293FT cells Invitrogen 2016. HCC38 MDA-MB-231 breast cancer cells provided Dr Jennifer A Pietenpol Vanderbilt University Medical Center authenticated tested Mycoplasma contamination MCF7-MB-175-VII-134VI BT-474-MB-231 HEK293 293FT cells DMEM/10% fetal bovine serum Antibiotic HCC38 HCC1428 RPMI 1640/10% FBS/1% AA MCF10A DMEM/F12 5% horse serum 20 ng/mL EGF 10 μg/mL insulin 0.5 μg/mL hydrocortisone 0.1 μg/mL cholera toxin 1% AA Long-term estrogen-deprived) cell lines described fulvestrant-resistant MCF7 cells concentrations fulvestrant 50 nM resistant 1 μM fulvestrant tamoxifen-resistant MCF7 HCC1428 tamoxifen 500 nMCells resistant 2 μM tamoxifen resistant cells removed drug 24 h treatment MCF7 cells PIK3CA E545K allele corrected wild provided by Dr. Ben Ho Park Vanderbilt University Medical Center.Xenograft mice maintained Care Laboratory Animals US National Institutes of Health Institutional Animal Care Use Committee procedures approved Institutional Ethics Review Committee University of Texas Southwestern Medical Center Eight-week old female ovariectomized athymic mice implanted 14-day 17β-estradiol pellet 107 MCF7 cells doxycycline-inducible suspended in IMEM Matrigel) injected subcutaneously right flank 4 weeks later mice tumors ≥250 mm3 vehicle (0.9% NaCl) doxycycline (10 mg/kg/daily Tumor diameters measured volume calculated = width2 × length/2 After 4 weeks tumors harvested homogenized TissueLyser II immunoblot analysis set signature analysesSingle-sample gene enrichment 125 breast cancer-related gene expression signatures) < 0.01 activated pathways between PRR11 high low tumorsGene enrichment analysis interface Broad Institute h.all.v6.2.ymbols.gmt gene sets-fixed paraffin 4-μm tumor sections deparaffinized Antigen retrieval citrate buffer (pH 6) decloaking chamber Endogen peroxidase blocked 3% H2O2 protein block Tumor sections incubated PRR11 antibody dilution 1:200 overnight 4 °C Envision visualization DAB hematoxylin counterstain Whole sections acquired AxioScan Z1 slide scanner ×20 Automated-quantitative scoring QuPath Color deconvolution stains set cell segmentation determined hematoxylin OD object classification tumor stroma Percentage PRR11+ cells calculated cell detection algorithm cytoplasm DAB OD mean region assessed non tumor.Fluorescence situ hybridizationFour-μm tissue sections charged slides hybridized overnight PRR11 FISH probe centromere 17 control probe Twenty to sixty tumor cell nuclei evaluated 100× oil immersion objective green PRR11 orange centromere 17 signalsPRR11/CEN17 ratio ≥2.0 amplification transfectionCells 6-well plates 60-mm dishes transfected 20 40 pmole siRNAs Lipofectamine reagent Control siRNA (4390843) PRR11 siRNAs PIK3CA siRNA (4390824-S10520) purchased ThermoFisher Scientific PRR11 ORF Lite cloned pLX302 pINDUCER20 LR ClonaseTM II Enzyme Mix pLX302-PRR11-ΔPR motif predicted Motif Scan removed pLX302-PRR11 Q5-Site-Directed Mutagenesis Kit pLX304-PRR11-ΔPR generated pLX302-PRR11-ΔPR cloning SMART doxycycline-inducible PRR11 shRNA 3’UTR purchased Dharmacon pLenti7.3-PIK3R1-Flag-PIK3R1-HA provided Dr. Gordon Mills Oregon Health Sciences UniversitypLenti6.3-PIK3R1-Flag-PIK3R1-HA cloned pDORNTM221 ClonaseTM II Enzyme pDONR221-PIK3R1-Flag cloned pLenti6.3/V5-DSETTM vector pINDUCER20 II Enzyme Mix pLX304-PRR11 -BRIP1 -SMARCD2 -TACO1 purchased DNASU.Lentiviral 1 μg plasmids co-transfected 0.75 μg psPAX2 0.5 μg pMD2.G 293FT cells Lipofectamine2000 Cell medium changed fresh 24 h post cells collected 48 h later Lentiviral particles sgRNA PRR11 purchased Dharmacon Virus-containing medium 8 μg/mL polybrene Puromycin selection cells transduced SMART doxycycline-inducible PRR11 pLX302-PRR11 pLX302-LacZ Blasticidin selection cells transduced pLX304-GFP -PRR11 -BRIP1 -SMARCD2 -TACO1 pLenti6.3-PIK3R1-FlagMCF7 LTED cells pIND-PIK3R1-HA selected G418 sulfate proliferation 24 h transfection HCC1428 MCF7 cells seeded 6-well plates trypsinized counted 3 days 6 days Z2 coulter counter analyzer-qPCRRNA extracted Maxwell RSC Cells Kit cDNA synthesized iSCRIPT cDNA synthesis Kit PCR PowerUpTM SYBRTM Green Master Mix PIK3CA GAPDH primers QuantStudio3 Real-Time PCR System luciferase assayCells seeded 96-well plates transfected pGLB-MERE pCMV-Renilla control PRR11 siRNA Twenty-four h switched estrogen-free medium 24 h 1 nM estradiol 24 h Renilla firefly luciferase activities measured Dual-Luciferase® Reporter Assay System assaysMCF7 HCC1428 cells 12-well plate transfected control PRR11 grown ± 1 nM estradiol 1 μM tamoxifen 1 μM fulvestrant 10 daysMDA-MB-175VII cells pLX304-PRR11 -BRIP1 -SMARCD2 -TACO1 grown estradiol 14 days-175VII-134VI cells pLX302-PRR11 -LacZ grown estradiol 1 μM taselisib 1 μM alpelisib 14 days MCF10A pLX302-PRR11 -LacZ cells grown 1 μM taselisib 1 alpelisib 10 days replaced 4 days Cells fixed cold-iced methanol 10 min −20 °C stained 0.5% crystal violet solution 10 min Stained monolayers imaged dissolved 10% acetic acid 15 min Supernatants transferred 96-well plates 560 nm analysisCells lysed RIPA buffer protease inhibitors incubated ice 30 min Supernatants collected centrifugation 14,000 rpm 10 min Protein concentration measured BCA Protein Assay kit Thirty μg protein SDS-PAGE nitrocellulose membrane analysis antibodies Supplementary Table 6.Cell cycle 48 h transfection cells trypsinized fixed 70% ethanol 3 h −20 °CCells washed PBS resuspended PBS 100 μg/mL RNase A 40 μg/mL propidium iodide 10 min room temperature Stained cells analyzed BD LSRFORTESSA cell cycle phases Dean-Jett-Fox model FlowJo (ver.10) Gating strategy Supplementary Fig. 4i.PCR arrayRNA extracted MCF7 LTED cells transfected control siRNA PRR11 siRNA 48 h Maxwell® RSC simplyRNA Cells Kit Complementary DNA synthesized RT2 First Strand Kit subjected real-time PCR RT2 ProfilerTM PCR Array Human Cell Cycle) cell plasmid GFP-based AKT-PH biosensor PH domain AKT1 fused COOH terminus GFP MCF7 LTED cells transduced pEGFP-AKT-PH sorted for GFP expression flow cytometry three-dimensional imaging modified light-sheet illuminates specimen light-sheet beam scanned fluorescence descanned mirror galvanometer high-NA silicone immersion objective ×100 1.35 high air immersion objective ×40 bespoke glass-tipped objective-AGY v1.0_single_objective_lightsheet Images acquired plane high-speed scientific CMOS camera (Hamamatsu Flash 4.0 LabView software sheared frequency-domain Python rotated affine transform MATLAB or IMOD No deconvolution 8 positions 35 mm dish chosen random multiple cells Movies acquired volumetric imaging rate 0.1 Hz 10 min experiments DynabeadsTM Protein G Immunoprecipitation Kit) 10 μg antibodies pre-incubated 1.5 mg Dynabeads 10 min incubated overnight 4 °C 1 mg cell lysate Precipitates eluted bead-antibody-antigen complex NuPAGE 10 min 70 °C Eluted samples 20 μg input lysates immunoblot analysis.Proximity ligation assayMCF7 LTED (5 × 104/well cells seeded 8-well chamber slides (Lab-Tek HEK293 cells (5 seeded 24 h post-transfection pLX302-PRR11-V5 pLenti7.3-PIK3R1-Flag-PIK3R1-HA PRR11 p85 V5 HA Flag antibodies usedPLA performed Duolink In Situ Red Starter Kit Mouse/Rabbit imaged DMi8 inverted microscope PLA foci quantified Duolink Image Tool software 5 images per sample analyzed reproducibilityPearson correlation hazard ratio t-tests performed GraphPad Prism 8 Microsoft Excel 2016. Data mean SD experiments conducted three times p value < 0.05 significant false discovery rate computed Benjamini–Hochberg procedure R 3.5.2 R studio 1.1.463 used Nature Research Reporting Summary Movie | 50.1 | 1.080966 |
10.1038/s41467-020-16663-1 | PMC7298017 | Anomalously specular radar reflections (ASRR) from Titan’s tropical region were interpreted earlier as evidence for liquid surfaces, but the Cassini spacecraft did not observe lakes/seas at the anomalously specular locations. Here, the authors show that ASRR originate from one terrain unit, likely paleolakes/paleoseas. | Saturn’s moon Titan has a methane cycle with clouds, rain, rivers, lakes, and seas; it is the only world known to presently have a volatile cycle akin to Earth’s tropospheric water cycle. Anomalously specular radar reflections (ASRR) from Titan’s tropical region were observed with the Arecibo Observatory (AO) and Green Bank Telescope (GBT) and interpreted as evidence for liquid surfaces. The Cassini spacecraft discovered lakes/seas on Titan, however, it did not observe lakes/seas at the AO/GBT anomalously specular locations. A satisfactory explanation for the ASRR has been elusive for more than a decade. Here we show that the ASRR originate from one terrain unit, likely paleolakes/paleoseas. Titan observations provide ground-truth in the search for oceans on exoearths and an important lesson is that identifying liquid surfaces by specular reflections requires a stringent definition of specular; we propose a definition for this purpose. | IntroductionAnomalously specular radar reflections from the southern tropical region of Saturn’s moon Titan (equator to ≈27°S, Saturn and Titan have a solar obliquity of ≈27°) were observed with the Arecibo Observatory and Green Bank Telescope from 2000-2008 and interpreted as evidence for liquid surfaces1,2. The Cassini Saturn orbiter (e.g., ref. 3) imaged Titan’s surface at infrared and microwave wavelengths from 2004-2017 and discovered >500 lakes/seas4. It did not, however, observe liquid surfaces in the regions that are anomalously specular to AO/GBT2. Invoking transient liquids is one possible resolution of this apparent discrepancy as methane precipitation does occur on Titan5; a more satisfactory explanation has been elusive.Here we show that the AO/GBT ASRR are correlated to one terrain unit and that this terrain unit has both smoother surfaces and a greater dielectric constant (different composition) than its surroundings, both of which contribute to the anomalous reflections. Transient liquids are not necessary to explain the ASRR. We argue that the terrain unit is likely paleolakes/paleoseas (sites of former lakes/seas), a geomorphologic unit that is of interest for both geology and astrobiology and is all the more interesting given its unique properties relative to other geomorphologic units on Titan. Our conclusion that Titan’s ASRR originate from solid surfaces, suggests that to identify liquids on exoearths by specular reflections, a stringent definition of specular should be used. We recommend a definition based on the coherence of the reflected electromagnetic waves.ResultsArecibo Observatory and Green Bank Telescope anomalously specular radar reflectionsTitan’s surface is not resolved in AO/GBT radar observations, however, the Doppler shift from Titan’s rotation results in an echo spectrum that is equivalent to the scan of a slit across Titan’s disk1,2. Figure 1a–e show example AO/GBT echo spectra of Titan (from the 83 observations, acquired over eight oppositions, in Black et al.2). The central, 1-Hz wide, Doppler bin corresponds to a ≈14-km wide slit that includes the entire ≈5150 km (Titan’s diameter) height of the disk. Higher (or lower) frequency bins correspond to slits shifted laterally away from the center of Titan’s disk in the direction of the limb rotating toward (or away from) Earth. The spectra are in units of normalized radar cross section (NRCS), which is the ratio of the radar power backscattered to the receiver to the power that would have been received if the power incident on the surface had been scattered isotropically; effectively, NRCS is the brightness in radar observations (e.g., ref. 6). The NRCS of a surface generally decreases with increasing incidence angle (e.g., refs. 2,7,8). As a result, AO/GBT spectra generally decrease away from the central Doppler bin (slit over the center of the disk), because the central Doppler bin includes the subradar point, where the incidence angle is zero degrees.Fig. 1Arecibo Observatory and Green Bank Telescope observations of Saturn’s moon Titan.a–e Example AO/GBT echo spectra of Titan in the opposite circular polarization channel, adapted from Black et al.2. The observation date, subradar location on Titan, and maximum-NRCS are in the panel legends. The width of the echo from Titan is indicated by the green line on the x-axis; data beyond the green line are indicative of the noise. All AO/GBT spectra from Titan have a broad, diffuse component with an NRCS of approximately zero at the limits of the echo (endpoints of the green line on the x-axis) that increases toward the center. The NRCS of the central Doppler bin varies drastically between the spectra. Observations with anomalously high peaks (e.g., d and e) are the AO/GBT anomalously specular radar reflections. c Example demonstrating that the maximum-NRCS of some observations may be affected by noise. f Histograms of the maximum-NRCS of AO/GBT observations of Titan reported in Black et al.2. The black histogram includes all 83 observations and the red histogram includes only observations where the maximum-NRCS is ≥4 times the standard deviation of the noise (σ), to distinguish between observations with a high maximum-NRCS that may be due to greater noise and those that are confidently anomalous. Two observations have a maximum-NRCS ≥3.5 and are clearly separate from the distribution and thus anomalously specular. Another group of observations, with 1.5 ≈< maximum-NRCS ≤2, is also separate from most observations; these observations are also anomalously specular, albeit less so than the two observations with maximum-NRCS ≥3.5.The transmitted signal in the AO/GBT observations was circularly polarized and the echo power was measured in both the opposite and same circular senses as that transmitted1,2. A specular reflection would originate from the subradar point and increase the NRCS of the central Doppler bin of the opposite circular polarization spectrum. Thus, for the purpose of investigating the AO/GBT ASRR, we use the maximum NRCS of each AO/GBT opposite circular polarization spectrum of Titan (NRCS of the specular component of the spectrum). The AO/GBT ASRR are the observations with anomalously high maximum-NRCS. We use the maximum-NRCS rather than the surface root-mean-square slope parameter of a radar scattering model that was used in previous analyses1,2 because Titan’s complex surface is variable over many length scales. The model parameter depends on neighboring Doppler bins, so it is contaminated by longitudinal surroundings. The maximum-NRCS depends on only the central Doppler bin and thus has a greater weighting from the terrain at the subradar location2.Figure 1f shows histograms of the maximum-NRCS of the AO/GBT observations from Black et al.2. The noise in the AO/GBT spectra varied substantially between observations (e.g., Fig. 1a–e). To distinguish between observations with a high maximum-NRCS that may be due to greater noise and those that are confidently anomalous, Fig. 1f includes both the histogram for all of the AO/GBT observations (black) and the histogram for only the observations where the maximum-NRCS was ≥4 times the standard deviation of the noise (red). Two observations have a conspicuously greater maximum-NRCS than all other observations and thus are distinctly anomalous (they are also the two most specular observations using the surface root-mean-square slope parameter instead of maximum-NRCS). Another group of observations with maximum-NRCS of ≈1.5-2 are intermediate; their maximum-NRCS is greater than most other observations but also substantially less than the two extreme ASRR. We consider these observations also to be anomalously specular and argue below that they originate from the same terrain unit as the two distinctly anomalous observations.Arecibo Observatory, Green Bank Telescope, and Cassini observationsThe subradar locations of the AO/GBT observations are shown on a map of Titan’s surface in Fig. 2a, b. The AO is limited to observing zenith angles <≈20° and thus the radar system can only observe Titan during ≈1/3 of the oppositions over Saturn’s ≈29.5-year orbital period (i.e., from 2000 to 2008 but not 2009 to 2027). The AO-Titan geometry further limits the subradar locations to a latitude range of ≈20° within Titan’s southern tropical region (i.e., from 7 to 27°S). The GBT shares these limitations because it does not have a radar transmitter and its radar observations of Titan were restricted to receiving echoes from the AO transmitter.Fig. 2Map of Titan.a The monochrome swaths are Cassini RADAR11 images. The purple and red dots are AO/GBT subradar locations; red dots are locations where the maximum-NRCS was ≥4 standard deviations above the noise. Dot radii are linearly proportional to maximum-NRCS. Colored tracks are Cassini RADAR altimetry observations where color indicates NRCS. There is a high concentration of large, red dots at ≈70–135°W, 15–30°S and many Cassini altimetry observations in this region also have a high NRCS. b Boxed area enlarged. The colorbar applies to both a and b.Figure 2 is similar to Fig. 17 in Black et al.2, with four important distinctions. Firstly, the size of the AO/GBT dots indicates the maximum-NRCS rather than the surface root-mean-square slope parameter of a radar scattering model (discussed above). Secondly, the spin models (rate and pole) used to determine locations on Titan for the AO/GBT and Cassini datasets were inconsistent in Black et al.2. Furthermore, the Cassini dataset was internally inconsistent and some observations used an erroneous model9. We have corrected the locations of both datasets to the current model recommended by the International Astronomical Union10 (Methods, Updated AO/GBT Subradar Locations on Titan and Supplementary Data 1). These corrections correspond to relative changes of 3–14 km, with an average change of 11 km. These changes are noteworthy because Titan’s complex surface varies on these length scales.A third distinction, is that Fig. 2 includes the complete Cassini radar instrument (called RADAR; ref. 11) image dataset of Titan from all phases of the mission from 2004 to 2017. Compared to the dataset from the prime mission (2004–2008), this results in an ≈50% increase in the number of AO/GBT locations that were imaged by the RADAR instrument and also improves the time constraints on temporal changes. A fourth important distinction, is the addition of the Cassini RADAR altimetry observations (Methods, Cassini RADAR Altimetry Observations). The NRCS of Titan’s surface depends sensitively on incidence angle (e.g., refs. 2,7,8) and the AO/GBT ASRR are reflections from the surface at nadir (0°) incidence. The incidence angles of the Cassini RADAR images, of the AO/GBT locations, however, range from 10 to 77°. In contrast, Cassini RADAR altimetry observations have near-nadir incidences (deviations from which are due to imperfect pointing). Thus, comparison of AO/GBT maximum-NRCS with Cassini-altimetry NRCS is a nadir-to-nadir comparison where the substantial effects of incidence angle are mitigated. This comparison, however, still has systematic differences since the AO/GBT observations include backscatter from non-nadir incidence angles along the height of the slit, and are at a radar wavelength (λ) of 12.6 cm, different from the Cassini RADAR (λ = 2.2 cm).To distinguish between AO/GBT subradar locations with a high maximum-NRCS that may be due to greater noise and those that are confidently anomalous (e.g., Fig. 1a–e), locations where the maximum-NRCS was ≥4 times the standard deviation of the noise are colored red in Fig. 2. Thus, dots that are both red and large indicate locations that are confidently (i.e., low noise), anomalously specular (high maximum-NRCS) to AO/GBT. The large, red dots are concentrated at ≈70–135°W, 15–30°S. This region similarly has a relatively high NRCS in many Cassini altimetry observations (each altimetry track has many independent echoes).The black areas in the polar regions of Fig. 2a are Titan’s lakes/seas4. Their low NRCS and stark contrast with surrounding terrains is distinctive in Cassini RADAR images4, and no such features were observed at any of the AO/GBT subradar locations, anomalously specular or non-specular2. Additional hypotheses for the AO/GBT ASRR include transient liquids, dunes/interdunes, and paleolakes/paleoseas.Transient liquids hypothesisSurface darkening following the observation of nearby clouds, most likely from ponding of methane rain, was observed twice by Cassini. The first rain event was in 2004 at ≈80°S (ref. 12), far from the subradar locations of the AO/GBT observations, and the second in 2010 at ≈20°S (ref. 5), which postdated all the AO/GBT observations. Thus, transient liquids from these two known rain events cannot explain the ASRR. The surface darkened by the 2010 tropical rain event did not completely revert to its original spectrum until >1 year after its initial darkening13, demonstrating that surface changes from rain may be detectable long after the rainfall. Many other clouds were observed on Titan by both Cassini and Earth-based telescopes (e.g., refs. 14–17), but they have not been associated with subsequent surface changes. Clouds over Titan’s southern tropical region were uncommon during the period from 2004 to 2008 when both Cassini and AO/GBT observed Titan.Nine AO/GBT locations were imaged by Cassini RADAR both before and after the AO/GBT observation; no surface changes are detected between the RADAR before and after images. Four RADAR images are <1 year from the AO/GBT observation and 13 images have an interval of <2 years; none of the images show surface liquids. Thus, the Cassini RADAR images do not support the hypothesis of transient liquids but also do not strongly constrain their presence during the AO/GBT observations. The low frequency of rain events on Titan (two detected events over ≈13 years), long reversion timescale (>1 year13), low frequency of clouds over Titan’s southern tropical region from 2004 to 2008 (refs. 15–17), and RADAR image constraints all suggest that transient liquids are not responsible for the AO/GBT ASRR.Black et al.2 noted that the AO/GBT reflections depend on date and/or latitude but these variables are correlated in the AO/GBT observations and cannot be separated (see their Fig. 19). Comparison with Cassini altimetry in Fig. 3a–d indicates that latitude, not date, is responsible for the apparent dependence. Note that in Black et al.2 there is a trend with date that is lost when the y-axis is changed from percentage of observations modeled with a specular component to maximum-NRCS. The trend with date is neither consistent with the choice of y-axis nor between AO/GBT and Cassini (Fig. 3b, d) and thus any temporal change hypothesis (including transient liquids) is unlikely to explain the AO/GBT ASRR. An explanation that does not invoke temporal changes is presented below. In summary, temporal changes are an unlikely hypothesis.Fig. 3NRCS as a function of latitude and date for AO/GBT and Cassini altimetry observations.Panels a, b show AO/GBT observations, like Fig. 19 of Black et al.2 but the y-axis variable is maximum-NRCS. Panels c, d show Cassini altimetry observations. Panels a, c show latitude and panels b, d show date. Black dots are observations and orange-edged, cyan-filled rectangles are weighted means of the bins. Error bars show one standard deviation of the noise. There is an approximately consistent peak of the cyan/orange bins at ≈23°S between AO/GBT and Cassini and the latitude trends are somewhat similar. There is no agreement in the trends with date. Thus, the anomalously specular AO/GBT observations depend on location, not date.Dunes/interdunes hypothesisTitan has dunes analogous to sand dunes on Earth except that the grains are likely composed of hydrocarbons rather than silicates (e.g., ref. 18). Dune fields cover millions of square kilometers on Titan, primarily in its tropical region, and are one of Titan’s main terrain units. Although the dune fields have rough surfaces at the scale of individual dunes (≈1 km), and possibly also at the scale of individual grains (≈1 mm), the surface of each dune and interdune may be smooth, as on the Earth, at the scale of the AO/GBT observation wavelength (λ = 12.6 cm, ref. 19). The AO/GBT ASRR are from Titan’s tropical region. From this cursory perspective, it is tempting to relate the ASRR to the dunes or interdune areas (which are brighter than the dunes in Cassini altimetry observations19). On closer inspection, however, there is no correlation. Seven AO/GBT subradar locations are clearly dune/interdune regions in Cassini RADAR images and their AO/GBT maximum-NRCS range from 0.56 to 1.09. From Fig. 1f, these maximum-NRCS values are not anomalous and do not include the observations of interest. Thus, the dune fields are ruled out as the source of the AO/GBT ASRR.Hotei and Tui Regiones and paleolake/paleosea hypothesisThe AO/GBT ASRR are concentrated at ≈70–135°W, 15–30°S (Fig. 2); the southeastern part of this area is Hotei Regio, Tui Regio is the southwestern part (boundaries shown in Fig. 4a), and the area in-between is southern Xanadu. Hotei and Tui Regiones have very similar properties and together are distinct terrains in Titan’s tropical region in their infrared spectrum, morphology, topography, and as we show in this paper, nadir radar brightness. They have been hypothesized to be cryovolcanic terrains on the basis of their spectroscopy, morphology, and topography (e.g., refs. 20–22). On the other hand, they have been hypothesized to be paleoseas/paleolakes, also based on their spectroscopy23,24 and morphology and topography25. Spectrally, their most distinct signature is an unusually high brightness in Titan’s 5-μm atmospheric window. MacKenzie et al.24 mapped Titan’s 5-μm-bright areas using Cassini Visual and Infrared Mapping Spectrometer (VIMS)26 observations and we use their map as the definition of the extent of Hotei and Tui Regiones in this work. The map is shown in Fig. 4a with the AO/GBT observations as in Fig. 2.Fig. 4Hotei and Tui Regiones.a Black areas are Titan’s 5-μm-bright regions as mapped in MacKenzie et al.24. Hotei and Tui Regiones are the 5-μm-bright regions at ≈80°W, 25°S and ≈120°W, 25°S. AO/GBT observations are plotted as in Fig. 2. Hotei and Tui Regiones are the main source of the AO/GBT anomalously specular reflections. Two noteworthy exceptions to the strong correlation of anomalously specular AO/GBT observations to Hotei/Tui are labeled NE1 and NE2 and discussed in the Discussion section. b The 18-km subradar track (red line) of the highest maximum-NRCS (and most specular in Black et al.2) AO/GBT observation, located in Hotei Regio (largest red dot in Hotei Regio in a). c The stereo topography of the same area37. The AO/GBT reflection is from a bright area that is morphologically similar to paleolakes25 (b) and is topographically low (c). d The 18-km subradar track (red line) of the second highest maximum-NRCS (and second most specular in Black et al.2) AO/GBT observation, located in Tui Regio (largest red dot in Tui Regio in a). The AO/GBT reflection is again from a bright area that is morphologically similar to the comparably specular location in Hotei Regio and paleolakes. Stereo topography does not cover this track.Figure 4a shows that every AO/GBT observation with a subradar point in Hotei/Tui is anomalously specular, including the two observations that are distinctly brighter than all others (the two largest, red dots; one is in Hotei Regio and the other Tui Regio). Of the two highest (maximum-NRCS ≥3.5) plus seven high confidence (red), intermediate maximum-NRCS (1.5 ≤ maximum-NRCS ≤ 2, Fig. 1f) observations, six are in Hotei or Tui Regiones, one is on Hotei Regio’s boundary, and only two are not in or near either Regio. The two exceptions merit an extended discussion, which is deferred to the next section. Regardless, we conclude that Hotei and Tui Regiones are the primary source of the AO/GBT ASRR.The subradar locations of the two distinctly specular AO/GBT observations were imaged by Cassini RADAR. Figure 4b, d show that the AO/GBT subradar tracks (the subradar point moves as the echo is integrated over time, primarily from Titan’s rotation) are over bright areas. These bright areas are morphologically similar to paleolakes (called empty lake basins4) in Titan’s polar regions25. Figure 4c shows that the bright areas in Hotei Regio are topographic lows, consistent with the paleolake hypothesis (stereo topography does not cover the analogous location in Tui Regio). Hotei Regio also has a relatively high NRCS in many Cassini altimetry observations (Fig. 2; Tui Regio was not observed with Cassini altimetry). However, the highest-NRCS Cassini altimetry observations are of Titan’s liquid-filled lakes/seas: coherent specular reflections from the lakes/seas can be 100 times brighter than the diffuse echoes from surrounding solid surfaces27,28. Thus, in Cassini altimetry observations, Hotei Regio is not as bright as the liquid-filled lakes/seas but is brighter than many other terrains. Hotei Regio has a higher NRCS than other tropical regions in both AO/GBT and Cassini altimetry observations; it is bright in nadir radar observations. The paleolakes in Titan’s north polar region are similarly brighter than their surroundings in Cassini altimetry observations but not as bright as the liquid-filled lakes/seas29,30. Consequently, we add Cassini-altimetry-brightness to the list of characteristics that support the paleolake/paleosea hypothesis for Hotei Regio and, by analogy, Tui Regio. We favor this hypothesis but note that the cryovolcanic and paleolake/paleosea hypotheses are not mutually exclusive. We also note that our earlier conclusion that Hotei/Tui are the primary source of the AO/GBT ASRR is independent of the interpretation of their geomorphology; regardless of whether they are paleolakes/paleoseas, cryovolcanoes, or another terrain. Based on the above correlations we predict that: (1) the bright regions of Tui Regio in RADAR images are topographic lows and (2) Tui Regio will have an intermediate NRCS in RADAR nadir observations.The smoother a surface, the greater its NRCS in nadir radar observations. The high NRCS of Hotei/Tui in AO/GBT and Cassini altimetry observations, however, cannot be attributed solely to the smoothness of their surface. As shown in Fig. 4b, d, the subradar locations of the anomalous AO/GBT observations are also bright in Cassini RADAR images, which have nonzero incidence angles (the incidence angles for Fig. 4b, d were 18° and 23°). A greater dielectric constant and greater diffuse scattering both increase the NRCS at all incidence angles. The dielectric constant of a surface depends on its composition (and other parameters, e.g., porosity) and Hotei/Tui have a different composition than their surroundings; recall that they are spectrally distinct, particularly in Titan’s 5-μm atmospheric window (e.g., ref. 24). Thus, the surfaces of Hotei/Tui likely differ from other surfaces in Titan’s tropical region in both their smoothness and dielectric constant (composition) and both of these differences contribute to their high maximum-NRCS in AO/GBT observations. These differences may be due to lake/sea sediments that were deposited in the past, when Hotei/Tui were liquid-filled31,32. Further study of Hotei/Tui may be fertile research for understanding the geology and chemistry of Titan’s lake/sea sediments and the long-term evolution of its climate.DiscussionThere are two noteworthy exceptions to the strong correlation of AO/GBT ASRR to Hotei/Tui. The first has a subradar point at ≈47°W, 13°S (NE1 in Fig. 4a), was acquired on 1/30/2007, and has the third-highest maximum-NRCS of all the AO/GBT observations. Based on the conclusions above, that location is predicted to be morphologically similar to Hotei/Tui/paleolakes, topographically low, and bright in 5-μm and nadir radar observations. That location is at the edge of two low-resolution Cassini RADAR images but noise inhibits interpretation of the surface. It was observed by VIMS and intriguingly is at an intersection of the equatorial-bright (organic sediments) and dark-blue (water ice enriched) spectral units, not the 5-μm-bright (evaporite) spectral unit (e.g., refs. 33,34). The second exception has a subradar point at ≈163°W, 24°S (NE2 in Fig. 4a) and was acquired on 11/3/2000. We have dismissed other bright reflections centered on the 2001 and 2008 bins in Fig. 3 based on their greater noise (primarily a result of shorter integration times in the first and last oppositions of the AO/GBT campaign) but this observation was almost six standard deviations above the noise and cannot be dismissed. Nevertheless, we have some reservation about predicting the terrain at that location is similar to Hotei/Tui. This was the first observation of the AO/GBT campaign, the maximum NRCS is not from the central Doppler bin, and the spectral spike is only one bin wide. This location was not observed with RADAR but was observed by VIMS and its spectrum is a mix of the equatorial-bright and 5-μm-bright spectral units. The Cassini mission ended in 2017 so these locations cannot be imaged in high-resolution until another spacecraft explores Titan.The nadir-to-nadir comparison between AO/GBT and Cassini altimetry observations also leads to an important conclusion about the definition of specular in the context of liquids on a planetary surface. The AO/GBT observations of Hotei and Tui Regiones are specular in several senses: they have a spike in the central Doppler bin of their opposite circular polarization spectrum, they have a high NRCS, and they are best fit by radar scattering models that include a specular component. The Cassini altimetry observations of Hotei Regio also have a high NRCS, supporting the notion that Hotei and Tui Regiones are specular terrains. However, Cassini altimetry observations of Titan’s lakes/seas indicate that reflections from Titan’s liquid surfaces are markedly more specular (brighter). The reflections from the smooth lake/sea surfaces are so bright that the echoes must be coherent27,28. The received power in this case decreases with the square of the total (two-way) distance from transmitter to reflector to receiver ((2d)2 for monostatic observations, where d is distance from transmitter/receiver to reflector) rather than the square of the distance from transmitter to reflector multiplied by the square of the distance from reflector to receiver (d4 for monostatic observations) as it does for noncoherent reflections. Using Titan as a solar system ground-truth in the search for oceans on exoearths suggests that, to identify liquids by specular reflections, a stringent definition of specular should be used. We recommend a definition based on the coherence of the reflected electromagnetic waves rather than definitions based on combinations of relative brightness, incidence angle, location, and/or polarization; reflections that are specular in the latter senses are quasi-specular. For Titan, liquids were expected based on the planetary context and reflections that are specular in the relative brightness, incidence angle, expected location, and polarization senses were observed but the reflections were from solid, not liquid, surfaces. The smooth liquid surfaces on Titan, however, are specular in a coherent sense at microwave27,28 and infrared (e.g., refs. 35,36) wavelengths.MethodsUpdated AO/GBT subradar locations on TitanThe AO/GBT subradar locations on Titan were updated from Black et al.2 using the International Astronomical Union’s (IAU) current (2015) spin model10. The start and end times of each AO/GBT observation (provided in Supplementary Data 1) were input into the Navigation and Ancillary Information Facility’s (NAIF) SPICE system. The subradar points on Titan, at the time the radar echo departed from Titan’s surface, were calculated using the cspice_subpnt module. The Titan ellipsoid was a sphere with a radius of 2575 km. The updated locations are provided in Supplementary Data 1. Note that the Cassini RADAR images and altimetry observations were also located using the same spin model and ellipsoid as part of the pipeline processing of the mission datasets.Cassini RADAR altimetry observationsIn both Figs. 2 and 3, only Cassini RADAR altimetry observations with an incidence angle of <1° are included. The cut removes echoes acquired with substantial pointing errors as well as echoes at the beginning and end of the tracks. We verified that changing the cutoff does not qualitatively affect the results. The NRCS of each altimetry echo was corrected for deviations from nadir (0° incidence) pointing; the incidence angle correction is described in the Cassini RADAR Users Guide, which is available from the NASA Planetary Data System (PDS). We found that varying the incidence angle correction, even not including a correction, does not qualitatively affect the results.Supplementary information
Peer Review File
Description of Additional Supplementary Files
Supplementary Data 1 | nature communications | [
"Article"
] | [
"Planetary science",
"Geomorphology",
"Hydrology"
] | specular radar reflections southern tropical region Saturn’s Titan (equator ≈27°S ≈27° observed Arecibo Observatory Green Bank Telescope 2000-2008 interpreted evidence liquid surfaces1,2 Cassini Saturn orbiter imaged Titan’s surface infrared microwave 2004-2017 discovered >500 lakes/seas4 observe liquid surfaces regions specular to AO/GBT2. transient liquids resolution discrepancy methane precipitation on satisfactory explanation elusive AO/GBT ASRR correlated to terrain unit smoother surfaces greater dielectric constant contribute reflections Transient liquids not necessary ASRR terrain unit likely paleolakes/paleoseas former lakes geomorphologic unit interest for geology astrobiology unique properties Titan’s ASRR originate from solid surfaces liquids exoearths specular reflections stringent definition specular definition coherence reflected electromagnetic waves Observatory Green Bank Telescope specular radar reflectionsTitan’s surface not resolved AO/GBT radar observations Doppler shift from Titan’s rotation echo spectrum equivalent slit across Titan’s Figure 1a–e AO/GBT echo spectra Titan 83 observations Black.2)central 1-Hz Doppler bin ≈14-km wide slit ≈5150 km (Titan’s diameter height disk Higher frequency bins slits shifted laterally from center Titan’s disk toward Earth spectra in units normalized radar cross section (NRCS), ratio radar power backscattered to receiver to power NRCS brightness radar observations NRCS decreases with increasing incidence angle 2 AO/GBT spectra decrease from central Doppler bin includes subradar point incidence angle zero degrees.Fig. 1Arecibo Observatory Green Bank Telescope observations Saturn’s Titan AO/GBT echo spectra Titan opposite circular polarization channel Black et al observation date subradar location Titan maximum-NRCS in panel legends width echo indicated by green line x-axis data beyond indicative of noise AO/GBT spectra have broad diffuse component NRCS approximately zero at limits echo increases toward center NRCS central Doppler bin varies between spectra Observations with anomalously high peaks are AO/GBT anomalously specular radar reflections maximum-NRCS affected by noiseHistograms maximum-NRCS AO/GBT observations Titan Black et al.2. black histogram includes 83 observations red histogram maximum-NRCS ≥4 times standard deviation noise high maximum-NRCS noise anomalous. Two observations maximum-NRCS ≥3.5 separate anomalously specular 1.5 ≈< maximum-NRCS ≤2 separate anomalously specular less than-NRCS ≥3.5 transmitted signal AO/GBT circularly polarized echo power measured opposite circular senses specular reflection subradar point NRCS central Doppler bin opposite polarization spectrum investigating AO/GBT ASRR maximum NRCS each AO/GBT opposite circular polarization spectrum Titan specular AO/GBT ASRR observations anomalously high maximum-NRCS use maximum-NRCS surface root-mean-square slope parameter radar scattering model Titan’s complex surface variable scales model parameter depends neighboring Doppler bins maximum-NRCS depends central Doppler bin greater weighting from terrain subradar location2.Figure 1f shows histograms maximum-NRCS AO/GBT observations Black et al.2. noise AO/GBT spectra varied between observationsdistinguish high maximum-NRCS due noise anomalous Fig. 1f includes AO/GBT observations maximum-NRCS ≥4 times standard deviation noise (red). Two observations greater maximum-NRCS distinctly anomalous most specular observations maximum-NRCS intermediate maximum-NRCS greater less than two extreme ASRR observations anomalously specular originate from same terrain unit anomalous.Arecibo Observatory Green Bank Telescope Cassini subradar locations AO/GBT observations shown map Titan’s surface Fig. 2a, b. AO limited zenith angles <≈20° observe Titan during ≈1/3 oppositions over Saturn’s ≈29.5-year orbital period 2000 to 2008 not 2009 to 2027) AO-Titan geometry limits subradar locations latitude range ≈20° Titan’s southern tropical region 7 to 27°S). GBT shares limitations radar transmitter observations restricted echoes AO transmitter.Fig. 2Map of Titan monochrome swaths Cassini RADAR11 images purple red dots AO/GBT subradar locations red dots maximum-NRCS ≥4 standard deviations above noiseDot radii proportional to maximum-NRCS Colored tracks Cassini RADAR altimetry observations indicates NRCS high concentration of large red dots at ≈70–135°W 15–30°S Cassini altimetry observations high NRCS Boxed area enlarged colorbar applies to a and b.Figure 2 similar to Fig. 17 Black et al.2 four distinctions size AO/GBT dots indicates maximum-NRCS surface root-mean-square slope parameter radar scattering model spin models Titan AO/GBT Cassini datasets inconsistent in Black.2. Cassini dataset inconsistent observations used erroneous corrected locations datasets to current model International Astronomical AO/GBT Titan corrections correspond to relative changes of 3–14 km average change 11 km Titan’s complex surface varies Fig. 2 includes complete Cassini radar instrument dataset Titan phases 2004 to 2017. increase in AO/GBT locations improves time constraints on temporal changes fourth addition of Cassini RADAR altimetry observations NRCS Titan’s surface depends on incidence angleAO/GBT ASRR reflections surface at nadir (0° incidence incidence angles Cassini RADAR images range 10 to 77° Cassini RADAR altimetry observations near-nadir incidences due imperfect AO/GBT maximum-NRCS Cassini-altimetry NRCS nadir-to-nadir effects incidence angle mitigated systematic differences AO/GBT observations include backscatter non-nadir angles radar wavelength 12.6 cm different Cassini RADAR (λ = 2.2 AO/GBT subradar locations high maximum-NRCS anomalous maximum-NRCS ≥4 times standard deviation noise colored red Fig dots red large indicate locations low anomalously specular maximum-NRCS to AO/GBT large red dots at, 15–30°S region high NRCS in Cassini altimetry observations black areas polar regions Fig. 2a are Titan’s lakes/seas4 low NRCS contrast distinctive in Cassini RADAR no features observed at AO/GBT subradar locations hypotheses for AO/GBT ASRR include transient liquids dunes/interdunes paleolakes/paleoseasTransient liquids hypothesisSurface darkening following clouds likely from methane rain observed twice by Cassini first 2004 ≈80°S from subradar AO/GBT second 2010 5) postdated AO/GBT observations transient liquids from explain ASRR surface darkened 2010 rain to original spectrum >1 year after surface changes rain detectable after rainfall other clouds observed on Titan by Cassini Earth telescopes 14–17) not associated with surface changes Clouds over southern tropical region uncommon 2004 to 2008.Nine AO/GBT locations imaged by Cassini RADAR before after no surface changes Four RADAR images <1 year from AO/GBT 13 interval <2 years none show surface liquids Cassini RADAR images support hypothesis transient liquids constrain presence during AO/GBT observations low frequency rain events on Titan (two over long reversion timescale>1 low frequency clouds southern 2004 to 2008 15–17) RADAR image constraints suggest transient liquids not responsible for AO/GBT ASRR AO/GBT reflections depend on date latitude correlated separatedComparison with Cassini altimetry Fig. 3a–d indicates latitude not date Black et al.2 trend with date lost when y-axis changed from to maximum-NRCS trend date consistent with y-axis between AO/GBT Cassini temporal change hypothesis unlikely explain AO/GBT ASRR explanation temporal changes unlikely.Fig. 3NRCS function latitude date for AO/GBT Cassini altimetry observations.Panels a b show AO/GBT observations y-axis maximum-NRCS Panels c d show Cassini altimetry observations latitude date Black dots observations orange-edged cyan-filled rectangles weighted bins Error bars show standard deviation noise consistent peak cyan/orange bins at ≈23°S between AO/GBT Cassini latitude trends similar no agreement in trends with date specular AO/GBT observations depend on location not date.Dunes/interdunes hypothesisTitan has dunes analogous sand dunes grains hydrocarbons Dune fields cover millions kilometers tropical region main terraindune fields rough surfaces at dunes (≈1 grains (≈1 mm), interdune smooth AO/GBT observation wavelength (λ = 12.6 cm AO/GBT ASRR from Titan’s tropical region tempting relate ASRR to dunes interdune areas brighter than dunes Cassini altimetry no correlation Seven AO/GBT subradar locations dune/interdune regions in Cassini RADAR images AO/GBT maximum-NRCS range 0.56 to 1.09. maximum-NRCS values not anomalous include observations interest dune fields ruled out source AO/GBT ASRR.Hotei Tui Regiones paleolake/paleosea AO/GBT ASRR concentrated at ≈70–135°W, 15–30°S (Fig. 2); southeastern Hotei Regio Tui Regio southwestern in-between southern Xanadu Hotei Tui Regiones similar properties distinct terrains in Titan’s tropical region infrared spectrum topography nadir radar brightness hypothesized to be cryovolcanic terrains hypothesized be paleoseas/paleolakes distinct signature high brightness in Titan’s 5-μm atmospheric window.MacKenzie et al.24 mapped Titan’s 5-μm-bright areas using Cassini Visual Infrared Mapping Spectrometer observations map Hotei and Tui Regiones map Fig. 4a AO/GBT observations Fig. 4Hotei Tui Regiones Black areas Titan’s 5-μm-bright regions mapped MacKenzie et al.24. at ≈80°W, 25°S ≈120°W, 25°S AO/GBT observations plotted Fig. 2. Hotei Tui Regiones main source AO/GBT anomalously specular reflections exceptions to correlation Hotei labeled NE1 and NE2 discussed 18-km subradar track highest maximum-NRCS most specular AO/GBT observation Hotei Regio stereo topography AO/GBT reflection from bright area similar to paleolakes25 topographically low 18-km subradar track second highest maximum-NRCS specular AO/GBT observation in Tui Regio AO/GBT reflection from bright area morphologically similar to paleolakes Stereo topography cover this trackFigure 4a shows AO/GBT observation Hotei/Tui specular including two brighter two red dots one Hotei Tui two highest-NRCS ≥3.5 seven high intermediate maximum-NRCS 2 observations six in Hotei or Tui Regiones one on Hotei Regio’s two not near Regio two exceptions discussion deferred next Hotei and Tui Regiones primary source AO/GBT ASRR subradar locations two specular AO/GBT observations imaged by Cassini RADAR Figure 4b d AO/GBT subradar tracks Titan’s rotation over bright areas similar to paleolakes Titan’s polar Figure 4c bright areas in Hotei Regio topographic lows consistent paleolake hypothesis cover location Tui Hotei Regio high NRCS in Cassini altimetry observations 2 Tui not observed highest-NRCS Cassini altimetry observations Titan’s liquid-filled lakes/seas reflections 100 times brighter than echoes solid Cassini altimetry observations Hotei Regio not bright liquid-filled lakes/seas brighter than other terrainsHotei Regio higher NRCS tropical regions AO/GBT Cassini altimetry bright nadir radar paleolakes Titan’s north polar region brighter Cassini altimetry not liquid-filled lakes/seas29 add Cassini-altimetry-brightness paleolake/paleosea hypothesis for Hotei Regio Tui Regio cryovolcanic paleolake/paleosea hypotheses not mutually exclusive Hotei/Tui primary source AO/GBT ASRR independent geomorphology paleolakes/paleoseas cryovolcanoes correlations predict bright regions Tui Regio RADAR topographic lows intermediate NRCS in RADAR nadir observations smoother surface greater NRCS nadir radar high NRCS Hotei/Tui in AO/GBT Cassini altimetry attributed solely smoothness surface Fig. 4b subradar locations anomalous AO/GBT observations bright in Cassini RADAR images nonzero incidence angles 18° greater dielectric constant diffuse scattering increase NRCS all incidence angles dielectric constant surface depends on composition parameters porosity Hotei/Tui different composition spectrally distinct in Titan’s 5-μm atmospheric windowsurfaces Hotei/Tui differ Titan’s tropical region smoothness dielectric constant contribute to high maximum-NRCS AO/GBT observations differences due to lake/sea sediments deposited past liquid study Hotei/Tui geology chemistry Titan’s lake/sea sediments long evolution climate two exceptions correlation AO/GBT ASRR to Hotei/Tui first subradar point at ≈47°W, 13°S (NE1 Fig. acquired 1/30/2007 third-highest maximum-NRCS AO/GBT observations location similar to Hotei/Tui/paleolakes topographically low bright in 5-μm nadir radar observations edge of two low-resolution Cassini RADAR images noise inhibits interpretation observed by VIMS intersection equatorial-bright dark-blue spectral units not 5-μm-bright second exception subradar point at ≈163°W, 24°S (NE2 Fig. 4a acquired 11/3/2000 dismissed other bright reflections 2001 2008 bins Fig. 3 greater noise shorter integration times observation almost six standard deviations above noise be dismissedreservation about predicting terrain location similar to Hotei/Tui first observation AO/GBT campaign maximum NRCS not from central Doppler bin spectral spike one bin wide location not observed with RADAR observed by VIMS spectrum mix of equatorial-bright 5-μm-bright units Cassini mission ended 2017 locations imaged in high-resolution until another spacecraft explores Titan nadir-to-nadir comparison between AO/GBT Cassini altimetry observations leads conclusion definition of specular context liquids on planetary surface AO/GBT observations of Hotei and Tui Regiones specular spike in central Doppler bin opposite circular polarization spectrum high NRCS best fit by radar scattering models specular Cassini altimetry observations Hotei high NRCS specular terrains Cassini altimetry observations Titan’s lakes/seas indicate reflections from liquid surfaces more specular (brighter). echoes must be coherent27 received power decreases with total distance from transmitter to reflector to receiver noncoherent reflectionsTitan solar system search for oceans exoearths suggests liquids by specular reflections stringent definition recommend definition coherence of reflected electromagnetic waves than relative brightness incidence angle location polarization reflections specular quasi-specular Titan liquids expected planetary context reflections specular brightness polarization observed reflections from solid not liquid surfaces smooth liquid surfaces on Titan specular at microwave27 infrared 35,36) wavelengths AO/GBT subradar locations on updated from Black et al.2 using International Astronomical Union’s current (2015) spin model10 start end times of AO/GBT observation Supplementary Data 1) into SPICE system subradar points on Titan calculated using cspice_subpnt module Titan ellipsoid radius 2575 km. updated locations in Supplementary Data 1. Cassini RADAR images altimetry observations located using same spin model ellipsoid.Cassini RADAR altimetry observationsIn Figs. 2 and 3 only with incidence angle <1° included. cut removes echoes substantial pointing errors beginning end tracks changing cutoff affect resultsNRCS altimetry echo corrected deviations from nadir (0° incidence angle correction Cassini RADAR Users Guide NASA Planetary Data System varying incidence angle correction affect results.Supplementary information Peer Review File Supplementary Files Data | 50.8 | 0.55421 |
10.1038/s41467-020-15066-6 | PMC7062830 | Prematurely born babies need extra oxygen to survive, but this can cause damage to the eyes and lead to infant blindness. Here the authors show that this hyperoxia changes the metabolism of Müller cells in the retina such that they use up, rather than produce, glutamine and secrete excess ammonium. | Although supplemental oxygen is required to promote survival of severely premature infants, hyperoxia is simultaneously harmful to premature developing tissues such as in the retina. Here we report the effect of hyperoxia on central carbon metabolism in primary mouse Müller glial cells and a human Müller glia cell line (M10-M1 cells). We found decreased flux from glycolysis entering the tricarboxylic acid cycle in Müller cells accompanied by increased glutamine consumption in response to hyperoxia. In hyperoxia, anaplerotic catabolism of glutamine by Müller cells increased ammonium release two-fold. Hyperoxia induces glutamine-fueled anaplerosis that reverses basal Müller cell metabolism from production to consumption of glutamine. | IntroductionPremature infants require oxygen supplementation to survive, but excess oxygen causes retinovascular growth suppression that underlies the leading cause of infant blindness known as retinopathy of prematurity (ROP)1. We analyzed changes in intermediary metabolism during hyperoxia in human retinal endothelial cells (RECs) and human retinal Müller glia, which coexist through glutamine consumption and production, respectively2. Using a stable isotope labeling technique in human RECs, primary mouse Müller glial cells and a human Müller glial cell line (MIO-M1) in culture, here we show that hyperoxia decreases entry of glycolytic carbon into the tricarboxylic acid cycle (TCAC) and induces utilization of glutaminolytic carbon in Müller cells. In hyperoxia, catabolism of glutamine increased ammonium release by twofold. Hyperoxia induces glutamine-fueled anaplerosis that reverses basal Müller cell metabolism from production to consumption of glutamine.Retinal Müller cells are linked functionally to RECs through the synthesis of glutamine3. Müller cells convert lactate and aspartate to glutamine via the TCAC flux4. Glutamine produced by Müller cells is essential to REC proliferation and migration5,6. REC-specific glutamine lyase (GLS) knockout mice exhibit compromised blood vessels5. GLS1 and GLS2 isoforms are differentially expressed in RECs; the latter is more concentrated in endothelial tip cells necessary to new blood vessel formation7. Glutamine synthetase (GS) ablation prevents normal development of retinal vasculature, further confirming the importance of glutamine to endothelial cell growth and development8. Previous studies have also reported that RECs express the glutamine transporter SLC1A59.The high glycolytic rate in cancer cells and endothelial cells has been proposed to be beneficial in supporting quick production of ATP and to produce biosynthetic molecules for the serine and pentose phosphate pathways6. However, previous experiments in human umbilical vein endothelial cells (HUVECs) demonstrate that 90% of glucose-derived carbon is released from the cells as lactate and therefore very little amount of glucose is actually used for biomass synthesis. In contrast, measurement of glutamine consumption in HUVECs reveals that 90% of carbon from glutamine remains in endothelial cells implying its importance for biomass production5. This suggests that glutamine is required for proliferation and development of endothelial cells10. The dominant source of glutamine in the retina is the Müller cell.The effect of hyperoxia on metabolism has mostly been linked to the loss of the mitochondrial complexes of the electron transport chain11,12. These studies have used indirect measures of metabolic function such as oxygen consumption or absence of the mitochondrial complex subunit protein levels13. Here, we present the effects of hyperoxia on intermediary metabolism using stable isotope-labeled substrates to demonstrate (1) the loss of pyruvate-derived citrate production in hyperoxic Müller cells, (2) hyperoxia-induced consumption of glutamine to feed the TCAC through anaplerosis, and (3) increased release of ammonium by hyperoxic Müller cells in culture.ResultsHyperoxia inhibits glucose-derived glutamine productionSince Müller cells are known to produce glutamine for other cell types in the retina2, we used [13C6]glucose to compare synthesis of glutamine from glycolytic carbon in normoxia and hyperoxia (Fig. 1a). We first used MIO-M1, an immortalized cell line, to study the effect of hyperoxia on metabolism. MIO-M1 cells were isolated from human eye and behave like primary Müller cells14. MIO-M1 cells were first cultured in normoxia with [13C6]glucose media for 24 h to establish an isotopic steady state followed by either normoxia or hyperoxia for 8 h (Fig. 1b) or 24 h (Fig. 1d). We first ensured that the isotopic steady state was established for at least 20 h before treating part of the cells with hyperoxia. Isotopic steady state here means that the isotopic enrichment for the metabolites have reached stable labeling with little or no change after 24 h and stays consistent over the next 24 h of measurement in normoxic condition (Supplementary Figs. 1 and 2). Any changes reported in isotopic steady state should reflect relative increase or decrease in the fluxes in the pathways where these metabolites are located. The percentage decrease or increase in labeling has been calculated as difference in enrichment. The isotopic steady-state pilot experiments were performed only with [13C6]glucose. Metabolites of glycolysis and TCAC were in isotopic steady state post 23 h after adding the labeled glucose (Supplementary Fig. 1).Fig. 1[13C6]Glucose and [13C5]glutamine labeling of MIO-M1 and primary Müller cells.[13C6]Glucose and [13C5]glutamine labeling of MIO-M1 and primary Müller cells demonstrates hyperoxia-induced decreased flux from pyruvate to citrate and glucose to glutamate, but increased glutaminolytic flux into TCAC via oxidative decarboxylation, and decreased malic enzyme flux. MIO-M1 cells were cultivated in [13C6]glucose media for 24 h to reach isotopic steady state, then incubated in normoxia (21% O2) or hyperoxia (75% O2) for 8 or 24 h. a Schema of first round labeling of [13C6]glucose carbon through glycolysis and TCAC. b Fractional enrichment of 13C-labeled metabolites after 8 h hyperoxic treatment (n = 6, t-test p values: M3 lactate = 0.0001; M3 pyruvate < 0.0001; M2 citrate = 0.0006; M2 glutamate p < 0.0001). c Total (sum all MIDs) glutamate in normoxic vs. hyperoxic cells, after 8 h hyperoxic treatment (n = 6, t-test, mean ± SEM, p values = 0.0002). d Fractional enrichment of 13C-labeled metabolites after 24 h of hyperoxic treatment (n = 6, t-test p values: M3 lactate = 0.2365; M3 pyruvate = 0.2862, M2 citrate < 0.0001, M5 glutamate < 0.0001). e Mass isotopomer distributions of citrate and glutamate between normoxia and hyperoxia. Mass isotopomer distributions were corrected for natural isotope abundances for data represented in this figure and subsequent figures. f Schema of [13C5]glutamine carbon atoms transition through TCAC, malic enzyme, pyruvate carboxylase, and glycolytic pyruvate entry into TCAC. MIO-M1 or primary Müller cells were cultured in [13C5]glutamine media for 24 h, then incubated further in normoxia (21% O2) or hyperoxia (75% O2) for 24 h. g Fractional enrichment of 13C-labeled metabolites after 24 h hyperoxic treatment (n = 6, t-test p values: M3 lactate < 0.0001; M2 citrate < 0.0001; M5 citrate < 0.1198; M4/M5 citrate < 0.0001; M3 pyruvate < 0.0001; M5 glutamate < 0.0001; M4 fumarate < 0.0001; M4 aspartate < 0.0001). h Comparison of mass isotopomer distributions of citrate and glutamate between normoxia and hyperoxia. i Fractional enrichment of 13C-labeled metabolites in primary Müller cells after 24 h hyperoxic treatment (n = 6 per condition; t-test p values: M0 citrate < 0.027; M5 glutamate < 0.0001; M4 fumarate < 0.0007; M4 aspartate < 0.0001; M4 citrate = 0.0005; M5 citrate = 0.0016; M4/M5 citrate < 0.0001). j Fractional enrichment of 13C-labeled metabolites in primary astrocytes after 24 h hyperoxia. N normoxia, H hyperoxia, AUC area under curve. Box plots extend from 25 to 75th percentiles. Middle box line = median; whiskers represent minimal/maximal values for Fig. 1 and all subsequent box plots in Figs. 2 and 3. p values = two-sided unpaired t-test.After 8 h, hyperoxia caused slight but statistically significant decrease in isotopic enrichments of M3 lactate and M3 pyruvate by 7% and 8%, respectively, while M2 citrate and M2 glutamate were decreased by 75% and 80%, respectively (Fig. 1b). This shows that hyperoxia only moderately suppresses glycolysis, but drastically decreases the entry of glycolytic carbon into the TCAC. We also compared total glutamate amounts in hyperoxia vs. normoxia by summing all mass isotopomers of glutamate. There was 32% decrease in total glutamate (sum of the areas of all isotopomers) in hyperoxic condition as compared with normoxic condition (Fig. 1c).A longer exposure to hyperoxia (from 24 h to 48 h) resulted in similar M3 lactate and M3 pyruvate derived from [13C6]glucose, i.e., 1.5% and 4% increase, respectively, which were not statistically significant from the 8 h time point (Fig. 1d). In contrast, M2 citrate enrichment fell even further, down to 30% after 24-h of hyperoxia as compared with normoxia (Fig. 1d). M2 glutamate was completely lost from MIO-M1 cells in hyperoxic condition (Fig. 1d) confirming the oxygen-induced decreased pyruvate entry into the TCAC and the complete block of glutamate formation from glycolytic carbon. Mass isotopomer distributions (MIDs) of citrate and glutamate highlight the relationship between glycolytic carbon entry into the TCAC and glutamate synthesis in normoxia compared with hyperoxia (Fig. 1e).Hyperoxia induces glutamine consumption in Müller cellsOur [13C6]glucose labeling experiment indicated that MIO-M1 cells were unable to produce glutamate from glycolytic carbon in hyperoxia. However, their total glutamate levels remained only 30% lower than in normoxia suggesting that hyperoxic MIO-M1 cells derive their glutamate from an alternative source. We used [13C5]glutamine (Fig. 1f) to test if glutamine to glutamate conversion is increased in hyperoxia. M5 glutamate production from [13C5]glutamine was 14% higher in hyperoxia (Fig. 1g). Glutamine-derived M3 lactate and M3 pyruvate labeling decreased in hyperoxia, as did M2 and M4/M5 citrate (Fig. 1g). The M3 labeling of lactate and pyruvate are not related to glycolytic flux, but rather to isotopic exchanges via malic enzymes.The TCAC metabolites produced from [13C5]glutamine are partly M4 labeled. We, therefore, looked at the M4 labeling of fumarate and aspartate, and found increased labeling of M4 fumarate and M4 aspartate, both by 11%, (Fig. 1g), further confirming the relatively accelerated rate of glutaminolysis in hyperoxia. The overall increase in M5 glutamate (Fig. 1h) confirmed increased relative contribution of glutamine deamidation to the glutamate pool in hyperoxia.Oxidative vs. reductive carboxylation in MIO-M1In order to determine whether variations in labeling from M5 glutamine of TCAC intermediates in hyperoxia originated by oxidative decarboxylation or reductive carboxylation, we also measured M4 vs. M5 citrate labeled from [13C5]glutamine. Citrate can be produced from glutamate by reductive carboxylation, which converts glutamate to citrate by back flux via isocitrate dehydrogenase or by directly converting glutamate to α-keto-glutarate (α-KG) to succinyl-CoA by oxidative decarboxylation. The former reaction yields M5 citrate, whereas the latter proceeds through the forward reactions of the entire TCAC and hence produces M4 citrate (Fig. 1f). We found that the M4/M5 ratio of citrate was decreased in hyperoxic conditions; yet overall M5 citrate trended higher without statistical significance between normoxia and hyperoxia (Fig. 1g, h). Glutamine derived, M4 labeled TCAC metabolites such as fumarate and aspartate (Fig. 1g) were increased in hyperoxia compared with normoxia. At steady state, these findings imply that hyperoxia induces oxidative decarboxylation of α-KG yet the resulting oxaloacetate does not increase M4 citrate. Although the decrease of M4 citrate and increase in M4 fumarate confirms that α-KG is oxidatively decarboxylated, concurrently, hyperoxia might induce a change in malic enzyme flux (Fig. 1f). To get a reflection of the malic enzyme flux, we measured M3 labeled lactate and pyruvate formed by [13C5]glutamine and found that hyperoxia decreased M3 pyruvate and M3 lactate, implying higher malic enzyme relative flux in the normoxic condition compared with hyperoxic conditions (Fig. 1g). In addition, as expected, M2 citrate was higher in normoxic conditions, which can further enhance the M4 citrate produced from M2 acetyl-CoA and M2 oxaloacetate (Fig. 1g).Hyperoxia induces glutamine consumption in primary Müller cellsTo determine whether oxygen-induced glutamine-fueled anaplerosis described above in MIO-M1 cells also occurs in primary Müller cells, we isolated primary Müller cells from P11 (postnatal day 11) mice. ROP is caused by oxygen supplementation necessary to resuscitating severely premature infants that unfortunately creates retinovascular growth attenuation and vasoobliteration that is the hallmark of phase 1, which subsequently leads to profound ischemia and abnormal angiogenesis in phase 2. Therefore, we chose specifically the hyperoxic phase 1 in the experimental correlate of ROP, the murine oxygen-induced retinopathy model (OIR)15, to test both primary Müller cells and retinal explants. Hyperoxic phase 1 in the mouse model of OIR is from P7 to P12, and to be consistent with the model, we have only used cells or retinal explants from mice within the phase 1 of the model. Lysates of MIO-M1 cells, primary Müller cells, retinal explants, and primary human astrocyte cultures were compared by western blotting the levels of GS and cellular retinaldehyde binding protein (CRALBP) to ensure that cultured Müller cells and astrocytes were differentiated glia (Supplementary Fig. 3A, B). Primary Müller cells expressed similar ratios of CRALBP/GS as was found in glia from retinal explants. Primary Müller cells were cultured in 12-well plates and then incubated in the media containing [13C5]glutamine for 24 h to establish isotopic steady state. After 24 h, the cells were incubated either in normoxic or hyperoxic incubator for another 24 h. Intracellular metabolites were extracted and analyzed by GCMS. Like in MIO-M1 cells, we saw a similar reduction in the proportion of M4 citrate in hyperoxia, implying reduction in flux entering from glycolysis to the TCAC (Fig. 1i). In addition, we also found increased M5 glutamate, M4 fumarate, and M4 aspartate consistent with oxygen-induced increased glutaminolytic flux (Fig. 1i). These findings corroborate well our findings in MIO-M1 cells. Primary Müller cells differ from MIO-M1 cells in that they have a higher ratio of M4/M5 citrate in hyperoxia (Fig. 1i).Primary astrocytes increase glutamine catabolism in hyperoxiaTo further determine whether glutamine-fueled anaplerosis might occur in other types of glia within the central nervous system, we used [13C5]glutamine to study the effect of hyperoxia on glutaminolytic flux in primary cortical astrocytes. Cells were again cultured in six-well plates and then incubated in [13C5]glutamine for 24 h to establish isotopic steady state, after which cells were incubated into normoxic or hyperoxic incubators for another 24 h. As with cultured human MIO-M1 and primary mouse Müller cells, M5 glutamate was statistically significantly higher in hyperoxic condition, implying higher rate of glutaminolysis in hyperoxia (Fig. 1j). However, primary astrocytes exhibit an interesting difference in the accumulation of metabolites downstream of αKG. In contrast to all Müller cell lines, M4 aspartate and M4 fumarate were lower in hyperoxic condition (Fig. 1j). This observation can be explained by the fact that astrocytes but not Müller cells express the AGC1 transporter protein, which allows aspartate and glutamate exchange between mitochondria and cytosol4. The difference in M4 aspartate and M4 fumarate enrichments in response to hyperoxia also might be due to decrease in partial isotopic dilution by cytosolic aspartate derived from proteolysis.Glycolytic flux in RECs is unchanged in hyperoxiaIn order to compare and contrast the effect of hyperoxia on RECs to MIO-M1 or primary Müller cells, we repeated the same labeling experiments in RECs as were undertaken in MIO-M1 and/or primary Müller cells above. Using [13C6]glucose (Fig. 2a), RECs demonstrated almost no change in M2 citrate (Fig. 2b), implying no decrease in the contribution of M6 glucose to citrate in hyperoxia, a distinct difference when compared with MIO-M1 or primary Müller cells treated identically. Unlike MIO-M1 or primary Müller cells, M3 lactate, and M3 pyruvate enrichments were increased in hyperoxia by 5% and 8%, respectively, while M2 glutamate enrichment was significantly decreased by 41% in hyperoxic conditions (Fig. 2b). Overall, [13C6]glucose labeling of RECs indicated very little change in glycolytic flux entry into the TCAC, however, glutamate production was decreased. MIDs for lactate, citrate, and glutamate are provided in Fig. 2c.Fig. 2[13C6]Glucose and [13C5]glutamine labeling of retinal endothelial cells in culture.In response to hyperoxia, [13C6]Glucose and [13C5]glutamine labeling of retinal endothelial cells in culture demonstrates no change in pyruvate to citrate flux, (a net decrease in glutamate production from glycolytic carbon, increased glutaminolytic flux feeding into TCAC via oxidative decarboxylation route, and decreased malic enzyme flux. Retinal endothelial cells were cultivated in [13C6]glucose containing media for 24 h to reach isotopic steady state, following which they were either incubated further in normoxia (21% O2) or hyperoxia (75% O2) for 24 h. a Schematic of the first round of [13C6]glucose carbon atom transition through glycolysis and TCAC. b Fractional enrichment of 13C-labeled metabolites after 24 h of hyperoxic treatment (n = 6, t-test p values: M3 lactate = 0.0086; M3 pyruvate = 0.0138; M2 citrate = 0.7974; M2 glutamate < 0.0001). c Comparison of mass isotopomer distributions of lactate, citrate and glutamate between normoxia and hyperoxia. d REC cells were cultivated in [13C5]glutamine containing media for 24 h to reach isotopic steady state, following which they were either incubated further in normoxia (21% O2) or hyperoxia (75% O2) for 24 h. e Fractional enrichment of 13C-labeled metabolites after 24 h of hyperoxic treatment (n = 6, t-test p values: M4 citrate = 0.0002; M5 citrate < 0.0001; M5 glutamate < 0.0001; M4 fumarate = 0.0070; M4 aspartate = 0.7713). f Comparison of mass isotopomer distributions of citrate and glutamate between normoxia and hyperoxia. N normoxia, H hyperoxia.Glutamine utilization in RECs also increases in hyperoxiaWe next measured labeling of intermediates from M5 glutamine in RECs incubated in normoxia and hyperoxia (Fig. 2d). M5 glutamate enrichment from glutaminolysis was increased in hyperoxia by 7%; M4 fumarate was increased by 4% suggesting increased deamidation of glutamine and subsequent entry of glutamate into the TCAC but in contrast to Müller cells, M4 aspartate and M4 fumarate were unchanged (Fig. 2e). Furthermore, the changes in citrate labeling (M4, via oxidative decarboxylation vs. M5, via reductive carboxylation) demonstrated that hyperoxia inhibits reductive carboxylation in RECs (Fig. 2f). Glutamate labeling of REC cells clearly demonstrated increased utilization of glutamine in hyperoxia to produce TCAC compounds as evident from increased production of M5 glutamate and M4 citrate from glutamine. When examining label channeling through malic enzyme in RECs, there was little back flux of label from glutamine into pyruvate and lactate.Quantitative comparison of metabolites in MIO-M1 and RECsTo understand the importance of these differences in metabolic fluxes between MIO-M1 and RECs, in normoxia and hyperoxia, we quantified the total amount of metabolites ([sum of all mass isotopomer areas of individual metabolites]/[area of M internal standard]) in incubations of MIO-M1 and RECs. Glucose and glutamine levels were almost equal, implying that both the cell lines had equal availability of these carbon sources (Fig. 3a, b). However, the relative lactate/pyruvate ratio, which increases in aerobic glycolysis, was higher in RECs as compared with MIO-M1 cells (Fig. 3c). In addition, relative fumarate and aspartate levels were lower in RECs as compared with MIO-M1 cells, implying lower TCAC flux (Fig. 3e, f). Glutamate levels overall were reduced in MIO-M1 cells in hyperoxia (Fig. 3g).Fig. 3Total metabolite levels of retinal endothelial cells and MIO-M1 cells; retinal explants incubated with M5 glutamine or M1 acetate.a–i Comparison of total metabolite levels between retinal endothelial cells vs. MIO-M1 cells, in normoxia vs. hyperoxia; evidence of higher aerobic glycolysis in retinal endothelial cells as compared with MIO-M1 cells. j, k Retinal explants incubated with M5 glutamine. l, m Retinal explants incubated with M1 acetate. a–i Metabolites were extracted from confluent cells incubated with M5 glutamine, spiked with M5 ribitol internal standard, extracted and assayed by GC-MS. The sum of all MIDs were normalized to M5 ribitol. Data are presented as histograms with SEM (N = 6). a Glucose consumption, b glutamine, c lactate to pyruvate ratio, d citrate, e fumarate, f aspartate, g glutamate, and h ammonium, levels in RECs vs. MIO-M1 cells. i Retinal endothelial cell GLS1 expression is increased (n = 6 per condition, p value 0.0002, two-sided unpaired t-test). GLS1 expression is also increased in MIO-M1 cells (n = 6 per condition, p value 0.09, two-sided unpaired t-test). (Panels j–k) Pulse-chase experiment on retinal explants incubated with [13C5]glutamine. Retinal explants were incubated in 5 mM [13C5]glutamine for 5 min and media was changed with unlabeled 5 mM glucose and 5 mM lactate containing media. j M5 glutamine was consumed faster in hyperoxic condition (n = 3 per time point per condition, mean ± SEM). The data shows that higher glutamine utilization rate in retinal explants in hyperoxia compared to normoxia. k M5 glutamate had a similar trend as glutamine with higher rate of utilization (n = 3 per time point per condition, mean ± SEM). l Schematic of the two rounds of [13C1]acetate carbon atom transition through the TCAC. m Fractional enrichment of 13C-labeled citrate and aspartate 30 min or 1 h after incubation in 1-13C acetate containing media (n = 6 for M1 citrate (1 h) and M1 aspartate (1h); n = 5 for M2 aspartate (30 min) per condition). L/P ratio lactate/pyruvate ratio, nREC normoxic retinal endothelial cells, REC retinal endothelial cells, MIO-M1 Immortalized Müller cell line.In MIO-M1 cells, hyperoxia decreases pyruvate to citrate conversion, but increases glutamine derived fumarate and aspartate. Given that conversion of 1 mole of glutamine to α-KG releases 1 or 2 moles of ammonium, we next measured ammonium from MIO-M1 in the same cells cultured in hyperoxia for 48 h. Ammonium production, normalized to cell number, increased twofold in response to hyperoxia (Fig. 3h). Ammonium could also result from degradation of amino acids in the media. To test for this possibility, we compared ammonium levels in spent vs. unspent media. The ammonium concentration in unspent medium was negligble (Supplementary Fig. 4A–C).Hyperoxia upregulates GLS and glutamine catabolismWe next measured the expression levels of glutaminase, which has two isoforms GLS1 and GLS2. In our culture conditions, MIO-M1 cells expressed GLS1; GLS2 expression was not detectible. GLS1 expression was slightly upregulated by hyperoxia in both MIO-M1 cells and RECs (Fig. 3i). RECs also expressed higher levels of GLS1 in response to hyperoxia (Fig. 3i), whereas GLS2 expression was not detectable in RECs.To test the effect of hyperoxia on glutamine uptake, we performed a pulse-chase experiment. P11 retinas were incubated ex vivo in KRB containing 5 mM [13C5]glutamine for 5 min (Fig. 3j, k). P11 retinas were used because this postnatal day coincides with the hyperoxic phase 1 of the OIR model (Postnatal days 7–12). Following 5 min incubation in media containing 5 mM M5 glutamine, the media was aspirated; retinas were washed with PBS, then incubated in KRB containing 5 mM unlabeled glucose and 5 mM lactate.Following media change, the M5 enrichment of glutamine and glutamate slowly decreased, presumably by formation of unlabeled glutamine from unlabeled glucose and lactate. These dilution rates increased in hyperoxic conditions (Fig. 3j, k).Others have reported that acetate is metabolised in the retina in the same way as in the brain16–21. We used [1-13C] acetate to test the effect of hyperoxia on the labeling of citrate by Müller cells in cultured retinal explants. Retina from P10 mice (i.e. during the hyperoxic phase 1 of the OIR model) were dissected and explants cultured in DMEM containing 5 mM glucose, 2 mM [1-13C] acetate and 1 mM glutamine (Fig. 3l). The explants were exposed to normoxia or hyperoxia for 30 min or 1 h, before extraction of metabolites. As in isolated MIO-M1 cell culture experiments (Fig. 1b, d, g), the M1 enrichment of citrate in retinal explants was decreased in response to hyperoxia at 30 min and 1 h (Fig. 3m). Thus hyperoxia decreases the contributions of glucose, glutamine and acetate to the citric acid cycle in retina.No change in hyperoxia mitochondrial number or phospho-PDHIt was reported that hypoxia leads to mitochondrial autophagy22. We measured concentrations of COX-IV protein, which is a marker for mitochondria, along with β-actin loading control by western blotting. We found no difference in the amount of COX-IV (Supplementary Fig. 5A), suggesting that the mitochondrial number does not change in response to 24 h of hyperoxia.Pyruvate entry into the TCAC is regulated by pyruvate dehydrogenase kinase (PDK) protein, which inactivates pyruvate dehydrogenase (PDH) through phosphorylation. PDK expression is upregulated by HIF1α. Using western blot, PDH levels were measured in the MIO-M1 cells using β-actin as an internal control. PDH protein levels were not altered by hyperoxia (Supplementary Fig. 5B). We then measured the amount of phosphorylated PDH in the same samples, along with cells treated with HIF1α stabilizer FG-4592 as a positive control. Phospho-PDH levels were higher in HIF1α stabilized condition, but were same in normoxic and hyperoxic samples (Supplementary Fig. 5C, D). Therefore, the decrease in citrate formation in hyperoxia (Fig. 1b, d, g) is not secondary to decreased levels of PDH or phosphorylation of PDH.Immunohistochemistry of mitochondria (Supplementary Fig. 6A, B) demonstrates a difference in mitochondrial morphology that occurs in MIO-M1 cells in hyperoxia, changing the staining pattern without affecting overall mitochondrial density. In contrast, no significant difference was observed in hyperoxic vs. normoxic RECs (Supplementary Fig. 6C, D). Quantification of mitochondrial morphology in both RECs and MIO-M1 cells is provided in Supplementary Fig. 6E–H. It demonstrates equal density of mitochondria but definite change in morphology due to hyperoxia in MIO-M1 cells and RECs. The latter have clumped mitochondria in hyperoxia reflecting smaller fragmented and reduced branching of the mitochondrial networks. Pyruvate kinase, required for the conversion of phosphoenolpyruvate to pyruvate, has two isoforms present in the retina, PKM1 and PKM2. PKM1 and PKM2 are present in photoreceptors, whereas there is weak expression of PKM2 in the other layers of the retina4. PKM1 is expressed in MIO-M1 cells in culture, and its expression decreases in response to 24 h of exposure to hyperoxia (Supplementary Fig. 7).DiscussionOur data obtained using both mouse Müller glial cells and MIO-M1 (an immortalized human Müller glia-like cell line), demonstrate that Müller cells in hyperoxia have a higher flux through the glutaminolytic branch to the TCAC rather than from glycolysis (Fig. 1b, d, g). The reversal of glutamine production in Müller cells in normoxia to consumption in hyperoxia has profound implications for the mechanism of oxygen-induced retinovascular growth attenuation, the causation of ROP. First, energy-dependent production of glutamine is necessary for endothelial cell survival and proliferation. Second, glutamine synthesis detoxifies ammonium, in lung epithelia, in pericentral hepatocytes, and in brain2,23,24. Third, neurons rely on glutamine produced by Müller cells to make glutamate for synaptic transmission25. High oxygen saturations therefore may interrupt the basic functions of glia critical to endothelial cell metabolism and neuronal homeostasis. The oxygen-induced release of ammonium and inhibition of glutamine synthesis could contribute to the phenomenon of hyperammonemia in low birth weight/premature infants treated with oxygen supplementation26,27. The schema in Fig. 4 demonstrates how oxygen excess affects the overall flow of carbon and nitrogen in both Müller cells and RECs.Fig. 4Model depicting the re-programming of metabolism of Müller and retinal endothelial cells in response to hyperoxia.In normal conditions, Müller cells remove ammonium produced by other cell types in two steps by converting α-KG to glutamate and then glutamate to glutamine. Müller cells are the only cell type in retina known to fulfill all the glutamine requirements of the retina and are the only cells known to remove toxic ammonium formed by other cells in the retina. Retinal endothelial cells use aerobic glycolysis to produce energy and glutamine for growth. The lower panel shows the hyperoxic effect on metabolism of these cell types. In hyperoxia Müller cells stop producing glutamine and utilize it at increased rates for their energy needs by oxidative decarboxylation. Müller cells fulfill their energy needs by anaplerosis from glutamine when glycolytic flux entry to TCAC is blocked in hyperoxic conditions. This can lead to low glutamine levels in the retina for other cells types, including retinal endothelial cells. Glutaminolysis in retinal endothelial cells is also increased in response to hyperoxia, whereas, IDH back flux of citrate is decreased in response to hyperoxia. Overall hyperglutaminolysis in response to hyperoxia can lead to accumulation of toxic ammonium as well as a glutamine deficit. This would cause an overall metabolic imbalance leading to oxygen-induced growth suppression.The metabolic consequences of hyperoxia are simply not the opposite of hypoxia. HIF blocks entry of pyruvate into the TCAC28. Given that HIF concentration is decreased in hyperoxia, one would hypothesize that the flux through PDH should increase. This is not the case as shown in a previous report in lung epithelium29. We measured the PDH and phospho-PDH levels in hyperoxia vs. normoxia cells, and found no difference. There are four paralogs of PDK in humans, PDK1, PDK 2, PDK 3 and PDK430. Phosphorylation of PDK1 is under HIF regulation via 12 phosphorylation sites. Out of those, the most common one studied in HIF stabilized conditions is p-Ser23231. We specifically tested for this phosphorylated form and found no difference in hyperoxic vs. normoxic condition. Therefore, in Müller cells, there may be a mechanism other than phosphorylation of PDH controlling the entry of pyruvate into the TCAC. It is also possible that a different phosphorylation site on PDH is responsible for the PDH flux blockage in hyperoxic conditions.There are other cells that show high rates of glutamine catabolism. For example, cancer cells oxidize glutamine to α-KG; if a monoalleleic mutation in isocitrate dehydrogenase I is present, α-KG is further reduced to the oncometabolite 2-hydroxyglutarate32,33. Glutamine catabolism in hypoxia is regulated by multiple pathways, such as HIF1α and c-Myc; of these, hypoxia and glucose deprivation is considered to down-regulate c-Myc expression34–37. In conclusion, we demonstrate that hyperoxia stimulates glutamine-fueled anaplerosis in retinal Müller cells. Such glutamine deprivation may perturb the growth and functions of retinal endothelial cells.MethodsEthics Approval StatementAll treatments of animals (mice) were approved by the Cleveland Clinic Institutional Animal Care and Use Committee (IACUC) under study protocol number 2019-2183.ChemicalsAll the chemicals were purchased from Sigma-Aldrich (St. Louis, MO, USA) unless otherwise stated. The derivatization reagent MSTFA+1%TMCS was purchased from Thermo-scientific (Bellefonte, PA, USA). Stable isotope-labeled compounds were purchased from Cambridge Isotope Laboratories (Andover, MA, USA) and were reported as 99% pure. LC-MS grade methanol (HiPerSolv Chromanorm; BDH VWR International, Radnor, PA, USA) and chloroform (LiChrosolv; MilliporeSigma, Burlington, MA, USA) were used for all the metabolite extractions. DMEM media was purchased from Cleveland Clinic Media Lab. Endothelial cell media, Complete Classic Medium with Serum and CultureBoost™, was purchased from Cell Systems (Kirkland, WA, USA). Iodine was purchased from Acros organics (Bellefonte, PA, USA). CyQuant NF kit was purchased from Invitrogen (Bellefonte, PA, USA).Cell cultureRECs were procured from Cell Systems (Kirkland, WA, USA). The immortalized Müller cell line MIO-M1 was a kind gift from Dr. Limb38. DMEM (without glucose, glutamine and pyruvate) was supplemented with 15mM glucose, 5 mM glutamine, 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. In glucose labeling experiments, unlabeled glucose was replaced with [13C6]glucose. Similarly, in glutamine labeling experiment, unlabeled glutamine was replaced with [13C5]glutamine. This media was used directly for MIO-M1 or primary Müller cells. For the endothelial cell experiments, this media was diluted 1:1 with Classic Medium containing Serum and CultureBoost™ (Cell systems, Kirkland, WA, USA). For isotopic labeling studies, cells were cultured in 6-well plates to 90% confluence and then media were replaced with media containing labeled substrate(s). Following media change, cells were incubated for 24 h in a normoxic cell culture incubator set at 5% CO2 and 37 °C, to establish isotopic steady state. Once isotopic steady state was reached, cells were either incubated in a normoxic or hyperoxic (75% O2 with 5% CO2) incubator for 8 or 24 h as stated in the Results section.Human cortical astrocytes were cultured in Astrocyte medium (Sciencell; catalog no 1801). All cells purchased from Sciencell have the following information listed regarding informed consent: “Human tissue used for the isolation of primary cells is derived from donors who have signed informed consent by the donor themselves or an authorized agent acting on the donor’s behalf”. For labeling experiments, media was replaced with DMEM (without glucose, glutamine and pyruvate) supplemented with 15 mM glucose, 5 mM glutamine, 10 % FBS and 1% penicillin/streptomycin. For glucose labeling experiments, [12C6]glucose was replaced with [13C6]glucose. Similarly, for glutamine labeling experiments, [12C5]glutamine was replaced with [13C5]glutamine. Same media composition was used for primary Müller cell labeling experiment.Primary Müller cell isolation and cultureEnucleated eyes from P11-P12 mice were held in dissecting medium (EBSS + 1% Penicillin/Streptomycin) on ice. Dissected retinas were minced and transferred to a collection tube with minimal volume of dissecting medium, held on ice. Retinal tissue was dissociated using the Worthington Papain Dissociation System (#LK003150). Briefly, a vial of papain was reconstituted in EBSS equilibrated with sterile 95% O2 : 5% CO2 and incubated at 37 °C for 10 min until fully dissolved. A vial of DNase I was reconstituted in equilibrated EBSS, and 250 μl was added to the activated papain. This papain/DNase solution was added to the retinal tissue and triturated gently with a P1000 pipette. The tissues were equilibrated in sterile 95% O2 : 5% CO2 and incubated at 37 °C with 160 rpm agitation for a total of 35 min, then divided up into 15 min, 15 min, and 5 min incubations, triturating gently after each. The cell suspension was centrifuged at 550 × g for 5 min at room temperature and the pellet resuspended in ovomucoid inhibitor solution, prepared according to the manufacturer’s instructions. Cells were again pelleted at 550 × g for 5 min at room temperature and passed through a 30 μm filter after resuspending in 1 ml primary Müller glia cell culture media: DMEM-high glucose, without L-glutamine, with sodium pyruvate + 1% GlutaMax, 1% Penicillin/Streptomycin, and 10% FBS.Cells obtained from the dissociation were counted, and a trypan blue dye exclusion assay was used to determine a final viability of about 83%. Retinal cells were plated at a density of 1.1 × 106 cells/ cm2 onto plates coated with 10 μg/cm2 laminin and maintained and maintained at 37 °C in primary Müller glia cell culture media. The first media change occurred after four days, followed by media changes every two days. Cells were harvested after 8 days in culture, at which point they are primarily comprised of Müller glia.Metabolite extractionTo extract metabolites, the media were aspirated from the 6-well plates with adherent cells and cells were quickly washed with 1ml of room temperature normal saline. To each well, 400 µl of −20 °C cold methanol and 400 µl of cold water was added. Cells were scraped with a cell scraper, while maintaining plates on ice. The resulting cell suspension was added to 400 µl of −20 °C cold chloroform. Tubes were agitated on thermomixer for 20 min at 4 °C and 1400 rpm. Tubes were then centrifuged at 15000 × g for 5 min at 4 °C. 300 µl of upper layer containing polar metabolites was dried under vacuum at −4 °C. In glutamine labeling experiment, 10 µl of 0.05 mg/ml of [13C5]ribitol was added as internal and recovery standard. Dried extracts were derivatized with 25 µl of 40 mg/ml of methoxylamine in pyridine for 30 min on a thermomixer set at 45 °C and 1000 rpm. These samples were further derivatized with 75 µl of MSTFA + 1% TMCS.Ammonium extraction/derivatization for GC-MS measurementCells cultured in 6-well plates were either incubated in normoxia or hyperoxia. Ammonium was measured with the method described in Yang et al.39 with slight modifications40. After 48 h of incubation, 400 µl of media sample was taken and added to 400 µl of 10 N NaOH in 1.5 ml tube. To this preparation, 400 µl of formaldehyde (36.5–38% in water) was added and left overnight at room temperature in the fume hood. Formaldehyde reacts with ammonium to produce a stable compound hexamethylenetetramine (HMT). To this HMT preparation, 200 µl of 10% iodine was added to form the extractible HMT-iodine adduct. Tubes were incubated on a thermomixer at 40 °C for 15 min with 1000 rpm agitation speed. Tubes were cooled down to room temperature and the whole mixture was added to 5 ml of chloroform, to extract HMTiodine from the solution. Tubes were centrifuged at 1000 × g at room temperature for 1 min to separate phases. 100 µl of lower phase containing HMTiodine was taken out into a fresh tube and 1 µl of alkane mix (C10-C40 50 mg/L, even chain alkanes) standard was added to the solution as internal standard. Sample 1 µl was then injected into the mass spectrometer and m/z 140 corresponding to mass of HMT was used for quantitation. The HMT area was normalized to that of one of the alkanes with m/z 198, which corresponds to mass of C14H30.Retinal explant labelingMouse were euthanized with a lethal dose of ketamine/xylazine, eyes enucleated and retinas were dissected in high glucose DMEM. For labeling from [1-13C]acetate, retinas were briefly washed with room temperature normal saline and immediately added to 1 ml of culture media in 12-well plates. Media used for the tracer experiment was DMEM containing 5 mM glucose, 1 mM glutamine, 2 mM [1-13C]acetate, 10% FBS and 1% Penicillin/Streptomycin. Retinas were cultured for 30 min or 1 h, after which they were washed with room temperature normal saline and then snap frozen in liquid nitrogen. Metabolites from retina were extracted using 500 µl of −20 °C cold 80% methanol, brief sonication and centrifugation at 15,000 × g for 5 min at 4 °C. Supernatant 350 µl was taken into a fresh tube and 10 µl of 0.05 mg/ml of [13C5]ribitol was added to each sample before drying for GCMS measurements. Metabolites were dried, derivatized, measured and data analyzed as described for astrocytes.For glutamine labeling, retinas were dissected as described above for acetate labeling; dissections were performed in KRB containing 5 mM glucose and 1% penicillin/streptomycin. Retinas were then incubated in KRB media containing 5 mM of [13C5]glutamine, 10% dialyzed FBS and 1% penicillin/streptomycin for 5 min. After 5 min of incubation in labeling media, retinas were removed and washed with 1 ml of normal saline. After this step, 1 ml of KRB media containing 10% FBS, 1% Penicillin/Streptomycin, 5 mM lactate and 5 mM glucose was added to each well of the 12-well plates containing retinal explants. Metabolites were extracted from the retinal explants using methanol/chloroform/water protocol as described earlier. Ten microliters of 0.05 mg/ml [13C5]ribitol was added to 300 µl of supernatant. Samples were dried under vacuum and derivatized as described earlier and measured on GCMS.Cell countingMedia from the cells cultured in 6-well plates in normoxic or hyperoxic incubator were removed. Cells were washed with sterile 2 ml PBS. Cells were trypsinized with 100 µl of 0.05% trypsin containing 0.53 mM EDTA and resuspended in 1 ml of fresh media. Tubes containing cells were centrifuged at 300 × g at room temperature for 5 min. Media were removed and cells were resuspended in 1 ml of HBSS. Cells were again washed and fresh 1 ml of HBSS was added. 50 µl of this was added to 50 µl of 2x CyQuant cell counting reagent and samples were incubated at 37 °C in cell culture incubator for 1 h, following which fluorescence was measured on Victor X2 plate reader (PerkinElmer, Waltham, MA) with excitation at 485 nm and emission detection at 535 nm with exposure time of 0.1 s.GC-EI-MS analysis of metabolitesDerivatives from glucose and glutamine labeling experiment were measured on 7890B GC coupled to EI/CI 5977 mass selective detector Agilent mass spectrometer. Electron impact ionization was used for all the measurements. One µl of sample was injected in splitless mode. Full scan method was used for glucose labeling experiment, with scan widow from 50 to 800 m/z. Front inlet heater was set to 250 °C, septum purge flow 3 ml/min. Injections were made in splitless mode onto HP-5ms inert 30 m × 250 µm × 0.25 µm column connected to MSD. Samples were run in a constant flow mode with flow of helium set to 1.1 ml/min. The oven temperature was set to a starting 60 °C for 1 min and then ramped at 10 °C/min to 325 °C with a final hold time of 10 min. The column was re-equilibrated at 60 °C for 1 min.Primary Müller cells and primary astrocyte metabolites were assayed separately on a similar column, DB-5 ms GC Column 30 m × 0.25 mm × 0.25 µm with DuraGuard 10 m. Scan method was similar with solvent delay of 6.6 min and Selected ion monitoring (SIM) method used was also similar with solvent delay of 6.6 min. Ions used in SIM method were same with different time windows to account for differences in retention time on the new column.SIM method was used for metabolite measurement for glutamine labeling assays. SIM windows used are provided below in Table 1.Table 1Single ion monitoring (SIM) windows.Time windowMetaboliteIons (m/z)6.0–6.16Pyruvate173–1786.16–9.00Lactate218–2239.0–9.65Proline243–2529.65–10Succinate246–25210–12.06Fumarate244–25012.06–12.40Malate334–34012.40–13.57Aspartate333–33913.57–14.90Glutamate347–35414.90–15.29U-13C Ribitol323, 336–34215.29–15.80Glutamine346–35315.80–16.60Citrate464–47416.60–37.50Glucose553–561Time window, ions measured between these time limits; Metabolites, name of the compounds; Ions (m/z), mass-to-charge ratio measured.Calculations of isotopic enrichments and concentrations of metabolitesMetabolite Detector and Mass Hunter software were used for deconvolution and annotation of the metabolites41. IsoCor was used to correct for natural abundance of isotopes and to derive true mass isotopomer distribution for each metabolite analyzed42. Unlabeled extracts from MIO-M1 cells and authentic metabolite standards were used to annotate metabolites in all the samples. NIST 2017 library was further used to confirm the compounds in the extracts from all cell lines and retinal explants.Western blotMIO-M1 cells were treated similarly as for metabolite labeling experiment, except the media used only contained unlabeled glucose and glutamine. The cells were scraped in RIPA buffer (Sigma-Aldrich, St. Louis, MO, USA) with protease inhibitor cocktail Complete and phosphatase inhibitor cocktail PhosSTOP (both from Roche Diagnostics, Mannheim, Germany). Protein concentration was measured with Pierce BCA protein assay reagent (Thermo-scientific, Bellefonte, PA, USA). Protein 8 or 16 µg were added 20 mM DTT (final concentration), volume was adjusted to 30 µl with 2× Sample Tris-Glycine SDS buffer from Novex Invitrogen (Bellefonte, PA, USA), and were heated at 95 °C for 3 min. The quantity of proteins in all the samples used in a single Western blot were equal. 25 µl of this sample preparation was then loaded onto the Tris-glycine gel 4–20% precast gels (Invitrogen, Bellefonte, PA, USA) and ran at 125 V for 1 h 45 min. Proteins were transferred to PVDF membrane using a wet transfer method and membrane was dried in for 1 h by sandwiching the membrane between two chromatography papers. Membrane was then blocked with Odyssey buffer (LI-COR, Lincoln, Nebraska, USA), for 1 h at room temperature, following which membranes were incubated overnight with primary antibody at a dilution recommended by the vendor. After overnight incubation with primary antibody, membrane was washed with TBST buffer and incubated with secondary antibody for 1 h at room temperature in dark.Following primary antibodies were used:PDH E1 component subunit alpha rabbit polyclonal Anti-PDH antibody catalog number ABS2082 from Millipore (Burlington, MA, USA). Dilution used 1:1000.PhosphoDetectTM Anti-PDH-E1α (pSer232) Rabbit pAb catalog number AP1063 from Millipore (Burlington, MA, USA). Dilution used 1:1000.COXIV antibody catalog number 4844 from Cell Signaling Technology (Danvers, MA, USA) was used to probe mitochondria in the samples. Dilution used 1:1000.β-Actin mAbs catalog #3700 and #4970 Cell Signaling Technology (Danvers, MA, USA) as loading control. Dilution used 1:2000.GS mouse mAb #610517 (BD Bioscience, San Jose, CA)—marker of Müller glia. Dilution used 1:1000.CRALBP rabbit pAb UW55 originally raised by Jack Saari (University of Washington) and gifted to us by John W. Crabb (Cleveland Clinic). Dilution used 1:1000.Secondary antibodies used were:IRDye® 680RD Donkey (polyclonal) anti-mouse IgG (H+L), catalog number 925-68072 from LI-COR (Lincoln, Nebraska, USA) Dilution used 1:1000.IRDye® 800CW Donkey (polyclonal) anti-Rabbit IgG (H+L), catalog number 925-32213 from LI-COR (Lincoln, Nebraska, USA) Dilution used 1:1000.Mitochondria stainingMitochondria morphology was revealed using Invitrogen CellLight Reagent BacMam 2.0 Mitochondria-GFP (ThermoFisher), which represents a baculoviral vector construct encoding a signal peptide, i.e., the leader sequence of E1α pyruvate dehydrogenase, fused to emerald GFP. MIO-M1 or REC cells were grown at 70% confluence in four-well Nunc Lab-Tek II Chamber Slides (ThermoFisher) were transfected overnight with Mitochondria-GFP at 3 × 105 or 10 × 105 particles/well, respectively. Next day cells continued to culture in the regular humidified CO2 incubator or were placed in the humidified hyperoxic cell culture chamber with 75% oxygen and 5% CO2 conditions controlled by ProOx Model C21 (Biospherix, Parish, NY) and connected to the sources of CO2 and O2. After overnight incubation cells were quickly rinsed with warm PBS, fixed in warm freshly prepared 4% paraformaldehyde in PBS, and mounted in VectaShield with DAPI (Vector Labs, Burlingame, CA). Fluorescent images were taken using a Zeiss AxioImager.Z1 fluorescent microscope equipped with 63× objective, AxioCam 503 mono camera and ApoTome.2 adapter. Quantitative analyses of mitochondrial networks were performed using two custom macros for NIH ImageJ v1.40 g image analysis software Mitophagy and MiNA43,44.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary | nature communications | [
"Article"
] | [
"Metabolomics",
"Molecular medicine"
] | infants require oxygen excess oxygen causes retinovascular growth suppression infant blindness retinopathy prematurity analyzed metabolism during hyperoxia in human retinal endothelial cells retinal Müller glia glutamine consumption production hyperoxia decreases glycolytic carbon induces utilization glutaminolytic carbon Müller cells catabolism glutamine ammonium release twofold induces glutamine-fueled anaplerosis reverses Müller cell metabolism to consumption Müller cells linked to RECs synthesis convert lactate aspartate to glutamine Glutamine essential to REC proliferation REC-specific glutamine lyase) knockout mice exhibit compromised blood GLS1 GLS2 isoforms differentially expressed in RECs concentrated endothelial tip cells new blood vessel Glutamine synthetase ablation prevents development retinal vasculature importance glutamine to endothelial cell growth RECs express glutamine transporter SLC1A59 high glycolytic rate in cancer cells endothelial cells production ATP biosynthetic molecules for serine pentoseexperiments in human umbilical vein endothelial cells 90% glucose-derived carbon released as lactate little glucose used for biomass synthesis glutamine consumption 90% carbon glutamine remains in importance for biomass glutamine required for proliferation development dominant source glutamine in retina Müller cell hyperoxia on metabolism linked to loss mitochondrial complexes studies used indirect measures oxygen consumption mitochondrial complex subunit protein effects hyperoxia on metabolism-labeled substrates loss of pyruvate-derived citrate production in hyperoxic Müller cells-induced consumption glutamine increased release of ammonium Müller cells inhibits glucose-derived glutamine Müller cells produce glutamine used [13C6]glucose to compare synthesis glutamine in normoxia hyperoxia used MIO-M1 cell line effect hyperoxia metabolism isolated like primary Müller cultured in normoxia with [13C6]glucose for 24 h isotopic steady state followed by normoxia or hyperoxia for 8 h 24ensured isotopic steady state 20 h before treating cells hyperoxia enrichment stable labeling change after 24 h consistent 24 h normoxic Figs 1 2) changes steady state reflect increase percentage decrease labeling calculated as difference enrichment pilot experiments with [13C6]glucose Metabolites glycolysis TCAC steady state post 23 h after glucose[13C6]Glucose [13C5]glutamine labeling MIO-M1 primary Müller cells decreased flux pyruvate to citrate glucose to glutamate increased glutaminolytic flux into TCAC decreased malic enzyme flux-M1 cells cultivated in [13C6]glucose 24 h state incubated normoxia (21% O2) or hyperoxia (75% O2) 8 or 24 h labeling [13C6]glucose glycolysis TCAC Fractional enrichment 13C-labeled metabolites after 8 h hyperoxic treatment = 6 values M3 lactate = 0.0001 pyruvate < 0.0001 citrate = 0.0006 glutamate < Total glutamate normoxic vs8 h treatment 6 enrichment 13C metabolites 24 h 6 M3 lactate 0.2365 M3 pyruvate 0.2862 M2 citrate 0.0001 M5 glutamate distributions citrate glutamate normoxia hyperoxia corrected [13C5]glutamine TCAC malic enzyme pyruvate carboxylase glycolytic pyruvate cells cultured [13C5]glutamine 24 h normoxia (21% hyperoxia (75% O2) 24 h Fractional enrichment 13C metabolites 24 h treatment 6 M3 lactate 0.0001 M2 citrate 0.0001 M5 citrate 0.1198 M4/M5 citrate 0.0001 M3 pyruvate 0.0001 M5 glutamate 0.0001 M4 fumarate 0.0001 aspartate distributions citrate glutamate normoxia hyperoxia Fractional enrichment 13C metabolites cells 24 h treatment 6 M0 citrate < 0.027 M5 glutamate 0.0001 M4 fumarate 0.0007 aspartate 0.0001 citrate M5 citrate M4/M5 citrateFractional enrichment 13C metabolites astrocytes after 24 h hyperoxia N normoxia H hyperoxia AUC area curve Box plots 25 to 75th percentiles Middle box line median whiskers minimal/maximal values Fig. 1 plots Figs. 2 3. values two-sided t-test 8 h hyperoxia enrichments M3 lactate M3 pyruvate 7% 8% M2 citrate M2 glutamate decreased 75% 80% hyperoxia moderately suppresses glycolysis decreases entry glycolytic carbon TCAC compared glutamate amounts hyperoxia normoxia 32% decrease total glutamate hyperoxic longer exposure hyperoxia 24 h to 48 h M3 lactate M3 pyruvate [13C6]glucose 1.5% 4% increase not significant 8 h M2 citrate enrichment fell 30% after 24-h hyperoxia M2 glutamate lost from MIO-M1 cells hyperoxic decreased pyruvate entry TCAC block glutamate formation glycolytic distributions citrate glutamate highlight relationship glycolytic carbon entry glutamate synthesis normoxiainduces glutamine consumption in Müller [13C6]glucose labeling experiment MIO-M1 cells produce glutamate glycolytic carbon hyperoxia total glutamate levels 30% lower normoxia hyperoxic cells derive glutamate alternative source used [13C5]glutamine glutamate conversion hyperoxia M5 glutamate production from 14% higher Glutamine-derived M3 lactate M3 pyruvate labeling decreased M2 M4/M5 citrate labeling not related to glycolytic flux isotopic exchanges malic enzymes TCAC metabolites from [13C5]glutamine partly M4 labeled increased labeling fumarate 11% accelerated rate glutaminolysis hyperoxia increase in M5 glutamate increased contribution glutamine deamidation glutamate pool hyperoxia.Oxidative vs reductive carboxylation in MIO variations labeling M5 glutamine measured M4 vs. M5 citrate from [13C5]glutamineCitrate produced from glutamate by reductive carboxylation converts to isocitrate dehydrogenase or to α-keto-glutarate succinyl-CoA oxidative decarboxylation former yields M5 citrate latter proceeds produces M4 citrate M4/M5 ratio citrate decreased in hyperoxic conditions M5 citrate higher without significance normoxia hyperoxia Glutamine derived M4 TCAC metabolites fumarate aspartate increased in hyperoxia steady state hyperoxia induces oxidative decarboxylation of α-KG oxaloacetate increase M4 citrate decrease M4 citrate increase M4 fumarate α-KG decarboxylated hyperoxia might malic enzyme flux measured M3 lactate pyruvate]glutamine hyperoxia decreased M3 pyruvate M3 lactate higher malic enzyme flux normoxic M2 citrate higher in normoxic conditions M4 citrate from M2 acetyl-CoA M2 oxaloacetateinduces glutamine consumption in primary Müller oxygen-induced glutamine-fueled anaplerosis in MIO-M1 cells primary Müller cells isolated from P11 11 mice ROP caused by oxygen supplementation premature infants creates retinovascular growth attenuation vasoobliteration leads to abnormal angiogenesis in phase 2. chose hyperoxic phase 1 retinopathy test primary Müller cells retinal explants Hyperoxic phase 1 P7 to P12 used cells retinal explants from mice phase 1 MIO-M1 cells primary Müller cells retinal explants human astrocyte cultures compared by levels GS cellular protein) Primary Müller cells expressed similar ratios of CRALBP/GS in glia retinal explants cultured in 12-well plates incubated in media [13C5]glutamine for 24 h state After 24 h incubated in normoxic or hyperoxic 24 h Intracellular metabolites extracted analyzed by GCMS similar reduction in M4 citrate in hyperoxia reduction flux glycolysis to TCACfound increased M5 glutamate M4 fumarate M4 aspartate oxygen-induced glutaminolytic flux findings corroborate MIO-M1 cells Primary Müller cells differ higher ratio M4/M5 citrate in hyperoxia astrocytes increase glutamine catabolism in hyperoxiaTo glutamine-fueled anaplerosis glia central used [13C5]glutamine effect hyperoxia glutaminolytic flux primary cortical astrocytes Cells cultured six-well plates incubated in [13C5]glutamine 24 h incubated normoxic hyperoxic incubators 24 h M5 glutamate higher in hyperoxic condition higher rate glutaminolysis primary astrocytes difference accumulation metabolites downstream αKG M4 aspartate M4 fumarate lower in hyperoxic astrocytes express AGC1 transporter protein aspartate glutamate exchange mitochondria difference in M4 aspartate fumarate enrichments hyperoxia due to decrease partial isotopic dilution cytosolic aspartate proteolysisGlycolytic flux in RECs unchanged in hyperoxia RECs MIO-M1 Müller cells repeated labeling experiments [13C6]glucose RECs no change M2 citrate no decrease M6 glucose citrate hyperoxia MIO-M1 primary Müller cells M3 lactate M3 pyruvate enrichments increased hyperoxia by 5% 8% M2 glutamate enrichment decreased by 41% in hyperoxic conditions [13C6]glucose labeling little change glycolytic flux TCAC glutamate production decreased MIDs for lactate citrate glutamate in Fig. 2c[13C6]Glucose [13C5]glutamine labeling retinal endothelial cells culture hyperoxia no change pyruvate to citrate flux decrease glutamate production increased glutaminolytic flux decreased malic enzyme flux cells cultivated in [13C6]glucose media for 24 h incubated in normoxia (21% O2) or hyperoxia (75% O2) for 24 h [13C6]glucose carbon atom transition through glycolysis TCACenrichment 13C metabolites after 24 h hyperoxic treatment 6 M3 lactate M3 pyruvate 0.0138 M2 citrate 0.7974 M2 glutamate < Comparison lactate citrate glutamate normoxia hyperoxia REC cells cultivated [13C5]glutamine media 24 h incubated normoxia (21% O2) or hyperoxia (75% O2) 24 h Fractional enrichment 13C metabolites after 24 h hyperoxic treatment 6 M4 citrate 0.0002 M5 citrate < 0.0001 M5 glutamate < 0.0001 M4 fumarate = 0.0070 M4 aspartate = 0.7713) Comparison distributions citrate glutamate normoxia hyperoxia utilization increases labeling intermediates M5 glutamine RECs normoxia hyperoxia M5 glutamate enrichment increased 7% M4 fumarate increased 4% deamidation glutamine TCAC M4 aspartate fumarate unchanged changes citrate labeling hyperoxia inhibits reductive carboxylation Glutamate labeling increased utilization glutamine hyperoxia production M5 glutamate M4 citrateexamining label channeling malic enzyme in RECs little back flux from glutamine into pyruvate lactate comparison metabolites in MIO-M1 differences metabolic fluxes MIO-M1 RECs quantified total metabolites in incubations MIO-M1 RECs Glucose glutamine levels equal availability carbon sources (Fig. 3a lactate/pyruvate ratio higher in RECs MIO-M1 fumarate aspartate levels lower in lower TCAC flux Glutamate levels reduced in MIO-M1 cells in hyperoxia (Fig. metabolite levels of retinal endothelial cells MIO-M1 cells incubated with M5 glutamine M1 acetate Comparison metabolite levels MIO-M1 normoxia hyperoxia higher aerobic glycolysis in retinal cells explants incubated with M5 glutamine M1 acetate Metabolites extracted from confluent cells M5 glutamine spiked with M5 ribitol assayed by-MS MIDs normalized to M5 ribitol Data presented as histograms with SEM (N = 6)Glucose consumption glutamine lactate pyruvate ratio citrate fumarate aspartate glutamate ammonium RECs MIO-M1 cells Retinal endothelial cell GLS1 expression increased 6 0.0002 increased MIO-M1 cells 6 0.09 Pulse-chase experiment retinal explants incubated [13C5]glutamine 5 mM [13C5]glutamine 5 min changed unlabeled 5 mM glucose 5 mM lactate M5 glutamine consumed faster hyperoxic condition = 3 higher glutamine utilization rate hyperoxia M5 glutamate similar higher utilization 3 [13C1]acetate transition TCAC Fractional enrichment 13C-labeled citrate aspartate 30 min 1 h after incubation 1-13C acetate media (n = 6 for M1 citrate M1 aspartate 5 M2 aspartate L/P ratio lactate/pyruvate ratio normoxic retinal cells MIO-M1 cell line hyperoxia decreases pyruvate citrate conversion increases glutamine fumarate aspartateconversion 1 mole glutamine to α-KG releases 1 or 2 moles ammonium measured ammonium from MIO-M1 cells hyperoxia 48 h Ammonium production increased twofold hyperoxia Ammonium from degradation amino acids media compared ammonium levels spent unspent media ammonium concentration unspent negligble upregulates GLS glutamine measured expression glutaminase GLS1 GLS2. MIO-M1 cells expressed GLS1 GLS2 not detectible GLS1 expression upregulated by hyperoxia in MIO-M1 cells RECs RECs expressed higher levels GLS1 GLS2 not detectable effect hyperoxia glutamine uptake pulse-chase experiment P11 retinas incubated in KRB 5 mM [13C5]glutamine 5 min hyperoxic phase 1 OIR incubation media aspirated retinas washed with PBS incubated in KRB 5 mM unlabeled glucose 5 mM lactate M5 enrichment glutamine decreased unlabeled glutamine lactate dilution rates increased in hyperoxic conditionsacetate metabolised retina used [1-13C] acetate effect hyperoxia labeling citrate cells retinal explants Retina P10 mice hyperoxic phase 1 dissected explants cultured in DMEM 5 mM glucose 2 mM [1-13C] acetate 1 mM glutamine explants exposed to normoxia or hyperoxia 30 min or 1 h before extraction metabolites M1 enrichment citrate decreased hyperoxia 30 min 1 h hyperoxia decreases contributions glucose glutamine acetate to citric acid cycle retina change hyperoxia mitochondrial number-PDHIt leads mitochondrial autophagy22 measured concentrations COX-IV protein marker β-actin loading control western blotting no difference COX-IV mitochondrial number change 24 h hyperoxia.Pyruvate entry TCAC regulated by dehydrogenase) protein inactivates PDK expression upregulated by HIF1α PDH levels measured MIO-M1 cells β-actin PDH protein levels not altered by hyperoxiameasured phosphorylated PDH samples cells treated HIF1α stabilizer FG-4592 Phospho-PDH levels higher in HIF1α stabilized condition same normoxic hyperoxic samples 5C decrease citrate formation hyperoxia not secondary to decreased PDH phosphorylation.Immunohistochemistry mitochondria Fig 6A B difference mitochondrial MIO-M1 cells hyperoxia staining pattern without affecting mitochondrial density no significant difference hyperoxic normoxic RECs 6C mitochondrial morphology MIO-M1 cells Fig. 6E–H equal density change due hyperoxia-M1 cells RECs clumped mitochondria hyperoxia smaller fragmented reduced branching networks Pyruvate kinase conversion phosphoenolpyruvate to pyruvate two isoforms retina PKM1 PKM2. present photoreceptors weak expression PKM2 other layers PKM1 expressed in MIO-M1 cells decreases 24 h exposure hyperoxia Fig 7)data mouse Müller glial cells MIO-M1 demonstrate Müller cells in hyperoxia higher flux through glutaminolytic branch to TCAC glycolysis (Fig. 1b reversal of glutamine production Müller normoxia to hyperoxia for oxygen-induced retinovascular growth attenuation ROP energy production glutamine necessary for endothelial cell survival proliferation glutamine synthesis detoxifies ammonium lung epithelia hepatocytes neurons rely on glutamine for synaptic High oxygen saturations interrupt functions glia metabolism homeostasis oxygen-induced release ammonium inhibition glutamine synthesis hyperammonemia in low birth weight/premature infants oxygen Fig. 4 oxygen excess affects flow carbon nitrogen in Müller cells RECs. 4Model re-programming metabolism Müller retinal endothelial cells hyperoxia Müller cells remove ammonium α-KG to glutamate glutamate glutamine fulfill glutamine requirements remove toxic ammonium cells use aerobic glycolysis produce energy glutamine for growth hyperoxic effect on metabolism In hyperoxia Müller cells stop glutamine utilize for energy by oxidative decarboxylationMüller cells energy from glutamine when glycolytic flux to TCAC blocked in hyperoxic conditions to low glutamine levels in retina endothelial Glutaminolysis increased IDH back flux citrate decreased hyperglutaminolysis accumulation ammonium glutamine deficit metabolic imbalance oxygen-induced growth suppression metabolic consequences hyperoxia not opposite hypoxia HIF blocks pyruvate into TCAC28 HIF concentration decreased flux PDH increase not case measured PDH phospho-PDH levels in hyperoxia normoxia cells no difference four paralogs of PDK in PDK1 2 3 PDK430 Phosphorylation of PDK1 under HIF via 12 phosphorylation sites common p-Ser23231 no difference in hyperoxic normoxic Müller cells mechanism other than phosphorylation PDH entry pyruvate TCAC different phosphorylation site on PDH for PDH flux blockage in hyperoxic other cells show high glutamine catabolism cancer cells oxidize glutamine to α-KG mutation in isocitrate dehydrogenase I α-KG to oncometabolite 2-hydroxyglutarate32Glutamine catabolism hypoxia regulated by pathways HIF1α c-Myc hypoxia glucose deprivation down-regulate c-Myc hyperoxia stimulates glutamine-fueled anaplerosis in retinal Müller cells glutamine deprivation growth functions retinal endothelial cells treatments approved Cleveland Clinic Institutional Animal Care Committee study protocol 2019-2183 chemicals purchased from Sigma-Aldrich Louis derivatization reagent MSTFA+1%TMCS Thermo-scientific Stable isotope-labeled compounds Cambridge Isotope Laboratories 99% pure-MS grade methanol chloroform used for metabolite extractions DMEM media Cleveland Clinic Media Lab Endothelial cell media Classic Medium Serum CultureBoostTM Cell Systems Iodine Acros organics CyQuant kit Invitrogen (Bellefonte cultureRECs from Cell Systems (Kirkland immortalized Müller cell line MIO-M1 from Dr. DMEM supplemented with 15mM glucose 5 mM glutamine 10% fetal bovine serum) 1% penicillin/streptomycinglucose unlabeled glucose replaced with [13C6]glucose glutamine [13C5]glutamine media used MIO-M1 primary Müller cells endothelial cell experiments diluted 1:1 with Classic Medium Serum CultureBoostTM Kirkland WA isotopic labeling cells cultured 6-well plates 90% confluence replaced labeled substrate cells incubated 24 h normoxic incubator 5% CO2 37 °C isotopic steady state incubated normoxic or hyperoxic (75% O2 5% CO2) incubator 8 or 24 h cortical astrocytes cultured Astrocyte medium (Sciencell 1801) cells consent tissue donors informed consent labeling media replaced with DMEM glucose glutamine 15 mM glucose 5 mM glutamine 10 % FBS 1% penicillin/streptomycin glucose [12C6]glucose replaced [13C6]glucose [12C5]glutamine [13C5]glutamine Same primary Müller cell labeling cell isolation cultureEnucleated eyes P11-P12 mice dissecting medium (EBSS + 1% Penicillin/Streptomycin iceDissected retinas transferred collection tube held ice Retinal tissue dissociated Worthington Papain Dissociation System vial papain reconstituted 95% O2 5% CO2 incubated 37 °C 10 min vial DNase I reconstituted 250 μl added activated papain retinal tissue triturated P1000 pipette tissues 95% O2 5% CO2 incubated 37 °C 160 rpm agitation 35 min divided 15 5 min incubations cell suspension centrifuged 550 × g 5 min resuspended ovomucoid inhibitor solution Cells pelleted 550 × g 5 min passed 30 μm filter 1 ml Müller glia cell culture media-high glucose L-glutamine sodium pyruvate 1% GlutaMax 1% Penicillin/Streptomycin 10% FBS.Cells counted trypan blue dye exclusion assay viability 83% Retinal cells plated 1.1 × 106 cells/ cm2 coated 10 μg/cm2 laminin maintained 37 °C Müller glia culture media first change four days two days Cells harvested after 8 days Müller gliaMetabolite media aspirated from 6-well plates washed with 1ml room saline 400 μl −20 °C methanol cold water added Cells scraped cell suspension added to 400 μl −20 °C chloroform Tubes agitated thermomixer 20 min 4 °C 1400 rpm centrifuged 15000 g 5 min 4 °C 300 μl metabolites dried vacuum −4 °C 10 μl 0.05 mg/ml [13C5]ribitol added Dried extracts derivatized with 25 μl 40 mg/ml methoxylamine pyridine 30 min thermomixer 45 °C 1000 rpm derivatized 75 μl MSTFA + 1% TMCS.Ammonium extraction/derivatization measurementCells 6-well plates incubated normoxia hyperoxia Ammonium measured Yang et al.39 48 h 400 μl media sample added to 400 μl 10 N NaOH 1.5 ml tube 400 μl formaldehyde (36.5–38% water added left overnight 200 μl 10% iodine added HMT-iodine Tubes incubated thermomixer 40 °C 15 min 1000 rpmTubes cooled added 5 ml chloroform extract HMTiodine centrifuged 1000 × g 1 min separate phases 100 μl lower phase HMTiodine fresh tube 1 μl alkane mix-C40 50 mg/L added 1 μl injected mass spectrometer m/z 140 HMT quantitation HMT normalized m/z 198 C14H30.Retinal explant labelingMouse euthanized ketamine/xylazine eyes enucleated retinas dissected high glucose DMEM retinas washed saline added 1 ml culture media 12-well plates DMEM 5 mM glucose 1 mM glutamine 2 mM [1-13C]acetate 10% FBS 1% Penicillin/Streptomycin Retinas cultured 30 min 1 h washed saline frozen liquid nitrogen Metabolites extracted 500 μl −20 °C 80% methanol sonication centrifugation 15,000 × g 5 min °C Supernatant 350 μl tube 10 μl 0.05 mg/ml [13C5]ribitol added sample measurements Metabolites dried derivatized measured data analyzed glutamine labeling retinas dissected KRB 5 mM glucose 1% penicillin/streptomycinRetinas incubated in KRB media 5 mM [13C5]glutamine 10% FBS 1% penicillin/streptomycin 5 min After removed washed with 1 ml saline 1 ml KRB media 10% FBS 1% Penicillin 5 mM lactate 5 mM glucose added to each 12-well plates Metabolites extracted methanol/chloroform/water Ten 0.05 mg/ml [13C5]ribitol added to 300 μl supernatant Samples dried derivatized measured on.Cell countingMedia removed washed 2 ml PBS trypsinized 100 μl 0.05% trypsin 0.53 mM EDTA resuspended 1 ml fresh media Tubes centrifuged 300 × g temperature 5 min resuspended 1 ml HBSS washed 1 ml HBSS added 50 μl to 50 μl 2x CyQuant cell reagent incubated at 37 °C 1 h fluorescence measured Victor X2 plate reader excitation 485 nm emission detection 535 nm time 0.1 s analysis glucose glutamine measured 7890B GC 5977 spectrometer Electron impact ionizationμl sample injected splitless scan glucose labeling 50 to 800 m/z inlet heater 250 °C septum purge 3 ml/min Injections splitless HP-5ms inert 30 m 250 μm 0.25 μm column MSD Samples constant flow helium 1.1 ml/min oven temperature 60 °C 1 min ramped 325 °C hold 10 min column re-equilibrated 60 °C 1 min Müller cells astrocyte metabolites assayed column DB-5 ms GC Column 30 m 0.25 0.25 μm DuraGuard 10 m Scan method solvent delay min ion monitoring method Ions different time windows metabolite measurement glutamine labeling assays windows Table windowMetaboliteIons)6.16Pyruvate173–1786.16–9.00Lactate218–2239.0–9.65Proline243–2529.06Fumarate244–25012.40Malate334–34012.57Aspartate333–33913.90Glutamate347–35414.29U-13C Ribitol323.80Glutamine346–35315.60Citrate464–4741650Glucose553–561Time window ions measured Metabolites Ions (m mass-to-charge ratio isotopic enrichments concentrations metabolitesMetabolite Detector Mass Hunter software deconvolution annotation IsoCor natural abundance isotopes mass isotopomer distribution metabolite Unlabeled extracts MIO-M1 cells metabolite standards annotate metabolites samples NIST 2017 library compounds extracts cell lines retinal explants blotMIO-M1 cells treated media unlabeled glucose glutamine cells scraped in RIPA buffer protease inhibitor cocktail Complete inhibitor cocktail PhosSTOP Protein concentration measured Pierce BCA protein assay reagent Protein 8 or 16 μg added 20 mM DTT volume adjusted to 30 μl 2× Sample Tris-Glycine SDS buffer heated at 95 °C 3 min proteins samples equal 25 μl sample preparation loaded Tris-glycine gel 4–20% precast gels ran at 125 V 1 h 45 min Proteins transferred to PVDF membrane dried 1 h chromatography papersMembrane blocked Odyssey buffer-COR Lincoln Nebraska 1 h room temperature incubated overnight primary antibody washed TBST buffer incubated secondary antibody 1 h room temperature primary antibodies used:PDH E1 alpha rabbit polyclonal Anti-PDH ABS2082 Millipore (Burlington MA Dilution 1:1000.PhosphoDetectTM Anti-PDH-E1α Rabbit AP1063 1:1000.COXIV antibody 4844 Cell Signaling Technology MA probe mitochondria 1:1000.β-Actin mAbs #3700 #4970 loading control 1:2000.GS mouse mAb #610517 Bioscience San Jose Müller glia 1:1000.CRALBP rabbit pAb UW55 Jack Saari Washington gifted John W. Crabb (Cleveland.Secondary antibodies:IRDye® Donkey (polyclonal) anti-mouse IgG (H 925-68072 LI-COR 1:1000.IRDye® 800CW Donkey anti-Rabbit IgG 925-32213 LI-COR 1:1000.Mitochondria morphology revealed Invitrogen CellLight Reagent BacMam 2.0 Mitochondria-GFP baculoviral vector signal peptide sequence E1α pyruvate dehydrogenase emerald GFP MIO-M1 cells grown 70% confluence four-well Lab-Tek II Chamber transfected overnight Mitochondria-GFP 3 × 105 10 × 105 particles/well Next day cells humidified CO2 incubator hyperoxic cell culture chamber 75% oxygen 5% CO2 ProOx Model C21 connected CO2 O2. incubation cells rinsed PBS fixed 4% paraformaldehyde VectaShield DAPI images Zeiss AxioImager.Z1 fluorescent microscope 63× objective AxioCam 503 mono camera ApoTome.2 adapter analyses mitochondrial networks NIH ImageJ v1.40 Mitophagy MiNA43,44 Nature Research Reporting Summary information | 49 | 0.875299 |
10.1038/s41467-021-21015-8 | PMC7878795 | Autoimmune Addison’s disease is a rare complex disease, which has not yet been characterized by non-biased genetic studies. Here, the authors perform the first GWAS for the disease, identifying nine loci including two coding variants in the gene Autoimmune Regulator (AIRE). | Autoimmune Addison’s disease (AAD) is characterized by the autoimmune destruction of the adrenal cortex. Low prevalence and complex inheritance have long hindered successful genetic studies. We here report the first genome-wide association study on AAD, which identifies nine independent risk loci (P < 5 × 10−8). In addition to loci implicated in lymphocyte function and development shared with other autoimmune diseases such as HLA, BACH2, PTPN22 and CTLA4, we associate two protein-coding alterations in Autoimmune Regulator (AIRE) with AAD. The strongest, p.R471C (rs74203920, OR = 3.4 (2.7–4.3), P = 9.0 × 10−25) introduces an additional cysteine residue in the zinc-finger motif of the second PHD domain of the AIRE protein. This unbiased elucidation of the genetic contribution to development of AAD points to the importance of central immunological tolerance, and explains 35–41% of heritability (h2). | IntroductionAutoimmune Addison’s disease (AAD) is the most common cause of primary adrenal failure in the Western world1. It is a rare disease, affecting from five individuals per million in Japan, to more than 200 per million in the Nordic countries2,3. The disease requires lifelong steroid hormone replacement therapy and is fatal if untreated. Autoimmune etiology is often apparent from the presence of other associated autoimmune diseases4, and is confirmed by the presence of autoantibodies against the adrenal enzyme 21-hydroxylase5.Despite the high heritability of AAD, amounting to 97% in a Swedish twin study (95% CI 0.88–0.99)6, genetic factors contributing to disease development have remained poorly defined. Due to the limited size of previously studied cohorts, candidate gene studies have for long been the only feasible option, even though the approach is known to be biased and many results fail to replicate. Targeted investigations have associated AAD with variation in the human leukocyte antigen (HLA) region on chromosome 6p217–9, and have also implicated other well-established autoimmune disease susceptibility genes such as PTPN2210, CTLA411–13, and CLEC16A14. Targeted sequencing studies have further identified BACH215 and AIRE16 as risk loci in AAD, but were limited to preselected gene panels and small sample sizes. A genome-wide association study (GWAS) in AAD has hitherto not been possible due to the insufficient size of available cohorts.Here we utilize the two largest Addison’s disease biobanks in the world17,18, enabling us to uncover both known and novel associations. Most intriguingly, we link AAD to protein-coding risk variants in AIRE, a gene crucial for antigen presentation in the thymus and for central immunological tolerance.ResultsGWAS of autoimmune Addison’s diseaseOur initial sample of 1457 unrelated cases was further filtered to ensure a homogenous cohort of patients with autoimmune adrenal failure. We selected only cases with serum autoantibodies against 21-hydroxylase, and removed cases with clinical manifestations indicating other disease etiologies. Individuals with autoimmune polyendocrine syndrome type-1 (APS-1)4 were identified and excluded using clinical criteria, cytokine autoantibodies, and AIRE gene sequencing. The main analysis encompassed 1223 cases with AAD and 4097 healthy controls (Supplementary Table 1 and Supplementary Note 1). Genotyping was performed on a single occasion on the Illumina Infinium Global Screening Array followed by phasing and imputation. We imputed genotypes from the Haplotype Reference Consortium, retaining more than 7 million variants with minor allele frequency (MAF) ≥ 1%. The case–control association was performed using logistic regression on allele dosages, with sex and the first five principal components as covariates (Supplementary Fig. 1). The genomic inflation factor (λGC) was 1.05 and the linkage disequilibrium (LD) score regression coefficient 1.02 (SE 0.007) indicating a low inflation in test statistics, mostly due to polygenicity (Supplementary Fig. 2).We assessed the proportion of heritability explained by additive effects of SNPs using the genome-wide complex trait analysis GCTA (genome-wide complex trait analysis) software19. To account for ascertainment bias, i.e., the enrichment of cases in our sample compared to the general population, we included disease prevalence in the calculation. Reports from Scandinavia have indicated a prevalence between 13 and 22 cases per 100,000 inhabitants, which corresponded to an SNP heritability rate for AAD between 34 and 40%3,17,20. In other words, 35–41% of the heritability estimated in twins (h2 ≈ 0.97) is explained by the SNPs covered in this GWAS6.Genome-wide significant risk lociOur genome-wide analysis identified nine risk loci that exceeded the genome-wide significance (P ≤ 5 × 10−8; Fig. 1 and Table 1). Besides the HLA region, which stood out as the major risk locus (top SNP rs3998178, P < 10−179), we discovered AAD associations with variants in or adjacent to PTPN22, CTLA4, LPP, BACH2, SH2B3, SIGLEC5, UBASH3A, and AIRE (Supplementary Fig. 3). Of these associated loci, five had previously been implicated in AAD (PTPN22, CTLA4, HLA, AIRE, and BACH2), underlining the reliability of our results, whereas four loci were novel: LPP, SH2B3, SIGLEC5, and UBASH3A. To identify any further independent signals within the association peaks, we performed conditional regression analysis on the most significant SNP in each peak.Fig. 1Manhattan plot for the genome-wide association study of autoimmune Addison’s disease with 1223 cases and 4097 controls.The –log10
P values from logistic regression on the y-axis are plotted against their physical chromosomal position on the x-axis for all SNPs across chromosomes 1–22 and X. Labels correspond to the prioritized or nearest genes. The dotted red bar marks the genome-wide significance level (P ≤ 5 × 10−8). The y-axis has been gapped to include the top SNP in the HLA region.Table 1Autoimmune Addison’s disease risk loci.Risk allele frequencyAssociationLocusaChr: PositionbSNPTypecRisk/alt alleledCasesControlsOR (95% CI)PPTPN221: 114377568rs2476601p.R620WA/G0.170.111.74 (1.53–1.98)6.3 × 10−17CTLA42: 204707138rs1157130325 kbG/A0.690.611.39 (1.26–1.53)5.0 × 10−11LPP3: 188112554rs1464510intronicA/C0.510.421.37 (1.25–1.5)7.3 × 10−12HLA-DQB16: 32623371rs39981784 kbT/C0.510.195.98 (5.29–6.76)3.5 × 10−179BACH26: 90926612rs108064255′ UTRA/C0.500.371.69 (1.53–1.85)2.8 × 10−27SH2B312: 111932800rs713782843 kbC/T0.530.461.3 (1.18–1.42)4.9 × 10−8SIGLEC519: 52204248rs811214370 kbA/G0.0730.0471.88 (1.51–2.34)1.4 × 10−8UBASH3A21: 43836186rs11203203intronicA/G0.420.351.35 (1.22–1.48)8.6 × 10−10AIRE21: 45714294rs74203920p.R471CT/C0.0650.0203.42 (2.71–4.32)9.0 × 10−25AIREe21: 45709153rs2075876intronicG/A0.950.902.17 (1.77–2.66)7.8 × 10−14Odds ratios and P values were estimated using logistic regression in 1223 cases diagnosed with autoimmune Addison’s disease and 4097 controls. Only results with P < 5 × 10−8 were reported to adjust for multiple testing.a The association peaks in chromosomes 1, 2, 12, and 19, span more than one gene, and the prioritized genes are reported. For HLA, the gene closest to the top SNP is reported.b Base-pair coordinates according to human reference genome GRCh37.c Functional characterization of SNPs overlapping prioritized genes, or distance from the SNP to the prioritized gene.d The risk allele indicates the effect allele for the OR, the second position gives the alternative allele.e Results for AIRE rs2075876 are from an analysis conditioning on the genotypes of the top SNP in AIRE, rs74203920.We also carried out a fine-mapping analysis of each association peak, bar that centered on HLA, for which the results are summarized in Supplementary Data 1. Most loci had only one credible configuration, and those that had several had highly overlapping ones (2:3 SNPs common to all configurations). When limited to a single causal SNP, most loci had many SNPs in the credible set (range 7–43). Only SNPs with a log10(Bayes Factor) > 2, indicating strong support for causality versus the null hypothesis, are reported in Supplementary Data 1.Association with the Autoimmune Regulator geneGiven that mutations in AIRE cause the monogenic disease APS-1 (OMIM #240300), of which AAD is a major component, this association peak was investigated in particular detail. Conditioning on the top AIRE SNP rs74203920, we found a second independent signal in AIRE (rs2075876, Pcond. < 7.8 × 10−14), which was not in LD with the covariate rs74203920 (r2 < 0.01) (Fig. 2). Of these two independent associations, the top SNP rs74203920 was a novel association, whereas rs2075876 has been investigated previously16.Fig. 2Two coding variants in the Autoimmune Regulator gene (AIRE) are independently associated with autoimmune Addison’s disease.a GWAS results without (upper panel) and with (lower panel) conditioning, on the top SNP rs74203920. The secondary association peak, including rs1800520, remains equally significant after conditioning reflecting its independent association. The –log10
P values from logistic regression of 1223 cases and 4097 controls are plotted against their physical chromosomal position. The red bars represent the genome-wide significance level (5 × 10−8). b The location and consequences of the coding change p.R471C in the PHD2 domain of AIRE. The additional charge from the cysteine residue (red) is in close proximity to the zinc ion (teal). Arginine is marked in green, histidine in orange, and wildtype (WT) cysteines in yellow.As we had carefully excluded cases of APS-1 using clinical data, serology, and genetic information, the strong association with the lead SNP in AIRE was a striking finding (rs74203920, OR = 3.4 (2.7–4.3), P = 9.0 × 10−25). Comparing carriers with non-carriers of the p.R471C variant in our group of cases (n = 1223), we could not detect any differences in age of disease onset or presence of autoantibodies (Supplementary Table 2). Furthermore, to test whether the effect of rs74203920-T was associated with AAD alone or with autoimmune comorbidities, we divided cases into isolated AAD (n = 443), and those with AAD and type 1 diabetes or autoimmune thyroid disease, i.e., Autoimmune Polyendocrine Syndrome type-2 (n = 682) (Table 2)4. The risk allele rs74203920-T was equally enriched in both categories, and exceeded genome-wide significance regardless of autoimmune comorbidity.Table 2Risk allele frequencies (RAF) in isolated autoimmune Addison’s disease and autoimmune polyendocrine syndrome type 2 (APS-2).LocusaSNPRAF controlsRAF isolated AADORPRAF APS-2OR (95% CI)PPTPN22rs24766010.110.181.72 (1.42–2.08)1.8 × 10−80.171.72 (1.46–2.02)7.9 × 10−11CTLA4rs115713030.610.671.28 (1.11–1.48)8.8 × 10−40.701.46 (1.29–1.66)4.4 × 10−9LPPrs14645100.430.531.46 (1.27–1.67)6.4 × 10−80.501.33 (1.19–1.5)8.1 × 10−7HLA-DQB1rs39981780.190.515.72 (4.81–6.81)6.5 × 10−860.526.31 (5.42–7.35)2.6 × 10−123BACH2rs108064250.370.501.66 (1.44–1.91)2.0 × 10−120.511.71 (1.52–1.93)1.7 × 10−18SH2B3rs71378280.460.521.29 (1.12–1.49)4.0 × 10−40.521.28 (1.13–1.44)5.9 × 10−5SIGLEC5rs81121430.0470.0772.04 (1.49–2.8)8.3 × 10−60.0711.76 (1.33–2.31)5.7 × 10−5UBASH3Ars112032030.350.401.24 (1.07–1.43)3.6 × 10−30.441.4 (1.24–1.58)5.9 × 10−8AIRErs742039200.0200.0723.73 (2.74–5.09)8.0 × 10−170.0633.24 (2.43–4.31)7.7 × 10−16AIRErs20758760.900.941.85 (1.38–2.47)3.5 × 10−50.962.67 (2.01–3.55)9.7 × 10−12Odds ratios and P values were estimated using logistic regression in isolated AAD (n = 443) and APS-2 (n = 682), respectively, compared to 4097 controls. Only loci with P < 5 x 10−8 in the overall analysis were tested.a The association peaks in chromosomes 1, 2, 12, and 19, span more than one gene, and the prioritized genes are reported. For HLA, the gene closest to the top SNP is reported.Since APS-1 is a recessive disorder, we formally tested but could not find support for rs74203920 and/or rs2075876 causing AAD with recessive inheritance (Supplementary Table 3). Rather, the risk effects of both SNPs were best described by an additive model. Lastly, we tested for differential association with other loci in carriers versus non-carriers of rs74203920 and rs2075876, respectively, but found no differences between the groups (Supplementary Tables 4 and 5). Taken together, the associated AIRE variants exert their risk effect independently from each other and from other risk loci.The risk allele rs74203920-T was uncommon in our Swedish and Norwegian controls (2.0%) and in the non-Finnish European population (1.4% GnomAD v2.1.1), but was strongly enriched among cases with AAD (MAF = 6.5%). The SNP is located in exon 12 and the risk allele encodes an arginine to cysteine substitution at amino acid residue 471 in the well-conserved zinc ion binding motif of the second PHD domain (PHD2) (Fig. 2a). The PHD2 domain of AIRE is stabilized by a zinc finger with two zinc ions, one of which is coordinated by amino acid residues C446, C449, C472, and C47521. Each zinc-binding residue is essential, as exemplified by the missense mutation p.C446G found in patients with APS-1, which destroys the structural fold of the PHD2 domain. By introducing an additional cysteine in the zinc-binding motif, it is possible that p.R471C alters the binding of the zinc ion and the structure of the PHD2 domain (Fig. 2b and Supplementary Fig. 4).Within the second, independent association peak in AIRE, the top SNP rs2075876 was in LD with the coding SNP rs1800520 (r2 = 0.83). This variant, a serine to arginine substitution of amino acid residue 278 (p.S278R), is located in the SAND domain. Hence, two coding changes were independently associated with AAD. In a functional assay of AIRE function, neither .R471C nor p.S278R interfered with AIRE-dependent transcription (Supplementary Fig. 5 and Supplementary Note 2)22. With two independent associations with AIRE, AIRE-dependent antigen presentation, and central immune tolerance appears to play an important role in the development of AAD.Dissection of the HLA associationThe HLA region is by far the strongest risk locus in AAD, but due to long-range LD and genetic heterogeneity, the dissection of risk within the region is challenging. To define the key components of genetic risk attributable to HLA, we imputed classical HLA alleles and their corresponding amino acids across HLA class I and class II, and constructed a general logistic model for AAD risk (Fig. 3 and Supplementary Note 3). We report seven independent alleles and amino acids associated with AAD at the genome-wide significance level (P value <5 × 10−8; Table 3). Consistent with previous studies, we found that the risk was dominated by HLA-DQB1*02:01 (OR = 7.3, P = 1.9 × 10−45, under the full model including all reported effects), and HLA-DQB1*03:02 (OR = 2.3, P = 1.4 × 10−21), that tag the well-established risk haplotypes corresponding to serotypes DR3-DQ2 and DR4-DQ8, respectively9,16,17. We also found largely additive risks for DQB1 position 30 Tyr, B pos. 74 Asp, B pos. 156 Asp, and DQA1*01:04. The tyrosine residue in position 74 of HLA-B was the first representation of HLA class I to be included in the model, and had only a weak correlation with HLA class II (r2 = 0.22 with HLA-DQB1*02:01). Comparing cases (n = 232) and controls (n = 2719) that carry neither of the major two risk haplotypes, the two strongest of remaining risk haplotypes contained DQB1*03:01 and DQB1*04:02, both of which encode a tyrosine residue in DQB1 position 30.Fig. 3Stepwise regression of the HLA association identifies the major genetic determinants of autoimmune Addison’s disease.The figure displays the results from the first six steps of regression modeling of the HLA risk effects—alleles and amino acids. Starting with a baseline model comprising sex and five principal components as covariates, we tested every allele and amino acid in turn for association with AAD (Supplementary Note). Additive, recessive, dominant, overdominant, and general variable encodings were compared with likelihood ratio tests and/or Bayesian information criterion. The allele or amino acid residue with most compelling evidence for association was included in the model at every step, and reconsidered at all subsequent steps. Downstream regression models were conditioned on the effects selected from previous models. The y-axes show the –log10
P values from stepwise logistic regressions of 1223 cases and 4097 controls. The dashed horizontal lines indicate genome-wide significance (P < 5 × 10−8). Diamonds mark the most significant effect. Blue color indicates strong linkage disequilibrium (r2) with the most significant effect, gray color indicates no correlation.Table 3HLA alleles associated to autoimmune Addison’s disease.Frequency aAssociationHLA allele or amino acidParameterCasesControlsOR (95% CI)bP valuecHLA alleles in LDdDQB1*02:01Additive effect0.390.127.29 (5.54–9.6)1.9 × 10−45DQA1*05:01, DRB1*03:01 (r2 > 0.95), B*08:01, C*07:01 (r2 > 0.5)DQB1*03:02Additive effect0.280.142.25 (1.91–2.66)1.4 × 10−21DQA1*03:01 (r2 > 0.95), DRB1*04:04 (r2 = 0.43)DQB1 pos. 30 TyrAdditive effect0.540.523.64 (2.9–4.59)3.5 × 10−28DQB1*06:02 (r2 = 0.16), DRB1*15:01 (r2 = 0.16)B pos. 74 AspAdditive effect0.610.371.97 (1.71–2.26)1.5 × 10−21B*08:01 (r2 = 0.30), C*07:02 (r2 = 0.23)DRB1*04:04Allelic interaction with DQB1*02:010.160.0446.66 (4.55–9.74)1.4 × 10−22DQA1*03:01 (r2 = 0.43), DQB1*03:02 (r2 = 0.43)B pos. 156 AspAdditive effect0.460.251.69 (1.45–1.97)2.9 × 10−11B*08:01 (r2 > 0.5), C*07:01 (r2 = 0.41)DQA1*01:04Additive effect0.0120.0213.85 (2.39–6.2)3.0 × 10−8DQB1*05:03 (r2 > 0.95), DRB1*14:01 (r2 > 0.95)Odds ratios and P values were estimated using stepwise logistic regression in 1223 cases and 4097 controls. Only results with P < 5 × 10−8 were reported to adjust for multiple testing.a Allele frequency and amino acid frequency, respectively.b Estimated odds ratio from the full model.c
P value from the full model. Alleles and amino acids are presented in order of inclusion.d HLA alleles with LD r2 > 0.5 are presented in categories of r2 > 0.5, >0.8, >0.9, and >0.95. Maximum 2 HLA alleles with r2 ≤ 0.5 are presented.After defining the major risk effects, we next investigated the presence of interactions between HLA alleles and amino acids included in the model, and all other alleles and residues in the dataset. We identified a strong risk effect for DRB1*04:04 in the presence of DQB1*02:01 (OR = 6.7, P = 1.4 × 10−22). Beside this interaction, no other pairs of alleles and/or residues passed the significance threshold for inclusion in the full model.To avoid the risk of conditioning out critical amino acids in local LD with major risk alleles, we extracted and investigated the most significantly associated residues from each round of the stepwise regression (Supplementary Table 6). The emerging amino acids were the arginine in position 52 of DQA1 (OR = 7.8, P = 2 × 10−157) and the alanine in position 57 of DQB1 (OR = 4.3, P = 2.6 × 10−152). The latter is a distinctive feature of the allele DQB1*02:01. Even though the long-ranging LD in the HLA region makes it difficult to pinpoint causal variation, it is striking that also the third independent amino acid resides in the binding pocket of the HLA-DQ heterodimer (DQB1 pos. 30 tyrosine, OR = 3.6, P = 3.8 × 10−35) (Fig. 4).Fig. 4Associated amino acids in the HLA-DQ heterodimer.Two amino acids in HLA-DQB1 and one amino acid in HLA-DQA1 were found to be associated with autoimmune Addison’s disease. A tyrosine at the 30th position and an alanine at the 57th position of HLA-DQB1 (top) and an arginine at the 52nd position of HLA-DQA1 (bottom) have been marked in orange. To visualize the binding pocket, a peptide ligand (gliadin) from the original crystal structure has been marked in pink.Established loci in AADThe PTPN22, CTLA4, and BACH2 loci are well-known drivers of autoimmune disease and we identified the variants and haplotype blocks that have previously been described in AAD and common autoimmune comorbidities (Fig. 5 and Supplementary Figs. 6 and 7). The fine-mapping analysis largely confirmed the missense variant in PTPN22, thought to be causal in a range of different autoimmune diseases. The credible set also included two eQTLs for BCL2L15, a weakly proapoptotic protein associated with autoimmune thyroid disease and type 1 diabetes23,24, in T helper cells. This differential regulation might constitute a secondary causal effect in addition to the canonical p.R620W variant in PTPN22.Fig. 5Shared genetic features.a Human diseases and traits studied in GWAS were clustered to reveal shared genetic risk factors. Diseases/traits are ordered by unsupervised hierarchical clustering, and color scale indicates genetic correlation. b Loci implicated in autoimmune Addison’s disease in order of decreasing effect size (odds ratio and 95% CI), 1223 cases, and 4097 controls. The horizontal, dashed line marks OR = 1. Blue squares indicate genome-wide significant associations for the diseases and loci/variants, respectively. c Circos plot representing the loci associated with AAD (boxes) and other autoimmune diseases (dots). The AAD track is highlighted in yellow, and the yellow wedge on chromosome 21 is magnified ×15. SLE-systemic lupus erythematosus, RA-rheumatoid arthritis, PSO-psoriasis, MS-multiple sclerosis, T1D-type 1 diabetes, VIT-vitiligo, CD-coeliac disease. d Side-by-side comparison of association statistics at the PTPN22 locus across a selection of autoimmune diseases. AAD statistics were calculated using logistic regression for 1223 cases and 4097 controls.For the chromosome 2 peak, no obviously immune relevant signals were identified in the credible set/configuration, though a couple of variants were located in enhancer-like sequences, and a GTEx eQTL gene for our prioritized CTLA4. In chromosome 6, almost all the variants in the credible set and configurations were BACH2 eQTLs for naive T-cell populations in particular, several also have enhancer-like signatures and exhibit H3K27Ac-marks in some ENCODE cell lines, strengthening the postulation that AAD is driven by differential regulation of BACH2 in cases versus controls.Novel loci in AADWe also discovered four novel loci that achieved genome-wide significance for association with AAD. The ORs were lower than for loci detected previously, commensurate with the improved statistical power in this study. All the SNPs in the credible set and configuration for the chromosome 3 peak were located in introns of LPP, as is the lead SNP from the GWAS. With the exception of some weak transcription factor binding sites, there were no functional categories located at any of these SNPs.The association peaks in chromosomes 12 and 19 encompassed several genes, but were ascribed to SH2B3 and SIGLEC5 after gene prioritization. The lead SNPs were barely genome-wide significant, and the most significant genotyped markers, as opposed to imputed markers, had P values of 7.8 × 10−8 and 4.3 × 10−7 for SH2B3 and SIGLEC5, respectively.The broad association at the SH2B3 locus looked similar in studies of type 1 diabetes and vitiligo, giving credibility to the result, but making it challenging to appoint a single candidate gene. The credible set and configuration for chromosome 12 were eQTLs for a number of tissues and a handful of genes (mostly the same for each variant) and H3K27Ac-marks in several cell lines, but none that appeared particularly relevant to autoimmunity. However, one of the credible set SNPs is both located in an enhancer-like sequence and is itself a missense variant in SH2B3. This variant (p.W262R, MAF ≈ 0.5) is not predicted to be deleterious (by SIFT/PolyPhen). In the chromosome 19 credible set/configuration, whole blood eQTLs were present in one variant each for SIGLEC14 (a MAPK/AKT-activating SIGLEC, as opposed to the DEPICT-prioritizedSIGLEC5) and SPACA6.In contrast to the above broad association peaks, the peak at the UBASH3A locus was well-defined and in a haplotype block containing no other genes but UBASH3A. eQTLs for UBASH3A exist for all credible configuration SNPs for various T-cell subpopulations, but all the variants in this locus’s credible configurations and sets are yet more significant eQTLs for all T cells in TMPRSS3.Interestingly, most AAD GWAS peaks harbor a gene with a role in or near the immunological synapse, the connection between antigen-presenting cells and lymphocytes (P = 0.003) (Supplementary Fig. 8)25.Genetic correlations and loci shared with other autoimmune traitsMany autoimmune diseases co-occur in affected individuals and families, and share numerous genetic risk factors26. To explore the genetic architecture underlying AAD at large, we investigated the overlap of risk loci with diseases and traits previously investigated in other well-powered GWAS. By unsupervised clustering of overlapping risk loci, an extensive and complex sharing of risk loci among immunological diseases clearly emerged (Fig. 5a). Systemic autoimmune diseases, inflammatory bowel diseases, and organ-specific autoimmune diseases, respectively, formed distinct groups within this major cluster. The majority of patients with AAD develop at least one additional autoimmune disease, such as Hashimoto’s thyroiditis, type 1 diabetes, vitiligo, or Graves’ disease18. It was therefore interesting to note that AAD displayed a significant overlap of genetic risk loci with its most common comorbidities, reflecting the genetic risk factors shared between organ-specific autoimmune diseases.We searched the National Human Genome Research Institute—European Bioinformatics Institute (EBI) GWAS catalog using our genome-wide significant AAD risk loci for associations with other autoimmune diseases (Fig. 5b, c). The overlapping loci included UBASH3A and SH2B3 in type 1 diabetes and celiac disease, and LPP in autoimmune thyroid disease and vitiligo. In general, surveying autoimmune diseases with summary statistics available through the GWAS catalog and PhenoScanner, risk variants and haplotypes were strikingly often shared across diseases (Supplementary Figs. 6 and 7)27,28. In the case of PTPN22, which had data available for the largest number of diseases, confidence intervals of effect estimates were overlapping between diseases, indicating equivalent effects (Fig. 5d).Notably, the strongest of our novel risk alleles, our lead SNP in AIRE (rs74203920), has not been linked to other autoimmune diseases in GWAS. Although our second independent signal in AIRE has previously been associated with AAD, and also with rheumatoid arthritis in East Asia16,29,30, the allele that increases the susceptibility to AAD (rs2075876-G), appears associated with decreased risk of rheumatoid arthritis. Thus, risk alleles at most loci appear to be general drivers of autoimmunity, whereas the risk alleles in AIRE are more specifically associated with AAD.DiscussionThis GWAS of AAD identified nine genome-wide significant associations and explained 35–41% of the additive genetic heritability (h2). The results point to the complex network of antigen presentation and immunomodulation that underlie autoimmune disease development (Fig. 6). In particular, two independent associations in AIRE highlight the importance of central immune tolerance in AAD pathogenesis. AIRE is essential for thymic expression of otherwise tissue-specific proteins, and hence important for negative selection of autoreactive thymocytes and prevention of organ-specific autoimmune disease. As strongly deleterious mutations in AIRE cause APS-1, it is particularly interesting that we associate two LD-independent protein-coding variants in AIRE with sporadic AAD.Fig. 6T-cell regulation and AAD GWAS associated regions.a Graphic representation of selected aspects of T-cell regulation, with gene products implicated by GWAS association proximity in red (antigen-presenting cell, APC). b AIRE activity in medullary thymic epithelial cell (mTEC), promoting expression of tissue-restricted antigens (TRAs) for the education of T cells.Given the allele frequency of 1.5–2.0% in the general population, the effect of the variant with the strongest association, p.R471C, must be subtle compared to mutations known to cause APS-1. p.R471C has previously been reported in single cases of multi-organ autoimmunity, including patients with AAD and type 1 diabetes31, or AAD and autoimmune thyroiditis32. None of the reported cases had interferon autoantibodies, which would be expected in patients with APS-133. Thus, current evidence does not support p.R471C as a cause of APS-1, but instead points to an increased risk of AAD at the population level.p.R471 is located near two highly conserved cysteine residues that coordinate a critical zinc ion in the second PHD zinc finger (p.C472 and p.C475). PHD fingers maintain the structural integrity, read methylation states of histones, and regulate gene expression through formation of complexes with chromatin regulators and transcription factors34. Our transfection assay did not show an effect of rs74203920 or rs1800520 on AIRE-dependent transcription, for which there may be several reasons. While AIRE is predominantly expressed and exerts its main functions in the thymus, HEK293 cells have despite their renal origin shown large overlap in AIRE-regulated genes with primary thymic cells from mice35–37. It is thus likely that p.R471C has an effect on AAD susceptibility that differs from classical deleterious mutations that cause APS-1. Furthermore, though it cannot be excluded that variation in other nearby genes is involved, for instance the inducible T-cell costimulator (ICOSLG), the credible set from fine-mapping of the locus includes only p.R471C.The HLA region is implicated in autoimmune disease and confers by far the largest risk for AAD compared to other known risk loci. We dissected the HLA-mediated risk of AAD in detail, confirming known associations and suggesting additional susceptibility alleles8,9,15,38. Our results demonstrate that risk is dominated by alleles in HLA class II, in particular the two major risk haplotypes, and an interaction between the two indicates a shared mechanism. We also identified strong effects mediated by HLA class I. For instance, aspartic acid in residues 74 and 156 of HLA-B are both represented in HLA-B*08:01, one of the few alleles that have been successfully investigated in functional studies on antigen presentation of 21-hydroxylase in AAD patients39. Notably, we could not detect any interactions between HLA class I and class II, nor any interactions between pairs of alleles conferring risk and protection. With a larger sample size, however, additional effects could potentially be uncovered and incorporated in a similar model. These results provide a foundation for further work aimed at understanding the exact mechanisms underlying HLA-mediated risk and for functional studies of implicated HLA alleles in antigen presentation.The two independent associations with AIRE point to alterations in central immunological tolerance as an underlying mechanism in AAD development. The importance of a correct expression of AIRE for maintaining immunological tolerance is also exemplified by Down Syndrome in which the extra copy of AIRE, located on chromosome 21, has been coupled to altered expression of AIRE in the thymus, impaired central tolerance and an overrepresentation of autoimmune diseases40,41. Many of the other risk loci identified in this study harbor genes involved in antigen presentation and recognition, and hence in thymocyte maturation. Beside HLA class II that presents antigens to developing T cells, the turnover of the T-cell antigen receptor (TCR) complex is regulated by UBASH3A42, and the immune checkpoint CTLA4 modulates the co-stimulation required for T-cell activation43. Alternatively, or in addition to UBASH3A, the putative causal variants identified by fine-mapping suggest a role in AAD for TMPRSS3 in T cells; the risk alleles are linked to higher expression levels, an effect also seen in type 1 diabetes44. The tryptophan substitution in position 620 of PTPN22 (p.R620W) disrupts the formation of complexes between PTPN22 and CSK, and the inhibitory effect on TCR signaling is attenuated45. BACH2 has been shown to stimulate (CD8+) T-cell differentiation by controlling access of transcription factors to their enhancers and to promote differentiation of regulatory T cells46. SH2B3, suggested to be the causal entity behind the common autoimmune ATXN2/SH2B3 association47, like UBASH3A above, is an inhibitor of signaling cascades in lymphocytes47. While the most highly associated variants lie nearer the 5′ end of the gene, LPP harbors a microRNA, miR-28, that appears to be involved in posttranscriptional regulation of PD148,49 which has an important role in self-tolerance, restraining autoreactive T cells and promoting Tregs50.The association peak on chromosome 19 provides three different potentially causal units: SIGLEC5 (prioritized by DEPICT), SIGLEC14 (whole blood eQTL in the credible configuration), and SPACA6 (whole blood eQTL in the credible set). The latter harbors miR-125a, a miRNA that appears to be involved in posttranscriptional regulation of KLF1351, which in turn regulates CCL5, a chemokine that affects the activation, migration, and proliferation of T cells52. SIGLEC14, an activating receptor highly homologous to the nearby SIGLEC5, is hard to rule out, though the interpretation is further complicated by the fact that a common polymorphism leads to a SIGLEC14/5 fusion gene and effectively a SIGLEC14-null phenotype53. SIGLEC5, as it recognizes self-cell surface sialoglycans and mediates inhibitory signaling in T cells54, is the most likely candidate from a cell biology standpoint. Figure 6 summarizes and highlights these T-cell-related effects of the most likely functional gene products associated with our GWAS hits.We identified robust association signals despite relatively modest sample sizes compared to many other autoimmune disorders, which indicate a rather homogenous disease etiology with relatively low polygenicity compared to other diseases, underpinning the high heritability estimates from epidemiological studies. We believe that a strength of this study was the opportunity to recruit the majority of AAD patients in Norway and Sweden through national registries with geographically matched controls, using stringent exclusion criteria and only including those with 21-hydroxylase autoantibodies.To conclude, our results highlight the importance of central immune tolerance in the development of AAD. Dysregulation of antigen presentation in the setting of negative selection in the thymus may be one of the factors that makes AAD exceptional among organ-specific autoimmune diseases, and the pathways identified should be explored in the development of preventive treatment strategies.MethodsSubjectsCases were recruited from the Swedish and Norwegian Addison Registries and fulfilled clinical diagnostic criteria for primary adrenal insufficiency, i.e., low serum cortisol with a compensatory increase in plasma adrenocorticotropic hormone1,17,18. In case of doubt, a corticotropin stimulation test was performed. Autoimmune etiology was confirmed by the presence of highly specific autoantibodies targeting the adrenal-specific enzyme 21-hydroxylase, the major autoantigen in AAD5. Cases were screened for APS-1 using clinical criteria, autoantibodies against interferon-α, interferon-ω, or interleukin 22, and/or AIRE gene sequencing55,56. Healthy controls were recruited from blood donor centers across Sweden and Norway to match the geographical coverage of registry cases.All study subjects gave their informed consent. The study was performed in accordance with the Declaration of Helsinki and approved by the local ethics committees in Stockholm, Sweden (dnr 2008/296-31/2), and Western Norway (biobank 2013-1504, project 2017-624).DNA extractionBlood samples were kept at −80 °C until processed at the HUNT Laboratory (Levanger, Norway). DNA was isolated using the MasterPure™ DNA purification kit version II B1 (Epicenter®, Madison WI), and normalized to 50 ng/µl. In total, 200 ng of each sample was pipetted by robot to 96-well plates (Abgene Storage Plates, ThermoFisher). Swedish/Norwegian and case/control samples were distributed in equal proportions in the plates. Technical replicates were included to facilitate quality control and genotype concordance between plates.Genotyping, imputation, and quality controlGenome-wide genotyping of 692,367 markers was performed using the Illumina Infinum Global Screening Array 1.0 by the Human Genomics Facility at Erasmus MC (Rotterdam, the Netherlands). Markers and samples were filtered iteratively using PLINK version 1.9 (Supplementary Note 1)57. In short, markers were first excluded based on call rate <95% or deviation from GenomeStudio genotype clusters. Second, samples were excluded on the basis of sample call rates <98%. Third, in-depth marker quality control excluded markers with call rate <98%, discordant calls in technical replicates, or deviation from Hardy–Weinberg equilibrium (HWE, P < 10−6). For X chromosome markers, HWE tests were performed in females only. Finally, samples with accumulated heterozygosity >0.34 were excluded.Bi-allelic SNPs that passed the above QC thresholds and that were present in the Haplotype Reference Consortium panel58 were used for phasing and imputation. Genotypes were phased in-house to take advantage of available pedigree information (SHAPEIT version 2.r837)59, and non-typed variants imputed using the Sanger Imputation Service (PBWT) and the Haplotype Reference Consortium release 1.158. Markers with imputation quality score >0.5 and MAF > 0.01 were included in the GWAS.Global ancestry was inferred using the LASER/TRACE software60 with the Human Genome Diversity Project reference panel61. Samples estimated to be non-European were excluded. Genetic relatedness was evaluated using high-quality markers pruned for LD in PLINK. For each pair of related samples (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat \pi$$\end{document}π^ > 0.1), cases and males were preferentially selected, otherwise the sample with the highest call rate was retained. In total, data from 5320 samples and 7.1 million markers were kept for association testing.GWASMain axes of genetic variation, as a proxy for population substructure, were assessed using principal component analysis of high-quality markers with MAF > 0.05, pruned for LD (r2 < 0.2), and excluding the extended HLA region. Association statistics were calculated using logistic regression of disease status on genotype dosages. Sex and the top five principal components were included as covariates to account for potential sex differences and confounding population stratification.Conditional analysisLoci passing the genome-wide significance threshold for association were subject to conditional analysis to identify any independent associations. This was done by stepwise inclusion of imputed genotype dosages for the index variants as covariates in logistic regression.Gene prioritization at associated lociEnrichment of association signals in physiological systems, tissues, and cell types, as well as prioritization of genes for each association, was performed using the computational tool DEPICT from GWAS summary statistics (Data-driven Expression Prioritized Integration for Complex Traits, https://broadinstitute.org/mpg/depict/)25. We used default settings with 500 permutations to adjust for gene length differences, and 50 repetitions to compute false discovery rates. The false discovery rate was set to 5%. Associated loci were plotted with LocusZoom62.Fine-mappingFine-mapping was carried out using FINEMAP63, in two runs. In total, 1 Mb windows around the lead SNPs for each genome-wide association peak (except that of HLA) were assessed, allowing maximum one or three causal SNPs per window per run, and otherwise default settings. It must, however, be noted that for small studies, such as ours is in a modern GWAS context, the benefits of such stochastic fine-mapping are likely to be small64.HLA imputation and dissectionClassical HLA alleles were imputed from directly genotyped and phased SNPs using HIBAG kernel version 1.465 and SNP2HLA version 1.0.366. The classifiers and reference panels for European samples provided with each software were used to impute two field (four digit) alleles in HLA-A, -B, -C, -DQA1, -DQB1, and -DRB1. To improve the precision of DRB1-allele calls, we built a model for HLA imputation using a reference panel of 699 healthy Norwegian controls. The DRB1 alleles were called by combining predictions from the default and in-house model, both weighted by the size of their training data. Genotypes were kept and treated as fixed if their posterior probability was at least 0.5 and more than twice as likely as the second most probable call. For 370 of the case samples, laboratory typing of HLA-A, -B (most single field), -DQB1, and -DRB1 (most two field) was available. Concordance for the three former was 92–98% for both HIBAG and SNP2HLA, while it was 90% for DRB1.We aimed at identifying the main drivers of risk mediated by classical HLA alleles and amino acids across HLA class I and class II, with a method adapted from Moutsianas et al.67. For every HLA allele in turn, we constructed several logistic regression models and used Bayesian information criterion and likelihood ratio test to help decide whether the effect was best described as additive, recessive, dominant, overdominant, or general. See the Supplementary Note 3 for details of the selection procedure. In short, the allele or amino acid residue effect from the most significant model was considered for inclusion in the full model. Five PCs and sex were included as covariates in all models. PCs were calculated from SNPs genome-wide, not including the HLA region, i.e., the same covariates as in the GWAS analysis. We only considered the inclusion of alleles/amino acids that met genome-wide significance P < 5 × 10−8, both at time of inclusion, and in the final model. Backward elimination was performed by leaving previous variables out of the current model, one by one, and by subsequently testing the goodness of fit.Loci shared with common diseases and traitsGenetic overlap was investigated using a method adapted from Farh et al.68. In short, GWAS catalog data were obtained from the EMBL-EBI website (https://ebi.ac.uk/gwas/), current as of December 201969. Diseases/traits with at least six reported associations (P ≤ 1 × 10−6) were included. Because many diseases/traits have been subject to more than one independent GWAS, they had multiple associated SNPs at the same locus. For these loci, defined as a window of 500 kb, only the most significant index SNP was considered. In total, 1990 diseases/traits and 5349 SNPs fulfilled the criteria and were used to find associations where index SNPs of two diseases/traits were within 500 kb of each other. Overlapping risk loci were used to compute a disease-by-disease correlation matrix for every pair of diseases/traits. We selected 22 autoimmune and 19 representative non-autoimmune diseases/traits for comparison and visualization in Fig. 5a as well as summary statistics from four autoimmune diseases for in-depth locus comparison70–73.We also used LD score regression to calculate genetic correlation between our GWAS results and 844 predefined traits at the LD Hub (http://ldsc.broadinstitute.org)74. LD score regression uses GWAS summary statistics of complex traits and diseases but excludes the HLA region, limiting its capacity for estimating genetic correlations between traits with a large proportion of their heritability explained by variation in the HLA.3D models of the second PHD domain in AIRETo visualize the associated amino acids in the HLA-DQA1 and HLA-DQB1 genes, the x-ray diffraction model of HLA-DQ2.3 (DQA1*03:01/DQB1*02:01) was used (Protein Data Bank 20 id: 4D8P)75. In order to generate protein models of the PHD2 domain and the potential structural impact of AIRE variants associated in this GWAS, we used the previously published NMR structure for PHD2 as a template (Protein Data Bank 20 id: 2LRI)21. This domain structure was subsequently mutated at the p.R471 position to a cysteine and visualized using PyMOL (https://pymol.org/).Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting SummaryDescription of Additional Supplementary FilesSupplementary Data 1 | nature communications | [
"Article"
] | [
"Disease genetics",
"Adrenal gland diseases"
] | IntroductionAutoimmune Addison’s disease (AAD common cause primary adrenal failure Western rare disease affecting five per million Japan to 200 per million Nordic requires lifelong steroid hormone replacement therapy fatal untreated Autoimmune etiology other autoimmune confirmed autoantibodies against adrenal enzyme 21-hydroxylase5 high heritability 97% Swedish twin study (95% CI 0.88–0.99)6 genetic factors poorly defined limited size cohorts candidate gene studies only feasible option biased investigations associated AAD with variation human leukocyte antigen (HLA) region chromosome 6p217–9 implicated genes PTPN2210 CTLA411–13 CLEC16A14 sequencing studies identified BACH215 AIRE16 risk loci AAD limited to preselected gene panels small sample sizes genome-wide association study) possible insufficient size cohorts two largest Addison’s disease biobanks associations link AAD to protein-coding risk variants in AIRE crucial for antigen presentation thymus central immunological tolerance initial sample 1457 unrelated cases filtered homogenous cohortselected cases with serum autoantibodies against 21-hydroxylase removed other disease etiologies autoimmune polyendocrine syndrome type-1 identified excluded using clinical criteria cytokine autoantibodies AIRE gene sequencing analysis 1223 cases with AAD 4097 healthy controls Genotyping Illumina Infinium Global Screening Array phasing imputation imputed genotypes from Haplotype Reference Consortium 7 million variants minor allele frequency) 1% case–control association logistic regression on allele dosages sex first five principal components covariates genomic inflation factor) 1.05 linkage disequilibrium) score regression coefficient 1.02 low inflation due to polygenicity assessed heritability explained by additive effects SNPs genome-wide trait analysis ascertainment bias included disease prevalence Scandinavia prevalence 13 22 cases per 100,000 inhabitants SNP heritability rate for AAD between 34 35–41% heritability in twins explained by SNPs significant risk analysis identified nine risk loci exceeded genome-wide significance (P ≤ 5 × 10−8HLA region major risk locus SNP rs3998178 P < AAD associations variants PTPN22 CTLA4 LPP BACH2 SH2B3 SIGLEC5 UBASH3A AIRE five implicated reliability four novel LPP SH2B3 SIGLEC5 UBASH3A conditional regression analysis significant SNP. plot genome-wide association study autoimmune Addison’s disease 1223 cases 4097 controls –log10 P values regression y chromosomal position SNPs chromosomes 1–22 X Labels prioritized genes dotted red bar genome-wide significance level (P 5 × 10−8) y-axis gapped top SNP HLA 1Autoimmune Addison’s disease risk loci allele)PPTPN221: 114377568rs2476601p.R620WA.111.74 (1.53–1.98)6.3 × 10−17CTLA42 204707138rs1157130325.39 (1.26–1.53)5.0 10−11LPP3/C0.510.421.37 (1)7.3 × 10−12HLA-DQB16 32623371rs39981784.98 (5.29–6.76)3.5 90926612rs108064255′.371.69 (1.53–1.85)2.8 111932800rs713782843.461.3 (1.18–1.42)4.9 10−8SIGLEC519 52204248rs811214370/G0.0730.0471.88 (1.51–2.34)1.4.351.35 (1.22–1.48)8.6 45714294rs74203920p/C0.0650.0203.42 (2.71–4.32)9.0 45709153rs2075876intronicG/A0.950.902.17 (1.77–2.66)7.8 ratios P values estimated 1223 cases autoimmune Addison’s disease 4097 controls results P < 5 × 10−8 testing association peaks chromosomes 1 2 12 19, gene prioritized genes gene closest top SNP Base-pair coordinates genome SNPs risk allele second alternative allele Results AIRE rs2075876 top SNP rs74203920carried fine-mapping analysis association peak HLA results summarized Supplementary Data 1. loci one credible configuration overlapping (2:3 SNPs limited single causal SNP loci many SNPs credible set 7–43) SNPs log10(Bayes Factor) > 2 causality reported Supplementary Data 1.Association Autoimmune Regulator mutations AIRE cause monogenic disease APS-1 AAD major peak investigated top AIRE SNP rs74203920 second independent signal (rs2075876 Pcond. < × 10−14) not LD covariate rs74203920 (r2 < 0.01) top SNP rs74203920 novel rs2075876 investigated coding variants Autoimmune Regulator gene associated autoimmune Addison’s disease results top SNP rs74203920 secondary association peak rs1800520 significant after conditioning –log10 P values logistic regression 1223 cases 4097 controls plotted chromosomal position red bars genome-wide significance level (5 × 10−8) location consequences coding change p.R471C PHD2 domain AIREcharge cysteine residue zinc ion Arginine green histidine orange wildtype cysteines yellow excluded APS-1 association lead SNP AIRE (rs74203920 OR = 3.4 (2.7–4.3), P = 9.0 × 10−25) carriers non-carriers p.R471C 1223) differences age disease onset autoantibodies Table 2) effect rs74203920-T autoimmune comorbidities divided cases isolated AAD 443) type 1 diabetes autoimmune thyroid disease 682) risk allele rs74203920-T enriched both categories exceeded genome-wide significance autoimmune comorbidity 2Risk allele frequencies autoimmune Addison’s disease polyendocrine syndrome type 2 controlsRAF (95% CI.110.181.72 (1.42–2.08)1.8 × 10−80.171.72 (1.46–2.02)7.9.610.671.28 (1.48)8.8 10−40.701.46 (1.66)4.4.430.531.46 (1.67)6.4 × 10−80.501.33 (1)8.1-DQB1rs39981780.190.515.72 (4.81–6.81)6.5 10−860.526.31 (5.42–7.35)2.6.370.501.66 (1.71 (1.52–1.93)1.7.460.521.29 (1.49)4.0 10−40.521.28 (1.13–1.44)5.9.04 (1)8.3.0711.76 (1.33–2.31)5.7.350.401.24 (1.07–1.43)3.6.441.4 (1.24–1.58)5.9.0723.73 (2.74–5.09)8.24 (2.43–4.31)7.7.900.941.85 (1.38–2.47)3.5 10−50.962.67 (2.01–3.55)9.7 ratios P values estimated regression 443) APS-2 682) 4097 controls P < 5 x 10−8 tested association peaks chromosomes 1 2 12 19, gene prioritized genes gene closest top SNP APS-1 recessive rs74203920 rs2075876 AAD risk effects SNPs additive modeltested association carriers non-carriers rs74203920 rs2075876 found no differences Tables 4 5) AIRE variants exert risk effect risk allele rs74203920-T uncommon in Swedish Norwegian controls (2.0%) non-Finnish European population (1.4% enriched among AAD (MAF = 6.5%) SNP in exon 12 risk allele encodes arginine to cysteine substitution amino acid residue 471 zinc ion binding motif second PHD domain stabilized by zinc finger two ions coordinated amino acid residues C446 C449 C472 C47521 Each zinc-binding residue essential mutation p.C446G APS-1 destroys PHD2 domain introducing additional cysteine p.R471C alters binding zinc structure PHD2 domain (Fig. 2b top SNP rs2075876 LD with coding SNP rs1800520 (r2 = 0.83) variant serine to arginine substitution amino acid residue 278 in SAND domain two coding changes associated with AAD neither .R471C p.S278R interfered with AIRE-dependent transcription 5associations AIRE antigen presentation central immune tolerance AAD HLA HLA strongest risk locus AAD long-range LD genetic heterogeneity dissection risk challenging risk HLA imputed HLA alleles amino acids HLA class I II logistic model AAD risk. 3 seven alleles amino acids AAD genome-wide (P value <5 × 10−8 risk dominated HLA-DQB1*02:01 (OR = P = 1.9 × 10−45 HLA-DQB1*03:02 (OR 2.3 P = 1.4 × 10−21) risk haplotypes DR3-DQ2 DR4-DQ8 additive risks DQB1 position 30 Tyr B pos. 74 Asp. 156 Asp DQA1*01:04 tyrosine residue position 74 HLA-B first HLA class I weak correlation HLA class II (r2 = 0.22 HLA-DQB1 cases (n = 232) controls = 2719) haplotypes strongest risk haplotypes DQB1*03:01 DQB1*04:02 encode tyrosine residue DQB1 position 30.Fig. regression HLA association genetic determinants autoimmune Addison’s diseasefigure displays results six steps regression modeling HLA risk amino acids baseline model sex components tested allele amino acid association Additive recessive dominant variable encodings compared likelihood ratio tests Bayesian information criterion allele compelling evidence association included reconsidered Downstream regression models conditioned effects previous models y-axes show –log10 P values regressions 1223 cases 4097 controls dashed horizontal lines indicate genome-wide significance (P < 5 × 10−8) Diamonds significant effect Blue linkage disequilibrium gray no correlation.Table 3HLA alleles autoimmune Addison’s disease allele amino alleles*02:01Additive effect0.390.127.29 (5.54–9.6)1.9 × 10−45DQA1 > 0.95) effect0.280.142.25 (1.91–2.66)1.4 × 10−21DQA1 0 0 effect0.540.523.64 (2.9–4.59)3.5 × 10−28DQB1*06:02 0.16) DRB1*15:01 effect0.610.371.9710−21B*08:01 (r2 0.30) C*07:02 0.23)DRB1 interaction DQB1*02.160.0446.66 (4.55–9.74)1.4 10−22DQA1*03:01 (r2 0.43) DQB1*03:02 effect0.460.251.69 (1.97)2.9 10−11B*08:01 (r2 > C*07:01 (r2 0.41)DQA1.0213.85 (2.39–6.2)3.0 × 10−8DQB1*05:03 (r2 > 0.95) DRB1*14:01 (r2 0.95)Odds ratios P values estimated regression 1223 cases 4097 controls results P < 5 × 10−8 testing Allele amino acid frequency odds ratio P value Alleles amino acids inclusion HLA alleles r2 > 0.5 0.5 >0.8 >0.9 >0.95 Maximum 2 alleles r2 0.5 interactions HLA alleles amino acids strong risk effect DRB1*04:04 DQB1*02:01 (OR = P = 1.4 × 10−22) no other alleles significance threshold model amino acids extracted associated residues regression Tableemerging amino acids arginine position 52 DQA1 (OR 7.8 P 2 × 10−157 alanine 57 DQB1 (OR 4.3 P 2.6 × 10−152). distinctive feature allele DQB1*02:01 long-ranging LD HLA region causal variation third amino acid binding pocket HLA-DQ heterodimer (DQB1 30 tyrosine OR 3.6 P 3.8 × 10−35) (Fig. 4) amino acids HLA-DQ heterodimer amino acids HLA-DQB1 one HLA-DQA1 associated autoimmune Addison’s disease tyrosine 30th alanine 57th position-DQB1 arginine 52nd marked orange peptide ligand (gliadin) original marked pink loci PTPN22 CTLA4 BACH2 loci drivers autoimmune disease identified variants haplotype blocks autoimmune comorbidities (Fig. 5 fine-mapping analysis confirmed missense variant PTPN22 causal autoimmune diseases set two eQTLs BCL2L15 proapoptotic protein autoimmune thyroid disease type 1 cells differential regulation secondary causal effect p.R620W variant PTPN22genetic features Human diseases traits GWAS clustered genetic risk factors ordered clustering color scale genetic correlation Loci autoimmune Addison’s disease decreasing effect size 1223 cases 4097 controls horizontal line marks OR = 1. Blue genome associations Circos plot loci AAD autoimmune diseases AAD track highlighted yellow wedge chromosome 21 magnified SLE lupus erythematosus RA-rheumatoid arthritis PSO-psoriasis MS-multiple sclerosis T1D-type 1 diabetes VIT-vitiligo CD-coeliac disease comparison association statistics PTPN22 locus autoimmune diseases AAD statistics calculated 1223 cases 4097 controls chromosome 2 peak no immune relevant signals variants enhancer-like sequences GTEx eQTL gene CTLA4 chromosome 6 variants BACH2 eQTLs T-cell populations enhancer-like signatures H3K27Ac-marks cell lines AAD driven by differential regulation BACH2 loci four novel loci genome-wide significance association AADORs lower improved statistical power SNPs chromosome 3 peak in introns LPP lead SNP GWAS weak transcription factor binding sites no functional categories at SNPs association peaks in chromosomes 12 19 genes ascribed to SH2B3 SIGLEC5 after gene prioritization lead SNPs barely genome-wide significant significant genotyped markers P values × 10−8 4.3 × 10−7 for SH2B3 SIGLEC5 association at SH2B3 locus similar in type 1 diabetes vitiligo credibility challenging single candidate gene credible set configuration chromosome 12 eQTLs for tissues handful genes H3K27Ac-marks in cell lines none relevant to autoimmunity one credible set in enhancer-like sequence missense variant in SH2B3 variant (p.W262R, MAF ≈ 0.5) not deleterious chromosome 19 whole blood eQTLs in one variant for SIGLEC14 MAPK/AKT-activating SPACA6 peak at UBASH3A locus well-defined haplotype block no other genes but UBASH3AeQTLs UBASH3A exist SNPs T-cell subpopulations variants significant T cells TMPRSS3 AAD GWAS peaks harbor gene immunological synapse antigen cells lymphocytes (P = 0.003) correlations loci autoimmune autoimmune diseases co-occur individuals share genetic risk AAD investigated overlap risk loci with diseases traits sharing risk loci among diseases emerged (Fig. Systemic autoimmune diseases inflammatory bowel diseases organ-specific autoimmune diseases groups cluster majority patients with AAD develop one additional autoimmune disease Hashimoto’s thyroiditis type 1 diabetes vitiligo Graves’ AAD overlap genetic risk loci with common comorbidities risk factors diseases searched National Human Genome Research Institute—European Bioinformatics Institute) GWAS catalog AAD risk loci for associations autoimmune diseases overlapping loci included UBASH3A SH2B3 in type 1 diabetes celiac disease LPP autoimmune thyroid disease vitiligosurveying autoimmune diseases GWAS catalog PhenoScanner risk variants haplotypes shared across diseases Figs. 6 PTPN22 largest diseases confidence intervals effect equivalent effects (Fig. strongest risk lead SNP AIRE (rs74203920) not linked autoimmune diseases second signal AIRE associated with AAD rheumatoid arthritis allele susceptibility AAD (rs2075876 decreased risk rheumatoid arthritis risk alleles general drivers autoimmunity AIRE associated AAD GWAS AAD identified nine genome-wide associations 35–41% additive genetic heritability results complex network antigen presentation immunomodulation autoimmune disease development (Fig. 6) two associations in AIRE highlight importance central immune AAD pathogenesis essential for thymic expression negative selection autoreactive thymocytes prevention organ-specific autoimmune disease deleterious mutations AIRE cause APS-1 two LD-independent protein-coding variants AIRE with sporadic AAD.Fig. 6T-cell regulation AAD GWAS associated regionsGraphic representation T-cell regulation gene products GWAS association in red AIRE activity medullary thymic epithelial cell promoting expression tissue-restricted antigens (TRAs) education T cells allele frequency 1.5–2.0% population effect variant strongest association p.R471C subtle compared mutations APS-1 p.R471C reported in cases multi-organ autoimmunity type 1 AAD autoimmune thyroiditis32 cases interferon autoantibodies expected APS-133 current evidence support p.R471C cause APS-1 points increased risk AAD population.p.R471 near cysteine residues zinc ion second PHD zinc finger (p.C472 and p.C475) PHD fingers maintain structural integrity methylation states regulate gene expression chromatin regulators transcription transfection assay effect rs74203920 or rs1800520 on AIRE-dependent transcription AIRE expressed thymus HEK293 cells overlap AIRE-regulated genes with primary thymic cells likely p.R471C effect AAD susceptibility from mutations APS-1 variation other nearby genes involved inducible T-cell costimulator (ICOSLG), credible set includes only p.HLA region in autoimmune disease largest risk for AAD dissected HLA-mediated risk associations additional susceptibility risk dominated by alleles HLA class II two major risk haplotypes interaction indicates shared mechanism identified effects by HLA class I aspartic acid in residues 74 156 HLA-B represented in HLA-B*08:01 investigated in studies antigen presentation 21-hydroxylase AAD interactions between HLA class I class II between pairs alleles risk protection larger sample size additional effects could uncovered results provide foundation for work mechanisms HLA-mediated risk functional studies of HLA alleles antigen presentation associations with AIRE to alterations central immunological tolerance AAD development importance correct expression AIRE by Down Syndrome extra copy AIRE chromosome 21, altered expression impaired central tolerance overrepresentation of autoimmune diseases40 other risk loci harbor genes involved in antigen presentation recognition thymocyte maturation HLA class II presents antigens to T cells T-cell antigen receptor) complex regulated by UBASH3A42 immune checkpoint CTLA4 modulates T-cell activation43UBASH3A causal variants suggest role in AAD for TMPRSS3 T cells risk alleles linked to higher expression levels type 1 tryptophan substitution in position 620 PTPN22 disrupts complexes between PTPN22 CSK inhibitory effect TCR signaling attenuated45 BACH2 (CD8+) T-cell differentiation transcription factors differentiation regulatory T cells46 SH2B3 causal autoimmune ATXN2/SH2B3 like inhibitor signaling cascades lymphocytes47 associated variants nearer 5′ end gene LPP harbors microRNA miR-28 posttranscriptional regulation PD148 role self-tolerance restraining autoreactive T cells promoting Tregs50 association peak on chromosome 19 provides three causal units SIGLEC5 SIGLEC14 SPACA6 latter harbors miR-125a posttranscriptional regulation KLF1351 regulates CCL5 activation migration proliferation T cells52 SIGLEC14 activating receptor homologous to SIGLEC5 hard to rule out common polymorphism leads to SIGLEC14/5 fusion gene SIGLEC14-null phenotype53SIGLEC5 recognizes self-cell surface sialoglycans mediates inhibitory signaling T likely candidate Figure 6 T-cell-related effects functional gene products GWAS hits identified robust association signals modest sample sizes homogenous disease etiology low polygenicity high heritability estimates strength study majority AAD patients Norway Sweden national registries geographically matched controls stringent exclusion criteria including 21-hydroxylase autoantibodies results highlight importance central immune AAD Dysregulation antigen presentation negative selection thymus AAD exceptional pathways preventive treatment strategies recruited from Swedish Norwegian Addison Registries fulfilled diagnostic criteria primary adrenal insufficiency low serum cortisol increase plasma adrenocorticotropic corticotropin stimulation test performed Autoimmune etiology confirmed specific autoantibodies targeting adrenal-specific enzyme 21-hydroxylase major autoantigen AAD5 Cases screened for APS-1 autoantibodies against interferon-α interferon-ω interleukin 22, AIRE gene sequencing55 Healthy controls recruited from blood donor centers Sweden Norway coverage subjects gave informed consentstudy performed Declaration of Helsinki approved ethics committees Stockholm Sweden 2008 Western Norway 2013-1504 2017-624) extractionBlood samples kept −80 °C HUNT Laboratory (Levanger DNA isolated MasterPureTM DNA kit normalized to 50 ng/μl 200 ng sample pipetted to 96-well plates Swedish/Norwegian case/control samples distributed Technical replicates included quality control genotype concordance.Genotyping imputation quality controlGenome-wide genotyping 692,367 markers Illumina Infinum Global Screening Array 1.0 Human Genomics Facility Erasmus MC filtered PLINK 1.9 excluded call rate <95% deviation from GenomeStudio genotype clusters call rates <98% excluded call rate <98%, discordant calls replicates deviation from Hardy–Weinberg X chromosome markers tests females only samples accumulated heterozygosity >0.34 excluded.Bi-allelic SNPs QC thresholds Haplotype Reference Consortium panel58 used for phasing imputationGenotypes phased in-house pedigree information (SHAPEIT version 2.r837 non-typed variants imputed Sanger Imputation Service Haplotype Reference Consortium release 1.158 Markers imputation quality score >0.5 MAF > 0.01 included GWAS ancestry inferred LASER/TRACE Human Genome Diversity Project reference Samples non-European excluded Genetic relatedness evaluated high markers pruned LD PLINK cases males selected sample highest call rate retained data from 5320 samples 7.1 million markers kept association testing axes genetic variation assessed principal component analysis high-quality markers MAF > 0.05 pruned for LD (r2 < excluding extended HLA region Association statistics calculated logistic regression disease status genotype dosages Sex top five principal components included covariates differences population stratification analysisLoci passing genome-wide significance threshold subject to conditional analysis associations imputed genotype dosages variants covariates logistic regressionGene prioritization lociEnrichment association signals systems tissues cell types prioritization genes performed using tool DEPICT GWAS statistics used default settings 500 permutations gene length differences 50 repetitions false discovery rates false discovery rate set 5% loci plotted with LocusZoom62 FINEMAP63 two runs 1 Mb windows lead SNPs each genome-wide association peak HLA assessed maximum one or three causal SNPs per window per run default settings small studies benefits fine-mapping imputation dissectionClassical HLA alleles imputed from genotyped SNPs HIBAG kernel version 1.465 SNP2HLA version 1.0.366 classifiers reference panels European samples alleles HLA-A -B -C -DQA1 -DQB1 -DRB1 precision DRB1-allele calls built model HLA imputation 699 healthy Norwegian controls DRB1 alleles called predictions default in-house model weighted by size training data Genotypes fixed if posterior probability 0.5 twice likely second most probable call370 case samples laboratory typing HLA-A -B -DQB1 -DRB1) available Concordance 92–98% HIBAG SNP2HLA 90% DRB1 aimed main drivers risk HLA alleles amino acids HLA class I II method adapted from Moutsianas et al.67 every HLA allele constructed logistic regression models used Bayesian information criterion likelihood ratio test effect additive recessive dominant Supplementary Note 3 selection procedure allele amino acid residue effect significant model considered inclusion full model Five PCs sex included covariates models PCs calculated from SNPs genome-wide not including HLA region same covariates GWAS analysis considered inclusion alleles/amino acids genome-wide significance P < 5 × 10−8 model Backward elimination previous variables out testing goodness of fit common diseases traitsGenetic overlap investigated method Farh et al.68 GWAS catalog data EMBL-EBI website December 201969 Diseases/traits six reported associations (P 1 × 10−6) included many diseases GWAS multiple associated SNPs same locus most significant index SNP considered1990 diseases/traits 5349 SNPs fulfilled criteria used associations within 500 kb Overlapping risk loci used disease-by correlation matrix selected 22 autoimmune 19 non-autoimmune diseases/traits for comparison in Fig. 5a statistics four autoimmune diseases for locus used LD score regression genetic correlation between GWAS results 844 traits at LD Hub regression uses GWAS statistics excludes HLA region estimating genetic correlations heritability variation HLA.3D models PHD domain acids HLA-DQA1 HLA-DQB1 genes x-ray diffraction model of HLA-DQ2.3 used (Protein Data Bank 20 4D8P protein models of PHD2 domain potential structural impact of AIRE variants used NMR structure for PHD2 template (Protein Data Bank 2LRI domain structure mutated at p.R471 position to cysteine visualized using PyMOL information research design Nature Research Reporting Summary.Supplementary Review FileReporting Additional Supplementary Data | 49.9 | 0.72098 |
10.1038/s41467-020-14591-8 | PMC7026101 | Imagers capable of reconstructing three-dimensional scenes in the presence of strong background noise are desirable for many remote sensing and imaging applications. Here, the authors report an imager operating in photon-starved and noise-polluted environments through quantum parametric mode sorting. | Active imagers capable of reconstructing 3-dimensional (3D) scenes in the presence of strong background noise are highly desirable for many sensing and imaging applications. A key to this capability is the time-resolving photon detection that distinguishes true signal photons from the noise. To this end, quantum parametric mode sorting (QPMS) can achieve signal to noise exceeding by far what is possible with typical linear optics filters, with outstanding performance in isolating temporally and spectrally overlapping noise. Here, we report a QPMS-based 3D imager with exceptional detection sensitivity and noise tolerance. With only 0.0006 detected signal photons per pulse, we reliably reconstruct the 3D profile of an obscured scene, despite 34-fold spectral-temporally overlapping noise photons, within the 6 ps detection window (amounting to 113,000 times noise per 20 ns detection period). Our results highlight a viable approach to suppress background noise and measurement errors of single photon imager operation in high-noise environments. | IntroductionThree-dimensional (3D) imaging technology has found applications across diverse disciplines, including machine-vision and ranging1,2, terrestrial mapping3, remote sensing, and environmental monitoring4–6. Active 3D imaging, which captures spatial and temporal information by detecting the reflected probe signal from the scene of interest, is becoming an important tool for extending human’s visual perspectives. For such, the detector characteristics of a 3D imaging system, like the operation mode, detection sensitivity, timing resolution, and dynamic range, are critical to the scene reconstruction1. The latest single photon detection techniques7 can faithfully detect light in its smallest quanta (i.e., single photons), enabling active imaging upon very low flux of backscattered photons8, extending its range to tens of kilometer ranges9. Combining single photon detection with computational techniques—such as image reconstruction algorithms10–13, signal processing8,14,15, and artificial intelligence assisted imaging16,17—has provided imaging capabilities under extreme conditions. For example, leveraging the spatial correlation of the target and the physics of low-flux measurement, first-photon imaging was developed to realize 3D imaging from only one detected photon per pixel, even with the presence of ambient noise photons15. More recently, low-flux 3D imaging beyond the line-of-sight has been demonstrated18,19, where the need for computationally intensive reconstruction algorithms has been significantly relieved by the use of light-cone transform enabled confocal scanning technique20.Under practical circumstances, the signal photons usually return along with strong background noise spanning the same spectral and temporal domains, making them indistinguishable to the detector14,21. Conventional approaches of isolating signal from noise, including those by matched time-frequency filters, are intrinsically limited by the trade-off between signal detection efficiency and noise rejection22,23. This limit applies to any linear-optical filtering approach, including those widely employed by computational enhanced imaging techniques21,24. Recognizing this, quantum parametric mode sorting (QPMS) has been proposed for mode-selective quantum frequency conversion22, following the pioneering studies of quantum pulse gating25 and quantum optical arbitrary waveform generation and measurement26. It is implemented by driving the conversion with pump pulses whose spectral width is comparable to its phase matching bandwidth, i.e., at the edge of phase matching. Under this condition, only signal photons in a single spatiotemporal mode, whose profile can be flexibly tailored by shaping the pump pulses, can be converted efficiently. Undesirable photons in other modes, even if they spectrally and temporally overlap with the signal, are converted with much less efficiency27,28. This exotic mode selectivity thus realizes superior nonlinear-optical filtering, which was demonstrated to achieve detection signal to noise more than 40 dB over a linear-optical filtering and detection system, and beat the theoretical limit of an ideal matched filter by 11 dB22,29.Active imaging also faces challenges caused by photons backscattered before the scenes of interests. As such, the confocal non-line-of-sight and first-photon imaging techniques would struggle to image a fog-besieged or highly-obscured target. This is because a free running single photon detector is likely to be saturated by photons scattered from the thick fog or the obscurant20, blinding it to the photons carrying information about the target. Even though estimation algorithms were recently demonstrated to improve the accuracy in arrival-time and reflectivity estimations14,24,30, those post-processing methods are computationally expensive and incapable of eliminating the fundamental uncertainties arising from Poisson (photon number) noise or distortions induced by the pileup effect.In this work, we address those limitations and extend active 3D imaging to robust operations in photon-starved and noise-polluted environments through the use of QPMS22,29. It harvests the maximum noise rejection by QPMS and efficient photon detection by a silicon avalanche photodiode (Si-APD). The signal to noise advantage is 36 dB over direct detection using an InGaAs single photon detector with a 1 ns gated detection window29, or 7 dB above the theoretical limit of ideal linear-optical matched filters22. This allows active 3D imaging in a scenario where the background is orders of magnitude stronger than the backscattered signal, with 34 times more spectral-temporally overlapping noise than signal photons. In addition, we highlight the extreme ranging resolution of our imaging technique by imaging through a highly reflective obscurant without being impeded by the pileup distortions, dead time, and count rate saturation issues that plague many other single photon imagers.ResultsHigh resolution single photon sensitive 3D imagingOur 3D imager utilizes an upconversion single photon detector (USPD) capable of QPMS due to carefully selected pump and probe pulses. The setup is shown in Fig. 1 (see Methods section for details). The pump and probe pulses are carved from a 50 MHz femtosecond mode-locked laser (MLL) using separate sets of cascaded spectral filters. The probe pulses are sent out to raster scan the target scene via an optical transceiver and programmable scanning MEMS mirror. Meanwhile, the pump pulses are sent to a programmable optical delay line (ODL). They are then combined with the backscattered probe signal and sent to the USPD. The internal conversion efficiency of the waveguide is 121% W−1 cm−2, and the total detection efficiency of the USPD is 3.6% with total dark count rate of 250 Hz (as compared with about 1000 Hz for a typical 1-ns gated InGaAs APD)29. The dark counts of USPD are primarily attributed to Raman noise photons generated in the upconversion waveguide22 as the Si-APDs 50 Hz dark count level is negligible. Note that Raman noise photons are maximally filtered out by QPMS due to the mode selectivity of our imager29,31. The robust all-fiber design of our imager reduces the number and footprint of required optical components while simplifying the optical alignment procedure. Finally, a field-programmable-gate-array (FPGA) is employed as the central processor for controlling the MEMS mirror and ODL, and to collect data from the USPD.Fig. 1Experimental setup of noise tolerant 3D single photon imaging.MLL, mode-locked fiber laser (Repetition rate = 50 MHz, Center wavelength = 1560 nm); MEMS, micro-electro-mechanical systems; ODL, optical delay line; ASE, amplified spontaneous emission (1520–1570 nm); USPD, upconversion single photon detector; Si-APD, silicon avalanche photodiode; FPGA, field-programmable-gate-array. Pump (1565.5 nm) and probe (1554.1 nm) pulses are carved out from the MLL using off-the-shelf 200 GHz telecom wavelengths dense-wavelength-division-multiplexing (DWDM) filters. The FPGA is the central processor for controlling the programmable MEMS mirror and programmable ODL, and acquiring the data from the USPD. The noise, which is covering the identical spectral-temporal region as probe signal, is carved out from an ASE source with a DWDM filter.To reconstruct the 3D image of the target scene, we measure the time-resolving upconversion signal to retrieve the time-of-flight (ToF) information of the backscattered signal while raster scanning the probe beam across the scene. We use a pixelwise maximum-likelihood value (MLV) approach where a time-resolving measurement is performed for each pixel by counting the number of upconverted photons as a function of the temporal delay between the synchronous probe and pump pulses. The temporal delay is scanned using the programmable ODL where backscattered signal photons that are temporally aligned with the pump pulses in the PPLN will be upconverted efficiently and detected by the Si-APD. In contrast, photons distributed in orthogonal time-frequency modes will be upconverted with very low efficiency, even when they temporally overlap with the pump pulses29,32. The scanning of the ODL creates a photon-counting histogram for the ODL points, and the MLV estimate is the point with the most detections. For post-processing, a standard MATLAB median filter is applied to smooth the reconstructed image. The filter converts every pixel value to the median of a 3 × 3 region consisting of the pixel and its eight nearest neighbors. This commonly used image processing tool is preferred over mean filtering because it is robust against bias from outlier values. An example of a complete time-resolved (longitudinal) upconversion signal from two different transverse locations is shown in Fig. 2a. Note that the time-resolving upconversion photon counting reflects the intensity-correlation of pump and backscattered probe pulses, where the measured width is about 9 ps (versus ~500 ps for typical InGaAs APDs). This shows the ultrahigh timing resolution of our system upon a single-detection event, which is defined by the phase matching of the upconversion and the full-width-half-maximum (FWHM) of optical pulses. It represents almost one order of magnitude improvement over the resolution achievable with Si-APDs whose timing jitter is typically capped at 50 ps. Sub-picosecond timing resolution is achievable with shorter optical pulses that can be created using spectrally wider filters and waveguides with broader phase matching33.Fig. 2High resolution single photon sensitive 3D imaging.a Time-resolving photon counting from two different longitudinal positions. b MLV histogram for two regions (A and B) within the dotted lines in (c). c 5 cm × 7 cm CNC machined aluminum depth resolution calibration chart. d Reconstructed image of the depth resolution chart in (c).Our system can achieve even higher ranging resolution by using the ODL to acquire the full upconversion histogram. We experimentally determine that a peak of 150 detections (equivalent to 0.006 photon detections per pulse over 500 μs dwell time) is sufficient to reach the minimum standard deviation of 0.5 ps for a single time-resolving measurement (see Supplementary Note 5). Reduced standard deviation in temporal measurement is translated into improvement in the ranging resolution15 as depicted in reconstructed 3D image (Fig. 2d).To assess the depth resolving capability of our imager, we carry out 3D imaging with 2400 pixels on a machined depth resolution chart placed 1.5 m from the transceiver. The chart is a 50 mm × 70 mm aluminum block with 20 depth varying circles of 7 mm diameter (see Fig. 2c). The bottom row contains four reference circles of 1 mm depth. The depth of the remaining circles ranges from 1 to 2.5 mm in steps of 0.1 mm. The reconstructed 3D image is shown in Fig. 2d, where the ODL was scanned in 1 ps steps over a range of 30 ps. The dwell time for each pixel is 1 ms per ODL sample, rendering a total data acquisition time of 72 s. This time can be substantially reduced by using shorter dwell time and fast ODLs, such as those based on solid-state switches. Shown in Fig. 2b, we plot the MLV histograms for all pixels contained within the dotted lines (regions A and B) of Fig. 2c. Each histogram contains two features, with the left-most from pixels on the surface around the indented circle and the right-most from pixels within the indented circle. The separation between these two features gives a measured ToF difference corresponding to the depth of the circle compared to the surrounding surface. Additionally, the depth profiling accuracy of our 3D imager is determined to be approximately 0.09 mm by measuring it against a calibrated linear translational stage and a certified gauge (see Supplementary Note 4). The transverse spatial resolution of this imager is currently limited by the collimated optical beam diameter (2.2 mm).Imaging through an obscuring objectOur 3D imager is advantageous in discriminating objects of interest in a complex environment with multiple reflecting surfaces owing to the picosecond pump pulses that are physically time-gating the backscattered photons. Therefore, we manage to accept photons from the target while discounting photons arriving at different times, even as close as several picoseconds away. This also effectively removes the distorting pileup effect14 that can arise from repeated detection of undesired photons within a larger detection window. The performance of our technique shows improvement in terms of sensitivity, timing resolution, accuracy, bias, and dead time14.We perform 3D imaging through an obscurant where a ceramic mannequin head is located 2 mm behind an obscuring aluminum wire mesh with 1 mm2 holes, as depicted in Fig. 3a, c. Note that the ToF difference between the mesh and the target is well below the 50 ps timing resolution of a typical Si-APD. The highly reflective mesh reduces the backscattered photons from the mannequin by an average of 5 dB while inducing backscattering ahead of the desired target. Nonetheless, our 3D imager is capable of resolving both the mannequin head and the aluminum mesh as shown in Fig. 3b. Additionally, time resolving feature of our imager allows us to isolate mannequin head as depicted in Fig. 3d. Our system distinguishes between the two reflecting surfaces by identifying separate peaks in the upconversion signal. This is demonstrated in Fig. 3e–g which show the time-resolving photon counting measurements for different pixels along the nose of the mannequin head. The left peak represents the backscatter from the mesh and the right peak represents the backscattered signal from the mannequin. The short time-gate created by picosecond pulses enables the immaculate retrieval of 3D information from the mannequin head despite the presence of backscatter from the obscurring material. The optical-gating advantage of our 3D imaging technique can be extended for use in distinguishing objects in a complex environment19,20 and other applications where precise detection of faint optical pulses is needed34,35.Fig. 3Imaging through obscuring object.a Close-up photo showing the milimetric distance of obscuring aluminum wire mesh (1 mm2 opening) and target (head mannequin). b Reconstructed 3D image showing the obscurant and target, resembling a. c Obscuring aluminum grid and target looking at probe signal's propagation direction. d Reconstructed 3D image of target isolated from obscurant. e–g Time-resolving photon counting obtained at different raster scanning position, showing distinguishable backscattering signal peaks from the obscurant and target.Noise-tolerant imagingTo demonstrate the noise tolerance of our technique, we perform 3D imaging in a scenario where the backscattered signal photons spectrally and temporally overlap with strong background noise. The noise photons are generated by using a DWDM filter to spectrally carve the amplified spontaneous emission (ASE) from an erbium doped fiber amplifier (EDFA). The filter is similar to the one used to carve the signal from the MLL to ensure that the signal and noise match well in spectrum. Then, we mix the noise with the backscattered signal using a 99:1 fiber beam splitter, as shown in Fig. 1.We test active imaging under two different levels of background noise, marked by the number of ASE noise photons per period of the signal pulses (20 ns). The results for 350 and 1700 background photons per period are shown in Fig. 4. These noise levels amount to incident signal to background ratios of 1/23,000 and 1/113,000, which is well beyond the capabilities of computational imaging post-processing36. For a typical direct detection system using an InGaAs APD with 1 MHz maximum counting rate, 7.5% detection efficiency, and 1 ns time-gating window, those levels correspond to 1.3 and 6.4 mean detected photons per detection window. In both cases the InGaAs detector would be saturated by the noise and therefore unable to detect the low-flux signal. In contrast, the mean photon number detected by QPMS per pulse is only 0.0004 and 0.0016, respectively. Hence, the current QPMS implementation gives an average of about 36 dB advantage in background suppression over direct detection.Fig. 4Noise-tolerant imaging.a Noise free raw 3D image of target, acquired with mean photon number of 15 detections per pixel. b Reconstructed 3D image acquired with 34 times noise photons after a median filter has been applied. c Error map of reconstructed 3D image (b), with noise free 3D image (a) as ground truth. d, e The raw 3D images acquired with the presence of 7 and 34 times noise. f Time-resolving photon counting at different noise levels, where the dashed lines mark the average number of 7 (green) and 34 (red) times noise photons counted at each ODL point.This substantial advantage is attributed to the reduced detection mode number, polarization sensitive upconversion, and mode selectivity against spectrally and temporally overlapping noise17,28,29. In the case of direct detection, the DWDM filter has the spectral bandwidth of B = 250 GHz and detector’s gating window T = 1 ns, which combined to a total detected time-frequency modes of πBT∕2 ≈ 392. For QPMS, the phase matching profile of upconversion waveguide limits the spectral bandwidth to B = 90 GHz while the effective detection window of T = 6 ps is defined by the FWHM of pump pulse. Evidently, the QPMS reduces the number of detection modes to ≈1, resulting a total noise suppression advantage of 25.9 dB. Theoretically, similar noise suppression can be achieved by using a combination of ideally matched time-frequency filters. However, it will be difficult to implement in practice for such ultrashort and faint optical pulses. Additionally, upconversion is efficient only along a single polarization, which offers another 3 dB noise suppression by rejecting half of randomly polarized noise photons. Finally, the mode selectivity contributes an additional 7.1 dB to the total 36 dB advantage in background suppression observed, agreeing well with the 7.4 dB mode selectivity predicted by simulation result (see Supplementary Note 3). Even higher selectivity can be achieved by using a longer PPLN waveguide, tailored phase matching profile or optimally designed pump pulses27,28,37.The exceptional noise rejection by QPMS not only defeats the detector saturation, but also helps accurate image reconstruction. This is because high levels of detected noise necessarily leads to image distortion and errors by virtue of its Poissonian variance. Figure 4a, d, e shows how the number of erroneous pixels on the raw 3D images grows quickly with the background level. Figure 4c shows that considerable features on the mannequin head are concealed by the noisy pixels. This highlights the importance of background rejection when the backscattered signal photons are scarce.Thanks to QPMS’ exceptional signal to noise, our 3D imager is able to recover the salient details of the target scene without the use of any post-selective filtering, despite overwhelming noise. In Fig. 4a, b, we compare the reconstructed 3D image without noise and with 1700 background photons per period, acquired with the same signal photon detection rate. Figure 4c shows the depth error (standard deviation) map between these two figures. As seen, the error occurs mainly around high contrast features of the mannequin, e.g., the nose and edge of the face, whose average depth (longitudinal) error is 1 mm. In general, the higher error in those locations is caused by the sharp angle of incidence and thus the dramatically reduced backscattered signal. Consequently, noisy pixels that are not reflecting the true surface morphology are dominant in those areas. Figure 4f gives an example of the time-scanned photon counts acquired by QPMS, where the photon counting peak is still identifiable despite orders of magnitude higher noise. Suppressing noise on the detection end will reduce the number of noisy pixels, allowing for image reconstruction with reduced error. Nevertheless, the errors in the reconstructed 3D image can be further minimized through post-selective filtering that distinguishes between signal and background detections by exploiting the scene’s transverse spatial correlation15.DiscussionWe demonstrate a noise-tolerant 3D imager that combines the exceptional noise rejection capabilities of QPMS with efficient single photon counting using a Si-APD. The use of QPMS gives a 36 dB advantage in noise rejection over typical direct photon detection with linear-optical filters, or 7.1 dB above the theoretical limit of a matched time-frequency filter. This advantage, made solely on the detection end, allows us to perform 3D imaging with weak returning signal at 0.0006 mean photon detection per pulse despite orders of magnitude stronger background, including the presence of 34-fold spectral-temporally overlapping noise photons. As such, our technique can be particularly useful for extending high-resolution 3D imaging over long distance and strong noise, such as bright solar background9. The same advantage may assist deep space communications38.Furthermore, with single photon detection of picosecond resolution, our 3D imager achieves 100 µm ranging resolution without any use of computationally intensive image reconstruction algorithms. It thus circumvents the limiting factors of detector time-jittering and pileup distortion, and may find impactful applications in non-line-of-sight and obscured environment imaging. Finally, the present LiDAR technique based on QPMS is applicable to a wide range of wavelengths, including those in the mid-IR regime31.MethodsImaging setupTwo nearly transform-limited, 6 ps pulses at 1554.1 nm (probe) and 1565.5 nm (pump) are carved out from the MLL using a pair of cascaded 200 GHz dense-wavelength-division-multiplexing (DWDM) filters each. We measure the pulse’s intensity and phase profile in both spectral and temporal domains using a frequency resolved optical gating (FROG) pulse analyzer, thereby allowing us to quantify the mode selectivity of upconversion detection22,29 (See Supplementary Note 3). Collimated signal probe pulses (Gaussian beam diameter: 2.2 mm) at 1554.1 nm are transmitted toward the scene through a transceiver. A fiber circulator separates the outgoing signal pulses and the incoming backscattered photons with a minimum isolation ratio of 55 dB. The transceiver of the imager is based on a simple monostatic coaxial arrangement using off-the-shelf telecom-grade optical components. The backscattered signal photon will be recombined with pump pulse via another DWDM and subsequently fiber coupled into the mode-selective upconversion detector. Details about the fiber pigtailed detector are explained in Supplementary Note 1. An FPGA is used as central processor for controlling the MEMS mirror and ODL, and acquiring the data from the USPD.Supplementary information
Supplementary Information
Peer Review File | nature communications | [
"Article"
] | [
"Single photons and quantum effects",
"Imaging and sensing",
"Optical metrology"
] | -dimensional (3D imaging technology applications across disciplines machine-vision terrestrial remote sensing environmental Active 3D imaging captures spatial temporal information detecting reflected probe signal scene important extending visual perspectives detector characteristics operation mode detection sensitivity timing resolution dynamic range critical to scene reconstruction1 single photon detection detect light smallest active imaging low flux backscattered extending range to tens kilometer Combining single photon detection with computational image reconstruction signal artificial imaging capabilities under extreme conditions spatial correlation low-flux measurement first-photon imaging 3D imaging from one detected photon per pixel ambient noise low-flux 3D imaging beyond line-of-sight demonstrated18 need intensive reconstruction algorithms relieved by light-cone transform confocal scanning signal photons return with background noise indistinguishable to detector14 Conventional approaches isolating signal from noise time-frequency filters limited by trade-off signal detection efficiency noise rejection22limit applies linear-optical filtering including computational enhanced imaging techniques21 quantum parametric mode sorting) proposed for mode-selective quantum frequency conversion22 following studies quantum pulse gating25 quantum optical arbitrary waveform generation measurement26 implemented by conversion with pump pulses spectral width comparable to phase matching bandwidth signal photons in single spatiotemporal mode tailored converted efficiently Undesirable photons in other modes converted less efficiency27 exotic mode selectivity realizes superior nonlinear-optical filtering detection signal to noise 40 dB over linear-optical limit of ideal matched filter by 11 dB22.Active imaging faces challenges photons backscattered before confocal non-line-of-sight first-photon imaging techniques struggle image fog-besieged highly-obscured target single photon detector saturated by photons fog blinding photons estimation algorithms improve accuracy in arrival-time reflectivity estimations14 computationally expensive eliminating uncertainties Poisson (photon number) noise distortions pileup effectlimitations extend active 3D imaging photon-starved noise-polluted environments QPMS22 maximum noise QPMS efficient photon detection silicon avalanche photodiode signal to noise advantage 36 dB over direct detection InGaAs single photon detector 1 ns 7 dB above limit linear-optical matched allows active 3D imaging background stronger backscattered signal 34 times more overlapping noise signal photons highlight extreme ranging resolution reflective obscurant without pileup distortions dead time count rate saturation issues resolution single photon sensitive 3D upconversion single photon detector (USPD QPMS pump probe pulses setup Fig. 1 pump probe pulses 50 MHz femtosecond mode-locked laser cascaded spectral filters probe pulses scan target scene optical transceiver MEMS mirror pump pulses programmable optical delay line combined with backscattered probe signal sent USPD internal conversion efficiency 121% W−1 cm−2 total detection efficiency USPD 3.6% dark count rate 250 Hz 1000 Hz typical 1-ns gated InGaAs APDdark counts USPD attributed to Raman noise photons upconversion Si-APDs 50 Hz dark count negligible Raman noise photons filtered by QPMS mode selectivity all-fiber design reduces optical components optical alignment field-programmable-gate-array (FPGA) MEMS mirror ODL data USPD. 1Experimental setup noise tolerant 3D single photon imaging.MLL mode-locked fiber laser rate 50 MHz wavelength 1560 nm); MEMS-electro systems ODL optical delay line ASE amplified spontaneous emission (1520–1570 nm); USPD upconversion single photon detector Si-APD silicon avalanche photodiode FPGA-programmable-gate-array Pump (1565.5 nm probe (1554.1 nm) pulses carved from MLL 200 GHz-multiplexing filters FPGA MEMS mirror ODL data USPD noise carved out from ASE source DWDM filter 3D image time-resolving upconversion signal-flight information probe beamuse pixelwise maximum-likelihood value (MLV) approach time-resolving measurement pixel counting upconverted photons temporal delay between probe pump pulses temporal delay scanned programmable ODL backscattered signal photons aligned with pump pulses upconverted efficiently detected Si-APD photons orthogonal time-frequency modes upconverted low efficiency pump scanning ODL creates photon-counting histogram points MLV estimate point most detections post-processing MATLAB median filter applied reconstructed image converts pixel value to median 3 × 3 region eight neighbors preferred mean filtering robust against bias example complete time-resolved upconversion signal two locations in Fig. 2a time-resolving upconversion photon counting reflects intensity-correlation pump probe pulses measured 9 ps (versus ~500 ps typical shows ultrahigh timing resolution system single-detection event defined by phase matching upconversion full-half-maximum) optical pulses improvement over resolution Si-APDs timing capped 50 ps Sub-picosecond timing resolution achievable with shorter optical pulses wider filters waveguides broader phaseFig. 2High resolution single photon 3D imaging Time-resolving photon counting two positions MLV histogram regions B dotted lines 5 cm × 7 cm CNC aluminum depth resolution calibration chart Reconstructed image depth chart system higher resolution ODL full upconversion histogram 150 detections 0.006 photon per pulse over 500 μs dwell time minimum standard deviation 0.5 ps time-resolving measurement Reduced standard deviation improvement ranging reconstructed 3D image (Fig. depth resolving 3D imaging 2400 pixels machined depth resolution chart 1.5 m from transceiver chart 50 mm × 70 mm aluminum block 20 depth varying circles 7 mm diameter bottom row four reference circles 1 mm depth 1 to 2.5 mm steps 0.1 mm reconstructed 3D image Fig. 2d ODL scanned 1 ps steps 30 ps dwell time pixel 1 ms per ODL total data acquisition time 72 s shorter dwell time fast ODLs Fig. 2b MLV histograms pixels dotted lines (regions A B 2c Each histogram two features left-most right-mostseparation between features gives ToF difference to depth circle surface depth profiling accuracy of 3D imager approximately 0.09 mm against linear translational stage certified gauge transverse spatial resolution limited by collimated optical beam diameter (2.2 mm).Imaging through obscuring 3D imager objects in complex environment with multiple reflecting surfaces picosecond pump pulses time-gating photons photons from target photons different times removes distorting pileup repeated detection photons shows sensitivity timing resolution accuracy bias dead 3D imaging through obscurant ceramic mannequin head 2 mm behind obscuring aluminum wire mesh with 1 mm2 holes Fig. 3a, c ToF difference between mesh target below 50 ps timing resolution of typical Si-APD reflective mesh reduces backscattered photons from by 5 dB backscattering ahead target 3D imager both mannequin head and aluminum mesh Fig. 3b time resolving feature mannequin head Fig. 3d system distinguishes reflecting surfaces separate peaks in upconversion signal demonstrated in Fig.3e–g show time-resolving photon counting measurements pixels nose mannequin head left peak backscatter mesh right peak backscattered signal mannequin short time-gate picosecond pulses enables retrieval 3D information from mannequin head despite backscatter material optical-gating advantage extended distinguishing objects complex precise detection faint optical pulses.Fig. 3Imaging through obscuring object Close-up photo milimetric distance obscuring aluminum wire mesh target Reconstructed 3D image obscurant target aluminum grid target signal 3D image target isolated from obscurant. e–g Time-resolving photon counting raster scanning position backscattering signal peaks from obscurant target.Noise-tolerant 3D imaging backscattered signal photons overlap with background noise noise photons generated DWDM filter amplified spontaneous emission) from erbium doped fiber amplifier filter similar noise with backscattered signal using 99:1 fiber beam splitter Fig. test active imaging under two levels background noise ASE noise photons per period signal pulses results for 350 1700 background photons per period shown in Fig.noise levels signal background ratios 1/23,000 1/113,000 beyond computational imaging direct detection system InGaAs APD 1 MHz counting rate 7.5% detection efficiency 1 ns time-gating window 1.3 6.4 photons per window InGaAs detector saturated noise low-flux signal mean photon QPMS per pulse 0.0004 0.0016 QPMS 36 dB advantage background suppression over direct detection. 4Noise-tolerant imaging Noise free 3D image 15 detections per pixel Reconstructed 3D image 34 times noise photons filter Error map 3D image noise free 3D images 7 34 times noise Time-resolving photon counting noise levels lines average 7 34 times noise photons each point advantage reduced detection mode number polarization sensitive upconversion mode selectivity against overlapping DWDM filter spectral bandwidth B = 250 GHz gating window T = 1 ns detected time-frequency modes πBT∕2 ≈ 392 QPMS limits spectral bandwidth to B = 90 GHz effective detection window T = 6 ps defined by FWHM pump pulseQPMS reduces detection modes to ≈1 noise suppression advantage 25.9 dB similar noise suppression matched time-frequency filters difficult to implement for ultrashort faint optical pulses upconversion efficient single polarization 3 dB noise suppression rejecting half randomly polarized noise photons mode selectivity contributes 7.1 dB to 36 dB advantage background suppression with 7.4 dB mode selectivity simulation Supplementary Note higher selectivity longer PPLN waveguide tailored phase matching profile designed pump exceptional noise QPMS defeats detector saturation helps image reconstruction high noise leads to image distortion errors Poissonian variance Figure erroneous pixels 3D images grows with background level features on mannequin head concealed by noisy pixels importance background rejection when backscattered signal photons scarce QPMS’ signal to noise 3D imager salient details scene without post-selective filtering despite noise. reconstructed 3D image without noise 1700 background photons per period Figure 4c error map error occurs around high contrast features mannequinnose edge face average depth error 1 mm higher error caused by sharp angle reduced backscattered signal noisy pixels not reflecting surface morphology dominant Figure 4f time-scanned photon counts by QPMS photon peak identifiable despite higher noise Suppressing noise detection end noisy pixels image reconstruction reduced error errors 3D image minimized through post-selective filtering signal background spatial noise-tolerant 3D imager noise rejection QPMS single photon counting Si-APD QPMS 36 dB advantage noise rejection over direct photon detection 7.1 dB above limit matched time-frequency filter advantage 3D imaging with weak returning signal at 0.0006 mean photon detection per pulse despite stronger background 34-fold overlapping noise photons technique useful for high-resolution 3D imaging over long distance strong noise solar deep space communications38 single photon detection picosecond resolution 3D imager achieves 100 μm resolution without intensive image reconstruction algorithmscircumvents detector time-jittering pileup distortion applications non-line-sight obscured environment imaging LiDAR technique QPMS applicable wavelengths including mid-IR setupTwo transform-limited 6 ps pulses at 1554.1 nm (probe 1565.5 nm (pump from MLL cascaded 200 GHz-wavelength-division-multiplexing (DWDM filters measure intensity phase profile spectral frequency (FROG) pulse analyzer quantify mode selectivity upconversion Supplementary Collimated signal probe pulses diameter 2.2 mm at 1554.1 nm transmitted scene through transceiver fiber circulator separates pulses photons minimum ratio 55 dB transceiver monostatic coaxial arrangement telecom-grade optical components backscattered signal photon recombined with pump pulse DWDM fiber coupled into mode-selective upconversion detector Supplementary Note 1. FPGA controlling MEMS mirror ODL acquiring data USPD.Supplementary information | 48.5 | 0.458014 |
10.1038/s41467-020-19044-w | PMC7566455 | CuP2 has a puzzling thermal transport behavior, with low thermal conductivity but quite large mean sound speeds. Here, the authors conduct a systematical study of the atomic structure and lattice dynamics of CuP2 to reveal the origin, finding a dimer rattling behavior. | A solid with larger sound speeds usually exhibits higher lattice thermal conductivity. Here, we report an exception that CuP2 has a quite large mean sound speed of 4155 m s−1, comparable to GaAs, but single crystals show very low lattice thermal conductivity of about 4 W m−1 K−1 at room temperature, one order of magnitude smaller than GaAs. To understand such a puzzling thermal transport behavior, we have thoroughly investigated the atomic structures and lattice dynamics by combining neutron scattering techniques with first-principles simulations. This compound crystallizes in a layered structure where Cu atoms forming dimers are sandwiched in between P atomic networks. In this work, we reveal that Cu atomic dimers vibrate as a rattling mode with frequency around 11 meV, which is manifested to be remarkably anharmonic and strongly scatters acoustic phonons to achieve the low lattice thermal conductivity. | IntroductionThermal conduction is one of the most fundamental physical properties of materials1. Materials with low thermal conductivity are desirable for a great variety of applications such as thermal insulation2, phase transition memory devices3, and thermoelectric energy conversion4. In electrically insulating nonmagnetic systems, phonons are the major heat carrier, and lattice thermal conductivity is proportional to the product of the square of sound speeds and phonon lifetimes5–7. To decrease the thermal conductivity, a general approach is to reduce the phonon lifetimes through enhancing phonon–phonon and phonon-disorder scattering. Phonon–phonon scattering, also known as phonon anharmonic interactions, dominates the thermal conductivity of disorder-free systems like PbTe8 and SnSe9. There is a special case called rattling, i.e., a localized vibrational mode scattering acoustic phonons with an indication of anti-crossing points in the phonon dispersions10. This idea has been broadly applied to rationalize the low thermal conductivity in phonon-glasses thermoelectric materials like filled skutterudites and clathrates11–13. In addition, spatially hierarchical atomic disorder has also been widely used, including solid solutions of alloys14, nanostructures15,16, liquid-like disorder17,18 and so forth, to lower the thermal conductivity through phonon-disorder scattering. For a variety of disorder-free compounds, we account for the relationship between thermal conductivity and sound speeds in Fig. 1a (ref. 19). In the logarithmic scale, most of the compounds are distributed around an empirical straight line. For a given thermal conductivity value the sound speeds are the dominating factor for materials above the line while the lattice anharmonicity governs thermal transport of materials below the line. Located in the line, single-crystalline diamond20,21 has the highest lattice thermal conductivity among those materials, 2400 W m−1 K−1, accommodating a very high mean sound speed of 14,400 m s−1. GaAs is also situated near the line with a mean sound speed of 3627 m s−1, whereas CuP2 has a similar mean sound speed, but its thermal conductivity is almost ten times lower22,23.Fig. 1Lattice thermal conductivity and crystal structure of CuP2.a A survey of lattice thermal conductivity of disorder-free materials vs. mean sound speed vm. The detailed data are listed in Supplementary Table 1 (for polycrystalline samples, please refer to Supplementary Fig. 1 and Table 2). b Temperature dependence of the lattice thermal conductivity of the CuP2 single crystal is shown from 50 to 300 K measured by a steady-state comparative method. The inset shows the data near room temperature. The error bars indicate the uncertainty (see “Methods”). c The layered structure with Cu dimer layers and P network layers are highlighted. The arrows in orange on Cu atoms represent the vibrational directions at the Γ point of the optical phonon mode located at about 11 meV (see Fig. 4). The atomic motions of this mode are also displayed in Supplementary Movie 1. d The isolated Cu dimer layers with intra- and inter-dimer distances labeled.The temperature dependencies of lattice thermal conductivity (κlat) of CuP2 single crystals along bc plane and out of bc plane are shown in Fig. 1b in the temperature region from 50 to 300 K, respectively. The κlat values in both directions increase as temperature decreases. At 300 K, κlat along the bc plane is 4.66 W m−1 K−1, while 3.57 W m−1 K−1 out of the bc plane. κlat for a polycrystalline sample is shown in Supplementary Fig. 1. At room temperature, it is about 0.62 W m−1 K−1. This value is greatly reduced compared to the single crystals, which might be attributed to the grain boundaries and defects scattering24. In order to understand the mechanism responsible for such a low thermal conductivity in CuP2, we conducted a systematical study of the atomic structure and lattice dynamics of this material.ResultsAtomic structuresThis compound crystallizes in a relatively simple monoclinic structure with space group P21/c (ref. 25). As shown in Fig. 1c, the structure is characteristic of a layered configuration where Cu layers (spacing 1.59 Å) and P network (spacing 3.85 Å) repeat alternatively along the a-axis. The Cu atoms forming a dimer occupy an equivalent position. The arrows in orange represent the vibration directions of Cu atoms associated with the rattling mode to be described later. As shown in Fig. 1c, d, the intra-dimer distance is 2.48 Å, while the inter-dimer distances are 3.85, 4.81, and 7.53 Å along a, b, and c directions, respectively. The electron localization function (ELF) calculation was used to support the weak bonding nature between Cu dimers as a direct theoretical evidence (see Supplementary Fig. 2). ELF around Cu atoms is around 0.2, which suggests a weak metallic bond between Cu atoms within Cu dimers, while ELF is almost zero between Cu dimers. This distinction along with the calculated charge density well supports the dimer nature. In consideration of the isolated nature of the Cu dimer, the vibration mode of the dimer as a whole unit can be treated as a localized phonon. Later we will show that these localized phonons play a crucial role in suppressing the thermal conduction of CuP2.Our neutron powder diffraction results confirm the crystal structure as previously reported25. Shown in Fig. 2a is the neutron powder diffraction pattern at room temperature (for data at lower temperatures, refer to Supplementary Fig. 3). The Rietveld refinement analysis suggests that the majority phase is monoclinic CuP2 while there is a minority phase of Cu3P, whose volumetric fraction is only 2.79%. This powder sample, crushed from single crystals, is strongly textured along (100) because of the excellent ductility associated with the layered lattice structure. It is important to notice that the background of the diffraction data is quite flat with no obvious diffuse scattering observed, which indicates no discernable atomic disorder in the sample. The detailed structural parameters determined in the refinements are listed in Supplementary Table 3. The Debye–Waller factors, in particular, U22 and U33 of Cu atoms are much larger than those of P atoms. This indicates that there is a significant thermal motion of Cu atoms confined within the layers. The full temperature dependencies of the lattice parameters and Debye–Waller factors are summarized in Supplementary Fig. 4. It should be noted that the fitting of temperature dependencies of Debye–Waller factors to the Einstein model suggests there is negligible atomic disorder. In addition to the powder diffraction, our single-crystal X-ray diffraction also confirms the structure. While the average structural study suggests the right phase and negligible disorder, we further employ the pair distribution function (PDF) analysis to directly confirm the disorder-free nature of the system. As shown in Fig. 2b, the X-ray structure factor SX(Q) is well normalized to 1 at high Q at 300 K. Reduced PDF, GX(r), is subsequently obtained via Fourier transform of SX(Q). The inset of Fig. 2b presents the experimental GX(r) that is well reproduced by the P21/c crystal model, supporting the disorder-free nature.Fig. 2Average and local structures of CuP2.a Neutron powder diffraction pattern and Rietveld refinement analysis at 300 K. b The structure factor, SX(Q), obtained in X-ray total scattering at 300 K. The inset is the reduced PDF, GX(r), as well as the real-space refinement based on the P21/c crystal model.Lattice dynamicsThe above structural study excludes disorder scattering as an origin of the low thermal conductivity. Now, we move to the lattice dynamics study for lattice anharmonicity. We begin with a survey over a large reciprocal space of the dynamic structure function S(Q, E) vs. momentum transfer (Q) and energy transfer (E) using the time-of-flight neutron spectrometer-Pelican. The measurements are carried out in the scattering plane defined by [H, 0, 0] and [0, L, L]. Shown in Fig. 3a is the phonon dispersion along [H, H, H] direction of the Brillouin zone G = (111). The branches of acoustic phonons are clearly observed, in agreement with the results from the triple-axis spectrometer-Taipan and the density functional theory (DFT) calculations. Unfortunately, the optical phonon modes above 10 meV cannot be observed due to limited accessible energy range and energy resolution. For more details of dispersions of different Brillouin zones, please refer to Supplementary Fig. 5. Before we rationally perform detailed INS measurements at Taipan, the lattice dynamics are fully exploited using DFT calculations. The DFT calculated phonon dispersions of Brillouin zones G = (200), G = (022), and G = (111) are plotted in Fig. 3b up to 16 meV while the complete dispersions are plotted in Supplementary Fig. 6. We can see the steep acoustic phonon branches that originate from the individual Γ point and reach the zone boundaries at about 10 meV. Above the acoustic phonon branches, there are a few flat optical phonon bands up to 16 meV.Fig. 3Phonon dispersions of CuP2.a The phonon dispersion obtained at the time-of-flight neutron spectrometer-Pelican along [H, H, H] direction of Brillouin zone G = (111). The circles are the corresponding results obtained from the triple-axis spectrometer—Taipan. The orange lines are the calculated dispersions. b Calculated phonon dispersions of Brillouin zones G = (200), G = (022), and G = (111). The size of the shadow bubbles represents the magnitude of Grüneisen parameter (γ) of the related branches. c–h Phonon dispersions of G = (200), G = (022), and G = (111) Brillouin zones collected at Taipan. The circles are the phonon energies determined by the spectral fitting to a Lorentzian function. The red dashed lines represent the linear fitting of phonon branches approaching the Brillouin zone centers to determine the sound speeds.Then, we focus on detailed investigations by conducting constant-E as well as constant-Q scans for Brillouin zones of G = (200), G = (022), and G = (111) using the thermal neutron triple-axis spectrometer-Taipan in the same scattering plane as on Pelican. The scan directions are summarized in Supplementary Fig. 7. Figure 3c, d shows the dispersions along [H, 0, 0] and [0, L, L] directions of Brillouin zone G = (200) at 300 K obtained by constant-E scans, respectively. Note that while Fig. 3c represents the longitudinal component of G = (200) Brillouin zone, some contributions of the longitudinal phonons are also included in Fig. 3d in addition to the major transverse components (see Supplementary Fig. 7). The phonon intensity decays quickly with departure from the Brillouin zone center. To accurately determine the dispersion relationships, the phonon energies are determined by a spectral fitting to a Lorentzian function. The obtained peak positions are plotted on the contour plots as circles with error bars. Then, we fit the phonon energies to a linear function approaching the zone centers to obtain the experimental sound speeds (see the dash lines). The derived values are listed in Supplementary Table 4. It can be seen that the dispersion along [H, 0, 0] direction is much steeper than that along [2, L, L] direction, indicating larger sound speeds. A similar procedure is applied to the cases of G = (022) and G = (111), as shown in Fig. 3e–h, respectively.Lattice anharmonicityThe temperature-dependent lattice dynamics are considered first on the phonon density of state (PDOS) measurements on a powder sample in a wide temperature region. Shown in Fig. 4a are PDOS at 200, 300, 400, 500, and 600 K, respectively. As temperature rises, it is clear that the whole profile is significantly broadened and several peaks are remarkably softened as an indication of strong anharmonicity. Compared with the DFT calculated PDOS, as plotted in Fig. 4b, it is identified that the Cu atoms mainly participate in the low-energy modes below 15 meV, while the high-energy optical modes are dominated by P atoms. This might be related to the fact that the covalent interactions of P networks are much stronger, in addition to the lighter atomic mass of P than that of the Cu dimer. Of particular interest is the experimentally observed mode located at about 11 meV. Compared with the DFT calculations of the phonon dispersion (Fig. 3b) and PDOS (Fig. 4b), this mode is associated with Cu and has a flat dispersion, thus it must be a rattling mode contributed by the Cu dimer (justified later). The strong anharmonicity of this rattling mode is demonstrated by the large softening from 11.40 meV at 200 K to 10.67 meV at 600 K. Since the acoustic phonons and the nearby low-frequency optical phonons are the main contributors to the thermal conductivity, we further investigate the anharmonic property of this Cu rattling mode in the following sections.Fig. 4Highly anharmonic optical mode at 11 meV.a The experimental PDOS, vertically shifted for clarity. b. Calculated total and partial PDOS. c. Phonon dispersion of G = (022) Brillouin zone along [H,0,0] at 100 K. d–f Temperature dependence of phonon spectra for constant-Q scans of G = (022) Brillouin zone along [H,0,0] with q = 0, 0.2, 0.3 at 100, 300 and 450 K, respectively. The green dash lines highlight the energy shift of the peaks at heating.Focusing on the significantly softened mode at about 11 meV, detailed constant-Q scans were performed at 100, 300, and 450 K, respectively. Figure 4c shows the dispersion along [H, 0, 0] direction of G = (022) at 100 K (the results at 300 and 450 K are shown in Supplementary Fig. 8). Just above the intense acoustic phonon branch, a very flat optical phonon branch originates from the Brillouin zone center at about 12 meV to the Brillouin zone boundary at about 10 meV. This is the rattling mode of the Cu dimer. At higher energies, three more flat branches are further observed. The temperature dependencies of these modes are well tracked at the constant-Q scans at q = 0, 0.2, 0.3, as shown in Fig. 4d–f, respectively. Data at q = 0 clearly indicate a distinct behavior for the three optical modes identified. The energy of the first mode (rattling mode) is reduced from 11.16 meV at 100 K to 10.24 meV at 450 K, at a similar rate of −0.002 meV K−1 as observed with the PDOS results. In contrast, the other two modes at 14.53 and 15.53 meV at 100 K hardly shift with increasing temperature. At q = 0.2 and 0.3 shown in Fig. 4e, f, it is further noticed that the acoustic mode around 6 meV shows no noticeable shift in energy upon temperature increase in contrast to the rattling mode. This behavior is also shown on more data as presented in Supplementary Figs. 9 and 10 for the Brillouin zone G = (022). This feature is in sharp contrast to most materials, such as FeSi26, where phonon modes are not selectively softened with increasing temperature.Apart from the remarkable temperature-induced softening of the rattling mode as discussed above, the strong anharmonicity is also manifested by the huge Grüneisen parameter (γ)27,28 calculated by DFT. In Fig. 3b, the size of the shadow bubbles represents the magnitude of γ of a specific mode. Both the rattling mode and the acoustic modes have very large γ. The largest γ for the rattling mode is about 3.3 at q = 0.5, corresponding to the G = (022) Brillouin zone boundary, while γ is almost larger than 2 in all reciprocal space covered. In particular, γ is strikingly large for the longitudinal acoustic (LA) branches (blue color) near the zone boundaries, where they tend to overlap with the rattling mode. In contrast, the Grüneisen parameter of GaAs is almost less than 1 in the whole temperature region below 300 K (ref. 29). As a result, the anharmonicity of CuP2 is much stronger than that of GaAs with the same magnitude of sound speed, resulting in the huge difference in the lattice thermal conductivity between the two materials. Moreover, our single-crystal X-ray diffraction measurement also suggests that Cu atoms have substantial third-order Debye–Waller factors, which provides further evidence of the strong anharmonicity of Cu-involved vibrations, as listed in Table 1. The values of U222 and U333 are 0.0025(6) and 0.00090(13) Å3 while U111 is as small as 0.0006(4) Å3. Such a difference is similar to the second-order Debye–Waller factors plotted in Supplementary Fig. 4d. As compared to Cu, the third-order Debye–Waller factors of P atoms are almost negligible within the error bars.Table 1Third-order Debye–Waller factors of Cu and P atoms determined in refinements of single-crystal X-ray diffraction data.U (Å3)CuPCu1P1P2U1110.0006(4)0.0001(7)0.0009(13)U2220.0025(6)0.0011(9)0.00021(20)U3330.00090(13)0.00021(20)−0.0001(2)Rattling mode and thermal conductivityHenceforward, we discuss the nature of the highly anharmonic rattling mode at 11 meV and its impact on the system. Atomic motions of this mode are plotted at the Γ point, as shown in Fig. 1c, d, which involve the vibrations of Cu dimers in between the P layers. The atomic motions are recorded in Supplementary Movie 1 and those of the nearby two modes are also present in Supplementary Movie 2 and Supplementary Movie 3, respectively. The justification of calling the vibration of the Cu dimers a rattling mode is further articulated here. Firstly, the Cu dimers behaving as rattlers have very weak bonding to the others as crystallographically and electronically confirmed above. This is also evidenced by the significant Debye–Waller factors of Cu atoms that represent large-amplitude anharmonic vibrations in bc plane, in agreement with the well-known skutterudites and clathrates compounds10,30,31. Secondly, the dispersion of this mode is quite flat as the group velocity is smaller than 10 m s−1 showing an uncorrelated and localized feature. This is in close analogy to the ideal rattling scenario, appearing as a dispersionless Einstein mode with a constant vibration frequency. Thirdly, the anti-crossing or avoided-crossing of the acoustic mode with the rattling mode is also theoretically predicted in the compound, though not experimentally observed due to experiment limitations, as highlighted in Supplementary Fig. 6. The theory predicted five anti-crossing points located at q = 0.27 of G = (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\raise0.5ex\hbox{$\scriptstyle 1$}\kern-0.1em/\kern-0.15em \lower0.25ex\hbox{$\scriptstyle 2$}}{\raise0.5ex\hbox{$\scriptstyle 1$}\kern-0.1em/\kern-0.15em \lower0.25ex\hbox{$\scriptstyle 2$}}0$$\end{document}1/21/20), q = 0.32 of G = (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\raise0.5ex\hbox{$\scriptstyle 1$}\kern-0.1em/\kern-0.15em \lower0.25ex\hbox{$\scriptstyle 2$}}00$$\end{document}1/200), q = 0.37 of G = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(00{\raise0.5ex\hbox{$\scriptstyle 1$}\kern-0.1em/\kern-0.15em \lower0.25ex\hbox{$\scriptstyle 2$}})$$\end{document}(001/2), q = 0.13 of G = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\overline {{\raise0.5ex\hbox{$\scriptstyle 1$}\kern-0.1em/\kern-0.15em \lower0.25ex\hbox{$\scriptstyle 2$}}} {\raise0.5ex\hbox{$\scriptstyle 1$}\kern-0.1em/\kern-0.15em \lower0.25ex\hbox{$\scriptstyle 2$}}0)$$\end{document}(1/2¯1/20), and q = 0.32 of G = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\overline {{\raise0.5ex\hbox{$\scriptstyle 1$}\kern-0.1em/\kern-0.15em \lower0.25ex\hbox{$\scriptstyle 2$}}} 00)$$\end{document}(1/2¯00). This behavior also gives rise to the decrease of the group velocity of the acoustic mode around the zone boundaries. This rattling mode differs from the previous ones reported in skutterudites and clathrates systems. These compounds have typical cage-like frames with the heavy atoms filled into the lattice void, whereas CuP2 crystallizes in a layered structure with Cu dimer as the rattler10–13,32. To distinguish with the conventional rattling mode, we term the behavior of CuP2 as dimer rattling.This rattling mode is expected to play a dominating role in the thermal transport. We have systematically determined the sound speeds both experimentally and theoretically, as summarized in Supplementary Table 4. For example, vLA = 6243 m s−1 and vTA = 3192 m s−1 are obtained from DFT calculations for the LA and transverse acoustic (TA) branches of G = (200), respectively. This vLA value is in excellent agreement with that determined by both fitting the INS experimental phonon dispersions and the Brillouin light scattering spectrum (Supplementary Fig. 11), which are 6364 m s−1 and 6275 m s−1, respectively. The mean sound speed vm = 4155 m s−1 is estimated using the equation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$3v_{\mathrm{m}}^{ - 3} = v_{{\mathrm{LA}}}^{ - 3} + 2v_{{\mathrm{TA}}}^{ - 3}$$\end{document}3vm−3=vLA−3+2vTA−3 (ref. 13) based on the theoretical results. As shown in Fig. 1d, unlike the ordinary materials that are distributed around the empirical line, CuP2 stands quite exceptional with the thermal conductivity one order of magnitude lower than GaAs that has a similar mean sound speed as CuP222,23.DiscussionWe have discovered the suppressed thermal transport property of CuP2 and established a profound understanding on the fundamental physical mechanism by a comprehensive study on the atomic structures and lattice dynamics through neutron, X-ray scattering techniques, and complementary DFT calculations. This system is manifested to be very anharmonic. The Cu atoms participate in a dimer rattling mode, which strongly scatters the LA phonons and leads to anti-crossing phenomena in the dispersion relationships. It is this mode that dominates the reduced thermal conductivity, counteracting the contribution of large mean sound speeds. The observed dimer rattling behavior in the open layered structures might represent an emerging opportunity to rationally tailor thermal transport properties of solids. The combined excellent acoustic conduction and thermal insulation properties may find CuP2 a promising material in some nontrivial applications requiring both excellent mechanical sound transmission (or mechanical rigidity) and heat insulation.MethodsSample preparationCuP2 single crystals were grown through a flux method33. Starting materials of Cu (purity: 99.999%), P (purity: 99.999%), and Sn (purity: 99.999%) in a molar ratio of 1:1:3 were placed in an alumina crucible and then sealed into an evacuated quartz tube. The mixture was placed into a box furnace and heated at 1233 K for 6 h. Then, it was cooled down to 873 K at a rate of 3 K/h. The excessive Cu and Sn eutectic flux were decanted in a centrifuge at 873 K. Single crystals were mechanically cleaved from the ingots. Single crystals were crushed for neutron and X-ray powder diffraction as well as for thermal conductivity measurement where a carbon-coated pellet (ϕ = 10 mm) was used.Thermal conductivity measurementThe total thermal conductivity (κtot) of the polycrystalline sample was obtained based on the formula κtot = DCpρ, where D is the thermal diffusion coefficient measured using the laser flash method (LFA457, NETZSCH, Germany), Cp is the Dulong–Petit specific heat capacity, and ρ is the density calculated from the geometrical dimensions and mass. Thermal conductivity of single crystals was measured in the temperature range from about 50 to 300 K using a steady-state comparative method34. The bar-shaped CuP2 samples with a dimension of about 0.5 × 0.5 × 3 mm3 were cut from as-grown crystals. The reference was a rod of constantan alloy with a diameter of 0.5 mm. The differential thermocouple was made of copper and constantan wires. In this method, the thermal conductivity of the sample is obtained by measuring the temperature difference between the heat source and heat sink under a steady heat flow through the bar-shaped sample (consist of CuP2 single crystal and the reference). The uncertainty of the steady state comparative method is related to the error in measuring sample dimensions, contact thermal resistance, and heat loss. The uncertainty is usually about 10–20% (ref. 35). The total thermal conductivity is expressed as a sum of lattice contribution (κlat) and electronic contribution (κel) for this compound1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\kappa _{{\mathrm{tot}}}=\kappa _{{\mathrm{lat}}}+\kappa _{{\mathrm{el}}}.$$\end{document}κtot=κlat+κel.The electronic part κel is proportional to the electrical conductivity σ through the Wiedemann–Franz relation2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\kappa _{{\mathrm{el}}} = L\sigma T,$$\end{document}κel=LσT,where L is Lorenz number36. The lattice thermal conductivity κlat can be estimated by subtracting the electrical contribution κel. The electrical conductivity of CuP2 is only about 8.09 Ω−1 cm−1 at 300 K (ref. 37). κel of CuP2 is about 0.0059 W m−1 K−1 by assuming L = 2.45 W Ω K−2 (ref. 38). In general, the true Lorenz number L for most thermoelectric materials is lower than this value so that the electronic contribution to the thermal conductivity is negligible in this system39.Neutron powder diffractionThe neutron powder diffraction was performed at the time-of-flight powder diffractometer at the Spallation Neutron Source of Oak Ridge National Laboratory, USA40. The powder sample with the mass around 3 g was loaded into a vanadium container of 8 mm diameter and measured in a Powgen Automatic Changer covering the temperature region of 10–300 K. The data were collected with neutrons of central wavelength 0.8 Å. Constant temperature scans were conducted at 10, 50, 100, 200, and 300 K, respectively. GSAS41 was used to refine the neutron powder diffraction patterns and the results are summarized in Fig. 2a, Supplementary Figs. 3 and 4 as well as Table 3. The Rietveld method employs a nonlinear least-square method to fit the profile that includes all structural and instrumental parameters42. Several figure-of-merits including Rp, Rwp, and χ2 that mean profile residual (reliability factor), weighted profile residual, and goodness of fit are used to quantify the quality of the fit.Synchrotron X-ray total scatteringThe high-energy X-ray powder diffraction experiment was carried out at the Beamline BL04B2 of SPring-8, Japan43. The CuP2 powder was put into a quartz capillary (ϕ = 1 mm) and then fixed into the sample holder. An empty capillary was measured as a background. X-ray energy was fixed at 61.4 keV with a Si (220) monochromator. The energy resolutions ΔE/E was approximately 5 × 10−3. A vacuum chamber with a Kapton window was used for minimizing the scattering background. Six point-detectors were arranged horizontally to obtain 2θ value up to 49o (Q range up to 25 Å−1). The real-space refinement was performed using PDFgui44.Single-crystal X-ray diffraction analysisThe single-crystal X-ray diffraction data were collected at a Pilatus CCD diffractometer equipped with graphite-monochromated Mo–Kα radiation (λ = 0.71073 Å) at 293 K. The crystal structure of CuP2 was solved, and three-order atomic thermal displacement factors of all atoms were refined through full-matrix least-square technique on F2. All of the calculations were performed using Jana2006 (ref. 45).INS measurementsINS experiments were first performed on the time-of-flight cold-neutron spectrometer-Pelican46,47, at the Australian Nuclear Science and Technology Organization (ANSTO), Australia. For the PDOS measurements, a powder sample was mounted in an annular aluminum sample can with 1 mm gap. The sample can was attached to the cold head of a closed-cycle refrigerator which is capable of achieving sample temperature from 1.5 to 800 K. The instrument was aligned for 4.69 Å (3.7 meV) incident neutrons. The resolution at the elastic line was 135 µeV. The intensity of an empty can was subtracted as the background contribution and the data were normalized to a vanadium standard that had the same geometry as the sample can. All data manipulations were performed using the Large Array Manipulation Program (LAMP)48. The scattering function S(Q,E), as a function of scattering wave vectors (Q) and phonon energy (E), were measured on energy gain mode over a wide temperature range and then transformed to a generalized PDOS using formula (3), where kB is Boltzmann’s constant and T is temperature3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g\left( E \right) = {\int} {\frac{E}{{Q^2}}S\left( {Q,E} \right)\left( {1 - e^{ - \frac{E}{{k_BT}}}} \right)} dQ.$$\end{document}gE=∫EQ2SQ,E1−e−EkBTdQ.For phonon dispersion measurements, a single crystal weighted 0.7 g was used and pre-orientated such that the scattering plane was defined by [H, 0, 0] and [0, L, L]. A large reciprocal space was covered by rotating the sample over 100° in a step of 1° around the axis perpendicular to the scattering plane. The incident neutron wavelength is 2.345 Å (14.8 meV), corresponding to the second-order reflection of the Pelican highly oriented pyrolytic graphite monochromators configured at 4.69 Å. The sample temperature was set at 300 K. The data were reduced by LAMP and the whole S(Q,E) was generated using HORACE49, which was also used to visualize the phonon dispersions. After the general survey over many Brillouin zones using the Pelican instrument, we focused on several selected Brillouin zones G = (200), G = (022), and G = (111) on the thermal neutron triple-axis spectrometer-Taipan at ANSTO50,51. The energies of the incident and scattered neutrons were defined by a double-focused PG (002) monochromator and analyzer. We used the open geometry of the neutron beam with a virtual source width of 10 mm at the beam exit from the reactor and a slit of 20 mm at the detector. The measurements were performed with fixed final neutron energy of 14.87 meV. After the sample, a HOPG filter was placed to remove higher-order reflections from the scattered beam. The sample was aligned in the same scattering plane as in the measurements at Pelican. The phonons were measured along [H, 0, 0], [H, H, H], and [0, L, L] directions. A standard cryofurnace was used to access the temperatures of 100, 300, and 450 K. The experimental data were fitted to a Lorentzian function using PAN of DAVE52.Brillouin light scatteringBrillouin light scattering measurements were performed at room temperature using a Sandercock-type six-pass tandem Fabre–Perot (TFP-2) as a spectrometer and a 532 nm laser as a light source. It is a well-established technique for measuring sound velocities53. In backscattering geometry, LA mode sound velocity (vLA) can be calculated as4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{\mathrm{LA}} = \frac{{f_{\mathrm{B}}\lambda _0}}{{2n}}.$$\end{document}vLA=fBλ02n.From the measured Brillouin frequency shift fB, using this equation, the refractive index n = 4.04 and the laser wavelength λ0 = 532 nm, the LA-wave velocity was determined to be vLA = 6275 m s−1.DFT calculationsAll the calculations in this work were performed using the ab initio DFT as implemented in the VASP code54. The projector augmented wave pseudopotentials55 and the Perdew–Burke–Ernzerhof56 functional within the general gradient approximation were used to take care of electron-ion and inter-electron exchange-correlation interactions, respectively. The wave functions were expanded using plane waves with an energy cutoff of 500 eV, and the electronic energy convergence was set to be 10−8 eV. The Brillouin zone of the primitive unit cell was sampled in the Γ-centered 8 × 10 × 15 k-point mesh for structural optimization until all the atomic force is less than 0.001 eV Å−1. The phonon dispersion relation was calculated using Phonopy package57 combining with VASP using 2 × 2 × 2 supercell with 96 atoms. The Γ-centered 2 × 2 × 2 q-point mesh was used. To unravel the chemical bonding nature of the Cu–Cu atomic pair, we also calculated the ELF58 and electronic charge transfer Δρ = ρ(CuP2) − ρ(atom), which is defined as the electron charge density difference between CuP2 and the constituent individual atoms.Supplementary informationSupplementary InformationDescription of Additional Supplementary FilesSupplementary Movie 1Supplementary Movie 2Supplementary Movie 3 | nature communications | [
"Article"
] | [
"Thermoelectric devices and materials",
"Two-dimensional materials"
] | IntroductionThermal conduction fundamental low thermal conductivity desirable for applications thermal insulation2 phase transition memory thermoelectric energy insulating nonmagnetic systems phonons major heat carrier lattice thermal conductivity proportional to sound speeds phonon decrease thermal conductivity reduce phonon lifetimes enhancing phonon–phonon phonon-disorder scattering Phonon–phonon scattering anharmonic interactions dominates thermal conductivity disorder-free systems like PbTe8 SnSe9 special case rattling localized vibrational mode scattering acoustic phonons anti-crossing points in phonon applied low thermal conductivity in phonon-glasses thermoelectric materials like skutterudites spatially hierarchical atomic disorder used lower thermal conductivity phonon-disorder scattering disorder-free compounds relationship thermal conductivity sound speeds in Fig. 1a compounds distributed around straight line sound speeds dominating for materials above line lattice anharmonicity governs thermal transport materials below lineline single-crystalline highest lattice thermal conductivity 2400 W m−1 K−1 high sound speed 14,400 m s−1 GaAs near line sound speed 3627 m s−1 CuP2 similar speed thermal conductivity ten times. 1Lattice thermal conductivity crystal structure survey-free materials sound speed detailed data Supplementary Table 1 polycrystalline samples Fig. 1 Table 2) Temperature dependence conductivity CuP2 crystal 50 to 300 K steady-state comparative method room temperature error bars uncertainty layered structure Cu dimer layers P network layers highlighted orange atoms vibrational directions Γ point optical phonon mode 11 meV atomic motions Supplementary Movie 1. isolated Cu dimer layers intra- inter-dimer distances temperature dependencies lattice thermal conductivity CuP2 crystals Fig. 1b 50 to 300 K values increase temperature decreases 300 K κlat bc plane 4.66 W m−1 K−1 3.57 W m−1 out polycrystalline sample Supplementary Fig. 1. room temperature 0.62 W m−1 K−1value reduced compared single crystals attributed to grain boundaries defects scattering24. understand low thermal conductivity in CuP2 conducted study atomic structure lattice dynamics compound crystallizes simple monoclinic structure space group P21/c (ref. 25). Fig. 1c structure layered configuration Cu layers 1.59 Å P network 3.85 Å repeat along a-axis Cu atoms forming dimer equivalent position orange represent vibration directions Cu atoms rattling mode Fig. 1c intra-dimer distance 2.48 Å inter-dimer distances 3.85 4.81 7.53 Å a b c directions electron localization function (ELF) calculation weak bonding nature between Cu dimers Fig. 2) ELF around Cu atoms 0.2 weak metallic bond ELF almost zero between dimers calculated charge density supports dimer nature dimer vibration mode localized phonon phonons thermal conduction neutron powder diffraction results confirm crystal structure Fig. 2a neutron powder diffraction pattern at room temperature lower temperatures Supplementary Fig. 3)Rietveld refinement analysis majority phase monoclinic CuP2 minority Cu3P volumetric fraction 2.79% powder sample single crystals textured along (100) layered lattice structure background diffraction data flat no diffuse scattering no atomic disorder detailed structural parameters Supplementary Table 3. Debye–Waller factors U22 U33 Cu atoms larger than P atoms significant thermal motion Cu atoms layers temperature dependencies lattice parameters Debye–Waller factors summarized Supplementary Fig. 4. temperature Einstein model negligible atomic disorder single-crystal X-ray diffraction confirms structure average study suggests right phase negligible disorder pair distribution function (PDF) analysis disorder-free Fig. 2b X-ray structure factor SX(Q) normalized to 1 at 300 K Reduced PDF GX obtained via Fourier transform Fig. 2b experimental GX(r) reproduced P21/c crystal model disorder-free nature. 2Average local structures CuP2.a Neutron powder diffraction pattern Rietveld refinement analysis at 300 K structure factor SX X-ray scattering at 300 K reduced PDF GX real-space refinement P21/c crystal modelLattice study excludes scattering low thermal conductivity lattice dynamics study anharmonicity survey reciprocal space dynamic structure function S(Q E) momentum transfer energy transfer time-of-flight neutron spectrometer-Pelican measurements scattering plane [H, L Fig. 3a phonon dispersion direction Brillouin zone G = (111) branches acoustic observed triple-axis spectrometer-Taipan density functional theory (DFT) calculations optical phonon modes above 10 meV limited energy range resolution Supplementary Fig. 5. INS measurements Taipan lattice dynamics exploited DFT calculations calculated phonon dispersions Brillouin zones G = (200), = (022) (111) plotted Fig. 3b up to 16 meV complete dispersions Supplementary Fig. 6. steep acoustic phonon branches originate from Γ point reach zone boundaries 10 meV flat optical phonon bands up to 16 meV. 3Phonon dispersions time-of-flight neutron spectrometer-Pelican Brillouin zone G = (111) circles results triple-axis spectrometer—Taipan orange lines calculated dispersionsCalculated phonon dispersions Brillouin zones G (200), (022) (111) size shadow bubbles represents Grüneisen parameter (γ) branches Phonon dispersions G (022) (111) collected Taipan circles phonon energies determined spectral fitting Lorentzian function red dashed lines linear fitting phonon branches zone centers sound speeds investigations constant-E constant-Q scans G (022) (111) thermal neutron triple-axis spectrometer-Taipan scattering plane Pelican scan directions summarized Supplementary Fig. 7. Figure 3c dispersions [H, 0, 0] L, L] directions Brillouin zone G (200) 300 K constant-E scans Fig. 3c longitudinal G (200) Brillouin zone longitudinal Fig. 3d phonon intensity decays departure Brillouin zone center dispersion phonon energies determined spectral fitting Lorentzian function peak positions plotted contour plots circles error bars fit phonon energies linear function zone centers experimental sound speeds derived values Supplementary Table 4. dispersion [H, 0, 0] direction steeper [2, L, L] direction larger sound speedssimilar procedure G = (022) G = (111) Fig. 3e–h.Lattice temperature-dependent lattice dynamics considered phonon density state (PDOS) measurements powder sample wide temperature region Fig. 4a PDOS at 200 300 400 500 600 K temperature rises profile broadened peaks softened strong anharmonicity DFT calculated PDOS Fig. 4b Cu atoms participate low-energy modes below 15 meV high-energy modes dominated P atoms covalent interactions P networks stronger lighter atomic mass P than Cu dimer experimentally observed mode at 11 meV calculations mode associated Cu flat dispersion rattling mode contributed Cu dimer strong anharmonicity softening from 11.40 meV at 200 K to 10.67 meV at 600 K acoustic low-frequency optical main contributors thermal conductivity investigate anharmonic property Cu rattling mode.Fig. 4Highly anharmonic optical mode at 11 meV experimental PDOS Calculated total partial PDOS Phonon dispersion of G = (022) Brillouin zone [H] at 100 K.Temperature phonon spectra constant-Q scans G = (022) Brillouin zone q = 0.2 0.3 100 300 450 K green lines highlight energy shift heating softened mode 11 meV constant-Q scans 100 300 450 K Figure 4c dispersion G 100 K results 300 450 K Supplementary Fig. acoustic phonon flat optical phonon branch Brillouin zone center 12 meV Brillouin zone 10 meV rattling mode Cu dimer higher energies three flat branches observed temperature dependencies tracked constant-Q scans q = 0 0.2 0.3 Fig. 4d–f Data q = 0 distinct behavior three optical modes energy first mode (rattling mode reduced 11.16 meV 100 K to 10.24 meV 450 K −0.002 meV K−1 other two modes 14.53 15.53 meV 100 K shift increasing temperature q = 0.2 0.3. acoustic mode 6 meV no shift energy temperature increase rattling mode data Supplementary Figs. 9 10 Brillouin zone G = (022) materials FeSi26 phonon modes not softened increasing temperaturetemperature-induced softening rattling mode strong anharmonicity Grüneisen parameter (γ)27,28 calculated DFT Fig. 3b size shadow bubbles γ mode rattling acoustic modes large γ largest γ rattling mode 3.3 at q = 0.5 G = Brillouin zone γ larger than 2 in reciprocal space γ large longitudinal acoustic branches zone boundaries overlap with rattling mode Grüneisen parameter GaAs less than 1 temperature below 300 K anharmonicity CuP2 stronger than GaAs difference lattice thermal conductivity single-crystal X-ray diffraction Cu atoms have third-order Debye–Waller factors strong anharmonicity Cu-involved vibrations Table 1. values U222 U333 0.0025(6) 0.00090(13) Å3 U111 small 0.0006(4) Å3 difference similar second-order Debye–Waller factors Fig. 4d third-order Debye–Waller factors P atoms negligible error bars 1Third-order Debye–Waller factors Cu P atoms single-crystal X-ray diffraction data(Å3)CuPCu1P1P2U1110.0006(4)0.0001(7)0.0009(13)U2220.0025(6)0.0011(9)0.00021(20)U3330.00090(13)0.00021(20)−0.0001(2)Rattling mode thermal discuss anharmonic rattling mode at 11 meV impact system Atomic motions plotted at Γ point Fig. 1c d vibrations Cu dimers between P layers motions recorded in Supplementary Movie 1 Movie 2 3 justification vibration Cu dimers rattling mode Cu dimers weak bonding crystallographically Debye–Waller factors Cu atoms large-amplitude anharmonic vibrations plane skutterudites clathrates compounds10 dispersion flat group velocity smaller than 10 m s−1 uncorrelated localized feature ideal rattling scenario dispersionless Einstein mode constant vibration frequency anti-crossing acoustic mode with rattling mode theoretically predicted compound not experimentally observed limitations Supplementary Fig.theory predicted five anti-crossing points at q = 0.27 of G =[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{upgreek\oddsidemargin-69pt}{document}\raise0.5ex 1\kern-0.1em/\kern-0.15em\lower0.25ex\scriptstyle 2-0.1em.15em 2{document}1/21/20) q = 0.32 of G =[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}\oddsidemargin-69pt}{document}\raise0.5ex\scriptstyle 1\kern-0.1em/\kern-0.15em\lower0.25ex\hbox\scriptstyle 2\end{document}1/200), q = 0.37 G[12pt]{minimal{amsmath\oddsidemargin-69pt}{document}\raise0.5ex 1\kern-0.1em/\kern-0.15em\lower0.25ex 2$}}\end{document}(001/2), q = 0.13 G[12pt]{minimal}{amsmath{wasysym\oddsidemargin-69pt}{document}\raise0.5ex\hbox 1\kern-0.1em/\kern-0.15em \lower0.25ex\hbox\scriptstyle 2$\raise0.5ex\hbox 1$\kern-0.1em/\kern-0.15em\lower0.25ex\hbox\scriptstyle 2{document}(1/2 ̄1/20) q = 0.32 G\documentclass[12pt]{minimal}\usepackage{amsmath{upgreek\oddsidemargin-69pt}\begin{document}\raise0.5ex\scriptstyle 1$\kern-0.1em.15em\lower0.25ex 2${document}(1/2 ̄00). behavior decrease group velocity acoustic mode zone boundaries rattling mode differs skutterudites clathrates systems compounds cage-like frames heavy atoms lattice void CuP2 crystallizes layered structure Cu dimer rattler10–13 CuP2 dimer rattling mode dominating thermal transport determined sound speeds experimentally theoretically Supplementary Table 4. vLA = 6243 m s−1 vTA = 3192 m s−1 DFT calculations LA transverse acoustic (TA branches G = (200), value INS experimental phonon dispersions Brillouin light scattering spectrum Fig. 6364 m s−1 and 6275 m s−1mean sound speed vm = 4155 m s−1 estimated equation\documentclass[12pt{minimal\usepackage{amsmath\oddsidemargin-69pt}{document}\mathrm{m}} - 3} ={LA - 3} + 2v{TA - 3}\end{document}3vm−3=vLA−3+2vTA−3 (ref. 13) theoretical results Fig. 1d CuP2 exceptional thermal conductivity lower than GaAs similar mean sound speed as CuP222,23 discovered suppressed thermal transport property of CuP2 understanding physical mechanism atomic structures lattice dynamics neutron X-ray scattering DFT calculations system anharmonic Cu atoms dimer rattling mode scatters LA phonons anti-crossing phenomena dispersion relationships dominates reduced thermal conductivity large mean sound speeds dimer rattling behavior open layered structures opportunity tailor thermal transport properties solids excellent acoustic conduction thermal insulation properties CuP2 promising material applications mechanical sound transmission heat insulation.preparationCuP2 crystals grown flux Starting materials Cu P Sn 99.999%) 1:1:3 in alumina crucible sealed into evacuated quartz tube mixture box furnace heated at 1233 K 6 h cooled to 873 K 3 K/h excessive Cu Sn decanted in centrifuge at 873 K crystals cleaved from ingots crushed for neutron X-ray powder diffraction thermal conductivity measurement carbon-coated pellet (φ = 10 mm) used.Thermal conductivity total thermal conductivity (κtot) sample obtained formula κtot = DCpρ D thermal diffusion coefficient Cp Dulong–Petit heat capacity ρ density Thermal conductivity measured 50 to 300 K steady-state comparative method34 bar-shaped CuP2 samples 0.5 × 0.5 × 3 mm3 cut from as-grown crystals reference rod of constantan alloy diameter 0.5 mm differential thermocouple copper constantan wires thermal conductivity temperature difference between heat source heat sink steady heat flow uncertainty related to error in measuring sample dimensions contact thermal resistance heat lossuncertainty 10–20% (ref. 35). total thermal conductivity sum lattice electronic contribution (κel\documentclass[12pt{minimal\usepackage{amsmath{wasysym-69pt{tot{lat=κlat+κel electronic part κel proportional to electrical conductivity σ Wiedemann–Franz relation2[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin-69pt}\kappa\mathrm{el = L\sigma T}κel=LσT L Lorenz number36 lattice thermal conductivity κlat estimated subtracting electrical contribution κel electrical conductivity CuP2 8.09 Ω−1 cm−1 at 300 K (ref. 37). κel of CuP2 0.0059 W m−1 K−1 L = 2.45 W Ω K−2Lorenz number L thermoelectric materials lower electronic contribution thermal conductivity negligible powder time-of-flight powder diffractometer Spallation Neutron Source Oak Ridge National Laboratory powder sample 3 g vanadium container 8 mm diameter measured Powgen Automatic Changer 10–300 K data collected neutrons wavelength Å temperature scans at 10 50 100 200 300 K GSAS41 diffraction patterns results summarized in Fig. 2a Supplementary Figs. 3 4 Table 3. Rietveld method nonlinear least-square method profile structural instrumental figure-of-merits Rp Rwp χ2 goodness fit quantify quality.Synchrotron X-ray high X-ray powder diffraction Beamline BL04B2 SPring-8 CuP2 powder quartz capillary 1 fixed sample holder empty capillary measured background X-ray energy fixed 61.4 keV Si (220) monochromator energy resolutions ΔE/E 5 × 10−3 vacuum chamber Kapton window scattering background Six point-detectors 2θ value up to 49o range 25 Å−1) real-space refinement using PDFgui44Single-crystal X-ray diffraction data Pilatus CCD diffractometer graphite-monochromated Mo–Kα radiation (λ 0.71073 Å at 293 K crystal structure CuP2 solved atomic thermal displacement factors refined full-matrix technique calculations using Jana2006 experiments time-of-flight cold-neutron spectrometer-Pelican46 Australian Nuclear Science Technology Organization powder sample annular aluminum sample can 1 mm gap attached cold head closed-cycle refrigerator 1.5 to 800 K instrument aligned for 4.69 Å (3.7 meV) neutrons resolution elastic line 135 μeV intensity empty can subtracted data normalized to vanadium standard data manipulations Large Array Manipulation Program (LAMPscattering function S(Q wave vectors phonon energy measured energy gain mode temperature range transformed to generalized PDOS formula (3) kB Boltzmann’s constant T temperature3[12pt{minimal{amsmath\oddsidemargin-69pt E\frac{E}{{Q^2}}S -{E=∫EQ2SQ,E1−e−EkBTdQ phonon dispersion measurements single crystal 0.7 g pre-orientated scattering plane defined by [H, 0, 0] L,] large reciprocal space covered rotating sample over 100° 1° around axis perpendicular scattering plane incident neutron wavelength 2.345 Å (14.8 meV), second-order reflection Pelican highly oriented pyrolytic monochromators 4.69 Å sample temperature set 300 K data reduced by LAMP whole S(Q,E) generated using HORACE49 phonon dispersionssurvey Brillouin zones Pelican instrument focused zones G = (200), (022) (111) thermal neutron triple-axis spectrometer-Taipan ANSTO50,51 energies incident scattered neutrons defined double-focused PG (002) monochromator analyzer open geometry neutron beam virtual source width 10 mm exit slit 20 mm detector measurements final neutron energy 14.87 meV HOPG filter reflections sample aligned same scattering plane Pelican phonons measured [H, 0, 0 L] directions standard cryofurnace temperatures 100 300 450 K data fitted Lorentzian function PAN DAVE52.Brillouin light measurements room temperature Sandercock-type six-pass tandem Fabre–Perot (TFP-2) spectrometer 532 nm laser light source-established technique measuring sound velocities53backscattering geometry LA mode sound velocity calculated-69pt}vLA=fBλ02n measured Brillouin frequency shift fB refractive index n = 4.04 laser wavelength λ0 532 nm LA-wave velocity vLA = 6275 m s−1.DFT ab initio DFT VASP code54 projector augmented wave pseudopotentials55 Perdew–Burke–Ernzerhof56 electron-ion inter-electron exchange-correlation interactions wave functions expanded plane waves energy cutoff 500 eV electronic energy convergence 10−8 eV Brillouin zone primitive unit cell sampled Γ-centered 8 × 10 × 15 k-point mesh structural optimization atomic force less than 0.001 eV phonon dispersion relation calculated Phonopy package57 VASP 2 × 2 × 2 supercell 96 atoms Γ-centered 2 × 2 × 2 q-point mesh usedunravel chemical bonding Cu–Cu atomic pair calculated ELF58 electronic charge transfer = ρ(CuP2) − ρ(atom), electron charge density difference CuP2 atoms.Supplementary Additional Movie 3 | 48.7 | 0.804107 |
10.1038/s41467-020-16889-z | PMC7314798 | Detection of LPS derived from Gram-negative bacteria by innate immune receptors is a critical step in the host response. Here Santos and colleagues show human GBP1 binds to LPS resulting in non-canonical inflammasome activation. | The human non-canonical inflammasome controls caspase-4 activation and gasdermin-D-dependent pyroptosis in response to cytosolic bacterial lipopolysaccharide (LPS). Since LPS binds and oligomerizes caspase-4, the pathway is thought to proceed without dedicated LPS sensors or an activation platform. Here we report that interferon-induced guanylate-binding proteins (GBPs) are required for non-canonical inflammasome activation by cytosolic Salmonella or upon cytosolic delivery of LPS. GBP1 associates with the surface of cytosolic Salmonella seconds after bacterial escape from their vacuole, initiating the recruitment of GBP2-4 to assemble a GBP coat. The GBP coat then promotes the recruitment of caspase-4 to the bacterial surface and caspase activation, in absence of bacteriolysis. Mechanistically, GBP1 binds LPS with high affinity through electrostatic interactions. Our findings indicate that in human epithelial cells GBP1 acts as a cytosolic LPS sensor and assembles a platform for caspase-4 recruitment and activation at LPS-containing membranes as the first step of non-canonical inflammasome signaling. | IntroductionDetection of lipopolysaccharide (LPS) is central to host defense against Gram-negative bacterial infections and to the pathogenesis of sepsis. Extracellular LPS is sensed by Toll-like receptor 4 (TLR4), which induces the production of cytokines via the MyD88 and TRIF signaling pathways1. Cytosolic LPS, on the other hand, is detected by the so-called non-canonical inflammasome, which controls the activation of caspase-4/−5 in humans and caspase-11 in mice2–6. These caspases cleave the pore-forming cell death effector gasdermin D (GSDMD) to induce pyroptosis and cytokine release. While the activation of other caspases requires their recruitment to multiprotein platforms formed by dedicated sensor and adaptor proteins (e.g., DISC, apoptosome and canonical inflammasome), no comparable platform has yet been reported for caspase-4/−5 or −11. Instead, their activation appears to involve a new mode of pattern recognition in which caspase-4/−11 act both as sensor and executor without the need for additional adaptor proteins or co-factors5. This model was proposed based on the observation that caspases-4/−11 binds the highly hydrophobic lipid A moiety of LPS through their CARD (caspase recruitment domain), resulting in their oligomerization and activation5. However, since LPS is hydrophobic and normally present within bacterial membranes, it is conceivable that cytosolic LPS sensing could require accessory factors in analogy to LPS-binding protein (LBP) or cluster of differentiation 14 (CD14) that are required for TLR4 signaling. LBP binds to LPS-containing outer membrane of bacteria and promotes the transfer of LPS onto CD14, which then delivers LPS to the MD-2/TLR4 complex7,8.Caspase-11 activation in mouse macrophages transfected with LPS or infected with Gram-negative bacteria requires the expression of interferon (IFN)-inducible GTPases, which include the GBPs (guanylate-binding proteins) or IRGs (immunity-related GTPases)9–12. These GTPases are highly upregulated after type-I or type-II IFN priming, and essential for cell-autonomous immunity against a variety of viruses, bacteria and parasites13. In macrophages, several GBPs as well as Irgb10 were found to target intracellular Gram-negative bacteria, such as Salmonella enterica serovar Typhimurium (referred to as Salmonella), Francisella novicida and Escherichia coli. Since this recruitment correlated with bacterial lysis and the activation of caspase-11, it gave rise to a model in which GBPs recruit Irgb10 towards bacterial membranes, thereby unleashing an Irgb10-dependent membranolytic activity that kills the pathogen and concomitantly liberates LPS for the activation of non-canonical inflammasome11. However, since humans lack the IRG family (except for a truncated IRGM copy and IRGC), and GBPs are nevertheless required for LPS-induced caspase-4 activation, the current model needs to be confirmed in human cells14,15.Here we report that IFNγ priming and the induction of GBPs are necessary for caspase-4 activation in human epithelial cells and monocytes/macrophages during infection with the Gram-negative bacterium Salmonella or after cytosolic LPS delivery by transfection or electroporation. We show that human GBP1 targets cytosolic Salmonella seconds after the bacteria escape from the vacuole and enter into the cytosol, and that GBP1 initiates the hierarchical recruitment of GBP2-4 and the assembly of a GBP coat on cytosolic bacteria. This GBP coat does not induce bacteriolysis, but instead initiates the recruitment and activation of caspase-4 to the surface of cytosolic bacteria. Human GBPs play distinct functional roles in this process: GBP1 together with GBP4 recruit caspase-4, whereas GBP3 is mainly required for caspase-4 activation. Investigating the mechanism by which GBP1 recognizes cytosol-exposed bacteria, we demonstrate that LPS associates with GBP1 in pyroptotic cells and that recombinant GBP1 binds LPS with high affinity. Monomeric GBP1 associates with LPS micelles to form a high-molecular weight complex upon incubation with LPS, and this association occurs via electrostatic interactions involving negative charges on LPS. Consistently, mutagenesis of GBP1 shows that positively charged residues are necessary for LPS binding and recruitment to bacteria. In conclusion, we show that GBP1 acts as a bona-fide cytosolic LPS sensor that detects and targets the LPS-containing membranes of Gram-negative bacteria, where it assembles a platform that promotes caspase-4 recruitment and activation.ResultsSalmonella-induced caspase-4 activation requires IFNγ primingTo study the human non-canonical inflammasome and its modulation by priming, we infected naive or IFNγ-primed HeLa cells, which lack canonical inflammasome pathways, with the facultative intracellular bacterium Salmonella. Since HeLa cells express TLR4 but not MD-2 and are thus not responsive to extracellular bacterial LPS16, we primed the cells with IFNγ, a cytokine that also plays a critical role in intestinal immunity against Salmonella17. Salmonella replicated rapidly in naive HeLa but was strongly restricted in IFNγ-primed cells (Fig. 1a), despite similar levels of bacterial invasion (Supplementary Fig. 1a). Strikingly, IFNγ-primed HeLa cells underwent lytic cell death with typical features of pyroptosis, such as plasma membrane swelling and ballooning, and nuclear condensation (Fig. 1b, Supplementary Fig. 1b–e and Supplementary Movies 1, 2), and released mature IL-18 (Supplementary Fig. 1f). Since in epithelial cells a subset of Salmonella escape from the Salmonella-containing vacuole (SCV) into the cytosol within the first hour after entry18,19, we hypothesized that Salmonella could activate the non-canonical inflammasome as previously observed in mouse macrophages20. To test this, we infected naive or IFNγ-primed wild-type, CASP4–/– and GSDMD–/– HeLa (Supplementary Fig. 1g). Deletion of CASP4 or GSDMD did not alter bacterial invasion, but abrogated Salmonella-induced IFNγ-dependent cell death (Fig. 1c and Supplementary Fig. 1h–k), confirming that Salmonella infection of HeLa cells activates the non-canonical inflammasome in an IFNγ-dependent manner. While bacterial replication was increased, IFNγ priming still partially reduced intracellular bacterial replication in CASP4–/– and GSDMD–/– HeLa (Supplementary Fig. 1l), suggesting that cell death was not the only mechanism by which IFNγ restricts bacterial growth. This finding was confirmed using Salmonella expressing PuhpT-GFP, a reporter for cytosolic replication (e.g., GFP under the control of the hexose phosphate transporter promoter, which responds to exogenous glucose-6-phosphate21 found exclusively in the host cytosol) (Supplementary Fig. 1m–o). Furthermore, using a chloroquine (CHQ)-resistance assay, an antimicrobial agent that only reaches bactericidal levels when concentrated within endocytic compartments22,23, we found that IFNγ priming mainly restricted cytosolic Salmonella (Supplementary Fig. 1p, q) thus reducing hyper-replication of the cytosolic population of Salmonella (Supplementary Fig. 1r, s)24,25. Thus, IFNγ controls a major caspase-4- and GSDMD-dependent mechanism that restricts cytosolic Salmonella replication by inducing host cell pyroptosis, and a minor mechanism that acts independently of cell death.Fig. 1IFNγ priming is required for LPS-induced caspase-4 activation in human epithelial cells.a–c Intracellular bacterial fold-replication (a) and release of LDH (b, c) in naive or IFNγ-primed wild-type, CASP4–/– or GSDMD–/– HeLa cells, at 1 or 6-h post-infection (p.i.) with Salmonella. Cells in 96-well plates were infected for 30 min, washed and gentamicin was added to kill extracellular bacteria. At the indicated time points supernatant was collected to determine the release of LDH, and then cells were lysed and the number of viable intracellular bacteria was determined by counting colony forming units (CFUs). The bacterial fold-replication was calculated relative to 1 h p.i. d Percentage of CHQ-resistant cytosolic Salmonella in naive or IFNγ-primed HeLa at 1.5 h p.i. Cells were infected for 30 min as in (a) and then treated with gentamicin ± CHQ for an additional 1 h before cells were lysed and bacteria counted by CFUs. The percentage of cytosolic bacteria was calculated as the ratio of (CHQ + gentamicinresistant / gentamicinresistant). e–g Release of LDH from naive or IFNγ-primed cells, 5 h after transfection with LPS (2.5 µg/50,000 cells) or 3–4 h after electroporation with LPS (300 ng/50,0000 cells). h Western blot analysis of full length (p43) and cleaved (p32) caspase-4 in the supernatants and cell lysates from naive or IFNγ-primed HaCaT cells, upon transfection with E. coli LPS LPS (2.5 µg/50,000 cells). i Streptavidin pull-down assay of the binding of biotin-conjugated LPS to endogenous caspase-4 from the lysates of naive or IFNγ-primed HBEC3-KT. Cells in 6-well plates were transfected with LPS-biotin (10 µg) or left untransfected, and biotinylated substrate was pulled down using equal amounts of streptavidin magnetic beads, which were then eluted in equal volumes of SDS-PAGE reducing sample buffer. Streptavidin-bound and -unbound fractions were analyzed by immunoblotting for caspase-4. Graphs show the mean ± SD, and data are pooled from two to six independent experiments performed in triplicate (a–g) or representative of two (h, i) independent experiments. *** P < 0.001; ns, not significant; two-tailed t-test.We next determined whether IFNγ was necessary to promote access of Salmonella or their LPS to the cytosol by quantifying bacterial resistance to CHQ at 1.5 h post-infection (p.i.). When compared with naive cells, IFNγ priming did not induce Salmonella escape to the host cytosol (Fig. 1d), indicating that IFNγ controlled the detection of LPS after bacterial entry into the cytosol. To confirm this, we next transfected cells with ultrapure E. coli or Salmonella LPS. Similarly to Salmonella infection, LPS transfection only caused cell death in IFNγ-primed HeLa and was completely abrogated by deletion of CASP4 or GSDMD (Fig. 1e and Supplementary Fig. 2a). IFNγ-priming was also required for pyroptosis and IL-18 release after LPS transfection (Fig. 1f and Supplementary Fig. 2b–d) and after LPS electroporation (Fig. 1g and Supplementary Fig. 2e) in a panel of human cell lines and primary cells, including human small intestinal epithelial cells (HIEC-6). To further substantiate that caspase-4 activation requires IFNγ, we transfected LPS into naive or IFNγ-primed cells and pulled down active caspase-4 using a cell-permeable pan-caspase activity probe, biotin-VAD(Ome)-fmk (bVAD-fmk)26. Active caspase-4 was only pulled down when cells where first primed with IFNγ and then transfected with LPS (Supplementary Fig. 2f). In accordance, LPS transfection only induced caspase-4 and GSDMD cleavage in IFNγ-primed cells (Fig. 1h and Supplementary Fig. 2g). Importantly, IFNγ-priming had no impact on the level of caspase-4 expression, since unlike murine caspase-11, caspase-4 was not induced by IFNγ (Fig. 1h, i and Supplementary Fig. 2c, g, h). To assess if IFNγ controlled caspase-4 activation upstream or downstream of LPS binding, we prepared lysates from HeLa or HBEC3-KT cells transfected with biotinylated LPS and pulled down LPS-interacting proteins with streptavidin-coupled beads. In both cell types caspase-4 could only be pulled-down with LPS in IFNγ-primed but not in naive cells (Fig. 1i and Supplementary Fig. 2h). Altogether, these findings suggest that in human epithelial cells one or several IFNγ-induced proteins are required for LPS-induced caspase-4 activation.GBP1 is required for non-canonical inflammasome activationSince caspase-11 activation in mouse macrophages requires IFN-induced GTPases, we speculated that human GBPs were necessary for LPS-induced caspase-4 activation. In agreement with previous studies27, we found that GBPs expression in HeLa cells was strongly upregulated by IFNγ priming (Supplementary Fig. 3a). RNA interference-mediated silencing of GBPs expression revealed a consistent reduction of LDH release in cells lacking GBP1 both after Salmonella infection as well as LPS transfection (Supplementary Fig. 3b–g). To confirm the phenotype, we generated GBP1–/– HeLa cells by CRISPR-Cas9 genome targeting (Fig. 2a) and found that GBP1-deficiency completely abrogated LDH release down to the background levels that were observed in naive cells after Salmonella infection or LPS transfection (Fig. 2b, c and Supplementary Fig. 3h, i), without altering bacterial entry (Supplementary Fig. 3j–m). GBP1-deficient cells were also unable to cleave and activate caspase-4 upon LPS transfection or Salmonella infection (Fig. 2d, e, p32 fragment). Furthermore, GBP1 was strongly required for LDH release when LPS was delivered by electroporation (Fig. 2f). Similarly to the knockout of CASP4 or GSDMD, GBP1-deficiency or knock-down of individual GBPs in HeLa only resulted in a partial loss of IFNγ-dependent restriction of cytosolic Salmonella replication (compare Supplementary Figs. 1l-s and 3n-p). Finally, GBP1 knock-down in HBEC3-KT and HaCaT cells also reduced LDH release upon LPS transfection (Fig. 2g and Supplementary Fig. 3q), demonstrating that GBP1 is important to regulate LPS-induced cell death in several human epithelial cell lines.Fig. 2GBP1 is required for Salmonella- and LPS-induced caspase-4 activation to induce pyroptosis in epithelial cells.a Immunoblots for GBP1, caspase-4 and GAPDH (loading control) in cell lysates from IFNγ-primed wild-type or GBP1–/– HeLa. b, c Release of LDH from naive or IFNγ-primed wild-type or GBP1–/– HeLa after Salmonella infection (b) or after 5 h transfection with E. coli LPS (2.5 µg / 50,000 cells) (c). d, e Immunoblots for full length (p43) and cleaved (p32) caspase-4 in combined supernatants and cell lysates from naive or IFNγ-primed wild-type and GBP1–/– HeLa, upon transfection with E. coli LPS for 5 h (d) or Salmonella infection (e). f Release of LDH from IFNγ-primed wild-type or GBP1–/– HeLa, 3 h after electroporation with LPS (300 ng/50,000 cells). g Release of LDH in IFNγ-primed HBEC3-KT or HaCaT cells treated with non-targeting control siRNA (NT) or with siRNAs targeting CASP4, GSDMD or GBP1, after E. coli LPS transfection. Cells were treated with siRNAs for 24 h and transfected with LPS (2.5 µg / 50,000 cells) for 5 h. Graphs show the mean ± SD, and data are pooled from two (c, f), three (g) or four (b) independent experiments performed in triplicate, or representative of three independent experiments (d, e). ***P < 0.001; ns, not significant; two-tailed t-test.GBP1 targets intracellular Salmonella to recruit GBP2-4Having demonstrated a role for GBPs in caspase-4 activation, we next expressed fluorescently tagged GBPs individually in naive or primed cells to determine if they target intracellular Salmonella. We found that in IFNγ-primed HeLa, GBP1, −2, −3 and −4 coated around 20–30% of intracellular Salmonella at 1 h p.i., whereas only very few bacteria were positive for GBP5, −6 or −7 (Fig. 3a, b and Supplementary Fig. 4a). Recruitment of tagged GBP2, −3 and −4 was strongly dependent on IFNγ priming, since these GBPs only poorly associated with Salmonella when expressed in naive cells. By contrast, eGFP-GBP1 associated with Salmonella even when expressed in naive cells, albeit at lower levels than in primed cells.Fig. 3GBP1 targets Salmonella and controls recruitment of GBP2-4.a Fluorescence confocal microscopy of naive or IFNγ-primed HeLa expressing N-terminal tagged eGFP-GBP1-7 (green) and infected with Salmonella-dsRed (red) for 1 h. DNA was stained with Hoechst (blue). Representative confocal images are shown and scale bars correspond to 10 µm. b Percentage of intracellular Salmonella positive for eGFP-GBP1-7 in naive or IFNγ-primed HeLa, at 1 h p.i. At least 200–300 bacteria were counted per coverslip. c Fluorescence confocal microscopy of IFNγ-primed wild-type or GBP1–/– HeLa expressing eGFP-GBP2-4 (green) and infected with Salmonella-dsRed (red) for 1 h. Representative confocal images are shown, and scale bar corresponds to 10 µm. d Schematic representation of wild-type GBP1 and a ΔCaaX mutant. e, f Fluorescence confocal microscopy of naive GBP1–/– HeLa expressing HA-GBP1wt or HA-GBP1ΔCaaX (e) or co-expressing mCherry-GBP1wt and HA-GBP1wt or HA-GBP1ΔCaaX (f) and infected with Salmonella for 1 h. HA-tagged GBP1 was visualized by immunostaining with an anti-HA antibody. Representative confocal images are shown and scale bars correspond to 5 µm. The percentage of HA-GBP1 positive bacteria was quantified by counting around 100 bacteria per coverslip. Graphs show the mean ± SD, and data are pooled from two (b, e, f) independent experiments performed in duplicate or representative of two (a, c, e, f) independent experiments.GBPs are known to homo- and hetero-oligomerize, forming a coat when recruited to intracellular pathogens28. We hypothesized that GBP1 had the ability to target Salmonella independently of other GBPs, whereas GBP2-4 required GBP1 for recruitment27. Indeed, recruitment of GBP2, −3 and −4 to Salmonella was completely abrogated in IFNγ-primed GBP1–/– HeLa cells (Fig. 3c). Furthermore, GBP1 co-expression in naive cells was sufficient to induce recruitment of tagged GBP2-4 to intracellular Salmonella (Supplementary Fig. 4b, c) to levels similar to those observed in IFNγ-primed cells (Fig. 3b). GBP1 targeting of intracellular Salmonella was also observed in primary human small intestinal epithelial cells (Supplementary Fig. 4d). Finally, time-lapse confocal microscopy of infected cells co-expressing GBP1 and either GBP2, −3 or −4, revealed that GBP1 and GBP2 were recruited simultaneously to the bacteria (Supplementary Fig. 4e and Supplementary Movie 3), whereas GBP3 and GBP4 were only recruited to bacteria minutes after GBP1 recruitment (Supplementary Fig. 4f, g and Supplementary Movies 4, 5). These data show that GBP1 is the first to target intracellular Salmonella and orchestrates the hierarchical recruitment of additional human GBP family members, namely GBP2-4. To gain further mechanistic insights on how GBP1 accumulates on intracellular Salmonella we analyzed different GBP1 mutants (Fig. 3d and Supplementary Fig. 5a, b)29. GTP hydrolysis, a triple-arginine polybasic motif (584-586) and the C-terminal CaaX box-dependent farnesylation were all required for proper GBP1 recruitment/accumulation around intracellular Salmonella (Fig. 3e and Supplementary Fig. 5c). Interestingly, HA-GBP1ΔCaaX was still not recruited to Salmonella even when co-expressed with mCherry-GBP1wt (Fig. 3f), suggesting that GBP1 oligomers around bacteria are only formed if monomers are farnesylated and capable of properly inserting into membranes.GBPs target cytosolic Salmonella seconds after SCV ruptureSince mouse GBPs were previously associated with vacuolar rupture10, we next asked if human GBPs associate with the SCV or cytosolic Salmonella by using galectin-3, a protein that binds to β-galactosides found on the inner leaflet of vacuolar membranes, as a marker for ruptured vacuoles30. GBP1-4 were indeed only found in the vicinity of galectin-3-positive bacteria (Fig. 4a), but closer analysis revealed that GBPs targeted the cytosol-exposed part of the bacteria and not the galectin-3-positive ruptured vacuoles (Fig. 4a, insets). Consistently, GBPs did not co-localize with LAMP1 (Supplementary Fig. 6a), a known marker of SCV membranes31. Furthermore, we did not observe a reduction in the percentage of cytosolic (Fig. 4b) or galectin-3-positive Salmonella (Supplementary Fig. 6b) in GBP1–/– cells, indicating that human GBP1-4 did not promote the escape of Salmonella from the SCV in HeLa cells, but associate with bacteria upon cytosolic entry. GBP1 also associated with Shigella flexneri after its escape from the endocytic vacuole but not with cytosolic Listeria monocytogenes (Supplementary Fig. 6c)27,29.Fig. 4GBP1 targets cytosolic Salmonella and is required for caspase-4 recruitment to the bacterial surface.a Fluorescence confocal microscopy of IFNγ-primed wild-type HeLa co-expressing galectin-3-eGFP (green) and mCherry-GBP1-7 (red) and infected with Salmonella for 1 h. b Percentage of CHQ-resistant cytosolic Salmonella in naive or IFNγ-primed wild-type or GBP1–/– HeLa at 1.5 h p.i. Cells in triplicate wells were infected for 30 min and then treated with gentamicin ± CHQ for an additional 1 h before lysing the cells and determining CFUs. The percentage of cytosolic bacteria was calculated as the ratio of (CHQ + gentamicinresistant/gentamicinresistant). c Fluorescence confocal microscopy of naive or IFNγ-primed HeLa expressing caspase-4-eGFP (green) and infected with Salmonella-dsRed for 1 h. d. Percentage of caspase-4-eGFP positive Salmonella at 1 h p.i., quantified by counting 100–200 bacteria per coverslip. nd, not detected. e, f Fluorescence confocal microscopy of IFNγ-primed HeLa co-expressing galectin-3-mOrange (red) and caspase-4-eGFP (green) and infected with Salmonella for 1 h. g Time-lapse fluorescence confocal microscopy of IFNγ-primed HeLa expressing caspase-4-eGFP (green) and galectin-3-mOrange (red) and infected with Salmonella. h. Mean normalized fluorescence intensities of galectin-3-mOrange and caspase-4-eGFP over time. Fluorescence intensities were quantified in a region of interest as exemplified in the figure, containing an event of caspase-4 and galectin-3 recruitment to an individual bacterium. The relative intensity signals were aligned using the time point of onset of galectin-3 recruitment as zero and the mean and SD of six different events were plotted. i Time-lapse fluorescence confocal microscopy of IFNγ-primed HeLa co-expressing caspase-4-eGFP (green) and mCherry-GBP1 (red) and infected with Salmonella. Images were acquired every 60 s. DIC, differential interference contrast. j Percentage of caspase-4-eGFP positive Salmonella at 1 h p.i., in IFNγ-primed wild-type and GBP1–/– HeLa. 100–200 bacteria were counted per coverslip. Representative confocal images are shown and scale bars correspond to 1 µm (f), 5 µm (c, e, g) or 10 µm (a, i). Graphs show the mean ± SD, and data are pooled from two (b) or three (d, j) independent experiments performed in triplicate or representative from at least three independent experiments (a, c, e–i). ***P < 0.001; ns, not significant, two-tailed t-test.We next used time-lapse confocal microscopy to follow the kinetics of SCV membrane rupture, GBP recruitment and pyroptosis. In Salmonella-infected cells, SCV rupture and galectin-3 recruitment was followed by a rapid and massive recruitment of GBP1, often occurring less than 30 seconds upon detectable galectin-3 appearance (Supplementary Fig. 6d and Supplementary Movie 6). GBP1 recruitment often started at one region of the bacterium, presumably the part exposed to the cytosol, then formed a coat around bacteria. GBP1 recruitment to cytosolic Salmonella was followed by pyroptotic cell death (Supplementary Fig. 6e and Supplementary Movie 7), and consistently, the majority of pyroptotic cells featured GBP1-positive Salmonella (Supplementary Fig. 6f).In mouse macrophages, GBP targeting of bacteria mediates recruitment of Irgb10, which correlates with the lysis of targeted bacteria and caspase-11 activation11. To determine if human GBPs lysed cytosolic Salmonella in HeLa cells, as a mechanism for LPS release and caspase-4 activation, we monitored GBP recruitment in GSDMD–/– cells infected with Salmonella-dsRed. Live-cell imaging revealed that while the bacteria were rapidly targeted by GBP1, they continued to divide in the host cell cytosol, showing no signs of lysis or viability loss (Supplementary Fig. 7a–d and Supplementary Movies 8–11). Therefore, GBP1 targets the surface of Salmonella within seconds upon rupture of the SCV membrane and bacterial escape to the host cytosol, and is followed by rapid induction of caspase-4-dependent pyroptosis independently of bacteriolysis.GBPs control caspase-4 recruitment to cytosolic SalmonellaSince GBP recruitment did not lyse bacteria, we presumed that GBPs control LPS-dependent caspase-4 activation by another mechanism. When probing for the intracellular localization of caspase-4 we found that caspase-4-eGFP was recruited on Salmonella, often covering the entire bacterium (Fig. 4c). Caspase-4 recruitment onto Salmonella absolutely required IFNγ priming as it was not detectable in naive cells (Fig. 4c, d and Supplementary Fig. 8a) despite high expression levels of the caspase. Time-lapse confocal microscopy of caspase-4-eGFP-expressing IFNγ-primed HeLa infected with Salmonella-dsRed showed that caspase-4, although initially diffused in the cytosol, was recruited to intracellular Salmonella within minutes, and that caspase-4 recruitment to Salmonella was rapidly followed by pyroptotic cell death in the majority of cells (Supplementary Fig. 8b, c and Supplementary Movie 12). Consistently, the few cells that did not recruit caspase-4 to bacteria did not initiate pyroptosis (Supplementary Fig. 8d and Supplementary Movie 13).Recruitment of caspase-4 to Salmonella correlated with SCV lysis because caspase-4-positive bacteria were found in the vicinity of galectin-3-positive ruptured vacuoles (Fig. 4e and Supplementary Fig. 8e). Similarly to GBP1-4, caspase-4 did not co-localize with LAMP1-positive SCVs (Supplementary Fig. 8f), but targeted the cytosol-exposed part of the bacterium and not the lysed vacuole (Fig. 4e, arrows and arrowheads, and Supplementary Fig. 8e, inset images). Super-resolution microscopy further confirmed that caspase-4 accumulated on the bacterial surface but not on ruptured SCVs (Fig. 4f and Supplementary Fig. 8g). Finally, time-lapse confocal microscopy of infected cells showed that caspase-4 was recruited to Salmonella 5–10 min after the first appearance of a galectin-3 signal (Fig. 4g, h and Supplementary Movie 14), altogether demonstrating that caspase-4 targets cytosolic Salmonella after SCV rupture.Since GBP1 also targeted cytosolic Salmonella and was required for caspase-4 activation, we hypothesized that GBPs might either control the recruitment of caspase-4 to cytosolic bacteria and/or the activation of caspase-4 at the bacterial surface. Time-lapse microscopy of HeLa cells co-expressing mCherry-GBP1 and caspase-4-eGFP showed that GBP1 recruitment preceded caspase-4 recruitment to the same bacterium by several minutes, and this was followed by pyroptosis (Fig. 4i and Supplementary Movie 15). This is consistent with the faster recruitment of GBP1 upon SCV rupture compared with the slower recruitment of caspase-4 (Supplementary Fig. 6d and Fig. 4g, h). Remarkably, we also observed a complete reduction in caspase-4 recruitment to Salmonella in GBP1–/– HeLa (Fig. 4j), while GBP1 was still recruited to bacteria in CASP4–/– cells (Supplementary Fig. 9). Together with the observation that GBP1-deficient cells were not able to activate caspase-4 (Fig. 2), the data suggest that GBP1, either directly or by controlling GBP2-4 recruitment, initiates the recruitment of caspase-4 to the bacterial surface.GBP1/3/4 are sufficient for LPS-induced caspase-4 activationWe next addressed the individual functions of GBP1-4 in caspase-4 recruitment and activation. GBP1–/– HeLa neither recruit GBP2-4 nor caspase-4 to cytosolic bacteria (Figs. 3c and 4j), thus making it impossible to determine whether GBP1 controls caspase recruitment directly or via other GBPs. We therefore co-expressed caspase-4-eGFP and mCherry-GBP1 in naive or IFNγ-primed cells, and determined caspase targeting to cytosolic Salmonella. While GBP1 targeted cytosolic Salmonella regardless of IFNγ priming, caspase-4 was only recruited to bacteria in primed cells (Fig. 5a). Thus, other GBPs and/or an unknown IFNγ-induced factor are necessary for GBP1-dependent caspase-4 recruitment. We thus co-expressed caspase-4 and mCherry-GBP1 with either doxycycline (Dox)-inducible eGFP-tagged GBP2, GBP3 or GBP4 in naive HeLa cells and visualized caspase recruitment by confocal microscopy (Supplementary Fig. 10a and Fig. 5b). GBP1 alone, or GBP1 together with GBP2 did not restore caspase-4 recruitment to cytosolic Salmonella in naive cells. On the other hand, co-expression of GBP1 with GBP4 and to a lesser degree with GBP3 was sufficient to induce recruitment of caspase-4 to cytosolic bacteria (Fig. 5b, c). The same was observed when using different vectors to co-express the proteins (Supplementary Fig. 10b), which excluded vector-biased caspase-4 recruitment and indicated that GBP1 controls recruitment of caspase-4 mainly via GBP4 and partially via GBP3.Fig. 5GBP1/4 are sufficient to recruit caspase-4 to Salmonella and together with GBP3 activate the non-canonical inflammasome in naive human epithelial cells.a Fluorescence confocal microscopy of naive and IFNγ-primed HeLa co-expressing caspase-4-eGFP (green) and mCherry-GBP1 (red) and infected with Salmonella for 1 h. DNA was stained with Hoechst (blue). Representative confocal images are shown and scale bar corresponds to 5 µm. b Fluorescence confocal microscopy of IFNγ-primed or naive HeLa cells co-expressing mCherry-GBP1 (red), Dox-inducible eGFP-GBP1, −2, −3 or −4 (green) and caspase-4-V5 (gray), and infected with Salmonella for 1 h. DNA was stained with Hoechst (blue). eGFP-GBPs were expressed by inducing cells with 1 µg/mL Dox for 16 h. Caspase-4-V5 was visualized by immunostaining with an anti-V5 antibody. Representative confocal images are shown and scale bar corresponds to 10 µm. c Percentage of caspase-4-V5 positive Salmonella at 1 h p.i., quantified out of the mCherry-GBP1-positive bacteria. At least 50 GBP1-positive bacteria were counted per coverslip. d Percentage of cell death in HeLa cells co-expressing constitutive mCherry-GBP1 and Dox-inducible eGFP or eGFP-GBP1, −2, −3 or −4. FLAG-GBP3 and HA-GBP4 were constitutively expressed together using a bicistronic plasmid. Cells were transfected with the indicated plasmids for 24 h. eGFP-GBPs were induced for 16 h with 1 µg/mL Dox, whereas eGFP was induced for 3 h. Cells were then transfected with E. coli-derived LPS (2.5 µg/50,000 cells) for 6 h and cell death values were normalized considering IFNγ-primed HeLa as 100% and naive cells co-expressing mCherry-GBP1 and eGFP as 0%. Graphs show the mean ± SD, and data are pooled from two independent experiments performed in duplicate (c), pooled from three independent experiments performed in triplicate (d) or are representative from two (b) or three (a) independent experiments. **P < 0.01; ***P < 0.001; one-way ANOVA.We next asked if expression of single or multiple GBPs in naive cells also restores caspase-4-dependent pyroptosis. Individual expression of GBP1-7 was not sufficient to induce LPS-induced pyroptosis in naive cells (Supplementary Fig. 10c, d). By contrast, LPS transfection induced significantly elevated levels of LDH release in naive cells co-expressing either GBP1/3/4 or GBP1/2/3/4 (Fig. 5d and Supplementary Fig. 10e, f). Interestingly, co-expression of only GBP1/3 already restored some LPS transfection-induced pyroptosis, whereas co-expression of GBP1/4 had no effect even though it was sufficient to promote caspase-4 recruitment on Salmonella (Fig. 5b, c and Supplementary Fig. 10b). Thus, while GBP1 drives caspase-4 recruitment to cytosolic Salmonella mainly by GBP4, GBP3 is nevertheless necessary to yield caspase-4 activation (Fig. 5b, d). In conclusion, these experiments indicate that a complex formed by GBP1, −3 and −4 promotes caspase-4 recruitment and activation following LPS detection without the requirement for other IFNγ-induced genes.GBP1 directly binds LPSSince GBPs showed a recruitment to cytosolic Gram-negative bacteria, OMVs or even transfected LPS (Fig. 3, Supplementary Fig. 6c and ref. 12), we tested if GBPs can directly bind LPS. Similarly to caspase-4 (Fig. 1i and Supplementary Fig. 2g), biotin-LPS was able to pull-down eGFP-GBP1 and to a lesser degree eGFP-GBP3, but not tagged GBP2 or GBP4 (Fig. 6a) from HeLa lysates. To assess if this interaction was direct, we next purified LPS-free recombinant His-GBP1 from CleanColi® BL21 (DE3) bacteria that produce Lipid IVA (which does not activate caspase-4/−11) instead of LPS followed by a lipid removal protocol32 and tested GBP1-LPS binding by surface plasmon resonance (SPR). SPR showed a direct binding of LPS to immobilized GBP1 with a KD of ~60 nM, which is comparable to the published KD of the LPS-caspase-4 and LPS-caspase-11 interaction (Fig. 6b–d). Kinetic analysis also showed that GBP1-LPS binding best fitted with a two-state-reaction model that describes a situation where initial binding is followed by a conformational change that stabilizes the complex. The complementary experiment with immobilized LPS also yielded similar a KD (Supplementary Fig. 11a, b), however the response was relatively weak since only a small amount of LPS absorbed on the chip surface. Moreover, microscale thermophoresis (MST) of GBP1 and FITC-LPS confirmed the interaction yielding a comparable KD value (Supplementary Fig. 11c).Fig. 6LPS binds to GBP1 to induce formation of a high-molecular weight protein complex.a Streptavidin pull-down assay for eGFP-GBP1-4 using biotin or biotin-conjugated LPS. HeLa cells stably expressing Dox-inducible eGFP-GBP1, −2, −3, or −4 were primed with IFNγ and 1 µg/mL Dox was added for 16 h. 1 million cells were lysed and incubated with 2 µg LPS-biotin or biotin, and the biotinylated substrates were pulled down using equal amounts of streptavidin magnetic beads, which were then eluted in equal volumes of SDS-PAGE reducing sample buffer. Streptavidin-bound and -unbound fractions were analyzed by western blot using an antibody against GFP. b SPR sensorgram of E. coli LPS (O111:B4) binding to human GBP1 immobilized on a CM5 chip surface. Sensorgram was obtained by using different LPS concentrations (47, 94, 188, 375, 750, and 1500 nM). Gray lines correspond to SPR data and orange lines to model fits using a two-state-reaction model. c Saturation curve of the titration of LPS on GBP1 immobilized on a CM5 chip. d Calculated dissociation constants (KD) for LPS binding to immobilized GBP1 (GBP1im) or GBP1 binding to immobilized E. coli LPS (LPSim). Dissociation constants for LPS-caspase-4 and LPS-caspase-11 were previously published by Shi et al.5. e, f Size exclusion chromatograms of recombinant, LPS-free His-tagged GBP1 incubated with various LPS derivatives. Following purification, GBP1 (1 µM) was incubated on ice with LPS (2 µM) for 5 h before being subjected to size-exclusion analysis on a Superdex 200 10/30 GL column. Protein size was estimated using molecular weight standards. Curves were corrected by subtracting LPS-specific absorbance at 280 nm. Individual fractions were run on a 12% acrylamide gel and immunoblotted against His6 to confirm the presence of GBP1 in elution peaks (f). g GTPase activity analysis of recombinant GBP1. GBP1 (500 nM) was incubated with GTP (5 µM) with or without ultrapure LPS (5 µM) for 30 min before the reaction was stopped. Luminescence was normalized to a buffer-only control. Graphs show the mean ± SD, and data are representative from three (a–d, g) or five (e, f) independent experiments performed with at least three independently expressed and purified batches of recombinant His-GBP1.We next investigated the consequences of the LPS-GBP1 interaction, since LPS was suggested to induce caspase-11/−4 oligomerization5. LPS-free recombinant GBP1 ran as a single monomeric peak on size-exclusion chromatography (SEC), close to its predicted size of 68.5 kDa (Fig. 6e, f, Supplementary Fig. 11d and Table 1), while E. coli and Salmonella Typhimurium LPS formed micelles eluting at around 1000 kDa (void volume). When GBP1 was incubated with LPS, the elution profile changed, resulting in a shift of the majority of GBP1 to higher molecular weight peaks from the range of 400 kDa to over 1000 kDa (Fig. 6e, f, Supplementary Fig. 11d and Table 1). The most prominent peak was found to be at 1000 kDa, indicating that GBP1 bound to LPS micelles, but smaller complexes were detected as well. Interestingly, even LPS from Rhodobacter sphaeroides induced a shift of GBP1 to higher molecular weight peaks (Supplementary Fig. 11d, e), despite acting as an antagonist of caspase-11 and not being able to induce oligomerization of the caspase5. Conversely, incubation of GBP1 with lipoteichoic acid (LTA) or peptidoglycan (Supplementary Fig. 11f, g) did not result in a similar shift. LBP and ovalbumin were used as positive and negative LPS-binding controls, respectively (Supplementary Fig. 12a)33. We next tested what part of LPS is required for the binding of GBP1 to LPS micelles. LPS from E. coli Ra, Rc, Rd, Re mutant strains, which lack the O-antigen and outer core, respectively, was still bound by GBP1 as it resulted in a shift to high-molecular weight peaks (Supplementary Fig. 12b, c). Consistently, Salmonella ΔwaaL or ΔwaaG mutants (lacking the O-antigen or outer core, respectively) still recruited GBP1 and caspase-4 in infected cells (Supplementary Fig. 12d–f). Since the LPS-GBP1 interaction was highly sensitive to detergents, the binding to Lipid A could not be tested since this requires solubilization of Lipid A with Tween-20 or other detergents5. Since GBP1 displays an oligomerization-dependent activation of GTP hydrolysis34, we finally assessed the impact of LPS on GBP1 GTPase activity (Fig. 6g). As reported previously, GBP1 had some intrinsic ability to hydrolyze GTP on its own, but its GTPase activity was significantly increased when incubated with LPS, supporting the fact that GBP1 interacts with LPS, and suggesting that LPS binding might promote an oligomeric state. In conclusion these findings indicated that GBP1 was able to bind directly to LPS, and that the LPS Lipid A and inner core region were sufficient for LPS-GBP1 interaction.Table 1Observed molecular weights of GBP1 peaks after incubation with various LPS.SampleTheoretical size (kDa)Observed size (kDa)GBP167.0178.7GBP1 + LPS-E. coli1028; 401.8; 78.7GBP1 + LPS-Salmonella1140.5, 78.7GBP1 + LPS-R. sph1318.6; 364GBP1-LPS interaction involves electrostatic interactionsEach LPS molecule from E. coli has 6–8 negatively charged groups, from phosphates and acid groups in the Lipid A and inner core (Supplementary Fig. 12b)35. Given that human GBP1 lacks the hydrophobic pockets that comprise the LPS binding sites in CD14 and MD-2, we hypothesized that GBP1 binds to LPS by electrostatic interactions, similarly to LBP36,37. Consistently, GBP1 binding to LPS micelles was disrupted by incubation with cations (Ca2+), which neutralize the negative charges on LPS35,38, or by incubation with polymyxin B, which interacts with the LPS Lipid A and inner core region through ionic and hydrophobic forces, resulting in more monomeric GBP1 and reduced levels of GBP1-LPS association (Fig. 7a and Supplementary Fig. 13). Furthermore, dephosphorylating LPS with alkaline phosphatase reduced GBP1-LPS binding and cell death upon LPS transfection (Fig. 7a, b and Supplementary Fig. 13), indicating that the phosphate groups found on Lipid A and inner core sugars of LPS (Supplementary Fig. 12b) play an essential role in promoting GBP1-LPS interaction and subsequent activation of the non-canonical inflammasome pathway. Basic residues in the N-terminal domain of LBP mediate binding to LPS micelles39. We thus mutated several positively charged surface patches in GBP1 and tested the impact of the mutations on GBP1-LPS interaction by SEC (Fig. 7c, d). While most mutations resulted in no or minor effect on the binding of GBP1 to LPS micelles, mutation of the triple-lysines 61-63 to alanines (A patch) notably reduced the formation of the higher molecular weight GBP1 peaks (~40%) and increasing the levels of monomeric GBP1, thus suggesting that these residues are required for binding (Fig. 7d and Supplementary Fig. 14a). Furthermore, expression of GBP1KKK61-63AAA (A patch) resulted in a significant reduction of targeting to cytosolic Salmonella compared with either GBP1WT or GBP1KK87-88AA (B patch) (Fig. 7e, f and Supplementary Fig. 14b). In summary, the results demonstrate that the GBP1-LPS interaction involves electrostatic forces and that disrupting the binding by dephosphorylating LPS or mutating GBP1 results in reduced caspase-4-induced pyroptosis upon LPS transfection or impaired GBP1 targeting of the bacterial surface.Fig. 7GBP1 is recruited to the bacterial surface and binds LPS through electrostatic interactions.a Size exclusion chromatograms of recombinant His-tagged GBP1 incubated with E. coli LPS, or with E. coli LPS pre-treated with CaCl2 (5 mM), Polymyxin B (10 µg/mL) or with alkaline phosphatase. Curves were corrected by subtracting the respective LPS-specific absorbance at 280 nm. Black curves representing control condition were overlayed. b Release of LDH from IFNγ-primed HeLa 5 h after transfection with E. coli LPS, or after transfection with LPS previously treated with alkaline phosphatase. c 3D structure of human GBP1 (PDB 1f5n), highlighting five different negatively charged patches (A to E). Residues comprising patch E are only visible in PDB 6k1z. For each patch, the indicated residues were all mutated to alanines and analyzed for GBP1-LPS interaction by size exclusion chromatography. Purple indicates GTPase domain, green indicates helical domain. d Size exclusion chromatograms of different His-tagged GBP1 mutants incubated with E. coli LPS. Curves were corrected by subtracting LPS-specific absorbance at 280 nm. e Fluorescence confocal microscopy of naive HeLa expressing eGFP-GBP1wt, eGFP-GBP1KKK61-63AAA or eGFP-GBP1KK87-88AA and infected with Salmonella-dsRed for 1 h. DNA was stained with Hoechst (blue). Representative confocal images are shown and scale bar corresponds to 5 µm. f Percentage of eGFP-GBP1 positive Salmonella at 1 h p.i., as quantified by counting between 100–200 bacteria per coverslip. Graphs show the mean ± SD, and data are pooled from three independent experiments performed in duplicate (f), four independent experiments performed in triplicate (b), or are representative from three (a, d, e) independent experiments. ***P < 0.001; ns, not significant, two-tailed t-test.DiscussionHere we report that GBP1 functions as an LPS sensor that recognizes Gram-negative bacteria in the cytosol of human epithelial cells, and that GBP1-LPS interaction involves electrostatic forces (Supplementary Fig. 15). Given that GBP1 is necessary for LPS-induced caspase-4 activation in various human cell types and after various LPS delivery methods (electroporation, chemical transfection, Gram-negative bacteria infection), our data imply that GBP1 is the very first protein in the non-canonical inflammasome pathway that interacts with LPS. This places GBP1 upstream of caspase-4 in cytosolic LPS sensing, raising the question if it might act similarly to LBP which acts as a co-factor for extracellular LPS detection by CD14 and MD-2/TLR4. Indeed, the finding that GBP1-LPS requires negative charges on LPS Lipid A and the inner core sugars as well as a positively charged surface patch in GBP1 is reminiscent of the mode of LPS binding by LBP, which involves two positively charged patches at the tip of the LBP N-terminal domain. From a functional point of view LBP and GBP1 will most likely differ. LBP binds to LPS micelles and to CD14 protein, and catalyzes multiple rounds of LPS transfer to CD14, which will then transfer a single bound LPS molecule to MD-2/TLR4. GBP1 on the other hand, does not function alone, but as part of a GBP1-4 complex that assembles on LPS-containing membranes. It is possible that this complex recruits caspase-4 and then transfers LPS onto caspase-4 thus promoting its activation. Alternatively, it is also possible that the assembly of the GBP complex and insertion of the GBPs into the bacterial membranes via lipid anchors results in a partial weakening of membrane integrity, thus allowing caspase-4 to bind the Lipid A moiety of LPS. Additional studies aimed at determining the structure and composition of the GBP complex will be necessary to understand how it promotes caspase-4 recruitment and activation.While it is not yet clear if the GBP1-LPS interaction also requires structural determinants in LPS, the ability of GBP1 to recognize negatively charged pathogen-derived molecules might extend its function beyond the recognition of Gram-negative bacteria. Indeed, GBP1 is known to be recruited to both the surface of cytosolic parasites, such as T. gondii, as well as to the membrane of the T. gondii parasitophorous vacuole, and to assemble a GBP coat in a similar manner25,28. Interestingly, in this case the GBP coat does not result in recruitment of caspase-4 (in line with the fact that parasites do not feature LPS), but in the induction of caspase-8-dependent apoptosis40. It is thus likely that parasites or their vacuoles feature molecules with similar chemical properties as the LPS Lipid A and core polysaccharides in their cell membrane or the membrane of the parasitophorous vacuole. Identification of additional ligands that bind GBP1 or other GBP family member will enhance our understanding of host innate immunity and establish new paradigms for pattern recognition.The human-specific mechanism reported here is in contrast to previous models that proposed that, in mice, GBPs promote non-canonical inflammasome activation by facilitating vacuolar escape and inducing bacterial membrane destruction10,41. While it is unlikely that mouse GBPs function fundamentally differently from human GBPs in the mechanism by which they recognize pathogens, it is possible that the existence of IRGs in mouse enhances their downstream effector functions. The lysis of bacteria and vacuoles reported in mouse cells (but not detected in human cells), is most likely a consequence of GBP-mediated recruitment of IRGs, such as Irgb10 that is reported to have antimicrobial properties11. Thus, in addition to recruiting caspase-11 directly in analogy to human GBPs, mouse GBPs might also mediate access to LPS and LPS liberation through the membranolytic activity of IRGs.To our knowledge, our study is the first to report a ligand of GBP1 and to characterize the mode of this interaction. While our findings still need to be validated in primary human cells, it nevertheless provides the first evidence that GBP1 and possibly other GBPs function as direct innate immune receptors for pathogen-associated molecular patterns, expanding the ever-increasing repertoire of cytosolic innate immune defense pathways.MethodsBacterial strains and mammalian cell cultureAll bacteria were grown at 37 °C in an orbital shaker. Salmonella enterica serovar Typhimurium strain SL1344 was grown in lysogeny broth (LB) medium supplemented with 10 g/L NaCl and streptomycin (50 µg/mL). Salmonella expressing dsRed (Salmonella-dsRed) was grown by supplementing LB medium with ampicillin (50 µg/mL). Salmonella enterica serovar Typhimurium strain 4/74 and their isogenic ΔwaaG or ΔwaaL mutants were a kind gift from Jay Hinton (University of Liverpool, Liverpool). Shigella flexneri M90T expressing the adhesin AfaI was a kind gift from Jost Enninga (Institut Pasteur, Paris) and was grown in tryptic soy broth (TSB) supplemented with ampicillin (50 µg/mL). Listeria monocytogenes strain EGD was a kind gift from Pascale Cossart (Institut Pasteur, Paris) and was grown in brain-heart infusion (BHI) medium. Unless stated otherwise, the HeLa clone CCL-2 from ATCC was used. Human epithelial HT-29 and HeLa (CCL-2 or Kyoto clones) cells were cultured in DMEM (Gibco) supplemented with 10% Fetal Calf Serum (FCS). Caco-2/TC-7 were cultured in DMEM supplemented with 20% FCS. HT-29 and Caco-2 cells were a kind gift from Shaynoor Dramsi (Institut Pasteur, Paris). THP-1 and U937 cells were cultured in RPMI supplemented with 10% FCS. HBEC3-KT cells were obtained from ATCC and were grown in Bronchial/Tracheal Epithelial Cell Growth Medium (Cell Applications, Inc.). HaCaT cells were obtained from CLS Cell Lines Service GmbH, and were grown in DMEM supplemented 10% FCS. HIEC-6 cells were obtained from ATCC and grown in Opti-MEM (Gibco) supplemented with 4% FCS, 10 mM Glutamine and 10 ng/mL of epidermal growth factor (EGF). Human primary monocyte-derived macrophages (hMDMs) were purified from buffy-coat obtained from the Swiss Red-Cross and purified and cultured as described previously42. All cells were grown at 37 °C, 5% CO2.Generation of CRISPR/Cas9 knockout cell linesKnock-out HeLa cells lines were generated using the Alt-R CRISPR-Cas9 System (Integrated DNA Technologies, IDT), by using a mix of a sequence-specific CRISPR RNA (crRNA), a conserved, transactivating crRNA (tracrRNA) and recombinant Alt-R S. pyogenes Cas9 (IDT). crRNA and tracrRNA were mixed to 1 µM, heated 5 min at 95 °C and cooled to room temperature. 1 µM Alt-R Cas9 was mixed and incubated at room temperature for 5 min. Lipofectamine RNAiMax transfection reagent (Invitrogen) was then added and the mixture was incubated for 20 min at room temperature. 40,000 cells/well were reversely transfected with the previous mixture in 96-well plates, to achieve a concentration of 10 nM ribonucleoprotein complex. After incubation for 2 days at 37 °C, 5% CO2, single clones were generated by serial dilutions and the desired gene knockouts were screened by performing the T7 endonuclease I assay, verified by sequencing of the PCR fragments and confirmed by western blotting. The following crRNAs were used: AGGGATTCCAACACCTTAAG (for CASP4), CCACGTACACGTTGTCCCCG (for GSDMD) and GAACACTAATGGGCGACTGA (for GBP1).Plasmids, siRNAs, and cell transfectionPlasmids expressing N-terminal fluorescently tagged GBPs were generated by inserting the GBPs coding sequences at the XhoI/HindIII sites of pEGFP-C1 and pmCherry-C1 (Clontech). pmiRFP703-GBP1 was generated by using the pEGFP-C1 plasmid and replacing eGFP by miRFP703 (addgene 8000143 was used as PCR amplification template) at the NheI/BglII sites, and then inserting the GBP1 coding sequence at the XhoI/HindIII sites. Caspase-4-eGFP was generated by fusing the amplified PCR products of caspase-4 and eGFP and inserting the coding sequence into the NheI/HindIII sites of pEGFP-C1. The pAIP vectors expressing HA-GBP1 or HA-GBP2 were a kind gift from T. Henry (CIRI, Lyon) and were used to generate HA-tagged GBPs, by replacing GBP1 by the GBP3 or −4 coding sequences at the EcoRI sites. The bicistronic plasmids encoding FLAG-GBP3 + HA-GBP4 were generated by inserting GBP3 or GBP4 at the NotI/PmeI sites of the pBud-EGFP vector (addgene 23027)44. Doxycycline-inducible eGFP-GBP1, −2, −3, −4 were generated by amplifying eGFP-GBPs generated above by PCR and inserting the coding sequences at the BamHI site of the pLVX-Puro vector (Clontech). All cloning was performed using In-Fusion cloning technology (Clontech) and plasmids were verified by sequencing. Plasmids encoding eGFP or mOrange tagged galectin-3 were a kind gift from Jost Enninga30,45. HeLa cells were either plated onto 8-well µ-slides (Ibidi) at a density of 1.5 × 104 cells/well for live imaging, onto 24-well plates containing glass coverslips at a density of 1.0 × 105 cells/well, onto 96-well glass bottom plates (Greiner) or onto 96-well plates (Eppendorf) at a density of 8.0 × 104 cells/well 24 h before transfection. Cells were then transfected with one, two or three expression plasmids using X-tremeGENE 9 DNA transfection reagent (Roche) for 16–48 h, according to the manufacturer’s instructions. A list of the plasmids and primers used in this study is provided in Supplementary Table 1. For siRNA knock-down experiments, cells were seeded onto 96-well plates at a density of 9.0 × 104 cells/well and on the following day transfected with 3 pmol (25 nM) Stealth RNAiTM siRNAs (Thermo Fisher Scientific) using Lipofectamine RNAiMax (a list of the siRNAs used in this study is shown in Supplementary Table 2). After 8 h, cells were incubated with IFNγ for an additional 16 h and experiments were then performed.Infection assays and transfection of cells with LPSWhen indicated, cells were primed with 10 ng/mL human IFNγ (Peprotech) for 16 hours. Overnight Salmonella cultures were sub-cultured 1/50 and grown until late exponential/early stationary phase (OD600 = 1.5–1.8). Overnight Shigella or Listeria cultures were sub-cultured 1/100 and grown until mid-exponential phase (OD600 = 0.5–0.7). Before infection, bacteria were collected by centrifugation, gently washed and resuspended in DMEM. Salmonella was added to HeLa cells in 96-well plates (approximately 50,000 cells per well) at a multiplicity of infection (MOI) of 50 and incubated for 30 min at 37 °C. For infection with Shigella or Listeria, bacteria were added to cells at a MOI of 20 and incubated for 30 min at 37 °C. Non-internalized bacteria were removed by three washes with warm DMEM and cells were incubated with DMEM containing 100 µg/mL gentamicin for 1 h to kill extracellular bacteria. Medium was then changed to DMEM containing 10 µg/mL gentamicin and 10% FCS for the remained of the experiment. At the desired time points p.i., cells were either processed for LDH release, enumeration of intracellular bacteria or fixed for immunofluorescence assays. To enumerate intracellular bacteria, infected cells were gently washed with PBS and lysed with water containing 0.2% Triton X-100. Bacteria were then serially diluted and plated onto LB agar. To quantify the percentage of cytosolic Salmonella in the total population, we used a CHQ resistance assay. Briefly, infected cells were incubated with 200 µg/mL CHQ (Sigma-Aldrich) and gentamicin for 1 h (CHQ-resistant bacteria) or with gentamicin only (total bacteria). Cells were washed, lysed and bacteria were plated as described above. The percentage of cytosolic bacteria was calculated by the ratio of (CHQ + gentamicinresistant/gentamicinresistant). U937 and THP-1 cells were seeded and differentiated with 100 ng/mL PMA for 48 h, followed by a 24 h resting period.Transfection of cells with smooth LPS from E. coli O111:B4 (Invivogen) or Salmonella (Sigma, L6143) was done at a concentration of 2.5 µg/50,000 cells, or 2.5 µg/80,000 cells (for THP-1 or U937), using Lipofectamine 2000 (Invitrogen). Briefly, LPS was diluted in Opti-MEM and incubated with Lipofectamine 2000 (1.0 µl/50,000 cells) for 20 min at room temperature. 75 µl of Opti-MEM was added to cells on 96-well plates and then 75 µl of LPS mixture was added on top. Plates were centrifuged for 5 min at 211 × g and then incubated at 37 °C for the indicated time points. For electroporation of HeLa or HBEC3-KT cells, the Neon Transfection System (Life Technologies) was used. Briefly, naive or IFNγ-primed cells were harvested, resuspended in resuspension buffer T and electroporated with LPS from E. coli O111:B4 at a concentration of 300 ng/50,0000 cells, using electrolytic buffer E and 1 pulse of 1300 V for 20 ms. Cells were then added to 200 µl pre-warmed Opti-MEM in a 96-well plate, centrifuged for 5 min at 211 × g and then incubated at 37 °C. Mock electroporation and non-electroporated cells were used as controls.Microscopy, time-lapse imaging, and image analysisInfected cells were washed once with PBS and fixed in 4% PFA for 20 min. Cells were then washed three times, permeabilized with 0.05% saponin and blocked with 1% BSA. Coverslips were incubated with antibodies when indicated and with Hoechst (1:1000) in PBS, and then mounted in ProLong Gold Antifade (Life Technologies) for confocal microscopy. Samples were imaged with a Zeiss LSM800 confocal laser scanning microscope using a 63×/1.4 NA oil objective, by acquiring Z-stacks of 300 nm step size. For live imaging, infection assays were performed in EM buffer (120 mM NaCl, 7 mM KCl, 1.8 mM CaCl2, 0.8 mM MgCl2, 5 mM glucose, 25 mM HEPES, pH 7.3). Cells were infected for 10 min as previously described, extracellular bacteria were removed by washing with warm EM buffer, and time-lapse microscopy of living cells was performed at 37 °C using a motorized xyz stage with autofocus. Super-resolution was performed using the Zeiss LSM800 Airyscan super-resolution system using the same objective and super-resolution images were calculated using the Zeiss ZEN software. Data were further analyzed and processed using FiJi software, and all fluorescence derived images shown here correspond to maximum 3D projections.LDH release, PI uptake, IL-18 release, and western blottingCell death was quantified by measuring LDH release to the supernatant, using the LDH cytotoxicity detection kit (Takara, Clontech). To normalize for spontaneous cell lysis, the percentage of cell death was calculated as follows: (LDHsample – LDHnegative control)/(LDHpositive control – LDHnegative control) × 100. PI influx measurement was performed as previously described46. The levels of IL-18 were measured by ELISA (R&D Sytems), according to the manufacturer’s instructions. For western blotting analysis, cell lysates were prepared and supernatants were precipitated. Mouse anti-caspase-4 4B9 (ADI-AAM-114-E, Enzo Life Sciences, 1:750), rabbit anti-GSDMD (ab210070, abcam, 1:1000), rabbit anti-GBP1 (ab121039, abcam, 1:1000), mouse anti-GAPDH (AM4300, Thermo Scientific, 1:1000), mouse anti-V5 (R960-25, Thermo Scientific, 1:2000), mouse anti-GFP (632381, Clonetech, clone JL-8, 1:5000), mouse anti-HA (ENZ-ABS-118-0200, Enzo Life Sciences, 1:2000), mouse anti-tubulin (ab40742, Abcam, 1:2000) were used and detected with horseradish peroxidase-conjugated secondary antibodies (1:5000, Southern Biotech).Active caspase pull-downHeLa cells were seeded onto 6-well plates and primed for 16 h with 10 ng/mL human IFNγ. Approximately 3 × 106 cells were then treated with 10 µM of biotin-VAD-fmk and transfected with 20 µg of E. coli LPS for 3 hours. Cells were lysed and incubated overnight with 20 µl of pre-washed streptavidin magnetic beads (Thermo Scientific). The beads were washed as described elsewhere47 and streptavidin-bound and left-over fractions (unbound) were analyzed on a 12% acrylamide gel and blotted against caspase-4.Streptavidin pull-down assaysApproximately 2 × 106 cells were collected and lysed in pull-down buffer (50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 5 mM EDTA, 1% NP40, 0.05% Na-deoxycolate and complete protease inhibitors). 700 µg of protein (as determined by BCA assay (Thermo Scientific) from the total cell lysate was incubated with 2 µg biotinylated LPS or with biotinylated Pam3CSK4 (Invivogen) at room temperature for two hours with rocking. After incubation, 20 µl of pre-washed streptavidin magnetic beads (Thermo Scientific) were added and incubated for 1 h at room temperature with constant rocking. The beads were washed three times in PBS with 0.05% Tween-20 and once with PBS and the precipitates were eluted in equal volumes of SDS-PAGE reducing sample buffer followed by western blotting analysis. 5% the of initial cell lysate (input) and equal volumes of pull-down were analyzed. For GFP-GBP pulldown assay, cells were lysed at a concentration of 20 × 106 cells/mL of lysis buffer (50 mM Tris pH 7.4, 150 mM NaCl, 10 mM MgCl2, 5 mM GTP, 300 μM AlF, 100 μg/mL digitonin (Sigma) and allowed to lyse on ice for 15 min. The cells were then spun 15 min at 6000 × g, 4 °C. The soluble extract was then incubated with 2 µg LPS/million cells equivalent and incubated at RT for 2 h with rotation. After incubation, streptavidin magnetic beads were added to the mix and incubated for an additional hour with rotation. The beads were than washed three times (30 min wash) with lysis buffer. The beads where than resuspended in reducing western blot loading buffer before being analyzed by immunoblotting.Quantitative PCR (qPCR)Total mRNA was extracted from HeLa cells using the RNeasy Mini kit (Qiagen) and up to 400 ng were reversed transcribed into cDNA using the Verso cDNA Synthesis kit (Thermo Fisher Scientific). Gene expression levels were quantified by qPCR using a LightCycler 480 (Roche) and LightCycler 480 SYBR Green I Master (Roche), according to standard protocols, by normalizing each sample to the respective levels of the housekeeping mRNA HPRT. The list of primers used for qPCR is shown in Supplementary Table 3.Purification of recombinant proteinsFull-length human GBP1 was cloned in pET-28a to generate an N-terminally His-tagged hGBP1 construct. pET-28a-hGBP1 was transformed into CleanColi BL21 (Lucigen), and the bacteria were grown in 2xYT medium until an OD600 of 0.5-0.7. Protein expression was then induced at 30 °C for 5 h with 0.2 mM IPTG. The bacterial pellet was resuspended in resuspension buffer (50 mM Tris pH 7.4, 150 mM NaCl, 1% Tween 20) and frozen at −80 °C until purification. For most assay, protein was freshly purified on a Ni-NTA affinity column using standard protocols48. Protein yield was quantified using Beer-Lambert law. After purification on a Ni-NTA columns, GBP1 was further purified on a size exclusion chromatography column (Superdex 200 10/30 GL, GE Healthcare) in running buffer (50 mM Tris pH 7.4, 150 mM NaCl) and concentrated using Amicon Ultra4 10 kDa (Millipore).Size exclusion chromatography of GBP1 and LPSFreshly purified GBP1 (1 µM) was incubated on ice alone or with a two-fold molar excess of LPS or LPS-derivatives for 5 h. Ultrapure O11:B4 E.coli LPS (Invivogen), Salmonella Typhimurium Smooth LPS (Enzo Life Science), Rhodobacter sphaeroides ultrapure LPS (Invivogen), E. coli F585 diphosphoryl Lipid A (Sigma-Aldrich), Salmonella minnesota 595 Lipid A (Invivogen), synthetic monophosphorylated Lipid A (Invivogen), E.coli EH100 LPS Ra mutant (Sigma-Aldrich), E.coli J5 LPS Rc mutant (Sigma-Aldrich), E.coli F583 LPS Rd mutant (Sigma-Aldrich), E. coli R515 LPS Re mutant (Enzo Life Science) was used. After incubation, GBP1 alone or GBP-LPS incubations were injected into a Superdex 200 10/30 GL column and run in running buffer for 1 column volume. Individual fractions (500 µL), were collected, precipitated with methanol and chloroform49, separated on a 12% acrylamide gel and analyzed by immunoblotting using an antibody against His6 tag. Experimental molecular weights of the peaks were approximated using a gel filtration standard (1511901; Bio-Rad). Where indicated, LPS was pre-incubated with 5 mM CaCl2 for 5 min on ice before being added to GBP1, as indicated above, or LPS was pre-incubated with polymyxin B (10 µg/mL) for 5 min at room temperature.Surface plasmon resonance (SPR)SPR measurements were performed on the Biacore T200 (GE Healthcare Life Sciences). GPB1 was immobilized on a CM5 sensor chip (GE Healthcare) using the amine coupling procedure (immobilization response was 310 RU or 0.31 ng/mm2). Then it was equilibrated in PBS buffer (pH 7.2; Gibco, Life Sciences), followed by the injection of the increasing concentrations of LPS (47, 94, 188, 375, 750, 1500 nM) into the flow channels. In the reverse experiment, when immobilizing LPS on the CM5 sensor chip, increasing concentrations of GBP1 (21.5, 43, 86, 172, 343, 688, 1375, 2750 nM) were used. Data were analyzed using BiacoreT200 Evaluation software 3.0. An equilibrium analysis was done using Langmuir isotherm fit with one equilibrium dissociation constant (KD). The best fit for the Kinetic curves was obtained with the two-state-reaction model that assume a possible structural re-arrangement after the initial binding.Microscale thermophoresis (MST)MTS was performed on 50 nM of FITC-labeled E. coli LPS (Sigma-Aldrich) using freshly purified hGBP1 expressed recombinantly (as described above) or BSA as a control. Serial dilutions of GBP1 or BSA were analyzed in assay buffer (50 mM Tris pH 7.4, 150 mM NaCl). Experiment was performed on a Nanotemper Monolith NT.115 microscale electrophoresis instrument with medium MST power. Data were fitted to a 1:1 binding model with the MO.Affinity Analysis software.GTPase activity assayGTPase activity assay was performed using the GTPase-GloTM kit (Promega) according to the manufacturer use. Recombinant human GBP1 (500 nM) was incubated with 5 µM LPS in GEF buffer (Promega) for 30 min at room temperature before assessing GTP hydrolysis. Luminescence values were normalized to a no-GBP1 control.Data analysisData analysis was performed using the following software: Gen5, GraphPad Prism v8 and Microsoft Excel. Statistical significances are referred as *, ** or *** for P-values <0.05, <0.01 or <0.001, respectively. For comparison of two groups, a two-tailed t-test was used, whereas for comparison of three or more groups P-values were determined using the two-way analysis of variance for multiple comparisons.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Information
Supplementary Movie 1
Supplementary Movie 2
Supplementary Movie 3
Supplementary Movie 4
Supplementary Movie 5
Supplementary Movie 6
Supplementary Movie 7
Supplementary Movie 8
Supplementary Movie 9
Supplementary Movie 10
Supplementary Movie 11
Supplementary Movie 12
Supplementary Movie 13
Supplementary Movie 14
Supplementary Movie 15
Reporting Summary | nature communications | [
"Article"
] | [
"Infection",
"Inflammasome",
"Innate immunity",
"Pattern recognition receptors"
] | lipopolysaccharide (LPS central to defense against Gram-negative infections pathogenesis sepsis Extracellular LPS sensed by Toll-like receptor 4 (TLR4) induces production cytokines MyD88 TRIF Cytosolic LPS detected by non-canonical inflammasome controls activation caspase-4/−5 humans caspase-11 in caspases cleave pore cell death effector gasdermin D) pyroptosis cytokine release activation other caspases requires platforms no comparable platform for caspase-4/−5 −11 activation new pattern recognition caspase-4/−11 sensor executor without additional adaptor proteins caspases-4/−11 binds hydrophobic lipid LPS through oligomerization LPS hydrophobic present within bacterial membranes cytosolic LPS sensing could require accessory factors LPS-binding protein (LBP) cluster differentiation 14 TLR4 signaling LBP binds LPS membrane promotes transfer LPS CD14 delivers LPS to MD-2/TLR4Caspase-11 activation in mouse macrophages transfected LPS infected Gram-negative bacteria requires interferon-inducible GTPases GBPs IRGs GTPases upregulated after IFN priming essential for cell immunity against viruses bacteria macrophages GBPs Irgb10 target intracellular Gram-negative bacteria Salmonella enterica Typhimurium Francisella novicida Escherichia coli recruitment correlated with bacterial lysis activation caspase-11 GBPs recruit Irgb10 bacterial membranes-dependent membranolytic activity pathogen liberates LPS non-canonical humans lack IRG family GBPs required for LPS-induced caspase-4 activation model needs in human cells14 IFNγ priming induction GBPs necessary for caspase-4 activation in human epithelial cells monocytes/macrophages during infection Salmonella after cytosolic LPS delivery transfection human GBP1 targets cytosolic Salmonella after cytosol GBP1 initiates recruitment GBP2-4 GBP coat cytosolic bacteriaGBP coat bacteriolysis initiates recruitment activation caspase-4 cytosolic bacteria Human GBPs roles GBP1 GBP4 recruit caspase-4 GBP3 required for caspase-4 activation GBP1 recognizes cytosol-exposed bacteria LPS associates with GBP1 pyroptotic cells recombinant GBP1 binds LPS high affinity GBP1 associates with LPS micelles high-molecular weight complex incubation association via electrostatic interactions negative charges LPS mutagenesis GBP1 positively charged residues necessary for LPS binding recruitment GBP1 acts cytosolic LPS sensor detects targets LPS-containing membranes Gram-negative bacteria caspase-4 recruitment activation-induced caspase-4 activation requires IFNγ infected naive IFNγ-primed HeLa cells with Salmonella HeLa cells express TLR4 MD-2 not responsive to bacterial primed cells with IFNγ intestinal immunity against Salmonella replicated in naive HeLa restricted in IFNγ-primed cells invasionIFNγ-primed HeLa cells lytic death pyroptosis plasma membrane swelling ballooning nuclear condensation (Fig. 1b released mature IL-18 epithelial cells Salmonella cytosol hour hypothesized Salmonella activate non-canonical inflammasome mouse infected IFNγ-primed wild-type CASP4–/– GSDMD–/– HeLa 1g). Deletion CASP4 GSDMD alter bacterial invasion abrogated Salmonella-induced IFNγ-dependent cell death (Fig. 1c Salmonella infection HeLa activates non-canonical inflammasome IFNγ bacterial replication increased IFNγ priming partially reduced intracellular bacterial replication in CASP4– GSDMD– HeLa cell death not only IFNγ restricts bacterial growth confirmed Salmonella expressing PuhpT-GFP reporter cytosolic replication hexose transporter promoter exogenous glucose-6-phosphate21 cytosol Figchloroquine (CHQ)-resistance assay antimicrobial agent bactericidal levels endocytic IFNγ priming restricted cytosolic Salmonella reducing hyper-replication IFNγ controls caspase-4 GSDMD-dependent cytosolic Salmonella replication cell pyroptosis minor cell death 1IFNγ priming LPS-induced caspase-4 activation human epithelial cells Intracellular bacterial fold-replication release LDH IFNγ-primed wild-type CASP4– GSDMD– HeLa cells 1 or 6-h post-infection Salmonella Cells infected 30 min washed gentamicin added kill bacteria supernatant collected release LDH cells lysed viable intracellular bacteria determined colony units bacterial-replication calculated relative to 1 h p.i Percentage CHQ-resistant cytosolic Salmonella IFNγ-primed HeLa 1.5 h p.i Cells infected 30 min treated gentamicin ± CHQ 1 h before lysed counted CFUspercentage cytosolic bacteria calculated ratio (CHQ + gentamicinresistant Release LDH naive IFNγ-primed cells 5 h after transfection LPS (2.5 μg/50,000 3–4 h after electroporation LPS (300 ng/50,0000 blot analysis full cleaved caspase-4 supernatants cell lysates IFNγ cells transfection E. coli LPS μg/50,000 Streptavidin biotin-conjugated LPS caspase-4 lysates HBEC3-KT Cells 6-well plates transfected LPS-biotin (10 μg untransfected biotinylated substrate pulled streptavidin magnetic beads eluted SDS-PAGE buffer Streptavidin-bound -unbound fractions analyzed immunoblotting caspase-4 Graphs mean ± SD data pooled two six experiments P < 0.001 two-tailed t-test determined IFNγ access Salmonella LPS cytosol bacterial resistance CHQ 1.5 h post-infection IFNγ priming induce Salmonella escape cytosol controlled detection LPS transfected cells ultrapure E. coli Salmonella LPSSalmonella infection LPS transfection caused cell death in IFNγ-primed HeLa abrogated by deletion CASP4 GSDMD IFNγ-priming required for pyroptosis IL-18 release after LPS transfection LPS electroporation in human small intestinal epithelial cells caspase-4 activation requires IFNγ transfected LPS into naive IFNγ-primed cells pulled down active caspase-4 using-caspase probe Active caspase-4 pulled down cells primed with IFNγ transfected with LPS LPS transfection induced caspase-4 GSDMD cleavage in IFNγ-primed cells IFNγ-priming impact caspase-4 expression not induced by IFNγ IFNγ caspase-4 activation prepared lysates from HeLa HBEC3-KT cells transfected with biotinylated LPS pulled down LPS-interacting proteins with streptavidin-coupled beads caspase-4 pulled-down with LPS in IFNγ-primed not naive cellsfindings suggest human epithelial cells IFNγ-induced proteins required for LPS-induced caspase-4 activation.GBP1 required for non-canonical inflammasome caspase-11 activation requires IFN-induced GTPases speculated human GBPs necessary for LPS-induced caspase-4 activation GBPs expression in HeLa cells upregulated by IFNγ priming interference silencing GBPs reduction LDH release cells lacking GBP1 after Salmonella infection LPS transfection generated GBP1– HeLa cells by CRISPR-Cas9 genome targeting GBP1-deficiency abrogated LDH release after Salmonella infection LPS transfection without altering bacterial entry GBP1-deficient cells unable activate caspase-4 LPS transfection Salmonella infection GBP1 required for LDH release when LPS delivered electroporation GBP1-deficiency partial loss of IFNγ-dependent restriction cytosolic Salmonella replication GBP1 knock-down in HBEC3-KT HaCaT cells reduced LDH release upon LPS transfectionFig. GBP1 LPS cell death human epithelial cell lines Salmonella LPS caspase-4 activation pyroptosis Immunoblots GBP1 caspase-4 GAPDH cell lysates IFNγ wild-type GBP1–/– HeLa Release LDH Salmonella infection 5 h transfection E. LPS (2.5 μg / 50,000 cells Immunoblots full length cleaved (p32) caspase-4 supernatants cell lysates-type GBP1– HeLa transfection LPS 5 h Salmonella infection Release LDH IFNγ-type GBP1– HeLa 3 h after electroporation LPS (300 ng/50,000 Release LDH IFNγ-primed HBEC3-KT HaCaT cells treated non control siRNA CASP4 GSDMD GBP1 after E. LPS transfection Cells treated 24 h transfected LPS (2.5 μg / 50,000 cells 5 h Graphs mean ± SD data pooled four experiments < 0.001 two-tailed t-testGBP1 targets Salmonella role caspase-4 activation expressed GBPs naive primed cells IFNγ-primed HeLa GBP1 −2 −3 −4 coated 20–30% intracellular Salmonella at 1 h p. few positive GBP5 −6 −7 (Fig. 3a Recruitment GBP2 −3 −4 dependent IFNγ priming associated Salmonella naive cells eGFP-GBP1 associated Salmonella naive cells lower levels. 3GBP1 targets Salmonella controls recruitment GBP2-4 confocal microscopy-primed HeLa expressing eGFP-GBP1-7 infected Salmonella-dsRed 1 h DNA stained Hoechst confocal images scale bars 10 μm Percentage intracellular Salmonella positive eGFP-GBP1-7-primed HeLa 1 h p 200–300 bacteria counted per coverslip microscopy IFNγ-primed HeLa expressing eGFP-GBP2-4 infected Salmonella-dsRed 1 h images scale bar 10 μm Schematic representation wild-type GBP1 ΔCaaX mutantFluorescence confocal microscopy GBP1–/– HeLa expressing HA-GBP1wt-GBP1ΔCaaX-expressing mCherry-GBP1wt-GBP1wt infected Salmonella 1 h HA-tagged GBP1 visualized immunostaining anti-HA antibody confocal images scale bars 5 μm percentage HA-GBP1 positive bacteria 100 bacteria per coverslip Graphs show mean SD data pooled two independent experiments.GBPs homo- hetero-oligomerize forming coat recruited intracellular pathogens28 hypothesized GBP1 target Salmonella independently GBP2-4 required GBP1 recruitment GBP2, −3 −4 Salmonella abrogated IFNγ-primed GBP1–/– HeLa cells GBP1 co-expression cells recruitment GBP2-4 intracellular Salmonella similar IFNγ-primed cells GBP1 targeting intracellular Salmonella observed human small intestinal epithelial cells time-lapse confocal microscopy infected cells co-expressing GBP1 GBP2 GBP1 GBP2 recruited simultaneously bacteriaMovie 3) GBP3 GBP4 recruited bacteria minutes after GBP1 recruitment Fig. 4f g Movies 4 5) data show GBP1 first intracellular Salmonella orchestrates recruitment human GBP family members GBP2-4 analyzed GBP1 mutants (Fig. 3d Fig 5a b GTP hydrolysis triple-arginine polybasic motif C-terminal CaaX box-dependent farnesylation required for GBP1 recruitment/accumulation Salmonella (Fig. 3e Fig. 5c). HA-GBP1ΔCaaX not recruited Salmonella co-expressed with mCherry-GBP1wt (Fig. GBP1 oligomers formed if farnesylated membranes target cytosolic Salmonella seconds after SCV mouse GBPs associated with vacuolar human GBPs associate with SCV cytosolic Salmonella galectin-3 marker ruptured GBP1-4 found galectin-3-positive bacteria (Fig. analysis targeted cytosol-exposed part not galectin-3-positive ruptured vacuoles. GBPs co-localize with LAMP1 Fig. marker SCVreduction cytosolic galectin-3-positive Salmonella in GBP1–/– cells GBP1-4 promote escape Salmonella bacteria cytosolic entry GBP1 associated with Shigella not cytosolic Listeria monocytogenes 4GBP1 targets cytosolic Salmonella required for caspase-4 recruitment microscopy IFNγ-primed wild-type HeLa-expressing galectin-3-eGFP-GBP1-7 infected Salmonella 1 h Percentage CHQ-resistant cytosolic Salmonella HeLa at 1.5 h p Cells infected 30 min treated gentamicin ± CHQ 1 h CFUs percentage cytosolic bacteria calculated ratio (CHQ + gentamicinresistant Fluorescence microscopy IFNγ-primed HeLa caspase-4-eGFP infected Salmonella 1 h Percentage caspase-4-eGFP positive Salmonella at 1 h p.i. 100–200 bacteria per coverslip not detectedFluorescence microscopy IFNγ-primed HeLa galectin-3-mOrange caspase-4-eGFP infected Salmonella 1 h Time-lapse fluorescence microscopy IFNγ HeLa galectin-3 infected Salmonella fluorescence intensities galectin-3-mOrange caspase-4-eGFP time quantified region interest caspase-4 galectin-3 recruitment bacterium intensity signals aligned time point onset galectin-3 recruitment zero mean SD six events plotted Time-lapse fluorescence microscopy IFNγ-primed HeLa caspase-4-eGFP mCherry-GBP1 infected Salmonella Images 60 s Percentage caspase-4-eGFP positive Salmonella 1 h IFNγ-primed wild GBP1– HeLa 100–200 bacteria counted per coverslip Representative confocal images scale bars 1 μm 5 μm 10 μm Graphs mean ± SD data pooled two three experiments ***P < 0.001 two-tailed t-test time-lapse confocal microscopy SCV membrane rupture GBP recruitment pyroptosisSalmonella-infected cells SCV rupture galectin-3 recruitment followed rapid GBP1 less than 30 seconds galectin-3 appearance Fig 6d 6) GBP1 recruitment started region bacterium cytosol formed coat around bacteria recruitment to cytosolic Salmonella followed by pyroptotic cell death 6e 7) majority pyroptotic cells featured GBP1-positive Salmonella mouse macrophages GBP targeting mediates recruitment Irgb10 correlates lysis caspase-11 human GBPs lysed cytosolic Salmonella monitored GBP recruitment in GSDMD–/– cells infected Salmonella-dsRed imaging bacteria targeted GBP1 cell cytosol no lysis viability loss Fig. 7a–d Movies 8–11) GBP1 targets surface Salmonella seconds rupture SCV membrane escape cytosol followed rapid caspase-4-dependent pyroptosis.GBPs control caspase-4 recruitment to cytosolic bacteria presumed GBPs control LPS-dependent caspase-4 activation caspase-4-eGFP recruited on Salmonella covering entire bacteriumCaspase-4 recruitment Salmonella required IFNγ priming not detectable in naive cells (Fig. 4c 8a despite high expression levels Time-lapse microscopy of caspase-4-eGFP-expressing IFNγ-primed HeLa infected Salmonella-dsRed diffused in cytosol recruited intracellular Salmonella within minutes followed by pyroptotic cell death majority Fig 8b c few cells recruit caspase-4 initiate pyroptosis 8d 13).Recruitment caspase-4 Salmonella correlated with SCV lysis-positive bacteria galectin-3-positive ruptured vacuoles (Fig. 4e caspase-4 co-localize with LAMP1-positive SCVs targeted cytosol-exposed part bacterium not lysed vacuole Super-resolution microscopy confirmed caspase-4 accumulated on bacterial surface not ruptured SCVs time-lapse microscopy cells caspase-4 recruited to Salmonella 5–10 min after first galectin-3 signal targets cytosolic Salmonella after SCV ruptureGBP1 targeted cytosolic Salmonella required for caspase-4 activation hypothesized GBPs control recruitment caspase-4 bacteria Time-lapse microscopy HeLa cells-GBP1 GBP1 preceded caspase-4 minutes followed by pyroptosis (Fig. 4i consistent with faster recruitment GBP1 SCV rupture slower caspase-4 observed reduction in caspase-4 recruitment to Salmonella in GBP1–/– HeLa GBP1 recruited to bacteria in CASP4–/– cells GBP1-deficient cells activate caspase-4 GBP1 initiates caspase-4 bacterial surface.GBP1/3/4 for LPS-induced caspase-4 functions GBP1-4 in caspase-4 recruitment activation GBP1–/– HeLa recruit GBP2-4 caspase-4 to cytosolic bacteria (Figs. 3c impossible to determine GBP1 controls recruitment co-expressed caspase-4-eGFP mCherry-GBP1 in-primed cells determined caspase targeting to cytosolic SalmonellaGBP1 targeted cytosolic Salmonella IFNγ caspase-4 recruited primed cells other GBPs unknown IFNγ factor necessary GBP1-dependent caspase-4 recruitment co-expressed caspase-4 mCherry-GBP1 with doxycycline eGFP GBP2 GBP3 GBP4 naive HeLa cells visualized recruitment confocal microscopy 10a GBP1 restore caspase-4 recruitment cytosolic Salmonella co-expression GBP1 GBP4 GBP3 recruitment caspase-4 cytosolic bacteria 5b vectors excluded vector-biased caspase-4 recruitment GBP1 controls recruitment GBP4 partially GBP3 5GBP1/4 recruit caspase-4 Salmonella GBP3 activate non inflammasome naive human epithelial cells confocal microscopy-primed HeLa co-expressing caspase-4-eGFP mCherry-GBP1 infected Salmonella 1 h DNA stained Hoechst confocal images scale bar 5 μmFluorescence microscopy HeLa cells-expressing mCherry-GBP1 eGFP-GBP1 caspase-4-V5 infected Salmonella 1 h DNA stained Hoechst eGFP-GBPs expressed 1 μg/mL Dox 16 h Caspase-4-V5 immunostaining anti-V5 antibody confocal images scale bar 10 μm Percentage caspase-4-V5 positive Salmonella 1 h mCherry-GBP1-positive bacteria 50 bacteria coverslip Percentage cell death HeLa cells-expressing mCherry-GBP1 Dox eGFP-GBP1 FLAG-GBP3 HA-GBP4 expressed transfected plasmids 24 h eGFP-GBPs induced 16 h 1 μg/mL Dox eGFP 3 h transfected E. LPS (2.5 μg/50,000 cells 6 h cell death values normalized IFNγ-primed HeLa 100% naive cells co-expressing mCherry-GBP1 eGFP 0% Graphs mean ± SD data pooled two three **P < 0.01 ***P < 0.001 one-way ANOVAasked if expression GBPs in cells restores caspase-4-dependent pyroptosis expression GBP1-7 sufficient LPS-induced pyroptosis LPS transfection induced elevated LDH release cells co-expressing GBP1/3/4 or GBP1/2/3/4 5d co-expression GBP1/3 restored LPS transfection-induced pyroptosis GBP1/4 no effect caspase-4 recruitment on Salmonella 5b GBP1 drives caspase-4 recruitment Salmonella GBP4 GBP3 necessary caspase-4 activation 5b experiments indicate complex GBP1, −3 −4 promotes caspase-4 recruitment activation following LPS detection without other IFNγ-induced genes.GBP1 binds GBPs recruitment to cytosolic Gram-negative bacteria OMVs transfected LPS tested if bind LPS biotin-LPS pull-down eGFP-GBP1 eGFP-GBP3 not tagged GBP2 or GBP4 from HeLa lysatesinteraction purified LPS-free His-GBP1 from CleanColi® BL21 bacteria Lipid IVA caspase-4/−11) lipid removal tested GBP1-LPS binding surface plasmon resonance direct binding LPS to GBP1 KD ~60 nM comparable KD LPS-caspase-4 LPS-caspase-11 interaction (Fig. Kinetic analysis GBP1-LPS binding two-reaction model experiment immobilized LPS yielded similar KD response weak small LPS absorbed microscale thermophoresis) GBP1 FITC-LPS confirmed interaction comparable KD value 6LPS binds to GBP1 high-molecular weight protein complex Streptavidin pull-down assay eGFP-GBP1-4 biotin-conjugated LPS HeLa cells-GBP1 primed with IFNγ 1 μg/mL Dox 16 h 1 million cells lysed incubated with 2 μg LPS-biotin biotinylated substrates pulled down streptavidin magnetic beads eluted SDS-PAGE sample buffer Streptavidin-bound -unbound fractions analyzed by western blot antibody against GFP SPR sensorgram of E.coli LPS human GBP1 CM5 chip Sensorgram LPS concentrations (47 94 188 375 750 1500 Gray lines SPR data orange lines model two-state-reaction model Saturation curve LPS GBP1 CM5 chip dissociation constants LPS GBP1 E. coli LPS LPS-caspase-4-caspase-11 published Shi et Size exclusion chromatograms recombinant LPS-free His GBP1 incubated LPS derivatives GBP1 (1 μM incubated ice LPS (2 μM) 5 h size-exclusion analysis Superdex 200 10/30 GL column Protein size estimated molecular weight standards corrected subtracting LPS-specific absorbance 280 nm 12% acrylamide gel immunoblotted His6 GBP1 elution peaks GTPase activity analysis recombinant GBP1 (500 nM incubated GTP (5 μM) ultrapure LPS 30 min Luminescence normalized buffer-only control Graphs mean ± SD data three five experiments batches recombinant His-GBP1 consequences LPS-GBP1 interaction caspase-11/−4 oligomerization5LPS-free GBP1 monomeric peak on size chromatography close predicted size 68.5 kDa (Fig. 6e Table E. coli Salmonella Typhimurium LPS formed micelles around 1000 kDa GBP1 incubated with LPS elution profile changed majority GBP1 to higher molecular weight peaks 400 kDa to over 1000 kDa. 6e prominent peak at 1000 kDa GBP1 bound to LPS micelles smaller complexes detected LPS from Rhodobacter sphaeroides induced shift GBP1 higher weight peaks antagonist caspase-11 oligomerization caspase5 incubation GBP1 with lipoteichoic acid) peptidoglycan similar shift LBP ovalbumin positive negative LPS-binding controls 12a tested LPS for binding GBP1 to LPS micelles LPS from E. coli Ra Rc Rd Re mutant strains bound by GBP1 high-molecular weight peaks 12b Salmonella ΔwaaL or ΔwaaG mutants recruited GBP1 caspase-4 in infected cellsLPS-GBP1 interaction sensitive to detergents binding to Lipid A solubilization with Tween-20 GBP1 oligomerization-dependent activation GTP assessed impact LPS on GBP1 GTPase activity (Fig. GBP1 hydrolyze GTP GTPase activity increased incubated with LPS GBP1 LPS LPS binding oligomeric state findings GBP1 to LPS LPS Lipid A inner core region sufficient for LPS-GBP1 interaction molecular weights GBP1 after incubation with LPS size)GBP167.0178.7GBP1 + LPS-E. LPS-Salmonella1140.5 LPS-R. interaction involves electrostatic LPS molecule E. coli has 6–8 negatively charged groups Lipid A inner core 12b human GBP1 lacks hydrophobic pockets LPS binding sites in CD14 MD-2 hypothesized binds to LPS by electrostatic interactionsGBP1 binding to LPS micelles disrupted by incubation with cations charges LPS35 incubation with polymyxin B interacts LPS Lipid A inner core monomeric GBP1 reduced GBP1-LPS association 7a dephosphorylating LPS with alkaline phosphatase reduced GBP1-LPS binding cell death LPS transfection groups on Lipid A inner core sugars LPS GBP1-LPS interaction activation non-canonical inflammasome pathway residues in N-terminal domain LBP mediate binding to LPS mutated positively charged patches in GBP1 tested impact on GBP1-LPS interaction (Fig. 7c most mutations no effect binding GBP1 LPS micelles mutation triple-lysines 61-63 to alanines reduced formation higher molecular weight GBP1 peaks monomeric GBP1 residues required for binding (Fig. 7d expression of GBP1KKK61-63AAA targeting cytosolic Salmonella GBP1KK87-88AA. 7eresults demonstrate GBP1-LPS interaction involves electrostatic forces GBP1 reduced caspase-4 pyroptosis impaired GBP1 targeting. 7GBP1 binds LPS electrostatic Size exclusion chromatograms recombinant GBP1 E. coli LPS CaCl2 (5 Polymyxin B alkaline corrected subtracting LPS-specific absorbance at 280 nm Black curves overlayed Release LDH from IFNγ-primed HeLa 5 h after transfection LPS 3D structure human GBP1 five negatively charged patches Residues patch E visible in PDB 6k1z mutated to alanines analyzed for GBP1-LPS interaction Purple GTPase domain green helical domain chromatograms His-tagged GBP1 mutants E. coli LPS corrected subtracting LPS-specific absorbance at 280 nm Fluorescence confocal microscopy of naive HeLa eGFP-GBP1wt-88AA infected with Salmonella-dsRed 1 h DNA stained with Hoechst Representative confocal images scale bar 5 μm.Percentage eGFP-GBP1 positive Salmonella at 1 h p.i. quantified counting 100–200 bacteria per coverslip. Graphs show mean ± SD data pooled from three four three. ***P < 0.001 two-tailed t-test GBP1 LPS sensor Gram-negative bacteria cytosol human cells GBP1-LPS interaction involves electrostatic forces Fig. GBP1 necessary for LPS-induced caspase-4 activation cell types LPS delivery methods data imply GBP1 first protein non-canonical inflammasome pathway interacts with LPS GBP1 upstream caspase-4 cytosolic LPS sensing question act LBP co-factor extracellular LPS detection by CD14 MD-2/TLR4 GBP1-LPS requires negative charges on LPS Lipid A inner core sugars positively charged surface patch GBP1 LPS binding LBP LBP GBP1 differ LBP binds to LPS micelles CD14 protein catalyzes LPS transfer to CD14 single LPS molecule MD-2/TLR4 GBP1 function alone part of GBP1-4 complex on LPS-containing membranespossible complex recruits caspase-4 transfers LPS activation assembly GBP complex insertion into bacterial membranes membrane integrity caspase-4 bind Lipid A LPS Additional studies structure composition GBP complex necessary caspase-4 recruitment activation not clear GBP1-LPS interaction requires structural determinants LPS GBP1 recognize negatively charged pathogen-derived molecules might function beyond Gram-negative bacteria GBP1 recruited to surface cytosolic parasites T. gondii membrane T. gondii parasitophorous vacuole GBP coat GBP coat in recruitment caspase-4 caspase-8-dependent apoptosis40 likely parasites vacuoles feature molecules similar chemical properties as LPS Lipid A core polysaccharides in cell membrane parasitophorous vacuole Identification additional ligands bind GBP1 GBP family member enhance understanding host innate immunity new paradigms for pattern recognition human-specific mechanism contrast to previous models mice GBPs promote non-canonical inflammasome activation vacuolar escape bacterial membrane destruction10unlikely mouse GBPs function differently human pathogens possible IRGs mouse enhances effector functions lysis of bacteria vacuoles mouse cells not human likely GBP-mediated recruitment IRGs Irgb10 antimicrobial recruiting caspase-11 mouse GBPs might mediate access LPS liberation membranolytic activity IRGs study first to report ligand GBP1 characterize mode interaction findings need in human cells provides first evidence GBP1 other GBPs function innate immune receptors for pathogen-associated molecular patterns expanding cytosolic innate immune defense pathways.MethodsBacterial strains mammalian cell bacteria grown at 37 °C orbital shakerSalmonella enterica Typhimurium SL1344 grown lysogeny broth 10 g/L NaCl streptomycin (50 μg Salmonella dsRed LB ampicillin (50 μg Typhimurium strain 4/74 ΔwaaG mutants Jay Hinton Liverpool Shigella M90T Jost Enninga Pasteur grown tryptic soy broth ampicillin (50 μg Listeria monocytogenes EGD Pascale Cossart Pasteur grown brain-heart infusion medium HeLa clone CCL-2 ATCC Human epithelial HT-29 HeLa cells cultured DMEM 10% Fetal Calf Serum Caco-2/TC-7 DMEM 20% FCS HT-29 Caco-2 cells Shaynoor Dramsi Pasteur THP-1 U937 cells cultured RPMI 10% FCS HBEC3-KT cells ATCC grown Bronchial/Tracheal Epithelial Cell Growth Medium HaCaT cells CLS Cell Lines Service grown DMEM 10% FCSHIEC-6 cells ATCC grown Opti-MEM 4% FCS 10 mM Glutamine 10 ng/mL epidermal growth factor Human primary monocyte macrophages purified buffy-coat Swiss Red-Cross cultured cells grown 37 °C 5% CRISPR/Cas9 knockout cell HeLa cells Alt-R CRISPR-Cas9 System sequence-specific CRISPR RNA recombinant Alt-R S. pyogenes Cas9 mixed 1 μM heated 5 min 95 °C cooled room temperature 1 μM Alt-R Cas9 incubated 5 min Lipofectamine RNAiMax transfection reagent) added incubated 20 min cells/well transfected 96-well plates 10 nM ribonucleoprotein complex incubation 2 days 37 °C 5% CO2 single clones generated serial dilutions gene knockouts screened T7 endonuclease I assay verified sequencing PCR confirmed western blotting crRNAs used CASP4) GSDMD GBP1)Plasmids siRNAs cell transfectionPlasmids expressing N-terminal GBPs generated inserting XhoI/HindIII sites pEGFP-C1 pmCherry-C1 pmiRFP703-GBP1 generated pEGFP-C1 plasmid miRFP703 NheI/BglII sites GBP1 XhoI/HindIII Caspase-4-eGFP generated fusing PCR caspase-4 eGFP NheI/HindIII sites pEGFP-C1 pAIP vectors expressing HA-GBP1 HA-GBP2 T. Henry GBPs GBP1 GBP3 −4 EcoRI sites bicistronic plasmids encoding FLAG-GBP3 HA-GBP4 inserting GBP3 GBP4 NotI/PmeI sites pBud-EGFP vector Doxycycline-inducible eGFP-GBP1 −2 −3 −4 generated amplifying eGFP-GBPs inserting coding sequences BamHI site pLVX-Puro vector cloning In-Fusion cloning plasmids verified sequencing Plasmids encoding eGFP mOrange galectin-3 JostHeLa cells plated 8-well μ-slides 1.5 × 104 cells/well imaging 24-well plates 1.0 × 105 cells/well 96-well glass plates (Eppendorf 8.0 × 104 cells/well 24 h before transfection transfected plasmids X-tremeGENE 9 DNA transfection reagent 16–48 h plasmids primers Supplementary Table 1. siRNA knock-down experiments cells seeded 96-well plates 9.0 × 104 cells/well transfected 3 pmol (25 nM) Stealth RNAiTM siRNAs (Thermo Fisher Scientific Lipofectamine RNAiMax Supplementary Table 2) After 8 h cells incubated IFNγ 16 h experiments performed.Infection assays transfection primed 10 ng/mL human IFNγ) 16 hours Salmonella cultures sub-cultured 1/50 grown Shigella Listeria sub-cultured 1/100 mid-exponential bacteria collected washed resuspended DMEM Salmonella added to HeLa cells 96-well plates 50,000 cells per multiplicity 50 incubated 30 min at 37 °Cinfection Shigella Listeria bacteria added cells MOI 20 incubated 30 min 37 °C Non-internalized bacteria removed washes DMEM incubated 100 μg/mL gentamicin 1 h bacteria changed DMEM 10 μg/mL gentamicin 10% FCS cells processed LDH release bacteria immunofluorescence assays cells washed PBS lysed 0.2% Triton X-100 Bacteria diluted plated LB agar cytosolic Salmonella CHQ resistance assay infected cells incubated 200 μg/mL CHQ gentamicin 1 h-resistant bacteria gentamicin Cells washed lysed plated percentage cytosolic bacteria calculated ratio (CHQ + gentamicinresistant U937 THP-1 cells seeded differentiated 100 ng/mL PMA 48 h 24 h resting cells LPS E. O111:B4 Salmonella 2.5 μg/50,000 cells 2.5 μg/80,000 cells Lipofectamine 2000 LPS diluted Opti-MEM incubated Lipofectamine 2000 (1.0 μl/50,000 cells 20 min room temperature75 μl Opti-MEM added 96-well plates 75 LPS mixture centrifuged 5 min 211 incubated 37 °C electroporation Neon Transfection System naive IFNγ-primed cells harvested resuspended electroporated LPS E. O111:B4 300 ng/50,0000 cells electrolytic buffer E pulse 1300 V 20 ms added 200 μl-warmed Opti-MEM 96-well plate centrifuged 5 min 211 incubated 37 °C Mock electroporation non cells controls imaging cells washed PBS fixed 4% PFA 20 min three times permeabilized 0.05% saponin blocked 1% BSA incubated antibodies Hoechst (1:1000) mounted ProLong Gold Antifade confocal microscopy Samples imaged Zeiss LSM800 confocal laser microscope 63×/1.4 Z-stacks 300 nmlive imaging infection assays EM buffer (120 mM NaCl 7 mM KCl 1.8 mM CaCl2 0.8 mM MgCl2 5 mM glucose 25 mM HEPES pH Cells infected 10 min bacteria removed EM buffer time-lapse microscopy 37 °C motorized xyz stage autofocus Super-resolution Zeiss LSM800 Airyscan images Zeiss ZEN software Data analyzed processed FiJi software fluorescence images maximum 3D projections.LDH release PI uptake IL-18 release western blottingCell death LDH release supernatant LDH cytotoxicity detection kit cell percentage cell death calculated × 100 PI influx measurement levels IL-18 measured ELISA western blotting cell lysates prepared supernatants precipitatedMouse anti-caspase-4-AAM-114-E Enzo Life Sciences anti-GSDMD-GBP1-GAPDH Thermo Scientific-V5 anti-GFP anti-HA-ABS-118-0200 anti-tubulin 1:2000 detected horseradish peroxidase-conjugated secondary antibodies (1:5000 Southern caspase pull cells seeded 6-well plates primed 16 h 10 ng/mL human IFNγ 3 × 106 cells treated 10 μM biotin-VAD-fmk transfected 20 μg E. LPS 3 hours lysed incubated overnight 20 μl pre-washed streptavidin magnetic beads beads washed-bound analyzed 12% acrylamide gel blotted against caspase-4.Streptavidin pull-down 2 × 106 cells collected lysed pull-down buffer (50 mM Tris-HCl 150 mM NaCl 5 mM EDTA 1% NP40 0.05% Na-deoxycolate protease 700 μg protein incubated 2 μg biotinylated LPS Pam3CSK4 room temperature two hoursincubation 20 μl-washed streptavidin magnetic beads Scientific incubated 1 h room temperature rocking beads washed three times 0.05% Tween-20 PBS precipitates eluted SDS-PAGE buffer western blotting analysis 5% initial cell lysate pull-down analyzed GFP-GBP cells lysed 20 × 106 cells/mL lysis buffer (50 mM Tris pH 7.4 150 mM NaCl 10 mM MgCl2 5 mM GTP 300 μM AlF 100 μg/mL digitonin lyse ice 15 min cells spun 15 min 6000 × g 4 °C soluble extract incubated 2 μg LPS/million cells 2 h streptavidin beads added incubated additional hour washed three times lysis buffer resuspended western blot buffer analyzed immunoblotting PCR mRNA extracted HeLa cells RNeasy Mini kit 400 ng transcribed cDNA Verso cDNA Synthesis kit Gene expression levels quantified qPCR LightCycler 480 housekeeping mRNA HPRTprimers qPCR Supplementary Table 3.Purification recombinant GBP1 cloned pET-28a N-terminally His-tagged hGBP1-hGBP1 transformed into CleanColi BL21 grown 2xYT medium OD600 0.5-0.7 Protein expression induced 30 °C 5 h 0.2 mM IPTG pellet resuspended buffer (50 pH 7.4 150 mM NaCl 1% Tween 20 frozen −80 °C until purification protein purified Ni-NTA affinity column Protein yield quantified Beer-Lambert law GBP1 purified size exclusion chromatography column 200 buffer (50 mM Tris pH 7.4 150 mM NaCl concentrated Amicon Ultra4 10 kDa chromatography GBP1 purified GBP1 (1 μM incubated ice LPS 5 h E.coli Salmonella Typhimurium Rhodobacter sphaeroides E. F585 diphosphoryl Lipid Salmonella minnesota 595 synthetic monophosphorylated Lipid E.coli EH100 J5coli F583 LPS mutant E. coli R515 LPS mutant Life Science incubation GBP1 GBP-LPS incubations injected Superdex 200 10/30 GL column buffer 1 column volume fractions (500 collected precipitated methanol separated 12% acrylamide gel analyzed immunoblotting antibody His6 molecular weights approximated gel filtration standard LPS pre-incubated 5 mM CaCl2 5 min GBP1 pre-incubated polymyxin B (10 μg/mL 5 min room temperature resonance measurements Biacore T200 GPB1 immobilized CM5 sensor chip amine coupling 310 RU 0.31 ng/mm2) equilibrated PBS buffer 7.2 injection concentrations LPS (47 94 188 375 750 1500 nM flow channels LPS concentrations GBP1 (21.5 43 86 172 343 688 1375 2750 nM analyzed BiacoreT200 Evaluation software 3.0 equilibrium analysis Langmuir isotherm fit dissociation constant best fit Kinetic curves two-state-reaction model structural re-arrangementMicroscale thermophoresis performed on 50 nM FITC-labeled E. coli LPS) purified hGBP1 or BSA control Serial dilutions GBP1 or BSA analyzed in assay buffer (50 mM Tris pH 7.4, 150 mM NaCl). Nanotemper Monolith NT.115 microscale electrophoresis instrument medium MST power Data fitted 1:1 model MO.Affinity Analysis software.GTPase activity GTPase-GloTM kit (Promega) Recombinant human GBP1 (500 nM) incubated with 5 μM LPS in GEF buffer (Promega 30 min room temperature before GTP hydrolysis Luminescence values normalized to no-GBP1 control.Data software Gen5 GraphPad Prism v8 Microsoft Excel Statistical significances referred * ** *** for P-values <0.05, <0.01 <0.001 comparison two groups two-tailed t-test three or more groups P-values two-way analysis of variance comparisons Nature Research Reporting Summary.Supplementary information 1 | 50.7 | 1.172199 |
10.1038/s41467-020-20873-y | PMC7840769 | Bacterial protein FtsZ polymerizes at mid-cell and exhibits treadmilling dynamics, driving the movement of enzymes that synthesize septal peptidoglycan. Here, McCausland et al. combine theoretical modelling with single-molecule imaging of bacteria to show that FtsZ treadmilling drives enzyme movement via a Brownian ratchet mechanism. | The FtsZ protein is a central component of the bacterial cell division machinery. It polymerizes at mid-cell and recruits more than 30 proteins to assemble into a macromolecular complex to direct cell wall constriction. FtsZ polymers exhibit treadmilling dynamics, driving the processive movement of enzymes that synthesize septal peptidoglycan (sPG). Here, we combine theoretical modelling with single-molecule imaging of live bacterial cells to show that FtsZ’s treadmilling drives the directional movement of sPG enzymes via a Brownian ratchet mechanism. The processivity of the directional movement depends on the binding potential between FtsZ and the sPG enzyme, and on a balance between the enzyme’s diffusion and FtsZ’s treadmilling speed. We propose that this interplay may provide a mechanism to control the spatiotemporal distribution of active sPG enzymes, explaining the distinct roles of FtsZ treadmilling in modulating cell wall constriction rate observed in different bacteria. | IntroductionDuring cell wall constriction in most Gram-negative bacteria, new septal peptidoglycan (sPG) synthesis and old cell wall degradation occur simultaneously1. A large number of the cell wall enzymes involved in this process and their regulators have been identified. However, it remains unclear how these proteins are orchestrated in time and space to achieve successful cytokinesis, and at the same time maintain the structural integrity of the septal cell wall2,3. Perturbations of PG remodeling at septum compromise cell division and often lead to cell lysis4.Recent studies have indicated that FtsZ, an essential component of the bacterial cell division machinery, may play a central part in regulating the spatiotemporal coordination of sPG synthesis enzymes. FtsZ is a highly conserved bacterial tubulin homolog and GTPase5–7. During cell division, FtsZ polymerizes at the cytoplasmic face of the inner membrane to form a ring-like structure (Z-ring) at mid-cell8–10. The Z-ring then locally recruits an ensemble of more than 30 proteins, many of which are sPG-remodeling enzymes1,11, to initiate septal cell wall constriction. New studies employing super-resolution and single-molecule imaging in vitro and in vivo have demonstrated that the FtsZ polymers exhibit GTP hydrolysis-driven treadmilling dynamics, which are the continuous polymerization at one end and depolymerization at the other end, with individual FtsZ monomers remaining stationary in the middle12–15. Most interestingly, it was found that FtsZ’s treadmilling dynamics drive processive movements of the essential sPG transpeptidase (TPase, FtsI in E. coli and PBP2B in B. subtilis)12,13 and glycosyltransferase FtsW16. Consequently, it was proposed that FtsZ’s treadmilling dynamics spatially and temporally distribute sPG synthesis enzymes along the septum plane to ensure smooth septum morphogenesis13. However, it is unknown how FtsZ’s treadmilling dynamics with stationary monomers in the cytoplasm are transduced into the periplasm to drive the persistent and directional movement of cell wall synthesis enzymes. The role of FtsZ’s treadmilling dynamics in modulating sPG synthesis activity also remains elusive, as it was shown that the cell wall constriction rate is dependent on FtsZ’s treadmilling speed in B. subtilis12 but not in E. coli13, or S. aureus17.In this work, we combined agent-based theoretical modeling with single-molecule imaging-based experimental testing to address the mechanism of the FtsZ treadmilling-dependent processive movement of sPG enzymes, and its associated role in bacterial cell division. We found that a Brownian ratchet mechanism underlies the persistent and directional movement of single sPG synthesis enzyme molecules driven by FtsZ’s treadmilling dynamics. Using FtsI as a model sPG enzyme, we found that the processivity of the Brownian ratchet is dependent on the (indirect) binding potential between FtsI and FtsZ and modulated by the balance between FtsI’s random diffusion and FtsZ’s treadmilling speed. This finding offers predictions about how different bacterial species could harness the same FtsZ treadmilling machinery to achieve distinct processivities of sPG enzymes, so that the available level of sPG synthases for cell wall constriction can be controlled differentially. Given the lack of linear stepping motors in the prokaryotic world, our work suggests a general framework for how polymer dynamics coupled with Brownian ratcheting could underlie directional transport of cargos, and be shaped by evolution to meet the needs of different cellular milieus.ResultsModel descriptionOur model is based on the concept of a Brownian ratchet, where FtsZ’s treadmilling introduces an asymmetry to bias the random diffusion of FtsI molecules in the periplasm, upon which FtsI persistently follows the shrinking end of a treadmilling FtsZ filament (Fig. 1). The quantitative details of the model are rooted in the physical and chemical properties of key components of the system, which can be characterized by experiments.Fig. 1Model description.a Schematic representation of an sPG synthase complex’s interaction with FtsZ treadmilling. FtsZ resides in the cytoplasm. FtsI is a transmembrane protein that does not dissociate from the membrane even when it dissociates from FstZ. b Model simplification of FtsZ–FtsI interaction at the septum. The FtsZ filament (purple) undergoes treadmilling by dissociating an FtsZ subunit from the left end and associating new ones on the right end. While FtsI (gray) intrinsically diffuses around, it has a binding affinity to FtsZ subunits. c Schematics of FtsZ–FtsI binding potentials. Here, the binding potential is assumed to be harmonic and short-ranged (~5 nm), which is about the size of an individual FtsZ subunit. Note that there is no energy barrier for FtsI to bind to FtsZ, because the binding potential is attractive. Once FtsI binds to an FtsZ subunit, however, the binding potential presents an energetic barrier preventing FtsI from diffusing away.As shown in Fig. 1, the model describes the movement of a free FtsI molecule at the septum as quasi-1D. The model assumes that FtsI, an essential TPase with a single transmembrane domain and a cytoplasmic tail, can freely diffuse along the inner membrane at the septum or interact indirectly with a treadmilling FtsZ filament underneath the inner membrane (Fig. 1a). The dynamics of a single FtsI molecule at the septum are therefore determined by three parameters: the constant of FtsI’s free diffusion (D), the treadmilling speed (VZ) of FtsZ filaments, and the attraction force determined by the binding potential (U) between FtsI and FtsZ (Fig. 1b, c).To set the ranges of the three parameters, we consider the following. First, we set the diffusion constant to range from 10−3 to 10−1 μm2/s, which is of a typical inner membrane protein in bacterial cells18,19. For example, PBP2, the counterpart of FtsI in cell wall elongation, was measured at ~0.06 μm2/s18,19. Second, the average treadmilling speed of FtsZ was at ~20–40 nm/s in vivo but can be a few-fold faster in vitro, therefore we set a large range of 10–100 nm/s14,15. Third, FtsI interacts with FtsZ at the septum indirectly through a relay of protein–protein interactions that include FtsN, FtsA, and/or FtsEX1,20. For simplicity we omit the details of the protein-protein interaction relay and refer to it as the interaction between FtsI and FtsZ. The indirect interaction between an FtsZ monomer and a nearby FtsI molecule constitutes an attractive force for each other, and can be described as a short-ranged harmonic binding potential (Fig. 1c). We assumed the potential range was ±2.5 nm, commensurate with the size of an FtsZ monomer (~5 nm)21–24. The potential’s magnitude was ~10 kBT, corresponding to a Kd in the μM range, which is typical for protein–protein interactions in the bacterial divisome system25–31. Note that here we use FtsI in E. coli as the model sPG enzyme, but the same analysis can be applied to any other sPG enzyme or divisome proteins as well.We first simulate the treadmilling of an FtsZ filament (Fig. 1b). The model depicts the filament shrinking and growing according to Eqs. (1) and (2), which respectively describe how the positions of the shrinking and growing ends of a treadmilling FtsZ filament at time t, xS(t) and xG(t) are related to the treadmilling rate, VZ:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{\rm{d}}x_S\left( t \right)}}{{{\rm{d}}t}} = V_{\rm{Z}},$$\end{document}dxStdt=VZ,2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{\rm{d}}x_{\rm{G}}\left( t \right)}}{{{\rm{d}}t}} = V_{\rm{Z}}.$$\end{document}dxGtdt=VZ.The model simulates the treadmilling events in a discretized sense, which occur every 5 nm/(Vz·Δt) simulation time step. Each time an FtsZ subunit falls off the shrinking end of the filament, the associated binding potential vanishes with it; likewise, when an FtsZ subunit adds onto the growing end, the associated binding potential appears with it. The treadmilling speed Vz of each FtsZ filament is drawn stochastically from an experimentally measured distribution13. Note that we do not model explicitly the stochastic on-reactions and off-reactions of individual subunits of an FtsZ filament, because such stochasticity is reflected in the treadmilling speed distribution of the filaments. The FtsZ filament length is set at 50 monomers (250 nm) and the treadmilling speed is independent of the filament length, in accordance to previous biochemical studies and a recent in vivo study20,32–35. To discern principal interactions, the model considers one FtsI molecule and one FtsZ filament in a self-contained septal section. It can be easily expanded to include multiple FtsI molecules per FtsZ filament (Supplementary Fig. 1).Next, to numerically compute the model, we describe the dynamics of FtsI by a Langevin-type equation (Eq. 3), where the viscous drag force on the molecule is in balance with a driving force f and a force from thermal noise ξ:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda \frac{{{\rm{d}}x\left( t \right)}}{{{\rm{d}}t}} = f\left( {x\left( t \right)} \right) + \xi \left( t \right).$$\end{document}λdxtdt=fxt+ξt.Here, x(t) represents the location of an FtsI molecule at time t along the 1D septum. λ is the effective viscous drag coefficient for FtsI’s movement with λ = kBT/D, where D is the diffusion constant of free FtsI molecules on the inner membrane when it is not interacting with FtsZ. f(x(t)) is the attractive force exerted upon the FtsI molecule by the FtsZ’s binding potential, U(x, t) (Fig. 1c). Specifically, f(x(t)) = –∂U(x, t)/∂x at time t. The last term ξ(t) reflects the random diffusive motion of FtsI on the inner membrane with 〈ξ(t)·ξ(t′)〉 = 2D·Δt·δ(t–t′), where Δt is the unitary time step in simulation. In the simulation, the FtsI molecule diffuses after the underlying FtsZ subunit falls off from the shrinking end, as it experiences a flat 5-nm local potential. However, the FtsI molecule has a probability to diffuse to the next FtsZ subunit in the row (i.e., another ~5 nm to the right), and associate there due to the presence of the binding potential of that FtsZ subunit. It could also diffuse away and dissociate from the FtsZ filament. This process is not deterministic but stochastic as we show below.Finally, assuming that the FtsZ filament treadmills from left to right with a steady-state length of ~250 nm, the model implements open boundary conditions on the FtsI molecule at both the left and right edges of the system and the right-ward FtsZ treadmilling is not limited. The model results presented below reflect the nominal case, whose essence remains robust against variations of model parameters within the physical range constrained by existing experimental data.A Brownian ratchet links directional movement to treadmillingAs we described above, Brownian ratcheting hinges on the diffusion of FtsI, its interaction with FtsZ, and FtsZ’s treadmilling speed. To examine how the movement of FtsI depends on FtsI’s diffusion and the binding potential between FtsI and FtsZ, we kept FtsZ’s treadmilling speed constant at an experimentally measured speed of 25 nm/s and carried out a phase diagram study using stochastic simulations. As described above, we chose a parameter range of 0.0001 to 0.1 μm2/s for FtsI’s diffusion18,19. The upper limit of the binding potential was set to be ~20 kBT, which corresponded to a dissociation constant Kd in the nM-range.We considered an initial condition in which both the shrinking end of an FtsZ filament and an FtsI molecule were at the left boundary of the septal section. To be commensurate with our experimental analysis, we counted an FtsI trajectory as moving directionally if it tracked the shrinking end of a treadmilling FtsZ filament persistently and unidirectionally for at least 4 s. Because of the stochastic nature of Brownian ratcheting, we characterized the state of FtsI under this parameter set condition as persistent end-tracking in the phase diagram if 50% or more of simulated FtsI trajectories displayed such a persistent directional movement.As shown in the phase diagram in Fig. 2a, the model revealed that when the binding potential between FtsZ and FtsI was weak (<5 kBT, ~mM Kd), FtsI largely displayed random diffusion without directional movements along the septum. When the attraction potential was sufficiently strong (>5 kBT), strong binding quenched free diffusion and confined FtsI to the end of an FtsZ filament. As the FtsZ subunit at the shrinking end of the filament fell off, FtsI was free to diffuse. However, the next FtsZ in the row presented the binding potential just a short distance away (~5 nm). By chance, the FtsI molecule diffused closer to the new end subunit of the FtsZ filament, where it was trapped again (the inset of Fig. 2b). Effectively, FtsI was “pulled” to the right by ~5 nm. With the subsequent FtsZ subunits falling off one after the other from the shrinking end, the FtsI molecule ratcheted forward and persistently tracked the end of the treadmilling FtsZ filament. These consecutive movements resulted in a persistent and directional trajectory of FtsI (Fig. 2b). Note that the directional movement of FtsI is not deterministic but rather probabilistic due to the stochastic nature of Brownian ratcheting; an FtsI molecule may decouple from a treadmilling FtsZ filament at any time (e.g., at a time point of ~14 s in the sample trajectory shown in Fig. 2b). Nevertheless, the average speed of FtsI molecules determined from their directional movement was tightly coupled to FtsZ’s treadmilling speed (Fig. 2c), recapitulating the experimentally measured near-linear correlation between FtsI’s directional motion with FtsZ’s treadmilling speeds in both wildtype and FtsZ GTPase mutants13.Fig. 2An FtsZ treadmilling-mediated Brownian ratchet mechanism drives FtsI’s directional movement.a Phase diagram depicting the dependence of FtsI’s persistent end-tracking (blue-shaded region) on FtsI’s diffusion constant and FtsI–FtsZ binding potential. b A representative simulated trajectory of an FtsI molecule persistently end-tracking with a treadmilling FtsZ filament. Inset: zoomed-in view of a boxed region of the trajectory. The model parameters for this simulation are: FtsZ’s treadmilling speed VZ = 25 nm/s, FtsI diffusion constant D = 0.04 μm2/s, FtsZ–FtsI binding potential U = 10 kBT, and the simulation time step = 5 × 10−6 s. The full trajectory is plotted every 5 × 10−2 s, and the zoomed-in inset is plotted every 5 × 10−6 s. c FtsI’s directional speed tightly couples with FtsZ’s treadmilling speed. Each of the data points was the average of >80 independent stochastic simulation trajectories using the segments that undergo directional movement. FtsZ’s treadmilling speed was fixed within each trajectory but varied across the ensemble following a Gaussian distribution with a standard deviation (SD) of 30%, in line with the experimental measurements in FtsZ WT and GTPase mutants13,16.The phase diagram (Fig. 2a) also showed that at a constant binding potential between FtsZ and FtsI, persistent end-tracking of FtsI required an appropriate range of diffusion constants. If FtsI diffused too rapidly, it could not be confined by the binding potential of the shrinking end of the FtsZ filament. Conversely, when FtsI diffused too slowly, it was not able to keep up with the speed of departing FtsZ subunits at the shrinking end. Once it fell behind, the FtsI molecule lost contact with the left most FtsZ subunits permanently.We note that an alternative initial condition, in which an FtsI molecule binds in the middle of an FtsZ filament, will result in the same end-tracking behavior. The bound FtsI molecule, if it has not dissociated, will start end-tracking when the shrinking end of the FtsZ polymer approaches and mobilizes it (Supplementary Fig. 1).Our modeling also found that coupling to the growing end of an FtsZ filament is unable to produce the directional movement of FtsI molecules (Supplementary Fig. 2). As there is no biochemical evidence showing that the FtsI-binding potential of a newly added FtsZ subunit at the growing tip will be higher than the ones in the middle of the filament, the addition of a new FtsZ subunit at the growing end does not bias the diffusion of the FtsI molecule bound at the original tip to dissociate and re-associate with the new FtsZ subunit. Consequently, slow-diffusing FtsI molecules will be stuck in the local binding potential, unable to catch up with the addition of new FtsZ subunits (Supplementary Fig. 2a), whereas fast-diffusing FtsI molecules have a high probability of escaping from the tip, because there are no FtsZ subunits beyond the growing tip to keep it within the vicinity as that in the shrinking tip-tracking scenario (Supplementary Fig. 2b). Therefore, FtsI cannot persistently track the growing end of an FtsZ filament.Taken together, our analysis showed that the end-tracking Brownian ratchet mechanism was able to couple FtsI’s directional movement to FtsZ’s shrinking end within the parameter range that is well consistent with experimentally measured data. Furthermore, the same model could explain the nondirectional movement of the cytoplasmic tail of FtsN, another divisome protein, in a recent in vitro study25. In this study, the cytoplasmic tail of FtsN was reported to follow the tracks of treadmilling FtsZ filaments on a supported lipid bilayer at the ensemble level. At the single molecule level, however, the FtsN tail only binds and unbinds FtsZ filaments transiently but does not exhibit directional movement25. Such a scenario could be explained by to our Brownian ratchet model in that the diffusion of a free FtsN cytoplasmic tail anchored on the membrane was too large (0.3–0.6 µm2/s)25.FtsZ treadmilling speed modulates FtsI’s end-tracking processivityNext, we investigated how FtsZ’s treadmilling speed impacts the processivity of FtsI’s directional movement at the shrinking end. Addressing this question will help us understand the role of FtsZ’s treadmilling dynamics in the spatial organization and/or regulation of sPG synthesis activity. We focused on three features that collectively define the processivity of FtsI’s end-tracking: (1) the propensity, (2) the run distance, and (3) the duration time of persistent end-tracking trajectories.We first examined how the relative propensity of FtsI’s persistent end-tracking was modulated by FtsZ treadmilling speed. The relative propensity is defined as the percentage of the number of FtsI persistent end-tracking trajectories at each FtsZ treadmilling speed, normalized by the total number of FtsI persistent end-tracking trajectories of all the simulated FtsZ treadmilling speeds. Keeping the diffusion constant of FtsI at 0.04 μm2/s and the binding potential at 10 kBT, stochastic simulations of the Brownian ratchet model predicted that the relative propensity of persistent end-tracking trajectories of FtsI dropped off with increasing FtsZ’s treadmilling speed (Fig. 3a). That is, when FtsZ treadmills too fast, FtsI could not persistently track the FtsZ shrinking end in most cases and became largely diffusive.Fig. 3Model predictions for the processivity of FtsI directional movement modulated by FtsZ treadmilling speed.a Predicted relative propensity of FtsI persistent end-tracking as a function of FtsZ’s treadmilling speed. For each FtsZ treadmilling speed, 100 independent stochastic FtsI trajectories were simulated, from which the number of FtsI persistent end-tracking trajectories were counted. In line with our experimental measurements, the criteria for persistent end-tracking were as follows: (1) the distance between the FtsZ shrinking end and FtsI is less than 100 nm and (2) FtsI persistently follows the FtsZ shrinking end for greater than 4 s. A total of 664 out of 1000 simulated FtsI trajectories were scored as persistent end-tracking, and the number of trajectories at each FtsZ treadmilling speed was normalized as a relative probability of end-tracking. b Calculated phase diagram of FtsI persistent end-tracking characterized by FtsI diffusion constant and FtsZ treadmilling speed. c Predicted FtsZ treadmilling speed-dependence of the duration of FtsI persistent end-tracking. d Predicted FtsZ treadmilling speed-dependence of run distance of FtsI persistent end-tracking. The shaded regions emphasize the rise and decay of the maximum persistent run distance dependent on FtsZ’s treadmilling speed. For the model calculations in a–d, the FtsZ–FtsI binding potential is set to be 10 kBT.To further this point, we calculated the phase diagram of FtsI’s persistent end-tracking propensity as a function of both FtsI’s diffusion constant and FtsZ’s treadmilling speed (Fig. 3b), while keeping the binding potential fixed at 10 kBT. Again, we used a threshold of 50% FtsI persistent end-tracking trajectories as the criterion for the phase boundary. As shown in Fig. 3b, for a fixed diffusion constant of FtsI, there was an upper limit of FtsZ’s treadmilling speed that FtsI could persistently track. Conversely, for a fixed FtsZ treadmilling speed, persistent end-tracking of FtsI required an appropriate range of diffusion constants. Importantly, very large diffusion constants of FtsI (>0.1 μm2/s) did not support persistent end-tracking irrespective of FtsZ’s treadmilling speed. These results were consistent with the phase diagram in Fig. 2a and again the recent in vitro study of FtsN’s cytoplasmic tail25.Next, we investigated how FtsZ’s treadmilling speed modulates the run distance and duration time of FtsI’s persistent end-tracking. The Brownian ratchet model predicted that both the run length and duration time of FtsI’s persistent end-tracking should display broad distributions, due to the stochastic nature of FtsI’s diffusion and the interaction between FtsI and FtsZ. Moreover, the model predicts that when FtsZ’s treadmilling speed increases, the duration time of FtsI’s persistent end-tracking will decrease (Fig. 3c), whereas the run distance will display a biphasic dependence—it increases to peak around an intermediate FtsZ treadmilling speed (~30 nm/s at the current parameter setting), and then decreases when FtsZ’s treadmilling speed increases further (Fig. 3d). Importantly, such distinctive dependences of duration time and run distance on FtsZ’s treadmilling speed is a natural consequence of the Brownian ratchet mechanism (see “Methods” section).Qualitatively speaking, when an FtsZ subunit falls off to the cytoplasm from the shrinking end of the FtsZ filament, the associated FtsI molecule will dissociate from the FtsZ subunit, either diffuse away on the membrane, or catch up with the next FtsZ subunit in the row to continue end-tracking, the latter depending on how fast FtsZ treadmills. When FtsZ treadmills too fast (for example >30 nm/s), it will be difficult for FtsI to catch up (Fig. 3a), resulting in early termination of end-tracking, and hence both the persistent run distance and duration time will be short (right sides of Fig. 3c, d). When FtsZ treadmills relatively slowly (<30 nm/s), the probability of FtsI catching up with the shrinking end of the FtsZ filament is high (Fig. 3a). Therefore, the slower FtsZ treadmills, the fewer number of dissociation events an end-tracking FtsI molecule would face, and hence the lower the chance for FtsI to diffuse away, leading to a longer time duration of the persistent run (Fig. 3c). Within the same time window, however, the persistent run distance will be proportional to FtsZ’s treadmilling speed as predicted in Fig. 3d, that is, the slower FtsZ treadmills, the shorter FtsI’s persistent run distance is. One can imagine in one extreme case where FtsZ does not treadmill at all (Vz = 0), the duration time of persistent runs would then be the longest and mainly dictated by the intrinsic dissociation rate of FtsI from FtsZ, and the persistent run distance would be the shortest (i.e., the size of a single FtsZ subunit). In the other extreme case where FtsZ treadmills infinitely fast (Vz = ∞), the duration time and distance of persistent runs would both be zero because FtsI could never track it. An analytical proof of these relationships is provided in the “Methods” section and Supplementary Fig. 3.Single-molecule tracking of FtsI confirms model predictionsTo experimentally examine the model’s predictions on the modulation of the processivity of FtsI’s directional movement by FtsZ’s treadmilling speed, we performed single-molecule tracking (SMT) of a functional sandwich fusion protein Halo-FtsISW labeled with JF646 in live E. coli cells36,37. To avoid disrupting the cytoplasmic interactions of FtsI’s N-terminal tail with other divisome proteins, we inserted the Halo tag between the last residue (18) of the N-terminal cytoplasmic tail and the first residue of the inner membrane helix (19) of FtsI (Fig. 4a). We integrated the halo-ftsISW fusion gene into the chromosome replacing the endogenous ftsI gene, and showed that it was expressed as a full-length fusion protein and supported normal cell division as a sole cellular source of FtsI similar to wild-type (WT) cells (Supplementary Fig. 4 and Supplementary Table 2).Fig. 4Experimental characterization of FtsI directional motion.a Schematic of the functional sandwich fusion of FtsI. The Halo tag is inserted between residues 18 and 19 of FtsI, immediately before the first residue of the transmembrane (TM) domain. b Diagram of individual E. coli cells loaded in microholes made by nanopillars. c Brightfield and fluorescence images of microholes loaded with E. coli cells labeled with GFP-ZapA and Halo-FtsISW fusion proteins. Inset shows the zoomed image of one cell in the yellow box. Blue circles indicated the pixels used for the fit of ZapA-GFP circle. Scale bar is 5 µm. Experiment repeated 18 times with similar results. d Circle-fitting of GFP-ZapA image super-imposed with the trajectory of a single FtsI molecule, colored in time. e the unwrapped trajectory from D with fitted lines at each segment to extract directional speeds, persistent run duration and distance. f Distributions of FtsI’s directional movement speeds (n = 77 trajectories, 74 directional events) and FtsZ’s treadmilling speed (from Yang et al.13). For the purpose of this study, we only plotted the fast-moving population of FtsI that follows FtsZ’s treadmilling and leave out the slow-moving population of FtsI that we show to be independent of FtsZ’s treadmilling13. Both FtsI’s and FtsZ’s histograms were bootstrapped 100 times to provide the shaded standard error bars (mean ± S.E.M.). A two-sample Kolmogrov–Smirnov test comparing the two datasets showed no difference in their distributions (p = 0.1852). g Dependence of the duration of FtsI’s persistent run on its speed. h Dependence of the distance of FtsI’s persistent run on its speed.To obtain precise measurements of the persistent run distance and duration time of single Halo-FtsISW molecules, we trapped individual E. coli cells vertically in agarose microholes made using cell-shaped nanopillar molds as previously described38,39, so that the entire circumference of the septum could be visualized at the same focal plane (Fig. 4b). To determine whether a Halo-FtsISW molecule was at a septum, we labeled the FtsZ-ring using an ectopically expressed GFP-ZapA fusion protein, which we and others have previously shown as a faithful marker of Z-ring localization and dynamics40. The GFP-ZapA image also allowed us to unwrap the circular trajectories of FtsI-Halo molecules along the septum (Fig. 4d) to linear displacements along the circumference of the septum (Fig. 4e), from which we could measure the persistent run speed, distance, and duration time (Fig. 4f, g, h).As shown in Fig. 4f, the directional motion speed of Halo-FtsI exhibited a wide distribution, similar to what we previously observed for FtsZ’s treadmilling. The similarity between FtsI’s directional motion speed distribution and FtsZ’s treadmilling speed distribution suggests that at these speed ranges, E. coli FtsI can faithfully end-track treadmilling FtsZ filaments as the model predicted, a point that will become important in the section below. Most importantly, the persistence run distance and duration time exhibited largely the same trends as what were predicted by the model: while the run duration time decreased monotonically (Fig. 4g), the persistence run distance increased and then decreased when FtsI’s speed increased (Fig. 4h). Note here that we inferred FtsZ’s treadmilling speed from FtsI’s directional motion speed due to the difficulty of a two-color co-tracking experiment in the same cells, and because we have demonstrated previously that these two were linearly coupled13. Another potential caveat in these experiments was that a very fast FtsZ treadmilling speed (i.e., >80 nm/s) is rare in wildtype E. coli cells as we showed previously. Therefore, given the relatively small dataset for high speed FtsZ treadmilling, our data cannot definitively determine whether FtsI could effectively end-track FtsZ filaments of very fast treadmilling speeds. Nevertheless, the agreement of our experimental measurements with theoretical predictions supported the validity of the Brownian ratchet model.The dependence of sPG enzymatic activity on FtsZ treadmillingIn E. coli, the total amount of septal PG synthesis and the septum constriction rate are insensitive to perturbations in FtsZ’s treadmilling speed from ~8 to ~30 nm/s in a series of FtsZ GTPase mutants13. This insensitivity suggests that end-tracking, directionally-moving FtsI molecules were inactive in sPG synthesis. Indeed, a second, slow-moving population of FtsW and FtsI (~8 nm/s) is found to move independently of FtsZ’s treadmilling, and likely corresponds to the active population of sPG synthesis in E. coli16. Similarly, in. S. pneumoniae, FtsW and its cognate TPase PBP2x were found to move completely independently of FtsZ’s treadmilling41, likely representing the active population of sPG synthase as that in E. coli. In B. subtilis, however, it was shown that the cell wall constriction speed is positively correlated with FtsZ’s treadmilling speed, suggesting that the faster FtsZ treadmills, the higher the sPG synthesis activity12. How could the same FtsZ treadmilling dynamics result in different sPG synthesis activity in different species?We propose that since FtsI molecules tracking FtsZ filaments are most likely inactive16, the population of FtsI molecules not tracking with treadmilling FtsZ polymers is then available for sPG synthesis. Therefore, FtsI’s off-rate, or the reciprocal of the time an FtsI molecule spends bound in the middle of FtsZ polymers and/or persistently end-tracking, represents the rate at which an FtsI molecule becomes available for sPG synthesis, and therefore is proportional to the sPG synthesis rate. As such, FtsZ’s treadmilling speed could modulate the rate of sPG synthesis in different bacterial species depending on the unique combination of the enzyme’s diffusion coefficient and binding potentials. This modulation could explain the difference observed between E. coli and B. subtilis.As shown in Fig. 5a, the model suggests that when FtsI diffuses relatively fast (~0.05 μm2/s, blue line), the lifetime of FtsZ-bound FtsI is largely insensitive to FtsZ’s treadmilling speed in a 3-fold range from ~8 to 25 nm/s. In contrast, when FtsI diffuses relatively slowly, the lifetime of FtsZ-bound FtsI is critically dependent on FtsZ treadmilling speed. For example, at a diffusion constant of 0.005 μm2/s, the relative lifetime of FtsI molecules decreased by ~70% when FtsZ treadmilling speed increased from ~8 to 25 nm/s (Fig. 5a, green line).Fig. 5Diffusion and binding potential modulate sPG synthase’s processivity on FtsZ’s treadmilling speed.a Predicted dependence of FtsZ-bound FtsI lifetime on FtsZ treadmilling speed with a binding potential of 10 kBT at different FtsI diffusion coefficients. b A representative trajectory of persistent end-tracking when FtsI diffusion is fast (0.05 μm2/s). c A representative trajectory of persistent end-tracking when FtsI diffusion is slow (0.005 μm2/s). d Similar to a, but with a weaker binding potential (8 kBT) at different FtsI diffusion coefficients. e Similar to a, but with a stronger binding potential (12 kBT) at different FtsI diffusion coefficients. For a, d, and e, FtsI is initially randomly positioned along an FtsZ filament. The lifetime of FtsZ-bound FtsI was defined as the first time that FtsI escapes from either end of the FtsZ filament for greater than 100 nm. For b and c, the model results are plotted every 2 × 10−2 s with a simulation time step of 10−5 s. f Mean-squared displacement (MSD) of FtsI, PBP2b and FtsW outside the septum in wildtype E. coli, B. subtilis, and S. pneumoniae. The fitted diffusion constants of FtsI, PBP2b, and FtsW were 0.041, 0.038, and 0.028 µm2/s, with α = 0.29, 0.51, and 0.71, respectively. Error bars represent standard error of the mean. g Model fitting of the dependence of sPG synthesis activity on FtsZ treadmilling speed in E. coli. The relative experimental data of sPG synthesis activity was taken from previous constriction rate measurements in Coltharp et al.42 and Yang et al.13 (mean ± S.E.M., each paper reported at least n > 30 cells). h Model fitting of the dependence of sPG synthesis activity on FtsZ treadmilling speed in B. subtilis. The relative experimental data of sPG synthesis activity was taken from previous cell division time measurements in Bisson-Filho et al.12 (mean ± standard deviation. n = 100 cells for cell constriction measurements, n > 50 cells for treadmilling measurements). For g and h, the model fitting used the measured diffusion constants from f, with varying binding potentials determined by lower and upper limits (E. coli: 8 and 9 kBT; B. subtilis: 10 and 12 kBT), marking the boundaries of the shaded regions.The physical reason behind this drastic difference between fast and slow FtsI diffusion lies at the core of Brownian ratchet mechanism. Fast diffusion will allow FtsI to catch up with the shrinking end of an FtsZ filament in a very short time (Fig. 5b). When FtsI’s diffusion becomes slower and slower, it eventually becomes the rate-limiting factor in the Brownian ratchet—a slow FtsI molecule falls behind the FtsZ shrinking end and takes a long time to catch up with the departing FtsZ filament, or simply diffuses away and become lost (Fig. 5c). As such, further increasing the FtsZ treadmilling speed in the latter case will significantly reduce the chance of FtsI keeping up with the FtsZ shrinking end and, hence the lifetime of the FtsZ-bound FtsI. Crucially, this diffusion-modulated dependence of sPG synthesis on FtsZ treadmilling speed hinges on the binding between FtsI and FtsZ. When the binding potential is reduced (i.e., weaker binding), the lifetime of FtsZ-bound FtsI is less sensitive to FtsZ speed than its higher-potential counterpart (compare Fig. 5a, d, e). Therefore, how FtsZ’s treadmilling speed modulates the rate of sPG synthesis depends on the combined effects of the sPG synthase’s diffusion coefficient and FtsZ-binding potential.To examine this hypothesis, we performed fast frame-rate single-molecule tracking to measure the diffusion coefficients of free sPG synthase molecules outside the septum (FtsI in E. coli and PBP2b in B. subtilis, Fig. 5f). As FtsZ only localizes to mid-cell during cell division, sPG synthase molecules not localized to mid-cell are considered free and not interacting with FtsZ. Using the measured diffusion coefficients, we then fit the experimentally observed dependence of cell wall constriction rate on FtsZ treadmilling speed in E. coli and B. subtilis12,42 with the normalized off-rate calculated from the model. The only free parameter in the model fitting is the binding potential, which was not possible to measure accurately in live cells with available experimental methods. As shown in Fig. 5g, h, we found that with the apparent diffusion constants of FtsI in E. coli and PBP2b in B. subtilis measured at ~0.041 ± 0.0051 (mean ± S.E.M., N = 5049 trajectories) and 0.038 ± 0.0019 µm2/s (mean ± S.E.M., N = 6765 trajectories) respectively, and binding potentials set at 8–9 and 10–12 kBT, respectively, the model quantitatively recapitulated the differential dependence of cell wall constriction rate on FtsZ’s treadmilling speed in the two species as previously measured. The higher binding potential of PBP2b to FtsZ in B. subtilis, likely due to the significantly different protein–protein interactions in Gram-positive bacteria, renders tighter coupling between end-tracking PBP2b molecules with FtsZ than that in E. coli, hence the fraction of end-tracking FtsI can be sensitively modulated by FtsZ’s treadmilling speed in B. subtilis, but that of FtsI in E. coli. cannot.As a comparison, we also measured the diffusion of FtsW in S. pneumoniae (Fig. 5f). We found that the apparent diffusion coefficient of FtsW was at 0.028 ± 0.0004 µm2/s (mean ± S.E.M., N = 21 trajectories), in the same order of magnitude as that of E. coli and B. subtilis. As FtsW does not follow the treadmilling of FtsZ at all in S. pneumoniae, it is most likely that the binding potential between FtsW and FtsZ is significantly lower than 5 KBT as predicted by the model (Fig. 2a) under the experimental condition. It remains interesting to investigate in S. pneumoniae whether other divisome proteins are also independent of FtsZ’s treadmilling, or they exhibit conditional dependence once the protein–protein interactions of the divisome are altered due to the presence or depletion of their binding partners. These possibilities will be further investigated in our future work.DiscussionIn this work, we presented data supporting a Brownian ratchet model that couples the directional movements of sPG synthases to FtsZ’s treadmilling, and underlies the differential sensitivity of sPG synthesis to FtsZ’s treadmilling speed in E. coli and B. subtilis.We first show that an sPG synthase molecule (here using FtsI as the model enzyme) can follow a treadmilling FtsZ polymer by end-tracking its shrinking tip, but not the growing tip, due to the intricate interplay between the sPG synthase molecule’s diffusion and its binding potential to FtsZ. Only within a particular diffusion range (~0.001–0.1 μm2/s) and at a sufficient binding potential (>5 kBT), an sPG synthase molecule can exhibit FtsZ-treadmilling-dependent directional movement (Fig. 2). Furthermore, we show that the persistent run duration and distance of FtsI exhibit different dependence on FtsZ’s treadmilling speed, as predicted by the Brownian ratchet model (Fig. 3). Using single-molecule tracking, we confirmed these model predictions (Fig. 4). The ability of treadmilling FtsZ polymers to modulate the persistent run duration and distance of sPG synthase molecules could play an important role in regulating the spatial distribution of sPG synthases to ensure the correct septum shape. Finally, we show that the Brownian ratchet model could explain the differential dependence of sPG synthesis activity on FtsZ’s treadmilling speed in E. coli, B. subtilis and S. pneumoniae.Given experimentally measured diffusion coefficients of sPG synthases in different bacterial species, the Brownian ratchet model predicts that the tighter binding between PBP2b and FtsZ in B. subtilis could cause tighter coupling between them. Hence, the fraction of time a PBP2b molecule spends on FtsZ, and consequently the fraction of time it is off FtsZ to become available for sPG synthesis, can be sensitively modulated by FtsZ’s treadmilling speed. In E. coli or S. pneumoniae, the binding potential between FtsI/FtsW and FtsZ may not be as high as that in B. subtilis, and hence the fraction of time an FtsI/FtsW molecule remains end-tracking FtsZ exhibits much less sensitivity or no sensitivity at all to FtsZ’s treadmilling speed. Different binding potentials between these different bacterial species are likely, as detailed molecular interactions among the septal ring complexes are distinct in each species. These results suggest that the same Brownian ratchet machinery may be at work, but operate in distinct regimes of the parameter space in different species. Consequently, sPG synthesis depends on FtsZ treadmilling differentially, reflecting different strategies to meet different functional needs.We note that additional factors could also be at play. For example, the expression level of sPG synthase relative to that of FtsZ in B. subtilis is significantly higher than that in E. coli or S. pneumoniae43–51 (Supplementary Table 1). This suggests that the number of sPG synthases per FtsZ filament in B. subtilis would be higher than that in E. coli or S. pneumoniae. Our Brownian ratchet model predicts that this condition further enhances the sensitivity of sPG synthesis activity to FtsZ’s treadmilling speed, as sPG synthase molecules bound to the inner positions of an FtsZ polymer would “knock” the end-tracking one off FtsZ (or vice versa, Supplementary Fig. 1). Therefore, the faster FtsZ treadmills, the faster the FtsZ shrinking end catches up to the sPG synthase in the middle and the more sPG synthase molecules will be dissociated from FtsZ to become available for sPG synthesis. This is a stark contrast to the case of a single sPG synthase per FtsZ filament, which is mostly likely the case in E. coli or S. pneumoniae (Supplementary Fig. 1).Moreover, the level of cell wall synthesis precursors, for example, could be another important factor. It is possible that across bacterial species, sPG synthase molecules are coupled to FtsZ’s treadmilling dynamics and their lifetime on FtsZ can be sensitively modulated by FtsZ’s treadmilling speed. However, if the level of a cell wall synthesis precursor is limiting, which is likely the case in E. coli, such a sensitivity could be further masked by the limited level of the precursor52–55. In S. pneumoniae, besides a low binding potential between an sPG synthase and FtsZ, cell wall synthesis precursor levels could also play a role in the independence of FtsZ’s treadmilling. High enough levels of PG precursors could saturate all sPG synthase molecules so that no free ones are available to track with FtsZ polymers.In summary, given the lack of linear stepper motors in prokaryotic world, Brownian ratcheting appears to be an ancient mechanism for directed cargo transportation in bacteria—another salient example is ParA-mediated DNA partitioning56–58. Interestingly, a similar Brownian ratchet mechanism also underlies the directional movement of mitotic chromosomes by end-tracking spindle microtubule in eukaryotes59. Can we distill unified fundamental principle(s) by which evolution shapes the same Brownian ratchet mechanism to meet distinct needs under different contexts? We will relegate these exciting questions to our future study.MethodsAnalytic solutionWe aim to obtain the analytical solution for the FtsZ treadmilling speed-dependence of run distance and duration of FtsI persistent end-tracking. An FtsI’s persistent end-tracking trajectory can be decomposed into repeating steps, each of which consists of two consecutive processes. Process (1) is “stay-on”: The FtsI molecule stays inside the binding potential of the FtsZ end subunit. Process (2) is “catch-up”: The FtsI molecule catches up with the next FtsZ subunit in the row when the FtsZ end subunit dissociates. We next calculate the probabilities of the two processes.For FtsI in the “stay-on” process (Supplementary Fig. 3a), it can undergo two reactions in parallel: FtsI can escape from the binding potential of the FtsZ end subunit with the rate of 1/τD, and the FtsZ end subunit can dissociate with the rate of 1/τz. Here, τz is the average lifetime of an FtsZ subunit, inversely proportional to the FtsZ treadmilling speed, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _{\rm{Z}} = \frac{{5\,{\rm{nm}}}}{{V_{\rm{Z}}}}$$\end{document}τZ=5nmVZ. τD is the average duration of FtsI staying in the binding potential if the FtsZ subunit never falls off. The probability that FtsI is still in “stay-on” at time, t, is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_1 = {\rm{exp}}\left( { - \left( {\frac{1}{{\tau _{\rm{D}}}} + \frac{1}{{\tau _{\rm{Z}}}}} \right)t} \right)$$\end{document}P1=exp−1τD+1τZt. Thus, by the time the FtsZ end subunit dissociates, P1 scales as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_1\sim {\mathrm{exp}}\left( { - \frac{{\tau _{\rm{Z}}}}{{\tau _{\rm{D}}}}} \right)$$\end{document}P1~exp−τZτD. Likewise, the catch-up probability of the FtsI molecule P2 scales as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_2\sim \exp \left( { - \frac{{\tau _{\rm{C}}}}{{\tau _{\rm{Z}}}}} \right)$$\end{document}P2~exp−τCτZ. Here, τC is the average catch-up time (Supplementary Fig. 3b). Taken together, the probability for FtsI end-tracking in each repeating step is:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P = P_1P_2\sim \exp \left( { - \left( {\frac{{\tau _{\rm{Z}}}}{{\tau _{\rm{D}}}} + \frac{{\tau _{\rm{C}}}}{{\tau _{\rm{Z}}}}} \right)} \right)$$\end{document}P=P1P2~exp−τZτD+τCτZIt follows that the probability of persistent end-tracking exact N-repeating steps is PN (1 − P). Consequently, the average number of FtsI persistent end-tracking steps is:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\langle N \right\rangle = \frac{{\mathop {\sum }\nolimits_{{\rm{N}} = 0}^{{\infty }} N \cdot P^N \cdot \left( {1 - P} \right)}}{{\mathop {\sum }\nolimits_{N = 0}^\infty P^N \cdot \left( {1 - P} \right)}}$$\end{document}N=∑N=0∞N⋅PN⋅1−P∑N=0∞PN⋅1−PTaking the continuum limit, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\langle N \right\rangle = \frac{{\mathop {\smallint }\nolimits_0^\infty \left\{ {N \cdot P^N \cdot \left( {1 - P} \right)} \right\}{\rm{d}}N}}{{\mathop {\smallint }\nolimits_0^\infty \left\{ {P^N \cdot \left( {1 - P} \right)} \right\}{\rm{d}}N}}$$\end{document}N=∫0∞N⋅PN⋅1−PdN∫0∞PN⋅1−PdN, which yields \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\langle N \right\rangle = \frac{{\tau _{\rm{D}}\tau _{\rm{Z}}}}{{\tau _{\rm{Z}}^2 + \tau _{\rm{C}}\tau _{\rm{D}}}}$$\end{document}N=τDτZτZ2+τCτD. The corresponding average run length and duration are:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\langle L \right\rangle = L_0\frac{{\tau _{\rm{D}}\tau _{\rm{Z}}}}{{\tau _{\rm{Z}}^2 + \tau _{\rm{C}}\tau _{\rm{D}}}},$$\end{document}L=L0τDτZτZ2+τCτD,and7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\langle T \right\rangle = \frac{{\tau _{\rm{D}}\tau _{\rm{Z}}}}{{\tau _{\rm{Z}}^2 + \tau _{\rm{C}}\tau _{\rm{D}}}}\tau _{\rm{Z}}$$\end{document}T=τDτZτZ2+τCτDτZHere, L0 = 5 nm, the length of FtsZ subunit.Specifically, persistent end-tracking entails that τD >> τZ >> τC. According to our numerical simulation, normally τD ∼ 60 s and τC ∼ 0.001 s. Within this parameter range, Eqs. 6–7 can quantitatively recapitulate the distinctive dependences of run length and duration on FtsZ treadmilling speed. Herein, Supplementary Fig. 3c presents a representative result of the analytic solution with τD = 60 s and τC = 0.0003 s.Qualitatively speaking, the longer the lifetime of an FtsZ subunit, FtsI in process (1) will have a higher chance to escape the binding potential of an FtsZ end subunit, decreasing the stay-on probability. In contrast, a longer lifetime of an FtsZ subunit will allow a longer time for FtsI in process (2) to catch up to the next FtsZ subunit in the row, increasing the catch-up probability. Therefore, the balance between the stay-on and catch-up processes defines an optimal lifetime of an FtsZ subunit—and, hence, the optimal FtsZ treadmilling speed—that maximizes the probability of FtsI end-tracking per repeating step. This determines the maximum of the average number of repeating steps, and explains the biphasic dependence of FtsI persistent run length on FtsZ treadmilling speed. On the other hand, the duration per repeating step is approximately the lifetime of an FtsZ subunit. As the FtsZ treadmilling speed increases, the decrease in the lifetime of the FtsZ subunit outweighs the corresponding variation in the average number of repeating steps, leading to the decrease in the overall duration of persistent end-tracking.Media, bacterial strains, and plasmidsCells were grown in M9 minimal medium or lysogeny broth (LB) (10% tryptone, 10% NaCl, and 5% yeast extract). Fresh LB plates of strains were struck with appropriate antibiotics (detailed below) once-per-week from frozen stocks, and all cultures were started with a single colony. Bacterial strains and plasmids used are detailed in Supplementary Table 1.Growth curveThree biological replicates from TB28 and JM136 were grown from single colonies in M9+Glucose (M9) [1× M9 salts (Sigma-Aldrich M9956), 1× Amino Acids (Sigma-Aldrich M5550), 0.4% glucose, 1× Vitamins (Sigma-Aldrich M6895), 2 mM MgSO4, 100 µM CaCl2] as overnights at 30 °C. The following day, the overnights’ OD600 were measured with a nanodrop and diluted to an OD600 of 0.1 in 200 μl M9 using a Corning Costar sterile 96-well plate. The 96-well plate was incubated in a Tecan Infinite M200 Pro set at 30 °C, where it would measure the OD600 of designated wells once every 30 min for 23.5 h, shaking the plate for 3 min at 220 rpm before measuring. To obtain the growth rate, the linear phase of the log-transformed growth curve data was fitted to a straight line. The slope of that line was used to calculate the doubling time through the equation below.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{Doubling}}\,{\rm{time}} = \frac{{\log 10(2)}}{{{\rm{slope}}}}$$\end{document}Doublingtime=log10(2)slopeWestern blot of FtsITB28 and JM136 were grown as overnights from single colonies in M9 at 37 °C. They were propagated 1:50 in 3 ml M9 and grown an additional 16 h at 25 °C until their OD600 reached ~0.5, at which point 500 μl was harvested. These samples were pelleted, flash-frozen, and stored at −80 °C for one hour. These pellets were then resuspended with PBS and mixed with 2× SDS buffer (100 mM Tris-HCL, 4% SDS, 0.2% bromophenol blue, 20% glycerol, 100 mM DTT). Cells were incubated for 10 min at 95 °C in a thermocycler, then 20 μl of each were loaded into a 10% BioRad polyacrylamide gel with 5 μl of PageRuler as a ladder. After electrophoresis, the gel was transferred at 25 V for 2 h to a nitrocellulose membrane. Membranes were checked for even transfer with a Ponceau stain before blotting. Rabbit α-FtsI, generously donated by Dr. David Weiss, was used as the primary antibody with a 1/50,000 dilution. Blots were then stained with 1/50,000 HRP goat antirabbit secondary antibody. Bands were visualized with a Clarity Western ECL Substrate BioRad kit and recorded using Blu-C autoradiography film for 10 s. This western was repeated twice with four more biological replicates overall with similar results.Guide RNA and repair oligo design for the Halo-FtsISW fusionThe chromosomal ftsI gene coordinates were first located using EcoCyc, which were then used to obtain the full ftsI gene sequence from NCBI (MG1655 genome, accession number NC_000913). This sequence was copied into ChopChop60 to identify candidate protospacers near the N-terminus of FtsI covering residues 18 and 19 to make the sandwich fusion. The repair oligo was designed for λ-Red insertion61 of HaloTag to break apart the protospacer, with 50 nt overlap on either end of the codons for residues 18–19 (Supplementary Table 3). Silent point mutations with comparable codon usage were also picked in this homology arm to edit the sequences of the protospacer and PAM to help select for a successful insertion of HaloTag.Plasmid constructionpJM44 was constructed through using InFusion Cloning. Briefly, a pACYC-sgRNA backbone was obtained from Dr. Glenn Hauk and inverse PCR was used to create a backbone (Supplementary Table 3). A short oligo was designed to insert the 20 bp protospacer in front of the guide RNA sequence, and an InFusion reaction was run to insert the sequence based on 15 nt homology overlap with the template. The plasmid was then transformed into chemically competent Stellar competent cells (E. coli HST08) and outgrown in SOC for 1 h at 37 °C. Cells were plated on LB + 150 μg/ml chloramphenicol (CAM) and incubated overnight at 37 °C. Colonies were screened by colony PCR of the sgRNA and confirmed via sanger sequencing. The plasmid was purified via ThermoScientific’s GeneJET Plasmid Miniprep Kit.JM136 constructionCRISPR/Cas9 was performed with a method similar to previous work62. TB28 harboring pKD46 and pJM25 was grown at 30 °C overnight in LB + 60 μg/ml carbenicillin (CB) + 50 μg/ml kanamycin (KAN). Those cells were diluted the following day 1:100 in 50 ml LB + 60 μg/ml CB + 50 μg/ml KAN and grown for 2 h or until their OD600 reached ~0.5. Once the culture hit the proper OD600, 0.2% arabinose was added to the culture to induce Cas9 for 1 h. After induction, these cells were prepared to be electrocompetent such that fresh, concentrated E. coli could be immediately electroporated with 2 μl concentrated sgRNA plasmid pJM44 and 10 μl repair oligo. Cells were outgrown for 1.5 h at 30 °C in SOC media then plated on LB + 50 μg/ml CB + 50 μg/ml KAN + 150 μg/ml CAM to incubate for 2 days at 30 °C.Colonies were screened by colony PCR of the chromosomal locus of ftsI. Overnight cultures were started of any hits at 37 °C in LB + 0.2% arabinose + 6% sucrose to kick out all plasmids. Subsequently, these overnights were serial-diluted to a factor of 10−6 and plated on LB plates that were then incubated overnight at 37 °C. Twenty random colonies were selected and screened to ensure all plasmids were excised. A final colony with all three plasmids kicked out would be checked by PCR once more and sequenced to ensure that the locus was correct.pRM027 constructionThe plasmid pRM027 (PT5-lac::meos3.2-ftsI, aph) was constructed by amplifying meos3.263 and ftsI genes from pJB106 and pVS155-FtsI40, respectively, and inserted to the linearized backbone from vector pCH02740 using the Infusion protocol (Clontech Inc.). The cat cassette was then replaced by an aph cassette amplified from pKD13.Preparing cells for single molecule trackingJM136 harboring pXY018 was grown overnight from a single colony at 37 °C in M9 + 150 μg/ml CAM. The following day, the culture was propagated 1:100 into fresh M9 + 150 μg/ml CAM and grown for 16 h at 25 °C. The following morning, 3 ml of culture was harvested at mid-log phase. Cells were concentrated to 100 μl and incubated with 10 nM of Janeliafluor 646 (JF646) for 30 min. After labeling, JM136 was washed three times with M9 medium without vitamins (M9−) and concentrated to 50 μl.EC812 harboring pRM027 was grown overnight from a single colony at 37 °C in LB + 150 μg/ml CAM and 50 μg/ml KAN with 0.2% l-arabinose. The culture was reinoculated 1:100 into fresh M9 media overnight at room temperature to log-phase for imaging. mEos3.2-FtsI supports cell growth and single molecule tracking at basal-level expression.A 3% agar pad was prepared using a nanopillar chip with pillars of a diameter range of 1.2–1.4 μm and a length of 4.5 μm. After cooling for 30 min, cells were added to the agar pad and incubated for 2 min. The agar pad was then washed by adding 1 ml M9 and incubating the agar pad for 2 min. The media was aspirated off and the agar pad set out to dry at room temperature for ~20 min. The agar pad was sandwiched with a coverslip, sealed in a Bioptechs FSC2 chamber, and taken to the microscope for imaging.Microscope and imaging setupJM136 harboring pXY018 was imaged using two split channels in widefield on an Olympus IX-71 microscope. JF646 was imaged with 647 set to ~50 W/cm2 and GFP-ZapA was imaged with 488 set to ~5 W/cm2. The channels were split with an Optosplit II system containing 600 rdc, with a 700/55 emission filter for JF646 and 540/30 emission filter for GFP-ZapA. We used an Andor iXon 897 Ultra EM-CCD camera with an APON100×OTIRF objective (1.49 NA/oil) and engaged 1.6× optivar. Our camera’s EM-Gain was turned on to 300 with a pre-amplifier gain setting on 3 and digitizer set to 16-bit. Baseline clamp was activated, with baseline offset set to 100.Phase contrast imagingTB28 and JM136 were grown overnight from single colonies at 37 °C in M9. The following day, the cultures were propagated 1:100 into fresh M9 and grown for 16 h at 25 °C until mid-log phase. Five hundred microliter of each culture was pelleted and concentrated in 50 μl. 0.5 μl of each strain was added to separate 3% M9 agar pads (pre-set for 30 min before use), sealed in a BiopTechs FSC2 chamber, and taken to the microscope for imaging. Cells were imaged with phase contrast on an Olympus IX-71 microscope with a 100×/1.30 NA Oil Ph3 objective and engaged 1.6× optivar. Images were recorded at a full 512 × 512 region on an Andor iXon Ultra Em-CCD camera (above) at 100 nm/pixel. Phase contrast images were recorded with 100 ms exposures and processed with Oufti to measure cell length64.Single molecule tracking of Halo-FtsI::JF646All microscopy was recorded using Metamorph software. A region of 300 pixels wide and 512 pixels tall was used to capture both channels. Samples were put on the microscope and acclimated for 30–60 min before imaging began to minimize axial drifting. Twenty-five regions were selected with cells in microholes, and an automated journal would run to (1) autofocus on cells lying on the surface of the agar pad, then (2) move up into the sample 2 μm to find the Z-ring of cells in microholes and (3) autofocus on the GFP-ZapA ring. Once a focal plane was set, the journal would record a 400-frame movie, where each frame encapsulated 1 s (500 ms exposure, 500 ms dark time).Aligning channels to correct for chromatic aberrationTo correct for chromatic aberration, we imaged TetraSpeck fluorescent beads in both channels using a 50 ms exposure with the same 300 × 512 region. 647 remained set to ~50 W/cm2. 488 was set to ~40 W/cm2. To align the channels, we cropped the two using a custom Matlab script that then used the imregister function to align the 488 (ZapA-GFP) channel to the 647 channel (JF646). The dimensions of the crop and the transformation matrix from this alignment were used to crop all channels and align the GFP-ZapA channel to JF646.Analyzing single molecule trajectoriesCropped 647 channels were first processed with ThunderSTORM65, a plug-in for ImageJ66. Image filtering used a wavelet filter (B-spline) with an order of 3 and scale of 2.0. A local maximum was used to localize the molecules with 1.5*std(Wave.F1) used to identify peak intensity threshold and a connectivity of 8-neigborhood. Sub-pixel localization used a guassian PSF with a 3-pixel fitting radius and an initial sigma of 1 pixel. The post-processed data was filtered to exclude intensity values less than 300 and a sigma bandpass filter of 60–300 nm. All analysis thereafter used custom scripts in Matlab R2019a. The localizations were linked to trajectories using a nearest-neighbor algorithm modified from Sbalzarini and Koumoutsakos67. To link molecules which may have blinked across frames or left the focal plane, a time threshold of 15 frames was applied. The distance threshold was 300 nm/1 frame, which approximates to a diffusion coefficient of ~0.05 μm2/s, or a max speed of 300 nm/s. Only trajectories with a corresponding GFP-ZapA ring were chosen for the next step.GFP-ZapA stacks served as both a marker to autofocus for imaging as well as for estimating the Z-ring diameter. Maximum-intensity projections were taken of movies with cells expressing GFP-ZapA and then subsequently aligned to the 647 channel as described above. Cells of interest were cropped out and a circle was then fit to the intensity profile of the GFP signal. Using the diameter, the real position of FtsI along the cell envelope can be back calculated and estimated. The trajectories were then manually segmented into mobile states only when segments were a minimum of four frames (4 s) long with processive displacements in one direction. The selected segments were fit to a straight line to minimize noise and classified as “processive” or “stationary” based on the classification description detailed in the following section. Processive segments were used to obtain the velocity, dwell time, and persistent length measurements.Segmentation classificationThe trajectories were first segmented manually (a segment contains at least four data points and with consistent noise). After segmenting trajectories (Supplementary Fig. 5), we classify whether the segments move processively or remain stationary. Included with our segmentation are a set of observables, {v, d, l, r}, where v is the slope of a linear fit of the segment, d is the total displacement, l is the trajectory length, and r is the standard deviation of all positions against the linear fit. Note these four parameters are not independent since d = l · v. We subsequently combine the parameters to {v, l, R}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R = \frac{r}{d}$$\end{document}R=rd, a dimensionless quantity representing the relative level of the residual to displacement (a noise to signal parameter). Through manual inspection, we determined to classify segments as processive based on a threshold of R ≤ 0.4.On average, one nanopillar experiment in a day (a biological replicate) will yield 0–5 trajectories. Single cells can give up to 1–3 usable trajectories for analysis. The 77 total trajectories used in this paper come from 18 biological replicates and 49 total cells. They yielded 232 total segments, 139 of which were marked as processive by this R threshold.Deconvolution of the FtsI fast populationThe data presented in Fig. 4f, g, h were representative of the fast, inactive population of FtsI from a two-population fit to a log-normal CDF (Supplementary Fig. 5b–d)16.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f\left( x \right) = P\left( {\frac{1}{{x\sigma _1\sqrt {2\pi } }}} \right)e^{ - \frac{1}{2}\left( {\frac{{\ln \left( x \right) - \mu _1}}{{\sigma _1}}} \right)^2} + \left( {1 - P} \right)\left( {\frac{1}{{x\sigma _2\sqrt {2\pi } }}} \right)e^{ - \frac{1}{2}\left( {\frac{{\ln \left( x \right) - \mu _2}}{{\sigma _2}}} \right)^2},$$\end{document}fx=P1xσ12πe−12lnx−μ1σ12+1−P1xσ22πe−12lnx−μ2σ22,where P is the percent makeup of the respective population. Briefly, we bootstrapped the FtsI data 200 times, iteratively fitting the CDF to obtain estimates for each fit parameter. A global fit of the mean bootstrapped values is presented in Supplementary Fig. 5b, and individual fits for each population are shown with respect to FtsZ (from Yang et al.13) in Supplementary Fig. 5c. We then used these parameters to overlay the raw FtsI histogram of 139 directional segments (Supplementary Fig. 5d). Using the slow population fit, we resampled the FtsI data 100 times by removing segments that would be selected for a bin determined by the slow population’s percent makeup compared to the original histogram. The final histogram and scatterplots in Fig. 4f, g, h are representative plots from the resampling, where 74 segments remain that are likely to be fast-moving.Three-dimensional (3D) PALM single-molecule tracking and analysisAstigmatism-based 3D single-molecule tracking was performed on the same microscope. The 568 nm excitation laser power was set to 500 W/cm2 with a 30 ms exposure time for 5000–10,000 frames of continuous acquisition. During the imaging, 0–1 W/cm2 405 nm activation light was increased stepwise and applied to activate mEos3.2-FtsI molecules. The UV power was tuned from sample to sample to maintain a low enough number of red-emitting molecules in each cell (<1 spot/frame/cell) for single-molecule localization and tracking.Localization of single molecules was determined using the ThunderSTORM plugin in ImageJ65,66. Molecules with anomalous brightness or uncertainty (>3σ) were filtered out. Molecules were tracked across frames using custom Matlab scripts implementing the tracking algorithm described in Sbalzarini et al.67. To calculate the MSD, all trajectories longer than four frames were selected, and the squared displacements were calculated in 2D. Consecutive frames in the trajectories were used for displacement calculation. All three MSD curves were fitted by the anomalous diffusion function MSD2D = 4Dtα + D0 since these molcules diffuse on a 2D curved membrane surface. D0 reflects the localization uncertainty under the imaging conditions.Single molecule tracking of PBP2b in B. subtilisB. subtilis strain bGS28, in which the native Pbp2B has been replaced with an IPTG-inducible, Halo-tagged Pbp2B, was imaged as described in Bisson-Filho et al.12. Briefly, cells were grown in CH media, Pbp2B was induced with 20 μM IPTG and labeled with 100 nM JF549 conjugated to HaloTag ligand, and cells were immobilized under an agarose pad for imaging. Cells were imaged in TIRF on a Nikon N-STORM microscope; time lapses were acquired with streaming 30 ms exposures for 1 min. Particle tracking was performed in TrackMate using the simple LAP tracker with the following settings: particle diameter was 300 nm, maximum linking distance was 300 nm, and no frame gaps were allowed. Tracks between 5 and 25 frames were analyzed further using custom MATLAB code12. MSD vs. t was calculated for each track, and the diffusion coefficient was computed by the MSD equation noted above.Single molecule tracking of FtsW in Streptococcus pneumoniaeS. pneumoniae strain IU15096 (D39 ∆cps rpsL1 ftsW-L0-ht-PC-erm) expresses a derivative of FtsW from its native chromosomal locus, with the carboxyl-terminus of FtsW fused to a 10-amino acid linker (L0) connected to the HaloTag (HT) domain41. Frozen cultures were inoculated into 4 mL of brain heart infusion (BHI; BD Bacto, 237500) broth and serially diluted. Cultures were incubated statically without shaking at 37 °C in 5% CO2 for 12–16 h. Following incubation, cultures with an optical density at 620 nm (OD620) between 0.1 and 0.4 were washed in C+Y, pH 7.1 medium, and diluted to OD620 = 0.01 in 5 ml of C+Y, pH 7.1 medium. Growth was monitored by measuring OD620 every 45 min. While cells are growing, microscope slides (VWR, 16004-368) were cleaned with 70% (v/v) ethanol and a Gene Frame (Fisher, AB0576) was affixed. A 1.5% (w/v) mixture of agarose (Sigma BioReagent, A9414) in C+Y, pH 7.1 medium was melted, and agarose pads were constructed by filling the middle of the Gene Frame with the melted agarose mixture, covering with a second microscope slide, and placing at 4 °C for 45 min. After the agarose pad had solidified, and immediately prior to coverslip attachment, the top slide was removed, and a ≈ 2 mm strip running down the center of the pad was cut out and removed with a clean razor blade. When the cell culture OD620 reached ≈0.1, 500 µL of cells were transferred to a 1.5 ml tube and Janelia Fluor® 549 HaloTag ligand was added to a final concentration of 120 pM. Cells were vortexed, spun briefly to collect liquid, and incubated for 15 min at 37 °C in the dark with shaking at 300 rpm. Cells were centrifuged at 21,000 × g for 5 min at room temperature (RT), washed twice with spinning in 0.9 ml of 37 °C–C+Y, pH 7.1 medium, and resuspended in 500 µl of 37 °C-C+Y, pH 7.1 medium. 1.2 µl of cells were pipetted onto coverslips (VWR, 48366-227) that had previously been soaked in a solution of 25% (v/v) concentrated HCl, 25% (v/v) sterile distilled H2O, and 50% (v/v) ethanol for 12–24 h, then submerged 5× in sterile distilled H2O, rinsed with fresh sterile distilled H2O, and dried with lens paper (VWR, 52846-001). Cells were allowed to dry on the coverslips for 1–2 min, then the coverslip was attached to the Gene Frame with the cells aligned over the agarose pad. Slides were incubated in the dark at 37 °C for 30 min prior to imaging.TIRFm was performed using the Deltavision OMX Super Resolution microscope (GE Systems), using a PCO.edge 4.2 (CMOS) camera (Kelheim Germany) and an Apo N60X/1.49 TIRF objective (Olympus). The laser line was 561 nm (emission filter 609–654) for FtsW-HT. DIC images were collected on a separate channel. Images were collected every 50 ms (20 FPS) for 100–200 s. The 561 nm (FtsW-HT) channel used a 21 ms exposure, 100% T, and a TIRF angle of 91.2°, while the DIC channel used a 3 ms exposure, 50% T, and an angle of 0° (epi). Deltavision SoftWoRx (GE Healthcare) was used to align the channels after image acquisition. Single cells were cropped from an image, and the FtsW-HT channel was processed using the FIJI plug-in ThunderSTORM65,66. Image filtering used a Wavelet filter (B-Spline) with an order of 3 and a scale of 2. The localization method was Local Maximum, with a peak intensity threshold of 1.5*std(Wave.F1) and 8-neighborhood connectivity. Sub-pixel localization used a PSF: Gaussian method with a fitting radius of 3 pixels, fitting method of Weighted Least Squares, and an Initial sigma of 1 pixel.Images depicting single molecule localizations of FtsW-HT from a single cell were visually inspected to ensure each frame had a maximum of one localization. Additional localizations resulting from single pixel noise were manually removed. A maximum of five consecutive frames without a localization was allowed within a single trajectory. From individual cells, time averaged MSDs were calculated as noted above on trajectories longer than 15 framesReporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Single-molecule biophysics",
"Cellular microbiology"
] | cell wall constriction in Gram-negative bacteria new septal peptidoglycan (sPG) synthesis old cell wall degradation occur cell wall enzymes regulators identified unclear how proteins cytokinesis maintain structural integrity septal cell wall2,3 Perturbations of PG remodeling at septum compromise cell division lead to cell lysis4 studies FtsZ essential bacterial cell division machinery coordination sPG synthesis enzymes conserved bacterial tubulin homolog GTPase5–7 cell division polymerizes at cytoplasmic face inner membrane ring-like structure (Z-ring) at mid-cell8–10 Z-ring recruits 30 proteins sPG-remodeling enzymes1 to initiate septal cell wall constriction studies FtsZ polymers exhibit GTP hydrolysis-driven treadmilling dynamics continuous polymerization depolymerization monomers stationary FtsZ’s dynamics drive movements of sPG transpeptidase glycosyltransferase FtsW16 proposed FtsZ’s dynamics distribute sPG synthesis enzymes along septum plane smooth septum morphogenesis13unknown FtsZ’s treadmilling dynamics cytoplasm periplasm movement cell wall synthesis enzymes role sPG synthesis elusive cell wall constriction rate on FtsZ’s treadmilling speed in B. not E. coli13 S. aureus17 combined agent theoretical modeling single-molecule imaging testing FtsZ treadmilling-dependent movement sPG enzymes role bacterial cell division Brownian ratchet mechanism underlies movement sPG synthesis enzyme molecules driven FtsZ’s dynamics FtsI model enzyme processivity Brownian ratchet on potential between FtsI FtsZ modulated by balance FtsI’s diffusion FtsZ’s treadmilling speed predictions bacterial species FtsZ treadmilling machinery distinct processivities sPG enzymes sPG cell wall constriction lack linear stepping motors prokaryotic world work suggests framework for polymer dynamics Brownian ratcheting directional transport cargos shaped by evolution cellular milieusResultsModel based on Brownian ratchet FtsZ’s treadmilling asymmetry diffusion FtsI molecules periplasm FtsI follows shrinking treadmilling FtsZ filament (Fig. 1) quantitative details rooted in physical chemical properties components system characterized experiments. 1Model sPG synthase interaction with FtsZ treadmilling FtsZ in cytoplasm FtsI transmembrane protein dissociate from membrane FtsZ–FtsI interaction at septum FtsZ filament (purple FtsZ subunit associating new right FtsI (gray) diffuses to FtsZ subunits FtsZ–FtsI binding potentials binding potential harmonic short-ranged (~5 size FtsZ subunit no energy barrier for FtsI to bind to FtsZ FtsI binds to FtsZ subunit binding potential energetic barrier preventing FtsI diffusing Fig. 1 describes movement free FtsI molecule at septum quasi-1D assumes FtsI single transmembrane domain cytoplasmic tail can diffuse inner membrane or interact with treadmilling FtsZ filamentdynamics FtsI molecule at septum determined by three parameters constant FtsI’s free diffusion treadmilling speed of FtsZ filaments attraction force potential (U between FtsI FtsZ (Fig. 1b, ranges parameters diffusion constant 10−3 to 10−1 μm2/s typical inner membrane protein in bacterial cells18 PBP2 FtsI measured ~0.06 μm2/s18 average treadmilling speed FtsZ ~20–40 nm/s in vivo faster in vitro range 10–100 nm/s14 FtsI interacts with FtsZ indirectly through protein–protein interactions FtsN FtsA FtsEX1 interaction between FtsI and FtsZ indirect interaction between FtsZ monomer FtsI molecule attractive force short-ranged harmonic binding potential (Fig. 1c). potential range ±2.5 nm size FtsZ monomer (~5 nm magnitude ~10 kBT Kd in μM range typical for protein–protein interactions bacterial divisome FtsI in E. coli model sPG enzyme analysis other sPG enzyme proteinssimulate treadmilling FtsZ filament (Fig. 1b). model depicts filament shrinking growing Eqs. (1) (2) positions shrinking growing ends filament time t xS(t) xG(t) related treadmilling rate VZ:1\documentclass[12pt]{minimal{amsmath{upgreek-69pt}\frac{{{\rm{d}}x_S\left t \right)}} = V_{\rm{Z}}{document}dxStdt=VZ,2\documentclass[12pt]{minimal}{amsmath{upgreek}\oddsidemargin-69pt}{{{\rm{d}}x\rm{G}}\left t \right)}}{d}}t}} = V_{\rm{Z}}{document}dxGtdt=VZ model simulates treadmilling events every 5 nm/(Vz·Δt) time stepFtsZ subunit falls off shrinking end filament binding potential vanishes adds growing end potential appears treadmilling speed FtsZ filament drawn from measured model stochastic on-reactions off-reactions subunits FtsZ filament reflected in treadmilling speed distribution FtsZ filament length at 50 monomers (250 nm) treadmilling speed independent of filament length previous biochemical studies recent in vivo model considers one FtsI molecule one FtsZ filament in septal section expanded multiple FtsI molecules per FtsZ filament Fig. dynamics of FtsI by Langevin-type equationviscous drag force molecule with driving force f thermal noise\documentclass[12pt{amsmath\oddsidemargin-69pt\lambda\left = f\left +{document}λdxtdt=fxt+ξt x(t) represents location FtsI molecule time t 1D septum λ viscous drag coefficient FtsI’s movement λ = kBT/D D diffusion constant FtsI molecules f(x(t)) attractive force FtsI molecule FtsZ’s potential U(x, t) (Fig. 1c). f(x(t)) = –∂U(x, t)/∂x at time t last term ξ(t) reflects random diffusive motion FtsI inner membrane 〈ξ(t)·ξ(t′ = 2D·Δt·δ(t–t′), Δt unitary time step simulation FtsI molecule diffuses after FtsZ subunit falls off end flat 5-nm local potentialFtsI molecule diffuse to next FtsZ subunit ~5 nm associate due to binding potential could diffuse dissociate from FtsZ filament process not deterministic stochastic FtsZ filament treadmills left to right length ~250 nm model implements open boundary conditions on FtsI molecule at left and right edges right-ward FtsZ treadmilling not limited model results reflect nominal case robust against variations data Brownian ratchet links directional movement to treadmillingAs hinges on diffusion FtsI interaction with FtsZ FtsZ’s treadmilling speed movement diffusion binding potential kept FtsZ’s treadmilling speed constant 25 nm/s phase diagram study simulations chose parameter range 0.0001 to 0.1 μm2/s for FtsI’s diffusion18 upper limit binding potential ~20 kBT dissociation constant Kd in nM-range initial condition shrinking end FtsZ filament FtsI molecule at left boundary septal section FtsI trajectory moving directionally if tracked shrinking end FtsZ filament persistently for 4 s.stochastic Brownian ratcheting characterized FtsI as persistent end-tracking if 50% FtsI trajectories displayed movement. 2a binding potential between FtsZ and FtsI weak (<5 kBT FtsI displayed random diffusion without directional movements attraction potential strong (>5 kBT), strong binding quenched diffusion confined FtsI to end FtsZ filament FtsZ subunit at fell off FtsI to diffuse next FtsZ binding potential short distance away (~5 nm). FtsI molecule diffused closer to new end subunit trapped FtsI “pulled” to right by ~5 nm subsequent FtsZ subunits falling off FtsI molecule ratcheted forward tracked end of FtsZ filament resulted in persistent directional trajectory of FtsI (Fig. 2b). directional movement FtsI not deterministic probabilistic due to stochastic Brownian ratcheting FtsI molecule may decouple from FtsZ filament average speed of FtsI molecules coupled to FtsZ’s treadmilling speed (Figcorrelation between FtsI’s directional motion treadmilling speeds in wildtype GTPase FtsZ treadmilling ratchet mechanism drives movement Phase diagram FtsI’s end-tracking on diffusion constant binding potential trajectory FtsI molecule treadmilling FtsZ filament zoomed-in view model parameters FtsZ’s treadmilling speed VZ 25 nm/s FtsI diffusion constant D = 0.04 μm2/s FtsZ–FtsI binding potential U = 10 kBT simulation time step 5 × 10−6 s full trajectory plotted every 5 10−2 s zoomed-in inset 5 × 10−6 s FtsI’s directional speed couples with FtsZ’s treadmilling speed >80 trajectories FtsZ’s treadmilling speed fixed varied across ensemble standard deviation) 30% measurements GTPase constant binding potential between FtsZ FtsI persistent end-tracking FtsI diffusion constants If FtsI rapidly confined by binding potential shrinking FtsZ filamentFtsI diffused slowly keep up with speed departing FtsZ subunits at shrinking end FtsI molecule lost contact with FtsZ subunits alternative condition FtsI molecule binds in middle of FtsZ filament same end-tracking behavior bound FtsI molecule end-tracking when shrinking end FtsZ polymer mobilizes modeling coupling to growing end FtsZ filament produce directional movement of FtsI molecules no biochemical evidence FtsI-binding potential of newly added FtsZ subunit at growing tip higher than middle filament addition new FtsZ subunit bias diffusion FtsI molecule original tip slow-diffusing FtsI molecules stuck in local binding potential unable catch up with addition new FtsZ subunits fast-diffusing FtsI molecules high probability escaping from tip no FtsZ subunits beyond growing tip FtsI track growing end of FtsZ filament analysis showed end-tracking Brownian ratchet mechanism FtsI’s directional movement to FtsZ’s shrinking end consistent with experimentally measured datamodel nondirectional movement cytoplasmic tail FtsN divisome protein in recent in vitro tail FtsN treadmilling FtsZ filaments lipid bilayer ensemble level single molecule level FtsN tail binds unbinds FtsZ filaments transiently directional Brownian ratchet model diffusion FtsN tail membrane too large (0.3–0.6 μm2/s treadmilling speed modulates FtsI’s end-tracking investigated FtsZ’s treadmilling speed impacts processivity FtsI’s directional movement shrinking end role FtsZ’s treadmilling dynamics sPG synthesis activity focused on three features FtsI’s end-tracking propensity run distance duration time end-tracking trajectories examined relative propensity FtsI’s end-tracking by FtsZ treadmilling speed propensity defined percentage FtsI end-tracking trajectories at each FtsZ treadmilling speed normalized by total number trajectories speedsdiffusion FtsI 0.04 μm2/s binding potential 10 kBT simulations Brownian ratchet model persistent end-tracking trajectories FtsI treadmilling speed. FtsZ fast FtsI track diffusive predictions FtsI movement modulated FtsZ treadmilling speed Predicted propensity FtsI persistent end-tracking treadmilling speed FtsZ treadmilling speed 100 independent FtsI trajectories simulated FtsI persistent end-tracking trajectories counted criteria persistent end-tracking distance between FtsZ shrinking end FtsI less than 100 nm FtsI follows FtsZ greater than 4 s 664 out of 1000 FtsI trajectories persistent end-tracking FtsZ speed normalized probability end-tracking phase diagram FtsI persistent end-tracking diffusion constant treadmilling speed Predicted FtsZ treadmilling speed-dependence duration FtsI persistent end-tracking FtsZ treadmilling speed-dependence run distance FtsI persistent end-tracking shaded regions maximum persistent run distance dependent treadmilling speedmodel calculations FtsZ–FtsI binding potential 10 kBT calculated phase diagram FtsI’s persistent end-tracking propensity function diffusion constant FtsZ’s treadmilling speed (Fig. binding potential fixed 10 kBT used 50% FtsI persistent end-tracking trajectories criterion phase boundary fixed diffusion constant FtsI upper limit FtsZ’s treadmilling speed track fixed FtsZ treadmilling speed end-tracking appropriate range diffusion constants large diffusion constants (>0.1 μm2/s) support persistent end-tracking treadmilling speed results consistent with phase diagram Fig. 2a in vitro study FtsN’s cytoplasmic tail25 investigated FtsZ’s treadmilling speed modulates run distance duration time persistent end-tracking Brownian ratchet model predicted run length duration time broad distributions stochastic FtsI’s diffusion interaction FtsZ’s treadmilling speed increases duration time end-tracking run distance biphasic increases intermediate FtsZ treadmilling speed decreases FtsZ’s treadmilling speed increasesof duration time run distance on FtsZ’s treadmilling speed consequence of Brownian ratchet mechanism FtsZ subunit falls off to cytoplasm from filament FtsI molecule dissociate diffuse or catch up with next FtsZ subunit on FtsZ treadmills When FtsZ treadmills too fast >30 nm/s), difficult for FtsI to catch up in early termination of end-tracking persistent run distance and duration time short When FtsZ treadmills slowly (<30 nm/s), probability of FtsI catching up with high slower FtsZ treadmills fewer dissociation events lower chance for FtsI to diffuse away longer duration of persistent run persistent run distance proportional to FtsZ’s treadmilling speed slower FtsZ treadmills shorter FtsI’s persistent run distance extreme case FtsZ treadmill (Vz = duration time persistent runs longest dictated by dissociation rate of FtsI from FtsZ persistent run distance shortestsize FtsZ subunit). FtsZ treadmills fast (Vz = duration distance persistent runs zero FtsI track analytical proof “Methods” section Supplementary Fig. 3.Single-molecule tracking FtsI confirms model FtsI’s movement treadmilling speed single-molecule tracking functional sandwich fusion protein Halo-FtsISW JF646 live E. coli interactions FtsI’s-terminal tail inserted Halo tag between last residue (18) first residue inner membrane helix (19) FtsI (Fig. integrated halo-ftsISW fusion gene chromosome ftsI gene full fusion protein supported normal cell division sole source FtsI (Supplementary Fig. 4 Table 2).Fig. 4Experimental characterization FtsI directional motion Schematic functional sandwich fusion FtsI Halo tag inserted between residues 18 19 first residue transmembrane Diagram E. coli cells microholes nanopillars Brightfield fluorescence images microholes E. coli cells GFP-ZapA Halo-FtsISW fusion proteins zoomed image one cell yellow boxBlue circles pixels fit ZapA-GFP circle Scale bar 5 μm Experiment repeated 18 times similar results Circle-fitting GFP-ZapA image super trajectory single FtsI molecule colored time unwrapped trajectory D fitted lines segment directional speeds run duration distance Distributions FtsI’s speeds 77 trajectories 74 events FtsZ’s treadmilling speed Yang et al.13) plotted fast-moving population FtsI slow-moving population independent FtsI’s FtsZ’s histograms bootstrapped 100 times shaded standard error bars two-sample Kolmogrov–Smirnov test no difference distributions (p = 0.1852) Dependence duration FtsI’s persistent run speed distance precise measurements run distance duration Halo-FtsISW molecules trapped E. cells vertically agarose microholes cell-shaped nanopillar molds entire circumference septum visualized focal plane (Fig. 4b). Halo-FtsISW molecule septum labeled FtsZ-ring ectopically expressed GFP-ZapA fusion protein marker Z-ring localizationGFP-ZapA image circular trajectories FtsI-Halo molecules septum (Fig. 4d to linear displacements circumference (Fig. persistent run speed distance duration time (Fig. 4f g Fig. 4f directional motion speed Halo-FtsI wide distribution similar FtsZ’s treadmilling similarity suggests E. FtsI can end-track treadmilling FtsZ filaments persistence run distance duration time same trends model run duration time decreased (Fig. persistence run distance increased decreased when FtsI’s speed increased (Fig. 4h). inferred FtsZ’s treadmilling speed from FtsI’s directional motion speed difficulty two-color co-tracking experiment linearly very fast FtsZ treadmilling speed >80 nm/s) rare in wildtype E. coli cells small dataset high speed FtsZ treadmilling data determine FtsI end-track FtsZ filaments fast treadmilling speeds agreement experimental measurements with theoretical predictions supported validity Brownian ratchet model sPG enzymatic activity on FtsZ treadmillingIncoli septal PG synthesis septum constriction rate insensitive to perturbations FtsZ’s treadmilling speed ~8 to nm/s in FtsZ GTPase insensitivity suggests end-tracking FtsI molecules inactive in sPG synthesis second slow-moving population FtsW FtsI (~8 nm/s) independently FtsZ’s treadmilling active sPG synthesis in E. in S. pneumoniae FtsW TPase PBP2x independently FtsZ’s active population sPG synthase In B. subtilis cell wall constriction speed correlated with FtsZ’s treadmilling speed faster FtsZ treadmills higher sPG synthesis FtsZ treadmilling dynamics different sPG synthesis activity in species FtsI molecules tracking FtsZ filaments likely not available for sPG synthesis FtsI’s off-rate represents for sPG synthesis proportional to sPG synthesis rate FtsZ’s treadmilling speed could modulate sPG synthesis in bacterial species enzyme’s diffusion coefficient binding potentialsmodulation difference between E. coli B. subtilis Fig. 5a model suggests FtsI diffuses fast~0.05 μm2/s lifetime FtsZ-bound FtsI insensitive to treadmilling speed ~8 to 25 nm/s FtsI diffuses slowly lifetime dependent on FtsZ treadmilling speed diffusion 0.005 μm2/s lifetime FtsI decreased ~70% speed increased ~8 to 25 nm/s (Fig. 5a 5Diffusion binding potential modulate sPG processivity treadmilling speed FtsI lifetime on treadmilling speed binding potential 10 kBT FtsI diffusion coefficients diffusion fast (0.05 μm2/s). slow (0.005 μm2/s). weaker binding potential (8 kBT) stronger binding potential (12 kBT) FtsI randomly positioned along FtsZ filament lifetime of FtsZ-bound FtsI defined as first time escapes from filament than 100 nm model results plotted every 2 × 10−2 s simulation time step 10−5 sMean-squared displacement of FtsI PBP2b FtsW outside septum in E. coli B. subtilis S. pneumoniae diffusion constants FtsI 0.041 0.038 0.028 μm2/s α = 0.29 0.51 0.71 Error bars standard error mean Model sPG synthesis FtsZ treadmilling speed in E. coli data Coltharp et al.42 Yang et al.13 > 30 sPG synthesis FtsZ treadmilling speed B. subtilis data Bisson-Filho et al.12 standard deviation n = 100 cells > 50 cells g h used diffusion constants varying binding potentials lower upper limits (E. coli: 8 9 kBT B. subtilis: 10 12 kBT), shaded regions difference between fast slow FtsI diffusion Brownian ratchet mechanism Fast diffusion FtsI catch up shrinking FtsZ filament short time (Fig.FtsI’s diffusion slower rate-limiting factor in Brownian slow FtsI molecule falls behind FtsZ shrinking end diffuses lost (Fig. increasing FtsZ treadmilling speed FtsI keeping up FtsZ lifetime FtsZ-bound FtsI diffusion dependence sPG synthesis on FtsZ speed hinges on binding between FtsI FtsZ binding potential reduced lifetime FtsZ-bound FtsI less sensitive to FtsZ speed Fig. 5a FtsZ’s treadmilling speed sPG synthesis depends on sPG synthase’s diffusion coefficient FtsZ-binding potential performed fast frame-rate single-molecule tracking diffusion coefficients of free sPG synthase molecules outside septum B. subtilis FtsZ localizes mid-cell during cell division sPG synthase molecules not free not interacting with FtsZ measured diffusion coefficients dependence cell wall constriction rate on FtsZ treadmilling speed in E. coli B. with normalized off-rate model only free parameter binding potential not possible to measure in live cells Figdiffusion constants FtsI in E. coli PBP2b in B. subtilis measured ~0.041 ± 0.0051 5049 0.038 ± 0.0019 μm2/s 6765 binding potentials at 8–9 10–12 kBT model recapitulated differential dependence cell wall constriction rate on FtsZ’s treadmilling speed species higher potential PBP2b to FtsZ in B. subtilis interactions tighter coupling between PBP2b molecules with FtsZ E. coli FtsI modulated by FtsZ’s treadmilling speed B. subtilis FtsI in E. coli measured diffusion FtsW in S. pneumoniae apparent diffusion coefficient FtsW 0.028 ± 0.0004 μm2/s 21 same as E. coli B. subtilis FtsW follow treadmilling FtsZ in S. pneumoniae likely binding potential between FtsW FtsZ lower than 5 KBT predicted model interesting to investigate other divisome proteins independent of FtsZ’s treadmilling conditional dependence interactions altered bindingpossibilities investigated future work presented data Brownian ratchet model movements sPG synthases FtsZ’s treadmilling differential sensitivity sPG synthesis treadmilling speed E. coli B. subtilis sPG synthase molecule FtsI follow treadmilling FtsZ polymer shrinking tip not growing tip interplay diffusion potential FtsZ diffusion range (~0.001–0.1 μm2/s sufficient binding potential (>5 exhibit FtsZ-treadmilling-dependent directional movement persistent run duration distance FtsI FtsZ’s treadmilling speed predicted Brownian ratchet model single-molecule tracking confirmed model predictions treadmilling FtsZ polymers modulate run duration distance regulating spatial distribution synthases correct septum shape Brownian ratchet model explain differential dependence sPG synthesis activity FtsZ’s treadmilling speed in E. coli B. subtilis S. pneumoniae Brownian ratchet model predicts tighter binding PBP2b FtsZ B. subtilis cause tighter couplingPBP2b molecule on FtsZ off for sPG synthesis modulated by FtsZ’s treadmilling speed In E. coli or S. pneumoniae binding potential between FtsI/FtsW and FtsZ high B. subtilis FtsI/FtsW molecule end-tracking FtsZ less sensitivity or to treadmilling speed Different binding potentials between bacterial species likely molecular interactions septal ring complexes distinct same Brownian ratchet machinery in distinct regimes in different species sPG synthesis depends on FtsZ treadmilling differentially different strategies additional factors expression level of sPG synthase relative to FtsZ in B. subtilis higher than E. coli or S. sPG synthases per FtsZ filament in B. subtilis higher Brownian ratchet model enhances sPG synthesis to FtsZ’s treadmilling speed sPG synthase molecules to FtsZ polymer end-tracking off FtsZ faster FtsZ treadmills faster end sPG synthase more sPG synthase molecules dissociated from FtsZ available for sPG synthesiscontrast to single sPG synthase per FtsZ filament likely in E. coli or S. pneumoniae level cell wall synthesis precursors factor sPG synthase molecules coupled to FtsZ’s treadmilling dynamics lifetime modulated by speed if cell wall synthesis precursor likely in E. coli masked by limited level In S. pneumoniae low binding potential between sPG synthase FtsZ cell wall synthesis precursor levels independence FtsZ’s treadmilling High levels PG precursors saturate sPG synthase molecules no free ones FtsZ polymers lack of linear stepper motors Brownian ratcheting ancient mechanism for cargo transportation in example ParA-mediated DNA similar ratchet mechanism movement of mitotic chromosomes in eukaryotes59 principle evolution shapes Brownian ratchet mechanism? future study aim to obtain solution for FtsZ treadmilling speed-dependence of run distance duration of FtsI persistent end-tracking FtsI’s end-tracking trajectory decomposed into repeating steps two processesProcess (1) “stay-on” FtsI molecule stays inside binding potential FtsZ end subunit Process (2) “catch-up” FtsI catches up with next FtsZ subunit when dissociates calculate probabilities two processes FtsI “stay-on” process Fig. undergo two reactions escape from binding potential FtsZ end subunit 1/τD 1/τz τz average lifetime of FtsZ subunit proportional to FtsZ treadmilling speed\documentclass[12pt]{minimal}{amsmath}-69pt}{document}\rm{Z}} =}τZ=5nmVZ τD average duration of FtsI staying in binding potential if FtsZ subunit never falls offprobability FtsI “stay-on” time\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{upgreek\oddsidemargin-69pt}\begin{document}$$P_1 = {\rm{exp}}\left\frac{1}{{\rm{D}}}} +\frac{1}{{{\rm{Z}}}}}\end{document}P1=exp−1τD+1τZt FtsZ end subunit dissociates P1 scales\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin-69pt}\begin{document}$$P_1\mathrm{exp}}\left\frac{{\tau{\rm{Z}}}}{{\rm{D\end{document}P1~exp−τZτD.catch-up probability FtsI molecule P2 scales\documentclass[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}$$P_2\sim\left\frac{{\tau{\rm{C}}}}{{\rm{Z}}}}}\right\end{document}P2~exp−τCτZ τC average catch-up time (Supplementary Fig.3b). probability for FtsI end-tracking each repeating step is:4\documentclass[12pt]{minimal{amsmath-69pt{document}$$P = P_1P_2\sim \left\frac{{\tau{\rm{Z}}}}{{\rm{D}}}}{C}}}}{{{document}P=P1P2~exp−τZτD probability of persistent end-tracking N-repeating steps is PN (1 − P). average number of FtsI persistent end-tracking steps is:5\documentclass[12pt]{minimal}{amsmath{upgreek}}{-69pt}{document}$$\left\langle \right\rangle = \frac{{\mathop\sum\nolimits\rm{N}} = 0}\infty N P^N\left {1 - P\mathop {\sum\nolimits_{N = 0} P^N {1 - P\right\end{document}N continuum limit[12pt]{minimal\usepackage{amsmath{wasysym\oddsidemargin}{-69pt}{document}\left\langle N \right\rangle =\frac{{\mathop\smallint\nolimits_0^\infty \left {N P^N \left {1 - P}\right\rm\nolimits^N {1 - P}\right\rm\end{document}N=∫0∞N⋅PN⋅1−PdN∫0∞PN⋅1−PdN[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}{document\left\langle N \right\rangle = \frac{{\tau{\rm{D}}\rm^2 +\rm{C}}{\rm{D}}}}{document}N=τDτZτZ2+τCτD average run length duration:6[12pt]{minimal{amsmath\oddsidemargin-69pt}\begin{document}\left\langle L\right\rangle = L_0\frac{{\tau{\rm{D}}\rm{Z}}^2 +{C}}{D}}}}{document}L=L0τDτZτZ2+τCτD[12pt]{minimal}{amsmath\oddsidemargin}{-69pt}\begin{document}\left\langle T \right\rangle = \frac{{\tau _{\rm{D}}\rm{Z}}^2 +{C}}{D}}}}{{Z}}{document}T=τDτZτZ2+τCτDτZHere L0 = 5 nm length FtsZ subunit persistent end-tracking entails τD >> τZ >> τC.numerical simulation τD 60 s τC 0.001 s Eqs. 6–7 recapitulate dependences run length duration on FtsZ treadmilling speed Supplementary Fig. 3c τD = 60 s τC = 0.0003 s longer lifetime FtsZ subunit FtsI (1) higher chance escape binding potential decreasing stay-on probability longer lifetime longer time FtsI catch up increasing catch-up probability balance stay-on catch-up defines optimal lifetime FtsZ treadmilling probability FtsI end-tracking per repeating step determines average number repeating steps explains biphasic dependence FtsI run length on FtsZ treadmilling speed duration per repeating step approximately lifetime FtsZ subunit treadmilling speed increases decrease lifetime outweighs variation average number repeating steps decrease duration persistent end-tracking bacterial strains plasmidsCells grown in M9 minimal medium broth) (10% tryptone 10% NaCl 5% yeast Fresh LB plates struck with antibiotics once-per-week cultures started with single colonyBacterial strains plasmids detailed in Supplementary Table 1.Growth curveThree replicates from TB28 JM136 grown from colonies in M9+Glucose M9 salts Amino Acids 0.4% glucose Vitamins 2 mM MgSO4 100 μM CaCl2] overnights at 30 °C overnights’ OD600 measured nanodrop diluted to 0.1 in 200 μl M9 Corning Costar sterile 96-well plate incubated in Tecan Infinite M200 Pro at 30 °C OD600 every 30 min for 23.5 h shaking plate 3 min at 220 rpm before growth linear phase log-transformed growth curve data fitted to straight line slope doubling time-69pt = 10(2)}}=log10(2)slopeWestern FtsITB28 JM136 grown as overnights from colonies in M9 at 37 °Cpropagated 3 ml M9 grown 16 h 25 °C OD600 ~0.5 500 μl harvested samples pelleted flash-frozen stored −80 °C hour pellets resuspended PBS mixed 2× SDS buffer (100 mM Tris-HCL 4% SDS 0.2% bromophenol 20% glycerol 100 mM Cells incubated 10 min 95 °C thermocycler 20 μl 10% BioRad polyacrylamide gel 5 μl PageRuler transferred 25 V 2 h nitrocellulose membrane checked Ponceau stain blotting Rabbit α-FtsI. David Weiss primary antibody 1/50,000 dilution Blots stained 1/50,000 HRP goat antirabbit secondary antibody Bands visualized Clarity Western ECL BioRad recorded Blu-C autoradiography 10 s repeated twice four replicates similar results RNA repair oligo Halo-FtsISW chromosomal ftsI gene coordinates located EcoCyc gene sequence NCBI copied ChopChop60 protospacers-terminus 18 19 fusion repair oligo λ-Red HaloTag 50 nt overlap residues 18–19Silent point mutations codon usage picked sequences protospacer PAM insertion HaloTag.Plasmid constructionpJM44 InFusion Cloning pACYC-sgRNA backbone obtained Dr. Glenn Hauk inverse PCR short oligo insert 20 bp protospacer guide RNA sequence InFusion reaction sequence homology overlap plasmid transformed into competent Stellar competent cells (E. coli HST08) outgrown SOC 1 h 37 °C Cells plated LB 150 μg/ml chloramphenicol incubated overnight 37 °C Colonies screened colony PCR confirmed sanger sequencing plasmid purified GeneJET Plasmid Miniprep Kit.JM136 constructionCRISPR/Cas9 TB28 pKD46 pJM25 grown 30 °C overnight LB 60 carbenicillin 50 kanamycin cells diluted 1:100 50 ml LB CB KAN grown 2 h OD600 ~0.5 0.2% arabinose added induce Cas9 1 h cells prepared electrocompetent E. coli electroporated 2 μl sgRNA plasmid pJM44 10 μl repair oligoCells outgrown 1.5 h 30 °C SOC plated LB 50 CB KAN 150 CAM incubate 2 days screened PCR chromosomal locus ftsI Overnight cultures started 37 °C LB 0.2% arabinose 6% sucrose plasmids serial-diluted factor 10−6 plated LB plates incubated overnight 37 °C Twenty colonies selected screened plasmids excised final colony three plasmids checked PCR sequenced locus-ftsI constructed amplifying meos3.263 ftsI genes pJB106 pVS155-FtsI40 inserted backbone vector pCH02740 protocol cassette replaced aph cassette amplified pKD13 single molecule trackingJM136 pXY018 grown overnight colony 37 °C M9 150 μg/ml CAM propagated 1:100 M9 150 μg CAM grown 16 h 25 °C morning 3 ml culture harvested mid-log phase Cells concentrated 100 μl incubated 10 nM Janeliafluor 646) 30 min washed three times M9 medium concentrated 50 μlEC812 pRM027 grown overnight colony 37 °C LB 150 μg/ml CAM 50 μg/ml KAN 0.2% l-arabinose reinoculated 1:100 M9 media log-phase imaging mEos3.2-FtsI supports cell growth single molecule tracking-level 3% agar pad nanopillar chip 1.2–1.4 μm length 4.5 μm 30 min cells added incubated 2 min washed 1 ml M9 2 min media aspirated room temperature ~20 min sandwiched sealed Bioptechs FSC2 chamber microscope imaging setupJM136 pXY018 imaged split channels Olympus IX-71 microscope JF646 647 ~50 W/cm2 GFP-ZapA 488 ~5 W/cm2. Optosplit II system 600 rdc 700/55 emission filter JF646 540/30 GFP-ZapA Andor iXon 897 Ultra EM-CCD camera APON100×OTIRF objective (1.49 NA/oil 1.6× optivar EM-Gain 300 pre-amplifier gain 3 digitizer 16-bit Baseline clamp offset 100 JM136 grown overnight colonies 37 °C M9 propagated 1:100 M9 grown 16 h 25 °C mid-log phasemicroliter culture pelleted concentrated 50 μl 0.5 μl strain added 3% M9 agar pads-set 30 min sealed BiopTechs chamber microscope imaging Cells imaged phase contrast Olympus IX-71 microscope 100×/1.30 NA Oil Ph3 objective 1.6× optivar recorded 512 × 512 region Andor iXon Ultra Em-CCD camera 100 nm/pixel Phase contrast images 100 ms exposures processed Oufti cell molecule tracking Halo-FtsI: microscopy recorded Metamorph software region 300 pixels wide 512 pixels tall channels Samples microscope acclimated 30–60 min before imaging minimize axial drifting Twenty-five regions selected cells microholes automated journal autofocus cells agar μm GFP-ZapA ring 400-frame movie 1 s (500 ms exposure 500 ms dark channels chromatic imaged TetraSpeck fluorescent beads both channels 50 ms exposure 300 × 512 region647 ~50 W/cm2. 488 ~40 W/cm2. channels cropped Matlab script imregister 488 (ZapA-GFP channel to 647 channel (JF646). dimensions crop transformation matrix channels GFP-ZapA channel JF646 single molecule trajectoriesCropped 647 channels processed ThunderSTORM65 plug-in ImageJ66 filtering wavelet filter order 3 scale 2.0 local maximum molecules 1.5*std(Wave.F1) peak intensity threshold connectivity 8-neigborhood Sub-pixel localization guassian PSF 3-pixel fitting radius initial sigma 1 pixel post-processed data filtered intensity values less than 300 sigma bandpass filter 60–300 nm analysis custom scripts Matlab R2019a localizations linked trajectories nearest-neighbor algorithm molecules time threshold 15 frames distance threshold 300 nm/1 frame diffusion coefficient ~0.05 μm2/s max speed 300 nm/s trajectories corresponding GFP-ZapA ring chosen next step.GFP-ZapA stacks autofocus Z-ring diameter Maximum-intensity projections taken movies cells GFP-ZapA aligned to 647 channelCells cropped circle fit to intensity profile GFP signal diameter real position of FtsI along cell envelope calculated estimated trajectories manually segmented into mobile states when minimum four frames (4 s) long processive displacements one direction selected segments fit to straight line minimize noise classified as “processive” or “stationary” classification description Processive segments used velocity dwell time persistent length measurements.Segmentation classificationThe trajectories segmented manually four data points consistent noise). After segmenting classify segments move processively or stationary Included segmentation observables {v, d l r} v slope linear fit segment d total displacement l trajectory length r standard deviation of positions against linear fit parameters not independent d = l · v combine parameters to {v, l, R} dimensionless quantity relative level residual to displacement noise to signal parameter). manual inspection classify segments as processive threshold R ≤ 0.4.one nanopillar experiment 0–5 trajectories Single cells 1–3 trajectories 77 trajectories from 18 replicates 49 cells yielded 232 segments 139 marked processive R threshold.Deconvolution FtsI fast data Fig. 4f g h fast inactive population FtsI from two-population fit to log-normal CDF Fig. 5b–d[12pt{minimal{amsmath\oddsidemargin-69pt{document}\left( x = P\frac{1}{{x\sigma _1\sqrt {2\pi -{1}{2} - _1^2} + {1 - P}\frac{1}{{x\sigma _2\sqrt {2\pi -\frac{1}{2} - _2}}{{\sigma _2^2}\end{document}fx=P1xσ12πe−12lnx−μ1σ12+1−P1xσ22πe−12lnx−μ2σ22 P percent makeup population bootstrapped FtsI data 200 times fitting CDF estimates each fit parameter global fit mean bootstrapped values Supplementary Fig. 5b individual fits population FtsZ Yang et al.13) Supplementary Fig. 5c used parameters overlay raw FtsI histogram 139 directional segments Figslow population resampled FtsI data 100 times segments makeup final histogram scatterplots Fig. 4f g h resampling 74 segments likely fast-moving.Three-dimensional (3D PALM single-molecule tracking analysisAstigmatism 3D single tracking microscope 568 nm laser power 500 W/cm2 30 ms exposure time 5000–10,000 frames 0–1 W/cm2 405 nm activation light increased mEos3.2-FtsI molecules UV power tuned low red-emitting molecules<1 spot/frame single-molecule localization tracking determined ThunderSTORM plugin ImageJ65,66 Molecules anomalous brightness uncertainty>3σ filtered out Molecules tracked across frames Matlab scripts Sbalzarini et al.67 trajectories longer four frames selected displacements calculated 2D Consecutive frames displacement three MSD curves fitted anomalous diffusion function MSD2D = 4Dtα + D0 curved membrane D0 reflects localization uncertainty imaging molecule tracking PBP2b B. subtilisBstrain bGS28 Pbp2B replaced IPTG-inducible Halo-tagged Pbp2B Bisson-Filho et al.12. cells grown CH media Pbp2B induced 20 μM IPTG labeled 100 nM JF549 HaloTag ligand immobilized agarose pad imaging Cells imaged TIRF Nikon N-STORM microscope time lapses 30 ms exposures 1 min Particle tracking TrackMate LAP tracker particle diameter 300 nm linking distance 300 nm no frame gaps Tracks 5 25 frames analyzed MATLAB MSD vs. t calculated track diffusion coefficient computed MSD equation molecule tracking FtsW Streptococcus strain IU15096 derivative FtsW chromosomal carboxyl-terminus fused 10-amino acid linker (L0) HaloTag Frozen cultures inoculated 4 mL brain heart broth diluted incubated 37 °C 5% CO2 12–16 h cultures optical density 620 nm washed C+Y pH 7.1 diluted OD620 = 5 ml Growth monitored OD620 every 45 min.cells growing microscope slides (VWR 16004-368 cleaned 70% ethanol Gene Frame AB0576) affixed 1.5% agarose (Sigma BioReagent A9414) pH 7.1 medium melted agarose pads constructed Gene Frame agarose second microscope slide 4 °C 45 min agarose pad solidified top slide removed 2 mm strip removed razor cell culture OD620 ≈0.1 500 μL cells transferred 1.5 ml tube Janelia Fluor® 549 HaloTag ligand added 120 pM Cells vortexed spun incubated 15 min 37 °C shaking 300 Cells centrifuged 21,000 × g 5 min washed 0.9 ml 37 pH 7.1 medium resuspended 500 μl 37 °C+Y pH 7.1 medium 1.2 μl cells pipetted coverslips (VWR 48366-227) soaked 25% HCl 25% H2O 50% ethanol 12–24 h submerged 5× H2O rinsed dried lens paper 52846-001) Cells dry coverslips 1–2 min coverslip attached Gene Frame cells aligned agarose padSlides incubated dark 37 °C 30 min imaging.TIRFm performed Deltavision OMX Super Resolution microscope PCO.edge 4.2 (CMOS camera Germany Apo N60X/1.49 TIRF objective laser line 561 nm (emission filter 609–654) FtsW-HT DIC images collected separate channel every 50 ms (20 FPS 100–200 s 561 nm channel 21 ms exposure 100% T TIRF angle 91.2° DIC channel 3 ms exposure 50% T angle 0° Deltavision SoftWoRx channels cells cropped FtsW-HT channel processed FIJI plug-in ThunderSTORM65,66 filtering Wavelet filter order 3 scale 2. localization Local Maximum peak intensity 1.5*std(Wave.F1) 8-neighborhood connectivity Sub-pixel localization PSF: Gaussian method fitting radius 3 pixels Weighted Least Squares Initial sigma 1 pixel.Images single molecule localizations inspected one localization Additional localizations removed five frames without localization trajectory time averaged MSDs calculated trajectories longer 15 Nature Research Reporting Summary.Supplementary | 50.4 | 1.32325 |
10.1038/s41467-020-16066-2 | PMC7195456 | It is unclear if a common EMT expression program exists. Here, the authors perform multiplexed single-cell RNA sequencing across 12 EMT time courses and 16 kinase inhibitor screens, and find that EMT transcriptional responses are context specific and EMT is not a single, linear transition. | Epithelial–mesenchymal plasticity contributes to many biological processes, including tumor progression. Various epithelial–mesenchymal transition (EMT) responses have been reported and no common, EMT-defining gene expression program has been identified. Here, we have performed a comparative analysis of the EMT response, leveraging highly multiplexed single-cell RNA sequencing (scRNA-seq) to measure expression profiles of 103,999 cells from 960 samples, comprising 12 EMT time course experiments and independent kinase inhibitor screens for each. We demonstrate that the EMT is vastly context specific, with an average of only 22% of response genes being shared between any two conditions, and over half of all response genes were restricted to 1–2 time course experiments. Further, kinase inhibitor screens revealed signaling dependencies and modularity of these responses. These findings suggest that the EMT is not simply a single, linear process, but is highly variable and modular, warranting quantitative frameworks for understanding nuances of the transition. | IntroductionEpithelial–mesenchymal (E/M) plasticity is ubiquitous within all epithelial tissues and the reversible transition between these two states contributes to a variety of biological processes, including tumor progression1. During the epithelial–mesenchymal transition (EMT), epithelial cells lose defining characteristics, such as stable cell–cell junctions, and gain the capacity to migrate and invade through extracellular matrices1. While the EMT has been extensively studied, a variety of EMT responses have been reported and no common, EMT-defining gene expression program has been identified2. The transition has historically been depicted as a simple conversion between discrete epithelial and mesenchymal states, but reports of individual cells co-expressing epithelial and mesenchymal genes have since introduced the concept of a partial EMT. This hybrid state has been shown to provide optimal stem cell traits to cancer cells3,4, allow for collective tumor cell migration and the formation of circulating tumor cell clusters5–8, and is associated with metastatic tumors9.Complicating the definition of epithelial, mesenchymal, and hybrid states, most studies have relied on bulk expression measurements of a small subset of marker genes from a static population of cells. Markers of epithelial and mesenchymal states are likely context specific, and thus relying on a small subset of these markers may lead to erroneous conclusions about the relative E/M status of cells. Single-cell RNA sequencing (scRNA-seq) analysis of head and neck squamous cell tumors demonstrated that while E/M plasticity was evident in many tumors, the specific partial EMT gene signature between tumors was variable9. Experimental induction of the EMT can also be variable: microarray analysis of three cell lines exposed to a combination of TGFB1 and TNF alpha (TNF) resulted in EMT responses with only 10–30% of differentially expressed genes shared between conditions10. And in a single mammary epithelial cell line, TGFB1 treatment and a spontaneous EMT induction model resulted in different EMT response trajectories, with only approximately a 50% overlap in differentially expressed genes11. This variability is also not limited to transcriptomic data, and canonical E/M proteins also co-occur inconsistently12.The extent of variability among EMT programs and the regulatory networks that drive them is still unclear given that most evidence spans multiple independent studies, and few have performed controlled comparisons. Here, we provide a thorough comparison of experimentally induced EMTs, spanning multiple cell types, and EMT inducers. We leverage highly multiplexed scRNA-seq to assess context specificity of the EMT and to compare regulatory features of the transition, assessing 103,999 cells from 960 samples, comprising 12 EMT time course experiments and kinase inhibitor screens for each.ResultsMultiplexed scRNA-seq enables comparative analysis of the EMTTo assess transcriptional dynamics of the EMT across a variety of contexts, we used MULTI-seq13 to generate scRNA-seq data from 12 distinct EMT time course experiments. We assessed four different cancer cell lines capable of undergoing an EMT (A549, lung; DU145, prostate; MCF7, breast; and OVCA420, ovarian) and exposed each to known EMT-inducing factors: TGFB1, EGF, and TNF. These cell lines were chosen because they all have an epithelial morphology in culture, have been shown to undergo an EMT in previous studies14–20, and represent four distinct cancer types. The specific inducers were chosen as they all have been previously shown to promote an EMT in different cell lines—including those used in this study in most cases14–20—and their binding to each of their cell surface receptors initiates independent signaling pathways. In response to these factors, each cell line exhibited morphological changes, consistent with an EMT (Supplementary Fig. 1a). We note different inducers can promote different morphologies in the same cell line (e.g., DU145 with TGFB1 or TNF), and some changes were modest in comparison to others (e.g., MCF7 cells treated with EGF). Lacking a typical spindle-shaped morphology, however, does not preclude other EMT traits. For example, at higher doses, EGF has been shown to promote an EMT associated with a circular morphology21. Ultimately, it is likely that these differences arise from subtleties in the expression programs initiated by each inducer, and particularly the different expression dynamics of various cytoskeletal and extracellular matrix proteins.For each of the 12 conditions, samples were collected at five distinct time points from 8 hs to 1 week after treatment, and three additional time points from 8 h to 3 days after the EMT-inducing stimulus had been removed (Fig. 1a). The 3-day withdrawal time point was chosen based on preliminary data, suggesting transcriptional reversion in as few as 3 days. In the aggregated data, expression profiles clustered dominantly by cell line, and after demultiplexing, the majority of cell line annotations (95.8% on average) were restricted to a dominant cluster, demonstrating robust multiplexing (Fig. 1b, Supplementary Fig. 1b). In total, we annotated 58,088 single cells from across 576 samples, comprising six replicates of the 12 time course experiments (Fig. 1c, Supplementary Fig. 1c). Replicates were highly correlated, supporting the consistency of the experimental procedures and processing workflow (Supplementary Fig. 2).Fig. 1Multiplexed scRNA-seq profiling of 12 EMT time course experiments.a Schematic of the 96-well experimental design for the 12 EMT time course experiments (left), and t-SNE embeddings of the MULTI-seq barcode counts, demonstrating strong signal for demultiplexing (right). b UMAP embedding of aggregated expression data of all data, colored by unsupervised clustering (top), and a graph showing the relative proportion of annotations for each cell line assigned to each cluster after demultiplexing (bottom). c Graph showing the number of cells captured for each time course experiment. d UMAP embeddings of each of the 12 time course experiments. Grey dots correspond to individual cells, shaded regions represent the related sample density for each time point, and colored dots correspond to the maxima of the density function. e UpSet plot showing the intersections of the top 1000 variable genes of each time course experiment. f GSEA plots showing the NES for the EMT hallmark genes in the variance-ranked genes for all conditions.Transcriptional dynamics of the EMT are context specificWe next assessed the temporal progression of each time course. In each case, time-dependent shifts in cells’ expression profiles were evident, and withdrawal samples showed reversion back toward the untreated state (Fig. 1d). In each cell line, receptors for the three EMT inducers were detectable, explaining these dynamics (Supplementary Fig. 3). While the top 1000 variable genes for each time course showed some expression patterns conserved across cell lines, context-dependent gene sets were dominant (Fig. 1e). Gene set enrichment analysis (GSEA) of variance-ranked genes for each time course did, however, demonstrate enrichment for the MSigDB hallmark EMT gene set in all conditions22 (Fig. 1f). This is consistent with the morphological changes we observed, and further supports that these changes are associated with an EMT response. The minimal overlap of top variable genes among conditions suggests that the specific EMT genes involved in the response may vary.To specifically compare temporal dynamics of the EMT, we first pseudotemporally ordered the cells from each condition (Fig. 2a, b). In each time course, cells progressively transitioned throughout the full 7 days of EMT induction, and withdrawal of the EMT stimulus led to a near-complete reversion after as few as 3 days (Fig. 2b). We note that it is possible that the cells could have continued to transition following day 7. It will be important for future studies to assess the temporal limits of the EMT response. We then assessed gene expression dynamics throughout the pseudotemporal trajectories. In all cases, transitions were not simply linear processes of two opposing E/M expression programs. Rather, all involved combinations of discrete transcriptional events (Supplementary Fig. 4), suggesting that the EMT may be a multistep process. We found that each condition, with the exception of A549 cells induced with EGF and OVCA420 treated with TNF, was associated with an average increase in the expression of the EMT hallmark gene set22, with TGFB1 often producing the most potent effects (Fig. 2c). GSEA revealed, however, that differentially expressed genes from these two conditions, along with all others, were enriched for the hallmark gene set (Supplementary Fig. 5a), but in these two specific conditions, several EMT hallmark genes are repressed, resulting in a net neutral EMT score (Supplementary Fig. 5b).Fig. 2EMT transcriptional responses are largely context specific.a UMAP embeddings of A549 cells treated with TGFB1. Each point represents an individual cell, and colors correspond to time point (top) or pseudotime value (bottom). b Sina plot showing the distribution of pseudotime values across time points for all 12 time course experiments, with time points colored the same as in a. Horizontal black bars represent mean expression values for each group and each point corresponds to a single cell. c Smoothed model of the EMT hallmark gene set score throughout the pseudotime. Shaded bands for each line correspond to the standard error for each model. d Clustered heatmap of all pairwise Jaccard similarity values for the differentially expressed genes in each condition. e Counts of how frequently each gene is differentially expressed among time course experiments. f Heatmap showing EMT-associated expression changes associated with a gene set of all genes that are differentially expressed in at least eight time course experiments. The colormap corresponds to the pseudotime beta coefficient of the linear model for each gene.Surprisingly, responses of individual cell lines to different stimuli were more similar than the responses of different cell lines to the same stimulus, but importantly, all pairwise comparisons show very little overlap in their differentially expressed genes (average Jaccard index of 0.22; Fig. 2d). Of all genes differentially expressed across conditions, the majority changed in as few as one to two conditions, suggesting that the global expression programs associated with the EMT are remarkably context specific (Fig. 2e, Supplementary Data 1). A small subset of canonical EMT genes, including TGFB1, CD44, and FN1, along with less-reported genes, such as TGM2 and PMEPA1, were differentially expressed in most conditions. The majority of the MSigDB hallmark EMT gene set was differentially expressed in only a small number of conditions, with only 49/200 hallmark genes being differentially expressed across the majority of conditions (Supplementary Fig. 6). Extracellular matrix proteins, proteases, and integrins from the hallmark gene set are variably affected across conditions, which could explain the differences in morphological changes observed (Supplementary Fig. 6). This may reflect that the hallmark genes were derived from various founder gene sets that may have been driven by fibroblast expression rather than an EMT (ref. 23). Interestingly, however, many canonical EMT genes, including SNAI1, CDH1 (E-cadherin), and CDH2 (N-cadherin) differentially expressed in only a small number of conditions (Fig. 2e).To identify signatures that may not have been represented in the hallmark gene set, we took all genes that were differentially expressed in at least eight (defined as two-thirds of our conditions as to not be too restrictive) of our experimental conditions, and compiled our own gene set of 86 conserved upregulated genes and 17 downregulated genes (Fig. 2f, Supplementary Data 2). While no gene represents a universal marker of the transition, this list contains those that were most frequently changed. Common epithelial-associated (downregulated) genes included various keratins (KRT8, KRT18, and KRT19), consistent with morphological changes and the loss of epithelial features. While the conserved mesenchymal-associated (upregulated) genes contain several canonical EMT genes, many are not typically associated with the transition. These upregulated genes, however, do enrich for GO terms associated typical EMT-associated traits, including extracellular matrix disassembly (p = 5.0e−4) and organization (p = 3.7e−4), cell migration (p = 2.0e−3), and negative regulation of apoptosis (p = 4.4e−11; Supplementary Fig. 7a). Regulatory regions of the 86 mesenchymal-associated genes are also enriched in binding sites for AP-1, MYC, MEF2, and KLF transcription factors (Supplementary Fig. 7b). These factors have all been implicated in the EMT and could represent conserved regulators of the transition24–28. We also confirmed that these 86 mesenchymal-associated genes have variable expression levels among cancer cells from individual human lung tumors and syngeneic mouse tumor models, as well as in scRNA-seq data of healthy epithelium from various mouse tissues (Supplementary Fig. 8a). Further, in each of these data sets, the 86 mesenchymal-associated genes are highly correlated (Supplementary Fig. 8b). Together, this suggests that this expression program is not simply an artifact of culture experiments, but are coexpressed in vivo and may contribute to an E/M heterogeneity program in these tissues.The EMT can be coordinated by diverse transcription factor networksWhile many of the most conserved EMT genes are regulated by shared regulatory factors (Supplementary Fig. 7b), these conserved genes only represent a small fraction of differentially expressed genes. We next sought to determine if the remainder of EMT-associated expression dynamics are driven by a common regulatory program that perhaps gives rise to distinct expression patterns due to cells’ epigenetic or mutational profiles, for example. Across the experimental conditions we assessed, most canonical EMT transcription factors—other than SNAI2—were rarely differentially expressed (Fig. 3a). While in some cases (e.g., TWIST1) the transcription factors were not detected, perhaps owing to insufficient sensitivity to lowly expressed genes, canonical EMT transcription factors were often readily detectable, but did not show dynamics throughout the EMT response (Supplementary Fig. 9). We scored each cell for the coexpression of transcription factors and their putative target genes (regulons), and identified those that showed differential activity throughout the EMT. We found that transcription factor activity is also remarkably context specific, with most being restricted to one to two of our time course experiments (Fig. 3b). Several factors were fairly well conserved, however. Consistent with our list of conserved genes, AP-1 (JUN, JUNB), the NFkB-associated RELB, ATF4, SOX4, and KLF6 regulons showed frequent activation, whereas ELF3 and MYBL2 activity often decreased (Fig. 3c). These factors have all been previously implicated in the EMT, but are not typically considered canonical EMT regulators29–34. To assess the accuracy of these results, we used ATAC-seq to assess the accessibility of transcription factor motifs throughout the EMT and compared accessibility dynamics to the inferred regulon activity. For the purpose of validation, we chose to assess the OVCA420 TGFB1 time course (Fig. 3d). This was the smallest data set in our scRNA-seq cohort, so we chose to validate the approach on the condition with the lowest power for inferring transcription factor activity. We found that in many cases, motif accessibility throughout the EMT-mirrored regulon activity measured from scRNA-seq data alone (Fig. 3e). This supports that the regulon activity inference provides an accurate representation of the transcription factor activity throughout each of the conditions assessed.Fig. 3Inferring transcription factor activity throughout the EMT.a Plot showing in which time course experiments various canonical EMT transcription factors are differentially expressed. b Counts of how frequently various transcription factors and their associated regulons are differentially active among time course experiments. c Heatmap showing EMT-associated changes of regulons that are differentially active in at least six time course experiments. The colormap corresponds to the pseudotime beta coefficient of a linear model for each regulon. d Differential accessibility of transcription factor motifs from ATAC-seq data of OVCA420 cells treated with TGFB1 for 0, 1, 3, or 7 days. The colormap represents the accessibility Z-score for each transcription factor motif. Examples of transcription factors from each cluster are listed. e Regulon activity score of the same transcription factors listed in d inferred from the OVCA420 TGFB1 time course experiment. Each dot represents a single cell, colored by time point. The black line corresponds to the modeled trend from a generalized additive model. Shaded bands for each line correspond to the standard error for each model.Kinase inhibitor screens reveal signaling dependencies in a variety of EMT responsesParacrine signaling is another regulatory feature likely to coordinate the EMT across a population of cells35–38. In fact, we found that the expression of secreted factors spanning a variety of signaling pathways broadly increased in each of our 12 time courses (Fig. 4a, b). Given this, we next established an experimental design to mechanistically assess the dependence of the EMT on multiple signaling pathways and compare these dependencies across contexts. We curated a selection of 22 small molecule inhibitors targeting a variety of kinases and treated cell lines alone for 7 days, or in combination with one of the three EMT inducers previously used (Fig. 4c). Leveraging MULTI-seq to multiplex samples, we ultimately generated scRNA-seq profiles for 45,911 cells across the 384 distinct conditions.Fig. 4Kinase inhibitor screens identify signaling dependencies of the EMT.a Gene set score of the KEGG pathway “cytokine–cytokine receptor interaction” over pseudotime for each time course experiment. Shaded bands for each line correspond to the standard error for each model. b Heatmap showing EMT-associated changes of the individual genes of the same gene set as in a, only listing those with a significant change in at least one time course experiment. c Schematic of the 384-sample experimental design for the kinase inhibitor screen. d Heatmap showing the number of cells annotated to each condition after demultiplexing the scRNA-seq data. e Summary of the number of genes that are differentially expressed in each cell line exposed to the inhibitors without EMT induction. f Average pseudotime values calculated for each condition. g Boxplots showing the distribution of pseudotime values for A549 cells treated with the inhibitors alone (grey) or in combination with TGFB1 (orange). The horizontal black line of the boxplot represents the median value, the box spans the 25th and 75th percentiles, and whiskers correspond to 1.5 times the interquartile range. h UMAP embeddings of untreated A549 cells with those had been treated with TGFB1 alone or in combination with the TGFBR1 inhibitor LY364947 (top), or the RIPK1 inhibitor necrostatin-5 (bottom). i Heatmap showing expression (Z-score) of genes differentially expressed in A549 cells by TGFB1 in untreated A549 cells, as well as those treated with TGFB1 alone or in combination with necrostatin-5. j Difference in normalized enrichment scores for transcription factor targets in genes that are successfully inhibited by necrostatin-5 compared to those that are not. Positive values correspond to regulons with that are enriched in necrostatin-5-inhibited genes, whereas negative values represent those are not affected by necrostatin-5.From retrieved cell counts alone, drop-out patterns from cell line-dependent and -independent cytotoxic/cytostatic effects can be observed (Fig. 4d, e). To assess the impact of these inhibitors on EMT progression, however, we calculated pseudotime values for the inhibited cells using the models built from corresponding time course experiments of the same cell line and EMT inducer (Fig. 4f). From this, we could identify inhibitors that reduced cells’ pseudotime values at 7 days compared to uninhibited controls, therefore dampening the EMT response. LY364947 (TGFBR1 inhibitor), for example, abrogated TGFB1-induced EMTs (Fig. 4f–h), and erlotinib and gefitinib (EGFR inhibitors) consistently blocked the effects of EGF (Fig. 4f).The effects of these inhibitors, however, weren’t limited to blocking the direct signaling of the EMT-inducing factor. For example, TGFBR1 inhibition partially blocked EMT progression in a variety of conditions, including EGF-treated A549 and OVCA420 cells, and TNF-treated A549 and MCF7 cells (Fig. 4f). This suggests that the activation of paracrine TGFB1 signaling may be critical for EMT progression, following a variety of initial stimuli, supporting previous work showing the dependence of the EMT on transcription-factor-activated TGFB1 autocrine loops39–41.Effects of direct EGFR inhibition with erlotinib and gefitinib were largely restricted to EGF-treated EMT responses, but inhibition of its downstream kinase MEK (with PD 0325901) hindered the EMT response in TGFB1-treated A549 and MCF7 cells. Non-canonical TGFBR1 signaling through MEK/ERK has been previously reported42,43, and two recent studies have proposed a MEK-dependent regulatory checkpoint in the EMT (refs. 11,44). While our data for TGFB1-treated A549 and MCF7 cells are in agreement with these findings, it also demonstrates that this checkpoint is not universal, even among other TGFB1-induced EMT responses, as DU145 and OVCA420 cells are not susceptible to MEK inhibition (Fig. 4f).Inhibition of RIPK1—a kinase involved in activating NFkB and necroptosis pathways—with necrostatin-5 (Nec-5) blocked EMT progression in all of the same conditions as TGFBR1 inhibition. Nec-5-treated cells, however, consistently had higher pseudotime values than TGFBR1-inhibited cells, suggesting a partial EMT response (Fig. 4f, g). To determine if the partial response corresponds to reduced magnitude of gene expression changes, or a selective inhibition of a subset of genes, we assessed expression levels of all genes differentially expressed following TGFB1 treatment. In each case, RIPK1 inhibition only abrogated a subset of TGFB1-induced expression changes, producing a partial EMT response (Fig. 4h, i). Importantly, we note that this partial response with RIPK1 inhibition is not due to a temporal block in EMT progression (i.e., preventing late EMT dynamics), as inhibition does not exclusively prevent late response genes (Supplementary Fig. 10). This suggests that the EMT involves multiple independent regulatory modules that can be perturbed without impacting others.To our knowledge, no direct cross-talk between the TGFB1 signaling and RIPK1 has been documented, but loss of RIPK1 has been previously associated with an enhanced epithelial phenotype, reduced ERK1/2 phosphorylation, and reduced transcriptional activity of the AP-1 complex45,46. To determine if RIPK1 inhibition prevents the activation of AP-1 targets in our EMT models, we assessed the enrichment of transcription factor binding motifs in the promoters of genes that failed to change throughout the EMT in RIPK1-inhibited cells. We found that the AP-1 binding site (JUN/FOS, BACH2 motifs) was the most enriched in promoters of genes that failed to become upregulated in Nec-5-treated cells in response to TGFB1 (Fig. 4j). Other enriched motifs include EGR1 and PAX4 binding sites. Both AP-1 and EGR1 can be activated through ERK1/2 signaling, providing a possible mechanistic link between RIPK1 and these transcriptional changes46,47. As ERK1/2 is a downstream effector of MEK, this may also explain the previously proposed MEK checkpoint of the EMT (refs. 11,44). While it is still unclear how RIPK1 becomes activated, this regulatory axis is conserved in every condition we assessed that is also dependent on TGFB1 signaling (based on similarity to TGFBR1 inhibition), and may represent a common, though not universal, regulatory network of the EMT.DiscussionHere, we have demonstrated that the EMT is a complex cellular process, driven by independent regulatory networks that ultimately give rise to incredible context specificity. Given these findings, we argue that the common paradigm of cells simply undergoing a linear transition between well-defined epithelial and mesenchymal programs is an oversimplification that can lead to erroneous conclusions. Given the variety of EMT responses that can be elicited—each with remarkable dissimilarity—a single mesenchymal gene expression program simply does not exist. The variety of possible responses also makes the full EMT indefinable, as the combination of all is likely to never occur. For the same reason, the number of possible partial EMT states is likely inumerable. This partial state has historically been defined as cells co-expressing both epithelial and mesenchymal markers, but studies have often relied on a small number of canonical markers to make this designation, and we have shown that most markers are inconsistently involved in the transition. This does not discount the likely importance of gradation along some axis of epithelial and mesenchymal phenotypes, but a more comprehensive definition of intermediate and polar states is required.In this study, we have begun to take steps toward understanding the complexity of E/M plasticity. As single-cell technologies are becoming increasingly scalable, it may soon be possible to learn the complete manifold of all possible states related to E/M plasticity for a given cell type. Unique environment and developmental history will likely mean that this manifold will vary for each cell type, but it may be possible to learn models for their prediction or alignment across settings. It will also be critical to understand how positions along the manifold are associated with phenotypic traits, and how cell perturbations promote dynamics within it. With a comprehensive model of the E/M plasticity, we will gain a quantitative understanding of nuanced cellular heterogeneity, improving our understanding of development, tissue homeostasis, and disease progression. This information will also help inform new strategies to therapeutically modulate cellular phenotypes in disease.MethodsCell cultureA549, DU145, and MCF7 cells were obtained from ATCC (CCL-185, HTB-81, and HTB-22, respectively). OVCA420 cells were kindly provided by Dr. Gordon Mills. All cells were cultured in Dulbecco’s Modified Eagle Medium with 4.5 g/L glucose, L-glutamine, and sodium pyruvate (Corning, 10-013-CV), supplemented with 10% of fetal bovine serum and cultured at 37 °C with 5% CO2.EMT time course experimentsFor each cell line, 10,000 cells were plated into each well of a 96-well plate according to the schematic in Fig. 1a. The addition of TGFB1, EGF, and TNF were scheduled such that all time points completed at the same time for collection. Cells were treated with 10 ng/mL TGFB1 (R&D Systems, #240-B-010), 30 ng/mL EGF (Invitrogen, #PHG0311), or 10 ng/mL TNF (Invitrogen, #PHC3015). Media was changed and fresh TGFB1, EGF, or TNF were added every 2 days to ensure relatively constant concentrations of these factors. To avoid over-confluence throughout the experiments, cells were passaged as required, but not within the last 2 days of the time course to avoid artifacts at the time of collection. After the scheduled treatments, cells were immediately processed for scRNA-seq multiplexing.The time course experiments were performed twice independently. Each time, the two time course replicates were performed in parallel, and on the second time through the experiment, two 10x libraries were generated for each plate replicate. Samples from the first replicate are labeled “Mix1” and “Mix2”, corresponding to the two plates running in parallel. Samples from the second replicate are labeled “Mix3a/b” and “Mix4a/b”.Kinase inhibitor screenFor each cell line, 10,000 cells were plated into four 96-well plates according to the schematic in Fig. 4c. Cells were simultaneously treated with small molecule kinase inhibitors (listed in Fig. 4c) and either 10 ng/mL TGFB1, 30 ng/mL EGF, or 10 ng/mL of TNF. No-inhibitor and No-EMT-inducer controls were also included for all conditions. All inhibitors were used at a final concentration of 1 µM (Cayman Chemical Kinase Screening Library, Item No. 10505, Batch No. 0537554). EMT inducers and kinase inhibitors were refreshed daily after replacing the culture media. After 7 days of treatment, all samples were immediately processed for scRNA-seq multiplexing.Multiplexing individual samples for scRNA-seqMultiplexing was performed according to the MULTI-seq protocol13, and reagents were kindly provided by Dr. Zev Gartner. Briefly, culture media was removed and each well was washed with 1× Dulbecco’s phosphate-buffered saline (PBS; Corning, #21-031-CV). Next, a lipid-modified DNA oligonucleotide and a unique sample barcode oligonucleotide were added at 200 nM to 0.05% trypsin with 0.53 mM EDTA. This was added to each sample to be multiplexed, with each sample receiving a different sample barcode. Cells were incubated with this trypsin mixture for 5 min at 37 °C, and plates were gently mixed periodically. After 5 min, a common lipid-modified co-anchor was added to each well at 200 nM to stabilize the membrane residence of the barcodes. Cells were incubated for an additional 5 min at 37 °C with periodic mixing. After this labeling time, all cells were in suspension, lifted from the plate. The trypsin was then neutralized with cultured media, and the cells were mixed by pipetting to ensure a single-cell suspension. Samples were then transferred to V-bottom 96-well plates, and pelleted at 400 × g for 5 min. Barcode-containing media was removed, and the cells were then washed with PBS + 1% bovine serum albumin (BSA). Washes were performed twice, and after the final wash, cells were resuspended in PBS + 1% BSA, pooled together, repelleted, and resuspended in PBS + 1% BSA. Viability and cell counts were then performed, before preparation of the scRNA-seq libraries.scRNA-seq library preparation and sequencingSingle-cell suspensions were processed using the 10× Genomics Single Cell 3′ RNA-seq kit (v2 for time course experiments, v3 for kinase inhibition). Gene expression libraries were prepared according to the manufacturer’s protocol. MULTI-seq barcode libraries were retrieved from the samples and libraries were prepared independently according to the MULTI-seq library preparation protocol13. Briefly, barcode libraries are separated from the cDNA libraries during the first round of size selection in the 10× Genomics library preparation protocol and PCR-amplified prior to sequencing13. Final libraries were sequenced on a NextSeq500 (Illumina). Expression libraries were sequenced so that time course libraries reached an approximate depth of 40,000–50,000 reads per cell (for the v2 scRNA-seq kit), and 20,000–25,000 reads per cell for the kinase inhibitor experiment (v3 scRNA-seq kit). For the time course data, we detected a median of 3649 genes and 17,330 UMIs per cell, and for the kinase inhibitor screens, we detected a median of 2360 genes and 7634 UMIs.Processing of raw sequencing readsRaw sequencing reads from the gene expression libraries were processed using CellRanger v2.2.0 for the time course data, and v3.0.2 for the kinase inhibitor data. The GRCh38 build of the human genome was used for both. Except for explicitly setting --expect-cells = 25,000, default parameters were used for all samples. MULTI-seq barcode libraries were simply trimmed to 26 bp (v2 kit) or 28 bp (v3 kit) using Trimmomatic48 (v0.36) prior to demultiplexing.Demultiplexing expression data with MULTI-seq barcode librariesDemultiplexing was performed using the deMULTIplex R package (v1.0.2) (https://github.com/chris-mcginnis-ucsf/MULTI-seq). The key concepts for demultiplexing are described in McGinnis et al.13. Briefly, the tool takes the barcode sequencing reads and counts the number of times each of the 96 barcodes appears for each cell. Then, for each barcode, it assesses the distribution of counts in cells and determines an optimal quantile threshold to deem a cell positive for a given barcode. Cells positive for more than one barcode are classified as doublets and are removed. Only cells positive for a single barcode are retained for downstream analysis. As each barcode corresponds to a specific sample in the experiment, the sample annotations can then be added to all cells in the data set.Data quality control and processingQuality control was first performed independently on each 10× Genomic library, and all main processing steps were performed with Seurat v3.0.2 (ref. 49). Expression matrices for each sample were loaded into R as seurat objects, only retaining cells with >200 genes detected. Cells with a high percentage of mitochondrial gene expression were also removed. We then subsetted the data, making independent seurat objects for each time course or kinase inhibition experiment (i.e., for all independent cell line and EMT inducer combinations). Each condition was then processed independently with a standard workflow. We first removed genes detected in <1% of the cells for the given experiment. The expression values were then normalized with standard library size scaling and log transformation. The top 3000 variable genes were detected using the variance-stabilizing transformation (vst) selection method in Seurat. Expression values were scaled and the following technical factors were regressed out: percentage of mitochondrial reads, number of RNA molecules detected, cycle cycle scores, and for the time course data, batch was also included. For initial exploration, PCA was run on the variable genes, but all UMAP embeddings included in figures are based on PCA run on genes used for pseudotemporal ordering of cells. UMAP embeddings were calculated from the first 30 principal components.Pseudotemporal ordering of cellsPseudotime models for each time course experiment were built using the R package psupertime v0.2.1 (ref. 50) on the top 3000 variable genes from each condition. Psupertime is based on ordinal logistic regression, taking scRNA-seq data with sequential labels and identifying a linear combination of genes that places the cells in the specified label order. To build the pseudotime model for each time course, we first omitted the treatment withdrawal samples. Because psupertime is based on regression; however, pseudotime values for new data can be calculated by simply performing matrix multiplication between the coefficient matrix of the pseudotime model and the expression matrix of the new data. We used this approach to calculate pseudotime values for both the treatment withdrawal samples of the time course experiment. We also used the time course models to calculate pseudotime values for the respective kinase inhibition experiments. As the range of pseudotime values can vary between conditions, we simply rescaled the values from 0 to 1 in cases where multiple models were compared in the same figure.Differential expression analysisFor time course experiments, expression dynamics of each gene, or transcription factor regulon score, as a function of pseudotime was modeled using the generalized additive model function provided by the R package mgcv with the model exp ~ s(pseudotime, k = 4) + batch, with the smoothing parameter estimation method set to restricted maximum likelihood (method = “REML”). The number of basis functions (k) was chosen such that the residuals were randomly distributed. P-values associated with the smoothed pseudotime function for each gene were adjusted using the p.adjust() function in R with the Benjamini–Hochberg method. As many genes may significantly vary throughout pseudotime but have low effect sizes, we only evaluated significant genes (adjusted p < 0.05) that are also within the top 2000 variable genes of each time course experiment. While others may be biologically relevant, their signal in the data is often too low to assess reliably.When assessing transcription factor activity (Fig. 3) and cytokine production (Fig. 4), we were more generally interested in assessing the directionality of change over pseudotime, so in these cases, we used the same approach, but removed the smoothing function from the model. This allowed us to report the single coefficient associated with the pseudotime covariate, representing whether activity generally increased or decreased throughout the transition.For the kinase inhibition experiment, we assessed the number of differentially expressed genes in cell lines treated with a kinase inhibitor, but no EMT inducer. For this, we still used the gam() function provided by the mgcv package with the model exp ~ inhibitor, setting the no-inhibitor controls as the intercept. We then quantified the number of genes with an adjusted p < 0.05.Calculating smoothed expression trendsTo calculate smoothed expression trends over pseudotime, we used models used for differential expression, but calculated the fit values for 200 evenly spaced pseudotime values ranging between the minimum and maximum pseudotime values.Gene set enrichment analysisGSEA was performed using the R package fgsea51. Input genes were ranked either by their variance values after the vst, computed by Seurat’s FindVariableFeatures() function, or by adjusted p-value from the differential expression analysis. Reference gene sets were collected from the Molecular Signatures Database (MSigDB) v6.2.Gene set scoringGene set scoring of the EMT hallmark gene set and the KEGG pathway “cytokine–cytokine receptor interaction” was performed using the AddModuleScore() function provided by the Seurat package. Default parameters were used.Transcription factor regulon scoring of single cellsRegulon scores for individual cells were computed using the SCENIC workflow52. Log-transformed expression values for each time course experiment were used as input into the command-line interface functions of pySCENIC. First, gene regulatory networks were computed using the grnboost2 method in the grn function. Next, enriched motifs were identified using the ctx function, providing the cisTarget v9 databases of regulatory features 500 bp upstream, 5 kb centered on the TSS, and 10 kb centered on the TSS. Finally, individual cells were scored for motifs using the aucell function.Identifying over-represented transcription factor motifs in gene listsThe R package RcisTarget52 was used to identify enriched transcription factor motifs associated with gene lists, using the cisTarget v9 transcription factor motif annotations and the hg19-tss-centered-10kb-10species.mc9nr database of motif rankings. To compare enrichment between two gene lists, we calculated the difference in normalized enrichment scores (NES) for motifs between the two lists and ranked motifs to identify uniquely enriched motifs.ATAC-seq sample preparation and analysisATAC-seq samples were prepared from OVCA420 cells treated with 10 ng/mL of TGFB1 for 0, 1, 3, or 7 days, and the experiment was performed independently twice. Sample preparation was performed as described by Buenrostro et al.53. Briefly, nuclei were extracted from 50,000 cells per sample and chromatin was tagmented using the TDE1 transposase provided in the Nextera DNA Library Preparation Kit (Illumina). While the original protocol recommended 2.5 µL of enzyme, we found that optimal tagmentation of these samples required 5 µL of enzyme at 37 °C for 30 min with gentle mixing. Finally, ATAC libraries were amplified and sequenced on a NextSeq500 150-cycle high output run, yielding ~50 M reads per sample.Raw reads were aligned to the hg38 build of the human genome using Bowtie2 (ref. 54) and peaks were called using MACS2 (ref. 55) with the following parameters: -q 0.01 --nomodel --shift -100 --extsize 200 -B --SPMR --broad. Differential motif accessibility was calculated using the R package chromVAR (ref. 56). Briefly, the summits of peaks from all samples were merged, and expanded to a 250 bp window, centered on the summit. Motifs from the human_pwms_v2 list included with the package were mapped to the peaks using the matchMotifs() function and then deviations across samples were computed. Significant deviations in motif accessibility were identified using the differentialDeviations() function.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Data 1
Supplementary Data 2
Reporting Summary | nature communications | [
"Article"
] | [
"Metastasis",
"Tumour heterogeneity",
"High-throughput screening"
] | IntroductionEpithelial–mesenchymal plasticity ubiquitous tissues reversible transition contributes to biological processes including tumor During transition cells lose characteristics migrate invade through extracellular matrices1 EMT extensively studied variety of responses reported no common EMT-defining gene expression program transition simple conversion between epithelial mesenchymal states cells co-expressing epithelial mesenchymal genes introduced partial EMT hybrid state optimal stem cell traits to cancer cells3,4 collective tumor cell migration formation circulating tumor cell associated with metastatic tumors9 definition epithelial hybrid states studies relied on bulk expression measurements small marker genes population Markers context specific may erroneous conclusions E/M status-cell RNA sequencing analysis of tumors E/M plasticity evident specific partial EMT gene signature variable9 Experimental induction EMT variable microarray analysis of three cell lines exposed to TGFB1 TNF alpha) resulted in EMT responses with 10–30% differentially expressed genes shared betweensingle mammary epithelial cell line TGFB1 treatment spontaneous EMT induction model different response trajectories 50% overlap in differentially expressed genes11 variability not limited to transcriptomic data canonical E/M proteins co-occur inconsistently12 variability among EMT programs regulatory networks unclear evidence spans multiple studies few controlled comparisons comparison of experimentally induced EMTs multiple cell types EMT inducers multiplexed scRNA-seq assess context specificity EMT regulatory features transition 103,999 cells from 960 samples 12 EMT experiments kinase inhibitor screens scRNA-seq comparative analysis MULTI-seq13 data 12 EMT experiments assessed four cancer cell lines EMT (A549 DU145 MCF7 OVCA420 exposed to EMT-inducing factors TGFB1 EGF TNF cell lines chosen epithelial morphology shown undergo EMT represent four cancer types inducers chosen promote EMT different cell binding cell surface receptors independent signaling pathways each cell line exhibited morphological changes consistent with EMT different inducers promote different morphologies same cell lineDU145 TGFB1 changes modest MCF7 cells treated EGF). Lacking spindle-shaped morphology preclude EMT traits higher doses EGF EMT circular differences arise from expression programs inducer different expression dynamics cytoskeletal extracellular matrix proteins 12 conditions samples collected five time points 8 hs to 1 week after treatment three 8 h to 3 days after EMT stimulus removed (Fig. 3-day withdrawal point chosen preliminary data transcriptional reversion in 3 days expression profiles clustered by cell line after demultiplexing majority cell line annotations (95.8% restricted to dominant cluster robust multiplexing 1b annotated 58,088 cells 576 samples six replicates of 12 experiments Replicates correlated consistency experimental procedures processing workflow. 1Multiplexed scRNA-seq profiling of 12 EMT experimentsSchematic 96-well experimental design 12 EMT experiments t-SNE embeddings MULTI-seq barcode counts strong signal demultiplexing UMAP embedding aggregated expression data colored unsupervised clustering graph proportion annotations each cell line after demultiplexing Graph number cells captured each experiment UMAP embeddings 12 experiments Grey dots cells shaded regions sample density colored dots maxima density function UpSet plot intersections top 1000 variable genes GSEA plots NES EMT hallmark genes variance-ranked genes conditions.Transcriptional dynamics EMT context temporal progression time-dependent shifts expression profiles evident withdrawal samples reversion untreated state (Fig. 1d). each cell line receptors three EMT inducers detectable explaining dynamics Fig. 3) top 1000 variable genes expression patterns lines context-dependent gene sets dominant (Fig. Gene set enrichment analysis) variance genes enrichment MSigDB hallmark EMT gene set all conditions22 (Fig. consistent changes supports changes associated EMT response minimal overlap top variable genes suggests specific EMT genes may varytemporal dynamics EMT pseudotemporally ordered cells each condition (Fig. 2a b). cells transitioned 7 days EMT induction withdrawal led near reversion after 3 days (Fig. 2b). cells following day 7. important future studies assess temporal limits EMT response assessed gene expression dynamics pseudotemporal trajectories transitions not linear processes discrete transcriptional events Fig 4) EMT multistep process each condition A549 cells induced EGF OVCA420 treated TNF associated average increase expression EMT hallmark gene set22 TGFB1 most effects (Fig. 2c). GSEA revealed differentially expressed genes two conditions enriched for hallmark gene set Fig 5a), two conditions several EMT hallmark genes repressed net neutral EMT score 5b).Fig. 2EMT transcriptional responses context specific UMAP embeddings A549 cells treated TGFB1 Each point represents cell colors time point pseudotime value Sina plot distribution pseudotime values across points 12 experiments Horizontal black bars represent mean expression values point single cell Smoothed model EMT hallmark gene set score Shaded bands standard errorheatmap Jaccard similarity values differentially expressed genes gene expressed experiments Heatmap EMT-associated expression changes differentially expressed eight experiments colormap corresponds pseudotime beta coefficient linear model each gene responses cell lines different stimuli similar same stimulus comparisons show little overlap differentially expressed genes (average Jaccard index 0.22 Fig. 2d). genes differentially expressed majority changed few one to two conditions global expression programs EMT context specific (Fig. 2e small subset canonical EMT genes TGFB1 CD44 FN1 less-reported genes TGM2 PMEPA1 differentially expressed most conditions majority MSigDB hallmark EMT gene set differentially expressed small number conditions 49/200 hallmark genes differentially expressed across majority conditions Extracellular matrix proteins proteases integrins hallmark gene set variably affected across conditions differences changes hallmark genes derived from founder gene sets driven fibroblast expression EMT canonical EMT genes SNAI1 CDH1 CDH2 differentially expressed small number conditionsidentify signatures hallmark gene set took genes differentially expressed in eight two-thirds experimental conditions compiled gene set 86 conserved upregulated 17 downregulated (Fig. 2f Supplementary Data 2) no gene universal marker transition list contains frequently changed Common epithelial-associated (downregulated genes included keratins (KRT8 KRT18 KRT19) consistent with changes loss epithelial features conserved mesenchymal-associated (upregulated genes contain canonical EMT genes not associated with transition upregulated genes enrich typical EMT-associated traits extracellular matrix disassembly organization cell migration negative regulation apoptosis 4.4e−11 Fig Regulatory regions 86 genes enriched in binding sites for AP-1 MYC MEF2 KLF transcription factors factors implicated in EMT represent conserved regulators 86 genes variable expression levels among cancer cells human lung tumors syngeneic mouse tumor models scRNA-seq data healthy epithelium mouse tissues 86 genes highly correlatedsuggests expression program not artifact culture experiments coexpressed in vivo contribute to E/M heterogeneity EMT coordinated by diverse transcription factor conserved EMT genes regulated by shared regulatory factors Fig represent small fraction of differentially expressed genes sought determine if remainder EMT expression dynamics driven by common regulatory program distinct expression patterns canonical EMT transcription rarely differentially expressed (Fig. 3a). some not detected insufficient genes detectable show dynamics throughout EMT response Fig 9) scored cell for coexpression of transcription factors target genes identified differential activity EMT transcription factor activity context specific restricted to one to two time course experiments (Fig. 3b). Several factors well conserved AP-1 (JUN NFkB-associated RELB ATF4 SOX4 KLF6 regulons showed frequent activation ELF3 MYBL2 activity decreased (Fig. 3c). factors implicated in EMT not typically canonical EMT regulators29–34accuracy used ATAC-seq accessibility transcription factor motifs compared accessibility dynamics inferred regulon activity validation OVCA420 TGFB1 time course (Fig. smallest data set scRNA-seq cohort approach condition lowest power inferring transcription factor activity found motif accessibility EMT-mirrored regulon activity measured scRNA-seq data (Fig. 3e). supports regulon activity inference accurate representation transcription factor activity.Fig. 3Inferring transcription factor activity EMT Plot EMT transcription factors differentially expressed transcription factors active Heatmap EMT-associated changes regulons differentially active six course experiments colormap pseudotime beta coefficient model each regulon Differential accessibility transcription factor motifs ATAC-seq data OVCA420 cells treated TGFB1 0 1 3 7 days colormap accessibility Z-score each transcription factor motif Examples transcription factors Regulon activity score transcription factors OVCA420 TGFB1 time course experiment dot represents cell colored time point black line modeled trend model Shaded bands standard error.Kinase inhibitor screens reveal signaling dependencies EMT responsesParacrine signaling regulatory feature EMTexpression secreted factors signaling pathways increased 12 time courses (Fig. 4a established experimental design assess EMT on signaling pathways compare contexts curated 22 small molecule inhibitors targeting kinases treated cell lines 7 days three EMT inducers (Fig. MULTI-seq multiplex samples generated scRNA-seq profiles for 45,911 cells 384 conditions.Fig. 4Kinase inhibitor screens identify signaling dependencies EMT Gene set score KEGG pathway “cytokine–cytokine receptor pseudotime each Shaded bands standard error Heatmap EMT-associated changes genes significant change Schematic 384-sample design kinase inhibitor screen Heatmap cells annotated each condition after demultiplexing scRNA-seq data Summary genes differentially expressed each cell line exposed inhibitors without EMT induction Average pseudotime values each condition Boxplots distribution pseudotime values A549 cells treated inhibitors or TGFB1 horizontal black line median value box spans 25th 75th percentiles whiskers 1.5 times interquartile rangeUMAP embeddings untreated A549 cells treated TGFB1 TGFBR1 inhibitor LY364947 RIPK1 inhibitor necrostatin-5 Heatmap expression-score genes expressed A549 cells TGFB1 treated TGFB1 necrostatin-5 Difference normalized enrichment scores transcription factor targets genes inhibited necrostatin-5 Positive values correspond regulons enriched necrostatin-5-inhibited genes negative values affected necrostatin-5 retrieved cell counts drop-out patterns cell line-dependent -independent cytotoxic effects observed (Fig. 4d impact inhibitors EMT progression calculated pseudotime values inhibited cells models inhibitors reduced pseudotime values 7 days uninhibited controls dampening EMT response (TGFBR1 abrogated TGFB1-induced EMTs erlotinib gefitinib (EGFR inhibitors blocked effects EGF effects inhibitors limited blocking signaling EMT-inducing factor TGFBR1 inhibition partially blocked EMT progression conditions EGF-treated A549 OVCA420 cells TNF-treated A549 MCF7 cellssuggests activation paracrine TGFB1 signaling critical for EMT progression supporting work EMT on-activated TGFB1 autocrine EGFR inhibition with erlotinib gefitinib restricted to EGF-treated EMT responses inhibition downstream kinase MEK PD 0325901 hindered EMT response in TGFB1-treated A549 MCF7 cells Non-canonical TGFBR1 signaling through MEK/ERK studies proposed MEK-dependent regulatory checkpoint EMT data TGFB1-treated A549 MCF7 cells checkpoint not universal DU145 OVCA420 cells not susceptible to MEK inhibition 4f).Inhibition kinase activating NFkB necroptosis necrostatin-5 blocked EMT progression TGFBR1 inhibition Nec-5-treated cells had higher pseudotime values than TGFBR1-inhibited cells suggesting partial EMT response reduced gene expression changes inhibition assessed expression levels all genes TGFB1 treatment RIPK1 inhibition abrogated TGFB1-induced expression changes partial EMT response partial response not due to temporal block EMT progressionpreventing late EMT inhibition prevent late response genes Fig. EMT involves multiple regulatory modules without no direct cross-talk between TGFB1 signaling RIPK1 loss of RIPK1 associated with enhanced epithelial phenotype reduced ERK1/2 phosphorylation reduced transcriptional activity AP-1 RIPK1 inhibition prevents activation AP-1 targets EMT models assessed transcription factor binding motifs in genes RIPK1-inhibited cells AP-1 binding site most enriched in genes Nec-5-treated cells TGFB1 (Fig. Other enriched motifs include EGR1 PAX4 binding sites AP-1 EGR1 activated through ERK1/2 signaling possible mechanistic link between RIPK1 transcriptional changes46 ERK1/2 downstream effector of MEK may explain MEK checkpoint of EMT unclear how RIPK1 regulatory axis conserved condition dependent on TGFB1 signaling may represent common regulatory network of EMT EMT complex cellular process independent regulatory networks context specificityfindings argue paradigm of cells linear transition between epithelial mesenchymal programs oversimplification erroneous conclusions variety of EMT responses single mesenchymal gene expression program exist variety responses makes full EMT indefinable combination likely partial EMT states likely inumerable partial state defined as cells co-expressing epithelial mesenchymal markers studies relied on small markers most markers inconsistently involved in transition importance of gradation epithelial mesenchymal phenotypes comprehensive definition of intermediate polar states required study steps understanding complexity of E/M plasticity single-cell technologies scalable learn complete manifold of states E/M plasticity for cell type Unique environment developmental history manifold vary cell possible learn models for prediction alignment across settings critical to understand positions along manifold with phenotypic traits cell perturbations promote dynamics comprehensive model of E/M plasticity gain understanding of cellular heterogeneity understanding development tissue homeostasis disease progression information strategies to modulate cellular phenotypes in diseaseMethodsCell cultureA549 DU145 MCF7 cells obtained from ATCC (CCL-185 HTB-81 HTB-22 cells provided by Dr. Gordon Mills cells cultured in Dulbecco’s Modified Eagle Medium with 4.5 g/L glucose L-glutamine sodium pyruvate supplemented 10% fetal bovine serum cultured at 37 °C 5% CO2.EMT time course cell 10,000 cells plated into 96-well plate Fig. 1a addition TGFB1 EGF TNF scheduled Cells treated with 10 ng/mL TGFB1 30 ng/mL EGF 10 ng/mL TNF Media fresh TGFB1 EGF TNF added every 2 days constant concentrations over-confluence cells passaged not last 2 days avoid artifacts After cells processed for scRNA-seq multiplexing experiments performed twice two replicates parallel two 10x libraries generated Samples first replicate labeled “Mix1” “Mix2”, second labeled “Mix3a/b” “Mix4a/b”.Kinase inhibitor 10,000 cells plated into four 96-well plates Fig. 4cCells treated with small molecule kinase inhibitors Fig. 4c 10 ng/mL TGFB1 30 ng/mL EGF 10 ng TNF No-inhibitor No-EMT-inducer controls included inhibitors concentration 1 μM (Cayman Chemical Kinase Screening Library Item 10505 Batch 0537554) EMT inducers kinase inhibitors refreshed daily after culture media After 7 days samples processed for scRNA-seq multiplexing-seq reagents provided by Dr. Zev Gartner culture media removed washed with 1× Dulbecco’s-buffered saline lipid-modified DNA oligonucleotide unique sample barcode oligonucleotide added 200 nM to 0.05% trypsin 0.53 mM EDTA Cells incubated 5 min at 37 °C mixed After 5 lipid-modified co-anchor added 200 nM incubated additional 5 min 37 °C cells lifted trypsin neutralized with cultured media cells mixed single-cell suspension Samples transferred to V-bottom 96-well plates pelleted at 400 × g 5 minBarcode media removed cells washed with PBS + 1% bovine serum albumin Washes performed twice cells resuspended in PBS + 1% BSA pooled repelleted PBS + 1% BSA Viability cell counts performed before preparation scRNA-seq libraries library preparation sequencingSingle-cell suspensions processed using 10× Genomics Single Cell 3′ RNA-seq kit (v2 time course v3 kinase Gene expression libraries prepared manufacturer’s protocol MULTI-seq barcode libraries retrieved prepared independently barcode libraries separated from cDNA libraries size selection PCR-amplified prior Final libraries sequenced on NextSeq500 (Illumina). course reads per cell v2 20,000–25,000 reads per kinase inhibitor detected median 3649 genes 17,330 UMIs per cell kinase inhibitor 2360 genes 7634 UMIs raw sequencing using CellRanger v2.2.0 v3.0.2 inhibitor GRCh38 build human genome used both --expect-cells = 25,000 default parameters used for samplesMULTI-seq barcode libraries trimmed to 26 or 28 bp (v3 using Trimmomatic48 (v0.36) prior demultiplexing deMULTIplex R package (v1.0.2) concepts in McGinnis et al.13 tool barcode sequencing reads counts 96 barcodes assesses distribution determines optimal quantile threshold cell positive Cells positive more than one barcode doublets removed cells positive for single barcode retained for downstream analysis each barcode corresponds sample sample annotations added to all cells quality control performed independently on each 10× Genomic library with Seurat v3.0.2 (ref Expression matrices loaded into R as seurat objects retaining cells with >200 genes detected Cells with high percentage mitochondrial gene expression removed subsetted data independent seurat objects for each kinase inhibition experiment Each condition processed independently removed genes in <1% cells expression values normalized with library size scaling log transformation top 3000 variable genes detected using variance-stabilizing transformation selection method SeuratExpression values scaled technical factors regressed mitochondrial reads RNA molecules detected cycle scores time course data batch included initial PCA run on variable genes UMAP embeddings based on PCA genes pseudotemporal ordering UMAP embeddings calculated from first 30 components.Pseudotemporal ordering cellsPseudotime models built using R package psupertime v0.2.1 (ref on top 3000 variable genes each condition based on logistic regression scRNA-seq data identifying linear combination genes cells label order omitted treatment withdrawal samples pseudotime values for new data calculated matrix multiplication between matrix pseudotime model expression matrix new data used values treatment withdrawal samples kinase inhibition experiments rescaled values from 0 to 1 in multiple models.Differential expression analysisFor expression dynamics of each gene transcription factor regulon score function pseudotime modeled using generalized additive model function R package mgcv model exp ~ s(pseudotime, k = 4) + batch smoothing parameter estimation method to restricted maximum likelihood basis functions (k) chosen residuals randomly distributed P-values smoothed pseudotime function gene adjusted using p.adjust() function R Benjamini–Hochberg method genes vary pseudotime low effect sizes evaluated significant genes p < 0.05) top 2000 variable genes each experiment others relevant signal data low assessing transcription factor activity cytokine production directionality change pseudotime same approach removed smoothing function model single coefficient pseudotime covariate activity increased decreased transition kinase inhibition experiment assessed differentially expressed genes cell lines treated kinase inhibitor no EMT inducer used gam() function mgcv package model inhibitor no-inhibitor controls intercept quantified number genes adjusted p < 0.05.Calculating smoothed expression used models differential expression calculated fit values 200 pseudotime values minimum maximum.Gene set enrichment R package fgsea51 Input genes ranked variance values vst FindVariableFeatures() or adjusted p-value differential expression analysis Reference gene sets collected Molecular Signatures Database (MSigDB) v6.2.Gene set EMT hallmark gene set KEGG pathway “cytokine–cytokine receptor interaction” AddModuleScore() function Seurat package Default parameters usedTranscription factor regulon scoring scores computed SCENIC Log-transformed expression values pySCENIC gene regulatory networks computed grnboost2 method enriched motifs identified ctx function cisTarget v9 databases regulatory features 500 5 kb 10 kb cells scored aucell function over-represented transcription factor motifs gene R package RcisTarget52 enriched transcription factor motifs cisTarget v9 transcription factor motif annotations hg19-tss-centered-10kb-10species.mc9nr database motif rankings enrichment gene lists calculated difference normalized enrichment scores) enriched motifs-seq sample preparation samples prepared OVCA420 cells treated 10 ng/mL TGFB1 0 1 3 7 days performed independently twice preparation Buenrostro et al nuclei extracted cells chromatin tagmented TDE1 transposase Nextera DNA Library Preparation Kit protocol 2.5 μL enzyme optimal tagmentation required 5 μL enzyme 37 °C 30 min mixing ATAC libraries amplified sequenced NextSeq500 150-cycle run ~50 M reads per samplereads aligned hg38 human genome Bowtie2 (ref. 54) peaks called MACS2. 55) parameters -q 0.01 --nomodel --shift --extsize 200 --SPMR --broad Differential motif accessibility calculated R package chromVAR (ref. 56). summits peaks merged expanded 250 bp window summit Motifs human_pwms_v2 list mapped peaks matchMotifs() deviations computed deviations identified differentialDeviations() Nature Research Reporting Summary.Supplementary information Peer Review File Supplementary Files Data 1 2 Reporting Summary | 49.4 | 0.66913 |
10.1038/s41467-021-21188-2 | PMC7873277 | Histone phosphorylation is a ubiquitous post-translational modification. Here the authors present a programmable chromatin kinase, dCas9-dMSK1, that enables controlled histone phosphorylation and specific gene activation. | Histone phosphorylation is a ubiquitous post-translational modification that allows eukaryotic cells to rapidly respond to environmental stimuli. Despite correlative evidence linking histone phosphorylation to changes in gene expression, establishing the causal role of this key epigenomic modification at diverse loci within native chromatin has been hampered by a lack of technologies enabling robust, locus-specific deposition of endogenous histone phosphorylation. To address this technological gap, here we build a programmable chromatin kinase, called dCas9-dMSK1, by directly fusing nuclease-null CRISPR/Cas9 to a hyperactive, truncated variant of the human MSK1 histone kinase. Targeting dCas9-dMSK1 to human promoters results in increased target histone phosphorylation and gene activation and demonstrates that hyperphosphorylation of histone H3 serine 28 (H3S28ph) in particular plays a causal role in the transactivation of human promoters. In addition, we uncover mediators of resistance to the BRAF V600E inhibitor PLX-4720 in human melanoma cells using genome-scale screening with dCas9-dMSK1. Collectively, our findings enable a facile way to reshape human chromatin using CRISPR/Cas9-based epigenome editing and further define the causal link between histone phosphorylation and human gene activation. | IntroductionDynamic epigenomic regulatory forces, including DNA methylation and post-translational modifications (PTMs) to histones, harmonize to control human gene expression1–3. Histone phosphorylation at serine residues 10 and 28 on histone subunit H3 (H3S10ph and H3S28ph, respectively), is one type of histone PTM that has been correlated with stimulus-dependent gene expression4–9. Despite this correlation, defining the causal function of endogenous H3S10ph and H3S28ph has been challenging due to a lack of technologies to manipulate histone phosphorylation at diverse loci within native chromatin contexts.The mitogen- and stress-activated protein kinase 1 (MSK1) is one of nine human proteins known to catalyze H3S10ph and H3S28ph in vitro10. MSK1 is primarily localized to the nucleus, where it can be activated by ERK or p38 mitogen-activated protein kinase (MAPK)-mediated phosphorylation and subsequent autophosphorylation11,12. MSK1-driven H3S10ph/H3S28ph has been correlated with the transactivation of stimulus-responsive genes through chromatin immunoprecipitation (ChIP) assays4,5. Furthermore, genome-wide analyses of H3S28ph levels suggest that MSK1-mediated histone phosphorylation is tightly linked to gene expression from most stress-responsive human promoters8. In addition, the artificial recruitment of MSK1 to endogenous NF1 transcription factor binding sites using an NF1-MSK1 fusion protein results in hyperphosphorylation of nearby histone H3S10 and H3S28 residues and increased gene expression from adjacent promoters13.The CRISPR/Cas9 system has been repurposed for programmable genome editing in human cells14–16. In parallel, nuclease-null deactivated CRISPR/Cas platforms have also been developed to manipulate endogenous histone PTMs at targeted loci17–22. However, no CRISPR/Cas-based tools have been created that permit the locus-specific modification of histone phosphorylation.Here we construct a fusion protein, called dCas9-dMSK1, that consists of the deactivated Cas9 protein from Streptococcus pyogenes (dCas9) and a hyperactive variant of human MSK1. We show that dCas9-dMSK1 permits locus-specific manipulation of H3S10ph and H3S28ph at targeted human loci and in turn the activation of gene expression from human promoters. Our work demonstrates that histone phosphorylation plays a causal role in the activation of human promoters, establishes a new way to engineer the human epigenome, and expands the dCas9-based epigenome editing arsenal.ResultsDevelopment of a CRISPR/Cas9-based histone kinaseHuman MSK1 has been linked to the deposition of H3S10ph/H3S28ph and transcriptional activation7,9–13, and our preliminary results showed that knockout of MSK1 causes changes to RNA polymerase II-mediated transcription in human cells (Supplementary Fig. 1, Supplementary Tables 1 and 2). Therefore, we hypothesized that dCas9 could be used to recruit MSK1 to individual human loci and clarify the casual role that locus-specific histone phosphorylation plays in human gene expression. To test this hypothesis, we constructed three fusion proteins between the C-terminus of dCas9 and the N-terminus of different MSK1 variants (Fig. 1a). Specifically, we fused dCas9 to full-length WT MSK1 (dCas9-MSK1), to MSK1 lacking an N-terminal inhibitory domain (NID; Supplementary Fig. 2) that has previously been observed to limit catalytic activity on chromatin assembled in vitro (dCas9-dMSK1), and to a catalytically inactivated version of dMSK1 (dCas9-ddMSK1)10. dCas9-MSK1 and dCas9-dMSK1 both displayed hallmarks of natural MSK1 catalytic activity (autophosphorylation of MSK1 serine residues 212 and 376)12 when transfected into human cells, whereas the catalytically inactive dCas9-ddMSK1 fusion did not (Fig. 1b). Moreover, both dCas9-MSK1 and dCas9-dMSK1 were able catalyze H3S10ph and H3S28ph when incubated with in vitro assembled human histone octamers (Fig. 1c), however, dCas9-dMSK1 exhibited a significantly (P = 0.029) greater level of enzymatic activity on H3S28 than dCas9-MSK1 (Fig. 1d). Together these results demonstrate that dCas9-MSK1 and dCas9-dMSK1 are catalytically active in human cells, able to phosphorylate human histones in vitro in the absence of auxiliary cellular cofactors, and that dCas9-dMSK1 harbors hyperactive histone H3S28 phosphorylation kinase activity relative to dCas9-MSK1.Fig. 1Development of a CRISPR/Cas9-based histone kinase.a Schematics of dCas9, dCas9 fused to WT MSK1 (dCas9-MSK1), dCas9 fused to a hyperactive truncated MSK1 (dCas9-dMSK1), and dCas9 fused to a catalytically inactivated version of dMSK1 (dCas9-ddMSK1). b dCas9 and dCas9-MSK1 fusion variants were transiently transfected into HEK293T cells and autophosphorylation levels were detected (at serine residues 212 and 376 of MSK1) by Western blot 72 h post-transfection. Data are representative of three independent biological experiments. c Purified dCas9, dCas9-MSK1, dCas9-dMSK1, and dCas9-ddMSK1 were incubated with human histone octamers in vitro and resulting levels of histone H3 phosphorylation at serine residues 10 and 28 (H3S10ph and H3S28ph, respectively) were measured after incubation for 1 h. d The relative levels of H3S10ph and H3S28ph were quantified using densitometry from three different in vitro histone kinase assays. Two-sided t-test, *P < 0.05; n = 3 independent experiments in panel (d); error bars, s.e.m.; ns, not significant; kDa, kilodaltons. Source data are available in the Source data file.dCas9-dMSK1 activates natural MSK1 targetsTo test whether the catalytically active dCas9-MSK1 and dCas9-dMSK1 fusion proteins could modulate endogenous human histone phosphorylation and/or gene expression, we first identified the natural targets of MSK1 using comparative RNA-seq. Two independent MSK1 knockout (KO) clones were generated by disrupting conserved exons in MSK1 (exons 1 and 2, respectively) using Cas9-mediated KO (Supplementary Fig. 1a,b). Each clonal MSK1 KO cell line displayed strikingly similar transcriptome-wide changes compared to WT HEK293T cells (Supplementary Fig. 3). Together, comparative RNA-seq between WT and MSK1 knockout cell lines resulted in 41 and 24 shared (between both knockout lines) downregulated and upregulated genes, respectively (Fig. 2a).Fig. 2dCas9-dMSK1 activates natural MSK1 target genes.a Differentially expressed genes in MSK1 knockout HEK293T cell lines compared to WT HEK293T cells are shown (blue circles, downregulated genes; red circles, upregulated genes). Data were analyzed using the Wald test and the adjusted P value (Padj) was calculated using the Benjamini and Hochberg method. b mRNA levels for the top five downregulated genes were measured by RT-qPCR at 72 h post-transfection of indicated dCas9-fusion proteins (or dCas9 control) and corresponding gRNAs. c, d ChIP-qPCR for H3S10ph and H3S28ph at the PRKCB and BMP2 promoters, respectively, 72 h post-transfection with the indicated dCas9-fusion proteins or dCas9 control. Two-sided t-test, *P < 0.05; n = 3 independent experiments for panels (b–d); error bars; s.e.m.; ns, not significant. Source data are available in the Source data file.We next designed guide RNAs (gRNAs) to recruit each fusion protein to the promoter regions of the top five most significantly (adjusted P value <0.01) downregulated genes (PRKCB, BMP2, SHROOM2, ZNF462, and GDF6) in MSK1 KO HEK293T cells (Supplementary Fig. 4) to test whether dCas9-MSK1 and/or dCas9-dMSK1 could activate natural MSK1 targets (i.e., genes downregulated upon the loss of MSK1). RT-qPCR revealed that dCas9-dMSK1, but not dCas9-MSK1, significantly activated BMP2 (P = 0.002) and GDF6 (P = 0.008) relative to a dCas9 control when targeted to each respective promoter region using specific gRNAs (Fig. 2b). Notably, dCas9-MSK1 was unable to activate any target genes, which we attribute to MSK1’s reliance upon co-activators (e.g., ATF and/or CREB) to penetrate endogenous human chromatin and phosphorylate histones, a requirement that dMSK1 (lacking the NID) does not appear to share10. This capability of dCas9-dMSK1 could be instrumental in the analysis of regulatory mechanisms controlling stimulus-induced transcription, for instance, by studying the role of CREB/ATF1 phosphorylation and recruitment of associated reader proteins in response to MSK1 targeting. Not all genes that were downregulated upon the loss of MSK1 were responsive to dCas9-dMSK1-mediated gene activation (Fig. 2b) suggesting that some downregulated genes were either indirect targets of MSK1 (i.e., regulated instead by direct targets of MSK1) and/or were not influenced by dCas9-dMSK1 mediated epigenetic changes at the gRNA sites targeted. To measure MSK1 occupancy at PRKCB, BMP2, SHROOM2, GDF6, and ZNF462, we performed ChIP-qPCR at the promoter region (Supplementary Fig. 4) of each gene in WT and MSK1 KO HEK293T cells. These assays demonstrate that MSK1 is significantly enriched at the promoter regions of BMP2 (P = 0.0001) and GDF6 (P = 0.0009), but not at the promoter regions of PRKCB, SHROOM2, or ZNF462 (Supplementary Fig. 5). Coupled with the fact that BMP2 and GDF6 can be activated by dCas9-dMSK1, whereas PRKCB, SHROOM2, and ZNF462 cannot, these results support the hypothesis that BMP2 and GDF6 are direct targets of MSK1, whereas PRKCB, SHROOM2, and ZNF462 appear to be indirectly affected by the loss of MSK1.To further investigate the mechanisms underlying dCas9-dMSK-mediated gene activation at responsive (BMP2) and non-responsive (PRKCB) loci, we used ChIP-qPCR to measure histone phosphorylation levels. Interestingly, despite high levels of dCas9-dMSK1 binding (Supplementary Fig. 6), the PRKCB promoter displayed no changes in H3S10ph nor H3S28ph in response to dCas9-dMSK1 targeting (Fig. 2c). In contrast, dCas9-MSK1 and dCas9-dMSK1 both significantly (P = 0.012 and P = 0.008, respectively) elevated H3S10ph levels at the BMP2 promoter compared to targeting with a dCas9 control (Fig. 2d). However, only dCas9-dMSK1 efficiently phosphorylated both H3S10 and H3S28 at the BMP2 promoter (Fig. 2d). These results at endogenous histones mirror the findings using histone octamers in vitro (Fig. 1d) and demonstrate that dCas9-dMSK1 is more efficient than dCas9-MSK1 at phosphorylating histone H3S28 both in vitro and within native human chromatin. Furthermore, these results indicate that targeted, locus-specific H3S28ph is causal for the activation of gene expression from human promoters that are sensitive to histone phosphorylation.dCas9-dMSK1 activates human promoters with high specificityTo determine the effects of dCas9-dMSK1 on an expanded set of target genes and genomic regulatory elements, we delivered dCas9-fusion proteins to the distal enhancer (DE), proximal enhancer (PE), and proximal promoter (PP) of OCT4 (Supplementary Fig. 7a). RT-qPCR 72 h after transient transfection in HEK293T cells showed that dCas9-dMSK1 was capable of potently activating OCT4 expression (Fig. 3a), but that this potency was concentrated to ~260–103 bp upstream of the OCT4 transcription start site (TSS; Supplementary Fig. 7b). We also targeted dCas9 and dCas9-MSK1 fusion proteins to the distal regulatory region (DRR), core enhancer (CE), and promoter of MYOD (Supplementary Fig. 8a) and observed a similar potency of dCas9-dMSK1 driven gene activation that was localized to the MYOD promoter region (~273–35 bp upstream of the MYOD TSS; Fig. 3b, Supplementary Fig. 8b). These data show that dCas9-dMSK1 can synthetically activate non-natural MSK1 target genes and that human promoters appear to be more sensitive to the regulatory effects of endogenous histone phosphorylation than distal enhancers.Fig. 3dCas9-dMSK1 activates endogenous human promoters with high genome and transcriptome-wide specificity.a, b RT-qPCR for OCT4 or MYOD mRNA levels 72 h post-transfection of dCas9, dCas9-MSK1, dCas9-dMSK1, or dCas9-ddMSK1 and 4 corresponding promoter-targeting gRNAs. Two-sided t-test, *P < 0.05; n = 3 independent experiments for both panels; error bars, s.e.m. c DESeq2 analysis of FLAG ChIP-seq binding data from HEK293T cells transiently co-transfected with dCas9-dMSK1 and four OCT4 promoter-targeting gRNAs compared to HEK293T cells transiently co-transfected with dCas9-dMSK1 and a non-targeting gRNA. Data were analyzed using the Wald test. Red circles indicate false discovery rate (FDR) < 0.01; d DESeq2 analysis of RNA-seq data from HEK293T cells transiently co-transfected with dCas9-dMSK1 and four OCT4 promoter-targeting gRNAs compared to HEK293T cells transiently co-transfected with dCas9-dMSK1 and a non-targeting gRNA. Data were analyzed using the Wald test and the adjusted P value (Padj) was calculated using the Benjamini and Hochberg method. Significantly upregulated mRNAs (OCT4 isoforms) are shown as red circles (Padj < 0.05) and a significantly downregulated gene (SEPT7P3) is designated with a blue circle (Padj < 0.05). Source data are available in the Source data file.To assess the specificity of dCas9-dMSK1, we performed ChIP-seq and RNA-seq in HEK293T cells co-transfected with dCas9-dMSK1 and four OCT4-targeting gRNAs or with dCas9-dMSK1 and a non-targeting gRNA control used previously23. Our ChIP-seq results showed that dCas9-dMSK1 binding to the OCT4 promoter was highly specific (FDR = 0.0006) across the human genome (Fig. 3c), with only one significant (FDR = 0.0086) off-target within an intron of CROCC2 identified. RNA-seq demonstrated that dCas9-dMSK1 targeted to the OCT4 promoter specifically and significantly activated two different isoforms of OCT4 (ENST00000638788.1 and ENST00000461401.1; Padj of 0.046 and 0.008, respectively) across the human transcriptome (Fig. 3d). One off-target transcript was detected upon dCas9-dMSK1 targeting to the OCT4 promoter in HEK293T cells; SEPT7P3 (Septin 7 pseudogene 3), which was significantly (Padj = 0.011) downregulated. Nevertheless, our results collectively demonstrate that dCas9-dMSK1 binding and activation of promoters are highly specific across the human genome and transcriptome, respectively.Consistent with our observations above, dCas9-MSK1 and dCas9-dMSK1 both significantly (P value <0.05) increased H3S10ph levels compared to dCas9 when targeted to the OCT4 and MYOD promoters (Fig. 4a, b). However, only dCas9-dMSK1 resulted in increased H3S10ph and H3S28ph levels at these targeted promoters, which was coincident with the activation of gene expression (Fig. 3a, b). H3S28ph ChIP-seq analysis confirmed our ChIP-qPCR observations that H3S28ph was enriched at the OCT4 promoter subsequent to dCas9-dMSK1 targeting, however, this H3S28ph enrichment was not statistically significant above background on a genome-wide scale (using cutoffs of log2 fold change >1 and FDR < 0.01; Supplementary Fig. 9). Interestingly, previous studies suggest that MSK1 phosphorylates either H3S10 or H3S28 but not both upon the same histone tail24–26. To evaluate if dCas9-dMSK1 could artificially deposit both H3S10ph and H3S28ph on the same histone tail, we performed re-ChIP analysis (Fig. 4c, d). In agreement with previous findings, our re-ChIP analyses indicated that dCas9-dMSK1-mediated phosphorylation of H3S10 occurred independently of H3S28 phosphorylation at the OCT4 and MYOD promoters (Fig. 4c, d, respectively), as evidenced by the lack of enrichment of H3S28ph after specific enrichment of H3S10ph via ChIP. Collectively our results show that H3S28ph in particular plays a causal role in the activation of gene expression from natural and non-natural MSK1 target genes, and moreover, highlights the specificity and mechanistic utility of dCas9-dMSK1.Fig. 4dCas9-dMSK1 induces target gene activation via H3S28ph.a, b ChIP-qPCR enrichment for H3S10ph and H3S28ph, at the OCT4 (panel a) or MYOD (panel b) promoters 72 h post-transfection of dCas9 or indicated dCas9-fusion protein and corresponding gRNAs. c, d Re-ChIP-qPCR enrichment for H3S10ph and H3S28ph at OCT4 (panel c) or MYOD (panel d) promoters 72 h post-transfection of dCas9 or indicated dCas9-fusion protein and corresponding gRNAs. Two-sided t-test, *P < 0.05; n = 3 independent experiments for all panels; error bars, s.e.m.; ns, not significant. Source data are available in the Source data file.dCas9-dMSK1 driven H3S28ph influences local H3K27ac statusH3S28ph has previously been shown to influence, and be influenced by, the dynamics of surrounding histone PTMs, especially the acetylation status of histone H3 lysine 27 (H3K27ac)4,8,9,13,26. Therefore, we also measured H3K27ac levels after targeting the OCT4 and MYOD promoters with dCas9 and dCas9-MSK1 fusion protein variants. Despite having no intrinsic histone acetyltransferase activity, targeting dCas9-dMSK1 to the OCT4 and MYOD promoters resulted in increased H3K27ac levels at both loci (Fig. 5a, b). Similarly, H3K27ac was enriched when dCas9-dMSK1 was targeted to the BMP2 promoter, but not when targeted to the PRKCB promoter (Supplementary Fig. 10). These results show that crosstalk exists between H3S28ph and H3K27ac at endogenous human loci and suggest that the interplay between H3S28ph and H3K27ac can result in the activation of gene expression from human promoters. Moreover, these observations are consistent with reports showing that H3S28ph promotes CBP/p300-dependent transcription in cell-free systems9.Fig. 5dCas9-dMSK1 influences H3K27 acetylation status at targeted human promoters.a, b ChIP-qPCR enrichment for H3K27ac at the OCT4 (panel a) or MYOD (panel b) promoters 72 h post-transfection of dCas9 or indicated dCas9-fusion protein and corresponding gRNAs. c, d RT-qPCR for OCT4 (panel c) or MYOD (panel d) in the presence of DMSO or the CBP/p300 inhibitor A485 72 h post-transfection of dCas9 or indicated dCas9-fusion protein and corresponding gRNAs. 20 µM of A485 or an equal volume of DMSO was added to cells 12 h post-transfection. Two-sided t-test, *P < 0.05; n = 3 independent experiments for all panels; error bars, s.e.m.; ns, not significant. Source data are available in the Source data file.To further investigate the relationship between histone phosphorylation and other histone PTMs, we measured how chemical inhibition of CBP/p300 histone acetyltransferases impacted dCas9-dMSK1 driven gene activation. Specific inhibition of CBP/p300 activity using A485 significantly reduced the efficacy of dCas9-dMSK1-mediated transactivation of OCT4 (P = 0.008) and MYOD (P = 0.0008) promoters (Fig. 5c, d), demonstrating that the activity of CBP/p300 appears to be functionally linked to the transactivation capacity of dCas9-dMSK1 at endogenous human promoters. In addition, WT MSK1 has been shown to physically and functionally interact with the KMT2A/MLL1 methyltransferase complex, which can catalyze methylation of H3K4 (ref. 27). Therefore, we measured the enrichment of H3K4me3 after targeting the OCT4 and MYOD promoters with dCas9 or dCas9-MSK1 fusion proteins (Supplementary Fig. 11). Although we did not observe any significant (P < 0.05) increases in H3K4me3 enrichment after targeting these promoters with dCas9-MSK1 nor dCas9-dMSK1, we note that as synthetic, programmable dCas9-based histone kinases, dCas9-MSK1 and dCas9-dMSK1 may not mechanistically function in exactly the same way(s) as WT MSK1 in human cells.We also targeted dCas9-fusion proteins to the HBA1, SOX2, and KLF4 promoter regions (Supplementary Fig. 12a–c) in HEK293T cells and measured gene expression 72 h post transient transfection (Supplementary Fig. 12d–f). dCas9-dMSK1 significantly (P value <0.05) activated gene expression from each targeted promoter, although to a lesser extent than OCT4 or MYOD. The reduced potency at these loci may be due in part, to differences in basal gene expression levels (Supplementary Table 3), as has been observed previously with dCas9-based activators28. Interestingly, each additional gRNA used to recruit dCas9-dMSK1 to a target promoter resulted in additive increases in gene activation, suggesting that local levels of histone phosphorylation are directly proportional to gene expression at responsive promoters (Supplementary Fig. 13). We also found that dCas9-dMSK1 was able to activate gene expression in A549 cells (Supplementary Fig. 14), and furthermore, that another human histone kinase (Aurora B) could also activate gene expression when fused to dCas9 and targeted to a human promoter (Supplementary Fig. 15). Altogether, these results show that dCas9-dMSK1 can be used to engineer the phosphorylation status of endogenous human histones which in turn can result in the activation of gene expression from human promoters. This capability is functional at diverse human promoters and in different human cell types, and can be achieved using other histone kinases, further establishing the causal role and functional importance histone phosphorylation at endogenous human loci.Genome-scale screening for mediators of PLX-4720 resistancedCas9-based tools can be used in unbiased approaches to screen the noncoding genome in high-throughput28,29. To test the efficacy of dCas9-dMSK1 at thousands of human promoters in high-throughput and to evaluate if targeted histone phosphorylation could uncover mediators of pathological gene expression, we combined dCas9-dMSK1 with a genome-scale gRNA library to identify genes that when synthetically overexpressed by dCas9-dMSK1, would result in resistance to the BRAF V600E inhibitor PLX-4720 (refs. 28,30). Briefly, we stably transduced A375 melanoma cells with a dCas9-dMSK lentiviral expression vector and then transduced these cells with a previously developed CRISPR activation (CRISPRa) gRNA library28 at a multiplicity of infection (MOI) of 0.2. Cells were then subjected to vehicle (DMSO) or PLX-4720 treatment for 16 days, after which we used next-generation sequencing to recover gRNA sequences enriched in the PLX-4720 treated population relative to DMSO control-treated cells (Fig. 6a).Fig. 6Genome-scale screening using dCas9-dMSK1 identifies mediators of BRAF V600E inhibitor resistance.a Flow-chart of the genome-scale screening regime using dCas9-dMSK1. Puro; puromycin. b Box plots showing the distribution of gRNA frequencies after lentiviral transduction in DMSO or PLX-4720 treated A375 cells. In all box plots, the green horizontal line represents the median and the upper and lower bounds of each box correspond to the 25th and 75th percentiles of each sample, respectively. The upper and lower whiskers extend to the maximum and minimum values, respectively, within 1.5-fold of the interquartile range of the box bounds. Outliers are plotted as green dots. n = 2 independent experiments. c Volcano plot displaying gRNA counts in PLX-4720 versus DMSO treated A375 cells. Data were analyzed using robust rank aggregation (RRA) with a cutoff of P value <0.01. Blue circles indicate depleted gRNAs, red circles indicate enriched gRNAs, and gray circles indicate gRNAs with no change. d mRNA levels of the top 10 most enriched genes in A375 cells stably expressing dCas9 or dCas9-dMSK1 when targeted with corresponding gRNAs identified from the screen. Two-sided t-test, *P < 0.05; n = 3 independent experiments for panel (d); error bars, s.e.m. Source data are available in the Source data file.dCas9-dMSK1 was well-expressed and catalytically active in A375 cells (Supplementary Fig. 16), and in both experimental replicates, we observed increased enrichment of gRNAs in PLX-4720 treated samples relative to DMSO treated control cells (Fig. 6b). Based on the ratio of normalized gRNA counts in PLX-4720 vs. DMSO treated cells, there were 5757 significantly (P value <0.01) depleted gRNAs (corresponding to 4889 targeted genes/promoters) and 315 significantly (P value <0.01) enriched gRNAs (corresponding to 314 targeted genes/promoters) in the PLX-4720 treated cells (Fig. 6c, Supplementary Table 4). Eighteen of the 314 enriched genes were identified previously using the dCas9-SAM platform28. The differences in enriched gRNAs between dCas9-SAM and dCas9-dMSK1 could be due to the lower relative potency of dCas9-dMSK1 compared to dCas9-SAM (Supplementary Fig. 17), and/or unique effects associated with histone phosphorylation at the targeted promoters identified in this study.Three of the top 10 genes identified as mediators of PLX-4720 resistance in our screen; EPDR1, AFF2, and ERC2, have been linked to the BRAF protein in colorectal cancer31, autism spectrum disorders32, and ganglioma33, respectively. MRPS15, TRAT1, LACC1, AGL, TDRP, MIPOL1, and LELP1 are mediators of PLX-4720 resistance that have not been previously reported. To validate our screening results, we targeted dCas9-dMSK1 (or a dCas9 control) to the promoters of EPDR1, AFF2, ERC2, MRPS15, TRAT1, LACC1, AGL, TDRP, MIPOL1, and LELP1 using the gRNAs enriched from the screen (Fig. 6d). In each case, dCas9-dMSK1 significantly (P value <0.05) increased the expression of each targeted gene. To confirm that the dCas9-dMSK1-mediated induction of hits identified in our screen, we used dCas9-dMSK1 (and corresponding gRNAs) to induce the overexpression of EPDR1 or AFF2 (2 of the top hits from our screen) in A375 cells and then treated cells with either DMSO (control) or 3.5 µM PLX-4720 for 72 h. Microscopy analysis and MTT assays indeed revealed that dCas9-dMSK1-mediated upregulation of EPDR1 or AFF2 results in improved cell fitness and higher viability compared to dCas9 control-treated A375 cells when challenged with PLX-4720 (Supplementary Fig. 18). These results demonstrate that dCas9-dMSK1-mediated histone phosphorylation and gene activation are compatible with high-throughput gRNA screening approaches, and that programmable histone phosphorylation can be leveraged to uncover unique insights into human gene regulation and pathology.In summary, we have built a CRISPR/Cas9-based epigenome editing tool, called dCas9-dMSK1, that permits writing histone phosphorylation at specific endogenous loci within native human chromatin. Using this capability, we have shown that histone phosphorylation plays a causal role in the activation of human promoters, and that H3S28ph is pivotal for this function. Furthermore, we have utilized dCas9-dMSK1 to demonstrate that functional crosstalk exists between histone phosphorylation and acetylation at endogenous human histones, and that this interplay is linked to the activation of gene expression from human promoters. Finally, we used dCas9-dMSK1 in combination with high-throughput genome-scale screening to uncover genes involved in the development of therapeutic resistance. Our work expands the CRISPR/Cas9-based epigenome editing toolkit, clarifies the mechanistic role that histone phosphorylation plays in regulating human genes, and provides an efficient way to synthetically manipulate an important epigenomic modification that is ubiquitous across eukaryotes.MethodsCell culture and plasmid constructionHEK293T cells (ATCC, CRL-11268), A549 cells (ATCC, CCL-185), and A375 cells (ATCC, CRL-1619) were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco, 31-053-028) supplemented with 10% FBS (Sigma, F2442) and 1% penicillin/streptomycin and maintained at 37 °C and 5% CO2. The cloning backbone pLV-dCas9-p300-P2A-Puro (dCas9-P300, Addgene, 83889) has been described previously29. N-terminal MSK1 (1–1644 nt) was amplified from the pDONR223-RPS6KA5 (Addgene, 23758) and C-terminal MSK1 (1645–2406 nt) was synthesized as a gBlock gene fragment (Integrated DNA Technologies). The pLV-dCas9-MSK1-P2A-Puro construct (dCas9-MSK1) was created by subcloning full-length MSK1 into a BamHI-digested pLV-dCas9-p300-P2A-Puro backbone via NEBuilder HiFi DNA Assembly (NEB, E2621). The pLV-dCas9-dMSK1-P2A-Puro (dCas9-dMSK1) was generated by amplifying amino acids 42–801 of MSK1 from dCas9-MSK1 and then subcloning Δ41MSK1 into a BamHI-digested pLV-dCas9-p300-P2A-Puro backbone. The pLV-dCas9-ddMSK1-P2A-Puro (dCas9-ddMSK1) was created by amplification of dMSK1 from dCas9-dMSK1 in overlapping fragments with primer sets harboring the specified nucleic acid mutations (D195A and D565A), and then cloning this PCR fragment into the BamHI-digested pLV-dCas9-p300-P2A-Puro backbone using NEBuilder HiFi DNA Assembly (NEB, E2621). Protein sequences of all dCas9 constructs are shown in Supplementary Note 1. All gRNAs used for non-screening experiments were cloned into the pSPgRNA backbone (Addgene, 47108)34. All gRNA protospacer targets are listed in Supplementary Table 5. gRNA library plasmids28 for genome-scale screening were purchased from Addgene (1000000057).TransfectionTransient transfections were performed in 24-well plates using 375 ng of respective dCas9 expression vector and 125 ng of single gRNA vectors or equimolar pooled gRNA expression vectors. Plasmids were mixed with Lipofectamine 3000 (Invitrogen, L3000015) as per manufacturer’s instruction. For ChIP-qPCR experiments, HEK293T cells were transfected in 15 cm dishes with Lipofectamine 3000 and 37.5 μg of respective dCas9 expression vector and 12.5 μg of equimolar pooled gRNA expression vectors as per manufacturer’s instruction.Western blottingTwenty micrograms of protein was loaded for SDS-PAGE and transferred onto a PVDF membrane for Western blots. Primary antibodies (α-FLAG; Sigma-Aldrich, F1804; α-MSK1; Abcam, ab99412; α-MSK1S376ph; Abcam, ab32190; α-MSK1S212ph; Abcam, ab79499) were used at a 1:1000 dilution in 1X Tris Buffered Saline with 1% Casein (Bio-Rad, 1610782EDU). Secondary α-mouse HRP (Sigma-Aldrich, A6154) or α-rabbit HRP (Abcam, ab6721) were used at a 1:3000 dilution in 1X Tris Buffered Saline with 1% Casein (Bio-Rad, 1610782EDU). Membranes were exposed after addition of ECL (Bio-Rad, 170-5060). Tubulin was detected with hFAB™ Rhodamine Anti-Tubulin Primary Antibody (Bio-Rad, 12004166; 1:3000 dilution).Histone octamer in vitro kinase assaydCas9 and dCas9-fusion proteins were purified from transfected HEK293T cells using ANTI-FLAG® M2 Affinity Gel (Sigma, A2220) per manufacturer’s instruction. Kinase assays were performed at 30 °C for 1 h in 30 μl reactions containing 500 ng of assembled human histone octamer, 150 nM of dCas9 or dCas9-fusion protein, 10 mM Tris-HCl, pH 8.0, 100 mM KCl, 10 mM MgCl2, 0.1 M ATP and 1X phosphatase inhibitor (Thermo Scientific, 78425). The resulting reaction was suspended in SDS-PAGE sample buffer (Bio-Rad, 1610747), boiled at 95 °C for 10 min, separated by SDS-PAGE, and then transferred onto a PVDF membrane for Western blotting. After blocking with 1% casein for 1 h, phosphorylated histones were detected by immunoblotting with an appropriate phosphor-specific antibody at a 1:1000 dilution (α-H3S10ph; Abcam, 5176, or α-H3S28ph; Abcam, ab32388). Non-phosphorylated histones were detected using α-H3 (Abcam, ab1791; 1:1000 dilution). Secondary α-rabbit HRP (Abcam, ab6721; 1:3000 dilution) was used and membranes were exposed and imaged after the addition of ECL (Bio-Rad, 170-5060). Densitometry analysis was carried out using ImageJ.MSK1 knockout cell linesIn total, 2 gRNAs (listed in Supplementary Table 5) targeting either exon 1 or exon 2 of human MSK1 were cloned into the LentiCRISPR v2 vector (Addgene, 52961). The resulting gRNA constructs were transiently transfected to HEK293T cells. Forty-eight hours post-transfection, cells were passaged and 1 μg/ml puromycin was added 3 h after plating. After 4 days of puromycin selection, cells were harvested, diluted 1000-fold, and replated in 15-cm plates with complete medium supplemented with 1 μg/ml puromycin. Two weeks later, single colonies were picked and cultured within 96-well plates and after 1 week of growth, cells were passaged into 24-well plates. MSK1 knockout lines were confirmed by western blot analysis with α-MSK1 (Abcam, ab99412; 1:1000 dilution) and secondary α-rabbit HRP (Abcam, ab6721; 1:3000 dilution). The LentiCRISPR v2 empty vector was used as a control in WT HEK293T cells.RNA sequencingRNA sequencing (RNA-seq) was performed in duplicate for each experimental condition. RNA was isolated from transfected cells using the RNeasy Plus mini kit (Qiagen, 74136). RNA-seq libraries were constructed using the TruSeq Stranded Total RNA Gold (Illumina, RS-122-2303). The qualities of RNA-seq libraries were verified using the Tape Station D1000 assay (Tape Station 2200, Agilent Technologies) and the quantities of RNA-seq libraries were checked again using real-time PCR (QuantStudio 6 Flex Real time PCR System, Applied Biosystem). Libraries were normalized and pooled and then 75 bp paired-end reads were sequenced on the Hiseq3000 platform (Illumina). Reads were aligned to the Hg38 transcriptome using HISAT2(2.1.0)35. Transcript abundance was calculated using feature Counts from the subread package (v2.0.0)36, and differential expression was determined in R studio (v1.2.13) using the DESeq2 (v1.28.1) analysis package with default parameters37. Gene ontology analysis was performed using DAVID Functional Annotation Bioinformatics Microarray Analysis38 at https://david.ncifcrf.gov/. All RNA-seq reads have been uploaded to the NCBI single read archive and made publicly available upon publication acceptance.Reverse-transcription quantitative PCR (RT-qPCR)RNA was isolated from transfected cells using the RNeasy Plus mini kit (Qiagen, 74136) and 1 µg of purified RNA was used as template for cDNA synthesis (Bio-Rad, 1725038). Real-time quantitative PCR was performed using SYBR Green (Bio-Rad, 1725275) and a CFX96 Real-Time PCR Detection System with a C1000 Thermal Cycler (Bio-Rad, 1855195). Baselines were subtracted using the baseline subtraction curve fit analysis mode and thresholds were automatically calculated using the Bio-Rad CFX Manager software version 2.1. Results are expressed as fold change above cells transfected with an empty vector plasmid (Addgene, 47108) after normalization to GAPDH expression using the ΔΔCt method. All qPCR primers and conditions are listed in Supplementary Table 6.ChIP-qPCRHEK293T cells were co-transfected with indicated dCas9-fusion expression vectors and gRNA constructs in 15 cm plates in biological triplicates for each condition tested. Cells were cross-linked for 10 min at RT using 1% formaldehyde (Sigma F8775-25ML) and then the reaction was stopped by the addition of glycine to a final concentration of 125 mM. Cells were harvested and washed with ice-cold 1X PBS and suspended in Farnham lysis buffer (5 mM PIPES pH 8.0, 85 mM KCl, 0.5% NP-40) supplemented with 1X protease and phosphatase inhibitor (Thermo Scientific, 78426). Cells were then pelleted and resuspended in RIPA buffer (1X PBS, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS) supplemented with 1X protease and phosphatase inhibitor. Approximately 2.5e7 cells were used for each H3S10ph and H3S28ph ChIP experiment. Chromatin in RIPA buffer was sheared to a median fragment size of 250 bp using a Bioruptor XL (Diagenode). Two micrograms of each antibody (α-FLAG; Sigma-Aldrich, F1804, Mouse IgG; Abcam, ab18413, Rabbit IgG; Abcam, ab171870, α-H3S10ph; Abcam, ab17246, α-H3S28ph; Abcam, ab32388, α-H3K4me3; Abcam, ab8580, or α-MSK1; Abcam, ab99412) was incubated with 50 μl Sheep anti-Rabbit or Mouse IgG magnetic beads (Life Technologies, 11203D or 11202D) for 16 h at 4 °C, respectively. Antibody-linked magnetic beads were washed three times with PBS/BSA buffer (1X PBS and 5 mg/ml BSA) and then sheared chromatin was incubated with corresponding antibody-linked magnetic beads at 4 °C overnight and then washed five times with LiCl IP wash buffer (100 mM Tris pH 7.5, 500 mM LiCl, 1% NP-40, 1% sodium deoxycholate). Cross-links were then reversed via overnight incubation at 65 °C and DNA was purified using QIAquick PCR purification kit (Qiagen, 28106) for ChIP-qPCR or ChIP-seq. Input DNA was prepared from ~1.0e6 cells. Ten nanograms of DNA was used for subsequent qPCR reactions using a CFX96 Real-Time PCR Detection System with a C1000 Thermal Cycler (Bio-Rad, 1855195). Baselines were subtracted using the baseline subtraction curve fit analysis mode and thresholds were automatically calculated using the Bio-Rad CFX Manager software version 2.1. The ChIP-qPCR data is plotted relative to percent input. All ChIP-qPCR primers and conditions are listed in Supplementary Table 6.ChIP sequencingChIP-sequencing (ChIP-seq) libraries were prepared using the KAPA Hyper Prep Kit (Roche) following the manufacturer’s protocol. Briefly, 3 ng of fragmented DNA was used to repair the ends before ligation to a diluted NEXTflex DNA adapter (PerkinElmer). The adaptor-ligated DNA was then enriched using 10 cycles of PCR (1 cycle at 98 °C for 45 s; 10 cycles at 98 °C for 15 s, 60 °C for 30 s, 72 °C for 30 s; and 1 cycle at 72 °C for 1 min). DNA was purified using AMPure XP beads (Beckman Coulter) after adaptor ligation and PCR enrichment, and the libraries were run on a 2200 Tape Station (Agilent Technologies) for quality control. A KAPA Library Quantification Kit (KAPA Biosystems) was utilized to quantify the libraries for pooling, and a final concentration of 1.5 nM was loaded onto an Illumina cBOT for cluster generation before sequencing on an Illumina HiSeq3000 for sequencing using a Single Read 50 bp run. For ChIP-seq data analysis, reads were aligned to the hg38 reference genome using Bowtie2 (v2.3.4.2)39 and duplicates were removed using the rmdup tool from SAMtools (v1.9)40. Reads from each duplicate for each condition were combined, and peaks were called using MACS2 (v2.1.2.1)41. Resulting peaks from each condition with a q-value ≤ 0.01 were merged using the mergeBed tool from BEDTools (v2.27.1)42. Read numbers in peaks were estimated using feature Counts from the subread package (v2.0.0)36. The difference in binding was assessed with R studio (v1.2.13) and DESeq2 (v1.28.1)37 with an FDR cutoff of ≤ 0.01.Re-ChIPFor re-ChIP experiments43, ~5e7 cells were used for the first ChIP and chromatin in RIPA buffer was sheared to a median fragment size of 250 bp using a Bioruptor XL (Diagenode). The first ChIP was performed by incubation of sheared chromatin with corresponding antibody-linked magnetic beads at 4 °C overnight. Samples were washed three times with Re-ChIP wash buffer (50 mM Tris-HCl pH 8.0, 500 mM NaCl, 0.1% SDS, 1% NP-40, 2 mM EDTA) and then chromatin-antibody complexes were eluted in Re-ChIP elution buffer (10 mM Tris-HCl pH 8.0, 2% SDS, 15 mM DTT) at 37 °C for 30 min and subsequently diluted 1/20 in ChIP dilution buffer (16.7 mM Tris-HCl pH 8.0, 167 mM NaCl, 1.2 mM EDTA, 1.1% Triton X-100, 0.01% SDS). The second antibody-linked-magnetic beads were added and incubated overnight on a rotary shaker at 4 °C. The second-round chromatin-antibody complexes were captured, washed, and eluted similarly and then DNA was purified for qPCR as above.Lentivirus productionOne day before transfection, HEK293T cells were seeded at ∼40% confluency in a 10-cm plate. The next day cells were transfected at ∼80–90% confluency. For each transfection, 10 μg of plasmid containing the vector of interest, 10 μg of pMD2.G (Addgene, 12259), and 15 μg of psPAX2 (Addgene, 12260) were transfected using calcium phosphate. Five hours post-transfection the media was changed. Supernatant was harvested 24 and 48 h post-transfection and filtered with a 0.45-μm PVDF filter (Millipore, SLGVM33RS), and then virus was concentrated using Lenti-X™ Concentrator (Takara, 631232), aliquoted and stored at −80 °C. Lentiviral titers were measured by the Lenti-X™ qRT-PCR Titration Kit (Takara, 631232).Lentiviral transduction for stable cell linesA375 cells were transduced with lentiviruses encoding dCas9 or dCas9-dMSK1 in 6-well plates at a MOI of 10. Briefly, 1e6 cells in 2 ml of media supplemented with 8 μg per ml polybrene (Sigma, TR-1003-G) were added to each well. Then, 48 h post-transduction, cells were passaged and 1 μg/ml puromycin was added 3 h after plating. Media was exchanged 2 days post-transduction and cells were passaged every other day starting 4 days after initial replating. Puromycin selection was maintained for a total of 7 days.CRISPR screening assayA375 cells stably transduced with dCas9-dMSK1 were transduced with gRNA libraries at a MOI of 0.2, with a minimal representation of 500 transduced cells per gRNA. Cells were maintained at >500 cells per guide during subsequent passaging. At 7 days post-transduction cells were split into DMSO and PLX-4720 conditions (2 μM PLX-4720 dissolved in DMSO; Selleckchem). Cells were then passaged every 3 days for a total of 16 days of drug or vehicle treatment. Greater than 500 cells per guide were harvested at 23 days post-transduction (16 days post-treatment) for genomic DNA extraction using the Qiagen Quick-gDNA midi kit (Qiagen, 13343). Eight micrograms of genomic DNA was used as a template across eight 100 μl PCR reactions to amplify gRNAs using Q5 hot start polymerase (NEB, M0493L). Amplification was carried out as per manufacturer’s instructions, using 25 cycles at an annealing temperature of 60 °C with the following primers:Fwd: 5′-AATGATACGGCGACCACCGAGATCTACACNNNNNNNNACACTCTTTCCCTACACGACGCTCTTCCGATCTGGACTATCATATGCTTACCGTAACTTG (8-bp index read barcode indicated by italics); Rev: 5′-CAAGCAGAAGACGGCATACGAGATNNNNNNNNGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTGCCAAGTTGATAACGGACTAGCCTT (8-bp index read barcode indicated by italics).Amplified libraries were purified using the QIAquick Gel Extraction Kit (Qiagen, 28706). Each sample was quantified after purification with the Qubit dsDNA High Sensitivity assay kit (Thermo Fisher, Q32854). Samples were pooled and sequenced using the MiSeq platform (Illumina). All reads from gRNA screening will be uploaded to the NCBI single read archive and made publicly available upon publication acceptance.Data analysisSamples were analyzed using the MAGECK-VISPR algorithm44 (quality control, modeling, and visualization of CRISPR screens with MAGeCK-VISPR) using Robust Rank Aggregation (RRA) default parameters. NGS data were de-multiplexed using uniquely indexed reads. Guide counts were determined based on perfectly matched sequencing reads only. For each condition, guide counts were normalized to the total number of counts per condition, and log2 counts were calculated based on these values. Ratios of counts between conditions were calculated as log2 ((count 1 + 1)/(count 2 + 1)) based on normalized counts. Associated box plots and gRNA expression change volcano plots were created using ggplot2 (https://ggplot2.tidyverse.org).Validation of BRAF(V600E) inhibitor (PLX-4720) candidate genesTen of the most enriched candidate gRNAs were cloned into Lenti_sgRNA (MS2) zeo backbone (Addgene, 61427). Lentiviruses were then produced, harvested, and transduced into A375 cells stably transduced with dCas9 control or dCas9-dMSK1. Cells were then put under 200 μg/ml Zeocin selection for 7 days, after which mRNA was harvested and measured using RT-qPCR as described above. All gene expression levels in A375 cells were normalized to A375 cells stably transduced with dCas9. Stably transduced cell lines with an EPDR1 or AFF2 gRNA were treated with 3.5 μM PLX-4720 for 72 h and cell morphology was observed using microscopy and cell viability was determined using MTT assays.MTT viability assayCells were plated at 2e3 per well in 96-well plates and incubated at 37 °C for 12 h before treatment with either DMSO or PLX-4720, in 200 µl. MTT reagent (3-(4,5-dimethythiazol-2-yl)-2,5-diphenyl tetrazolium bromide; Cayman, 21795-1) was then added to a final concentration of 0.5 mg/ml. The reaction was incubated at 37 °C for 3 h and then media and MTT reagent were removed. Fifty microliters of DMSO was then added and cells were incubated at 37 °C for 30 min to 2 h, until cells had lysed and purple crystals had dissolved. Medium only wells were used as blanks. Absorbance was then measured at 570 nanometers and then cell viability was calculated using the formula: (absPLX4720 – absBlank)/(absDMSO – absBlank) × 100.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Kinases",
"Synthetic biology",
"Gene regulation",
"CRISPR-Cas9 genome editing"
] | epigenomic regulatory forces DNA methylation post-translational modifications histones human gene Histone phosphorylation at serine residues 10 28 histone subunit H3 correlated with stimulus-dependent gene defining causal function of endogenous H3S10ph H3S28ph challenging technologies phosphorylation-activated protein kinase 1 H3S10ph H3S28ph in localized to nucleus activated by ERK p38 protein kinase phosphorylation autophosphorylation11 MSK1-driven H3S10ph/H3S28ph correlated with transactivation of stimulus-responsive genes analyses MSK1-mediated histone phosphorylation linked to gene expression from stress-responsive artificial recruitment of MSK1 to endogenous NF1 transcription factor binding sites in hyperphosphorylation of histone H3S10 H3S28 residues increased gene expression from CRISPR/Cas9 system repurposed for genome editing in human nuclease-null deactivated CRISPR/Cas platforms developed manipulate endogenous histone PTMs atno CRISPR/Cas tools locus-specific modification histone phosphorylation construct fusion protein dCas9-dMSK1 deactivated Cas9 protein Streptococcus pyogenes (dCas9) hyperactive variant human MSK1 dCas9-dMSK1 permits locus-specific manipulation H3S10ph H3S28ph human loci activation gene expression demonstrates histone phosphorylation activation promoters new way engineer epigenome expands dCas9-based epigenome editing arsenal CRISPR/Cas9-based histone kinaseHuman MSK1 linked to deposition H3S10ph/H3S28ph transcriptional preliminary results knockout MSK1 causes changes RNA polymerase II transcription cells hypothesized dCas9 recruit MSK1 to human loci clarify locus-specific histone phosphorylation gene expression constructed three fusion proteins between C-terminus dCas9 N-terminus MSK1 variants fused dCas9 to full-length WT MSK1 MSK1 lacking N-terminal inhibitory domain activity inactivated version dMSK1 (dCas9dCas9-MSK1-dMSK1 displayed MSK1 catalytic activity (autophosphorylation MSK1 serine residues 212 transfected cells inactive dCas9-ddMSK1 fusion dCas9-MSK1-dMSK1 H3S10ph H3S28ph incubated with human histone octamers dCas9-dMSK1 0.029) greater enzymatic activity on H3S28 results demonstrate dCas9-MSK1-dMSK1 catalytically active in human cells phosphorylate histones in vitro dCas9-dMSK1 harbors hyperactive histone H3S28 phosphorylation kinase activity-MSK1. CRISPR/Cas9-based histone kinase fused MSK1 MSK1 inactivated dMSK1 dCas9 dCas9-MSK1 fusion variants transfected into HEK293T cells autophosphorylation levels detected serine residues 212 376 MSK1) by Western blot 72 h post-transfection Data three experimentsPurified dCas9-MSK1-ddMSK1 incubated human histone octamers vitro histone H3 serine residues 10 28 measured after incubation 1 h levels H3S10ph H3S28ph quantified densitometry three vitro histone kinase assays Two-sided t-test *P < 0.05 3 experiments panel error bars s.m. ns not significant kDa kilodaltons Source data-dMSK1 activates natural MSK1 dCas9-MSK1-dMSK1 fusion proteins modulate histone phosphorylation gene expression identified natural targets MSK1 comparative RNA-seq Two MSK1 clones generated exons MSK1 Cas9-mediated KO clonal MSK1 cell line similar transcriptome-wide changes WT HEK293T cells comparative RNA-seq resulted 41 24 shared downregulated upregulated genes (Fig.-dMSK1 activates MSK1 target genesDifferentially expressed genes MSK1 HEK293T cell lines WT HEK293T downregulated red circles upregulated analyzed Wald test adjusted P value calculated Benjamini Hochberg method mRNA levels top five downregulated genes measured RT-qPCR 72 h post-transfection dCas9-fusion proteins ChIP-qPCR H3S10ph H3S28ph PRKCB BMP2 promoters 72 h post-transfection dCas9-fusion proteins Two-sided t-test *P < 0.05 3 experiments error bars Source data guide RNAs recruit fusion protein regions top five downregulated genes (PRKCB BMP2 SHROOM2 ZNF462 GDF6) MSK1 HEK293T cells test dCas9-MSK1 dCas9-dMSK1 activate MSK1 targets RT-qPCR dCas9-dMSK1 not dCas9-MSK1 activated BMP2 (P = 0.002) GDF6 (P = 0.008) dCas9 control targeted regiondCas9-MSK1 activate target genes reliance co-activators ATF CREB penetrate human chromatin histones requirement dMSK1 dCas9-dMSK1 could instrumental analysis regulatory mechanisms controlling stimulus-induced transcription CREB/ATF1 phosphorylation recruitment reader proteins MSK1 targeting Not all genes downregulated loss MSK1 responsive to dCas9-dMSK1-mediated activation 2b some downregulated genes indirect targets MSK1 not influenced by dCas9-dMSK1 epigenetic changes gRNA sites measure MSK1 occupancy at PRKCB BMP2 SHROOM2 GDF6 ZNF462 performed ChIP-qPCR at promoter region each gene in WT MSK1 KO HEK293T cells MSK1 enriched at regions BMP2 (P = GDF6 not PRKCB SHROOM2 ZNF462BMP2 GDF6 activated by dCas9-dMSK1 PRKCB SHROOM2 ZNF462 support hypothesis BMP2 GDF6 targets MSK1 PRKCB SHROOM2 ZNF462 affected loss MSK1 dCas9-dMSK gene activation responsive non loci used ChIP-qPCR histone phosphorylation high dCas9-dMSK1 binding PRKCB promoter no changes in H3S10ph H3S28ph dCas9-dMSK1 targeting dCas9-MSK1 dCas9-dMSK1 (P = 0.012 0.008 elevated H3S10ph levels BMP2 promoter dCas9 control dCas9-dMSK1 phosphorylated H3S10 H3S28 at BMP2 promoter results endogenous histones dCas9-dMSK1 efficient dCas9-MSK1 phosphorylating histone H3S28 vitro chromatin indicate targeted locus-specific H3S28ph causal for activation gene expression promoters sensitive histone phosphorylationdCas9-dMSK1 activates human promoters effects-dMSK1 target genes genomic elements delivered dCas9-fusion proteins distal proximal OCT4 RT-qPCR 72 h transfection HEK293T cells dCas9-dMSK1 OCT4 expression potency ~260–103 bp upstream OCT4 transcription start site targeted dCas9 dCas9-MSK1 proteins distal regulatory region core enhancer promoter MYOD similar dCas9-dMSK1 activation MYOD promoter region~273–35 bp upstream TSS show dCas9-dMSK1 non-natural MSK1 target genes human promoters sensitive effects endogenous histone phosphorylation distal enhancers 3dCas9-dMSK1 activates endogenous human promoters high genome transcriptome-wide specificity RT-qPCR OCT4 MYOD mRNA levels 72 h post-transfection dCas9 dCas9-MSK1-dMSK1 promoter-targeting gRNAs Two-sided t-test *P < 0.05 n = 3 experiments error barsFLAG ChIP-seq binding HEK293T cells-transfected dCas9-dMSK1 OCT4-targeting gRNAs HEK293T dCas9-dMSK1 non-targeting gRNA Wald test Red circles false discovery rate (FDR < 0.01 RNA-seq data HEK293T cells-transfected dCas9-dMSK1 OCT4 gRNAs HEK293T Wald test adjusted P value calculated Benjamini Hochberg method upregulated mRNAs isoforms red circles (Padj < 0.05) downregulated gene (SEPT7P3) blue circle (Padj < 0.05) Source data dCas9-dMSK1 performed ChIP-seq RNA-seq HEK293T cells-transfected dCas9-dMSK1 OCT4-targeting gRNAs-dMSK1 non-targeting gRNA-seq results dCas9-dMSK1 binding OCT4 promoter specific (FDR = 0.0006) human genome. one (FDR = 0.0086) off-target CROCC2RNA-seq dCas9-dMSK1 targeted OCT4 promoter activated two isoforms OCT4 (ENST00000638788.1 ENST00000461401.1 Padj 0.046 0.008 human transcriptome (Fig. 3d). One off-target transcript detected-dMSK1 OCT4 HEK293T cells SEPT7P3 (Padj = 0.011) downregulated results dCas9-dMSK1 binding activation specific human genome transcriptome dCas9-MSK1 dCas9-dMSK1 <0.05) increased H3S10ph levels OCT4 MYOD promoters (Fig. 4a dCas9-dMSK1 increased H3S10ph H3S28ph levels targeted promoters coincident activation gene expression (Fig. 3a H3S28ph ChIP-seq analysis confirmed enriched OCT4 promoter-dMSK1 targeting enrichment not statistically significant genome-wide studies MSK1 phosphorylates H3S10 or H3S28 not same histone dCas9-dMSK1 deposit H3S10ph H3S28ph same histone tail re-ChIP analysis (Figre-ChIP analyses dCas9-dMSK1-mediated phosphorylation H3S10 independently H3S28 OCT4 MYOD promoters. lack enrichment H3S28ph after enrichment H3S10ph ChIP results show H3S28ph activation gene expression MSK1 target genes highlights specificity utility dCas9-dMSK1 4dCas9-dMSK1 induces gene activation H3S28ph ChIP-qPCR enrichment H3S10ph H3S28ph OCT4 MYOD promoters 72 h post-transfection dCas9-fusion protein Re-ChIP-qPCR enrichment H3S10ph H3S28ph OCT4 MYOD promoters 72 h post-transfection dCas9 Two-sided t-test *P < 0.05 n = 3 experiments error bars not significant Source data.dCas9-dMSK1 driven H3S28ph influences H3K27ac statusH3S28ph histone PTMs acetylation status histone H3 lysine 27 (H3K27ac)4,8,9,13,26.measured H3K27ac levels OCT4 MYOD promoters dCas9-MSK1 fusion protein variants no activity targeting dCas9-dMSK1 OCT4 MYOD promoters increased H3K27ac levels (Fig. 5a H3K27ac enriched dCas9-dMSK1 targeted BMP2 promoter not PRKCB promoter results crosstalk between H3S28ph H3K27ac human loci interplay activation gene expression consistent H3S28ph promotes CBP/p300-dependent transcription cell-free 5dCas9-dMSK1 influences H3K27 acetylation status targeted human promoters ChIP-qPCR enrichment H3K27ac OCT4 MYOD promoters 72 h post-transfection dCas9-fusion protein RT-qPCR OCT4 MYOD DMSO CBP/p300 inhibitor A485 72 h post-transfection 20 μM A485 DMSO added cells 12 h post-transfection Two-sided t-test *P < 0.05 n = 3 experiments panels error bars not significant Source data filerelationship histone phosphorylation PTMs measured inhibition CBP/p300 acetyltransferases dCas9-dMSK1 gene activation inhibition CBP/p300 A485 reduced efficacy dCas9-dMSK1-mediated transactivation OCT4 = MYOD (P 0.0008) promoters (Fig. 5c CBP/p300 linked transactivation dCas9-dMSK1 human promoters MSK1 KMT2A/MLL1 methyltransferase complex methylation H3K4 measured enrichment H3K4me3 targeting OCT4 MYOD promoters dCas9-MSK1 fusion proteins significant < 0.05) increases H3K4me3 enrichment dCas9-MSK1 dCas9-dMSK1 dCas9 kinases dCas9-MSK1-dMSK1 not function WT MSK1 human cells targeted dCas9-fusion proteins HBA1 SOX2 KLF4 promoter regions in HEK293T cells measured gene expression 72 h post transfection dCas9-dMSK1 <0.05) activated gene expression targeted promoter lesser than OCT4 MYODreduced potency due to differences basal gene expression levels observed with dCas9-based additional gRNA dCas9-dMSK1 promoter gene activation local levels histone phosphorylation proportional to gene expression at promoters dCas9-dMSK1 gene expression in A549 cells another human histone kinase (Aurora B) activate gene expression fused to dCas9 targeted promoter results show dCas9-dMSK1 phosphorylation status of endogenous human histones gene expression promoters functional at diverse promoters different cell types achieved using other histone kinases causal role importance histone phosphorylation at endogenous loci.Genome-scale screening for mediators PLX-4720 resistancedCas9-based tools screen noncoding genome in high efficacy dCas9-dMSK1 at human promoters pathological gene expression combined dCas9-dMSK1 with genome-scale gRNA library identify genes overexpressed by dCas9-dMSK1 resistance to BRAF V600E inhibitor PLX-4720transduced A375 melanoma cells dCas9-dMSK lentiviral expression vector CRISPR gRNA multiplicity infection) 0.2 Cells subjected PLX-4720 treatment 16 days next-generation sequencing gRNA sequences PLX-4720 treated population (Fig. screening dCas9-dMSK1 BRAF V600E inhibitor resistance Flow-chart screening dCas9-dMSK1 Box plots distribution gRNA frequencies after lentiviral transduction DMSO PLX-4720 treated A375 cells green horizontal line median upper lower bounds 25th 75th percentiles whiskers extend maximum minimum values within 1.5-fold interquartile range Outliers green dots n = 2 experiments Volcano plot gRNA counts PLX-4720 versus DMSO treated A375 cells analyzed rank aggregation) cutoff P value <0.01 Blue circles depleted red enriched gray circles no change levels top 10 enriched genes A375 cells dCas9 dCas9-dMSK1 when targeted Two-sided t-test *P < 0.05 n = 3 experiments error barsSource data file.dCas9-dMSK1 active in A375 cells Fig increased enrichment in PLX-4720 samples DMSO control cells (Fig gRNA 5757 <0.01 depleted gRNAs 4889 targeted genes 315 <0.01 enriched gRNAs 314 genes in PLX-4720 cells (Fig. 6c Table 4) Eighteen of 314 enriched genes identified dCas9-SAM differences in gRNAs between dCas9-SAM-dMSK1 lower potency-dMSK1 unique effects histone phosphorylation targeted promoters.Three top 10 genes mediators PLX-4720 resistance EPDR1 AFF2 ERC2 linked to BRAF protein in colorectal autism spectrum ganglioma33 MRPS15 TRAT1 LACC1 AGL TDRP MIPOL1 LELP1 mediators PLX-4720 resistance not previously reported targeted dCas9-dMSK1 to promoters EPDR1 AFF2 MRPS15 TRAT1 LACC1 AGL TDRP MIPOL1 LELP1 using gRNAs enriched from screen (FigdCas9-dMSK1 <0.05) increased expression targeted gene-dMSK1-mediated used-dMSK1 overexpression EPDR1 or AFF2 in A375 cells treated with DMSO or 3.5 μM PLX-4720 for 72 h Microscopy analysis MTT assays dCas9-dMSK1-mediated upregulation EPDR1 or AFF2 improved cell fitness higher viability control-treated A375 cells Fig. results demonstrate dCas9-dMSK1-mediated histone phosphorylation gene activation compatible with high-throughput gRNA screening programmable histone phosphorylation human gene regulation pathology built CRISPR/Cas9-based epigenome editing tool dCas9-dMSK1 permits writing histone phosphorylation at endogenous loci human chromatin histone phosphorylation activation promoters H3S28ph pivotal dCas9-dMSK1 functional crosstalk between histone phosphorylation acetylation endogenous histones linked to activation gene expression used dCas9-dMSK1 high-throughput genome-scale screening uncover genes therapeutic resistancework expands CRISPR/Cas9 epigenome editing toolkit clarifies histone phosphorylation genes epigenomic modification eukaryotes culture plasmid constructionHEK293T A549 A375 cultured Dulbecco’s Eagle’s medium 10% FBS 1% penicillin/streptomycin maintained 37 °C 5% CO2. cloning backbone pLV-dCas9-p300-P2A-Puro described N-terminal MSK1 (1–1644 amplified from pDONR223-RPS6KA5 C-terminal MSK1 (1645–2406 synthesized gBlock gene fragment pLV-dCas9-MSK1-P2A-Puro created subcloning MSK1 BamHI-digested pLV-p300 backbone NEBuilder HiFi DNA Assembly pLV-dCas9-dMSK1-P2A-Puro generated amplifying amino acids 42–801 MSK1 subcloning Δ41MSK1 BamHI-digested pLV-dCas9-p300-P2A-Puro backbonepLV-dCas9-ddMSK1-P2A-Puro created fragments acid mutations (D195A PCR-digested pLV-dCas9-p300-P2A-Puro backbone NEBuilder HiFi DNA Assembly Protein sequences dCas9 constructs Supplementary Note 1. gRNAs non experiments cloned pSPgRNA backbone gRNA protospacer targets Supplementary Table 5. gRNA library genome-scale screening purchased Addgene transfections 24-well plates 375 ng dCas9 expression vector 125 ng single gRNA Plasmids mixed Lipofectamine 3000 ChIP-qPCR experiments HEK293T cells transfected 15 cm dishes Lipofectamine 3000 37.5 μg dCas9 vector 12.5 μg equimolar pooled gRNA vectors blottingTwenty micrograms protein loaded SDS-PAGE transferred PVDF membrane Western blotsantibodies-FLAG-Aldrich-MSK1 1:1000 dilution Tris Buffered Saline 1% Casein α-mouse HRP-Aldrich α-rabbit HRP 1:3000 dilution Tris Saline 1% Casein Membranes exposed ECL 170-5060) Tubulin detected hFABTM Rhodamine Anti-Tubulin Primary Antibody 1:3000 octamer kinase dCas9-fusion proteins purified transfected HEK293T cells ANTI-FLAG® M2 Affinity Gel Kinase assays 30 °C 1 h 30 μl reactions 500 ng histone octamer 150 nM dCas9 protein 10 mM Tris-HCl pH 8.0 100 mM KCl 10 mM MgCl2 0.1 M ATP 1X inhibitor Scientific reaction suspended SDS-PAGE buffer boiled 95 °C 10 min transferred PVDF membrane Western blottingblocking 1% casein 1 h histones detected immunoblotting phosphor-specific antibody 1:1000 dilution-H3S10ph α-H3S28ph Non-phosphorylated histones detected α-H3 1:1000 Secondary α-rabbit HRP 1:3000 membranes exposed imaged ECL Densitometry analysis ImageJ.MSK1 knockout cell 2 gRNAs exon 1 2 MSK1 cloned LentiCRISPR v2 vector transfected HEK293T cells Forty-eight hours-transfection cells passaged 1 μg/ml puromycin added 3 h plating 4 days cells harvested diluted 1000-fold replated 15-cm plates medium 1 μg/ml puromycin Two weeks colonies picked cultured 96-well plates 1 week passaged 24-well plates MSK1 knockout lines confirmed blot analysis α-MSK1 1:1000 dilution secondary α-rabbit HRP LentiCRISPR v2 empty vector control HEK293T cells performed each conditionRNA isolated from transfected cells RNeasy Plus kit RNA-seq libraries constructed TruSeq Stranded Total RNA Gold qualities verified Tape Station D1000 assay quantities checked real-time PCR 6 PCR Libraries normalized pooled 75 bp paired-end reads sequenced Hiseq3000 platform Reads aligned to Hg38 transcriptome HISAT2(2.1.0 Transcript abundance calculated Counts differential expression determined R studio (v1.2.13) DESeq2 (v1.28.1 analysis Gene ontology analysis DAVID Functional Annotation Bioinformatics Microarray RNA-seq reads uploaded NCBI single read archive available publication-transcription quantitative PCR)RNA isolated from transfected cells RNeasy Plus kit 1 μg purified RNA template for cDNA synthesis Real-time PCR SYBR Green CFX96 Real-Time PCR Detection System C1000 Thermal Cycler Baselines subtracted baseline subtraction curve thresholds calculated Bio-Rad CFX Manager software 2.1Results change cells transfected empty vector plasmid normalization GAPDH method qPCR primers conditions Supplementary Table 6.ChIP-qPCRHEK293T cells co-transfected dCas9-fusion vectors gRNA constructs 15 cm plates triplicates condition Cells cross-linked 10 min 1% formaldehyde stopped glycine 125 mM Cells harvested washed ice 1X PBS suspended Farnham lysis buffer (5 mM PIPES pH 8.0 85 mM KCl 0.5% NP-40 1X protease phosphatase inhibitor pelleted resuspended RIPA buffer (1X PBS 1% NP-40 0.5% sodium deoxycholate 0.1% SDS 1X protease phosphatase inhibitor 2.5e7 cells H3S10ph H3S28ph ChIP experiment Chromatin sheared fragment size 250 bp Bioruptor XL Two micrograms antibody-FLAG Mouse IgG incubated 50 μl Sheep anti-Rabbit Mouse IgG magnetic beads Technologies 16 h at 4 °CAntibody-linked magnetic beads washed three times PBS/BSA buffer (1X PBS 5 mg/ml BSA sheared chromatin incubated beads 4 °C overnight washed five times LiCl IP wash buffer (100 mM pH 7.5 500 mM LiCl 1% NP-40 1% sodium deoxycholate). Cross-links reversed overnight incubation 65 °C DNA purified QIAquick PCR purification kit for ChIP-qPCR Input DNA from ~1.0e6 cells Ten nanograms DNA qPCR reactions CFX96 Real-Time PCR Detection System C1000 Thermal Cycler Baselines subtracted baseline subtraction curve thresholds calculated Bio-Rad CFX Manager software 2.1. ChIP-qPCR data plotted relative percent input primers conditions Supplementary Table libraries prepared KAPA Hyper Prep Kit 3 ng fragmented DNA ligation NEXTflex DNA adapter-ligated DNA enriched 10 cycles PCR 98 °C 72 purified AMPure XP beads after libraries run on 2200 Tape Station quality controlKAPA Library Quantification Kit libraries pooling concentration 1.5 nM loaded Illumina cBOT cluster before sequencing Illumina HiSeq3000 Single Read 50 bp run ChIP-seq analysis reads aligned hg38 genome Bowtie2.4.2 duplicates removed rmdup SAMtools Reads combined peaks called MACS2 (v2.1.2.1 peaks q-value ≤ 0.01 merged mergeBed tool Read numbers estimated Counts (v2.0 difference binding assessed R studio (v1.2.13) DESeq2 (v1.28.1 FDR cutoff ≤ 0.01.Re ~5e7 cells first ChIP chromatin RIPA buffer sheared fragment size 250 bp Bioruptor XL ChIP incubation sheared chromatin antibody-linked magnetic beads 4 °C overnightSamples washed Re-ChIP buffer (50 Tris-HCl 500 NaCl 0.1% SDS 1% NP-40 2 mM EDTA chromatin-antibody complexes eluted buffer (10 mM-HCl 2% SDS 15 mM DTT 37 °C 30 min diluted 1/20 buffer (16.7 mM Tris-HCl 167 mM NaCl 1.2 mM EDTA 1.1% Triton X-100 0.01% antibody-magnetic beads incubated overnight shaker 4 °C second chromatin-antibody complexes captured washed eluted DNA purified qPCR transfection HEK293T cells seeded confluency 10-cm transfected confluency 10 μg plasmid pMD2.G 15 μg psPAX2 media changed Supernatant harvested 24 48 h filtered 0.45-μm PVDF filter concentrated Lenti-XTM Concentrator stored −80 °C titers measured Lenti-XTM qRT-PCR Titration KitLentiviral transduction cells transduced 6-well plates MOI 10. 1e6 cells 2 ml media 8 μg per ml polybrene well 48 h post-transduction cells passaged 1 μg/ml puromycin added 3 h plating Media exchanged 2 days passaged day 4 days replating Puromycin selection maintained 7 days screening assayA375 cells transduced dCas9-dMSK1 gRNA MOI 0.2 500 cells per gRNA maintained >500 cells per guide 7 days post-transduction split DMSO PLX-4720 conditions (2 μM PLX-4720 DMSO passaged 3 days 16 days treatment 500 cells harvested 23 days post-transduction genomic DNA extraction Qiagen Quick-gDNA kitEight micrograms genomic DNA template eight 100 μl PCR reactions amplify gRNAs Q5 hot start polymerase M0493L). Amplification manufacturer’s instructions 25 cycles annealing temperature 60 °C primers:Fwd 5′ (8-bp index barcode Rev: 5′ italics).Amplified libraries purified QIAquick Gel Extraction Kit sample quantified after purification Qubit dsDNA High Sensitivity assay kit Samples pooled sequenced MiSeq platform reads gRNA screening uploaded NCBI single read publicly available publication acceptance.Data analysisSamples analyzed MAGECK-VISPR algorithm44 Robust Rank Aggregation (RRA) default parameters NGS data de-multiplexed uniquely indexed reads Guide counts determined matched sequencing reads guide counts normalized total log2 counts calculated Ratios counts between conditions calculated log2 ((count 1 + 1)/(count 2 + 1)) counts Associated box plots gRNA expression change volcano plots created ggplot2BRAF(V600E inhibitor (PLX-4720) genesTen enriched gRNAs cloned Lenti_sgRNA backbone Lentiviruses transduced A375 cells dCas9 Cells under 200 μg/ml Zeocin selection 7 days mRNA harvested measured RT-qPCR gene expression levels normalized transduced cell lines EPDR1 treated 3.5 μM PLX-4720 72 h cell observed microscopy viability determined MTT assays plated 2e3 96-well plates incubated 37 °C 12 h treatment DMSO PLX-4720 200 μl MTT reagent (3-(4,5-dimethythiazol-2-yl-2,5-diphenyl tetrazolium bromide added 0.5 mg/ml incubated 37 °C 3 h MTT reagent removed Fifty microliters DMSO added cells incubated 37 °C 30 min to 2 h purple crystals dissolved Medium wells Absorbance measured 570 nanometers cell viability calculated (absPLX4720 – absBlank)/(absDMSO – absBlank) × 100 Nature Research Reporting Summary | 50.3 | 0.855647 |
10.1038/s41467-020-19153-6 | PMC7584637 | How plant biosynthetic gene clusters (BGCs) form and evolve remains unclear. Here, via examining available genomes within and between Arabidopsis species, the authors show that the thalianol BGC has evolved recently and is still dynamically evolving through involvement of auxiliary genes and chromosomal inversions. | Numerous examples of biosynthetic gene clusters (BGCs), including for compounds of agricultural and medicinal importance, have now been discovered in plant genomes. However, little is known about how these complex traits are assembled and diversified. Here, we examine a large number of variants within and between species for a paradigm BGC (the thalianol cluster), which has evolved recently in a common ancestor of the Arabidopsis genus. Comparisons at the species level reveal differences in BGC organization and involvement of auxiliary genes, resulting in production of species-specific triterpenes. Within species, the thalianol cluster is primarily fixed, showing a low frequency of deleterious haplotypes. We further identify chromosomal inversion as a molecular mechanism that may shuffle more distant genes into the cluster, so enabling cluster compaction. Antagonistic natural selection pressures are likely involved in shaping the occurrence and maintenance of this BGC. Our work sheds light on the birth, life and death of complex genetic and metabolic traits in plants. | IntroductionNonrandom gene organization in eukaryotes plays a significant role in genome evolution and function. Although most eukaryotic genomes lack operons, they do contain groups of physically clustered genes that are related in function despite being unrelated in sequence. Striking examples of this are biosynthetic gene clusters (BGCs)1,2. BGCs have been studied extensively in fungi1,2. Over the last few years numerous BGCs have also been reported from plants, including for compounds that determine agricultural traits such as disease resistance and drought tolerance3,4 or have important pharmaceutical applications5,6. There is compelling evidence that these plant clusters have not arisen by horizontal gene transfer from microbes, but instead are likely to be generated through gene duplication, neofunctionalization, and genomic relocation1,2. However, very little is known about how these gene clusters form. Several examples of syntenic BGCs that make variants of the same types of molecules have been reported when different species are compared, including for steroidal glycoalkaloids in the Solanaceae and triterpenes associated with bitterness in the Cucurbitaceae7,8. However, large-scale analysis of the evolutionary dynamics of individual BGCs within and across species within appropriate taxonomic windows has been hindered by the lack of available sequence resources.Understanding the clustering of functionally related nonhomologous genes is crucial to our understanding of the relationship between genome organization and the evolution of complex adaptive traits. Over 10 years ago we characterized an “operon-like” gene cluster from the model plant species Arabidopsis thaliana (the thalianol BGC)9. More recently we have shown that this cluster plays an important role in shaping the root microbiota10.Here, we take advantage of the rich resource of recently generated sequence data for A. thaliana and its relatives to investigate the occurrence, nature and evolution of this paradigm BGC at both the vertical (between species) and horizontal (within species) levels. Our results reveal that the thalianol BGC has evolved specifically in the Arabidopsis genus, and is still dynamically evolving between sister species through involvement of distant auxiliary genes. Within the A. thaliana species, the thalianol BGC is primarily fixed, as demonstrated by its conserved biochemical function and evidence of pervasive purifying selection imposed on individual BGC genes. In contrast, signatures for localized positive selection are identified when considering the cluster region as a whole. Analysis of de novo genome sequence assemblies of A. thaliana accessions from around the world further reveals compaction of the thalianol BGC through chromosomal inversions. The divergent and polymorphic cluster variants that we identify provide new insights into the birth, life and death of BGCs in plants.ResultsThe biochemical function and evolutionary origin of the thalianol clusterIn our original discovery of the thalianol BGC in A. thaliana accession Col-0 we reported four physically adjacent coexpressed genes (THAS, THAH, THAO, and THAA1) for the biosynthesis of thalianol-derived triterpenes9. Subsequent analysis revealed three additional genes (a linked gene THAA2, and two unlinked genes THAR1 and THAR2) that are required for synthesis of the final pathway end product, thalianin (T10)10 (Fig. 1a). The original four core cluster genes are conserved between A. thaliana Col-0 and A. lyrata, both gene sets producing the acetylated thalianol derivative (-)-3β,7β-dihydroxy-16-keto-thalian-15-yl acetate (T7) when transiently expressed in Nicotiana benthamiana10,11 (Fig. 1a). While the order of the four core genes is the same in these sister species, the THAA2 ortholog AL8G20050 in A. lyrata has a different orientation (Fig. 1b). In addition, there are intervening genes between THAA1 and THAA2 in the two species, which are not implicated in plant metabolism and are not syntenic (Fig. 1b).Fig. 1The emergence of the thalianol cluster in the Arabidopsis genus.a The characterized thalianin pathway in A. thaliana (modified from ref. 14). Genes are indicated by arrows: OSC, oxidosqualene cyclase; CYP, cytochrome P450; ACT_IIIa, acyltransferase subfamily IIIa. The compounds are numbered as in ref. 14. b Comparison of the thalianol cluster from A. thaliana and A. lyrata. Orthologous genes identified by reciprocal blast analysis between the two species are connected with gray lines. c Species tree indicating presence (solid line) or absence (dashed line) of the thalianol cluster. The times of divergence are from Hohmann et al.12.Thalianin belongs to the large and structurally diverse family of plant natural products known as triterpenes. In a previous study, we investigated the occurrence of triterpene BGCs in the sequenced genomes of 13 different Brassicaceae species11. We identified a superficially similar triterpene BGC in the genome of the related species Capsella rubella, which diverged from A. thaliana around 8–9 million years ago12 (Fig. 1c). This cluster transpired to be the result of an independent evolutionary event11. Other than the C. rubella example, thalianol-like clusters were not detected in any of the other species included in this analysis11. To further investigate the origins of the thalianol cluster, we next interrogated the newly generated genome sequence assemblies of two other Arabidopsis species, Arabidopsis halleri13 and Arabidopsis arenosa (this study; see Methods section). The first committed step in triterpene biosynthesis is the cyclization of 2,3-oxidosqualene to different triterpene scaffolds, a reaction carried out by enzymes known as oxidosqualene cyclases (OSCs). Systematic mining of the four Arabidopsis genomes identified a total of 55 predicted OSC genes (Supplementary Fig. 1). Phylogenetic analysis, in combination with investigation of the genomic locations of these OSC genes, revealed that the occurrence of bona fide syntenic thalianol synthase (THAS) genes was restricted to a monophyletic branch including only A. thaliana and A. lyrata OSC genes (marked in red in Supplementary Fig. 1). Consistent with this, analyses of the four Arabidopsis genomes using plantiSMASH analysis, a BGC mining algorithm designed for plants14, detected thalianol BGCs only in A. thaliana and A. lyrata (Supplementary Fig. 1). Thus, the thalianol cluster likely evolved in a common ancestor of the Arabidopsis genus around 5.8 MYA12 (Fig. 1c).Comparison of the thalianol cluster between speciesAlthough the full thalianin pathway has been extensively studied in A. thaliana accession Col-09,10, the full pathway in A. lyrata has not been characterized11. We therefore cloned the putative THAA2 ortholog from A. lyrata (AL8G20050; 90.4% predicted protein identity to AtTHAA2). Coexpression of this gene with the four other A. lyrata thalianol cluster genes in N. benthamiana yielded epi-thalianin (T17) (Fig. 2a), indicating that these five clustered genes are functionally conserved between A. thaliana and A. lyrata. In line with this, all five genes are coexpressed in the roots in both A. thaliana and A. lyrata (Fig. 2b). In A. thaliana, two auxiliary oxidoreductase genes THAR1 (AT3G29250) and THAR2 (AT1G66800) are required for epimerization of the C3 hydroxy moiety of T7 to give the final pathway end product, thalianin (T10)10 (Fig. 1a). Surprisingly, in A. lyrata, targeted metabolite analysis of root extracts did not identify T10 but T17 (Fig. 2a). The two isomers were not detectable in C. rubella and A. arenosa, both of which lack the thalianol BGC (Figs. 1c and 2a). Together our results indicate that the thalianol BGC has evolved specifically in the Arabidopsis genus, but is still dynamically evolving when A. thaliana and A. lyrata are compared. The presence of T17 in A. lyrata root extracts further implies that the thalianol BGC in A. lyrata per se encodes a full biosynthetic pathway without the involvement of two auxiliary genes as in A. thaliana. In agreement with this, reciprocal BLASTP and synteny analysis were unable to identify putative orthologs of THAR1 and THAR2 in A. lyrata. Transcriptome analysis revealed that the most closely related gene to A. thaliana THAR2 in A. lyrata (AL1G20010; 70.5% predicted protein identity) was preferentially expressed in the flowers rather than the roots, although the gene most closely related to A. thaliana THAR1 (AL4G46670; 81.5% predicted protein identity) had a similar expression pattern to the thalianol cluster genes (Fig. 2b). In addition to being part of the thalianin pathway in A. thaliana, the THAR2 gene is also required for the biosynthesis of arabidin, a triterpene derived from another A. thaliana Col-0 BGC10,15 (Supplementary Fig. 2a). However, the arabidin gene cluster is not present in A. lyrata16 (Supplementary Fig. 2b). Altogether, these results suggest that although the thalianol cluster genes share conserved expression patterns and enzymatic functions in A. thaliana and A. lyrata, in A. lyrata the five clustered genes encode a complete biosynthetic pathway for epi-thalianin (T17), while in A. thaliana the two additional unlinked genes THAR1 and THAR2 have been recruited to convert the T7 pathway intermediate into thalianin (T10) (Fig. 2c).Fig. 2Involvement of auxiliary genes drives diversification of thalianol BGC between species.a LC-MS analysis of extracts from Nicotiana benthamiana leaves co-expressing thalianol pathway genes (top three traces) and of root extracts from A. thaliana accession Col-0 and A. lyrata accession VLH6. TIC, total ion chromatogram; EIC, extracted ion chromatogram. The mass fragment 579.3622 was used to detect T10 and T17. The MS spectra underlying the EIC peaks are as shown in Supplementary Fig. 3b. Each experiment was independently repeated three times. b Comparison of the expression patterns of the thalianol cluster-related genes in different organs in A. thaliana and A. lyrata. Transcripts per kilobase million (TPM) values were used to generate the heatmap. The colors indicate the level of gene expression from low (red) to high (green). The Z-score represents the deviation from the mean by standard deviation units. c Divergence of the thalianol pathways in A. thaliana and A. lyrata. The black arrows show the directions of the biosynthetic pathways, starting from THAS. The numbering of the products is as in Fig. 1a. The structural differences between the final pathway end products T10 and T17 are highlighted. Source data underlying b are provided as a Source Data file.Variation in the thalianol cluster within speciesWe next systematically examined polymorphisms, including SNPs, small indels and large gene deletion variations, in the thalianol cluster genes based on genome sequences generated with short reads from 1135 A. thaliana accessions17 (https://tools.1001genomes.org/polymorph/). Overall, SNPs in the thalianol cluster genes were the most common types of polymorphisms (97%), followed by small indels (55% and 36% for insertions and deletions, respectively) (Supplementary Table 1). These polymorphisms were mostly benign, with only two accessions having SNPs that are predicted to have a large effect (i.e., premature stop codons) on the THAS gene and most accessions having small indels in the UTR regions of the cluster genes (Supplementary Table 1). Only ~2% (21/1135) of accessions were found to have large gene deletions involving one or more cluster genes. Of those, 16 accessions lacked the THAS, THAO, and THAA1 genes and a large 3′-UTR region of the THAA2 gene (~6 kb) (Fig. 3a). Other types of deletions were also found at much lower frequencies (Fig. 3a). The two unlinked genes (THAR1 and THAR2) are present in nearly all of the A. thaliana accessions (Supplementary Fig. 3a). In accordance with this, analysis of transcriptome data for seven A. thaliana accessions revealed similar expression patterns for the thalianol cluster genes as in Col-09 (Supplementary Fig. 3b). Metabolite analysis confirmed that T10 could be detected in root extracts from accessions that had the thalianol cluster genes but not in those that lacked them (e.g., Cerv-1, Dr-0, and Lillö-1) (Supplementary Fig. 4). Taken together, we conclude that this cluster is likely to be functional in the vast majority of A. thaliana accessions.Fig. 3Variation in the thalianol cluster within A. thaliana species.a
A. thaliana accessions identified from the Arabidopsis 1001 genome sequence database as lacking homologs to thalianol cluster genes (aligned with the Col-0 reference genome). Deletions are indicated by gray bars. The three different types of deletion patterns (lacking one or more cluster genes) are indicated by colored triangles. A fully assembled region from long-read re-sequencing of the accession Lillö-1 (asterisked) showing an inversion in this region is illustrated at the bottom. b Chromosomal inversions in the thalianol region from genomes of representative de novo assembled A. thaliana accessions. c Frequencies of Col-0-like and contiguous thalianol clusters in the genomes of 21 de novo assembled A. thaliana accessions. Accessions Sha and Eri are newly PacBio-sequenced accessions that were not included in the 1135 short read sequenced collections17. d Phylogenetic tree for 1135 A. thaliana accessions. Phylogenetic distances were calculated based on 28,775 filtered SNPs using RAxML40. Tree node support (>90%) is indicated by pink dots on branches (1000 bootstrap replicates). Accessions with thalianol pathway gene deletions and chromosomal inversion variations are indicated. Other monophyletic groups are collapsed.To gain insights into polymorphisms in gene order for the thalianol BGC, we analyzed 22 A. thaliana genome sequence assemblies generated with PacBio long reads (see Methods section). These analyses confirmed the deletion profile observed for the Lillö-1 accession (Fig. 3a). They further showed that, while the order of the four core genes is well conserved, that of the intervening genes between THAA1 and THAA2 is not. Of the 21 assembled A. thaliana genomes (excluded Lillö-1, since it lacked most of the core genes), only three had the same gene order as the Col-0 thalianol cluster. Strikingly, in the remaining 17 accessions the THAS, THAH, THAO, THAA1, and THAA2 genes were contiguous, with no predicted genes between THAA1 and THAA2. The reversed orientation of the three rearranged genes and detailed alignment analysis suggests that these natural accessions have experienced chromosomal inversions to form compact thalianol clusters relatively recently (Fig. 3b, c and Supplementary Fig. 5). Although our sample size for de novo sequenced genomes is relatively small (21 out of 1135 accessions), the contiguous forms of the thalianol cluster constitute the majority (17 out of 22; 77%) of these accessions and are phylogenetically widely distributed (Fig. 3d). We hypothesize that chromosomal inversion may facilitate the compaction of the thalianol BGC in natural populations.Analysis of selection on the thalianol pathwayTo investigate what types of selection may shape the within and between species variations of the thalianol BGC, we simulated variant data at within-species and/or between-species levels. Pairwise between-species nonsynonymous/synonymous substitution (dN/dS) analysis identified relaxed purifying selection for the thalianol cluster genes when compared with phytosterol pathway genes (which are highly conserved across land plants), reflecting the Arabidopsis-genus specific distribution of this gene cluster (Fig. 4a). Notably, the later thalianol pathway genes were far less constrained than the early pathway genes (Supplementary Fig. 6). This is consistent with different roles for THAA2 in A. thaliana and A. lyrata as shown in Fig. 2c. Branch-site model analysis (see Methods section) did not detect gene-wide positive selection for any of the thalianol cluster genes (Supplementary Table 2), although a site model analysis (see Methods section) did detect episodic positive selection on several sites across the clustered genes (Supplementary Table 3). These analyses are in agreement with the conserved enzymatic functions of the thalianol pathway genes between species. We next performed several versions of the McDonald–Kreitman (MK) test18 by using both polymorphism and divergence to test for positive selection in more recent evolutionary time (see Methods section). These analyses found no signature of current positive selection imposed on the thalianol cluster genes (Supplementary Table 4). Altogether, our results suggest that individual genes for thalianol BGC are mostly under purifying selection, presumably due to their conserved enzymatic functions and expression patterns within and between species.Fig. 4Molecular evolution analysis within and between species.a Pairwise dN/dS analysis between A. thaliana and A. lyrata for the thalianol pathway genes. The phytosterol pathway genes were used as controls for strong purifying selection. The arrows illustrate the directions of the biosynthetic pathways, starting with cycloartenol synthase (CAS1) for sterol biosynthesis and THAS for thalianol-derived triterpenes. Two-sided Student t-test, df = 8. b Comparison of LD (r2 > 0.1) numbers simulated from the A. thaliana 1001 genomes SNP data across genome neighborhoods for clustered and non-clustered OSC genes. Genome neighborhoods were defined as 50 Kb either side of each OSC gene. Although the At1g78500 gene product PEN6 is phylogenetically close to the four OSCs that each form BGCs9 (Supplementary Fig. 1), analysis of the expression patterns of the genes neighboring At1g78500 indicate that it is unlikely to form part of a functional BGC (Supplementary Fig. 7). Therefore, the PEN6-centric genomic neighborhood (GN) was considered as non-clustered. Two-sided Student t test, df = 7. c Plot of LD values (r2 > 0.1) in the THAS-centric GN along with the genomic distances of biallelic SNPs. The two LDs located in the region encompassing the four cluster genes shown in b are marked in red. The black line is a smoothed trend fitted by local regression (LOESS).Having carried out evolutionary tests on single genes, we then expanded our tests for positive selection to consider the thalianol cluster as a whole by performing genome-wide linkage disequilibrium (LD) analysis using SNP data from the 1135 A. thaliana genomes. The genome of A. thaliana accession Col-0 contains 13 OSC genes, each with different functions9. These genes are located in a total of nine genomic neighborhoods (GNs), tandem duplicated OSC genes being treated as one GN (see Methods section). We systematically examined the LD extensions of these nine OSC GNs. To do this, we counted the number of squared allele frequency correlation (r2) values (r2 > 0.1) for biallelic sites located within the OSC-centric GNs (spanning 50 Kb on each side of an OSC locus; see Fig. 4b). Our results demonstrate that GNs bearing OSC gene clusters have significantly more nonrandom associations of biallelic sites than GNs containing OSC genes that are not part of BGCs (Student t-test; Fig. 4b). In addition, nonrandom associations of biallelic sites in the THAS-centric GN are mostly located in the flanking regions of the thalianol cluster and decay with increasing genetic distance (Fig. 4c), only two being detected in the region encompassing the four cluster genes shown in Fig. 4c (indicated by the red circles). This pattern suggests that the thalianol cluster is currently undergoing or has recently undergone a selective sweep, which has reduced the diversity in the cluster region. The elevated LDs in the flanking regions are presumably due to the hitch-hiking effect19. This interpretation is further supported by the low levels of single feature polymorphism (SFPs) across the thalianol core cluster in comparison to the flanking regions (Supplementary Fig. 8), suggesting that the thalianol cluster as a whole is under directional positive selection.DiscussionHere, we have carried out large-scale investigation of the evolutionary dynamics of a plant BGC, the thalianol cluster. Our study suggests that an ancestral thalianol cluster evolved relatively recently in evolutionary time, before the split of A. thaliana and A. lyrata. The cluster was not detected in the A. halleri and A. arenosa accessions included in our study. Analysis of additional accessions as more genome sequences become available will establish whether lack of the thalianol BGC is a consistent feature of these two species. Comparisons of the thalianol clusters in A. thaliana and A. lyrata reveal differences in cluster organization and function. In A. lyrata, five clustered genes are necessary and sufficient for biosynthesis of the pathway end-product epi-thalianin (T17), while in A. thaliana a further two unlinked genes (THAR1 and THAR2) are required to give the corresponding pathway end-product, thalianin (T10) (Fig. 5). Thus the interplay between the core BGC and unlinked auxiliary genes may provide a mechanism for diversification of BGCs between plant species.Fig. 5Schematic illustrating the cluster variants and polymorphisms identified for the Arabidopsis thalianol cluster in this study.The ancestral state for the two unlinked genes THAR1 and THAR2 is ambiguous based on the current data. The divergence (between species) and polymorphisms (within species) for the thalianol cluster are shown.Future experiments will establish whether epi-thalianin contributes to the establishment of microbial communities in and around A. lyrata roots, and if so how distinct these communities are from those of A. thaliana10.SNPs and genomic deletion/insertion events have been implicated in the maintenance and diversification of BGCs in fungi, along with horizontal gene transfer from taxonomically distant fungal species20. Our investigations of the genomes of over a thousand A. thaliana accessions from around the world reveal that SNPs and small indels are the most common sequence polymorphisms in the thalianol cluster genes, followed by chromosomal inversions and gene loss (Fig. 5). Overall, the frequency of predicted deleterious mutations was low, suggesting that the thalianol cluster is fixed and functional in the vast majority of accessions.Our results provide evidence to suggest that chromosomal inversions may lead to cluster compaction (i.e., by bringing the THAA2 gene into the core thalianol cluster) (Fig. 3b, c). Although the role of chromosomal inversions in evolution in animals has been extensively investigated21,22, examples of such studies in plants are few23. Chromosomal inversions suppress recombination within and around the inversion region, and so may insulate clustered coadapted alleles against dispersal24 (Fig. 6). Indeed, our systematic LD analysis identified significantly more coinherited biallelic SNPs in the genomic neighborhoods (GNs) of OSC genes that form parts of BGCs compared to those of nonclustered OSCs (Fig. 4b). In line with this, a recent population genomic study also identified evolutionarily recent inversions associated with adapted quantitative trait loci (QTLs) in Drummond’s rockcress (Boechera stricta)25. The high frequency of compact thalianol BGCs in natural A. thaliana populations (Fig. 3b–d) is thus in accordance with the signatures of positive selection detected on the thalianol BGC as a whole, suggesting that the chromosomal inversions that we have identified are more likely to have driven cluster assembly rather than cluster break-up. Closer physical colocalization of genes will reduce the likelihood of loss of single pathway genes during recombination2,26. Altogether, our results suggest that chromosomal inversion may be a genetic mechanism enabling growth and compaction of BGCs in plants.Fig. 6A proposed model for compaction/assembly of plant BGCs through chromosomal inversions.Schematic inspired by Wellenreuther and Bernatchez23. Genes encoding enzymes are indicated with colored arrows.The reasons for the existence of BGCs in eukaryotic genomes remain hotly debated. Several scenarios, i.e., coinheritance, coregulation, and avoidance of toxic effects from pathway intermediates have been proposed2. These various hypotheses are not mutually exclusive. The overriding point, however, is that the formation and maintenance of BGCs must inevitably be a consequence of natural selection. Analysis within and between species for the thalianol BGC suggest that both positive and purifying selections are likely involved in shaping the formation and maintenance of this cluster (Figs. 3c and 4). These antagonistic selection pressures may further reduce the local recombination between physically linked enzymatic genes26. As we learn more about the dynamics of BGCs within and between species, we will be able to gain further insights into the forces and mechanisms shaping genome architecture and adaptive evolution.MethodsPlant materials and growth conditionsSeeds of A. thaliana accessions were obtained from Professor Jian Hua (Cornell University) and NASC (http://arabidopsis.info/). Seeds of A. arenosa were obtained from Prof. Kirsten Bomblies (ETH, Zurich). All seeds were surface sterilized in 5% sodium hypochlorite for 7 min, followed by incubation in 70% ethanol for 5 min, and then washed by ddH2O for five times. The seeds were placed on ¼ MS medium, subjected to cold treatment (4 °C, 3 days) before moving to long-day conditions (22 °C; 16 h light/8 h dark). For A. lyrata, 3 weeks of cold treatment were applied. N. benthamiana plants were grown in a greenhouse under long-day conditions (22 °C; 16 h light/8 h dark).Genome sequencing and assembly of A. arenosaWe used 0.4 g of A. arenosa leaf material from the accession SNO (Strecno, Slovakia, 49.17417N, 18.86167E). The frozen leaf material was ground in liquid nitrogen, and then 10 mL of CTAB DNA extraction buffer was added (Tris-HCl 100 mM (Fisher Scientific), CTAB 2% (w/v) (Sigma), NaCl 1.4 M (Sigma), EDTA 20 mM (Sigma-Aldrich)) and 20 μL of Proteinase K at 20 mg/mL (Qiagen). The mixture was incubated at 55 °C for 1 h then cooled on ice. Chloroform (0.5× volume; Fisher Scientific) was added and inverted to mix. Material was centrifuged at 1008×g for 30 min, and the upper phase recovered. Then 1× volume of phenol:chloroform:isoamyl alcohol (25:24:1; Sigma) was added and the mixture centrifuged for 30 min at 1008×g. The upper phase was recovered and 10% volume NaOAc at 3 M (Sigma) was added along with 2.5× volume of ice cold 100% ethanol (Sigma-Aldrich). The tubes were inverted to mix and incubated on ice for 30 min before centrifugation for 30 min at 1008×g (4 °C). The pellets were washed in 4 mL of ice cold 70% ethanol (Sigma-Aldrich). Tubes were spun for 10 min at 1008×g at 4 °C and the 70% ethanol wash repeated twice more. The pellets were air dried and resuspended in 300 μL nuclease-free water (Qiagen) with 3 μL RNase A at 4 mg/mL (Qiagen). DNA concentrations were determined on a Qubit Fluorometer 2.0 (Invitrogen) using the Qubit dsDNA HS Assay kit. Fragment sizes were assessed using a Q-card (OpGen Argus) and the Genomic DNA TapeStation assay (Agilent).The DNA was diluted to 0.5 ng/μL with EB (Qiagen) and checked with a Qubit Fluorometer 2.0 (Invitrogen) using the Qubit dsDNA HS Assay kit. The Chromium User Guide was followed as per the manufacturer’s instructions (10× Genomics, CG00043, Rev A). The final library was quantified using qPCR (KAPA Library Quant kit, Illumina; ABI Prism qPCR Mix, Kapa Biosystems). The sizes of the library fragments were monitored using a Bioanalyzer (High Sensitivity DNA Reagents, Agilent). Samples were pooled based on the molarities calculated using the two QC measurements.Sample SNO was sequenced on an Illumina HiSeq2500, using Rapid Run V2 mode150 PE, generating 82.10 Mb reads. These were assembled using Supernova 2.0.027, giving a raw coverage of 57.91× and effective coverage of 45.30×. The molecule length was 26.58 Kb.The assembly scaffold N50 was 2.19 Mb and overall size (counting only scaffolds longer than 10 Kb) was 127.02 Mb. Gene content was estimated using BUSCO v228 by searching database embryophyte_odb9. This showed that the gene space of the A. arenosa assembly was nearly complete, with 97.5% of the plant-specific BUSCOs present and 1.4% missing. Of those present, 4.7% were duplicate copies (Supplementary Table 5).RNA and cDNA preparationA. lyrata plants postflowering were used for total RNA isolation. Roots were recovered from soil, washed thoroughly with sterile water and dried with a clean tissue. Around 10 mg of ground roots were extracted with 600 μL of TRIzol reagent (ThermoFisher, Carlsbad, CA, USA). Chloroform (120 μL) was added to remove proteins and other impurities. After centrifugation (13,000×g; 10 min), an aliquot (400 µL) of supernatant was recovered and an equal volume of isopropanol added. Following mixing the pellet was recovered by centrifugation (10,000×g; 10 min) and washed with 70% ethanol. After centrifugation (10,000×g; 5 min) the pellet was dried at room temperature for 10 min. Sterilized water (30 μL) was then added to dissolve the total RNA. Reverse transcription reactions were performed using ≤2 μg of total RNA, random primers and a reverse transcription kit (Agilent, Santa Clara, CA, USA). The cDNA library was then used for amplification of the coding sequence of AlTHAA2 (AL8G20050). A pair of gene-specific primers attached with GATEWAY att sites were used (Fwd: GGGGACAAGTTTGTACAAAAAAGCAGGCTTCATGGACACCATGAAGGTTGAAATC; Rev: GGGGACCACTTTGTACAAGAAAGCTGGGTTTTAGATCAATACACTTGGATTTG). cDNAs were cloned into the GATEWAY entry vector pDONR207 (Invitrogen).Transient plant expressionThe pEAQ-Dest-1 expression vector was used for transient plant expressing experiments29. Constructs for transient expression of other A. lyrata cluster genes and A. thaliana thalianol cluster genes were adopted from previous work10,11. All constructs were verified by sequencing and introduced into A. tumefaciens strain LBA4404. Agro-infiltration of N. benthamiana was carried out as described previously30. N. benthamiana leaves were harvested 6 days after infiltration.LC-MS-IT-TOF analysisTen mg of freeze-dried N. benthamiana leaves were homogenized with a tungsten bead (3 mm, Qiagen) and extracted with 200 µL of pure methanol (sonication 30 min). The extracts were centrifuged at 16,000×g for 1 min. Aliquot of (170 µL) supernatant was recovered and filtered through 0.45 µm PTFE spin columns before injection onto LC. For root metabolites profiling, ten-days roots grown on vertical ¼ MS plates were harvested and flash frozen with liquid nitrogen. Samples were then freeze-dried. For comparison, 10 mg of sample were weighed out and extracted as described for tobacco samples. LC-MS analysis was carried out on a Prominence/Nexera UPLC system attached to an Ion-trap Tof mass spectrometer (Shimadzu). Separation was performed on a 100 × 2.1 mm Kinetex reverse column (1.7 µm C18 100 Å), using the following gradient of acetonitrile (solvent B) versus 0.1% formic acid in miliQ water (solvent A): 0.5 min, 2% B; 5 min, 10% B; 17 min, 30% B; 33 min, 90% B; 35.8 min, 100% B; 43 min, 100% B; 45 min 2% B. The flow rate was 0.5 mL/min and the run temperature was set at 40 °C. Injection volume was 1 µL. The MS data were acquired in both positive and negative modes. Spectra were collected from m/z 200–2000 with automatic sensitivity control set to a target of 70% optimal base peak intensity. The instrument was calibrated immediately before analysis using sodium trifluoroacetate cluster ions, according to the manufacturer’s instructions. The standards (T10 and T17) were generated in previous work10.In silico analysis of publicly available transcriptome dataRNA-seq data (SRA files) for five tissues (root, leaf, stem, flower, and silique) from A. thaliana (Col-0) and A. lyrata were downloaded from the National Center for Biotechnology Information (NCBI) BioProject PRJNA33605331. RNA sequence data for roots and shoots for A. thaliana accessions Col-0, Bay-0, Cvi-0, Ita-0, Kas-1, LP2-6, and Ws-2 were downloaded from NCBI BioProject PRJEB1409232. The sequencing quality was checked by FastQC v0.10.133. Trimmomatic v0.3934 was used to trim adapters and overrepresented sequences. Sequence data were then aligned to reference genomes (PHYTOZOME v.12.0; https://phytozome.jgi.doe.gov/pz/portal.html) by HISAT2 v2.1.035. Abundance was called using StringTie v1.3.536. The transcripts per million (TPM) values were then extracted as expression levels for genes of interest.Genome mining and plantiSMASH analysisThe Arabidopsis genome sequences (A. thaliana accession Col-0, A. lyrata, A. halleri) were retrieved from PHYTOZOME v.12.0. The genome of A. arenosa was generated de novo as part of this work. OSC genes were identified using HMMER337 with HMMER profiles (pHMMs) PF13243 and PF13249 downloaded from the PFAM library38. The cut_tc (trusted cut-off) option was used. Pseudogenes for OSCs (predicted protein sequence < 600 amino acids) were discarded from the analysis. The HMMER output was aligned using MUSCLE39. The alignment was used to infer the maximum likelihood tree by RaxML40. The OSC tree topology was consistent with a previous study that included OSCs from the wider Brassicaceae11. For plantiSMASH analysis14, the genome files (FASTA format) and annotation files (GFF3 format) were converted to GenBank format as input files.Long read assemblies for the thalianol cluster genomic regionLong read genome sequences for the thalianol cluster region from 22 A. thaliana accessions were retrieved from the 1001 Genomes Plus project. Seven used in this study (Nd-1, An-1, Cvi-0, Eri-1, Kyo, Ler-0, and Sha) were previously published41,42.Pair-wise dN/dS analysisThe pairwise dN/dS (ω) analysis averages the dN/dS ratio across all sites. To determine the estimates of ω, pairwise CDS (coding sequence) alignments of A. thaliana accession Col-0 and A. lyrata for the thalianol cluster genes and phytosterol pathway genes43 were used as an input for the codeml program from PAML 4.944. To determine if the dN/dS ratio was significantly different from 1, the program was executed twice with different control files. To obtain the maximum likelihood estimates of ω, the control file was set as: runmode = −2, model = 0, NSsites = 0, Fix_omega = 0. To obtain the likelihood with ω = 1, the control file was set as: runmode = −2, model = 0, NSsites = 0, Fix_omega = 1, Omega =1. The log likelihood values from the two executions were subtracted. The negative of twice of this value was used for likelihood ratio test (LR test). The p value was determined by comparison to χ2 with one degree of freedom. The estimated ω values with p value < 0.001 were plotted.Codon-based positive selection analysisThe protein-coding DNA sequences were aligned with TranslatorX45. In order to minimize negative impacts due to sequence size, gene sequences less than (for THAS min = 2200 bp; for THAO min = 1450 bp; for THAH min = 1400 bp; for THAA min = 1200 bp) were not included for the analysis. The sequences were aligned by MUSCLE39. Poorly aligned regions in the alignment were removed by Gblocks46. The Gblocks filtered nucleotide sequence were further translated to protein sequence and used to infer phylogenetic tree. The final Gblocks nucleotide sequence alignment and the protein phylogenetic tree were used for various evolutionary tests using HyPhy47. The input file for HyPhy analysis is shown in Supplementary Data 1. BUSTED analysis (similar to “branch-site” model in PAML) was used to infer whether a gene has experienced positive selection at at least one site on at least one branch48. MEME analysis (similar to “site” model in PAML) was used to detect individual sites evolving under positive selection in a proportion of branches (labeled as {foreground} in the input file)49. For both analyses, “universal genetic code” were selected and p < 0.05 was set for significance.Tests for neutral evolutionMcDonald and Kreitman analysis (MK test) is less sensitive to demographic factors such as migration, making it useful for detecting selection18. It captures infrequent adaptive mutations which fix fast relatively to regular neutral mutations, therefore elevating the divergence ratio (between species) over the polymorphism ratio (within species), a signature of positive selection. To examine departures from the neutral theory of molecular evolution, MK test was performed on a publicly available web interface (iMKT) (https://imkt.uab.cat/)50. The input sequences for polymorphism were downloaded from the 1001 Arabidopsis genome browser (http://signal.salk.edu/atg1001/3.0/gebrowser.php). The corresponding outgroup gene sequences from A. lyrata were taken from either11 for the thalianol core cluster genes or retrieved via reciprocal blast of the Col-0 reference gene on the Phytozome website (https://phytozome.jgi.doe.gov/pz/portal.html) for THAA2. Several versions of the MK test were performed to compute the polymorphism data from 1001 accession genomes and the outputs were compared with divergence data between A. thaliana and A. lyrata. The standard MK test can detect positive selection if adaptive mutations rapidly reach fixation and thus contribute relative more to divergence than to polymorphism when compared to neutral mutations. In this case, a positive alpha (α) indicates the proportion of nonsynonymous substitutions that have been fixed by positive selection. To avoid bias from the segregation of slightly deleterious nonsynonymous substitutions, we also applied Fay, Wycoff, and Wu (FWW) correction51 on the standard MKT by considering only those polymorphic sites with a frequency above the cutoff of 0.05. Sometimes when relaxed purifying selection reaches neutral, it can cause an increased level of polymorphism relative to divergence, therefore adaptive estimation (α) will be underestimated. By taking account of bias from both segregation of weakly deleterious variants and relaxed purifying selection, we further applied Extended MKT52 to the thalianol pathway genes.Construction of a phylogenetic tree for A. thaliana accessions using SNP dataThe 1001_SNP_MATRIX was downloaded from the 1001 genomes data center (https://1001genomes.org/data/GMI-MPI/releases/v3.1/SNP_matrix_imputed_hdf5/).SNPs with a minor allele frequency (MAF) of 0.05 or greater (28,775 variants) were then computed by SNPhylo53 to generate a maximum likelihood tree.Linkage disequilibrium analysisThe 1001_SNP_MATRIX was used to simulate the genome-wide linkage disequilibrium (squared allele frequency correlation, r2) using TASSEL 554. Only the top 50 r2 values are reported. The p-values for biallelic sites were calculated using Fisher’s exact test in Tassel. r2 = 0.1 was used as the threshold. r2 values from biallelic sites that are both located within the defined OSC gene-linked genomic regions (50 kb before and after the coding region of an OSC gene) were extracted. There are 13 OSC loci in the A. thaliana Col-0 genome9. Five OSC genes were located in clustered OSC gene-centric genomic regions: THAS (At5g48010)9, MRN (At5g42600)55, PEN3 (At5g36150)56, and ABDS (At4g15340, At4g15370)15. The nonclustered OSC genes11 were: LAS1 (At3g45130), PEN6 (At1g78500), At1g66960, At1g78950-970 (At1g78950, At1g78955, At1g78960, At1g78970), and CAS1 (At2g07050). The tandem duplicates in the At1g78950-970 GN were considered as a single locus, so the 50 kb on the left was taken before At1g78950 and the 50 kb on the right was taken after At1g78970 gene.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationReporting SummaryDescription of Additional Supplementary FilesSupplementary Data 1 | nature communications | [
"Article"
] | [
"Evolutionary genetics",
"Plant genetics",
"Secondary metabolism"
] | gene organization in eukaryotes genome evolution function genomes lack operons contain clustered genes related in function sequence examples are biosynthetic gene clusters (BGCs studied in reported from plants compounds agricultural traits disease resistance pharmaceutical evidence plant clusters not by horizontal gene transfer generated through gene duplication neofunctionalization genomic relocation1,2 little known about how gene clusters form examples syntenic BGCs variants same molecules reported when species compared for steroidal glycoalkaloids in Solanaceae triterpenes bitterness in Cucurbitaceae7,8 large-scale analysis of evolutionary dynamics of BGCs hindered by lack of sequence resources.Understanding clustering of functionally related nonhomologous genes crucial to relationship genome organization evolution of complex adaptive traits 10 years ago characterized “operon-like” gene cluster from Arabidopsis thaliana thalianol BGC)9 cluster in shaping root microbiota10 sequence data for A. thaliana to investigate occurrence nature evolution of paradigm BGC at vertical horizontal levelsresults reveal thalianol BGC evolved in Arabidopsis genus evolving between species genes A. thaliana thalianol BGC fixed conserved biochemical function purifying selection genes signatures localized positive selection identified cluster region Analysis genome thaliana reveals compaction thalianol BGC chromosomal inversions divergent polymorphic cluster variants provide insights into birth life death BGCs biochemical function evolutionary origin thalianol original discovery thalianol BGC in A. thaliana Col-0 reported four adjacent coexpressed genes (THAS THAH THAO THAA1) biosynthesis thalianol analysis revealed three additional genes linked THAA2 unlinked THAR1 THAR2) for synthesis thalianin original four cluster genes conserved between A. thaliana Col-0 A. lyrata producing acetylated thalianol derivative-dihydroxy-16-keto-thalian-15-yl acetate expressed in Nicotiana order four core genes same species THAA2 ortholog in A. lyrata differentintervening genes between THAA1 THAA2 species not implicated in plant metabolism not syntenic (Fig. 1b).Fig emergence thalianol cluster Arabidopsis genus thalianin pathway in A. thaliana from ref. Genes indicated by OSC oxidosqualene cyclase CYP cytochrome P450 ACT_IIIa acyltransferase subfamily IIIa compounds numbered ref. 14. Comparison thalianol cluster A. thaliana A. lyrata Orthologous genes connected with gray lines Species tree presence thalianol cluster times divergence from Hohmann et al.12.Thalianin triterpenes triterpene BGCs 13 Brassicaceae identified similar triterpene BGC Capsella rubella diverged from A. thaliana 8–9 million years (Fig. 1c). cluster independent evolutionary thalianol-like clusters not detected in other speciesorigins thalianol cluster interrogated genome Arabidopsis species halleri13 arenosa first step triterpene biosynthesis 2,3-oxidosqualene to triterpene scaffolds oxidosqualene cyclases Systematic mining four Arabidopsis genomes identified 55 OSC genes Phylogenetic analysis genomic locations OSC genes occurrence syntenic thalianol synthase) genes restricted monophyletic branch A. thaliana A. lyrata OSC genes analyses four Arabidopsis genomes plantiSMASH analysis detected thalianol BGCs in A. thaliana A. lyrata thalianol cluster likely evolved in common ancestor Arabidopsis around 5.8 MYA12 1c).Comparison thalianol cluster between full thalianin pathway studied in A. thaliana A. lyrata not cloned putative THAA2 ortholog from A. lyrata Coexpression with four A. lyrata thalianol cluster genes in N. benthamiana yielded epi-thalianin (T17) (Fig five clustered genes conserved between A. thaliana A. lyratafive genes coexpressed in roots A. thaliana A. lyrata (Fig. 2b). A. thaliana oxidoreductase genes THAR1 THAR2 (AT1G66800 required for epimerization C3 hydroxy T7 thalianin (Fig. 1a). A. lyrata T10 but T17 (Fig. isomers detectable in C. rubella A. arenosa lack thalianol BGC (Figs. 1c 2a). thalianol BGC evolved in Arabidopsis genus dynamically evolving when A. thaliana A. lyrata presence T17 in A. lyrata root extracts implies thalianol BGC encodes full biosynthetic pathway without genes BLASTP synteny analysis identify orthologs of THAR1 THAR2 in A. lyrata Transcriptome analysis related gene thaliana THAR2 (AL1G20010 70.5% expressed in flowers roots related thaliana THAR1 (AL4G46670 81.5% similar expression thalianol (Fig. 2b). THAR2 gene required for biosynthesis arabidin triterpene from A. thaliana Col-0 BGC10,15arabidin gene cluster not present in A. Fig. results suggest thalianol cluster genes share expression patterns functions in A. thaliana A. lyrata five clustered genes encode biosynthetic pathway epi-thalianin (T17) A. thaliana genes THAR1 THAR2 convert T7 into thalianin (T10) (Fig. auxiliary genes drives diversification thalianol BGC between species LC-MS analysis extracts Nicotiana benthamiana leaves-expressing thalianol genes root extracts A. thaliana A. lyrata mass fragment.3622 detect T10 T17 MS spectra EIC peaks Supplementary Fig. 3b experiment repeated three times Comparison expression patterns thalianol cluster-related genes organs A. thaliana A. lyrata Transcripts per million) values heatmap colors indicate gene expression Z-score deviation mean Divergence thalianol pathways in A. thaliana A. lyrata black arrows show directions pathways from THAS numbering products Fig. 1a structural differences between products T10 T17 highlighted Source dataVariation thalianol cluster examined polymorphisms SNPs small indels large gene deletion variations in thalianol cluster genes sequences 1135 A. thaliana accessions17 SNPs genes most common (97%), followed small indels (55% 36% polymorphisms mostly benign two accessions SNPs predicted large effect on THAS gene most small indels in UTR regions ~2% (21/1135) accessions large gene deletions 16 accessions lacked THAS THAO THAA1 genes large 3′-UTR region of THAA2 gene~6 Other deletions found at lower frequencies two unlinked genes (THAR1 and THAR2) present in nearly all A. thaliana accessions analysis transcriptome data for seven A. thaliana accessions revealed similar expression patterns for thalianol cluster genes Col-09 Metabolite analysis confirmed T10 detected in root extracts from accessions thalianol cluster genes not cluster likely functional in majority of A. thaliana accessions.Fig. 3Variation in thalianol cluster within A. thaliana speciesthaliana accessions Arabidopsis 1001 homologs thalianol genes Col-0 Deletions indicated gray bars three types deletion patterns colored triangles fully region accession Lillö-1 inversion illustrated bottom Chromosomal inversions thalianol region A. thaliana accessions Frequencies Col-0-like contiguous thalianol clusters 21 A. thaliana accessions Sha Eri newly PacBio-sequenced not included 1135 Phylogenetic tree 1135 A. thaliana accessions distances calculated 28,775 filtered SNPs RAxML40 Tree node support (>90%) pink dots branches Accessions thalianol pathway gene deletions chromosomal inversion variations indicated monophyletic groups collapsed polymorphisms gene order thalianol BGC analyzed 22 A. thaliana genome sequence assemblies PacBio confirmed deletion profile Lillö-1 accession order four core genes conserved intervening genes between THAA1 and THAA2 not 21 A. thaliana genomes Lillö-1 three same gene order as Col-0 thalianol clusterremaining 17 accessions THAS THAH THAO THAA1 THAA2 genes contiguous no predicted genes between THAA1 THAA2. reversed orientation genes alignment accessions experienced chromosomal inversions compact thalianol clusters recently (Fig. 3b c Fig 5) sample size small (21 out of 1135 contiguous forms thalianol cluster majority (17 out of 22; 77%) widely distributed (Fig. hypothesize chromosomal inversion facilitate compaction thalianol BGC populations.Analysis selection thalianol variations thalianol BGC simulated variant data levels nonsynonymous substitution) analysis identified relaxed purifying selection thalianol cluster genes pathway genes conserved Arabidopsis-genus distribution (Fig. 4a). later thalianol pathway genes less constrained early 6) consistent roles THAA2 in A. thaliana A. lyrata Fig. 2c Branch-site model analysis gene-wide positive selection thalianol cluster genes site model analysis episodic positive selection sites across clustered genes analyses agreement with conserved enzymatic functions thalianol pathway genes between speciesperformed McDonald–Kreitman) polymorphism divergence positive selection recent analyses found no positive selection thalianol cluster genes Table 4) results suggest genes thalianol BGC purifying selection conserved enzymatic functions expression patterns species. 4Molecular evolution analysis species Pairwise dN/dS analysis A. thaliana A. lyrata thalianol pathway genes pathway genes controls purifying selection arrows illustrate biosynthetic pathways synthase sterol biosynthesis THAS thalianol-derived triterpenes Two-sided Student t-test 8. Comparison LD (r2 > 0.1) numbers A. thaliana 1001 genomes data neighborhoods clustered non-clustered OSC genes defined 50 Kb OSC gene At1g78500 gene PEN6 close four OSCs expression patterns genes neighboring At1g78500 unlikely functional BGC PEN6-centric genomic neighborhood) non-clustered Two-sided Student t test = 7. Plot LD values (r2 > 0.1) THAS-centric GN genomic distances biallelic SNPs two LDs four cluster genes marked redblack line smoothed trend local regression evolutionary tests on single genes expanded tests for positive selection thalianol cluster genome-wide linkage disequilibrium (LD) analysis using SNP data 1135 A. thaliana genomes genome A. thaliana accession Col-0 contains 13 OSC genes different functions9 in nine genomic neighborhoods duplicated OSC genes as one GN examined LD extensions of nine OSC GNs counted allele frequency correlation (r2) values for biallelic sites OSC-centric GNs 50 Kb Fig. results demonstrate GNs OSC gene clusters more nonrandom associations biallelic sites than GNs OSC genes nonrandom associations biallelic sites in THAS-centric GN in flanking regions thalianol cluster decay with increasing genetic distance (Fig. two detected in region four cluster genes 4c suggests thalianol cluster undergoing selective sweep reduced diversity elevated LDs in flanking regions due to hitch-hiking effect19 low levels single feature polymorphism across thalianol core cluster thalianol cluster under directional positive selectionlarge investigation evolutionary dynamics plant BGC thalianol cluster study suggests ancestral thalianol cluster evolved recently before split A. thaliana A. lyrata cluster not detected in A. halleri A. arenosa accessions Analysis additional accessions lack thalianol BGC species Comparisons thalianol clusters A. thaliana A. lyrata reveal differences organization function A. lyrata five clustered genes necessary for biosynthesis epi-thalianin (T17) A. thaliana two genes (THAR1 THAR2)-product thalianin (T10) interplay core BGC unlinked auxiliary genes diversification BGCs between species cluster variants polymorphisms Arabidopsis thalianol cluster ancestral state unlinked genes THAR1 THAR2 ambiguous divergence species polymorphisms (within species thalianol cluster shown.Future experiments establish epi-thalianin microbial communities A. lyrata roots A. genomic deletion/insertion events implicated in maintenance diversification BGCs in fungi gene transfer from distant fungal investigations genomes thousand A.thaliana accessions reveal SNPs small indels common sequence polymorphisms in thalianol cluster genes followed chromosomal inversions gene loss. 5) frequency deleterious mutations low thalianol cluster fixed functional majority accessions results evidence chromosomal inversions lead to cluster compaction THAA2 gene into core thalianol cluster (Fig. 3b role inversions in evolution animals investigated21 in plants inversions suppress recombination insulate clustered alleles against dispersal24 systematic LD analysis identified more coinherited biallelic SNPs in OSC genes BGCs nonclustered OSCs. recent population study identified recent inversions adapted quantitative trait loci in Drummond’s rockcress high frequency of compact thalianol BGCs in natural A. thaliana populations. accordance with signatures positive selection thalianol BGC chromosomal inversions driven cluster assembly break-up Closer colocalization genes loss single pathway genes during recombination2 results suggest chromosomal inversion enabling growth compaction BGCs in plantsproposed model compaction plant BGCs chromosomal inversions inspired Wellenreuther Bernatchez23 Genes encoding enzymes indicated colored arrows reasons BGCs eukaryotic genomes debated scenarios coinheritance coregulation avoidance effects hypotheses not exclusive overriding formation maintenance BGCs natural selection Analysis thalianol BGC positive purifying selections formation maintenance cluster (Figs. 3c 4) antagonistic selection pressures reduce recombination enzymatic dynamics BGCs species insights genome adaptive evolution materials growth conditionsSeeds A. thaliana obtained from Professor Jian Hua (Cornell University NASC Seeds A. arenosa Prof. Kirsten Bomblies (ETH seeds sterilized 5% sodium hypochlorite 7 min incubation 70% ethanol 5 min washed ddH2O five times seeds 1⁄4 MS medium cold treatment (4 °C 3 days long-day conditions (22 A. lyrata 3 weeks cold treatment N. benthamiana plants grown greenhouse long-day conditions (22 °C dark).Genome sequencing assembly A. used 0.4 g A.leaf material SNO Slovakia 49.17417N frozen leaf ground liquid nitrogen 10 mL CTAB DNA buffer added-HCl 100 mM CTAB 2% NaCl 1.4 M EDTA 20 mM 20 μL Proteinase K 20 mg/mL incubated 55 °C 1 h cooled ice Chloroform (0.5× added centrifuged 1008×g 30 min upper phase recovered 1× volume:chloroform:isoamyl alcohol added centrifuged 30 min upper phase recovered 10% NaOAc 3 M 2.5× ice cold 100% ethanol tubes incubated ice 30 min pellets washed 4 mL ice cold 70% ethanol Tubes spun 10 min 1008×g 70% ethanol pellets air dried resuspended 300 μL nuclease-free water 3 μL RNase A 4 mg/mL DNA concentrations determined Qubit Fluorometer 2.0 dsDNA HS Assay kitFragment sizes assessed Q-card Genomic DNA TapeStation assay DNA diluted 0.5 ng/μL EB checked Qubit Fluorometer 2.0 dsDNA HS Assay Chromium User Guide followed Genomics library quantified qPCR Library ABI Prism qPCR Mix sizes monitored Bioanalyzer Sensitivity DNA Reagents Samples pooled molarities QC measurements SNO sequenced Illumina HiSeq2500 Rapid Run V2 82.10 Mb reads Supernova 2.0.027 raw coverage 57.91× effective coverage 45.30× molecule length 26.58 Kb assembly scaffold N50 2.19 Mb overall size 127.02 Mb Gene content estimated BUSCO v228 embryophyte_odb9 gene space A. arenosa assembly nearly complete 97.5% plant BUSCOs present 1.4% missing 4.7% duplicate copies Table 5).RNA cDNA preparationA. lyrata plants RNA isolation Roots recovered washed dried 10 mg roots extracted 600 μL TRIzol reagent Chloroform (120 μL added proteins impuritiescentrifugation (13,000×g 10 (400 μL supernatant recovered isopropanol added pellet recovered 10 min washed 70% ethanol 5 dried room temperature 10 min Sterilized water (30 μL added dissolve RNA Reverse transcription reactions ≤2 μg total RNA random primers reverse transcription kit (Agilent Santa Clara CA cDNA library amplification coding sequence AlTHAA2 (AL8G20050) gene-specific primers GATEWAY sites used cDNAs cloned GATEWAY entry vector pDONR207 plant pEAQ-Dest-1 expression vector Constructs expression A. lyrata A. thaliana genes adopted previous constructs verified introduced A. tumefaciens strain LBA4404. Agro-infiltration N. benthamiana leaves harvested 6 days after infiltration mg freeze-dried N. benthamiana leaves homogenized tungsten bead extracted 200 μL pure methanol 30 extracts centrifuged 16,000×g 1 min (170 μL supernatant recovered filtered through 0.μm PTFE spin columns injection LC root metabolites ten-days roots 1⁄4 MS plates harvested frozen liquid nitrogen Samples freeze-dried 10 mg sample extracted tobacco samples LC-MS analysis Prominence/Nexera UPLC system Ion-trap Tof mass spectrometer Separation 100 × 2.1 mm Kinetex reverse column (1.7 μm C18 acetonitrile B 0.1% formic acid water 0.5 min 2% B 5 10% B 17 30% B 33 90% B 35.8 min 100% B 43 min 100% B 45 min 2% B flow rate 0.5 mL/min run temperature 40 °C Injection volume 1 μL MS data positive negative modes Spectra collected m/z 200–2000 control 70% base peak intensity instrument calibrated sodium trifluoroacetate cluster ions standards (T10 T17) generated silico analysis transcriptome-seq data tissues leaf stem A. thaliana lyrata BioProject RNA sequence data roots shoots A. thaliana BioProject sequencing quality checked FastQC v0.10.133.Trimmomatic v0.3934 trim adapters sequences data aligned genomes (PHYTOZOME v.12.0 HISAT2 v2.1.035 Abundance called StringTie v1.3.536 transcripts per million values extracted expression levels genes.Genome mining plantiSMASH Arabidopsis genome sequences (A. thaliana. lyrata. halleri retrieved from PHYTOZOME v.12.0 genome A. arenosa generated de novo OSC genes identified HMMER337 profiles PF13243 PF13249 PFAM cut_tc option used Pseudogenes OSCs discarded HMMER output aligned MUSCLE39 maximum likelihood tree RaxML40 OSC tree topology consistent previous study Brassicaceae11 plantiSMASH genome files annotation converted to GenBank format read assemblies thalianol cluster sequences 22 A. thaliana accessions retrieved 1001 Genomes Plus project Seven (Nd-1 An-1 Cvi-0 Eri-1 Kyo Ler-0 Sha previously-wise dN/dS averages dN/dS ratio sitesestimates ω CDS alignments A. thaliana Col-0 A. lyrata thalianol cluster phytosterol pathway codeml program PAML 4.944 dN/dS ratio different 1 program executed twice control files maximum likelihood estimates ω control file runmode = −2 model 0 NSsites 0 Fix_omega 0 ω = 1 runmode −2 model 0 0 Fix_omega 1 Omega =1 log likelihood values subtracted negative used for likelihood ratio test p value determined comparison χ2 one degree freedom estimated ω values with p value < 0.001 plotted.Codon-based positive selection protein-coding DNA sequences aligned with TranslatorX45 gene sequences less than THAS 2200 THAO 1450 1200 not included sequences aligned by MUSCLE39 Poorly aligned regions removed by Gblocks46 Gblocks filtered nucleotide sequence translated to protein sequence phylogenetic tree final Gblocks sequence alignment protein phylogenetic tree used for evolutionary tests HyPhy47 input file HyPhy analysis Supplementary Data 1. BUSTED analysis PAML gene positive selection one site branch48MEME analysis model PAML sites evolving under positive selection branches “universal genetic code” selected p < 0.05 significance neutral evolutionMcDonald Kreitman analysis test less sensitive to demographic factors migration useful captures infrequent adaptive mutations fix fast neutral mutations divergence ratio polymorphism ratio signature positive selection departures neutral theory evolution MK test performed web interface (iMKT input sequences polymorphism downloaded from 1001 Arabidopsis genome browser outgroup gene sequences A. lyrata taken thalianol retrieved reciprocal blast Col-0 gene Phytozome website versions MK test polymorphism data 1001 genomes outputs compared with divergence data between A. thaliana A. lyrata standard MK test positive selection adaptive mutations rapidly reach fixation contribute more to divergence than polymorphism neutral mutations positive alpha (α) indicates proportion nonsynonymous substitutions fixed by positive selectionbias deleterious nonsynonymous substitutions applied Fay Wycoff Wu) correction51 standard MKT polymorphic sites frequency above cutoff 0.05. relaxed purifying selection neutral polymorphism adaptive estimation (α) underestimated bias deleterious variants applied Extended MKT52 thalianol pathway genes tree A. thaliana accessions SNP 1001_SNP_MATRIX downloaded 1001 genomes data center minor allele frequency 0.05 or greater (28,775 variants computed SNPhylo53 maximum likelihood tree.Linkage disequilibrium 1001_SNP_MATRIX genome-wide linkage disequilibrium frequency TASSEL 554 top 50 r2 values reported p-values biallelic sites calculated Fisher’s exact test Tassel r2 = 0.1 threshold r2 values biallelic sites defined OSC gene-linked genomic regions extracted 13 OSC loci A. thaliana Col-0 genome9 Five OSC genes clustered OSC gene-centric genomic regions THAS MRN (At5g42600 PEN3 (At5g36150 ABDSnonclustered OSC genes11 LAS1 (At3g45130), PEN6 (At1g78500), At1g66960 At1g78950-970 CAS1 (At2g07050). tandem duplicates At1g78950-970 GN single locus 50 kb left before At1g78950 50 right after At1g78970.Reporting Nature Research Reporting Summary.Supplementary Additional Supplementary Data 1 | 50.1 | 0.655536 |
10.1038/s41467-020-17835-9 | PMC7438325 | To optimize operating conditions of large scale semiconductor quantum devices, a large parameter space has to be explored. Here, the authors report a machine learning algorithm to navigate the entire parameter space of gate-defined quantum dot devices, showing about 180 times faster than a pure random search. | Variability is a problem for the scalability of semiconductor quantum devices. The parameter space is large, and the operating range is small. Our statistical tuning algorithm searches for specific electron transport features in gate-defined quantum dot devices with a gate voltage space of up to eight dimensions. Starting from the full range of each gate voltage, our machine learning algorithm can tune each device to optimal performance in a median time of under 70 minutes. This performance surpassed our best human benchmark (although both human and machine performance can be improved). The algorithm is approximately 180 times faster than an automated random search of the parameter space, and is suitable for different material systems and device architectures. Our results yield a quantitative measurement of device variability, from one device to another and after thermal cycling. Our machine learning algorithm can be extended to higher dimensions and other technologies. | IntroductionGate defined quantum dots are promising candidates for scalable quantum computation and simulation1,2. They can be completely controlled electrically and are more compact than superconducting qubit implementations1. These devices operate as transistors, in which electrons are controlled by applied gate voltages. If these gate voltages are set correctly, quantum dots are created, enabling single-electron control. If two such quantum dots are created in close proximity, the double quantum dot can be used to define robust spin qubits from the singlet and triplet states of two electrons3,4. Due to device variability, caused by charge traps and other device defects, the combination of gate voltage settings which defines a double quantum dot varies unpredictably from device to device, and even in the same device after a thermal cycle. This variability is one of the key challenges that must be overcome in order to create scalable quantum circuits for technological applications such as quantum computing. Typical devices require several gate electrodes, creating a high-dimensional parameter space difficult for humans to navigate. Tuning is thus a time-consuming activity and we are reaching the limits of our ability to do this manually in arrays of quantum devices. To find, in a multidimensional space, the gate voltages which render the device operational is referred to in the literature as coarse tuning5,6.Here, we present a statistical algorithm which is able to explore the entire multidimensional gate voltage space available for electrostatically defined double quantum dots, with the aim of automatically tuning them and studying their variability. Until this work, coarse tuning required manual input7 or was restricted to a small gate voltage subspace8. We demonstrate a completely automated algorithm which is able to tune different devices with up to eight gate electrodes. This is a challenging endeavour because the desired transport features are only present in small regions of gate-voltage space. For most gate voltage settings, the device is either pinched off (meaning that the charge carriers are completely depleted so that no current flows) or too open (meaning that the tunnel barriers are too weakly defined for single-electron charge transport to occur). Moreover, the transport features that indicate the device is tuned as a double quantum dot are time-consuming to measure and difficult to parametrise. Machine learning techniques and other automated approaches have been used for tuning quantum devices5–14. These techniques are limited to small regions of the device parameter space or require information about the device characteristics. We believe our work significantly improves the state-of-the-art: our algorithm models the entire parameter space and tunes a device completely automatically (without human input), in approximately 70 min, faster than the typical tuning by a human expert.Our algorithm explores the gate-voltage space by measuring the current flowing through the device, and its design makes only a few assumptions, allowing it to be readily applied to other device architectures. Our quantum dot devices are defined in a two-dimensional electron gas in a GaAs/AlGaAs heterostructure by Ti/Au gate electrodes. DC voltages applied to these gate electrodes, V1–V8, create a lateral confinement potential for electrons. Particularly important are the two plunger gate voltages V3 and V7, which mainly tune the electron occupation of the left and right dots. A bias voltage Vbias is applied to ohmic contacts to drive a current (I) through the device. The device schematic, designed for precise control of the confinement potential15–17, is shown in Fig. 1a. Measurements were performed at 50 mK.Fig. 1Overview of device, and gate voltage space.a Schematic of a gate-defined double quantum dot device. b Left: Boundary hypersurface measured as a function of V2, V5, and V8, with fixed values of V1, V3, V4, V6, and V7. The current threshold considered to define this hypersurface is 20% of the maximum measured current. The gate voltage parameter space, restricted to 3D for illustration, contains small regions in which double and single quantum dot transport features can be found. These regions typically appear darker in this representation because they produce complex boundaries. Right: For particular gate voltage locations marked with green crosses, the current as a function of V7 and V3 is displayed. The top and bottom current maps display double and single quantum dot transport features, respectively.We consider the space defined by up to eight gate voltages between 0 and −2 V. This range was chosen to avoid leakage currents. In this parameter space, the algorithm has to find the desirable transport features within tens of mV. Identifying these features is slow because it is requires measuring a two-dimensional current map, i.e., a plot of I as a function of the two plunger gate voltages. Although other techniques for measuring the double quantum dot exist, such as charge sensing and dispersive readout, they also require other parameters to be retuned when the gate voltages vary and are therefore not suitable for automated measurements. Our algorithm is thus designed to minimize the number of current maps that it requires to find the transport features in question.We make two observations. Firstly, that for very negative gate voltages, no current will flow through the device, i.e., the device is pinched-off. Conversely, for very positive gate voltages, full current will flow and single electron transport will not be achieved. This means that transport features are expected to be found near the hypersurface that separates low and high current regions in parameter space. The second observation is that to achieve single-electron transport, a confinement potential is needed. The particular transport features that evidence single-electron transport are Coulomb peaks, which are peaks in the current flowing through the device as a function of a single plunger gate voltage. These observations lead us to only two modelling assumptions: (i) single and double quantum dot transport features are embedded near a boundary hypersurface, shown in Fig. 1b, which separates regions in which a measurable current flows, from regions in which the current vanishes; (ii) large regions of this hypersurface do not display transport features.The algorithm consists of two parts: a sampling stage that generates candidate locations on the hypersurface, and an investigation stage in which we collect data in the vicinity of each candidate location, i.e., close to the candidate location in gate voltage space (see Section “Investigating nearby voltage space, for precise definitions of the size of the regions explored around candidate locations”), to evaluate transport features (Fig. 2). The results of the investigation stage feed back into the sampler, which chooses a new candidate location in the light of this information. The purpose of the sampler is to produce candidate locations in gate voltage space for which the device operates as a double quantum dot. A block diagram of the algorithm is displayed in Fig. 3. Our modelling assumptions are based on the physics of gate defined devices leading to minimal constraints; we do not assume a particular shape for the hypersurface, and we instead allow measurements to define it by fitting the data with a Gaussian process. Overall, the algorithm minimises tuning times by identifying candidate locations on a hypersurface model that is updated with each measurement; by prioritising the most promising of these locations; and by avoiding the acquisition of two-dimensional current maps which do not correspond to a double quantum dot regime.Fig. 2Overview of the algorithm.The sampling phase stage produces candidate locations in gate voltage space, which are on the boundary hypersurface (pink surface). The distance between a candidate location (red cross) and the origin of the gate voltage space is marked with a dashed line. The investigation stage evaluates the local region by, for example, measuring current maps which are evaluated by a score function. (The current map displayed is an example of a measurement performed by the algorithm. It uses a colour scale running from red, the highest current measured, to blue, the lowest current). Evaluation results are fed back to the sampling stage.Fig. 3Flow diagram of the algorithm.(See text and Fig. 4 for a full description.) Each step is annotated with the corresponding panel in Fig. 4. Steps that involve interaction with the device are shaded brown, and entirely computational steps are grey. In the ablation studies of Section “Ablation study”, the different modules that constitute the algorithm are studied in terms of their contribution to the algorithm’s performance; these modules are marked by the blue background regions. The steps belonging to the initialisation, sampling, and investigation steps are indicated on the right.We demonstrate over several runs, in two different devices and over multiple thermal cycles, that the algorithm successfully finds transport features corresponding to double quantum dots. We perform an ablation study, which identifies the relative contribution of each of the modules that constitute the algorithm, justifying its design. Finally, we demonstrate that our algorithm is capable of quantifying device variability, which has only been theoretically explored so far18. We have done this by comparing the hypersurfaces found for different devices and for a single device in different thermal cycles.Automating experimental science has the impact to significantly accelerate the process of discovery. In this work we show that a combination of simple physical principles and flexible probabilistic machine learning models can be used to efficiently characterise and tune a device. We envisage that in the near future such judicious application of machine learning will have tremendous impact even in areas where only small amounts of data are available and no clear fitness functions can be defined.ResultsDescription of the algorithmThe algorithm starts with an initialization stage. This stage begins with setting Vbias. The current is then measured at the two extremes of the gate voltage space, Vj = 0 and Vj = −2 V for j = 1, . . . , N, where N is the number of gate electrodes. For the most negative extreme, the measured current should be 0, but current offsets might change this value for different measurement setups. The difference between the currents at these two extremes is the full-scale current which is used to set the threshold that defines the hypersurface. The search range was chosen as the typical gate voltage range used when tuning similar devices from scratch.The algorithm then begins an iterative process during which it alternates between the sampling and investigation stages. In each iteration, the sampling stage identifies a candidate location on the hypersurface in voltage space, attempting to select locations with a high probability of desirable transport features. The investigation stage then explores the nearby region of voltage space, attempting to identify whether current maps measured in this region show Coulomb peaks and honeycomb patterns. The presence of Coulomb peaks is reported back to the sampling stage as an evaluation result, which it uses in future iterations to inform its selection of new candidates. The steps that make up each iteration will now be described in detail.Searching for the hypersurfaceIn each iteration, the algorithm first locates the hypersurface in gate voltage space. To do this, it selects a search direction, specified by a unit vector u which during the first 30 iterations of the algorithm is selected randomly from a hypersphere, restricted to the octant where all gate voltages are negative. The gate voltages are then scanned along a ray beginning at the origin o and parallel to u (Fig. 4a). During this scan, the current is monitored; when it falls below a threshold of 20% of full scale, this is taken as defining a location v(u) on the hypersurface.Fig. 4Characterising the boundary hypersurface using machine learning.Each panel illustrates a step of the algorithm presented in Fig. 3. The gate voltage space, restricted to two dimensions for illustration, is divided into regions of near-zero (blue) and non-zero current (pink), separated by a boundary hypersurface. a Locating the hypersurface. The gate voltages are scanned along a ray (violet arrow) starting at the origin (white circle) and defined by direction u. By monitoring the current, the intersection with the hypersurface is measured. b To determine whether a region should be pruned, the algorithm scans each gate voltage individually toward the bottom of its range from a location just inside the hypersurface as shown. If only one scan intersects the hypersurface (as in the inset), future exploration of that region is inhibited by displacing the origin as shown. c Based on a short 1D scan, the location is classified according to whether it shows current peaks indicating Coulomb blockade. d If peaks are found, a 2D scan (violet square) is performed in the plane of V3 and V7, and is possibly repeated at higher resolution. e From the first thirty measurements (green and yellow circles), the algorithm builds a model of the hypersurface and assigns a probability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tilde{P}}_{{\rm{peak}}}$$\end{document}P~peak that peaks will be found. f To refine the model, the algorithm generates a set of candidate search locations (squares), each weighted by its corresponding value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tilde{P}}_{{\rm{peak}}}$$\end{document}P~peak, and selects one at random. A new scan is then performed in the corresponding direction to generate a new measurement of the hypersurface location. Steps d–f are then repeated indefinitely. Inset: Scheme for numerically sampling the hypersurface using simulated Brownian motion. Each point represents a simulated particle moving inside the enclosed volume. The collisions between the particles and the modelled hypersurface generate a set of candidate search locations.While this procedure correctly identifies locations for which current through the device is pinched off, it does not recognise whether the device is “tunable” in the sense that every gate voltage strongly affects the current. We find that for some locations, most gate voltages have little effect, which suggests that the measured current is not being determined by the potential in the quantum dot. With such a combination of gate voltages, a double quantum dot cannot be usefully formed. To reduce the amount of time spent exploring such regions of the hypersurface, we implemented the following heuristic pruning process (Fig. 4b), applied in each of the first 30 iterations. From the hypersurface intersection v(u), all voltages are stepped upwards to a location vδ(u) ≡v(u) + δ, where δ is a step-back vector with each component chosen to be +100 mV. Each voltage in turn is then swept downwards towards the bottom of its range or until the hypersurface is encountered. If the hypersurface is encountered only along one voltage axis k, then the origin for subsequent iterations is moved so that its k-component is equal to the k-component of vδ (Fig. 4b, inset). Over several iterations, this process prunes away search paths for which the hypersurface is not intersected within the chosen range.Investigating nearby voltage spaceHaving located the hypersurface, the algorithm then proceeds to investigate the nearby region of voltage space to determine whether a double quantum dot is formed. The investigation is carried out in the plane containing v(u) and defined by varying the two plunger gate voltages V3 and V7. These gates, selected before running the algorithm, are the ones that should predominantly shift the electrochemical potential in the left and right dots. If a double quantum dot is formed, the current should therefore show a honeycomb pattern in this plane, similar to Fig. 2.First, a one-dimensional scan is made in this investigation plane, starting at v(u) and running along the diagonal axis \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{V}}_{e}\equiv \frac{1}{\sqrt{2}}({\hat{V}}_{3}+{\hat{V}}_{7})$$\end{document}V^e≡12(V^3+V^7), where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{{V}_{i}}$$\end{document}Vi^ indicates a unit vector in voltage space (Fig. 4c). This scan is chosen to have length 128 mV and resolution 1 mV. A peak detection routine identifies the presence or absence of Coulomb peaks. If Coulomb peaks are absent, investigation here ceases and a new iteration begins.Next, if Coulomb peaks are present in this diagonal scan a two-dimensional scan is made (Fig. 4d). The scanning region is a square oriented along \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{V}}_{e}$$\end{document}V^e and its orthogonal axis \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{V}}_{a}\equiv \frac{1}{\sqrt{2}}({\hat{V}}_{3}-{\hat{V}}_{7})$$\end{document}V^a≡12(V^3−V^7). This square is bounded by v(u), and its side length is chosen to be 3.5 times the average peak spacing identified in the diagonal scan. (If the diagonal scan shows less than 3 peaks, the side length is set to be 100 mV.) The scan is made first at low resolution (16 × 16 pixels), and a score is assigned to the resulting current map. The score function (see Supplementary Methods, Score function) is a predefined mathematical expression designed to reward specific transport features that correspond to the visual features typically looked for by humans when manually tuning a device. In particular, it is designed to identify honeycomb patterns similar to Fig. 2 indicating the formation of a double quantum dot. It rewards current maps containing sharp and curved lines.If the score function of the low-resolution scan is high, it is repeated at high resolution (48 × 48 pixels). The score threshold is dynamically adjusted throughout the experiment so that 15% of low-resolution scans are repeated. (See Supplementary Methods, Optimal threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha ^{\prime}$$\end{document}α′, for a statistical analysis of the optimal threshold.) The high-resolution maps, scanned in regions of voltage space identified as showing desirable double-dot behaviour, constitute the output of the tuning algorithm.Searching efficiently by learning about the hypersurfaceTo more rapidly locate the hypersurface, and to increase the fraction of time spent exploring regions of gate space containing Coulomb peaks, the algorithm improves the search process of Section Searching for the hypersurface by incorporating information from its measurements. It applies this information beginning with the 31st iteration. To do this, it starts each iteration by using the measured locations of the hypersurface to generate a model hypersurface spanning the entire voltage space (Fig. 4e). The model is generated using a Gaussian process19 incorporating the uncertainty of the measured locations as explained in the Supplementary Methods, Gaussian process models. To each candidate search direction u, the model assigns an estimated distance to the hypersurface m(u) with uncertainty s(u). Furthermore, the model uses information on whether current peaks were identified in previous searches to assign to each point on the model hypersurface a probability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tilde{P}}_{{\rm{peak}}}$$\end{document}P~peak of expecting peaks.Using this model, the algorithm can now select new search directions u more efficiently. It is desirable to select search directions associated with a high probability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tilde{P}}_{{\rm{peak}}}$$\end{document}P~peak, while also occasionally exploring less promising regions of the hypersurface. To achieve this trade-off, the algorithm first generates a set of candidate search locations on the hypersurface (Fig. 4f). To generate a set that is approximately uniform despite the convoluted shape of the hypersurface, we adopt a selection routine based on simulated Brownian motion20; a set of “particles” is simulated inside the hypersurface, and each encounter with the hypersurface contributes one candidate location (Fig. 4f, inset). To each of these locations, the algorithm then assigns a weight proportional to the corresponding value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tilde{P}}_{{\rm{peak}}}$$\end{document}P~peak, and selects one location at random (i.e., using Thompson sampling). This location then defines a new search direction u.The model hypersurface is also used to improve the efficiency of the search. Instead of beginning at the origin (as in Fig. 4a), the new search scan begins at the location g(u) ≡ o + (m(u) − 2s(u))u, which should lie just inside the hypersurface (as in Fig. 4f). If m(u) − 2s(u) < 0, the search scan begins at o.Occasionally, the measured current at the beginning of the scan is below threshold, indicating that g(u) is already in the pinched-off region. In these cases, the algorithm scans in the opposite direction, along—u. Once the measured current increases above 0.8 of the value at o, the algorithm reverts to measuring in the u direction to locate the hypersurface in the usual way. Over many iterations, the algorithm thus builds up the required set of high-resolution current maps, measured with constantly improved efficiency.Experimental resultsThe performance of our algorithm is assessed by a statistical analysis of the expected success time μt. This is defined as the time it takes the algorithm to acquire a high-resolution current map that is confirmed a posteriori by humans as containing double quantum dot features. Note that this confirmation is only needed to assess the performance of the algorithm. Because human labelling is subjective, three different researchers labelled all current maps, deciding in each case if they could identify features corresponding to the double quantum dot regime, with no other information available. See Supplementary Methods, Bayesian statistics, for details of the multilabeller statistical analysis.Device tuningTo benchmark the tuning speed of our algorithm, we ran it several times on two different devices with identical gate architecture, Devices 1 and 2, and we compared its performance with a Pure random algorithm. The Pure random algorithm searches the whole gate voltage parameter space by producing a uniform distribution of candidate locations. Unlike our algorithm, which we will call Full decision, it does not include hypersurface weighting or pruning rules, but uses peak detection in its investigation stage. All Full decision runs presented in this section for Device 1 and Device 2 were performed during a single cool down (cool down 1). The Pure random runs in each device were performed in a different cool down (cool down 2).As mentioned in the introduction, we consider a gate voltage space whose dimension is defined by the number of working gate electrodes, and we provide a gate voltage range that avoids leakage currents. While for Device 1 we considered the eight-dimensional parameter space defined by all its gate electrodes, for Device 2 we excluded gate electrode 6 by setting V6 = 0 mV due to observed leakage currents associated with this gate.We define the average count \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{C}$$\end{document}C¯ as the number of current maps labelled by humans as displaying double quantum dot features divided by the number of labellers. For a run of the Pure random algorithm in Device 2 and five runs of our algorithm in Devices 1 and 2, we calculated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{C}$$\end{document}C¯ as a function of laboratory time (Fig. 5a, b). We observe that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{C}$$\end{document}C¯ is vastly superior for our algorithm compared with Pure random, illustrating the magnitude of the parameter space.Fig. 5Algorithm’s performance.a–d Average number of current maps displaying double quantum dot features, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{C}$$\end{document}C¯, and P(peaks) as a function of laboratory time. Current maps are labelled by humans a posteriori, i.e., after the algorithm is stopped. a, c, b, d correspond to one run of Pure random and five runs of our algorithm, respectively. All algorithm runs displayed in main panels were performed in Device 2, while insets show runs of our algorithm in Device 1. e, f High resolution current maps measured in Device 2 by Pure random and one of our algorithm runs, respectively. We indicate the time the algorithm had been running for before they were acquired and the number of labellers, C, that identified them as displaying double quantum dot features. Current maps are ordered from left to right in decreasing order of C, and maps that have the same values of C are displayed in the order at which they were sampled. Each panel uses an independent colour scale running from red (highest current measured) to blue (lowest current).The labellers considered a total of 2048 current maps produced in different runs, including those of the ablation study in Section Ablation study. The labellers had no information of the run in which each current map was produced, the device or the algorithm used. For the Pure random approach, the labelled set was composed of 51 current maps produced by the algorithm and 100 randomly selected from the set of 2048.The time μt is estimated by the multi-labeller statistics. The multi-labeller statistics uses an average likelihood of μt over multiple labellers and produces an aggregated posterior distribution (see Supplementary Methods, Bayesian statistics). From this distribution, the median and 80% (equal-tailed) credible interval of μt is 2.8 h and (1.9, 7.3) h for Device 1 and 1.1 h and (0.9, 1.6) h for Device 2. Experienced humans require approximately 3 h to tune a device of similar characteristics into exhibiting double quantum dot features (F. Kuemmeth, personal communication). Our algorithm’s performance might therefore be considered super human. Due to device variability, the hypersurfaces of these two devices are significantly different, showing our algorithm is capable of coping with those differences.In Fig. 5c, d, we compare the probability of measuring Coulomb peaks in the vicinity of a given v(u), P(peaks), for Pure random and different runs of our algorithm. We calculate P(peaks) as the number of sampled locations in the vicinity of which Coulomb peaks were detected over n. In this way, we confirm that P(peaks) is significantly increased by our algorithm. It has a rapid growth followed by saturation. Fig. 5e, f shows the high resolution current maps produced for Device 2 by Pure random and one of our algorithm runs. We observe that our algorithm produces high resolution current maps which are recognized by all labellers as displaying double quantum dot features within 1.53 h. The three current maps in Fig. 5f correspond to double quantum dot regimes found by our algorithm in different regions of the gate voltage space. The number of labellers C who identify the current maps produced by Pure random as corresponding to double quantum dots, C, is 0 or 1. This demonstrates our algorithm finds double quantum dot regimes, which can be later fine tuned to reach optimal operation conditions fully automatically21.To significantly reduce tuning times, we then modified our algorithm to group gate electrodes that perform similar functions. The algorithm assigns equal gate voltages to gate electrodes in the same group. For Device 1, we organized the eight gate electrodes into four groups: G1 = (V1), G2 = (V2, V8), G3 = (V3, V7), and G4 = (V4–V6). In this case, the median and 80% credible interval of μt improve to 0.6 h and (0.4, 1.1) h (see Supplementary Fig. 2 for a plot of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{C}$$\end{document}C¯). This approach, by exploiting knowledge of the device architecture, reduces μt by more than four times.Ablation studyOur algorithm combines a sampling stage, which integrates the hypersurface sampling with weighting and pruning, and an investigation stage that includes peak detection and score function decisions. Each of these modules, illustrated in Fig. 3, contributes to the algorithm’s performance. An ablation study identifies the relative contributions of each module, justifying the algorithm’s architecture. For this ablation study we chose to compare our algorithm, Full decision, with three reduced versions that combine different modules; Pure random, uniform surface, and peak weighting (see Table 1).Table 1Comparison of algorithms used in the ablation study.AlgorithmSS: Hypersurface samplingSS: Weighting and pruningIS: Peak detectionIS: Score function decisionPure random××✓×Uniform surface✓×✓×Peak weighting✓✓✓×Full decision✓✓✓✓Modules in the sampling stage (SS) and the investigation stage (IS) are indicated with a tick if included and with a cross if excluded.Pure random, defined in the previous section, produces a uniform distribution of candidate locations over the whole gate voltage space. It excludes the sampling and pruning rules. Uniform surface makes use of the hypersurface sampling, but no weighting or pruning rules are considered. Peak weighting combines the hypersurface sampling with weighting and pruning rules. These three algorithms use peak detection in their investigation stage, but none of them use the score function decision. For the ablation study, we define low (high) resolution as 20 × 20 (60 × 60) pixels.To analyse the algorithm’s performance, we estimate P(peaks) and the probability of success, i.e., the probability to acquire a high-resolution current map labelled as containing double quantum dot features, given Coulomb peak measurements P(success∣peaks). To take measurement times into consideration, we define t500 as the time to sample and investigate 500 locations in gate voltage space. The ablation study was performed in Device 1 keeping investigation stage parameters fixed. The cool down cycle was the same as in Section “Device tuning” (cool down 1), except for Pure random, which was performed in a new thermal cycle (cool down 2). Results are displayed in Fig. 6.Fig. 6Ablation study.a, b Bar charts and corresponding data points comparing μt (light green), t500 (dark green), P(peaks) (dark blue) and P(success∣peaks) (light blue) for the different algorithms considered. Error bars represent 80% (equal-tailed) credible intervals. Due to a measurement problem, 459 sampling iterations instead of 500 were considered for the Full decision algorithm. c–f High resolution current maps sampled by pure random, uniform surface, peak weighting, and full decision, respectively. In each panel, we indicate C, the number of human labellers that identified a map as displaying double quantum dot features. Current maps with identical values of C are displayed in the order in which they were sampled, from top to bottom. Current maps with C: 0 were randomly selected. Each panel uses an independent colour scale running from red (highest current measured) to blue (lowest current). Figure 6a shows that the introduction of the hypersurface sampling, and weighting and pruning, increases t500. This is because P(peaks) increases with these modules (Fig. 6b), and thus the number of low and high resolution current maps required by the investigation stage is larger. Within uncertainty, P(success∣peaks) remains mostly constant for the different algorithms considered. The result is a decreasing μt from Pure random to Peak weighting within experimental uncertainty. See “Methods”, “Mathematical analysis of ablation study results, for a mathematical analysis of these results”.The reason behind the use of peak detection in all the algorithms considered for this ablation study is the vast amount of measurement time that would have been required otherwise. Without peak detection, the posterior median estimate of μt for Pure random is 680 h.To complete the ablation study, we compare the considered algorithms with the grouped gates approach described in the previous section, keeping parameters such as the current map resolutions are equal. We found μt = 80.5 min (see Supplementary Fig. 3 for a plot comparing these algorithms).In summary, comparing Pure random and Uniform surface, we show the importance of hypersurface sampling. The difference between Uniform surface and Peak weighting highlights the importance of weighting and pruning. The improved performance of Full decision with respect to Peak weighting evidences the tuning speedup achieved by the introduction of the score function. These results demonstrate Full decision exhibits the shortest μt and imply an improvement over Pure random without peak detection of approximately 180 times.Device variabilityThe variability of electrostatically defined quantum devices has not been quantitatively studied so far. We have been able to exploit our algorithms for this purpose. Using the uniform surface algorithm only (no investigation stage), we obtain a set of locations on the hypersurface va. Changes occurring to this hypersurface are detected by running the algorithm again and comparing the new set of locations, vb, with va. This comparison can be done by a point set registration method, which allows us to find a transformation between point sets, i.e., between the hypersurface locations.Affine transformations have proven adequate to find useful combinations of gate voltages for device tuning9,10. To find a measure of device variability, understood as changes occurring to a device’s hypersurface, we thus use an affine transformation vt = Bvb, with B a matrix which is a function of the transformation’s parameters. We are looking for a transformation of coordinates that converts vb into a set of locations vt which is as similar to va as possible.The particular point set registration method we used is coherent point drift registration22. This method works with an affine transformation which includes a translation vector. We have modified the method to set this translation vector to zero, as the transformation between hypersurfaces can be fully characterized by the matrix B (see Supplementary Methods, Point set registration).We have used this approach to quantify the variability between Devices 1 and 2, and the effect of a thermal cycle in the hypersurface of Device 2. Figure 7 displays the matrix Bc = B − I for each case, quantifying how much B, the transformation that converts a set of locations from one hypersurface onto the other, differs from the identity matrix (I). Nonzero elements of Bc thus indicate device variability. Diagonal elements of Bc are responsible for scale transformations and can be interpreted as a capacitance change for a given gate electrode. Off-diagonal elements are responsible for shear transformations and can be interpreted as a change in cross-capacitance between a pair of gate electrodes.Fig. 7Learning about device variability.Bc matrices obtained using point set registration. Indices are the gate voltage locations vb and vt. V6 = 0 mV was fixed in Device 2 to prevent leakage currents. a Transformation between the hypersurface of Device 2 before and after a thermal cycle. b Transformation between the hyperfsurfaces of Device 1 and Device 2. Figure 7a shows Bc corresponding to the changes in the hypersurface of Device 2 after a thermal cycle (cool down 1 vs. cool down 3). This transformation shows that device variability in a thermal cycle is dominated by a uniform change in capacitance for all gate electrodes. We have also measured Bc for a thermal cycle of Device 1 (see Supplementary Fig. 4). Figure 7b displays Bc comparing the hypersurface of Device 1 (cool down 1) with the hypersurface of Device 2 (cool down 3). We observe that the variability between these devices, which share a similar gate architecture, is given by nonuniform changes in gate electrode capacitance, as well as by changes in cross-capacitance. This variability is attributed to charge traps and other device defects, such as a small differences in the patterning of gate electrodes.DiscussionWe demonstrated an algorithm capable of tuning a quantum device with multiple gate electrodes in approximately 70 min. This was achieved by efficiently navigating a multidimensional parameter space without manual input or previous knowledge about the device architecture. This tuning time was reproduced in different runs of the algorithm, and in a different device with a similar gate architecture. Our tuning algorithm is able to tune devices with different number of gate electrodes with no modifications. We showed that gate electrodes with similar functions can be grouped to reduce the dimensionality of the gate voltage space and reduce tuning times to 36 min. Tuning times might be further improved with efficient measurement techniques23, as measurement and gate voltage ramping times were found to be the limiting factor. The use of charge sensors and RF readout could also be implemented to improve tuning times, although these techniques would require to be automatically tuned to their optimum operating configuration, and would be restricted to small regions of the gate voltage space. We analysed our algorithm design through an ablation study, which allowed us to justify and highlight the importance of each of its modules. The improvement over the pure random search without peak detection is estimated to be 179 times.We showed that device variability can be quantified using point set registration by uniform sampling of the hypersurface separating regions of high and low current in gate voltage space. We found that variability between devices with similar gate architectures is given by nonuniform changes in gate capacitances and cross-capacitances. Variability across thermal cycles is only given by a uniform change in gate capacitances.Other device architectures might use the sampling stage of our algorithm as a first tuning step, and the investigation stage can be adapted to tune quantum devices into more diverse configurations. To achieve full automated tuning of a singlet–triplet qubit, it will be necessary to go beyond this work by tuning the quantum dot tunnel barriers, identifying spin-selective transitions, and configuring the device for single-shot readout.MethodsThe score function as a classifierOne of key strength of the proposed algorithm is that it does not require an ideal score function. It is important to highlight that we are using the score function just as a classifier, instead of aiming at finding the gate voltage configuration that maximises the score. The reason for this is threefold; (i) the score function is not always a smooth function; (ii) it does not always capture the quality of the transport features; (iii) it is just designed for a particular transport regime, in this case, honeycomb patterns. Therefore, the score threshold acts as a parameter that just controls the characteristics of the classifier. If the threshold is low, many high resolution scans not leading to double quantum dot transport features are produced. If the threshold is too high, then promising gate voltage windows are missed. The optimal threshold can be estimated by minimising the time required to produce a high-resolution current map that is labelled by humans as containing double quantum dot features.Mathematical analysis of ablation study resultsThe results in the ablation study can be verified under a few assumptions by a mathematical derivation of μt (see Supplementary Methods, Mathematical derivation of μt). From this derivation, we can compare the expected times \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{t}^{{\rm{abl}}}$$\end{document}μtabl for Pure random, Uniform surface, and Peak weighting:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{t}^{{\rm{abl}}}=\frac{{\mu }_{i}^{{\rm{abl}}}}{P({\rm{success}})},$$\end{document}μtabl=μiablP(success),where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{i}^{{\rm{abl}}}$$\end{document}μiabl is the expected time per each iteration of the algorithm, and P(success) =P(peaks)P(success∣peaks) is the probability that double quantum dot transport features are observed in a high resolution scan at a given iteration of the algorithm. For each iteration, time is required for a low resolution scan t2D−L, a high resolution scan t2D−H, and for the rest of the investigation and sampling tothers, including ramping gate voltages, peak detection, and computation time. The simulation of the Brownian particles is conducted in parallel with the investigation stage of the location proposed by the sampler in a previous run, and it does not increase tothers. As a result, the expected time for each iteration is2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{i}^{{\rm{abl}}}={t}_{{\rm{others}}}+P\left({\rm{peaks}}\right){t}_{{\rm{2D}}},$$\end{document}μiabl=tothers+Ppeakst2D,where t2D = t2D−L + t2D−H. Note that 2D scans are acquired with probability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P\left({\rm{peaks}}\right)$$\end{document}Ppeaks. If the score function decision is not included, high resolution current maps are always acquired when Coulomb peaks are detected. In this case, low resolution current maps are not useful, but we have still included t2D−L in t2D to keep the comparison between algorithms consistent.For all methods in Table 1 except Pure random, the time for 2D scans is the same, t2D−L ≈ 33 s and t2D−H ≈ 273 s, and tothers ≈ 35 s. Therefore, the difference on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{t}^{{\rm{abl}}}$$\end{document}μtabl across methods is given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P\left({\rm{peaks}}\right)$$\end{document}Ppeaks and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P\left({\rm{success}}| {\rm{peaks}}\right)$$\end{document}Psuccess∣peaks. In Fig. 6b, we can see that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P\left({\rm{success}}| {\rm{peaks}}\right)$$\end{document}Psuccess∣peaks is similar across the different algorithms, but \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P\left({\rm{peaks}}\right)$$\end{document}Ppeaks is different. In conclusion, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P\left({\rm{peaks}}\right)$$\end{document}Ppeaks in Eqs. ((1)) and ((2)) determines t500 and μt in Fig. 6a.Rearranging \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{t}^{{\rm{abl}}}$$\end{document}μtabl yields\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{t}^{{\rm{a}}bl}=\left(\frac{{t}_{{\rm{others}}}}{P({\rm{peaks}})}+{t}_{{\rm{2}}D}\right)\frac{1}{P({\rm{success}}| {\rm{peaks}})},$$\end{document}μtabl=tothersP(peaks)+t2D1P(success∣peaks),and this implies that t2D has a significant weight when P(peaks) is large, motivating the introduction of the score function.The expected time for Full decision algorithm is\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{l}{\mu }_{i}^{{\rm{full}}}={t}_{{\rm{others}}}+P\left({\rm{peaks}}\right){t}_{2{\rm{D}}-{\rm{L}}}+P\left({\rm{highres}}\right){t}_{2{\rm{D}}-{\rm{H}}}\\ {\mu }_{t}^{{\rm{full}}}=\frac{{\mu }_{i}^{{\rm{full}}}}{P({\rm{success}})},\end{array}$$\end{document}μifull=tothers+Ppeakst2D−L+Phighrest2D−Hμtfull=μifullP(success),where P(highres) is the probability of acquiring a high resolution current map given a score. The score function decision always makes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{i}^{{\rm{full}}}$$\end{document}μifull smaller than \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{i}^{{\rm{abl}}}$$\end{document}μiabl, because \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{i}^{{\rm{a}}bl}-{\mu }_{i}^{{\rm{full}}}=P({\rm{peaks}})(1-P({\rm{highres}}| {\rm{peaks}})){t}_{2{\rm{D}}-{\rm{H}}}$$\end{document}μiabl−μifull=P(peaks)(1−P(highres∣peaks))t2D−H and P(highres∣peaks) < 1. This is experimentally verified in Fig. 6 from the fact that t500 of Full decision is smaller than that of Peak weighting.Comparisons between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{t}^{{\rm{abl}}}$$\end{document}μtabl and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{t}^{{\rm{full}}}$$\end{document}μtfull can be affected by the dependence of P(success∣peaks) on the score function threshold. In Fig. 6b, however, we observe that P(success∣peaks) is similar for Peak weighting and Full decision. This implies that the introduction of a score function threshold does not reduce the probability of success.In this case,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{t}^{{\rm{abl}}}-{\mu }_{t}^{{\rm{full}}}=\frac{1-P({\rm{highres}}| {\rm{peaks}})}{P({\rm{success}}| {\rm{peaks}})}{t}_{2{\rm{D}}-{\rm{H}}}.$$\end{document}μtabl−μtfull=1−P(highres∣peaks)P(success∣peaks)t2D−H.This equation confirms that that the score function reduces μt in the case that the score function threshold does not degrade P(success∣peaks). Further analysis on the optimal threshold, i.e, the threshold that minimizes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{t}^{{\rm{full}}}$$\end{document}μtfull, can be found in Supplementary Methods, Optimal threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha ^{\prime}$$\end{document}α′.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2 | nature communications | [
"Article"
] | [
"Quantum dots",
"Computational science",
"Electronic devices"
] | IntroductionGate defined quantum dots for scalable quantum computation simulation1,2 controlled electrically more compact than superconducting qubit implementations1 devices operate as transistors electrons controlled by gate voltages If voltages set correctly quantum dots created single-electron control If two quantum dots created double quantum dot robust spin qubits from singlet triplet states device variability charge traps defects gate voltage settings double quantum dot varies to cycle variability scalable quantum circuits for computing devices require several gate electrodes high-dimensional parameter space difficult for Tuning time-consuming reaching limits of ability in quantum devices gate voltages render device operational as coarse tuning5,6 statistical algorithm multidimensional gate voltage space for electrostatically defined double quantum dots automatically tuning studying variability coarse tuning required manual restricted to small gate voltage subspace8 automated algorithm devices with up to eight gate electrodes challenging endeavour desired transport features present in small regions of gate-voltage spacegate voltage settings device pinched charge carriers depleted no current or open tunnel barriers weakly defined for single-electron charge transport transport features double quantum dot time-consuming difficult parametrise Machine learning automated approaches used tuning quantum limited small regions require information characteristics our work improves algorithm models parameter space tunes device automatically 70 min faster typical tuning algorithm explores gate-voltage space current design makes few assumptions other device quantum dot devices defined two-dimensional electron gas GaAs/AlGaAs heterostructure by Ti/Au gate electrodes DC voltages V1–V8 create lateral confinement potential for electrons important plunger gate voltages V3 V7 tune electron occupation left right dots bias voltage applied to ohmic contacts current device schematic precise control confinement Fig. 1a Measurements at 50 mK. 1Overview device gate voltage space Schematic gate-defined double quantum dot device Boundary hypersurface measured function of V2 V5 V8 fixed values V1 V3 V4 V6 V7 current threshold 20% of maximum measured currentgate voltage parameter space to 3D contains regions double single quantum dot transport features regions appear darker complex boundaries gate voltage locations marked green crosses current function of V7 and V3 displayed top bottom current maps display double single quantum dot transport features space defined by eight gate voltages between 0 and −2 V avoid leakage currents algorithm desirable transport features within tens of mV Identifying features slow requires two-dimensional current map plot gate voltages other techniques measuring double quantum dot charge sensing dispersive readout require parameters voltages vary not suitable for automated measurements algorithm designed to minimize current maps transport features two observations negative gate voltages no current through device pinched-off positive gate voltages full current single electron transport achieved transport features near hypersurface low high current regions single-electron transport confinement potential needed transport features single-electron transport are Coulomb peaks current flowing function single plunger gate voltage observations lead two modelling assumptions single double quantum dot transport features embedded near hypersurface Fig.separates regions current flows from current vanishes large regions hypersurface display transport features algorithm two parts sampling stage generates candidate locations hypersurface investigation stage data vicinity candidate location close to voltage space Section “Investigating nearby voltage space evaluate transport features (Fig. 2) results feed back into sampler chooses new candidate location purpose produce locations in gate voltage space device operates as double quantum dot block diagram algorithm displayed in Fig. 3. modelling assumptions based on physics of gate defined devices minimal constraints assume particular shape for hypersurface allow measurements to define by fitting data with Gaussian process algorithm minimises tuning times identifying locations promising avoiding acquisition two-dimensional current maps double quantum dot regime.Fig. 2Overview algorithm sampling produces candidate locations in gate voltage space on boundary hypersurface distance between candidate location and origin gate voltage space marked with dashed line investigation stage evaluates local region measuring current maps evaluated by score function current map colour scale from red to blue lowest Evaluation results fed back to sampling stage.Fig.3Flow diagram algorithm(See text Fig. 4 for full description Each step annotated with panel Fig. 4. Steps interaction with device shaded brown computational steps grey ablation studies Section “Ablation modules algorithm studied contribution to performance marked by blue background regions steps initialisation sampling investigation steps indicated right demonstrate over several runs two devices multiple thermal cycles algorithm finds transport features double quantum dots ablation study identifies relative contribution modules justifying design algorithm device variability theoretically explored comparing hypersurfaces for different devices single device in different thermal cycles.Automating experimental science discovery work simple physical principles flexible probabilistic machine learning models characterise tune device envisage near future application machine learning impact areas small data available no clear fitness functions defined.ResultsDescription starts initialization stage setting Vbias current measured at two extremes gate voltage space Vj = 0 Vj = −2 V for j = 1, N N number of gate electrodes negative extreme measured current should 0 current offsets might change for measurement setups difference between currents extremes is full-scale current threshold hypersurfacesearch range chosen typical gate voltage range tuning devices algorithm begins iterative process alternates between sampling investigation stages sampling stage identifies candidate location on hypersurface high probability transport features investigation stage explores nearby region current maps show Coulomb peaks honeycomb patterns presence of Coulomb peaks reported to sampling stage evaluation result uses future selection new candidates steps each iteration described in detail.Searching for hypersurfaceIn algorithm locates hypersurface selects search direction specified by unit vector u 30 selected randomly from hypersphere gate voltages negative gate voltages scanned along ray at origin o parallel to u (Fig. 4a). current monitored below 20% of full scale location v(u on hypersurface.Fig. 4Characterising boundary hypersurface using machine learning illustrates step algorithm Fig. 3. gate voltage space divided into regions near-zero) and non-zero current separated by boundary hypersurface Locating hypersurface gate voltages scanned along ray at origin defined by direction u intersection with hypersurface measureddetermine region pruned algorithm scans each gate voltage toward bottom range from location inside hypersurface If one scan intersects hypersurface future exploration inhibited by displacing origin short 1D scan location classified current peaks indicating Coulomb blockade If peaks found 2D scan (violet square performed in plane of V3 and V7 possibly repeated at higher resolution. From first thirty measurements (green yellow algorithm builds model hypersurface assigns probability\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}-69pt peaks found. refine model algorithm generates candidate search locations (squares), each weighted by corresponding value of\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}-69pt selects one at randomnew scan performed new measurement hypersurface location Steps d–f repeated indefinitely Scheme sampling hypersurface using simulated Brownian motion Each point represents particle inside volume collisions between particles hypersurface generate candidate search locations procedure locations current device pinched off recognise “tunable” every gate voltage affects current some most gate voltages little effect measured current not determined by potential quantum dot double quantum dot formed reduce time exploring implemented heuristic pruning process (Fig. first 30 iterations From hypersurface intersection v(u), voltages stepped upwards to vδ(u) ≡v(u) + δ δ step-back vector +100 mV Each voltage swept downwards towards bottom range until hypersurface encountered If hypersurface encountered voltage axis k origin iterations moved k-component equal to vδ (Fig. 4b iterations process prunes search paths hypersurface not intersected.Investigating nearby voltage located hypersurface algorithm nearby region voltage space determine double quantum dot investigation in plane containing v(u) varying two plunger gate voltages V3 and V7.gates selected before algorithm shift electrochemical potential in left right dots If double quantum dot formed current honeycomb pattern similar Fig. one-dimensional scan investigation plane starting at v(u) along diagonal axis\documentclass[12pt{minimal{amsmath{wasysym-69pt{document}\hat{V}}_{e}\equiv \frac{1}\sqrt{2}}(\hat{V}}_{3}+{V}}_{7}\end{document}V^e≡12(V^3+V^7)\documentclass[12pt{minimal}{amsmath{upgreek-69pt}\hat{{V}_\end{document}Vi^ indicates unit vector in voltage space (Fig. 4c). scan length 128 mV resolution 1 mV peak detection routine identifies presence absence of Coulomb peaks If absent investigation ceases new iteration begins if Coulomb peaks present two-dimensional scan (Fig.4d). scanning region square oriented along\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin{-69pt}\begin{document}\hat{V}}_{e}\end{document}V^e orthogonal axis[12pt{minimal}{amsmath{wasysym{upgreek\oddsidemargin{-69pt}\begin{document}\hat{V}}_{a}\equiv \frac{1}{\sqrt{2}}(\hat{V}}_{3}-{V}}_{7}\end{document}V^a≡12(V^3−V^7) square bounded by v(u), side length 3.5 times average peak spacing diagonal scan scan less than 3 peaks side length 100 mV. scan first at low resolution (16 × 16 pixels), score assigned to resulting current map score function mathematical expression transport features visual features identify honeycomb patterns Fig. 2 double quantum dot rewards maps sharp curved linesscore function low-resolution scan high repeated at high resolution (48 × 48 pixels). score threshold dynamically adjusted 15% low-resolution scans repeated Supplementary Methods Optimal threshold\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin-69pt}\alpha statistical analysis optimal threshold high-resolution maps scanned regions voltage space desirable double-dot behaviour output tuning algorithm.Searching efficiently learning hypersurfaceTo increase time exploring gate space Coulomb peaks algorithm improves search process hypersurface incorporating information measurements applies information 31st iteration starts iteration measured locations hypersurface generate model hypersurface spanning entire voltage space (Fig. 4e). model generated Gaussian process19 incorporating uncertainty measured locations Supplementary Methods Gaussian process modelseach candidate search direction u model assigns estimated distance to hypersurface m(u) with uncertainty s(u). model uses information current peaks previous searches assign each point hypersurface probability\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs}{upgreek}\oddsidemargin-69pt}{document}{peak{document expecting peaks model algorithm select new search directions efficiently desirable select search directions high probability\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts{amssymb{mathrsfs}{upgreek}\oddsidemargin-69pt}{document{peak{document}P~peak occasionally exploring less promising regions hypersurface algorithm generates candidate search locations hypersurface (Fig. 4f). set uniform selection routine based simulated Brownian motion20 inside hypersurface each encounter contributes one candidate location (Fig4f locations algorithm assigns weight proportional to value\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin{-69pt}{document}\tilde{P}}{peak\end{document}P~peak selects one location at random Thompson sampling). location defines new search direction model hypersurface efficiency search origin new search scan begins at location g(u) ≡ o + (m(u) − 2s(u))u inside hypersurface Fig. 4f). If m(u) − 2s(u) < 0 search scan begins at o.Occasionally measured current below threshold g(u) in pinched-off region algorithm scans opposite direction along—u Once current increases above 0.8 at o reverts to u direction hypersurface iterations algorithm builds high-resolution current maps improved efficiency performance algorithm assessed by statistical analysis expected success time μt acquire high-resolution current map confirmed double quantum dot features confirmation only needed to assess performancehuman labelling subjective three researchers labelled current maps deciding features double quantum dot regime no other information See Supplementary Methods, Bayesian statistics for details multilabeller statistical analysis.Device tuningTo tuning speed algorithm ran on two devices identical gate architecture Devices 1 and 2 compared performance with Pure random algorithm Pure random algorithm searches gate voltage space uniform distribution candidate locations Unlike algorithm decision hypersurface weighting pruning rules uses peak detection Full decision runs Device 1 2 performed during single cool down 1) Pure random runs performed different cool down 2) consider gate voltage space defined by number working gate electrodes provide gate voltage range avoids leakage currents Device 1 considered eight-dimensional parameter space Device 2 excluded gate electrode 6 V6 = 0 mV due to observed leakage currents define average count number of current maps labelled displaying double quantum dot features divided by number of labellersrun Pure random algorithm Device 2 five runs our algorithm Devices 1 2 calculated\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts}{amssymb{amsbsy{mathrsfs{upgreek}\oddsidemargin-69pt} function laboratory time (Fig. 5a, b).[12pt{amsmath{wasysym{amsfonts{amsbsy{mathrsfs{upgreek\oddsidemargin}{-69pt} superior our algorithm Pure random magnitude parameter space.Fig. 5Algorithm’s performance.a–d Average number current maps displaying double quantum dot features\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy{mathrsfs}{upgreek}\oddsidemargin}-69pt} P(peaks) function laboratory time Current maps labelled humans a posteriori after algorithm stoppeda c b d correspond one Pure random five runs our algorithm algorithm runs main panels performed Device 2 insets show runs our Device 1. e f High resolution current maps measured Device 2 by Pure random algorithm indicate time algorithm before acquired number labellers C displaying double quantum dot features maps ordered left to right decreasing order C same values displayed order sampled Each panel independent colour scale red to blue (lowest labellers considered 2048 current maps different runs including ablation study no information run device algorithm used Pure random approach labelled set 51 maps 100 randomly selected from 2048 time μt estimated by multi-labeller statistics average likelihood labellers posterior distribution median 80% credible interval μt 2.8 h (1.9, 7.3) h for Device 1 1.1 h (0.9, 1.6) h Device 2. humans require 3 h to tune device exhibiting double quantum dot features Our algorithm’s performance super human device variability hypersurfaces two devices different algorithm differences Fig. 5c, d probability measuring Coulomb peaks v(u), P for Pure random different runs our algorithmcalculate P(peaks) as sampled locations Coulomb peaks detected over n P(peaks increased by algorithm rapid growth followed by saturation Fig. 5e high resolution current maps for Device 2 by Pure random algorithm algorithm produces maps recognized by labellers double quantum dot features within 1.53 h three current maps in Fig. 5f correspond to double quantum dot regimes in different regions gate voltage space number of labellers identify maps double quantum dots is 0 or 1. algorithm finds double quantum dot regimes fine tuned to optimal operation conditions reduce tuning times modified algorithm to group gate electrodes similar functions assigns equal gate voltages to same group Device 1 organized eight gate electrodes into four groups G1 G2 G3 V7) G4 (V4–V6) median 80% credible interval of μt improve to 0.6 h and (0.4, 1.1) h Supplementary Fig. 2 plot approach reduces μt by four timesAblation algorithm combines sampling stage hypersurface weighting pruning investigation stage peak detection score function decisions Fig. 3 contributes to performance ablation study contributions algorithm Full decision with three reduced versions Pure random uniform surface peak weighting Table 1) 1Comparison algorithms ablation study Hypersurface samplingSS Weighting pruningIS Peak Score function decisionPure investigation stage indicated with tick included cross if excluded.Pure random produces uniform distribution candidate locations gate voltage space excludes sampling pruning rules Uniform surface hypersurface sampling no weighting pruning rules Peak weighting combines hypersurface sampling weighting pruning rules three algorithms use peak detection investigation stage none use score function decision low (high) resolution as 20 × 20 (60 × 60) pixels performance estimate P(peaks) probability of success high-resolution current map double quantum dot features t500 as time to sample investigate 500 locations in gate voltage space ablation study performed in Device 1 investigation stage parameters fixedcool down cycle same Section “Device tuning” 1) except Pure random new thermal cycle 2) Results Fig. 6Ablation study Bar charts data points comparing μt t500 P(peaks) P(success∣peaks) algorithms Error bars represent 80% credible intervals measurement problem sampling iterations instead of 500 for Full decision algorithm High resolution current maps sampled by pure random uniform surface peak weighting full decision C human labellers map double quantum dot features maps with identical values C displayed order sampled top to bottom maps with C: 0 randomly selected Each panel independent colour scale red to blue introduction hypersurface sampling weighting pruning increases t500 P(peaks) increases number low high resolution current maps required larger P(success∣peaks) remains constant for algorithms result decreasing μt from Pure random to Peak weighting within experimental uncertainty “Mathematical analysis of ablation study results peak detection algorithms measurement time required Without peak detection median estimate μt for Pure random 680 h.ablation study compare algorithms with grouped gates approach current map resolutions equal found μt = 80.5 min (see Supplementary Fig. 3 plot comparing Pure random Uniform surface importance of hypersurface sampling difference between Uniform surface Peak weighting highlights importance of weighting pruning improved performance of Full decision Peak weighting tuning speedup score function results Full decision shortest μt improvement over Pure random without peak detection approximately 180 times.Device variabilityThe variability of electrostatically defined quantum devices not quantitatively studied algorithms uniform surface algorithm obtain locations on hypersurface va Changes detected by comparing locations vb with va comparison by point set registration method transformation between locations.Affine transformations adequate find combinations gate voltages for device tuning9 device variability use affine transformation vt = Bvb B matrix function of transformation’s parameters for transformation converts vb into locations vt similar to va point registration method coherent point drift registration22 works with affine transformation includes translation vectormodified method translation vector to zero transformation between hypersurfaces characterized by matrix B Supplementary Methods Point set used approach variability between Devices 1 2 effect thermal cycle in hypersurface Device 2. Figure 7 displays matrix Bc = B − I from matrix Nonzero elements Bc indicate device variability Diagonal elements scale transformations capacitance change for gate electrode Off-diagonal elements for shear transformations change cross between gate electrodes.Fig. device variability.Bc matrices obtained point set registration Indices are gate voltage locations vb vt V6 = 0 mV fixed in Device 2 prevent leakage Transformation between hypersurface Device 2 before after thermal cycle 1 2. Figure 7a shows Bc changes hypersurface Device 2 after thermal cycle device variability thermal dominated by uniform change capacitance for all gate electrodes measured Bc for thermal cycle Device 1 Fig. 4) Figure 7b displays Bc Device 1 with 2 variability between devices by nonuniform changes in gate electrode capacitance changes cross-capacitance variability attributed to charge traps device defects differences patterningdemonstrated algorithm tuning quantum device with multiple gate electrodes in 70 min achieved multidimensional parameter space without manual input knowledge device architecture tuning time reproduced in different runs different device similar gate architecture tuning algorithm devices different gate electrodes no modifications gate electrodes similar functions reduce dimensionality voltage tuning times to 36 min Tuning times with efficient measurement voltage ramping times charge sensors RF readout improve tuning times restricted to small regions gate voltage analysed algorithm through ablation study improvement over random search without peak detection 179 times device variability quantified using point set registration uniform sampling hypersurface regions high low current gate voltage space variability between devices similar gate architectures given by nonuniform changes in gate capacitances cross-capacitances Variability across thermal cycles uniform change gate capacitances device architectures use sampling stage first tuning step investigation stage adapted to tune devices diverse configurations full automated tuning of singlet–triplet qubit necessary tuning quantum dot tunnel barriers spin-selective transitions configuring device for single-shot readout score function algorithm require ideal score functionimportant highlight using score function as classifier instead of finding gate voltage configuration score reason threefold score function not always smooth always capture quality transport features designed for particular transport regime honeycomb patterns. score threshold controls characteristics classifier threshold low high resolution scans not leading double quantum dot transport features produced too high promising gate voltage windows missed optimal threshold estimated by minimising time produce high-resolution current map labelled double quantum dot featuresMathematical analysis ablation study results ablation study verified mathematical derivation μt Supplementary Methods Mathematical derivation μt). compare expected times[12pt]{minimal}{amsmath}{wasysym{upgreek}\oddsidemargin{-69pt}{document}{\mu\rm{abl}}}\end{document}μtabl Pure random Uniform surface Peak weighting:1[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}{upgreek}\oddsidemargin}{-69pt}{document}{\mu{t\rm{abl}}}\mu{i}\rm{abl}}}}{P(\rm{success}})}\end{document}μtabl=μiablP(success),where[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$${\mu }_{i}^{{\rm{abl}}}end{document expected time each iteration algorithm P(success) =P(peaks probability double quantum dot transport features high resolution scan iteration each iteration time required low resolution scan t2D−L high resolution scan t2D−H investigation sampling tothers ramping gate voltages peak detection computation time simulation Brownian particles parallel investigation stage increase tothers expected time each iteration is2\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{upgreek\setlength\oddsidemargin{-69pt}\begin{document}\rm{abl}}}=\rm{others+P\left{peaks}}\right{2D\end{document}μiabl=tothers+Ppeakst2D t2D = t2D−L + t2D−H.2D scans acquired probability[12pt{minimal{amsmath{wasysym\oddsidemargin-69pt}{document}\left\end{document}Ppeaks score function decision not included high resolution maps acquired when Coulomb peaks detected low resolution maps not useful included t2D−L in t2D comparison methods Table 1 except random time 2D scans same t2D−L ≈ 33 s t2D−H ≈ 273 s tothers 35 sdifference on\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek}\oddsidemargin}{-69pt}\begin{document}{abl\end{document}μtabl across methods given by[12pt]{minimal}{amsmath{wasysym}{upgreek}\oddsidemargin}{-69pt}\begin{document}$$P\left(\rm{peaks}}\right\end{document}Ppeaks[12pt]{minimal}{amsmath}{wasysym{amsfonts}{amssymb}}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$P\left(\rm{success}}{peaks}}\right)\end{document}Psuccess∣peaks. Fig.6b\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin{-69pt}\begin{document}$$P\left(\rm{success}}{peaks}}\right\end{document}Psuccess∣peaks similar algorithms[12pt]{minimal{amsmath{wasysym\oddsidemargin}{-69pt}{document}$$P\left({peaks}}\right\end{document}Ppeaks different\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{amsbsy{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$P\left(\rm{peaks}}\right\end{document}Ppeaks Eqs. ((1)) ((2)) determines t500 μt Fig. 6a.Rearranging[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}\mu\rm{abl\end{document}μtabl yields[12pt]{minimal}{amsmath{wasysym{mathrsfs}{upgreek}\oddsidemargin-69pt}{document\mu{t\rm{a}}bl}\left(\frac{{t\rm{others}}}}{P(\rm{peaks}})}+\rm{2}}D}\right)\frac{1}{P({\rm{success}}\rm{peaks}})}\end{document}μtabl=tothersP(peaks)+t2D1P implies t2D significant weight when P(peaks) large introduction score functionexpected time Full decision algorithm[12pt\usepackage{amsmath\oddsidemargin-69pt\begin{document}{array\mu\rm{full}}}\rm{others}}}+P\left\rm{peaks}}\right\rm{D}}-\rm{L}}}+P\left\rm{highres}}\right\rm{D}}-{\rm{H\rm{full}}}\rm{full}}}}{P\rm{success}})\end{array}{document}μifull=tothers+Ppeakst2D−L+Phighrest2D−Hμtfull=μifullP P(highres) probability acquiring high resolution current map scorescore function decision makes[12pt]{minimal{amsmath{wasysym{-69pt}\begin{document}${\mu{i}\rm{full}}}\end{document smaller than[12pt]{minimal{amsmath\oddsidemargin}{-69pt}\begin{document}$${\mu_{i}^{{\rm{abl}}}\end{document}μiabl[12pt]{minimal}{amsmath}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}${\mu }_{i}^{{\rm{a}}bl}-{\mu{i}^{{\rm{full}}}=P({\rm{peaks}})(1-P({\rm{highres}}{peaks}})){t}_{2{\rm{D}}-{\rm{H}}}\end{document}μiabl−μifull=P(peaks)(1−P(highres∣peaks))t2D−H and P(highres∣peaks) < 1. experimentally verified in Fig. 6 t500 of Full decision smaller than Peak weighting.Comparisons between\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amsbsy{mathrsfs{upgreek}\oddsidemargin-69pt}{document[12pt]{minimal}{amsmath{wasysym{amsfonts{amsbsy{mathrsfs{upgreek}\oddsidemargin-69pt}{full affected by dependence P(success∣peaks) on score function threshold Fig. 6b P(success∣peaks) similar for Peak weighting and Full decision introduction of score function threshold reduce probability of success.[12pt{minimal}\usepackage{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}\begin{document}\mu\rm{abl}}}-\rm{full}}}={1-P(\rm{highres}}{peaks}}{success}}{t{2{\rm{D}}-{\rm{H}}}\end{document}μtabl−μtfull=1−P(highres∣peaks)P(success∣peaks)t2D−H equation confirms score function reduces μt score threshold degrade P(success∣peaks). analysis optimal thresholdthreshold minimizes[12pt]{minimal{amsmath{wasysym{amsfonts{mathrsfs{upgreek{\oddsidemargin}{-69pt}{document}\mu{full\end{document}μtfull Supplementary Methods Optimal threshold[12pt]{minimal}{amsmath}{wasysym}{amsfonts{amssymb}{amsbsy}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}{document}\alpha\prime\end{document}α′.Supplementary Review FileDescription Additional Supplementary FilesSupplementary Data 2 | 48.4 | 1.020055 |
10.1038/s41467-020-16906-1 | PMC7305221 | Understanding the sub-picosecond dynamics of driven-dissipative condensates of interacting bosons is challenging. Here the authors combine a lattice of plasmonic nanoparticles with a dye molecule solution in strong coupling and reveal distinct lasing, stimulated thermalization, and condensation regimes. | Bosonic condensates offer exciting prospects for studies of non-equilibrium quantum dynamics. Understanding the dynamics is particularly challenging in the sub-picosecond timescales typical for room temperature luminous driven-dissipative condensates. Here we combine a lattice of plasmonic nanoparticles with dye molecule solution at the strong coupling regime, and pump the molecules optically. The emitted light reveals three distinct regimes: one-dimensional lasing, incomplete stimulated thermalization, and two-dimensional multimode condensation. The condensate is achieved by matching the thermalization rate with the lattice size and occurs only for pump pulse durations below a critical value. Our results give access to control and monitoring of thermalization processes and condensate formation at sub-picosecond timescale. | IntroductionIdeal Bose–Einstein condensation (BEC) means accumulation of macroscopic population to a single ground state in an equilibrium system, with emergence of long-range order. Bosonic condensation in non- or quasi-equilibrium and driven-dissipative systems extends this concept and offers plentiful new phenomena, such as loss of algebraically decaying phase order1,2, generalized BEC into multiple states3, rich phase diagrams of lasing, condensation and superradiance phenomena4,5, and quantum simulation of the XY model that is at the heart of many optimization problems6. Such condensates may also be a powerful system to explore dynamical quantum phase transitions7. Each presently available condensate system offers different advantages and limitations concerning studies of non- and quasi-equilibrium dynamics. The ability to tune interactions precisely over a wide range is the major advantage of ultracold gases8,9. Polariton condensates10–18 offer high critical temperatures compared to ultracold gases. In photon condensates19,20, the thermal bath is easily controlled21,22. Recently periodic two-dimensional (2D) arrays of metal nanoparticles, so-called plasmonic lattices or crystals23, have emerged as a multifaceted platform for room temperature lasing and condensation at weak24–28 and strong coupling29–31 regimes.In plasmonic lattices, the lattice geometry and periodicity, the size and shape of the nanoparticles, and the overall size of the lattice can be controlled with nanometer accuracy and independent of one another. The energy where condensation or lasing occurs is given by the band edge energy that depends on the period of the array. Remarkably the band edge energy and the dispersion are extremely constant over large lattices (accuracy 0.1%28). In semiconductor polariton condensates, disorder in the samples often leads to traps and fragmentation15, or condensates may be trapped by geometry16,19. Thus plasmonic lattices offer a feature complementary to other condensate systems, namely that propagation of excitations over the lattice can be used for monitoring time-evolution of such processes as thermalization: each position in the array can be related to time via the group velocity, and there are no spurious effects due to non-uniformity of the sample. Spatially resolved luminescence was utilized in this way in the first observation of a BEC in a plasmonic lattice28.Here, we show that formation of a condensate with a pronounced thermal distribution is possible at a 200 fs timescale and attribute this strikingly fast thermalization to partially coherent dynamics due to stimulated processes and strong coupling. We observe a unique double-threshold phenomenon where one-dimensional (1D) lasing occurs for lower pump fluences and 2D multimode condensation, associated with thermalization, at higher fluences. The transition between lasing and condensation shown in our work is different from previous condensates16,28,32–36: it relies on matching the system size, propagation of excitations, and the thermalization dynamics. Importantly we find a peculiar intermediate regime showing features of a thermalization process but no macroscopic population at the lowest energy states. This regime allows us to reveal the stimulated nature of the thermalization process by the behavior of the luminescence in lattices of different sizes. As a direct evidence of the ultrafast character of the thermalization and condensation process, we show that it occurs only for pump pulse durations below a critical value of 100–250 fs. In the following, we first present characterization, such as luminescence spectra and spatial coherence, of the lasing and condensation phenomena and then focus on the main results: the stimulated nature of the thermalization process and the dramatic effect of the pump pulse duration.ResultsSystemOur system consists of cylindrical gold nanoparticles in a rectangular lattice overlaid with a solution of organic dye molecule IR-792 (see details in the Section “Methods: Samples”). The lattice supports dispersive modes, so-called surface lattice resonances (SLRs), which are hybrid modes composed of localized surface plasmon resonances at the nanoparticles and the diffracted orders of the periodic structure23,37. The electric field of the SLR modes is confined to the lattice plane in which the SLR excitations can propagate. An SLR excitation can be considered a bosonic quasiparticle that consists (mostly) of a photon and of collective electron oscillation in individual metal particles.The SLR modes are classified to transverse magnetic (TM) or transverse electric (TE) depending on the polarization and propagation direction, as defined in Fig. 1b. The measured dispersions are displayed in Fig. 1c–e. In the presence of dye molecules, the SLR dispersion shifts downwards in energy with respect to the initial diffracted order crossing. Moreover the TE modes begin to bend when approaching the molecular absorption line at 1.53 eV. These observations indicate strong coupling between the SLR and molecular excitations38; a coupled modes fit to the data gives a Rabi splitting of 164 meV (larger than the average line width of the molecule (150 meV) and SLR (10 meV)) and an exciton part of 23% at k = 0. In the following, we refer to the hybrids of the SLR mode excitations and molecular excitons as polaritons, for brevity. The coherence length of the polaritons (samples that are not pumped) is 24 μm, as obtained from the observed dispersions. Here, we have a high molecule concentration in a liquid gain solution, in contrast to previous studies26,29. The plasmonic lattice, optimized for creating the condensate, has a particle diameter of 100 nm and height of 50 nm, the period in y- and x-direction of py = 570 nm and px = 620 nm, dye concentration of 80 mM, and a lattice size of 100 × 100 μm2. The period py is varied between 520 and 590 nm, and the lattice size between 40 × 40 and 200 × 200 μm2.Fig. 1Schematic of the system and dispersion of the modes.a Artistic illustration of the experimental configuration. b Light cones for the diffracted orders (0, −1) and (0, +1) that arise for x-polarized nanoparticles. Crosscut along ky (kx = 0) is called the TE mode (red solid lines) and along kx (ky = 0) the TM mode (red dashed line). For the TE mode, the polarization is perpendicular to the propagation direction (ex, ky), and for the TM mode parallel (ex, kx). Crosscuts of the SLR dispersions are experimentally obtained by measuring the transmission (c) without the molecule or (d, e) reflection with 80 mM solution of IR-792. In d the SLR dispersion exhibits a red shift and an avoided crossing with the absorption transition of the molecule. The absorption line and the uncoupled SLR mode are depicted with white dashed lines and the lower-polariton branch given by a coupled-modes-model fit is shown with a black line. Absorption and emission spectra of IR-792 are displayed in the left panel with blue and red lines, respectively. Scanning electron micrograph of the nanoparticle array is shown as an inset in e, the scale bar is 500 nm.We excite the sample with laser pulses at 1 kHz repetition rate and central wavelength of 800 nm, and resolve the luminescence spectrally as a function of angle and spatial position on the array, Fig. 1a (for details see Section “Methods: Transmission, reflection, and luminescence measurement setup”). The pump does not directly couple to the SLR modes, and only a small fraction of the photons are coupled to the single particle resonance and/or absorbed by molecules within a near vicinity of the nanoparticle lattice. Active region of the dye molecules lies within a few hundred nanometers from the lattice plane, shown experimentally in refs. 27,39–41, and molecules further away are unlikely to couple to the SLR modes.Transition as a function of pump fluenceWe study the luminescence properties of the plasmonic lattice as a function of pump fluence, that is, the energy per unit area per excitation pulse, and find a prominent double-threshold behavior. The system is excited with an x-polarized 50 fs laser pulse that has a flat intensity profile and a size larger than the lattice. Excited polariton modes continuously leak through radiative loss, and therefore the observed luminescence intensity is directly proportional to the population of the polaritons. We record real space and momentum (k-)space intensity distributions and the corresponding spectra, the photon energy is E = hc/λ0 and the in-plane wave vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{x,y}=2\pi /{\lambda }_{0}\sin ({\theta }_{x,y})$$\end{document}kx,y=2π/λ0sin(θx,y).The sample luminescence as a function of pump fluence is presented in Fig. 2. The total luminescence intensity reveals two non-linear thresholds and a linear intermediate regime, shown in Fig. 2a. The line spectra, shown as insets, are obtained by integrating the real space spectra along the y-axis and unveil the population of polaritons as a function of energy. At the first threshold, Fig. 2b, c, lasing (or polariton lasing/polariton condensation) typical for nanoparticle arrays24,27,29,30 is observed throughout the array. Increasing the pump fluence beyond the first threshold, Fig. 2d, e, the luminescence becomes more intense in the central part compared to the top and bottom parts of the array. Moreover luminescence at the center takes place at a lower energy than at regions closer to the edges. We interpret this red shift as a signature of polariton population undergoing a thermalization process and propagating along the array in +y and −y directions, discussed below. At the second threshold, Fig. 2f, g, the system undergoes a transition into a condensate (that presents a Maxwell–Boltzmann (MB) distribution at higher energies): the real space intensity distribution shows uniform luminescence in the central part of the array, and the line spectrum (Fig. 2a, top inset) has a narrow peak at the band edge and a long thermalized tail at higher energies. A fit of the tail to the MB distribution (dashed line) gives the room temperature, T = 313 ± 2 K (for more information see Section “Methods: Fits to the MB distribution”). We use the existence of a long (ranging over several kBT, where kB is the Boltzmann constant) tail that fits the MB distribution as the criterion for distinguishing between what we call the condensate and lasing regimes. By the term condensate we thus refer here to the existence of a MB tail in addition to narrow-peaked population at low energy; peaked population alone, without the tail, is referred to as lasing. Since we are at the strong coupling regime, the lasing that we see is actually polariton lasing which in the literature is called also polariton condensation16,42; we refer to this just by the words lasing regime, for brevity. As will be shown below, also the momentum-space confinement and spatial coherence properties of our condensate and lasing regimes differ dramatically, further confirming that the two phenomena are distinct.Fig. 2Pump fluence dependence, real space images and spectra.a Double-threshold curve of the pump fluence dependence of the total luminescence intensity. Insets: Line spectra obtained by integrating over the real space spectra in the y-direction (between the white lines). The FWHM of the spectral peaks is marked in the insets: 12, 72, and 4.0 meV with increasing pump fluence. The dashed line in the top inset is a fit of the tail of the distribution to the Maxwell–Boltzmann distribution which gives a temperature of 313 ± 2 K (95% confidence bounds). b–g Left column: Real space images of the plasmonic lattice. Right column: Spectral information of the luminescence as a function of y-position. The pump fluence is b, c 0.87 mJ cm−2, d, e 1.5 mJ cm−2, and f, g 3.5 mJ cm−2. The wave fronts in d, e arise from standing waves related to the momentum of the modes, for more information see Supplementary Fig. 1 and Note 1. The real space images are recorded for single pump pulses but the corresponding spectra are integrated over 500 (c, e) and 70 pulses (g). The results for the full range of pump fluences are presented in Supplementary Movie 1.The spectrometer counts per emitted condensate pulse correspond to a photon number of ~109 (see Section “Methods: Estimation of photon number in the condensate”), which is roughly 105 times more than in the first BEC in a plasmonic lattice28. The increased luminescence is attributed to stimulated processes and differences in the sample as well as the pump and detection geometry (Supplementary Note 2). This tremendous improvement has increased the signal-to-noise ratio so that we can now observe a prominent thermal tail (it is likely to be even longer but the data is cut due to filtering out the pump pulse) which is an important signature of the efficiency of the thermalization process even in ultrafast timescales. The upgrade of luminescence intensity is also crucial for future fundamental studies and applications of this type of condensate. For instance, thermodynamic quantities can be determined using the observed photon distribution43. To produce a condensate, the periodicity must be tuned with respect to the thermalization rate and the array size (Supplementary Note 1).Three distinct regimes are also observed in the k-space intensity distributions. Figure 3 presents the k-space images and spectra for the same sample and pump parameters as in Fig. 2. Figure 3a–c shows that lasing spreads in the TM mode to large k (i.e., large emission angles), whereas thermalization of the polaritons occurs mostly along the TE mode, see Fig. 3d–f. At the condensation threshold, Fig. 3g–i, we observe confinement in both kx and ky, implying that the condensate has a 2D nature, in contrast to the lasing regime where confinement is observed only in ky. Figure 3j shows line spectra obtained by integrating along ky from TE mode crosscuts (Fig. 3c, f, i); the spectrometer slit width of 500 μm corresponds to ±1.3∘ around θx = 0 in the 2D k-space images. Intriguingly, at the condensation threshold, we see multiple modes highly occupied at ky = 0. The line spectrum at 3.49 mJ cm−2 shows three narrow peaks followed by a thermalized tail. The full-width at half-maximum (FWHM) of the highest peak is 3.3 meV, significantly narrower than the bare SLR mode (10 meV). The spacing of the multiple peaks decreases toward lower energy, ruling out the possibility of a trivial Fabry–Pérot interference. We have also observed that the spacing does not depend on the periodicity or lattice size in a straighforward manner, and is not caused by waveguide effects25,44. Based on T-matrix simulations, we find that the cylindrical shape and finite size of the nanoparticles leads to three distinct modes around the Γ-point (ky = kx = 0) energy in an infinite lattice, one of these modes being highly degenerate. This highlights the role of the nanoparticles beyond providing a periodic structure. The degeneracy is lifted by a finite lattice size, producing further distinct modes. The lattice size thus provides an additional means to tune the mode structure. For more information, see Supplementary Fig. 2 and Note 3. While at thermal equilibrium condensation occurs to the lowest energy, in driven-dissipative systems condensation to several modes at distinct energies is possible3,5. Since we observe a temporally integrated signal, we cannot rule out the possibility of a single-mode condensate temporally evolving between different states in the sub-picosecond scale.Fig. 3Momentum space images and spectra.First column: 2D momentum (k-)space images. Second and third column: k-space spectrum in the TM and TE mode directions, respectively. The spectra of the TM and TE modes correspond to horizontal and vertical slices of the 2D k-space spectrum, respectively. The pump fluence is a–c 0.85 mJ cm−2, d–f 1.5 mJ cm−2, g–i 3.5 mJ cm−2, as in Fig. 2. The images and the corresponding line spectra are integrated over 500 (c), 330 (f), and 20 pulses (i). The horizontal stretching of the emission peaks in h, i are CCD blooming artifacts. j Population distribution in the TE mode integrated from (c, f, i) along ky. FWHMs are indicated with dashed black lines: 6.3, 17, and 3.3 meV with increasing pump fluence. The results for the full range of pump fluences are presented in Supplementary Movie 2.Lasing and condensation transitions are expected to result in increased spatial and temporal coherence of the emitted light. Increase in temporal coherence was evidenced as the narrowing of spectral line width (Fig. 3j). To study the spatial coherence, we have performed a Michelson interferometer experiment as a function of pump fluence. In the Michelson interferometer, the real space image is split into two, one image is inverted and combined with the other one at the camera pixel array. The contrast of the observed interference fringes is extracted with a Fourier analysis of the spatial frequencies in the interfered image to exclude any artifacts produced by intensity variations in a single non-interfered image (Supplementary Figs. 3 and 4 and Note 4).In Fig. 4, the fringe contrast (proportional to the first-order correlation function g(1)) is shown as a function of pump fluence in both y- and x-directions of the lattice. High spatial coherence occurs in the y-direction over the array in both the lasing and condensation regimes (Fig. 4b, c). At the intermediate regime, the spatial coherence decreases, in agreement with the observation that the thermalizing population dominates the luminescence signal. In the x-direction, spatial coherence is lower than in the y-direction over the whole pump range. However, the condensate clearly exhibits high spatial coherence also in x-direction, in contrast to lasing, which shows separated emission stripes in the real space image (see Fig. 4d, Supplementary Fig. 3). The spatial coherence measurements are in line with the observations from the 2D k-space images (Fig. 3), where (1) lasing exhibits confined luminescence along ky (the direction of feedback) but spreads along kx, (2) condensation shows 2D k-space confinement. The coherence both in the lasing and condensation cases extends over the whole array (100 μm × 100 μm), thus the coherence length is at least four times larger than that of the samples without pumping (24 μm). In a future study, larger samples should be used for finding how the coherence decays (e.g., exponential and polynomial). Whether algebraically decaying phase order exists in 2D driven-dissipative systems is a subtle question1,2,45,46.Fig. 4Spatial coherence measurement along y- and x-axis of the plasmonic lattice.a The Michelson interferometer fringe contrast averaged over the region of interest (white boxes in b–e) as a function of pump fluence. For reference, the contrast is extracted also from non-interfered real space images, as light blue and yellow curves. b–e Interfered real space images, where one of the images is inverted with respect to the white dashed line, to obtain the spatial coherence between −y and +y positions (b, c) or −x and +x positions (d, e). The interference patterns are shown for pump fluences corresponding to the lasing and condensation regimes (dashed lines in a). To obtain the spatial coherence between −x and +x positions, instead of rotating the detection optics, we rotate the sample (the lattice) and pump polarization by 90∘. The vertical stripes in b and d are due to one-dimensional lasing in y-direction, while the horizontal stripes (fringes) in b, c and vertical in e arise from spatial coherence in the overlapped images in the Michelson interferometer. For more details see Methods (Spatial coherence measurements) and Supplementary Note 4. The real space images are recorded for single pump pulses.The lasing and condensation take place at energies 1.397 and 1.403 eV, respectively, lower than the band-edge energy of the system without molecules, 1.423 eV. However the energies are blue-shifted from the lower polariton energy 1.373 eV (band edge in reflection) or 1.382 eV (fitted lower polariton branch; Fig. 1d), which are experimentally obtained by a reflection measurement and coupled modes model38. Since the whole dispersion gradually blue shifts as a function of pump fluence, it may be associated to degradation of strong coupling instead of originating from Coulomb interactions. Such saturation-caused non-linearity can also be considered as effective polariton–polariton interaction15 (a rough estimate of the strength of such interaction in our case is given in Supplementary Note 8). Based on the band-edge locations (1.423−1.403) eV/(1.423−1.373) eV, the coupling in the condensation regime has decreased to ~40% of the case without pumping. Note that the observed double-threshold behavior is different from semiconductor microcavity polariton condensates where condensation has a lower threshold than photonic lasing associated with loss of strong coupling16,32.Stimulated nature of the thermalizationAt the intermediate regime, we observed red shift of the luminescence as a function of distance in y-direction, Fig. 2e. The trails of the red shift begin from the emission maximum of the dye molecule (~1.46 eV), at a certain distance from the array edges, and reach the band-edge energy (~1.40 eV) exactly at the center of the 100 × 100 μm2 array.To understand the red shift, we have recorded real space images and spatially resolved spectra for different lattice sizes at intermediate pump fluences sufficient to trigger the thermalization process, Fig. 5a–f. In a large array (Fig. 5e, f), we observe the trails of the red shift toward the center of the array, similarly as in the 100 × 100 μm2 array (Fig. 2d, e), but the red-shifting populations do not merge at the center. In a small array (Fig. 5a, b), the situation looks different at first glance since the red shift seems to occur from the center of the array toward the edges. However, careful comparison of the distance between the array edge and the location where the red shift begins (see Supplementary Fig. 10 for details) reveals that for all arrays, for the given pump fluence, the distance is the same (~25 μm for 2.2 mJ cm−2 pump fluence presented in Fig. 5a, c, e).Fig. 5Observation of stimulated emission pulse build-up in finite size lattices.a–f Real space images (left column) and spectra (middle column) at intermediate pump fluence (2.2 mJ cm−2) for lattice sizes a, b 40 × 40 μm2, c, d 60 × 60 μm2, e, f 150 × 150 μm2. Please note the different color scales in different panels. Line spectra for the different lattice sizes are presented in Supplementary Fig. 11. g Rate-equation simulation of stimulated emission. The pulse build-up time relevant for our system is marked by vertical dashed lines (the peaks of the population inversion N* and of the output pulse). Horizontal dotted line indicates the value at which N* overcomes the losses in the simulation. h Illustration of the sum (red dashed line) of spatial intensity profiles of the thermalizing pulses propagating to left (yellow) and right (blue) starting from everywhere along the y-axis of the array. The Gaussian shape approximates the increasing and decreasing intensity of the excitations as they gather gain and suffer losses under propagation. The results of such sum for different array sizes are shown as insets in a, c, e with the same false color as the real space images. i Measured distance from the array edge to the location where the red shift begins (indicated by the 25 μm scale bars in a, c, e) as a function of pump fluence for different lattice sizes. The legend indicates the square-array sizes in μm. The diagonal dashed line is the inverse of the pump fluence (50/P), and the horizontal dashed line indicates the saturation value (~18 μm). The inset shows the pulse build-up time (converted to distance by multiplying with the group velocity of the SLR mode) obtained from the rate-equation simulation.We explain this distance by stimulated emission pulse build-up time: the time between the maxima of the population inversion and the output pulse as defined in the rate-equation simulation in Fig. 5g (see39 and Supplementary Note 5 for description of the model; note that we use this model just to illustrate the concept of pulse build-up in the thermalization process, not to describe the condensate or lasing observations). Pulse build-up is a well-known phenomenon in Q-switched lasers47. In our system, the pump pulse excites the molecules, and the polaritons begin to propagate when the first photons populate the modes. The modes then gather gain while propagating, and therefore the peak of the stimulated emission pulse appears after a certain distance traveled along the array. This distance is seen as the dark zones in real space measurements, and it corresponds to the pulse build-up time. Note that this pulse is different to a lasing or condensate pulse since the line width of the luminescence is large and spatial coherence is small. By summing up such spatial intensity profiles of the thermalizing pulses at every point along the lattice (Fig. 5h), we can reconstruct the real space intensity distributions (insets of Fig. 5a, c, e). The dark zones appear because the edges do not receive propagating excitations from outside the lattice; the dark zones at the edges have approximately half the intensity of the central part. In the small lattice, intensity at the edges is similar to that of the larger lattices but in the center it is only half of that. Moreover the wavy interference patterns in the central part (Fig. 5c, e) indeed only appear for arrays larger than 40 × 40 μm2, where there are counter-propagating pulses that can interfere.We found that the width of the dark zones depends on pump fluence as predicted by the rate-equation simulation and the theory of Q-switched lasers, namely the build-up time is inversely proportional to the pump fluence. Fig. 5i shows that the dark-zone width follows the inverse of the pump fluence until it saturates at around 3 mJ cm−2 (corresponding to the BEC threshold) to a value below 20 μm (~100–140 fs). The inset in Fig. 5i shows the pulse build-up time extracted from our rate-equation simulation, and it displays a similar ~1/P dependence.We attribute the red shift to a thermalization process. At the intermediate regime of pump fluences, however, a thermal distribution is not reached before the population decays. For higher fluences, a condensate peak and a tail with MB distributed population emerges. A classical thermal MB distribution in logarithmic scale is a straight line. In contrast, a peaked feature at low energies, together with a MB tail, is a characteristic of the BE distribution. At low energies (E − μ < kBT), the BE distribution can be approximated as kBT/(E − μ). A distribution of this form appears in so-called classical condensation of waves (or Rayleigh–Jeans (RJ) condensation), resulting for instance from an interplay between random noise and suitable gain/loss profiles of an optical system48–52. Our system does not have the conditions typical for classical condensation and, most importantly, the observed linear-in-log-scale tail does not match a distribution of the form kBT/(E − μ). To rule out the RJ condensation by the distribution, one needs to observe it for energies E that are larger than the condensate peak energy by more than kBT, because for E − μ < kBT the BE and RJ distributions coincide. It is thus essential that we resolve the tail up to energies that are 75 meV above the condensate energy—three times larger than the room temperature kBT = 25 meV.As the observed distribution does not match with either classical MB or RJ distribution, one can ask how does a distribution resembling the equilibrium BE case form in a non-steady-state, driven-dissipative system. In the weak coupling regime, the answer is known. Our system is similar to the photon condensates19,33,53 in the sense that dye molecules with a vibrational level structure provide the thermalization mechanism. Differences to the photon condensates are our type of excitations (plasmonic-photonic lattice modes), ultrafast time-scales, and strong light-matter coupling. It has been shown both experimentally and theoretically that, in the weak coupling regime, recurrent absorption and emission processes of light with molecules whose vibrational manifold is coupled to an external bath lead to thermalization and condensation with a BE(-type) distribution. This occurs both for continuous wave19,33,36,53 and pulsed28,54,55 pumping. The vibrational manifold serves as the energy loss channel to move the photon population towards lower energies, and thermal population of the vibrational states provides the temperature for the BE distribution. Due to the vibrational energy loss, the molecules may emit at lower energy than they absorb: this provides an effective coupling between photons at different energies, thus the thermalization process does not need scattering originating from Coulomb interaction as in inorganic semiconductor polariton condensation. The thermalization requires that several52,56 (or just one20) absorption-emission cycles take place within the lifetime of the system. The speed of the thermalization process in general depends on the number of molecules, strength of the light-matter coupling, and the number of photons since stimulated processes may be important. It is plausible that this mechanism or its modified form provides the thermalization also in our present strongly coupled case. Although the lifetime of our system is extremely short, the emission-absorption processes are, as we show, highly stimulated during the red shift, also at the higher energies. This helps to fulfill the thermalization criterion of several emission-absorption cycles within the lifetime. The thermalization rates in the present case are higher (e.g., 0.20 eV/ps in Fig. 2e) than observed for the same molecular concentration in our previous work28 (0.08 eV/ps), further confirming that the larger number of photons and stimulated processes are contributing to the speedup.A discrete step of simultaneous creation of a low energy polariton and a vibrational quantum is often quoted as the route for organic semiconductor polariton condensation14,29,31,57,58. The microscopic foundation of this phenomenon is the same as that of photon condensation, but the parameter regimes differ. A discrete step is likely to be the adequate picture when the absorption and emission spectra of the molecules show distinct vibrational sub-peaks; in contrast, a smooth red shift process leading to a BE distribution is more likely for molecules whose vibrational states are not prominently visible in the spectra52. Our system which shows strong coupling (polaritons), and molecule spectra with no vibrational shoulders (Fig. 1d), is in the middle ground between the weak coupling photon condensation mechanism and the relaxation by a discrete step, and requires a theoretical description going beyond the approximations done in both. For a single molecule with one vibrational state coupled to a few light modes, one can theoretically describe how a process that is coherent, except for vibrational losses that provide the energy loss channel, leads to rapid red shift of emission (Supplementary Fig. 5 and Note 6). To rigorously describe our observations, one would need a model consisting of many molecules with several vibrational states coupled to a thermal reservoir, and multiple light modes at the multiphoton regime. The model should then be solved without resorting to weak coupling perturbation theory in the light-matter coupling. The current state-of-the-art theory59–63 forms a good starting point for this kind of advanced description. Such theory could predict how the condensation threshold depends on losses, thermalization speed, and competition with lasing, which are expected to play a role since the photon numbers emitted by the condensate that we observe here are several orders of magnitude larger than the equilibrium estimate for the critical number28.Effect of pulse durationThe spatial measurements have enabled an astounding, yet indirect, way to look into the dynamics of the system. To complement the spatial observations, we have probed the dynamics directly in the time domain by altering the excitation pulse duration. A 50 fs excitation pulse results in the double-threshold behavior with a distinct regime for lasing, an intermediate regime showing an incomplete stimulated thermalization process, and condensation, as explained above. Remarkably by using a longer pulse, we observe only the first (lasing) threshold, and the system does not undergo condensation even at higher pump fluences. The real space intensity distribution and spectrum remain almost unchanged from low to high pump fluence for a 500 fs pulse duration, Fig. 6. The intensity distributions and spectra resemble the lasing regime seen at low pump fluence with the 50 fs pulse (Fig. 2b, c), while the intermediate and condensation regimes are absent. Besides the luminescence intensity, the different threshold behavior is clearly visible in the FWHM curves of the spectral maximum (Supplementary Fig. 6 and Note 7). The FWHM is significantly decreased with both pulse durations at the first (lasing) threshold but only with the 50 fs pulse, the FWHM is decreased even further at the second (condensation) threshold. The k-space images and spectra for the 500 fs pulse (Supplementary Fig. 7, Movie 4) reveal that the luminescence from low to high pump fluences is spread widely in the TM mode, hence there is no 2D confinement.Fig. 6Pump fluence dependence, real space images and spectra for 500 fs pump pulse duration.a Pump fluence dependence of the total luminescence intensity. Insets: Line spectra obtained by integrating over the real space spectra in the y-direction (between the white lines). b–g Left column: Real space images of the plasmonic lattice. Right column: Spectral information of the luminescence as a function of position in the y-axis. The pump fluence is b, c 0.88 mJ cm−2, d, e 1.6 mJ cm−2, f, g 3.5 mJ cm−2. FWHM of the spectral peaks is marked in the insets. With increasing pump fluence, we obtain a FWHM of 23, 18, and 23 meV. The results for the full range of pump fluences are presented in Supplementary Movie 3.So far we have compared the results for 50 and 500 fs excitation pulses at the same pump fluences so that the total amount of energy injected to the system per pulse is the same for different pulse durations. However, triggering the stimulated thermalization process might depend on the instantaneous pump intensity rather than the fluence, which means that the condensation threshold could be reached also with longer pulses if the pump fluence was increased. To test this, we have further studied the dependence on pulse duration by several intermediate measurements (Fig. 7) that show condensation with 100 and 250 fs pulses, but not with longer pulses. The condensation threshold for the 100 fs pulse is equal to that of the 50 fs pulse (3.5 mJ cm−2), whereas for the 250 fs pulse it is higher (4.5 mJ cm−2). The resulting line spectra at the condensation threshold for the 100 and 250 fs pulse durations (blue and red solid lines in Fig. 7a) are similar to that measured for the 50 fs pulse presented in Fig. 2a: macroscopic population at the band edge followed by a linear distribution at the higher energies. Fit of the tail to the MB distribution gives 316 ± 2 and 331 ± 4 K for the 100 and 250 fs pulse, respectively (for more information see Section “Methods: Fits to the MB distribution”). With a longer pulse (350 fs), thermalization in the time-integrated signal remains incomplete (too much population in the higher energy states). With the longest excitation pulses (>350 fs), we observed no signs of thermalization or condensation even at the highest pump fluences that we could measure until damaging the samples. For all pulse lengths, the thermalization process competes over the same gain with the lasing. It seems that for a long excitation pulse the instantaneous population inversion does not reach a high enough value for the thermalization process to take over the lasing which is already triggered at the first threshold (see also Fig. 6).Fig. 7Line spectra and threshold pump fluences for different pulse durations.a Line spectra for different pulse durations. Condensation occurs for 100 and 250 fs pulses, but not for longer ones. For the 100 and 250 fs pulses, the spectra are shown at the condensation threshold (3.5 and 4.5 mJ cm−2, respectively). The threshold in case of the 100 fs pump pulse is the same as for the 50 fs pulse. We have defined the condensation threshold such that a narrow peak is observed at the band edge together with a linear distribution at high energies in the time-integrated signal. For a 350 fs pulse no condensation is observed, but we show a spectrum for which the tail of the distribution is closest to linear (at 6.0 mJ cm−2). For a 500 fs pulse the spectrum is shown at the highest measured pump fluence (6.3 mJ cm−2). b Total luminescence intensity as a function of pump fluence. Increasing the pump pulse duration increases slightly the first (lasing) threshold, whereas the second (condensation) threshold is visible only for the 100 and 250 fs pulses. For the 350 fs pulse, we increased the pump fluence up to 10 mJ cm−2, after which the nanoparticle array starts to damage, to confirm that no condensation threshold occurs even at higher fluences. Rather a saturation and degradation of the luminescence signal can be observed. The highlighted data points correspond to the line spectra in a. c Threshold pump fluence for lasing and condensation, as well as the beginning of the incomplete stimulated thermalization regime, as a function of pulse duration. In the lasing regime, there is one spectral intensity maximum corresponding to the band edge (this is the lasing peak), and another one (smaller in intensity) at higher energy. First, for increasing pump fluence, the lasing peak grows and the higher energy peak diminishes. At a certain pump fluence, however, the higher energy peak starts growing. We determine this pump fluence value from the line spectra and define it as the threshold for the incomplete thermalization regime. At this pump fluence, the interference patterns (such as in Fig. 2e) become visible in the real space spectra. The lines are guides to the eye.The observations for different pulse durations are summarized by Fig. 7c, which shows the thresholds for lasing and condensation, as well as the start of the incomplete stimulated thermalization regime, as a function of pulse duration. The dependence of the condensation and incomplete stimulated thermalization thresholds on the pulse length is stronger than in the lasing case where the threshold is quite monotonous as a function of the pulse duration. Sensitivity to the excitation pulse duration highlights the ultrafast nature of plasmonic systems and endorses the sub-picosecond dynamics of the thermalization process.Interestingly the critical pulse duration for observing condensation is similar to or smaller than the time (~250–350 fs) in which the polaritons propagate from the edges to the center in the 100 × 100 μm2 array (see Section “Methods: Samples for discussion on the group velocity”). Besides the critical pulse duration, observing the condensation requires that the thermalization time matches the propagation time so that the polaritons have red-shifted to the band-edge energy while still having large population density. This is achieved by an optimal balance between the dye concentration and pump fluence (thermalization speed), lattice size (distance that the excitations need to propagate from the array edges to the center), and lattice period (band-edge energy). The condensation is also sensitive to the incident angle of the pump. To obtain a clear tail with MB distribution in the time-integrated luminescence signal, the pump pulse needs to come at normal incidence, and even a slight misalignment changes both the real space intensity pattern and the spectral distribution. Nonzero incidence angle results in a time difference of the excitation at the two edges of the array, and can cause an asymmetry in the populations of counter-propagating polariton modes.DiscussionFundamental questions on the dynamics of BEC in driven-dissipative systems are still largely open, despite years of research1,2,15,16. What is the nature of the energy relaxation and thermalization processes, how does the condensate form, and what are its quantum statistical and long-range coherence properties? These questions have been addressed for weakly coupled BECs, but become challenging to answer for strongly coupled room temperature condensates as higher energy scales imply faster dynamics. Here, we have shown that plasmonic lattices offer an impressive level of access and control to the sub-picosecond dynamics of condensate formation via propagation of excitations and the finite system size.We experimentally demonstrated that a bosonic condensate with a clear-cut thermal excited state population can form in a timescale of a few hundreds of femtoseconds. We propose that this extraordinary speed of thermalization is possible because the process is partially coherent due to strong light–matter coupling and stimulated emission. Strong light–matter coupling at the weak excitation limit was indicated by our reflection measurements. By varying the lattice size, we revealed the stimulated nature of the thermalization process. While strong light–matter coupling at the multiphoton regime is described by the Dicke model64, to explain both the red shift and the thermal distribution we observed, one would need to involve vibrational degrees of freedom that are (strongly) coupled to the electronic ones as well as to a thermal bath. Work toward surmounting such theoretical challenges has already begun since systems where light and molecular electronic and vibrational states are strongly coupled have promise for lasing and condensation, energy transfer, and even modification of chemical reactions65,66. We have shown here that plasmonic lattices offer a powerful platform for studies of such ultrafast light–matter interaction phenomena. The shape, size, and material of the nanoparticles can be accurately tuned, as well as the lattice geometry, composition of the unit cell, and the lattice size. This provides a large, controlled parameter space vital for testing and (dis)qualifying theoretical predictions. In particular, the dynamics can be accessed in complementary ways: through conventional time-domain techniques as well as indirectly via the propagation of excitations in the lattice.MethodsSamplesThe gold nanoparticle arrays are fabricated with electron beam lithography on glass substrates where 1 nm of titanium is used as an adhesion layer (see SEM image in Supplementary Fig. 8). The nominal dimensions of the plasmonic lattice, optimized for the condensation experiment, are the following: a nanoparticle diameter of 100 nm and height of 50 nm, the period in y- and x-direction of py = 570 and px = 620 nm, respectively, and a lattice size of 100 × 100 μm2. In reference measurements, the period py is varied between 520 and 590 nm and the lattice size between 40 × 40 and 200 × 200 μm2.Asymmetric periodicity separates the diffracted orders in the energy spectrum for the two orthogonal polarizations (ex and ey), and the SLR dispersions are correspondingly separated, which simplifies the measured spectra. For x-polarized nanoparticles (as in our experiments), the TE and TM modes correspond to combinations of (ex, ky) and (ex, kx), respectively. Under pumping, which SLR mode is mainly excited is determined by the pump polarization as the molecules are excited more efficiently via the single particle resonance at the plasmonic hot-spots of each nanoparticle41. In the experiment with different periods, px was kept always 50 nm larger than py. In the experiment where lattice size was varied, however, the lattice period in x and y directions was the same (px = py = 570 nm). We found that asymmetric periodicity does not play a crucial role in forming the condensate but just simplifies the data analysis of the experimental results.Group velocity for the TE mode is obtained close to the Γ-point, in the linear part of the dispersion, for samples both without and with the dye molecules. The group velocity is 0.65c for the uncoupled TE mode (Fig. 1c) and 0.48c for the strongly coupled polariton mode (Fig. 1d; c is the speed of light). We use the group velocity to convert propagation distance to time. In the experiments we see that the strongly coupled dispersion starts to resemble the uncoupled one at high pump fluences due to the saturation effects, as explained in the manuscript. We cannot exactly specify what the group velocity is at a certain pump fluence, therefore we present the time conversions with a group velocity range from 0.48c to 0.65c. This means that the propagation of 50 μm distance along the array takes 250–350 fs.The dye molecule solution is index-matched to the glass substrate (n = 1.52), the solution being a mixture of 1:2 DMSO:Benzyl Alcohol. The solution is sealed inside a Press-to-Seal silicone isolator chamber (Sigma-Aldrich) between the glass substrate and superstrate. The dye solution has a thickness of ~1 mm, which is very large compared to the extent of the SLR electric fields27,39,41. Given by the excess of the dye molecules and the natural circulation of the fluid, there are always fresh dye molecules available for consecutive measurements of the sample when scanning the pump fluence, which makes the sample extremely robust and long-lasting. IR-792 perchlorate (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{C}}}_{42}{{\rm{H}}}_{49}{{\rm{ClN}}}_{2}{{\rm{O}}}_{4}{\rm{S}}$$\end{document}C42H49ClN2O4S) was chosen as the dye molecule because it dissolves to the used solvent in very high concentrations, in contrast to many other dye molecules, e.g., IR-140 that has also been used by us27,39 and others24,26,41 as a gain medium in plasmonic nanoparticle array lasers. We have collected the information on tested dye molecules and clarify the reasoning behind the molecule choice in Supplementary Table 1.Transmission, reflection, and luminescence measurement setupA schematic of our experimental setup is depicted in Supplementary Fig. 9. The same setup is used for transmission, reflection and luminescence measurements with minor modifications. The spectrometer resolves the wavelength spectrum of light that goes through the entrance slit, and each pixel column in the 2D CCD camera corresponds to a free space wavelength, λ0, and each pixel row to a y-position at the slit. The y-position further corresponds to either an angle (k-space) or the y-position at the sample (real space). The photon energy is E = hc/λ0 and in the case of angle-resolved spectra (dispersions) the in-plane wave vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{x,y}={k}_{0}\sin ({\theta }_{x,y})=2\pi /{\lambda }_{0}\sin ({\theta }_{x,y})$$\end{document}kx,y=k0sin(θx,y)=2π/λ0sin(θx,y), where h is the Planck constant and c the speed of light in free space. Next, the three different experiment types are explained, starting with the luminescence measurement where the sample is optically excited with an external pump laser. The excitation pulse (or pump pulse) is generated by Coherent Astrella ultrafast Ti:Sapphire amplifier. The pulse has a central wavelength of 800 nm, and at the laser output, a duration of <35 fs with a bandwidth of 30 nm. The pump pulse is guided through a beam splitter and mirrors, and finally to the mirror M1 (see Supplementary Fig. 9), which directs the pump pulse to the excitation path of the setup. We have a band-pass filter in the excitation path that is used in combination with a long-pass filter in the detection path to filter out the pump pulse in the measured luminescence spectra. The pump pulse is linearly polarized, and to filter only the horizontal component we have placed a linear polarizer after the band-pass filter. The pump fluence is controlled with a metal-coated continuously variable neutral-density filter wheel (ND wheel). The pump pulse is spatially cropped with an adjustable iris and the iris is imaged onto the sample with a help of lens L1 and the microscope objective. The inverted design enables exciting the sample at normal incidence, which is crucial for simultaneous excitation of the dye molecules over the sample. Excitation at normal incidence also prevents any asymmetry in the spatial excitation of the molecules around the nanoparticles with respect to lattice plane. The inverted pumping scheme and accurate optical alignment were essential for repeatable and precise condensate formation.In the detection path, we have the long-pass filter and optionally a linear polarizer. An iris or pinhole acts as a spatial filter at the 1st image plane to restrict the imaged area at the sample. The 1st image plane is relayed onto the real-space camera (1st Cam.). In the k-space measurements, the back-focal plane of the objective (Fourier plane; containing the angular information of the collected light) is relayed onto the 2D k-space camera (2nd Cam.) as well as onto the spectrometer slit, with the tube lens and a k-space lens. For the real space measurements, the beam-splitter before the k-space lens is replaced with an additional real-space lens to produce the 2nd image plane of the sample to 2nd Cam. and onto the spectrometer slit. The spectrometer slit selects a vertical slice either of the 2D k-space image or the real space image. In the luminescence measurements, we use a slit width of 500 μm. In the k-space, it corresponds to ±1.3∘ around θx = 0, or to ±0.16 μm−1 around kx = 0 at E = 1.4 eV. Respectively in the real space, the slit opening of 500 μm corresponds to 27 μm slice at the sample.The dispersion of optical modes in the bare plasmonic lattice can be measured in transmission mode of the setup, where the sample is illuminated with a focused and diffused white light from a halogen lamp. The lattice modes are visible as transmission minima (extinction maxima) in the angle-resolved spectrum. When a thick layer of high-concentration dye molecule solution is added in the chamber on top of the nanoparticle array, the transmission measurement is not applicable due to a complete absorption by the molecules. To access the dispersion of the lattice modes in this case, we use the setup in reflection mode by utilizing the same inverted design as in the luminescence measurement. The halogen lamp is inserted before the iris in the excitation path, that is imaged onto the sample, and the dispersion of the lattice modes is revealed by reflection (scattering) maxima in the collected angle-resolved spectrum.The luminescence measurement as a function of pump fluence is automated with LabView. Predefined fluence steps are measured such that for each step: (1) the calibrated ND filter wheel is set to a correct position, (2) the shutter is opened, (3) the image is acquired with spectrometer and optionally with 1st and 2nd Cam., and (4) the shutter is closed. Integration time of the spectrometer is automatically adjusted during the measurement to avoid saturation at highly non-linear threshold regimes. The pump pulse duration is measured with a commercial autocorrelator (APE pulseCheck 50). In the pulse duration measurement, the pump pulse goes through the same optics as in the actual experiments. The pulse duration is changed by adjusting the stretcher-compressor of the external pump laser.Fits to the MB distributionWe fit the thermalized tail in the measured population distributions to MB distribution (Fig. 2, Supplementary Fig. 6). The fit function is given by1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{{\rm{MB}}}(E)=\frac{d(E)}{{{\rm{e}}}^{(E-\mu )/({k}_{{\rm{B}}}T)}},$$\end{document}fMB(E)=d(E)e(E−μ)/(kBT),where d(E) is the degeneracy of the modes as a function of energy E, μ is the chemical potential, kB is the Boltzmann constant, and T is temperature. The fit was done for the part of the distribution that is linear in logarithmic scale, at pump fluence at/above the threshold. We call this part of the distribution (between energies 1.41 and 1.47 eV) the “tail”. Fitting was performed with a nonlinear least squares method.The degeneracy d(E) was approximated by the density of states for light traveling in a 2D plane. The light dispersion in the xy plane forms a conical surface (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E=\hslash c/n\sqrt{{k}_{x}^{2}+{k}_{y}^{2}}$$\end{document}E=ℏc/nkx2+ky2). The dispersions of the SLR modes are well approximated by this everywhere except the very near vicinity of the k = 0 point67, and our fitted range starts from a finite k where the approximation is valid. This dispersion results in a linearly increasing but nearly constant density of states for the fitted energy range of ~60 meV (d(E) = 1. . . 1.05).The fit gives a temperature of 313 ± 2 K for a chosen pump fluence of P = 3.5 mJ cm−2, error limits representing the 95% confidence bounds for the fit. This fit is presented in manuscript Fig. 2a (top inset) and Supplementary Fig. 6a. The fitted pump fluence is chosen such that the fluence is the lowest showing a linear slope in the time-integrated population distribution (in logarithmic scale). The chosen fluence also corresponds to the narrowest FWHM of the highest condensate peak (see Supplementary Fig. 6c). The high-energy tail remains linear over pump fluences between ~3.5 and 4 mJ cm−2, with a slightly changing slope. For two higher pump fluences of 3.7 and 3.9 mJ cm−2, the fits to the the Maxwell-Boltzmann distribution give temperatures of 282 ± 2 and 250 ± 2 K, respectively. Goodness of fit is described by the square root of the variance of the residuals (RMSE) and the R-square value. The values of (RMSE, R-square) for the three pump levels 3.5, 3.7, and 3.9 mJ cm−2 are (107, 0.996), (98, 0.998), and (146, 0.998), respectively. For pump fluences above ~4 mJcm−2, the linear slope is distorted and the condensate degrades. This is evident also as a decrease of the spatial coherence (Fig. 4a) and an increase of the FWHM of the spectral maximum (Supplementary Fig. 6c).For the longer pulses, 100 and 250 fs, the fit gives 316 ± 2 and 331 ± 4 K at the condensation threshold, 3.5 and 4.5 mJ cm−2, respectively. The values of (RMSE, R-square) for the fits are (119, 0.997) and (219, 0.985), respectively. The fits are still quite good for these pulse durations, whereas for 350 fs and longer pulses no thermal MB distribution was observed.Estimation of photon number in the condensateThe photon number in the condensate is estimated from the measured luminescence intensity. A strongly attenuated beam from the external pump laser (800 nm, 1 kHz) is directed to the spectrometer slit, and the total counts given by the spectrometer CCD camera (Princeton Instruments PIXIS 400F) is compared to the average power measured with a power meter (Ophir Vega). The measured average power of 167 nW corresponds to 6.7 × 108 photons/pulse whereas the total counts in the CDD camera are 8.4 × 106, leading to a conversion factor of ~80 photons/count. In the condensation regime, the total counts per pulse is about 3 × 106 (manuscript Fig. 2a). The collection optics including the beam splitters reduce the signal roughly by a factor of 2.5, and as the slit width of 500 μm corresponds to 27 μm at the sample, we collect luminescence from an area that is about 1/4 of the 100 μm wide nanoparticle array. Finally, the sample is assumed to radiate equally to both sides, so the actual photon number per emitted condensate pulse becomes: nph ≈ 2.5 × 4 × 2 × 80 × 3 × 106 = 4.8 × 109.Spatial coherence measurementsSpatial coherence of the sample luminescence is measured with a Michelson interferometer. The real space image is split into two arms and the image in one of the arms is inverted in vertical direction with a hollow roof retro-reflector. Then the two images are combined again with a beam splitter and overlapped at the camera pixel array, simultaneously. With this design the spatial coherence (first-order correlation g(1)) can be measured separately along both x- and y-axis of the plasmonic lattice. The retro-reflector always inverts the image vertically, with respect to y = 0 in laboratory reference frame, but the sample and the pump polarization can be rotated 90∘ to measure g(1)(−x, x) instead of g(1)(−y, y) in the lattice coordinates. The first-order correlation function describing the degree of spatial coherence is given by2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${g}^{(1)}(-{\bf{y}},{\bf{y}})=\frac{\langle {E}^{* }(-{\bf{y}})E({\bf{y}})\rangle }{\sqrt{\langle E{\left(-{\bf{y}}\right)}^{2}\rangle \langle E{\left({\bf{y}}\right)}^{2}\rangle }},$$\end{document}g(1)(−y,y)=⟨E*(−y)E(y)⟩⟨E−y2⟩⟨Ey2⟩,where E(y) is the electric field at point y. The first-order correlation function relates to the interference fringe contrast C as follows:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C({\bf{y}},-{\bf{y}})=\frac{2\sqrt{I({\bf{y}})I(-{\bf{y}})}}{I({\bf{y}})+I(-{\bf{y}})}{g}^{(1)}({\bf{y}},-{\bf{y}}),$$\end{document}C(y,−y)=2I(y)I(−y)I(y)+I(−y)g(1)(y,−y),where I(y) is the luminescence intensity at point y of the lattice. The fringe contrast in the interfered images is extracted with a Fourier analysis as explained in Supplementary Note 4.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Movie 1
Supplementary Movie 2
Supplementary Movie 3
Supplementary Movie 4 | nature communications | [
"Article"
] | [
"Nanophotonics and plasmonics",
"Polaritons",
"Ultrafast photonics",
"Bose-Einstein condensates",
"Quantum mechanics"
] | Bose–Einstein condensation means accumulation macroscopic population single state system long-range order Bosonic condensation in non- quasi-equilibrium driven-dissipative systems extends concept offers new phenomena loss decaying phase generalized BEC into multiple rich phase diagrams lasing condensation superradiance quantum simulation of XY model optimization condensates explore quantum phase transitions7 Each condensate system offers advantages limitations non- quasi-equilibrium dynamics tune interactions wide range advantage ultracold Polariton offer high critical temperatures photon condensates19 thermal bath controlled21 periodic two-dimensional arrays metal nanoparticles plasmonic lattices for room temperature lasing condensation at strong regimes lattice geometry periodicity size shape nanoparticles size controlled nanometer accuracy energy condensation lasing by band edge energy on period array band edge energy dispersion constant over large lattices (accuracy 0.1%28) In semiconductor polariton condensates disorder samples leads to traps fragmentation15 trapped by geometry16plasmonic lattices offer propagation excitations time-evolution thermalization position related to time via group velocity no spurious effects due to non-uniformity sample Spatially resolved luminescence utilized first observation of BEC in plasmonic lattice28 formation condensate with pronounced thermal distribution possible at 200 fs timescale fast thermalization to coherent dynamics due to stimulated processes strong coupling double-threshold phenomenon one-dimensional (1D lasing for lower pump fluences 2D multimode condensation at higher fluences transition between lasing condensation different relies on system size propagation thermalization dynamics intermediate regime features thermalization process no macroscopic population at lowest energy states stimulated nature thermalization luminescence in lattices different sizes ultrafast character thermalization condensation occurs for pump pulse durations below 100–250 fs present characterization of lasing condensation phenomena focus on main results stimulated nature thermalization process dramatic effect of pump pulse durationcylindrical gold nanoparticles in rectangular lattice organic dye molecule IR-792 lattice supports dispersive modes surface lattice resonances hybrid modes localized surface plasmon resonances nanoparticles diffracted orders periodic electric field SLR confined to lattice plane SLR excitation bosonic quasiparticle photon collective electron oscillation metal particles SLR modes transverse magnetic or transverse electric) polarization propagation direction defined in Fig. 1b measured dispersions displayed in Fig. 1c–e dye molecules SLR dispersion shifts downwards TE modes bend approaching molecular absorption line at 1.53 eV observations indicate strong coupling between SLR molecular excitations38 coupled modes splitting 164 meV (larger than average line (150 meV SLR (10 meV exciton part 23% at k = 0 hybrids SLR excitations molecular excitons as polaritons coherence length polaritons 24 μm high molecule concentration in liquid gain solution contrast previousplasmonic lattice condensate particle diameter 100 nm height 50 nm period y x py = 570 nm px 620 nm dye concentration 80 mM lattice size 100 × 100 μm2. period py 520 nm lattice size 40 200 × μm2.Fig. system dispersion modes experimental configuration Light cones diffracted orders (0 −1) +1) x-polarized nanoparticles Crosscut ky (kx = 0 TE mode kx = 0 TM mode TE mode polarization perpendicular propagation direction TM mode parallel Crosscuts SLR dispersions obtained measuring transmission without molecule reflection 80 mM solution IR-792 SLR dispersion red shift avoided crossing absorption transition molecule absorption line uncoupled SLR mode white dashed lines lower-polariton branch black line Absorption emission spectra IR-792 left panel blue red lines Scanning electron micrograph nanoparticle array scale bar 500 nm excite sample laser pulses 1 kHz wavelength 800 nm resolve luminescence spectrally angle spatial position array1a details see Section “Methods Transmission reflection luminescence measurement setup”). pump couple SLR modes small fraction photons coupled single particle resonance absorbed by molecules nanoparticle lattice Active region dye molecules within few hundred nanometers from lattice plane shown refs. 27,39–41 molecules further away unlikely couple SLR modes.Transition function pump luminescence properties plasmonic lattice function pump fluence energy per unit area excitation pulse double-threshold behavior system excited with x-polarized 50 fs laser pulse flat intensity profile size larger than lattice Excited polariton modes leak radiative loss luminescence intensity proportional to population polaritons record real space intensity distributions spectra photon energy E = hc/λ0 in-plane wave vector\documentclass,y=2π/λ0sin(θx sample luminescence function pump fluence presented in Fig.total luminescence intensity reveals two non-linear thresholds linear intermediate regime in Fig. 2a line spectra obtained integrating real space spectra y-axis unveil population polaritons function of energy first threshold Fig. 2b, c lasing condensation) typical for nanoparticle arrays24 observed Increasing pump fluence luminescence intense central part top luminescence center lower energy than edges red shift signature of polariton population thermalization propagating in +y −y directions second threshold Fig. 2f, g system into condensate Maxwell–Boltzmann (MB) distribution at higher energies): space intensity distribution shows uniform luminescence in central part line spectrum 2a narrow peak at band edge long thermalized tail at higher energies fit tail to MB distribution gives room temperature, T = 313 ± 2 K Section “Methods Fits MB long tail fits MB distribution criterion for distinguishing between condensate and lasing regimes condensate MB tail narrow-peaked population at low energy peaked population without tail as lasingstrong coupling regime lasing polariton lasing polariton lasing regime momentum-space confinement spatial coherence properties condensate lasing regimes differ distinct.Fig. 2Pump fluence dependence real space images spectra Double-threshold curve pump fluence total luminescence intensity Line spectra integrating real space spectra y-direction FWHM spectral peaks 12, 72 4.0 meV increasing pump fluence dashed line top inset fit Maxwell–Boltzmann distribution temperature 313 ± 2 K (95% confidence bounds). Left Real space images plasmonic lattice Right Spectral information luminescence y-position pump fluence b, c 0.87 mJ cm−2 d e 1.5 mJ cm−2 f, g 3.5 mJ cm−2. wave fronts d e standing waves momentum Supplementary Fig. 1 Note 1. real space images single pump pulses spectra integrated over 500 e 70 pulses results full range pump fluences Supplementary Movie spectrometer counts per condensate pulse photon number ~109 105 times more than first BEC plasmonic lattice28increased luminescence attributed to stimulated processes differences in sample pump detection geometry 2) improvement increased signal-to-noise ratio prominent thermal tail data pump pulse important signature efficiency thermalization process ultrafast upgrade luminescence intensity crucial for future studies applications condensate thermodynamic quantities determined using photon periodicity tuned thermalization rate array size.Three distinct regimes observed in k-space intensity distributions Figure 3 presents k-space images spectra for sample pump parameters lasing spreads in TM mode to large k thermalization polaritons TE mode condensation threshold confinement in kx ky condensate 2D nature contrast to lasing regime confinement only in ky Figure 3j shows line spectra integrating ky from TE mode crosscuts spectrometer slit width 500 μm to ±1.3∘ around θx = 0 2D k-space images condensation threshold multiple modes occupied at ky = 0 line spectrum at 3.49 mJ cm−2 shows three narrow peaks thermalizedfull-width half-maximum highest peak 3.3 meV narrower than SLR mode (10 spacing peaks decreases lower energy out Fabry–Pérot interference spacing depend on periodicity lattice size not caused by waveguide T-matrix simulations cylindrical shape finite size nanoparticles three modes around Γ-point energy infinite lattice one degenerate role nanoparticles periodic structure degeneracy lifted by finite lattice size distinct modes lattice size tune mode structure Supplementary Fig. 2 Note 3. thermal condensation lowest energy driven-dissipative systems several modes distinct energies temporally integrated signal single-mode condensate evolving between states sub-picosecond. 3Momentum space images spectra 2D momentum (k-)space images third k-space spectrum TM TE mode directions spectra correspond horizontal vertical slices 2D k-space spectrum pump fluence a–c 0.85 mJ cm−2 d–f 1.5 mJ cm−2 g–i 3.5 mJ cm−2 images line spectra integrated over 500 330 20 pulses horizontal emission peaks h, i CCD blooming artifactsPopulation distribution TE mode (c f i) ky FWHMs indicated dashed black lines 6.3 17, 3.3 meV increasing pump fluence results full range pump fluences Supplementary Movie 2.Lasing condensation transitions increased spatial temporal coherence emitted light Increase temporal coherence narrowing spectral line width (Fig. 3j). coherence Michelson interferometer experiment function pump fluence space image split two inverted combined camera pixel array contrast interference fringes extracted Fourier analysis spatial frequencies Figs. 3 4 Note 4) Fig. 4 fringe contrast correlation function function pump fluence y- x-directions lattice High spatial coherence y-direction lasing condensation regimes (Fig. 4b intermediate regime coherence decreases thermalizing population dominates luminescence signal x-direction spatial coherence lower y-direction pump range condensate exhibits high spatial coherence x-direction contrast lasing separated emission stripes space image Fig. 4d Supplementary Fig. 3) spatial coherence measurements line observations 2D k-space imageslasing confined luminescence along ky spreads along kx condensation shows 2D k-space confinement coherence in lasing condensation extends over array (100 μm coherence length four times larger than samples without pumping (24 μm). future study larger samples coherence decays algebraically decaying phase order in 2D driven-dissipative systems. 4Spatial coherence measurement along y- x-axis plasmonic lattice Michelson interferometer fringe contrast region function of pump fluence contrast extracted from non-interfered real space images Interfered space images inverted spatial coherence between −y +y −x +x interference patterns for pump fluences lasing condensation regimes coherence −x +x positions rotate sample pump polarization by 90∘ vertical stripes in b d due to one-dimensional lasing y-direction horizontal stripes c e from spatial coherence in overlapped images Michelson interferometer Methods (Spatial coherence measurements Supplementary Note 4. real space images recorded for single pump pulseslasing condensation at energies 1.397 1.403 eV lower than band-edge energy system without 1.423 eV energies blue-shifted from lower energy 1.373 eV or 1.382 eV obtained by reflection measurement coupled modes dispersion blue shifts pump fluence to degradation strong coupling Coulomb interactions saturation-caused non-linearity polariton–polariton estimate Supplementary Note 8). band-edge locations (1.423−1.403) eV/(1.423−1.373) eV coupling in condensation decreased to ~40% without pumping double-threshold behavior different from semiconductor microcavity polariton condensates lower threshold loss.Stimulated thermalizationAt intermediate regime red shift luminescence function distance y Fig 2e trails begin from emission maximum dye molecule (~1.46 reach band-edge energy (~1.40 eV) at center of 100 × 100 μm2 array red shift recorded space images spatially resolved spectra for lattice sizes at intermediate pump fluences thermalization Fig. 5a–f large array trails red shift toward center 100 100 μm2 arrayred-shifting populations merge center small array (Fig. 5a, different red shift from center toward edges distance edge red shift Fig. 10 all arrays distance same (~25 μm 2.2 mJ cm−2 pump fluence Fig 5a c e).Fig. stimulated emission pulse build-up finite size lattices Real space images spectra intermediate pump fluence (2.2 mJ cm−2) lattice sizes a b 40 × μm2 d 60 × 60 μm2 e f 150 × 150 μm2. different color scales panels Line spectra lattice sizes Supplementary Fig. 11. Rate-equation simulation stimulated emission pulse build-up time marked by vertical dashed lines peaks population inversion N* output Horizontal dotted line indicates N* overcomes losses sum spatial intensity profiles thermalizing pulses propagating left right y-axis array Gaussian shape approximates increasing decreasing intensity excitations results array sizes shown as insets in a c e same false color space imagesMeasured distance from array edge to red shift begins 25 μm scale bars function of pump fluence for different lattice sizes legend indicates-array sizes in μm diagonal dashed line inverse of pump fluence (50 horizontal dashed line saturation value (~18 μm). inset shows pulse build-up time distance multiplying group velocity SLR mode from rate-equation simulation distance by stimulated emission pulse build-up time between population inversion output pulse simulation Fig. 5g pulse build-up thermalization not condensate lasing Pulse build-up in Q-switched lasers47 pump pulse excites molecules polaritons propagate photons populate modes gather gain peak stimulated emission pulse after distance traveled array distance seen as dark zones in space measurements corresponds to pulse build-up time pulse different to lasing condensate pulse line width luminescence large spatial coherence small summing up spatial intensity profiles thermalizing pulses lattice (Fig. reconstruct real space intensity distributions Fig. 5a c dark zones appear edges receive excitations lattice half intensity central partsmall lattice intensity edges similar larger center half wavy interference patterns central part (Fig. 5c appear arrays larger 40 × 40 μm2 counter-propagating pulses width dark zones depends pump fluence rate-equation simulation Q-switched lasers build-up time inversely proportional pump fluence Fig. 5i dark-zone width follows inverse pump fluence until saturates 3 mJ cm−2 BEC threshold below 20 μm (~100–140 fs). Fig. 5i pulse build-up time rate-equation simulation similar ~1/P dependence red shift thermalization process intermediate regime pump fluences thermal distribution reached population decays higher fluences condensate peak MB distributed population emerges classical thermal MB distribution straight line peaked feature low energies MB tail BE distribution low energies (E − μ < BE distribution approximated kBT/(E − μ). classical condensation waves Rayleigh–Jeans) interplay random noise gain/loss profiles opticalsystem conditions for classical condensation linear-log-scale tail match distribution kBT/(E − μ). rule out RJ condensation observe for energies E larger than condensate peak by kBT E − μ < kBT BE RJ distributions coincide resolve tail up to energies 75 meV above condensate energy—three times larger than room temperature kBT = 25 meV distribution match classical MB or RJ distribution distribution BE form in non-steady-state driven-dissipative system weak coupling regime answer known system similar to photon condensates19 dye molecules vibrational level structure provide thermalization Differences excitations ultrafast time-scales strong light-matter coupling weak coupling regime recurrent absorption emission light with molecules vibrational manifold lead to thermalization condensation with BE-type) distribution continuous pumping vibrational manifold energy loss channel photon population towards lower energies provides temperature for BE distribution molecules emit at lower energy than absorb effective coupling between photons energies thermalization need scattering Coulomb interaction semiconductor polariton condensationthermalization requires absorption-emission cycles within lifetime system speed depends on number molecules light-matter coupling photons plausible mechanism provides thermalization in our present coupled case lifetime system short emission-absorption processes stimulated during red shift at higher energies thermalization criterion several emission-absorption cycles within lifetime thermalization rates present case higher 0.20 eV/ps in Fig. 2e than same molecular concentration previous work28 (0.08 eV larger number photons stimulated processes to speedup discrete step creation low energy polariton vibrational quantum route for organic semiconductor polariton condensation14 microscopic foundation same as photon condensation parameter regimes differ discrete step when absorption emission spectra show distinct vibrational sub-peaks smooth red shift BE distribution likely for molecules vibrational states not visible Our system shows strong coupling molecule spectra no vibrational shoulders (Fig. 1d), middle ground between weak coupling photon condensation mechanism relaxation by discrete step requires theoretical descriptionsingle molecule one vibrational state few light modes describe process coherent vibrational losses leads to rapid red shift emission Fig. 5 Note 6) observations model many molecules several vibrational states thermal reservoir multiple light modes at multiphoton regime model solved without weak coupling perturbation theory light current theory59–63 starting point advanced description condensation threshold on losses thermalization speed competition with lasing photon numbers emitted condensate larger than equilibrium estimate critical number28.Effect of pulse durationThe spatial measurements dynamics system probed dynamics altering excitation pulse duration 50 fs excitation pulse double-threshold behavior distinct regime for lasing intermediate regime incomplete stimulated thermalization condensation longer pulse first (lasing threshold system condensation at higher pump fluences space intensity distribution spectrum unchanged from low to high pump fluence for 500 fs pulse duration Fig. 6. intensity distributions spectra resemble lasing regime at low pump fluence 50 fs pulse (Fig. 2b intermediate condensation regimes absent different threshold behavior visible in FWHM curves spectral maximum FigFWHM decreased pulse durations first threshold 50 fs pulse second threshold k-space images spectra 500 fs pulse luminescence from low to high pump fluences spread in TM mode no 2D confinement.Fig. 6Pump fluence dependence real space images spectra 500 fs pump pulse duration Pump fluence dependence total luminescence intensity Line spectra space spectra y-direction space images plasmonic lattice Spectral information luminescence position y-axis pump fluence 0.88 mJ cm−2 d 1.6 mJ cm−2 g 3.5 mJ cm−2. FWHM spectral peaks marked insets increasing pump fluence FWHM 23, 18 23 meV results full range pump fluences Supplementary Movie compared results 50 500 fs excitation pulses total energy injected per pulse same durations stimulated thermalization instantaneous pump intensity fluence condensation threshold reached longer pulses if pump fluence increased studied dependence on pulse duration intermediate measurements (Fig. 7) condensation with 100 250 fs pulses not longer pulsescondensation threshold 100 fs pulse equal 50 fs pulse (3.5 mJ cm−2) 250 fs higher (4.5 mJ cm−2) line spectra condensation threshold 100 250 fs pulse durations Fig. 7a similar 50 fs pulse Fig. 2a macroscopic population band edge linear distribution higher energies Fit MB distribution gives 316 ± 2 331 ± 4 K 100 250 fs pulse Section “Methods Fits MB longer pulse (350 fs), thermalization incomplete (too much population higher energy longest excitation pulses (>350 no signs thermalization condensation highest pump fluences thermalization process competes over same gain with lasing long excitation pulse instantaneous population inversion high thermalization lasing first threshold Fig. 6).Fig. 7Line spectra threshold pump fluences different pulse durations Condensation occurs 100 250 fs pulses not longer ones spectra shown at condensation threshold (3.5 4.5 mJ cm−2 threshold 100 fs same as 50 fs pulse defined condensation threshold narrow peak band edge linear distribution at high energies time-integrated signal350 fs pulse no condensation spectrum distribution closest linear 6.0 mJ cm−2) 500 fs pulse highest pump fluence (6.3 mJ cm−2) luminescence intensity function pump fluence Increasing pump pulse duration increases first (lasing threshold second (condensation threshold visible 100 250 fs pulses 350 fs pulse increased pump fluence 10 mJ cm−2 nanoparticle array no condensation threshold higher fluences saturation degradation luminescence highlighted data points line spectra Threshold pump fluence lasing condensation beginning incomplete stimulated thermalization regime function pulse duration lasing regime one spectral intensity maximum band edge lasing higher energy increasing pump fluence lasing peak grows higher energy peak diminishes pump fluence higher energy peak starts growing pump fluence value line spectra threshold incomplete thermalization regime interference patterns Fig. 2e visible space spectra lines guides eye observations pulse durations summarized Fig. 7c thresholds lasing condensation start incomplete stimulated thermalization regime pulse durationdependence condensation incomplete thermalization thresholds on pulse length stronger than lasing case threshold pulse duration Sensitivity to excitation pulse duration highlights ultrafast plasmonic systems sub-picosecond dynamics thermalization critical pulse duration for condensation similar to smaller than time (~250–350 fs) polaritons from edges to center in 100 × 100 μm2 array “Methods condensation requires thermalization time propagation time polaritons red-shifted to band-edge energy large population density achieved by balance between dye concentration pump fluence (thermalization lattice size (distance to lattice period condensation sensitive to incident angle pump clear MB distribution pump pulse needs normal incidence slight misalignment changes space intensity pattern spectral distribution Nonzero incidence angle time difference excitation at edges asymmetry in populations counter-propagating polariton modes questions on dynamics of BEC in driven-dissipative systems open despitenature energy relaxation thermalization processes condensate form quantum statistical long-range coherence properties? questions addressed for weakly coupled BECs challenging for strongly coupled room temperature condensates higher energy scales faster dynamics plasmonic lattices offer access control to sub-picosecond dynamics condensate formation via propagation excitations finite system size bosonic condensate thermal excited state hundreds femtoseconds speed thermalization due to strong light–matter coupling stimulated emission Strong light–matter coupling weak excitation limit indicated by reflection measurements varying lattice size revealed stimulated nature thermalization strong light–matter coupling multiphoton regime described by Dicke red shift thermal distribution vibrational degrees freedom coupled to electronic thermal bath Work surmounting theoretical challenges begun systems light electronic vibrational states strongly coupled for lasing condensation energy transfer modification chemical reactions65 plasmonic lattices offer platform for ultrafast light–matter interaction shape size material nanoparticles tuned lattice geometry composition unit cell size large controlled parameter space for testing)qualifying theoretical predictions dynamics accessed conventional time-domain techniques propagation excitations latticegold nanoparticle arrays fabricated electron beam lithography on glass substrates 1 nm titanium adhesion layer Fig. 8). nominal dimensions plasmonic lattice for condensation experiment nanoparticle diameter 100 nm height 50 nm period y- x-direction py = 570 px = 620 nm lattice size 100 × 100 μm2. period py varied between 520 and 590 nm lattice size 40 × 200 × 200 μm2.Asymmetric periodicity separates diffracted orders energy spectrum orthogonal polarizations (ex SLR dispersions separated simplifies measured spectra x-polarized nanoparticles TE TM modes correspond to (ex, ky) (ex, kx), pumping SLR mode excited determined by pump polarization plasmonic experiment px 50 nm larger than py period x y directions same (px = py = 570 nm). asymmetric periodicity forming condensate simplifies data analysis velocity for TE mode obtained close to Γ-point dispersion samples dye molecules 0.65c uncoupled TE mode 0.48c strongly coupled mode1d c speed light). group velocity propagation distance to time strongly coupled dispersion uncoupled at high pump fluences due saturation effects specify group velocity at pump fluence time conversions 0.48c to 0.65c propagation of 50 μm distance array takes 250–350 fs dye molecule solution index-matched to glass substrate (n = 1.52) mixture 1:2 DMSO:Benzyl Alcohol sealed Press-to-Seal silicone isolator chamber between glass substrate solution thickness ~1 mm large compared to SLR electric fields27 excess dye natural circulation fluid fresh dye molecules available for measurements pump fluence sample robust long-lastingIR-792 perchlorate[12pt{minimal\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}{C}}}{42{49{ClN{4{S}}}C42H49ClN2O4S chosen dye molecule dissolves solvent high concentrations contrast other dye molecules IR-140 used gain medium plasmonic nanoparticle array lasers collected information tested dye molecules reasoning molecule choice Supplementary Table 1.Transmission reflection luminescence measurement schematic experimental setup Supplementary Fig. 9. setup transmission reflection luminescence measurements minor modifications spectrometer resolves wavelength spectrum light entrance slit each pixel column 2D CCD camera corresponds free space wavelength λ0 pixel row y-position slity-position corresponds to angle (k-space or sample (real photon energy E = hc/λ0 angle-resolved spectra in-plane wave vector\documentclass[12pt{minimal}{amsmath{wasysym{upgreek\oddsidemargin{-69pt}{document}{k}_{x,y}={0}\sin\theta)=2\pi\lambda}kx,y=k0sin(θx,y)=2π/λ0sin(θx,y), h Planck constant c speed of light in space three experiment types luminescence measurement sample optically excited with external pump laser excitation pulse generated by Coherent Astrella ultrafast Ti:Sapphire amplifier central wavelength 800 nm duration <35 fs bandwidth 30 nm pump pulse guided through beam splitter mirrors mirror M1 Fig 9) directs to excitation path band-pass filter in excitation path with long-pass filter in detection path filter pump pulse luminescence spectra pump pulse linearly polarized filter horizontal component linear polarizer after band-pass filterpump fluence controlled with metal-coated variable neutral-density filter wheel pump pulse spatially cropped with adjustable iris imaged onto sample lens L1 microscope objective inverted design enables sample at normal incidence crucial for excitation dye molecules prevents asymmetry spatial excitation inverted pumping scheme accurate optical essential for condensate formation detection path long-pass filter linear polarizer iris pinhole spatial filter at 1st image plane imaged area relayed real-space camera k-space measurements back-focal plane objective relayed 2D k-space camera spectrometer slit tube lens k-space lens real space measurements beam-splitter before-space lens replaced with additional real-space lens 2nd image plane spectrometer slit spectrometer slit selects vertical slice 2D k-space or real space luminescence measurements slit width 500 μm to ±1.3∘ around θx = 0 ±0.16 μm−1 around kx = 0 at E = 1.4 eV real slit opening 500 μm to 27 μm slice at sampledispersion of optical modes in plasmonic lattice measured in transmission mode sample illuminated with white light halogen lamp lattice modes visible as transmission minima (extinction maxima) in angle-resolved spectrum high dye molecule solution added transmission measurement not applicable to absorption access dispersion setup in reflection mode inverted design luminescence measurement halogen lamp inserted before iris excitation path imaged onto sample dispersion modes revealed by reflection maxima in angle-resolved spectrum luminescence measurement pump fluence automated with LabView Predefined fluence steps measured ND filter wheel set shutter opened image acquired with spectrometer 1st 2nd Cam. shutter closed Integration time spectrometer adjusted avoid saturation at non threshold regimes pump pulse duration measured with commercial autocorrelator (APE pulseCheck same optics experiments changed by adjusting stretcher-compressor external pump laser.Fits to MB thermalized population distributions to MB distribution (Fig. 2 Supplementary Fig. 6)fit function[12pt{minimal\usepackage{amsmath{upgreek\oddsidemargin-69pt}{document}{MB(E)={d(E)}(E-\mu/}fMB(E)=d(E)e(E−μ)/ d(E) degeneracy of modes function of energy E μ chemical potential kB Boltzmann constant T temperature fit for distribution linear logarithmic scale pump fluence at/above threshold (between energies 1.41 and 1.47 eV “tail”. Fitting performed with nonlinear least method degeneracy d(E) approximated by density of states for light traveling 2D planelight dispersion xy plane forms conical surface[12pt{minimal{amsmath\oddsidemargin-69pt}=ħc/nkx2+ky2) dispersions SLR modes approximated except k = 0 point67 fitted range starts from finite k approximation valid dispersion results linearly increasing constant density of states energy range ~60 meV (d(E) = 1.05) fit gives temperature 313 ± 2 K pump fluence P = 3.5 mJ cm−2 95% confidence bounds presented manuscript Fig. 2a Supplementary Fig. 6a fitted pump fluence chosen lowest linear slope in time-integrated population distribution corresponds narrowest FWHM highest condensate peak high-energy tail linear over pump fluences between ~3.5 and 4 mJ cm−2 slightly changing slope two higher pump fluences 3.7 3.9 mJ cm−2 fits give temperatures 282 ± 2 and 250 ± 2 Kfit described by variance residuals (RMSE R-square value values R-square pump levels 3.5 3.7 3.9 mJ cm−2 are (107, 0.996) (98, 0.998) (146, 0.998) pump fluences above ~4 mJcm−2 linear slope distorted condensate degrades spatial coherence increase FWHM spectral maximum longer pulses 100 250 fs fit gives 316 ± 2 331 ± 4 K at condensation threshold 3.5 4.5 mJ cm−2 values (RMSE R-square) (119, 0.997) (219, 0.985) good 350 fs longer pulses no thermal MB distribution.Estimation photon number estimated from luminescence intensity attenuated beam pump laser (800 nm directed to spectrometer slit total counts compared to average power power 167 nW 6.7 × 108 photons/pulse total counts 8.4 × 106 conversion factor ~80 photons/count condensation regime total counts per pulse about 3 × 106collection optics beam splitters reduce signal factor 2.5 slit width 500 μm 27 μm sample collect luminescence 1/4 100 μm nanoparticle array sample equally both sides photon number per condensate pulse nph ≈ 2.5 × 4 × 2 × 80 × 3 × 106 = 4.8 × 109.Spatial coherence sample luminescence measured Michelson interferometer space image split into two arms inverted hollow roof retro-reflector images combined beam splitter overlapped at camera pixel array spatial coherence measured separately x- y-axis plasmonic lattice retro-reflector inverts image vertically sample pump polarization rotated 90∘ measure g(1)(−x, x) g(1)(−y,)first-order correlation function spatial coherence given\documentclass[12pt{minimal{amsmath{upgreek\oddsidemargin-69pt}\begin{document}(-{\bf{y}}{y}})=\frac{\langle\bf{y}}\bf{y}})\rangle{\langle E{\left(-{\bf{y}}\right)}{2}\rangle\end{document}g(1)(−y,y)=⟨E*(−y)E(y)⟩⟨E−y2⟩⟨Ey2⟩ E(y) electric field at point yfirst-order correlation function interference fringe contrast C\documentclass[12pt]{minimal}{amsmath\oddsidemargin{-69pt}{document}({\bf{y}},-{\bf{y}})=\frac{2\sqrt{I({\bf{y}})I+I}{(1)}\end{document}C(y,−y)=2I(y)I(−y)I(y)+I(−y)g(1)(y I(y) luminescence intensity point y lattice fringe contrast interfered images extracted Fourier analysis Supplementary Note 4.Supplementary information Peer Review File Additional Supplementary Files Movie 1 2 3 4 | 47.7 | 1.139835 |
10.1038/s41467-020-20727-7 | PMC7815879 | Multicomponent reactions enable the rapid construction of diverse molecular scaffolds with modularity and step economy. In this work, the authors report the use of boronic acids as carbon nucleophiles in a Passerini-type three-component coupling reaction towards an expanded inventory of α-hydroxyketones. | Multicomponent reactions (MCRs) facilitate the rapid and diverse construction of molecular scaffolds with modularity and step economy. In this work, engagement of boronic acids as carbon nucleophiles culminates in a Passerini-type three-component coupling reaction towards the synthesis of an expanded inventory of α-hydroxyketones with skeletal diversity. In addition to the appealing features of MCRs, this protocol portrays good functional group tolerance, broad substrate scope under mild conditions and operational simplicity. The utility of this chemistry is further demonstrated by amenable modifications of bioactive products and pharmaceuticals as well as in the functionalization of products to useful compounds. | Introductionα-Hydroxyketones (also known as acyloins) are structural units ubiquitously found in natural products1–5 and pharmaceuticals6,7. They are also oft-employed synthetic precursors in a panel of high-value transformations (Fig. 1a)8–13. The construction of these important molecules is therefore the subject of substantial synthetic efforts14. Traditional benzoin condensation method assembles α-hydroxyketones via condensation of different aldehydes, thus limits its applicability within this substrate class15–17. The alternate oxidative pathways that encompass α-hydroxylation of ketones18–22 and ketohydroxylation of olefins23–27 are certainly enabling, but continues to be challenged in terms of substrate diversity and poor selectivity. Hence, devising complementary routes towards these useful entities from readily available starting materials is highly relevant and desirable.Fig. 1Precedent works and proposed Passerini-type coupling reaction with boronic acids as nucleophilic agents.a
α-Hydroxyketones in bioactive moleculars or as synthetic precursors. b Classic Passerini or Ugi reaction. c Petasis boronic acid-Mannich reaction. d Passerini-type coupling reaction of boronic acids (this work).Multicomponent reactions (MCRs) are often prized for their concise and modular features in forging complex molecules with synthetic and biological interest28–36. The representative Passerini reaction37–46 or Ugi reaction47–54 efficiently assembles α-acyloxyamides or α-acylaminoamides from several reactant components via the intermediacy of nitrilium species in single-pot operation (Fig. 1b)55–58. Interception of this electrophilic intermediate in Passerini reaction pathway by carbon nucleophiles (in place of conventionally used carboxylic acids) would offer an intriguing access to α-hydroxyketone products; yet such synthetic maneuver remains underexplored59,60. Central to the successful establishment of this chemistry would lie in choosing suitable carbon nucleophiles that would not interfere with the formation of the nitrilium intermediate while possess sufficient nucleophilicity to capture this electrophile.On the other hand, boronic acids are easily available, benign and common building blocks for C-C bond cross-coupling reactions, in both transition-metal catalysis61,62 and metal-free catalysis regimes63–75. In boronic acid-Mannich reaction (or Petasis reaction), for instance, the nucleophilic feature of boronic acids effects the formation of boron “ate” complex, leading to functionalized amines following 1,3-metallate migration (Fig. 1c)76–80. To this end, a recent endeavor of our group has unraveled a 1,4-metallate shift of boron “ate” nitrilium species generated from nitrile oxide and arylboronic acid, thus mediating stereospecific formation of C-C bond between oxime chlorides and arylboronic acids under metal-free conditions81. Grounded in these knowledges, we envisioned that a boron “ate” nitrilium intermediate could be released from co-treatment of aldehyde, isocyanide, and boronic acid; 1,4-metallate shift of which will invoke C-C bond coupling and α-hydroxyketones could be revealed on hydrolysis (Fig. 1d). Here, we disclose the development of a Passerini-type coupling reaction, which afforded α-hydroxyketones from the combination of readily available aldehydes, isocyanides, and boronic acids (aryl, alkenylboronic acids, and alkynyl trifluoroborate salts) under transition-metal-free conditions. Mild reaction conditions, ease of execution, high functional group tolerance, broad substrate scope, and utility are practical features of this methodology.ResultsInvestigation of reaction conditionsExploratory investigations towards our envisioned Passerini-type reaction involving boronic acids were conducted with phenylpropyl aldehyde (1a), tertbutyl isocyanide (2a) and 4-methoxyphenyl boronic acid (3a) as test substrates (Table 1). To our delights, simple mixing of the three reactants (1a, 2a, and 3a) without any other additive in DCM furnished the desired α-hydroxyketone product 4a in 60% isolated yield (entry 1). A solvent screen of DCE, MeCN, toluene, MeOH, and THF revealed that the best reaction efficiency was endowed by CHCl3, whereas using MeOH caused a complete reaction inhibition (entries 1−7). As reaction temperature was decreased to 10 °C, the yield of 4a improved to 68% (entry 8). Binary mixture of CHCl3 and water in a ratio of 7:3 (entries 9−11) minimally but meaningfully enhanced the delivery of 4a to 72% yield (entry 11). This has guided our subsequent study of mixed solvent system with CHCl3 against various buffer solutions (entries 12−15) where the combination with pH = 8.0 buffer delightfully provided 81% yield of target product (entry 14). We reasoned that a basic reaction medium could sequester the byproduct B(OH)3 generated during reaction, thus promoting this boronic acid-involved Passerini-type reaction. It was further established that on replacement of tertbutyl isocyanide (2a) with cyclohexyl isocyanide (2b), benzyl isocyanide (2c), or ethyl 2-isocyanoacetate (2d), formation efficiency of α-hydroxyketone product 4a was diminished (entry 16). None of the other ratios of the three reagents resulted in higher yields (entries 17−18).Table 1Optimization of the reaction conditionsa.EntriesIsocyanide1a:2:3aSolventTemp. (°C)Yield (%)b12a1:1.5:1.8DCMrt6022a1:1.5:1.8DCErt5532a1:1.5:1.8CHCl3rt6442a1:1.5:1.8MeCNrt5052a1:1.5:1.8toluenert4862a1:1.5:1.8MeOHrtN.R72a1:1.5:1.8THFrt3082a1:1.5:1.8CHCl3106892a1:1.5:1.8CHCl3/H2O (3:7)1062102a1:1.5:1.8CHCl3/H2O (1:1)1067112a1:1.5:1.8CHCl3/H2O (7:3)1072122a1:1.5:1.8CHCl3/pH = 6.5 buffer (7:3)1066132a1:1.5:1.8CHCl3/pH = 7.8 buffer (7:3)1074142a1:1.5:1.8CHCl3/pH = 8.0 buffer (7:3)1081152a1:1.5:1.8CHCl3/pH = 9.0 buffer (7:3)1079162b/2c/2d1:1.5:1.8CHCl3/pH = 8.0 buffer (7:3)1055/21/trace172a1:1:1CHCl3/pH = 8.0 buffer (7:3)1052182a1:1.2:1.8CHCl3/pH = 8.0 buffer (7:3)1062aReaction conditions: 1a (0.2 mmol), 2a (0.3 mmol), 3a (0.36 mmol), and solvent (1 mL) under an argon atmosphere for 24 hours unless otherwise specified.bIsolated yield.Scope of aldehydesHaving optimized the model coupling of this Passerini-type reaction, we examined the generality of these conditions with respect to a range of aldehyde components (Fig. 2). Delightfully, diverse aliphatic aldehydes were aptly transformed in moderate to high yields. Phenylpropyl aldehydes with strong electron-withdrawing groups and 3-(furan-2-yl)propanal furnished the α-hydroxyketone products 4b–4d in 66% to 90% yields. The chain length of aldehydes posed no effect on the effectiveness of this reaction, providing respective α-hydroxyketones (4e–4g) in moderate yields. Primary aldehydes bearing ester, adamantyl, and benzyloxy moieties were tolerated well to yield 4h–4j in moderate efficiencies. Secondary aldehydes comprised of acyclic and cyclic analogs (cyclopropyl, cyclohexyl, piperidinyl) were incorporated in 4k–4q with moderate to good yields as well. The diastereomeric ratios (dr) of compounds 4l and 4n are 1.13:1 and 1.38:1. Comparable outcome was observed for a tertiary 1-phenylcyclobutane-1-carbaldehyde substrate, which afforded 4r in 54% yield. It merits mention that transformation of paraformaldehyde has given rise to 4s, which serves as versatile synthetic intermediate for a variety of bioactive molecules. More importantly, this reaction was well suited to diverse aromatic aldehydes when treated in concert with cyclohexyl isocyanide (2b). The electronic property and the position of substituents on the benzene ring had minimal bearing on the efficiency of this transformation. Neutral (4t), electron-rich (4u–4y), or electron-deficient (4z–4aa) functionalities found good compatibility and were left unscathed in respective molecular outputs. The accommodation of halogen substituents (4ab–4ae) signified potential structural elaborations from these handles. Fused ring reactants including 2-naphthaldehyde (4af) and 1-naphthaldehyde (4ag) were also suitable candidates for this MCR.Fig. 2Scope of aldehydesa.Reaction conditions: aaldehyde 1 (0.2 mmol, 1 equiv), 2a (0.3 mmol), 3a (0.36 mmol), and CHCl3/pH = 8 buffer (7:3, 1 mL) under an argon atmosphere for 24 hours unless otherwise specified; bThe dr was determined by 1H NMR analysis. caldehyde 1 (0.2 mmol, 1 equiv), 2b (0.3 mmol), 3a (0.36 mmol), and DCM/pH = 8 buffer (7:3, 1 mL) under an argon atmosphere for 24 hours.Scope of boronic acidsThis protocol featured an admirable scope with respect to arylboronic acid substrates (Fig. 3). For electron-rich congeners, good reactivities were exhibited. Arylboronic acids with electronically neutral meta-para-dimethyl, para-methyl, and para-tertbutyl substituents produced α-hydroxyketones 5a–5c in moderate yields. Analogs with electron-rich substituents such as acetal, alkoxy, and diphenylamino groups reacted smoothly towards products 5d–5l in 51–85% yields. Inclusion of alkenyl or alkynyl group was noteworthy; from which products 5j and 5k were acquired in 80% and 77% yield. This study was auspiciously and effortlessly extendable to a series of heteroarylboronic acids containing furan (5m), thiophene (5n–5p), benzofuran (5q), benzothiophen (5r), protected or unprotected indoles (5s, 5u), 7-azaindole (5t), dibenzothiophene (5w) and carbazole (5x) cores. Remarkably, both aryl and alkyl substituted alkenylboronic acids could rendered the corresponding α-hydroxy enones 5y and 5z in 82% and 63% yields, which broadly expand the scope of the products. For electron-deficient substituted boronic acids, such as the halobenzene boronic acids, only trace amounts of products could be obtained, which probably is due to their low nucleophilicity that cannot capture the nitrilium intermediates. Aliphatic boronic acids, such as phenethylboronic acid and cyclopentylboronic acid, do not react under our standard conditions, perhaps owing to the lack of π electrons which makes 1,4-alkyl shift difficult68.Fig. 3Scope of boronic acidsa.Reaction conditions: a1a (0.2 mmol, 1 equiv), 2a (0.3 mmol), 3 (0.36 mmol), and CHCl3/pH = 8 buffer (7:3, 1 mL) under an argon atmosphere for 24 hours unless otherwise specified. b1 (0.2 mmol, 1 equiv), 2a (0.3 mmol), 3 (0.36 mmol), and DCM/pH = 8 buffer (7:3, 1 mL) under an argon atmosphere for 24 hours.Passerini-type reaction of alkynylboron compoundsα-Hydroxy alkynylketones are important intermediates for the synthesis of natural products and drug molecules82,83. However, the synthesis of such α-hydroxyketones has faced significant challenges and usually multiple steps are required82,83. We sought to explore the Passerini-type reaction on alkynylboron compounds, if successful, a straightforward and efficient method could be disclosed for the synthesis of α-hydroxy alkynylketones, which further demonstrates the strengths and capability of our protocol (Fig. 4). Alkynyl trifluoroborate salt was employed as the source of alkyne in our transformation owing to the instability of alkynylboronic acid. To our delight, the Passerini-type reaction of alkynyl trifluoroborate salt could proceed smoothly under the action of Lewis acid (Sc(OTf)3), and the target products (7a–7d) could be obtained in a moderate yield. This reaction could not occur without Lewis acid (Sc(OTf)3), probably because a four-coordinated boron “ate” nitrilium intermediate could not be generated from potassium phenyltrifluoroborate, aldehyde, and isocyanide. Lewis acid may promote the conversion of potassium phenyltrifluoroborate (6) into phenyldifluoroborane, which could form a four-coordinated boron intermediate84.Fig. 4Passerini-type reaction of alkynyl trifluoroborate salt.Reaction conditions: 1 (0.2 mmol, 1 equiv), 2a (0.5 mmol), 6 (0.6 mmol), and THF (1.5 mL) under an argon atmosphere for 12 hours unless otherwise specified.Late-stage modifications of complex moleculesThe excellent functional group compatibility prompted our endeavors to extrapolate this synthesis scheme to late-stage modification of bioactive or therapeutic agents (Fig. 5). A series of bioactive or drug molecules (Ibuprofen, Naproxen, Ketoprofen, Gemfibrozil, Indometacin, L-Menthol and L-Borneol, and Cholesterol) were derivatized into corresponding aldehydes which, upon treatment with 4-methoxyphenyl boronic acid under established Passerini-type coupling conditions, were smoothly incorporated in eventual α-hydroxyketone derivatives 8a–8h. Futhermore, conversions of arylboronic acids that were derived from drug molecules such as Epiandrosterone and Clofibrate had brought forth drug analogs 8i and 8j in moderate yields. It was thus envisioned that this method would simplify access to discover other bioactive molecules. The previous MCRs involving isocyanide exhibit poor stereoselectivity. This Passerini-type reaction of boronic acids showed similar results in terms of stereochemical control. In most cases (4l, 4n, 8a–8c, and 8f–8i), the dr values remained between 1:1 and 2.5:1 (see the Supplementary Information for details).Fig. 5Late-stage modifications of bioactive or drug molecules.Reaction conditions: a1a (0.2 mmol, 1 equiv), 2a (0.3 mmol), 3 (0.36 mmol), and CHCl3/pH = 8 buffer (7:3, 1 mL) under an argon atmosphere for 24 hours unless otherwise specified. bThe dr was determined by 1H NMR analysis. c1 (0.2 mmol, 1 equiv), 2a (0.3 mmol), 3 (0.36 mmol) and DCM/pH = 8 buffer (7:3, 1 mL) under an argon atmosphere for 24 hours. dThe dr was determined by HPLC analysis.Gram-scale synthesis and synthetic applicationsThe practical constraint of this Passerini-type MCR with boronic acids was next evaluated through translation to gram-scale synthesis. As shown in Fig. 6a, reaction efficiencies were preserved on 2 gram-scale (10 mmol, 50 times), thus implying the application potential for industrial production of the α-hydroxyketones.Fig. 6Gram-scale synthesis and synthetic applications.a Gram-scale synthesis. b Transformations of α-hydroxyketones. c Synthesis of Harmandianone. d Syntheis of α-acyloxy lactone. e Synthesis of α, β-unsaturated lactone.The readiness of α-hydroxyketone products for chemical manipulations was pronounced in production of 1,2-diol (9), 1,2-dione (10), quinoxaline (11), cyclic sulfamate imine (12), and poly-substituted oxazole (13) (Fig. 6b). The innate step economy of MCRs has also presented an abbreviated route towards (±) Harmandianone, a phenylpropanoid derivative isolated from Clausena Harmandiana fruits85, from simple building blocks (Fig. 6c). Of further significance, products could be precursors for entities that constitute the structural core of bioactive compounds such as α-acyloxy lactone and α,β-unsaturated lactone. The former (19) was fabricated upon lactonization of α-hydroxyketone 18 formed from methyl 4-oxobutanoate (17) (Fig. 6d). A two-step olefination and ring-closing olefin metathesis of 4a afforded α,β-unsaturated lactone product (±) 22 (Fig. 6e). The high diastereoselectivity of compound 20 may originate from the chelation of hydroxyl, carbonyl oxygen, and magnesium86.In conclusion, we have realized the application of boronic acid as carbon nucleophiles in the manifold of Passerini reaction. Accordingly, this protocol provided simplified modular access of α-hydroxyketones from aldehydes, isocyanide, and boronic acids. The functional group tolerance of this chemistry has supported late-stage diversifications of bioactive products and pharmaceuticals through this three-component coupling reaction. The wealth of follow-up chemical conversions that could be performed on procured α-hydroxyketones has additionally illustrated the utility of this method.MethodsGeneral procedure A for the synthesis of α-hydroxyketones from alkylaldehydesIn air, a 10 mL schlenk tube was charged with arylboronic acids (0.36 mmol, 1.8 equiv). The tube was evacuated and filled with argon for three cycles. Then, chloroform (0.7 mL), pH = 8 buffer (0.3 mL), alkylaldehydes (0.20 mmol, 1 equiv), tertbutyl isocyanide (34 μl, 0.30 mmol, 1.5 equiv) were added under argon. The reaction was allowed to stir at corresponding temperature for 24 hours. Upon completion, proper amount of silica gel was added to the reaction mixture. After removal of the solvent, the crude reaction mixture was purified on silica gel (petroleum ether and ethyl acetate) to afford the desired products.General procedure B for the synthesis of α-hydroxyketones from arylaldehydesIn air, a 10 mL schlenk tube was charged with arylboronic acids (0.36 mmol, 1.8 equiv). The tube was evacuated and filled with argon for three cycles. Then, dichloromethane (0.7 mL), pH = 8 buffer (0.3 mL), arylaldehydes (0.20 mmol, 1 equiv), cyclohexyl isocyanide (37 μl, 0.30 mmol, 1.5 equiv) were added under argon. The reaction was allowed to stir at room temperature for 24 hours. Upon completion, proper amount of silica gel was added to the reaction mixture. After removal of the solvent, the crude reaction mixture was purified on silica gel (petroleum ether and ethyl acetate) to afford the desired products.General procedure C for the synthesis of α-hydroxyketones from alkynyl trifluoroborate saltIn air, a 10 mL schlenk tube was charged with alkynyl trifluoroborate salt (0.60 mmol, 3 equiv) and Sc(OTf)3 (30.0 mg, 0.06 mmol, 0.3 equiv). The tube was evacuated and filled with argon for three cycles. Then, THF (1.5 mL), aldehydes (0.20 mmol, 1 equiv), and tertbutyl isocyanide (57 μl, 0.50 mmol, 2.5 equiv) were added under argon. The reaction was allowed to stir at room temperature for 12 hours. Upon completion, proper amount of silica gel was added to the reaction mixture. After removal of the solvent, the crude reaction mixture was purified on silica gel (petroleum ether and ethyl acetate) to afford the desired products.Supplementary informationSupplementary InformationPeer Review File | nature communications | [
"Article"
] | [
"Synthetic chemistry methodology",
"Combinatorial libraries",
"Reaction mechanisms"
] | -Hydroxyketones acyloins found in natural-employed synthetic precursors high-value transformations. 1a construction synthetic Traditional benzoin condensation method assembles α-hydroxyketones limits applicability substrate alternate oxidative pathways α-hydroxylation of ketones18–22 ketohydroxylation of olefins23–27 challenged substrate diversity poor selectivity complementary routes towards entities from starting materials relevant.Fig. Passerini-type coupling reaction with boronic acids as nucleophilic agents α-Hydroxyketones in bioactive synthetic precursors Passerini Ugi reaction boronic acid-Mannich reaction Passerini-type coupling reaction boronic acids reactions concise modular features complex molecules Passerini reaction37–46 Ugi reaction47–54 α-acyloxyamides α-acylaminoamides from reactant components nitrilium species single-pot operation (Fig. 1b)55–58 Interception electrophilic intermediate in Passerini reaction by carbon nucleophiles carboxylic acids access to α-hydroxyketone products underexplored59chemistry choosing carbon nucleophiles nitrilium intermediate capture electrophile boronic acids available blocks C-C bond cross-coupling reactions transition-metal metal-free catalysis boronic acid-Mannich reaction nucleophilic feature effects boron “ate” complex functionalized amines 1,3-metallate migration (Fig. 1c recent endeavor unraveled 1,4-metallate shift boron “ate” nitrilium nitrile oxide arylboronic acid mediating formation C-C bond between oxime chlorides arylboronic acids metal-free envisioned boron “ate” nitrilium intermediate released from co aldehyde isocyanide boronic acid 1,4-metallate shift C-C bond coupling α-hydroxyketones revealed on hydrolysis (Fig. development Passerini-type coupling reaction α-hydroxyketones from aldehydes isocyanides boronic acids under transition-metal-free conditions Mild reaction conditions ease execution high functional group tolerance broad substrate utility features methodologyResultsInvestigation reaction Passerini-type reaction conducted with phenylpropyl aldehyde tertbutyl isocyanide (2a 4-methoxyphenyl boronic acid (3a) substrates (Table 1) mixing reactants (1a 2a 3a) additive α-hydroxyketone product 4a 60% yield solvent screen DCE MeCN toluene MeOH THF best reaction efficiency CHCl3 MeOH inhibition 1−7) temperature decreased 10 °C yield 4a improved 68% mixture CHCl3 water 7:3 9−11) enhanced 4a 72% yield guided study mixed solvent system CHCl3 buffer solutions 12−15) pH = 8.0 buffer provided 81% yield reaction medium sequester byproduct B(OH)3 boronic acid Passerini-type reaction replacement tertbutyl isocyanide (2a) with isocyanide benzyl isocyanide ethyl 2-isocyanoacetate (2d), efficiency α-hydroxyketone product 4a diminished other ratios higher yields 17−18).Table 1Optimization reaction:2:3aSolventTemp.(°C)Yield (3:1.8CHCl3/H2O (7:1.8CHCl3/pH 6.5 buffer (7:3)1066132a1:1.5 buffer (7 8.0 (7 9.0 buffer (7:3)1079162b/2c/2d1:1.5:1.8CHCl3/pH 8.0 buffer (7:3)1055/21/trace172a1:1:1CHCl3/pH 8.0 buffer (7:3)1052182a1:1.2:1.8CHCl3/pH 8.0 buffer (7 conditions 1a (0.2 mmol), 2a (0.3 3a (0.36 solvent (1 mL) argon atmosphere 24 hours yield optimized coupling examined conditions aldehyde components (Fig. 2) diverse aliphatic aldehydes transformed moderate to high yields Phenylpropyl aldehydes electron-withdrawing groups 3-(furan-2-yl)propanal α-hydroxyketone products 4b–4d 66% to 90% yieldschain length aldehydes α-hydroxyketones (4e–4g moderate yields Primary aldehydes ester adamantyl benzyloxy 4h–4j moderate Secondary aldehydes acyclic cyclic piperidinyl 4k–4q moderate good yields diastereomeric ratios 4l 4n 1.13:1 1.38:1 Comparable outcome tertiary 1-phenylcyclobutane-1-carbaldehyde substrate 4r 54% yield transformation paraformaldehyde 4s versatile synthetic intermediate bioactive molecules reaction suited aromatic aldehydes cyclohexyl isocyanide electronic property position substituents benzene ring minimal efficiency Neutral electron-rich electron-deficient (4z–4aa functionalities compatibility unscathed halogen substituents (4ab–4ae potential structural elaborations Fused ring reactants 2-naphthaldehyde (4af 1-naphthaldehyde (4ag suitable.Fig. 2Scope conditions aaldehyde 1 (0.2 2a (0.3 3a (0.36 CHCl3/pH = 8 buffer (7:3 1 mL) argon atmosphere 24 hours dr determined 1H NMR analysiscaldehyde 1 (0.2 mmol 2b (0.3 3a (0.36 DCM/pH 8 buffer (7:3 1 mL argon atmosphere 24 hours boronic protocol admirable scope arylboronic acid substrates (Fig. 3) electron-rich congeners good reactivities Arylboronic acids neutral-para-dimethyl-methyl-tertbutyl substituents produced α-hydroxyketones 5a–5c moderate yields Analogs electron-rich acetal alkoxy diphenylamino products 5d–5l 51–85% yields alkenyl alkynyl group products 5j 5k 80% 77% yield study extendable heteroarylboronic acids furan thiophene benzofuran benzothiophen indoles 7-azaindole dibenzothiophene carbazole aryl alkyl substituted alkenylboronic acids α-hydroxy enones 5y 5z 82% 63% yields scope electron-deficient acids halobenzene trace amounts low nucleophilicity nitrilium intermediates Aliphatic boronic acids react standard conditions lack π electrons 1,4-alkyl shiftboronic conditions a1a (0.2 mmol 2a (0.3 3 (0.36 CHCl3/pH = 8 buffer (7:3, 1 mL argon atmosphere 24 hours b1 (0.2 mmol 2a (0.3 3 (0.36 DCM/pH = 8 buffer (7:3 1 mL argon 24 hours-type reaction alkynylboron compoundsα-Hydroxy alkynylketones synthesis natural products drug synthesis challenges multiple steps Passerini-type reaction alkynylboron compounds successful efficient method synthesis α-hydroxy alkynylketones strengths protocol (Fig. 4) Alkynyl trifluoroborate salt source alkyne instability alkynylboronic acid Passerini-type reaction Lewis acid (Sc)3) target products (7a–7d) obtained moderate yield without Lewis acid four-coordinated boron nitrilium intermediate generated potassium phenyltrifluoroborate aldehyde isocyanide Lewis acid conversion potassium phenyltrifluoroborate four-coordinated boron. 4Passerini-type reaction alkynyl trifluoroborate saltReaction conditions 1 (0.2 2a (0.5 6 (0.6 THF (1.5 mL argon 12 hours-stage modifications complex functional compatibility synthesis bioactive agents molecules (Ibuprofen Naproxen Ketoprofen Gemfibrozil Indometacin L-Menthol L-Borneol Cholesterol derivatized aldehydes 4-methoxyphenyl boronic acid coupling incorporated α-hydroxyketone derivatives 8a–8h conversions arylboronic acids Epiandrosterone Clofibrate drug analogs 8i 8j moderate method bioactive molecules MCRs isocyanide poor stereoselectivity Passerini-type reaction similar results stereochemical control cases (4l 4n 8a–8c dr values 1:1 2.5:1. 5Late-stage modifications bioactive drug molecules conditions (0.2 mmol 2a (0.3 3 (0.36 CHCl3/pH = 8 buffer (7:3 1 mL argon 24 hours dr determined 1H NMR analysis (0.2 mmol 2a (0.3 3 (0.36 DCM/pH = 8 buffer (7:3 1 mL argon 24 hoursdetermined by HPLC analysis-scale synthesis constraint Passerini-type MCR boronic acids evaluated translation gram-scale synthesis. 6a reaction efficiencies preserved 2 gram-scale (10 mmol 50 potential industrial production α-hydroxyketones-scale synthesis synthetic applications Gram-scale synthesis Transformations α-hydroxyketones Synthesis Harmandianone α-acyloxy lactone α β-unsaturated lactone readiness α-hydroxyketone products chemical manipulations production 1,2-diol 1,2-dione quinoxaline sulfamate imine poly-substituted oxazole step economy MCRs abbreviated route towards Harmandianone derivative Clausena Harmandiana products precursors for bioactive compounds α-acyloxy lactone α,β-unsaturated lactone fabricated lactonization α-hydroxyketone 18 methyl 4-oxobutanoate two-step olefination-closing olefin metathesis α,β-unsaturated lactone product high diastereoselectivity compound 20 chelation hydroxyl carbonyl oxygen application boronic acid carbon nucleophiles in Passerini reactionprotocol provided access α-hydroxyketones from aldehydes isocyanide boronic acids functional group tolerance late-stage diversifications bioactive products chemical conversions on α-hydroxyketones utility method procedure synthesis α-hydroxyketones from 10 mL tube charged with arylboronic acids (0.36 mmol 1.8 filled argon three cycles chloroform (0.7 mL), pH = 8 buffer (0.3 alkylaldehydes (0.20 mmol 1 tertbutyl isocyanide (34 μl 0.30 1.5 added under argon 24 hours silica gel added mixture purified on silica gel ether ethyl acetate) B synthesis α-hydroxyketones from 10 mL tube charged with arylboronic acids (0.36 mmol 1.8 filled argon three cycles dichloromethane (0.7 mL), pH = 8 buffer (0.3 arylaldehydes (0.20 mmol 1 isocyanide (37 μl 0.30 1.5 added argon temperature 24 hours silica gel addedremoval solvent crude mixture purified silica gel ether ethyl acetate products procedure synthesis α-hydroxyketones alkynyl trifluoroborate 10 mL tube charged alkynyl trifluoroborate salt (0.60 mmol 3 Sc(OTf)3 (30.0 mg 0.06 mmol 0.3 evacuated filled argon three cycles THF (1.5 aldehydes (0.20 mmol 1 tertbutyl isocyanide (57 μl 0.50 mmol 2.5 equiv added argon room temperature 12 hours silica gel added purified silica gel products Review File | 51.5 | 0.321891 |
10.1038/s41467-020-17212-6 | PMC7359353 | Melanosomes traffic along F-actin in melanocytes. Here, the authors show that Rab27a coordinates SPIRE/FMN actin assembly and MyoVa motor proteins to generate a cell-wide actin/myosin network that links melanosomes and allows the collective activity of these force generators to drive their traffic. | Cell biologists generally consider that microtubules and actin play complementary roles in long- and short-distance transport in animal cells. On the contrary, using melanosomes of melanocytes as a model, we recently discovered that the motor protein myosin-Va works with dynamic actin tracks to drive long-range organelle dispersion in opposition to microtubules. This suggests that in animals, as in yeast and plants, myosin/actin can drive long-range transport. Here, we show that the SPIRE-type actin nucleators (predominantly SPIRE1) are Rab27a effectors that co-operate with formin-1 to generate actin tracks required for myosin-Va-dependent transport in melanocytes. Thus, in addition to melanophilin/myosin-Va, Rab27a can recruit SPIREs to melanosomes, thereby integrating motor and track assembly activity at the organelle membrane. Based on this, we suggest a model in which organelles and force generators (motors and track assemblers) are linked, forming an organelle-based, cell-wide network that allows their collective activity to rapidly disperse the population of organelles long-distance throughout the cytoplasm. | IntroductionIn animal cells, unlike plants and yeast, microtubules (MTs) and actin filaments (AFs) are thought to regulate transport in a manner akin to the infrastructure of a developed nation1–4. This ‘highways and local roads’ model suggests that MTs are tracks for long-range transport (highways) between the cell centre and periphery, driven by kinesin and dynein motors. Meanwhile AFs (local roads) and myosin motors work downstream picking up cargo at the periphery, and transporting it for the ‘last μm’ to its final destination. This model makes intuitive sense as MTs in many cultured animal cells form a polarised radial network of tracks spanning >10 μm from the centrally located centrosome to the periphery and appear ideally distributed for long-range transport. Meanwhile, with some exceptions in which AFs form uniformly polarised arrays, e.g., lamellipodia, filopodia and dendritic spines, AF architecture appears much more complex. In many cells AF appear to comprise populations of short (1–2 μm length), with random or anti-parallel filament polarity, and not an obvious system of tracks for directed transport5,6.This view is exemplified by the co-operative capture (CC) model of melanosome transport in melanocytes7,8. Skin melanocytes make pigmented melanosomes and then distribute them, via dendrites, to adjacent keratinocytes, thus providing pigmentation and photo-protection (reviewed in ref. 9). The CC model proposes that transport of melanosomes into dendrites occurs by sequential long-distance transport from the cell body into dendrites along MTs (propelled by kinesin/dynein motors), followed by AF/myosin-Va-dependent tethering in the dendrites. Consistent with this, in myosin-Va-null cells melanosomes move bidirectionally along MTs into dendrites, but do not accumulate therein, and instead cluster in the cell body7,10. This defect results in partial albinism in mammals due to uneven pigment transfer from melanocytes to keratinocytes (e.g., dilute mutant mouse and human Griscelli syndrome (GS) type I patients; Fig. 1a)11,12. Subsequent studies revealed similar defects in mutant mice (and human GS types II and III patients) lacking the small GTPase Rab27a (ashen) and its effector melanophilin (Mlph; leaden) which recruit and activate myosin-Va on melanosomes8,13,14.Fig. 1FMN1- and SPIRE1/2-deficient melanocytes show perinuclear melanosome clustering.a A schematic representation of the strategy used to identify regulators of myosin-Va/dynamic AF-dependent melanosome transport. In wild-type melanocytes melanosomes are dispersed throughout the cytoplasm. Loss of myosin-Va (or its regulatory proteins, e.g., Rab27a) or pharmacological disruption of dynamic AFs blocks melanosome dispersion, resulting in perinuclear melanosome clustering 15. To identify AF-associated proteins working with myosin-Va, we used siRNA to deplete known AF regulators and screened for targets whose depletion phenocopied the loss of myosin-Va/depletion of dynamic AFs, i.e., caused perinuclear clustering of melanosomes. b melan-a cells were transfected with the indicated siRNA fixed 72 h cells later and imaged using bright-field optics to observe melanosome distribution (see Experimental procedures). c A bee swarm plot showing the extent of melanosome dispersion in individual cells transfected with the indicated siRNA (n (cells) = 28 (NT), 18 (Rab27a), 20 (FMN1) and 22 (SPIRE1/2)). Horizontal bars indicate the populations that are being compared. n.s. indicates no significant difference between the populations as determined by one-way ANOVA. All other comparisons yielded highly statistically significant differences p = <0.0001. d A western blot showing the expression of FMN1 and GAPDH (loading control) in whole cell lysates of melan-a and melan-f melanocytes. e A bee swarm plot showing the extent of melanosome dispersion in adenovirus-infected melan-f cells expressing the indicated proteins (n (cells) = 14 for all conditions). **** indicates a statistically significant difference (p = <0.0001) between this population and the others. No other statistically significant differences were observed. f Confocal micrographs showing the distribution and effect of GFP-FMN1 expression on melanosome distribution in melan-f cells. White dotted boxes in images indicates the region shown in high-magnification overlay image (GFP-FMN1 in green and melanosomes in magenta). g melan-a cells were transfected with the indicated siRNA. After 72 h cells were infected with adenovirus expressing GFP or GFP-SPIRE1/2 (human), fixed 24 h later and processed for immunofluorescence. Cells were then imaged using bright-field and fluorescence optics to observe melanosome and GFP distribution. Asterisks indicate cells with hyper-dispersed melanosome distribution. h A bee swarm plot showing the percentage of SPIRE1/2-depleted/adenovirus-infected melan-a cells in low-magnification (10×) fields of view, in which melanosomes were dispersed and hyper-dispersed. Scale bars = 50 μm (b, g) and 21 μm (g magnified portion), 10 μm (f) and 4 μm (f magnified portion). c, e, h **** and *** indicates statistical significance of differences of p = <0.0001 and p = <0.001 as determined by one-way ANOVA. Significance indicators above datasets indicate differences compared with GFP control. The horizontal bar indicates the datasets that are being compared. No other significant differences were observed. Bars indicate the mean and 25th and 75th percentile of data. Source data for c, d, e and h are provided in the Supplementary Source data file.Previously, we tested the CC model by using cell normalisation technology to quantitatively examine the contribution of MTs and AF/myosin-Va to melanosome transport15. Surprisingly, our results indicated that MTs are essential for perinuclear clustering, but not peripheral dispersion of melanosomes. Instead we found that MTs retard dispersion, which is dependent upon myosin-Va and a population of dynamic AFs. Functional analysis of mutant proteins suggested that myosin-Va works as a processive motor dispersing melanosomes along AFs whose +/barbed ends are oriented away from melanosomes and towards the cell periphery. Finally, using an activatable motor to directly monitor melanosome dispersion in myosin-Va-null cells in real time, we found that myosin-Va can disperse melanosomes rapidly (~1 μm/min) into peripheral dendrites (>10 μm) even in MT-depleted cells. Overall our data highlighted the role of myosin-Va and dynamic AFs in long-range transport, rather than tethering, and suggest that melanosome distribution is determined by the balance of MT-dependent clustering and long-range AF/myosin-Va-dependent dispersion. However, studies of AFs organisation in melanocytes have not revealed the existence of a polarised network that would seem requisite for myosin-Va-driven transport and melanosome dispersion7. Thus, the mechanism of this process remained unclear.Here, we investigated this issue and identify SPIRE1/2 and formin-1 (FMN1) AF assembly proteins as essential regulators of myosin-Va-driven melanosome dispersion. FMN1 is one of 15 mammalian formins that nucleate and elongate unbranched AFs (refs. 16,17). FMN1 and FMN2 comprise the FMN subfamily of formins18. FMN1 function is linked to limb development, neurogenesis and spermatogenesis, while FMN2 acts in oocyte development, and memory and learning19–25. Like other formins, FMNs contains two formin homology domains (FH1 and FH2) that drive AF assembly17. Dimeric FH2 forms a ring that encircles the +/barbed ends of AFs, while the proline rich FH1 interacts with profilin in complex with G-actin promoting filament elongation26–28. The FMNs are characterised by a short (~30 amino acids) C-terminus FH2 tail that mediates the interaction with SPIRE proteins (termed FSI; formin/SPIRE interaction sequence)29–31. The N-terminal regions of the FMN1/2 formins are large (859 amino acids in murine FMN1), but contain no conserved sequence motifs.Mammalian genomes encode two SPIRE genes, SPIRE1 and SPIRE2. The SPIRE actin nucleators are modular proteins that contain an N-terminus AF nucleation module, comprising a KIND (kinase non-catalytic C-lobe domain) that interacts with the FSI motif of FMN1/2, and four G-actin-binding WH2 (WASP-homology 2) domains29,32,33. This is coupled to a C-terminus membrane-binding domain, comprised of SB (SPIRE box, conserved among SPIRE proteins) and FYVE-type zinc finger (Fab1p, YOTB, Vac1 and EEA1) domains34,35. The SPIRE box has sequences similarities to the N-terminal a-helix of Rab GTPase-binding domain (RBD) of the synaptic vesicle transport regulator Rabphilin-3A (RPH3A). Thus, FMN and SPIRE proteins may collaborate to assemble AFs at organelle membranes22,23. Previously, their combined function has been implicated in regulating oocytes development and repair of DNA damage22,36,37. In mouse oocytes, SPIRE1 and SPIRE2 cooperate with FMN2 to generate AFs for myosin-Vb-driven cortical transport of Rab11 vesicles22,23. More recently work has identified a myosin-V globular tail domain binding motif (GTBM) located between the N- and C-terminus modules of SPIRE proteins, which may co-ordinate the recruitment of myosin-V and AF assembly to Rab11-positive intracellular membranes38.Here, we present evidence that the myosin-Va-mediated melanosome transport/dispersion in melanocytes is dependent upon AF assembly activities of FMN1 and SPIRE1/2, and that SPIRE1/2 (predominantly SPIRE1) can be recruited to melanosomes by Rab27a. Based on these findings, we propose a cargo-driven model of organelle dispersion in which Rab27a plays a central role co-ordinating the function of both AF motors and assembly proteins.ResultsFMN1, SPIRE1 and SPIRE2 are required for melanosome dispersion in melanocytesTo better understand myosin-Va/dynamic AF-based melanosome dispersion in melanocytes, we used siRNA knockdown to test the involvement of known AF regulatory proteins in this process. For this we transfected wild-type melanocytes (melan-a) with an siRNA mini-library comprised of 130 pools (four target-specific oligonucleotide duplexes/pool) directed against the transcripts of known AF regulators (Supplementary Table 1). We then visually screened the transfected cells to identify siRNA that induced perinuclear melanosome clustering, reasoning that knockdown of proteins working with myosin-Va/dynamic AF should result in dispersion defects like those seen in myosin-Va/dynamic AF-deficient cells (Fig. 1a). Consistently (5/5 transfections), we found that knockdown of FMN1, and double knockdown of its interacting partners SPIRE1 and SPIRE2, induced melanosome clustering like that seen for Rab27a knockdown (part of a myosin-Va receptor at the melanosome), albeit that the extent of clustering was significantly lower (Fig. 1b, c; mean pigment area (% total); NT = 86 ± 10.28%, Rab27a = 31.28 ± 8.242%, FMN1 = 45.07 ± 8.061% and SPIRE1/2 = 52.56 ± 6.868%). [Transfection with five of the other siRNA pools caused melanosomes clustering in melanocytes in >2/5 experiments. These targets were not investigated further.]Interestingly, although quantitative real-time-PCR (Q-RT-PCR) confirmed that single siRNA transfection reduced mRNA for SPIRE1 and SPIRE2, only SPIRE1 knockdown resulted in a significant decrease in the proportion of cells with dispersed melanosomes compared with control (NT) siRNA-transfected cells (Supplementary Fig. 1a–e; dispersed melanosomes (% cells); NT = 86.43 ± 4.887% versus SPIRE1 = 46.93 ± 10.64% and SPIRE2 = 82.57 ± 13.26%). Nevertheless, the effect of SPIRE2 depletion could be seen by the significantly lower proportion of cells with dispersed melanosomes seen in SPIRE1/2 versus SPIRE1 alone depletion (Supplementary Fig. 1b–e; mean % of cells with dispersed melanosomes; SPIRE1/2 = 12.73 ± 5.878%). In addition, in a subset of SPIRE1-depleted cells we saw that melanosomes were cleared from the cell centre and enriched at the periphery. This pattern (termed ‘hyper-dispersed’ (HD)) differed from the uniform dispersion pattern seen in control NT siRNA-transfected cells, SPIRE2 or SPIRE1/2-depleted cells (Supplementary Fig. 1d, e; mean HD melanosomes (% total cells) = 6.324 ± 2.645). These data indicate that SPIRE1 plays a dominant role in establishing the uniform (physiological) melanosome dispersion pattern seen in wild-type melanocytes. Consistent with this Q-RT-PCR showed that FMN1 and SPIRE1 mRNA expression levels were comparable and exceeded that of SPIRE2 by fivefold (Supplementary Fig. 1f; mRNA copies (×103)/50 ng total RNA; SPIRE1 = 17.33 ± 1.883, SPIRE2 = 3.468 ± 0.5726 and FMN1 = 27.2 ± 4.17). We were unable to detect FMN2 expression using Q-RT-PCR, indicating that FMN2 is unlikely to be expressed in melanocytes. This suggests that FMN1 and SPIREs (predominantly SPIRE1) cooperate to disperse melanosomes in melanocytes.We next used add-back experiments, in which human SPIRE1 and SPIRE2 (siRNA resistant), and mouse FMN1 proteins were expressed in SPIRE1/2 depleted melan-a and immortal FMN1-deficient melanocytes (melan-f) to confirm that melanosome clustering in siRNA experiments was specific for depletion of the expected target (Fig. 1d–h; Fig. 1e, mean pigment area (% total), –protein expression = 50.15 ± 9.228, GFP = 48.66 ± 8.618, FMN1 = 88.43 ± 7.972; Fig. 1h, mean % of cells with dispersed melanosomes (total); SPIRE1/2 knockdown with GFP expression = 11.83 ± 8.02, SPIRE1 = 92.58 ± 7.072, SPIRE2 = 94.67 ± 11.36). Interestingly, we saw that in cells expressing lower levels of SPIRE2 melanosomes were hyper-, rather than uniformly, dispersed (Fig. 1h, Supplementary Fig. 1g; mean % of cells with HD melanosomes; GFP = 3.088 ± 2.748, SPIRE1 = 0.4695 ± 0.8132, SPIRE2 = 49.96 ± 11.44). This indicates that SPIRE2 is less efficient in uniformly dispersing melanosomes compared with SPIRE1.SPIRE1/2 and FMN1 generate dynamic AFs that are essential for melanosome dispersalAs FMN and SPIRE cooperate in AF assembly, we hypothesised that they collaborate in melanocytes to assemble dynamic AFs used by myosin-Va to disperse melanosomes18. To test this, we used latrunculin-A to deplete dynamic AFs in SPIRE1/2 and control NT siRNA-transfected melan-a and melan-f cells, and then observed the effects on melanosome distribution and melanosome-associated AF content. We found that this treatment reduced melanosome dispersion and AF content in control melan-a cells (as before), but not further reduce these parameters in melan-f and SPIRE1/2-depleted melan-a cells15 (Fig. 2a–d; mean pigment area (% total) ± SD in the absence and presence of latrunculin-A; melan-a = 68.26 ± 9.956 versus 51.56 ± 12.14; melan-f = 45.48 ± 11.9 versus 45.89 ± 13.29; NT-transfected melan-a = 85.18 ± 8.549 versus 43.94 ± 16.09; SPIRE1/2-depleted melan-a 36.9 ± 9.668 versus 35.28 ± 10.26; Fig. 2e, f; melanosome-associated AF content integrated density (AU) ± SD; melan-a = 413.7 ± 186.1 versus 167.6 ± 75.5; melan-f = 241.6 ± 112.3 versus 241.9 ± 93.26; NT-transfected melan-a = 767.3 ± 331.9 versus 304.8 ± 237.8; SPIRE1/2 siRNA-transfected melan-a = 304.7 ± 145.7 versus 213.2 ± 74.68). Conversely, re-expression of GFP-Fmn1, but not GFP, increased AF content in melan-f cells (Fig. 2g, h; integrated density (AU) ± SD; GFP-Fmn1 = 128.6 ± 71.79 versus GFP = 47.13 ± 21.52). These data indicate that FMN1 (and SPIRE1/2) are important factors for the assembly of dynamic AFs that support myosin-Va-dependent melanosome dispersion.Fig. 2FMN1 and SPIRE1/2 generate latrunculin-A-sensitive AFs essential for melanosome dispersion.melan-a and melan-f cells were plated onto glass coverslips, and transfected with siRNA as indicated: c, d, f, infected with GFP-FMN1 expressing virus g, h and/or incubated with latrunculin-A (lat-A) for 60 min a–f as indicated. Cells were then fixed and stained with fluorescent phalloidin to reveal AFs (see Experimental procedures). a, c, g Fluorescent and bright-field images showing the distribution of AFs and melanosomes in melanocytes. Scale bar = 15 μm. b, d, e, f, h Bee swarm plots showing the extent of melanosome dispersion (b, d) and AF abundance (e, f, h) in melanocytes, in the presence and absence of latrunculin-A. b and e, and d and f show data from the same population of cells. The number of cells measured in each case is indicated in brackets in the bee swarm plot associated with that data (b, d, f, h). Numbers of cells in e and f are the same as in b and d. **** and *** indicate significant difference p = <0.0001 and p = <0.001 as determined by one-way ANOVA. n.s. indicates no significant difference. Data are from one of three independent experiments. Bars within each dataset indicate the mean and 25th and 75th percentile of data. Bars linking datasets indicate the pairs that are being compared for similarity. Source data for b, d–f, h are provided in the supplementary Source data file.To further investigate this, we used high-resolution field emission scanning electron microscopy (FESEM), and replica transmission EM to analyse the cytoskeleton network surrounding the melanosomes in melan-a cells and variants. In line with the results of fluorescence microscopy, high-magnification FESEM images revealed that melanosomes in melan-a cells were surrounded by a meshwork of filaments compatible with AFs on the basis of their diameter (mean filament diameter ± SD = 8.6 ± 0.3 nm, n = 87) and morphology (dense bundles and branched networks, Fig. 3a–c). Melanosomes are fully embedded in the filament network, with filaments visible above and below melanosomes (Fig. 3g, coloured yellow and orange, respectively), and in many cases appear as integral part of the network as multiple AFs can be seen originating from the surface of melanosomes, bridging adjacent organelles (Fig. 3g, coloured blue; Fig. 3k, arrows). In contrast in melan-f cells, melanosome-associated filaments were almost entirely absent and instead melanosomes were decorated with short filaments or ‘stubs’ (Fig. 3f, l, arrowheads). FESEM immuno-electron microscopy revealed that most of the filament network on, above and below the melanosomes was labelled by phalloidin, confirming that those filaments are part of a complex actin network (Fig. 3h, i). This was further confirmed by rapid freeze/freeze dry/replica TEM (ref. 39), which revealed that myosin S1 decorated filaments surrounding melanosomes (Fig. 3j, arrowheads). To quantify the differences in the melanosome-associated AFs, we further measured filament length of filaments emanating from melanosomes in FESEM images (Fig. 3k–m), confirming significantly shorter melanosome-associated filaments in melan-f cells. (Fig. 3m; mean filament length in nm ± SD: melan-a = 261.7 ± 160.6 (median = 223, n = 433) versus melan-f = 116.1 ± 72.32 (median = 103; n = 443); p < 0.0001). Similar differences in AF density and dimensions were observed in melan-a cells depleted of SPIRE1/2 and Rab27a (Supplementary Fig. 2). Given their dimensions, we suggest that these stubs correspond to short AFs that require FMN1 and SPIRE1/2 for extension into a network. These data support the hypothesis that SPIRE1/2 and FMN1 assemble AFs used by myosin-Va to disperse melanosomes and extend this idea by suggesting that these AFs are constructed at the melanosome membrane.Fig. 3High-resolution electron microscopy reveals a reduction in melanosome-associated AFs in FMN1-deficient melanocytes compared with controls.a–f Wild-type (melan-a, a–c) and FMN1-deficient (melan-f, b–f) cells were prepared for field emission scanning electron microscopy (FESEM; see Experimental procedures). Cells in a and d are shown with increased magnification, with high magnification of the insets (yellow) in c and f. The red line in d indicates the cell outline. Arrows in c point at a loose network of filaments around melanosomes; arrowheads show melanosomes on top of a dense filament network. g Colourised filaments in high-magnification FESEM images of melan-a melanosomes (pink) indicate filaments linking (blue), above (orange) and below (yellow) melanosomes. h, i Immuno-electron microscopy of melan-a cells viewed with FESEM/backscatter showing phalloidin labelling (10 nm gold particles) of actin filaments around and over melanosomes, as well as inter-melanosome filaments (arrowheads). Higher magnification in i shows colourised filaments (cyan) and gold particle labelling (yellow). j TEM of rapid freeze/freeze dry metal replica showing myosin S1 decoration of filaments (arrowheads) around melanosomes. g, k High-magnification FESEM showing melanosomes with multiple AFs emerging from them in melan-a cells (g, arrows) or AF stubs on melan-f melanosomes (l, arrowheads). m Bee swarm plot showing size distribution for AFs emanating from melanosomes, as measured on FESEM images. n (actin filaments) = 433 (melan-a) and 443 (melan-f). **** indicates significant difference p = <0.0001 between melan-a and melan-f cells as determined by Mann–Whitney test. Source data for i are provided in the supplementary Source data file. Bars indicate the mean and 25th and 75th percentile of data. Scale bars: a, b, d, 10 μm; c, e–g, 1 μm; h, k, l, 200 nm; i, j, 100 nm.The N-terminus AF nucleation (WH2) and FMN interaction (KIND) activities are essential for SPIRE function in melanosome transportTo examine further their role in melanocytes, we tested which SPIRE protein domains are essential for melanosome dispersion. We found that neither the AF nucleation module (KW: FMN interaction (KIND) and G-actin binding (WH2 cluster)) or the membrane-binding module (MSFH: myosin-V-binding domain (M), SPIRE box (S), FYVE (F) and C-terminus flanking sequence (H)) fragments uniformly dispersed melanosomes as efficiently as intact SPIRE1 (p = <0.001) in SPIRE1/2-depleted melanocytes. However KW dispersed melanosomes to a significantly greater extent than either MSFH or GFP alone (Fig. 4a–c; mean % of cells with dispersed melanosomes, KW = 58.85 ± 10.91 (p = <0.001), MSFH = 18.69 ± 4.37 and GFP = 4.65 ± 3.443%). Thus while both modules are essential for optimal SPIRE1 function, the AF nucleation module (KW) is the more significant element in driving melanosome dispersion. Interestingly, we noted that KW behaved similarly to SPIRE2 in promoting melanosome hyper-dispersion in a significant subset of cells (Fig. 4; mean % of cells with HD melanosomes KW = 36.69 ± 7.746; SPIRE2 = 35.91 ± 10.72; SPIRE1 = 0). This indicates that the SPIRE1 membrane targeting domain is essential for uniform cytoplasmic melanosome dispersion, and that SPIRE1 and SPIRE2 differ in their association with membranes.Fig. 4The FMN interaction (KIND) and AF nucleation (WH2) domains of SPIRE1 are essential for melanosome dispersion.a A schematic representation of the domain structure of human SPIRE1, and the correspondence with mutant and chimeric proteins used in functional studies (b, c). Numbers indicate amino acid boundaries. K KIND, W WH2, M GTBM globular tail domain binding motif, S SB SPIRE box, F FYVE-type zinc finger, H C-terminal flanking sequences similar to H2 of Slp/Slac-proteins. b melan-a cells were depleted of SPIRE1/2 by siRNA transfection and 72 h later infected with adenoviruses expressing the indicated proteins. Cells were fixed 24 h later, processed for immunofluorescence and imaged using bright-field and fluorescence optics to observe melanosome and protein distribution/expression (see Experimental procedures). Scale bars = 100, 20 and 3 μm in low, medium and high-magnification images. Boxes in KW-Rab27a images indicate the region shown below. For the merged image green = KW-Rab27a and magenta = melanosomes. c Is a bee swarm plot showing the percentage of human SPIRE1/2 expressing SPIRE1/2-depleted melan-a cells (50 cells for each condition in each experiment), in which melanosomes are classed as dispersed and/or hyper-dispersed. Results shown are from of three independent experiments. Source data for c are provided in the Supplementary Source data file.To investigate AF assembly module further, we tested WMSFH and KMSFH truncations lacking either the KIND (FMN interaction) or WH2 (G-actin interaction) domains. We found that neither truncation dispersed melanosomes in SPIRE1/2-depleted cells to a greater extent than GFP (Fig. 4b, c; mean % of cells with dispersed (total) melanosomes; WMSFH = 19.01 ± 1.693 and KMSFH = 23.08 ± 2.374). This indicates that cooperation between SPIRE1 and FMN1 is required for AF assembly in melanocytes. Similar results were seen in experiments using human SPIRE2 to rescue SPIRE1/2 depletion (Supplementary Fig. 3). The integrity of the expressed SPIRE1 and SPIRE2 proteins (and FMN1 proteins below) was confirmed by western blotting (Supplementary Fig. 4a–c).AF assembly and SPIRE interaction domains are required for FMN1 function in melanosome transportWe then used a similar approach to examine the role of FMN1 in melanocytes. We found that expression of a C-terminus fragment, encompassing the conserved FH1 and FH2 domains, and the FSI motif (FH1-FH2-FSI), restored peripheral melanosome distribution in melan-f cells to a similar extent as intact FMN1, but a reciprocal N-terminus fragment did not (Fig. 5; mean pigment area (% total); GFP = 50.69 ± 14.29%, FMN1 = 89.92 ± 6.598%, N-term = 51.44 ± 10.17%, FH1-FH2-FSI = 89.28 ± 8.512%). Truncations that removed the FSI or FH1 were unable to fully disperse melanosomes (Fig. 5; mean pigment area (% total); ΔFSI = 79.52 ± 12.34%, FH2-FSI = 42.77 ± 13.89%). Also we found that point mutations predicted to disrupt either the FH2 contact with AF +/barbed ends (I1074A and K1229D) or the electrostatic FSI/SPIRE-KIND interaction (K1418E) significantly reduced FMN1 function in transport (Fig. 5; mean pigment area (% total), I1074A = 57.77 ± 12.73%, K1229D = 77.75 ± 12.53 and K1418E = 72.1 ± 17.34 %)26,31. These results show that the FH1 and FH2 domains, and the FSI motif are essential for FMN1 function and indicate that FMN1 assembles AFs in collaboration with SPIRE in melanocytes.Fig. 5The AF assembly (FH1-FH2) and SPIRE interaction (FSI) domains of FMN1 are essential for melanosome dispersion.a A schematic representation of the domain structure of murine FMN1 and the composition of truncation mutants, and chimeric proteins used in functional studies (b, c). White asterisks indicates the site of point mutations. Numbers indicate amino acid boundaries. b melan-f cells were plated on glass coverslips and infected with adenoviruses expressing the indicated proteins. Cells were fixed 24 h later, processed for immunofluorescence and the intracellular distribution of expressed protein and melanosomes (bright-field) was observed using a fluorescence microscope (see Experimental procedures). Scale bars = 20 μm and 2 μm for magnified region. Dashed boxes in FH1-FH2-Rab27a images indicate the region of the image shown in high magnification below. The merged image show melanosomes and FH1-FH2-Rab27a coloured magenta and green. c A bee swarm plot showing the extent of pigment dispersion in cells expressing the indicated proteins. n = 30 (GFP), 59 (FMN1), 18 (N-term), 70 (FH1-FH2-FSI), 31 (ΔFSI), 20 (FH2-FSI), 37 (I1074A), 35 (K1229D), 28 (K1418E) and 51 (FH1-FH2-Rab27a). ****, **, * and n.s. indicate significant differences of p = <0.0001, 0.01, 0.05 and no significant difference as determined by one-way ANOVA. Significance indicators above and below each dataset indicate differences between that dataset and the positive (FMN1 wild type) and negative (GFP alone) controls. Results shown are representative of three independent experiments. Bars indicate the mean and 25th and 75th percentile of data. FH formin homology, FSI formin-SPIRE interaction sequence. Source data for c are provided in the Supplementary Source data file.The membrane-binding module of SPIRE is related to the Rab-binding domain of Rab27/3 effectorsWithin the mammalian Rab family (>60 genes), proteins of the Rab27/3/8 branch regulate the transport of exocytic vesicles towards the plasma membrane40. The Slp/Slac (synaptotagmin-like protein/Slp lacking C2 domain) class of effectors of these GTPases have a common Rab binding domain consisting of a FYVE-type zinc finger flanked by α-helical regions (termed H1 and H2 hereafter) that make direct and essential with contacts Rab3/27 (refs. 41,42). Previous studies reported sequence similarity between the SPIRE box (SB) of SPIRE proteins and the H1 helix of the of the Rab27/3 effector RPH3A (ref. 34). These observations together with evidence of Rab3a:SPIRE1 interaction in vitro and the presence of an adjacent FYVE-type zinc finger in the membrane-binding region suggest that SPIREs could be Rab27 effectors (Fig. 4a)43. Consistent with this our sequence alignments showed a high similarity between the SB and the H1 regions of several Rab27/3 effectors, including Mlph and conservation of residues important in Rab27/3 interaction among this group (Supplementary Fig. 5a coloured asterisks). Phylogenetic analysis of the sequences of the putative RBDs of SPIRE proteins with those of other containing Rab27/3/8 effectors grouped SPIRE proteins into this family (Supplementary Fig. 5b). These observations support the idea that SPIRE protein function in melanosome transport could be regulated by Rab27a.GTP-dependent interaction of SPIRE proteins with active Rab27aTo test this possibility, we performed GST pull-down experiments using bacterially expressed, purified GST-Rab27a fusion proteins (GTP-locked Rab27a-Q78L and GDP-locked Rab27a-T23N mutant proteins) and lysates of HEK293 cells transiently expressing Myc-epitope-tagged SPIRE1 and SPIRE2 (Fig. 6a–c). Western blotting revealed that GST-Rab27a-Q78L pulled-down greater quantities of Myc-tagged SPIRE1 and SPIRE2 compared with GST-Rab27a-T23N, and that SPIRE1 interacted more strongly with Rab27a-Q78L compared with SPIRE2. Similar results were obtained in experiments using wild-type GST-Rab27a loaded with GTPγS compared with GDPγS to pull-down GFP-SPIRE1-MSFH (Supplementary Fig. 6).Fig. 6SPIRE1/2 interact with active Rab27a via their membrane-binding C-termini.a A schematic representation of the domain structure of SPIRE1/2, Mlph and truncations used in interaction studies (b, c, e, f). Interaction of SPIRE1/2 and Rab27a was investigated using GST pull-down (b, c, e, f) and BiFC (d) assays (see Experimental procedures). b, c, e Western blots and Ponceau S stained filters showing the results of pull-down assays measuring the interaction of GST-Rab27a (active Q78L and inactive T23N mutants) with SPIRE1/2 and the indicated truncations (Myc-tagged (b, c) and GFP-tagged (e) in vitro). c is a contrast enhanced version of the section of b showing interaction of SPIRE2 with Rab27a-Q78L and Rab27a-T23N. d Fluorescence images and a bee swarm plot (upper and lower panels) showing the results of the BiFC assay, reporting the interaction of Rab27a with SPIRE1 and SPIRE2 in HEK293a cells (see Experimental procedures). Images of mCherry indicate transfection efficiency and vYFP indicates BiFC, i.e., interaction. The bee swarm plot shows the BiFC signal for populations of cells expressing SPIRE1/2 with and without active and inactive Rab27a mutants. Data shown are from three independent experiments. ****, *** and ** indicate significant differences of p = <0.0001, p = <0.001 and p = <0.01 between the adjacent dataset and Rab27a wild-type/SPIRE1/2 expressing cells as determined by one-way ANOVA of data for SPIRE1 and SPIRE2 with different Rab27a proteins. No other significant differences were observed. Two-way ANOVA comparison of BiFC signal for the Rab27a proteins with different SPIREs revealed no significant differences. Bars indicate the mean and 25th and 75th percentile of data. Scale bar = 250 μm. f Line plots showing the extent of binding of GFP-SPIRE1-MSFH (n = 4) and GFP-Mlph-RBD (n = 3) as a function of increasing GST-Rab27a-Q78L concentrations. Data are presented as mean values ± SEM and the equilibrium dissociation constants (Kd) are provided. K KIND, W WH2, M GTBM globular tail domain binding motif, S SB SPIRE box, F FYVE-type zinc finger, H C-terminal flanking sequences similar to H2 of Slp/Slac-proteins, WB western blotting. Source data b–f are provided in the Supplementary Source data file.We then investigated Rab27a:SPIRE1/2 interaction in mammalian cells using a bimolecular fluorescence complementation (BiFC) assay, in which Rab27a and SPIRE1/2 were transiently co-expressed in HEK293 cells as fusions to C- and N-terminus fragments of the Venus yellow fluorescent protein (vYFP). We observed significantly higher BiFC signal in cells transfected with wild-type Rab27a and the Q78L mutant compared with the inactive mutants T23N and N133I, in which the mean BiFC signal was similar to cells transfected with the vYFP fragments alone (Fig. 6d; mean normalised BiFC (% max); SPIRE1/SPIRE2 WT = 89.35/85.41%, T23N = 28.35/5.969%, Q78L = 75.21/85.23%, N133I = 24.71/6.601% and vYC alone = 8.637/16.65%). To examine Rab27a:SPIRE1/2 interaction in pigment cells, we expressed GFP-SPIRE1 (human) in B16-F1 cells, immuno-precipitated endogenous Rab27a from lysates using Rab27a specific antibodies and tested for co-immunoprecipitation of GFP-SPIRE1. Using mass spectrometry, we identified Rab27a (13 unique peptides), confirming the efficiency of the IP, and SPIRE1 (18 unique peptides) confirming the interaction of Rab27a and SPIRE1 in pigment cells (Supplementary Fig. 7, Supplementary Tables 2 and 3). These data concur with the results of pull-down assays and other approaches used here to test this interaction (Figs. 7 and 8, Supplementary Figs. 5–8) and further indicate that SPIRE1/2 are Rab27a effectors. Consistent with this recent proximity proteomic studies identified an interaction between Rab27a and endogenous SPIRE1 in HUVEC endothelial cells44.Fig. 7Rab27a recruits SPIRE1 to melanosomes in melanocytes.Melanocytes were transfected with plasmids allowing expression of the indicated proteins as fusions to the C-terminus of EGFP. Cells were fixed after 48 h, stained with GFP-specific antibodies to detect the expressed proteins, and the intracellular distribution of expressed protein and melanosomes was observed using a confocal microscope (see Experimental procedures). a–e Single confocal z-sections of the distribution of each protein, pigmented melanosomes (transmitted light/phase contract images) and merge images (melanosomes pseudo-coloured magenta). Upper panels show whole cells. Boxes indicate regions shown in lower panels at high magnification allowing comparison of the distribution of melanosomes and fluorescent protein. Line plots are fluorescence intensity profile plots of the boxed regions in high-magnification images averaged along the vertical axis. a–c, d and e are melan-a, melan-ln and melan-ash cells. d, e Yellow lines indicate the borders of transfected cells. Scale bars = 20 μm and 3 μm in main images and magnified portions. Source data are provided in the Supplementary Source data file.Fig. 8Functional evidence that SPIRE1/2 interact with melanosome-associated Rab27a in melanocytes.melan-ln (Mlph null) and melan-ash (Rab27a null) melanocytes were infected with viruses expressing the indicated proteins. Cells were fixed after 24 h, stained with GFP-specific antibodies, and the intracellular distribution of GFP-fusion proteins and melanosomes was observed using a confocal microscope (see Experimental procedures). a A schematic representation of the structure of mini-Va and myoSPIRE1/2 proteins, and their interaction with membrane-associated endogenous Rab27a. R27BD Rab27-binding domain, SB SPIRE box, GTBM globular tail binding domain. b Single confocal z-sections showing the distribution of expressed proteins, melanosomes and their colocalisation (from left to right). In merge images melanosomes are false-coloured magenta. For myoSPIRE1/2 upper and lower panels are low and high-magnification images. Boxes in low-magnification images indicate the area presented in the high-magnification images below. Scale bars = 10 μm in main images and 2 μm in magnified portions. Arrows indicate colocalisation of spots of myoSPIRE with melanosomes. Cell outlines are shown by white lines in phase contrast (melanosome) images. c A bee swarm plot showing the effect of expression of myoSPIRE and other proteins on melanosome distribution in melanocytes (melan-ln/Mlph null and melan-ash/Rab27a null). Data presented are mean pigment area measurements from four independent experiments. In each case, pigment area was measured for ten cells expressing each of the different proteins. ****, *** and n.s. indicate significant difference p = <0.0001, 0.001 and none between the datasets linked by horizontal bars as determined by one-way ANOVA. Bars within datasets indicate the mean and 25th and 75th percentile of data. Source data for c are provided in the supplementary Source data file.The C-terminus membrane-binding module of SPIRE proteins interacts with Rab27aTo map the Rab27a-binding site(s) in SPIRE proteins, we tested the ability of SPIRE1 truncations to interact with active Rab27a-Q78L using the GST- pull-down assay. Consistent with the distribution of sequence similarity between SPIREs and other Rab27 effectors, we found an interaction of GST-Rab27a-Q78L with Myc-SPIRE1-MSFH (and Myc-SPIRE2-MSFH) proteins but not Myc-SPIRE1-KWM (Fig. 6b). To better map the Rab27a-binding site within the SPIRE-MSFH, we generated further truncations and tested the Rab27a-Q78L interaction (Fig. 6a). We found that SPIRE1-SFH interacted strongly with Rab27a-Q78L as did the FYVE-only and SF proteins, albeit to a lesser extent (Fig. 6e). This indicates that the C-terminus SFH fragment of SPIRE proteins interacts with Rab27a, and suggests that the interaction mechanism is conserved with other Rab27 effectors.SPIRE proteins interact with Rab27a with lower affinity than MlphTo characterise Rab27a:SPIRE interaction further, we compared the affinity of this interaction with other Rab27a:effector interactions, e.g., Mlph. To do this, we quantified the affinity of interaction of GFP-SPIRE-MSFH and GFP-Mlph-RBD with GST-Rab27a-Q78L in pull-down assays by measuring GFP depletion from HEK293 cell lysates (Fig. 6a, f). This revealed dissociation constants (Kd) of 143 (±25) nM and 707 (±155) nM, and maximum binding levels of 39.4% and 12.0% for the GFP-Mlph-RBD and GFP-SPIRE1-MSFH. These data correspond to the previously determined Kd of 112 nM Rab27a-Q78L:Mlph RBD interaction and indicate that SPIRE1 is a weaker Rab27a interactor than Mlph45. By this method, we were unable to determine the Kd of SPIRE2:Rab27a interaction. These data align with pull-down assays results and suggest that Rab27a interacts more strongly with SPIRE1 than SPIRE2 (Fig. 6b).SPIRE1/2 associate with melanosomes by a Rab27a-dependent mechanismThe above findings indicate that SPIRE proteins are Rab27 effectors whose expression is required for melanosome dispersion. As Rab27a is present on the cytoplasmic face of the melanosome membrane, this suggests that SPIRE proteins associate with melanosomes in a Rab27a-dependent manner. To test this, we expressed GFP-tagged SPIRE proteins and Rab27a in melan-a cells, and used confocal microscopy to examine their intracellular localisation. We observed that all three proteins were distributed in a punctate pattern throughout the cytoplasm (Fig. 7a–c). Consistent with Rab27a:SPIRE interaction studies, high-magnification imaging and intensity profile plots revealed that spots of SPIRE1 and Rab27a, but not SPIRE2, often overlapped with melanosomes (visible in phase contrast images; Fig. 7a–c; Pearson linear correlation coefficient = 0.823 for Rab27a, 0.623 for SPIRE1 and −0.081 for SPIRE2). We also saw that Myc-SPIRE1-MSFH overlapped with melanosome resident protein tyrosinase-related protein 1 (Trp1) and other melanosome-targeted proteins (co-expressed mRuby3-Rab27a and GFP-myosin-Va-CC-GTD (that includes the melanocyte-specific coiled coil and globular tail domains; Supplementary Fig. 8)). These observations indicate that SPIRE1 associates with melanosomes.To test whether this association was Rab27a dependent, we repeated the above experiment using melanocytes deficient in Mlph (melan-ln, in which Rab27a is retained on the melanosome membrane) and Rab27a (melan-ash)46,47. This revealed that Rab27a associated with melanosomes in both cell types (and dispersed melanosomes in melan-ash cells), but that SPIRE1 associated with perinuclear clustered melanosomes in melan-ln cells only (Fig. 7d, e; Pearson linear correlation coefficient = 0.601 and 0.739 for Rab27a, and = −0.230 and 0.617 for SPIRE1 in melan-ash and melan-ln). This indicates that association of SPIRE1 with melanosomes is dependent upon interaction with endogenous Rab27a, and that SPIRE1 is a Rab27 effector.To further investigate Rab27a:SPIRE2 interaction in melanocytes, we developed a, sensitive, functional assay that measures changes in melanosome distribution in melan-ln melanocytes as a read-out of the interaction of candidate proteins with endogenous melanosome-associated Rab27a (Fig. 8a). We modified the previously described ‘minimyosin’ protein (that couples an active motor-lever arm (S1) fragment of myosin-Va to the RBD of the Rab27 effector synaptotagmin-like protein 2-a (Slp2-a)) by replacing the Slp2-a RBD fragment with SPIRE proteins to create ‘myoSPIRE’ fusions (Fig. 8a). We then tested the ability of these to disperse clustered melanosomes in melan-ln and melan-ash cells (i.e., in the presence/absence of endogenous Rab27a). In melan-ln, we found that both myoSPIRE proteins, like minimyosin but not Rab27a, dispersed melanosomes compared with GFP alone (Fig. 8b, c; mean pigment area (% total); GFP = 26.3 ± 7.43%, myoSPIRE1 = 63.4 ± 7.96%, myoSPIRE2 = 66.6 ± 3.76%, mini-Va = 62.3 ± 10.8 % for and Rab27a = 31.3 ± 10.7%). Interestingly SPIRE1 did not rescue melanosome transport, even though it can interact with Rab27a and myosin-Va like Mlph (Supplementary Fig. 9; mean pigment area (% total); GFP = 24.8 ± 7.91, SPIRE1 = 18.5 ± 3.95 and Mlph = 75.4 ± 18.8). One possibility is that the affinity of SPIRE1:myosin-Va/Rab27a interactions may be too low to recruit sufficient active myosin-Va to melanosomes to drive their dispersal. Consistent with this our data reveal the low affinity of Rab27a:SPIRE1/2 interaction relative to Rab27a:Mlph (Fig. 6f). Meanwhile previous studies showed that SPIRE2 interacts with myosin-Va-GTD (0.9 ± 0.11 μM) with lower affinity compared with Mlph (0.5 μM)38,48. In melan-ash, although Rab27a expression dispersed melanosomes, neither minimyosin nor myoSPIRE proteins did so to a significantly greater extent than GFP alone (Fig. 8c; mean pigment area (% total); GFP = 31.9 ± 6.80%, myoSPIRE1 = 37.8 ± 5.84%, myoSPIRE2 = 38.6 ± 5.97%, mini-Va = 39.2 ± 1.97% and Rab27a = 66.9 ± 6.20%). This shows that SPIRE1 and SPIRE2 can interact with endogenous Rab27a at the melanosome membrane in melanocytes. Consistent with this using confocal microscopy we observed that spots of both myoSPIREs were often located adjacent to melanosomes in melan-ln cells (arrows in Fig. 8b).Rab27 dependent targeting of SPIRE proteins to membrane is important for uniform melanosome dispersion in melanocytesNext, we used two approaches to test the functional importance of SPIRE:Rab27a interaction. Firstly, we exploited sequence conservation between SPIREs and other effectors to generated mutants that reduce SPIRE1:Rab27a interaction41,49. Using the GST Rab27a-Q78L pull-down assay, we found that mutation of the highly conserved glutamic acid residue at codon 548 of human SPIRE1 to alanine or lysine (E548A/K) blocked interaction with Rab27a (Supplementary Fig. 5c). This is consistent with the possible electrostatic contact of the equivalent residue (E14/54) in Mlph/RPH3A and the basic side chain of arginine 80/83 located in the switch II region of Rab27b/Rab3a (Supplementary Fig. 5d)41,42. Expression of SPIRE1E548K in SPIRE1/2 depleted melan-a cells resulted in a significant increase in the proportion of cells with dispersed melanosomes compared with GFP (Fig. 4b, c; mean % of cells with dispersed melanosomes = 97.61 ± 2.044). However, as seen for SPIRE1-KW and SPIRE2, in a significant proportion of cells expressing SPIRE1E548K melanosomes were hyper- rather than uniformly dispersed, as seen for intact SPIRE1 (Fig. 4b, c; mean % of cells with HD melanosomes = 56.85 ± 1.985). Also spots of SPIRE1E548K protein, like SPIRE2, were less frequently associated with melanosomes compared with wild-type SPIRE1 (Supplementary Fig. 5e arrows). Secondly, we investigated the functional effect of tethering the active SPIRE2 (KW) fragment to melanosomes, by expressing it as a fusion to the melanosome-targeted, non-functional, Rab27aSF1F4 mutant (hereafter KW-Rab27a; Fig. 4a). Recent FRAP studies found that association of Rab27a with melanosomes is relatively stable indicating that the chimeric KW-Rab27a is likely to stably associate with melanosomes50. Accordingly, we found that KW-Rab27a, localised to melanosomes (unlike SPIRE2-KW) and restored their uniform dispersion in SPIRE1/2-depleted cells to a greater extent than SPIRE2-KW alone (Fig. 4a, b ((high magnification inset), Fig. 4c; mean % cells with uniformly dispersed melanosomes KW-Rab27a = 98.01 ± 2.0%). These findings indicate that association with Rab27a at the melanosome membrane enhances the ability of SPIRE proteins to uniformly disperse melanosomes, and supports the idea that differences in the Rab27a interaction affinity of SPIRE1 and SPIRE2 underlie different patterns of melanosome dispersal seen in cells expressing either protein alone.Membrane targeting of FMN formins reduces their function in melanosome transportGiven the importance of FMN1/SPIRE interaction in melanosome transport and the finding that SPIRE proteins can associate with melanosomes, we next investigated whether FMN1 associates with melanosomes29. Confocal analysis of localisation in melan-f cells revealed that GFP-Fmn1 was distributed throughout the cytoplasm and not enriched adjacent to melanosomes, indicating that FMN1, unlike SPIRE1, does not strongly associate with melanosomes (Fig. 1f (inset)). This suggests that FMN1 may interact transiently, rather than stably, with melanosome-associated SPIRE proteins, as proposed by in vitro studies of AF assembly by SPIRE1 and Fmn2 (ref. 51). To test this possibility, we used the Rab27aSF1F4 mutant fusion strategy to generate the FH1-FH2-Rab27a fusion protein, which stably targets the FH1 and FH2 domains of FMN1 to melanosomes, and measured its activity in dispersing melanosomes in melan-f cells (Fig. 5a). We saw that although FH1-FH2-Rab27a localised to melanosomes and increased melanosome dispersion compared with GFP, it was significantly less effective than intact FMN1 or the FH1-FH2-FSI fragment (Fig. 5a, b (high magnification inset), Fig. 5c; mean pigment area (% total), FH1-FH2-Rab27a = 55.42 ± 17.49). This indicates that stable association with melanosomes reduces FMN1 function and supports idea that FMN1 transiently associates with melanosome-associated SPIRE1/2.To further investigate the importance of the subcellular positioning of SPIRE1/2 and FMN1, we tested the ability of FMN1 related protein, FMN2, to rescue melanosome transport in melan-f cells. FMN2, like FMN1, contains FH1 and FH2 domains, and the C-terminus FSI, but unlike FMN1, FMN2 may be co-translationally myristoylated and targeted to the cell membrane52,53. Thus, FMN2 may generate AF near the cell membrane, i.e., distant from the perinuclear clustered melanosomes in melan-f cells, rather than throughout the cytoplasm like FMN1 (Figs. 1f and 5b). We found that both intact FMN2 (hereafter GFP-FMN2) and the FMN2 FH1-FH2-FSI fragment expressed as C-terminus fusions to GFP were distributed throughout the cytoplasm, and restored melanosome dispersion with similar efficiency to GFP-FMN1 (Supplementary Fig. 10a–c; mean pigment area (% total); GFP-FMN1 = 84.11 ± 9.544, GFP-FMN2 = 81.72 ± 8.011 and GFP-FMN2 FH1-FH2-FSI = 86.42 ± 10.84). Meanwhile, although the N-terminus fusion of FMN2 to GFP (FMN2-GFP) dispersed melanosomes more efficiently than GFP alone, it did so with lower efficiency than GFP-FMN2 (Supplementary Fig. 10b, c; mean pigment area (% total); GFP = 44.17 ± 9.31, FMN2-GFP = 58.37 ± 7.818). To probe the role of N-myristoylation and membrane targeting in reducing the function FMN2-GFP versus GFP-FMN2, we blocked this by two approaches; (1) expression of the non-N-myristoylatable Fmn2-G2A mutant, and (2) treatment with the N-myristoyltransferase inhibitor (NMTi) IMP-1088 (refs. 53,54). We found that both strategies redistributed FMN2-GFP to a cytoplasmic pattern, similar to GFP-FMN2 and GFP-FMN1, and enhanced its function in transport (Supplementary Fig. 10b, c; mean pigment area (% total); FMN2-[G2A]-GFP = 79.71 ± 11.72%, FMN2-GFP + NMTi = 79.03 ± 9.488%). Using a myristic acid analogue/ alkyne (YnMyr) labelling approach, we confirmed the N-myristoylation status of FMN2-GFP, and the effects of the G2A mutant and NMTi in blocking this modification (Supplementary Fig. 10d). These observations support the idea that Fmn1 does not stably associate with membranes in melanocytes.DiscussionHere we investigated the mechanism of myosin-Va-dependent melanosome dispersion in melanocytes, focusing on the role of dynamic AFs that we previously identified as essential for this process15.Our data indicate: (1) that FMN1 and SPIRE proteins regulate melanosome dispersal in melanocytes, (2) that FMN1 and SPIRE1/2 extend AFs from the membranes of melanosomes to generate a cell-wide network that links adjacent melanosomes and (3) that Rab27a can recruit SPIRE1/2 to melanosomes as effectors, along with myosin-Va and Mlph. The latter point raises the interesting possibility that Rab27a drives melanosome dispersion in melanocytes by co-ordinating the activity of AF motors (i.e., myosin-Va) and AF assembly machinery (i.e., SPIRE1/2 and FMN1).Based on these findings, we propose a model by which this might occur (Fig. 9). Firstly, we suggest that Rab27a-GTP recruits two effector complexes to melanosomes; (i) myosin-Va/melanophilin and (ii) SPIRE1/2 (Fig. 9a). The recently identified myosin-Va:SPIRE1/2 interaction may also stabilise SPIRE1/2 at the melanosome membrane and/or integrate SPIRE1/2 activity with myosin-Va38. We expect that the recruitment of two effector complexes requires two pools of Rab27a, as the Rab27 interaction domain of SPIRE1/2 appears conserved with other effectors, e.g., Mlph and Slp2 (refs. 41,49). Interaction with these effectors obscures large portions of the surface of Rab27, including the switch regions, meaning that it is unlikely that a single Rab27 protein could interact with both effectors simultaneously. Rab27a interaction may activate SPIRE1/2 at the melanosome membrane by disrupting the intramolecular regulatory KIND/FYVE interaction35. Active SPIRE1/2 may then transiently recruit FMN1, via SPIRE1/2-KIND:FMN1-FSI interaction, thereby allowing FMN1 with SPIREs to associate with, and elongate, the +/barbed ends of short AFs located at the melanosome membrane. This would result in the formation of an AF network that is polarised with +/barbed ends oriented away from the melanosome membrane (i.e., towards other melanosomes and the cell periphery; Fig. 9b). This is consistent with the observation that melan-a, but not melan-f or SPIRE1/2 depleted, melanocytes harbour a population of AFs that emanate from, and link adjacent melanosomes. Thus myosin-Va attached to an adjacent melanosome may then ‘walk’ towards the +/barbed ends of one of these AFs thereby dispersing melanosomes from one another (Fig. 9c).Fig. 9A model indicating how Rab27a could regulate AF-driven melanosome transport by integrating the activity of myosin-Va, and SPIRE1/2 and FMN1.See main text for details. In a, a model melanosome, myosin-Va, SPIRE and an actin filament are shown, Rab27a is not shown for clarity but is present on the melanosome membrane. In b, the activity of FMN1 in extending the +end of melanosomes from melanosomes is shown. In c, grey and black colours indicate AF and myosin-Va associated with melanosomes of the same colour. In d, grey and black colours indicate the position of numbered melanosomes at the start and finish of an episode of dispersive transport.Extrapolating from this two melanosome scenario to the cellular level, we propose that this pattern of myosin-Va/AF-dependent inter-organelle repulsion could drive and sustain dispersal of melanosomes from one another, resulting in their dispersal throughout the cytoplasm, as seen in wild-type melanocytes (Fig. 9d). We suggest that such an inter-connected, three-dimensional network of organelles, tracks and motors could allow cell-wide coupling of local organelle-associated force generators (i.e., myosin-Va motors and FMN1/SPIRE1/2 AF assembly machinery). And that this arrangement could allow the generation of sufficient force to rapidly disperse these large, rigid and numerous organelles (1000 s/cell), and maintain this distribution, more efficiently than conventional transport models, which envisage single organelles pulled along a limited number of single tracks by motors independently of other cargo, motors and tracks.As a preliminary to test of our model, we used the open-source modelling framework Cytosim to simulate melanosome dispersion by myosin-Va motors and FMN1/SPIRE1/2 AF builders. Based on our data, we modelled AFs polarised with their −/pointed ends attached (via Rab27a/SPIRE interaction) to melanosomes and +/barbed ends growing (extended by FMN1) into the cytoplasm, and myosin-Va anchored with their cargo-binding tail (and Rab27a/Mlph) to the melanosome and motor domains free in the cytoplasm (hereafter termed ‘melanosim-attached’, see Supplementary Code 1). Running the simulation from a starting point, in which melanosomes were clustered in the centre of the model cell, we saw that as AFs emerged from melanosomes myosin-Va motors on adjacent melanosomes transiently attached to the filaments and the linked melanosomes moved apart relative to one another (Supplementary Fig. 11a, Supplementary Movies 1 and 2, see Supplementary Fig. 11b for examples). This resulted in rapid global dispersion of the clustered population of melanosomes throughout the cytoplasm of the cell, as predicted by our mode. This closely resembled previous live-cell observation of melanosome dispersion in melan-d cells, in response to acute activation of a pharmacologically activatable myosin-Va motor15. The extent of dispersion was reduced in simulations using free AFs whose pointed ends were not attached to melanosomes (‘melanosim-detached’, Supplementary Fig. 11a, Supplementary Movie 3). These observations support the hypothesis that a melanosome-associated Rab27a can generate a network of organelle linked motors and filament builders, and disperse rapidly and uniformly organelles throughout the cytoplasm of a eukaryotic cell.Several pieces of data presented here, and previously, support this possibility. Firstly, melan-f cells and SPIRE1/2 depleted melan-a cells show significantly fewer melanosome-associated AFs compared with melan-a cells. Secondly, fusion of active SPIRE2 and FMN1 to melanosomes via the Rab27aSF1F4 mutant boosts the function of SPIRE2-KW and reduces the function of FMN1-FH1-FH2, consistent with stable and transient modes of association of SPIRE1/2 and FMN1 with melanosomes. Thirdly, consistent with the predicted requirement for proximity between melanosomes and AFs, re-targeting of FMN (FMN2-GFP) to the cell membrane via N-myristoylation reduces its function. Fourthly, we previously found that the processive +/barbed end directed transport activity of myosin-Va was essential for its function in melanosome dispersion15.Beyond melanocytes evidence of such non-conventional or network-like organisations of the acto-myosin cytoskeleton exist in other cell types, e.g., the sarcomere of muscle cells and stress fibres of migrating fibroblasts represent. These structures allow the amplification of force generated by myosin-II motors to drive cell shape changes and movement55,56. Meanwhile recent work in oocytes reveal that myosin-Vb, SPIRE1/2 and FMN2 form a meshwork of short AFs that drive rapid, long-range outward flow of a large population of Rab11-positive vesicles to the plasma membrane, independently of MTs (refs. 23,57). With our data, this suggests that similar AF-dependent organelle transport systems might be widespread in somatic mammalian cells.Our work additionally highlights differences between SPIRE1 and SPIRE2 function. In particular, we found that SPIRE1 has higher affinity for interaction with Rab27a, localised more strongly to melanosomes and more efficiently drove uniform melanosome dispersion compared with SPIRE2. In contrast uniform melanosome dispersion required higher levels of SPIRE2 expression and lower levels resulted in hyper-dispersion. The KW-Rab27a fusion protein uniformly dispersed melanosomes more efficiently than SPIRE2-KW, conversely SPIRE1/2 mutants compromised in membrane/Rab27a interaction promoted hyper-dispersion. This suggests that interaction with Rab27a is important in ensuring uniform dispersion, as seen in wild-type melanocytes, by toning down and localising the hyper-dispersive activity of the KW to melanosomes. In parallel, we found that the SPIRE1 gene was more highly expressed than SPIRE2 and that single depletion of SPIRE1, but not SPIRE2, reduced dispersion leading to perinuclear clustering and hyper-dispersion of melanosomes. Overall, these data suggest that SPIRE1 plays a more significant role in melanosome transport than SPIRE2. Currently the mechanism by which SPIRE2 hyper-disperses melanosomes at low expression levels is unclear. One possibility is that at low concentration SPIRE2 is activated by interaction with factors e.g. other Rabs, associated with peripherally located compartments, and assembles dynamic AFs at the periphery that allow local accumulation of melanosomes. Meanwhile at higher concentrations SPIRE2 may be activated by Rab27a on melanosomes triggering uniform melanosome dispersal via the mechanism proposed in Fig. 9. Future studies should address this question.Finally, we suggest that our data have broader implication for transport in the many other cell types in which Rab27, related Rab proteins, SPIRE1/2, FMN and myosin-V proteins are expressed. In particular we note the similarity between the proteins identified here and those shown to regulate the vesicle associated AF meshwork that drives asymmetric spindle positioning during meiosis in mouse oocytes (SPIRE1/2, FMN2, Rab11 and myosin-Vb)22,23. The finding that similar groups of proteins drive similar activities i.e. rapid, long-range AF-dependent organelle transport, in different cell types suggests that related machineries may drive transport in other animal cell types. Consistent with this in mammals Rab27 regulates the dispersal and secretion of organelle contents in many specialised ‘secretory’ cell types e.g. inflammatory mediator containing granules in mast cells, lytic granules in cytotoxic lymphocytes and exosome containing MVBs in cancer cells58–60. The oligodactylism phenotype of the FMN1 knockout mouse also supports the possible function of FMN1 in secretory transport processes, since proper limb development requires concerted communication between cells, which is vitally dependent on secreted proteins25. Also, the behavioural phenotypes of FMN2 and SPIRE1 mutant mice may be caused by altered secretion of neuropeptides or growth factors61,62. Thus, it will be interesting to test whether Rab27a contributes similarly to co-ordinate the AF-dependent transport of other organelles in other cell types.MethodsAnimal proceduresAll animal procedures with mice reported in this work were first validated by the National Centre for Biotechnology Ethics Committee on Animal Experimentation, thereafter approved by the National CSIC Ethics Committee and eventually authorised by the Autonomous Community of Madrid, acting as the competent authority, according to the Spanish legislation (L32/2007, RD53/2013, L6/2013, OM ECC/566/2015 and RD1386/2018) and the European Directive 2010/63/EU, with the approval reference PROEX 343/15. All mice were housed at the registered CNB animal facility, fed and provided water and regular rodent chow ad libitum, with a light/dark cycle 08:00–20:00, according to the European and Spanish norms, and the animal welfare recommendations.Derivation and maintenance of cultured cellsCultures of immortal FMN1-deficient melanocytes (melan-f) were derived essentially63. In brief, mice carrying a previously generated FMN1 loss of function allele (FMN1pro) were crossed with Ink4a-Arf mutant mice, in order to generate pups homozygous for the FMN1pro mutant allele and heterozygous for Ink4a-Arf mutant allele. Genotyping of the embryos was carried out by PCR screening25,63. Melanocytes were then derived from the dorsal skin of newborn mutant mice64. Cultures of immortal melan-f, melan-ash, melan-ln and melan-a melanocytes, and HEK293a were maintained, infected with adenovirus expression vectors and transfected with siRNA oligonucleotides at a final concentration of 66.7 fmol/μl of transfection mixture and Oligofectamine transfection reagent diluted 300-fold in optiMEM medium (both Thermo-Fisher)65,66. The melanocyte cell lines described here are available from the Wellcome Trust Functional Genomics Cell Bank http://www.sgul.ac.uk/depts/anatomy/pages/WTFGCB.htm.Transfection of cultured cellsCultured cells were transfected with siRNA as described above and plasmid DNA at a final concentration of 1.67 ng/μl using Fugene 6 transfection reagent (producer) diluted 300-fold66. The sequence of siRNA oligonucleotides used in this study are are indicated in Supplementary Table 4. siRNA oligonucleotides were purchased from Dharmacon Thermo-Fisher and Sigma Genosys UK.Quantitative real-time PCRPrimers and the probes for Q-RT-PCR targets (from Sigma Genosys, Cambridge, UK) were designed using Primer Express software (Life Technologies). Probes were labelled at the 5′- and 3′-ends with fluorophore 6-FAM (6-carboxyfluorescein) and quencher TAMRA (tetramethylrhodamine), respectively (Supplementary Table 5). To generate mRNA samples, pools of melan-a cells grown in six-well plates (1 × 105/well) were transfected with siRNA in triplicate as described previously66. Seventy two hours later cells were harvested and mRNA extracted using the RNeasy Mini RNA extraction kit (Qiagen). cDNA was generated using Moloney Murine Leukaemia Virus M-MLV reverse transcriptase (Promega) using random primers. To generate a standard curve of signal:template concentration for each Q-RT-PCR assay a pool containing 5% of each of the cDNA samples analysed was generated. This pool was serially diluted in DEPC water (1:4, 1:16, 1:64, 1:256) and these were used as template in target gene and GAPDH Q-RT-PCR assays. (The shape of the standard curve indicates the relationship between signal and template concentration. For all assays standard curves gave straight lines with R2 > 0.99, indicating that there is a linear relationship between signal and template). To measure the expression of targets in siRNA-transfected cells each neat cDNA was diluted 1:32 in DEPC water and the following reagents were added per well of a 96-well plate: 6.5 μl TaqMan Fast 2× PCR Master Mix (Life Technologies); 0.4 μl forward primer (10 μM); 0.4 μl reverse primer (10 μM); 0.25 μl probe (10 μM); 3 μl cDNA; 2.45 μl DEPC water. For each sample, three technical repeats were performed. Reaction plates were sealed with optically clear adhesive film, centrifuged, and Q-RT-PCR performed using a StepOnePlus Real-Time PCR system (Applied Biosystems) using the’fast’ mode. CT values for each reaction were determined by the StepOne software. The slope (S), intercept (I) and R2 values were calculated for the standard curve of each Q-RT-PCR assay. CT values from siRNA-transfected samples were then processed to generate a ‘quantity value’ for each CT value as follows; (1) (CT-I)/S = LQ, (2) 10LQ = Q, (3) Q × (1/MNT) = GOIP (where MNT = mean non-targeted quantity value) and (4) GOIP/GP = normalised expression of target relative to GAPDH (where GP is the normalised quantity value for the GAPDH primer). Expression of FMN1 was detected using Taqman assay Mm01327668_m1 from Thermo-Fisher Scientific, UK.For measurement of the expression levels of SPIRE, FMN and MLPH genes, the RNA isolation kit of Macherey-Nagel (Düren, Germany) was used. To reduce amount of melanin and increase purity of total RNA step two of the protocol was repeated. In order to quantify the amount of total RNA isolated, spectrophotometric determination was used. cDNA was generated employing the Qiagen QuantiNova® Reverse Transcription Kit (according the manufacturer’s protocol). An internal control RNA was used to verify successful reverse transcription. PCR primer sets were designed using primer3, Primer-Blast and were constructed by Sigma-Aldrich (Taufkirchen, Germany). The primer sets and templates used are shown in Supplementary Table 6. For primer testing generated cDNA from melan-a total RNA and plasmid DNAs as positive controls were used as templates. For absolute quantification, an external plasmid-cDNA standard was used to create a standard curve with known copy number concentrations. Linearised plasmid DNA was serially diluted in water. The copy number of standard DNA molecules could be determined by the following formula: ((X g/μl DNA)/[plasmid length in basepairs × 660]) × 6.022 × 1023 = Y molecules/μl). The absolute quantification by Q-RT-PCR was performed with the Rotor-Gene Q (Qiagen) thermocycler. For each reaction triplets were performed, and CT values were determined with the help of the Qiagen Rotor-Gene Q Series Software. CT values of the samples are compared to the standard curve with known concentrations to calculate the amount of target in the samples. For this purpose, the QuantiNova® SYBR® Green PCR Kit (GE Healthcare Life Sciences, Freiburg, Germany) and the associated protocol were employed.ImmunoblottingSDS–PAGE was carried out using 4–15% gradient SDS–PAGE gels (Bio-rad #4561085) and a Mini-PROTEAN Tetra Cell (Bio-rad 1658000EDU). Molecular weight standards were BLUeye Prestained Protein Ladder (Sigma-Aldrich 94964-500UL). Proteins were and transferred to PVDF membrane (Merck Millipore HVLP04700) using a Hoefer™ TE22 Mini Tank Blotting Unit (Fisher Scientific 10380895)38,66. Details of primary and secondary antibodies used in this study are in Supplementary Tables 7 and 8. Uncropped western blots are available in the Supplementary Information file.Signal was detected using a Li-Cor infrared scanner (Odyssey; Fig. 1d, Supplementary Fig. 4) or by chemiluminescence (Luminata Forte Western HRP substrate; Merck Millipore, Darmstadt, Germany; Fig. 6, Supplementary Figs. 5 and 6) recorded with an Image Quant LAS4000 system (GE Healthcare Life Sciences; Supplementary Fig. 10b).Multiple protein sequence alignments and phylogenetic tree generationMultiple protein sequence alignments for C-terminal regions of human SPIRE1 and SPIRE2, and N-terminal regions of human MLPH, MYRIP, Rabphilin-3A and NOC2 amino acid sequences, were performed using the Clustal Omega Multiple Sequence Alignment tool (EMBL-EBI, Hinxton, UK). Respective cDNA sequences were obtained from the NCBI Gene database with the following accession numbers: Homo sapiens (Hs)-SPIRE1: NM_020148.2, Hs-SPIRE2: AJ422077, Hs-MLPH: NM_024101.6, Hs-MYRIP: NM_001284423.1, Hs-RPH3A (Rabphilin-3A): NM_001143854.1, Hs-RPH3AL (NOC2): NM_006987.3. Generation of phylogenetic trees for Rab27 interacting proteins was based on multiple amino acid sequence alignments using Clustal Omega as described above and manual confinements. Respective cDNA sequences were obtained from the NCBI Gene database with the following accession numbers: Hs-RIMS1: NM_014989.5, Hs-RIMS2: NM_001100117.2, Hs-RPH3A (Rabphilin-3A): NM_001143854.1, Hs-RPH3AL (NOC2): NM_006987.3, Hs-MLPH (SLAC2-A): NM_024101.6, Hs-EXPH5 (SLAC2-B): NM_015065.2, Hs-MYRIP (Slac2-c): NM_001284423.1, Hs-SYTL1 (SLP1): XM_005246022.3, Hs-SYTL2 (SLP2-A): NM_032943.4, Hs-SYTL3 (SLP3-A): NM_001242384.1, Hs-SYTL4 (SLP4, granuphilin): NM_080737.2, Hs-SYTL5 (SLP5): NM_001163334.1, Hs-SPIRE1: NM_020148.2, Hs-SPIRE2: AJ422077. Evolutionary analyses and phylogenetic tree formation were conducted in MEGA7 (ref. 67). The evolutionary history was inferred by using the Maximum Likelihood method based on the JTT matrix-based model68. The bootstrap consensus tree inferred from 500 replicates is taken to represent the evolutionary history of the taxa analysed69. Branches corresponding to partitions reproduced in <50% bootstrap replicates are collapsed. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (500 replicates) are shown next to the branches69. Initial tree(s) for the heuristic search were obtained automatically by applying Neighbour-Join and BioNJ algorithms to a matrix of pairwise distances estimated using a JTT model, and then selecting the topology with superior log likelihood value. The analysis involved14 amino acid sequences. There was a total of 188 positions in the final dataset.Generation of bacterial and mammalian protein expression vectorsExpression vectors were generated by standard cloning techniques using AccuPrime Pfx (Thermo-Fisher, Waltham, MA, USA) or Pfu (Promega, Mannheim, Germany) DNA polymerases, restriction endonucleases and T4 DNA ligase (both New England Biolabs, Frankfurt am Main, Germany). Point mutants were generated using the In-Fusion HD cloning kit (TakaraBio/Clontech). DNA sequencing was carried out by LGC Genomics (Berlin, Germany) and Source Bioscience (Nottingham, UK). Supplementary Table 9 shows details of vectors used in this study. Amino acid boundaries are related to the following cDNA sequences: human SPIRE1 (NM_020148.2), human SPIRE2 (AJ422077), murine FMN1 (XP_011237597), murine FMN2 (NP_062318.2), murine myosin-Va (XM_006510827.3) and murine Mlph (NP_443748.2). To generate mRuby3 fusion protein encoding expression vectors, the pKanCMV-mClover3-mRuby3 vector was used as template to PCR amplify the mRuby3 cDNA sequence for further subcloning. This vector was a gift from Michael Lin and was provided by Addgene (plasmid #74252)70.Recombinant protein expression and purificationRecombinant GST-Rab27a-WT, Q78L mutant and T23N mutant proteins were expressed in Escherichia coli Rosetta bacterial cells (Merck Millipore, Novagen). Bacteria were cultured in LB medium (100 mg/l ampicillin, 34 mg/l chloramphenicol) at 37 °C until an A600 nm of OD 0.6–0.8. Protein expression was induced by 0.2 mM isopropyl-β-D-thiogalactopyranoside (Sigma-Aldrich, Taufkirchen, Germany) and continued at 16–20 °C for 18–20 h. Bacteria were harvested and lysed by ultra-sonication. Soluble proteins were purified by an ÄKTApurifier system (GE Healthcare Life Sciences) using GSH-Sepharose 4B (GE Healthcare Life Sciences) beads and size-exclusion chromatography (High Load 16/60 Superdex 200; GE Healthcare Life Sciences). Proteins were concentrated by ultrafiltration using Amicon Ultra centrifugal filters (Merck Millipore) with respective molecular weight cut offs. The final protein purity was estimated by SDS–PAGE and Coomassie staining.Nucleotide loading of GST-Rab27aIn order to analyse the interaction of GST-Rab27a and SPIRE1 depending on its nucleotide loading (GTP versus GDP), GST-Rab27a proteins were loaded with the non-hydrolysable GDP (GDPβS) and GTP (GTPγS) analogues (both from Sigma-Aldrich), respectively, prior to GST pull-down assays. For each nucleotide exchange 5.8 mg of purified GST-Rab27a fusion protein was used, corresponding to 2.5 mg pure Rab27a protein, and mixed with a fourfold molar excess of GDPβS and GTPγS (1.13 mM each), respectively, 12.5 units alkaline phosphatase (CIAP; Roche, Penzberg, Germany) and CIAP buffer (20 mM Tris-HCl pH 7.4, 200 mM NaCl, 1 mM MgCl2, 2 mM DTE, 200 mM (NH4)2SO4, 100 µM ZnCl2) in a total volume of 800 µl. The mixture was incubated for 17 h at 4 °C on a rotating wheel. On the next day, CIAP buffer was exchanged by RAB polarisation buffer (20 mM Tris-HCl pH 7.4, 200 mM NaCl, 5 mM MgCl2, 1 mM DTE, 5 µM GDPβS/GTPγS) using NAP-10 columns (GE Healthcare Lifesciences) according to manual.GST pull-down from HEK293 cell lysatesHEK293 cells were cultured and transfected with plasmid DNA as described previously38. To ensure equal input of GFP-tagged or Myc-epitope-tagged SPIRE1 and SPIRE2 deletion mutants for GST pull-downs, all SPIRE mutants were first expressed in HEK293 cells and respective protein expression levels were analysed by western blotting (anti-GFP, anti-c-Myc). Protein bands were quantified using ImageQuant TL (GE Healthcare Life Sciences) and normalised to the lowest signal band. According to quantification, the amount of cell lysates employed in subsequent pull-downs was adjusted. For pull-downs, HEK293 cells were transfected with expression vectors encoding fluorescent protein or Myc-epitope-tagged proteins. Forty eight hour post transfection, cells were lysed in lysis buffer (25 mM Tris-HCl pH 7.4, 150 mM NaCl, 5 mM MgCl2, 10% (v/v) glycerol, 0.1% (v/v) Nonidet P-40, 1 mM PMSF, protease inhibitor cocktail) and centrifuged at 20,000 × g, 4 °C, 20 min to remove insoluble debris. For GST pull-down assays 50 µg GST-RAB27A (-WT, -Q78L, T23N, -GTPγS, -GDPβS) proteins and 25 µg GST control was coupled to GSH-Sepharose 4B beads (1:1 suspension) for 1 h, 4 °C on a rotating wheel. Beads were washed twice with pull-down buffer (25 mM Tris-HCl pH 7.4, 150 mM NaCl, 5 mM MgCl2, 10% (v/v) glycerol, 0.1% (v/v) Nonidet P-40) and subsequently incubated with cell lysates for 2 h at 4 °C on a rotating wheel. Here, 1 ml lysates with the lowest expressed protein was employed and lysates with higher protein abundance were diluted with pull-down buffer to 1 ml, according to prior quantification. Beads were washed four times with pull-down buffer and bound proteins were eluted with 1× Laemmli buffer, denatured at 95 °C for 10 min, and subsequently analysed by immunoblotting.Quantitative GST pull-down assaysGST pull-down assays were performed as described above from HEK293 cell lysates with increasing concentrations of GST-Rab27a-Q78L protein. According to prior quantification, the concentration of ectopic expressed GFP-SPIRE1-MSFH was ~50 nM. Cell lysates were pooled and equally distributed to beads coupled GST-Rab27a-Q78L protein. Beads were pelleted and the cell lysate supernatant was diluted 1:4 with aqua. dest. Each sample was allowed to adapt to 20 °C for 10 min in a water bath. The concentration-dependent binding of Rab27a-Q78L to C-terminal GFP-SPIRE1-MSFH was determined by fluorospectrometric analysis using FluoroMax- 4 Spectrofluorometer (Horiba Jobin Yvon, Bensheim, Germany). The same experiment was performed employing GFP-Mlph-RBD expressed in HEK293 cells and purified GST-Rab27a-Q78L proteins. The AcGFP1 green fluorescent protein was excited at 488 nm (slit = 5 nm) and the emission at 507 nm was recorded (emission maximum, slit = 5 nm) with an integration time of 0.1 s. The data were calculated as fraction bound (%) (y) using the equation (1)y = ((yo – yc)/yo) × 100, where yo and yc equal the fluorescence signal of the supernatant of pull-down experiments carried out in the absence and presence of GST-Rab27a-Q78L protein coupled to beads.Furthermore, data were analysed in SigmaPlot 12.3 software (Systat Software, Erkrath, Germany). Equilibrium binding data were fitted according to the equation (2)y = (Bmax × x)/(Kd × x) with Bmax representing the maximal amplitude, Kd representing the equilibrium constant and x representing the concentration of GST-Rab27a-Q78L.Microscopy and image analysisCells (1 × 104) for immunofluorescence were cultured on 13 mm diameter 1.5 thickness glass coverslips (SLS, Nottingham, UK. 4616-324139) for at least 24 h, prior to infection with adenovirus or transfection with plasmid DNA. Twenty-four hours later cells were paraformaldehyde fixed, stained, and fluorescence and transmitted light images were then collected using a Zeiss LSM710 confocal microscope fitted with a 63× 1.4NA oil immersion Apochromat lens or a Zeiss Axiovert 100 S inverted microscope fitted with 10× and 40× objective lenses, and an Axiocam MR3 CCD camera. Primary and secondary antibodies used in this study are listed in Supplementary Tables 7 and 8, F-actin was detected using texas-red-phalloidin (Sigma P1951; 100 nM).Analysis of melanosome dispersion/transportFor analysis of the effects of siRNA on melanosome distribution transfections were carried out in triplicate, i.e., three wells of a 24-well plate for each siRNA in each experiment. Seventy two hours later phase contrast images of three different randomly selected low power (using 10× objective) fields of cells in each well were captured using Axiovision 4.8 software associated with a Zeiss Axiovert 100 S inverted microscope. Images were then randomised and the number of cells with clustered melanosomes in each field was counted by a researcher blinded to the identity of the siRNA transfected into each field of cells. Cells in which pigmented melanosomes were contained within the perinuclear cytoplasm (<50% of the total cytoplasmic area) were defined as containing clustered melanosomes. Measurement of the function of adenovirus expressed FMN1 and SPIRE1/2 proteins in melanosome transport was based on manual measurement of the proportion of cell area occupied by pigmented melanosomes65.Bimolecular fluorescence complementationFor BiFC HEK293a cells were transfected with plasmids allowing expression of mCherry (control for transfection efficiency), vYNE-SPIRE1/2 (human SPIRE1/2 fused to the C-terminus of enhanced vYFP (accession CAO91538.1) N-terminus fragment (amino acids 1–173)) and vYCE-Rab27a wild type and mutants (Rab27a (rat)) fused to the C-terminus of enhanced vYFP C-terminus fragment (amino acids 155–239) using Fugene 6 transfection reagent (Promega E2691). Forty-eight hours later mCherry and YFP (BiFC) fluorescence was recorded from living cells in low power fields of view for each condition using the Zeiss Axiovert 100 S microscope, using the same exposure setting for each channel and each condition. Normalised BiFC signal was determined using ImageJ software. Briefly mCherry images were converted to binary and ROIs corresponding to transfected cells were saved to the regions manager. Mean vYFP for each ROI was extracted and normalised BiFC determined for each ROI by dividing the vYFP signal by the corresponding mCherry signal to normalise for differences in the level of transfection in each cell. Median BiFC signal for ROI was determined using Graphpad Prism7 software (Graphpad, La Jolla, USA).Colocalisation analysisFixed cells were analysed with a Leica AF6000LX fluorescence microscope, equipped with a Leica HCX PL APO 63×/1.3 GLYC objective and a Leica DFC350 FX digital camera (1392 × 1040 pixels, 6.45 μm × 6.45 μm pixel size; all from Leica, Wetzlar, Germany). 3D stacks were recorded and processed with the Leica deconvolution software module. Images were recorded using the Leica LASX software and further processed with Adobe Photoshop, and subsequently assembled with Adobe Illustrator. The extent of colocalisation of myosin-Va, Rab27a and SPIRE1 proteins at intracellular melanocyte membranes, as well as the localisation of Rab27a and SPIRE1 C-terminal fragments, respectively, at melanosome membranes was analysed using the ImageJ (V2.0.0) plug-in Coloc2. Here, the colocalisation rate is indicated by the Pearson’s correlation coefficient (PCC) as a statistical measure to unravel a linear correlation between the intensity of different fluorescent signals. A PCC value of 1 indicates a perfect colocalisation, 0 indicates a random colocalisation and a PCC value of −1 indicates a mutually exclusive localisation of the analysed signals. To take the noise of each image into account and to gain an objective evaluation of PCC significance, a Costes significance test was performed. To do so, the pixels in one image were scrambled randomly and the correlation with the other (unscrambled) image was measured. Significance regarding correlation was observed when at least 95% of randomised images show a PCC less than that of the original image, meaning that the probability for the measured correlation of two colours is significantly greater than the correlation of random overlap71,72.Cytoskeleton preparations for FESEMmelan-a and melan-f cells were plated on 7 mm glass coverslips and grown for 24 h. The medium was aspirated and the cells were extracted for 1 min with 0.25% Triton X-100 (Sigma, UK) and 5 μM phallacidin (Sigma) in cytoskeletal stabilisation buffer (CSB; 5 mM KCl, 137 mM NaCl, 4 mM NaHCO3, 0.4 mM KH2PO4, 1.1 mM Na2HPO4, 2 mM MgCl2, 5 mM Pipes, 2 mM EGTA and 5.5 mM glucose, pH 6.1; ref. 73). After a quick wash in CSB, the samples were fixed with 1% glutaraldehyde (Sigma) in CSB with 5 μM phallicidin for 15 min. The coverslips were transferred to HPLC-grade water and incubated with 2% osmium tetroxide in water for 1 hour. After three 15 min washes in distilled water, the preparations were dehydrated through successive immersion in 30, 50, 70, 90 and 100% ethanol solutions (three times 10 min each), followed by two 20 min incubation in methanol. The coverslips were carefully picked from the methanol solution and gently immersed vertically in HMDS (Hexamethyldisilazane, Sigma) in glass bottles for 30 s twice and left to air-dry for at least an hour before mounting them on 10 mm diameter specimen stubs. The coverslips were then coated with silver paint and sputter coated with a 5–6 nm layer of platinum an Edwards S150B sputter coater. Samples were imaged in a JEOL JSM-6700F scanning electron microscope by secondary electron detection. For phalloidin labelling of AFs, melan-a cells were permeabilised for 45 s with 0.25% Triton in CSB, briefly rinsed in CSB and incubated with 0.5% glutaraldehyde in CBS for 10 min. The coverslips were then rinsed once rapidly, transferred on to parafilm, and incubated with 0.5 μM biotylinated phalloidin (Biotin-XX phalloidin, Thermo-Fisher Scientific/Molecular Probes) in PBS with 1% BSA and 0.1% fish gelatin (CWFS gelatin, Aurion) for 1 h, followed by 50 mM NH4Cl in PBS (15 min in total with a change to fresh solution after 7 min; ref. 39) and 0.1% fish gelatin in PBS (30 min in total with a change to fresh solution after 15 min). The coverslips were then incubated for 3 h in a humidifying chamber with 10 nm gold-coupled anti-biotin antibodies (Aurion) in 0.1% fish gelatin in PBS and 0.01% BSA-c (Aurion). The coverslips were washed three times 10 min each in gelatin (0.1% in PBS), followed by 1 min in 0.05 % Triton in PBS and four times 1 min washes in PBS. The coverslips were then post-fixed with 1% glutaraldehyde/5 µM phallacidin in CBS for 15 min, transferred to distilled water and processed for osmium/HMDS procedure as above. Samples were imaged by secondary electron detection for SEM and backscatter for gold particle labelling, and combined for image acquisition.FESEM colorisation and filament measurementsFilaments were highlighted using the Brush tool in Adobe Photoshop with 20–50% opacity and pseudo-colours were applied using the Hue/Saturation tool in Photoshop to avoid obscuring the structural details. Filament diameter was measured using the line profiling function in FIJI/ImageJ software on four different FESEM images of melan-a cells at two different resolution (25 K and 40 K). Filaments emanating from melanosomes were measured by racing them individually on FESEM images of melan-a (n = 6) and melan-f (n = 3) cells. Statistical significance was determined by Mann–Whitney test after evaluating data distribution normality by D’Agostino and Pearson normality test using GraphPad Software.Cytoskeleton preparations for rapid freeze/freeze dry/metal replicaFor metal replica microscopy39, melan-a cells were grown on customised right trapezoid-shaped small glass coverslips for 24 h, and cytoskeletons were extracted with 0.25% Triton in CBS for 45 s. The cytoskeleton were rinsed briefly in CBS, incubated with 0.2 mg/ml myosin S1 (Cytoskeleton Inc., Denver) in PEM buffer (100 mM PIPES-KOH, pH 6.9, 1 mM MgCl2 and 1 mM EGTA) for 20 min and fixed in with 1% glutaraldehyde in CBS for 15 min. The coverslips were washed twice for 5 min each with distilled water and then plunge-frozen in a liquid nitrogen-cooled 1:4 mixture of isopentane/propane. The samples were transferred to the specimen mount of a freeze fracture unit (Balzers BAF400D) and rotary shadowed at a 45 degree angle with 1.2–1.3 nm tantalum-tungsten or 1.5 nm platinum, followed by respectively 2.5 or 5 nm carbon at 90 degree angle. The replicas were separated from the glass coverslips with 8% hydrofluoric acid, washed into distilled water and picked up onto the surface of formvar-coated copper grids. Samples were examined using a JEOL 1200 EXII transmission electron microscope operating at 80 kV. Images were inverted and processed for increased contrast in Photoshop.Metabolic labelling and purification of Fmn2 with myristate analogue YnMyrBriefly, HEK293a cells were transfected with plasmids allowing expression of the N-terminus 153 amino acids of FMN2 (with/without the G2A mutation) as an N-terminus fusion to GFP, and incubated with NMTi IMP-1088 (100 nM) and YnMyr (tetradic-13-ynoic acid, 1 μM) as indicated for 24 h. Cells were harvested by trypsinisation and washed with 1× PBS. Pellets were homogenised by sonication in 120 μl lysisbuffer containing 1% (v/v) Triton X-100, 0.1% (w/v) SDS and EDTA-free protease inhibitor cocktail, in PBS (pH 7.4). Homogenates were centrifuged (10,000 × g, 4 min, 4 °C), supernatant was collected, protein concentrations were determined. Bio-orthogonal CLICK-ligation of 350 μg protein with capture reagent AzTB was performed as described earlier74. Aliquots containing 20 μg protein were kept aside to compare AzTB labelling and abundance of proteins at various stages, namely input, after enrichment with Neutravidin-agarose beads (enriched from 300 μg) and the pull-down supernatant (20 μg). Protein samples were resolved by a standard 12% SDS–PAGE gel running at 90 V, whereafter AzTB-clicked YnMyr-labelled proteins were visualised by fluorescence scanning on a Typhoon Variable Mode Imager 9500 (GE Healthcare). Hereafter, proteins were immobilised on a nitrocellulose membrane using wet blotting (Bio-RAD). Proteins Fmn2-GFP and Fmn2[G2A]-GFP were detected by western blotting with GFP antibody (1:1000, mouse), HSP90 (1:1000, mouse-anti-human Santa Cruz #sc-69703) and ARL1 (1:2000, rabbit-anti-human Proteintech 16012-1-AP), bound subsequently by HRP-conjugated secondary antibodies (1:10,000, Advansta anti-mouse R-05071-500 and anti-rabbit R-05072-500) and detected, using HRP Luminata (Merck) and an ImageQuant LAS4000 (GE Healthcare).Immunoprecipitation of Rab27aEight, 14 and 28 × 10 cm dishes of 80% confluent B16-F1 mouse melanoma cells, infected with adenovirus vectors expressing GFP, GFP-Mlph or GFP-hsSPIRE1, and cultured for a 10 days to allow expression of proteins. Cells were then washed with cold PBS, trypsinised to detach from plates, harvested by centrifugation and lyzed in 500 μl of buffer containing 20 mM Tris-HCl, pH 7.5, 50 mM NaCl, 1 mM EGTA, 1% CHAPS, 1× protease inhibitor cocktail (Roche, Mannheim Germany. Product 05 892 970 001) 1 mM GTP. Lysates were centrifuged at 10,000 × g for 15 min and the supernatant was then incubated with goat anti-Rab27a polyclonal antibodies (1 mg; Sicgen, Portugal, Product Ab0023-200) previously immobilised on 50 μl of protein A/G UltraLink Resin (Thermos scientific Rockford IL, Product # 53132). After overnight incubation at 4 °C, the beads were precipitated by low speed centrifugation, washed with the same buffer and analysed using mass spectrometry to identify co-precipitating proteins.Sample preparation and mass spectrometryEach IP sample was washed in 200 µL 50 mM triethyl ammonium bicarbonate (TEAB; Sigma, UK, Product #T7408) before centrifugation (3000 × g, 30 s) and careful removal of the supernatant. The samples were then resuspended in 100 µL 50 mM TEAB and were reduced with dithiothreitol (5 mM DTT, 56 °C, 20 min; Sigma, UK, Product #D9779) and alkylated using iodoacetamide (55 mM IAA, room temperature, 15 min in the dark; Sigma, UK, Product #I1149). Tryptic digestion was performed as follows; 1 µg trypsin (Promega, UK, Product #V5111) was added to each sample along with 1 % protease max (20 ng/µL; Promega, UK, Product #V207A) and each sample was incubated at 37 °C in a thermomixer (650 r.p.m.) for 2 h. After 2 h a further 1 µg of trypsin was added and incubation at 37 °C in a thermomixer (650 rpm) continued for 16 h. A further 1 µg of trypsin was added to each sample after 16 h incubation and the samples were incubated at 47 °C in a thermomixer (650 r.p.m.) for 1 h. Samples were concentrated and de-salted using HyperSep C18 spin tips (10–200 µL size; Thermo Scientific, UK, Product #60109-412) using the manufacturers protocol. The samples were concentrated using a vacuum concentrator before resuspension in 5% acetonitrile + 0.1% formic acid (Merck, UK, Product #1.59002.2500). Each sample was analysed on a Sciex TripleTOF 6600 mass spectrometer coupled in line with a eksigent ekspert nanoLC 425 system running in micro flow (5 µL/min) mobile phase B (100% acetonitrile + 0.1% formic acid; Merck, UK, Product #1.59002.2500) over mobile phase A (0.1% formic acid; Merck, UK, Product #1.59013.2500). Samples (4 μl) were injected and trapped onto a YMC Triart C18 pre-column (5 mm, 3 µm, 300 µm ID; mobile phase A; 0.1% formic acid, B; acetonitrile with 0.1 % formic acid) at a flow rate of 10 µL/min. The sample was then eluted off the pre-column onto the YMC Triart C18 analytical column (15 cm, 3 µm, 300 µm ID) in line to a Sciex TripleTOF 6600 Duospray Source using a 50 μm electrode, positive mode +5500 V. Samples were analysed in IDA (Information Dependent Acquisition) mode. The following linear gradient was used: mobile phase B increasing from 3 to 30% over 68 min; 30 to 40% over 5 min, 40 to 80% for column wash and re-equilibration over 2 min (total run time 87 min). IDA acquisition mode was used with a top30 ion fragmentation (TOFMS m/z 400–1250; product ion 100–1500) followed by 15 s exclusion using rolling collision energy, 250/50 ms accumulation time (TOFMS/product ion; 1.8 s cycle). IDA data were searched using ProteinPilot 5.0.2 (iodoacetamide alkylation, thorough search with emphasis on biological modifications) against the SwissProt mouse database (January 2019) with the relevant human sequence added.Statistics and reproducibilityUnless otherwise stated all experiments were repeated independently at least three times and yielded similar results. In cases where data from one experiment are shown this is representative of the results of all three experiments.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Movie 1
Supplementary Movie 2
Supplementary Movie 3
Supplementary Code 1
Reporting Summary | nature communications | [
"Article"
] | [
"Cell biology",
"Cytoskeleton",
"Actin"
] | animal cells microtubules (MTs) actin filaments (AFs regulate transport infrastructure developed local model suggests MTs tracks for long-range transport cell centre periphery driven by kinesin dynein motors AFs myosin motors work downstream cargo to final destination MTs cells form polarised radial network tracks >10 μm from centrosome to periphery for long-range transport exceptions polarised AF complex AF short (1–2 μm random filament polarity not tracks for exemplified by co-operative capture) model of melanosome transport in melanocytes7,8 Skin melanocytes make pigmented melanosomes distribute to adjacent keratinocytes pigmentation photo-protection model proposes transport melanosomes into dendrites by sequential long-distance transport from cell body into dendrites along MTs (propelled by kinesin/dynein followed AF/myosin-Va-dependent tethering in dendrites in myosin-Va-null cells melanosomes move bidirectionally along MTs into dendrites accumulate cluster in cell body7defect partial albinism in mammals uneven pigment transfer melanocytes to keratinocytes mutant mouse human Griscelli syndrome type I patients studies revealed defects in mutant mice human GS types II III patients lacking GTPase Rab27a effector melanophilin activate myosin-Va on. 1FMN1- SPIRE1/2-deficient melanocytes show perinuclear melanosome clustering strategy regulators myosin-Va AF-dependent melanosome transport wild-type melanocytes melanosomes dispersed cytoplasm Loss myosin-Va proteins disruption blocks melanosome dispersion perinuclear melanosome clustering AF proteins-Va used siRNA deplete AF regulators screened for targets perinuclear clustering cells transfected with siRNA 72 h later imaged bright-field optics melanosome distribution bee swarm plot melanosome dispersion in cells transfected siRNA 28 18 20 22 Horizontal bars indicate populations compared no significant difference comparisons yielded significant differences p = <0.0001.western blot FMN1 GAPDH lysates melan-a-f melanocytes bee swarm plot melanosome dispersion adenovirus-infected melan-f cells 14 significant difference (p = <0.0001) population No other differences Confocal micrographs GFP-FMN1 expression melanosome distribution melan-f cells White boxes high-magnification overlay image (GFP-FMN1 green melanosomes melan-a cells transfected siRNA 72 h cells infected adenovirus GFP-SPIRE1/2 fixed 24 h processed immunofluorescence Cells imaged bright-field fluorescence optics melanosome GFP distribution Asterisks hyper-dispersed melanosome distribution bee swarm plot SPIRE1/2-depleted-infected melan-a cells low-magnification melanosomes dispersed hyper-dispersed Scale bars 50 μm 21 μm 10 μm 4 μm **** *** statistical significance differences p = <0.0001 p = <0.001 one-way ANOVA Significance indicators differences GFP control horizontal bar datasets No other significant differences Bars mean 25th 75th percentile data Source data c d e h Supplementary Source data filetested CC model cell normalisation technology MTs AF/myosin-Va melanosome results MTs essential for perinuclear clustering not peripheral dispersion MTs retard dispersion dependent myosin-Va dynamic AFs analysis proteins myosin-Va motor dispersing melanosomes AFs cell periphery melanosome dispersion myosin-Va-null cells myosin-Va melanosomes rapidly~1 μm/min into peripheral dendrites μm even MT-depleted cells data highlighted role myosin-Va dynamic AFs long-range transport melanosome distribution determined MT-dependent clustering AF/myosin-Va-dependent dispersion studies AFs organisation melanocytes polarised network for myosin-Va-driven transport melanosome mechanism unclear investigated SPIRE1/2 formin-1 (FMN1) AF assembly proteins essential regulators myosin-Va-driven melanosome dispersion FMN1 15 mammalian formins nucleate elongate unbranched AFs FMN1 FMN2 FMN subfamily FMN1 linked to limb development neurogenesis spermatogenesis FMN2 oocyte development memoryFMNs two domains (FH1 FH2) AF FH2 ends AFs proline FH1 interacts with profilin G-actin filament FMNs short~30 amino acids C-terminus FH2 SPIRE proteins N-terminal regions FMN1/2 formins large (859 amino acids no conserved sequence motifs genomes encode SPIRE genes SPIRE1 SPIRE2. SPIRE actin nucleators modular proteins N-terminus AF nucleation module KIND-catalytic C-lobe domain) interacts FSI motif FMN1/2 four G-actin-binding WH2 coupled C-terminus membrane-binding domain SB (SPIRE FYVE-type zinc finger SPIRE box N-terminal a-helix Rab GTPase-binding domain) synaptic vesicle transport regulator Rabphilin-3A FMN SPIRE proteins collaborate assemble AFs organelle combined function oocytes development repair DNA mouse oocytes SPIRE1 SPIRE2 cooperate with FMN2 generate AFs myosin-Vb-driven cortical transport Rab11identified myosin-V globular tail domain binding motif) between N- C-terminus modules SPIRE proteins co-ordinate recruitment myosin-V AF assembly to Rab11-positive intracellular evidence myosin-Va-mediated melanosome transport/dispersion in upon AF assembly activities FMN1 SPIRE1/2 SPIRE1/2 SPIRE1) recruited to melanosomes by Rab27a propose cargo-driven model of organelle dispersion Rab27a AF motors assembly proteins SPIRE1 SPIRE2 required for melanosome dispersion in myosin-Va/dynamic AF-based melanosome dispersion used siRNA knockdown test involvement AF regulatory proteins transfected wild-type melanocytes with siRNA mini-library 130 pools against transcripts AF regulators visually screened transfected cells to identify siRNA perinuclear melanosome clustering knockdown of proteins myosin-Va AF dispersion defects cells knockdown of FMN1 double SPIRE1 SPIRE2 induced melanosome clustering like Rab27a knockdown extent lower1b pigment area NT 86 ± 10.28%, Rab27a 31.28 ± 8.242% FMN1 45.07 ± 8.061% SPIRE1/2 52.56 ± 6.868%) five siRNA pools caused melanosomes >2/5 experiments targets not investigated single siRNA transfection reduced mRNA SPIRE1 SPIRE2 SPIRE1 dispersed melanosomes control cells Fig. 1a–e dispersed melanosomes% NT 86.43 ± 4.887% SPIRE1 46.93 ± 10.64% SPIRE2 82.57 ± 13.26%) SPIRE2 depletion lower proportion dispersed melanosomes SPIRE1/2 versus SPIRE1 depletion Fig 1b–e % dispersed melanosomes SPIRE1/2 12.73 ± 5.878%) SPIRE1-depleted cells melanosomes cleared centre enriched periphery control NT cells SPIRE2 SPIRE1/2-depleted Fig. 1d e HD melanosomes% 6.324 ± 2.645) SPIRE1 uniform melanosome dispersion pattern wild-type melanocytesQ-RT-PCR FMN1 SPIRE1 mRNA expression exceeded SPIRE2 fivefold Fig. 1f SPIRE1 17.33 ± 1.883 SPIRE2 3.468 ± 0.5726 FMN1 27.2 ± 4.17) FMN2 expression unlikely melanocytes FMN1 SPIREs disperse melanosomes add-back experiments human SPIRE1 SPIRE2 mouse FMN1 proteins expressed SPIRE1/2 depleted FMN1-deficient melanocytes melanosome clustering depletion (Fig. 1d–h pigment area –protein expression 50.15 ± 9.228 GFP 48.66 ± 8.618 FMN1 88.43 ± 7.972 % cells dispersed melanosomes SPIRE1/2 11.83 ± 8.02 SPIRE1 92.58 ± 7.072 SPIRE2 94.67 ± 11.36) cells lower levels SPIRE2 melanosomes hyper dispersed melanosomes GFP 3.088 ± 2.748 SPIRE1 0.4695 ± 0.8132 SPIRE2 49.96 ± 11.44) SPIRE2 less efficient dispersing melanosomesSPIRE1/2 FMN1 generate melanosome FMN SPIRE cooperate AF assembly used latrunculin-A AFs SPIRE1/2-transfected melan-a-f cells observed melanosome distribution AF content reduced melanosome dispersion AF content-a cells melan-f SPIRE1/2-depleted melan. 2a–d pigment area latrunculin-A melan-a 68.26 ± 9.956 51.56 12.14 melan-f 45.48 ± 11.9 45.89 13.29 NT-transfected melan-a 85.18 ± 8.549 43.94 16.09 SPIRE1/2-depleted melan-a 36.9 ± 9.668 35.28 ± 10.26 melan-a 413.7 ± 186.1 167.6 75.5-f 241.6 112.3 241.9 93.26 NT-transfected melan-a 767.3 ± 331.9 304.8 ± 237.8 SPIRE1/2 siRNA-transfected melan 304.7 ± 145.7 213.2 ± 74.68) re-expression GFP-Fmn1 increased AF content melan-f cells 2g 128.6 71.79 GFP 47.13 21.52)data indicate FMN1 SPIRE1/2 dynamic AFs myosin-Va-dependent melanosome dispersion. 2FMN1 SPIRE1/2 generate latrunculin-A-sensitive AFs melanosome dispersion.melan-a melan-f cells plated glass transfected with siRNA infected GFP-FMN1 incubated latrunculin-A 60 min Cells fixed stained fluorescent phalloidin reveal AFs Experimental Fluorescent bright-field images distribution AFs melanosomes melanocytes Scale bar 15 μm d e f h Bee swarm plots melanosome dispersion AF abundance melanocytes presence latrunculin-A f show data same population number cells measured indicated brackets plot Numbers e f same **** indicate significant difference p = <0.0001 <0.001 ANOVA no significant difference Data from three experiments Bars indicate mean 25th 75th percentile data Bars linking datasets indicate pairs compared similarity Source data for b d–f h supplementary Source data file used high-resolution field emission microscopy replica transmission EM analyse cytoskeleton network surrounding melanosomes melan-a cells variantsresults fluorescence microscopy high-magnification FESEM images revealed melanosomes in melan-a cells surrounded by filaments compatible with AFs diameter (mean filament diameter 8.6 ± 0.3 nm n = 87) morphology (dense bundles branched networks Fig. 3a–c). Melanosomes embedded in filament network filaments visible above below (Fig. 3g AFs surface organelles. 3g melan-f cells melanosome filaments absent decorated with short filaments (Fig. 3f FESEM immuno-electron microscopy filament network labelled by phalloidin part complex actin network (Fig. 3h confirmed rapid freeze dry/replica TEM myosin S1 decorated filaments surrounding melanosomes. 3j differences melanosome AFs measured filament length FESEM images (Fig. shorter melanosome filaments in melan-f cells. mean filament length nm melan-a = 261.7 ± 160.6 versus melan-f = 116.1 ± 72.32 p < 0.0001)differences in AF density dimensions in melan-a cells depleted SPIRE1/2 Rab27a Fig 2) short AFs FMN1 SPIRE1/2 for extension network data support hypothesis SPIRE1/2 FMN1 assemble AFs myosin-Va constructed at melanosome membrane.Fig. 3High-resolution electron microscopy reduction in melanosome-associated AFs in FMN1-deficient melanocytes controls Wild-type-a FMN1-deficient cells prepared for electron microscopy Cells a d increased magnification high magnification insets c f red line d indicates cell outline Arrows c point loose network filaments around melanosomes arrowheads show melanosomes dense filament network Colourised filaments in indicate linking melanosomes Immuno-electron microscopy melan-a cells phalloidin labelling gold particles actin filaments around melanosomes inter-melanosome filaments Higher magnification shows colourised filaments gold particle labelling myosin S1 decoration filaments around melanosomesFESEM melanosomes multiple AFs melan-a cells stubs melan-f melanosomes Bee swarm plot size distribution AFs melanosomes FESEM images (actin filaments 433 (melan-a) 443 (melan-f). difference p = <0.0001 melan-a-f cells Mann–Whitney test Source data supplementary Source data file Bars mean 25th 75th percentile data Scale bars a b d 10 μm c e–g 1 μm h k l 200 nm i j 100 nm N-terminus AF nucleation FMN interaction essential SPIRE melanosome tested SPIRE protein domains melanosome dispersion AF nucleation membrane-binding module dispersed melanosomes intact SPIRE1 <0.001) SPIRE1/2-depleted melanocytes KW dispersed melanosomes greater than MSFH GFP (Fig. 4a–c mean % cells dispersed melanosomes KW 58.85 ± 10.91 MSFH 18.69 ± 4.37 GFP 4.65 ± 3.443%)modules essential SPIRE1 AF nucleation module melanosome dispersion KW SPIRE2 melanosome hyper-dispersion (Fig. 4 mean % HD melanosomes KW 36.69 ± 7.746 SPIRE2 35.91 ± 10.72 SPIRE1 = SPIRE1 membrane targeting domain essential for cytoplasmic melanosome dispersion differ association membranes. FMN AF nucleation domains SPIRE1 essential melanosome dispersion domain structure human SPIRE1 correspondence with mutant chimeric proteins Numbers amino acid boundaries globular tail domain SPIRE box FYVE-type zinc C-terminal flanking sequences-proteins cells depleted of SPIRE1/2 infected with adenoviruses fixed 24 h processed for immunofluorescence imaged bright-field fluorescence optics melanosome protein distribution/expression Scale bars = 100 20 3 μm-magnification images Boxes in KW-Rab27a images indicate region green = KW-Rab27a magenta = melanosomesbee swarm plot human SPIRE1/2-depleted cells (50 each melanosomes dispersed hyper-dispersed Results three experiments Source data Supplementary Source data file tested WMSFH KMSFH truncations lacking KIND WH2 domains neither truncation dispersed melanosomes SPIRE1/2-depleted cells (Fig. 4b % dispersed melanosomes WMSFH 19.01 ± 1.693 KMSFH 23.08 ± 2.374) cooperation SPIRE1 FMN1 required AF assembly melanocytes Similar results human SPIRE2 depletion integrity expressed SPIRE1 SPIRE2 proteins FMN1 proteins confirmed by western blotting assembly SPIRE interaction required FMN1 function melanosome similar approach role FMN1 expression C-terminus fragment FH1 FH2 restored peripheral melanosome distribution intact FMN1 reciprocal N-terminus fragment (Fig. 5 GFP = 50.69 ± 14.29%, FMN1 = 89.92 ± 6.598%, N-term = 51.44 ± 10.17% FH1-FH2-FSI = 89.28 ± 8.512%)Truncations FSI FH1 disperse melanosomes. 5 ΔFSI 79.52 ± 12.34% FH2-FSI 42.77 ± 13.89%) point mutations FH2 contact AF FSI/SPIRE-KIND interaction reduced FMN1 function. 5 pigment area I1074A 57.77 ± 12.73%, K1229D 77.75 ± 12.53 K1418E 72.1 17.34 FH1 FH2 domains FSI motif essential for FMN1 function assembles AFs SPIRE melanocytes. AF assembly SPIRE interaction domains FMN1 essential melanosome dispersion structure FMN1 composition truncation mutants chimeric proteins asterisks point mutations Numbers amino acid boundaries cells plated infected adenoviruses proteins Cells fixed 24 h processed immunofluorescence intracellular distribution expressed protein melanosomes observed fluorescence microscope Scale bars 20 μm 2 μm magnified region Dashed boxes FH1-FH2-Rab27a images indicate region merged image melanosomes FH1-FH2-Rab27a coloured magenta greenbee swarm plot pigment dispersion cells proteins n 30 59 (FMN1) 18 (N 70 (FH1-FH2-FSI), 31 20 37 35 28 (K1418E 51 (FH1-FH2-Rab27a). **** ** * differences p = <0.0001 0.05 no difference one-way ANOVA Significance indicators differences positive (FMN1 negative (GFP controls Results three experiments Bars mean 25th 75th percentile data FH formin homology FSI formin-SPIRE interaction sequence Source data Supplementary Source data file membrane-binding module SPIRE related Rab-binding domain Rab27/3 mammalian Rab family>60 proteins Rab27/3/8 branch regulate transport exocytic vesicles plasma Slp/Slac effectors common Rab binding domain FYVE-type zinc finger α-helical regions H1 H2 Rab3/27 studies sequence similarity SPIRE proteins H1 helix Rab27/3 effector RPH3A Rab3a:SPIRE1 interaction FYVE-type zinc finger membrane-binding suggest SPIREs Rab27 effectors (Fig. 4asequence alignments showed similarity SB H1 regions Rab27/3 effectors Mlph conservation residues interaction Fig. 5a Phylogenetic analysis SPIRE proteins Rab27/3/8 effectors grouped family Fig. 5b). support SPIRE protein function melanosome regulated by Rab27a.GTP-dependent interaction SPIRE proteins GST pull-down experiments bacterially expressed purified GST-Rab27a fusion proteins-locked Rab27a-Q78L Rab27a-T23N lysates HEK293 cells expressing Myc-epitope-tagged SPIRE1 SPIRE2 (Fig. 6a–c). blotting GST-Rab27a-Q78L pulled SPIRE1 SPIRE2-T23N SPIRE1 interacted with Rab27a-Q78L results wild-type GST-Rab27a GTPγS GFP-SPIRE1-MSFH Fig. 6) 6SPIRE1/2 Rab27a membrane-binding C-terminischematic domain structure SPIRE1/2 Mlph truncations interaction studies Interaction SPIRE1/2 Rab27a investigated GST pull-down BiFC assays c e Western blots Ponceau S stained filters pull-down assays interaction GST-Rab27a Q78L inactive T23N mutants SPIRE1/2 truncations (Myc-tagged GFP-tagged contrast enhanced interaction SPIRE2 Rab27a-Q78L-T23N Fluorescence images bee swarm plot BiFC assay interaction Rab27a SPIRE1 SPIRE2 HEK293a cells mCherry transfection efficiency vYFP BiFC interaction bee swarm plot BiFC signal cells SPIRE1/2 without active inactive Rab27a mutants Data three experiments ** differences p = <0.0001 <0.001 <0.01 adjacent dataset Rab27a wild-type/SPIRE1/2 expressing cells one-way ANOVA No other significant differences Two-way ANOVA comparison BiFC signal Rab27a proteins SPIREs no significant differences Bars mean 25th 75th percentile data Scale bar 250 μmLine plots binding GFP-SPIRE1-MSFH 4)-Mlph-RBD 3) GST-Rab27a-Q78L concentrations Data mean values ± SEM equilibrium dissociation constants globular tail binding SPIRE box zinc finger C-terminal flanking sequences Slp/Slac-proteins western blotting Source data Supplementary Source data file investigated Rab27a:SPIRE1/2 interaction mammalian cells Rab27a co-expressed HEK293 cells C- N-terminus fragments protein higher BiFC signal cells Rab27a Q78L mutant inactive mutants T23N N133I mean BiFC signal similar vYFP. 6d BiFC SPIRE1/SPIRE2 89.35/85.41%, T23N 28.35/5.969%, Q78L 75.21/85.23% N133I 24.71/6.601% vYC 8.637/16.65%) Rab27a:SPIRE1/2 interaction pigment cells expressed GFP-SPIRE1 B16-F1 cells immuno-precipitated Rab27a tested co-immunoprecipitation GFP-SPIRE1mass spectrometry identified Rab27a (13 SPIRE1 (18 peptides interaction pigment cells Fig. 7 Tables 2 3) data concur pull-down assays approaches (Figs. 7 8 5–8) indicate SPIRE1/2 Rab27a effectors proteomic studies interaction Rab27a SPIRE1 in HUVEC endothelial. 7Rab27a recruits SPIRE1 to melanosomes melanocytes.Melanocytes transfected plasmids expression proteins C-terminus EGFP Cells fixed after 48 h stained GFP-specific antibodies proteins intracellular distribution protein melanosomes observed confocal microscope confocal z-sections distribution protein pigmented melanosomes images Upper panels whole cells Boxes regions high magnification distribution protein Line plots fluorescence intensity profile plots regions a–c d e melan-a melan-ln melan-ash cells Yellow lines borders transfected cells Scale bars = 20 μm 3 μm images magnified portions Source data Supplementary Source data file.Fig. SPIRE1/2 with melanosome Rab27a in melanocytesmelan-ln melan-ash melanocytes infected viruses expressing proteins Cells fixed after 24 h stained GFP-specific antibodies intracellular distribution GFP-fusion proteins melanosomes observed confocal microscope schematic structure mini-Va myoSPIRE1/2 proteins interaction with membrane Rab27a-binding domain SPIRE globular binding domain confocal z-sections distribution expressed proteins melanosomes colocalisation melanosomes false-coloured magenta myoSPIRE1/2 low high-magnification images Boxes low-magnification indicate area high-magnification Scale bars 10 μm main images 2 μm magnified Arrows indicate colocalisation myoSPIRE melanosomes Cell outlines white lines phase contrast images bee swarm plot effect expression myoSPIRE proteins melanosome distribution-ash mean pigment area measurements from four experiments ten cells expressing proteins difference p = <0.0001 0.001 none between datasets bars Bars indicate mean 25th 75th percentile data Source data supplementary Source data fileC-terminus module SPIRE proteins interacts with-binding tested SPIRE1 truncations Rab27a-Q78L GST- pull-down assay interaction GST-Rab27a-Q78L with Myc-SPIRE1-MSFH-SPIRE2-MSFH proteins not Myc-SPIRE1-KWM Rab27a-binding site SPIRE generated truncations tested Rab27a-Q78L interaction SPIRE1-SFH interacted with Rab27a-Q78L FYVE-only SF proteins lesser extent C-terminus SFH fragment SPIRE interacts with Rab27a interaction mechanism conserved Rab27 effectors.SPIRE proteins interact Rab27a lower affinity than compared affinity with other Rab27a interactions quantified affinity GFP-SPIRE-MSFH GFP-Mlph-RBD with GST-Rab27a-Q78L GFP depletion from HEK293 cell lysates dissociation constants 143 (±25) nM 707 (±155) nM maximum binding levels 39.4% 12.0% for GFP-Mlph-RBD GFP-SPIRE1-MSFHdata correspond Kd 112 nM Rab27a-Q78L:Mlph RBD interaction SPIRE1 weaker Rab27a interactor Mlph45 determine Kd SPIRE2:Rab27a interaction data align with assays Rab27a interacts with SPIRE1 6b).SPIRE1/2 associate with melanosomes Rab27a-dependent SPIRE proteins Rab27 effectors required for melanosome dispersion Rab27a present cytoplasmic face melanosome membrane associate melanosomes expressed GFP-tagged SPIRE proteins Rab27a in melan cells confocal microscopy localisation proteins distributed punctate pattern cytoplasm (Fig. high-magnification imaging plots SPIRE1 Rab27a SPIRE2 overlapped with melanosomes Pearson linear correlation coefficient 0.823 for Rab27a 0.623 SPIRE1 −0.081 SPIRE2) Myc-SPIRE1-MSFH overlapped with melanosome protein tyrosinase-related protein 1 melanosome-targeted proteins (co-expressed mRuby3-Rab27a GFP-myosin SPIRE1 associates with melanosomestest association Rab27a dependent repeated experiment using melanocytes deficient Mlph Rab27a-ash revealed Rab27a associated melanosomes both cell types dispersed melan-ash SPIRE1 associated perinuclear clustered melanosomes melan-ln cells (Fig. 7d e Pearson linear correlation coefficient = 0.601 0.739 for Rab27a −0.230 SPIRE1 melan-ash melan-ln). association SPIRE1 melanosomes dependent interaction endogenous Rab27a SPIRE1 Rab27 effector investigate Rab27a:SPIRE2 interaction melanocytes developed assay changes melanosome distribution melan-ln melanocytes interaction Rab27a (Fig. modified ‘minimyosin’ protein Rab27 effector with SPIRE proteins create ‘myoSPIRE’ fusions (Fig. tested disperse clustered melanosomes in melan-ln melan-ash cells presence endogenous Rab27a). melan-ln both myoSPIRE proteins not Rab27a dispersed melanosomes compared GFP alone (Fig.pigment area GFP 26.3 ± 7.43%, myoSPIRE1 63.4 7.96% 66.6 3.76% mini-Va 62.3 ± 10.8 % Rab27a 31.3 ± SPIRE1 rescue melanosome transport Rab27a myosin-Va Mlph Fig. 9 pigment GFP 24.8 ± 7.91 SPIRE1 18.5 ± 3.95 Mlph 75.4 ± 18.8) affinity SPIRE1:myosin-Va/Rab27a low myosin-Va low affinity Rab27a:SPIRE1/2 SPIRE2 interacts myosin-Va-GTD (0.9 ± 0.11 μM lower affinity Mlph melan-ash Rab27a dispersed melanosomes minimyosin myoSPIRE GFP. 8c pigment area GFP 31.9 ± 6.80% myoSPIRE1 37.8 ± 38.6 ± 5.97% mini-Va 39.2 ± 1.97% Rab27a 66.9 ± 6.20%). SPIRE1 SPIRE2 interact Rab27a melanosome membrane spots myoSPIREs adjacent melanosomes melan cellsdependent targeting SPIRE proteins to membrane important for uniform melanosome dispersion used two approaches test SPIRE:Rab27a interaction exploited sequence conservation between SPIREs effectors mutants SPIRE1:Rab27a GST Rab27a-Q78L pull-down assay mutation glutamic acid residue codon human SPIRE1 to alanine lysine blocked interaction Rab27a consistent with electrostatic contact equivalent residue (E14/54) in Mlph/RPH3A chain arginine 80/83 switch II region Rab27b/Rab3a 5d Expression SPIRE1E548K in SPIRE1/2 depleted melan-a cells dispersed melanosomes compared GFP 97.61 2.044) cells expressing SPIRE1E548K melanosomes hyper- uniformly dispersed SPIRE1 56.85 ± 1.985) spots SPIRE1E548K protein less frequently associated with melanosomes wild-type SPIRE1 5e investigated functional effect tethering active SPIRE2 (KW) fragment to melanosomes fusion melanosome-targeted non-functional Rab27aSF1F4 mutantFRAP studies found association Rab27a with melanosomes stable KW-Rab27a likely KW-Rab27a localised to melanosomes SPIRE2-KW restored uniform dispersion in SPIRE1/2-depleted cells than SPIRE2-KW (Fig. 4a % cells dispersed melanosomes KW-Rab27a = 98.01 2.0%) association with Rab27a melanosome membrane enhances SPIRE proteins disperse melanosomes supports differences in Rab27a interaction affinity SPIRE1 SPIRE2 underlie different patterns melanosome dispersal targeting FMN reduces function melanosome importance FMN1/SPIRE interaction investigated FMN1 with GFP-Fmn1 distributed throughout cytoplasm not enriched adjacent to melanosomes FMN1 SPIRE1 associate with melanosomes (Fig. 1f FMN1 may interact transiently with melanosome-associated SPIRE proteins vitro studies AF assembly SPIRE1 Fmn2used Rab27aSF1F4 mutant fusion strategy FH1-FH2-Rab27a fusion protein targets FH1 FH2 domains FMN1 to melanosomes measured activity melanosomes melan-f cells (Fig. FH1-FH2-Rab27a increased melanosome dispersion less effective than intact FMN1 FH1-FH2-FSI fragment (Fig. mean pigment area FH1-FH2-Rab27a = 55.42 ± 17.49) association with melanosomes reduces FMN1 function FMN1 associates with melanosome-associated SPIRE1/2 tested FMN1 related protein FMN2 melanosome transport melan-f cells FMN2 contains FH1 FH2 domains C-terminus FSI co-translationally myristoylated targeted cell FMN2 generate AF near cell membrane distant from melanosomes melan-f cells (Figs. 1f intact FMN2 FMN2 FH1-FH2-FSI fragment distributed throughout cytoplasm restored melanosome dispersion similar efficiency GFP-FMN1 10a–c mean pigment area GFP-FMN1 = 84.9.544 GFP-FMN2 81.72 ± 8.011 FH1-FH2-FSI 86.42 ± 10.84). N-terminus fusion FMN2 to GFP dispersed melanosomes lower efficiency GFP-FMN2 Fig. 10b pigment area GFP 44.17 ± 9.31 FMN2-GFP 58.37 ± 7.818) N-myristoylation membrane targeting function FMN2-GFP blocked approaches non-N-myristoylatable Fmn2-G2A mutant treatment N-myristoyltransferase inhibitor (NMTi) IMP-1088 strategies redistributed FMN2-GFP cytoplasmic pattern enhanced function transport Fig 10b pigment area FMN2-[G2A-GFP 79.71 ± 11.72%, FMN2-GFP + NMTi = 79.03 ± 9.488%) labelling confirmed N-myristoylation status FMN2-GFP G2A mutant NMTi blocking modification support Fmn1 associate with membranes melanocytes investigated myosin-Va-dependent melanosome dispersion melanocytes role dynamic AFsdata indicate FMN1 SPIRE proteins regulate melanosome dispersal melanocytes FMN1 SPIRE1/2 extend AFs from membranes cell-wide network melanosomes Rab27a recruit SPIRE1/2 myosin-Va Mlph Rab27a drives melanosome dispersion AF motors myosin-Va AF machinery SPIRE1/2 FMN1) propose model (Fig. Rab27a-GTP recruits two effector complexes to melanosomes myosin-Va/melanophilin SPIRE1/2 myosin-Va:SPIRE1/2 interaction may stabilise SPIRE1/2 melanosome membrane integrate SPIRE1/2 activity with myosin recruitment two effector complexes requires two pools of Rab27a interaction domain SPIRE1/2 with other effectors Mlph Slp2 Interaction effectors obscures surface Rab27 unlikely single Rab27 protein interact with both effectors simultaneously Rab27a interaction may activate SPIRE1/2 membrane KIND/FYVE Active SPIRE1/2 may recruit FMN1 FMN1 associate elongate ends short AFs melanosome membraneformation AF network polarised +/barbed ends oriented from melanosome membrane towards melanosomes cell periphery Fig. 9b). consistent with melan-a not melan-f SPIRE1/2 depleted melanocytes harbour AFs from adjacent melanosomes myosin-Va attached to adjacent melanosome towards +/barbed ends dispersing melanosomes (Fig. 9c).Fig. 9A model Rab27a AF-driven melanosome transport myosin-Va SPIRE1/2 FMN1 model melanosome myosin-Va SPIRE actin filament Rab27a present on melanosome membrane activity FMN1 melanosomes grey black colours indicate AF myosin-Va associated with melanosomes same colour grey black colours indicate position melanosomes start finish dispersive transport pattern myosin-Va/AF-dependent inter-organelle repulsion drive dispersal melanosomes throughout cytoplasm wild-type melanocytes (Fig. 9d). inter-connected three-dimensional network organelles tracks motors allow cell-wide coupling of local organelle-associated force generatorsmyosin-Va motors FMN1/SPIRE1/2 AF assembly machinery). arrangement force disperse large rigid organelles (1000 s maintain distribution than conventional transport models single organelles pulled limited tracks motors used open-source modelling framework Cytosim simulate melanosome dispersion by myosin-Va motors FMN1/SPIRE1/2 AF builders modelled AFs polarised −/pointed ends attached Rab27a/SPIRE to melanosomes +/barbed ends growing FMN1) into cytoplasm myosin-Va anchored Rab27a/Mlph to melanosome motor domains cytoplasm ‘melanosim-attached’ melanosomes clustered cell AFs emerged from melanosomes myosin-Va motors melanosomes attached to filaments linked melanosomes moved apart Fig. 11a Movies 1 2 resulted in rapid global dispersion of clustered melanosomes cytoplasm predicted resembled previous live-cell melanosome dispersion cells activation activatable myosin-Va dispersion reduced in simulations free AFs pointed ends not attached to melanosomes (‘melanosim-detached’ 11a 3)observations support hypothesis melanosome Rab27a generate organelle motors filament builders disperse organelles cytoplasm eukaryotic cell data support possibility melan-f cells SPIRE1/2 depleted melan-a cells show fewer melanosome AFs fusion of SPIRE2 FMN1 to melanosomes via Rab27aSF1F4 mutant boosts SPIRE2-KW reduces FMN1-FH1-FH2 SPIRE1/2 re-targeting FMN) to cell membrane-myristoylation reduces function processive transport activity myosin-Va essential for melanosome dispersion15 evidence non-conventional organisations acto-myosin cytoskeleton exist in other cell types sarcomere muscle cells stress fibres migrating fibroblasts structures allow amplification force myosin-II motors cell shape changes work in oocytes myosin-Vb, SPIRE1/2 FMN2 form meshwork AFs drive flow Rab11-positive vesicles to plasma membrane suggests similar AF-dependent organelle transport systems in somatic mammalian cells work highlights differences between SPIRE1 SPIRE2 functionSPIRE1 interaction Rab27a localised to melanosomes drove uniform melanosome dispersion SPIRE2. uniform melanosome dispersion required higher SPIRE2 expression lower levels hyper-dispersion KW-Rab27a fusion protein dispersed melanosomes efficiently SPIRE2-KW SPIRE1/2 mutants compromised promoted hyper-dispersion interaction Rab27a important uniform dispersion hyper-dispersive activity SPIRE1 gene highly expressed than SPIRE2 depletion of SPIRE1 SPIRE2 reduced dispersion perinuclear clustering hyper-dispersion data suggest SPIRE1 significant in melanosome transport SPIRE2. mechanism SPIRE2 hyper-disperses melanosomes low expression levels unclear low concentration SPIRE2 activated by interaction with Rabs AFs local accumulation melanosomes higher concentrations activated by Rab27a triggering uniform melanosome dispersal Fig. Future studies address suggest data implication for transport in cell types Rab27 Rab proteins SPIRE1/2 FMN myosin-V proteins expressedsimilarity between proteins vesicle AF meshwork asymmetric spindle positioning meiosis in mouse oocytes (SPIRE1/2 FMN2 Rab11 myosin-Vb similar proteins drive rapid-range AF-dependent organelle transport different cell types suggests machineries transport other animal cell types Rab27 regulates dispersal secretion organelle contents in cell types inflammatory mast cells lytic cytotoxic lymphocytes MVBs cancer oligodactylism phenotype FMN1 knockout mouse supports function FMN1 secretory transport limb development requires communication cells dependent on secreted proteins25 behavioural phenotypes FMN2 SPIRE1 mutant mice caused by altered secretion neuropeptides growth factors61 test Rab27a AF-dependent transport organelles other cell types procedures mice validated by National Centre for Biotechnology Ethics Committee approved National CSIC Ethics Committee authorised by Autonomous Community of Madrid Spanish legislation European Directive 2010/63/EU approval reference PROEX 343/15.mice housed CNB animal facility fed water rodent light/dark cycle 08:00–20:00 European Spanish norms animal welfare recommendations.Derivation maintenance cultured FMN1-deficient melanocytes derived mice FMN1 loss allele crossed with Ink4a-Arf mutant mice pups homozygous FMN1pro heterozygous Ink4a-Arf Genotyping embryos PCR Melanocytes derived from dorsal skin newborn mutant Cultures melan-f-ash-ln-a melanocytes HEK293a maintained infected adenovirus vectors transfected siRNA oligonucleotides 66.7 fmol/μl transfection mixture Oligofectamine transfection reagent 300-fold optiMEM medium melanocyte cell lines Wellcome Trust Functional Genomics Cell Bank cultured transfected siRNA plasmid DNA 1.67 ng/μl Fugene 6 transfection reagent diluted 300-fold66 siRNA oligonucleotides Supplementary Table 4. purchased from Dharmacon Thermo-Fisher Sigma Genosys UKreal-time PCRPrimers probes Q-RT-PCR targets Sigma Genosys designed Primer Express software Probes labelled 5′ 3′-ends fluorophore 6-FAM (6-carboxyfluorescein TAMRA Table 5) mRNA melan-a cells six-well plates transfected siRNA Seventy two hours later cells harvested mRNA extracted RNeasy Mini extraction kit cDNA generated Moloney Murine Leukaemia Virus M-MLV reverse transcriptase) random primers signal-PCR pool 5% cDNA samples generated diluted DEPC water (1:4 template target gene GAPDH Q-RT-PCR assays standard curve signal template concentration straight lines R2 > 0.99 linear relationship expression targets siRNA-transfected cells cDNA diluted 1:32 DEPC water reagents added per 96-well plate 6.5 μl TaqMan Fast 2× PCR Master Mix 0.4 μl forward primer reverse primer 0.25 μl probe 3 μl cDNA 2.45 μl DEPC water three technical repeats performedReaction plates sealed film centrifuged Q-RT-PCR performed StepOnePlus Real-Time PCR system mode CT values determined StepOne software slope intercept R2 values calculated standard curve-PCR CT values siRNA-transfected samples processed ‘quantity (CT-I)/S = LQ 10LQ = Q Q × (1/MNT) = GOIP MNT non quantity value GOIP/GP = normalised expression target relative GAPDH Expression FMN1 detected Taqman assay Mm01327668_m1 Thermo-Fisher Scientific measurement expression SPIRE FMN MLPH genes RNA isolation kit Macherey-Nagel used reduce melanin increase purity total RNA step protocol repeated RNA spectrophotometric determination cDNA generated Qiagen QuantiNova® Reverse Transcription Kit internal control RNA reverse transcription PCR primer sets designed primer3 Primer-Blast constructed Sigma-Aldrich templates Supplementary Table 6. testing generated cDNA from melan-a total RNA plasmid DNAs positive controls templates external plasmid-cDNA standard curve concentrations Linearised plasmid DNA diluted in watercopy number standard DNA molecules determined formula ((X g/μl DNA)/[plasmid length × 660]) × 6.022 × 1023 = Y molecules/μl). absolute quantification Q-RT-PCR Rotor-Gene Q thermocycler reaction triplets performed CT values determined Qiagen Rotor-Gene Q Series Software CT values compared standard curve concentrations target QuantiNova® SYBR® Green PCR Kit (GE Healthcare Life Sciences protocol.ImmunoblottingSDS–PAGE 4–15% gradient SDS–PAGE gels Mini-PROTEAN Tetra Cell Molecular weight standards BLUeye Prestained Protein Ladder Proteins transferred to PVDF membrane (Merck Millipore) HoeferTM TE22 Mini Tank Blotting Unit primary secondary antibodies Supplementary Tables 7 8. Uncropped western blots Supplementary Information file.Signal detected Li-Cor infrared scanner chemiluminescence Forte recorded Image Quant LAS4000 systemprotein sequence alignments tree C-terminal SPIRE1 SPIRE2 N-terminal MLPH MYRIP Rabphilin-3A NOC2 Clustal Omega Multiple Sequence Alignment tool Hinxton cDNA sequences NCBI Gene database accession numbers sapiens-SPIRE1: NM_020148.2-SPIRE2 AJ422077-MLPH_024101.6-MYRIP_001284423.1-RPH3A_001143854.1-RPH3AL (NOC2) NM_006987.3. Generation phylogenetic trees Rab27 proteins alignments Clustal Omega manual confinementscDNA sequences NCBI Gene database accession numbers Hs-RIMS1: NM_014989.5-RIMS2 NM_001100117.2 Hs-RPH3A NM_001143854.1-RPH3AL (NOC2) NM_006987.3 Hs-MLPH NM_024101.6-EXPH5 NM_015065.2 Hs-MYRIP NM_001284423.1 Hs-SYTL1_005246022.3-SYTL2 NM_032943.4-SYTL3 NM_001242384.1-SYTL4 NM_080737.2-SYTL5 NM_001163334.1 Hs-SPIRE1: NM_020148.2 Hs-SPIRE2: AJ422077 Evolutionary analyses phylogenetic tree formation in MEGA7 evolutionary history inferred Maximum Likelihood method JTT matrix-based bootstrap consensus tree from 500 replicates evolutionary history Branches <50% replicates collapsed percentage trees Initial tree(s) heuristic search obtained Neighbour-Join BioNJ algorithms distances topology superior log likelihood value analysis involved14 amino acid sequences 188 positions final datasetbacterial mammalian protein expression generated cloning AccuPrime Pfx Waltham Pfu DNA polymerases endonucleases T4 DNA ligase New England Biolabs Frankfurt mutants In-Fusion HD cloning kit DNA sequencing LGC Genomics Source Bioscience (Nottingham Table 9 vectors Amino acid boundaries related cDNA sequences human SPIRE1 SPIRE2 murine FMN1 FMN2 myosin-Va.3) Mlph mRuby3 fusion protein encoding vectors pKanCMV-mClover3-mRuby3 vector cDNA sequence Michael Lin provided Addgene protein expression GST-Rab27a-WT Q78L T23N mutant proteins Escherichia Rosetta bacterial cells Bacteria cultured LB medium (100 mg/l ampicillin 34 mg/l chloramphenicol 37 °C A600 nm OD 0.6–0.8 Protein expression induced 0.2 mM isopropyl-β-D-thiogalactopyranoside-Aldrich continued 16–20 °C 18–20 h Bacteria harvested lysed ultra-sonicationproteins purified ÄKTApurifier system GSH-Sepharose 4B beads size-exclusion chromatography Proteins concentrated ultrafiltration Amicon Ultra centrifugal filters molecular weight cut offs final protein purity estimated SDS–PAGE Coomassie staining.Nucleotide loading GST interaction GST-Rab27a SPIRE1 proteins loaded non-hydrolysable GDP GTP) analogues Sigma assays nucleotide exchange 5.8 mg purified GST-Rab27a fusion protein 2.5 mg pure Rab27a protein mixed fourfold molar excess GDPβS GTPγS (1.13 mM 12.5 units alkaline phosphatase buffer (20-HCl pH 7.4 200 NaCl 1 MgCl2 2 DTE (NH4)2SO4 100 ZnCl2) volume 800 μl mixture incubated 17 h 4 °C wheel next CIAP buffer exchanged by RAB polarisation buffer (20-HCl NaCl 5 MgCl2 1 DTE 5 GDPβS/GTPγS NAP-10 columnsGST pull-down HEK293 cell cells cultured transfected plasmid DNA input GFP Myc SPIRE1 deletion mutants pull-downs mutants expressed HEK293 cells protein expression levels analysed western blotting-GFP anti-c Protein bands quantified ImageQuant TL normalised lowest signal band cell lysates pull-downs adjusted cells transfected vectors fluorescent protein Myc-epitope-tagged proteins transfection cells lysed buffer (25 Tris-HCl pH 7.4 150 NaCl 5 MgCl2 glycerol Nonidet P-40 1 mM PMSF protease inhibitor centrifuged × g 4 °C 20 min remove debris GST pull-down assays 50 μg GST-RAB27A proteins 25 μg GST control GSH-Sepharose 4B beads 1 h 4 °C rotating wheel Beads washed pull-down buffer-HCl NaCl MgCl2 glycerol P-40 incubated cell lysates 2 h 4 °C wheel1 ml lysates lowest protein higher diluted buffer 1 ml Beads washed four times proteins eluted 1× Laemmli buffer denatured 95 °C 10 min analysed immunoblotting GST pull-down HEK293 cell lysates increasing concentrations GST-Rab27a-Q78L protein concentration GFP-SPIRE1-MSFH ~50 nM lysates pooled distributed beads GST-Rab27a-Q78L protein Beads pelleted cell lysate supernatant diluted 1:4 aqua dest sample 20 °C 10 min concentration-dependent binding Rab27a-Q78L GFP-SPIRE1-MSFH determined fluorospectrometric analysis FluoroMax- 4 Spectrofluorometer GFP-Mlph-RBD HEK293 cells purified GST-Rab27a-Q78L proteins AcGFP1 green fluorescent protein excited 488 nm emission 507 nm integration time 0.1 s data calculated fraction bound (%) (y ((yo – yc)/yo) × 100 fluorescence signal supernatant experiments absence GST-Rab27a-Q78L protein beadsdata analysed SigmaPlot 12.3 Software Erkrath Equilibrium binding data fitted (2)y = (Bmax × x)/(Kd × x Bmax amplitude Kd constant x concentration GST-Rab27a-Q78L.Microscopy image analysisCells 104 immunofluorescence cultured 13 mm glass coverslips (SLS Nottingham 24 h infection adenovirus transfection plasmid DNA Twenty-four hours later cells paraformaldehyde fixed stained fluorescence light images collected Zeiss LSM710 confocal microscope 63× 1.4NA Apochromat lens Zeiss Axiovert 100 S inverted microscope 10× 40× objective lenses Axiocam MR3 CCD camera Primary secondary antibodies Supplementary Tables 7 8 F-actin detected texas-red-phalloidin (Sigma P1951 100 nM).Analysis melanosome dispersion melanosome distribution transfections triplicate Seventy two hours later contrast images three fields captured Axiovision 4.8 software Zeiss Axiovert 100 S microscope Images randomised cells melanosomes counted researcherCells pigmented melanosomes perinuclear cytoplasm<50% cytoplasmic area clustered Measurement function adenovirus FMN1 SPIRE1/2 proteins melanosome transport measurement proportion cell area pigmented melanosomes65.Bimolecular fluorescence BiFC HEK293a cells transfected with plasmids mCherry vYNE-SPIRE1/2-terminus vYFP 1–173 vYCE-Rab27a-terminus vYFP acids 155–239) Fugene 6 transfection reagent E2691) Forty-eight hours later mCherry YFP (BiFC fluorescence recorded from living cells low power fields Zeiss Axiovert 100 S microscope same exposure setting Normalised BiFC signal determined ImageJ software mCherry images converted to binary ROIs transfected cells saved regions manager vYFP ROI extracted normalised BiFC determined dividing vYFP signal mCherry signal differences level transfectionMedian BiFC signal ROI determined Graphpad Prism7 software La Jolla USA).Colocalisation cells analysed Leica AF6000LX fluorescence microscope Leica HCX PL APO 63×/1.3 GLYC objective Leica DFC350 FX digital camera (1392 × 1040 pixels 6.45 μm size Leica Wetzlar Germany). 3D stacks recorded processed Leica deconvolution software Images recorded Leica LASX software processed Adobe Photoshop Adobe Illustrator colocalisation myosin-Va Rab27a SPIRE1 proteins intracellular melanocyte membranes Rab27a SPIRE1 C-terminal fragments melanosome analysed ImageJ (V2.0.0) plug-in Coloc2. colocalisation rate indicated Pearson’s correlation coefficient) 1 perfect colocalisation 0 random −1 exclusive localisation Costes significance test performed pixels image scrambled randomly correlation measured correlation observed 95% randomised images show PCC less than original image probability correlation two colours greater than random overlap71.Cytoskeleton preparations FESEMmelan-a melan-f cells plated 7 mm glass coverslips grown 24 h.medium aspirated cells extracted 1 min 0.25% Triton X-100 5 μM phallacidin buffer 5 mM KCl 137 mM NaCl 4 mM NaHCO3 0.4 mM KH2PO4 1.1 mM Na2HPO4 2 mM MgCl2 5 mM Pipes 2 EGTA 5.5 mM glucose pH 6.1 wash samples fixed 1% glutaraldehyde 5 μM phallicidin 15 min coverslips HPLC water incubated 2% osmium tetroxide 1 hour 15 min washes dehydrated 30 50 70 90 100% ethanol solutions 20 min methanol coverslips picked immersed HMDS 30 s air-dry hour 10 mm specimen stubs coated silver paint 5–6 nm platinum Edwards S150B sputter coater Samples imaged JEOL JSM-6700F electron microscope melan-a cells permeabilised 45 s 0.25% Triton rinsed incubated 0.5% glutaraldehyde 10 mincoverslips rinsed transferred parafilm incubated 0.5 μM biotylinated phalloidin PBS 1% BSA 0.1% fish gelatin 1 h 50 mM NH4Cl PBS (15 min fresh solution 7 0.1% fish gelatin PBS (30 min incubated 3 h humidifying chamber 10 nm gold-coupled anti antibodies 0.1% fish gelatin PBS 0.01% BSA-c washed three times 10 min gelatin (0.1% 1 min 0.05 % Triton PBS four 1 min washes PBS post-fixed 1% glutaraldehyde/5 μM phallacidin CBS 15 min transferred distilled water processed osmium/HMDS Samples imaged by secondary electron detection SEM backscatter gold particle labelling combined image.FESEM colorisation filament measurementsFilaments highlighted Brush Photoshop 20–50% opacity pseudo-colours applied Hue/Saturation Filament diameter measured line profiling FIJI/ImageJ four FESEM images melan-a (25 40 Filaments melanosomes measured Statistical significance determined by Mann–Whitney test data distribution normality D’Agostino Pearson normality test GraphPadCytoskeleton freeze melan-a cells grown glass coverslips 24 h cytoskeletons extracted 0.25% Triton CBS 45 s rinsed incubated 0.2 mg/ml myosin S1 Denver PEM buffer (100 mM PIPES-KOH pH 6.9 1 mM MgCl2 1 mM EGTA 20 min fixed 1% glutaraldehyde CBS 15 min washed 5 min distilled water plunge-frozen nitrogen-cooled 1:4 isopentane/propane samples transferred freeze fracture unit rotary shadowed 45 degree 1.2–1.3 nm tantalum-tungsten 1.5 nm platinum 2.5 5 nm carbon 90 degree angle replicas separated 8% hydrofluoric acid washed distilled water picked formvar-coated copper grids examined JEOL 1200 EXII transmission electron microscope 80 kV Images inverted processed contrast Photoshop labelling purification Fmn2 myristate analogue cells transfected plasmids incubated NMTi IMP-1088 (100 nM) YnMyr 24 h Cells harvested washed 1× PBSPellets homogenised 120 μl lysisbuffer 1% Triton X-100 0.1% SDS EDTA-free protease inhibitor PBS (pH 7.4) Homogenates centrifuged 4 min 4 supernatant collected protein concentrations determined Bio-orthogonal CLICK-ligation 350 μg protein AzTB Aliquots 20 μg protein compare AzTB labelling abundance enrichment Neutravidin-agarose beads pull-down supernatant (20 Protein samples resolved 12% SDS–PAGE gel 90 V AzTB-clicked-labelled proteins visualised fluorescence scanning Typhoon Variable Mode Imager 9500 proteins immobilised nitrocellulose membrane wet blotting Proteins Fmn2-GFP Fmn2[G2A]-GFP detected western blotting GFP antibody HSP90 ARL1 HRP-conjugated secondary antibodies HRP Luminata ImageQuant LAS4000 Rab27aEight 14 28 × 10 cm dishes 80% B16-F1 mouse melanoma cells infected adenovirus vectors GFP cultured 10 days expression proteinsCells washed cold PBS trypsinised harvested centrifugation lyzed 500 μl buffer 20 mM Tris-HCl pH 7.5 50 mM NaCl 1 mM EGTA 1% CHAPS protease inhibitor cocktail (Roche Mannheim 1 mM GTP Lysates centrifuged 10,000 × g 15 min supernatant incubated goat anti-Rab27a antibodies (1 Sicgen Portugal 50 μl protein A/G UltraLink Resin Rockford IL overnight incubation 4 °C beads precipitated low centrifugation washed buffer analysed mass spectrometry co-precipitating proteins preparation sample washed 200 μL 50 mM triethyl ammonium bicarbonate centrifugation 30 s removal supernatant samples resuspended 100 μL 50 mM TEAB reduced dithiothreitol (5 mM DTT 56 °C alkylated iodoacetamide (55 mM IAA Tryptic digestion 1 μg trypsin 1 % protease max incubated 37 °C thermomixer 2 h After 1 μg trypsin added incubation 16 h1 μg trypsin added sample 16 h incubation 47 °C thermomixer 1 h concentrated de-salted HyperSep C18 spin tips (10–200 μL Thermo Scientific concentrated vacuum concentrator resuspension 5% acetonitrile 0.1% formic acid Product #1 analysed Sciex TripleTOF 6600 mass spectrometer nanoLC 425 system micro flow (5 μL/min phase B acetonitrile 0.1% formic acid phase A (0.1% formic acid (4 μl injected trapped YMC Triart C18 pre-column (5 mm 3 μm 300 μm ID 0.1% flow rate 10 μL/min eluted YMC Triart C18 column (15 cm 3 μm 300 Sciex TripleTOF 6600 Duospray Source 50 μm electrode mode +5500 V analysed IDA Acquisition modelinear gradient mobile phase B 3 to 30% 68 min 30 to 40% 5 min 40 to 80% column wash re-equilibration 2 min run time 87 IDA acquisition mode top30 ion fragmentation (TOFMS m/z 400–1250 product ion 100–1500 15 s exclusion rolling collision energy 250/50 ms accumulation time/product ion 1.8 s IDA data searched ProteinPilot 5.0.2 alkylation biological SwissProt mouse database 2019 human sequence.Statistics experiments repeated three times similar results data one experiment representative three Nature Research Reporting Summary.Supplementary information Review Supplementary Files Movie 1 2 3 | 49.7 | 1.902053 |
10.1038/s41467-021-22113-3 | PMC7985376 | Mitochondria-lysosome contact sites mediate cross-talk between the two organelles. Here, the authors show mitochondria-lysosome contacts are prolonged and defective in heterozygous mutant GBA1 neurons, which is caused by defective modulation of TBC1D15 due to decreased GBA1 activity. | Mitochondria-lysosome contacts are recently identified sites for mediating crosstalk between both organelles, but their role in normal and diseased human neurons remains unknown. In this study, we demonstrate that mitochondria-lysosome contacts can dynamically form in the soma, axons, and dendrites of human neurons, allowing for their bidirectional crosstalk. Parkinson’s disease patient derived neurons harboring mutant GBA1 exhibited prolonged mitochondria-lysosome contacts due to defective modulation of the untethering protein TBC1D15, which mediates Rab7 GTP hydrolysis for contact untethering. This dysregulation was due to decreased GBA1 (β-glucocerebrosidase (GCase)) lysosomal enzyme activity in patient derived neurons, and could be rescued by increasing enzyme activity with a GCase modulator. These defects resulted in disrupted mitochondrial distribution and function, and could be further rescued by TBC1D15 in Parkinson’s patient derived GBA1-linked neurons. Together, our work demonstrates a potential role of mitochondria-lysosome contacts as an upstream regulator of mitochondrial function and dynamics in midbrain dopaminergic neurons in GBA1-linked Parkinson’s disease. | IntroductionMitochondria–lysosome (M–L) contact sites were recently identified as inter-organelle membrane contacts involving the dynamic tethering of mitochondria with lysosomes1,2. Importantly, M–L contacts allow for the bidirectional regulation of both mitochondrial and lysosomal network dynamics, and mediate their direct interaction in a pathway distinct from mitophagy or lysosomal degradation of mitochondrial-derived vesicles2. M–L contact sites are further regulated by TBC1D15, a Rab7 GAP recruited to mitochondria via the outer mitochondrial membrane protein Fis13–6, which mediates Rab7 GTP hydrolysis to promote M–L contact untethering dynamics2. However, the formation and role of M–L contacts in human neurons have not been previously investigated.Parkinson’s disease (PD) is the most common neurodegenerative movement disorder characterized by progressive loss of dopaminergic neurons in the substantia nigra pars compacta (SNc), leading to the cardinal motor symptoms of PD7,8. Interestingly, both mitochondrial and lysosomal defects have been genetically and functionally linked to PD9,10. Indeed, the lysosomal enzyme GBA1 which encodes for β-glucocerebrosidase (GCase) and catalyzes the hydrolysis of glucosylceramide (GlcCer) to glucose and ceramide, represents the greatest genetic risk factor for PD, with GBA1 mutation carriers exhibiting more severe cognitive symptoms11,12. GBA1-PD patient-derived mutant neurons demonstrate both lower GCase protein levels and reduced GCase activity13–16. Moreover, wild-type GCase activity is decreased in both idiopathic and multiple types of familial PD patient neurons9,16–20, highlighting targeting of GBA1 as a potential therapeutic approach for PD14,16,21 and a key role for GCase activity in PD pathogenesis. However, whether M–L contact dynamics are disrupted in PD patient neurons and further modulated by GCase activity is not known.Here we investigated the impact of M–L contact sites in neuronal function and GBA1-linked PD pathogenesis. Using super-resolution time-lapse imaging, we found that M–L contact sites dynamically formed in the soma, axon, and dendrites of human iPSC-derived dopaminergic neurons. We further reported a role for M–L contacts in PD patient-derived dopaminergic neurons, as contact untethering was disrupted in patient neurons expressing heterozygous mutant GBA1. Finally, we showed that defective M–L contact untethering was caused by decreased GCase lysosomal enzymatic activity leading to GlcCer accumulation. This resulted in decreased levels of the untethering protein TBC1D15 which normally mediates Rab7 GTP hydrolysis for contact untethering. Ultimately, these defects resulted in disrupted mitochondrial distribution and function, but could be rescued by TBC1D15 in Parkinson’s patient GBA1-linked neurons. Together, our work identifies an important role for M–L contact sites in human neurons and in GBA1-linked PD pathophysiology.ResultsM–L contact sites dynamically form in human neuronsTo investigate M–L contact sites in human neurons, we utilized induced pluripotent stem cell (iPSC) technology (Supplementary Fig. 1a) to generate human midbrain dopaminergic neurons from healthy controls using previously established protocols16,22. Differentiated neurons expressed the neural specific marker β-III-tubulin (TUJ1) and midbrain dopaminergic neuron-specific markers: tyrosine hydroxylase (TH), forkhead box protein A2 (FOXA2), and LIM Homeobox Transcription Factor 1 Alpha (LMX1A) (Supplementary Fig. 2a, b), and were subsequently imaged for mitochondrial and lysosomal dynamics in live neurons (Supplementary Fig. 3a, b).Using 3D super-resolution structured illumination microscopy (3D N-SIM), we found that mitochondria and lysosomes (Mito-RFP, Lyso-GFP) formed stable inter-organelle contacts in human neurons (Fig. 1a). Neuronal M–L contacts were further confirmed to be <10 nm apart using electron microscopy (EM) (Fig. 1b), consistent with other contact sites2,23. Next, we conducted confocal time-lapse microscopy of live neurons, and found that M–L contacts dynamically formed over time and remained tethered together (yellow arrows) before subsequently untethering from one another (white arrows) (Fig. 1c and Supplemental Movies 1–3). Mitochondria in contact with lysosomes maintained their membrane potential, as visualized by TMRM imaging (Supplementary Fig. 3c–e), and dynamic M–L contact tethering could be further observed in human neurons by imaging outer mitochondrial membrane and lysosomal membrane proteins (TOM20-RFP, LAMP1-GFP) (Supplementary Fig. 3f), as well as by imaging mitochondrial and lysosomal-targeted dyes (MitoTracker, LysoTracker) (Supplementary Fig. 3g). In addition, we could also visualize M–L contacts in human neurons by proximity ligation assay (PLA) imaging of TOM20 and LAMP1 on the outer mitochondrial and lysosomal membrane (Supplementary Fig. 4a, b). M–L contacts in human neurons remained tethered for an average duration of 88.07 ± 5.05 s (Fig. 1d, e), with ~18% of lysosomes in contact with mitochondria at any point in time (Fig. 1f). Thus, M–L contacts can dynamically form to mediate crosstalk between mitochondria and lysosomes in human neurons.Fig. 1Mitochondria–lysosome contacts dynamically form in human neurons.a Representative 3D structured illumination microscopy (N-SIM) images of M–L contacts (yellow arrows) in wild-type human dopaminergic iPSC-derived neurons (mitochondria: red, Mito–RFP; lysosomes: green, Lyso-GFP) (30 neurons from N = 3 independent experiments were imaged). b Representative electron microscopy (EM) images of M–L contacts (yellow arrows) with distance between membranes <10 nm (mitochondria, M; lysosomes, L). c Representative time-lapse confocal images of dynamic contacts between mitochondria (red, Mito-RFP) and lysosomes (green, Lyso-GFP). Time-lapse recordings were taken at 2 s intervals for 3–5 min. Yellow arrows mark stable M–L contacts. White arrows mark the site of M–L contacts before contact formation or after contact untethering. Black line shows duration of contacts (see Supplementary Movies 1 and 2). d, e Quantification of duration of stable M–L contacts from confocal images (n = 156 contacts from 26 neurons, N = 3 independent experiments). d Average minimum duration of neuronal M–L contacts. e Relative frequency (percentages) distribution of neuronal M–L contacts. X-axis represents bin centers. Bin width = 60 s. f Percentage of lysosomes contacting mitochondria (for >20 s) (n = 41 neurons, N = 3 independent experiments). For all quantifications, data are means ± S.E.M. Scale bars, 500 nm (a, c); 100 nm (b).Spatial compartmentalization of neuronal M–L contact dynamicsNext, we investigated the spatial compartmentalization of neuronal M–L contacts in the soma, dendrites, and axons of human neurons, as both organelles are localized throughout multiple neuronal compartments. Dendrites and axons were identified as being positive for MAP2 or Tau, respectively (Supplementary Fig. 2c). Importantly, we found that mitochondria and lysosomes tethered at contact sites (yellow arrows) within all three neuronal compartments in the soma (Fig. 2a (left) and Supplemental Movie 4), dendrites (Fig. 2a (middle), Supplemental Movie 5), and axons (Fig. 2a (right) and Supplemental Movie 6). M–L contacts in dendrites and axons exhibited decreased mitochondrial and lysosomal motility (yellow arrows) (Fig. 2b). To further examine whether the dynamics of contacts differed by spatial compartmentalization, we quantified the duration of M–L contacts and percentage of lysosomes in contact with mitochondria across different neuronal compartments. Interestingly, live-cell imaging analysis revealed that the average minimum duration of M–L contacts was significantly increased in axons compared to those in the soma (soma: 80.6 ± 6.2 s; dendrites: 82.4 ± 9.7 s; axons: 112.5 ± 13.2 s) (Fig. 2c). In contrast, the percentage of lysosomes contacting mitochondria in the soma, dendrites, and axons did not differ (Fig. 2d). Together, these results demonstrate that M–L contacts are able to form with varying dynamics across multiple neuronal compartments in human neurons.Fig. 2Spatial compartmentalization of neuronal mitochondria–lysosome contact dynamics.a Representative time-lapse confocal images of contacts between mitochondria (red, Mito-RFP) and lysosomes (green, Lyso-GFP) in soma, dendrites, and axons of wild-type human dopaminergic iPSC-derived neurons. Time-lapse recordings were taken at 2 s intervals for 5 min. Yellow arrows mark stable M–L contacts. White arrows mark the site of M–L contacts before contact formation or after contact untethering. Black line shows duration of contacts. Scale bar = 500 nm (see Supplementary Movies 4–6). b Representative frames of live-cell imaging (left) and dual color kymographs (right) of M–L contacts in dendrites and axons (mitochondria: red, Mito–RFP; lysosomes: green, Lyso-GFP). In confocal image frames (left), contact sites are denoted by yellow arrows. In kymographs: white scale bar = 1 μm, yellow vertical bar = 30 s. Yellow arrows in kymograph point to start and end timepoints of M–L contact tethering. Left black line shows duration of contacts. c, d Quantification of M–L contacts across different neuronal compartments. One-way ANOVA followed by Tukey’s multiple comparisons test. c Comparison of average minimum durations of M–L contacts in soma, dendrites, and axons (n = 86 (soma), n = 35 (dendrite), n = 34 (axon) contacts from 26 neurons, N = 3 independent experiments), *p = 0.0335. d Percentage of lysosomes contacting mitochondria in soma, dendrites, and axons (for >20 s) (n = 14 (soma), n = 13 (dendrite), n = 14 (axon) neurons; N = 3 independent experiments). For all quantifications, data are means ± S.E.M; *p ≤ 0.05, ns: not significant.Loss of GCase activity disrupts M–L contact untethering in GBA1-PD patient dopaminergic neuronsPD has been genetically and functionally linked to both mitochondrial and lysosomal defects9,10, but whether M–L contacts are disrupted in PD has not been previously investigated. As mutations in the lysosomal enzyme GCase (GBA1) represent the greatest genetic risk factor for PD12, and GCase activity is decreased in both idiopathic and multiple types of familial PD patient neurons9,16–18, we examined M–L contact dynamics in GBA1-PD (Δhet 84GG) patient neurons.Using PD patient fibroblasts harboring mutant GBA1, we generated human iPSCs and their isogenic controls by CRISPR-Cas916 (Supplementary Fig. 1a) which were subsequently differentiated into midbrain dopaminergic neurons (mutant GBA1 (∆GBA); isogenic control (Corr)) (Supplementary Fig. 2a–c). Both mutant GBA1 and its isogenic control did not affect the efficiencies of fibroblast reprogramming to iPSCs or differentiation into midbrain dopaminergic neurons from iPSCs16.We first confirmed that PD patient-derived mutant GBA1 dopaminergic neurons exhibited decreased GCase protein levels (Fig. 3a, b). We also conducted live-cell GCase activity assays and showed that mutant GBA1 patient neurons had reduced total GCase enzymatic activity compared to isogenic control neurons (Fig. 3c–e). To further compare lysosomal and non-lysosomal GCase activity separately, we used Bafilomycin A1 (BafA1), an inhibitor of lysosomal acidification, as an established protocol for examining lysosomal GCase activity24 and also found a significant decrease in lysosomal GCase activity in patient-derived mutant GBA1 neurons compared to CRISPR-corrected isogenic control neurons (Fig. 3c, d, f), consistent with previous findings16.Fig. 3Loss of GCase activity disrupts mitochondria–lysosome contact untethering in GBA1-PD patient dopaminergic neurons.a, b Western blot analysis of PD patient-derived mutant GBA1 dopaminergic neurons (∆GBA; het 84GG) and its CRISPR-corrected isogenic control (Corr) neurons. GCase level was significantly reduced in ∆GBA neurons (N = 4 independent experiments). Paired two-sided Student’s t-test; **p = 0.0083. c–f ∆GBA and Corr neurons were treated with either DMSO or BafA1 and subjected to live-cell GCase activity analysis. e Quantification of the area under each curve (AUC) demonstrates decreased total GCase activity in ∆GBA neurons. Paired two-sided Student’s t-test; *p = 0.0183. f Lysosomal GCase activity was calculated by subtracting BafA1 values from DMSO. Values are expressed as fold-change compared to isogenic controls (N = 3 independent experiments). Paired two-sided Student’s t-test; *p = 0.0352. g Representative time-lapse confocal images of contacts between mitochondria (red, Mito-RFP) and lysosomes (green, Lyso-GFP) in Corr (left) and ∆GBA (right) human neurons. Yellow arrows mark stable M–L contacts. White arrows mark the site of M–L contacts after contact untethering. Black line shows duration of contacts. Scale bar = 500 nm (see Supplementary Movies 7 and 8). h Quantification of average minimum duration (left) and relative frequency distribution of the duration of stable M–L contacts (right), showing increased duration of stable M–L contacts in ∆GBA neurons (n = 48 contacts from Corr and n = 71 contacts from ∆GBA, N = 3 independent experiments). X-axis of the histogram represents bin centers. Bin width = 60 s. Unpaired two-sided Student’s t-test; *p = 0.0204. i Percentage of lysosomes contacting mitochondria (for >20 s) (n = 12 Corr and n = 36 ∆GBA neurons, N = 3 independent experiments). Unpaired two-sided Student’s t-test. For all quantifications, data are means ± S.E.M.; *p ≤ 0.05, **p ≤ 0.01, ns: not significant.We next investigated whether lysosomal contacts with mitochondria were disrupted by loss of lysosomal GCase activity by conducting confocal live-cell microscopy of mitochondria and lysosomes in mutant GBA1 and CRISPR-corrected isogenic control neurons. Interestingly, while M–L contacts dynamically formed in both conditions (yellow arrows) (Fig. 3g and Supplemental Movies 7, 8), the average duration of M–L contact tethering was significantly increased in mutant GBA1 neurons, indicative of inefficient untethering events (Fig. 3h). However, the percentage of lysosomes in contacts was similar between conditions (Fig. 3i), suggesting that the subsequent untethering but not initial formation of M–L contact tethering was disrupted in mutant GBA1 neurons.To further examine M–L contacts, we conducted PLA imaging of M–L contacts in mutant GBA1 and CRISPR-corrected isogenic control neurons (Supplementary Fig. 4c), and similarly found that M–L contacts could still form in both conditions (Supplementary Fig. 4d). In addition, we conducted EM imaging of M–L contacts in mutant GBA1 and CRISPR-corrected isogenic control neurons (Supplementary Fig. 4e), and also found that M–L contacts tethered together in both conditions, with the length of membrane contact between mitochondria and lysosomes not altered in mutant GBA1 neurons (Supplementary Fig. 4f). Of note, ER–mitochondria contact and ER–lysosome contact formation were also not disrupted in mutant GBA1 neurons (Supplementary Fig. 5a–f). Together, our results suggest that loss of GCase activity does not disrupt M–L contact formation, but preferentially disrupts the untethering of M–L contact sites, resulting in prolonged contact site tethering between mitochondria and lysosomes.We further examined whether these changes in M–L contact dynamics might be specific to loss of GCase activity rather than general lysosomal enzyme dysfunction. Mutant GBA1 neurons had altered lysosomal morphology, including enlarged lysosomes (Fig. 3g and Supplementary Fig. 3h, i). However, inhibition of other lysosomal enzymes in human neurons which also led to enlarged lysosomal morphology (Supplementary Fig. 6a) including inhibition of lysosomal acid ceramidase (carmofur treatment), cysteine proteases (E64D treatment), or aspartyl proteases (pepstatin A treatment) did not disrupt M–L contact dynamics (Supplementary Fig. 6b). Thus, our results suggest that this pathway is selectively disrupted by loss of GCase activity rather than general lysosomal defects or enzyme dysfunction.In addition, we assessed whether rescuing GCase activity in mutant GBA1 neurons was sufficient to restore defective M–L contact dynamics. The modulator S-181 was recently found to increase GCase activity in mutant GBA1 neurons16. Importantly, S-181 treatment (Supplementary Fig. 6c) rescued the prolonged M–L contact tethering in mutant GBA1 neurons (Supplementary Fig. 6d), further highlighting the role of GCase activity in regulating M–L contact dynamics.Tethering proteins mediating M–L contact are disrupted in GBA1-PD patient dopaminergic neuronsWe subsequently examined the potential mechanism through which mutant GBA1 might disrupt M–L contact dynamics. We recently showed that M–L contact untethering is mechanistically regulated by lysosomal RAB7-GTP hydrolysis from GTP to GDP-bound Rab72, and is driven by the GAP (GTPase-activating protein) activity of mitochondrial TBC1D15 (Rab7 GAP)3,5, which is recruited to the outer mitochondrial membrane by Fis14,6. Thus, we investigated whether RAB7-GTP hydrolysis might be disrupted by mutant GBA1 in patient neurons. We first measured the total protein levels of RAB7 together with Fis1 and TBC1D15 in PD patient-derived mutant GBA1 dopaminergic neurons compared to isogenic controls (Fig. 4a–d). We observed no differences in total Rab7 levels (Fig. 4c) or Fis1 levels (Fig. 4d), and Fis1 levels were not altered even after normalizing for mitochondrial levels (Fis1/Tom20: Corr vs GBA1, p = 0.23 (not significant)) (Supplementary Fig. 3j). Of note, TBC1D15 and Fis1 localization to mitochondria (Supplementary Fig. 7a, b), and Rab7 localization to lysosomes (Supplementary Fig. 7c), as well as the levels of the Rab7 GEF (guanine nucleotide exchange factor) complex proteins Mon1 and Ccz1 (Supplementary Fig. 7d) were also not altered in mutant GBA1 neurons. However, we surprisingly found that TBC1C15 levels were significantly decreased in mutant GBA1 patient-derived neurons (*p ≤ 0.05) (Fig. 4b).Fig. 4Tethering proteins mediating mitochondria–lysosome contact are disrupted in GBA1-PD patient dopaminergic neurons.a–d Western blot analysis of b TBC1D15, c Rab7, and d Fis1 levels in PD patient-derived mutant GBA1 dopaminergic neurons (∆GBA) and its CRISPR-corrected isogenic control (Corr) neurons. Protein levels were normalized to loading control GAPDH. Values are expressed as fold-change compared to Corr (N = 3 independent experiments). Paired two-sided Student’s t-test; *p = 0.0442. e, f GST-RILP pull-down to measure GTP-bound Rab7 levels in ∆GBA and Corr neurons. f Rab7-GTP levels were normalized to total Rab7 normalized to GAPDH. Values are expressed as fold-change compared to Corr (N = 3 independent experiments). Paired two-sided Student’s t-test; *p = 0.0478. For all quantifications, data are means ± S.E.M., *p ≤ 0.05.Based on these findings, we hypothesized that disrupted M–L contact untethering in mutant GBA1 neurons might be due to defective Rab7-GTP hydrolysis as a consequence of reduced TBC1D15 levels (Rab7 GAP). To test this, we performed GST-RILP (Glutathione Transferase-Rab Interacting Lysosomal Protein) pull-down assays25 to measure GTP-bound Rab7 levels, as RILP preferentially binds to GTP-bound Rab726,27 (Fig. 4e). Importantly, we found that PD patient-derived mutant GBA1 dopaminergic neurons demonstrated significantly increased levels of RAB7-GTP/total Rab7 compared to CRISPR-corrected isogenic control neurons (*p ≤ 0.05) (Fig. 4e, f). Together, these results suggest that decreased TBC1D15 levels in mutant GBA1 neurons disrupt Rab7 GTP hydrolysis, resulting in increased GTP-bound Rab7 and prolonged M–L contact tethering dynamics.To further investigate the downregulation of TBC1D15 in mutant GBA1 neurons, we analyzed TBC1D15 by qPCR and showed that decreased TBC1D15 protein levels in mutant GBA1 neurons were not due to lower TBC1D15 transcripts levels (Supplementary Fig. 8a). Rather, decreased TBC1D15 protein levels resulted from elevated proteasomal degradation of TBC1D15, which could be inhibited by lactacystin, leading to similar TBC1D15 protein levels between mutant GBA1 and isogenic control neurons (Supplementary Fig. 8b). We next examined whether mitochondrial dysfunction was sufficient to disrupt TBC1D15 expression levels. Treatment of either the mitochondrial uncoupler carbonyl cyanide p-trifluoromethoxyphenylhydrazone (CCCP), Complex I inhibitor rotenone, Complex III inhibitor Antimycin, or the ATP synthase inhibitor Oligomycin (Supplementary Fig. 8c–f) did not disrupt TBC1D15 expression levels in HeLa cells. In contrast, increasing GCase activity in mutant GBA1 neurons using the modulator S-18116 was sufficient to rescue the decreased levels of TBC1D15 (Supplementary Fig. 6e, f). Thus, these findings suggest that TBC1D15 levels are downregulated at the protein level by proteosomal degradation in mutant GBA1 neurons, due to the loss of GCase activity rather than general mitochondrial dysfunction.Finally, we examined if loss of TBC1D15 was able to disrupt M–L contact dynamics in human neurons. Knockdown of TBC1D15 in wild-type iPSC-derived neurons (Supplementary Fig. 9a) increased M–L contact tethering duration (Supplementary Fig. 9b), consistent with findings in non-neuronal cells2 and further supporting our results that decreased TBC1D15 levels in PD GBA1-linked patient neurons disrupt M–L contact dynamics.Inhibition of GCase activity disrupts M–L contact untetheringTo further examine whether M–L contact untethering defects were indeed dependent on loss of GCase activity, we treated wild-type human midbrain dopaminergic neurons expressing wild-type GCase with the GCase inhibitor conduritol-b-epoxide (CBE) (50 μM; 7 days) which decreases neuronal GCase activity and promotes GlcCer accumulation14,16,28. Using confocal live-cell time-lapse imaging of mitochondria and lysosomes in both wild-type (Ctrl) and CBE-treated (+CBE) human neurons, we found that M–L contacts dynamically formed in both conditions (yellow arrows) (Fig. 5a). In addition, CBE treatment was effective in increasing lysosomal size as has been previously reported29 (Fig. 5b, c). Importantly, we found that inhibition of GCase activity by CBE treatment resulted in significantly prolonged M–L contact tethering duration (***p ≤ 0.001) (Fig. 5a, d), consistent with what we observed in mutant GBA1 patient-derived neurons, and further confirming the effect of decreased GCase activity on disrupting M–L contact untethering.Fig. 5Inhibition of GCase activity disrupts mitochondria–lysosome contact untethering in dopaminergic neurons.a Healthy control WT human iPSC-derived dopaminergic neurons were treated with vehicle (Ctrl) or CBE (50 μM; 7 days) to inhibit GCase activity. Representative time-lapse confocal images of contacts between mitochondria (red, Mito-RFP) and lysosomes (green, Lyso-GFP) in Ctrl (left) and CBE-treated (right) human neurons. Yellow arrows mark stable M–L contacts. White arrows mark the site of M–L contacts after contact untethering. Black line shows duration of contacts. Scale bar = 500 nm. b Representative confocal images of lysosomes (LAMP1–GFP) in the Ctrl (left) and CBE-treated (right) human neurons. Scale bar = 1 μm. c The effects of CBE treatment on lysosomal morphology were confirmed by quantification of the percentage of lysosomes that were enlarged (diameter >0.5 μm) in M–L contacts (n = 12 neurons per condition, N = 3 independent experiments). Unpaired two-sided Student’s t-test; ***p = 0.0001. d CBE treatment in control neurons increased the average minimum duration of stable M–L contacts (n = 60 contacts per condition, N = 3 independent experiments). Unpaired two-sided Student’s t-test; ***p = 0.0001. e Representative confocal images of immunocytochemistry of GlcCer in the Ctrl (left) and CBE-treated (right) human neurons. Scale bar = 5 μm. f Quantification of CBE treatment leading to increased GlcCer levels as measured by immunofluorescence signal of GlcCer (n = 21 neurons per condition, N = 3 independent experiments). g Exogenous GlcCer treatment in control neurons increased the average minimum duration of stable M–L contacts (n = 93 contacts from Ctrl, n = 94 contacts from +GlcCer neurons, N = 3 independent experiments). f, g Unpaired two-sided Student’s t-test; ***p < 0.0001. h–k Western blot analysis of i TBC1D15, j Rab7, and k Fis1 levels in Ctrl and CBE-treated neurons. Protein levels were normalized to loading control GAPDH. Values are expressed as fold-change compared to Ctrl (N = 5 independent experiments). Paired two-sided Student’s t-test; *p = 0.0104. l, m GST-RILP pull-down to measure GTP-bound Rab7 levels in Ctrl and CBE-treated neurons. m Rab7-GTP levels were normalized to total Rab7 normalized to GAPDH. Values are expressed as fold-change compared to Ctrl (N = 3 independent experiments). Paired two-sided Student’s t-test; *p = 0.0242. For all quantifications, data are means ± S.E.M., *p ≤ 0.05, ***p ≤ 0.001.We additionally validated these findings in two other wild-type iPSC-derived neuronal lines which we characterized for iPSC and midbrain dopaminergic neuron-specific markers (Supplementary Fig. 10a–c). Inhibition of GCase activity with CBE treatment in both lines also led to significantly prolonged M–L contact tethering duration (Supplementary Fig. 10d, e), further supporting our results that inhibition of GCase activity disrupts M–L contact dynamics in human neurons.As loss of GCase activity results in significantly increased GlcCer levels (Fig. 5e, f), we then asked whether treatment of exogenous GlcCer in wild-type iPSC-derived neurons could be sufficient to disrupt M–L contact dynamics. Using live-cell microscopy of mitochondria and lysosomes in wild-type iPSC-derived neurons treated with exogenous GlcCer, we found that this also led to significantly prolonged contact site tethering duration (***p < 0.001) (Fig. 5g), highlighting a role for increased GlcCer levels in mutant GBA1 neurons in dysregulating M–L contacts.Next, we investigated whether inhibition of GCase activity also disrupted M–L contact untethering machinery. We measured the total protein levels of Rab7 together with TBC1D15 and Fis1 in wild-type (Ctrl) and CBE-treated (+CBE) humans neurons (Fig. 5h–k). Indeed, CBE-treated neurons also showed significantly decreased TBC1C15 levels (*p ≤ 0.05) (Fig. 5h, i), without changes in total Rab7 or Fis1 levels (Fig. 5j, k), consistent with mutant GBA1 patient-derived neurons. To further support these findings, wild-type iPSC-derived neurons treated with exogenous GlcCer also showed reduced TBC1D15 levels (Supplementary Fig. 10f, g).We then examined whether inhibition of GCase activity by CBE treatment led to similar defects in Rab7 GTP hydrolysis, resulting in elevated GTP-bound Rab7 levels. Using the GST-RILP pull-down assay in CBE-treated neurons (Fig. 5l), we further found that GCase inhibition resulted in increased levels of RAB7-GTP/total Rab7 (Fig. 5l, m), consistent with what we observed in mutant GBA1 neurons. In summary, our results from both mutant GBA1 and CBE-treated neurons support the hypothesis that loss of GCase activity, which increases GlcCer levels, leads to disruption of TBC1D15 levels and Rab7-GTP hydrolysis in neurons, resulting in the misregulation of M–L contact untethering dynamics.Mitochondrial dysfunction due to prolonged M–L contacts is partially rescued by TBC1D15 expressionM–L contacts mediate the bidirectional regulation of both mitochondrial and lysosomal network dynamics, and importantly are able to directly regulate both mitochondrial dynamics and motility2,30–32. We thus investigated whether mitochondrial dynamics might be disrupted in PD patient-derived mutant GBA1 dopaminergic neurons, consistent with defective M–L contact untethering, by analyzing the distribution of mitochondria in the soma and axons. Healthy mitochondria with intact mitochondrial membrane potential were imaged by live-cell imaging in patient-derived mutant GBA1 (∆GBA) and CRISPR-corrected isogenic control (Corr) neurons (Supplementary Fig. 3c). Interestingly, while there was no difference in mitochondrial density in the soma (Fig. 6a, b and Supplementary Fig. 3k), we found that axonal mitochondrial density was significantly decreased in mutant GBA1 neurons compared to axons of isogenic control neurons (Fig. 6a, c), suggesting that mutant GBA1 preferentially disrupts axonal mitochondrial distribution in Parkinson’s patient neurons. In addition, we also observed defective mitochondrial respiration as measured by decreased oxygen consumption rate (OCR) in mutant GBA1 neurons (Fig. 6d), as well as decreased AMPK activation (Fig. 6e), compared to isogenic control neurons. We also measured cellular ATP level and found significantly decreased ATP levels in mutant GBA1 neurons compared to isogenic controls (Fig. 6f), even after normalization for mitochondrial mass (***p < 0.001; Corr vs GBA [ATP intensity/TOM20 levels]). Consistent with our findings that TBC1D15 levels were decreased in mutant GBA1 neurons, we also observed mitochondrial dysfunction in TBC1D15 knockdown neurons (Supplementary Fig. 9c).Fig. 6Mitochondrial dysfunction due to prolonged mitochondria–lysosome contacts is partially rescued by expression of TBC1D15 in GBA1-PD patient dopaminergic neurons.a Live-cell distribution of mitochondria (TMRM, red) in the soma and axons from PD patient-derived mutant GBA1 dopaminergic neurons (∆GBA) and its CRISPR-corrected isogenic control (Corr) neurons. White arrows mark mitochondria in a single axon. Scale bar, 10 μm. b Quantified mitochondrial density in the soma (TMRM-positive pixels/pixels of soma). Corr (n = 47 neurons), ∆GBA (n = 47 neurons) (N = 3 independent experiments). c Quantified mitochondrial density in axons (mitochondria count/length of axon (μm)). Corr (n = 122 axons), ∆GBA (n = 114 axons) (N = 3 independent experiments). b, c Unpaired two-sided Student’s t-test; ***p < 0.0001. d Oxygen consumption rate (OCR) was measured and normalized to total protein content. Corr (n = 9 samples), ∆GBA (n = 9 samples) (N = 3 independent experiments). e Western blot analysis of phospho-AMPKα and AMPKα levels in Corr and ∆GBA neurons. Ratio of p-AMPKα/AMPKα are expressed as fold-change compared to Corr (N = 3 independent experiments). Paired two-sided Student’s t-test; *p = 0.0354. f Total cellular ATP content was measured and normalized to total protein content (ng/μl). Corr (n = 40 samples), ∆GBA (n = 40 samples) (N = 3 independent experiments). Unpaired two-sided Student’s t-test; ***p < 0.0001. g Live-cell distribution of mitochondria (TMRM, red) in the soma and axons from Corr and ∆GBA human neurons with or without human TBC1D15 lentiviral expression. White arrows mark mitochondria in a single axon. Scale bar, 10 μm. h Quantification of average minimum duration of M–L contacts in Corr-vehicle (n = 62 contacts from 18 neurons), ∆GBA-vehicle (n = 60 contacts from 15 neurons), ∆GBA-TBC1D15 (n = 63 contacts from 19 neurons) (N = 3 independent experiments); ***p < 0.0001. i Quantified mitochondrial density in the soma (TMRM-positive pixels/pixels of soma). Corr-vehicle (n = 32 neurons), ∆GBA-vehicle (n = 31 neurons), ∆GBA-TBC1D15 (n = 38 neurons) (N = 3 independent experiments). j Quantified mitochondrial density in axons (mitochondria count/length of axon (μm)). Corr-vehicle (n = 124 axons), ∆GBA-vehicle (n = 127 axons), ∆GBA-TBC1D15 (n = 127 axons) (N = 3 independent experiments), ***p < 0.0001. k Total cellular ATP concentration was measured and normalized to total protein content (ng/μl). Corr-vehicle (n = 19 samples), ∆GBA-vehicle (n = 19 samples), ∆GBA-TBC1D15 (n = 19 samples) (N = 3 independent experiments), ***p = 0.0002, *p = 0.0493. h–k One-way ANOVA followed by Tukey’s multiple comparisons test. For all quantifications, data are the means ± S.E.M.; *p ≤ 0.05, ***p ≤ 0.001, ns: not significant.Based on these findings, we hypothesized that dysregulated M–L contact untethering dynamics in mutant GBA1 neurons might contribute to these defects in mitochondrial distribution and function. Thus, to test if these defects could be rescued by promoting M–L contact untethering in mutant GBA1 neurons, we expressed human TBC1D15 in mutant GBA1 neurons by lentiviral transduction (Fig. 6g and Supplementary Fig. 11a). The expression of exogenous TBC1D15 in neurons was confirmed by western blot analysis and fluorescence imaging (Supplementary Fig. 11b, c). Importantly, consistent with our previous findings that TBC1D15 promotes M–L contact untethering2, transduction of wild-type TBC1D15 in mutant GBA1 neurons (∆GBA+TBC1D15) was able to promote M–L contact untethering, resulting in significantly decreased M–L contact durations compared to lentiviral vehicle-treated neurons (∆GBA+veh) (***p ≤ 0.001) (Fig. 6h).We further investigated the effect of rescuing M–L contact dynamics on axonal mitochondrial density in mutant GBA1 neurons. Interestingly, while TBC1D15 expression did not alter mitochondrial densities in the soma of mutant GBA1 neurons (Fig. 6g, i), it significantly rescued the decreased mitochondrial density in axons of mutant GBA1 neurons (Fig. 6g, j). Moreover, TBC1D15 expression was also able to partially rescue ATP levels in mutant GBA1 neurons (Fig. 6k). Together, our results suggest that upregulation of TBC1D15 in PD patient mutant GBA1 neurons is sufficient to rescue the misregulation of M–L contact untethering, as well as downstream defects in mitochondrial dynamics and function.Finally, we extended our analysis to additional PD mutant GBA1 patient neurons (Δhet N370S) and CRISPR-edited isogenic control neurons (Supplementary Fig. 12a–c). Mutant GBA1 (N370S) patient neurons also showed decreased GCase levels and enzymatic activity (Supplementary Fig. 12d–g). Importantly, they also had significantly increased GTP-bound Rab7 due to lower TBC1D15 levels (Supplementary Fig. 12h–m), which resulted in prolonged M–L contact tethering (Supplementary Fig. 12n). Moreover, S-181 modulator treatment in mutant GBA1 (N370S) neurons was sufficient to rescue both M–L contact dynamics (Supplementary Fig. 12n) as well as TBC1D15 levels (Supplementary Fig. 12o). In addition, mutant GBA1 (N370S) neurons also demonstrated reduced mitochondrial function (Supplementary Fig. 12p). Thus, these findings further highlight the role of defective GCase activity in PD GBA1-linked patient neurons on disrupting Rab7 GTP hydrolysis machinery and M–L contact site dynamics.DiscussionOur study demonstrates that M–L contact sites dynamically form in human neurons, and further investigates their role in neurons from patients with GBA1-linked PD. We found that loss of lysosomal GCase enzymatic activity in PD patient-derived dopaminergic neurons led to prolonged M–L contact tethering dynamics due to defective contact untethering machinery, and resulted in misregulated axonal distribution of mitochondria and decreased ATP levels. Importantly, we showed that rescuing M–L contact site dynamics in PD patient neurons is sufficient to ameliorate defects in mitochondrial distribution and function, thus highlighting a potential role for M–L contact site dysregulation in PD pathogenesis.Multiple genes linked to mitochondria or lysosomes have been identified as causative or risk genes of PD33,34. Moreover, both mitochondrial and lysosomal dysfunction have been implicated in PD9,10,15,35–40, suggesting a functional crosstalk between these two organelles. We previously found that oxidized dopamine mediates convergence of mitochondrial and lysosomal dysfunction in human but not in mouse dopaminergic neurons9, highlighting the importance of patient-derived dopaminergic neurons for studies of PD pathogenesis.Despite the previously studied role of mitochondria and lysosomes in PD pathogenesis, a direct homeostatic relationship between these two organelles that is independent of eventual lysosomal degradation of mitochondria has not been examined. In this context, our work provides evidence for the role of M–L contacts not only in the homeostasis of dopaminergic neurons but also as a link between mitochondrial and lysosomal dysfunction in PD pathogenesis. We specifically focused on GBA1 mutations which represent the greatest genetic risk factor for PD11,12. Importantly, wild-type GCase enzyme activity is also reduced in patient neurons with genetic or idiopathic PD who do not harbor GBA1 mutations9,16,18, suggesting that loss of GCase activity is an important contributor to PD pathogenesis. Previously, we reported that the loss of GCase function in patient neurons compromises lysosomal protein degradation15 which contributes to other key PD pathogenic phenotypes including α-synuclein accumulation13,15,24,41. In addition to its primary lysosomal dysfunction, GBA1 mutations and abnormal GCase activity have also been linked to mitochondrial dysfunction9,28,42–46.In this study, we show that GBA1 mutant PD patient neurons have defective M–L contact dynamics, resulting in prolonged contact tethering. While previous studies did not examine the GTP-bound state of Rab747,48, we found that loss of GCase activity resulted in an increased percentage of active GTP-bound lysosomal Rab7, which directly mediates M–L contact tethering2. This is likely due to decreased levels of TBC1D15 (Rab7 GAP) in mutant GBA1 patient neurons, as TBC1D15’s GAP activity promotes Rab7 GTP hydrolysis and subsequent M–L contact untethering2. Indeed, we found that increasing TBC1D15 expression was sufficient to rescue the prolonged M–L contact tethering we observed in mutant GBA1 neurons. In addition, we further observed mitochondrial dysfunction in patient neurons, including abnormal mitochondrial distribution in axons, decreased mitochondrial respiration and lower ATP levels. Of note, TBC1D15 expression was also able to rescue mitochondrial dysfunction. Thus, dysregulation of M–L contacts may play an important role in GBA1-linked PD pathogenesis, and targeting contact machinery may help ameliorate downstream mitochondrial dysfunction. Given that mitochondria play key roles as energy suppliers especially in the synapses of active neurons, we hypothesize that such abnormal distribution of mitochondria may further contribute to synaptic dysfunction in PD10.While inter-organelle contact sites have been found to be essential subdomains for modulating cellular function and homeostasis49–51, only recently have studies reported the function and molecular architecture of inter-organelle contacts in neurons, such as those between the mitochondria and endoplasmic reticulum40,52,53. In addition, dysfunction of inter-organelle contacts in disease have been shown to be key contributors in the development of various diseases1,30,31,40,54–60. Moreover, the recent identification of M–L contact sites has shed new light on the direct relationship between mitochondria and lysosomes in a pathway independent of lysosomal degradation of mitochondria2,31,61–64, allowing for direct crosstalk and regulation of both organelles in a dynamic manner1. In our previous study, we showed the formation of M–L contacts in non-neuronal cells and identified protein mediators responsible for contact untethering1,2. Here, we demonstrate that M–L contacts are also key contributors in human neurons and that contacts dynamically form in multiple neuronal compartments, suggesting that they act as important sites for the neuronal regulation of mitochondrial and lysosomal dynamics. Together, our findings not only provide insights into inter-organelle contacts in maintaining the cellular homeostasis of human neurons, but also suggest the importance of M–L contacts as a potential target for therapeutic development in PD.MethodsHuman iPSC culture and characterizationDetailed procedures for iPSC culture and neuronal differentiation have been described previously16. Healthy control and PD patient skin fibroblasts16 (GBA1 heterozygous 84GG mutation (c.84dupG frameshift mutation) which prevents the expression of the mutant allele, resulting in reduced GCase levels arising from a single wild-type copy of GBA1; and GBA1 heterozygous p.N370S mutation) were reprogrammed into iPSCs through Northwestern University Stem Cell Core Facility, using Sendai virus (SeV)-based delivery of four Yamanaka factors (Oct, Sox2, Klf4, and c-Myc). We also generated an isogenic control iPSC line by correcting the mutation using CRISPR/Cas9 protocols65 as recently described16. Guide RNAs targeting the mutation were cloned into vector PX461 (Addgene #48140) carrying the cDNA encoding for GFP-tagged Cas9 nuclease. The plasmid was electroporated into GBA1 84GG mutant iPSCs together with ssODN carrying the corrected sequence. After 48 h, generation of an isogenic control line was confirmed by FACS-sorting and sequencing. All iPSCs were maintained either in mTeSR™1 or mTeSR™ Plus media (Stemcell Technologies, #85850, #05825) and cells were passaged as small chunks every 6–8 days depending on confluence. All iPSC lines have been routinely characterized for expression of pluripotency markers. Cells were plated on PDL-coated coverslips and subjected to immunofluorescence staining for NANOG, OCT4, SSEA4, and TRA1-81. Genomic integrity was confirmed as described previously16. Mycoplasma tests were performed on a monthly basis to maintain qualified iPSC lines.Generation and characterization of hiPSC-derived dopaminergic neuronsHuman dopaminergic iPSC-derived neurons were differentiated using previously established protocol22 for analysis of M–L contacts. Briefly, after plating single iPSCs, cells were treated with factors according to the original protocol. At day 13 of differentiation, cells were passaged en bloc (size of 1–2 mm) onto 10-cm culture dishes pre-coated with poly-d-lysine (PDL) (Sigma, #P1149) / laminin (Invitrogen, #23017-015). At day 25, neurons were treated with accutase and passaged onto PDL/laminin-coated culture dishes, and subjected to dopaminergic marker characterization through immunocytochemistry (ICC) using TH, FOXA2, LMX1A together with neural specific marker β-III-tubulin (TUJ1) at day 30. After day 50, neurons were considered mature and maintained in Neurobasal media (Life Technologies, # 21103049) containing Neurocult SM1 (Stemcell Technologies, #5711). Neurons at day 50–60 were used for immunostaining, EM, live-cell time-lapse confocal imaging, live-cell or fixed-cell super-resolution SIM. Neurons from N ≥ 3 independent experiments (biological replicates; batches of neuron differentiation) per condition were used for each experiment.ImmunocytochemistryCells were immunostained for multiple antibodies depending on the purpose of each experiment. Neurons were plated on PDL/laminin-coated coverslips and fixed in 4% paraformaldehyde in PBS for 20 min and permeablized with 10% FBS and 0.1% saponin in PBS at room temperature. For GlcCer staining, 10% FBS was substituted with 2% gelatin (Sigma, #G1393). Cells were then immuno-labeled with the following primary antibodies: Oct4 (Abcam, #19857, 1:100); SSEA1 (Millipore, MAB#4304, 1:100); Nanog (R&D systems, #AF1997, 1:50); Tra-1-81 (Millipore, #MAB4381, 1:100); TH (Calbiochem, #657012, 1:500); β-III-tubulin (Biolegend, #801202, #802001, 1:1000); Lmx1a (Milipore, MAB#10533, 1:500); FoxA2 (Santa Cruz, #sc-101060, 1:100); Tom20 (Abcam, #78547, 1:100); Lamp1 (Santa Cruz, #sc-20011, 1:100); Map2 (Novus Biological, #NB300213, 1:3000); Tau (DAKO, #A002401-2, 1:300); GlcCer (Glycobiotech, #RAS_0011, 1:100). After overnight incubation at 4 °C, coverslips were washed three times with PBS for 5 min each, incubated in Alexa-conjugated secondary antibodies (Invitrogen, #A21206, #A21202, #A11029, #A11011, #A10042, #A10037, #A21449, 1:1000) for 1 h at room temperature, washed three times and mounted onto Superfrost Plus microscope slides (Fisherbrand, #12-550-15) with VECTASHIELD HardSet Antifade Mounting Medium (Vector Labs, #H-1400). Alternatively, ProLong™ Diamond Antifade Mountant (Thermo Fisher Scientific, #P36959) was used for SIM sample preparation. Images were obtained on either a Leica DMI4000B confocal microscope using Leica Application Suite X (Leica) or a Nikon A1R laser scanning confocal microscope with GaAsp detectors using NIS-Elements (Nikon). Colocalization was measured using the EzColocalization plugin in ImageJ (National Institutes of Health (NIH))66.Proximity ligation assayProximity between outer mitochondrial membrane protein Tom20 and lysosomal membrane protein Lamp1 was detected using Duolink™ PLA kit (Sigma Aldrich, #DUO92101) according to the manufacturer’s protocol. Neurons were plated on PDL/laminin-coated coverslips and cells were immuno-labeled with Tom20 (Abcam, #78547, 1:100) and Lamp1 (Santa Cruz, #sc-20011, 1:100). Images of red PLA signals were collected using Nikon A1R laser scanning confocal microscope with GaAsp detectors.Electron microscopyFor EM, neurons were grown on PDL/laminin-coated glass coverslips. Neurons were fixed with 2.5% glutaraldehyde and 2% paraformaldehyde in 0.1 M cacodylate buffer, pH 7.3 for 30 min at room temperature, and then submitted to the Northwestern EM core facility for subsequent processing. After post-fixation in 1% osmium tetroxide and 3% uranyl acetate in PBS, cells were dehydrated in an ethanol series, embedded in Epon resin and polymerized for 48 h at 60 °C. Ultrathin sections were made using a UCT ultramicrotome (Leica Microsystems) and contrasted with 4% uranyl acetate and Reynolds’s lead citrate. Samples were imaged on a FEI Tecnai Spirit G2 transmission electron microscope (FEI) operated at 80 kV. Images were captured with an Eagle 4k HR 200kV CCD camera and analyzed using ImageJ (NIH).Live-cell time-lapse confocal microscopyLive-cell time-lapse confocal imaging was conducted of mitochondria and lysosomes labeled with the following reagents: Lentivirus expressing Tom20-RFP or Lamp1-GFP (MOI = 3, 3 days); CellLight® BacMam 2.0 baculovirus (PPC (particles per cell) = 40, 2 days) [Thermo Fisher: Lysosomes-GFP (#C10596), Lysosomes-RFP (#C10597), Mitochondria-RFP (# C10601), ER-GFP (#C10590)]; LysoTracker™ Red DND-99 (2 μM, 45 min; Thermo Fisher, #L7528); MitoTracker™ Green FM (0.125 μM, 45 min; Thermo Fisher, #M7514); and TMRM (0.1 μM, 45 min; Fisher Scientific, #T669), which is a cell-permeant dye that accumulates in active mitochondria with intact membrane potential (ΔΨm). For analysis of lysosomal-drug-treated cultures, neurons were treated with GCase inhibitor CBE (50 μM, 7 days) (Cayman Chemicals, #15216), carmofur (7.5 μM, 2 days) (Cayman Chemicals, #14243), E-64D (10 μM, 24 h) (Cayman Chemicals, #13533), or Pepstatin A (10 μM, 24 h) (Milipore, #516481). For live imaging, neurons were grown on glass-bottomed culture dishes (MatTek, #P35G-1.5-14-C) in Neurobasal media (Life Technologies, #21103049) containing Neurocult SM1 (Stemcell Technologies, #5711). Samples were imaged on a Nikon A1R laser scanning confocal microscope with GaAsP detectors using a Plan Apo λ 100 × 1.45 NA oil immersion objective (Nikon) using NIS-Elements (Nikon). During the imaging, culture dishes were kept in a temperature-controlled chamber (37 °C) at 5% CO2.Structured illumination microscopyM–L contact sites in both fixed and live neurons were imaged using super-resolution SIM. For SIM analysis, neurons were cultured on nitric-acid-treated, PDL/laminin-coated High Precision Glass Cover Slip (Bioscience Tools, #CSHP-No1.5-12). Samples were prepared using the same protocol for regular ICC. Super-resolution images were taken on a Nikon N-SIM system with a 100× oil immersion objective lens, 1.49 NA (Nikon). Images were captured and reconstructed using Nikon NIS-Elements.Neuronal compartmentalization of M–L contactsDendrites and axons were identified as positive for either MAP2 or Tau immunostaining, respectively, in fixed cells. In live-cell imaging, axons and dendrites were distinguished using Map2-GFP lentiviral expression. Neuron cultures had dendrites that were thicker and shorter than axons, with many axons branching from dendrites, consistent with previous observations in dopaminergic neurons67. Using these characteristic morphologies, neurites were further characterized as dendrites in subsequent imaging experiments involving live or fixed neurons.Image analysisTo quantify contacts, live neurons were imaged at 2-s intervals for 3–6 min. Contacts were defined as mitochondria and lysosomes in close proximity at the beginning of the video and which lasted for greater than 20 s. The duration, number of contacts, and the density of mitochondria were analyzed manually in NIS-Elements (Nikon) for further examination. The minimum duration of contacts was quantified as the time before contact termination and dissociation (mitochondria and lysosomes detaching from one another) over the course of 5-min videos. Any contacts that lasted beyond 5 min were categorized as 300 s in bar graphs and as >300 s in histograms for the minimum duration of M–L contacts. The percentage of lysosomes in contacts was quantified as the percentage of lysosomes in contact with mitochondria for greater than 20 s divided by the total number of Lyso-GFP-positive vesicles in the region of interest. The length of neurites and the area of cell bodies were measured using a built-in function of ImageJ (NIH).Generation and transduction of lentiviral constructsHuman MAP2 Lentiviral cDNA ORF vector was purchased from Sino Biological Inc. (#HG13690-ACGLN). Human TBC1D15 ORF in the plasmid generated by Vector Builder was subjected to Q5® Site-Directed Mutagenesis (New England Biolabs, #E0554S) to insert NheI enzyme cutting site (5ʹ-GCTAGC-3ʹ) and then cloned into lentiviral vector including BFP tag (Addgene, #117136). mApple-Tom20 (Addgene, #54955) and LAMP1-mGFP (Addgene, #34831) were obtained from Addgene plasmids using the NheI/NotI cut sites and subcloned into pER4 lentiviral expression vector. For TBC1D15 knockdown, shRNA constructs against TBC1D15 were obtained from Horizon Discovery (#RHS4533-EG64786) together with TRC Lentiviral Non-targeting shRNA Control (#RHS6848). To generate the lentivirus control, the nuclear localization signal (NLS) of SV40 large T antigen was introduced after BFP coding sequence in the same lentiviral plasmid (Addgene #117136) used for hTBC1D15 cloning. Lentiviral vectors were packaged and transfected into HEK 293FT cells using X-treme Gene HP DNA transfection (Roche, #06366236001) together with helper plasmids psPAX2 (Addgene, #12260) and pLP3 (Invitrogen). Quantitation of retroviral antigens was determined using ZeptoMetrix Corporation RETROtek HIV-1 p24 Antigen ELISA kits (Fisher Scientific, #22-156-700). Concentrated viruses were aliquoted and kept at −80 °C for future use. For Map2 expression, neurons were transduced with Map2 Lentivirus and incubated 4 days before live-cell confocal imaging or fixation. For TBC1D15 expression, neurons at day 35 were transduced with hTBC1D15 Lentivirus, incubated 15 days (MOI = 8) until day 50 and were subjected to further analysis. Expression of constructs delivered by lentiviral infection was verified by either immunoblot analysis or imaging of the fluorescent label.Live-cell GCase activity assayThe enzymatic activity of GCase in live human neurons was measured as previously described15 with minor modifications. The neuron culture media was changed to phenol red-free Neurobasal media (Thermo Fisher, #12348017) containing Neurocult SM1 (Stemcell Technologies, #5711) a day before the measurement. The next day, half of the neurons were treated with DMSO and 50 μg/ml fluorescein di-β-d-glucopyranoside (PFB-FDGluc) (Fisher Scientific, #P11947) in phenol red-free Neurobasal media. Another half were treated with 400 nM BafA1 (Cayman chemicals, #11038) and 50 μg/ml PFB-FDGluc in phenol red-free Neurobasal media. Neurons were incubated for 1 h at 37 °C in the dark to allow for the accumulation of PFB-FDGluc, a substrate of GCase, in lysosomes. GCase activity was quantified for 4.5 h with 30 min intervals by measuring fluorescence upon the cleavage of PFB-FDGluc over time in a Spectramax i3 plate reader (Molecular Devices) (Ex = 485 nm, Em = 525 nm). Activity within the lysosomal compartment was determined by measuring the response to BafA1. Non-lysosomal GCase activity was interpreted as the activity that was not responsive to BafA1. After the last measurement, neurons were washed three times with PBS, and were lysed in RIPA lysis buffer. BCA protein assay was done according to the manufacturer’s protocol (Thermo Fisher, #23225) to measure the total protein amount required for normalization. Total GCase activity was quantified by calculating the area under the DMSO curve (AUC) using Prism7 (GraphPad) software. Lysosomal GCase activity was obtained by measuring the area between BafA1 and DMSO curves.SDS-PAGE and Western blottingNeurons were collected in ice-cold PBS and centrifuged at 400×g for 5 min at specific time points (e.g. day 30 for neuron characterization, day 40–70 for biochemical analysis of organelle contacts). Pellets were lysed in N-PER™ Neuronal Protein Extraction Reagent (Thermo Scientific, #87792) with cOmplete protease inhibitor cocktail (Sigma, #11836170001), and lysates were collected according to the manufacturer’s protocol. After boiling for 20 min in 4× Laemmeli sample buffer, protein samples were separated on 4–20% Tris-glycine precasted gel (Invitrogen, #XP04202BOX) and transferred to PVDF or nitrocellulose membranes using Trans-blot TurboTM transfer system (Biorad). Membranes were blocked with 5% milk in 1× Tris-buffered saline (50 mM Tris, pH 7.4, 150 mM NaCl) with 0.1% Tween (TBST) for 1 h at room temperature and incubated with a primary antibody, 4 °C, overnight: Rab7 (Cell Signaling, #D95F2, 1:1000; Abcam, #ab137029, 1:1000); TBC1D15 (Sigma Aldrich, #SAB2701508, 1:500); Fis1 (Alexis, #ALX-210-1037-0100, 1:1000); GAPDH (Millipore, #MAB374, 1:2000); GBA (Abnova, #H00002629-M01, 1:500); TH (Calbiochem, #657012, 1:1000); Synapsin (Santa Cruz, #sc-398849, 1:1000); β-III-tubulin (TUJ1) (Biolegend, #801202, 1:5000); AMPKα (Cell Signaling, #2532, 1:1000); Phospho-AMPKα (Cell Signaling, #2535, 1:1000); Ccz1 (Santa Cruz, #sc-514290, 1:1000); Mon1 (Abcam, #ab103919, 1:500); Tom20 (Abcam, #ab56783, 1:1000); Lamp1 (Santa Cruz, #sc-20011, 1:500). The next day, after three times of washing with 1× TBST, membranes were incubated in secondary goat anti-mouse and goat anti-rabbit HRP antibody (Jackson Immuno Research Lab, #115-035-146, #111-035-144, 1:10,000) diluted in 5% milk in 1× TBST for 1 h and washed three times with 1× TBST. HRP signal was developed using Clarity chemiluminescence substrate (Biorad, #170-5061) or Lumigen ECL Ultra (Lumigen, #TMA-100), and images were taken on the ChemiDoc XRS+ imaging station (Biorad). Protein levels were normalized against GAPDH. Quantification was done using ImageJ (NIH).Quantitative RT-PCRTotal RNA was purified from neurons using RNeasy Micro Kit (Qiagen, #74004) and transcribed to cDNA using the High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher, #4368814). Quantitative RT-PCR was performed using the 7500 Fast Real-Time PCR system (Applied Biosystems), with ssoAdvanced universal SYBR Green Supermix (BioRad, #1725271). The following pre-designed primer sets were used: TBC1D15 (PrimerBank ID #226342866c3) and GAPDH (PrimerBank ID #378404907c2) (Supplementary Table 1).GST-RILP pull-down assayTo determine active GTP-bound Rab7 levels in neurons, GST-RILP pull-down assay was performed as previously described25 with minor modifications. Plasmids were gifts from Aimee Edinger (Addgene plasmid #79149). GST-control and GST-fused Rab7 binding domain of RILP protein (nucleotides 658–897) were expressed in BL21 bacteria. Bacteria were collected and lysed in B-PER™ Bacterial Protein Extraction Reagent (Thermo Scientific, #78248) with 1,4-Dithiothreitol (DTT, 1 mM), EDTA (1 mM), and cOmplete protease inhibitor cocktail (Sigma, #11836170001). Proteins were purified using pre-equilibrated 50% slurry of glutathione-Sepharose 4B beads (GE Healthcare) and quantified using the BCA assay. Neurons to be analyzed in the pull-down assay were pelleted and lysed in pull-down buffer (20 mM HEPES, 100 mM NaCl, 5 mM MgCl2, 1% TritonX-100, and protease inhibitors). GST-control beads and GST-RILP beads were added to the neuron lysates and the samples were rocked overnight at 4 °C, followed by washing with cold pull-down buffer. Bound proteins were eluted by boiling in 2× Laemmli sample buffer (Sigma Aldrich, #S3401) and used for western blot analysis.Seahorse XF Cell Mito Stress TestOCR of neurons was analyzed using an XF 24 Extracellular Flux Analyzer (Seahorse Biosciences) according to the manufacturer’s protocol; 1.25 × 105 neurons were plated on one well of XF24 cell culture microplates. Four empty wells without neurons were used as background control for temperature-sensitive fluctuations in OCR analysis. Before the assay, culture medium was replaced with Seahorse XF medium (Seahorse Bioscience, #103575-100) supplemented with 1 mM sodium pyruvate (Corning®, #25-000-CI), 10 mM d-glucose (Sigma Aldrich, #G8769), and 2 mM glutamine (Gibco, #25030-081) and incubated for 1 h in a CO2-free incubator. OCR was measured, after sequential injection of 1 µM oligomycin, CCCP, and Antimycin A. After the assay, neurons were lysed and subjected to BCA protein assay (Thermo Fisher, #23225) for normalization.Total cellular ATP assayTotal cellular ATP content was measured using the ATPlite kit (PerkinElmer, #6016943) according to the manufacturer’s protocol. Neurons were plated on black 96-well plates (Thermo Fisher, #237108) at a density of 5 × 104/well. Neurons were lysed in lysis solution provided in the kit and luminescence was measured in a Spectramax i3 plate reader (Molecular Devices). Neuron lysates were subjected to BCA protein assay (Thermo Fisher, #23225) for normalization.Mitochondrial drug treatmentHeLa cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco, #11995-065) supplemented with 10% FBS, 100 U/ml Penicillin–Streptomycin (Gibco, #15140-122). A day after passaging, cells were treated for 0, 4, and 8 h with CCCP (10 μM) (Sigma Aldrich, #C2759), Rotenone (100 nM) (Sigma Aldrich, #R8875), Antimycin (1 μM) (Sigma Aldrich, #A8674), or Oligomycin (1 μM) (Sigma Aldrich, #O4876). Treated cells were lysed in RIPA buffer (Boston Bioproduct, #BP-115-5x) with cOmplete protease inhibitor cocktail (Sigma, #11836170001), and lysates were subjected to BCA protein assay (Thermo Fisher, #23225) and SDS-PAGE.Statistical analysisFor all statistical tests, cells from N ≥ 3 independent experiments (biological replicates) per condition were used (see text and figure legends for details). Data were analyzed using unpaired two-tailed Student’s t-test (for two datasets) or one-way ANOVA with Tukey’s post-hoc test (for multiple datasets). Bar graphs presented are in the form of means ± SEM. Statistics and graphing were performed using Prism 7 (GraphPad) software. Videos and images were processed using NIS-Elements (Nikon) and assembled using ImageJ. All figures were assembled in Adobe Illustrator.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationDescription of Additional Supplementary FilesSupplementary Movie 1Supplementary Movie 2Supplementary Movie 3Supplementary Movie 4Supplementary Movie 5Supplementary Movie 6Supplementary Movie 7Supplementary Movie 8Reporting Summary | nature communications | [
"Article"
] | [
"Lysosomes",
"Mitochondria",
"Parkinson's disease"
] | IntroductionMitochondria–lysosome contact sites identified inter-organelle membrane tethering mitochondria lysosomes1,2 contacts allow bidirectional regulation mitochondrial lysosomal network dynamics direct interaction distinct from mitophagy lysosomal degradation regulated by TBC1D15 Rab7 GAP recruited to mitochondria mediates Rab7 GTP hydrolysis untethering formation role of M–L contacts in human neurons not investigated.Parkinson’s disease (PD) common neurodegenerative movement disorder loss of dopaminergic neurons motor symptoms mitochondrial lysosomal defects genetically linked to lysosomal enzyme GBA1 for hydrolysis genetic risk factor for PD GBA1 mutation carriers severe cognitive GBA1-PD mutant neurons lower GCase protein levels reduced GCase wild-type GCase activity decreased in idiopathic familial PD patient targeting GBA1 potential therapeutic approach for key role for GCase activity in PD pathogenesis M–L contact dynamics disrupted in PD neurons modulated by GCase activity not knowninvestigated impact M–L contact sites neuronal function GBA1-linked PD pathogenesis super-resolution-lapse imaging M–L contact sites formed soma axon dendrites human iPSC-derived dopaminergic neurons role contacts PD patient dopaminergic neurons untethering neurons GBA1 defective M–L contact untethering caused decreased GCase lysosomal enzymatic activity GlcCer accumulation decreased levels untethering protein TBC1D15 GTP hydrolysis defects disrupted mitochondrial distribution function rescued by TBC1D15 Parkinson’s patient GBA1-linked neurons role M–L contact sites human neurons GBA1-linked PD pathophysiology contact sites form utilized induced pluripotent stem cell) technology midbrain dopaminergic neurons healthy controls Differentiated neurons expressed marker β-III-tubulin (TUJ1) dopaminergic-specific markers tyrosine hydroxylase Factor 1 Alpha imaged for mitochondrial lysosomal dynamics live neurons3D microscopy found mitochondria lysosomes-RFP formed stable inter-organelle contacts in human neurons (Fig. M–L contacts confirmed <10 nm apart electron microscopy (Fig. consistent contact conducted confocal time-lapse microscopy live neurons M–L contacts formed remained tethered (yellow (white (Fig. 1c 1–3) Mitochondria lysosomes maintained membrane potential visualized TMRM imaging dynamic M–L contact tethering observed imaging outer mitochondrial membrane lysosomal membrane proteins imaging mitochondrial lysosomal-targeted dyes visualize M–L contacts proximity ligation assay (PLA) imaging TOM20 LAMP1 mitochondrial lysosomal membrane Fig 4a M–L contacts remained tethered average duration 88.07 ± 5.05 s (Fig. 1d ~18% lysosomes contact with mitochondria (Fig. M–L contacts form crosstalk between mitochondria lysosomes human neurons. 1Mitochondria–lysosome contacts human neurons3D microscopy images M–L contacts wild-type human dopaminergic iPSC-derived neurons (mitochondria red Mito–RFP lysosomes green Lyso-GFP (30 neurons 3 experiments electron microscopy images M–L contacts distance membranes <10 nm (mitochondria M lysosomes time-lapse confocal images contacts mitochondria lysosomes 2 s intervals 3–5 min Yellow stable M–L contacts White arrows contacts before Black line shows duration duration M–L contacts confocal images = 156 contacts 26 neurons 3 experiments). Average minimum duration M–L contacts Relative frequency distribution contacts X-axis bin centers width 60 s Percentage lysosomes contacting mitochondria >20 s = 41 neurons 3 experiments). data means ± S.E.M. Scale bars 500 nm 100 nm (b).Spatial compartmentalization neuronal M–L contact investigated compartmentalization M–L contacts soma dendrites axons human neurons Dendrites axons positive for MAP2 or Taufound mitochondria lysosomes tethered at contact sites neuronal compartments soma dendrites axons M–L contacts in dendrites axons decreased mitochondrial lysosomal motility spatial compartmentalization quantified duration M–L percentage lysosomes contact mitochondria across neuronal compartments live-cell imaging average minimum duration M–L increased in axons soma 80.6 ± 6.2 s 82.4 ± 9.7 s 112.5 ± 13.2 s percentage lysosomes contacting mitochondria in soma dendrites axons differ results M–L contacts form varying dynamics across neuronal compartments human. 2Spatial compartmentalization neuronal mitochondria–lysosome contact dynamics time-lapse images of contacts between mitochondria lysosomes in soma dendrites axons human dopaminergic iPSC-derived neurons recordings 2 s intervals for 5 min Yellow arrows mark stable M–L contacts White contacts before formation after untethering Black line shows durationScale bar 500 nm Supplementary Movies 4–6) live-cell imaging dual color kymographs M–L contacts dendrites axons red Mito–RFP lysosomes green Lyso confocal image frames contact sites yellow arrows kymographs white scale bar 1 μm yellow vertical bar 30 s Yellow start end M–L contact tethering black line duration Quantification M–L contacts neuronal compartments One-way ANOVA Tukey’s multiple comparisons test average minimum durations M–L contacts soma dendrites axons 86 35 34 26 neurons 3 *p 0.0335 Percentage lysosomes contacting mitochondria axons >20 s 14 13 neurons 3 data ± S.E.M *p ≤ 0.05 not GCase activity disrupts M–L contact GBA1-PD patient dopaminergic neuronsPD linked mitochondrial lysosomal M–L contacts investigatedmutations in lysosomal enzyme GCase (GBA1) risk for PD12 GCase activity decreased in familial PD patient examined M–L contact dynamics in GBA1-PD patient neurons PD patient fibroblasts GBA1 generated human iPSCs isogenic controls by CRISPR-Cas916 differentiated into midbrain dopaminergic neurons mutant GBA1 isogenic control affect fibroblast reprogramming to differentiation neurons PD patient mutant GBA1 dopaminergic neurons decreased GCase protein levels live-cell GCase activity assays mutant GBA1 neurons reduced GCase enzymatic activity isogenic control neurons activity used Bafilomycin A1 inhibitor lysosomal acidification decrease in GCase activity in mutant GBA1 neurons to CRISPR-corrected isogenic control neurons 3Loss of GCase activity disrupts mitochondria–lysosome contact in GBA1-PD patient dopaminergic neuronsWestern blot analysis PD mutant GBA1 dopaminergic neurons CRISPR-corrected control) neurons GCase level reduced ∆GBA neurons = 4 t-test = 0.0083 ∆GBA Corr neurons treated DMSO BafA1 live-cell GCase activity analysis area under curve decreased GCase activity neurons = Lysosomal GCase activity calculated subtracting BafA1 values DMSO Values-change isogenic controls (N = 3 experiments). t-test *p = 0.0352. time-lapse images contacts mitochondria lysosomes Corr ∆GBA neurons Yellow arrows stable M–L contacts White arrows M–L contacts untethering Black line duration contacts Scale bar 500 nm average minimum duration relative frequency distribution stable M–L contacts increased duration ∆GBA neurons (n = 48 contacts Corr 71 ∆GBA 3 X-axis bin centers Bin width 60 s Student’s t-test *p = 0.0204.Percentage lysosomes contacting mitochondria >20 s) (n = 12 Corr 36 ∆GBA neurons 3 experiments). Unpaired two-sided Student’s t-test quantifications data means ± S.E.M. *p ≤ 0.05 **p ≤ 0.01 ns not significant investigated lysosomal contacts mitochondria disrupted loss lysosomal GCase activity confocal live-cell microscopy lysosomes mutant GBA1 CRISPR-corrected isogenic control neurons M–L contacts formed conditions average duration tethering increased mutant GBA1 neurons inefficient untethering events percentage lysosomes contacts similar between conditions untethering formation M–L contact tethering disrupted mutant GBA1 neurons PLA imaging mutant GBA1 CRISPR neurons contacts form both conditions EM imaging contacts mutant GBA1 CRISPR-corrected neurons contacts tethered both conditions length membrane contact lysosomes not altered mutant GBA1 neurons ER–mitochondria ER–lysosome contact formation not disrupted mutant GBA1 neurons5a–f). results suggest loss of GCase activity M–L contact formation disrupts untethering prolonged tethering between mitochondria lysosomes examined changes in M–L contact dynamics specific to loss GCase activity lysosomal enzyme dysfunction Mutant GBA1 neurons altered lysosomal morphology enlarged lysosomes 3g inhibition of lysosomal enzymes acid ceramidase disrupt M–L contact dynamics results suggest pathway disrupted by loss GCase activity lysosomal defects enzyme dysfunction assessed rescuing GCase activity in mutant GBA1 neurons restore defective M–L contact dynamics modulator S-181 GCase activity S-181 treatment rescued prolonged M–L contact tethering in mutant GBA1 neurons highlighting role GCase activity regulating.Tethering proteins disrupted in GBA1-PD patient dopaminergic examined potential mechanism mutant GBA1 disrupt M–L contact dynamicsshowed M–L contact untethering regulated by lysosomal RAB7-GTP hydrolysis from to GDP-bound Rab72 driven by activity mitochondrial TBC1D15 recruited outer mitochondrial membrane by Fis14,6 investigated RAB7-GTP hydrolysis disrupted by mutant GBA1 in patient neurons measured protein levels RAB7 Fis1 TBC1D15 in PD patient-derived mutant GBA1 dopaminergic neurons controls no differences in Rab7 Fis1 levels Fis1 levels altered after normalizing TBC1D15 Fis1 localization to mitochondria Rab7 to lysosomes Rab7 GEF proteins altered in mutant GBA1 neurons TBC1C15 levels decreased in mutant GBA1 patient neurons (*p 0.05) (Fig 4Tethering proteins mitochondria–lysosome contact disrupted in GBA1-PD patient dopaminergic neurons analysis of TBC1D15 Rab7 Fis1 levels in PD patient mutant GBA1 dopaminergic neurons-correctedProtein levels normalized GAPDH Values-change compared Corr 3-sided Student’s t-test *p = 0.0442 GST-RILP GTP-bound Rab7 levels ∆GBA Corr neurons Rab7-GTP levels normalized Rab7 GAPDH Values fold-change Corr 3 Student’s t-test *p = 0.0478 data ± S.E.M. *p ≤ hypothesized disrupted M–L contact untethering mutant GBA1 neurons due defective Rab7-GTP hydrolysis reduced TBC1D15 levels performed-RILP GTP-bound Rab7 levels PD patient-derived mutant GBA1 dopaminergic neurons increased levels RAB7-GTP/total Rab7 CRISPR-corrected isogenic control neurons (*p ≤ 0.05) decreased TBC1D15 levels GBA1 neurons disrupt Rab7 GTP hydrolysis increased GTP-bound Rab7 prolonged M–L contact tethering analyzed TBC1D15 qPCR decreased TBC1D15 protein levels neurons not due lower TBC1D15 transcripts levelsdecreased TBC1D15 protein levels proteasomal degradation inhibited by lactacystin similar levels between mutant GBA1 isogenic control neurons mitochondrial dysfunction expression Treatment-trifluoromethoxyphenylhydrazone Complex I rotenone Antimycin ATP inhibitor Oligomycin disrupt TBC1D15 expression in cells increasing GCase activity in mutant GBA1 neurons modulator S-18116 decreased levels TBC1D15 suggest TBC1D15 levels downregulated by proteosomal degradation in mutant GBA1 neurons loss GCase activity mitochondrial dysfunction loss of TBC1D15 M–L contact dynamics in human neurons Knockdown TBC1D15 in wild-type iPSC-derived neurons increased M–L contact tethering duration decreased TBC1D15 levels in GBA1-linked patient neurons disrupt M–L contact dynamicsInhibition GCase activity disrupts M–L contact untetheringTo defects loss GCase activity treated wild-type human midbrain dopaminergic neurons-b-epoxide (CBE) (50 μM 7 days decreases GCase activity promotes GlcCer-cell time-lapse imaging mitochondria lysosomes wild-type CBE-treated neurons M–L contacts formed. CBE treatment lysosomal size inhibition GCase activity prolonged M–L contact tethering duration (***p ≤ 0.001) consistent mutant GBA1 neurons decreased GCase activity contact untethering. 5Inhibition GCase activity disrupts mitochondria–lysosome contact untethering dopaminergic neurons Healthy human dopaminergic neurons treated with CBE (50 μM 7 days GCase activity time-lapse confocal images contacts mitochondria lysosomes Ctrl CBE-treated neurons Yellow arrows mark stable contacts White arrows contacts after untethering Black line shows duration contacts Scale bar = 500 nmconfocal images lysosomes (LAMP1–GFP Ctrl CBE-treated neurons Scale bar 1 μm effects CBE treatment lysosomal confirmed lysosomes enlarged >0.5 μm M–L contacts 12 neurons per condition 3 t-test ***p = 0.0001 CBE treatment increased duration M–L contacts 60 contacts per condition 3 = 0.0001 confocal images immunocytochemistry GlcCer Ctrl CBE-treated neurons Scale bar 5 μm CBE treatment increased GlcCer levels 21 neurons per condition 3 experiments). Exogenous GlcCer treatment increased duration M–L contacts 93 contacts Ctrl 94 +GlcCer 3 Student’s t-test ***p < 0.0001 analysis TBC1D15 Rab7 Fis1 levels Ctrl CBE-treated neurons Protein levels normalized control GAPDH Values-change compared Ctrl (N = 5 experiments). Paired-sided Student’s t-test *p = 0.0104. GST-RILP pull-down GTP-bound Rab7 levels Ctrl CBE-treated neurons Rab7-GTP levels normalized GAPDHValues fold-change Ctrl 3 Paired two-sided Student’s t-test *p = 0.0242. quantifications data ± S.E.M. *p ≤ 0.05 ***p ≤ 0.001 validated findings two wild-type iPSC-derived neuronal lines characterized iPSC midbrain dopaminergic neuron-specific markers Fig Inhibition activity CBE treatment prolonged M–L contact tethering duration disrupts contact loss GCase activity increased GlcCer levels asked treatment exogenous GlcCer wild-type neurons disrupt M–L contact dynamics live-cell microscopy neurons GlcCer led prolonged contact tethering duration (***p < 0.001) increased GlcCer levels mutant GBA1 neurons investigated inhibition M–L contact untethering measured protein levels Rab7 TBC1D15 Fis1 wild-type CBE-treated neurons CBE-treated neurons showed decreased TBC1C15 levels (*p ≤ 0.05) without changes Rab7 Fis1 levels consistent with mutant GBA1 patient-derived neuronsfindings wild-type iPSC-derived neurons treated with exogenous GlcCer showed reduced TBC1D15 levels Fig. 10f examined inhibition activity by CBE treatment defects Rab7 GTP hydrolysis elevated GTP-bound Rab7 levels GST-RILP assay CBE-treated neurons GCase inhibition increased RAB7-GTP Rab7 consistent mutant GBA1 neurons results support hypothesis loss GCase activity increases GlcCer levels TBC1D15 levels Rab7-GTP hydrolysis misregulation M–L contact untethering dynamics.Mitochondrial dysfunction prolonged M–L contacts by TBC1D15 contacts mitochondrial lysosomal network dynamics investigated mitochondrial dynamics in PD patient-derived mutant GBA1 dopaminergic neurons defective contact untethering distribution mitochondria soma axons Healthy mitochondria imaged-cell imaging in mutant GBA1 CRISPR-corrected isogenic control neurons Fig. no difference in mitochondrial density in soma. 6aaxonal mitochondrial density decreased in mutant GBA1 neurons isogenic control neurons mutant GBA1 disrupts axonal mitochondrial distribution Parkinson’s neurons observed defective mitochondrial respiration decreased oxygen consumption rate mutant neurons decreased AMPK activation measured cellular ATP level decreased levels mutant GBA1 neurons normalization mitochondrial mass < 0.001 TBC1D15 levels decreased mutant GBA1 neurons observed mitochondrial dysfunction in TBC1D15 knockdown neurons 6Mitochondrial dysfunction mitochondria–lysosome contacts by expression TBC1D15 in GBA1-PD patient dopaminergic neurons Live-cell distribution mitochondria soma axons mutant GBA1 dopaminergic neurons CRISPR-corrected isogenic control neurons White arrows single axonmitochondrial density soma (TMRM-positive pixels Corr 47 ∆GBA 47 3 mitochondrial density axons count/length axon Corr 122 ∆GBA 114 3 Unpaired Student’s t-test ***p < 0.0001 Oxygen consumption rate (OCR) measured normalized protein content Corr 9 ∆GBA 9 3 analysis-AMPKα AMPKα levels Corr ∆GBA neurons Ratio p-AMPKα/AMPKα-change Corr 3 t-test *p = 0.0354. Total cellular ATP content measured normalized protein content (ng/μl). Corr 40 ∆GBA 40 3 t-test ***p < 0.0001 Live-cell distribution mitochondria (TMRM soma axons Corr ∆GBA neurons with TBC1D15 lentiviral expression White arrows mark mitochondria single axon Scale bar 10 μm average minimum duration M–L contacts Corr-vehicle 62 18 ∆GBA-vehicle 60 15 ∆GBA 63 19 3 ***p < 0.0001.Quantified mitochondrial density soma (TMRM-positive pixels Corr-vehicle 32 ∆GBA-vehicle ∆GBA-TBC1D15 38 3 Quantified mitochondrial density axons (mitochondria count/length axon Corr-vehicle 124 ∆GBA-TBC1D15 3 ***p < 0.0001 cellular ATP concentration measured normalized to protein content (ng/μl). Corr-vehicle ∆GBA ***p = 0.0002 *p = 0.0493 One-way ANOVA Tukey’s multiple comparisons test data means ± S.E.M. *p ≤ 0.05, ***p ≤ 0.001 ns not significant hypothesized dysregulated M–L contact untethering dynamics mutant GBA1 neurons defects mitochondrial distribution function test contact expressed human TBC1D15 mutant neurons by lentiviral transduction (Fig. 6g expression exogenous TBC1D15 confirmed by western blot analysis fluorescence imagingconsistent TBC1D15 promotes M–L contact transduction TBC1D15 in mutant GBA1 neurons contact untethering decreased M–L contact durations compared to lentiviral vehicle-treated neurons ≤ 0.001) investigated effect rescuing M–L contact dynamics on axonal mitochondrial density in mutant GBA1 neurons TBC1D15 alter mitochondrial densities rescued decreased mitochondrial density TBC1D15 partially ATP levels mutant GBA1 neurons results suggest upregulation TBC1D15 in PD patient mutant GBA1 neurons misregulation M–L contact untethering defects mitochondrial dynamics function extended analysis to additional PD mutant GBA1 patient neurons CRISPR-edited isogenic control neurons neurons showed decreased GCase levels enzymatic activity increased GTP-bound Rab7 due to lower TBC1D15 levels prolonged M–L contact tetheringS-181 treatment in mutant GBA1 (N370S) neurons M–L contact dynamics TBC1D15 levels mutant GBA1 neurons reduced mitochondrial function findings highlight defective GCase activity PD neurons Rab7 GTP hydrolysis machinery M–L contact site dynamics study M–L contact sites form in human neurons GBA1-linked PD loss of lysosomal GCase activity in PD patient neurons led prolonged M–L contact tethering dynamics misregulated axonal distribution decreased ATP levels M–L contact site dynamics PD neurons defects mitochondrial distribution function potential role M–L contact site dysregulation in PD pathogenesis genes linked to mitochondria lysosomes causative mitochondrial lysosomal dysfunction implicated in functional crosstalk oxidized dopamine mediates mitochondrial lysosomal dysfunction in human not mouse dopaminergic importance of patient-derived dopaminergic neurons for PD pathogenesisrole mitochondria lysosomes in PD pathogenesis direct homeostatic relationship lysosomal degradation not examined work evidence M–L contacts homeostasis dopaminergic neurons link mitochondrial lysosomal dysfunction PD pathogenesis focused on GBA1 mutations genetic risk factor for wild-type GCase enzyme activity reduced in patient neurons with PD GBA1 mutations9 loss of GCase activity PD pathogenesis loss of GCase function compromises lysosomal protein contributes to PD pathogenic phenotypes α-synuclein accumulation13 GBA1 mutations abnormal GCase activity linked to mitochondrial dysfunction9 GBA1 mutant PD patient neurons have defective M–L contact dynamics prolonged contact tethering loss of GCase activity increased active GTP-bound lysosomal Rab7 M–L contact tethering2. due to decreased levels TBC1D15 GAP) in mutant GBA1 neurons Rab7 GTP hydrolysis M–L contact increasing TBC1D15 expression prolonged M–L contact tethering in mutant GBA1 neuronsobserved mitochondrial dysfunction in patient neurons abnormal distribution in axons decreased respiration lower ATP levels TBC1D15 expression mitochondrial dysfunction dysregulation of M–L contacts in GBA1-linked PD pathogenesis targeting contact machinery may mitochondrial dysfunction mitochondria energy suppliers in synapses active neurons abnormal distribution may to synaptic dysfunction in inter-organelle contact sites essential for cellular function recently studies reported function molecular architecture of contacts in neurons dysfunction of inter-organelle contacts in key diseases1 recent identification of M–L contact sites on relationship between mitochondria lysosomes independent of degradation direct crosstalk regulation previous study showed formation of M–L contacts in non-neuronal cells identified protein mediators for contact untethering1,2 M–L contacts key contributors in human neurons in multiple neuronal compartments important for regulation mitochondrial lysosomal dynamics findings provide insights into inter-organelle contacts cellular homeostasis suggest importance of M–L contacts potential for therapeutic development in PDiPSC culture procedures described Healthy patient skin (GBA1 heterozygous 84GG mutation levels p.N370S mutation reprogrammed into iPSCs Northwestern University Stem Cell Core Facility Sendai virus Yamanaka factors (Oct Sox2 Klf4 c generated isogenic control iPSC line mutation CRISPR/Cas9 cloned into vector PX461 (Addgene #48140) cDNA encoding GFP-tagged Cas9 nuclease plasmid electroporated into GBA1 84GG mutant iPSCs ssODN corrected sequence After 48 h isogenic control line confirmed FACS-sorting sequencing iPSCs maintained in mTeSRTM1 mTeSRTM Plus media cells passaged every 6–8 days iPSC lines characterized for pluripotency markers Cells plated PDL-coated coverslips subjected immunofluorescence staining for NANOG OCT4 SSEA4 TRA1-81 Genomic integrity confirmed Mycoplasma tests performed monthly iPSC lines characterization hiPSC-derived dopaminergic neurons differentiated M–L contactsplating iPSCs cells treated factors original protocol day 13 cells passaged 1–2 10-cm culture dishes poly-d-lysine (PDL laminin (Invitrogen day 25 neurons treated accutase PDL/laminin-coated dishes dopaminergic marker characterization TH FOXA2 LMX1A neural marker β-III-tubulin day 30 day 50 neurons mature maintained Neurobasal media Neurocult SM1 (Stemcell Technologies #5711) Neurons day 50–60 immunostaining EM live-cell time-lapse imaging super-resolution SIM Neurons 3 experiments immunostained multiple antibodies plated PDL/laminin-coated fixed 4% paraformaldehyde 20 min permeablized 10% FBS 0.1% saponin room temperature staining 10% FBS substituted 2% gelatin (Sigma #G1393)Cells immuno-labeled antibodies Oct4 (Abcam #19857 SSEA1 MAB#4304 Nanog&D systems Tra-1-81 #MAB4381 TH (Calbiochem #657012 β-III-tubulin (Biolegend #801202 Lmx1a (Milipore MAB#10533 FoxA2 Cruz-101060 Tom20 #78547 Lamp1 Cruz #sc-20011 Map2 (Novus Biological #NB300213 Tau (DAKO #A002401-2 GlcCer (Glycobiotech_0011 overnight incubation 4 °C coverslips washed three PBS 5 min incubated Alexa-conjugated secondary antibodies #A21206 1:1000 1 h room temperature washed three times Superfrost Plus microscope slides VECTASHIELD HardSet Antifade Mounting Medium ProLongTM Diamond Antifade Mountant sample Images obtained Leica DMI4000B confocal microscope Nikon A1R laser scanning confocal microscope Colocalization measured EzColocalization plugin ImageJ HealthProximity ligation mitochondrial protein Tom20 lysosomal Lamp1 detected PLA kit Neurons plated PDL/laminin-coated coverslips immuno-labeled Tom20 Lamp1 red PLA signals Nikon A1R laser microscope neurons grown PDL/laminin coverslips fixed 2.5% glutaraldehyde 2% paraformaldehyde 0.1 M cacodylate buffer pH 7.3 30 min submitted Northwestern EM facility processing-fixation 1% osmium tetroxide 3% uranyl acetate cells dehydrated ethanol Epon resin polymerized 48 h 60 °C Ultrathin sections UCT ultramicrotome contrasted 4% uranyl acetate Reynolds’s lead citrate Samples imaged Tecnai Spirit G2 electron microscope 80 kVImages captured Eagle 4k HR 200kV CCD camera analyzed ImageJ (NIH).Live-cell time-lapse confocal mitochondria lysosomes reagents Lentivirus Tom20-RFP Lamp1-GFP (MOI 3 CellLight® BacMam 2.0 baculovirus 40 2 days Fisher Lysosomes-GFP#C10596)-RFP#C10597) Mitochondria-RFP ER-GFP#C10590) LysoTrackerTM Red DND-99 (2 μM MitoTrackerTM Green FM (0.125 μM 45 min TMRM (0.1 μM 45 min cell-permeant dye active mitochondria lysosomal-drug-treated cultures neurons treated inhibitor CBE (50 μM 7 days carmofur (7.5 μM 2 days E-64D (10 μM 24 h Pepstatin A (10 μM 24 h neurons grown glass-bottomed culture dishes Neurobasal media Neurocult SM1Samples imaged Nikon A1R laser microscope GaAsP detectors Apo 100 × 1.45 NA oil immersion objective NIS-Elements culture dishes temperature-controlled chamber (37 °C) 5% CO2.Structured illumination microscopyM–L contact sites fixed neurons imaged super-resolution SIM neurons cultured nitric-acid-treated PDL/laminin-coated High Precision Glass Cover Slip Samples prepared protocol ICC Super-resolution images Nikon N-SIM system 100× oil immersion objective lens 1.49 captured reconstructed NIS-Elements.Neuronal compartmentalization M–L contactsDendrites axons identified positive MAP2 Tau immunostaining fixed cells live-cell imaging axons dendrites distinguished Map2-GFP lentiviral expression cultures dendrites thicker shorter axons branching dendrites dopaminergic neurites characterized dendrites imaging experiments live neurons imaged 2-s intervals 3–6 min Contacts defined mitochondria lysosomes 20 s duration density analyzed manually NIS-Elementsminimum duration contacts before termination dissociation lysosomes over 5-min videos contacts beyond 5 min categorized 300 s bar >300 s histograms M–L contacts percentage lysosomes in contacts than 20 s divided by Lyso-GFP-positive vesicles region length neurites area cell bodies measured ImageJ transduction lentiviral constructsHuman MAP2 Lentiviral cDNA ORF vector purchased from Sino Biological Human TBC1D15 ORF subjected to Q5® Site-Directed Mutagenesis cloned into lentiviral vector BFP tag (Addgene #117136) mApple-Tom20 LAMP1-mGFP #34831) obtained from Addgene plasmids subcloned into pER4 lentiviral expression vector TBC1D15 knockdown shRNA constructs obtained from Horizon Discovery TRC Lentiviral Non-targeting shRNA Control (#RHS6848) nuclear signal) SV40 large T antigen introduced after BFP coding sequence lentiviral plasmid hTBC1D15 cloningLentiviral vectors transfected HEK 293FT cells X Gene HP DNA transfection plasmids psPAX2 pLP3 retroviral antigens determined ZeptoMetrix Corporation RETROtek HIV-1 p24 Antigen ELISA kits Scientific Concentrated viruses aliquoted kept −80 °C Map2 expression neurons transduced Lentivirus incubated 4 days before live-cell imaging TBC1D15 expression neurons day 35 transduced hTBC1D15 Lentivirus incubated 15 days until day 50 subjected analysis Expression lentiviral infection verified immunoblot analysis imaging fluorescent label.Live-cell GCase activity enzymatic activity GCase neurons measured modifications neuron culture media changed phenol red-free Neurobasal media Neurocult SM1 half neurons treated DMSO 50 μg/ml fluorescein-β-d-glucopyranoside half treated 400 nM BafA1 50 μg/ml PFB-FDGluc Neurons incubated 1 h at 37 °C accumulation PFB-FDGluc lysosomesGCase activity quantified 4.5 h 30 min intervals measuring fluorescence PFB-FDGluc Spectramax i3 plate reader 485 Em 525 Activity lysosomal compartment response BafA1 Non-lysosomal GCase activity responsive BafA1 neurons washed three times PBS lysed RIPA lysis buffer BCA protein assay manufacturer’s protocol protein amount normalization Total GCase activity quantified calculating area DMSO curve Prism7 Lysosomal GCase activity obtained area between BafA1 DMSO curves blottingNeurons collected ice-cold PBS centrifuged 400×g 5 min day 30 day 40–70 Pellets lysed N-PERTM Neuronal Protein Extraction Reagent cOmplete protease inhibitor cocktail lysates collected manufacturer’s protocolboiling 20 min 4× Laemmeli buffer protein samples separated 4–20% Tris-glycine gel transferred PVDF nitrocellulose membranes Trans-blot TurboTM transfer system Membranes blocked 5% milk 1× Tris-buffered saline (50 mM Tris pH 7.4 150 mM NaCl 0.1% Tween (TBST 1 h temperature incubated primary antibody 4 °C overnight Rab7 Abcam TBC1D15 Fis1 GAPDH GBA TH (Calbiochem Synapsin Cruz-398849 β-III-tubulin (TUJ1) (Biolegend #801202 AMPKα Phospho-AMPKα Ccz1 Mon1 Tom20 Lamp1 Cruz #sc-20011 next day three times washing 1× TBST membranes incubated secondary goat anti-mouse anti-rabbit HRP antibody Immuno Lab 1:10,000 diluted 5% milk 1× TBST 1 h washed three times 1× TBSTHRP signal developed Clarity chemiluminescence substrate Lumigen ECL Ultra images ChemiDoc XRS+ imaging station Protein levels normalized GAPDH Quantification ImageJ RT RNA purified neurons RNeasy Micro Kit transcribed cDNA High-Capacity cDNA Reverse Transcription Kit Fisher #4368814) Quantitative RT-PCR 7500 Fast Real-Time PCR system SYBR Green Supermix #1725271) pre-designed primer sets TBC1D15 GAPDH #378404907c2)-RILP pull-down GTP-bound Rab7 levels neurons assay modifications Plasmids Aimee Edinger GST-control-fused Rab7 binding domain RILP protein expressed BL21 bacteria Bacteria collected lysed B-PERTM Bacterial Protein Extraction Reagent 1,4-Dithiothreitol EDTA cOmplete protease inhibitor cocktail Proteins purified-equilibrated 50% glutathione-Sepharose 4B beads quantified BCA assayNeurons pelleted lysed buffer (20 mM HEPES 100 mM NaCl 5 mM MgCl2 1% TritonX-100 protease inhibitors). GST-control GST-RILP beads added lysates samples rocked overnight 4 °C cold buffer Bound proteins eluted 2× Laemmli buffer western blot analysis.Seahorse XF Cell Mito Stress TestOCR neurons analyzed XF 24 Extracellular Flux Analyzer 1.25 × 105 neurons plated XF24 cell culture microplates Four empty wells without neurons control fluctuations culture medium replaced Seahorse XF medium supplemented 1 mM sodium pyruvate 10 mM d-glucose 2 mM glutamine incubated 1 h CO2-free incubator OCR measured injection 1 μM oligomycin CCCP Antimycin A neurons lysed BCA protein assay normalization cellular ATP measured ATPlite kit Neurons plated on 96-well plates density 5 × 104/well.Neurons lysed in solution luminescence measured Spectramax i3 plate reader lysates subjected to BCA protein assay #23225) for normalization.Mitochondrial drug treatmentHeLa cells cultured in Dulbecco’s Eagle’s medium supplemented with 10% FBS, 100 U/ml Penicillin–Streptomycin after cells treated 0 4 8 h with CCCP (10 μM Rotenone (100 nM Antimycin (1 Oligomycin (1 μM Treated cells lysed in RIPA buffer cOmplete protease inhibitor cocktail lysates subjected to BCA protein assay SDS-PAGE.Statistical cells from N ≥ 3 experiments per used Data analyzed using unpaired two-tailed Student’s t-test one-way ANOVA with Tukey’s post test Bar graphs means ± SEM Statistics Prism 7) software Videos images processed NIS-Elements assembled ImageJ figures assembled in Adobe Illustrator information Nature Research Reporting SummarySupplementary informationSupplementary InformationDescription Additional Supplementary FilesSupplementary Movie 4Supplementary 6Supplementary 8Reporting Summary | 46.8 | 1.073671 |
10.1038/s41467-020-16062-6 | PMC7203111 | Bottom-up design of functional device components based on nanometer-sized building blocks relies on accurate control of their self-assembly behavior. Here, the authors demonstrate a solvent-mediated polymerization of atom-precise gold–silver nanoclusters into macroscopic single crystals with highly anisotropic p-type semiconducting characteristics. | Bottom-up design of functional device components based on nanometer-sized building blocks relies on accurate control of their self-assembly behavior. Atom-precise metal nanoclusters are well-characterizable building blocks for designing tunable nanomaterials, but it has been challenging to achieve directed assembly to macroscopic functional cluster-based materials with highly anisotropic properties. Here, we discover a solvent-mediated assembly of 34-atom intermetallic gold–silver clusters protected by 20 1-ethynyladamantanes into 1D polymers with Ag–Au–Ag bonds between neighboring clusters as shown directly by the atomic structure from single-crystal X-ray diffraction analysis. Density functional theory calculations predict that the single crystals of cluster polymers have a band gap of about 1.3 eV. Field-effect transistors fabricated with single crystals of cluster polymers feature highly anisotropic p-type semiconductor properties with ≈1800-fold conductivity in the direction of the polymer as compared to cross directions, hole mobility of ≈0.02 cm2 V−1 s−1, and an ON/OFF ratio up to ≈4000. This performance holds promise for further design of functional cluster-based materials with highly anisotropic semiconducting properties. | IntroductionThe remarkable progress in the synthesis, structural discovery, functionalization and theoretical understanding of ligand-stabilized, atom-precise metal nanoclusters has opened fascinating opportunities to use these well-defined nanometer-size building blocks for designing nanomaterials with tunable properties1–9. The bottom-up fabrication of nanomaterials relies on spontaneous or directed self-assembly. Self-assembly based on supramolecular weak interactions or designed linkers was introduced for bottom-up fabrication of nanomaterials, consisting of colloidal metal nanoparticles, since mid-1990’s. Early successful realizations included gold nanoparticle assemblies where individual particles were linked by dithiols10, DNA11,12, or using viral proteins as scaffolds13,14. Gold nanoclusters stabilized by thiolates15 have for a long time served as a prototype, and the first successful structural determinations of their atomic structures led to breakthroughs in understanding the structure of the gold–ligand interface, the interactions in the ligand layer, the organic surface, and the principles that govern their electronic structure and optical response16–21, This has inspired efforts towards nanoscale and macroscale assemblies of gold22–26, silver27,28, and copper clusters29–32 that might have a variety of applications in the fields of (bio)chemical sensing33–35, and optoelectronics29–32,36–38.Up to date, cluster self-assembly has been demonstrated mostly for three-dimensional (metal-organic framework-like or spherical capsids) and 2D (film) systems, while successful realizations of anisotropic one-dimensional (1D) cluster materials have been scarce until very recently26,38. Here, we demonstrate a solvent-mediated assembly of 1-ethynyladamantane (A-Adm) protected, intermetallic 34-atom Au–Ag clusters (hereafter labeled as (AuAg)34 and A-Adm noted as ligand L) into cluster polymers that feature direct metal–metal bonds connecting the clusters into 1D chains. This material makes macroscopic single crystals with anisotropic semiconducting properties. Density functional theory (DFT) calculations of the crystal predict semiconductor-type electronic band structure with a band gap of about 1.3 eV. This material is amenable for testing as a device component for a field-effect transistor (FET) showing p-type behavior, highly anisotropic electrical conductivity with about 1800-fold difference along/across the polymer chain, hole mobility of ≈0.02 cm2 V-1 s−1, and an ON/OFF current ratio up to 4000 of the FET device.ResultsSynthesis of the (AuAg)34 clusters and observation of polymerizationThe syntheses of (AuAg)34 and (AuAg)34n nanoclusters in different solvents are summarized in Fig. 1. In a typical synthesis, AuSMe2Cl was first dissolved in a mixture solvent of chloroform and methanol. A-Adm was then added to the solution and stirred for 20 min. Subsequently, Ag acetate and tert-butylamineborane were added to the mixture under vigorous stirring. The reaction was aged for 10 h at room temperature. During the period, the color of the reaction mixture gradually changed from yellow to dark red. The solution was then centrifuged for 2 min at 10,000 r min–1 to give a dark red solution. After being stored at 25 oC, black block crystals were formed in ca. 7 days. When dichloromethane was used in otherwise similar synthesis, subsequent crystallization yielded single crystals containing the cluster polymers. Furthermore, we observed that when the crystals of the monomeric (AuAg)34 were re-dissolved in a mixture of methanol and CH2Cl2, crystals containing the cluster polymers were obtained.Fig. 1Schematics of solvent-mediated self-assembly of the cluster polymers.Depending of the solvent, either monomeric clusters or cluster polymers are formed. Colors: golden and green, Au/Ag; gray, C. All hydrogen atoms are omitted for clarity.It is intriguing to observe such dramatic effect in our synthesis caused by an apparently minute change in the synthesis conditions by two closely related solvents. On the other hand, it is well documented that solvent can play a pivotal role in the synthesis of metal nanoclusters39,40. Changing the chemical environment can drive structural transformation of Au/Ag nanoclusters affecting their physico-chemical properties41–46.Atomic structure of the building block (AuAg)34 and the cluster polymerThe total structures of (AuAg)34 and the cluster polymer (AuAg)34n were determined by single-crystal X-ray diffraction. Both of (AuAg)34 and (AuAg)34n crystallize into C2/c space group (crystal data shown in Supplementary Table 1). Similar to the cases of thiolate-protected (AuM)38 (M = Ag or Cu) nanoclusters47–50, positional disorders of Au and Ag atoms also occur in the kernels of (AuAg)34 and (AuAg)34n. As shown in Supplementary Tables 2 and 3, based upon least-squares refinement of the X-ray data, the compositions of (AuAg)34 and (AuAg)34n are Au21.3Ag12.7L20 and [Au21.4Ag12.6L20]n, respectively. In addition, we have used inductively coupled plasma mass spectrometric (ICP-MS) and energy-dispersive X-ray spectroscopic (EDS) to get insight into their metallic distributions. The X-ray results are confirmed by independent compositional assignments made using of ICP-MS (1:1.69 and 1:1.73 molar ratio of Ag/Au for total 34 metallic atoms in (AuAg)34 and (AuAg)34n respectively) and EDS (1:1.67 and 1:1.74 molar ratio of Ag/Au for total 34 metallic atoms in (AuAg)34 and (AuAg)34n, respectively) (Supplementary Figs. 1 and 2; and Supplementary Table 4).The metal frameworks of (AuAg)34 and (AuAg)34n are described in Fig. 2 and Supplementary Figs. 3, 4, 5a, 5b. They are very similar, differing only by the decoration of the outermost three atoms, namely, AuAg2 (Fig. 3). The anatomy of the metal structure of (AuAg)34n is shown in Fig. 2. It is interesting to note that (AuAg)34 and (AuAg)34n share the same (AuAg)31L18 unit (Fig. 2 and Supplementary Figs. 1 and 5). Specifically, (AuAg)31 can be described as bi-centered-icosahedra sharing a fusion M3 face (13 × 2–3 = 23), capped with a hexagonal ring of exterior shell Au6 in the “equatorial” plane of the shared fusion M3 face and two apical Ag atoms (Fig. 2). The metal site Au/Ag disorders, as obtained by least-squares refinement of the occupancies, are indicated in Supplementary Figs. 1 and 5b and Supplementary Tables 2 and 3. Finally, the exterior metal shell comprises an Ag–Au–Ag unit (Fig. 3 and Supplementary Figs. 4b, c and 5b) with one Ag atom bridging one of the nonbonding edges of the “equatorial” hexagonal ring of exterior shell Au6 (Fig. 3). The Ag–Au–Ag unit is bent in the monomeric (AuAg)34 (Fig. 3a and Supplementary Figs. 4b and 5b) but linear in the polymeric (AuAg)34n (Fig. 3b and Supplementary Fig. 4c), which holds the key to the structural transformation from (AuAg)34 to (AuAg)34n (vide infra). The Ag–Au–Ag unit in the monomeric nanocluster is disorder because of the symmetry of the crystal structure (Supplementary Fig. 5b). For the three possibilities of the nearest adjacent nanoclusters in the real unit cell, the nearest Au/Ag–Au/Ag distances should be 5.348, 6.814, and 10.241 Å, respectively (Supplementary Fig. 5c–e). The Au atom in the Ag–Au–Ag linkage of the polymeric (AuAg)34n nanocluster is disordered with 5 possible positions arranged in a same plane (Supplementary Fig. 6). In comparison, because of their π-type bonding with Ag, the disordering of the associated ligands is not significant.Fig. 2Anatomy of the core structure of (AuAg)34 and (AuAg)34n.Colors: golden and green, Au/Ag; gray, C. All hydrogen atoms are omitted for clarity.Fig. 3Different “Ag–L–Au–L–Ag” units of (AuAg)34 and (AuAg)34n.The Au–Au distances in the distorted Au6 hexagon and Ag–Ag distance in the “Ag–Au–Ag” unit of a
(AuAg)34 and b
(AuAg)34n. Colors: golden and green, Au/Ag; gray, C. All hydrogen atoms are omitted for clarity.A detailed structural analysis of the bi-icosahedral (AuAg)23 cores in both (AuAg)34 and (AuAg)34n is shown in Supplementary Fig. 7. Metal–metal bond lengths between the central atom and the first-shell M (M = Au or Ag) atoms range from 2.71 to 2.83 Å, whereas the M···M (M = Au or Ag) distances in the icosahedral shells (including M atoms in the shared triangular face) are in the range of 2.77–3.12 Å. Between the two M6 (M = Au or Ag) corrugated planes, three nearest M···M (M = Au or Ag) pairs are significantly longer (average 3.28 Å) and are evenly distributed at the waist of the (AuAg)23 rod. Bond lengths between each exterior Ag atom and the Au3 (100% Au occupancy) faces range from 2.95 to 3.02 Å (2.99 Å (av) for (AuAg)34n and 2.98 Å (av) for (AuAg)34).In the plane defined by the shared M3 triangle, six exterior shell Au (100% Au occupancy) atoms are bonded, respectively, to the three M atoms of the fusion plane in a radial fashion, with an average M···M distance of 3.08 Å. These six exterior Au atoms form a distorted hexagon with alternating long and short nonbonding Au···Au distances (5.24 (av) and 4.00 (av) Å for (AuAg)34n; 5.09 (av) and 4.18 (av) Å for (AuAg)34). The latter (short) distances play a key role in the (AuAg)34 to (AuAg)34n transformation, as we shall see next (Fig. 3).There are also 18 ligands on the surface of (AuAg)31 metal framework. Every Au atom of the hexagonal arrangement Au6 shell is linearly bound by two ligands (Supplementary Fig. 4b). Each “apical” Ag atom is coordinated to the C–C π-bonds of three ligands (Supplementary Fig. 4a).The key difference between (AuAg)34 and (AuAg)34n lies in the connectivity of the cluster building blocks of (AuAg)31L18 via the “linker hinge” unit “Ag(a)–L–Au–L–Ag(b)” highlighted in Fig. 3. In the monomer (AuAg)34 (top), one end of the linker, Ag(a), is anchored on a short edge of the exterior (nonbonding) Au hexagon of a given cluster (left). The other end, Ag(b), instead, bends back to bind to an adjacent Au atom of the same cluster unit, giving rise to an Ag(a)–Au–Ag(b) angle of 84.50 degree. Upon transformation to the polymer form, the Ag(a)–Au–Ag(b) angle of the linker is straightened to 180.00 degree, with the Ag(b) atom being inserted into the short Au···Au edge of the exterior (nonbonding) Au hexagon of the nearest adjacent cluster unit (right). This chain reaction propagates to produce the polymer (AuAg)34n. As a result, each (AuAg)31L18 cluster unit in the polymer has two linear linker hinges attached to it as shown in Fig. 4). In the single crystal, the polymeric chains of (AuAg)34 clusters are stacked in the unit cell as shown in Fig. 5, indicative of highly anisotropic crystal properties as will be discussed later.Fig. 4Atomic structure of the cluster polymer.a Total structure viewed approximately orthogonal to the c-axis (parallel direction). b Space-filling view of the metal atoms (viewed from the top direction in a). Colors: golden and green, Au/Ag; gray, C. All hydrogen atoms are omitted for clarity.Fig. 5Packing of cluster polymer chains in the monoclinic unit cell.a Packing inside a unit cell. The rods indicate the polymerization direction (c-axis). b Zoomed-up image of the marked area in a. The pink dots denote the centroids of the A-Adm groups. The average center-to-center distance and H···H contact distances between adamantyl moieties are 6.7 and 2.3 Å, respectively. Only weak van der Waals interactions are observed between the metallic chains along crystallographic a- or b-axes. Colors: golden and green, Au/Ag; gray, C. All hydrogen atoms are omitted for clarity.Contrasting the two structures, given the exact same chemical formulations and virtually identical (AuAg)31L18 building blocks, the driving force for the monomer to polymer transformation can be attributed to an increase of a single Ag–Au bond (viz. two in the monomer and three in the polymer as shown in Fig. 3, top, two Ag(b)–Au bonds of 2.86 Å (av) and bottom, three Ag(b)–Au bonds of 2.93 Å (av)). Of the three short Au···Au edges of the exterior (nonbonding) Au hexagon of the (AuAg)31L18 building block, only one is occupied by Ag(a) in the monomer but two are occupied by Ag(a) and Ag(b) (a and b are equivalent) in the polymer.Optical properties and electronic structureThe solid-state optical diffuse reflectance spectra of (AuAg)34 clusters both in monomeric and polymeric forms were measured (Fig. 6 and Supplementary Fig. 8). The spectra show two relatively weak peaks/shoulders in the range of 2.4–2.7 eV and a rather long tail towards the optical gap, which can be estimated from the extrapolated low-intensity absorptions as about 1.4 eV both for the monomeric (AuAg)34 and the cluster polymer.Fig. 6UV–Vis absorption spectra.Absorption spectra of both monomeric and polymeric crystals. Computed spectrum for one of the isolated (AuAg)34L20 model clusters (see details of the models in the Methods) is shown as well. The intensity of the computed spectrum is scaled to the experimental peak at 2.7 eV of the monomeric cluster but no shifts have been applied in the energy axis.Optical properties and the electronic structure of the isolated (AuAg)34L20 cluster and the cluster polymer crystal were studied by extensive DFT computations (technical details discussed in Methods). Since the experimental crystal structure contained positional Au/Ag disorder in 18 out of the 34 metal sites, we created four models for the isolated cluster: two models spanning both extremes of Au/Ag occupation (Au19Ag15L20 and Au26Ag8L20) and two models for the composition Au21Ag13L20, which is close to the measured Au21.3Ag12.7L20, and where the disordered sites were occupied randomly by gold or silver weighted by the experimental probability (Supplementary Table 2). We studied the electronic ground state structure of all these four model clusters by using the DFT implementation in the GPAW code51 and computed the optical absorption spectra by using the linear-response (LR) formalism of the time-dependent DFT (LR-TDDFT)52. In all calculations, the electron–electron interactions were treated by using the Perdew–Burke–Ernzerhof (PBE) exchange-correlation potential53, which we have found to be an acceptable compromise between accuracy and computational efficiency in numerous previous studies of ligand-stabilized noble metal clusters. The electronic structure was analyzed for both non-relaxed (all atom coordinates taken directly from the crystal data) and computationally relaxed clusters. The atomic coordinates of the DFT-relaxed models 1–4 described above are given as Supplementary Datasets 1–4.The computed energy gaps between the highest occupied and lowest unoccupied molecular orbitals (HOMO-LUMO gaps) are consistently in the range of 1.31–1.40 eV for non-relaxed clusters and 1.28–1.46 eV for relaxed ones. LR-TDDFT calculations gave optical spectra that are compared to the experimental data in Supplementary Fig. 9. Comparing the spectra in energy scale one sees that all the four models yield absorption spectra that reproduce the shape of the experimental spectrum rather well up to about 3 eV, albeit a slight underestimation of the optical gap. Particularly one of the models where Au/Ag occupations were drawn randomly from the experimental distribution (“model 4”) reproduces the full shape of the experimental spectrum extremely well up to 3 eV, including the two peaks/shoulders around 2.4–2.7 eV (comparison shown also in Fig. 6). Shapes of the spectra for relaxed clusters change slightly, most likely due to the fact that the PBE overestimates the metal–metal bonds by 2–4 %.The non-relaxed cluster model 4 was then used to build the model for the cluster polymer crystal, with four clusters in the periodic crystal unit cell (Supplementary Fig. 10, and the coordinates given as Supplementary Dataset 5) and the electronic structure sampled by 4 × 4 × 4 Monkhorst–Pack k-point mesh54 in the reciprocal space. The electronic density of states shown in Fig. 7 features broad valence (E < 0) and conduction (E > 0) bands with a fine structure. The apparent band gap, determined as the peak separation closest to the gap, is about 1.3 eV. This gap is comparable to the HOMO-LUMO and optical gap for the isolated cluster. The magnitude of the band gap indicates that the cluster polymer crystal should behave as a semiconductor-type material.Fig. 7DFT-computed electronic density of states (DOS) of the cluster polymer crystal.The cluster model 4 was used (as in Fig. 6) to build the periodic crystal, and the integration over the Brilloin zone was done in a 4 × 4 × 4 Monkhorst–Pack k-point mesh. The band gap is centered around zero.To gain qualitative insight into the characteristics of the valence and conduction bands, we formed a summed density from the occupied/unoccupied electron states forming the peaks closest to the band gap in Fig. 7 (top of the valence band at about –0.9 eV, and bottom of the conduction band at about +0.75 eV). This data is visualized in Supplementary Fig. 11 showing that appreciable density overlap between the neighboring clusters in the polymer exists only in the direction of the polymer axis. This implies that the electrical conductivity of the crystal should be anisotropic.Highly anisotropic p-type semiconductivity of the polymer crystalAs Fig. 5 indicated, the (AuAg)34n clusters form polymeric chains parallel to the c-axis of the single crystal, and these chains are separated in a- and b-directions by the bulky A-Adm ligands (Fig. 5b). This structural anisotropy together with the DFT-predicted electronic structure (discussed above) of the polymer crystal gives reasons to anticipate highly anisotropic (semi)conductivity.To prove this, FET devices were fabricated to measure the direction-dependent conductivity of the polymer crystals (representative crystals shown in Supplementary Figs. 12 and 13) as discussed in Methods. The schematics of the single-crystal nanocluster FET chip are depicted in Fig. 8a. Detailed fabrication processes are described in Supplementary Figs. 14 and 15. In brief, four-terminal electrode sets (5 nm Cr/50 nm Au) were deposited on silicon wafer with 300 nm silicon dioxide using photolithography, followed by e-beam evaporation. Then a single crystal of cluster polymer was transferred onto the pre-patterned gold electrodes, with the device channel aligned with the a- or c-crystallographic axis, and by using silver paste to create the electrical contacts between the crystal and the electrode. The highly doped silicon wafer serves as a global backgate for the devices.Fig. 8Electrical transport properties of the cluster polymer crystals.a The structure of the polymer crystal FET. b
I–V plot of the polymer crystal along a-axis and c-axis, respectively, with the range of corresponding conductivity values shown in the inset. c Transistor transfer (VSD = 8 to –16 V in –4 V steps) and d output characteristics (VG = 16 to –16 V in –4 V steps) measured along c-axis of the polymer crystal. Channel length, width = 80, 50 μm.More than ten devices were studied in this work, each device comprising two orthogonal electrode pairs along the a- and c-crystallographic axes. The electrical conductivity was determined from the slope of the linear I–V curve. All the field-effect transistors showed the anisotropy of electrical conductivity, as illustrated in Fig. 8b. The averaged electrical conductivity along the c-crystallographic axis of the crystal at room temperature and relative humidity of 56% is 1.49 × 10-5 S m−1, which is 1800 times of the electrical conductivity along the a-crystallographic axis (values averaged over six single crystals). Six measurements, plus the average and standard deviation are shown in Supplementary Table 5. Blank controls without crystal transferred was also measured, showing only instrument noise levels (Supplementary Fig. 16), which means that the conductivity is contributed by the crystal itself. Furthermore, the averaged conductivity of monomeric crystals is around 6.27 × 10-8 S m−1, near the conductivity along a-axis and much lower than the conductivity along c-axis of polymeric crystal (Supplementary Fig. 17). These results confirm that the direct linking of the clusters by the –Ag–Au–Ag– chains are beneficial for carrier transport.We then probed the semiconductor properties of single crystal along c-crystallographic axis. The transfer and output characteristic curves are shown in Fig. 8c, d. The transfer curves show that at negative gate voltage (VG), the source-drain currents increase with more negative VG, demonstrating p-type field effect. This indicates a hole conduction mechanism. The ON/OFF current ratio is around 4000 and the charge carrier mobility reaches 2.46 × 10−2 cm2 V−1 s−1 at source-drain voltage (VSD) = –16 V calculated using the standard method in the unsaturation regime55. The exponential behavior in the output curves can be attributed to Schottky barrier between the electrode and the polymer crystal. Moreover, the transistor characteristic is reproducible as shown in the supporting information (Supplementary Fig. 18). As a comparison, recently reported photoconductive two-dimensional (2D) films of phosphine-thiolate-stabilized Au25 clusters36 showed ON/OFF ratio of about 50 000 for VSD = 6 V, charge carrier mobility approaching 10−5 cm2 V−1 s−1 at VSD = 20 V, and n-type field effect. The mobility of our polymeric crystal of nanoclusters is in the range of traditional p-type single-crystal organic semiconductors56 and close to the mobility of supercrystal of CdSe Quantum Dots (Supplementary Table 6)57,58. We further note that the conductivity (1.49 × 10–5 S m−1, see above) of our crystals in the c-crystallographic axis is one to three orders of magnitude higher than the values reported recently for thiolate-stabilized 1D assemblies of Au21 clusters26 where 1D “nanofibrils” of the clusters were formed by modulating the weak interactions between the ligand layers of the clusters. These comparisons indicate that the conductivity and charge carrier mobility is increased by several orders of magnitude in our macroscopic cluster-based materials via direct linking of the clusters by the –Ag–Au–Ag– chains in the cluster polymer crystal.DiscussionWe have demonstrated that the use of bulky adamantane alkyne facilitates the solvent-mediated self-assembly of atomically precise intermetallic (AuAg)34 nanoclusters into 1D cluster polymers that grow into macroscopic crystals with up to hundreds of microns in length, making the single crystals readily available for materials characterizations. The materials have highly anisotropic structure where the neighboring clusters are directly connected by –Ag–Au–Ag– linkages in the c-direction of the crystals but have insulating adamantane layers between the polymeric chains. We have tested the performance of this material as a component of a FET device and found that the device properties show about 1800-fold anisotropy in the conductance properties along and across the polymer chain, ON/OFF ratio of about 4000, p-type field effect, and hole carrier mobility of up to 2.46 × 10−2 cm2 V−1 s−1. This work offers a new insight into the structural factors for controlling the electron transport in assemblies of clusters and nanoparticles and enhances our abilities to create new hierarchical nanoparticle assemblies with desired structures and properties.MethodsMaterialsAll materials were purchased from Alfa Aesar without further purification. Silver acetate (AgOAc, purity 98%), Borane-tert-butylamine complex [(CH3)3CNH2•BH3, purity 97%], 1-Ethynyladamantane (C12H16, purity 95%), Sodium methoxide (CH3ONa, powder, purity 99%), Potassium methoxide (CH3OK, powder, purity 99%), dichloromethane (CH2Cl2, analytical grade) and methanol (CH3OH, analytical grade) were purchased from Sinopharm Chemical Reagent Co. Ltd. (Shanghai, China). Leitsilber 200 Silver Paint was puchased from Tedpella. All reagents were used as received without further purification. The water used in all experiments was ultrapure. All reagents were used as received without further purification. AuSMe2Cl was prepared according to literature methods59.Physical measurementsUV–Vis absorption spectra were recorded on a Varian Cary 5000 spectrophotometer. Energy-dispersive X-ray spectroscopy (EDS) was performed on a scanning electron microscopy (SEM, Hitachi S4800) equipped with energy-dispersive X-ray spectroscopy operated at 15 kV. The compositions of (AuAg)34n and (AuAg)34 were determined by inductively coupled plasma mass spectrometry (ICP-MS, Agilent 7700).Single-crystal analysisThe diffraction data were collected on an Agilent SuperNova X-Ray single-crystal diffractometer using Cu Kα (λ = 1.54184 Å) and Mo Kα (λ = 0.71073 Å) micro-focus X-ray sources at 100 K. The data were processed using CrysAlisPro. The structure was solved and refined using Full-matrix least-squares based on F2 with programs SHELXT and SHELXL60 within OLEX261.Synthesis of (AuAg)34nSix milligrams of AuSMe2Cl was first dissolved in the mixture solution of dichloromethane and methanol. In all, 3.5 mg 1-Ethynyladamantane and 5 mg sodium methoxide were then added to the solution and stirred for 20 min. Five milligrams of AgCH3COO was added and stirred for 5 min. The reducing agent, tert-butylamineborane (3.6 mg), was added to the mixture under vigorous stirring. The reaction was aged for 12 h at room temperature. The solution was centrifuged for 4 min at 10,000 r min–1 . The brown supernatant was subjected to natural evaporation in the dark. Red block crystals were obtained within 1 week in a yield of ~35% (based on Au).Synthesis of (AuAg)34Six milligrams of AuSMe2Cl was first dissolved in the mixture solution of chloroform and methanol. In alll, 3.5 mg 1-Ethynyladamantane and 5 mg sodium methoxide were then added to the solution and stirred for 20 min. Five millihrams of AgCH3COO was added and stirred for 5 min. The reducing agent, tert-butylamineborane (3.6 mg), was added to the mixture under vigorous stirring. The reaction was aged for 12 h at room temperature. The solution was centrifuged for 4 min at 10,000 r min–1. The brown supernatant was subjected to natural evaporation in the dark. Red block crystals were obtained within 1 week in a yield of ~25% (based on Au).DFT computationsThe density functional theory (DFT) computations were obtained with the GGA-PBE exchange-correlation functional53 employing a uniform real-space grid for representing the wave functions as coded in the GPAW software51. The grid-spacing was 0.2 Å for all the calculations. The optical spectra for the monomers were calculated with the linear-response time-dependent DFT (LR-TDDFT) as implemented in GPAW52.The effect of the ratio of gold to silver atoms was tested using four different compositions of the metal core for the monomers: Au19Ag15, Au26Ag8 and two different isomers of Au21Ag13. The atomic coordinates with fractional occupancies were assigned employing the probabilities from Supplementary Fig. 1. For Au19Ag15, the element with the larger probability was always chosen (model 1). For Au26Ag8, the opposite choice was made (model 2). The metal atoms for the trial clusters with a core of Au21Ag13 were randomly assigned using the probabilities as weights (model 3, 4). The calculation box was chosen for each system so that 6 Å of vacuum was left on each side. All four structures were also relaxed, requiring the final structure to have forces of <0.05 eV Å–1 on each atom.For all the monomers, the UV–Vis spectrum was calculated for the coordinates taken directly from the crystal structure, as well as the relaxed structure, and these spectra were compared to the measured one. The electron-hole excitations with an energy up to 4.75 eV (≈260 nm) were included in the calculations. A Gaussian broadening of 0.1 eV for individual oscillator strengths was used to plot the spectra.For the polymer, the model cluster 4 was chosen, since it had the best fit between the experimental and computational UV–Vis spectrum for the monomer. A calculation with periodic boundary conditions was performed using the positions and the unit cell determined from the crystal structure. The monoclinic unit cell has four clusters inside. Monkhorst–Pack sampling54 with 4 x 4 x 4 k-points was used for the Brillouin-zone. The total density of states (DOS) was calculated using a Gaussian broadening of 0.1 eV.Fabrication of single-crystal nanoclusters-based transistorsThe fabrication procedure is illustrated in Supplementary Fig. 14. First, a layer of photoresist (AZ5214E) was spin-coated on 300 nm SiO2 covered Si wafer and metal electrodes (5 nm Cr/50 nm Au) were patterned by photolithography, followed with electron beam evaporation and lift-off process. Then a single crystal of cluster polymers was carefully transferred onto contact with the pre-patterned electrodes, served as source-drain electrode. The heavily doped silicon wafer serves as backgate electrode. Conductive silver epoxy was introduced to glue them ensuring a better contact. After the conductive silver epoxy was completely dry for about 2 h, we could conduct the electrical properties measurement.Electrical properties measurementThe electrical conductivity of single crystal along different crystallographic axes was measured at room temperature at relative humidity of 56% using a Keithley 4200-SCS source meter and a MPS150 manual probe station.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Data 1
Supplementary Data 2
Supplementary Data 3
Supplementary Data 4
Supplementary Data 5
Supplementary Data 6
Supplementary Data 7 | nature communications | [
"Article"
] | [
"Electronic properties and materials",
"Nanoparticles",
"Nanowires"
] | progress in synthesis structural discovery functionalization understanding of ligand-stabilized atom-precise metal nanoclusters opened opportunities for designing nanomaterials with tunable fabrication nanomaterials relies on self-assembly Self-assembly on supramolecular weak interactions linkers introduced metal since mid-1990’s realizations included gold nanoparticle assemblies particles linked by dithiols10 viral proteins as scaffolds13 Gold nanoclusters stabilized by thiolates15 served as prototype structural determinations led to breakthroughs in understanding structure gold–ligand interface interactions ligand layer organic surface electronic structure optical inspired efforts towards nanoscale macroscale assemblies of copper applications in (bio)chemical cluster self-assembly demonstrated for three-dimensional 2D systems anisotropic one-dimensional (1D) cluster materials scarce until solvent-mediated assembly of 1-ethynyladamantane (A-Adm) protected intermetallic 34-atom Au–Ag clusters into cluster polymers direct metal–metal bonds connecting into 1D chainsmaterial macroscopic crystals anisotropic semiconducting properties calculations predict semiconductor-type electronic band structure band gap 1.3 eV material amenable field-effect transistor (FET) p-type behavior anisotropic electrical conductivity 1800-fold difference polymer hole mobility ≈0.02 cm2 V-1 s−1 ON/OFF current ratio up 4000 (AuAg)34 clusters)34 nanoclusters solvents summarized Fig. 1. AuSMe2Cl dissolved chloroform methanol A-Adm added stirred 20 min Ag acetate tert-butylamineborane added reaction aged 10 h temperature color changed yellow to dark red centrifuged 2 min 10,000 r min–1 dark red solution stored 25 oC black block crystals formed 7 days dichloromethane used crystallization yielded crystals cluster polymers crystals (AuAg)34 re-dissolved methanol CH2Cl2 crystals cluster polymers obtained. 1Schematics solvent-mediated self-assembly cluster polymers solvent monomeric clusters cluster polymers formed Colors golden green Au/Ag gray C hydrogen atoms omitted clarityintriguing observe dramatic effect synthesis minute change solvents solvent pivotal synthesis metal nanoclusters39 Changing chemical environment structural transformation Au/Ag nanoclusters affecting-chemical.Atomic structure building block (AuAg)34 cluster structures (AuAg)34 cluster polymer (AuAg)34n determined by single-crystal X-ray diffraction crystallize into C2/c space group data Supplementary Table 1) Similar thiolate-protected (AuM)38 positional disorders Au Ag atoms occur in kernels (AuAg)34)34n Supplementary Tables 2 3 compositions (AuAg)34)34n are Au21.3Ag12.7L20 [Au21.4Ag12.6L20]n used inductively coupled plasma mass spectrometric) energy-dispersive X-ray spectroscopic (EDS) metallic distributions X-ray results confirmed by compositional assignments ICP-MS (1:1.69 1:1.73 molar ratio Ag/Au 34 EDS (1:1.67 1:1.74 molar ratio (Supplementary Figs. 1 2 Supplementary Table 4) metal frameworks of (AuAg)34)34n described in Fig.Supplementary Figs. 3 4 5a 5b similar differing decoration outermost three atoms AuAg2 (Fig. 3) anatomy metal structure (AuAg)34n Fig. 2. (AuAg)34 (AuAg)34n share (AuAg)31L18 unit (Fig. 2 1 5) (AuAg)31 bi-centered-icosahedra fusion M3 face (13 × 2–3 = hexagonal ring exterior Au6 two apical Ag atoms (Fig 2) metal site Au/Ag disorders Supplementary Figs. 1 5b Tables 2 3. exterior metal shell Ag–Au–Ag unit (Fig. 3 4b one Ag atom bridging hexagonal ring shell Au6 Ag–Au–Ag unit bent in monomeric (AuAg)34 (Fig. 3a 4b 5b linear in polymeric (AuAg)34n 3b key structural transformation from (AuAg)34 to (AuAg)34n Ag–Au–Ag unit monomeric nanocluster symmetry crystal structure Fig. 5b). nearest nanoclusters nearest Au/Ag–Au/Ag distances 5.348, 6.814 10.241 ÅAu atom Ag–Au–Ag linkage)34n nanocluster disordered 5 positions plane Fig. 6) π-type bonding Ag disordering ligands not significant.Fig. core structure (AuAg)34)34n.Colors golden green Au/Ag gray C hydrogen atoms omitted.Fig. “Ag–L–Au–L–Ag” units (AuAg)34)34n Au–Au distances distorted Au6 hexagon Ag–Ag distance “Ag–Au–Ag” unit)34n Colors golden green Au/Ag gray C hydrogen atoms omitted detailed structural analysis bi-icosahedral (AuAg)23 cores (AuAg)34)34n Supplementary Fig. 7. Metal–metal bond lengths central atom first-shell M Au Ag atoms 2.71 to 2.83 Å M···M Ag) distances icosahedral 2.77–3.12 Å three nearest M···M) pairs longer (average 3.28 Å) evenly distributed waist (AuAg)23 rod Bond lengths exterior Ag atom Au3 faces 2.95 to 3.02 Å (2.99)34n 2.98)34)plane M3 triangle six exterior Au atoms bonded to three M atoms fusion plane average M···M distance 3.08 Å six atoms form distorted hexagon alternating long short nonbonding Au···Au distances (5.24 4.00) Å for (AuAg)34n 5.09 4.18 Å (AuAg)34) distances (AuAg)34 to (AuAg)34n transformation 18 ligands (AuAg)31 metal framework Every Au atom bound by two ligands Each Ag atom coordinated to C–C π-bonds three ligands difference between (AuAg)34)34n connectivity cluster building blocks (AuAg)31L18 “linker hinge” unit “Ag(a)–L–Au–L–Ag(b)” Fig. 3. (AuAg)34 one end linker Ag(a), anchored edge Au hexagon other end Ag bends back adjacent Au atom Ag(a)–Au–Ag(b) angle 84.50 degree.transformation to polymer form Ag(a)–Au–Ag(b) angle linker straightened to 180.00 degree Ag(b) atom inserted into edge Au hexagon cluster unit chain reaction polymer (AuAg)34n each (AuAg)31L18 cluster unit has two linear linker hinges Fig. single crystal polymeric chains (AuAg)34 clusters stacked in unit cell Fig. 5 anisotropic crystal properties. 4Atomic structure cluster polymer orthogonal to c-axis metal atoms Colors golden green Au/Ag gray, C hydrogen atoms omitted. 5Packing cluster polymer chains in monoclinic unit cell rods polymerization direction (c Zoomed-up image pink dots centroids A-Adm groups average center-to-center distance H···H contact distances between adamantyl moieties 2.3 Å weak van der Waals interactions between metallic chains crystallographic a- b-axes Colors golden green/Ag gray, C hydrogen atoms omitted chemical formulations identical (AuAg)31L18 building blocks monomer to polymer transformation increase single Ag–Au bond two three polymertop two Ag(b)–Au bonds 2.86 Å bottom three Ag(b)–Au bonds 2.93 Å three short edges exterior hexagon (AuAg)31L18 building block one occupied Ag(a) monomer two Ag(a) Ag(b) polymer properties electronic solid-state optical diffuse reflectance spectra (AuAg)34 clusters monomeric polymeric forms measured (Fig. 6 Fig show weak peaks/shoulders 2.4–2.7 eV long optical gap 1.4 eV monomeric (AuAg)34 cluster polymer. 6UV–Vis absorption spectra monomeric polymeric crystals Computed spectrum isolated (AuAg)34L20 intensity scaled experimental peak 2.7 eV monomeric no shifts energy axis.Optical properties electronic structure isolated (AuAg)34L20 cluster polymer crystal studied DFT computations experimental crystal structure positional Au/Ag disorder 18 34 metal sites created four models cluster two extremes Au/Ag occupation (Au19Ag15L20 Au26Ag8L20) composition Au21Ag13L20 close measured Au21.3Ag12.disordered sites occupied randomly by gold or silver weighted experimental probability (Supplementary Table 2) studied electronic ground state structure four model clusters DFT GPAW computed optical absorption spectra linear-response formalism time-dependent DFT electron–electron interactions treated Perdew–Burke–Ernzerhof) exchange-correlation acceptable compromise accuracy computational efficiency ligand-stabilized metal electronic structure analyzed for non-relaxed computationally relaxed clusters atomic coordinates DFT-relaxed models 1–4 Supplementary Datasets 1–4 computed energy gaps between highest occupied lowest unoccupied molecular orbitals 1.31–1.40 eV for non-relaxed 1.28–1.46 eV for relaxed LR-TDDFT calculations optical spectra compared to experimental data Supplementary Fig. 9. four models absorption spectra reproduce shape experimental spectrum up to 3 eV underestimation optical gap Au/Ag (“model reproduces full shape spectrum up to 3 eV peaks/shoulders 2.4–2.7 eV Fig. 6)spectra relaxed clusters change PBE overestimates metal–metal bonds 2–4 non-relaxed cluster model 4 used cluster polymer crystal four clusters periodic crystal unit cell Fig. 10 coordinates Supplementary Dataset 5) electronic structure sampled 4 × × 4 Monkhorst–Pack k-point mesh54 reciprocal space electronic density Fig. 7 broad valence (E < 0) conduction (E > 0) bands fine structure band gap 1.3 eV comparable to HOMO-LUMO optical gap isolated cluster magnitude indicates cluster polymer crystal semiconductor-type material.Fig. 7DFT-computed electronic density states polymer crystal cluster model 4 used periodic crystal integration Brilloin zone 4 × 4 × 4 Monkhorst–Pack k-point mesh band gap centered around zero formed summed density from occupied/unoccupied electron states peaks closest band gap Fig. 7 valence –0.9 eV conduction +0.75 eV). Supplementary Fig. 11 density overlap clusters polymer axis electrical conductivity crystal anisotropic.Highly anisotropic p-type semiconductivity polymer(AuAg)34n clusters form polymeric chains parallel c-axis crystal separated by A-Adm ligands (Fig. structural anisotropy DFT-predicted electronic structure polymer crystal anisotropic (semi)conductivity FET devices fabricated measure direction-dependent conductivity polymer crystals Figs. 12 13 schematics single nanocluster FET chip Fig. 8a fabrication processes Figs. 14 15. four-terminal electrode sets (5 nm Cr/50 nm Au) deposited on silicon wafer 300 nm silicon dioxide photolithography e-beam evaporation single crystal polymer transferred pre-patterned gold electrodes channel aligned with a- c-crystallographic axis silver paste electrical contacts doped silicon wafer backgate. 8Electrical transport properties cluster polymer crystals structure polymer crystal FET I–V plot polymer crystal a c-axis conductivity values Transistor transfer 8 to –16 V output characteristics 16 to –16 V along c-axis Channel length 80, 50 μm ten devices studied two orthogonal electrode pairs a- c-crystallographic axeselectrical conductivity determined from I–V curve field-effect transistors showed anisotropy Fig. 8b averaged conductivity c-crystallographic axis at room temperature humidity 1.49 × 10-5 S m−1 1800 times a-crystallographic axis six measurements deviation Supplementary Table 5. Blank controls without crystal measured instrument noise levels conductivity contributed crystal averaged conductivity monomeric crystals 6.27 × 10-8 S m−1 near a-axis lower than c-axis polymeric crystal direct linking clusters –Ag–Au–Ag– chains beneficial for carrier transport probed semiconductor properties single crystal along c-crystallographic axis transfer output curves Fig. 8c d negative voltage source-drain currents increase p-type field effect hole conduction mechanism ON/OFF current ratio 4000 charge carrier mobility 2.46 × 10−2 cm2 V−1 s−1 at source-drain voltage = –16 V exponential behavior output curves Schottky barrier between electrode polymer crystal transistor characteristic reproducible supporting informationphotoconductive films-thiolate-stabilized Au25 clusters36 showed ON/OFF ratio 50 000 VSD = 6 V charge carrier mobility 10−5 cm2 V−1 s−1 at VSD = 20 V n-type field effect mobility polymeric crystal nanoclusters traditional p-type single-crystal organic semiconductors56 close supercrystal CdSe Quantum Dots conductivity (1.49 × 10–5 S m−1 crystals c-crystallographic axis to three orders higher than thiolate-stabilized 1D Au21 clusters26 interactions ligand layers conductivity charge carrier mobility increased in macroscopic cluster-based materials linking clusters –Ag–Au–Ag– chains crystal bulky adamantane alkyne facilitates self precise intermetallic (AuAg)34 nanoclusters into 1D cluster polymers macroscopic crystals hundreds microns crystals available for materials characterizations materials anisotropic structure clusters connected by –Ag–Au–Ag– linkages insulating adamantane layers between chainstested performance material FET device properties show 1800-fold anisotropy conductance polymer chain ON/OFF ratio 4000 p-type field effect hole carrier mobility 2.46 × 10−2 cm2 V−1 s−1 work offers insight structural factors electron transport nanoparticles enhances nanoparticle assemblies materials purchased Alfa Aesar purification Silver acetate (AgOAc Borane-tert-butylamine complex(CH3)3CNH2•BH3 purity 97% 1-Ethynyladamantane (C12H16 Sodium methoxide (CH3ONa Potassium methoxide (CH3OK dichloromethane (CH2Cl2 methanol (CH3OH purchased Sinopharm Chemical Reagent Co Leitsilber 200 Silver Paint Tedpella reagents purification water ultrapure reagents AuSMe2Cl prepared literature measurementsUV–Vis absorption spectra recorded Varian Cary 5000 spectrophotometer Energy-dispersive X-ray spectroscopy scanning electron microscopy S4800 15 kVcompositions (AuAg)34n determined plasma mass spectrometry Agilent 7700).Single-crystal diffraction data collected Agilent SuperNova X-Ray single-crystal diffractometer Cu Kα 1.54184 Mo Kα micro-focus X-ray sources 100 K data processed CrysAlisPro structure solved refined Full-matrix least-squares programs SHELXT SHELXL60 OLEX261.Synthesis (AuAg)34nSix milligrams AuSMe2Cl dissolved dichloromethane methanol 3.5 mg 1-Ethynyladamantane 5 mg sodium methoxide added stirred 20 min Five milligrams AgCH3COO added 5 min agent tert-butylamineborane (3.6 added aged 12 h centrifuged 4 min 10,000 r brown supernatant evaporation Red block crystals obtained 1 week yield ~35% (AuAg)34Six milligrams AuSMe2Cl dissolved chloroform methanol 3.5 mg 1-Ethynyladamantane 5 mg sodium methoxide AgCH3COO 5 min reducing agent tert-butylamineborane (3.6 added aged 12 h solution centrifuged 4 min 10,000 r min–1brown supernatant subjected evaporation dark Red block crystals obtained 1 week yield ~25% Au).DFT density functional theory computations obtained GGA-PBE exchange-correlation uniform real-space grid wave functions GPAW grid-spacing 0.2 Å optical spectra monomers calculated linear-response time-dependent DFT) GPAW52 ratio gold to silver atoms tested four compositions metal core Au19Ag15 Au26Ag8 two Au21Ag13 atomic coordinates occupancies assigned probabilities Supplementary Fig. 1. Au19Ag15 larger probability chosen Au26Ag8 opposite 2) metal atoms trial clusters core Au21Ag13 randomly assigned probabilities 3 calculation box chosen system 6 Å vacuum left each side four structures relaxed final forces <0.05 eV Å–1 each atom UV–Vis spectrum calculated coordinates crystal structure relaxed structure spectra compared measured one electron-hole excitations energy up to 4.75 eV (≈260 nm) included calculations Gaussian broadening 0.1 eV oscillator strengths spectra polymer model cluster 4 chosen best fit experimental computational UV–Vis spectrum calculation periodic boundary conditions positions unit cell crystal structuremonoclinic unit cell four clusters Monkhorst–Pack sampling54 4 x k-points Brillouin-zone total density states (DOS) calculated Gaussian broadening 0.1 eV.Fabrication single-crystal nanoclusters procedure Supplementary Fig. 14. photoresist (AZ5214E) spin-coated 300 nm SiO2 Si wafer metal electrodes (5 nm Cr/50 nm Au patterned photolithography electron beam evaporation lift-off process single crystal cluster polymers transferred pre-patterned electrodes source-drain electrode doped silicon wafer backgate electrode Conductive silver epoxy better contact epoxy 2 h electrical properties measurement conductivity single crystal measured room temperature relative humidity Keithley 4200-SCS source meter MPS150 manual probe station.Supplementary information Review Additional Supplementary Files | 51.1 | 0.49402 |
10.1038/s41467-020-18584-5 | PMC7524734 | Living cells and tissues experience various complex modes of forces but how different force modes impact gene expression is elusive. Here authors apply forces via magnetic beads to integrins on a cell surface and observe force-mode dependent chromatin stretching and gene upregulation in cells and identify underlying mechanisms. | Living cells and tissues experience various complex modes of forces that are important in physiology and disease. However, how different force modes impact gene expression is elusive. Here we apply local forces of different modes via a magnetic bead bound to the integrins on a cell and quantified cell stiffness, chromatin deformation, and DHFR (dihydrofolate reductase) gene transcription. In-plane stresses result in lower cell stiffness than out-of-plane stresses that lead to bead rolling along the cell long axis (i.e., alignment of actin stress fibers) or at different angles (90° or 45°). However, chromatin stretching and ensuing DHFR gene upregulation by the in-plane mode are similar to those induced by the 45° stress mode. Disrupting stress fibers abolishes differences in cell stiffness, chromatin stretching, and DHFR gene upregulation under different force modes and inhibiting myosin II decreases cell stiffness, chromatin deformation, and gene upregulation. Theoretical modeling using discrete anisotropic stress fibers recapitulates experimental results and reveals underlying mechanisms of force-mode dependence. Our findings suggest that forces impact biological responses of living cells such as gene transcription via previously underappreciated means. | IntroductionIt is well established that living cells and tissues respond to mechanical force stimulation such as shear stresses in the blood or interstitial flow1,2, tractional forces at cell–matrix contacts3, and cell–cell contacts4. Mechanical properties of the living cells and tissues are also known to be critical in regulating their biological responses in physiology and diseases5–8. However, how different modes of forces impact gene expression is elusive. For example, while it is well known that shear stresses at the cell apical surface produce different signals and cellular effects than do stretching forces at the base or sides of endothelial cells, as vessel diameter changes during dilation or constriction9,10, the underlying mechanisms of how gene transcription is altered by various force modes remain unclear. There are only a few well characterized technologies that can apply controlled local mechanical forces to a single living cell. One such method is the atomic force microscopy (AFM) that in general applies an indentation force on the apical surface of the cell11. Another method is the optical tweezer that uses a focused laser light to trap a micrometer-sized particle to apply a lateral force on the cell apical surface12. The third method is the method of magnetic twisting cytometry (MTC) or magnetic gradient pullers. In magnetic gradient pullers, an electromagnetic inhomogeneous field is generated at the tip of the tweezer to pull on a magnetic bead attached to the apical surface of the cell13. In MTC, a ferromagnetic bead is magnetized with a strong magnetic field pulse in the horizontal direction (e.g., along Y-axis) and then a weak homogenous magnetic field is applied in the vertical direction (along Z-axis) to rotate the bead about the X-axis14,15. This early version of the MTC can only rotate the magnetic bead about one axis—the X-axis and is thus called one dimensional (1D) MTC. Because of the technical difficulty to compare mechanical responses of the living cell to various mechanical probes16, it is not clear how surface force modes influence cellular responses and functions. Here we describe a strategy of using the three-dimensional MTC (3D MTC)17,18 to apply forces in any desired directions to the same living cell by rotating the magnetic bead about any axes. We find that for a given stress amplitude, a living cell responds to a local in-plane stress differently from responding to a local out-of-plane complex stress by stretching chromatin and upregulating gene transcription to different levels. Experimental results in living cells and theoretical modeling analyses using discrete anisotropic elements reveal that stress fiber anisotropy determines force-mode dependent cell stiffness and chromatin stretching and thus regulates gene upregulation.ResultsA strategy of applying local in-plane stressThe 1D MTC generated a bead rotation that was out-of-the X–Y plane and thus generated a complex stress on the cell surface: as the bead rotated in the Y–Z plane, the bead edge that moved upward stretched the cell membrane and the bead edge that moved downward compressed the cell membrane (Fig. 1a, b). In contrast, the 3D MTC could magnetize a magnetic bead in any desired direction17,18 (X, Y, or Z) and applied a homogeneous twisting field in any desired direction (X, Y, or Z) (Fig. 1c). When a magnetic bead was magnetized in Z direction and then twisted in the Y direction or in the X direction, an out-of-plane complex stress was applied as the bead rotated about the X-axis or the Y-axis, respectively (Fig. 1c). Here we describe a strategy to apply local in-plane stresses with the 3D MTC: when a magnetic bead was magnetized in X direction and then twisted in Y direction (Fig. 1d, left), a local stress was applied to the cell surface as the bead rotated about the Z-axis (Fig. 1d, middle, right).Fig. 1Strategies of applying different stress modes to a cell.a Schematic of the cell (blue color) with its nucleus (yellow color); the big black dot was the ferromagnetic bead. The cell was placed in a culture dish in the center of the coils and the focal plane of the microscope. Not drawn to scale. b Schematic of one dimensional magnetic twisting device (1D MTC) to apply force on the cell. The bead rotated in the Y–Z plane about the X-axis (torque was in the x direction) and generated a local complex stress. c Schematic of three-dimensional magnetic twisting device (3D MTC) to apply force on the cell. The bead rotated in the X–Z plane (out-of-plane) about the Y-axis (the right-hand rule), the Y–Z plane (out-of-plane) about the X-axis, or the X–Y plane (in-plane) about the Z-axis. d Schematic of magnetizing the magnetic bead in the X direction and twisting it in the Y direction. The bead rotated in the X–Y plane (about the Z-axis) which was parallel to the substrate surface of cell spreading and generated a local stress.Cell stiffness and chromatin stretching under different stress modesUsing the strategy of in-plane stress, we applied local stresses to the cell surface via specific receptors like integrins using an Arg-Lyn-Asp (RGD) peptides coated magnetic bead (Fig. 2a). The peak amplitude of the sinusoidal wave was kept at 15 Pa and loading frequency was kept at 0.3 Hz. We measured the bead rotation angles (radians) when the in-plane stress was applied (Fig. 2b). We also measured 2D projection of center displacements of the bead in the X–Y plane on the surface of the same cell as the bead rolled either along the long axis of the cell (0° stress mode) or transverse the long axis of the cell (90° stress mode) when the out-of-plane complex stress was applied (Fig. 2b). As expected from the anisotropic mechanical behaviors of the living cell17, the bead displacements were much less for the 0° mode (i.e., along the long axis of the cell and thus the direction of most stress fibers) than for the 90° mode (Fig. 2b). Using the published method of computing cell stiffness by taking into account the bead–cell contact area19, we computed cell stiffness of the same living cell under the condition of different stress modes. Cell stiffness was twice as much for the 0° mode as for the 90° mode (Fig. 2c). Interestingly, cell stiffness was lowest when the in-plane stress was applied (Fig. 2c). To determine how the local surface stresses deform the chromatin, we quantified deformation of chromatin domains where green fluorescent protein (GFP) labeled transgene DHFR (dihydrofolate reductase) resided20 (Fig. 2d). Mean Square Displacements (MSDs) of the GFP spots (Fig. 2e) and the changes in distances between any two GFP spots (chromatin deformation) (Fig. 2f) were highest for the 90° stress mode, intermediate for the in-plane stress mode, and lowest for the 0° mode. We then computed stretching of the chromatin domain containing the DHFR gene by quantifying the tensile strains and the shear strains of the chromatin21 and found that the in-plane stress mode resulted in the strains that were higher than the 0° mode and lower than the 90° mode (Fig. 2g, h). The data showed that tensile strains were about twice as much as the shear strains, suggesting that the dominant form of the chromatin deformation was tensile (i.e., stretching) for the in-plane mode. This result was unexpected. Since in contrast to the out-of-plane stress modes that resulted in predominantly normal strains for the 0° mode and similar magnitudes of normal and shear strains for the 90° mode at the cell cortex, the in-plane stress mode caused mostly shear strains at the cell cortex (Supplementary Fig. 1; Supplementary Table 1), but inside the nucleus the chromatin domain deformation via the in-plane mode was mainly tensile, suggesting that the complex structural arrangements of the cytoskeleton, linker of nucleoskeleton and cytoskeleton, and the nuclear lamins propagate the surface stress into the nucleus as a complex stress to result in mainly tensile deformation in the chromatin. Next we examined how stress amplitudes of the in-plane mode would impact chromatin deformation and cell stiffness. MSDs of individual GFP spots increased with the stress amplitude (Supplementary Fig. 2a–d). The bead twisting angle increased linearly with stress amplitude within the range of stress that we had applied (Supplementary Fig. 2e). As a result, cell stiffness did not change with the amplitude of the in-plane stress (Supplementary Fig. 2f). Furthermore, chromatin deformation increased with the in-plane stress amplitude (Supplementary Fig. 2g), suggesting that the structural changes in the chromatin domain are direct mechanical responses. Together these data show that in-plane stress induces different responses on cellular mechanical properties (i.e., cell stiffness) and chromatin deformation from the out-of-plane complex stresses.Fig. 2Mechanical anisotropy of the cell and of the chromatin induced by different force modes.a A representative image of adherent and elongated CHO (Chinese hamster ovary) cell with the GFP labeled chromatin domain (green dots; see the enlarged image within the dashed white lines). A 4-μm RGD-coated ferromagnetic bead (the solid black ball) was attached to the cell surface via integrins. Theta represents the angle of bead rolling direction with respect to the long axis of the cell (this notation applies to all cells in all figures). The bead and GFP spots in the white box were enlarged and shown on the right. Scale bar, 3 μm. The in-plane (X–Y plane) rotation has a direction along the Z-axis and the two out-of-plane rotations have directions along the X- and Y-axes, using the right-hand rule (see Fig. 1). b Peak 2D displacements of the center of the magnetic bead in the X–Y plane at stress angles 0° or 90° for the out-of-plane stress mode or the peak bead twisting angles for the in-plane stress mode in the same cell. In all stress modes, the peak amplitudes of the sinusoidal magnetic fields were maintained at 15 Pa and 0.3 Hz. Each triangle or dot represents one cell. c Cell stiffness computed with different stress modes. P = 0.0079 between 90° and in-plane stress modes; P < 0.001 between 0° and 90°, 0° and in-plane stress modes. d Fluorescent image of the three GFP spots in the same chromatin of a representative cell. Scale bar, 1 μm. e Normalized mean squared displacement (MSD) of all individual GFP spots when the stress (15 Pa at 0.3 Hz) was applied at 0° or 90° or in-plane mode. No stress data represent the spontaneous GFR spots movements in the absence of force application. f Chromatin deformation (i.e., changes of distances between any two GFP spots in the same chromatin domain) depends on stress modes. P < 0.001 between each different stress modes. g Tensile strains of the chromatin were computed from chromatin deformation21. P = 0.0055 between 0° and in-plane stress modes; P = 0.0017 between in-plane and 90° stress modes; P < 0.001 between 0° and 90° stress modes. h Shear strains were computed from the same chromatin deformation. P = 0.00701 between 0° and in-plane stress modes; P < 0.001 between in-plane and 90° and 0° and 90° stress modes. For b, c, and e–h, mean ± s.e.m.; n = 39 cells, 29 independent experiments; **P < 0.01; ***P < 0.001. P values were calculated and corrected using two-tailed Student’s t-test and Bonferroni correction. Source data are provided as a Source data file.Gene upregulation varies with stress modesNext we examined how DHFR gene transcription might change in response to different stress modes at the same stress amplitude and frequency. To minimize contributions of potential confounding factors in the cytoplasm to DHFR transcription, we employed a published strategy of using fluorescently labeled 5′-end probes that could detect the transcription of the first 1700 bps DHFR mRNA20. This way we could detect processes of transcription by reading the partial transcripts as early as seconds to a few minutes before the full transcript of DHFR was completed (DHFR gene is 34 kb long and it takes ~10 min to complete one transcript). When the stress was applied at 15 Pa and 0.3 Hz for 2 min, DHFR transcription was upregulated from the baseline: the in-plane stress mode caused an upregulation that was higher than the 0° mode and lower than the 90° mode (Fig. 3a). Extending the duration of stress application to 60 min, we found that all three stress modes resulted in a time-dependent elevation of transcription of DHFR (Fig. 3b). In line with the published results, the in-plane mode also led to a stress-amplitude dependent elevation of DHFR transcription (Fig. 3c). Since the extent of the chromatin deformation and DHFR upregulation under the in-plane mode was between those under the 0° mode and the 90° mode, we wondered if the 45° stress mode (i.e., the bead rotation was along the direction that was diagonal between the long axis and the short axis of the cell) would cause similar responses from the cell as the in-plane mode. It is interesting that although cell stiffness probed via the 45° out-of-plane mode was ~3 times that via the in-plane mode (Fig. 3d), the chromatin deformations were very similar (Fig. 3e). In addition, there was no difference in DHFR transcription upregulation between the 45° mode and the in-plane mode, either after 2 min stress or after longer stress application (Fig. 3f, g), suggesting that although these two modes are quite distinct in cell surface deformation (one is out-of-plane and the other is in-plane) that led to different stiffness values of the same cell, their impacts on the chromatin deformation and DHFR transcription were similar.Fig. 3Transcription varies with stress modes and increases with stress duration.a Summarized data of DHFR transcription detected with 5′-end probes. Stress was applied to each cell only once at one particular angle for all FISH experiments. Fluorescently labeled 5′-end probes were used to detect expression of the first 1700 bps DHFR mRNA. Controls (No stress) were the cells in the same dish without attachment of magnetic beads. The stress was applied for 2 min at 15 Pa and 0.3 Hz. P < 0.001 between each different stress modes. Mean ± s.e.m.; n = 115, 97, 344, and 85 cells for no stress, 0°, in-plane, and 90° stress modes, respectively, in nine separate experiments; ***P < 0.001. b
DHFR transcription upregulation depends on stress (15 Pa at 0.3 Hz) duration for both in-plane and out-of-plane modes. P = 0.016 between 30 and 60 min under 0° stress mode; P = 0.025 between 15 and 30 min and P = 0.0015 between 30 and 60 min under in-plane stress mode; P = 0.0039 between 30 and 60 min under 90° stress mode; P < 0.001 between other conditions under each stress mode. Mean ± s.e.m.; 0°: n = 21, 33, and 52 cells at 15, 30, and 60 min; in-plane: n = 38, 66, and 60 cells at 15, 30, and 60 min; 90°: n = 26, 25, and 32 cells at 15, 30, and 60 min. Three independent experiments. *P < 0.05; **P < 0.01; ***P < 0.001. The dashed line was the no stress control. c
DHFR gene upregulation depends on stress amplitudes of the in-plane mode. All stresses were applied at 0.3 Hz for 30 min. n = 71 cells for 0 stress; n = 47, 50, 68, and 66 cells at 6, 9, 12, and 15 Pa, respectively. P = 0.032 between 0 stress and 6 Pa; P = 0.0033 between 6 and 9 Pa; P = 0.014 between 9 and 12 Pa; P < 0.001 between other different stress amplitudes. Mean ± s.e.m.; *P < 0.05; **P < 0.01; ***P < 0.001. d Cell stiffness computed from different stress modes. P = 0.036 between 0° and 45° stress modes; P = 0.039 between 45° and 90° stress modes; P = 0.037 between 90° and in-plane stress modes; P < 0.001 between 0° and in-plane stress modes as well as 45° and in-plane stress modes. Mean ± s.e.m.; n = 11 cells, eight independent experiments. *P < 0.05; ***P < 0.001. e Chromatin deformation (chromatin deformation, i.e., changes of distance between GPF spots) with different stress modes. No difference was found between the 45° mode and the in-plane mode. P = 0.97 between in-plane and 45° stress modes; P < 0.001 between other different stress modes. Mean ± s.e.m.; n = 33 GFP spots; 11 cells, eight independent experiments. ***P < 0.001; ns = not significantly different. f Comparison of DHFR upregulation (quantified with 5′-end probes) between the out-of-plane 45° mode and the in-plane mode. The stress was applied for 2 min at 15 Pa and 0.3 Hz. P = 0.069 between in-plane and 45° stress modes. Mean ± s.e.m.; n = 344 and 140 cells for in-plane and 45° stress modes; nine separate experiments; ns = not significantly different. The dashed line was the baseline transcription level in the absence of force. g
DHFR upregulation was similar at longer durations of stress (15 Pa and 0.3 Hz) between the 45° out-plane mode and the in-plane mode. P = 0.44, 0.14, and 0.4 between in-plane and 45° stress modes under 15, 30, and 60 min, respectively. Mean ± s.e.m.; 45°: n = 21, 29, and 39 cells at 15, 30, and 60 min, respectively; in-plane: n = 38, 66, and 60 cells at 15, 30, and 60 min, respectively. Three independent experiments; ns = not significantly different. The dashed line was the baseline transcription level in the absence of force. P values were calculated and corrected using two-tailed Student’s t-test and Bonferroni correction. Source data are provided as a Source data file.Stress fiber anisotropy determines chromatin deformation and gene upregulationNext we set out to determine what dominates anisotropy of cell stiffness, chromatin deformation, and gene transcription upregulation. Since filamentous actin (F-actin) is a major player in cell stiffness22, we used a specific F-actin inhibitor, Latrunculin A (LatA), to treat the cells for various times to disrupt actin stress fibers. Actin stress fibers number decreased when the cells were treated with LatA for 2 min; stress fibers became short, disorganized, aggregated structures after 5 min LatA treatment in the absence of obvious cell rounding (Supplementary Fig. 3a–c). As a result, cell stiffness decreased gradually at 2 min for all stress modes; by 5 min of LatA treatment, the differences in cell stiffness between various stress modes disappeared (Fig. 4a). The cell stiffness data are consistent with the loss of structural integrity of the stress fibers. Interestingly, MSDs of chromatin GFP spots, a measure of chromatin deformation, increased after 2 min of LatA treatment (compare Fig. 4c with Fig. 4b), likely due to the fact that stress fibers were only partially disrupted (Supplementary Fig. 3b) and that the remaining stress fibers were able to transmit forces to the nucleus that was softened by the reduction in the F-actin in the cytoplasm. Five minutes after LatA treatment, however, force-induced chromatin deformation completely disappeared under all stress modes (Fig. 4d), likely because stress fiber integrity and the force-transmission pathway to the chromatin were completely disrupted. Interestingly, the gene upregulation under various stress modes after LatA treatment followed the same trend as the chromatin deformation: 2 min after LatA, transcription was increased when compared to that without LatA treatment for all stress modes and there was anisotropy in changes in transcription under different stress modes; however, no transcription upregulation was observed after 5 min LatA treatment under any stress mode (Fig. 4e, f). Taken together, these results suggest that actin stress fibers are responsible for anisotropy in cell stiffness, chromatin deformation, and rapid gene upregulation under different stress modes.Fig. 4Actin stress fibers dominate cytoskeletal anisotropy-dependent cellular responses.a Cell stiffness under different stress modes before adding Latrunculin A (No LatA control), after adding Latrunculin A (1 μM) for 2 min (+LatA 2 min) or 5 min (+LatA 5 min). P = 0.0075 between No LatA and +LatA 2 min under 45° stress mode; P = 0.023 between No LatA and +LatA 2 min under 90° stress mode; P = 0.039 between No LatA and +LatA 2 min and P = 0.0032 between +LatA 2 min and +LatA 5 min under in-plane stress mode; P < 0.001 between other conditions under each stress mode. Mean ± s.e.m.; n = 12 cells; nine independent experiments. *P < 0.05; **P < 0.01; ***P < 0.001. Normalized mean squared displacement (MSD) of all individual GFP spots when the stress (15 Pa at 0.3 Hz) was applied at 0°, 45°, or 90° out-of-plane mode or in-plane mode before adding LatA (b), after adding LatA for 2 min (c), or for 5 min (d). Mean ± s.e.m.; n = 12 cells, nine independent experiments. e Representative images of RNA FISH at no stress, 0°, 45°, or 90° out-of-plane mode or in-plane mode. Stress (15 Pa at 0.3 Hz) was applied before adding Latrunculin A (No LatA control), after adding Latrunculin A for 2 min (+LatA 2 min), or 5 min (+LatA 5 min). The yellow dashed lines highlighted the area of the nucleus. The GFP chromatin domains (green color) and the DHFR FISH fluorescence (red color) in the white box of each image were enlarged and shown on the right. The brightfield image of each cell was shown in an inset at the left site and the scale bar was 3 μm within each inset. The black dot was the magnetic bead attached to the cell surface. The nucleus was highlighted with a dashed line. The scale bar on the top right panel was 3 μm. f
DHFR transcription varies with duration of LatA treatment for both in-plane and out-of-plane modes. P = 0.0023 between No LatA and +LatA 2 min under 0° stress mode; P = 0.0048 between No LatA and +LatA 2 min under in-plane stress mode; P < 0.001 between other conditions under each stress mode. Mean ± s.e.m.; 0°: n = 38, 45, and 47 cells; in-plane: n = 58, 89, and 41 cells; 45°: n = 35, 52, and 40 cells; 90°: n = 28, 31, and 62 cells for No LatA control, +LatA 2 min and +LatA 5 min, respectively; three independent experiments. Control was the cells applied with the same stress but not treated with LatA. **P < 0.01; ***P < 0.001. The dashed line indicated the DHFR gene expression levels without applying the stress. P values were calculated and corrected using two-tailed Student’s t-test and Bonferroni correction. Source data are provided as a Source data file.Mechanisms of stress-mode dependence of cell stiffness and chromatin deformationTo elucidate the underlying mechanism of why the in-plane stress mode leads to the lowest cell stiffness of all the stress modes but induces chromatin deformation comparable to that by the 45° out-of-plane complex stress mode, we employed the approach of finite element model (FEM) (Fig. 5a, b). The FEM of a typical anisotropic cell, whose number of actin stress fibers along the cell long axis was twice as much as the short axis (Fig. 5a), generated stress-mode dependent magnetic bead displacements and chromatin deformation (Fig. 5c–e). The displacement of the magnetic bead along the short axis of the cell (the 90° stress mode) was 1.6 times that along the long axis (the 0° stress mode), and the displacement of the bead at 45° stress mode was in-between these two values (Fig. 5d), indicating that the cell stiffness has distinct anisotropy. In addition, the rotation angle of the magnetic bead under the in-plane stress was ~0.5 rad, suggesting that the cell stiffness induced by the in-plane mode was lower than all out-of-plane stress modes. These results quantitatively reproduced what was measured experimentally in living cells (see Fig. 2 and Supplementary Fig. 4a). We then analyzed the effects of stress modes on chromatin deformation. We used the same approach as the live cell experiments, i.e., three points in the X–Y plane of a given height inside the nucleus to represent the points in the same chromatin domain (Fig. 5c). We then calculated the change of the distance between each two points before and after loading, and took the average of the three values as the chromatin deformation, with which the tensile and shear strain were calculated. The FEM calculation qualitatively reproduced the live cell experimental measurements (compare Fig. 5e–g with Fig. 2f–h and with Supplementary Fig. 4). Tensile strains and shear strains exhibited anisotropy and the tensile strains were ~twice as large as the shear strains. In addition, the chromatin strains induced by the in-plane stress mode were similar to those by the 45° out-of-plane stress mode.Fig. 5Finite element models (FEM) reveal anisotropic responses of the cell.a A 3D illustration of the location of bead and spatial arrangement of the nucleus and actin stress fibers in the FEM model of the cell–bead system: the magnetic bead is in light blue, the nucleus is in dark blue, the cytoplasm is in light brown, and the cell cortex is in brown. Not drawn to scale. b The cross-sectional illustration of cell–bead system, showing the dimension of the cell, nucleus, and the bead (see Methods section). The specific torque Tm (applied stress of 15 Pa) was applied on the cell via the bead. We estimated various parameter values for different components of the cell, obtained from the published report43. The Young’s moduli of cell membrane cortex, cytoplasm, and nucleus were 2, 0.25, and 1.0 kPa, respectively; Poisson’s ratios of cell membrane cortex, cytoplasm, and nucleus were 0.3, 0.49, and 0.3, respectively; the actin stress fiber’s stiffness was 20 nN μm−1. c Illustration of application of load for each stress mode, where theta represents the angle between the rolling direction of the magnetic bead and the long axis of the cell, and the magnified view of the boxed area of nucleus illustrating the three points for calculating the chromatin deformation. d Magnetic bead displacements and rotation angle. e Chromatin deformation. f Chromatin tensile strain. g Chromatin shear strain.To examine the role of structural integrity of F-actin in cell stiffness anisotropy and chromatin deformation, actin stress fibers were gradually disassembled in the FEM to simulate the inhibitory effect of LatA on F-actin in the live cell experiments. Based on the experimental results from live cells, when treated with LatA for 2 min, actin stress fibers were only partially disassembled, but when treated with LatA for 5 min, the stress fibers were completely destroyed (Fig. 6a–c). The displacements of the magnetic bead after LatA treatment under the three out-of-plane stress modes were shown in Fig. 6d: as stress fibers gradually disassembled, the differences of the bead displacements among the three stress modes were substantially reduced. These results recapitulate the cell experimental results that stress fibers play a key role in the anisotropy of cell stiffness and the average cell stiffness.Fig. 6FEM simulation of the effects of disrupting stress fibers on cell and chromatin deformation.a No LatA control. b +LatA 2 min. c +LatA 5 min. Cell shapes and the bead are not drawn to scale. d Effects of the LatA treatment on magnetic bead displacements under different stress modes. e Effects of LatA treatment on chromatin deformation under different stress modes.It is known from our earlier published reports that the long distance transmission of force into the nucleus depends on actin stress fibers21,23. In the FEM, stress fibers between the nucleus and the cell surface were partially (~30%) disrupted after 2 min LatA, leading to weakening the constraints on the nucleus to increase chromatin deformation, which was a result of the elevated stress concentration on the remaining stress fibers to propagate stresses to deform chromatin (Fig. 6e). In contrast, stress fibers between the magnetic bead and the nucleus in the FEM after 5 min LatA treatment were completely disrupted, no longer being able to transfer the load into the nucleus. Hence the chromatin deformation reduced dramatically and the impact of different stress modes on chromatin deformation no longer existed (Fig. 6e). The modeling data therefore are consistent with the experimental results from living cells (see Fig. 4b–d).Our FEM simulation results showed that under the out-of-plane complex stress modes most of the actin stress fibers were directly stretched or compressed by the rolling motion of the magnetic bead. In contrast, under the in-plane stress mode, most of the stress fibers were not directly stretched. Instead, a tangential deflection was produced which allowed a large rotation of the magnetic bead, resulting the larger extent of cell deformation and thus lower cell stiffness than those by out-of-plane stress modes.Why chromatin deformation under the in-plane stress mode is similar to that under the 45° out-of-plane stress mode is not obvious. The fact that the two different stress modes (i.e., 45° out-of-plane and in-plane) have similar impacts at a distance, i.e., inside the nucleus, is consistent with the idea that stresses and strains are possibly mediated by the anisotropic cytoskeletal structures and not by a continuous material. To determine the underlying mechanism, we performed additional FEM simulations. Comparing nuclear deformation between the in-plane and 45° stress modes (Supplementary Fig. 5a, b), we found that the nuclear deformation magnitudes and distributions between the two cases were very similar and the nuclear strains near the site of loading were similar as well (Supplementary Fig. 5c, d). In contrast, at the cell cortex near the magnetic bead, the 45° out-of-plane stress mode induced much higher normal strains than the in-plane stress mode whereas the in-plane stress mode induced higher shear strains than the 45° stress mode (Supplementary Fig. 1c, d; Supplementary Table 1). These simulation results suggest that although the cell deformations induced by these two different modes are quite distinct at the cell cortex, the stresses that are propagated into the nucleus by the in-plane and the 45° out-of-plane stress modes are quite similar, possibly due to the fact that all stresses in the cytoplasm have to be concentrated into the nucleus to deform the chromatin, leading to similar chromatin deformation.DiscussionIn the current study we compared the effects of applying (cell surface) out-of-plane complex stresses with those of applying a (cell surface) in-plane stress on cell mechanical behaviors and biological responses such as gene transcription. We find that there is substantial anisotropy in cell stiffness, chromatin deformation, and gene transcription under different stress modes. For out-of-plane stress modes, rolling the magnetic bead along the short axis of the cell (i.e., 90° from the direction of the most stress fibers) results in larger cell deformation than that along the long axis of the cell (0°) or 45° from the cell long axis. In contrast, applying an in-plane stress leads to more cell deformation than all of the out-of-plane stresses. FEM simulation shows that this is due to the fact that in-plane mode causes little compression or stretching of the stiff actin stress fibers and mainly tangential deflection of the stress fibers, resulting less resistance from the stiff stress fibers. Strikingly, the in-plane stress mode induces similar chromatin deformation as the 45° out-of-plane stress mode in living cells. This result is recapitulated by FEM simulations. To determine why the 45° mode and the in-plane mode induce similar strain maps in the nucleus and how gene transcription is altered by various force modes, we modeled the actin stress fiber as an elastic rod of 12 kPa modulus24 and mapped the stress fiber deformation near the nucleus and found that the 0° mode resulted in small stress fiber deformation and low nuclear strain magnitudes; the 90° mode resulted in the large stress fiber deformation and high nuclear strain magnitudes. In contrast, the 45° mode and the in-plane mode led to intermediate stress fiber deformation near the nucleus and medium nuclear strain magnitudes (Fig. 7). These modeling results show that although the 45° mode and the in-plane mode generate different cell cortex strains, the two force modes generate similar stress fiber deformation patterns near the nucleus and thus ensure similar nuclear strain maps, possibly as a result of complex stress distribution in the stress fibers. These modeling data demonstrate that chromatin tensile strain is low for the 0° mode, high for the 90° mode, and medium for both the 45° mode and the in-plane mode, consistent with the experimental data from living cells (see Figs. 2g and 3e). From the current study and a previously published report20, it is known that the level of rapid DHFR gene upregulation depends on the extent of chromatin stretching or chromatin tensile strain. Together these findings suggest that because the majority of the stress fibers aligns along the long axis of the cell, which results in an anisotropy of cell stiffness, different force modes cause different stress fiber deformation at the nucleus to deform the chromatin and to increase gene expression differently.Fig. 7A model of stress fiber anisotropy regulating chromatin strain and gene upregulation.The magnetic bead was bonded to the cell surface via integrins and focal complex-like structures. The schematics on the left were the Y–Z view and X–Y view of the bead (4 μm in diameter) and the cell. Each actin stress fiber was modeled as an elastic rod (Young’s modulus was 12 kPa24 with 0.4 μm equivalent diameter and Poisson’s ratio was 0.4). There were twice as many stress fibers aligned along the long axis of the cell as those aligned along the short axis of the cell. A constant stress of 15 Pa was applied from the bead to the cell under each force mode. The middle images were the stress fiber deformation and von Mises equivalent strain maps of the nucleus under different force modes. For visual clarity, the bead was removed from each image. The images on the right were chromatin strains under various force modes. Each wiggly line represents a stretched chromatin domain containing the DHFR gene locus (the pink dot), which was located at ~2 μm from the edge of the nuclear envelope37. From the color scale bar of the nuclear strain and the chromatin domain location, it was estimated that the chromatin domain strain was low (~5%) for the 0° mode (the bead rolling along the cell long axis), high (~25%) for the 90° mode, and medium (~10%) for both the 45° mode and the in-plane mode, comparable with the experimental data from living cells (see Figs. 2 and 3). This model suggests that the extent of chromatin stretching (tensile strain) and thus the level of gene upregulation (the number of the plus sign represents relative levels of gene upregulation) depend on nuclear strain, which is caused by different stress fiber deformation at the nucleus under different force modes. The color scale applies to all images.Is the out-of-plane complex stress or in-plane stress physiologically relevant modes of force application? It is well known that endothelial cells lining the blood vessels experience fluid shear stresses on their apical surfaces and these stresses are balanced at the basal surfaces at the focal adhesions25. In contrast, blood vessels experience blood pressure induced stretching that can cause complex strains in the cells of the vessels along the length of the vessels26. It is also known that the myosin-II mediated cellular tractions at the focal adhesions are complex stresses with both in-plane and out-of-plane components27. Therefore, exploring the effects of different stress modes on cellular functions is both physiologically relevant and important in understanding how forces impact living cells and tissues.A living cell’s cytoskeleton consists of three major filament systems: actin filaments, microtubules, and intermediate filaments. The actin filaments are bundled together to form actin stress fibers with myosin-II. We have identified the dominant role of actin stress fibers in anisotropic responses of the cells. Microtubules are known to play mechanical roles in cell functions28,29. However, when the cells were treated with colchicine, there were no observable significant changes in cell stiffness or chromatin deformation (Supplementary Fig. 6), suggesting that microtubules play a minor role, if any, in rapid gene upregulation under different stress modes in these CHO cells. Intermediate filaments of the cytoskeleton also play roles in mechanical and biological functions of living cells30,31. However, it is known that intermediate filaments play significant roles in cell stiffness only at relatively large deformations32. In the future, it will be interesting to explore the potential impact of intermediate filaments in cell responses under different stress modes. CHO cells, generally elongated and have numerous stress fibers after adhesion and spreading, are an epithelial cell line. It remains to be determined in the future whether our findings can be applied to other types of cells.When the stress fibers are disrupted by LatA, changes in geometry (i.e., bead–cell contact area) and/or bead–cell adhesion might influence cell stiffness measurement. We employed an approach of infecting living cells with talin-GFP, which labeled focal adhesions and focal complex-like structures at the cell basal surface (Supplementary Fig. 7) to examine this possibility. Bead embedding was calculated using the largest diameter of the X–Y projected image of talin-GFP fluorescence as a measure of the largest bead–cell contact interface; bead–cell contact area was estimated by measuring the X–Y projected area times the average intensity of the largest fluorescence ring-like structure of talin-GFP surrounding the bead, which was an index of the bead–cell contact area, after taking into account of temporal controls without LatA of laser scanning induced photobleaching (Supplementary Fig. 8a). We found that after 5 min LatA treatment there was only a slight decrease (~10%) both in the bead embedding and the bead–cell contact area from that before the F-actin disruption (Supplementary Fig. 8b, c). This finding suggests that for the same applied torque, the applied stress was increased by ~30% due to the 10% reduction in the contact area, which would be predicted to lead to a ~30% elevation in cell stiffness19. However, 5 min after LatA treatment to disrupt stress fibers, the cell stiffness exhibited the opposite trend: the cell stiffness was reduced dramatically for all force modes (see Fig. 4a). These data suggest that the change in cell stiffness after LatA treatment is most likely due to the increase in cell deformation as a result of stress fiber disruption and not due to changes in the bead–cell contact area. On the issue of the adhesion between the RGD-bead and integrins, it is known that a single ligand–integrin bond yielding force is ~100 pN33. There are hundreds of integrin receptors that interact with one 4-μm magnetic bead, which is coated with a saturating amount of RGD peptides, suggesting the bonds between the bead and the cell surface are strong. Moreover, we did not observe any bead peeling off the cell surface after stress fiber disruption, suggesting that most bonds are stable. However, it is reported that the characteristic adhesion stress decreases by ~25% after F-actin disruption with cytochalasin D treatment for 5 min34. It will be interesting in the future to investigate whether the characteristic adhesion stress changes in various force modes and whether its change after stress fiber disruption impacts on cell stiffness measurement. Furthermore, the stretching distance of the RGD-bead and integrins is only on the order of 0.1–1 nm before the bond is broken33, too small to account for the measured bead displacements of several hundred nanometers in live cells (see Fig. 2b). The bead displacements increased by several folds when stress fibers are disrupted by LatA; these changes in bead displacements cannot be explained by the bead–cell bond adhesion alteration. From published articles14,17,20–22, it is known that when the RGD-bead is attached to the cell surface, the bead is attached to the cell via the integrin–actin linkages and the bead is tethered by tense F-actin bundles (i.e., stress fibers), providing majority of the resistance to bead rotation (or displacement) under stress. This resistance is an index of cell stiffness. When F-actin bundles are disrupted, this tethering is abolished and hence the resistance to bead rotation (or displacement) under applied stress is substantially reduced and the cell stiffness is dramatically decreased. Together all these suggest that the bead–cell contact area and adhesion play a minor role in cell stiffness after stress fiber disruption.To better understand the effect of long-range force propagation by the actin stress fibers, we performed simulation by treating the composite of cytoplasm and the actin cytoskeleton as anisotropic uniform elastic materials without explicitly modeling the actin cytoskeleton (denoted as continuous model) (Supplementary Fig. 9). In contrast, in the above analyses, we have employed an FEM (denoted as discrete model, see Fig. 5a) to simulate actin stress fibers as discrete stiff elements embedded in the cytoplasm and calculate their impact on cell and chromatin deformation. The boundary conditions and stress modes in the continuous model are the same as the discrete model. Our results show that although the continuous model could generate similar bead displacements or rotation angles (i.e., similar cell deformation) as the discrete model (compare Supplementary Fig. 9b with 9e) by choosing proper parameter values for the anisotropic properties of the cell, the magnitudes of the chromatin deformation in the continuous model are much smaller than those computed from the discrete model (compare Supplementary Fig. 9c, f), as a result of more localized deformation in the cytoplasm and the nucleus using the continuous model than the discrete model (compare Supplementary Fig. 9a, d). These results are mainly due to the fast decay of the deformation field in the uniform elastic body in the continuous model (stress decays rapidly as the square of the distance), as predicted by the St. Venant’s principle (i.e., a local force only causes local deformation), so that the load from the magnetic bead cannot be effectively transferred to the nucleus. In contrast, the stress fibers in the discrete model can effectively transfer the load to the nucleus to a long distance due to stress concentration, consistent with the previous report of a single stress fiber model23. Therefore, the discrete model is more appropriate for simulating long distance force propagation and chromatin deformation of an intact, spread living cell. Besides cytoskeletal filament stiffness, it is known that myosin-II dependent endogenous prestress is critical in the long distance force propagation in the cytoplasm and to the nucleus20–23,35. To examine the role of nonmuscle myosin-II, we treated the cells with blebbistatin to inhibit myosin-II. We found that for the same loading, cell stiffness decreased substantially after 50 μM blebbistatin treatment (Supplementary Fig. 10a) and chromatin deformation decreased dramatically (Supplementary Fig. 10b–d), consistent with a finding that myosin-II inhibition reduces cell stiffness36 and earlier reports that prestress mediates long force propagation in the cytoskeleton and into the nucleus21–23,35. Furthermore, gene upregulation by the applied stress was substantially inhibited when the myosin-II dependent endogenous stress was inhibited with blebbistatin for all force modes (Supplementary Fig. 10e), consistent with and extending the previous finding20 when myosin light chain kinase is inhibited by ML-7. In our current discrete FEM model, the impact of the endogenous prestress has been incorporated into the stress fibers to increase the stiffness of the stress fibers and thus is not explicitly shown. Complete disruption of stress fibers in the discrete FEM that removes the influence of myosin-II dependent prestress leads to almost total abolishment of chromatin deformation is consistent with the previous reports21,23. Furthermore, it is well established that living cells are viscoelastic and they respond to loading frequencies. However, a recent report finds that loading frequencies between 0.3 and 6 Hz have only modest impacts on chromatin deformation and gene upregulation; the article reveals that force-induced gene upregulation in living cells does not follow the weak power law of rheology but depends on H3K9 demethylation37. Although our current discrete FEM is only an elastic model, it has recapitulated the essential features of mechanical and transcriptional responses of living cells at a low loading frequency (0.3 Hz in the current study) and revealed the underlying mechanisms of cellular responses. In the future, the FEM can be extended to include viscoelastic elements in order to better simulate responses under high loading frequencies (e.g., 10–20 Hz). Recently a 3D cell model has been reported to describe the crosstalk among cell adhesions, the cytoskeleton, and the nucleus38. It will be interesting to find out if this model can simulate the effects of different stress modes on cell stiffness and direct chromatin stretching.In summary, we demonstrate that stress fiber anisotropy contributes to force-mode dependent cellular mechanical responses, chromatin deformation, and gene transcription. Our current study is a first attempt to provide insights on better understanding how living cells respond to different modes of complex forces in physiology and disease.MethodsCell culture and reagentsCHO DG44 DHFR D10 cells (provided by Dr. Andrew Belmont of University of Illinois who created this cell line) were cultured in Ham’s F12 media without thymidine and hypoxanthine (Shanghai Basal) with 10% fetal bovine serum (FBS, Gibco) and 1% Penicillin–Streptomycin (Hyclone)20. The authentication of the cell line was confirmed by directly visualizing the GFP-lac repressor staining patterns in these cells. Cells were passaged every 3 days by using TrypLE (Gibco). After centrifuging, cells were dispersed uniformly within 1 mL medium and 200 μL cell suspension was added to a sterile six-pore plate with 3 mL medium which was precoated with 0.1% gelatin (Cat. No. 354236, Corning). Cells were cultured in a 37 °C and 5% CO2 incubator for 2 days before experiment. DAPI (4′, 6-diamidino-2-phenylindole) staining of the cell nuclei was constantly performed to monitor these cells during the course of experiments for possible mycoplasma contamination and no signs of mycoplasma contamination were found. Cells in different dishes were randomly assigned during experiments. LatA was purchased from Dalian Meilum Biotech Co., Ltd (Cat. No. J0704A). Colchicine was from MedChem Express Co., Ltd (Cat. No. HY-16569). Blebbistatin was from ApexBio Tech LLC (Cat. No. B1387). F-actin was stained for 4 h before experiment by SiR-actin KIT from Cytoskeleton, Inc. (Cat. No. CY-SC001). Talin-GFP was from Thermo Fisher Scientific (Cat. No. C10611). For talin-F-actin double-staining, actin was labeled by SiR-actin KIT for 4 h after Talin-GFP reagent was added for 16 h. Reagents for RNA FISH: 20× SSC (saline sodium citrate) (Cat. No. AM9763, Thermo Fisher Scientific); 10× PBS (phosphate buffered saline) (Cat. No. AM9625, Thermo Fisher Scientific); BSA (Cat. No. AM2616, Thermo Fisher Scientific); Ribonucleic acid (Cat. No. R1753-2KU, Sigma); Dextran Sulfate Sodium (Cat. No. D8906-10G, Sigma); Deionized Formamide (Cat. No. AM9342, Thermo Fisher Scientific); 16% Formaldehyde (Cat. No. 28906, Thermo Fisher Scientific); Ethanol absolute (Cat. No. 10009218, Sinopharm Co., Ltd). Salmon fibrinogen (Cat. No. SEA-133) and thrombin (SEA-135) were purchased from Pfenex Inc. (CA, USA).3D magnetic twisting cytometry3D MTC18 can generate stresses in any defined direction (X, Y, Z) via rotational movements of ferromagnetic magnetic beads (Boston magnetic beads, purchased from J. Fredberg, Boston, MA) attached to the cell membrane. When the magnetic beads were magnetized in the Z direction and applied a homogeneous twisting magnetic field in the X or Y direction, beads would rotate out of X–Y plane and move along the short or long axis of the cells. This mode of loading was called the out-of-plane stress mode. When the beads were magnetized in the X direction and applied a twisting magnetic field in the Y direction (i.e., the beads were magnetized and twisted in the X–Y plane), the beads would rotate in X–Y plane and generate stresses. This model of loading was called the in-plane stress mode. In order to apply different modes of force to the same cell, the culture dish containing the cells and the beads were rotated so that the long axis of a cell with the magnetic bead on its surface was aligned along the Y-axis. Then the bead was magnetized along the Z-axis and a twisting field of 15 Pa was applied along the Y-axis. The bead rotation direction vector thus was along the X-axis, based on the right-hand rule. This was the 0° out-of-plane mode. The same bead was then re-magnetized along the Z-axis and a twisting field of 15 Pa was applied along the X-axis. The bead was rotated along the Y-axis. This was the 90° out-of-plane mode. The same bead was then re-magnetized along the Z-axis again, and simultaneously a twisting field of 10.6 Pa was applied along the X-axis and a twisting field of 10.6 Pa was applied along the Y-axis, such that the bead rotated at an angle of 45° between the X- and Y-axes and the sum of the two vector magnitudes was 15 Pa. This was the 45° out-of-plane mode. The in-plane mode was applied by re-magnetizing the same bead along the X-axis and applying the twisting field of 15 Pa along the Y-axis. Using this strategy, we were able to apply different modes of force of the same amplitude to the same cell and quantify its stiffness and chromatin deformation. For the out-of-plane modes, the bead displacements were measured by quantifying 2D projections of the center of the bead displacements in the X–Y plane. For the in-plane mode, the bead rotation angle was measured directly by quantifying the 2D rotation angle of the bead edge in the X–Y plane using the gray pixels of the cell surface right at the bead edge as the reference points. We measured the distance between the bottom of the bead and the rigid cell substrate. The distances were 3.42 ± 0.20 μm (maximum is 5.82 μm, and minimum is 1.32 μm) (Supplementary Fig. 11). If the distance is below 1 μm, the substrate stiffness starts to critically contribute to the apparent cell stiffness measurement. Therefore, the impact of the rigid substrate for our CHO cells in our experimental conditions does not appear to be substantial. Stress applied to cell surface was proportional to the magnitude of the twisting magnetic field. In our experiments, 10, 15, 20, or 25 Gauss magnetic field corresponded to 6, 9, 12, or 15 Pa stress to the cells, calibrated using a viscous fluid standard14,15, with which it was determined that the bead magnetic moment constant was 2 Pa Gauss−1. The applied stress was equal to the bead magnetic moment constant times the magnetic field divided by 6. One pair of MTC coils could generate 25 Gauss per 100 turns and each pair of the coils in X, Y, or Z direction in the 3D MCT had 180 turns of coils and hence there was a gain of 1.8 in the magnetic field for the coils of the 3D MTC. For example, for a 25 Gauss magnetic field, the stress applied by the 3D MTC was 15 Pa (2 Pa Gauss−1 times 25 Gauss times 1.8 divided by 6). To independently calibrate the bead magnetic moment constant, we chose an alternative method by fully embedding the individual magnetic bead in a uniform elastic material made of fibrin gels of known moduli. The salmon fibrinogen was 2 or 4 mg mL−1 and 50 μL of fibrinogen was activated by thrombin (1 μL of 100 U mL−1) to form the fibrin gels39, with the corresponding shear elastic modulus of 60 or 140 Pa, respectively39. The stress was applied at 6, 9, 12, or 15 Pa to the bead using the in-plane stress mode at 0.3 Hz and the resultant bead angular rotation (strain) was quantified. The calculated shear elastic modulus of the 2 mg mL−1 fibrin gel was 65 Pa and of the 4 mg mL−1 fibrin gel was 148 Pa (Supplementary Fig. 12), very close to the published shear modulus values of the fibrin gels39. These results suggest that the calibrated bead magnetic moment constant using the viscous standard was accurate and the stiffness measurement by the 3D MTC technology was reliable. Before experiments, the magnetic beads had been coated overnight with a saturating amount (50 μg mL−1 RGD per mg of beads) of Arg-Gly-Asp (RGD, PEPTIDE 2000, Cat. No. R4658, Sigma-Aldrich) peptides. Beads and cells were co-incubated for 20 min in the incubator to let the beads tightly attach to the integrins. To ensure there was only one bead per cell in 35 mm petri dish with 18 mm glass-bottomed well (Cat. No. GBD00003-200, Cell E&G), beads were added at a low amount (30 μL of 1 mg mL−1 beads) to each dish. Since the chromatin deformation was often very small in the 0° out-of-plane stress mode, we selected the beads whose distances were ~4–5 μm from the GFP spots such that we were able to quantify chromatin deformation at the same location of the same cell under all four stress modes (0°, 45°, and 90° out-of-plane modes and the in-plane mode). The frequency of sinusoidal magnetic field was fixed at 0.3 Hz. Cell stiffness was calculated from bead displacements or rotation angles19,20. As the beads might have different degrees of embedding with cells, different embedding coefficients were used to calculate cell stiffness19. For the beads whose degree of beads embedding were 36.3% ± 0.7% (n = 39 cells, max = 45%, min = 26%) (Supplementary Table 2), embedding coefficients β = 0.8, α = 0.3 were used to calculate cell stiffness of out-of-plane and in-plane modes, respectively.RNA FISHCustom Stellaris FISH Probes were designed against DHFR mRNA (https://www.ncbi.nlm.nih.gov/nuccore/NM_010049.3) by using Stellaris RNA FISH probe Designer (Bioresearch Technologies Inc., Petaluma, CA) to detect the DHFR genes transcription. CHO DG44 cells were applied requisite stress amplitude with a certain time, then fixed in 3.7% formaldehyde for 30 min and permeabilized in 70% ethanol for 12 h at 4 °C. Samples were hybridized with DHFR Stellaris FISH Probes and incubated in dark humidified incubator for 12 h. Washing 2× in 1 mL wash buffer at 37 °C for 30 min, then proceed to imaging by Leica DMI6000B microscopy after adding 2 mL 2× SSC. Image J was used to analyze the FISH images.Microscopy and live cell imagingLeica DMI6000B with 63 × 1.4 NA oil-immersion objective (Leica, Cat. No. 15506350) was used to image FISH fluorescence via Leica MM AF software. Leica SP8-STED microscopy with 100 × 1.4 NA oil-immersion objective was used to visualize chromatin GFP spots in CHO cells via Leica Application Suite X software. Chromatin deformation was measured following by quantifying changes of distances in the X–Y plane between chromatin GFP spots in the same chromatin domain20. The Z-direction displacement was not measured and hence the chromatin deformation and chromatin strains were an underestimate of the total deformation. However, since the CHO cells were very spread (see Fig. 2a), the cell height in Z-axis was much less than the dimensions in X–Y plane, the changes in distances in Z was very small. Leica SP8-STED microscopy excited GFP spots and Talin-GFP at 488 nm, SiR-actin at 633 nm. Fluorescence was detected by Hybrid detector.Quantification of bead–cell contact areaThe CHO cells were cultured for 48 h on glass-bottomed 35-mm dishes, precoated with 0.1% gelatin. Then 20 μL mL−1 medium of the reagent of CellLight® Talin-GFP (Thermo, C10611, BacMam 2.0) that was a fusion construct of the c terminus of human talin and emGFP were added to each 35-mm dish for 16 h before experiment. The fusion construct is packaged in the insect virus baculovirus, which does not replicate in mammalian cells, providing accurate and specific targeting to cellular talin-GFP, according to manufacturer’s product description. Bead embedding was calculated using the largest diameter of the X–Y projected image of talin-GFP fluorescence as a measure of the largest bead–cell contact interface; bead–cell contact area was estimated by measuring the X–Y projected area times the average intensity of the largest fluorescence ring-like structure of talin-GFP surrounding the bead, which was an index of the bead–cell contact area, after taking into account of temporal controls without LatA of laser scanning induced photobleaching. Adjustment of thresholds was made at the same level for all images to eliminate the background noise. Since bead–cell contact areas varied among different cells, each bead–cell contact area after LatA treatment was normalized by that before the drug treatment.Dynamic tracking and image analysisMATLAB software for data analyzing and image processing18 can be downloaded from https://pan.baidu.com/s/1qXCGozy/.Finite element modelThe FEM of the cell consists of four parts: the cell membrane cortex, the cytoplasm, the nucleus, and the cytoskeleton (Fig. 5a). The first three parts were discretized by 3D bulk elements, while the cytoskeleton, simplified by using contractile actin stress fibers40, was discretized by the spring elements41. In living CHO cells under the experimental conditions, we found that the cell length was 41.79 ± 0.98 μm, the width was 10.68 ± 0.34 μm, and the height was 8.19 ± 0.17 μm (n = 39 cells) (Supplementary Fig. 11). To be consistent with the observed values in live cell experiments, length, width, and thickness of the cell were chosen to be 40, 12, and 8 μm, respectively, and the bead embedding was set to be 38%, as the experimentally measured bead embedding was 36.3% (Supplementary Table 2). The thickness of cell membrane cortex42 was set to be 0.25 μm. The nucleus was modeled as an elastic ellipsoid with its long axis as 12 μm and short axis as 7 μm. In order to consider the anisotropic arrangement of the cytoskeletal structure caused by cell polarization, the number of actin stress fibers along the long axis (Y-axis) was set to be twice as those along the short axis (X-axis). The magnetic bead was also discretized by 3D bulk element, and its adhesion with cell membrane cortex was modeled by bonded contact interactions, which provided the anchor points of the actin cytoskeleton to the bead. The contact of the magnetic bead with the cell surface was chosen to be a few μms from the end of the nucleus. During the loading process, a stable adhesion was assumed between the cell and the substrate, which was regarded as a fixed connection. The mesh was locally refined near the magnetic bead (the bead’s Young’s modulus was assumed to be 2 × 105 MPa), generating a total of 129,543 elements and 208,144 nodes. To determine actin stress fiber deformation near the nucleus, the stress fibers were modeled as elastic rods which had an equivalent diameter of 0.4 μm, whose Poisson’s ratio was 0.4, and whose Young’s modulus was 12 kPa24. Other parameters were set to be the same as those used above when stress fibers were modeled as spring elements. von Mises equivalent strains of the nucleus were computed under various force modes.Statistical analysisTwo-tailed student’s t-test was used for all statistical analyses, except when multiple comparisons were carried out within a given experiment with Bonferroni correction.Supplementary informationSupplementary Information | nature communications | [
"Article"
] | [
"Biological techniques",
"Computational biophysics"
] | living cells tissues respond to mechanical force stimulation shear stresses blood tractional forces at cell–matrix cell–cell contacts4 Mechanical properties critical regulating biological responses physiology diseases5–8 different forces impact gene expression elusive shear stresses at cell apical surface produce different signals effects stretching forces at base sides endothelial cells mechanisms gene transcription force modes unclear few technologies apply controlled local mechanical forces to single living cell method atomic force microscopy (AFM) applies indentation force on apical surface method optical tweezer laser light micrometer-sized lateral force on cell apical surface12 third method magnetic twisting cytometry (MTC) magnetic gradient pullers electromagnetic inhomogeneous field generated tweezer pull on magnetic bead apical surface ferromagnetic bead magnetized with strong magnetic field horizontal direction Y weak homogenous magnetic field applied vertical direction rotate bead about X-axis14 early version MTC rotate magnetic bead about one X-axis called one (1D) MTCtechnical difficulty to compare mechanical responses living cell to probes16 clear surface force modes influence cellular responses functions strategy using three-dimensional MTC apply forces directions living cell rotating magnetic bead axes given stress amplitude living cell responds to local in-plane stress differently out-plane complex stress stretching chromatin upregulating gene transcription Experimental results theoretical modeling analyses stress fiber anisotropy determines force-mode cell stiffness chromatin stretching regulates gene upregulation strategy applying local in-plane 1D MTC generated bead rotation out-of X–Y plane generated complex stress on cell surface bead rotated plane edge upward stretched cell membrane downward compressed membrane (Fig. 1a 3D MTC magnetize magnetic bead (X Y Z applied homogeneous twisting field direction magnetic bead magnetized Z twisted Y out-of-plane complex stress applied as rotated X Y-axis strategy apply local in-plane stresses with 3D MTC magnetic bead magnetized X twisted Y direction local stress applied to cell surface as rotated Z-axis1Strategies applying stress modes cell Schematic cell (blue nucleus (yellow black dot ferromagnetic bead cell culture dish center coils focal plane microscope Not drawn scale Schematic one magnetic twisting device (1D MTC) force cell bead rotated Y–Z plane X-axis generated local complex stress Schematic three-dimensional magnetic twisting device (3D MTC) force bead rotated X–Z plane Y-axis X–Y Schematic magnetizing magnetic bead X direction twisting Y direction bead rotated X–Y plane parallel substrate surface cell generated local stress.Cell stiffness chromatin stretching under stress modesUsing strategy in-plane stress applied local stresses cell surface receptors integrins Arg-Lyn-Asp (RGD) peptides coated magnetic bead (Fig. 2a). peak amplitude sinusoidal wave 15 Pa loading frequency 0.3 Hz measured bead rotation angles in-plane stress applied (Fig. 2b). measured 2D projection center displacements bead X–Y plane surface same cell transverse (90° stress out-of-plane complex stress applied (Fig.2b). expected anisotropic living bead displacements less for 0° mode than 90° mode (Fig. 2b). cell stiffness computed cell stiffness same under different stress modes stiffness twice as much for 0° as 90° mode (Fig. 2c). stiffness lowest when in-plane stress stresses chromatin quantified deformation of chromatin domains where green fluorescent protein (GFP) transgene DHFR (dihydrofolate reductase) (Fig. 2d). Mean Square Displacements (MSDs) of GFP spots changes distances highest for 90° stress mode intermediate in-plane lowest 0° mode computed stretching of chromatin domain containing DHFR gene tensile strains shear strains in-plane stress mode strains higher than 0° lower than 90° mode (Fig. 2g data showed tensile strains twice as much as shear strains dominant form chromatin deformation tensile stretching for in-plane mode result unexpected.contrast out-of-plane stress modes normal strains shear 90° cell cortex in-plane stress caused shear strains cell cortex Fig. 1 Table 1) inside nucleus chromatin domain deformation mainly tensile complex structural arrangements cytoskeleton nuclear lamins propagate surface stress into nucleus tensile deformation chromatin examined stress amplitudes in-plane mode chromatin deformation cell stiffness MSDs GFP spots increased with stress amplitude Fig 2a–d). bead twisting angle increased with stress amplitude cell stiffness change with amplitude in-plane stress chromatin deformation increased with in-plane stress amplitude structural changes chromatin direct mechanical responses data show in-plane stress induces cellular mechanical properties cell stiffness chromatin deformation out-of-plane stresses.Fig. 2Mechanical anisotropy cell chromatin induced by different force modes image adherent elongated CHO hamster ovary) cell with GFP labeled chromatin domain 4-μm RGD-coated ferromagnetic bead attached to cell surface via integrinsTheta represents bead rolling direction axis cell applies cells bead GFP spots white box enlarged right Scale bar 3 μm in-plane rotation Z-axis out-of-plane rotations X- Y-axes right-hand rule Fig. 1) Peak 2D displacements magnetic bead X–Y plane stress angles 0° 90° out-plane peak bead twisting angles in-plane peak amplitudes magnetic fields 15 Pa 0.3 Hz triangle represents one cell Cell stiffness computed stress modes P = 0.0079 between 90° in-plane modes P < 0.001 0° 90° image three GFP spots chromatin cell Scale bar 1 μm Normalized mean squared displacement) GFP spots stress (15 Pa 0.3 Hz applied 0° 90° in-plane mode No data spontaneous GFR movements force Chromatin deformation changes GFP spots depends stress modes P < 0.001 Tensile strains chromatin computed P = 0.0055 between 0° in-plane modes P = 0.0017 in-plane 90° P < 0.001 between 0° 90°Shear strains computed chromatin deformation. P = 0.00701 between 0° in-plane modes P < 0.001 between 90°. b c e–h mean ± s.e.m. n = 39 cells 29 experiments **P < 0.01; ***P < 0.001 P values calculated corrected using two-tailed Student’s t-test Bonferroni correction. Source data provided file.Gene upregulation varies with stress examined DHFR gene transcription different stress modes factors employed fluorescently labeled 5′-end probes detect transcription first 1700 bps DHFR mRNA20 partial transcripts before transcript (DHFR gene 34 kb long ~10 min stress applied at 15 Pa 0.3 Hz for 2 min DHFR transcription upregulated in-plane stress mode upregulation higher than 0° lower than 90° mode stress to 60 min all three stress modes time-dependent elevation transcription DHFR in-plane mode led stress-amplitude dependent elevation DHFR transcription chromatin deformation DHFR upregulation in-plane mode between 0° 90° wondered 45° stress modebead rotation long short cell similar responses in-plane mode cell stiffness 45° out-of-plane mode ~3 times in-plane mode chromatin deformations similar (Fig no difference in DHFR transcription upregulation between 45° in-plane mode 2 min or longer stress modes distinct cell stiffness impacts chromatin deformation DHFR transcription similar 3Transcription varies with stress modes increases duration DHFR transcription detected with 5′-end probes Stress cell once one angle FISH experiments labeled 5′-end probes first 1700 bps DHFR mRNA Controls stress cells same dish without magnetic beads stress applied 2 min at 15 Pa 0.3 Hz P < 0.001 between stress modes n = 115 97 344 85 cells for no stress 0° in-plane 90° stress modes nine experiments < 0.001 DHFR transcription upregulation depends on stress (15 Pa 0.3 Hz duration in-plane out-of-plane modesP 0.016 30 60 min 0° 0.025 0.0015 in-plane 0.0039 90° P < 0.001 conditions 0° 21, 33 52 cells 15 30 60 min in-plane 38 66 60 cells 15 30 90° 26, 25 32 cells 15 60 Three independent experiments *P < 0.05 **P < < 0.001 dashed line no stress control DHFR gene upregulation depends stress amplitudes in-plane mode stresses 0.3 Hz 30 min 71 cells 0 stress 47 50 68 66 cells at 6 9 12 15 Pa P = 0.032 between 0 stress 6 Pa 6 9 Pa 0.014 9 12 Pa P < 0.001 stress amplitudes < 0.05 < 0.01 ***P < 0.001 Cell stiffness stress modes P = 0.036 0° 45° 0.039 45° 90° 0.037 90° in-plane P < 0.001 0°-plane n = 11 cells eight experiments *P < 0.05 < 0.001 Chromatin deformation different stress modes No difference 45° in-plane mode P = 0.in-plane 45° stress modes P < 0.001 n 33 GFP spots 11 cells eight experiments < 0.001 not different Comparison DHFR upregulation 5′-end probes out-of-plane 45° in-plane mode stress 2 min 15 Pa 0.3 Hz P = 0.069 n 344 140 cells nine experiments not different line baseline transcription level force DHFR upregulation similar longer durations stress (15 Pa 0.3 Hz 45°-plane in-plane mode P = 0.44 0.14 0.4 in-plane 45° stress modes 15 30 60 min 45° n = 21, 29, 39 cells 15 30 60 min in-plane 38 66 60 cells 15 30 60 min Three experiments not different dashed line baseline transcription level force P values calculated corrected two-tailed Student’s t-test Bonferroni correction Source data fiber anisotropy determines chromatin deformation gene anisotropy cell stiffness chromatin deformation gene transcription upregulation filamentous actin cell F-actin inhibitor Latrunculin A actin stress fibersActin stress fibers decreased treated LatA 2 min short disorganized after 5 min LatA treatment cell rounding cell stiffness decreased at 2 min all stress modes by 5 min differences stiffness modes disappeared cell stiffness data consistent with loss structural integrity stress fibers chromatin GFP spots deformation increased after 2 min LatA treatment due stress fibers partially disrupted remaining forces to nucleus reduction F-actin Five minutes after LatA treatment force-induced chromatin deformation disappeared stress modes stress fiber integrity force-transmission pathway disrupted gene upregulation stress modes after LatA trend chromatin deformation 2 min after LatA transcription increased all modes anisotropy transcription stress modes no transcription upregulation after 5 min LatA treatment 4e results suggest actin stress fibers responsible for anisotropy cell stiffness chromatin deformation rapid gene upregulation under stress modes 4Actin stress fibers dominate cytoskeletal anisotropy-dependent cellular responsesCell stiffness stress modes before Latrunculin after (1 μM 2 5 min P = 0.0075 between No LatA +LatA 2 min under 45° P = 0.023 90° P 0.039 +LatA 2 0.0032 +LatA 2 5 min in-plane P < 0.001 conditions mode Mean ± s. n = 12 cells nine experiments. *P < 0.05 **P < 0.01 ***P < 0.001 Normalized displacement (MSD) GFP spots stress (15 Pa at 0.3 Hz) applied at 0° 45° 90° before after 2 5 min Mean ± s.e.m. n = 12 cells nine experiments images RNA FISH at no stress 0° 45° 90° out-of-plane Stress (15 Pa 0.3 Hz applied before Latrunculin A after 2 or 5 min 5 yellow dashed lines nucleus GFP chromatin domains DHFR FISH fluorescence white enlarged brightfield image cell scale bar 3 μm black dot magnetic bead cell surface nucleus highlighted dashed line scale bar top right panel 3 μmDHFR transcription varies LatA treatment in-plane out-plane modes P = 0.0023 between No LatA +LatA 2 min 0° P 0.0048 LatA +LatA 2 min in-plane stress P < 0.001 other conditions mode 0° 38 45 47 cells in-plane 58 89 41 cells 45° 35 52 40 cells 90° 28, 31, 62 cells No LatA +LatA 2 min +LatA 5 min three independent experiments cells stress not treated LatA **P < 0.01; ***P < 0.001 dashed line DHFR gene expression levels without stress P values calculated corrected two-tailed Student’s t-test Bonferroni correction Source data stress-mode cell stiffness chromatin in-plane stress mode lowest cell stiffness induces chromatin deformation 45° out-plane stress mode employed finite element model anisotropic cell stress fibers generated stress-mode dependent magnetic bead displacements chromatin deformation displacement magnetic bead short axis 90° stress mode 1.6 times long axis 0° 45° stress mode in-between valuescell stiffness anisotropy rotation angle magnetic bead under in-plane stress ~0.5 rad cell stiffness lower than out-of-plane stress modes results reproduced living cells Fig. 2 Fig. analyzed effects stress modes on chromatin deformation used approach three points in X–Y plane nucleus chromatin domain (Fig. calculated distance between points before after loading average values chromatin deformation tensile shear strain calculated FEM calculation reproduced live cell measurements (compare Fig. 5e–g Fig. 2f–h Supplementary Fig. 4) Tensile strains shear strains exhibited anisotropy ~twice as large shear strains chromatin strains in-plane stress mode similar to 45° out-of-plane stress mode.Fig. 5Finite models anisotropic responses cell 3D illustration location bead arrangement nucleus actin stress fibers model cell–bead system magnetic bead light blue nucleus dark blue cytoplasm light brown cell cortex brown cross-sectional illustration cell–bead system cell nucleus bead specific torque Tm stress 15 Pa) applied cell via beadestimated parameter values components cell from Young’s moduli membrane cortex cytoplasm nucleus 2 0.25 1.0 kPa Poisson’s ratios 0.3 0.49 0.3 actin stress stiffness 20 nN μm−1 Illustration load stress mode theta angle rolling magnetic bead axis cell magnified view boxed area nucleus chromatin deformation Magnetic bead displacements rotation angle Chromatin deformation tensile strain shear strain role structural F-actin cell stiffness anisotropy chromatin deformation actin stress fibers disassembled FEM inhibitory effect LatA F-actin experiments results treated LatA 2 min actin stress fibers partially disassembled 5 min completely destroyed (Fig. 6a–c). displacements magnetic bead after LatA treatment three stress modes Fig 6d fibers differences reduced results stress fibers key anisotropy cell stiffness average cell stiffness.Fig. 6FEM simulation effects disrupting stress fibers cell chromatin deformation No LatA control +LatA 2 min +LatA 5 min Cell shapes bead not drawn scaleEffects LatA treatment on magnetic bead displacements under stress modes Effects on chromatin deformation long distance transmission force into nucleus depends on actin stress fibers21 FEM stress fibers between nucleus cell surface partially (~30% disrupted after 2 min LatA constraints nucleus chromatin deformation elevated stress concentration fibers (Fig. 6e). stress fibers between magnetic bead nucleus after 5 min LatA treatment disrupted transfer load into nucleus chromatin deformation reduced impact stress modes existed (Fig. modeling data consistent with experimental results living cells Fig. FEM simulation results under out-of-plane complex stress modes actin stress fibers stretched or compressed by rolling motion magnetic bead in-plane stress mode not stretched tangential deflection large rotation magnetic bead larger cell deformation lower cell stiffness chromatin deformation under in-plane stress mode similar to 45° out-of-plane stress mode not obvious two stress modes similar impacts at distancenucleus consistent with idea stresses strains mediated by anisotropic cytoskeletal structures not continuous material mechanism performed FEM simulations nuclear deformation between in-plane 45° stress modes 5a nuclear deformation magnitudes distributions similar strains near site loading similar cell cortex near magnetic bead 45° out-of-plane stress mode induced higher normal strains in-plane in-plane higher shear strains 45° Fig. 1c d Table 1) simulation results suggest cell deformations by modes distinct at cell cortex stresses propagated into nucleus by in-plane 45° out-of-plane similar due stresses concentrated into nucleus deform chromatin similar deformation compared effects of out-of-plane complex stresses with in-plane stress on cell mechanical behaviors biological responses gene transcription substantial anisotropy in cell stiffness chromatin deformation gene transcription under different stress modes out-of-plane stress modes rolling magnetic bead along short axis cell 90° from stress fibers larger cell deformation than long axis or 45°in-plane stress leads to more cell deformation than out-of-plane stresses FEM simulation to causes little compression stretching stiff actin stress fibers tangential deflection less resistance in-plane stress mode induces similar chromatin deformation as 45° out-of-plane stress mode in living cells recapitulated by FEM simulations 45° in-plane similar strain maps gene transcription force modes modeled actin stress fiber as elastic rod 12 kPa mapped deformation 0° mode small deformation low nuclear strain 90° mode large deformation high strain 45° mode and in-plane mode led to intermediate stress fiber deformation medium nuclear strain magnitudes 45° mode in-plane mode generate different cell cortex strains similar stress fiber deformation patterns ensure similar nuclear strain maps complex stress distribution chromatin tensile strain low for 0° high for 90° medium for 45° mode and in-plane mode consistent with experimental data from living cells Figs. 2g 3e). rapid DHFR gene upregulation depends on chromatin stretching or tensile strainfindings suggest stress fibers long axis cell anisotropy cell force modes cause stress fiber deformation chromatin increase gene expression. 7A model stress fiber anisotropy regulating chromatin strain gene upregulation magnetic bead bonded cell surface via integrins focal complex structures schematics Y–Z X–Y view bead (4 μm cell stress fiber modeled elastic rod (Young’s modulus 12 kPa24 μm diameter Poisson’s ratio 0.4). twice as many stress fibers aligned long axis cell as short axis constant stress 15 Pa applied bead cell each force mode images stress fiber deformation von Mises strain maps nucleus under force modes bead removed image images chromatin strains under force modes wiggly line represents stretched chromatin domain DHFR gene locus ~2 μm from edge nuclear chromatin domain strain low (~5%) 0° mode high (~25%) 90° medium (~10%) 45° in-plane mode comparable with experimental data living cells Figs. 2 3)model suggests chromatin stretching gene upregulation depend on nuclear strain caused by stress fiber deformation nucleus under force modes color scale applies to images out-of-plane in-plane stress relevant force? endothelial cells blood experience fluid shear stresses on apical surfaces balanced at basal surfaces focal adhesions25 blood vessels experience blood pressure stretching complex strains in cells myosin-II mediated cellular at focal adhesions are complex stresses in-plane out-of-plane exploring effects stress modes on cellular functions physiologically relevant important understanding living cells tissues living cell’s cytoskeleton consists three filament systems actin filaments microtubules intermediate filaments actin filaments form stress fibers with myosin-II dominant role actin stress fibers in anisotropic responses Microtubules play roles in cell cells treated with colchicine no significant changes in cell stiffness chromatin deformation microtubules minor role in gene upregulation under stress modes cells Intermediate filaments in mechanical biological functions in cell stiffness at large deformations32future explore impact intermediate filaments in cell responses stress modes CHO cells elongated numerous stress fibers after adhesion spreading epithelial cell line findings other types cells stress fibers disrupted by LatA changes geometry bead–cell contact area adhesion influence cell stiffness measurement employed infecting living cells with talin-GFP focal adhesions complex structures cell basal surface Bead embedding calculated using largest diameter X–Y projected image talin-GFP fluorescence largest bead–cell contact interface bead–cell contact area estimated average intensity largest fluorescence ring structure talin-GFP bead index bead–cell contact area controls without laser scanning photobleaching after 5 min LatA treatment slight decrease (~10%) bead bead–cell contact area before F-actin disruption suggests same applied torque stress increased ~30% due 10% reduction contact area predicted ~30% elevation cell stiffness19 5 min after LatA treatment cell stiffness opposite trend reduced dramatically all force modesdata suggest change cell stiffness after LatA treatment likely due to cell deformation stress fiber disruption not changes bead–cell contact area adhesion between RGD-bead integrins single ligand–integrin bond yielding force is ~100 pN33 hundreds of integrin receptors interact with one 4-μm magnetic bead coated with RGD peptides bonds cell strong bead peeling off cell surface after stress fiber disruption bonds stable adhesion stress decreases ~25% after F-actin disruption with cytochalasin D treatment for 5 to investigate adhesion stress changes in force modes after disruption cell stiffness stretching distance of RGD-bead and integrins 0.1–1 nm before bond small to for bead displacements nanometers in live cells bead displacements when stress fibers disrupted by LatA explained by bead–cell bond adhesion alteration RGD-bead attached to cell surface via integrin–actin linkages tethered by F-actin bundles stress resistance to bead rotation under stress resistance index of cell stiffnessF-actin bundles disrupted tethering abolished resistance to bead rotation under stress reduced cell stiffness decreased suggest bead–cell contact area adhesion minor role in cell stiffness after stress fiber disruption understand effect long-range force propagation by actin stress fibers performed simulation treating cytoplasm actin cytoskeleton as anisotropic uniform elastic materials without modeling actin cytoskeleton continuous model) Fig 9) employed FEM discrete model 5a) simulate actin stress fibers as discrete stiff elements in cytoplasm calculate impact on cell chromatin deformation boundary conditions stress modes in continuous model same as discrete model results show continuous model generate similar bead displacements rotation angles cell deformation magnitudes chromatin deformation continuous model smaller than discrete model more localized deformation in cytoplasm nucleus model results due to fast decay of deformation field in uniform elastic body continuous model decays predicted St. Venant’s principle local force causes local deformation), load from magnetic bead transferred to nucleusstress fibers in discrete model transfer load to nucleus long distance to concentration consistent with previous single stress fiber discrete model appropriate for simulating long distance force propagation chromatin deformation of intact living cell myosin-II dependent endogenous prestress critical in long distance force propagation in cytoplasm nucleus20–23 treated cells with blebbistatin inhibit cell stiffness decreased after 50 μM blebbistatin treatment chromatin deformation decreased consistent with myosin-II inhibition reduces cell stiffness36 prestress mediates long force propagation in cytoskeleton nucleus21–23 gene upregulation by stress inhibited when myosin-II dependent stress inhibited with blebbistatin consistent with In current discrete FEM model impact endogenous prestress incorporated into stiffness not explicitly shown Complete disruption of stress fibers influence myosin-II prestress leads to total abolishment of chromatin deformation consistent with previous reports21 living cells are viscoelastic respond to loading frequenciesrecent report finds loading frequencies 0.3 6 Hz modest chromatin deformation gene upregulation force-induced gene upregulation weak power law depends on H3K9 demethylation37 current FEM elastic model recapitulated mechanical transcriptional responses cells at low loading frequency (0.3 Hz mechanisms FEM can include viscoelastic elements high loading frequencies 10–20 3D cell model crosstalk among cell adhesions cytoskeleton nucleus38 model simulate effects stress modes on cell stiffness chromatin stretching stress fiber anisotropy contributes to force-mode dependent cellular mechanical responses chromatin deformation gene transcription current study first attempt living cells respond complex forces.MethodsCell culture reagentsCHO DG44 DHFR D10 cells Dr. Andrew Belmont University cultured in Ham’s F12 media without thymidine hypoxanthine 10% fetal bovine serum 1% Penicillin–Streptomycin authentication cell line confirmed by visualizing GFP-lac repressor staining patternsCells passaged 3 days TrypLE centrifuging cells dispersed 1 mL medium 200 μL cell suspension added sterile six-pore plate 3 mL medium precoated 0.1% gelatin. 354236 Cells cultured 37 °C 5% CO2 incubator 2 days before experiment DAPI (4′ 6-diamidino-2-phenylindole) staining cell nuclei mycoplasma contamination no signs mycoplasma Cells dishes randomly assigned LatA purchased Dalian Meilum Biotech Co. Colchicine MedChem Express Co. Blebbistatin ApexBio Tech LLC B1387) F-actin stained 4 h SiR-actin KIT Cytoskeleton Talin-GFP Thermo Fisher Scientific C10611) talin-F-actin double-staining labeled SiR-actin KIT 4 h after Talin-GFP reagent 16 h Reagents FISH 20× SSC (saline sodium citrate) AM9763 10× PBS (phosphate buffered saline BSA Ribonucleic acid Dextran Sulfate Sodium Deionized Formamide 16% Formaldehyde Ethanol absolute10009218 Sinopharm Co. Salmon fibrinogen SEA-133) (SEA-135) purchased from Pfenex Inc. (CA USA).3D magnetic twisting cytometry3D MTC18 stresses direction (X Y Z via rotational movements ferromagnetic beads (Boston magnetic beads purchased J. Fredberg, Boston MA attached cell membrane beads magnetized Z direction applied twisting magnetic field X or Y direction beads rotate X–Y plane short long axis cells out-of-plane stress mode beads magnetized X direction applied twisting field Y direction rotate X–Y plane generate stresses in-plane stress mode modes force cell culture dish cells beads rotated long axis cell magnetic bead aligned Y-axis bead magnetized Z-axis twisting field 15 Pa applied Y-axis bead rotation direction X-axis 0° out-of-plane mode same bead re-magnetized Z-axis twisting field 15 Pa applied X-axis bead rotated Y-axis 90° out-of-plane modebead re-magnetized Z-axis twisting field 10.6 Pa applied X-axis Y-axis bead rotated angle 45° X Y-axes sum two vector magnitudes 15 Pa 45° out-of-plane mode in-plane mode re-magnetizing bead X-axis twisting field 15 Pa Y-axis different modes force same same cell quantify stiffness chromatin deformation out-of-plane modes bead displacements measured quantifying 2D projections center X–Y plane in-plane mode bead rotation angle measured quantifying 2D rotation angle bead edge gray pixels cell surface reference measured distance between bottom bead rigid cell substrate distances 3.42 ± 0.20 μm (maximum 5.82 μm minimum 1.32 μm) Fig. distance below 1 μm substrate stiffness cell stiffness measurement impact rigid substrate for CHO cells not substantial Stress cell surface proportional to magnitude twisting magnetic field 10, 15 20 25 Gauss magnetic field corresponded to 6, 9 12 15 Pa stress cells calibrated viscous fluid standard14 bead magnetic moment constant was 2 Pa Gauss−1.applied stress equal to bead magnetic moment constant times field divided by 6. MTC coils generate 25 Gauss per 100 turns X Y Z direction 3D MCT 180 turns gain 1.8 in magnetic field for coils 25 Gauss magnetic field stress MTC 15 Pa (2 times 25 times 1.8 divided by 6) bead magnetic moment constant magnetic bead in elastic material fibrin gels salmon fibrinogen 2 or 4 mg mL−1 50 μL activated by (1 μL 100 shear elastic modulus 60 or 140 Pa stress applied at 6 9 12 15 Pa bead in-plane stress mode at 0.3 Hz bead angular rotation (strain) quantified calculated shear elastic modulus of 2 mg mL−1 fibrin gel 65 Pa 4 mg mL−1 fibrin gel 148 Pa close to published values results suggest calibrated bead magnetic moment constant accurate stiffness measurement 3D MTC reliable magnetic beads coated overnight with (50 μg mL−1 per Arg-Gly-Asp peptides Beads cells co-incubated for 20 minone bead per cell 35 mm petri dish 18 mm glass-bottomed well (Cat. No. GBD00003-200 added low (30 μL 1 mg mL−1 beads each chromatin deformation small 0° out-of-plane stress mode selected beads ~4–5 μm from GFP spots deformation four stress modes 45° frequency magnetic field 0.3 Hz Cell stiffness calculated from bead displacements rotation different degrees embedding embedding coefficients beads embedding 36.3% ± 0.7% (n = 39 cells 45% min = 26%) coefficients β = 0.8 α = 0.3 cell stiffness out in-plane modes Stellaris FISH Probes designed against DHFR mRNA DHFR genes transcription CHO DG44 cells applied stress amplitude fixed 3.7% formaldehyde 30 min permeabilized 70% ethanol 12 h 4 °C Samples hybridized with DHFR Stellaris FISH Probes incubated dark humidified incubator 12 hWashing 2× 1 mL buffer 37 °C 30 min imaging Leica DMI6000B microscopy 2 mL 2× SSC Image J FISH images imagingLeica DMI6000B 63 × 1.4 NA oil-immersion objective FISH fluorescence software Leica SP8-STED microscopy 100 × 1.4 NA oil-immersion objective visualize chromatin GFP spots CHO cells Leica Application Suite X software Chromatin deformation measured distances X–Y GFP spots Z-direction displacement not measured underestimate CHO cells spread cell height Z-axis less X–Y changes distances small Leica SP8-STED microscopy GFP spots Talin-GFP nm SiR-actin 633 nm Fluorescence detected Hybrid detector.Quantification bead–cell contact CHO cells cultured 48 h glass-bottomed 35-mm dishes precoated 0.1% gelatin 20 μL medium CellLight® Talin-GFP fusion human talin emGFP added each 35-mm dish 16 h experiment insect virus baculovirus mammalian cells accurate targeting cellular talin-GFPBead embedding calculated using largest diameter X–Y projected image talin-GFP fluorescence largest bead–cell contact interface contact area estimated X–Y area average intensity largest fluorescence ring structure-GFP surrounding bead index bead–cell contact area temporal controls without LatA laser scanning photobleaching Adjustment thresholds same level images eliminate background noise bead–cell contact areas varied cells after LatA treatment normalized before drug treatment.Dynamic tracking image analysisMATLAB software data image downloaded element FEM cell four parts membrane cortex cytoplasm nucleus cytoskeleton (Fig. first three parts discretized by 3D bulk elements cytoskeleton discretized by spring living CHO cells cell length 41.79 ± 0.98 μm 10.68 ± 0.34 μm height 8.19 ± 0.17 μm (n = 39 cells consistent length width thickness cell 40, 12, 8 μm bead embedding 38% experimentally embedding 36.3% thickness cell membrane cortex42 0.25 μmnucleus modeled elastic ellipsoid long axis 12 μm short 7 μm anisotropic arrangement cytoskeletal structure cell polarization actin stress fibers long axis twice short axis magnetic bead discretized by 3D bulk element adhesion with cell membrane cortex modeled bonded contact interactions anchor points actin cytoskeleton contact magnetic bead with cell surface few μms from end nucleus loading stable adhesion assumed between cell substrate fixed connection mesh refined near magnetic bead Young’s modulus 2 × 105 generating 129,543 elements 208,144 nodes actin stress fiber deformation modeled elastic rods diameter 0.4 μm Poisson’s ratio 0.4 Young’s modulus 12 kPa24. parameters same as elements von Mises equivalent strains nucleus computed under force modes.Statistical analysisTwo-tailed student’s t-test used for analyses except multiple comparisons Bonferroni correction.Supplementary | 47.8 | 1.059247 |
10.1038/s41467-021-21416-9 | PMC7902626 | Subduction of oceanic crust and sediments contributes to heterogeneities in the mantle, which are sampled by mantle plumes. Here, the authors find that extreme isotopic heterogeneity in Samoan clinopyroxenes can help constrain the composition of mantle sources containing sediment recycled into the Earth’s mantle. | Lavas erupted at hotspot volcanoes provide evidence of mantle heterogeneity. Samoan Island lavas with high 87Sr/86Sr (>0.706) typify a mantle source incorporating ancient subducted sediments. To further characterize this source, we target a single high 87Sr/86Sr lava from Savai’i Island, Samoa for detailed analyses of 87Sr/86Sr and 143Nd/144Nd isotopes and major and trace elements on individual magmatic clinopyroxenes. We show the clinopyroxenes exhibit a remarkable range of 87Sr/86Sr—including the highest observed in an oceanic hotspot lava—encompassing ~30% of the oceanic mantle’s total variability. These new isotopic data, data from other Samoan lavas, and magma mixing calculations are consistent with clinopyroxene 87Sr/86Sr variability resulting from magma mixing between a high silica, high 87Sr/86Sr (up to 0.7316) magma, and a low silica, low 87Sr/86Sr magma. Results provide insight into the composition of magmas derived from a sediment-infiltrated mantle source and document the fate of sediment recycled into Earth’s mantle. | IntroductionSubduction zones are the largest recycling systems on Earth where sediments, oceanic crust, mantle lithosphere, volatiles, and possibly blocks of continental crust return to the mantle. It is generally accepted that mantle plumes ‘sample’ such recycled materials leading to a diverse range of geochemical signatures reflected in ocean island basalts (OIBs)1–3. Mantle isotopic heterogeneities are inferred by examining global OIB lavas, but the scale, origin, and compositions of the sources and/or derivative melts remain under intense investigation. Many studies that have evaluated isotopic heterogeneity in OIBs have done so from the analysis of whole rock or bulk mineral separates4–8, but these types of measurements obscure possible broader isotopic heterogeneity present within individual crystals or melt inclusions within a single lava. Previous work has shown 87Sr/86Sr and 143Nd/144Nd isotopic heterogeneity in lower oceanic crust clinopyroxene and plagioclase cumulate crystals, and another work has reported 87Sr/86Sr heterogeneity in olivine-hosted melt inclusions from a single lava sample—these are a couple of examples among many studies showing isotopic heterogeneity in individual crystals or melt inclusions9–14. Related to this study, recent work demonstrated intra- and inter-crystal 87Sr/86Sr heterogeneity in plagioclase in Samoan lavas with an enriched mantle 2 (EM2)15 signature16. Collectively, these studies suggest that multiple isotopically distinct melts can contribute to a single lava, and thus magma mixing13 is an essential part of igneous petrogenesis.Subduction recycling of terrigenous sediments into the mantle has been cited to explain the EM2 geochemical signature recorded in lavas erupted at ocean island volcanoes7,17. Lavas with elevated 87Sr/86Sr (>0.706)18 paired with low 143Nd/144Nd, sampling an EM2 component, are relatively common among the volcanoes of the Samoan Islands in the South Pacific Ocean. Lavas erupted off the submarine flanks of Savai’i Island, western Samoa (Supplementary Fig. 1), have whole rock 87Sr/86Sr values up to 0.720469 (ref. 7), plagioclase up to 0.7224 ± 0.0003 (2SE), and plagioclase intra-crystal variability of 87Sr/86Sr >5000 p.p.m.16. These studies have invoked two-endmember mantle mixing as the mechanism generating heterogeneity in Samoan lavas, suggesting up to 7% addition of a continental crust-like sediment in the Samoa plume source16. This study traces 87Sr/86Sr heterogeneity to a finer scale by obtaining coupled, high-precision Sr and Nd isotopes and major and trace elements on individual clinopyroxene crystals from the second-most (at the time of this study) enriched (high 87Sr/86Sr) Samoan whole rock lava (whole rock 87Sr/86Sr = 0.718562).Here we show that extreme 87Sr/86Sr heterogeneity observed in individual clinopyroxenes from a single geochemically extreme oceanic lava is consistent with magma mixing of at least two isotopically distinct melts—one basaltic and one trachytic. Further, the clinopyroxenes studied are found within one of the two Samoan lava samples in which Edwards et al.16 measured plagioclase 87Sr/86Sr profiles from single porphyrocrysts. Consequently, a direct comparison of the extreme isotopic heterogeneity observed in the clinopyroxenes of this study and previously published plagioclase 87Sr/86Sr within a single hand specimen provides new constraints on the source of isotopic heterogeneity within the context of magma mixing and constraints from phase equilibria. Moreover, these new isotopic constraints help shed light on the origin of terrigenous sediment in the Samoan plume source and, thus, the deep-mantle residence of subducted terrigenous material in large low shear wave velocity provinces (LLSVPs)19,20.Results and discussionAn extreme EM2 sample from SamoaThe studied sample is a 5 Ma submarine lava21 ALIA-D115-18, dredged off the flanks of Savai’i Island, western Samoa (see Supplementary Fig. 1). This sample is a nepheline-normative trachyandesite with a bimodal crystal population. The larger crystal population is ≥25 μm in diameter and includes porphyrocrysts of clinopyroxene, plagioclase, and ilmenite as well as porphyroclasts consisting of the same three phases with modal abundances corresponding to a microgabbro. This study focuses on clinopyroxene porphyrocrysts, which were analyzed for major (electron microprobe) and trace elements (laser ablation inductively coupled mass spectrometry (LA-ICP-MS)) prior to dissolution and isotopic analysis by thermal ionization mass spectrometry (TIMS). Seventeen individual clinopyroxene crystals were analyzed for 87Sr/86Sr with resulting ratios ranging from 0.716833 (±0.000015, 2SE) to 0.723888 (±0.000015, 2SE), the highest value ever reported in an oceanic lava. Nine of the 17 crystals were also targeted for 143Nd/144Nd analysis, and single-crystal values range from 0.512238 (±0.000013, 2SE; εNd = −7.6 calculated using 143Nd/144Ndchondrite = 0.512630 (ref. 22)) to 0.512357 (±0.000013, 2SE; εNd = −5.3) (Supplementary Table 1). Together, the clinopyroxene Sr and Nd isotopes plot along an extension of the mantle array formed by global OIB, and including the new clinopyroxene data of this study, the 87Sr/86Sr variability in this single lava sample spans ~35% of the variability observed in all Samoan lavas, and ~30% of the entire oceanic mantle variability (Fig. 1).Fig. 1143Nd/144Nd versus 87Sr/86Sr of new ALIA-D115-18 single clinopyroxene data and prior data on the whole rock and bulk (100s of grains of) clinopyroxene.Fields for Samoan shield lavas (excluding lavas from Savaiʼi Island), and global OIB and mid-ocean ridge basalts, are shown in orange and gray fields, respectively. Other Samoan shield lavas from the ALIA dataset (i.e., submarine shield lavas from Savaiʼi Island) are shown for reference as well (i.e. ALIA-D114: orange triangles, ALIA-D115: red squares, ALIA-D118: orange circles, and ALIA-D128 lavas: orange diamonds, and the new clinopyroxene data, denoted as yellow crossed squares, collected in this study)7. The inset shows a zoomed view of the clinopyroxene data in addition to the bulk clinopyroxene analyses (representing 100s of pooled grains; gray squares with an “x”) and the ALIA-D115-18 whole rock (red square) published previously7,8. All error bars smaller than symbols except where shown and represent 2σ standard errors. Note that only nine of the 17 clinopyroxenes analyzed for 87Sr/86Sr also have 143Nd/144Nd data, so not all clinopyroxenes analyzed for 87Sr/86Sr can be shown on this figure. However, the nine clinopyroxenes analyzed for both Sr and Nd isotopes encompass the full range of 87Sr/86Sr observed in this study. Samoan shield lavas, ALIA lavas, and Savai’i xenolith compositions are from elsewhere5–7,69. The black dashed line is a theoretical magma mixing line derived in this study using binary mixing theory; mixing endmember compositions are provided in Supplementary Table 2, and methods for calculating endmembers are discussed in the text. Each hatch is a 10% increment except for the first two hatches, which are in 5% increments. The green star denotes the EM2-derived, high 87Sr/86Sr, trachytic mixing endmember calculated in this study (see Supplementary Table 2).Generation of isotopic heterogeneityData from clinopyroxene Sr and Nd isotopes from lava sample ALIA-D115-18 of this study, together with major and trace element and isotopic constraints from other Samoan whole rock lavas from the same submarine dredge as ALIA-D115-18 (referred to as the ALIA-D115 dredge), are used to constrain the composition of the magma mixing endmembers—including the extreme EM2 endmember—contributing to the isotopic variability observed in clinopyroxenes from ALIA-D115-18. This permits us to better define the origin and makeup of the high 87Sr/86Sr EM2 endmember in the Samoan plume, thereby constraining the petrogenesis of EM2 lavas.In 87Sr/86Sr and 143Nd/144Nd space, addition of clinopyroxene data obtained in this study to the existing Samoan whole rock data requires a mixing line that exhibits more significant curvature (i.e. r < 1, where r = [Nd/Sr]M/[Nd/Sr]S23; where M [mafic] and S [silicic] represent the low- and high-87Sr/86Sr endmembers, respectively21) than previous suggestions based on whole rock data alone (see ref. 4, Fig. 1). This implies that the Sr and Nd concentrations as well as the 87Sr/86Sr and 143Nd/144Nd values in the two endmembers need be re-evaluated. It should be noted, however, that the proposed mixing discussed here is not mixing between mantle sources, but mixing between magmas, and therefore differs from the approach in Jackson et al.7. Employing an inverse model approach to existing whole rock data from Samoa (see “Methods”), we argue for an EM2-derived mixing endmember that is not only high in 87Sr/86Sr, but also trachytic in composition. Mixing of the EM2 trachytic magma with more primitive Samoan basalt-like compositions reproduces the mixing trends observed in ALIA-D115 whole rock lavas (Fig. 1) as well as the range of 87Sr/86Sr and 143Nd/144Nd in Samoa clinopyroxene.The genesis of this high 87Sr/86Sr, trachytic magma is an outstanding question and is further explored in the discussion and Supplementary Text. This aside, results of this study are consistent with the observation of Jackson et al.7 who noted a distinct positive correlation between 87Sr/86Sr and SiO2 concentration among Samoan whole rocks from the ALIA-D115 dredge haul, suggesting two-endmember mixing between a highly evolved (high silica, low MgO) endmember with high 87Sr/86Sr and a less evolved (basaltic) endmember with low 87Sr/86Sr. In addition, Edwards et al.16 noted a negative correlation between 87Sr/86Sr and Sr concentration in plagioclase crystals suggesting the high 87Sr/86Sr (the “EM2” component) endmember likely experienced significant loss of Sr due to plagioclase fractionation prior to mixing, consistent with the “EM2” component being highly evolved prior to mixing. If extreme plagioclase fractionation of the magma with radiogenic 87Sr/86Sr did occur prior to mixing, then a negative correlation should also be observed in the clinopyroxene since they are a direct result of the mixing process. However, there is no clear correlation between clinopyroxene 87Sr/86Sr and Sr concentration, which can be best explained if plagioclase fractionation occurred after the onset of mixing. This conceptual model is consistent with MELTS24–26 and Magma Chamber Simulator (MCS)27 mixing modeling, which shows that clinopyroxene crystallization begins ~50 °C prior to plagioclase saturation upon mixing of a mafic, low 87Sr/86Sr melt and a silicic, high 87Sr/86Sr melt. This scenario is also consistent with the observation that the range of 87Sr/86Sr in the clinopyroxenes (0.71683–0.72389) exceeds that of the range of plagioclase 87Sr/86Sr values found in the same hand specimen (0.71791–0.72239) by Edwards et al.16 (Fig. 2), suggesting that upon plagioclase crystallization, the system may have been more homogenized relative to an earlier time when clinopyroxene crystallization initiated.Fig. 287Sr/86Sr values for single clinopyroxene grains measured in this study with LA-ICP-MS spot analyses from three plagioclase grains from the same sample (ALIA-D115-18)16.Previous measurements of bulk clinopyroxene8 (representing 100s of pooled grains; gray squares with an “x”) from this same sample are shown together with the whole rock value7 (where WR stand for whole rock; red square), clinopyroxenes measured in this study (pink squares with an “x”), and plagioclase analyses from Edwards et al.16 (blue diamonds, circles, and triangles); all error bars smaller than symbols except where shown and represent 2σ standard errors. In the histogram, the number of clinopyroxene observations (n = 17; red bars and red figure axis label) span a larger range of 87Sr/86Sr than the plagioclase 87Sr/86Sr analyses16 (n = 157; blue bars and blue figure axis label; error bars are 2 standard errors); where the clinopyroxene and plagioclase histograms overlap, a purple shade is used. The wider range of clinopyroxene 87Sr/86Sr values compared with the plagioclase from the same sample is consistent with magma mixing modeling using MELTS24–26 and the Magma Chamber Simulator27. The magma mixing models show that magma mixing between a mafic endmember and a more silicic, trachytic endmember (as derived in this study) precipitates clinopyroxene about 50 °C prior to plagioclase saturation and thus the mixing magmas may be slightly more homogenized at this point, explaining the reduced range of 87Sr/86Sr exhibited by plagioclase (see Main text and “Methods” for more detail).Figure 3 shows geochemical data from the ALIA-D115 dredge whole rock lavas (red symbols) in addition to shield lavas (orange symbols) and pillow glasses (purple symbols) from neighboring Samoan islands and seamounts. Mixing is clearly defined by the ALIA-D115 dredge lavas in 87Sr/86Sr versus element concentration space (Fig. 3), and in major and select trace element space (Supplementary Figs. 2 and 3), spanning compositions varying from trachybasaltic to trachyandesitic in a single dredge: curved lines define mixing in the ALIA-D115 lavas, while the rest of the Samoan lavas show zero-slope fractional crystallization trends.Fig. 387Sr/86Sr versus element concentration showing theoretical mixing curves of inferred endmember compositions.The direction of fractional crystallization is denoted by arrows. The green star denotes the EM2-derived mixing endmember composition (SiO2 = 65.5 wt%, MgO = 0.5 wt%, Al2O3 = 17.1 wt%, FeO = 3.8 wt%, CaO = 2.7 wt%, TiO2 = 1 wt%, and Na2O + K2O = 11 wt%; see Supplementary Table 2), which is calculated in this study (see “Methods”); the black hatched line shows the path of the calculated mixing curves, where each hatch increases by 10% mixing from the geochemically depleted mafic endmember to the more evolved EM2-derived endmember (black hatched line is derived from the same methodology as described briefly in Fig. 2 and in the “Methods”). Mixing is clearly defined by a curve with a non-zero slope, and is unlike the lavas that have been affected only by fractional crystallization (which have zero-slope trends). The ALIA-D115 lavas are represented by red squares, the other Samoan whole rock lavas by orange diamonds, and Samoan pillow glasses by purple circles. See Fig. 1 caption for data references; Samoa pillow glass data can be found in Workman et al.6. All errors bars smaller than symbols.The Sr, Nd, and major element oxide concentrations, as well as the 87Sr/86Sr and 143Nd/144Nd of the two endmembers that anchor the ALIA-D115 mixing array are reconstructed using a mixing analysis. Note that we do not use the clinopyroxenes for reconstructing the endmembers because they show ill-defined mixing trends with respect to major and trace elements. This is likely due to (1) fractionation of trace elements during crystal growth and (2) the clinopyroxenes being zoned such that averaged spot analyses of trace element concentrations (determined by averaging multiple LA-ICP-MS spots on each crystal; see “Methods”) will not correspond directly with grain surface major elements (obtained by extracting average major element concentrations from EPMA maps; see “Methods”) and will thus produce spurious trends, especially when compared to radiogenic isotopes (which were measured on dissolved crystals). However, these issues will not affect the isotopes, which is why the mixing trend is evident for 87Sr/86Sr and 143Nd/144Nd. The method for defining the endmember compositions using the ALIA-D115 lava trends (i.e., to reconstruct the composition of both endmembers), in terms of major elements and Sr and Nd concentrations, requires initial estimates for the MgO content of the endmembers and linear regression analysis of the ALIA whole rock bulk compositions in Ci/MgO versus 1/MgO space (where Ci is the concentration of a major element oxide or a trace element; see detailed description in the “Methods” section). MgO is estimated, ab initio, because it can be best constrained in both mixing endmembers given the limited variability at both ends of the mixing array formed by the ALIA-D115 lavas as observed when comparing SiO2 versus MgO, for example (see Supplementary Fig. 2). In comparing SiO2 versus MgO concentrations, the ALIA-D115 lavas show a linear mixing array that extends from a group of mafic Samoan shield lavas to a hypothetical endmember extending toward 0 wt% MgO. An averaged Samoan shield lava value is used for the MgO concentration of the mafic mixing endmember and we test three different MgO concentrations for the silicic mixing endmember from the lowest value observed in ALIA-D115 lavas (3.8 wt% MgO) to 0.5 wt% MgO (see “Methods” for more details). Further, according to mixing theory, 87Sr/86Sr versus Nd/Sr concentration and 143Nd/144Nd versus Sr/Nd concentration (companion plots to the hyperbolic mixing relation in 87Sr/86Sr versus 143Nd/144Nd space23) should give rise to linear arrays in a system described by two-component mixing. Thus, linear regressions were performed using these linear arrays, and given the derived Sr and Nd concentrations in each endmember, estimates of 87Sr/86Sr and 143Nd/144Nd were obtained for both endmembers (see “Methods” for more detail, Supplementary Figs. 4 and 5 for regressions, and Supplementary Table 2 for calculated endmember compositions). In addition, as proof of concept, this linear regression method was applied to two other select trace elements (Nb, Th) pertinent to generating a ratio useful for tracking source mantle characteristics (Nb/Th28; see Supplementary Fig. 8 and Supplementary Table 2 for concentrations in the endmembers).Given an initial assumption of 10 wt% and 0.5 wt% MgO in the low 87Sr/86Sr (mafic) and high 87Sr/86Sr (silicic) endmembers, respectively (see “Methods” for justification), we calculate 380 p.p.m. Sr and 63 p.p.m. Nd in the high 87Sr/86Sr EM2 endmember magma (see Supplementary Table 2 for composition of two other high 87Sr/86Sr endmember calculations with different starting MgO concentrations) and retrieve values of 87Sr/86Sr = 0.73158 and 143Nd/144Nd = 0.51213 (εNd = −9.8); the calculated major element composition of this EM2 endmember magma, which is trachytic, is provided in Supplementary Table 2. We calculate 712 p.p.m. Sr and 48 p.p.m. Nd in the low 87Sr/86Sr endmember and retrieve values of 87Sr/86Sr = 0.70610 and 143Nd/144Nd = 0.51280 (see Supplementary Table 2). Figures 1 and 3 illustrate mixing curves (black lines) using these values showing good agreement with the ALIA-D115 lavas and the clinopyroxenes of this study.The percentage of mixing observed in 87Sr/86Sr versus 143Nd/144Nd space (Fig. 1), consistent with the other major element and isotope plots (Fig. 3 and Supplementary Figs. 2 and 3), suggests that the most silica-rich ALIA-D115 lava (sample ALIA-D115-21) is composed of ~70–80% of the calculated EM2 endmember melt and ~20–30% of the mafic melt endmember.Preservation of isotopic heterogeneity in the magma chamberRecent work has shown zoning in pyroxene to be a reliable indicator of magma mixing events29,30. Here, Mg and Fe zoning in clinopyroxene is used to assess magmatic residence times post magma mixing (where magmatic residence time corresponds to the time needed to eradicate anomalous compositional lamella in clinopyroxene), given relevant temperatures and diffusivities. The eradication of compositional anomalies in the form of thin growth lamella can be computed following a standard one-dimensional diffusion analysis31 (details are presented in the “Methods” section). Attention is focused on the most mobile elements in clinopyroxene that exhibit zonation and for which diffusivity data exist at the inferred temperature range relevant to their petrogenesis. Fe/Mg inter-diffusion in clinopyroxene is ideally suited in this analysis, and we select two magmatic temperatures for analysis: 1200 °C (log10
DMg/Fe inter-diffusion = −18 m2/s)32, and 1150 °C (log10DMg/Fe inter-diffusion = −19.4 m2/s)33. The thinnest Fe and Mg lamella observed in clinopyroxene is about 15 μm thick (see Supplementary Fig. 6). The preservation of Fe and Mg lamellae in clinopyroxene suggest magmatic residence times of <45 years at a temperature of 1150 °C to <2 years at 1200 °C. Although there are uncertainties in the assumed temperature and in the experimental values of Fe/Mg inter-diffusion in clinopyroxene, we argue these estimates are accurate to within an order of magnitude, and potentially within 20–30%, based on experimentally determined activation energies and reasonable estimates for magma temperatures as constrained by phase equilibria34. The elemental concentration field observed in the clinopyroxenes exhibit an irregular zoning pattern, and along with the 87Sr/86Sr zoning in plagioclase from Edwards et al.16, are suggestive of at least one magma mixing event. The preservation of elemental zoning in clinopyroxene suggests that the eruption occurred relatively soon after magma mixing. The short timescales between mixing of geochemically distinct melts and the eruption of the mixture was sufficiently short for preservation of the heterogeneous 87Sr/86Sr and 143Nd/144Nd in the magmatic clinopyroxenes. Hence, the observed isotopic heterogeneity reflects the heterogeneity represented by the two magmas that contributed to the mixture.Thermodynamic modeling of the evolution ALIA-D115 lavasUsing the two endmember compositions described above—the mafic, low 87Sr/86Sr and the trachytic, high 87Sr/86Sr endmembers—thermodynamically constrained mixing simulations using MELTS24–26 and the MCS27 at ~0.2 GPa best capture the major element and isotope characteristics of the Samoan lavas (see Supplementary Figs. 2 and 3) and validate the mixing endmembers proposed in this study. In addition, the mixing calculations approximately reproduce the dominant mineralogy of clinopyroxene (Mg#71–80; natural clinopyroxene range from Mg#71–82 (ref. 8); see Supplementary Table 1) and plagioclase (~An45; natural plagioclase range from An50–60 (ref. 16)) as observed in the ALIA-D115-18 lava (see “Methods” for model parameterizations). It is important to note that the MCS mixing lines plotted in Supplementary Figs. 2 and 3 (plotted in gray) are not expected to follow the same path in compositional space as the closed-system binary mixing models, nor end at the “EM2”-derived endmember (see “Methods”). This is because the former tracks the predicted melt composition whereas the latter tracks the bulk composition of the mixture. It is expected that mixing of two (or more) magmas would produce crystal populations with heterogeneous major, trace, and 87Sr/86Sr and 143Nd/144Nd compositions since the hybridization (complete mixing or homogenization of two or more magmas) of the mixing endmember magmas is not instantaneous and most likely incomplete, even after decades of magma co-mingling13 (where co-mingling refers to incomplete mixing between two or more magmas producing discrete compositional bands in the actively mixing magmas).Origin of EM2 and derived meltsThe petrogenetic modeling presented here shows that a silicic endmember EM2 magma is consistent with the geochemical array formed by the ALIA-D115 lavas from Samoa. An important outcome of this effort is the determination that the extreme EM2 magma endmember in Samoa is trachytic in composition. The origin of the EM2 endmember is still an important question to be answered. Previous work has shown that even though EM2 has high 87Sr/86Sr and EM2 lavas at Samoa are relatively evolved, EM2 is not the result of shallow contamination of crustal material or sediments. Oceanic crust has insufficiently radiogenic 87Sr/86Sr to produce the 87Sr/86Sr observed in the ALIA-D115 lavas, and furthermore, Pb isotopic compositions of modern marine sediments from the Samoan region compared with the ALIA-D115 lavas form a non-overlapping, diverging trend such that the highest 87Sr/86Sr lavas are furthest from the marine sediment field7,16. Supplementary Fig. 8 shows that Nb/Th of the ALIA-D115 lavas decreases with increasing 87Sr/86Sr, consistent with magma mixing between a mafic, low 87Sr/86Sr (high Nb/Th) source and one that has high 87Sr/86Sr and low Nb/Th ratios (Nb/Th = 1.8 for the low MgO EM2-derived endmember of this study) similar to upper continental crust (Nb/Th = 1.1 (ref. 35)) or average sediments (Nb/Th = 1.2 (ref. 36)). However, as stated above, neither of which are modern materials, but rather ancient subducted materials.While the presence of this trachytic endmember (with 87Sr/86Sr up to 0.73158) is inferred from the mixing calculations presented in this study, it has not (yet) been found in pure form in lavas examined from the ALIA-D115 dataset. However, geochemical interrogation of the highest 87Sr/86Sr Samoan lavas—ALIA-D115-18 and ALIA-D115-21 (with whole rock 87Sr/86Sr of 0.718592 and 0.720469, respectively)7—reveals that single clinopyroxene crystals (87Sr/86Sr up to 0.723888; this study, Supplementary Table 1) and zones in plagioclase (87Sr/86Sr up to 0.7224)16 sample even more extreme EM compositions than the host whole rocks and trend toward the endmember EM2 melt calculated here. More intriguingly, reports of high-SiO2 (65 to 69 wt%) trachytic melt inclusions37 hosted in the same ALIA-D115-18 clinopyroxene grains examined in this study raise the possibility that the EM2 endmember may be preserved in relatively pure form as glassy melt inclusions in the clinopyroxenes from a high 87Sr/86Sr lava (see Supplementary Table 2 for example melt inclusion compositions). Unfortunately, these melt inclusions are too small for analysis of 87Sr/86Sr. Nonetheless, the presence of trachytic melts representing the EM2 endmember composition raises important questions about the origin of melts sampling this source. The fact that the model presented here indicates ~20% of a mafic magma mixed with ~80% trachytic EM2 endmember magma suggests that the potential conditions of mixing of silicic and mafic magmas may be due to an initially isolated silicic magma body that was recharged by a more mafic magma, where clinopyroxene and plagioclase crystallization (and melt inclusion entrapment) occurs during the mixing process. This scenario is consistent with the highly variable nature of 87Sr/86Sr observed in the clinopyroxene and plagioclase crystals. The question still remains, what is the origin of these silicic, trachytic melts?Previous studies provide evidence of silicic melts in the mantle derived by various processes, such as low-degree partial melting of anomalous mantle, reaction between CO2-rich fluids/melts or basaltic to silicic melts and peridotite, or as the result of extensive fractional crystallization5,38–48. One hypothesis tested in this study is partial melting of geochemically enriched (with high 87Sr/86Sr) mantle compositions followed by crystal fractionation to produce trachytic melts. Considering a scenario in which ancient subducted sediment has infiltrated and mixed with ambient peridotite, partial melting of this source could give rise to silicic melts with radiogenic 87Sr/86Sr. Results (described in detail in the Supplementary text) show that partial melts (at a range of pressures) of various mixed sediment+peridotite sources are consistently basanitic (high-degree partial melts) to phonolitic (low-degree melts) and subsequent fractionation of these melts does not produce trachytes. However, we acknowledge that we have not undertaken an exhaustive search of parameter space (e.g., different sediment compositions, different combinations of sediment±recycled oceanic crust±peridotite, various H2O and CO2 source concentrations, and a full range of fO2, pressures, and temperatures), and future work will focus on melting and fractional crystallization processes in the petrogenesis of compositions similar to the EM2 trachytes presented here.A scenario in which trachytic melts are formed directly through partial melting of a sediment-like source and subsequently mixed with more mafic Samoan melts to form the ALIA-D115 mixing trend is also possible. However, silicic melts are highly reactive with ultramafic bulk compositions and a transport scenario in which large volumes of this melt travel through 10s of kilometers of ocean mantle lithosphere to shallow magma chambers and remain isolated, requires further study (see Adams et al.49, for a further discussion).Trachytic melts have been shown to be common products of metasomatic processes such as reaction of ambient peridotite mantle with CO2-rich fluids or silicic melts4,5,43,47,48,50–52. Many workers have also suggested that partial melting of mantle already modified by metasomatic processes can produce disequilibrium silicic liquids that might vary substantially in composition depending on the source5,47,48,53. Evidence of carbonatitic and silicic melt metasomatism in Samoan xenoliths4,5,54,55 exists, and a trachytic glass found in one highly carbonatite metasomatized peridotite mantle xenolith from Savai’i Island—the same island from which the lavas of this study are derived—exhibits radiogenic 87Sr/86Sr, but is still much lower than the predicted 87Sr/86Sr of this study for the EM2-derived endmember. Further, as noted earlier, it remains to be determined how low melt fractions of trachyte melt observed in a xenolith can be scaled up to larger volumes and additionally, it is not clear how to transport this trachyte melt through the mantle, from the melt source to the magma mixing region. This type of petrogenetic mechanism, which invokes carbonatite metasomatism of peridotite to generate the trachytic melt source, may be a robust avenue for future research into the origin of EM2 that could be tested using thermodynamic phase equilibria modeling and/or experimental analysis (see Adams et al.49, for further discussion).The origin of EM2 and why it manifests in extreme form in Samoa remain important problems: EM2 is identified in lavas from Society and Marquesas Islands17,56, for example, but the magnitude of the geochemical enrichment at these localities is less pronounced than at Samoa. Nonetheless, the new radiogenic isotopic compositions derived in this study support models advocating for a large fraction of recycled terrigenous sediment into the Samoan plume7,16. This further suggests that, since EM mantle domains have been found to correspond geographically with the LLSVPs57,58, sediment-rich domains can survive in the LLSVPs over geologic timescales to be sampled by the Samoan plume.MethodsTIMS analysisSample preparationSeventeen clinopyroxene grains lacking visible attached groundmass or surface alteration were plucked from indium mounts. The clinopyroxenes were then leached in concentrated HNO3 at 90 °C for 2 min and then washed once with milli-Q (18.2 × 106 Ω cm) water; this process was repeated twice to ensure that any indium remaining on the grains was removed. The clinopyroxenes were then leached in 6 N HCl at 90 °C for 2 min to remove any remaining surface contamination, then they were rinsed three times with milli-Q water. Seventeen clinopyroxene samples, together with two aliquots of USGS Reference Material BCR2 hosting 10.8 ng Sr and 0.88 ng Nd (the first aliquot), and 5.6 ng Sr (the second aliquot, but Nd was not analyzed), respectively, were individually transferred to separate vials containing 150 μl of concentrated HNO3 and 225 μl of concentrated HF for dissolution. The samples were then set on a 120 °C hotplate for 2 days for dissolution. All clinopyroxenes and BCR2 aliquots were spiked with 84Sr and 150Nd for determination of the total amount of analyte in the sample by isotope dilution. Samples were then dried down and later brought up in a 500 μl concentrated HNO3 solution and placed on a 120 °C hotplate to flux for 24 h to eliminate fluorides. Following dry-down, the samples were brought up in 1 ml of 3 N HNO3 to load on to columns for chemical separation of Sr and Nd; column chemistry follows Koornneef et al.59. Total procedural blanks were processed together with sample clinopyroxenes and the BCR2 through all steps of sample dissolution (beginning with dissolution in HNO3 and HF), column separations, and mass spectrometry. Total procedural blanks (including all stages of wet chemistry, e.g., sample dissolution, column chemistry, etc., and loading on TIMS filaments) varied from 22 to 60 pg for Sr (and averaged 45 pg in one session and 37 pg in the other), and 2.2 to 2.7 pg for Nd. The Sr and Nd blanks are dwarfed by the total amount of Sr (~5.6 to 58.7 ng) and Nd (~0.88 to 25.6 ng), respectively, in each clinopyroxene grain and the two BCR2 aliquots. The [Sr]sample/[Sr]blank and [Nd]sample/[Nd]blank ratios for each sample are provided in Supplementary Table 1. Thus, while blanks corrections are applied to the samples and the BCR2 aliquots (assuming a lab blank 87Sr/86Sr of 0.711 and 143Nd/144Nd blank of 0.5118, obtained by pooling multiple blanks), blank corrections to the 87Sr/86Sr and 143Nd/144Nd ratios are negligible.87Sr/86Sr and 143Nd/144Nd isotope analysesAll of the clinopyroxene grains were processed through the same batch of column chemistry with a BCR2 and two total procedural blanks; two batches of chemistry were required for Sr isotopes (so two BCR2 analyses for Sr are reported) and one batch of chemistry was required for Nd isotopes (so one BCR2 analysis for Nd is reported). The clinopyroxene grains and the BCR2 were analyzed for 87Sr/86Sr and 143Nd/144Nd over two analytical sessions. The separated, dried Sr samples were brought up in 1 μl of HNO3 and each loaded onto outgassed, zone-refined rhenium (99.999%) filaments (H Cross, USA) along with a 1 μl TaCl5 emitter solution. For each analytical session, two total procedural blanks were also loaded onto rhenium filaments in addition to a BCR2, all of which passed through the same column chemistry as the clinopyroxene and BCR2 unknowns. In addition, eight filaments with 1 ng NBS987 were analyzed together with the clinopyroxenes and BCR2 aliquots: average 87Sr/86Sr = 0.710262 ± 0.000076 (2 SD, N = 8). All samples were analyzed by static Faraday collection without amplifier rotation on UCSB’s Triton Plus employing 1011 Ω amplifiers and using a 3.3 picoamp gainboard, and gains were measured every barrel. Samples were corrected for the offset between the preferred (0.710240) and the average measured NBS987 87Sr/86Sr from the same analytical session (i.e., the same barrel). Sr isotopes were corrected for mass bias assuming an exponential law and using canonical 86Sr/88Sr ratio of 0.1194, and isobaric interferences from Rb were corrected by monitoring mass 85, but changes to the 87Sr/86Sr ratios due to this correction were negligible.Only nine of the above 17 clinopyroxene grains were analyzed for 143Nd/144 Nd; these were selected after 87Sr/86Sr was analyzed and were chosen because they spanned the range of measured 87Sr/86Sr values. The dried down, separated Nd samples were brought up in 4 µl of 1 M HNO3 and loaded onto outgassed, zone-refined (99.999%, H Cross, USA) double rhenium filaments. Two blanks, six 1 ng JNDi’s (average 0.512106 ± 0.000033, 2 SD), one 500 ng JNDi (0.512112 ± 0.000011, 2SE), and a BCR2 (run through the columns) were also loaded onto rhenium filaments and run during the same analytical session (i.e., the same barrel) as the clinopyroxenes. Samples were corrected for the offset between the preferred (0.512099)60 and the average measured JNdi 143Nd/144Nd from the same analytical session. All samples were analyzed by static Faraday collection without amplifier rotation on UCSB’s Triton Plus employing 1013 Ω amplifiers and using a 3.3 picoamp gainboard (gains are measured once, at the start of the new barrel). Nd isotopes were corrected for mass bias assuming an exponential law and using a canonical 146Nd/144Nd ratio of 0.7219. Isobaric interferences from Sm were corrected by monitoring mass 147, but changes to the 143Nd/144Nd ratios due to this correction were negligible.At UCSB, the long-term reproducibility (up to and including this study) of 87Sr/86Sr for 1 ng and 500 ng loads of NBS987 by static multicollection (without amplifier rotation) using the same methods as this study are 0.710248 ± 0.000061 (2 SD, N = 69) and 0.710244 ± 0.000014 (2 SD, N = 39), respectively. For BCR2, multiple runs of aliquots hosting 5.6 to 10.8 ng Sr that were spiked and processed through column chemistry and mass spectrometry (following the same methods as samples here) yield an average 87Sr/86Sr value of 0.705018 ± 0.000065 (N = 9), which is in line with the 87Sr/86Sr of the two BCR2 aliquots analyzed in this study: the 10.8 ng aliquot 87Sr/86Sr = 0.704974 ± 0.000018 (2SE), and the 5.6 ng aliquot 87Sr/86Sr = 0.705027 ± 0.000045 (2SE). These values are comparable to an average 87Sr/86Sr of 0.705005 ± 0.000010 for BCR2 reported by Weis et al.61 (following normalization to the same NBS987 value used here). For 143Nd/144Nd, the long-term reproducibility of 1 ng loads of JNdi by static multi-collection (without amplifier rotation) using the same methods as this study (i.e., 1013 Ω amplifiers) is 0.512104 ± 0.000030 (2 SD, N = 27). For BCR2, 0.50 to 0.88 ng Nd aliquots spiked and processed through column chemistry and mass spectrometry (processed together with the samples here) yield an average 143Nd/144Nd value of 0.512618 ± 0.000023 (N = 4); the 143Nd/144Nd of the BCR2 aliquot analyzed here (0.88 ng Nd) was 0.512634 ± 0.000032 (2 SE). This is comparable to an average 143Nd/144Nd value of 0.512621 ± 0.000012 for BCR2 reported by Weis et al.61 (following normalization to the same JNdi value used here, and the La Jolla to JNdi conversion of Tanaka et al.62).Clinopyroxene major and trace element analysisClinopyroxene major elements reported in Supplementary Table 1 were collected by the electron probe micro-analyzer (EPMA) at the University of California Santa Barbara. An accelerating voltage of 15 kV was used with a 20-nA beam and a 2-µm spot size. Each oxide is the average of many 2-µm pixels across a fully quantitative x-ray map performed on each grain. All data for each element map were filtered to remove any inclusions within the clinopyroxenes prior to averaging the pixels. Low total measurements from individual pixels (<90 wt%) were also filtered out of the dataset prior to averaging.Clinopyroxene trace element concentrations were collected at the University of California Santa Barbara (Supplementary Table 1) by LA-ICP-MS using a Photon Machines Excite 193 Excimer laser coupled to an Agilent 7700 quadrupole ICP-MS. A 15-µm spot diameter was used. Unknowns were corrected relative to reference material NIST612 analyzed every 8–10 unknown analyses; analyses and the 2RSD on each element can be found in Supplementary Table 4. Each trace element measurement reported in Supplementary Table 1 is an average of three to four spot analyses spread across each clinopyroxene grain.Residence times from diffusionTo obtain an estimate of the residence time, defined as the time interval between crystal growth and lava eruption (quenching), of the clinopyroxene porphyrocrysts, the diffusive characteristic of lamellar compositional bands is considered. Examination of compositional profiles revealed that thin lamella of locally higher Mg/Fe were preserved (Supplementary Fig. 6). Based on Fe–Mg inter-diffusion, we used the thickness of these lamella to obtain an estimate of the maximum residence time of the crystal. That is, if diffusion was active for a period of time greater than the residence, the compositional anomaly would have been erased. ln this one-dimensional diffusion model, a lamella of greater Mg/Fe than its surroundings initially at composition C0 occupies a region between x = −b and x = +b. The composition outside the lamella is C1 at all times. The non-dimensional differential equation governing the decay of the lamella compositional anomaly is1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{\partial \widehat C}}{{\partial \widehat t}} = D\frac{{\partial ^2\widehat C}}{{\partial \widehat x^2}},$$\end{document}∂C^∂t^=D∂2C^∂x^2,where the non-dimensional variables are defined as:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat C = \frac{{C_1 - C}}{{C_1 - C_0}},\,\widehat x = \frac{x}{b},\,{\mathrm{and}}\,\,\,\widehat t = \frac{{Dt}}{{b^2}}$$\end{document}C^=C1−CC1−C0,x^=xb,andt^=Dtb2where 2b is the lamella thickness, D is species-appropriate diffusion coefficient, and t is the time since the start of diffusion. The initial and boundary conditions in terms of the non-dimensional parameters are:IC: at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t = 0$$\end{document}t=0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat C = 1$$\end{document}C^=1 for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$- 1 {\,\,}< {\,\,}\widehat x {\,\,}< {\,\,}1$$\end{document}−1<x^<1BC: at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat x = \pm {\!}1$$\end{document}x^=±1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat C = 0$$\end{document}C^=0 for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widehat t {\,\,}> {\,\,}0$$\end{document}t^>0The solution31 recast in dimensional variables is:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{C_1 - C}}{{C_1 - C_0}} = 2\mathop {\sum}\limits_{n = 0}^\infty {\frac{{ - 1^n}}{{(n + \frac{1}{2})\pi }}} {\mathrm{exp}}\left[ - \left( {n + \frac{1}{2}} \right)^2\pi ^2\frac{{Dt}}{{b^2}}\right]{\mathrm{cos}} \left(n + \frac{1}{2}\right)\frac{{\pi x}}{b}.$$\end{document}C1−CC1−C0=2∑n=0∞−1n(n+12)πexp−n+122π2Dtb2cosn+12πxb.Based on this solution, we computed the time necessary to eradicate 90% of the compositional anomaly at the centerline of the lamella (see ref. 18, pg. 101 for graphical solution) for the thinnest lamella in the clinopyroxenes: ~15 µm (b = 7.5 µm). Diffusion calculations were performed at 1150 °C (log10
DMg/Fe inter-diffusion = −19.4 m2/s)33 and 1200 °C (log10
DMg/Fe inter-diffusion = −18 m2/s)32. The effect of varying temperature ±50 K is modest, increasing or decreasing residence times by about 20–30%. The greatest uncertainty is in the laboratory determination of diffusivities.Retrieval of magma mixing endmembersThe principles of magma mixing are well established and have been used for decades to decipher magma mixing from assimilation and/or fractionation processes in natural magmatic systems23,63–66. The Samoan ALIA whole rock lavas discussed in this study are very likely a result of magma mixing as noted by examination of Fig. 3. The mixing trends can be used to estimate the composition of the mixing endmembers given the MgO content of the silicic (S; high 87Sr/86Sr) and mafic (M, low 87Sr/86Sr) mixing endmembers. Based on binary mixing theory23,65,66, for any two chemical species (major oxides, trace elements, isotopes), the mixing equations may be recast in the ratio-reciprocal form3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{C_2^H}}{{C_1^H}} = A + \frac{B}{{C_1^H}}$$\end{document}C2HC1H=A+BC1Hwhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_1^H$$\end{document}C1Hand \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_2^H$$\end{document}C2H are concentrations of chemical species 1 and 2 in the mixed (hybrid) magma, respectively, and A and B are constants related to the chemical species concentrations in endmembers M and S. The constants A and B in Eq. (3) are defined from mixing theory to be:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A = \frac{{C_2^{\mathrm{M}} - C_2^{\mathrm{S}}}}{{C_1^{\mathrm{M}} - C_1^{\mathrm{S}}}}$$\end{document}A=C2M−C2SC1M−C1Sand5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B = \frac{{(C_1^{\mathrm{M}}C_2^{\mathrm{S}} - C_1^{\mathrm{S}}C_2^{\mathrm{M}})}}{{C_1^{\mathrm{M}} - C_1^{\mathrm{S}}}}$$\end{document}B=(C1MC2S−C1SC2M)C1M−C1Swhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_1^{\mathrm{M}}$$\end{document}C1M and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_2^{\mathrm{M}}$$\end{document}C2M are concentrations of chemical species 1 and 2 in the mafic (M) endmember and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_1^{\mathrm{S}}$$\end{document}C1S and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_2^{\mathrm{S}}$$\end{document}C2S are the concentrations in the silicic (S) endmember. The linear regression of the petrochemical data for the ALIA-D115 lavas gives the intercept and slope that corresponds to A and B. A and B are two equations with 4 unknowns, thus if the concentration of chemical species 1 is known, or can be confidently estimated, in both endmembers, then the concentration of chemical species 2 in both endmembers can be calculated by simple rearrangement of Eqs. (4) and (5). The quality of the computed parameters depends on the quality of the fit of the data to Eq. (3). We have chosen to initially fix values of MgO in both M and S because there is relatively little variability in MgO concentrations in both endmembers based on the array defined by the ALIA-D115 lavas. In other words, a plot of Samoan lavas in SiO2 versus MgO space (see Supplementary Fig. 2), shows the linearity expected for mixing, as the ALIA-D115 lavas form a trend with the M endmember sitting in the cloud of data points of “other Samoan shield lavas” and the S endmember extending toward 0 wt% MgO, thus, the MgO in the S mixing endmember must lie between the value of the most evolved and high 87Sr/86Sr ALIA lava (ALIA-D115-21) and 0 wt% MgO (see Supplementary Fig. 2). The M endmember is assigned a value of 10 wt% MgO (similar to the MgO composition one would get if all “other Samoan shield lavas” with less than 50 wt% SiO2 were averaged) whereas for the S endmember, we have chosen to perform the mixing calculations for three different values of MgO: 3.8 wt% (from ALIA115-21), 2 wt%, and at the low end of the spectrum, 0.5 wt% (which is very close to the MgO content of clinopyroxene-hosted trachytic melt inclusions; see Supplementary Table 2) to cover the possible MgO contents. The method allows one to calculate the oxide compositions of M and S endmembers for each of the three assumed values of MgO in S, given a constant value in M. In detail, we used the A and B values from the regression of expressions of the following form: Ci/MgO versus 1/MgO space, where Ci are the following chemical species: SiO2, Al2O3, Na2O, K2O, and Nd (see Supplementary Fig. 4).From these plots, we computed an estimate of the M and S bulk compositions for SiO2, Al2O3, Na2O, K2O, and Nd (see Supplementary Fig. 4). As an example, with a linear least squares regression equation describing data in K2O/MgO versus 1/MgO space, with some re-arranging of Eqs. (4) and (5) we can solve for the concentration of K2O in both the M and S endmembers given our estimated initial values of MgO for each endmember as known values. For the chemical species other than SiO2, Al2O3, Na2O, K2O, and Nd, we used the computed K2O concentration as the new denominator value (instead of MgO) to calculate the concentrations of TiO2, FeO, CaO, MnO, P2O5, and Sr. The reason for this is that plots of these elements with K2O as chemical species 1 in Eq. (3) showed much better linearity compared to equivalent plots using MgO in the denominator—we need the best correlations possible to get a best estimate on the two endmember compositions.According to mixing theory, “companion plots” to the ratio–ratio plot 87Sr/86Sr versus 143Nd/144Nd should give linear trends and are of the form 87Sr/86Sr versus Nd/Sr and 143Nd/144Nd versus Sr/Nd (Supplementary Fig. 5). Thus, these plots are used again with ordinary least squares regression techniques and, given the inferred Sr and Nd concentrations in each endmember derived in the manner described above, we extract 87Sr/86Sr and 143Nd/144Nd in both endmembers (see Supplementary Table 2) using again, Eqs. (4) and (5). The last step of the process is to use the calculated 87Sr/86Sr and 143Nd/144Nd and Sr and Nd concentrations to obtain a self-consistent mixing curve to the ALIA-D115 lavas, in addition to the new clinopyroxene data of this study, in a plot of 87Sr/86Sr versus 143Nd/144Nd.The values of the reconstructed major oxide, Sr and Nd concentrations, and isotopic ratios for the mafic and silicic magma mixing endmembers for each of the three assumed MgO contents of the R endmember are given in Supplementary Table 2. The three high 87Sr/86Sr endmember reconstructed compositions (using the three different choices of MgO for the endmember) lie either in the trachyte (two of three) or trachyandesite (one of three) field on the total alkali versus silica diagram.Lastly, it is important to note that although some correlations in the ratio–ratio plots (see Supplementary Fig. 4) are not as highly correlated as others, likely related to crystal fractionation effects, it does not actually matter significantly what elements are used in the denominator in these calculations (we have chosen to use MgO and K2O, because these are the most well constrained by the data), the derived endmember compositions will be quite similar. The values change very slightly based on how good the correlations are. As an example, we used the ratio of Sr/K2O vs. 1/K2O to derive Sr concentrations in the two endmembers with a correlation coefficient of 0.96; the resultant Sr concentration in the mafic endmember was 712 p.p.m. and 380 p.p.m. in the silicic endmember. If we had instead performed the same calculation with Sr/MgO versus 1/MgO, the correlation coefficient is 0.85 and the Sr concentration in the mafic endmember would be 726 p.p.m. and the silicic endmember would be 360 p.p.m. These differences are comparable to analytical uncertainties for real measurements.MCS mixing modelingUsing the calculated mafic (M; low 87Sr/86Sr) and silicic (S; high 87Sr/86Sr) mixing endmembers from the above procedure, MELTS24,25 and the MCS27 were used to simultaneously model the effects of magmatic recharge and fractional crystallization processes on Samoan lavas (RFC in MCS jargon). The metric used to gauge model acceptability was how well the phase equilibria and geochemistry matches the petrology and petrochemistry of the observed Samoa ALIA-D115 lava suite. In the magma mixing calculations, the M endmember composition was the host magma and the S endmember composition, based upon an assumed MgO content of 0.5 wt% (see Supplementary Table 2 for details), was the recharge magma. The H2O content of both M and S endmembers was set at 0.1 wt% along the QFM (quartz-magnetite-fayalite) buffer establishing the ferric to ferrous iron ratio in each endmember. Since the porphyrocrystic clinopyroxenes of this study are in isotopic disequilibrium with the whole rock, clinopyroxene-liquid thermobarometry is not meaningful. Because an independent pressure estimate could not be made, MCS simulations were performed in the range 0.1 to 1 GPa. Upon mixing at low pressures (<0.5 GPa), the stable fractionating mineral assemblage consists of olivine, clinopyroxene, plagioclase, and spinel. Simulations were run for three cases where the high 87Sr/86Sr recharge magma (S endmember) was initially either: (1) 100% liquid, (2) possessed a crystallinity of about 15% (15% An22Ab63Or15 anorthoclase feldspar, modally), and (3) possessed a crystallinity of 60% (70% An8Ab48Or44 alkali feldspar, 23% An20Ab60Or20 plagioclase feldspar, 5% spinel, and 2% orthopyroxene, modally) upon addition to the host magma (M endmember). At constant pressure, the difference between these three initial recharge magma thermodynamic states is that as the crystallinity of the recharge magma increases the initial anorthite content and Mg# of the plagioclase and clinopyroxene, respectively, that crystallize from the host magma increases upon the initiation of mixing. As an example, at 0.2 GPa, when the recharge magma was 100% liquid, plagioclase (An45) was stabilized at 1107 °C whereas when the recharge magma was 60% crystalline, An50 plagioclase was stabilized at ~1119 °C. In the case of the ALIA-D115 lavas, the latter case more closely corresponds to observed plagioclase within the lavas (range An50–60)16. With increasing pressure, the first plagioclase to crystallize from the magma decreased in An content and there was increased stabilization of orthopyroxene in the assemblage over clinopyroxene, which is not observed in the ALIA-D115 lava suite. Thus, MCS simulations favor lower pressures of magma mixing and a partly crystalline recharge magma at the time of mixing. Overall, the modal amounts and phase compositions of the calculations agree quite well with the observed mineralogy. It is important to note that the MCS mixing lines plotted in Supplementary Figs. 2 and 3 are the evolutionary paths of the melt as a result of magma mixing and crystallization (crystals are removed from the system as soon as they form), so the bulk composition is not conserved and thus, one would not expect the mixing lines to end at the EM2-derived endmember, much different than the conservative mixing line based on closed system binary mixing. In addition, the MCS models are thermodynamically constrained models and so the terminus of the mixing line is dictated by where the simulation terminated. In any regard, the ALIA-D115 lavas can be explained reasonably well given the ubiquitous geological uncertainties in any modeling exercise.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2Supplementary Data 3Supplementary Data 4 | nature communications | [
"Article"
] | [
"Geochemistry",
"Petrology"
] | zones recycling systems sediments oceanic crust mantle lithosphere volatiles continental crust return mantle mantle plumes recycled materials diverse geochemical signatures ocean island basalts Mantle isotopic heterogeneities inferred global OIB lavas scale origin compositions sources melts investigation studies evaluated isotopic heterogeneity OIBs rock bulk mineral obscure broader isotopic heterogeneity crystals melt inclusions lava work shown 87Sr/86Sr 143Nd/144Nd isotopic heterogeneity in lower oceanic crust clinopyroxene plagioclase crystals 87Sr/86Sr heterogeneity olivine-hosted melt inclusions lava work intra inter-crystal 87Sr/86Sr heterogeneity in plagioclase Samoan lavas enriched mantle 2 studies suggest multiple isotopically distinct melts contribute single lava magma essential igneous petrogenesis recycling terrigenous sediments mantle EM2 geochemical signature ocean island Lavas elevated 87Sr/86Sr (>0.706)18 low 143Nd/144Nd EM2 component common among volcanoes Samoan Islands South Pacific OceanLavas erupted flanks Savai’i Island western Samoa rock 87Sr/86Sr values 0.720469 plagioclase 0.7224 ± 0.0003 intra-crystal variability 87Sr/86Sr >5000 p.p.m studies invoked two-endmember mantle mixing heterogeneity Samoan lavas 7% addition continental crust sediment Samoa plume study traces 87Sr/86Sr heterogeneity coupled high-precision Sr Nd isotopes trace elements clinopyroxene crystals enriched 87Sr/86Sr) Samoan rock lava 87Sr/86Sr = extreme 87Sr/86Sr heterogeneity clinopyroxenes lava consistent magma mixing two isotopically distinct basaltic trachytic clinopyroxenes two Samoan lava samples measured plagioclase 87Sr/86Sr profiles porphyrocrysts comparison extreme isotopic heterogeneity clinopyroxenes plagioclase 87Sr/86Sr constraints source isotopic heterogeneity magma mixing phase equilibrianew isotopic constraints origin terrigenous sediment Samoan plume source deep-mantle residence subducted terrigenous material low shear wave velocity provinces extreme EM2 sample 5 Ma submarine lava21 ALIA-D115-18 dredged flanks Savai’i Island western Samoa Fig nepheline-normative trachyandesite bimodal crystal population larger crystal population ≥25 μm includes porphyrocrysts clinopyroxene plagioclase ilmenite three phases modal abundances microgabbro study focuses clinopyroxene porphyrocrysts analyzed for major trace elements dissolution isotopic analysis thermal ionization mass spectrometry Seventeen clinopyroxene crystals analyzed for 87Sr/86Sr ratios 0.716833 to highest value reported in oceanic lava Nine of 17 crystals targeted for 143Nd/144Nd analysis single-crystal values range from 0.512238 to 0.512357.3) (Supplementary Table 1)clinopyroxene Sr Nd isotopes plot mantle array global OIB new data 87Sr/86Sr variability lava sample spans ~35% Samoan lavas ~30% oceanic mantle variability (Fig. 1). 1143Nd/144Nd versus 87Sr/86Sr ALIA-D115-18 single clinopyroxene data prior data whole rock bulk clinopyroxene Samoan lavas Savaiʼi global OIB mid-ocean ridge basalts shown orange gray fields Other Samoan shield lavas Savaiʼi Island ALIA-D114 orange triangles-D115 red squares orange circles-D128 orange diamonds new clinopyroxene data yellow crossed squares shows zoomed view clinopyroxene data bulk clinopyroxene analyses gray squares ALIA-D115-18 whole rock (red square) error bars smaller symbols represent 2σ standard errors nine of 17 clinopyroxenes for 87Sr/86Sr have 143Nd/144Nd data not all clinopyroxenesnine clinopyroxenes analyzed Sr Nd isotopes encompass range 87Sr/86Sr Samoan shield lavas ALIA lavas Savai’i xenolith compositions from elsewhere5–7,69. black dashed line theoretical magma mixing line binary mixing theory endmember compositions Supplementary Table 2 Each hatch 10% first two hatches 5% increments green star EM2-derived high 87Sr/86Sr trachytic mixing endmember Table isotopic heterogeneityData clinopyroxene Sr Nd isotopes lava sample ALIA-D115-18 isotopic constraints Samoan rock lavas constrain composition magma mixing endmembers—including extreme EM2 isotopic variability clinopyroxenes ALIA-D115-18 origin makeup high 87Sr/86Sr EM2 endmember in Samoan plume petrogenesis of EM2 lavas 87Sr/86Sr 143Nd/144Nd addition clinopyroxene data Samoan rock data requires mixing line significant curvaturer < 1 = [Nd/Sr]M/[Nd/Sr]S23 M [mafic S [silicic represent low- high-87Sr/86Sr endmembers Sr Nd concentrations 87Sr/86Sr 143Nd/144Nd values endmembers proposed mixing not mantle sources magmas differs from Jackson et al.7. inverse model approach for EM2-derived mixing endmember high in 87Sr/86Sr trachytic composition Mixing EM2 trachytic magma with primitive Samoan basalt compositions reproduces mixing trends ALIA-D115 rock lavas 87Sr/86Sr 143Nd/144Nd in Samoa clinopyroxene genesis high 87Sr/86Sr trachytic magma Supplementary Text results consistent with Jackson et al.7 positive correlation between 87Sr/86Sr SiO2 concentration Samoan rocks ALIA-D115 suggesting two-endmember mixing evolved (high silica low MgO endmember high 87Sr/86Sr less evolved (basaltic endmember low 87Sr/86Sr Edwards et al.negative correlation between 87Sr/86Sr Sr concentration in plagioclase crystals high 87Sr/86Sr “EM2” component endmember loss Sr plagioclase fractionation mixing extreme plagioclase fractionation 87Sr/86Sr negative correlation in clinopyroxene result mixing no clear correlation between clinopyroxene 87Sr/86Sr Sr concentration fractionation after mixing consistent with MELTS24–26 Magma Chamber Simulator (MCS)27 mixing modeling clinopyroxene crystallization begins ~50 °C prior plagioclase saturation mixing mafic low 87Sr/86Sr silicic high 87Sr/86Sr melt range 87Sr/86Sr in clinopyroxenes (0.71683–0.72389) exceeds plagioclase 87Sr/86Sr values specimen (0.71791–0.72239) Edwards et plagioclase crystallization system homogenized.Fig. 287Sr/86Sr values single clinopyroxene grains measured with LA-ICP-MS spot analyses three plagioclase grainsmeasurements bulk clinopyroxene8 pooled grains gray squares shown whole rock value7 red clinopyroxenes (pink squares plagioclase analyses Edwards et al.16 (blue diamonds circles triangles); error bars smaller 2σ standard errors clinopyroxene observations (n = 17 red bars span larger range 87Sr/86Sr plagioclase analyses16 (n = 157 blue bars 2 standard clinopyroxene plagioclase histograms overlap purple shade wider range clinopyroxene 87Sr/86Sr values plagioclase consistent magma mixing modeling MELTS24–26 Magma Chamber Simulator27 mixing mafic silicic trachytic endmember precipitates clinopyroxene 50 °C prior plagioclase saturation mixing magmas homogenized reduced range 87Sr/86Sr plagioclase detail).Figure 3 shows geochemical data ALIA-D115 dredge whole rock lavas symbols shield lavas pillow glasses Samoan islands seamounts Mixing defined ALIA-D115 dredge lavas 87Sr/86Sr versus element concentration spacemajor element space Figs. 2 3) compositions trachybasaltic to trachyandesitic single dredge curved lines define mixing ALIA-D115 lavas Samoan lavas zero-slope fractional crystallization. 387Sr/86Sr element concentration mixing curves direction fractional crystallization denoted arrows green star EM2-derived endmember composition (SiO2 65.5 MgO 0.5 Al2O3 17.1 FeO CaO 2.7 TiO2 1 Na2O + K2O 11% calculated study black hatched line mixing curves hatch increases 10% mixing depleted to EM2-derived endmember methodology Mixing defined curve non-zero slope unlike lavas affected fractional crystallization ALIA-D115 lavas represented red squares Samoan lavas orange diamonds Samoan pillow glasses purple circles Fig. 1 data references Samoa pillow glass data Workman et al.6. errors smaller symbols Sr Nd major element oxide concentrations 87Sr/86Sr 143Nd/144Nd endmembers ALIA-D115 mixing array reconstructed mixing analysisuse clinopyroxenes for reconstructing endmembers show ill-defined mixing trends major trace elements due to trace elements during crystal growth clinopyroxenes zoned averaged analyses trace element concentrations correspond with grain surface major elements EPMA maps produce spurious trends compared to radiogenic isotopes measured on dissolved issues affect isotopes mixing trend evident for 87Sr/86Sr 143Nd/144Nd method defining endmember compositions using ALIA-D115 lava trends elements Nd requires initial estimates for MgO content linear regression analysis of ALIA whole rock bulk compositions in Ci/MgO versus 1/MgO space Ci concentration major element oxide trace element MgO estimated ab initio constrained in both mixing endmembers limited variability at ends mixing array ALIA-D115 lavas comparing SiO2 versus MgO MgO ALIA-D115 lavas show linear mixing array from Samoan shield lavas to hypothetical endmember toward 0 wt% MgOaveraged Samoan shield lava MgO concentration mafic mixing endmember three MgO concentrations silicic mixing endmember lowest ALIA-D115 (3.8 wt% MgO to 0.5 wt% MgO mixing theory 87Sr/86Sr versus Nd/Sr 143Nd/144Nd Sr/Nd mixing linear arrays two-component mixing linear regressions performed arrays derived Sr Nd concentrations estimates 87Sr/86Sr 143Nd/144Nd obtained endmembers “Methods” Supplementary Figs. 4 5 Supplementary Table 2 linear regression method applied trace elements (Nb Th ratio tracking source mantle characteristics Supplementary Fig. 8 Table 2 initial assumption 10 wt% 0.5 wt% MgO low 87Sr/86Sr high 87Sr/86Sr (silicic endmembers calculate 380 p.p.m. Sr 63 p.p.m. Nd high 87Sr/86Sr EM2 endmember magma Supplementary Table 2 87Sr/86Sr calculations MgO concentrations retrieve values 87Sr/86Sr = 0.73158 143Nd/144Nd = 0.51213 (εNd = −9.8); major element composition EM2 endmember magma trachytic Supplementary Table 2. calculate 712 p.m. Sr 48 p.p.m. Nd low 87Sr/86Sr endmember values 87Sr/86Sr = 0.70610 143Nd/144Nd = 0.51280 Supplementary Table 2) Figures 1 3 illustrate mixing curves agreement ALIA-D115 lavas clinopyroxenes percentage mixing 87Sr/86Sr versus 143Nd/144Nd (Fig. 1) consistent major element isotope plots 3 silica-rich ALIA-D115 lava) ~70–80% EM2 endmember melt ~20–30% mafic melt endmember.Preservation isotopic heterogeneity magma zoning pyroxene indicator magma mixing Mg Fe zoning magmatic residence times post mixing anomalous compositional lamella eradication compositional anomalies thin growth lamella computed one-dimensional diffusion “Methods” Attention mobile elements zonation diffusivity data inferred temperature range relevant petrogenesisFe/Mg inter-diffusion in clinopyroxene suited two magmatic temperatures 1200 °C −18 m2/s 1150 °C −19.4 m2/s thinnest Fe and Mg lamella in clinopyroxene 15 μm thick Supplementary Fig. 6) preservation Fe Mg lamellae suggest magmatic residence times <45 years at 1150 °C to <2 years at 1200 °C uncertainties in temperature experimental values Fe/Mg inter-diffusion estimates accurate order magnitude 20–30% activation energies elemental concentration field clinopyroxenes irregular zoning pattern 87Sr/86Sr zoning suggestive of one magma mixing event preservation elemental zoning suggests eruption occurred soon after magma mixing short timescales between mixing eruption for preservation of heterogeneous 87Sr/86Sr 143Nd/144Nd in clinopyroxenes observed isotopic heterogeneity reflects two magmas mixtureThermodynamic modeling evolution ALIA-D115 two endmember compositions mafic low 87Sr/86Sr trachytic high 87Sr/86Sr endmembers—thermodynamically constrained mixing simulations MELTS24–26 MCS27 ~0.2 GPa capture major element isotope characteristics Samoan lavas Figs. 2 3) validate mixing endmembers mixing calculations reproduce dominant mineralogy clinopyroxene (Mg#71–80#71–82 plagioclase (~An45 An50–60 ALIA-D115-18 lava model MCS mixing lines Supplementary Figs. 2 3 not follow same path closed-system binary mixing models end “EM2”-derived endmember former tracks predicted composition latter tracks bulk composition mixtureexpected mixing two magmas crystal populations heterogeneous major trace 87Sr/86Sr 143Nd/144Nd compositions hybridization not instantaneous likely incomplete after decades magma co-mingling13 co-mingling incomplete mixing discrete compositional bands EM2 derived petrogenetic modeling shows silicic endmember EM2 magma consistent with geochemical array ALIA-D115 lavas Samoa extreme EM2 magma endmember Samoa trachytic composition origin EM2 endmember important work shown EM2 high 87Sr/86Sr lavas Samoa evolved not result shallow contamination crustal material sediments Oceanic crust insufficiently radiogenic 87Sr/86Sr produce ALIA-D115 lavas Pb isotopic compositions marine sediments Samoan ALIA-D115 form non-overlapping diverging trend highest 87Sr/86Sr lavas furthest from marine sediment field7 Supplementary Fig.Nb/Th ALIA-D115 lavas decreases 87Sr/86Sr consistent magma mixing low 87Sr/86Sr source high 87Sr/86Sr low Nb/Th ratios/Th = 1.8 low MgO EM2-derived endmember similar upper continental crust = 1.1 average sediments = 1.2 modern ancient subducted materials presence trachytic endmember 87Sr/86Sr up to 0.73158) inferred mixing not found pure form lavas ALIA-D115 dataset geochemical interrogation highest 87Sr/86Sr Samoan-D115-18-D115-21 87Sr/86Sr 0.718592 0.720469 single clinopyroxene crystals (87Sr/86Sr up to 0.723888 zones plagioclase (87Sr up to 0 sample extreme EM compositions trend toward EM2 melt high-SiO2 (65 to 69 wt%) trachytic melt ALIA-D115-18 clinopyroxene grains possibility EM2 endmember preserved pure melt inclusions high 87Sr/86Sr lava Table 2 melt inclusions too small for analysis 87Sr/86Sr.presence of trachytic melts EM2 endmember composition raises questions about origin model indicates ~20% mafic magma with ~80% trachytic EM2 magma potential mixing silicic mafic magmas due to isolated silicic magma body recharged by mafic magma clinopyroxene plagioclase crystallization during mixing consistent with variable 87Sr/86Sr in clinopyroxene plagioclase crystals origin of silicic trachytic melts studies evidence silicic melts in mantle by low partial melting mantle reaction between CO2-rich fluids fractional crystallization5 hypothesis is partial melting of geochemically enriched high 87Sr/86Sr) mantle compositions crystal fractionation produce trachytic melts ancient subducted sediment mixed with ambient peridotite partial melting could silicic melts with radiogenic 87Sr/86Sr Results show partial melts of mixed sediment+peridotite sources are consistently basanitic to phonolitic fractionation produce trachytes not exhaustive search of parameter spacesediment compositions combinations oceanic crust±peridotite H2O CO2 concentrations fO2 pressures future work on melting fractional crystallization petrogenesis EM2 trachytes trachytic melts formed through partial melting sediment mixed with mafic Samoan melts ALIA-D115 mixing trend possible silicic melts reactive ultramafic bulk compositions transport scenario large volumes travel ocean to shallow magma chambers requires further study Adams et al.49 melts common products metasomatic processes reaction peridotite mantle with CO2-rich fluids silicic melts4,5 partial melting mantle modified produce disequilibrium silicic liquids vary composition Evidence carbonatitic silicic melt metasomatism in Samoan xenoliths4,5 exists trachytic glass in carbonatite metasomatized peridotite mantle xenolith Savai’i radiogenic 87Sr/86Sr lower than predicted 87Sr/86Sr for EM2-derived endmemberlow melt fractions trachyte melt xenolith scaled to larger volumes clear transport trachyte melt to magma mixing region petrogenetic mechanism carbonatite metasomatism peridotite trachytic melt for future research origin EM2 thermodynamic phase equilibria modeling experimental analysis Adams et al.49 origin EM2 extreme form in Samoa important problems EM2 identified in lavas Society Marquesas Islands17 geochemical enrichment less pronounced than Samoa new radiogenic isotopic compositions support large fraction recycled terrigenous sediment into Samoan plume7 suggests EM mantle domains correspond with LLSVPs57 sediment-rich domains survive in LLSVPs over geologic timescales Samoan plume analysisSample preparationSeventeen clinopyroxene grains lacking groundmass surface alteration plucked from indium mounts leached in HNO3 at 90 °C 2 min washed with milli-Q water repeated twice leached in 6 N HCl at 90 °C 2 min contamination rinsed three times with milli-Q waterSeventeen clinopyroxene samples two aliquots USGS Reference Material BCR2 10.8 ng Sr 0.88 ng Nd 5.6 ng Sr transferred to vials 150 μl HNO3 225 μl HF for dissolution samples set on 120 °C hotplate 2 days dissolution BCR2 aliquots spiked with 84Sr 150Nd for total analyte Samples dried in 500 μl HNO3 solution 120 °C hotplate 24 h eliminate fluorides samples 1 ml 3 N HNO3 columns for chemical separation of Sr Nd follows Koornneef et al.59 procedural blanks processed with clinopyroxenes BCR2 through steps dissolution column separations mass spectrometry blanks varied 22 to 60 pg for Sr 45 2.2 to 2.7 pg for Nd Sr Nd blanks dwarfed by Sr (~5.6 to 58.7 ng Nd (~0.88 to 25.6 ng), in each clinopyroxene grain two BCR2 aliquots[Sr/ [Nd ratios in Supplementary Table 1. blanks corrections applied to samples BCR2 aliquots lab blank 87Sr/86Sr 0.711 143Nd/144Nd blank 0.5118 blank corrections to 87Sr/86Sr 143Nd/144Nd ratios negligible.87Sr/86Sr 143Nd/144Nd isotope clinopyroxene grains processed same column chemistry BCR2 two procedural blanks Sr isotopes Nd clinopyroxene grains BCR2 analyzed for 87Sr/86Sr 143Nd/144Nd over two analytical sessions separated dried Sr samples 1 μl HNO3 loaded onto outgassed zone-refined rhenium (99.999%) filaments USA 1 μl TaCl5 emitter solution two procedural blanks loaded onto rhenium filaments BCR2 passed same column chemistry BCR2eight filaments 1 ng NBS987 analyzed clinopyroxenes BCR2 aliquots average 87Sr/86Sr = 0.710262 ± 0.000076 (2 N = 8). samples analyzed static Faraday collection UCSB’s Triton Plus 1011 Ω amplifiers 3.3 picoamp gainboard gains measured every barrel corrected offset average NBS987 87Sr/86Sr isotopes corrected mass bias 86Sr/88Sr ratio 0.1194 isobaric interferences Rb corrected mass 85 changes/86Sr ratios negligible nine 17 clinopyroxene grains analyzed for 143Nd/144 Nd selected after 87Sr/86Sr values dried separated Nd samples 4 μl 1 M HNO3 loaded outgassed zone-refined (99.999%, H Cross USA) double rhenium filaments Two blanks six 1 ng JNDi’s (average 0.512106 ± 0.000033 one 500 ng JNDi (0.512112 ± BCR2 loaded filaments Samples corrected for offset preferred average 143Nd/144Ndsamples analyzed static Faraday collection without amplifier rotation UCSB’s Triton Plus 1013 Ω amplifiers 3.3 picoamp gainboard measured Nd isotopes corrected for mass bias exponential law 146Nd/144Nd ratio 0.7219 Isobaric interferences corrected mass 147 changes 143Nd/144Nd ratios negligible long-term reproducibility 87Sr/86Sr 1 500 ng loads NBS987 static multicollection 0.710248 ± 0.000061 69) and 0.710244 ± 0.000014 BCR2 aliquots 5.6 to 10.8 ng Sr column chemistry mass spectrometry average 87Sr/86Sr value 0.705018 ± 0.000065 (N = 9) line with 87Sr/86Sr two BCR2 aliquots 10.8 0.704974 ± 0.000018 5.6 0.705027 ± 0.000045 comparable to average 87Sr/86Sr 0.705005 ± 0.000010 for BCR2 Weis et al.61 NBS987 143Nd/144Nd long-term reproducibility 1 ng loads JNdi static multi-collection amplifier rotation1013 Ω amplifiers 0.512104 ± 0.000030 (2 SD N = BCR2 0.50 to 0.88 ng Nd aliquots column chemistry mass spectrometry yield average 143Nd/144Nd 0.512618 ± 0.000023 (N =/144Nd BCR2 aliquot (0.88 ng Nd 0.512634 ± 0.000032 (2 comparable average 143Nd/144Nd 0.512621 ± 0.000012 BCR2 Weis et al.61 normalization JNdi La Jolla JNdi conversion Tanaka et al.62).Clinopyroxene major trace element elements collected-analyzer University of California Santa Barbara accelerating voltage 15 kV 20-nA beam 2-μm spot size average 2-μm pixels x-ray map data filtered inclusions Low measurements<90 filtered trace element concentrations collected University of California Santa Barbara LA-ICP-MS Photon Machines Excite 193 laser Agilent 7700 quadrupole ICP-MS 15-μm spot diameter Unknowns corrected NIST612 every 8–10 unknown analyses Supplementary Table 4. trace element measurement three to four spot analyses clinopyroxene grainResidence times estimate residence time crystal growth lava eruption clinopyroxene porphyrocrysts diffusive characteristic lamellar compositional bands considered thin lamella higher Mg/Fe preserved Fig. 6) Fe–Mg inter-diffusion used thickness lamella maximum residence time crystal if diffusion active greater residence compositional anomaly erased one-dimensional diffusion model lamella greater Mg/Fe composition C0 occupies region between x = −b x = +b composition outside lamella C1non-dimensional differential equation decay lamella compositional anomaly[12pt]{minimal}{amsmath\oddsidemargin-69pt}}\frac{{\partial \widehat t}} = D ^2\widehat x^2}}{document}∂C^∂t^=D∂2C^∂x^2 non-dimensional variables defined[12pt]{minimal}{amsmath\oddsidemargin}-69pt}{document}\widehat C = \frac{{C_1 - - C_0}}\widehat x = \frac{x}{b} t = \frac{{Dt}}{{b^2}}\end{document}C^=C1−CC1−C0,x^=xb,andt^=Dtb2where 2b lamella thickness D species-appropriate diffusion coefficient t time since start diffusioninitial boundary conditions non-dimensional parameters\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts}{upgreek}\setlength\oddsidemargin}{-69pt}\begin{document}$$t = 0\end{document}t=0[12pt]{minimal}{amsmath{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}\widehat C = 1$$\end{document}C^=1\documentclass[12pt]{minimal}{amsmath}{wasysym}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$- 1 {\,\,}<\widehat x {}1$$\end{document}−1<x^<1BC\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt{document}\widehat x =\pm}x^=±1[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin}{-69pt}\widehat C = 0{document}C^=0[12pt]{minimal}{amsmath{wasysym}\oddsidemargin}{-69pt}\widehat t {\}> {}0{document}t^ solution31 recast variables:2[12pt]{minimal}{amsmath}{wasysym{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}\frac{{C_1 - C}}{{C_1 - C_0}} = 2\limits_{n = 0}\infty\frac{{ - 1^n}}{{ + \frac{1}{2})\pi\mathrm{exp}}\left[ - \left {n + \frac{1}{2}} \right)^2pi ^2\frac{{Dt}}{{b^2}}\mathrm\left(n +\frac{1{2} x}}{b}{document}C1−CC1−C0=2∑n=0∞−1n(n+12)πexp−n+122π2Dtb2cosn+12πxb computed time eradicate 90% compositional anomaly centerline lamella ref. 18 pg. 101 thinnest lamella clinopyroxenes ~15 μm (b = 7.5 μm). Diffusion calculations 1150 °C −19.4 m2/s 1200 °C −18 m2/s effect varying temperature ±50 K modest residence times 20–30% uncertainty laboratory determination diffusivities.Retrieval magma mixing principles magma mixing established decipher assimilation fractionation Samoan ALIA rock lavas likely result magma mixing Fig. 3. mixing trends estimate composition endmembers MgO content silicic mafic mixing endmembersbinary mixing theory23,65,66 two chemical species oxides trace elements mixing equations recast ratio-reciprocal[12pt]{minimal{amsmath{wasysym{upgreek\oddsidemargin{-69pt}}\frac{{C_2^H}}{{C_1^H}} = A + \frac{B}{{C_1^H}}{document}C2HC1H=A+BC1Hwhere[12pt]{minimal{amsmath{wasysym{upgreek\oddsidemargin{-69pt}}$$C_1^H$\end{document}C1Hand[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek{\oddsidemargin}{-69pt}}$$C_2^H$${document}C2H concentrations chemical species 1 2 mixed magma A B constants related chemical species concentrations endmembers M S constants A and B in Eq.defined mixing theory[12pt]{minimal\usepackage{amsmath{wasysym{mathrsfs\oddsidemargin-69pt}{document}$A =\frac{{C_2^{\mathrm{M}} - C_2^\mathrm\end{document}A=C2M−C2SC1M−C1Sand5[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}{amsbsy{mathrsfs}{upgreek}\setlength\oddsidemargin{-69pt}\begin{document}$B = \frac{{(C_1^{\mathrm{M}}C_2^{\mathrm{S}} - C_1^ C_1^{\end{document}B=(C1MC2S−C1SC2M)C1M−C1Swhere[12pt]{minimal}{amsmath{wasysym}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}}$$C_1^{\mathrm{M}}}C1M[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{upgreek}\oddsidemargin{-69pt}}$$C_2^{\mathrm{M}}}C2M concentrations chemical species 1 2 mafic (M) endmember[12pt]{minimal}{amsmath{wasysym{upgreek}\oddsidemargin{-69pt}{document}$$C_1^{\mathrm{S}}{document}C1S[12pt]{minimal}{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}}$$C_2^{\mathrm{S}}{document}C2S concentrations silicic (S) endmember linear regression petrochemical data ALIA-D115 lavas intercept slope A B.A and B equations with 4 unknowns if concentration chemical species 1 known in both endmembers concentration 2 calculated by rearrangement Eqs. (4) and (5) quality computed parameters depends on data to Eq. (3) fix values MgO in M and S little variability in ALIA-D115 lavas plot of Samoan lavas in SiO2 versus MgO shows linearity for mixing ALIA-D115 lavas form trend M endmember in Samoan S endmember extending toward 0 wt% MgO MgO in S mixing endmember must between evolved 87Sr/86Sr ALIA lava (ALIA-D115-21) and 0 wt% MgO M endmember assigned 10 wt% MgO Samoan shield lavas” with less than 50 wt% SiO2 averaged S endmember mixing calculations for three values MgO: 3.8 wt% 2 wt%, 0.5 wt% clinopyroxene-hosted trachytic melt inclusions method allows calculate oxide compositions of M and S endmembers for three assumed values MgO in constant value in M.used A B values regression Ci/MgO versus 1/MgO Ci chemical species SiO2 Al2O3 Na2O K2O Nd Supplementary Fig. 4) plots computed M S bulk compositions for SiO2 Al2O3 Na2O K2O Nd Fig linear least regression equation K2O/MgO versus 1/MgO re-arranging Eqs. (4) (5) concentration K2O M S endmembers estimated initial values MgO Al2O3 used computed K2O concentration new denominator concentrations TiO2 FeO CaO MnO P2O5 Sr plots with K2O chemical species 1 Eq. (3) better linearity MgO need best correlations estimate endmember compositions mixing theory “companion plots” plot 87Sr/86Sr versus 143Nd/144Nd give linear trends 87Sr/86Sr versus Nd/Sr 143Nd/144Nd versus Sr/Nd (Supplementary Fig. 5)plots used least regression techniques inferred Sr Nd concentrations endmember extract 87Sr/86Sr 143Nd/144Nd endmembers Supplementary Table 2) using Eqs. (4) (5) last step calculated 87Sr/86Sr 143Nd/144Nd Sr Nd concentrations self-consistent mixing curve ALIA-D115 lavas new clinopyroxene data plot 87Sr/86Sr versus 143Nd/144Nd reconstructed major oxide Sr Nd concentrations isotopic ratios mafic silicic magma mixing endmembers three assumed MgO contents in Supplementary Table 2. three high 87Sr/86Sr endmember compositions MgO in trachyte (two or trachyandesite (one) field on total alkali versus silica diagram correlations in ratio–ratio plots Supplementary Fig. 4) crystal effects elements MgO K2O constrained derived endmember compositions similar values change slightly based correlations used ratio Sr/K2O vs. 1/K2O derive Sr concentrations two endmembers correlation coefficient 0.96 resultant Sr concentration in mafic endmember 712 p.p.m. and 380 p.p.m.silicic endmember calculation Sr/MgO versus 1/MgO correlation coefficient 0.85 Sr concentration mafic 726 p.p.m. silicic 360 p.p.m differences comparable to analytical uncertainties real measurements.MCS mixing calculated mafic silicic mixing endmembers MELTS24,25 MCS27 model magmatic recharge crystallization on Samoan lavas model acceptability phase equilibria geochemistry petrology petrochemistry Samoa ALIA-D115 lava suite magma mixing M endmember host magma S endmember MgO 0.5 wt% recharge magma H2O content M S endmembers 0.1 wt% QFM buffer ferric to ferrous iron ratio porphyrocrystic clinopyroxenes isotopic disequilibrium with rock clinopyroxene-liquid thermobarometry not meaningful independent pressure estimate MCS simulations 0.1 to 1 GPa mixing low pressures (<0.5 stable fractionating mineral assemblage olivine clinopyroxene plagioclase spinelSimulations three cases high 87Sr/86Sr recharge magma 100% liquid crystallinity 15% An22Ab63Or15 anorthoclase feldspar crystallinity 60% (70% An8Ab48Or44 alkali feldspar 23% An20Ab60Or20 plagioclase feldspar 5% spinel 2% orthopyroxene addition host magma constant pressure crystallinity increases anorthite content Mg# plagioclase clinopyroxene increases mixing 0.2 GPa 100% liquid plagioclase (An45) stabilized 1107 °C 60% crystalline An50 plagioclase stabilized ~1119 °C ALIA-D115 lavas latter case corresponds to observed plagioclase (range An50–60)16 increasing pressure first plagioclase crystallize decreased increased stabilization orthopyroxene over clinopyroxene not observed ALIA-D115 lava suite MCS simulations favor lower pressures magma mixing partly crystalline recharge magma modal amounts phase compositions agree with observed mineralogy MCS mixing lines Supplementary Figs.2 3 evolutionary paths of melt magma mixing crystallization (crystals removed system bulk composition not conserved expect mixing lines end at EM2-derived endmember different than conservative mixing line closed system binary mixing MCS models thermodynamically constrained terminus mixing line dictated by simulation ALIA-D115 lavas explained well geological uncertainties modeling.Supplementary InformationPeer Review FileDescription Additional Supplementary FilesSupplementary Data 4 | 51.4 | 1.104852 |
10.1038/s41467-020-14933-6 | PMC7062842 | Protein motion in crystals causes diffuse X-ray scattering, which so far has been very challenging to measure and interpret. Here the authors present a finely sampled diffuse scattering map from triclinic lysozyme, which allows them to resolve inter- and intramolecular correlations and they further analyze the maps using all-atom molecular dynamics simulations and simple vibrational models, revealing the contribution of internal protein motion. | Protein dynamics are integral to biological function, yet few techniques are sensitive to collective atomic motions. A long-standing goal of X-ray crystallography has been to combine structural information from Bragg diffraction with dynamic information contained in the diffuse scattering background. However, the origin of macromolecular diffuse scattering has been poorly understood, limiting its applicability. We present a finely sampled diffuse scattering map from triclinic lysozyme with unprecedented accuracy and detail, clearly resolving both the inter- and intramolecular correlations. These correlations are studied theoretically using both all-atom molecular dynamics and simple vibrational models. Although lattice dynamics reproduce most of the diffuse pattern, protein internal dynamics, which include hinge-bending motions, are needed to explain the short-ranged correlations revealed by Patterson analysis. These insights lay the groundwork for animating crystal structures with biochemically relevant motions. | IntroductionConventional structure determination by X-ray crystallography relies on the intense spots recorded in diffraction images, known as Bragg peaks, that represent the average electron density of the unit cell. The average electron density is blurred when atoms are displaced from their average positions, leading to a decay in the Bragg intensities and giving rise to a second signal: a continuous pattern known as diffuse scattering1,2. Although the disorder is routinely modeled in structure refinement of Bragg data as atomic displacement parameters (ADPs) or B-factors3, information about whether groups of atoms move independently or collectively is contained only in the diffuse scattering (Supplementary Fig. 1). However, the diffuse signal is weak compared to Bragg data and challenging to accurately measure. Diffuse scattering has therefore been largely ignored in macromolecular crystallography, and instead, atomic motions have been inferred solely from Bragg data4–6.The potential of diffuse scattering as a probe of protein dynamics was envisioned over 30 years ago when Caspar et al.7 attributed the cloudy diffuse signal from an insulin crystal to liquid-like internal motions. More recently, it has been proposed that diffuse scattering can also disambiguate common structure refinement models that fit collective motions of atoms to ADPs8. Motivated by these key ideas, a number of models of protein motion have been proposed to explain macromolecular diffuse scattering2,9–13. However, in all cases to-date, agreement between measurement and simulation has been far from compelling14–20, and thus, the promise of diffuse scattering has not yet been realized.The main bottleneck in the field has been the lack of accurate data. In particular, the diffuse pattern is typically a small variation on top of a large background and is therefore easily corrupted by intense Bragg peaks. Thus, it has been common practice to heavily process images either by filtering or masking near-Bragg pixels14,17. However, this treatment suppresses features that are derived from long-ranged correlations extending beyond the unit cell and may also alter the information contained in the remaining signal. The emerging view is that long-ranged correlations must be considered2,19,21, but despite the advent of pixel array detectors that are newly enabling22,23, diffuse scattering data capable of testing such models have not been reported.To understand the fundamental origins of diffuse scattering from protein crystals, we analyzed the total scattering from the triclinic form of hen lysozyme (Fig. 1a) collected at ambient temperature using a photon-counting pixel array detector (Supplementary Fig. 2A). The triclinic crystals24 feature low mosaicity and importantly, one protein molecule per unit cell, ensuring that features between the Bragg peaks are fully resolved. By combining high-quality experimental data with new processing methods, we were able to construct a highly detailed map of diffuse scattering without filtering the images. This map reveals, for the first time, a surprisingly large contribution of long-ranged correlated motions across multiple unit cells, while also enabling detection of protein motions in a manner that is consistent with both Bragg diffraction and diffuse scattering.Fig. 1Diffuse scattering map of triclinic lysozyme with intensities on an absolute scale of electron units (Ie).a Ribbon diagram of lysozyme (top) and the triclinic unit cell containing one protein (bottom). b A highly detailed three-dimensional map of diffuse scattering was obtained. The outer sphere is drawn at 2 Å resolution. c The total scattering is made up of three components: inelastic Compton scattering (lower left), a broad isotropic ring that dominates the diffuse signal (upper left), and variational features in the diffuse scattering (right). Intense halos are visible in the layers containing Bragg peaks (l = 0 plane, upper right). Cloudy scattering is best visualized in the planes mid-way between the Bragg peaks (l = 1∕2 plane, lower right).ResultsConstruction of a three-dimensional reciprocal space mapFor accurate measurements of diffuse scattering at room temperature, the main challenges are to avoid contamination by Bragg peaks and background scattering and to achieve high signal-to-noise while avoiding radiation damage. Using well-collimated and monochromatic synchrotron radiation, we measured the angular broadening (apparent mosaicity) of our triclinic lysozyme crystals to be 0.02–0.03 degrees, which is as small as could be resolved by the diffraction instrument25. With such low mosaicity, the sharp, Gaussian-shaped Bragg peaks are readily distinguished from the underlying diffuse scattering (Supplementary Fig. 3A). To take advantage of this low mosaicity, data were collected with fine phi-slicing (0.1 deg). Crystals were held in low-background capillaries (Supplementary Fig. 2A), and low-dose partial datasets were collected from multiple sample volumes. In total, four crystals yielded 5500 images from 11 different sample volumes (Supplementary Fig. 2B, Supplementary Table 1). Using standard crystallography methods, we determined a structure to 1.21 Å (Supplementary Table 2) that agrees well with a previously reported room-temperature structure (PDB ID 4lzt24, 0.14 Å r.m.s.d.). Analysis of the structure and Bragg intensities shows that radiation damage effects were minimal (Supplementary Fig. 4).A three-dimensional diffuse map (Fig. 1b) was constructed from the same set of images (described in detail in the “Methods” section). Background scattering varied with spindle angle (Supplementary Figs. 2A and 5) and was therefore subtracted frame-by-frame (Supplementary Fig. 2C). Scale factors for each image pixel were calculated from first principles to account for X-ray beam polarization, detector absorption efficiency, solid angle, and attenuation by air. Additionally, we utilized the high data redundancy to correct for other experimental artifacts, including self-absorption of the crystal, changes in illuminated volume, differences in efficiency among the detector chips, and excess scattering from the loop and liquid on the surface of the crystal (Supplementary Fig. 6). Each of these corrections improved data quality (Supplementary Fig. 7). The data were accumulated on a fine reciprocal space grid such that the Bragg peaks were entirely contained within the voxels centered on the reciprocal lattice nodes (Supplementary Fig. 3B). In this grid, the reciprocal lattice vectors a*, b*, and c* are subdivided by 13, 11, and 11, respectively. The map had a maximum resolution of 1.25 Å, and Friedel pairs were averaged, for a total of ~50 million unique voxels.To enable rigorous comparison between simulations and experiment, we adapted the integral method of Krogh-Moe26,27 to place the map on an absolute scale of electron units per unit cell (Methods, Supplementary Fig. 8). By doing so, we are able to subtract the inelastic scattering contribution, which depends only on the atomic inventory and is insensitive to molecular structure (Fig. 1c, lower left). The final diffuse map thus represents the coherent scattering of interest (Fig. 1b) with features that depend on structure.Phonon-like scatteringThe diffuse scattering is dominated by a broad, isotropic scattering ring with a peak at ~3 Å (Fig. 1c, upper left). Although this ring is generally attributed to water, short-ranged protein disorder also contributes28,29. To better visualize the non-isotropic fluctuations, we resampled the full map mid-way between the Bragg peaks and defined the isotropic background as one sigma level below the mean scattering of this map in each resolution bin (Methods, Supplementary Fig. 9). Subtracting this background from the full map reveals clear non-isotropic features, hereafter referred to as “variational” (Fig. 1c, and Supplementary Movie 1). The most striking variational features are the intense halos (Fig. 1c, upper right) that appear to co-localize with Bragg peaks at the reciprocal lattice nodes (Fig. 1b), and are significantly asymmetric in certain directions (Supplementary Fig. 10, left). Overlaid with the halos is a cloudy pattern that is found throughout the map (Fig. 1c, lower right), which we estimate accounts for roughly half of the integrated variational intensity in most resolution bins (Supplementary Fig. 11).The presence of such intense halo scattering near the Bragg peaks was unexpected, as it implies that the correlations between atoms in different unit cells are significant and long-ranged. In protein crystallography, an outstanding question has been whether such correlations are dynamic in nature, and specifically, due to lattice vibrations7,9,15,21,28,30. The scattering intensity of a phonon (vibrational mode) is proportional to the mean squared amplitude of vibration and peaks at certain points in reciprocal space. In particular, a phonon with wavevector k makes the greatest contribution when the scattering vector q (with magnitude \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left|{\bf{q}}\right|=2\pi /d$$\end{document}q=2π∕d) is parallel to the phonon polarization and displaced from the nearest Bragg peak at q0 such that q − q0 = ±k31. The scattering of the so-called acoustic phonons, which are thermally excited at room temperature, is proportional to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v}_{s}^{-2}{\left|{\bf{k}}\right|}^{-2}$$\end{document}vs−2k−2, where vs is the speed of sound. Thus, at the Bragg peak locations, acoustic phonon scattering is expected to produce halos with a characteristic \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${|{\bf{q}}-{{\bf{q}}}_{0}|}^{-2}$$\end{document}∣q−q0∣−2 decay in intensity in any given direction.With our finely sampled diffuse map, the halo scattering can be inspected directly. We selected three symmetric and intense halos and plotted their intensities along the three reciprocal axes on a double-log scale, where a power law is a straight line (Fig. 2a, left). Both the power-law behavior and the characteristic exponent are fully consistent with acoustic phonon scattering. Furthermore, the fact that the plot remains linear as q approaches q0 implies that the lattice vibrations are coherent over at least \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\pi /| {{\bf{k}}}_{\min }| \sim 300$$\end{document}2π∕∣kmin∣~300 Å or ~10 unit cells. The characteristic exponent of approximately −2 is also found for other intense halos throughout the map (Fig. 2a, right). These results are highly suggestive of vibrational lattice dynamics.Fig. 2Evidence for long-ranged correlations in experimental maps and molecular dynamics (MD) simulations.a Throughout the diffuse map, intense halo scattering is observed around Bragg reflections. Halo profiles centered on three Bragg reflections (q0) show a power-law decay with an exponent close to −2 (gray line) along the directions (q − q0): a* (blue), b* (orange), and c* (green). Error bars represent the standard error of the mean. Histograms of the best-fit exponent along a*, b* and c* (top to bottom) for the 100 most-intense halos between 2 and 10 Å resolution also show that −2 is the most frequent value. b Halo features appear in simulated scattering from supercell MD as the simulation size is increased from 1 to 343 (7 × 7 × 7) unit cells. Each panel shows the variational component in the l = 0 plane. c Although increasing the supercell size improves agreement (green to orange), MD does not reproduce experiment on an absolute scale (black diamonds), as judged by the standard deviation profile of the diffuse intensity. In contrast, much better agreement is obtained with the lattice model described in Fig. 3 (blue). d MD displays a worse correlation (CC) with experiment (orange) compared to the lattice model (blue). The dashed line represents theoretical limit of the experimental data, CC*.All-atom molecular dynamics simulationsAlthough all-atom MD simulations have previously been used to investigate the contribution of protein dynamics to diffuse scattering12,16,29,32–34, the effect of long-ranged correlations due to lattice disorder has not been examined. We thus performed all-atom MD simulations of triclinic lysozyme crystals as a function of supercell size (Methods). Experimentally determined coordinates were used to define and initialize an array of proteins comprising the supercell, and periodic boundary conditions were imposed to remove edge effects. Supercells composed of 1, 27 (3 × 3 × 3), 125 (5 × 5 × 5), and 343 (7 × 7 × 7) unit cells were simulated for 5, 5, 2, and 1 μs, respectively. Guinier’s equation35 was used to calculate the diffuse intensity per unit cell from the simulation trajectory (Methods). Because the boundary conditions are periodic, the diffuse scattering was sampled at integer subdivisions of the reciprocal lattice (i.e., the number of unit cells in each direction).In the 1 unit-cell simulation (Fig. 2b), cloudy variational features are observed in rough qualitative agreement with the experiment (Fig. 1c, Supplementary Fig. 9B, D), suggesting that local protein and solvent dynamics contribute to the observed diffuse scattering. Unlike simpler models that do not include liquid correlations in the bulk solvent, MD provides a prediction for the isotropic component (Supplementary Fig. 9C). The overall correlation of the isotropic component is 0.9965 between 25 and 1.25 Å resolution, and the magnitude is also similar (Supplementary Fig. 9A, C). However, halos are absent, consistent with the lack of intermolecular disorder enforced by a 1 unit-cell simulation. As the size of the supercell is increased, the diffuse scattering pattern evolves in a complex manner with the halos becoming increasingly apparent (Fig. 2b), confirming that they depend on intermolecular correlations and lattice degrees of freedom. In the 343 unit-cell simulation, the r.m.s. displacement of each chain about its center of mass was 0.20–0.22 Å in each direction. Although this may seem to be a small motion, the intense halo signal is derived from constructive interference of scattered radiation from many proteins moving collectively.In the 343 unit-cell simulation (Fig. 2b), the simulated scattering contains both cloudy and halo features similar to those observed experimentally. However, the MD does not reproduce the experiment on an absolute scale (Fig. 2c, black diamonds). To make a quantitative comparison, we interpolated the experimental map on the simulation grid (7 × 7 × 7) and computed the Pearson correlation coefficient (CC) between the two in thin shells of constant resolution (Fig. 2d, orange). Although we obtain a reasonable CC of ~0.7 up to 2 Å resolution, the CC decreases at higher resolution. Moreover, there is a significant gap between CC (Fig. 2d, orange) and CC* (Fig. 2d, black dashed), which estimates the maximum CC a model can achieve, given the precision of the data36. This discrepancy indicates that model-data agreement is not limited by noise and instead points to shortcomings of the crystal model, including the current MD force fields. In particular, the accuracy of MD for diffuse scattering appears to be limited by errors in the average electron density (Supplementary Fig. 12). To gain insight into the underlying physics of the variational scattering features, we thus sought simpler dynamical models that can be refined to fit both the Bragg and diffuse data.Lattice dynamics refined against diffuse scatteringGiven the evidence for acoustic phonon scattering, we investigated whether vibrational models can capture the observed halo shapes and intensities. We developed a lattice dynamics model where each protein is able to move as a rigid body that is connected to neighboring molecules via spring-like interactions (detailed in Methods). The proteins were arranged in a 13 × 11 × 11 supercell to match the sampling of the experimental map (Fig. 3a). Residues of neighboring molecules that form lattice contacts were linked by a pair potential between alpha carbons (Fig. 3a, dark red lines), reflecting a restoring force that depends on the relative displacements of the two end-points. For generality, we allowed each pair potential to be a linear combination of two types of springs: Gaussian and directional. Gaussian springs37 have a restoring force that is independent of the direction of the displacement relative to the spring, and directional springs38 have a restoring force only along the vector between the end-points.Fig. 3Lattice dynamics model refined to diffuse scattering.a A lattice dynamics model was constructed with rigid protein units arranged in a 13 × 11 × 11 supercell with a linear combination of Gaussian and directional springs connecting the Cα atoms of residues involved in lattice contacts (dark red lines). Spring constants were refined to fit the variational scattering around 400 intense Bragg peaks between 2 and 2.5 Å resolution. b Comparison of predicted and measured halo intensity around the (1,2,13) Bragg reflection in the h = 1 plane. The plane perpendicular to the scattering vector is indicated by a dashed line. The model with equal Gaussian springs does not reproduce this shape as well as the fully-refined model. c The shape anisotropy of each of the 400 halos used for model fitting was quantified and mapped as an equal-area projection of the hemisphere centered on b*. Full refinement of the spring constants was needed to reproduce the pattern of halo anisotropy seen in experiment. d The simulated one-phonon scattering for the fully-refined lattice model (left) is compared with the measured variational scattering (right) in the l = 0 plane. The intensity scale is the same as Fig. 1c. Blue boxes surround halos that were included in the fit.The model was refined against a set of 400 intense halos between 2 and 2.5 Å resolution, consisting of a total of ~600,000 voxels. As there are three-dimensional halos associated with all 30,108 unique Bragg reflections, these 400 represent a small subset (1.3%). The spring constants were initially restrained to be all Gaussian and equal, and restraints were relaxed during subsequent stages of refinement. For a given set of springs, the equations of motion were solved by the Born/Von-Karman method31,39,40, and the diffuse scattering was calculated using the one-phonon approximation (detailed in Methods). At each refinement stage, we monitored the overall χ2 value between the experimental and simulated scattering (Supplementary Fig. 13A), as well as the ability of the model to reproduce the halo shape (Fig. 3b). To monitor agreement with halo anisotropy, we fit each of the halos to a function of the form \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I={[{({\bf{q}}-{{\bf{q}}}_{0})}^{{\rm{T}}}{\bf{G}}({\bf{q}}-{{\bf{q}}}_{0})]}^{-1}$$\end{document}I=[(q−q0)TG(q−q0)]−1, where G is a 3 × 3 positive definite matrix, and defined an anisotropy parameter, a = G⊥∕G∥ − 1, where G∥ is the component of G parallel to q0, and G⊥ is the average of the perpendicular components. The fully-parameterized model was necessary to reproduce the pattern of halo anisotropy (Fig. 3c, Supplementary Fig. 13B).After refining the lattice dynamics model using the working set of 400 halos (Fig. 3d, blue boxes), we simulated the complete diffuse scattering map over the full resolution range. Remarkably, the simulation reproduces many of the variational scattering features observed in experiment (Fig. 3d, right). Anisotropic halo shapes are reproduced even in regions of the map that were not used to refine the model (Fig. 3d, regions outside of blue boxes). Streaks in the pattern are also reproduced and can be attributed to a modulation of the halos by the molecular transform (Supplementary Fig. 10). Moreover, we find that the halos do not decay to zero mid-way between the Bragg peaks as previously expected9, giving rise to a cloudy pattern that resembles the cloudy variational scattering in the data (Fig. 3d, right). The standard deviations of intensity have very similar profiles and absolute magnitudes (Fig. 2c, blue solid and diamonds), suggesting that the lattice dynamics make the most significant contribution to the scattering variations. This conclusion is supported by the much smaller variations seen in the 1 unit-cell MD simulation (Fig. 2c, green), where lattice disorder is absent by construction.As before, the agreement between the experimental and simulated maps was assessed with CC and CC*. For the lattice dynamics model, the CC is excellent in regions of high signal-to-noise (CC ~ 0.9 between 2 and 5 Å resolution) (Supplementary Fig. 14, solid) and only limited by the experimental precision at higher resolution (Supplementary Fig. 14, dashed). To improve the signal-to-noise, the maps were interpolated on a 7 × 7 × 7 grid (Fig. 2d, blue), enabling direct comparison with the 343 unit-cell MD simulation (Fig. 2d, orange). Strikingly, the lattice dynamics model clearly outperforms all-atom MD in its ability to describe the variational component (Fig. 2c, d).The lattice model can be further assessed against existing biophysical data. Our model predicts that sound waves should propagate through the crystal (Supplementary Movie 2). Based on the calculated dispersion relations of the acoustic vibrational modes (Supplementary Fig. 15), we obtain longitudinal sound velocities of 1.0–1.3 km s−1 and corresponding transverse velocities that are slower by a factor of 1.3–2.1 depending on the propagation direction (Supplementary Table 3). Although few measurements of sound propagation have been made in protein crystals, longitudinal velocities have generally been reported to be ~2 km s−1 41–43, and transverse velocities are estimated to be 2–3 times slower41,44. Thus, our interpretation that the halo scattering arises from dynamic, rather than static, disorder appears physically reasonable.Contribution of lattice dynamics to atomic motionAs described earlier, the amount of apparent motion for each atom can be quantified from Bragg data by refining individual ADPs, the 6 components needed to describe a 3-dimensional Gaussian probability distribution. Our data quality was sufficient to refine full anisotropic ADPs for every non-H atom. To determine the extent to which lattice dynamics contribute to atomic motion, corresponding ADPs were calculated directly from the refined lattice model (Methods, Supplementary Table 4). In Fig. 4a, the full ADPs of the backbone atoms are reduced to a single isotropic B-factor per residue to facilitate visual comparison. Overall, the backbone B-factors for the lattice model (5.2 Å2 on average) fall below those of experiment (9.4 Å2 on average). The B-factors from the lattice model show small variations, which can be attributed to rigid-body rotational motion with an r.m.s. amplitude of 0.8∘ (Supplementary Table 4). However, the B-factor variations in the data are much more pronounced (Fig. 4a), particularly for side-chains (Supplementary Fig. 16A). These residual B-factors imply the existence of internal dynamics, in other words, that atoms within the protein undergo collective motions.Fig. 4Models of collective internal motions in lysozyme refined to Bragg data.a Apparent atomic motions can be evaluated by comparing the atomic displacement parameters (ADPs) obtained experimentally from Bragg data with those calculated from models. To facilitate visual comparison, the ADPs of backbone atoms are averaged to produce a single isotropic B-factor per residue. The lattice dynamics model in Fig. 3 (blue curve) underestimates the experimental B-factors (gray bars), but a good fit is obtained by combining lattice dynamics with the internal dynamics described in (b, c) (dark red symbols). b The model for internal dynamics was constructed using an elastic network with rigid residues. Both intermolecular (dark red lines) and intramolecular contacts (blue and green lines, corresponding to the α and β domains, respectively) were modeled as springs, and the spring constants were refined to fit the residual ADPs, i.e. the experimental ADPs that are unaccounted for by lattice dynamics. c In the full internal dynamics model, the Cα atoms in the α and β domains show negative directional correlations (dashed boxes), indicating that their motions are anti-correlated and consistent with hinge-bending. d The two domains have no correlations when their motions are suppressed in the model refinement.Protein dynamics refined against Bragg dataThe collective motions of lysozyme have been a topic of long-standing biophysical interest since hinge-bending motions between the two domains (Fig. 4b, blue and green) were first proposed as a mechanism for substrate binding and release45,46. To investigate the presence of such collective motions, we developed an elastic network model, in which each protein residue moves as a rigid body, and all non-H atoms within 4 Å are coupled with directional springs (Methods). As with the lattice model, the crystal environment was modeled with intermolecular springs and periodic boundary conditions, and the dynamics were calculated using the Born/Von-Karman method. In order to model only the internal protein dynamics, the Hessian matrix describing the restoring forces was modified to suppress rigid-body motion of the entire protein. The model was parametrized with one coupling constant per residue (i.e., 129 free parameters total) so that springs joining a residue pair were assigned a spring constant equal to the geometric mean of the coupling constants (Methods). The parameters were then refined by minimizing the least-squares difference between all components of the calculated (lattice + internal) and experimental ADPs derived from Bragg data.The refined model is able to reproduce the pattern of B-factors obtained experimentally (Fig. 4a and Supplementary Fig. 16A, B). To assess the importance of hinge-bending in the model, we examined the covariance matrices Cij for all alpha carbon pairs and calculated a “directional correlation”, which is the component of Cij along the inter-atomic vector normalized by the r.m.s. displacements of the two atoms (Methods). By this measure (Fig. 4c), the two domains are significantly anti-correlated as expected for hinge-bending motion.Contribution of protein dynamics to diffuse scatteringLattice dynamics account for the bulk of the variational diffuse scattering, as evaluated by CC and standard deviation (Fig. 2c, d). However, these statistics emphasize the most intense features in the signal, which in this case are the halos. To assess the more subtle contributions of internal protein motions, correlations in the signal should be separated based on length-scale. We thus calculated the diffuse Patterson (also known as 3D-ΔPDF), which is the Fourier transform of the diffuse scattering. The diffuse Patterson map represents the mean autocorrelation of the difference electron density, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \rho =\rho -\left\langle \rho \right\rangle$$\end{document}Δρ=ρ−ρ, such that a vector from the origin of the map corresponds to a vector between two points in the crystal. Thus, the central part of the diffuse Patterson is affected only by those correlations that are short-ranged.At large distances, the experimental diffuse Patterson displays peaks at the lattice nodes as expected (the Fourier transform of a lattice is also a lattice) (Fig. 5a, left), whereas continuous features are most intense at short distances (Fig. 5a, right). To determine whether lattice dynamics alone can account for the short-ranged correlations, the diffuse Patterson was calculated directly from the refined lattice model (Methods). Although the simulated and experimental maps share similar features (Fig. 5a, b), the amplitudes of the fluctuations are clearly underestimated for distances shorter than ~10 Å (Fig. 5f, blue curve vs. diamonds).Fig. 5Detection of internal motions by diffuse Patterson analysis.a The diffuse Patterson map represents the autocorrelation of the difference electron density and is a function of the vector r between points in the crystal (dashed circle corresponds to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left|{\bf{r}}\right|=10$$\end{document}r=10 Å). The experimental diffuse Patterson in the a-b plane contains peaks at the lattice nodes (left) and continuous fluctuations that are most intense near the origin (right). b–e Diffuse Patterson maps simulated from models. The lattice model underestimates the fluctuations at short length scales, but addition of full internal dynamics reproduces the experimental pattern. f The standard deviation of the diffuse Patterson maps in spherical shells of constant pair-distance for the experimental map (black diamonds), lattice model (blue), internal model (green), and the combined model (dark red). g The reciprocal space correlation coefficient (CC) between experiment and simulation in shells of constant resolution within the central part of the Patterson map (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\,{<}\,\left|{\bf{r}}\right|\,{<}\,25$$\end{document}2<r<25 Å) . The colors are the same as in (f). The 343-unit cell MD simulation is in orange. The dashed line shows CC*. h Gain in CC relative to the lattice dynamics model alone (blue) for the combined model (dark red) and the model in which domain motion was suppressed (black dashed line).In contrast, the diffuse Patterson calculated from the internal motion model, refined to the residual ADPs, shows prominent fluctuations for pair distances less than ~10 Å but very little outside this range (Fig. 5c, f, green). Assuming that the protein internal motions are independent of lattice motions, the diffuse Patterson maps can be added (Fig. 5d). The combined model displays remarkable agreement with the experimental map and reproduces the characteristic decay of fluctuation amplitude almost exactly (Fig. 5f, dark red curve vs. diamonds). To assess the agreement more quantitatively, the CC profile was calculated in reciprocal space (a Fourier transform of the 2 < r < 25 Å region). The combined model (Fig. 5g, h, dark red) displays a significant gain in CC over the lattice model alone (Fig. 5g, h, blue). The level of model-data agreement that we obtain is excellent (Fig. 5g, dark red), especially when compared to the all-atom MD simulation (Fig. 5e, g, orange) as well as all previously reported studies14,15,17–20,33.The question of model quality has consequence to protein crystallography, where it is common practice to fit models of collective motion to the B-factors, since this often increases the data-to-parameter ratio. Diffuse scattering has been proposed as a means of critically evaluating these models8. To explore this idea, we repeated refinement of the internal elastic network model with domain motions selectively suppressed. This restrained model has the same number of free parameters as the unrestrained model (Supplementary Note 1) and it is also able to reproduce the experimentally derived B-factors (R2 = 0.88 for both models, see Supplementary Fig. 16C). However, the internal dynamics are significantly different (Fig. 4d), underscoring the challenge of distinguishing differing models of protein motion from Bragg data alone. Yet, the two models are distinguishable by diffuse scattering: fluctuations in the diffuse Patterson decay more rapidly with domain motions suppressed (Supplementary Fig. 17), leading to a subtle but systematically worse CC, particularly at high resolution (Fig. 5h, dashed).DiscussionBy studying the total X-ray scattering from triclinic lysozyme crystals both experimentally and theoretically, we were able to obtain fundamental insight into the collective motions that produce macromolecular diffuse scattering. Simple vibrational models of the lattice and internal dynamics were developed that explain the electron density correlations spanning two orders of magnitude in length-scale. Vibrations of the entire protein in the lattice account for the shapes and magnitudes of the diffuse halo features and about half of the backbone ADPs, while internal motions of the protein make up the remainder. The collective nature of these internal motions was investigated by diffuse Patterson analysis, which separates correlations based on the inter-atomic vector. Remarkably, we found that two models that fit the ADPs equally well could be distinguished by their agreement to the experimental diffuse Patterson, experimentally demonstrating a key application of diffuse scattering proposed a decade ago8. Finally, although the MD was limited in its ability to reproduce the variational diffuse scattering, our results demonstrate that this signal provides an excellent experimental benchmark for improving simulations in the future.For over 30 years, the ultimate goal of diffuse scattering studies has been to capture internal protein motions from crystallographic data. The success of previous efforts has been limited primarily by data quality as well as the assumption that the variational scattering is largely due to internal motion. In fact, lattice disorder contributes significantly, further underscoring the need for detailed, high-quality data and realistic models. Despite the added challenges, we have also shown that by accounting for lattice dynamics, the remaining diffuse signal indeed contains information about internal motion and can be used to differentiate alternate models. With the initial goal of the diffuse scattering field realized, the next grand challenge of refining structural models that are consistent with the total scattering now appears within reach.MethodsCrystallizationTriclinic crystals of lysozyme were obtained by the micro-batch method with temperature-cycling to select this crystal form based on its phase diagram47. Lyophilized hen egg white lysozyme (Hampton Research) was dissolved in 20 mM sodium acetate (NaOAc) pH 4.6 at a stock concentration of 100 mg per mL, passed through a 0.2 μm filter, and used without further purification. All other reagents were purchased from Sigma, unless noted. Triclinic crystals were grown using a microbatch-under-oil technique with 8 μL drops containing 5–15 mg per mL protein, 224–300 mM NaNO3, and 50 mM NaOAc pH 4.5 that were covered with 20 μL paraffin oil (Hampton Research). Crystallization trays were set up at room temperature, moved to 4 ∘C for 8–12 h, and then returned to room temperature. Both triclinic and monoclinic crystals nucleate during the 4 ∘C incubation, however, after returning to room temperature the triclinic crystals grow at the expense of the monoclinic form47.Data collectionX-ray data were collected using the macromolecular crystallography beamline F1 at the Cornell High Energy Synchrotron Source (CHESS), which provided a 12.693 keV X-ray beam collimated to 0.1 mm diameter. Room-temperature data collection was performed using the plastic capillary sheathing method48. Crystals were harvested using low-scatter kapton loops (MicroLoops, MiTeGen), taking care to minimize the amount of solvent surrounding the crystal, and placed within 2 mm diameter, 25 μm wall poly(ethylene terephthalate) capillaries (MicroRT, MiTeGen) with 10 μL reservoir solution in the tip. During X-ray exposure, images were recorded every 0.1∘ using a pixel-array detector (Pilatus3 6M, Dectris) while rotating the sample at 1∘ s−1. A dose rate of 1.3 kGy s−1 was estimated assuming a flux of 2.5 × 1010 photons s−1 and a mass energy-absorption coefficient49 of μen ∕ ρ = 2.0 cm2 g−1. After 50 s of exposure (~65 kGy), the sample was refreshed by translating to a new spot or replacing the crystal. A background dataset was collected for each crystal by translating the sample out of the beam along the spindle axis and collecting 1 s exposures while rotating at 1∘ s−1.Structure determination from Bragg dataBragg data were integrated using xds50, with geometric parameters refined at 2∘ increments. The fitted peak profiles and mosaicity per frame were examined to verify that the crystal had not slipped or cracked. The best wedges were then scaled and merged using aimless51 (Supplementary Table 1). Model building and refinement were carried out using the ccp4 suite of programs52,53. The initial model was prepared from PDB ID 4lzt24, using the most probable (highest occupancy) protein atom coordinates only. Structure refinement was carried out using alternate runs of REFMAC554 and manual modeling in coot55, with atomic displacement parameters included in the final rounds. Alternate conformers were modeled when justified by the electron density and stereochemistry. The atomic coordinates and structure factors have been deposited in the Protein Data Bank under accession code 6o2h. Data collection and refinement statistics are shown in Supplementary Table 2.Overview of diffuse data processingReciprocal space maps were generated in Matlab (The Mathworks) as described in the following sections. Briefly, an integration mask was first produced to separate rapidly varying features (including Bragg peaks) from continuously varying features (including diffuse scattering) in three-dimensional reciprocal space. Following per-pixel image corrections, the integration mask was used to generate a coarse continuous scattering map and a Bragg map. Using the coarse continuous scattering map, a scaling model was refined to globally minimize the discrepancy between redundant observations. Bragg intensities were corrected and brought into agreement with the values used for structure determination in the previous section. The intensities were then placed on an absolute scale. Finally, continuous scattering intensities were accumulated on a fine grid to produce the final diffuse map. The scaling corrections from the previous step were applied during integration, and redundant observations were merged without further scaling. A detailed description of each operation is below.Construction of integration maskAs the Bragg intensities and continuous scattering require different corrections, a sensitive moving-window filter was first used to detect and mask out rapidly varying features. The filter algorithm compared the observed count distribution to that expected from Poisson statistics, as described below. Briefly, a voxel was masked out if its exclusion made the neighborhoods to which it contributes more Poisson-like according to the Kullback-Leibler (KL) divergence of the observed and ideal distributions. The unmasked voxels then describe a function that varies smoothly on the scale of the reciprocal space grid.X-ray images were processed in 2∘ wedges. Each pixel was mapped onto a provisional reciprocal space grid, where the reciprocal unit cell was subdivided by a factor of 5 in each direction. For each voxel, a histogram of counts per pixel was accumulated. Using these count histograms, the filtering algorithm proceeded as follows. The neighborhood (filter window) was defined as the set of voxels within a Euclidian distance of ≤2 grid units from the central voxel. For each neighborhood, the weighted median count rate rmedian was found, as well as the KL divergence of the total count histogram from the expected Poisson distribution with rate rmedian. A voxel was masked if its exclusion reduced the sum of KL divergences for all neighborhoods. First, the voxels were ranked by this change in KL divergences, ΔKL, in ascending order (worst offenders first). Then, voxels were masked progressively, and the ΔKL values of neighboring voxels were updated without re-sorting. The algorithm halted after encountering a voxel with ΔKL ≥ 0. In the resulting integration mask, the unmasked voxel grid represented the continuous scattering, whereas the masked voxel grid consisted mainly of Bragg peaks.Integration and scalingPer-pixel image corrections were applied prior to integration. Scale factors for each image pixel were calculated from first principles to account for X-ray beam polarization, detector absorption efficiency, solid angle, and attenuation by air (see Section 2 of the Supplementary Methods). The background count rate for each pixel was estimated from an exposure where the crystal was translated out of the beam along the spindle axis (Supplementary Fig. 2C).Using the mask generated in the previous step, the unmasked and masked voxels were then integrated separately in 2∘ wedges. To generate a coarse map of continuous scattering, the unmasked voxel grid was reduced to one sample per reciprocal lattice node. In addition, observations of the same voxel in adjacent wedges were combined. Then, the geometric and background corrections were applied to the photon counts to generate a map of Imeas for the continuous scattering (Equation 32 in the Supplementary Methods). The masked voxels containing Bragg peaks were integrated in a similar manner, except that the local diffuse background was subtracted and the Lorentz correction was applied (Equation 34 in the Supplementary Methods). For the background, the value of Imeas for the coarse continuous scattering map was used. The Bragg intensities were further filtered to remove partial observations. The total reciprocal space volume sampled by the detector during integration (the accumulation over contributing pixels of Equation 28 in the Supplementary Methods) was compared with the actual volume of the masked voxels. The peak was considered to be fully recorded if the volumes agreed within 5%. This rejects a large fraction of the recorded Bragg peaks, however, they are later replaced using more precise integration methods, described below.Using the coarse continuous scattering map, a scaling model was refined in order to minimize the discrepancy of redundant observations and correct for experimental artifacts. In this case, redundancy comes from Friedel symmetry and the fact that different wedges of data overlapped in reciprocal space. The scaling model related the expected intensity of an observation i to the merged intensity Imerge, in terms of four correction factors, as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{{\rm{pred}}}(i)=a({x}_{i},{y}_{i},{\phi }_{i})d({p}_{i})\left[b({\phi }_{i}){I}_{{\rm{merge}}}({{\bf{h}}}_{i})+c({s}_{i},{\phi }_{i})\right],$$\end{document}Ipred(i)=a(xi,yi,ϕi)d(pi)b(ϕi)Imerge(hi)+c(si,ϕi),where Ipred is the model’s prediction for the measured intensity, hi is the index of the symmetry-equivalent reflection in the asymmetric unit of reciprocal space, and a, b, c, and d are functions of the experimental geometry; ϕi is the spindle rotation angle, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${s}_{i}=\left|{{\bf{s}}}_{i}\right|$$\end{document}si=si is the scattering vector magnitude, pi is the detector chip index, and (xi, yi) is the position in the detector plane. Roughly speaking, a corrects for absorption, b corrects for overall changes in illuminated volume and beam intensity, c is strictly positive and corrects for excess isotropic scattering (which may occur if extra material, such as the sample loop, passes through the beam), and d corrects for detector chip efficiency (flat-field errors). The continuous functions a, b, and c were obtained by linear interpolation on multi-dimensional grids. A 9 × 9 grid was used for the detector plane position, 100 grid points were used for scattering vector (0 < s < 0.9132 Å−1), and 26 grid points were used for the spindle angle coordinate of each 50∘ data wedge. A set of 960 discrete values was used for d, corresponding to the 960 detector chips in the Pilatus 6M.The parameters of the scaling model were fit by minimizing the sum of the χ2 and regularization terms, as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal{H}}=\sum _{i}{\left({I}_{{\rm{meas}}}(i)-{I}_{{\rm{pred}}}(i)\right)}^{2}{\sigma }_{i}^{-2}+\sum _{j}{\lambda }_{j}{{\mathcal{B}}}_{j},$$\end{document}H= ∑iImeas(i)−Ipred(i)2σi−2+ ∑jλjBj,where σi is the uncertainty (standard error) estimate for Imeas(i), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathcal{B}}}_{j}$$\end{document}Bj are the regularization functions and λj are the corresponding weights (Lagrange multipliers). The regularization functions are used to stabilize refinement and to enforce smoothness of the correction factors. For the correction factors a, b, and c, smoothness was enforced by minimizing the second derivative. Discrete approximations56 of the following integrals were used: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int d\phi dxdy{\left|{\partial }_{\phi }^{2}a\right|}^{2}$$\end{document} ∫dϕdxdy∂ϕ2a2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int d\phi dxdy{\left|{\partial }_{x}^{2}a+{\partial }_{y}^{2}a\right|}^{2}$$\end{document} ∫dϕdxdy∂x2a+∂y2a2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int d\phi {\left|{\partial }_{\phi }^{2}b\right|}^{2}$$\end{document} ∫dϕ∂ϕ2b2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int d\phi ds{\left|{\partial }_{\phi }^{2}c\right|}^{2}$$\end{document} ∫dϕds∂ϕ2c2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int d\phi ds{\left|{\partial }_{s}^{2}c\right|}^{2}$$\end{document} ∫dϕds∂s2c2. In addition, the offset correction was forced to be positive, and to stabilize the refinement, its magnitude was minimized using a discrete approximation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\int d\phi ds{\left|c\right|}^{2}$$\end{document} ∫dϕdsc2. Finally, the detector correction factors were regularized using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sum _{p}{\left|d(p)-1\right|}^{2}$$\end{document} ∑pd(p)−12, which ensures d = 1 in the absence of data. The nonlinear minimization problem was solved iteratively by alternately minimizing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathcal{H}}$$\end{document}H with constant Imerge (a linear problem) and updating Imerge given the new scale factors57. To simplify the implementation, each set of parameters was refined individually (or in pairs) with the others held fixed. Satisfactory results were obtained by refining corrections in the following sequence: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{b,o,cb,o,c,a,d\right\}$$\end{document}b,o,cb,o,c,a,d, where cb refers to fitting the model for c followed by b, and o is an outlier rejection step (Supplementary Figs. 6, 7).After refining the scaling model, redundant observations in the continuous scattering map were merged. Observations more than 5σ from the mean were excluded. The estimated Bragg intensities, obtained from integration of the masked voxels, were also scaled and merged using the same model, except that the offset correction was omitted and an outlier cutoff of 2σ was used. The merged values were compared with the Bragg intensities integrated and merged using xds50 and aimless51. A single scale factor was found to bring the xds/aimless values into agreement with our Bragg intensity map. Since the intensities determined by xds and aimless are more accurate and complete than our estimates, the xds/aimless values were used instead for all subsequent analysis. Doing so also ensures that the Bragg intensities matched the values used for structure determination.Placement of intensities on an absolute scaleAfter merging, the intensities were placed on an absolute scale. The overall scale factor was found by adaptation of the total intensity method originally described by Krogh-Moe (K-M)26,27. The standard K-M method, described in Section 2.3 of the Supplementary Methods, involves predicting the contribution of individual atoms to the total intensity (Itotal,predicted, Equation 40 in the Supplementary Methods) and comparing the prediction to the measured value (Itotal,measured, Equation 41 in the Supplementary Methods) to determine a scale factor α as follows:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha =\frac{{I}_{{\rm{total,predicted}}}}{{I}_{{\rm{total,measured}}}}.$$\end{document}α=Itotal,predictedItotal,measured.To test for convergence, the scaling factor was calculated in two ways: first using the standard K-M method, and second using a modified K-M method to account for inter-atomic interference (Supplementary Fig. 8). For the standard K-M scaling method, an estimate of the atomic inventory of the unit cell was used to calculate the theoretical coherent and incoherent scattering for independent atoms. The theoretical scattering calculation included 290 water molecules, 6 nitrate ions, and 1 lysozyme molecule in the unit cell, for a total of 1546 H, 613 C, 199 N, 493 O, and 10 S atoms. The total number of electrons was Z = 10720. Then, Itotal,predicted was calculated using Equation 40 in the Supplementary Methods, integrating over the observed region of reciprocal space. The modified K-M method was identical to the standard K-M method, except that Itotal,predicted was modified to include the interference between all atom pairs whose average inter-atomic distance could be predicted from the chemical structure alone (i.e. the protein sequence and known structure of water and solutes). Both covalent bonding and torsional restraints were considered. Molecular coordinates were taken from the chemical component dictionary58, and pair distances between atoms of adjacent amino acids in the sequence were calculated assuming a planar peptide bond with a bond length of 1.33 Å. This resulted in 7405 pair distances for lysozyme, 6 per nitrate molecule, and 3 per water molecule. Then, the following bonding correction was calculated and added to the elastic scattering in Equation 40 in the Supplementary Methods:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{{\rm{bond}}}(s)=2\sum _{n< m}{f}_{n}(s){f}_{m}(s)\frac{\sin (2\pi s\ {r}_{nm})}{2\pi s\ {r}_{nm}},$$\end{document}Ibond(s)=2∑n<mfn(s)fm(s)sin(2πsrnm)2πsrnm,where f is the atomic scattering factor, the sum is over bonded atom pairs, and rnm is the inter-atomic distance.With synthetic data, both methods converge to the expected value of 1 for α (Supplementary Fig. 8, left). However, the modified method converges much more quickly and provides a more accurate scale factor at the resolutions (~2 Å) that are typical for macromolecular crystallography.Generation of the final diffuse mapA final map of the diffuse intensities ID was generated on a fine grid with 13, 11, and 11 subdivisions along the reciprocal unit cell vectors a*, b*, and c*, respectively. The resolution range of the map was 25–1.25 Å (scattering vector of 0.04–0.8 Å−1). Voxels containing Bragg peaks were excluded. Geometric and background corrections were applied (Equation 32 in the Supplementary Methods), and redundant observations were merged using the scaling model derived from the coarse map, described above. Errors were estimated using Poisson statistics and propagated through the correction, scaling and merging steps. When merging, observations with intensities more than 5σ from the mean were flagged as outliers and excluded. Intensities were placed on an absolute scale using the previously-determined scale factor α, and the theoretical incoherent scattering was subtracted (Equation 36 in the Supplementary Methods). To calculate the correlation coefficient for random half-datasets, the unmerged observations were randomly assigned using an algorithm that gave approximately equal statistical weight to each half-dataset, and the half-datasets were merged separately. The final map includes the isotropic scattering component due to elastic scattering.All-atom molecular dynamics (MD) simulationFour all-atom MD simulations of triclinic lysozyme crystals were performed with 1, 27 (3 × 3 × 3), 125 (5 ×5 × 5), and 343 (7 × 7 × 7) unit cells, similar to a 12 unit-cell simulation described previously59. The simulation was prepared using the AMBER suite, version 1860, using the ff14SB force field for the protein61,62, the SPC/E model for water63, and the general Amber force field (GAFF)64 parameters for the nitrate ion. The simulation boxes had dimensions equal to integer multiples of the experimentally determined room-temperature unit cell from PDB ID 4lzt (a = 27.24 Å, b = 31.87 Å, c = 34.23 Å, α = 88.52∘, β = 108.53∘, γ = 111.89∘). The simulation was initialized with the measured protein atom coordinates from PDB ID 4lzt (using the “A” alternate conformer)24, arranged in a supercell grid. Nine nitrate ions were added per unit cell to neutralize the charge, as well as 290 water molecules. The number of water molecules was manually adjusted in order to achieve ~1 atm pressure at 295 K, resulting in 290, 293, 284, and 270 waters per protein chain in the 1, 27, 125, and 343 unit cell simulations, respectively. The simulations were equilibrated for about 0.2 μs and continued for an additional 5, 5, 2, and 1 μs, respectively, saving coordinates every 0.4 ns. A time step of 4 fs was used, where non-water hydrogen masses are set to 3 amu, with a corresponding decrease in the mass of its bonded atom65.For each snapshot, structure factors were calculated from the atomic coordinates using the ccp452 program sfall with its default grid parameters, a resolution of 0.95 Å, and a VDWR parameter of 3.0. The B-factor was set to 15 Å2 for each atom. This value for a “snapshot” B-factor smooths the electron density distribution to allow the Fourier transforms used by sfall to obtain a converged result; this was tested by comparing to test calculations using twice as many grid points in each dimension and for test calculations in which the “snapshot” B-factor was varied between 5 and 20 Å2. Since every atom was assigned the same B-factor, its effect can be undone by multiplying the structure factors coming from the sfall run by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\exp \left(+B{s}^{2}/4\right)$$\end{document}exp+Bs2∕4. The Bragg intensity per unit cell (Equation 16 in the Supplementary Methods) then was calculated using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{B}={N}^{-2}{\left\langle F({h}_{0})\right\rangle }^{2}$$\end{document}IB=N−2F(h0)2 where F(h0) is the supercell structure factor evaluated at the Bragg positions h0, N is the number of unit cells, and brackets represent an average over all saved simulation frames (see Section 1 of the Supplementary Methods). Similarly, the diffuse scattering per unit cell (Equation 15 in the Supplementary Methods) was calculated using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{D}={N}^{-1}(\left\langle {F}^{2}\right\rangle -{\left\langle F\right\rangle }^{2})$$\end{document}ID=N−1(F2−F2). The whole procedure is encapsulated in the md2diffuse.sh script, distributed as a part of the AmberTools distribution (http://ambermd.org).Lattice dynamics simulationLattice dynamics simulations and model refinement were performed in Matlab. Protein molecules were modeled as rigid bodies, and the lattice contacts were modeled as an elastic network13,37,38,66,67 with pair-wise interactions between α carbons. The lattice contacts were identified in the all-atom structure determined in this study (PDB ID 6o2h). First, atoms with alternate conformers were assigned to their occupancy-weighted average positions. Then, the atomic coordinates of the 26 nearest neighbors in the lattice were generated by applying the crystal symmetry operators. A lattice contact was defined between any atom in the central protein chain that came within 4 Å of an atom belonging to a neighbor. Finally, the network was reduced to a Cα model. If any atoms belonging to a pair of residues formed a lattice contact, a spring was created between the Cα atoms in the network. A total of 100 intermolecular springs were modeled, of which 50 were unique due to crystal symmetry.Two types of pair potential were modeled: Gaussian and directional, as follows:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{j{j}^{\prime}}^{({\rm{Gauss.}})}=\frac{1}{2}{\gamma }_{j{j}^{\prime}}{\left|{{\bf{u}}}_{(j)}-{{\bf{u}}}_{({j}^{\prime})}\right|}^{2}$$\end{document}Vjj′(Gauss.)=12γjj′u(j)−u(j′)2and6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{j{j}^{\prime}}^{({\rm{dir.}})}=\frac{1}{2}{\gamma }_{j{j}^{\prime}}{\left(({{\bf{u}}}_{(j)}-{{\bf{u}}}_{({j}^{\prime})})\cdot {\hat{{\bf{r}}}}_{(j,{j}^{\prime})}\right)}^{2},$$\end{document}Vjj′(dir.)=12γjj′(u(j)−u(j′))⋅r^(j,j′)2,where j and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${j}^{\prime}$$\end{document}j′ are the node indices, u is the displacement vector of a node from its equilibrium position, γ is a spring constant, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{{\bf{r}}}}_{(j,{j}^{\prime})}$$\end{document}r^(j,j′) is the unit vector pointing from node j to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${j}^{\prime}$$\end{document}j′.The equations of motion were solved in a rigid-body vibrational coordinate system using the Born/Von-Karman method (Section 3 of the Supplementary Methods). The diffuse scattering was calculated for a 13 × 11 × 11 periodic supercell, chosen to match the level of detail in the experimental map, using the one-phonon approximation (Equation 62 in the Supplementary Methods). Terms in the one-phonon structure factor (Equation 63 in the Supplementary Methods) were calculated using the fast-Fourier transform-based method68 with form factors approximated by four Gaussians and a constant69,70. The scattering contribution from the mean solvent density was modeled using Babinet’s principle: since any constant can be added to the electron density without changing the structure factor (except at s = 0), a constant density of ρsolv. surrounding a protein can be equivalently modeled by a density of 0 and −ρsolv. in the solvent-excluded region. For reasons of computational convenience, the excluded solvent can then be represented by pseudo-atoms with Gaussian form factors. To calculate the Babinet representation, excluded voxels of the solvent mask from REFMAC554 were divided among the modeled atoms based on proximity. The constant mask density associated with each atom was approximated by a three-dimensional anisotropic Gaussian with the same first and second moments. The overall solvent scaling parameters ksolv. and Bsolv. were then adjusted to minimize the least-squares difference between Fobs. and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left|{F}_{{\rm{model}}}\right|$$\end{document}Fmodel, defined as follows:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{{\rm{model}}}={F}_{{\rm{calc.}}}+{k}_{{\rm{solv.}}}\exp (-{B}_{{\rm{solv.}}}{s}^{2}/4)\ {F}_{{\rm{solv.}}},$$\end{document}Fmodel=Fcalc.+ksolv.exp(−Bsolv.s2∕4)Fsolv.,where Fcalc. is the structure factor of the modeled atoms (protein and ordered solvent) and Fsolv. is the structure factor of the excluded solvent. The refined parameters ksolv. and Bsolv. were then applied to the excluded-solvent form factors. The resulting pseudo-atoms were included in the list of atoms occupying the unit cell and assigned to the same rigid group as the nearest protein atom.Spring constants in the model were refined in order to minimize the least-squares difference between the simulated one-phonon scattering and the measured variational scattering around the 400 most intense halos in the 2–2.5 Å resolution range. Although the limited resolution range was used for refinement, the agreement of the model was ultimately assessed throughout reciprocal space (Fig. 2d). The reduced χ2 for refinement was calculated as follows:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\chi }_{{\rm{red.}}}^{2}={\left(\sum _{n = 1}^{N}{M}_{n}\right)}^{-1}\sum _{n=1}^{N}\sum _{m=1}^{{M}_{n}}{\left(\frac{{I}_{n,m}^{({\rm{meas.}})}-{I}_{n,m}^{({\rm{calc.}})}-{b}_{n}}{{\sigma }_{n,m}}\right)}^{2},$$\end{document}χred.2=∑n=1NMn−1 ∑n=1N ∑m=1MnIn,m(meas.)−In,m(calc.)−bnσn,m2,where N = 400 is the number of halos fit, Mn is the number of measured voxels around the nth halo (typically Mn = 13 × 11 × 11 − 1 = 1572), bn is an arbitrary constant offset for each halo (determined separately by least-squares minimization for each n), and σ is the experimental uncertainty. The spring constants were refined in four stages. In the first stage, all springs were set to Gaussian springs and assigned the same spring constant. In the second stage, springs belonging to the same interface (those involving a particular neighbor) were given the same spring constant. In the third stage, the pair-potentials for each interface were allowed to be a linear combination of Gaussian and directional. Finally, each pair potential was refined individually with a linear combination of Gaussian and directional springs. The overall χ2 was monitored during refinement to assess whether adding the extra degrees of freedom to the model significantly improved the fit (Supplementary Fig. 13).After refining the model, the scattering was calculated throughout reciprocal space using the one-phonon approximation (Equation 62 in the Supplementary Methods). The Pearson correlation coefficient (CC) between the measured variational scattering map and the simulation was calculated within shells of constant resolution spanning 0.04 Å−1 to 0.80 Å−1 with a constant width of Δs = 0.02 Å−1. Within each resolution bin, CC was calculated as follows:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{CC}}=\frac{\sum _{n}\left({I}_{{\rm{meas.}}}({{\rm{s}}}_{n})-{\bar{I}}_{{\rm{meas.}}}\right)\left({I}_{{\rm{calc.}}}({{\rm{s}}}_{n})-{\bar{I}}_{{\rm{calc.}}}\right)}{\sqrt{\sum _{n}{\left({I}_{{\rm{meas.}}}({{\rm{s}}}_{n})-{\bar{I}}_{{\rm{meas.}}}\right)}^{2}\sum _{n}{\left({I}_{{\rm{calc.}}}({{\rm{s}}}_{n})-{\bar{I}}_{{\rm{calc.}}}\right)}^{2}}},$$\end{document}CC=∑nImeas.(sn)−Īmeas.Icalc.(sn)−Īcalc. ∑nImeas.(sn)−Īmeas.2 ∑nIcalc.(sn)−Īcalc.2,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bar{I}}_{{\rm{meas.}}}$$\end{document}Īmeas. and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bar{I}}_{{\rm{calc.}}}$$\end{document}Īcalc. are the mean intensities in that resolution bin, and the sums are over all measured voxels within the resolution bin (Supplementary Fig. 14). For comparison with the MD simulation, which was calculated on a coarser 7 × 7 × 7 sub-sampled reciprocal lattice, the full map was interpolated at the voxels of the coarser grid by least-squares fitting a 2nd order polynomial over all neighboring voxels (a 3 × 3 × 3 kernel). Voxels at the Bragg positions (those with integer Miller indices) were excluded. The CC was calculated, as above, between the interpolated simulated map and a similarly interpolated experimental map (Fig. 2d).Internal protein dynamics simulationInternal dynamics simulations and model refinement were performed in Matlab. The dynamics of lysozyme within the crystal environment were simulated using an all-atom elastic network where each residue was restrained to move as a rigid body, and lattice contacts were explicitly modeled. To generate the model, first the atoms with alternate conformers were assigned to their occupancy-weighted average positions. Then, springs were created between any pair of non-H protein atoms belonging to different residues within a cutoff distance of 4 Å. Intermolecular springs were modeled between atoms in the protein and those of its neighbors in the lattice within the 4 Å cutoff distance. All springs were of the directional type (Eq. (6)).The equations of motion for a single unit cell were solved using the Born/Von-Karman method as described for the lattice dynamics simulation, above, except that the potential energy function was modified in order to remove those modes associated with rigid-body motion of the entire protein. This was done by assigning the component of displacement associated with such motions a restoring force of zero. The normal modes associated with rigid-body displacements then have eigenvalues of zero and are eliminated during generalized inversion of the dynamical matrix (discussed in Section 3.3 of the Supplementary Methods). More specifically, components of the Hessian matrix (Equation 45 in the Supplementary Methods) were modified as follows:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{V}}}_{({\boldsymbol{l}},{{\boldsymbol{l}}}^{\prime})}:={{\bf{P}}}^{{\bf{T}}}{{\bf{V}}}_{({\boldsymbol{l}},{{\boldsymbol{l}}}^{\prime})}{\bf{P}},$$\end{document}V (l,l′):=PTV (l,l′)P,where P is an operator that projects out the rigid-body component of displacement,11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bf{P}}={\bf{I}}-\left({\bf{A}}\backslash {{\bf{A}}}_{0}\right)\left({{\bf{A}}}_{0}\backslash {\bf{A}}\right),$$\end{document}P=I−A\A0A0\A,I is a 6m × 6m identity matrix (m = 129 is the number of residues), A is a 3n × 6m matrix (n is the number of atoms in the protein) that transforms between the Cartesian atomic displacement coordinates, u, and the generalized coordinates of the internal dynamics model (Equation 43 in the Supplementary Methods), A0 is a 3n × 6 matrix that transforms between u and the generalized coordinates of the lattice dynamics model, and the forward slash signifies left matrix division (if X = A\A0, then X is the least squares solution to the system of equations AX = A0).The model was parameterized with one coupling constant per residue, so that a spring connecting a pair of atoms (j and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${j}^{\prime}$$\end{document}j′) has a spring constant equal to the geometric mean of the residues’ coupling constants gi and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${g}_{{i}^{\prime}}$$\end{document}gi′, as follows:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{j,{j}^{\prime}}=\sqrt{{g}_{i}{g}_{{i}^{\prime}}}.$$\end{document}γj,j′=gigi′.The parameters were optimized in order to minimize the χ2 between the measured and simulated atomic displacement parameters (ADPs), calculated as follows:13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\chi }^{2}=\sum _{j=1}^{N}\sum _{n=1}^{9}{\left({\left({{\bf{U}}}_{j}^{({\rm{meas.}})}\right)}_{n}-{\left({{\bf{U}}}_{j}^{({\rm{latt.}})}+{{\bf{U}}}_{j}^{({\rm{int.}})}\right)}_{n}\right)}^{2},$$\end{document}χ2= ∑j=1N ∑n=19Uj(meas.)n−Uj(latt.)+Uj(int.)n2,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left({{\bf{U}}}_{j}\right)}_{n}$$\end{document}Ujnis the nth component of the ADP for atom j (Uj is a symmetric 3 × 3 matrix with 9 components), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{U}}}_{j}^{({\rm{latt.}})}$$\end{document}Uj(latt.) is the calculated ADP for the fully-refined lattice dynamics model, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{U}}}_{j}^{({\rm{int.}})}$$\end{document}Uj(int.) is the calculated ADP for the internal dynamics model (Equation 56 in the Supplementary Methods).After refining the model, the displacement correlations were assessed using the directional correlation coefficient, defined as follows:14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{CC}}}_{j,{j}^{\prime}}=\frac{{\hat{{\bf{r}}}}_{j,{j}^{\prime}}^{{\rm{T}}}\left\langle {{\bf{u}}}_{j}{{\bf{u}}}_{{j}^{\prime}}^{{\rm{T}}}\right\rangle {\hat{{\bf{r}}}}_{j,{j}^{\prime}}}{\sqrt{({\rm{Tr}}{{\bf{U}}}_{j}/3)({\rm{Tr}}{{\bf{U}}}_{{j}^{\prime}}/3)}},$$\end{document}CCj,j′=r^j,j′Tujuj′Tr^j,j′(TrUj∕3)(TrUj′∕3),where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{{\bf{r}}}}_{j,{j}^{\prime}}$$\end{document}r^j,j′ is the unit vector pointing from atom j to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${j}^{\prime}$$\end{document}j′.We also defined an alternate model of internal protein motion where the modes associated with rigid-body displacements of individual domains are suppressed. Residues were assigned to three groups45 as follows: 5–36 and 98–129 to α, 40–94 to β, and those remaining to the hinge region. The Hessian matrix was modified as described above, except that the P operator appearing in Eq. (10) was calculated as follows:15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bf{P}}={\bf{I}}-\left({\bf{A}}\backslash {{\bf{A}}}_{1}\right)\left({{\bf{A}}}_{1}\backslash {\bf{A}}\right),$$\end{document}P=I−A\A1A1\A,where I is a 6m × 6m identity matrix (m = 129 is the number of residues), A1 is a 3n × 6d matrix (d = 3 is the number of groups) that projects from the generalized coordinate system of the 3-group model to the atomic displacements (Equation 43 in the Supplementary Methods). The model was parameterized and refined as described above for the unrestrained model, and the directional correlation was calculated using Eq. (14).Diffuse Patterson map calculationDiffuse Patterson maps were calculated in Matlab as the Fourier transform of the diffuse scattering (Section 1.3 in the Supplementary Methods), using a three-dimensional fast-Fourier transform (FFT).The experimental diffuse map was pre-processed before performing the FFT to compensate for missing data. First, missing voxels in the diffuse map were filled in with the mean values from neighboring voxels. Then, the mean intensity in each resolution shell was subtracted, and voxels beyond the resolution limit of the map were filled with zeros. Finally, the data array were zero-padded to yield a diffuse Patterson map with a real-space voxel approximately 0.3 Å on a side (the voxel dimensions were a ∕ 91, b ∕ 107, and c ∕ 115, where a, b, and c are the lattice constants).The diffuse Patterson map for the refined vibrational model (lattice + internal) was calculated without approximation in the central region where r < 25 Å. To perform the calculation efficiently, the scattering per unit cell (Equation 59 in the Supplementary Methods) was rearranged to single out a reference unit cell (l = 0):16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{D}=\sum _{j}{f}_{j}\left\{\sum _{{l}^{\prime},{j}^{\prime}}{f}_{{j}^{\prime}}{e}^{2\pi i{\bf{s}}\cdot \left({{\bf{r}}}_{j}-{{\bf{r}}}_{{j}^{\prime}}-{{\bf{r}}}_{{l}^{\prime}}+{{\bf{r}}}_{0}\right)}{T}_{j}{T}_{{j}^{\prime}}\left({T}_{0j,{l}^{\prime}{j}^{\prime}}-1\right)\right\},$$\end{document}ID= ∑jfj∑l′,j′fj′e2πis⋅rj−rj′−rl′+r0TjTj′T0j,l′j′−1,where the first sum runs over all atoms in the unit cell, fi is the atomic scattering factor, Tj is the Debye-Waller factor (Equation 60 in the Supplementary Methods), and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T}_{0j,{l}^{\prime}{j}^{\prime}}$$\end{document}T0j,l′j′ depends on the cross-terms of the covariance matrix (Equation 61 in the Supplementary Methods with l = 0). The term in the curly brackets resembles the standard structure factor equation for the primed atoms, except that the origin is shifted and the Debye-Waller factor is replaced by17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left({T}_{{\rm{eff}}}\right)}_{0j,{l}^{\prime}{j}^{\prime}}={T}_{j}{T}_{{j}^{\prime}}\left({T}_{0j,{l}^{\prime}{j}^{\prime}}-1\right).$$\end{document}Teff0j,l′j′=TjTj′T0j,l′j′−1.The effective Debye-Waller factor was separated into contributions from lattice and internal motion:18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T}_{{\rm{eff}}}={T}_{{\rm{eff}}}^{{\rm{latt}}}+{T}_{{\rm{eff}}}^{{\rm{int}}}.$$\end{document}Teff=Tefflatt+Teffint.The lattice term was calculated as follows:19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left({T}_{{\rm{eff}}}^{{\rm{latt}}}\right)}_{0j,{l}^{\prime}{j}^{\prime}}={T}_{j}{T}_{{j}^{\prime}}\left({T}_{0j,{l}^{\prime}{j}^{\prime}}^{{\rm{latt}}}-1\right),$$\end{document}Tefflatt0j,l′j′=TjTj′T0j,l′j′latt−1,where the experimentally determined ADPs were used in Tj and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T}_{{j}^{\prime}}$$\end{document}Tj′, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T}_{0j,{l}^{\prime}{j}^{\prime}}^{{\rm{latt}}}$$\end{document}T0j,l′j′latt was calculated from the refined lattice model (Equations 55 and 61 in the Supplementary Methods). This corresponds to the definition used in the one-phonon simulation (Equation 62 in the Supplementary Methods). For the internal motions, the effective Debye-Waller factor was calculated as follows:20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left({T}_{{\rm{eff}}}^{{\rm{int}}}\right)}_{0j,{l}^{\prime}{j}^{\prime}}={T}_{j}^{{\rm{latt}}}{T}_{j}^{{\rm{int}}}{T}_{{j}^{\prime}}^{{\rm{latt}}}{T}_{{j}^{\prime}}^{{\rm{int}}}{T}_{0j,{l}^{\prime}{j}^{\prime}}^{{\rm{latt}}}\left({T}_{0j,{l}^{\prime}{j}^{\prime}}^{{\rm{int}}}-1\right),$$\end{document}Teffint0j,l′j′=TjlattTjintTj′lattTj′intT0j,l′j′lattT0j,l′j′int−1,where the T’s are calculated from the covariance matrices of the lattice and internal dynamics simulations.Excluded-solvent effects were modeled by pseudo-atoms with Gaussian scattering factors, as described above for the lattice dynamics simulation. In Eq. (16), terms in curly brackets were calculated using the FFT-based method as described for the lattice dynamics simulation, except that each atom had an effective Debye-Waller factor (Eq. (18)) and coordinates relative to rj. Since only the central part of the Patterson was desired, the sum was carried out over all atoms in the unit cell and its 26 nearest neighbors that satisfied \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|{{\bf{r}}}_{j}-{{\bf{r}}}_{{j}^{\prime}}-{{\bf{r}}}_{{l}^{\prime}}+{{\bf{r}}}_{0}|\,{<}\,29$$\end{document}∣rj−rj′−rl′+r0∣<29 Å (the cutoff distance was chosen to be somewhat larger than the maximum distance of 25 Å to avoid truncation artifacts). After calculating the diffuse intensity map using Eq. 16, the mean intensity in each resolution shell was subtracted and voxels outside the experimental resolution limit of 1.25 Å were set to zero. Then, the map was zero-padded, and the Patterson function was calculated using the three-dimensional FFT, as described above for the experimental map.The reciprocal space correlation coefficients between diffuse Patterson maps were also calculated in Matlab. First, real space voxels with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left|{\bf{r}}\right|\,{<}\,2$$\end{document}r<2 Å or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left|{\bf{r}}\right|\,{> }\, 25$$\end{document}r>25 Å were set to zero, and maps were truncated at ∣x∣ < a, ∣y∣ < b and ∣z∣ < c so that the reciprocal space map would be oversampled by a factor of 2 in each direction. Then, the inverse FFT of each truncated map was calculated. The Pearson correlation coefficients (Eq. (9)) between the experimental and simulated intensity maps were calculated in shells of constant resolution spanning 0.04 to 0.80 Å−1 with bin widths of Δs = 0.04 Å−1.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Movie 1
Supplementary Movie 2 | nature communications | [
"Article"
] | [
"Structural biology",
"Biophysics"
] | structure determination by X-ray crystallography relies on intense spots in diffraction images Bragg peaks average electron density cell electron density blurred when atoms displaced from positions to decay in Bragg intensities second signal diffuse scattering1,2 disorder modeled in Bragg data as atomic displacement parameters (ADPs) B-factors3 information in diffuse scattering diffuse signal weak compared to Bragg data challenging to measure Diffuse scattering ignored in macromolecular crystallography atomic motions inferred from Bragg potential of diffuse scattering protein dynamics envisioned 30 years ago Caspar attributed signal insulin to liquid-like internal motions proposed diffuse scattering can disambiguate structure refinement models motions models of protein motion proposed to explain diffuse scattering2 agreement between measurement simulation promise of diffuse scattering not realized main bottleneck lack of accurate data diffuse pattern small variation large background easily by Bragg peaks to process images by filtering or masking near-Bragg pixels14treatment suppresses long-ranged correlations cell alter information remaining signal view long-ranged correlations pixel array detectors diffuse scattering data not reported origins diffuse scattering protein crystals analyzed scattering from triclinic form hen lysozyme ambient temperature photon-counting pixel array detector triclinic feature low mosaicity one protein molecule per unit cell features between Bragg peaks resolved data new processing methods detailed map of diffuse scattering without filtering map reveals large contribution long-ranged correlated motions across unit cells detection protein motions Bragg diffraction diffuse scattering.Fig. 1Diffuse scattering map of triclinic lysozyme intensities scale electron units Ribbon diagram of lysozyme triclinic unit cell one protein detailed three-dimensional map of diffuse scattering outer drawn at 2 Å resolutiontotal scattering three components inelastic Compton scattering broad isotropic ring diffuse signal variational features scattering Intense halos in layers Bragg peaks = 0 plane upper Cloudy scattering planes between Bragg peaks = 1∕2 plane lower three-dimensional reciprocal space measurements diffuse scattering room temperature challenges avoid contamination Bragg peaks background scattering high signal-to-noise radiation damage monochromatic synchrotron radiation measured angular broadening triclinic lysozyme crystals 0.02–0.03 degrees diffraction low mosaicity sharp Gaussian-shaped Bragg peaks distinguished from diffuse scattering Fig data collected with fine phi-slicing (0.1 Crystals held in low-background capillaries low-dose partial datasets from multiple sample volumes four crystals yielded 5500 images from 11 sample volumes 2B Table 1) determined structure to 1.21 Å 2) agrees with room-temperature structure 0.14 radiation damage effects minimal 4) three-dimensional diffuse map1b constructed from same images Background scattering varied with spindle angle 2A 5) subtracted frame-by-frame 2C). Scale factors calculated X-ray beam polarization detector absorption solid angle attenuation by air utilized high data redundancy artifacts self-absorption changes illuminated volume differences efficiency detector chips excess scattering liquid surface crystal 6) corrections improved data quality 7) data accumulated on reciprocal space grid Bragg peaks within voxels reciprocal lattice nodes Fig 3B). reciprocal lattice vectors a* b* c* subdivided by 13, 11 11 map maximum resolution 1.25 Å Friedel pairs averaged ~50 million unique voxels comparison adapted method Krogh-Moe26 map on scale electron units per cell inelastic scattering on atomic inventory insensitive to molecular structure (Fig. 1c final diffuse map represents coherent scattering (Fig. 1b) features on structure.Phonon-like dominated by broad isotropic scattering ring peak at ~3 Å (Fig1c upper ring attributed to water short-ranged protein disorder contributes28 non-isotropic fluctuations resampled map between Bragg peaks defined isotropic background one sigma level below mean scattering each resolution bin Supplementary Fig. 9) Subtracting background reveals non-isotropic features “variational” (Fig. 1c Supplementary 1) striking features intense halos (Fig. 1c upper right co-localize with Bragg peaks lattice nodes asymmetric directions Fig. 10 Overlaid with halos cloudy pattern map (Fig. 1c lower half integrated variational intensity resolution bins Fig intense halo scattering near Bragg peaks unexpected implies correlations between atoms cells significant long-ranged protein crystallography correlations dynamic due to lattice vibrations7,9 scattering intensity phonon proportional to mean amplitude vibration peaksphonon with wavevector k contribution when scattering vector q magnitude\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{amssymb{mathrsfs{upgreek-69pt}) parallel to phonon polarization displaced from Bragg peak at q0 q − q0 = ±k31 scattering of acoustic phonons thermally excited at room temperature proportional to\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amsbsy{mathrsfs{upgreek}\oddsidemargin}-69pt}{document}{v}{document}vs−2k−2 vs is speed of sound.Bragg peak locations acoustic phonon scattering expected produce halos characteristic\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek-69pt}\bf{q}} decay in intensity direction sampled diffuse map halo scattering inspected selected three symmetric intense halos plotted intensities along three reciprocal axes on double-log scale power law straight line (Fig. 2a, left). power-law behavior characteristic exponent consistent with acoustic phonon scattering plot remains linear as q approaches q0 implies lattice vibrations coherent[12pt]{minimal}{amsmath{wasysym{upgreek-69pt}}$2\bf{k 300{document}2π∕∣kmin∣~300 ~10 unit cells characteristic exponent of approximately −2 found for other intense halos (Fig. 2a, right). results suggestive of vibrational lattice dynamics.Fig.long-ranged correlations in experimental maps molecular dynamics simulations intense halo scattering around Bragg reflections Halo profiles on three Bragg reflections show power-law decay exponent close to −2 directions a* b* c* Error bars represent standard error mean Histograms best-fit exponent along a* b c* 100 most-intense halos 2 10 resolution show −2 most frequent value Halo features simulated scattering supercell MD size from 1 to 343 (7 cells panel shows variational component l = 0 plane increasing supercell size improves agreement MD reproduce experiment absolute scale better agreement with lattice model Fig. 3 MD displays worse correlation (CC with experiment lattice model dashed line represents theoretical limit experimental data CC*-atom molecular dynamics long-ranged correlations due lattice disorder not examined performed all-atom MD simulations of triclinic lysozyme crystals supercell size Experimentally determined coordinates define initialize proteins supercell periodic conditions remove edge effectsSupercells 1 27 (3 × 3 × 3) 125 (5 × 5 × 5) 343 (7 × 7 × 7) unit cells simulated for 5 5 2 1 μs Guinier’s equation35 diffuse intensity per unit cell conditions periodic diffuse scattering sampled at integer subdivisions reciprocal lattice 1 unit-cell simulation (Fig. 2b), cloudy variational features with experiment 1c local protein solvent dynamics contribute to diffuse scattering MD prediction for isotropic component correlation 0.9965 between 25 1.25 Å resolution magnitude similar halos absent lack intermolecular disorder 1 unit-cell simulation size supercell diffuse scattering pattern evolves halos apparent depend on intermolecular correlations lattice degrees freedom 343 unit-cell simulation r.m.s. displacement of each chain about center mass 0.20–0.22 Å each direction intense halo signal from interference scattered radiation from proteins moving 343 unit-cell simulation (Fig. 2b), scattering contains cloudy halo features similar to MD reproduce experiment absolute scale2c black diamonds). comparison interpolated experimental map simulation grid (7 × 7) computed Pearson correlation coefficient (CC) between two thin shells constant resolution (Fig. 2d reasonable CC ~0.7 up to 2 Å resolution decreases higher resolution significant gap between CC (Fig 2d and CC* maximum CC model discrepancy indicates model-data agreement not limited by noise points shortcomings crystal model MD force fields accuracy MD diffuse scattering limited by errors average electron density scattering sought simpler dynamical models fit Bragg diffuse data.Lattice dynamics against diffuse investigated vibrational models capture halo shapes intensities developed lattice dynamics model each protein rigid body connected to neighboring molecules spring-like interactions proteins arranged in 13 × 11 × 11 supercell experimental map (Fig. 3a). Residues molecules linked by pair potential between alpha carbons (Fig. 3a restoring force on relative displacements end-points pair potential linear combination two springs Gaussian directionalGaussian springs37 restoring force independent displacement directional springs38 force vector between end-points. 3Lattice dynamics model diffuse scattering rigid protein units 13 × 11 × 11 supercell Gaussian directional springs connecting Cα atoms Spring constants refined fit variational scattering around 400 Bragg peaks between 2 2.5 Å resolution Comparison predicted measured halo intensity (1,2,13) Bragg reflection h = 1 plane plane perpendicular scattering vector indicated dashed line model equal Gaussian springs reproduce shape fully model shape anisotropy of 400 halos quantified equal-area projection hemisphere centered b* Full refinement spring constants needed reproduce halo anisotropy simulated one-phonon scattering compared with measured variational scattering l = 0 plane intensity scale same Fig. 1c Blue boxes surround halos model refined against 400 intense halos 2 2.5 Å resolution ~600,000 voxels 30,108 Bragg reflections 400 small subset (1.3%). spring constants Gaussian equal relaxed refinementset springs equations motion solved Born/Von-Karman method31 diffuse scattering calculated one-phonon approximation Methods). refinement stage monitored χ2 value experimental simulated scattering Fig. model reproduce halo shape (Fig. 3b). monitor agreement halo anisotropy fit halos function form\documentclass[12pt{minimal}{amsmath}{wasysym{upgreek}\oddsidemargin-69pt}{document}=\bf{q}}}{document}I=(q−q0)TG]−1 G 3 × 3 positive definite matrix anisotropy parameter a = G⊥∕G∥ − 1 G∥ component G parallel to q0 G⊥ average perpendicular components fully-parameterized model necessary reproduce halo anisotropy (Fig. 3c Supplementary Fig. lattice dynamics model 400 halos (Fig. 3d simulated complete diffuse scattering map full resolution range simulation reproduces variational scattering features experiment (Fig3d Anisotropic halo shapes reproduced in regions map not used model (Fig. 3d outside blue boxes). Streaks in pattern reproduced attributed to modulation halos by molecular transform Fig. halos decay to zero between Bragg peaks cloudy pattern variational scattering (Fig. 3d standard deviations of intensity similar profiles magnitudes (Fig. 2c lattice dynamics to scattering variations supported by smaller variations in 1 unit-cell MD simulation (Fig. 2c lattice disorder absent agreement between experimental simulated maps assessed with CC CC* lattice dynamics model CC excellent in high signal-to-noise 14 limited by experimental precision at higher resolution maps interpolated on 7 × 7 × 7 (Fig. 2d comparison with 343 unit-cell MD simulation lattice dynamics model outperforms all-atom MD variational component (Fig. 2c lattice model assessed against biophysical data model predicts sound waves propagate through crystal calculated dispersion relations of acoustic vibrational modeslongitudinal sound velocities 1.0–1.3 km s−1 transverse velocities slower 1.3–2.1 propagation direction Table 3) few measurements sound propagation protein crystals longitudinal velocities ~2 km s−1 transverse velocities 2–3 times interpretation halo scattering from dynamic disorder reasonable lattice dynamics atomic apparent motion atom quantified Bragg data refining ADPs 3-dimensional Gaussian probability distribution data quality refine full anisotropic ADPs non-H atom lattice dynamics atomic motion ADPs calculated from refined lattice model Table 4) Fig. 4a full ADPs backbone atoms reduced single isotropic B-factor per residue backbone B-factors lattice model (5.2 Å2 below experiment (9.4 Å2 B-factors show small variations rigid-body rotational motion r.m.s. amplitude 0.8∘ Table 4) B-factor variations more pronounced (Fig. side-chains residual B-factors imply internal dynamics atoms collective motions. 4Models collective internal motions lysozyme refined Bragg dataatomic motions evaluated comparing displacement parameters (ADPs Bragg data with models ADPs backbone atoms averaged single isotropic B-factor per residue lattice dynamics model Fig. 3 underestimates experimental B-factors good fit combining lattice dynamics with internal dynamics (b c) model internal dynamics constructed elastic network with rigid residues intermolecular intramolecular contacts modeled as springs spring constants refined to fit residual ADPs unaccounted lattice dynamics full internal dynamics model Cα atoms in α β domains show negative directional correlations motions anti-correlated consistent with hinge-bending domains no correlations when motions suppressed in model refinement.Protein dynamics refined against Bragg collective motions of lysozyme biophysical interest since hinge-bending motions binding developed elastic network model each protein residue moves as rigid body non-H atoms within 4 Å coupled with directional springs crystal environment modeled with intermolecular springs periodic conditions dynamics calculated using Born/Von-Karman methodmodel internal protein dynamics Hessian matrix modified suppress rigid-body motion protein model parametrized with one coupling constant per residue 129 free parameters springs residue assigned spring constant equal to geometric mean coupling constants parameters refined least-squares difference between components experimental ADPs Bragg data refined model pattern B-factors (Fig. 4a Fig. 16A hinge-bending examined covariance matrices Cij alpha carbon pairs calculated “directional component Cij inter-atomic vector normalized by r.s. displacements atoms (Fig. domains anti-correlated for hinge-bending motion protein dynamics to diffuse scatteringLattice dynamics account variational diffuse scattering evaluated by CC standard deviation (Fig. 2c, statistics emphasize intense features halos subtle contributions internal protein motions correlations separated on length-scale calculated diffuse Patterson 3D-ΔPDF), Fourier transform of diffuse scatteringdiffuse Patterson map represents autocorrelation electron density[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}\Delta\rho\rangle{document}Δρ=ρ−ρ vector origin map corresponds to vector between two points crystal central part diffuse Patterson affected by correlations short-ranged large distances diffuse Patterson displays peaks at lattice nodes (Fig. 5a continuous features intense at short distances (Fig. 5a short-ranged correlations diffuse Patterson calculated from refined lattice model simulated experimental maps similar features. 5a amplitudes fluctuations underestimated for distances shorter than ~10 Å (Fig. 5f blue curve. 5Detection internal motions by diffuse Patterson analysisdiffuse Patterson map represents autocorrelation electron density function vector r between points crystal (dashed circle corresponds\documentclass[12pt]{minimal{amsmath{wasysym-69pt\left=10=10 experimental diffuse Patterson a-b plane contains peaks at lattice nodes continuous fluctuations intense near origin (right). Diffuse Patterson maps simulated from models lattice model underestimates fluctuations short length scales full internal dynamics reproduces experimental pattern standard deviation of diffuse Patterson maps in spherical shells constant pair-distance experimental map (black lattice model internal model combined model (dark red). reciprocal space correlation coefficient (CC) between experiment simulation in constant resolution central part Patterson map[12pt]{minimal}{amsmath{wasysym{amsfonts{upgreek-69pt$2,25$}2<r<25 Åcolors same (f). 343-unit cell MD simulation orange dashed line shows CC* Gain CC relative lattice dynamics model (blue combined model (dark red domain motion suppressed (black diffuse Patterson internal motion model residual ADPs shows fluctuations pair distances less than ~10 Å little outside range (Fig. 5c, f green). protein internal motions independent lattice motions diffuse Patterson maps added (Fig. 5d). combined model agreement experimental map reproduces decay fluctuation amplitude (Fig. 5f red curve CC profile calculated reciprocal space Fourier transform 2 < r < 25 Å combined model (Fig. 5g h significant gain CC over lattice model (Fig. 5g model-data agreement excellent 5g compared to all-atom MD simulation (Fig. 5e g reported studies14 model quality protein crystallography fit models collective motion B-factors increases data-to-parameter ratio Diffuse scattering proposed repeated refinement internal elastic network model domain motions suppressed restrained model same free parameters experimentally derived B-factors (R2 = 0.88internal dynamics different (Fig. challenge distinguishing models protein motion from Bragg data two models distinguishable by diffuse scattering fluctuations in diffuse Patterson decay rapidly domain motions suppressed Fig leading to worse CC at high resolution (Fig. 5h X-ray scattering from triclinic lysozyme crystals insight into motions macromolecular diffuse scattering vibrational models lattice internal dynamics explain electron density correlations two orders magnitude Vibrations protein lattice account for shapes magnitudes diffuse halo features half backbone ADPs internal motions protein remainder collective nature motions investigated by diffuse Patterson analysis separates correlations inter-atomic vector two models ADPs distinguished by agreement diffuse Patterson application diffuse scattering MD limited variational diffuse scattering results benchmark for improving simulations 30 years goal diffuse scattering capture internal protein motions from crystallographic data success limited by data quality assumption variational scattering due to internal motion lattice disorder contributes need for detailed high-quality data realistic modelschallenges accounting lattice dynamics diffuse signal contains internal motion models initial goal diffuse scattering realized refining structural models scattering reach crystals lysozyme obtained micro-batch temperature-cycling Lyophilized hen egg white lysozyme dissolved 20 mM sodium acetate pH 100 mg per mL passed 0.2 μm filter used without purification reagents purchased Sigma Triclinic crystals grown microbatch-under-oil 8 μL drops 5–15 mg per mL protein 224–300 mM NaNO3 50 mM NaOAc pH 4.5 covered 20 μL paraffin oil Crystallization trays room temperature moved 4 ∘C 8–12 h returned triclinic monoclinic crystals nucleate 4 incubation grow expense monoclinic collectionX-ray data macromolecular crystallography Cornell High Energy Synchrotron Source 12.693 keV X-ray beam collimated 0.1 mm diameter Room-temperature data collection plastic capillary sheathingCrystals harvested low-scatter kapton loops solvent placed within 2 mm 25 μm wall poly(ethylene terephthalate capillaries 10 μL reservoir solution X-ray exposure images recorded every 0.1∘ pixel-array detector sample at 1∘ s−1 dose rate 1.3 kGy s−1 estimated flux 2.5 × 1010 photons s−1 mass energy-absorption μen = 2.0 cm2 g−1 After 50 s exposure sample refreshed replacing crystal background dataset collected translating 1 s exposures 1∘ s−1.Structure determination Bragg integrated xds50 geometric parameters refined at 2∘ increments peak profiles mosaicity per frame examined crystal wedges scaled merged aimless51 Model building refinement ccp4 initial model from PDB ID 4lzt24 probable protein atom coordinates Structure refinement REFMAC554 manual modeling atomic displacement parameters final Alternate conformers modeled justified by electron density stereochemistry atomic coordinates structure factors deposited in Protein Data Bank accession code 6o2hData collection refinement statistics in Supplementary Table 2.Overview diffuse data processingReciprocal space maps generated in Matlab integration mask produced separate rapidly varying features Bragg peaks from continuously varying features diffuse scattering in reciprocal space corrections coarse continuous scattering Bragg map scaling model refined to minimize discrepancy between redundant observations Bragg intensities corrected with values structure determination intensities placed on absolute scale continuous scattering intensities accumulated on fine grid final diffuse map scaling corrections applied during integration redundant observations merged without scaling description operation.Construction of integration Bragg intensities continuous scattering corrections sensitive moving-window filter used to detect mask rapidly varying features filter algorithm compared observed count distribution to Poisson statistics voxel masked if Poisson-like unmasked voxels describe function smoothly on reciprocal space grid.X-ray images processed in 2∘ wedges Each pixel mapped onto provisional reciprocal space grid subdivided by factor 5 voxel histogram counts per pixel accumulated filtering algorithmneighborhood defined as voxels within Euclidian distance ≤2 grid units from central voxel weighted median count rate rmedian found KL divergence of total count histogram from Poisson distribution voxel masked if reduced KL divergences voxels ranked by change KL divergences ascending order (worst offenders voxels masked progressively ΔKL values updated without re-sorting algorithm halted after voxel with ΔKL ≥ 0 unmasked voxel grid represented continuous scattering masked Bragg peaks.Integration scalingPer-pixel image corrections applied integration Scale factors calculated X-ray beam polarization detector absorption solid angle attenuation air background count rate estimated from exposure crystal unmasked and masked voxels integrated separately in 2∘ wedges continuous scattering unmasked voxel grid reduced to one sample per reciprocal lattice node observations of same voxel in adjacent wedges combinedgeometric background corrections applied to photon counts map Imeas for continuous scattering (Equation 32 masked voxels Bragg peaks integrated local diffuse background subtracted Lorentz correction applied (Equation 34 background value Imeas for coarse continuous scattering map used Bragg intensities filtered remove partial observations total reciprocal space volume sampled compared with volume masked voxels peak fully recorded if volumes agreed within 5% rejects recorded Bragg peaks replaced using precise integration methods scattering scaling model refined minimize discrepancy redundant observations correct experimental artifacts redundancy from Friedel symmetry wedges data overlapped in reciprocal spacescaling model expected intensity observation i merged intensity Imerge four correction factors[12pt]{minimal}{amsmath-69pt}{document}{I{pred}}}(i)=a({x}_{i}{y}_{i}{\phi{i})d({p}_{i}\left[b({\phi{i}{I}\rm{merge}}}\bf{h}}}_{i})+c({s}_{i}{\phi }_{i})\right\end{document}Ipred(i)=a(xi,yi,φi)d(pi)b(φi)Imerge(hi)+c(si Ipred model’s prediction measured intensity hi index symmetry-equivalent reflection asymmetric reciprocal space a b c d functions experimental geometry φi spindle rotation angle[12pt]{minimal}{amsmath}}}{-69pt}{document}$${s}_{i}=\left|{{\bf{s}}}_{i}\right|$\end{documentsi scattering vector pi detector chip index (xi, yi) position detector plane a corrects absorption b changes illuminated volume beam intensity c excess isotropic scattering d detector chip efficiency-field errors). continuous functions a b c obtained linear interpolation on multi-dimensional grids 9 × 9 grid detector plane position 100 grid points scattering vector < 26 grid points spindle angle coordinate 50∘ data wedge 960 discrete values for d 960 detector chips in Pilatus 6M.parameters scaling model minimizing χ2 regularization terms\documentclass[12pt]{minimal}{amsmath{wasysym-69pt}\begin{document}\mathcal{H}}=\sum{i\left\rm{meas-{pred\right^{2}{\sigma{-2}+\sum{j}\lambda\mathcal{B}}}\end{document}H= ∑iImeas(i)−Ipred(i)2σi−2+ ∑jλjBj σi uncertainty error) estimate Imeas(i),\documentclass[12pt]{minimal}{amsmath}{wasysym{upgreek}\oddsidemargin-69pt}{document}$\mathcal{B}}}{j\end{document}Bj regularization functions λj corresponding weights (Lagrange multipliers). regularization functions stabilize refinement enforce smoothness correction factors correction factors a b c smoothness enforced minimizing second derivativeapproximations56 integrals used[12pt]{minimal}{amsmath{wasysym{upgreek\setlength{\oddsidemargin}{-69pt}\begin{document}$$\int d\phi dxdy\left\right\end{document ∫dφdxdy∂φ2a2[12pt]{minimal}{amsmath{wasysym}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$\int d\phi dxdy\left\partial{2}a\right\end{document} ∫dφdxdy∂x2a+∂y2a2[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{amsbsy}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$\int d\phi\left\partial\right\end{document} ∫dφ∂φ2b2[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document}$$\int d\phi ds{\left\right{2\end{document} ∫dφds∂φ2c2[12pt]{minimal}{amsmath{wasysym}{amsfonts{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$\int d\phi ds{\left\partial{2}c\right{2}$$\end{document} ∫dφds∂s2c2. offset correction positive magnitude minimized approximation[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$\int d\phi ds{\left\right{2}$$\end{document} ∫dφdsc2. detector correction factors regularized\documentclass[12ptminimal{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin{-69pt}\begin{document}\sum{p}\left|d(p)-1\right|}^{2}\end{document ∑pd(p)−12 ensures d = 1 absence data nonlinear minimization problem solved minimizing\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin}{-69pt}\begin{document\mathcal{H}}\end{document}H constant Imerge linear problem updating Imerge new scale factors57 parameters refined individually pairs others held fixedresults refining corrections sequence\documentclass[12pt{minimal\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}\left{b,o,cb,o,c,a,d\end{document}b,o,cb,o,c,a,d cb fitting model c b o outlier step (Supplementary Figs. 6, 7) scaling model redundant observations scattering map merged Observations more than 5σ from mean excluded estimated Bragg intensities masked voxels scaled merged offset correction omitted outlier cutoff 2σ used merged values compared with Bragg intensities xds50 aimless51 single scale factor found xds/aimless values Bragg intensity map intensities xds aimless accurate/aimless values used subsequent analysis Bragg intensities values structure determination.Placement intensities absolute scaleAfter merging intensities placed absolute scale overall scale factor found adaptation total intensity method Krogh-Moe (K-M)26,27.standard K-M method Section 2.3 Supplementary Methods contribution atoms to total intensity (Itotal,predicted Equation 40 comparing prediction to measured value Equation 41 determine scale factor α\documentclass[12pt]{minimal}{amsmath\oddsidemargin-69pt}{document}\alpha ={total,measured}}}}{document}α=Itotal,predictedItotal,measured convergence scaling factor calculated standard K-M method modified K-M method inter-atomic interference (Supplementary Fig. 8). standard K-M method estimate atomic inventory unit cell theoretical coherent incoherent scattering for atoms scattering included 290 water molecules 6 nitrate ions 1 lysozyme molecule total 1546 H, 613 C 199 N 493 O 10 S atoms total number electrons Z = 10720 Itotal,predicted calculated using Equation 40 Methods region reciprocal spacemodified K-M method identical to standard Itotal modified interference between atom pairs distance predicted from chemical structure protein sequence structure water covalent bonding torsional restraints considered Molecular coordinates from chemical component dictionary58 pair distances between atoms adjacent amino acids calculated assuming planar peptide bond bond length 1.33 Å resulted in 7405 pair distances for lysozyme 6 nitrate 3 per water molecule bonding correction calculated added to elastic scattering in Equation 40 Supplementary Methods\documentclass[12pt{amsmath-69pt{bond)=2\sum}Ibond)=2∑n<mfn(s)fm(s)sin(2πsrnm)2πsrnm f atomic scattering factor sum over bonded atom pairs rnm inter-atomic distance synthetic data methods converge to expected value 1 for α (Supplementary Fig.modified method converges quickly provides accurate scale factor resolutions~2 Å typical macromolecular crystallography final diffuse map intensities generated fine 13, 11 subdivisions cell vectors a b c* resolution range 25–1.25 Å (scattering vector 0.04–0.8 Å−1) Voxels Bragg peaks excluded Geometric background corrections applied redundant observations merged scaling model Errors estimated Poisson statistics propagated through correction scaling merging steps intensities more than 5σ from mean flagged outliers excluded Intensities placed on absolute scale scale factor α theoretical incoherent scattering subtracted 36 correlation coefficient half-datasets unmerged observations randomly assigned merged separately final map includes isotropic scattering component elastic scattering-atom molecular dynamics simulationFour simulations triclinic lysozyme crystals performed with 1 27 (3 125 (5 343 (7 × × unit cells 12-cell simulation simulation prepared AMBER suite version 1860 ff14SB force field SPC/E model general Amber force field parameters nitrate ionsimulation boxes room-temperature cell PDB ID 4lzt (a 27.24 b 31.87 c 34.23 Å α 88.52∘ β 108.53∘ γ 111.89∘). simulation initialized protein atom coordinates PDB ID 4lzt supercell grid Nine nitrate ions added per cell 290 water molecules adjusted ~1 atm pressure at 295 K 290 293 284 270 waters per protein chain 1 27, 125 343 cell simulations simulations equilibrated 0.2 μs continued 5 5 2 1 μs saving coordinates every ns time step 4 fs non-water hydrogen masses 3 amu decrease mass bonded structure factors calculated from atomic coordinates ccp452 program resolution 0.95 Å VDWR parameter 3.0 B-factor set 15 Å2 atom“snapshot” B-factor smooths electron density distribution Fourier transforms converged result tested calculations grid points B-factor between 5 20 Å2. every atom assigned same B-factor effect undone multiplying structure factors from sfall\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}\exp \left(+B{s}^{2}/4\right)\end{document}exp+Bs2∕4.Bragg intensity per unit cell (Equation 16 Supplementary Methods calculated using\documentclass[12pt]{minimal{amsmath{wasysym{upgreek-69pt}}{I}_{B}={N}^{-2} F({h}_{0}{2}}IB=N−2F(h0)2 F(h0) is supercell structure factor Bragg positions h0 N number unit cells brackets average over saved simulation frames Section 1 Supplementary Methods). diffuse scattering per unit cell (Equation 15 Methods calculated using\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek-69pt}}{I}_{D}={N}^{-1} {F}^{2}\rangle -^{2}){document}ID=N−1(F2−F2) procedure in md2diffuse.sh script AmberTools distributionorg).Lattice dynamics model refinement in Matlab Protein molecules modeled as rigid bodies lattice contacts as elastic network13,37,38,66,67 with pair-wise interactions between α carbons lattice contacts identified in all-atom structure (PDB ID 6o2h). atoms with alternate conformers assigned to occupancy-weighted average positions atomic coordinates of 26 nearest neighbors in lattice generated crystal symmetry operators lattice contact defined between atom in central protein chain within 4 Å of atom neighbor network reduced to Cα model If atoms residues formed lattice contact spring created between Cα atoms 100 intermolecular springs modeled 50 unique due to crystal symmetry.Two types of pair potential modeled Gaussian and directional\documentclass[12pt{minimal\usepackage{amsmath-69pt{document}}{j\prime}}{Gauss=12γjj′u(j(j′)2and6\documentclass[12pt]{minimal{amsmath{upgreek\oddsidemargin-69pt}{document}$${V}_{j{j}\prime}}{dir.}}\frac{1}{2}{\gamma{j{j}{\prime}}\left\bf{u}}}{(j)}{j\prime}\cdot\bf{r}}}}{(j{j}{\prime}}\right}^{2}\end{document}Vjj′(dir.)=12γjj′(u(j)−u(j′))⋅r^(j,j′)2 j\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek}{\oddsidemargin}{-69pt}{document}$${j}^{\prime\end{document}j′ node indices u displacement vector node from position γ spring constant[12pt]{minimal}{amsmath}{wasysym}{\oddsidemargin}{-69pt}{document}\bf{r}}}}{(j,{j}^{\prime}\end{document}r^(j,j′) unit vector pointing node j[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}}}{upgreek}{\oddsidemargin}{-69pt}{document}$${j}^{\prime}\end{document}j′.equations motion solved rigid-body vibrational coordinate system Born/Von-Karman method (Section 3 Supplementary diffuse scattering calculated 13 × 11 × 11 periodic supercell detail experimental map one-phonon approximation (Equation 62 one-phonon structure factor (Equation 63 calculated fast-Fourier transform form factors approximated four Gaussians constant69,70 scattering contribution solvent density modeled Babinet’s principle constant electron density changing structure factor = constant density ρsolv. protein modeled density 0 −ρsolv. solvent-excluded region excluded solvent represented pseudo-atoms Gaussian form factors excluded voxels solvent mask REFMAC554 divided atoms proximity constant mask density approximated three-dimensional anisotropic Gaussian solvent scaling parameters ksolv. Bsolv. adjusted minimize least-squares difference Fobs\documentclass[12pt]{minimal}{amsmath{-69pt}\begin{document}\left{F\rm{model\right\end{document}Fmodel defined[12pt]{minimal}{amsmath{wasysym}{upgreek}}{-69pt}{document}{F}_{{\rm{model}}}={F}_{{\rm{calc.}}}+{k\rm{solv.}}}-{B}_{{{solv.}}}{s}^{2}/4)\end{document}Fmodel=Fcalc.+ksolv.exp(−Bsolv.s2∕4)Fsolv. Fcalc. structure factor modeled atoms (protein solvent Fsolv. factor excluded solvent refined parameters ksolv. Bsolv. applied to excluded-solvent form factors resulting pseudo-atoms included in list atoms unit cell assigned same rigid group as nearest protein atomSpring constants refined minimize least-squares difference simulated one-phonon scattering measured variational scattering around 400 intense halos 2–2.5 Å resolution range limited resolution range used for refinement agreement model assessed throughout reciprocal space (Fig. 2d). reduced χ2 refinement calculated\documentclass[12pt{minimal}{amsmath{upgreek\oddsidemargin-69pt}\begin{document}{red.}\left\sum{n = 1\sum{m=1}\end{document}χred.2=∑n=1NMn−1 ∑m=1MnIn(meas.(calc.)−bnσn,m2 N = 400 number halos Mn measured voxels around nth halo (typically = 13 × 11 × 11 − 1 = 1572) bn arbitrary constant offset for each halo least-squares minimization σ experimental uncertainty.spring constants refined in four stages first stage springs set to Gaussian assigned same constant second stage springs same interface given same spring constant third stage pair-potentials each interface linear combination Gaussian directional Finally each pair potential refined individually with linear Gaussian directional springs overall χ2 monitored during refinement adding extra degrees freedom fit (Supplementary Fig. 13).After scattering calculated reciprocal space one-phonon approximation (Equation 62 Supplementary Methods). Pearson correlation coefficient (CC) between measured variational scattering map simulation calculated within constant resolution 0.04 Å−1 to 0.80 Å−1 constant width Δs = 0.02 Å−1 each resolution bin CC calculated\documentclass[12pt]{minimal}\usepackage{amsmath\oddsidemargin-69pt{document\rm{CC}}=\frac{\sum{n}\left{meas\right)\sqrt{\sum{n}{\left_{{\rm{meas.}}}-{\bar{I}}\right}{2}\sum{n{\left_{{\rm{calc.}}}{n-{\bar{I}}\right}^{2}}}\end{document}CC=∑nImeas.(sn.Icalc.\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{upgreek{\oddsidemargin}{-69pt}\begin{document}$${\bar{I}}_{{\rm{meas.}}}\end{document}Īmeas.\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$${\bar{I}}_{{\rm{calc.}}}\end{document}Īcalc.mean intensities in resolution bin sums over voxels Fig. 14). comparison MD simulation calculated on coarser 7 × 7 × 7 sub-sampled reciprocal lattice full map interpolated at voxels coarser fitting 2nd order polynomial over voxels 3 × 3 × 3 Voxels at Bragg positions integer Miller indices excluded CC calculated between interpolated simulated map experimental map (Fig. 2d).Internal protein dynamics refinement in Matlab dynamics lysozyme crystal environment simulated all-atom elastic network residue rigid body lattice contacts modeled atoms alternate conformers assigned occupancy-weighted positions springs created between non-H protein atoms cutoff distance 4 Å Intermolecular springs modeled between atoms protein lattice 4 Å springs directional type (Eq. equations of motion single unit cell solved Born/Von-Karman method lattice dynamics potential energy function modified remove modes rigid-body motion protein displacement restoring force of zeronormal modes rigid-body displacements have eigenvalues zero eliminated during inversion dynamical matrix Section 3.3 Supplementary Methods). components Hessian matrix (Equation 45 modified\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}}{mathrsfs}{upgreek}-69pt}{document}\bf{V}}}\boldsymbol{l}}={{\bf{P}}}\bf{T}}}{{{V{P}}\end{document}V (l,l′):=PTV (l,l′)P P operator projects out rigid-body component displacement\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}{amsbsy}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}{\bf{P}}={\bf{I}}-\left(\bf{A}}}\right\left\end{document}P=I−A\A0A0\A,I 6m × 6m identity matrix = 129 A 3n × 6m matrix atoms protein transforms Cartesian atomic displacement coordinates u internal dynamics model (Equation 43 Supplementary A0 3n × 6 matrix transforms u lattice dynamics model forward slash left matrix division X = A\A0 least solution equations AX A0)model parameterized one coupling constant per residue spring connecting atoms (j\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}$${j}^{\prime}\end{document}j′) has spring constant equal to geometric mean residues’ coupling constants gi[12pt{upgreek-69pt}\begin{document}$${g}_{{i}^{\prime}}\end{document}gi′\documentclass[12pt]{minimal}{amsmath{wasysym{amsbsy{mathrsfs{upgreek{\oddsidemargin}{-69pt}\begin{document}$${\gamma }_{j,{j}^{\prime}}=\sqrt{{g}_{i}{g}_{{i}^{\prime}}}\end{document}γj,j′=gigi′.parameters optimized minimize χ2 between measured simulated atomic displacement parameters calculated\documentclass[12pt]{minimal}\usepackage{amsmath{upgreek\oddsidemargin}{-69pt}\begin{document}$${\chi^{2}=\sum{j=1}^{N}\sum{n=1}^{9}{\left\bf{U}}}_{j}^{meas\right{n{latt{int\right{n}}^{2}\end{document}χ2= ∑j=1N ∑n=19Uj(meas.)n−Uj(latt.)+Uj(int.\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin}-69pt}{document}\left({{\bf{U}}}\right{n\end{document nth component ADP atom j (Uj 3 × 3 matrix 9[12pt]{minimal}{amsmath}{wasysym}}{upgreek}\oddsidemargin}{-69pt}{document}\bf{U}}}{j}{latt.}}\end{document}Uj(latt.) calculated ADP fully-refined lattice dynamics model[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\begin{document}${{\bf{U}}}{j}\rm{int.}}\end{document}Uj(int.calculated ADP internal dynamics model (Equation 56 Supplementary refining model displacement correlations assessed directional correlation coefficient defined\documentclass[12pt]{minimal{amsmath\oddsidemargin-69pt}}${{\rm{CC}}}{j,{j}{\prime}}\hat{{\bf{r}}}}_{j,{j}{\prime}}\rm{T}}}\left\langle\bf{u}}}{j\bf{u}}}{\prime}}{T\rangle\hat{{\bf{r}}}}{j,{j}{\prime}}}{\rm{Tr}}{{\bf{U}}}_{j}/3)(^{\prime}}/3\end{document}CCj,j′=r^j,j′Tujuj′Tr^j(TrUj∕3)\documentclass[12pt]{minimal}{amsmath}{wasysym}\oddsidemargin}-69pt}{document}\hat{{\bf{r}}}}{j,{j}^{\prime\end{document}r^j,j′ unit vector atom j[12pt]{minimal}\usepackage{amsmath}{wasysym{amsfonts{mathrsfs{upgreek{\oddsidemargin}{-69pt}{document}{j}^{\prime\end{document}j′ defined alternate model internal protein motion modes rigid-body displacements suppressed Residues assigned three 5–36 98–129 to α 40–94 to β remaining hinge region Hessian matrix modified P operator Eq.(10) calculated\documentclass[12pt{minimal\usepackage{amsmath{upgreek\oddsidemargin-69pt}\begin{document}\bf{P}}=\bf{I}}-\left\right\end{document}P=I−A\A1A1\A I 6m × 6m identity matrix (m = 129 number residues), A1 3n × 6d matrix (d = 3 number groups generalized coordinate system 3-group model to atomic displacements (Equation 43 model parameterized refined unrestrained model directional correlation calculated using Eq. (14).Diffuse Patterson map calculated in Matlab Fourier transform of diffuse scattering 1.3 three-dimensional fast-Fourier transform experimental diffuse map pre-processed compensate missing data missing voxels filled with mean values neighboring mean intensity resolution shell subtracted voxels beyond resolution limit filled with zerosdata array zero-padded diffuse Patterson map real-space voxel 0.3 Å voxel dimensions a 91 b 107 c 115 lattice diffuse Patterson map refined vibrational model calculated approximation central region r < 25 Åcalculation scattering per unit cell (Equation 59 rearranged single reference unit cell (l = 0[12pt{minimal{amsmath\oddsidemargin{-69pt}{document}{I}{D}=\sum{j}{f}_{j}\left\sum}{\prime}{\prime}}{f}{\prime}}{2\pi\bf{s}}({{\bf{r}}}{j\prime}}{\prime}}{0}{T}_{j}{T}_{{j{\prime}}\left({T}_{0j{l{\prime}{j{\prime}}-1\right\end{document}ID ∑jfj∑l′,j′fj′e2πis⋅rj−rj′−rl′+r0TjTj′T0j,l′j′−1 first sum runs over all atoms unit cell fi atomic scattering factor Tj Debye-Waller factor (Equation 60[12pt]{minimal}{amsmath}{mathrsfs}{upgreek}-69pt}}{T}{0j{l{\prime}{j^{document}T0j,l′j′ depends cross-terms covariance matrix (Equation 61 Supplementary Methods l = 0). term curly brackets resembles standard structure factor equation primed atoms origin shifted Debye-Waller factor replaced\documentclass[12pt]{minimal}{amsmath}{wasysym}}}}{mathrsfs}{upgreek}-69pt}}\left({T}{0j{l}{\prime}{j={T}{j{T{\prime}}{0j\prime{document}Teff0j,l′j′=TjTj′T0j,l′j′−1.Debye-Waller factor separated lattice internal motion:18\documentclass[12pt]{minimal}\usepackage{amsmath}}{upgreek}{\oddsidemargin-69pt}{document}{T}_{{\rm{eff}}}={T}\rm{eff{latt+{T{int}}}\end{document}Teff=Tefflatt+Teffint.lattice term calculated\documentclass[12pt]{minimal{amsmath{wasysym\oddsidemargin-69pt}{document}$${\left({T}_{{\rm{eff}}}{latt}}}\right{0j{l}{\prime}{j}{\prime}}={T}_{j}{T}\prime}}\left({T}{0j{l\prime\rm{latt}}}-1\right),\end{document}Tefflatt0j,l′j′=TjTj′T0j,l′j′latt−1 experimentally determined ADPs used Tj[12pt]{minimal}{amsmath}{wasysym}}}}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$${T}_{{j}^{\prime}}\end{document}Tj′[12pt]{minimal}{amsmath{wasysym}{upgreek}{\oddsidemargin}{-69pt}document$${T}_{0j{l}{\prime}{j}^\rm{latt}}}}T0j,l′j′latt calculated refined lattice model (Equations 55 61 corresponds definition one-phonon simulation (Equation 62 internal motions Debye-Waller factor calculated\documentclass[12pt]{minimal{amsmath\oddsidemargin{-69pt}{document}$${\left({T}\rm{eff{int\right}{0j,{l}{\prime}{j}{\prime}}={T}_{j}{latt{T}{j}{int{T}{\prime}}{latt{T{0j{l}{\prime}{j}{\prime}}{latt}}}\left({T}{0j{l}{\prime}{j}{{int}}}-1\end{document}Teffint0j,l′j′,l′j′lattT0j,l′j′int−1 T’s calculated covariance matrices latticedynamics simulations-solvent effects modeled by pseudo-atoms with Gaussian scattering factors for lattice dynamics simulation In Eq. (16) terms in curly brackets calculated using FFT-based method each atom had Debye-Waller factor coordinates relative to rj central part of Patterson desired sum carried out over all atoms in unit cell 26 nearest neighbors\documentclass[12pt{minimal{amsmath-69pt}{document\bf cutoff distance larger than maximum 25 Å to avoid truncation artifacts). calculating diffuse intensity map Eq. 16 mean intensity in each resolution shell subtracted voxels outside experimental resolution limit 1.25 Å set to zero map zero-padded Patterson function calculated using three-dimensional FFT reciprocal space correlation coefficients between diffuse Patterson maps calculated in Matlabreal space voxels[12pt{minimal{amsmath{wasysym\oddsidemargin{-69pt}}\left{document}r<2[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin}{-69pt} 25\end{document}r>25 Å set zero maps truncated ∣x∣ < a ∣y∣ < b ∣z∣ < c reciprocal space map oversampled factor 2 each direction inverse FFT each truncated map calculated Pearson correlation coefficients (Eq. (9) experimental simulated intensity maps calculated constant resolution 0.04 to 0.80 Å−1 bin widths Δs = 0.04 Å−1.Supplementary information Peer Review File Additional Supplementary Files Supplementary Movie 1 Movie 2 | 47.9 | 1.879448 |
10.1038/s41467-020-20674-3 | PMC7815847 | Suppression of terminal differentiation is essential for epidermal progenitor maintenance. Here, the authors show that intronic polyadenylation is dynamically regulated by the cooperation between CPSF and RNA-binding proteins to influence epidermal differentiation gene expression. | In self-renewing somatic tissue such as skin epidermis, terminal differentiation genes must be suppressed in progenitors to sustain regenerative capacity. Here we show that hundreds of intronic polyadenylation (IpA) sites are differentially used during keratinocyte differentiation, which is accompanied by downregulation of the Cleavage and Polyadenylation Specificity Factor (CPSF) complex. Sustained CPSF expression in undifferentiated keratinocytes requires the contribution from the transcription factor MYC. In keratinocytes cultured in undifferentiation condition, CSPF knockdown induces premature differentiation and partially affects dynamically used IpA sites. These sites include an IpA site located in the first intron of the differentiation activator GRHL3. CRISPR knockout of GRHL3 IpA increased full-length GRHL3 mRNA expression. Using a targeted genetic screen, we identify that HNRNPA3 interacts with CPSF and enhances GRHL3 IpA. Our data suggest a model where the interaction between CPSF and RNA-binding proteins, such as HNRNPA3, promotes site-specific IpA and suppresses premature differentiation in progenitors. | IntroductionSelf-renewing somatic tissue, such as epithelium, undergoes continuous turnover to compensate for wear and tear. In this dynamic regeneration process, terminal differentiation is essential for fulfilling the specialized tissue function; However, terminal differentiation genes must be suppressed in tissue progenitors to sustain their regenerative capacity1,2. The molecular mechanisms underlying this spatiotemporal regulation of differentiation genes, in somatic tissue homeostasis, still remains incompletely understood.Human skin epidermis, a stratified epithelium, is a highly accessible research platform for exploring gene regulatory mechanisms governing somatic tissue differentiation. Primary epidermal cells (keratinocytes) can be isolated and expanded in culture, in both undifferentiation and differentiation conditions, facilitating the integration of genomic, proteomic, and genetic tools into this platform1–5. Decades of research identified that epidermal differentiation involves the upregulation of multiple transcription factors, which bind to their chromatin targets and further activate barrier function5,6. One of these transcription activators is GRHL3, which is a selective late-differentiation activator that promotes protein crosslinking and cornified envelope formation7–9. GRHL3 is repressed in epidermal progenitors. Several distinct regulatory mechanisms, utilized by the progenitors to repress GRHL3, have been identified recently. These include PRMT1 binding at its promoter to influence its transcription, as well as EXOSC9 degrading mRNA post-transcriptionally2,10. In addition to the roles of transcription activators, epigenetic and post-transcriptional regulators as well as non-coding RNAs are also involved in regulating epidermal differentiation1,6,11–18. These findings highlight the coexistence of multiple gene regulatory mechanisms, at distinct steps of gene expression, to fine-tune the overall abundance of gene products.About ~30% of the human genome is composed of introns19. Despite often being viewed as “junk”, introns’ influence on gene expression is being increasingly appreciated. Notably, polyadenylation sites have been identified in introns in addition to the well-established 3′UTR regions20. Usage of these intronic polyadenylation (IpA) sites terminates transcription prematurely. This IpA mechanism, although still under-studied, has been explored in several developmental and physiological processes. In muscle regeneration after injury, an IpA event in platelet-derived growth factor (PDGFRα) generates a short-isoform decoy to suppress full-length gene function and attenuate muscle fibrosis21. During the late stage of spermatogenesis, sterol regulatory element binding transcription factor 2 (SREBF2) switches from full-length to a short isoform using IpA to control germ-cell specific gene expression22. Recent transcriptome-wide profiling further identified recurrent IpA sites that inactivate tumor suppressor genes in leukemia23, as well as the diverse IpA events in the immune system in response to various cellular environments24. How distinct IpA sites are being used in specific biological processes still remains unclear.Polyadenylation requires cleavage of nascent RNAs, a process that involves the participation of multiple protein complexes. The cleavage stimulation factor (CstF) as well as the cleavage factors CF I and CFII bind to upstream and downstream elements25. The polyadenylation process is catalyzed by the polyadenylate polymerase (PAP). Notably, a central player of this process is the Cleavage and Polyadenylation Specificity Factor (CPSF) complex26. CPSF binds to the AAUAAA consensus sequence, the most important cis-regulatory element in cleavage and polyadenylation, through its CPSF4 and WDR33 subunits. CPSF3 is the endonuclease subunit that directly cleaves nascent RNA. These components, as well as two other regulatory subunits CPSF2 and FIP1L1, are brought together by a large scaffolding subunit CPSF127–30. Intriguingly, the expression levels of CPSF subunits vary among different cell types31. Elevated CPSF expression was also observed in somatic cell reprogramming as well as in cancer32,33. How differential CPSF levels influence gene expression in somatic tissue, and what functions upstream to control CPSF expression, are still open questions.In this study, we show that hundreds of IpA sites are differentially used by the transcription machinery during keratinocyte differentiation, which is accompanied by the reduction of CPSF expression. The enriched expression of CPSF in the progenitor state is downstream to MYC, and is essential for suppressing premature differentiation. We find that CPSF expression level influences a subset of differential IpA sites during keratinocyte differentiation. These CPSF-dependent IpA sites include an IpA site located in the first intron of GRHL3. Our CRISPR KO experiments show that the usage of this GRHL3 IpA site suppresses full-length GRHL3 mRNA expression. Using a combination of protein complex purification coupled with a genetic screen, we further identified HNRNPA3 as a key CPSF-interacting RNA-binding protein that enhances GRHL3 IpA usage. Taken together, our findings support a working model where CPSF cooperates with distinct RNA-binding proteins to modulate IpA usage in a site-specific manner, shaping the transcriptome of undifferentiated keratinocytes and suppressing premature differentiation.ResultsKeratinocyte differentiation involves altered IpA usage and downregulation of CPSFWe leveraged the 3′READS+ technique, which features high sensitivity and strand specificity34, to map transcriptome-wide PolyA sites in both undifferentiated and differentiated human keratinocytes. For every PolyA site identified in each gene, “Fraction of PolyA site Usage (FPU)” was calculated as the counts at this site divided by the sum of the counts from all PolyA sites in this gene (Supplementary Fig. 1a). With this quantification method, FPU is internally normalized within each sequencing library, as it calculates the fractions within individual genes. This method circumvents the technical challenge of normalizing PolyA site usage across different libraries, especially when gene expression is different between the two conditions. In total, 14625 PolyA sites were identified to robustly associate with 7990 expressed genes in keratinocytes (Fig. 1a). In undifferentiated or differentiated keratinocytes, the FPU of these sites is highly correlated between the replicates (Fig. 1b, c). The FPU is more divergent comparing undifferentiated versus differentiated keratinocytes (Fig. 1d), suggesting genome-wide changes in PolyA site usage between these two conditions. Among all these PolyA sites, 2739 of these sites are located in intronic regions (Supplementary Fig. 1b), 17% of which overlaps with the IpA sites recently identified in the immune system24 (Supplementary Fig. 1c, d), highlighting the specificity of IpA events occurring in distinct tissue types. Both lists of IpA sites partially overlap with the PolyA_DB3 database35 (Supplementary Fig. 1e, f). To identify the differentially used IpA sites between undifferentiated and differentiated keratinocytes, we integrated 3′READS+ data analysis with strand-specific RNA-seq data validation. In total, 428 differentially used IpA sites were identified (fold change > = 2 from 3′READS+, fold change ≥1.5 from RNA-seq, Fig. 1e, Supplementary Data 1). These IpA sites are highly enriched with the AAUAAA motif (E value: 1.9 × 10−165). Notably, these differentially used IpA sites include more downregulated (60%, 256/428) than upregulated sites (40%, 172/428). Examples of these differentially used IpA sites include a downregulated IpA site in ESPN (cytoskeleton regulator), and an upregulated IpA site in the IQCK (EF-hand protein binding; Fig. 1f, g). Thus, dynamic IpA usage occurs during the cell-fate switch from the progenitor state to terminal differentiation.Fig. 1Keratinocyte differentiation involves IpA alterations and CPSF downregulation.a Illustration showing the workflow of analyzing IpA sites in undifferentiated (UD) and differentiated (DF) keratinocytes. b–d Scatter plots showing the correlation of the 3′READS+ data between replicates, and between undifferentiated and differentiated conditions. e Heat map showing fold change of the 428 differentially used IpA sites during keratinocyte differentiation, in both 3′READS+ data and RNA-seq data. f, g Genome browser tracks showing the differential IpA usage in both EPSN and IQCK during keratinocyte differentiation, with both 3′READS+ and RNA-seq data. (Beige highlight: 3′UTR; Green highlight: UD-enriched IpA; Red highlight: DF-enriched IpA). h Illustration showing the composition and function of the CPSF complex. h Heat map comparing the relative mRNA levels of CPSF subunits and POLR2A between undifferentiated and differentiated keratinocytes. i qRT-PCR comparing the mRNA expression of CPSF genes between undifferentiated versus differentiated keratinocytes. Dots represent data points in technical replicates. j Immunoblots comparing the expression of CPSF subunits at the protein level between undifferentiated and differentiated human keratinocytes. β-tubulin was used as loading control. Average fold change and SD quantified from 3 replicates are indicated blow each panel. Source data are provided as a Source Data file.In addition to these IpA sites, 11,506 sites were identified in 3′UTR from our 3′READS+ analysis. 4677 of these sites correspond to genes with single PolyA sites in the 3′UTR, while 6829 sites were associated with 2727 genes that feature 2 or more PolyA sites in the 3′UTRs. To determine if 3′UTR shortening or lengthening could be a feature of keratinocyte differentiation, we calculated the fraction of distal PolyA site usage within 3′UTR of these 2727 genes, by taking the total counts from the most distal site divided by the sum of counts from the 3′UTR of that gene. This identified 211 genes with increased distal PolyA usage and 457 genes with decreased distal usage (fold change > 1.5, Supplementary Fig. 1g–i). These data indicate that both lengthening and shortening of 3′UTR occur during keratinocyte differentiation, although shortening occurs in more genes. This trend is similar to the findings in spermatogenesis where shortening of the 3′UTR was observed36.The polyadenylation process involves the participation of multiple complexes, including CPSF, CstF, CF I, CF II and PAP. The expression of PAP remains relatively constant between undifferentiated and differentiated keratinocytes, based on our RNA-seq data. Interestingly, the subunits encoding other complexes in this process are more dynamic (Supplementary Fig. 1j), including the downregulation of CPSF, the central player of cleavage and polyadenylation. Using both qRT-PCR as well as western blotting, the downregulation of core CPSF complex subunits in differentiation was confirmed at both mRNA and protein levels (Fig. 1h–j). Among the CPSF antibodies we used for Western blotting, the CPSF2 antibody worked in immunostaining of human skin sections, which exhibited stronger signals in the basal progenitor layer of the epidermis and reduced signal in differentiated layers. In comparison, the immunostaining of RNA polymerase II (Pol II) showed strong signals throughout both undifferentiated and differentiated layers of the epidermal tissue (Supplementary Fig. 1k). These data suggest that CPSF downregulation occurs during keratinocyte differentiation.Suppression of CPSF in undifferentiated keratinocytes induces differentiationThe differential expression of CPSF raised the question of whether altered of CPSF level impacts keratinocyte function. Leveraging “ON TARGETplus” siRNA that simultaneously targets 4 different regions of CPSF1, we performed nucleofection in keratinocytes cultured in undifferentiated condition. A non-targeting control pool of four siRNAs was nucleofected in parallel as a negative control. The efficacy of CPSF1 RNAi was confirmed at both mRNA and protein levels (Fig. 2a and Supplementary Fig. 2a). In addition to CPSF1, the protein levels of other CPSF complex subunits were reduced. This was likely caused by protein degradation with loss of the CPSF1 scaffold, as the mRNA levels of these CPSF subunits were minimally affected (Supplementary Fig. 2b).Fig. 2CPSF Downregulation in progenitors impairs self-renewal and induces differentiation.a qRT-PCR quantification of CPSF1 knockdown efficiency. Dots represent data points in technical replicates. b Epidermal tissue regenerated by 50% CTRLi labelled by H2B-GFP (green), and 50% CTRLi or CPSFi labelled by H2B-mCherry (red). Scale bar: 100 μM. Representative images from 25 images per condition are shown. c Quantification of red:green ratio comparing tissue sections of CTRLi/CTRLi versus CTRLi/CPSFi (n = 25 images, p = 6.55 × 10−18, t-test, 2 tailed, error bars are presented as mean values ± SEM). d Diagram showing the design of CPSF1 CRISPRi. The locations of three independent sgRNAs (sg1, sg2, and sg3) are labelled relative to the transcription start site (TSS). e qRT-PCR quantification comparing the knockdown efficiency between CPSF CRISPRi versus control. Dots represent data points in technical replicates. f, g Clonogenic assay comparing keratinocytes with CPSF1 CRISPRi vs. non-targeting control. Colonies with diameter > 1/16 inch were quantified. Dots represent data points in technical replicates. h Venn diagram comparing the differentially expressed genes in CPSF siRNA vs. CPSF CRISPRi. These two data sets significantly overlap with each other (Fisher’s exact test, 2-Tail, p = 1.1 × 10−308). i Heat map showing fold change of 739 differentially expressed genes (fold change > 2, P < 0.05, Wald test) shared between CPSF1 RNAi and CRISPRi. j Bar graph showing the top GO term associated with the genes significantly altered by CPSF CRISPRi (P values: modified Fisher’s exact test). k, l qRT-PCR comparing mRNA levels of differentiation marker genes between CPSF1 knockdown vs controls. Dots represent data points in technical replicates. Source data are provided as a Source Data file.The regenerative capacity of keratinocytes with CPSF1 knockdown was assessed using a progenitor competition assay. In brief, epidermal tissue was regenerated using 50% keratinocytes expressing H2B-GFP and 50% of keratinocytes expressing H2B-mCherry. The H2B-GFP keratinocytes were treated with control siRNAs as an internal control; the H2B-mCherry keratinocytes were treated with CPSF1 siRNA or control siRNA. The regenerated epidermal was sectioned, and the ratio of mCherry-labelled nuclei versus GFP-labelled nuclei in the tissue sections was quantified using ImageJ37. CPSF1 knockdown strongly reduced the representation of mCherry-labelled cells as compared with control in the regenerated epidermal tissue (Fig. 2b, c). This reduction of mCherry-labelled cells was even more drastic in the bottom half of the epidermal tissue, as compared to the top more differentiated half (Supplementary Fig. 2c, d). These data suggest that CPSF1 knockdown impaired the regenerative capacity of keratinocytes. Consistent with the progenitor competition assay, epidermal tissue regenerated entirely using keratinocytes treated with CPSF1 RNAi was hypoplastic. The lipid stain Nile Red showed a significant increase in epidermal tissue with CPSF RNAi, suggesting enhanced barrier function to prevent water loss. No statistically significant difference was detected for apoptosis using the TUNEL assay (Supplementary Fig. 2e–h). In addition, CPSF RNAi also impaired keratinocyte migration in scratch wound healing assay (Supplementary Fig. 2i, j). Thus CPSF knockdown impaired multifaceted functions of keratinocytes critical for physiological processes such as tissue homeostasis and wound healing.In addition to RNAi, we used CRISPRi38 as an orthogonal method to suppress CPSF1 expression. This strategy involved the expression of the enhanced CRISPR repressor, which includes both KRAB and MeCP2 fused with the nuclease-inactive dCas9 for improved repression39. Three independent sgRNAs near the transcription start site (TSS) of CPSF1 were designed and expressed individually with dCas9-KRAB-MeCP2 in undifferentiated keratinocytes. All three independent sgRNAs successfully downregulated CPSF1 expression at both the mRNA and protein levels (Fig. 2d, e and Supplementary Fig. 2k). Using clonogenicity assay, the colony-forming ability of keratinocytes was assessed in CPSF CRISPRi versus non-targeting controls. The colony numbers were strongly reduced with CPSF1 loss (Fig. 2f, g), supporting that the intact function of CPSF is essential for epidermal progenitor self-renewal.The validation of two knockdown approaches, RNAi and CRISPRi, allowed us to identify CPSF target genes that are altered in both. RNA-seq analysis identified 1113 genes from RNAi and 1584 genes from CRISPRi (fold change >=2, p < 0.05). The 739 genes shared in these two approaches are termed as “CPSF core targets” (Fig. 2h, i and Supplementary Data 2). The top enriched Gene Ontology (GO) terms of the upregulated core targets include “Keratinization” and “Keratinocyte differentiation”, while the top GO terms of the downregulated genes include “DNA replication” and “Cell division” (Fig. 2j). These GO terms are very similar to the top GO terms associated with the genes altered during keratinocyte differentiation6. Using qRT-PCR, we validated several representative differentiation genes that were upregulated in CPSF suppression, including the differentiation activator GRHL3 as well as differentiation marker genes SPRR1B, S100A8, and S100A9 (Fig. 2k, l). These data identified an essential role of CPSF in suppressing premature differentiation in epidermal progenitor maintenance.Sustained CPSF expression in undifferentiated keratinocytes requires MYCAs CPSF expression is downregulated in differentiated keratinocytes, we searched for mechanisms that could influence CPSF expression. We first asked if general impairment of proliferation and induction of differentiation in keratinocyte is sufficient to downregulate CPSF expression. Two independent strategies were tested. CPSF expression was first examined in keratinocytes with PRMT1 knockdown. PRMT1 was recently demonstrated as an essential regulator for sustaining proliferation and suppressing differentiation marker genes in undifferentiated keratinocytes2; however PRMT1 knockdown did not affect CPSF expression (Supplementary Fig. 3a). CPSF expression was also compared between early- and late-passage keratinocytes. Late-passage keratinocytes, after prolonged culture, expressed an increased level of differentiation markers such as p16 and S100A9, as well as reduced levels of cell cycle markers such as Ki67 and AURKB (Supplementary Fig. 3b). However, the expression of the CPSF subunits was not dramatically altered (Supplementary Fig. 3c). Thus, impaired proliferation and induced differentiation are not sufficient to downregulate CPSF expression in primary human keratinocytes.To identify specific regulators that could be essential for maintaining CPSF1 expression in undifferentiated keratinocytes, we searched for transcription factors that can bind to the regulatory regions of the CPSF1 gene leveraging publicly available ChIP-seq data40. We found that the MYC ChIP-seq signal is enriched at the CPSF1 promoter, in both keratinocytes as well as the breast cancer cell line MCF7 (Fig. 3a). MYC is downregulated during keratinocyte differentiation (Fig. 3b). To characterize the role of MYC in regulating CPSF expression, we first evaluated whether MYC could be essential for sustaining CPSF expression in undifferentiated keratinocytes. Two independent shRNAs targeting MYC were designed and validated using both qRT-PCR and Western blotting (Fig. 3c, d). Keratinocytes expressing these shRNAs showed downregulation of CPSF1, as compared to the non-targeting control shRNA (Fig. 3e). Since MYC is downregulated in differentiation, we also tested whether overexpression of MYC might be sufficient to increase CPSF expression in the differentiation condition. Between keratinocytes infected with the pCDH-MYC overexpression construct41 versus the vector control, no drastic differences were observed in the mRNA level of CPSF1 (Supplementary Fig. 3d, e). These findings suggest that MYC is essential for sustaining CPSF expression in the progenitor state, yet it is not sufficient to drive high expression of CPSF in the differentiation state.Fig. 3Sustained CPSF expression in undifferentiated keratinocytes requires MYC.a Genome browser tracks showing the binding of MYC at CPSF1 promoter in keratinocytes as well as in MCF7 cells (GSE32883). b Western blots showing the downregulation of MYC during keratinocyte differentiation. Quantification of MYC protein level fold change is indicated below (n = 3). c Western blotting showing the knockdown efficiency of shRNAs targeting MYC. Quantification is indicated below (n = 3). d, e qRT-PCR quantification of MYC and CPSF1 comparing MYC knockdown versus control. Dots represent data points in technical replicates. Source data are provided as a Source Data file.CPSF downregulation alters IpA usageSince keratinocyte differentiation involves both altered usage of IpA sites as well as reduced CPSF levels, we tested whether CPSF knockdown could influence IpA usage using 3′READS+. We identified that 178 out of the 428 IpA sites differentially used in keratinocyte differentiation were altered in the same direction with CPSF knockdown (Fig. 4a, b). In particular, 74% of these IpA sites showed reduced usage in CPSF knockdown or in keratinocyte differentiation. Genes associated with these CPSF-dependent IpA sites include ALOX15B, EPSN, CRBN, as well as the differentiation activator GRHL3 (Fig. 4c–f). Genes associated with CPSF dependent or independent IpA sites did not show drastic differences in gene expression (Supplementary Fig. 4a).Fig. 4CPSF Downregulation Partially Accounts for the IpA Alterations in Differentiation.a Pie chart showing 178 out of the 428 IpA sites altered in differentiation are influenced by CPSF1 knockdown. b Heat map showing the fold changes of these 178 IpA sites, comparing DF versus CPSFi. c–f Genome browser tracks comparing the differential IpA usage in DF and in CPSFi. Locations of the qPCR primers designed for IpA usage quantification are labelled below (brown arrowheads: proximal primers, green arrowheads: distal primers, beige highlight: 3′UTR, green highlight: UD-enriched IpA). g–j qRT-PCR quantification of the relative usage of the IpAs is calculated using the Proximal:Distal usage ratio, comparing undifferentiated vs. differentiated keratinocytes, CSPF1 siRNA and CPSF CRISPRi vs. control conditions, and MYC knockdown vs. control. Dots represent data points in technical replicates. Source data are provided as a Source Data file.To validate the altered usage of these IpA sites, we developed a qRT-PCR strategy. In brief, two pairs of qPCR primers were used for each gene of interest, with the first pair (proximal) designed immediately before the IpA site, and the second pair (distal) designed for exons between the IpA and the 3′ end of the gene. The relative usage of IpA can be quantified as proximal: distal ratio by using both pairs of primers in qRT-PCR. With this approach, we validated that the usage of these IpA sites in representative genes was strongly decreased in the context of keratinocyte differentiation, CPSF siRNA knockdown, CPSF CRISPRi, as well as MYC knockdown (Fig. 4g–j). Notably, the fold change of IpA usage for these IpA sites was less drastic in CPSF knockdown as compared to differentiation, indicating that the differentially used IpA sites in keratinocyte differentiation is partially influenced by CPSF downregulation. Usage of these IpA sites was minimally altered in migrating keratinocytes versus control (Supplementary Fig. 4b), suggesting that distinct biological processes occurring in the same cell type may involve different sets of IpA sites.CPSF suppresses GRHL3 expression by promoting IpA usageTo explore whether altered IpA usage influenced by CPSF knockdown might be linked to gene expression, we intersected the CPSF core targets from RNA-seq with the 165 genes that are associated with the 178 CPSF-dependent IpA sites. This identified a total of 14 genes (Fig. 5a and Supplementary Data 3). The majority of these 14 genes show anticorrelation of fold change between RNA-seq and 3′READS+, in CPSF knockdown or in keratinocyte differentiation. Among them, the IpA site associated with GRHL3 stood out for a couple of reasons. First, this GRHL3 IpA site features the highest FPU in undifferentiated keratinocytes among all the IpA sites associated with these 14 genes (Fig. 5b), suggesting that the usage of this IpA site could play an important role in influencing the overall mRNA expression of its host gene. Second, GRHL3 is a transcriptional activator that can further modulate the expression of other epidermal differentiation marker genes. Using double knockdown with CPSF CRISPRi in combination with GRHL3 RNAi, we confirmed that GRHL3 RNAi suppressed the induction of a number of differentiation markers that were induced by CPSF CRISPRi alone, including both mid-epidermal-differentiation markers (SPRR1B, S100A9, SPRR1A, S100A8) and late-epidermal-differentiation markers (SBSN and CRCT1; Fig. 5c–e). In particular, SBSN and CRCT1 were validated as direct targets of GRHL3, and the other four genes were also found to be downstream to GRHL3 in keratinocyte differentiation42. These data suggest that GRHL3 is a key downstream target mediating CPSF’s role in suppressing differentiation in epidermal progenitors, and that IpA could play a role in modulating GRHL3 expression.Fig. 5CPSF suppresses GRHL3 expression by promoting IpA usage.a Venn diagram comparing genes associated with CPSF-dependent IpA sites and genes altered by CPSF knockdown. 14 genes are shared in these two lists of genes. b Heatmap showing the relative expression from RNA-seq, relative usage of IpA sites from 3′READS+, and the FPU of IpA in undifferentiated keratinocytes from 3′READS+. GRHL3 IpA has the FPU in undifferentiated keratinocytes. c–e qRT-PCR quantification of knockdown efficiency and differentiation marker gene expression in CPSF1-GRHL3 double knockdown. Downregulation of GRHL3 using siRNA in the context of CPSF1 CRISPRi largely restored the differentiation gene inhibition in keratinocytes. Dots represent data points in technical replicates. f Illustration showing the two CRISPR strategies to disrupt the GRHL3 IpA site. One strategy uses a single sgRNA (sg1) targeting the PAM site immediately adjacent to the AAUAAA consensus sequencing of the IpA. The second strategy uses two sgRNAs (sg2 and sg3) to delete 1-kb genomic sequence containing the IpA site. g, h qRT-PCR quantification of GRHL3 full-length mRNA level, comparing CRISPR KO vs control. Dots represent data points in technical replicates. i, j qRT-PCR quantification of differentiation marker gene expression comparing CRISPR KO vs control. Dots represent data points in technical replicates. Source data are provided as a Source Data file.This differentially used IpA site of GRHL3 is located in its first intron, about 4.3 kb downstream of the transcription start site. The usage of this IpA site is drastically downregulated in differentiated keratinocytes and in CPSF knockdown, while GRHL3 mRNA expression is strongly upregulated. To determine if usage of this IpA site could suppress the expression of full-length GRHL3, we took two different CRISPR approaches to knock out this IpA site (Fig. 5f). In the first approach, we leveraged a PAM sequence (AGG) directly adjacent to the AAUAAA (AATAAA of DNA sequence) CPSF-binding consensus sequence, and targeted this site with a single sgRNA (sg1) to create small indels. In a second approach, we co-expressed two sgRNAs (sg2 + sg3) that are designed to delete 1 Kb intronic genomic sequence containing this IpA site. Both approaches achieved an average of 50% knockout efficiency in primary human keratinocytes, as estimated using PCR as well as the TIDE algorithm43 (Supplementary Fig. 5a, b). As compared to non-targeting sgRNA controls, both approaches resulted in the upregulation of GRHL3 mRNA expression (Fig. 5g, h). These knockout cells also exhibited upregulation of differentiation marker genes that are downstream of GRHL3 (Fig. 5i, j). Thus this genomic sequence of this GRHL3 IpA site, preferentially used in undifferentiated keratinocytes, plays a critical role in suppressing full-length GRHL3 gene expression and GRHL3-mediated differentiation.Primary human keratinocytes only have a limited life span under the cell culture conditions without feeder cells. To further confirm the role of IpA in influencing GRHL3 expression, we expanded our scope to immortalized cell lines which allow isolation and expansion of single clones with CRISPR editing. Leveraging RNA-seq data of cell lines generated by the ENCODE project44, we found that the GRHL3 IpA site is also used in HCT116 cells (Supplementary Fig. 5c). Similar to keratinocytes, these two CRISPR KO strategies upregulated of GRHL3 in bulk HCT116 cells (Supplementary Fig. 5d). We subsequently expanded and characterized 60 clones of HCT116 cells derived from the “sg2 + sg3” strategy, which allowed rapid PCR to screen deletion in one or both alleles. In total, 9 clones showed deletion in one of the two alleles (HET), and 1 clone showed deletion in both alleles (KO; Supplementary Fig. 5e). The HETs only displayed mild upregulation of GRHL3, while the KO showed drastic upregulation of GRHL3 at nearly 20 fold (Supplementary Fig. 5f, g). Sanger sequencing confirmed the expected deletion of ~1 kb containing the IpA site, created by the combination of sg2 and sg3. We also noticed minor differences (up to 50 bp at each end) among these individual clones, even between the two alleles within the single KO clone (Supplementary Fig. 5h). Taken together, these data demonstrate that the usage of this GRHL3 IpA site suppresses full-length GRHL3 mRNA expression.HNRNPA3 cooperates with CPSF to promote GRHL3 IpACPSF downregulation alone could only partially explain the fold change and selectivity of the differentially used IpA sites during keratinocyte differentiation. In the example of GRHL3 IpA, the usage decreases more than 100-fold in differentiation, based on quantification by qRT-PCR. CPSF knockdown alone led to ~3–4-fold reduction of GRHL3 IpA, using the same quantification method. To identify other molecular mechanisms influencing IpA usage, synergistically with CPSF, a targeted screen was designed to identify potential cofactors enhancing GRHL3 IpA. As illustrated in Fig. 6a, this strategy involved double knockdown of CPSF in combination with a candidate cofactor, to determine if double knockdown reduces GRHL3 IpA usage more than CPSF knockdown alone.Fig. 6HNRNPA3 cooperates with CPSF to promote GRHL3 IpA.a Illustration showing the genetic screen for identifying factors that enhance GRHL3 IpA. b Heat map showing the knockdown efficiency of the two shRNAs targeting specific candidates that include potential CPSF1-interacting proteins as well as CSTF2. c Heat maps showing the relative fold change of GRHL3 IpA usage comparing double knockdown vs CPSF single knockdown. d qRT-PCR showing the fold change of GRHL3 IpA usage, comparing HNRNPA3-CPSF double knockdown, single knockdowns and control. Dots represent data points in technical replicates. e, f Co-immunoprecipitation showing CPSF1 associates with HNRNPA3 in undifferentiated condition, but not in differentiated condition. This association is not disrupted by RNase treatment. Quantification of immunoprecipitation efficiency is indicated below. Source data are provided as a Source Data file.Putative CPSF-interacting proteins were included as candidates in this genetic screen, based on our pilot mass spectrometry experiment comparing CPSF1 immunoprecipitation between undifferentiated and differentiated keratinocytes (Supplementary Fig. 6a, b). This experiment identified several proteins related to the function of RNA binding, associating with CPSF1 in the undifferentiated but not in the differentiated condition. In addition, we also looked into complexes that cooperate with CPSF in cleavage and polyadenylation, such as CSTF, CFI and CFII. We prioritized CSTF2 as a target in this screen, as CSTF2 was the most downregulated subunit of these complexes, according to our RNA-seq data of keratinocyte differentiation.Two shRNAs targeting each candidate gene were validated with their knockdown efficiency (Fig. 6b), and were introduced to keratinocytes in combination with CPSF siRNA. HNRNPA3 showed the highest reduction of GRHL3 IpA usage in double knockdown versus single knockdown (Fig. 6c, d). Using co-immunoprecipitation, we confirmed that HNRNPA3 associated with CPSF1 only in the undifferentiated condition (Fig. 6e). HNRNPA3 is expressed in both undifferentiated and differentiated keratinocytes, although the protein level is slightly reduced in differentiation (Supplementary Fig. 6c). Thus, the reduced interaction between CPSF1 and HNRNPA3 in differentiation could be a result of the reduced expression of both proteins. In undifferentiated keratinocytes, the association between CPSF1 and HNRNPA3 was not disrupted by RNase (Fig. 6f), suggesting that HNRNPA3 and CPSF1 do not require RNA to bridge their interaction. The usage of other differential IpA sites, such as the sites associated with ALOX15B and CRBN (Supplementary Fig. 6d, e), were not drastically enhanced by HNRNPA3 knockdown in the context of CPSF RNAi. These data suggest that the differentially used IpA sites in keratinocyte differentiation are regulated by diverse mechanisms, and HNRNPA3 selectively influences a subset of IpA sites such as GRHL3 IpA.HNRNPA3 suppresses keratinocyte differentiation and influences GRHL3 splicingThe cooperation between HNRNPA3 and CPSF1 in controlling GRHL3 IpA suggests that they could synergistically suppress differentiation in epidermal progenitors. RNA-seq data comparing HNRNPA3 knockdown versus control identified a total of 1490 differentially expressed genes (fold change >=2, p < 0.05, Supplementary Data 4 and Fig. 7a, b). The upregulated genes are highly enriched with GO terms such as “epidermal development” and “keratinocyte differentiation” (Fig. 7c). A total of 306 genes are affected by the downregulation of HNRNPA3 and CPSF1 (Fig. 7d). These overlapping genes include GRHL3 as well as GRHL3 target genes, such as SPRR1b and SBSN. Double knockdown of GRHL3 and CPSF drastically elevated the expression of these differentiation genes, as compared to single knockdowns (Fig. 7e–g). Thus, HNRNPA3 cooperates with CPSF to suppress a subset of differentiation genes, although HNRNPA3 also has CPSF-independent roles in gene regulation.Fig. 7HNRNPA3 suppresses differentiation and influences GRHL3 splicing.a Western blotting validating the knockdown efficiency of two shRNAs targeting HNRNPA3, at the protein level. Quantification is indicated below (n = 2). b Heat map showing the relative expression of 1490 genes that are differentially expressed between HNRNPA3 knockdown versus control. c Bar graph showing the top GO terms associated with these 1490 genes altered by HNRNPA3 knockdown (p value: modified Fisher’s exact test). d Venn diagram comparing HNRNPA3 target genes versus CPSF core target genes. e–g qRT-PCR quantification of GRHL3, SPRR1B and SBSN in double knockdown and single knockdown of HNRNPA3 and CPSF1. Dots represent data points in technical replicates. h RNA-seq and 3′READS + tracks showing enriched RNA-seq reads before the GRHL3 IpA site. These RNA-seq reads are likely to be enriched in a “hidden exon”. i Illustration showing the qRT-PCR strategies to quantify “inclusion” and “skipping” of the hidden exon, as well as the ratio of pre-mRNA versus total RNA. Arrow heads indicate the location of the primers. j Quantification of the relative ratios of “inclusion” versus “skipping” of the “hidden exon” comparing HNRNPA3 knockdown versus control. HNRNPA3 knockdown promoted the splicing between exon1 and exon2 and skipped the “hidden exon”. Dots represent data points in technical replicates. k, l Ratio of qRT-PCR amplification between Exon1-intron1 junction versus exon1, in HNRNPA3 knockdown or CPSF1 knockdown. Dots represent data points in technical replicates. Source data are provided as a Source Data file.We next explored the nature of this HNRNPA3-CPSF collaboration in controlling GRHL3 IpA. HNRNPA3 is part of the hnRNP A/B family. The proteins in this family are characterized by two tandem RNA recognition motifs (RRMs) in the N-terminal region45,46. Systematic analysis of their Drosophila homologs indicates that these hnRNP A/B proteins control overlapping but diverse targets in pre-mRNA processing47. The best-characterized human protein in this family is HNRNPA1, which binds to the consensus sequence UAGGGA/U and directly antagonizes splicing48,49. HNRNPA3 shares 94% similarity in the tandem RRMs with HNRNPA1, but differs in the C-terminal region. Although the function of HNRNPA3 is currently under-characterized, the high similarity of RRMs suggests that HNRNPA3 and HNRNPA1 could bind to a very similar RNA consensus sequence to influence pre-mRNA processing.RNA-seq data in undifferentiated keratinocytes, generated by us as well as by previous studies50, identified a potential “hidden exon” with enriched RNA-seq reads immediately before the GRHL3 IpA site (Fig. 7h), suggesting that GRHL3 IpA may involve the inclusion of this “hidden exon” through alternative splicing. To quantify potential alternative splicing events including or skipping this “hidden exon”, qPCR primers were designed to amplify the junction between exon 1-“hidden exon” versus exon1-exon2 (Fig. 7i). HNRNPA3 knockdown drastically increased the ratio of skipping versus inclusion of the “hidden exon” (Fig. 7j), suggesting that HNRNPA3 antagonizes the splicing between exon1-exon2 and promotes the connection between exon1-“hidden exon”. To further clarify the role of HNRNPA3 in splicing, additional primer pairs were designed to quantify the ratio between exon1-intron1 junction versus exon1. This ratio was strongly reduced in HNRNPA3 knockdown, but not in CPSF1 knockdown (Fig. 7k, l), suggesting that HNRNPA3 stabilizes the exon1-intron1 junction and suppresses splicing of intron 1.Taken together, these findings suggest a working model where CPSF and RNA-binding proteins cooperatively influence gene expression through controlling the usage of specific IpA sites. In the context of epidermal tissue progenitors, MYC functions upstream of CPSF to sustain its high level of expression. The physical interaction between CPSF and HNRNPA3 synergistically promotes GRHL3 IpA, through suppressing the junction between the adjacent exons and promoting cleavage of nascent RNA, to suppress the premature expression of terminal differentiation genes in epidermal progenitor maintenance (Fig. 8).Fig. 8Working model.Graphic illustration showing the working model of CPSF-HNRNPA3 cooperation in epidermal progenitor maintenance. MYC functions upstream to sustain CPSF expression in undifferentiated keratinocytes. In the first intron of GRHL3, CPSF and HNRNPA3 bind to nascent RNA to cooperatively suppress splicing and promote intronic polyadenylation, which suppresses full-length GRHL3 mRNA expression and inhibit premature differentiation. UD undifferentiated keratinocytes, DF differentiated keratinocytes.DiscussionIt remains incompletely understood how genomic information is being selectively accessed, to fine-tune spatiotemporal gene regulation in development and in somatic tissue homeostasis. In this study, we found that human keratinocyte differentiation involves differential usage of intronic polyadenylation sites. In particular, we characterized an IpA site located within the first intron of GRHL3, a key transcriptional activator of epidermal differentiation. Usage of GRHL3 IpA in epidermal progenitors contributes to suppressing the expression from GRHL3 as well as the terminal differentiation genes downstream to GRHL3. Both CPSF and its interacting protein HNRNPA3 are essential for promoting the usage of this IpA site.HNRNPA3 is one of the several RNA-binding proteins which were identified to associate with CPSF in undifferentiated, but not differentiated keratinocytes. In the genetic screen, HNRNPA3 stood out to have the strongest influence to enhance CPSF’s ability to suppress GRHL3 IpA. Although the molecular function of HNRNPA3 is not fully understood, its high homology to HNRNPA1 suggests its role in splicing. In the case of HNRNPA1, multiple copies of this protein can bind and spread on nascent RNA to suppress splicing by altering RNA secondary structure and displacing splicing-promoting proteins such as the serine/arginine (SR)-rich-family proteins51,52. Our qPCR quantification identified that the HNRNPA3 knockdown promoted exon1-exon2 junction and destabilized exon1-intron1 junction. In addition, motif search using RBPmap53 identified three HNRNPA1 motif sites (p < 0.01, Z-score > 2.5) within 2 kb upstream of the GRHL3 IpA site, suggesting that HNRNPA3 could bind to GRHL3 pre-mRNA and influence IpA usage through splicing. These data suggest that HNRNPA3 can involve at least two mechanisms to influence GRHL3 splicing and polyadenylation: HNRNPA3 can bind directly to pre-mRNA within the first intron to suppress the splicing between exon1 and exon2; HNRNPA3 can also bind to CPSF to stabilize CPSF binding to the AAUAAA motif, facilitating the full assembly of cleavage and polyadenylation machinery to promote IpA. Our “sg2 + sg3” CRISPR KO strategy, designed to remove the AAUAAA motif, did not affect these three putative HNRNPA3 binding sites. Future studies dissecting the contributions from HNRNPA3 in regulating this GRHL3 IpA, using additional CRISPR strategies to KO these three putative HNRNPA3 binding sites, can further elucidate the contribution from both HNRNPA3 and CPSF to regulating GRHL3 expression.In addition to HNRNPA3, a couple of other RNA binding proteins such as FUS and ELAVL1, also enhanced GRHL3 IpA usage in our screen, although to a lesser extent as compared to HNRNPA3. FUS had been previously demonstrated to bind to nascent RNA near the alternative PolyA sites54. ELAVL1 is also implicated in alternative splicing. For example, ELAVL1 loss was known to promote exon 11 skipping of the translation initiation factor Eif4enif155. ELAVL1’s association with TRA2-beta was demonstrated to promote the inclusion of exon256. Therefore the selectivity of GRHL3 IpA is likely to involve the cooperation from multiple RNA-binding proteins, although HNRNPA3 had the strongest effect from our target screen. As HNRNPA3 did not appear to enhance CPSF1’s regulation of ALOX15B IpA and CRBN IpA, the usage of different IpA sites may involve diverse regulatory mechanisms.Previous studies in different systems demonstrate that IpA can lead to truncated proteins21,23,24. In the case of GRHL3, this specific IpA is located within the first intron. If translated, this isoform would only retain 6 amino acids of the original GRHL3 protein. For the other IpA sites, their relative location varies among different genes. For example, the ALOX15B IpA is located in the 5th intron. The mRNA generated through ALOX15B IpA could be translated into a protein that misses ~80% of the lipoxygenase domain. A key technical barrier at present, for characterizing the roles of additional IpA sites, is the lack of high-quality antibodies that are specifically raised to target the N-terminal regions of these proteins.Among the 2739 IpA sites that we identified in keratinocytes, 17% of these sites overlap with the IpA sites cataloged in the immune system24. However, when we compared the host genes associated with these IpA sites in these two systems, the overlap increased to 45%. For example, in keratinocytes the AGO3 gene is associated with 3 IpA sites, and in the immune cells AGO3 is associated with 4 IpA sites. Only 2 of these IpA sites overlap. Thus the same gene can associate with shared and distinct IpA sites in different cell types.In summary, our work provides genome-wide profiling of polyadenylation in human keratinocyte differentiation, and sheds light into the regulatory mechanisms underlying the usage of specific IpA sites. This work also reveals the essential roles of CPSF and HNRNPA3 in regulating keratinocyte differentiation, highlighting the significance of pre-mRNA processing in influencing somatic tissue homeostasis.MethodsCell culturePrimary human keratinocytes were isolated from the surgically discarded fresh foreskin (obtained from Northwestern Skin Biology & Diseases Resource-Based Center). Tissue was collected under a protocol approved by the Northwestern University Institutional Review Board (IRB # STU00009443). Patient consent for neonatal foreskin tissue was not required as this tissue is de-identified and considered discarded material per IRB policy. Keratinocytes from at least three de-identified donors were mixed and cultured in 50% complete Keratinocyte-SFM (Life Technologies #17005-142) and 50% Medium 154 (Life Technologies #M-154-500). Keratinocyte differentiation was induced by adding 1.2 mM CaCl2 in full confluency for four days. HEK293T and phoenix cells were cultured in DMEM (Gibco) containing 10% fetal bovine serum (HyClone). HCT116 cells were cultured in McCoy’s 5A (Modified) Medium (Gibco) containing 10% fetal bovine serum.Plasmid constructionFor lentiviral CRISPRi, the pLEX_Cas9 plasmid (Addgene #117987) was modified by replacing its Cas9 sequence with KRAB-dCas9 from pHR-SFFV-KRAB-dCas9-P2A-mCherry (Addgene #60954) to make pLEX-KRAB-dCas9-BSD. Then its CMV enhancer and promoter were replaced by the UCOE (Ubiquitous Chromatin Opening Element)-SFFV promoter from pMH0001 (Addgene #85969) to generate pLEX-UCOE-SFFV-KRAB-dCas9-BSD.For retroviral CRISPRi, dCas-KRAB-MeCP2-BSD was PCR amplified from the pB-CAGGS-dCas9-KRAB-MeCP2 plasmid (Addgene #110824), and then cloned into the retroviral vector LZRS linearized by BamHI and NotI.For retroviral CRISPR, Cas9-BSD was PCR amplified from the pLEX_Cas9 plasmid (Addgene #117987), and then cloned into the retroviral vector LZRS linearized by BamHI and NotI.For sgRNA cloning, we linearized the pLentiGuide plasmid (Addgene #117986) by BsmBI, and ligated sgRNA sequence into it to make pLentiGuide-sg-mCherry.For GRHL3 and ESPN tandem sgRNA cloning, a double-stranded DNA block containing the sgRNA2-tRNA-sgRNA3 was synthesized from GENEWIZ, Inc., and then cloned into the pLentiGuide plasmid (Addgene #117986) linearized by BsmBI.For shRNA cloning, we linearized pLKO.1 puro plasmid (Addgene #8453) by AgeI and EcoRI, and ligated annealed shRNA oligos into it to make pLKO.1-sh-puro. The oligo sequences for generating shRNA constructs are included in Supplementary Data 5.Gene transferTransfection of HEK293T and phoenix cells was performed using Turbofect (Thermo Fisher) following the manufacturer’s instruction. High-titer virus was collected at 48 and 72 h post-transfection, added to wells of keratinocytes and centrifuged at 1250 rpm for 1 h at 32 °C. Keratinocytes were then selected using blasticidin (5 μg/mL) or puromycin (2 μg/mL) after infection for 48 h. HCT116 cells were virally infected and selected for blasticidin or puromycin resistance using the same methods as keratinocytes.siRNA knockdownON-TARGETplus siRNA- SMARTpool targeting CPSF1 (L-020395-00) and GRHL3 (L-014017-02) were ordered from Dharmacon. 4D-Nucleofector (Lonza) was used for nucleofection following the manufacturer’s instruction.Genomic DNA knockout analysisIn order to check CRISPR knockout efficiency of GRHL3 IpA in keratinocytes and HCT116 cells, total genomic DNA was extracted using Quick-DNA Miniprep Plus Kit (Zymo). For sg-1 disruption, the corresponding genomic region harboring the knockout site was amplified using Phusion High-Fidelity DNA Polymerase (Thermo Fisher), and gel extracted for Sanger sequencing (ACGT, INC.). Sequencing results for the mixed pool were analyzed using TIDE43 webtool (https://tide.deskgen.com/). For sg2-sg3 deletion, the corresponding genomic region harboring the knockout site was amplified using GoTaq® DNA Polymerase (Promega), resolved on an agarose gel, and quantified using Image J software (NIH). For analysis of the two alleles in the single-clone KO HCT116 cells, the PCR product was cloned into the pLZRS vector and individual clones were analyzed using Sanger sequencing.Protein expression and tissue analysisFor immunoblot analysis, 20–50 μg of cell lysate was loaded per lane for SDS-PAGE and transferred to PVDF membranes. The blots were scanned and quantified using the Li-COR Odyssey Clx imaging system (LI-COR). For immunofluorescence staining, tissue sections (7 µm thick) were fixed using either 50% acetone and 50% methanol, or 4% formaldehyde. Primary antibodies were incubated at 4 °C overnight and secondary antibodies were incubated at room temperature for 1 h. Images were captured by an EVOS FL Auto 2 fluorescent microscopy (Thermo Fisher) and processed by Image J software (NIH).Colony formation assayMouse fibroblast 3T3 cells were treated with 15 μg/mL mitomycin C (Sigma) in serum-free DMEM for 2 h, then trypsinized and plated at ~8 × 105 cells per well in a 6-well plate overnight. The media was changed to FAD media 1 h before seeding 1000 keratinocytes onto the feeder layer. The medium was changed every 2 days for eight days. Then the wells were washed with PBS to remove the 3T3 cells, and remaining keratinocytes were fixed in 1:1 acetone/methanol for 5 min. The plates were allowed to air dry for 5 min, and then colonies were stained with crystal violet.Wound healing assayIn 24-well plates, 0.2 × 106 CTRL or CPSF1 knockdown keratinocytes were seeded and allowed to grow to full confluence the next day. Keratinocytes were treated with 10 μg/mL mitomycin C for 2 h prior to wounding. Wounds were made using 100-μl filter pipette tip (Thermo Fisher) and the healing process was monitored under a microscope. For quantification, the surface area of the scratch at different time points was measured using the Image J software (NIH). For comparing IpA usage between migrating keratinocytes versus control, a grid of scratches were made to a confluent monolayer of keratinocytes as described previously9. RNA extraction and qPCR analysis were performed using scratched versus non-scratched plates.Organotypic human epidermis regenerationFor organotypic epidermal cultures, keratinocytes were nucleofected with CPSF1 siRNA or non-targeting control siRNA, trypsinized 4 days later and counted. The dermis was prepared from donated cadaver samples from the New York Firefighters Skin Bank. Usage of human dermis from de-identified donors for organotypic epidermal regeneration has been approved by Northwestern IRB. In all, 1.0 × 106 CTRL or CPSF1 knockdown keratinocytes were seeded onto the top of each piece of the devitalized dermis. The organotypic cultures were raised to the air/liquid interface to induce stratification and differentiation for 6 days. The regenerated tissue was then embedded OCT before cryosectioning and imaging. The TUNEL assay of CTRL and CPSF1 siRNA tissues was performed by In Situ Cell Death Detection Kit, TMR red (Roche) following the manufacturer’s instruction, and cells with positive signals were counted using Image J software (NIH). In the progenitor competition assay, keratinocytes were labeled by infection retrovirus expressing H2B-GFP or H2B-mCherry before the nucleofection of CTRL or CPSF1 siRNA. For quantification, the GFP or mCherry expressing keratinocytes in the regenerated epidermis were counted using the Image J software (NIH). For each image, quantification was performed for the entire thickness of epidermal tissue as well as the top and bottom halves of the tissue divided evenly using Image J.Nile Red stainingNile Red staining of tissue sections was performed similar to the method previously described57, with our experimental details listed below. The stock solution was prepared by dissolving Nile Red (Sigma 72485) first in acetone at 500 μg/mL then diluting to a working solution of 2.5ug/mL. Skin tissue sections were fixed in 10% formalin for 5 min, rinsed briefly in PBS, then incubated in the working solution for 10 min at room temperature. The slides were subsequently washed with PBS and stained by NucBlue Fixed Cell ReadyProbes Reagent (DAPI, Invitrogen) for 5 min, and were mounted with anti-Fade Fluorescence Mounting Medium (Abcam). For quantification, the thickness of Nile-Red-positive regions in each image was measured at 3 regions using the Image J software (NIH). The average thickness from 3 measures of each image were calculated, and a total of 26 images per condition were included for statistical analysis (T-test) using Prism.CPSF1 complex purification and protein identification using mass spectrometryKeratinocytes were trypsinized, washed in PBS, and resuspended in 200 μL hypotonic buffer (10 mM HEPES at pH 7.4, 1.5 mM MgCl2, 10 mM KCl, 1× protease inhibitor cocktail (Roche)) per million cells. Cells were lysed by adding an equal volume of hypotonic buffer with 0.4% NP-40 for 2 min. Nuclei were pelleted by centrifugation at 4000 rpm and lysed in ten cell pellet volumes of nucleus lysis buffer (50 mM Tris at pH 8.0, 0.05% igepal, 10% glycerol, 2 mM MgCl2, 250 mM NaCl, protease inhibitor cocktail (Roche)). Nuclei were sheared with a 27.5-gauge needle, and lysis proceeded for 30 min. Insoluble material was removed by centrifugation at 13,000 rpm for 10 min, and nuclear supernatant was used for purification. Dynabeads™ Protein G (Thermo Fisher) were conjugated with CPSF1 mouse monoclonal antibody (Santa Cruz Biotechnology, Inc.) or mouse IgG as control, added to nuclear supernatant for 4 °C incubation overnight, and washed five times with nucleus lysis buffer. Proteins were boiled off from beads and separated on SDS-PAGE for immunoblotting or mass spectrometry identification.For protein identification, immunoprecipitations were first separated on SDS-PAGE and stained with colloidal blue (Life Technologies). Gel slices (0.5 cm) were submitted to the Northwestern Proteomics Facility for mass spectrometry analyses.RNA-seqRNA was extracted using Quick-RNA™ MiniPrep (Zymo Research) with DNase I treatment. RNA-seq libraries were prepared using NEBNext Ultra™ Directional RNA Library Prep Kit for Illumina (New England BioLabs) with ribosomal-RNA depletion (New England BioLabs) or NEBNext® Poly(A) mRNA Magnetic Isolation Module (New England BioLabs). Libraries were sequenced as single-end 50-base-pair (bp) reads using the Illumina HiSeq 4000 platform by Northwestern University NUSeq Core facility.3′READS +library construction3′READS+ experiments were performed as described34 with minor modifications to enable multiplexing on the Illumina HiSeq 4000 platform, with our experimental details listed below. Poly(A)+ RNA from keratinocytes was captured from 15 μg total RNA using NEBNext® Poly(A) mRNA Magnetic Isolation Module (New England BioLabs). Fragmentation was performed on the beads using ShortCut® RNase III (New England BioLabs). The fragmented poly(A)+ RNA was ligated to 5′ adapter (5′ -CAGACGUGUGCUCUUCCGAUCUNNNN) on the beads with T4 RNA ligase I (New England BioLabs). The ligation products were captured and poly(A)-tail-trimmed by RNase H (New England BioLabs) on biotin-T15-(+TT)5 (Exiqon) bound to Dynabeads MyOne Streptavidin C1 (Thermo Fisher). The RNA fragments were then ligated to 3′ adapter (5′ -rApp/NNNN AGATCGGAAGAGCGTCGTGTAG/3ddC) with T4 RNA ligase 2, truncated KQ (New England BioLabs). The ligation products were then reverse transcribed using M-MLV reverse transcriptase (Promega), followed by PCR using Phusion high-fidelity DNA polymerase (Thermo Fisher) and NEBNext® Multiplex Oligos for Illumina® (New England BioLabs) for 15 cycles. PCR products were size-selected with AMPure XP beads (Beckman Coulter), and sequenced by Illumina HiSeq 4000 platform with 1x50bp by the NUSeq Core facility at Northwestern University.qRT-PCR expression analysisTotal RNA was extracted using Quick-RNA™ MiniPrep (Zymo Research), and reverse transcribed using the SuperScript VILO cDNA synthesis kit (Invitrogen). For quantification of the proximal/distal PolyA-site usage ratio, an additional step of poly(A)+ RNA isolation was performed using NEBNext® Poly(A) mRNA Magnetic Isolation Module (New England BioLabs). qPCR was performed using the SYBR Green Master Mix (Thermo Fisher) or EvaGreen Master Mix (Bullseye). Samples were run in duplicates and normalized to levels of 18S ribosomal RNA for each reaction. Statistical analysis such as one-way ANOVA and Two-tailed Student’s unpaired t-test was calculated using GraphPad Prism7. Bar graphs and their associated error bars are represented as mean ± standard deviation. Primer sequences used for qPCR are listed in Table S5.RNA-seq analysisRNA-seq libraries were aligned to the hg19 genome using Hisat258. Browser tracks were generated using the UCSC genome browser. Htseq59-count was used to generate counts tables at all genes. Differential expression analysis was done using DESeq260. Genome browser tracks of RNA-seq were normalized by sequencing depth.3′READS+ analysis3′READS+ data analysis was performed based on the protocol described in Zheng et al.34. 5′ adapters were removed using CutAdapt, followed by removal of 3′ adapter consisting of four random nucleotides. T’s corresponding to PolyA tails were removed from reads and saved. Up to one non-T base was allowed in T tails. Reads < 22 nucleotides were filtered out, and the remaining reads were aligned using bowtie2 in end-to-end mode. Aligned reads with MAPQ score less than 10 were filtered out. Sequence through 20 bp downstream of PolyA site was obtained. If the T tail of a read had a non-T base, and the sequence from the start of the T tail to the non-T base aligned to the genome, then that portion of the T tail was removed and considered part of the aligned sequence. Reads in which the T tail had at least two unmappable T’s were considered Poly-A Site supporting (PASS) reads. Poly-A sites from PASS reads were merged by 24 nucleotides. Sites corresponding to hg19 blacklist version 2 (Boyle lab), retrotransposons (ucscRetroInfo5 table), microRNA, and snoRNA (ucsc hg19 refseq table) were filtered out. We then filtered out poly-A sites that did not overlap with a gene according to the GENCODE version 17 annotation from UCSC. A table of counts was built at the remaining sites for each library.To identify differential usage of Poly-A sites, the sum of counts at each gene was calculated. The relative “Fraction of PolyA site Usage” (FPU) was then calculated as counts at each site divided by the total counts at the gene associated with that site. EdgeR was used to calculate differential usage and p-value based on the FPU.Several filters were applied to 3′READS+ PolyA site analyses to assure we focused on stably expressed genes and functional PolyA sites. PolyA sites met the following criteria (1) gene expressed in keratinocytes (at least 5 counts in either UD or DF RNA-seq libraries) (2) the sum of the counts from the two biological replicates ≥ 10, and (3) FPU from at least one library ≥10%.The 18173 remaining sites were annotated using GENCODE version 17. 1118 sites overlapped with multiple genes and were removed from further analysis. We designated Poly-A sites as Introns, UTRs, or Exons using bed files for those regions from GENCODE version 17. We prioritized introns to ensure that previously uncharacterized intronic sites would not be incorrectly annotated as UTRs. 4885 sites were designated as introns. From the remaining sites, 11734 were identified as UTRs. The final 436 sites were annotated as exons.To correct for potentially incorrect annotations of UTR’s as introns, we reasoned that sites associated with genes that only appeared once should be designated as UTRs. We identified 4525 of these sites. While 3566 of these were already correctly annotated as UTR’s, we also adjusted the annotation of the remaining 959 sites (54 exons, 905 introns) to UTR. Given that GENCODE tables used for annotation included regions previously filtered out, one final step was required to refine our list. From the annotated list of 17,055 sites, 2429 were ambiguous as they are located in overlapping genes. Even if one of these genes did not meet the filters previously stated, it was impossible to distinguish whether the reads corresponded to the filtered out gene or not. Removal of these sites left us with a robust final list of 14625 poly-A sites.To determine the 428 differentially used Intronic PolyA sites in Fig. 1a, we used a fold-change cut off of 2. Sites were validated using RNA-seq data. We combined three replicates of undifferentiated (UD) and three replicates of differentiated (DF) keratinocytes and built a table of counts from RNA-seq at PolyA site ±100 bp and calculated log2FC from UD to DF conditions. Sites with fold change >1.5 in the same direction in both 3′READS+ libraries and RNA-seq libraries and with more counts upstream than downstream the PolyA site, were considered alternatively used IpA sites. To identify which of these sites were regulated by CPSF, we built a table of counts from CPSF1-knockdown and control 3′READS+ data at PolyA sites and calculated FC. Sites with FC 1.5 that changed in the same direction as in UD DF 3′READS+ data were considered regulated by CPSF. The height of 3′READS+ genome browser tracks was normalized based on sequencing depth.For motif analysis of the PolyA sites, DNA sequences were extracted with an extension of 200 bp both upstream and downstream. MEME61,62 motif search was performed using the RNA (DNA-encoded) motif database “Ray2013 Homo sapiens (DNA-encoded)”, using a background model of 1-order that adjusts for dimer biases.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting SummaryDescription of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2Supplementary Data 3Supplementary Data 4Supplementary Data 5 | nature communications | [
"Article"
] | [
"Developmental biology",
"Differentiation",
"Self-renewal",
"Molecular biology"
] | IntroductionSelf-renewing somatic tissue epithelium undergoes turnover wear tear terminal differentiation essential for specialized tissue function genes suppressed in progenitors sustain regenerative capacity1,2 molecular mechanisms regulation differentiation genes incompletely understood.Human skin epidermis stratified epithelium accessible research platform for gene regulatory mechanisms tissue differentiation Primary epidermal cells (keratinocytes) isolated expanded in culture undifferentiation differentiation conditions integration genomic proteomic genetic tools research epidermal differentiation involves upregulation transcription factors chromatin targets activate barrier function5,6 GRHL3 late-differentiation promotes protein crosslinking cornified envelope GRHL3 repressed in epidermal progenitors regulatory mechanisms repress GRHL3 identified include PRMT1 promoter EXOSC9 degrading mRNA post-transcriptionally2 epigenetic post-transcriptional regulators non-coding RNAs epidermal differentiation1,6 findings highlight coexistence multiple gene regulatory mechanisms steps gene expression abundance gene products ~30% human genome introns19 introns’ influence on gene expression appreciatedpolyadenylation sites identified in introns 3′UTR terminates transcription prematurely mechanism under-studied explored in developmental physiological processes muscle regeneration after injury IpA event in growth factor generates short-isoform decoy full gene function muscle late spermatogenesis sterol transcription factor 2 switches from to short isoform IpA germ-cell gene profiling identified recurrent IpA sites inactivate tumor suppressor genes in diverse IpA events in immune system IpA sites in biological processes unclear.Polyadenylation requires cleavage of nascent RNAs multiple protein complexes cleavage stimulation factor) factors CF I CFII bind to upstream downstream polyadenylation catalyzed by polyadenylate polymerase central player Cleavage and Polyadenylation Specificity Factor (CPSF) CPSF binds to AAUAAA consensus sequence through CPSF4 WDR33 subunits CPSF3 cleaves nascent RNA components CPSF2 FIP1L1 by subunit CPSF127–30expression levels CPSF vary cell Elevated CPSF expression observed in somatic cell reprogramming CPSF levels influence gene expression tissue functions control expression open questions study IpA sites used by transcription machinery during keratinocyte differentiation reduction CPSF expression enriched expression CPSF progenitor state downstream to MYC essential suppressing premature differentiation CPSF expression influences differential IpA sites keratinocyte differentiation CPSF-dependent sites include first intron GRHL3 CRISPR experiments usage GRHL3 IpA site suppresses full-length GRHL3 mRNA expression protein complex purification genetic screen identified HNRNPA3 key CPSF-interacting RNA-binding protein enhances GRHL3 IpA usage findings support model CPSF cooperates with RNA proteins IpA usage transcriptome keratinocytes suppressing premature differentiation differentiation involves altered IpA usage downregulation 3′READS+ technique transcriptome-wide PolyA sites in undifferentiated differentiated human keratinocytes “Fraction PolyA site Usage calculated divided byquantification method FPU normalized each sequencing library fractions genes method circumvents normalizing PolyA site usage across libraries gene expression different 14625 PolyA sites associate with 7990 genes in keratinocytes undifferentiated keratinocytes FPU correlated between replicates (Fig 1b FPU divergent undifferentiated differentiated keratinocytes suggesting genome-wide changes in PolyA site usage 2739 in intronic regions 17% overlaps with IpA sites immune system24 specificity IpA events distinct tissue types IpA sites overlap with PolyA_DB3 database35 differentially used IpA sites between undifferentiated differentiated keratinocytes integrated 3′READS+ data analysis with strand-specific RNA-seq data validation 428 differentially used IpA sites identified (fold change > = 2 ≥1.5 sites enriched with AAUAAA motif (E value: 1.9 × 10−165). include more downregulated (60%, 256/428) than upregulated sites (40%, 172/428) Examples include downregulated in ESPN (cytoskeleton upregulated in IQCKdynamic IpA usage during cell switch progenitor to terminal differentiation. 1Keratinocyte differentiation involves IpA alterations CPSF downregulation Illustration analyzing IpA sites in undifferentiated differentiated keratinocytes plots correlation 3′READS+ data replicates undifferentiated differentiated conditions map change 428 differentially used IpA sites during differentiation Genome browser tracks differential IpA usage in EPSN IQCK highlight 3′UTR Green UD-enriched Red DF-enriched Illustration composition function CPSF complex Heat map mRNA levels CPSF subunits POLR2A between undifferentiated differentiated keratinocytes qRT-PCR mRNA expression CPSF genes Dots data Immunoblots expression CPSF subunits protein level keratinocytes β-tubulin loading control Average fold change SD from 3 replicates indicated panel Source data file 11,506 sites identified in 3′UTR+ analysis 4677 correspond to genes single PolyA sites 6829 sites associated with 2727 genes 2 or more PolyA sites3′UTR shortening keratinocyte differentiation calculated distal PolyA site usage within 3′UTR 2727 genes total counts distal site divided 3′UTR identified 211 genes increased distal PolyA usage 457 decreased usage (fold change > 1.5 data indicate lengthening shortening 3′UTR occur during keratinocyte differentiation shortening more genes trend similar spermatogenesis shortening polyadenylation involves multiple complexes CPSF CstF CF I CF II PAP expression PAP constant between undifferentiated differentiated keratinocytes subunits encoding other complexes dynamic downregulation CPSF qRT-PCR western blotting downregulation CPSF complex subunits differentiation confirmed at mRNA protein levels CPSF antibodies CPSF2 antibody worked immunostaining human skin sections stronger signals basal progenitor layer epidermis reduced differentiated layers immunostaining RNA polymerase II) showed strong signals undifferentiated differentiated layers epidermal tissue suggest CPSF downregulation occurs during keratinocyte differentiationSuppression CPSF undifferentiated keratinocytes induces differential expression CPSF CPSF level keratinocyte function Leveraging TARGETplus” siRNA 4 regions CPSF1 performed nucleofection keratinocytes undifferentiated non-targeting control pool four siRNAs nucleofected control efficacy CPSF1 confirmed mRNA protein levels (Fig. 2a protein levels other CPSF subunits reduced caused protein degradation loss CPSF1 scaffold mRNA levels minimally affected 2CPSF Downregulation impairs self induces differentiation qRT-PCR CPSF1 knockdown efficiency Epidermal tissue regenerated 50% CTRLi H2B-GFP 50% CTRLi CPSFi H2B-mCherry Scale bar 100 μM 25 red:green ratio tissue CTRLi versus/CPSFi 25 images p = 6.55 × 10−18 t-test 2 tailed error bars mean values Diagram design CPSF1 CRISPRi locations three sgRNAs labelled transcription start site qRT-PCR knockdown efficiency CPSF CRISPRi versus control Clonogenic assay keratinocytes CPSF1 CRISPRi-targeting control Colonies diameter > 1/16 inch quantified Dots data points replicates Venn diagram differentially expressed genes CPSF siRNA CRISPRi data sets overlap (Fisher’s test p = 1.1 × 10−308) Heat map change 739 differentially expressed genes 2 P < 0.05 CPSF1 RNAi CRISPRi Bar graph top GO term genes altered CPSF CRISPRi Fisher’s qRT-PCR mRNA levels differentiation marker genes CPSF1 knockdown controls Dots data points replicates Source data file regenerative capacity keratinocytes CPSF1 knockdown assessed progenitor competition assay epidermal tissue regenerated 50% H2B-GFP 50% H2B-mCherry H2B-GFP treated control siRNAs H2B-mCherry treated CPSF1 siRNA control regenerated epidermal sectioned ratio mCherry-labelled nuclei versus GFP-labelled nuclei quantified ImageJ37 CPSF1 knockdown reduced representation mCherry-labelled cells drastic bottom half CPSF1 knockdown impaired regenerative capacityprogenitor competition assay epidermal tissue regenerated keratinocytes CPSF1 RNAi hypoplastic lipid stain Nile Red increase tissue CPSF RNAi enhanced barrier function water loss No difference apoptosis TUNEL assay CPSF RNAi impaired keratinocyte migration scratch wound healing assay CPSF functions keratinocytes homeostasis wound healing used CRISPRi38 suppress CPSF1 expression enhanced CRISPR repressor KRAB MeCP2 nuclease-inactive dCas9 Three independent sgRNAs CPSF1 expressed dCas9-KRAB-MeCP2 keratinocytes downregulated CPSF1 expression mRNA protein levels. 2d clonogenicity assay colony-forming ability keratinocytes CPSF CRISPRi non-targeting controls colony numbers reduced with CPSF1 loss 2f intact function CPSF essential for epidermal progenitor self-renewal validation knockdown approaches RNAi CRISPRi CPSF target genes altered RNA-seq analysis identified 1113 genes RNAi 1584 genes CRISPRi < 0.05)739 genes approaches “CPSF core (Fig. 2h Supplementary Data 2) top terms upregulated targets include “Keratinization” “Keratinocyte downregulated genes include “DNA “Cell (Fig. similar to genes altered keratinocyte qRT-PCR validated differentiation genes upregulated CPSF suppression differentiation activator GRHL3 differentiation marker genes SPRR1B S100A8 S100A9 (Fig. 2k data role CPSF suppressing premature differentiation epidermal progenitor maintenance CPSF expression undifferentiated keratinocytes requires MYCAs downregulated differentiated keratinocytes searched mechanisms expression impairment proliferation induction differentiation downregulate CPSF expression Two strategies tested CPSF expression examined in keratinocytes PRMT1 knockdown essential sustaining proliferation suppressing differentiation genes knockdown affect CPSF expression Fig. expression compared between early- late-passage keratinocytes culture increased differentiation markers p16 S100A9 reduced cell cycle markers Ki67 AURKB Fig 3b). expression CPSF subunits not dramatically alteredimpaired proliferation induced differentiation not downregulate CPSF expression in human keratinocytes regulators maintaining CPSF1 expression undifferentiated keratinocytes searched for transcription factors regulatory regions CPSF1 gene ChIP-seq MYC ChIP-seq signal enriched at CPSF1 promoter keratinocytes breast cancer cell line MCF7 MYC downregulated during keratinocyte differentiation evaluated essential sustaining CPSF expression undifferentiated keratinocytes Two independent shRNAs targeting MYC validated qRT-PCR Western blotting Keratinocytes showed downregulation CPSF1 non-targeting control downregulated differentiation tested overexpression increase CPSF expression keratinocytes infected-MYC overexpression vector control no drastic differences mRNA level CPSF1 suggest MYC essential sustaining CPSF expression progenitor state not sufficient high expression differentiation state 3Sustained CPSF expression undifferentiated keratinocytes requires MYC Genome browser tracks binding MYC at CPSF1 promoter keratinocytes MCF7 cells Western blots downregulation MYC during keratinocyte differentiation MYC protein level fold changeWestern blotting knockdown efficiency shRNAs targeting MYC Quantification (n = 3) qRT-PCR quantification MYC CPSF1 knockdown control Dots data points replicates Source data.CPSF downregulation alters IpA keratinocyte differentiation involves altered IpA reduced CPSF levels tested CPSF knockdown IpA usage 3′READS+ 178 IpA sites altered CPSF knockdown. 4a 74% IpA sites reduced usage CPSF Genes CPSF-dependent sites include ALOX15B EPSN CRBN differentiation activator GRHL3 CPSF dependent independent IpA sites drastic differences expression 4CPSF Downregulation Accounts IpA Alterations Differentiation 178 428 IpA sites influenced CPSF1 knockdown Heat map changes 178 IpA sites DF versus CPSFi Genome browser tracks differential IpA usage DF CPSFi Locations qPCR primers IpA usage quantification labelled proximal distal UD-enriched qRT-PCR quantification relative usage IpAs calculated Proximal:Distal usage ratio undifferentiated vskeratinocytes CSPF1 siRNA CPSF CRISPRi MYC knockdown Dots represent data points replicates Source data file validate altered usage IpA sites developed qRT-PCR strategy two pairs qPCR primers for each gene first (proximal before IpA second between 3′ end relative usage IpA quantified as proximal: distal ratio usage IpA sites genes decreased keratinocyte differentiation CPSF siRNA knockdown CPSF CRISPRi MYC knockdown (Fig. 4g–j). fold change IpA usage less drastic in CPSF knockdown influenced by CPSF downregulation Usage minimally altered in migrating keratinocytes versus control biological processes cell type different sets IpA sites.CPSF suppresses GRHL3 expression IpA altered IpA usage CPSF knockdown gene expression intersected CPSF core targets from RNA-seq with 165 genes 178 CPSF-dependent IpA sites identified 14 genes (Fig. 5a majority genes show anticorrelation change between RNA-seq 3′READS+ in CPSF knockdown keratinocyte differentiationIpA site GRHL3 highest FPU in undifferentiated keratinocytes (Fig. usage mRNA expression host gene GRHL3 transcriptional activator expression epidermal differentiation marker genes double knockdown with CPSF CRISPRi GRHL3 RNAi induction differentiation markers mid late-epidermal (SBSN CRCT1 Fig. SBSN and CRCT1 direct targets of GRHL3 other four genes downstream to GRHL3 in keratinocyte differentiation42 suggest GRHL3 key downstream target suppressing differentiation epidermal progenitors IpA could GRHL3 expression.Fig. 5CPSF suppresses GRHL3 expression IpA usage Venn diagram comparing genes CPSF-dependent IpA sites genes altered by CPSF knockdown 14 genes shared Heatmap showing relative expression from RNA-seq usage of IpA sites from 3′READS+ FPU in undifferentiated keratinocytes GRHL3 IpA has FPU in undifferentiated keratinocytes. qRT-PCR quantification of knockdown efficiency differentiation marker gene expression in CPSF1-GRHL3 double knockdown.Downregulation GRHL3 using siRNA CPSF1 restored differentiation gene inhibition keratinocytes Dots data points replicates Illustration two CRISPR strategies disrupt GRHL3 IpA site One single sgRNA PAM site adjacent AAUAAA sequencing second strategy two sgRNAs delete 1-kb genomic sequence IpA site qRT-PCR quantification GRHL3 full-length mRNA level CRISPR KO vs control Dots data replicates qRT-PCR differentiation marker gene expression CRISPR control Dots Source data file used IpA site GRHL3 first intron 4.3 kb downstream transcription start site downregulated differentiated keratinocytes CPSF knockdown GRHL3 mRNA expression upregulated expression GRHL3 two CRISPR approaches knock IpA site first approach leveraged PAM sequence adjacent AAUAAA CPSF sequence targeted site single sgRNA (sg1) small indels second approach co-expressed two sgRNAs + delete 1 Kb genomic sequence IpA site approaches achieved 50% knockout efficiency in primary human keratinocytes estimated PCR TIDEcompared non-targeting sgRNA controls approaches GRHL3 mRNA expression (Fig. 5g knockout cells differentiation marker genes downstream GRHL3 (Fig. 5i j). genomic sequence GRHL3 IpA site used undifferentiated keratinocytes full-length GRHL3 gene expression differentiation human keratinocytes limited life span cell culture without feeder cells GRHL3 expression expanded to immortalized cell lines isolation expansion single clones with CRISPR editing RNA ENCODE GRHL3 IpA site used in HCT116 cells CRISPR KO strategies upregulated GRHL3 in HCT116 cells expanded characterized 60 clones HCT116 cells from “sg2 + sg3” strategy PCR deletion one or both alleles 9 clones showed deletion one 1 clone both alleles HETs mild upregulation GRHL3 KO drastic upregulation 20 fold Sanger sequencing confirmed expected deletion of ~1 kb IpA site sg2 sg3 minor differences (up to 50 bp each end among clones two alleles single KO clonedata demonstrate usage GRHL3 IpA site suppresses full-length GRHL3 mRNA expression.HNRNPA3 CPSF GRHL3 IpACPSF downregulation partially explain fold change selectivity used IpA sites during keratinocyte differentiation GRHL3 IpA usage decreases 100-fold in differentiation qRT-PCR CPSF knockdown ~3–4-fold reduction GRHL3 IpA mechanisms IpA usage targeted screen designed cofactors enhancing GRHL3 IpA strategy double knockdown CPSF candidate cofactor GRHL3 IpA usage 6HNRNPA3 CPSF GRHL3 IpA genetic screen factors GRHL3 IpA knockdown efficiency shRNAs targeting candidates CPSF1-interacting proteins fold change GRHL3 IpA usage double knockdown CPSF single knockdown qRT-PCR change GRHL3 IpA usage HNRNPA3-CPSF double knockdown single control Co-immunoprecipitation CPSF1 associates with HNRNPA3 undifferentiated not differentiated association not disrupted by RNase treatment Quantification immunoprecipitation efficiency Source data provided fileCPSF-interacting proteins in genetic screen pilot spectrometry experiment comparing CPSF1 immunoprecipitation undifferentiated differentiated keratinocytes experiment identified proteins related RNA binding associating with CPSF1 undifferentiated not differentiated complexes CPSF CSTF CFI CFII prioritized CSTF2 most downregulated subunit.Two shRNAs targeting gene validated knockdown efficiency introduced to keratinocytes with CPSF siRNA HNRNPA3 showed highest reduction GRHL3 IpA usage in double knockdown co-immunoprecipitation confirmed HNRNPA3 associated with CPSF1 in undifferentiated condition HNRNPA3 expressed in undifferentiated differentiated keratinocytes protein level slightly reduced in differentiation reduced interaction between CPSF1 HNRNPA3 differentiation reduced expression undifferentiated keratinocytes association between CPSF1 HNRNPA3 not disrupted by RNase HNRNPA3 CPSF1 require RNA bridge usage other differential IpA sites ALOX15B CRBN not enhanced by HNRNPA3 knockdown CPSF RNAidata suggest differentially used IpA sites keratinocyte differentiation regulated HNRNPA3 influences.HNRNPA3 suppresses differentiation influences GRHL3 cooperation HNRNPA3 CPSF1 GRHL3 differentiation epidermal progenitors RNA-seq data HNRNPA3 control identified 1490 differentially expressed genes p < 0.05 upregulated genes enriched with GO terms “epidermal “keratinocyte differentiation” 306 genes affected by downregulation HNRNPA3 CPSF1 overlapping genes include GRHL3 target genes Double knockdown of GRHL3 CPSF expression differentiation genes HNRNPA3 cooperates with CPSF differentiation genes CPSF-independent roles 7HNRNPA3 suppresses differentiation influences GRHL3 splicing Western blotting knockdown efficiency shRNAs targeting HNRNPA3 map expression 1490 genes differentially expressed between HNRNPA3 knockdown versus control Bar graph top GO terms 1490 genes altered by HNRNPA3 knockdown Venn diagram comparing HNRNPA3 target genes versus CPSF target genesqRT-PCR GRHL3 SPRR1B SBSN double single knockdown HNRNPA3 CPSF1 Dots data points replicates RNA-seq 3′READS tracks enriched before GRHL3 IpA site likely enriched “hidden exon”. Illustration qRT-PCR strategies “inclusion” “skipping” hidden exon pre-mRNA versus total RNA Arrow heads primers relative ratios “inclusion” “skipping” “hidden exon” HNRNPA3 knockdown control promoted splicing exon1 exon2 skipped “hidden exon”. Dots Ratio qRT-PCR amplification Exon1-intron1 junction exon1 HNRNPA3 CPSF1 knockdown Dots Source data file HNRNPA3-CPSF collaboration controlling GRHL3 IpA HNRNPA3 hnRNP A/B family proteins two tandem RNA recognition motifs) N-terminal proteins control diverse targets pre-mRNA processing47 best-characterized human protein HNRNPA1 binds consensus sequence UAGGGA/U antagonizes splicing48HNRNPA3 94% similarity tandem RRMs with HNRNPA1 differs in C-terminal region function under-characterized high similarity suggests could bind to similar RNA consensus sequence influence pre-mRNA processing.RNA-seq data in keratinocytes identified potential “hidden exon” enriched RNA-seq before GRHL3 IpA site GRHL3 IpA may involve inclusion through alternative splicing qPCR primers amplify junction between exon 1-“hidden exon” HNRNPA3 knockdown increased ratio skipping versus inclusion antagonizes splicing promotes connection between additional primer pairs quantify ratio between exon1-intron1 junction versus exon1 ratio reduced in HNRNPA3 knockdown not in CPSF1 knockdown HNRNPA3 stabilizes exon1-intron1 junction suppresses splicing intron findings suggest model CPSF and RNA-binding proteins influence gene expression specific IpA sites epidermal tissue MYC functions upstream of CPSF high level expressioninteraction between CPSF HNRNPA3 promotes GRHL3 IpA junction exons cleavage RNA premature expression terminal differentiation genes in epidermal progenitor maintenance (Fig. CPSF-HNRNPA3 cooperation epidermal progenitor maintenance MYC functions CPSF expression in undifferentiated keratinocytes first intron GRHL3 CPSF HNRNPA3 bind to nascent RNA suppress splicing intronic polyadenylation GRHL3 mRNA expression premature differentiation undifferentiated incompletely genomic information accessed gene regulation in development somatic tissue homeostasis human keratinocyte differentiation involves usage intronic polyadenylation sites characterized IpA site within first intron of GRHL3 key transcriptional activator epidermal differentiation Usage GRHL3 IpA expression GRHL3 terminal differentiation genes CPSF protein HNRNPA3 essential usage IpA site.HNRNPA3 RNA-binding proteins with CPSF in undifferentiated keratinocytes HNRNPA3 GRHL3 IpA HNRNPA3 high homology to HNRNPA1 suggests role in splicingHNRNPA1 copies bind spread RNA suppress splicing RNA structure displacing splicing-promoting proteins qPCR HNRNPA3 knockdown promoted exon1-exon2 junction destabilized exon1-intron1 junction motif search RBPmap53 identified three HNRNPA1 motif sites (p < 0.01 Z-score > 2.5) within 2 kb upstream GRHL3 IpA site HNRNPA3 bind GRHL3 pre-mRNA influence IpA usage splicing HNRNPA3 two mechanisms influence GRHL3 splicing polyadenylation bind pre-mRNA suppress splicing bind to CPSF stabilize CPSF binding AAUAAA motif cleavage polyadenylation IpA “sg2 + sg3” CRISPR KO strategy AAUAAA motif affect three HNRNPA3 binding sites Future studies HNRNPA3 GRHL3 IpA additional CRISPR strategies elucidate CPSF GRHL3 expression other RNA binding proteins FUS ELAVL1 enhanced GRHL3 IpA usage lesser extent FUS nascent RNA near alternative PolyA sites54 ELAVL1 implicated in alternative splicingELAVL1 loss exon 11 skipping translation ELAVL1’s association with TRA2-beta inclusion exon256 selectivity of GRHL3 IpA cooperation multiple RNA-binding proteins HNRNPA3 strongest effect target HNRNPA3 enhance CPSF1’s regulation of ALOX15B IpA and CRBN IpA usage different IpA sites may involve diverse regulatory mechanisms studies IpA to truncated proteins21 GRHL3 IpA first intron translated isoform 6 amino acids of original GRHL3 protein other IpA sites location varies among genes ALOX15B IpA in 5th intron mRNA could into protein ~80% lipoxygenase domain barrier IpA lack of high-quality antibodies N-terminal regions 2739 IpA sites identified in keratinocytes 17% overlap with immune system24. compared host genes overlap increased to 45% in keratinocytes AGO3 gene associated with 3 IpA sites immune cells 4 sites Only 2 sites overlap same gene can associate with IpA sites in different cell typeswork genome profiling polyadenylation human keratinocyte differentiation regulatory mechanisms IpA sites reveals roles CPSF HNRNPA3 keratinocyte differentiation pre-mRNA processing tissue homeostasis human keratinocytes isolated from discarded foreskin Northwestern Skin Biology Diseases collected Northwestern University Institutional Review Board Patient consent neonatal foreskin tissue required discarded Keratinocytes from three de-identified donors mixed cultured in 50% complete Keratinocyte-SFM 50% Medium 154 differentiation induced 1.2 mM CaCl2 full confluency four days HEK293T phoenix cells cultured in DMEM 10% fetal bovine serum HCT116 cells cultured McCoy’s 5A Medium) 10% fetal bovine serum lentiviral CRISPRi pLEX_Cas9 plasmid modified Cas9 sequence with KRAB-dCas9-KRAB pLEX-KRAB-dCas9-BSDCMV replaced-SFFV pMH0001 #85969) pLEX-UCOE-SFFV-KRAB-dCas9-BSD-KRAB-MeCP2-BSD amplified-CAGGS-dCas9-KRAB-MeCP2 plasmid #110824) cloned retroviral vector LZRS BamHI NotI CRISPR Cas9-BSD amplified pLEX_Cas9 plasmid #117987) cloned LZRS BamHI sgRNA cloning linearized pLentiGuide plasmid #117986) BsmBI ligated sgRNA-sg-mCherry GRHL3 ESPN tandem sgRNA cloning double-stranded DNA block synthesized GENEWIZ cloned pLentiGuide plasmid #117986) BsmBI shRNA cloning linearized pLKO.1 plasmid #8453) AgeI EcoRI ligated shRNA oligos pLKO.1-sh-puro sequences Supplementary Data transferTransfection HEK293T phoenix cells Turbofect High-titer virus collected 48 72 h post-transfection added keratinocytes centrifuged 1250 rpm 1 h 32 °CKeratinocytes selected blasticidin (5 μg/mL puromycin (2 μg/mL infection 48 h HCT116 cells infected selected blasticidin puromycin resistance CPSF1 (L-020395-00 GRHL3 (L-014017-02) ordered Dharmacon 4D-Nucleofector DNA knockout CRISPR efficiency GRHL3 keratinocytes HCT116 cells DNA extracted Quick-DNA Miniprep Plus Kit sg-1 disruption region amplified Phusion High-Fidelity DNA Polymerase extracted Sanger sequencing results analyzed TIDE43 webtool sg2-sg3 deletion genomic region amplified GoTaq® DNA Polymerase resolved agarose gel quantified Image J software alleles single-clone HCT116 cells PCR product cloned pLZRS vector clones analyzed Sanger sequencing expression tissue immunoblot 20–50 μg cell lysate loaded SDS-PAGE transferred PVDF membranes blots scanned quantified Li-COR immunofluorescence staining tissue sections fixed 50% acetone 50% methanol 4% formaldehydeantibodies incubated 4 °C secondary room temperature 1 h Images captured EVOS FL Auto 2 microscopy processed Image J software formation assayMouse fibroblast 3T3 cells treated 15 μg/mL mitomycin C serum-free DMEM 2 h trypsinized plated × 105 cells per well 6-well plate overnight changed FAD media 1 h before seeding 1000 keratinocytes medium changed 2 days eight days wells washed PBS 3T3 keratinocytes fixed 1:1 acetone/methanol 5 min plates air dry 5 min colonies stained crystal violet healing 24-well plates 0.2 × 106 CTRL CPSF1 keratinocytes seeded next day treated 10 μg/mL mitomycin C 2 h Wounds 100-μl filter pipette tip healing monitored microscope surface area measured Image J software scratches monolayer RNA extraction qPCR analysis scratched non-scratched plates human epidermis keratinocytes nucleofected CPSF1 trypsinized 4 days later counted dermis prepared donated cadaver samples New York Firefighters Skin Bankhuman dermis donors for organotypic epidermal regeneration approved by Northwestern IRB 1.0 × 106 CTRL CPSF1 keratinocytes seeded devitalized dermis organotypic cultures raised to air/liquid interface stratification differentiation 6 days regenerated tissue embedded OCT before cryosectioning imaging TUNEL assay CTRL CPSF1 by In Situ Cell Death Detection Kit cells positive signals counted using Image J software progenitor competition assay keratinocytes labeled by infection retrovirus expressing H2B-GFP or H2B-mCherry before nucleofection CTRL CPSF1 siRNA mCherry keratinocytes counted Image J software quantification top bottom halves Red stock solution Nile Red (Sigma 72485) in acetone 500 μg/mL to 2.5ug/mL tissue sections fixed in 10% formalin for 5 min rinsed in PBS incubated in working solution 10 min room temperatureslides washed stained NucBlue Cell ReadyProbes Reagent 5 mounted anti-Fade Fluorescence Mounting Medium Nile-Red-positive regions measured average thickness 26 images analysis.CPSF1 purification protein identification mass spectrometryKeratinocytes trypsinized washed PBS resuspended 200 μL hypotonic buffer (10 mM HEPES 7.4 1.5 mM MgCl2 KCl protease inhibitor Cells lysed hypotonic buffer 0.4% NP-40 2 min Nuclei pelleted 4000 rpm lysed ten nucleus lysis buffer (50 mM 8.0 0.05% igepal 10% glycerol 2 mM MgCl2 250 mM NaCl sheared 27.5-gauge needle lysis 30 min Insoluble material removed 13,000 10 min nuclear supernatant purification DynabeadsTM Protein G conjugated CPSF1 mouse monoclonal antibody IgG nuclear supernatant 4 °C washed five times nucleus lysis buffer Proteins boiled separated SDS-PAGE immunoblotting mass spectrometry identificationprotein identification immunoprecipitations separated SDS-PAGE stained colloidal blue Gel slices submitted Northwestern Proteomics Facility mass spectrometry analyses-seqRNA extracted Quick MiniPrep DNase I treatment RNA-seq libraries prepared NEBNext Directional RNA Library Prep Kit ribosomal-RNA depletion Poly mRNA Magnetic Isolation Module Libraries sequenced single-end 50-base reads Illumina HiSeq 4000 Northwestern University NUSeq experiments performed modifications multiplexing Illumina HiSeq 4000 detailsPoly(A)+ RNA keratinocytes captured 15 μg RNA NEBNext® Poly(A) mRNA Magnetic Isolation Module Fragmentation ShortCut® RNase III fragmented RNA ligated to 5′ adapter T4 RNA ligase I ligation products captured poly(A)-tail-trimmed RNase H biotin-T15-(+TT)5 Dynabeads MyOne Streptavidin C1 RNA fragments ligated to 3′ adapter T4 RNA ligase 2 ligation products reverse transcribed M-MLV reverse transcriptase PCR Phusion high-fidelity DNA polymerase NEBNext® Multiplex Oligos 15 cycles PCR products size-selected AMPure XP beads sequenced Illumina HiSeq 4000 platform 1x50bp NUSeq Core Northwestern University-PCR RNA extracted Quick-RNATM MiniPrep reverse transcribed SuperScript VILO cDNA synthesis kit proximal PolyA-site usage ratio poly(A)+ RNA isolation NEBNext® Poly(A) mRNA Magnetic Isolation Module England qPCR SYBR Green Master Mix EvaGreen Master Mix Samples run normalized 18S ribosomal RNA Statistical analysis one-way ANOVA Two-tailed Student’s unpaired t-test GraphPad Prism7 Bar graphs error bars mean ± standard deviation Primer sequences qPCR Table S5-seq libraries aligned hg19 genome Hisat258 Browser tracks generated UCSC genome browser Htseq59-count Differential expression analysis DESeq260 tracks normalized sequencing depth.3′READS+ Zheng et al.34 5′ adapters removed CutAdapt 3′ T’s PolyA tails removed saved one non-T base allowed T tails Reads < 22 nucleotides filtered remaining aligned bowtie2 reads MAPQ score less than 10 filtered out Sequence 20 bp downstream PolyA siteT tail non-T base sequence aligned to genome removed aligned sequence T tail two unmappable T’s considered Poly-A Site supporting) reads Poly-A sites merged by 24 nucleotides Sites corresponding to hg19 blacklist version 2 retrotransposons microRNA snoRNA filtered out filtered out poly-A sites with gene GENCODE version 17 annotation table of counts built remaining sites each sum of counts at each gene calculated “Fraction of PolyA site Usage” (FPU) calculated by total counts EdgeR used differential usage p-value filters applied to 3′READS+ PolyA site analyses stably expressed genes functional PolyA sites PolyA sites met criteria gene expressed in keratinocytes 5 counts in UD or DF RNA-seq libraries sum counts from two biological replicates ≥ 10 FPU from one library ≥10% 18173 remaining sites annotated using GENCODE version 17. 1118 sites overlapped with multiple genes removed from analysis designated Poly-A sites as Introns, UTRs Exons prioritized introns 4885 sites designated as intronsremaining sites 11734 identified as UTRs final 436 annotated as exons correct incorrect annotations sites genes once UTRs identified 4525 sites 3566 annotated as UTR’s adjusted remaining 959 sites (54 exons, 905 introns to UTR GENCODE tables annotation included regions filtered out list 17,055 sites 2429 ambiguous in overlapping genes impossible to distinguish filtered gene Removal left final list 14625 poly-A sites determine 428 differentially used Intronic PolyA sites Fig. 1a used fold-change cut off of 2. Sites validated using RNA-seq data combined three replicates undifferentiated three differentiated keratinocytes built table of counts calculated log2FC from UD to DF conditions Sites with fold change >1.5 same direction more counts upstream considered alternatively used IpA sites identify sites regulated by CPSF built table of counts from+ data calculated FC Sites with FC 1.5 same direction as UD DF+ data regulated by CPSF height of 3′READS+ genome browser tracks normalized based on sequencing depthmotif analysis PolyA sites DNA sequences extracted extension 200 bp upstream downstream MEME61,62 motif search RNA (DNA-encoded motif database “Ray2013 Homo sapiens (DNA-encoded background model 1-order dimer biases Nature Research Reporting Summary.Supplementary Review FileReporting Additional Supplementary FilesSupplementary Data 5 | 50.1 | 1.171151 |
10.1038/s41467-020-16782-9 | PMC7293246 | Ischemia and reperfusion damage contribute to early graft dysfunction and recipient’s death. Here the authors show the feasibility and safety of a non-ischemic heart preservation method for heart transplantation in a non-randomized trial. | Pre-clinical heart transplantation studies have shown that ex vivo non-ischemic heart preservation (NIHP) can be safely used for 24 h. Here we perform a prospective, open-label, non-randomized phase II study comparing NIHP to static cold preservation (SCS), the current standard for adult heart transplantation. All adult recipients on waiting lists for heart transplantation were included in the study, unless they met any exclusion criteria. The same standard acceptance criteria for donor hearts were used in both study arms. NIHP was scheduled in advance based on availability of device and trained team members. The primary endpoint was a composite of survival free of severe primary graft dysfunction, free of ECMO use within 7 days, and free of acute cellular rejection ≥2R within 180 days. Secondary endpoints were I/R-tissue injury, immediate graft function, and adverse events. Of the 31 eligible patients, six were assigned to NIHP and 25 to SCS. The median preservation time was 223 min (IQR, 202–263) for NIHP and 194 min (IQR, 164–223) for SCS. Over the first six months, all of the patients assigned to NIHP achieved event-free survival, compared with 18 of those assigned to SCS (Kaplan-Meier estimate of event free survival 72.0% [95% CI 50.0–86.0%]). CK-MB assessed 6 ± 2 h after ending perfusion was 76 (IQR, 50–101) ng/mL for NIHP compared with 138 (IQR, 72–198) ng/mL for SCS. Four deaths within six months after transplantation and three cardiac-related adverse events were reported in the SCS group compared with no deaths or cardiac-related adverse events in the NIHP group. This first-in-human study shows the feasibility and safety of NIHP for clinical use in heart transplantation. ClinicalTrial.gov, number NCT03150147 | IntroductionSurvival after heart transplantation (HT) has improved markedly over the past three decades, but graft dysfunction still remains the leading cause of early mortality. Only one-third of all donated hearts are used because of the risk of early and late graft dysfunction or logistical problems due to the limitations of acceptable allograft ischemic time1,2. Donors are generally older and have more comorbidities now than before3. Allograft ischemia lasting more than 4 h increases the risk of mortality, and marginal donors are less tolerant to ischemia4,5.Despite a general improvement in most aspects of HT, donor hearts are still preserved prior to transplantation with an ischemic static cold storage (SCS). Ischemia and reperfusion (I/R) damage contributes to early dysfunction of the donor heart and death of the recipient. Ischemia results in tissue hypoxia and microvascular dysfunction6–8. The subsequent reperfusion increases the activation of innate and adaptive immune responses, resulting in a cell death program7,9. The injured endothelium increases the risk of acute cellular rejection (ACR) and cardiac allograft vasculopathy (CAV)10. Together, these factors affect early and late survival11,12.With the SCS method, the heart is flushed with cold crystalloid solutions and transported on ice. The nonischemic heart-preservation (NIHP) system is instead a portable device approved for ground and airborne transportation (Fig. 1)13. The heart is continuously perfused with a cold (8 °C) oxygenated cardioplegic nutrition–hormone solution containing erythrocytes from the blood bank. This is in contrast to the organ care system which uses a warm, noncardioplegic preservation solution containing donor blood14.Fig. 1The nonischemic heart-preservation method (NIHP).Shown is a drawing of the NIHP method (a). The equipment consists of a reservoir, a pressure-controlled roller pump, an oxygenator, an arterial-leukocyte filter, a heater–cooler unit, oxygen and carbon dioxide containers, a gas mixer, sensors, and a programmable control system. The reservoir is filled with 2.5 L of the perfusion solution (b) plus ~500 mL compatible irradiated and leukocyte-reduced blood cells from the hospital blood bank, providing a hematocrit of ~15%. Perfusion is provided through the aortic cannula to the coronary vessels. The picture (c) shows the first human heart transplantation using the NIHP method. The heart is mounted and submerged into the heart-preservation solution, which is actively regulated to maintain a pH of ~7.4 and a temperature of 8 °C. The device software is adjusted to maintain a mean blood pressure of 20 mmHg in the aortic root, providing a coronary flow between 150 and 250 mL/min.Preclinical studies, using the NIHP system, have shown that the pig donor heart can be safely preserved for 24 h and that the endothelium contractile function can be preserved for at least 8 h8,13,15. In a recently published study of life-supporting porcine cardiac xenotransplantation using the same system, NIHP was one of two keys to the success16. Therefore, an NIHP system might allow the procurement of distant donor hearts and possibly enable resuscitation of marginal donor hearts, thereby expanding the donor pool. However, this state-of-the-art technology has never been applied to humans.Here we report the first-in-human use of the NIHP method in adult HT. In this nonrandomized phase II study, we investigate event-free survival and immediate graft function. We show a decrease of cardiac injury markers, less ACR, and no death or cardiac-related serious adverse events among recipients transplanted using the NIHP method. Our results show that NIHP is safe and feasible, encouraging further clinical investigations.ResultsRecruitmentBetween April 2, 2017 and September 25, 2018, 42 patients underwent HT, 11 patients were excluded because they met one of the exclusions criteria (4 patients), did not provide written informed consent (4 patients), or required an urgent transplantation (3 patients). Transplantation was planned in advance when the NIHP method could be used because the device and team members trained to use the system must be available. This resulted in the NIHP system being assigned to 6 patients out of the total 31 eligible patients (Fig. 2). The donor and recipient’s characteristics did not exclude any patient from being assigned to the NIHP group; however, they were excluded if they met any exclusion criteria. Following organ retrieval, all organs were used. All patients were followed-up for 6 months or until death, and no data on outcomes were missing. The latest follow-up occurred on March 25, 2019.Fig. 2CONSORT flow diagram.Modified CONSORT flow diagram for all recipients enrolled in the trial. MOF multi-organ failure; NIHP nonischemic heart preservation; SCS static cold storage.Donor, recipient, and preservation characteristicsTable 1 shows the baseline characteristics of the donors and recipients in the two study groups. Overall, eight (26%) recipients and nine (29%) donors were women. The median age was 54 years (interquartile range [IQR], 43–60) for the donors and 56 years (IQR, 46–64) for the recipients. Baseline characteristics, except for body size, were similar for those in the two groups. The donor size was similar in the two groups but the NIHP recipients were larger and had a median body mass index (BMI) of 30 kg/m2 (IQR, 29–32) compared with the SCS group, who had a median BMI of 26 kg/m2 (IQR, 23–28). This resulted in a larger and unfavorable size mismatch in the NIHP group (recipient/donor BMI, 1.2; IQR, 1.1–1.4) compared with the SCS group (recipient/donor BMI, 0.9; IQR 0.7–1.1). Furthermore, the median total preservation time was longer for the NIHP group (223 min; IQR, 202–263) than for the SCS group (194 min; IQR, 164–223), (Supplementary Fig. 1).Table 1Donor, recipient, and transplantation characteristics.Donor characteristicsNIHP (n = 6)SCS (n = 25) Median age (year)56 (46–68)53 (41–58) Female sex1 (17%)8 (32%) Median body mass index (kg/m2)24 (22–27)26 (23–33) Cause of death Cerebrovascular event2 (33%)16 (64%) Head trauma3 (50%)1 (4%) Other1 (17%)8 (32%) Blood group A4 (66%)7 (28%) AB03 (12%) B1 (17%)4 (16%) O1 (17%)11 (44%) History of smoking2 (33%)7 (37%) Hypertension2 (33%)9 (39%) Cytomegalovirus6 (100%)19 (79%)Recipient characteristicsNIHP (n = 6)SCS (n = 25) Median age (year)59 (56–64)55 (46–63) Female sex08 (32%) Median duration on waiting list (days)118 (105–222)114 (33–340) Median body mass index (kg/m2)30 (29–32)26 (23–28) Diagnosis Ischemic cardiomyopathy2 (33%)5 (20%) Nonischemic cardiomyopathy4 (67%)16 (64%) Other04 (16%) Blood group A5 (83%)7 (28%) AB04 (16%) B1 (17%)4 (16%) O010 (40%) Insulin-treated diabetes1 (17%)3 (12%) Peripheral vascular disease1 (17%)5 (20%) History of stroke1 (20%)1 (4%) Preoperative cytomegalovirus4 (67%)17 (68%) Median most recent creatinine (µmol/L)111 (104–136)95 (84–129) Median S-bilirubin (µmol/L)11 (4.0–14)12 (7.0–23) Median pulmonary vascular resistance (Wood units)2.4 (1.7–2.8)2.0 (1.7–2.2) Panel-reactive antibody level 0–10%3 (50%)9 (36%) 11–80%2 (33%)10 (40%) >80%1 (17%)6 (24%) Ventricular assist device3 (50%)12 (48%)Transplantation detailsNIHP (n = 6)SCS (n = 25) Median volume cardioplegia (L)1.2 (1.1–1.3)1.9 (1.8–2.0) Median total preservation time (minutes)223 (202–263)194 (164–223) Median R/D body mass index ratio1.2 (1.1–1.4)0.9 (0.7–1.1) Female donor to male recipient1 (17%)1 (4%)Data are n (%) or median (IQR).R/D recipient/donor, IQR interquartile range.Ex-vivo perfusion dataWe arrested the donor hearts in the NIHP group with the heart-preservation solution without erythrocytes. Then, we harvested the hearts in the same way as performed for the SCS group. We cannulated the distal ascending aorta from the device and submerged the heart in the preservation medium (Fig. 1c). The median preperfusion organ mounting time (ischemic time) was 24 min (IQR, 20–28 min) (Supplementary Fig. 2). The organ was perfused for a median 140 min (IQR, 109–162 min) with a pressure of 20 mmHg (IQR, 19–21 mmHg) resulting in coronary blood flow of 178 mL/min (IQR, 160–221 mL/min). The temperature was stable at 8 °C during the entire perfusion time (Supplementary Fig. 3). The median aB-lactate was 1.5 mmol/L preperfusion (IQR, 1.2–1.5) and 1.4 mmol/L (IQR, 1.3–1.5) after continuous perfusion (Supplementary Table 1).Event-free graft survival (primary outcome)During the first 6 months, all of the patients assigned to the NIHP group met the primary composite outcome of event-free survival (survival free of severe primary graft dysfunction (PGD) at 24 h, free of extracorporal mechanical support use within 7 days, and free of ACR ≥ 2R within 180 days); however, only 18 (72%) of those assigned to the SCS group achieved event-free survival (Kaplan–Meier estimate of event-free survival 72%; 95% confidence interval (CI), 50–86%) (Table 2 and Fig. 3). All patients survived the first 30 days after transplantation. No death or cardiac-related serious adverse events were reported within 6 months after transplantation in the NIHP group; however, four (16%) death and three (12%) cardiac-related serious adverse events occurred in the SCS group (Table 3).Table 2Patients outcomes.Primary outcomeNIHP (n = 6)SCS (n = 25)RR/ES (95% CI) Survival free of event within 180 days6 (100%)18 (72%)1.4 (1.1–1.8) First event that resulted in failure to reach the primary end point PGD within 24 h02 (8%)– ECMO within 7 days01 (4%)– ACR ≥ 2R within 180 days03 (12%)– Death within 180 days01 (4%)–Secondary outcomesNIHP (n = 6)SCS (n = 25)RR/ES (95% CI) Immediate graft function Reperfusion time (minutes)91 (83–95)89 (77–107)0.14 (−0.75 to 1.03) Inotropic score at 6 h post transplantation21 (9–24)30 (20–54)−0.55 (−1.5 to 0.40) LVEF < 40% within 24 h02 (9%)− RVEF < 40% within 24 h1 (17%)6 (27%)0.61 (0.090–4.1) I/R-tissue injury cTnI > 0.02 ng/mL at end of preservation1 (20%)15 (100%)0.20 (0.035–1.2) CK-MB > 4.3 ng/mL at end of preservation06 (33%)– CK-MB 6 ± 2 h after ending preservation (ng/mL)76 (54–101)138 (72–198)−1.18 (−2.2 to 0.10) CK-MB 12 ± 4 h after ending preservation (ng/mL)38 (30–67)53 (41–77)−0.84 (−1.8 to 0.16) CK-MB 24 ± 6 h after ending preservation (ng/mL)16 (10–24)15 (12–38)−0.41 (−1.3 to 0.51) Renal function Minimum creatinine clearance within 7 days33 (31–40)44 (34–59)−0.83 (−1.8 to 0.18) CRRT within 7 days3 (50%)4 (16%)3.1 (0.94–10) Liver function ASAT within 48 h (μkat/L)1.6 (1.4–2.1)2.6 (2.2–3.6)−1.3 (−2.3 to −0.19) ALAT within 48 h (μkat/L)0.4 (0.3–0.5)0.6 (0.4–0.8)−1.0 (−2.0 to −0.067) Time on ventilator (hours)32 (22–54)39 (22–52)−0.19 (−1.1 to 0.71) Acute rejection (ACR ≥ 1R) within 180 days2 (33%)15 (63%)0.56 (0.17–1.8) Duration of ICU stays (days)7.0 (5.4–17)6.0 (5.1–11)0.062 (−0.81 to 0.95)Data are n (%) or median (IQR).ACR acute cellular rejection, ALAT alanine transaminase, ASAT aspartate aminotransferase, cTnI cardiac troponin I, CK-MB creatinine kinase-muscle/brain, CRRT continuous renal replacement therapy, ECMO extracorporeal membrane oxygenation, ES effect size, ICU intensive care unit, IQR interquartile range, I/R ischemia and reperfusion, LVEF left ventricular ejection fraction, NIHP nonischemic heart preservation, RR relative risk, RVEF, right ventricular ejection fraction, SCS static cold storage.Fig. 3The probability of event-free survival.The Kaplan–Meier plot shows the probability of event-free survival (primary end point) defined as survival free of severe primary graft dysfunction at 24 h, survival free of extracorporeal mechanical support use at 7 days, and survival free of acute cellular rejection ≥2R at 180 days (cyan: NIHP group; red: SCS group). Kaplan–Meier estimate free of event was 72% [95% CI 50–86%] for the SCS group. NIHP (n = 6) nonischemic heart preservation; SCS (n = 25) static cold storage.Table 3Serious adverse events.Serious adverse eventsNIHP (n = 6)SCS (n = 25)RRAcute cardiac failure03 (12%)–Acute bleeding (BARC type IV)a06 (24%)–Respiratory failure2 (33%)8 (32%)1.04 (0.29–3.7)Acute kidney failure (KDIGO)a5 (83%)17 (68%)1.23 (0.78–1.9)Acute liver failure00–Permanent stroke00–Permanent pacemaker04 (16%)–Data are n (%).NIHP nonischemic heart preservation, RR relative risk, SCS static cold storage.aFor definition see online methods.Secondary outcomes of NIHP and SCS groupAlthough the NIHP group had a longer duration of preservation (out of body) and the recipients were matched with smaller donors compared with the SCS group, we did not observe any difference in terms of early organ dysfunction or the need for inotropic support. As shown in Table 2, the immediate graft function was similar for both groups. However, there was a difference in cardiac injury markers. One patient (20%) in the NIHP group had a pathological cardiac troponin I (cTnI) > 0.02 ng/mL at the end of preservation compared with all patients in the SCS group (Table 2). Furthermore, the median creatine kinase-muscle/brain (CK-MB) level, assessed 6 ± 2 h after ending perfusion were 76 ng/mL (IQR, 50–101) ng/mL for the NIHP group and 138 ng/mL (IQR, 72–198) for the SCS group (Fig. 4).Fig. 4Creatine kinase-muscle/brain level after preservation according to the treatment group.The box plot shows the creatine kinase-muscle/brain (CK-MB) level at different timepoints after ending preservation (T0). Data are represented as boxplots. The middle line is the median, the lower and upper axis correspond to the first and third quartiles, the upper whisker extends from the axis to the largest value no further than 1.5 × interquartile range (IQR) from the axis, and the lower whisker extends from the axis to the smallest value (at most 1.5 × IQR) of the axis. Data beyond the end of the whiskers are outlying points that are plotted individually. NIHP (n = 6) nonischemic heart preservation; SCS (n = 25) static cold storage.All patients followed a predefined protocol for surveillance and monitoring. During the first 6 months after transplantation, 2 patients (33%) in the NIHP group had an ACR ≥ 1R; however, 15 patients (60%) in the SCS group did (Supplementary Fig. 4). None of the patients in the NIHP group had an ACR ≥ 2R; however, 4 patients (16%) in the SCS group did.The NIHP group showed a tendency for reduced postoperative renal function compared with the SCS group (Table 2). The minimum creatinine clearance levels within 7 days after transplantation were 33 mmol/L (IQR, 31–40) for the NIHP group and 44 mmol/L (IQR, 34–59) for the SCS group. Half of the patients in the NIHP group needed continuous renal replacement therapy (CRRT) within the first 7 days after transplantation; however, only six (24%) patients in the SCS group needed CRRT. None of the patients required dialysis treatment at the last follow-up date. The median aspartate aminotransferase (ASAT) on postoperative day 1 was 1.6 (IQR 1.4–2.0) for the NIHP group; however, it was 2.6 (IQR, 2.2–3.6) for the SCS group (Table 2). None of the patients developed severe liver failure.Serious adverse eventsThe proportion of patients who had serious adverse events (cardiac, renal, pulmonary failure, bleeding complication, or the need for a permanent pacemaker) leading to an extended length of stay in the intensive care unit (ICU) were comparable for the two groups (Table 3). The most common adverse events in the two groups were acute renal failure (22 patients; 71%) and respiratory failure (defined as need for a ventilator for more than 48 h) (10 patients; 32%). The length of stay in the ICU was similar for both groups; 7.0 days (IQR, 5.4–17 days) and 6.0 days (5.1–11 days) for the NIHP and SCS groups, respectively (Table 2).DiscussionThis first-in-human study shows the feasibility and safety of NIHP method’s for HT. All patients in the NIHP group had an event-free survival at 6 months; however, only 72% of the patients in the SCS group had event-free survival at 6 months. Among NIHP patients, we did not observe any early mortality or cardiac-related serious complications. However, in the SCS group, three patients received extracorporeal membrane oxygenation and four patients had a moderate ACR.The pathogenesis of PGD is still unclear, but ischemia/reperfusion injury has been identified as a contributing risk factor5,17,18. In the present study, we found a decrease in the cardiac injury marker cTnI obtained immediately after preservation and in CK-MB levels after 6 h in the NIHP group. Troponin and CK-MB are sensitive markers of cardiac ischemia and myocardial damage6,19. Preclinical studies have shown that CK-MB levels correlate with ischemia/reperfusion tissue damage with HT. Schecter et al. reported that an increased level of cTnI in the preservation solution is associated with development of PGD17,20,21. These findings might indicate that the NIHP method reduces the myocardial damage better than the SCS method.During the first 6 months after transplantation, we also observed less ACR in the NIHP group than in the SCS group. Decreased allograft rejection may suggest that the endothelium was less damaged in the NIHP group; this has been demonstrated in preclinical studies, and is attributable to less ischemia/reperfusion injuries8,15,22. According to the latest International Society for Heart and Lung Transplantation registry report, treatment for rejection within the first year after transplantation was associated with an increased risk of CAV development and an increased mortality risk of up to 50% at 5 years12. Furthermore, ischemia/reperfusion tissue injury may enhance the activation of innate and adaptive immune responses, resulting in the initiation of a cell death program7,9.We noted more undersized and older donors in the NIHP group than in the SCS group. Unfavorable body size mismatch and older donors are well-known risk factors for PGD18. This observation may indicate that using nonischemic preservation for marginal donor hearts can make it possible to expand the donor pool in the future, which has been suggested by others23,24. However, a larger study is needed to confirm this observation.The NIHP method is a new type of technology for clinical use; therefore, a learning effect should be expected. However, all accepted donors were utilized and there were no device-related complications. Both groups had similar proportions of patients with serious adverse events leading to an extended length of stay in the hospital. The simplicity of the NIHP system is probably a significant contribution to these observations. An additional advantage of the method is its hypothermic environment for the heart. The hypothermic preservation provides increased safety and protection against external impacts on the system such as power failure. With normothermic preservation, an interruption in ex-vivo perfusion can result in irreversible damage to the heart. During the only randomized controlled trial evaluating normothermic preservation, five donor hearts were considered unacceptable for transplantation after the use of that preservation system14. Because these hearts were considered acceptable initially, it cannot be ruled out that something happened in transit that rendered these hearts unusable. Creating an artificial environment similar to the physiological state in which a warm beating heart is supposed to work is both complicated and risky. Moreover, it involves additional surgical and technical support and appropriate transport, inevitably resulting in more expensive management compared with what is needed for SCS. The future commercial NIHP system will not require extra personnel support.The potential benefits with NIHP system are an improved postoperative course and reduced total cost of the transplantation. Complications directly connected to the transplant result in increased costs; for example, if the recipient develops PGD requiring mechanical circulatory support, then the ICU stay will be prolonged. An extension of the allograft preservation time will make it possible to schedule transplantation during the day, when the highest competence will be available for these complex, high-risk cases. Furthermore, NIHP may make it possible to increase the donor pool by utilizing more marginal donors and enabling organ sharing across long distances (perhaps even between continents)25. Finally, a preservation system that can decrease the activation of innate and adaptive immune responses resulting in a downregulation of the immune system, might provide further benefits for organ transplantation7,9,24. This would most likely reduce the need for immunosuppression and decrease the occurrence of complications (for example, toxicity, infection, and malignancies).Our study has some limitations. Because it was a nonrandomized trial, bias in the selection of both donors and recipients could have affected the results. Another limitation of this study was its unblinded nature. Personnel involved in patient care could have favored the innovative NIHP treatment or favored the established SCS technique, thus leaving the direction of the potential bias open to speculation.In conclusion, this first-in-human study describes the clinical evaluation of a new technology for HT. It represents a first, necessary step in demonstrating that NIHP is feasible, safe, and effective in clinical practice. Because all patients in the NIHP group had an event-free survival at 6 months, further clinical investigations on the efficacy of machine perfusion in HT are warranted26. To confirm and extend the results of this study, a randomized trial is required and has been initiated.MethodsStudy designThis investigator-led prospective, open-label, nonrandomized trial of NIHP treatment of donor hearts for HT was performed at Skane University Hospital, Karolinska University Hospital, Linköping University Hospital, and Uppsala University Hospital, which cover two-thirds of the counties in Sweden. Six patients were permitted to be transplanted with donor hearts preserved with the nonischemic method. These were compared with contemporary control patients transplanted with hearts preserved according to standard procedures of SCS. The Swedish research ethics committee approved the trial (2016/603). Patients at the aforementioned centers underwent transplantation at Skane University Hospital; then, they returned to their initial centers for care after transplantation. Transplant candidates were discussed by transplant board members and cardiologists from the participating clinics, and patients accepted for transplantation were screened for the study. Patients accepted for transplantation, who did not fulfill any exclusion criteria, were included in this study after they signed written informed consent. Furthermore, patients on the waiting list were screened for the study (starting April 1, 2017) and those eligible were contacted and included after they signed written informed consent. Due to the delay in the trial registration, a control patient underwent transplantation before the clinical trial registration (ClinicalTrial.gov, number NCT03150147) was completed. The transplantation procedure and perioperative care were completed according to standard practices27. All patients were treated with antithymocyte globulin as induction therapy and triple immunosuppression (tacrolimus, mycophenolate mofetil, and glucocorticoids) as maintenance therapy. All participating hospitals followed the national protocol for surveillance and monitoring, which normally includes 14 visits for endomyocardial biopsies during the first year. The biopsies obtained (normally 3–5 biopsies) are sent for histologic evaluation. The grading of ACR (0R, 1R, 2R, or 3R) is done on the basis of an overall assessment of the biopsies according to the ISHLT guidelines28. No major amendments were made to the trial design after the start of recruitment.Eligibility and consentOrgan donors had to be 70 years or younger. Donors were excluded if any of the following criteria were fulfilled: insulin-treated diabetes, significant coronary artery disease, hepatitis B-positive or hepatitis C-positive serology; human immunodeficiency virus-positive serology; tuberculosis, malignancy; and abnormal ventricular function < 45%. All adult (aged 18 years or older) recipients on our waiting list for HT were eligible; however, we excluded those who previously underwent solid organ or bone marrow transplantation, had grown up congenital heart disease, had undergone four or more sternotomies, had known malignancy, had kidney failure (Iohexol plasma clearance < 30 at listing), had liver failure (ASAT, alanine transaminase, or total bilirubin more than five-times the upper limit of normal, or international normalized ratio > 2.0), had ongoing septicemia, and had urgent, and/or systemic inflammatory disorders treated with corticosteroids. Potential participants provided consented while on the waiting list, and that consent was affirmed on the day of transplantation. Consent included allowing the recording of anonymized data for trial purposes and the collection of biological samples for storage in the trial biobank.Study logisticsDonor hearts were offered to our heart transplant program through Scandiatransplant (http://www.scandiatransplant.org). Assessment of potential donor hearts were based on the usual constellation of clinical factors, including history, coronary angiography, echo assessment, and direct examination of the heart during procurement. The same standard criteria for donor hearts were used for the NIHP group. Transplantation was scheduled in advance when the NIHP method could be used, because the device and team members trained to use the system must be available. Donors and recipients were excluded from the NIHP method only if they met any of the exclusion criteria. Furthermore, initially, we could only use ground transportation which limited the pool of potential donors who could be assigned to the NIHP method. After the NIHP method was approved for air transport in April 2018, the system could be used without this restriction, which resulting in that the NIHP method being used for 5 of the total 11 transplantations performed over the next 5 months. In addition, as mandated by the local research ethics committees, safety and logistic feasibility were assessed after the first and third patients were subjected to the NIHP method. All patients eligible for transplantation, but not assigned to the NIHP method who had signed the written informed consent and did not fulfill any exclusion criteria, were included as controls during the study period.Nonischemic heart-preservation deviceThe device used during this study was made in-house and accepted for use by the Department of Medical Technology of Skane University Hospital in Lund, Sweden. The XVIVO Perfusion AB (Göteborg, Sweden) bought the patent to the device and will continue its development with the aim of making it a commercially available device. The device comprises a miniaturized and fully automated heart–lung machine, housed in a portable apparatus (height, 455 mm; length with handles 695 mm; width 415 mm; weight 32 kg), that enables transportation between hospitals (Supplementary Fig. 5). The equipment consists of a reservoir, a pressure-controlled roller pump, an oxygenator, an arterial-leukocyte filter, a heater–cooler unit, oxygen and carbon dioxide containers, a gas mixer, sensors, and a programmable control system. The reservoir (not shown in the figure) is filled with 2.5 L of the heart perfusion solution plus ~500 mL of compatible irradiated and leukocyte-reduced blood cells from the hospital blood bank, providing a hematocrit level of ~15 % (Fig. 1b). The NIHP device software is adjusted to maintain a mean blood pressure of 20 mmHg in the aortic root, providing a coronary flow between 150 and 250 mL/min (Fig. 1a).Nonischemic heart-preservation groupThe donor heart was arrested with the heart-preservation solution without erythrocytes (1200 mL) (Fig. 1b). Then the donor heart was then harvested using the same procedure as that used for the SCS group. Thereafter the distal ascending aorta was cannulated from the device with a special double-lumen cannula for easy deairing (Fig. 2a) and a soft 3/8-inch silicon tube was placed into the left ventricle through the atrium to maintain the ventricle in a decompressed state. This was performed to prevent inflation of the left ventricle if leakage of perfusate medium through the aortic valve were to occur during perfusion. The venae cavae and pulmonary artery were left open for a free outlet of perfusate from the coronary sinus. The double-lumen cannula supplying the aorta with the preservation medium was fixed in a vertical position and the heart was completely submerged in the preservation medium (Fig. 1c). Throughout the perfusion process with the NIHP device, the temperature, perfusion flow, and aortic root pressure were continuously monitored with the built-in sensors. During perfusion of the donor heart (NIHP group) blood samples were retrieved from the reservoir every 30 ± 10 min. After explantation of the recipient heart, the continuous perfusion was switched to intermittent perfusion. During the implantation of the heart, the aortic cannula was kept in the aortic root, thereby facilitating stability of the heart. Intermittent perfusions with 200–300 mL of the preservation solution was administrated through the cannula every 15 min during the implantation procedure to avoid ischemia. The cannula was withdrawn before the aortic anastomosis was performed. Blood samples were retrieved from the coronary sinus in the right atrium.When the NIHP device was used, a research fellow and a research engineer participate in the procedure. The research fellow and research engineer transported the machine-perfusion device to the donor hospital and assisted donor surgeons with connecting the heart to the machine. One of the senior staff surgeons performed the transplantation, and an attending surgeon performed the donor harvesting. No changes were made to the existing rules for organ allocation or transportation protocols.Static cold storage groupFor the SCS group, the donor heart was arrested with a crystalloid cardioplegic (1–2 L) solution (Plegisol; Pfizer, New York, NY). The heart was then stored on ice slush at a temperature of ~4 °C. On arrival to the hospital, 500–800 mL of blood cardioplegia was administrated to the donor heart, and blood samples from the coronary sinus were obtained and analyzed as described previously.Study outcomesThe primary end point was a composite of survival free of severe PGD at 24 h, free of extracorporal mechanical support use within 7 days, and free of ACR ≥ 2R within 180 days5,28. Secondary endpoints included the following: (1) ischemia/reperfusion tissue injury—differences in cTnI and CK-MB collected at end of preservation and 6 ± 2, 12 ± 4, and 24 ± 6 h after the end of preservation (Triage CARDIO3, Alere with Biosite Triage®MeterPro); (2) immediate graft function as indicated by any one of the following clinical indicators: (i) the need for inotropic support (as judged by inotrope score5) in the first 6 h after arrival to the ICU, (ii) reperfusion time (time from aortic cross-clamp release in the recipient to termination of cardio pulmonary bypass), (iii) left ventricular ejection fraction (EF) < 40% on days 1 post operatively, (iv) right ventricular EF < 40% on days 1 post operatively; (3) postoperative renal function (difference in estimated minimum creatinine clearance within 7 days post transplant and need for CRRT within 7 days after transplantation); (4) postoperative liver function, peak ASAT and peak alanine transaminase within 24 h after transplantation; (5) postoperative pulmonary function and ventilator requirement (number of hours); (6) ACR ≥ 1R within 6 months after transplantation; (7) length of stay in the ICU; (8) graft and patient survival at 6 months.During the study, we monitored recipient and donor demographics, medical history, vital signs, laboratory assessments, echocardiography, and right-sided cardiac catheterization. The volumes of the cardioplegic and preservation solutions were registered, as were total preservation and ischemic times. We defined the total preservation time as the donor heart’s out-of-body time of the donor heart (i.e., x-clamp on the donor aorta at donor hospital until release of x-clamp donor aorta at transplant center). Cold ischemia time refers to the length of time that the donor heart was kept cold without any continuous perfusion. The main endpoints, PGD and ACR, were blindly assessed.All endpoints described were included in the current trial registration and were prespecified in the study protocol, except for CK-MB and cTnI collected at end of preservation; these were added to the protocol on September 10, 2017. The timeframe for the primary end point was extended from 30 to 180 days because no events were observed in the NIHP group (study protocol update December 31, 2018). The collection of biological samples from the donor hearts for storage in the trial biobank has not yet been analyzed. Measurements of troponin postoperatively and CK-MB at two extra timepoints, cell-free donor DNA, and EQ-5D were added to the trial registration (NCT03150147) after completion of the nonrandomized part of this study.Serious adverse events(1) Acute cardiac-related events were defined as the need for an intra-aortic balloon pump and/or mechanical circulatory support within 7 days post transplantation; (2) acute bleeding was defined according to the Bleeding Academic Research Consortium (BARC) type IV criteria (>2000 mL/24 h and/or requiring re-operation for bleeding, and/or intracranial bleeding, and/or transfusion of >5 red blood cell concentrates/48 h)29; (3) respiratory failure was defined as impairment of respiratory function requiring re-intubation, requiring tracheostomy, or the inability to discontinue invasive ventilator support within 48 h after cardio pulmonary bypass due to respiratory issues and not due to sedation issues; (4) acute kidney failure was defined according to the Kidney Disease Improving Global Outcomes (KDIGO) criteria as an increase in serum creatinine of >27 μmol/L within 48 h or 1.5 times baseline within 7 days30; (5) acute liver failure was defined as the rapid development of hepatocellular dysfunction, specifically coagulopathy, and mental status changes (encephalopathy) in a patient without prior known liver disease; (6) permanent stroke was defined as an episode of a computed tomography-verified acute neurological dysfunction to be caused by ischemia or hemorrhage that persisted ≥24 h or until death; (7) permanent pacemaker was defined as need for a permanent pacemaker implantation 2 weeks after transplantation.Statistical analysisThe primary outcome (actuarial survival free of event) was analysed using the Kaplan–Meier method. The Kaplan–Meier estimate is presented with 95% CIs. For patients who had more than one event during follow-up that resulted in failure to reach the primary end point, the event that occurred first is the one included in the analysis. The relative risk and 95% CI were calculated for the outcome variables. The effective size was used to compare mean values. Data were assumed to have unequal variances and the approximate degree of freedom was obtained from Welch’s formula. Furthermore, continuous variables were log-transformed to fulfill normality assumptions. The baseline value was defined as the last assessment prior to the transplantation. Continuous variables were summarized using the median, and the IQR and categorical variables were summarized using frequency and percentage. Missing values were not imputed. Because of the small sample size in both groups, only descriptive statistics were performed. Data were collected in Microsoft Excel 16.35 (2019 Microsoft Corporation, Redmond, WA). Statistical analyses were performed using Stata MP statistical package version 16.0 (2019 StataCorp LP, College Station, TX).Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary | nature communications | [
"Article"
] | [
"Heart failure",
"Phase II trials"
] | after heart transplantation improved decades graft dysfunction leading cause early mortality one-third donated hearts used risk early graft dysfunction problems allograft ischemic Donors older have more comorbidities Allograft ischemia 4 h increases risk mortality marginal donors less tolerant to improvement HT donor hearts preserved with ischemic static cold storage Ischemia reperfusion damage early dysfunction death Ischemia results in tissue hypoxia microvascular reperfusion increases immune responses cell death injured endothelium increases risk acute cellular rejection cardiac allograft vasculopathy affect early SCS method heart flushed with cold solutions transported on ice nonischemic heart-preservation system portable for ground airborne transportation heart perfused with cold (8 °C) oxygenated cardioplegic nutrition–hormone solution erythrocytes contrast to system warm noncardioplegic preservation solution donornonischemic heart-preservation method drawing equipment reservoir pressure-controlled roller pump oxygenator arterial-leukocyte filter heater–cooler unit oxygen carbon dioxide containers gas mixer sensors programmable control system reservoir filled with 2.5 L perfusion solution ~500 mL irradiated leukocyte-reduced blood cells hospital blood bank hematocrit ~15% Perfusion through aortic cannula to coronary vessels first human heart transplantation NIHP heart submerged into-preservation solution pH ~7.4 temperature 8 °C device software mean blood pressure 20 mmHg coronary flow 150 250 mL/min.Preclinical studies pig donor heart preserved 24 h endothelium contractile function 8 porcine cardiac xenotransplantation NIHP keys to NIHP might procurement distant donor hearts resuscitation marginal donor hearts donor pool technology never applied to humans first-in-human use NIHP in adult HT nonrandomized phase II study event-free survival immediate graft function decrease cardiac injury markers less ACR no death serious adverse events NIHP results show NIHP safe feasible further investigationsApril 2 2017 September 25, 2018 42 patients underwent HT 11 excluded criteria (4 informed consent required urgent transplantation (3 Transplantation planned advance NIHP method used device team members available NIHP system assigned to 6 patients 31 eligible patients (Fig. 2) donor characteristics exclude NIHP group excluded if met exclusion criteria organ retrieval all organs used patients followed-up 6 months until death no data outcomes missing latest follow-up March 25, 2019.Fig. 2CONSORT flow diagram recipients MOF multi-organ failure NIHP nonischemic heart preservation SCS cold storage.Donor recipient preservation characteristicsTable 1 baseline characteristics donors recipients two study groups eight (26%) recipients nine (29%) donors women median age 54 years donors 56 years recipients characteristics body size similar donor size similar NIHP recipients larger median body mass index (BMI 30 kg/m2 SCS group median BMI 26 kg/m2size mismatch NIHP BMI 1.2 IQR 1.1–1.4 SCS IQR 0.7–1.1). median preservation time longer NIHP (223 min IQR 202–263) SCS (194 min IQR 164–223) 1Donor recipient transplantation characteristicsDonor 25 age (46–68)53 (41–58) Female (17%)8 (32%) body mass index/m2)24 (22–27)26 (23–33) death Cerebrovascular (33%)16 (64% Head (32%) Blood group A4 (66% (28%) (44% (33% (33% (39% Cytomegalovirus6 (79%)Recipient age (56–64)55 (46–63) Female (32%) waiting (105–222)114 (33–340) body mass index/m2)30 (29–32)26 (23–28) Ischemic (33%)5 (20%) Nonischemic (67%)16 (64% Blood A5 (83%)7 Insulin-treated Peripheral vascular Preoperative (67%)17 (68%) creatinine/L)111 (104–136)95 (84–129) S-bilirubin/L)11 (4.0–14)12 (7.0–23) pulmonary vascular resistance (1.7–2.8)2.0 Panel-reactive antibody level (36% 11–80%2 (33% (24%) Ventricular assist (50%)12 (48%)Transplantation cardioplegia)1.2 (1.1–1.3)1.9 (1.8–2.0)preservation time (minutes)223 (202–263)194 (164–223) Median R/D body mass index ratio1.2 (1.1–1.4)0.9 (0.7–1.1) Female donor to male recipient1 (17%)1 (4%)Data median/D/donor range-vivo perfusion arrested donor hearts NIHP group heart-preservation solution without erythrocytes harvested hearts group cannulated distal aorta submerged heart preservation medium median preperfusion organ mounting time 24 min 20–28 perfused 140 min pressure 20 mmHg coronary blood flow 178 mL/min temperature stable 8 °C perfusion median aB-lactate 1.5 mmol/L preperfusion 1.4 mmol/L after continuous perfusion-free graft survival first 6 months patients NIHP group met event-free survival severe primary graft dysfunction 24 h free extracorporal mechanical support use 7 days ACR 2R 180 18 (72%) SCS group achieved event-free survival survival 72% 95% interval 50–86%) patients survived first 30 days after transplantationdeath adverse events 6 months transplantation NIHP four (16%) death three (12%) events SCS outcomes outcomeNIHP 6)SCS 25 Survival 180 days6 (100%)18 (72%)1.4 (1.1–1.8 PGD 24 ECMO 7 ACR ≥ 2R 180 Death 180 outcomesNIHP graft function Reperfusion time (83–95)89 (77–107)0.14 (−0.75 to 1.03) Inotropic score 6 h post transplantation21 (20–54)−0.55 (−1.5 to LVEF < 40% 24 RVEF < 40% 24 h1 (17%)6 (27%)0.61-tissue injury > 0.02 ng/mL (20%)15 (100% CK-MB > 4.3 ng/mL (33%-MB 6 2 h.18 12 4 h (41–77)−0.84 (−1.8 24 6 h (12–38)−0.41 (−1.3 to 0.51) Renal function Minimum creatinine clearance 7 days33 (34–59)−0.83 (−1.8 to CRRT within 7 days3 (50%)4 (16%)3.1Liver function ASAT 48 h/L)1.6)2.6 (2.2–3.6)−1.3 (−2.3 −0.19) ALAT 48/L)0.4)0.6 (−2.0 −0.067) ventilator (22–54)39 (22–52)−0.19 (−1.1 0.71) Acute rejection 1R 180 (33% (63%)0.56 (0.17–1.8) ICU stays (days)7.0 (5.4–17)6.0 (5.1–11)0.062 (−0.81 0 cellular ALAT alanine transaminase ASAT aspartate aminotransferase cardiac troponin CK-MB creatinine kinase ECMO oxygenation effect ICU intensive care unit IQR interquartile range ischemia reperfusion ejection NIHP heart preservation RR risk static cold storage event-free survival Kaplan–Meier plot graft dysfunction 24 h extracorporeal mechanical support 7 days acute cellular rejection ≥2R 180 days 72% NIHP preservation SCS 25 static cold storage adverse eventsadverse eventsNIHP 6)SCS 25)RRAcute cardiac failure03 (12%)–Acute bleeding (BARC type IV (24%)–Respiratory failure2 (33%)8 (32%)1.04 (0.29–3.7)Acute kidney failure (KDIGO (83%)17 (68%)1.23 (0.78–1.9)Acute pacemaker04 nonischemic heart preservation RR risk SCS static cold storage outcomes NIHP SCS NIHP longer duration preservation smaller donors difference early organ dysfunction need inotropic support immediate graft function similar difference cardiac injury markers One patient (20%) NIHP pathological cardiac troponin I) > 0.02 ng/mL end preservation SCS group median creatine kinase-muscle/brain) level 6 ± 2 h perfusion 76 ng/mL NIHP 138 ng SCS group 4Creatine kinase-muscle/brain level after preservation treatment group plot level timepoints preservation Data boxplotsmiddle line median lower upper axis first third quartiles upper whisker extends to largest value 1.5 × interquartile range lower whisker extends to smallest value 1.5 × IQR) Data beyond end whiskers outlying points plotted individually NIHP = 6) nonischemic heart preservation SCS = 25) cold storage patients followed protocol surveillance monitoring first 6 months 2 patients (33%) NIHP ACR ≥ 1R 15 patients (60%) SCS group None NIHP ACR ≥ 2R 4 (16%) SCS NIHP group reduced postoperative renal function SCS group minimum creatinine clearance levels 7 days 33 mmol/L 31–40) NIHP 44 mmol/L 34–59) SCS group Half patients NIHP needed renal therapy first 7 days six (24%) SCS needed None required dialysis last follow-up date median aspartate aminotransferase) postoperative day 1 1.6 NIHP group 2.6 2.2–3.6) SCS group None developed severe liver failure.Serious adverse proportion serious adverse events bleeding permanent pacemaker extended stay intensive care unit comparable two groups (Tablecommon adverse events acute renal failure (22 71%) respiratory failure 48 (10 32%) length stay ICU similar 7.0 days 6.0 days (5.1–11 for NIHP SCS groups (Table 2) study shows feasibility safety NIHP for HT patients NIHP group event-free survival at 6 months 72% SCS group at 6 months NIHP early mortality cardiac-related serious complications SCS group three patients received extracorporeal membrane oxygenation four moderate ACR pathogenesis of PGD unclear ischemia/reperfusion injury contributing risk decrease cardiac injury marker cTnI CK-MB levels after 6 h in NIHP group sensitive markers of cardiac ischemia myocardial CK-MB levels correlate with ischemia/reperfusion damage HT increased cTnI associated with development PGD17 NIHP method reduces myocardial damage better than SCS first 6 months after transplantation less ACR in NIHP group SCS group Decreased allograft rejection endothelium less damaged in NIHP group attributable to less ischemia/reperfusion injuries8latest International Society for Heart and Lung Transplantation report treatment for rejection first year after transplantation increased risk CAV development mortality risk 50% at 5 years12 ischemia/reperfusion tissue injury enhance immune responses cell death program7,9 more undersized older donors in NIHP group than SCS group body size mismatch older donors risk factors for PGD18 indicate nonischemic preservation for marginal donor hearts can expand donor pool future suggested larger study needed to confirm NIHP method new technology learning effect expected all accepted donors utilized no device-related complications groups similar proportions of patients with serious adverse events extended stay in hospital simplicity of NIHP system to observations advantage hypothermic environment for heart hypothermic preservation provides safety protection against impacts power failure normothermic preservation interruption in ex-vivo perfusion can irreversible damage heart randomized controlled trial five donor hearts unacceptable for transplantation after preservation transit unusable Creating artificial environment similar heart complicated risky involves additional surgical technical support transport more expensive management future commercial NIHP system require extra personnel supportpotential benefits NIHP system are improved postoperative course reduced cost transplantation Complications connected transplant costs PGD support ICU stay prolonged extension of allograft preservation time transplantation during day highest competence for complex high-risk cases NIHP may increase donor pool marginal donors organ sharing across long distances continents preservation system activation adaptive immune responses might benefits for organ transplantation7,9 need for immunosuppression complications infection study has limitations nonrandomized trial bias could affected results unblinded nature Personnel could NIHP treatment or SCS technique potential bias open to speculation first-in-human study clinical evaluation new technology for HT first step demonstrating NIHP feasible safe effective clinical practice all patients NIHP group had event-free survival at 6 months further clinical investigations on efficacy machine perfusion HT warranted26 To confirm results randomized trial required initiated investigator-led open-label nonrandomized trial of NIHP treatment donor hearts for HT performed at Skane University Hospital Karolinska University Hospital Linköping University Hospital Uppsala University Hospital two-thirds of counties in SwedenSix patients transplanted with donor hearts nonischemic method compared with control patients standard procedures Swedish research ethics committee approved trial (2016/603) Patients underwent transplantation at Skane University Hospital returned initial centers after Transplant candidates discussed by board cardiologists patients accepted screened Patients accepted criteria included after informed consent patients waiting list screened April 1, 2017) eligible contacted included after consent delay trial registration control patient underwent transplantation before clinical trial registration transplantation procedure perioperative care completed standard practices27 patients treated with antithymocyte globulin triple immunosuppression (tacrolimus mofetil glucocorticoids maintenance hospitals followed national protocol surveillance monitoring 14 visits endomyocardial biopsies first year biopsies 3–5 sent for histologic evaluation grading of ACR (0R 1R 2R 3R) overall assessment biopsies ISHLT No major amendments trial design after recruitment.Eligibility consentOrgan donors 70 years or youngerDonors excluded if criteria insulin-treated diabetes significant coronary artery disease hepatitis B or C human immunodeficiency virus-positive tuberculosis malignancy abnormal ventricular function < 45% All adult 18 or older recipients HT eligible excluded previously solid organ bone marrow transplantation congenital heart disease four or more sternotomies known malignancy kidney failure (Iohexol plasma clearance < 30 failure bilirubin than five-times upper limit normal ratio > ongoing septicemia urgent systemic inflammatory disorders treated with corticosteroids participants consented waiting list affirmed day transplantation Consent recording anonymized data for collection of biological samples for biobank hearts offered transplant through Scandiatransplant Assessment based on clinical factors history coronary angiography echo assessment direct examination standard criteria NIHP group Transplantation scheduled in advance NIHP method device team members available Donors excluded NIHP if met exclusion criteria initially ground transportation limited pool potential donors NIHPNIHP method approved for air transport April 2018 used without restriction for 5 11 transplantations next 5 months safety logistic feasibility assessed after first third patients NIHP patients eligible for transplantation not NIHP consent exclusion criteria included as controls study.Nonischemic heart-preservation device made in-house accepted by Department of Medical Technology of Skane University Hospital Lund Sweden XVIVO Perfusion AB (Göteborg bought patent development commercially available miniaturized automated heart–lung machine portable apparatus 455 695 415 weight 32 enables transportation between hospitals equipment reservoir pressure-controlled roller pump oxygenator arterial-leukocyte filter heater–cooler unit oxygen carbon dioxide containers gas mixer sensors programmable control system reservoir filled with 2.5 L heart perfusion solution ~500 mL compatible irradiated leukocyte-reduced blood cells from hospital blood bank hematocrit level ~15 % NIHP software mean blood pressure 20 mmHg coronary flow between 150 and 250 mL/min 1a).Nonischemic heart-preservation donor heart arrested with without erythrocytes (1200 mL)donor heart harvested same procedure as SCS group distal ascending aorta cannulated with double-lumen cannula (Fig. 2a soft 3/8-inch silicon tube placed into left ventricle decompressed prevent inflation if leakage perfusate medium venae cavae pulmonary artery left open for outlet perfusate double-lumen cannula preservation medium fixed in vertical position heart submerged in preservation medium (Fig. 1c). perfusion process NIHP device temperature perfusion flow aortic root pressure monitored with sensors blood samples retrieved every 30 ± 10 min After explantation switched to intermittent perfusion implantation aortic cannula kept in aortic root stability Intermittent perfusions with 200–300 mL preservation solution every 15 min to avoid ischemia cannula withdrawn before aortic anastomosis Blood samples retrieved from coronary sinus right atrium NIHP device research fellow engineer transported machine-perfusion device to donor hospital assisted surgeons connecting heart machine senior staff surgeons performed transplantation attending surgeon performed donor harvesting No changes to rules for organ allocation or transportation protocolsStatic cold storage SCS group donor heart arrested with crystalloid cardioplegic (1–2 L solution (Plegisol Pfizer New York heart stored on ice slush ~4 °C hospital 500–800 mL blood cardioplegia administrated blood samples coronary sinus obtained analyzed primary end point survival free severe PGD at 24 h extracorporal mechanical support 7 days ACR ≥ 2R within 180 Secondary endpoints/reperfusion tissue injury—differences in cTnI CK-MB end preservation 6 ± 2 12 ± 4 24 ± 6 h after immediate graft function clinical indicators need inotropic support first 6 h reperfusion time left ventricular ejection fraction < 40% 1 right ventricular EF < 40% 1 postoperative renal function minimum creatinine clearance 7 days need CRRT 7 days liver function peak ASAT alanine transaminase 24 h pulmonary function ventilator requirement ACR ≥ 1R 6 months after transplantation length of stay ICU graft patient survival at 6 monthsstudy monitored donor demographics medical history vital signs laboratory assessments echocardiography right-sided cardiac catheterization volumes cardioplegic preservation solutions registered total preservation ischemic times defined total preservation time as donor heart’s out-of-body time Cold ischemia time heart cold without continuous perfusion main endpoints PGD ACR blindly assessed endpoints included current trial registration prespecified protocol except CK-MB cTnI end preservation added September 10, 2017. timeframe for primary end point extended from 30 to 180 days no events NIHP group December collection biological samples for storage biobank not analyzed Measurements troponin CK-MB cell-free donor DNA EQ-5D added to trial registration (NCT03150147 after nonrandomized partadverse Acute cardiac events need intra-aortic balloon pump mechanical circulatory support 7 days post transplantation acute bleeding defined type IV criteria (>2000 mL/24 h re-operation intracranial transfusion >5 red blood cell concentrates/48 h respiratory failure impairment respiratory function requiring re-intubation tracheostomy inability discontinue invasive ventilator support within 48 h after bypass acute kidney failure increase serum creatinine >27 μmol/L within 48 h or 1.5 times baseline within 7 days30 acute liver failure hepatocellular dysfunction coagulopathy mental status changes (encephalopathy patient without prior liver disease permanent stroke acute neurological dysfunction ischemia or hemorrhage ≥24 h until death permanent pacemaker need implantation 2 weeks after transplantation primary outcome survival free event analysed Kaplan–Meier method 95% CIs one event event first included analysis relative risk 95% CI calculated for outcome variables effective size mean values Data assumed unequal variances degree freedom obtained from Welch’s formulacontinuous variables log-transformed normality assumptions baseline value last assessment transplantation variables summarized median IQR categorical variables frequency percentage Missing values not imputed small sample size descriptive statistics performed Data collected Microsoft Excel 16.35 Redmond Statistical analyses Stata MP statistical package 16.0 (2019 StataCorp College Station Nature Research Reporting Summary.Supplementary information Peer Review File Reporting Summary | 47.5 | 0.612297 |
10.1038/s41467-020-16418-y | PMC7250864 | The DIS3L2 exonuclease degrades aberrant 7SL RNAs tagged by an oligouridine 3′-tail. Here the authors analyze DIS3L2 knockout mouse embryonic stem cells and suggest that DIS3L2-mediated quality control of 7SL RNA is important for ER-mediated translation and calcium ion homeostasis. | DIS3L2-mediated decay (DMD) is a surveillance pathway for certain non-coding RNAs (ncRNAs) including ribosomal RNAs (rRNAs), transfer RNAs (tRNAs), small nuclear RNAs (snRNAs), and RMRP. While mutations in DIS3L2 are associated with Perlman syndrome, the biological significance of impaired DMD is obscure and pathological RNAs have not been identified. Here, by ribosome profiling (Ribo-seq) we find specific dysregulation of endoplasmic reticulum (ER)-targeted mRNA translation in DIS3L2-deficient cells. Mechanistically, DMD functions in the quality control of the 7SL ncRNA component of the signal recognition particle (SRP) required for ER-targeted translation. Upon DIS3L2 loss, sustained 3’-end uridylation of aberrant 7SL RNA impacts ER-targeted translation and causes ER calcium leakage. Consequently, elevated intracellular calcium in DIS3L2-deficient cells activates calcium signaling response genes and perturbs ESC differentiation. Thus, DMD is required to safeguard ER-targeted mRNA translation, intracellular calcium homeostasis, and stem cell differentiation. | IntroductionWhile surveillance of protein-coding messenger RNAs (mRNAs) and mechanisms of nonsense-mediated decay (NMD) have been extensively investigated, little is known about quality control processes for non-coding RNAs (ncRNAs). DIS3L2-mediated decay (DMD) was recently identified as a surveillance pathway for certain ncRNAs, including ribosomal RNAs (rRNAs), transfer RNAs (tRNAs), small nuclear RNAs (snRNAs), microRNAs (miRNAs), and RNA component of mitochondrial RNA processing (RMRP)1–13. These aberrant ncRNAs are oligouridylated by the terminal uridyl transferases (TUTases) TUT4 (also known as ZCCHC11, and TENT3A) and TUT7 (ZCCHC6, or TENT3B), and subsequently degraded by the 3′–5′ exoribonuclease DIS3L2. Accordingly, DMD was implicated in a variety of cellular and physiological processes, including miRNA biogenesis and stability1,11,14–16, rRNA maturation9,13, mRNA degradation and apoptosis17–19, cell proliferation20–23, differentiation14,15, and gametogenesis24, as well as alternative splicing22. DIS3L2 binds to and degrades RNA species that are tagged by poly-uridine 3′ tails—a RNA modification that is catalyzed by the cytoplasmic TUTases1,5,7,11,25. Increasing evidence supports that this tag marks highly structured ncRNAs with imprecise 3′ ends for degradation, which could otherwise impair their proper folding and/or formation of multi-subunit ribonucleoprotein complexes7–10,12. One of the highly enriched ncRNAs in DIS3L2 immunoprecipitation (IP) experiments is the 7SL RNA component of the signal recognition particle (SRP) complex involved in endoplasmic reticulum (ER)-targeted mRNA translation7,12. While mutations in DIS3L2 are associated with Perlman syndrome20, the biological significance of impaired DMD for aberrant ncRNAs, including 7SL RNA, is obscure and pathological RNAs have not been identified18,21,24,26. Moreover, DMD contribution to mRNA translation has not been addressed so far.Eukaryotic ER bound to translating ribosome machineries (also referred to as rough ER, or RER) is the main organelle responsible for coordinated biogenesis, folding, post-translational modification, and sorting of membrane-associated, secretory, and extracellular proteins27–29. Moreover, ER, with its unique architecture stretching from the nuclear envelope to the cell membrane30 functions as a main intracellular storage reservoir for calcium ions (Ca2+), responds to environmental cues and developmental signals and is involved in stress sensing in eukaryotic cells31–34. The biogenesis of several secreted growth factors and hormones, as well as membrane-localized signaling receptors, metabolites and ion channels, rely on ER-associated mRNA translation (reviewed in ref. 35). Among other pathways, SRP-dependent recruitment of ribosome-bound mRNAs to the ER translocons is a major first step towards the final destination of the encoded proteins36–41. SRP itself is an evolutionarily conserved ribonucleoprotein complex comprising of the RNA polymerase III-encoded 7SL RNA as well as six protein subunits: SRPs 72, 68, 54, 19, 14, and 9 in eukaryotes. Notably, disruption of SRP complex results in dysregulation of ER-associated mRNA translation and secretory protein sorting39, suggesting the significance of intact SRP complex for normal secretory and membrane proteins. ER-targeted mRNA translation starts with cytosolic ribosomes bound to respective mRNAs that stall upon translation of the signal peptide in the amino-terminus of the nascent polypeptide40,42. Signal peptide recognition and binding by SRP is essential for this stall and for recruitment of the mRNA to the ER membrane. Perturbation of SRP abrogates ER-targeted mRNA translation and results in inhibition of protein sorting or protein secretion39,43,44, as well as increased calcium leakage from the ER translocon45,46.In this study, we reveal a key role for DMD-mediated quality control of 7SL RNA. In the absence of DIS3L2, the aberrant uridylated 7SL RNA inhibits the function of the SRP that leads to defective translation of secreted and transmembrane proteins at the ER and compromised ER-targeted calcium homeostasis. Consequently, embryonic stem cell (ESC) differentiation including that towards the renal lineage is perturbed, reminiscent of the renal abnormalities in Perlman syndrome patients20.ResultsDIS3L2 is specifically required for ER-targeted mRNA translationWe set out to simultaneously study mRNA expression and mRNA translation efficiency (TE) in DIS3L2 knockout mouse ESCs (mESCs) using ribosome profiling (Ribo-seq)47. Consistent with previous reports4,12, DIS3L2 loss did not affect global mRNA expression levels. Strikingly however, altered translation of many mRNAs was detected by changes in the abundance of ribosome-protected fragments (RPFs) (Fig. 1a, Supplementary Fig. 1a, and Supplementary Data 1). TEs of hundreds of mRNAs were significantly changed (at least 2-fold) in DIS3L2 knockout cells compared to control ESCs (Fig. 1b). Gene ontology (GO) analysis of translationally downregulated mRNAs showed enrichment of membrane- and ER-localized transcripts, both of which utilize ER-associated mRNA translation for protein synthesis, whereas translationally upregulated transcripts were associated with the mitochondria (Fig. 1c). Since the connection between DIS3L2 function and mitochondrial physiology has been previously established17,18, we chose to focus on the role of DIS3L2 in ER-associated protein synthesis. For the translationally downregulated mRNAs, we observed decreased RPFs throughout the coding regions (CDS) of the mRNAs, yet interestingly, we noticed a specific accumulation of RPFs at the 5′ end of the respective CDS (Fig. 1d), consistent with the observation that signal peptides are typically contained at the termini of respective proteins48. In contrast, RPFs of translationally upregulated mRNAs were uniformly increased throughout CDS (Supplementary Fig. 1b). Using three independent tools (Phobius49, SignalP50, TMHMM51), significantly more transmembrane or signal peptide-containing proteins were detected among the transcripts with downregulated translation in knockout cells (Fig. 1e). These results suggest that in DIS3L2 knockout cells, translation of some of the ER-associated mRNAs is stalled at the sequences that encode signal peptide.Fig. 1Dysregulated ER-targeted mRNA translation upon DIS3L2 loss.a Scatter plot representing mRNA expression (input) and translation efficiency [ribosome-protected fragments (RPFs) abundance in DIS3L2 knockout cells compared to heterozygote cells as control (CTRL)]. Translationally up- and downregulated genes (≥2-fold change) are marked in red and green, respectively. b Scatter plot representing significant perturbation of translation efficiency upon DIS3L2 loss. P values are from two-sided Wald tests (DESeq2) and P value adjustments were made for multiple comparisons. c Gene ontology analysis of translationally dysregulated mRNAs in the DIS3L2 knockout mESCs. d Overall distribution of RPFs among translationally downregulated mRNAs. e Significant over-representation of transmembrane and/or signal sequences containing transcripts among translationally downregulated mRNAs (P values are from two-sided Fisher’s exact tests).Metabolic labeling (35S methionine labeling) of de novo synthesized proteins revealed specifically reduced levels of secreted proteins in the culture media (but not in the cell lysates) of DIS3L2 knockout cells (Fig. 2a, Supplementary Fig. 2a). Chemical inhibition of the sarco/ER Ca2+ ATPase by thapsigargin treatment52 suppressed de novo protein synthesis to a greater extent in DIS3L2 knockout cells than in control cells (Fig. 2b), suggesting compromised ER-mediated calcium homeostasis upon DIS3L2 depletion. To directly assess ER-targeted mRNA translation, we used a specific luciferase reporter (GLuc) that is secreted from the cells and accumulates in the culture media53. Secreted luciferase (but not other luciferase reporters, Fig. 2f) was produced at a significantly lower level in DIS3L2-depleted mESCs (Fig. 2c). Re-expression of WT DIS3L2, but not the catalytic mutant protein, rescued the relative levels of secreted luciferase in the knockout cells (Fig. 2d). The specific requirement for DIS3L2 for expression of the secreted luciferase reporter was also detected in different human cell lines (Fig. 2e). Next, to investigate the signal peptide sequence requirement for defective ER-targeted translation in the absence of DIS3L2, we engineered dual luciferase reporters containing Firefly (as an internal control) and Renilla luciferases with or without ER-specific signal peptide (from Insulin mRNA) and an ER-retention signal, KDEL (http://signalpeptide.de). Assessing luciferase activities of these and the parental reporters in the lysates and the supernatant samples obtained from control and DIS3L2 knockout mESCs showed that defective protein translation in DIS3L2 knockout cells depends on the presence of the signal peptide-coding sequence at the 5′ end of the luciferase reporter. Also, removal of the KDEL signal from the 3′ end of the reporter direct luciferase reporters more effectively to the supernatant and diminishes their detection in the lysates, thereby underscoring the robustness and sensitivity of our reporter systems (Fig. 2f). Notably, no tangible changes were observed in the mRNA expression of transfected luciferase reporters in DIS3L2 knockout cells (Supplementary Fig. 2b–e). These results highlight the specific requirement of DIS3L2 for normal ER-targeted translation of a subset of mRNAs encoding transmembrane or secreted proteins.Fig. 2DIS3L2 is specifically required for ER-targeted mRNA translation.a Quantification of de novo protein synthesis in cell supernatants or lysate samples using metabolic labeling. Bars representing mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 6 independent experiments) are shown. b Quantification of the de novo protein synthesis after TG treatment using metabolic labeling. Bars represent mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 3 independent experiments) are indicated. In a, b, CPM represents counts per minute. c Normalized GLuc luciferase activity in the supernatant or cell lysates. Bars representing mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 9 independent experiments) are indicated. d Left panel showing Western blot and right panel showing normalized activity of secreted GLuc reporter after over-expression of DIS3L2 proteins. Bars representing mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 6 independent experiments) are indicated. e Upper panel: Western blot analysis of DIS3L2 expression in different human cell lines stably depleted of DIS3L2; lower panel: normalized activities of ER (GLuc) and cytosolic (Renilla) luciferases after DIS3L2 knockdown [n = 4 independent experiments (293T cells); n = 5 independent experiments (HCT116 cells) and n = 5 independent experiments for BJ cells]. Minimum, maximum, median, and boxes extending from the 25th to 75th percentiles are represented. P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals) are shown. Boxes mark minimum to maximum values. f Upper panel: Schematic representation of dual luciferase reporters psiCHECK-2 (parental), 5′-end signal peptide/3′-end KDEL-tagged (SP + KDEL), and only 5′-end signal peptide tagged (SP), respectively; lower panels: normalized activity of indicated luciferase reporter in the supernatants or lysate samples from control or DIS3L2 knockout mESCs. Bars representing mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 3 independent experiments) are indicated. Source data are provided as a Source data file.DIS3L2-mediated quality control of 7SL ncRNADMD is involved in the quality control of several structured ncRNAs4,7,10,12,21,24. While DIS3L2 has been implicated in the surveillance of rRNAs, since we found no change in steady-state rRNAs levels, bulk translation as measured by polysome analysis9 or metabolic labeling (Fig. 2a), or requirement for DIS3L2 for the translation of non-secreted luciferase reporters (Fig. 2f), we considered that rRNA is unlikely to account for the observed specific defect in ER-targeted translation. We next explored whether defects in certain tRNAs might contribute to the defective translation detected in DIS3L2-deficient cells. Using our Ribo-seq data and published CLIP-seq results on DIS3L2-targeted tRNAs12, we performed an analysis of codon usage frequencies in DIS3L2 knockout mESCs. This showed no codon exclusion or preference associated with these potentially misprocessed tRNAs (Supplementary Figs. 3 and 4), which suggests that it is unlikely that impaired ER-targeted translation in DIS3L2 knockout cells is caused by defective tRNA incorporation into ribosomes. Instead, due to a direct involvement of the 7SL RNA component of the SRP complex on ER-targeted translation40,42, we sought to systematically determine whether defective ER translation is due to a role for DIS3L2 exoribonuclease in the quality control of 7SL RNA. Rapid amplification of complementary DNA (cDNA) ends from circularized RNAs (cRACE)8 was used to precisely characterize the 3′ end of the 7SL RNA in DIS3L2-depleted cells. This showed extensive 3′-end uridylation in DIS3L2-bound transcripts (Fig. 3a), which are mostly truncated from the canonical 3′ end of 7SL RNAs (Fig. 3b). Logistic regression analysis of DIS3L2-bound 7SL RNAs determined a significantly positive correlation between the 7SL RNA 3′-end truncation and the probability of 3′-end uridylation, further underscoring the quality control function of DIS3L2, especially for truncated and thus aberrant 7SL RNAs (Fig. 3c).Fig. 3DIS3L2-targeted quality control of 7SL.a cRACE analysis of DIS3L2-bound 7SL RNAs in knockout cells revealing extensive 3′-end uridylation. b cRACE analysis of DIS3L2-bound 7SL RNAs in respect to its canonical 3′ end (underlined). Note that most of the reads end before the canonical end. Nucleotides from mature 7SL RNA or genomic extensions that are transcribed into uridine are marked in magenta and excluded from the analysis. c Logistic regression analysis of U-tailed occurrence in respect to 7SL RNA truncation or genomic extension. d Co-precipitation of 7SL RNA with ectopically expressed FLAG-SRP68 (F-SRP68) protein analyzed by qRT-PCR. Upper panel showing Western blot. Lower left panel: Equal enrichment of 7SL RNA in FLAG-SRP68 precipitates from knockout and control samples; lower right panel: relative uridylation of 7SL RNA in corresponding samples. Bars represent mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 4 independent experiments). e Enrichment of uridylated 7SL RNA in RPL23a-IP samples in DIS3L2 knockout samples. Upper panel showing Western blot and lower panel showing qRT-PCR. Bars representing mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 3 independent experiments) are indicated. f Northern blot analysis of 7SL RNA expression in total RNA samples from control and DIS3L2 knockout cells. Slow migrating uridylated 7SL RNA species are marked. This experiment was repeated thrice independently with similar results. g qRT-PCR analysis showing a significant accumulation of uridylated 7SL RNA in the subpolysome and specially polysome fraction of DIS3L2 knockout cells. Bars representing mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 3 independent experiments) are indicated. h Normalized activity of GLuc and Renilla (from TK vector) reporters in DIS3L2 knockout mESCs, and the effect of TUTases depletion on these two reporters. Bars represent mean ± SEM. P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals) are shown (n = at least three independent experiments). KO represents DIS3L2 knockout mESCs and CTRL represents heterozygous control mESCs. i Normalized secreted luciferase activity following forced expression of full-length (WT) or truncated 7SL RNAs (n = 6 independent experiments). Bars representing mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals) are shown. Source data are provided as a Source data file.IP of FLAG-SRP68 protein and quantitative real-time PCR (qRT-PCR) showed uridylated 7SL RNA to be associated with the SRP (Fig. 3d), and IP of a ribosomal protein showed that uridylated 7SL RNA accumulates in ribosomes (Fig. 3e) in knockout mESCs. Northern blot analysis of total RNA samples showed accumulation of uridylated 7SL RNA in DIS3L2-deficient mESCs (Fig. 3f). Moreover, sucrose gradient centrifugation and qRT-PCR showed the presence of uridylated 7SL RNA in subpolysome and especially in polysomes in DIS3L2-deficient ESCs (Fig. 3g). Notably, defective ER-targeted translation in DIS3L2 knockout cells was (at least partially) rescued by knockdown of the TUTases, TUT4/7, that mediate 7SL RNA uridylation (Fig. 3h, Supplementary Fig. 5a, b). Finally, forced expression of truncated 7SL RNA (Fig. 3i, Supplementary Fig. 5c–e) inhibited ER-targeted translation of the luciferase reporter mRNA, whereas over-expression of full-length [wild-type (WT)] 7SL RNA could rescue the defective ER-targeted luciferase translation in DIS3L2 knockout cells (Fig. 3i), which strongly supports the requirement of the DMD pathway in the surveillance of 7SL RNA and for ER-targeted mRNA translation.Impaired intracellular calcium storage and insulin release upon DIS3L2 lossConsistent with our Ribo-seq, metabolic labeling, and reporter assays showing defective ER-targeted translation, we found decreased calcium levels in the ER of DIS3L2 knockout cells with corresponding increased calcium ion (Ca2+) concentration in the cytosol. This dysregulation of calcium homeostasis mainly resulted from the increased calcium leakage from the ER. Inhibition of translation elongation by anisomycin treatment was shown to lock the translocon in a Ca2+-impermeable configuration45,54. Interestingly, anisomycin treatment alleviated ER calcium leakage in DIS3L2 knockout cells (Fig. 4a). Stimulation of ER calcium release using chemicals targeting different calcium channels further supports defective ER calcium mobilization in DIS3L2-depleted cells (Fig. 4b–e). While ER-targeted translation and ER function in intracellular calcium homeostasis are interconnected processes, our findings suggest that (i) DIS3L2 safeguards ER-targeted translation of transmembrane and secretory proteins and (ii) upon DIS3L2 depletion, the defective/delayed ER translation may cause ER calcium leakage an imbalanced intracellular localization of calcium.Fig. 4Impaired intracellular calcium storage upon DIS3L2 loss.a Left panel: Schematic representation of increased ER calcium leakage through ER membrane-localized translocon (shown in green) upon translation inhibition (perhaps by DIS3L2 loss). Basal free ER (middle) and cytosolic (right) Ca2+ levels are shown in the middle and right panels, respectively (n = 12 independent experiments). Bars representing mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals) are indicated. b–e ER calcium leakage in mESCs pretreated with or without anisomycin (n ≥ 8). In b, note the overlapping values for control and knockout cells treated with anisomycin. Normalized free cytosolic Ca2+ measured after treatment with thapsigargin (c) or ATP (d). e ER calcium release in mESCs upon IP3 stimulation (in b–e, n = 10 independent experiments). Red arrows indicate the time of chemical treatments. f Normalized luciferase activity of indicated reporters in Min6 cells (n = at least four independent experiments). g Relative insulin secretion levels in Min6 cells upon glucose or potassium stimulations (n = 6 independent experiments). h Basal free ER (left) and cytosolic (right) Ca2+ levels in Min6 cells transiently transfected with indicated siRNAs (n = 8 independent experiments). i ER calcium leakage in DIS3L2-depleted Min6 cells (n = 10 independent experiments); red arrow indicates the time of EGTA treatment. j Cytosolic calcium levels upon high glucose stimulation of siDIS3L2-transfected Min6 cells (n = 10 independent experiments); red arrow indicates the time of glucose stimulation. Bars represent mean ± SEM. P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals). Source data are provided as a Source data file.Insulin secretion is a complex physiological process that relies on delicate regulation of calcium intracellular concentration. To test the effect of DIS3L2 loss on hormone secretion, we used Min6 cells as an in vitro model of pancreatic β-cells55,56. Min6 cells sense increased glucose levels and metabolize glucose, which in turn leads to increased ATP/ADP level that triggers closure of a potassium channel. This causes cell membrane depolarization, calcium ion influx, and finally insulin secretion. Upon transient DIS3L2 depletion, ER-targeted translation (Fig. 4f) and glucose-stimulated insulin secretion (Fig. 4g) were attenuated in Min6 cells. Moreover, upon bypassing glucose-sensing steps in Min6 cells by KCl administration, again DIS3L2-depleted cells failed to effectively secrete insulin (Fig. 4g). DIS3L2 knockdown in Min6 cells caused ER calcium depletion and increased cytosolic calcium (Fig. 4h), as well as increased calcium leakage from the ER (Fig. 4i) similar to DIS3L2 knockout mESCs, suggesting a common defect in the regulation of calcium homeostasis. Finally, in contrast to control cells, DIS3L2-depleted Min6 cells failed to increase cytosolic calcium level upon glucose stimulation (Fig. 4j). This highlights the requirement of DIS3L2 for endogenous protein secretion, at least partially through its role in ER-targeted translation to prevent calcium leakage from the ER.DIS3L2 is required for normal embryonic stem cell differentiationRegulation of calcium signaling is critical for proper stem cell differentiation and organ development, and its perturbation leads to malignancy and transformation57–59. Moreover, phenotypic analysis of DIS3L2 knockout animal models or Perlman syndrome26 has suggested a differentiation failure in the renal lineage, although the molecular mechanism contributing to this phenotype is obscure. To elucidate the effect of DIS3L2 loss and calcium homeostasis in stem cells and during renal differentiation, we utilized spontaneous and renal lineage-directed in vitro differentiation of DIS3L2-depleted mESC line (Supplementary Fig. 6a) carrying renal-specific Osr1-GFP reporter60. Similar to DIS3L2 knockout ESCs, stable DIS3L2 knockdown using specific short hairpin RNA (shRNA) led to TUTase-dependent downregulation of ER-targeted translation as measured by GLuc reporter (Supplementary Fig. 6b). DIS3L2 knockdown had no tangible effect on pluripotency of the mESCs, as was evident by normal proliferation, pluripotency marker expression, and their clonogenicity as compared to isogenic control knockdown ESCs (Supplementary Fig. 6c–e). However, DIS3L2 deficiency resulted in the formation of larger embryoid bodies (EBs) during spontaneous ESC differentiation (Fig. 5a, b). High-throughput sequencing of PolyA+ RNAs, hierarchical clustering, and principal component analysis (PCA) of RNA-sequencing data revealed extensive changes in mRNA expression of DIS3L2-deficient cells, especially at the later stages of EB differentiation (Fig. 5c–e, Supplementary Data 2 and 3). Moreover, consistent with defective ESC differentiation, dissociation and replating of cells from EBs at day 12 (d12) in the presence of leukemia inhibitory factor (LIF) and serum generated ESC colonies in DIS3L2-depleted cultures, but not in control knockdown ESCs (Fig. 5f). Interestingly, GO analysis of differentially expressed genes (DEGs) at d10 (Supplementary Fig. 6f, g) and d12 (Fig. 5g, h) of EB differentiation marked enrichment of specific molecular functions (e.g., calcium ion binding) and biological processes (e.g., extracellular matrix organization) (Fig. 5h, Supplementary Fig. 6g). Moreover, DEGs at d12 were associated with “abnormal appendicular skeleton morphology,” “renal glomerular disease,” “spontaneous abortion,” and “wide anterior fontanel” (Fig. 5h), phenotypes that commonly occur in Perlman syndrome patients20 and/or are related to calcium homeostasis. Together, these data suggest that while ESC pluripotency is unaffected, DIS3L2 deficiency causes late differentiation defects that are associated with calcium homeostasis, as well as bone and renal development, recapitulating some of the developmental phenotypes that characterize Perlman syndrome patients.Fig. 5Impaired in vitro differentiation upon DIS3L2 loss.a The size of EBs during in vitro spontaneous differentiation. Minimum, maximum, median, and boxes extending from the 25th to 75th percentiles are shown (n = at least 30 EBs measured). b Representative images of EBs at d12. This experiment was repeated twice independently with similar results. Scale bar = 200 μm. c The number of differentially expressed genes (DEGs) during spontaneous differentiation of EBs. Hierarchical clustering (d) and PCA analysis (e) of mRNA sequencing data are shown. f Phase contrast (left panels) and AP staining (right panels) of d12 differentiated cells replated for 5 days in the presence of LIF and serum to support ESC colonization. This experiment was repeated twice independently with similar results. Scale bar = 200 μm. g Heatmap representation of differentially expressed genes at d12 during the entire course of differentiation. h Gene ontology analysis of DEGs from g at d12 of differentiation; numbers inside the bars indicate the number of DEGs associated with each biological terms.DIS3L2 loss perturbs calcium signaling pathways and renal differentiationTime-course analysis of gene expression during EB differentiation showed increased expression of Ca2+ signaling target/sensor genes in DIS3L2 knockdown cells. These include polycystic kidney disease-associated proteins (Pkd1, Pkd2), calcium/calmodulin-dependent protein kinases (Camk genes), and Regucalcin, but not randomly selected genes (Fig. 6a, Supplementary Data 3). This elevated expression of calcium signaling targets or sensors suggests the cellular response to perturbed calcium homeostasis in DIS3L2-depleted cells (Fig. 4). Moreover, in shDIS3L2 cells, the expression of renal-associated transcription factors was dysregulated during ESC differentiation (Fig. 6b, Supplementary Data 2 and 3). This prompted us to specifically investigate the renal differentiation propensity of DIS3L2-deficient cells using renal-directed differentiation60 (Fig. 6c). DIS3L2 knockdown caused larger renal differentiation aggregates and increased proportion of renal progenitor cells as evidenced by the increased number of Osr1-GFP-expressing cells (Fig. 6d, e, Supplementary Fig. 7), and also the increased expression of renal-specific markers of the metanephric mesenchyme progenitor stage (Fig. 6f, g). These observations are in line with organ—and especially kidney—overgrowth and developmental defects observed in Perlman syndrome patients. Finally, to address the casual and pathological effects of dysregulated 7SL RNA decay in DIS3L2-depleted cells, on renal differentiation, we stably over-expressed truncated 7SL RNA in control and DIS3L2 knockout ESCs and induced in vitro renal differentiation. qRT-PCR analysis of renal-specific markers at the end of differentiation showed that (1) similar to DIS3L2 knockdown cells, DIS3L2 knockout cells also express higher levels of renal progenitor markers compared to control cells; and more interestingly, (2) over-expression of truncated and therefore aberrant 7SL RNA worsened this defect in DIS3L2 knockout cells as is marked by elevated expression of Six2 and IGF2, as well as calcium-sensing protein Camk1g (Fig. 6h). This further highlights the essential function of DIS3L2 in the regulation of renal differentiation and also connects the pathological role of truncated 7SL RNA to the defective differentiation of DIS3L2-deficient cells. Thus, while DIS3L2-depleted undifferentiated ESCs behave normally, they manifest differentiation defects associated with kidney development/function and show overgrowth, especially during renal lineage differentiation. This is also consistent with organomegaly and kidney hypoplasia phenotypes seen in Perlman syndrome patients, and suggests that this is likely due to cell-autonomous abnormalities that can be recapitulated ex vivo.Fig. 6Altered calcium-sensing and renal differentiation propensity upon DIS3L2 loss.a Heatmap representation of differentially expressed genes associated with calcium signaling or calcium sensing. b Heatmap representation of differentially expressed genes associated with renal differentiation. c The protocol used during in vitro differentiation of mESCs to renal progenitors (metanephric mesenchyme). d Phase-contrast and fluorescent images of Osr1-GFPpositive renal progenitors at d8.5 of renal differentiation. Note the larger size of differentiation aggregate as well as increased intensity of Osr1-GFP signal in shDIS3L2 culture. This experiment was repeated trice independently with similar results. Scale bar = 100 μm. e Left panels: Flow cytometry analysis of Osr1-GFPpositive cells in samples from d; right panel, relative number of Osr1-GFPpositive cells at d8.5 of renal differentiation. Bars representing mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 3 independent experiments) are shown. f Western blotting analysis of samples from d; Met. Mes., metanephric mesenchyme stage, that is, 8.5 days after differentiation. Representative images of two independent experiments with similar results are provided. g qRT-PCR analysis of renal-specific markers for indicated samples at day 8.5 of differentiation, corresponding to the metanephric mesenchyme (MeMe) stage. Bars represent mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 3 independent experiments). h qRT-PCR analysis of indicated mRNAs samples at day 8.5 of differentiation, corresponding to the metanephric mesenchyme (MeMe) stage. Bars represent mean ± SEM and P values (unpaired two-tailed Student’s t test, 0.95 confidence intervals; n = 3 independent experiments). Control and DIS3L2 knockout cells were transduced with pLKO.1 empty vector (EV) or a vector stably expressing truncated 7SL RNA under its endogenous promoter. i A model describing the biological function of DIS3L2 in safeguarding ER-targeted translation, calcium homeostasis, and stem cell differentiation. Source data are provided as a Source data file.DiscussionIn this study, we used cellular models to examine the physiological requirement of DMD and to illuminate the underlying molecular mechanism. We conclude that DIS3L2 functions to eliminate aberrant 7SL ncRNA subunit of the SRP particle, and thereby safeguards the ER-targeted mRNA translation of membrane-targeted and/or secretory proteins. Moreover, in the absence of DIS3L2, cellular calcium homeostasis and ER function is impaired. Besides the housekeeping functions, ER-associated mRNA translation and ER calcium homeostasis are particularly critical for the proper function of organs that are specialized in physiological communication with other organs or with the environment. These include pancreas (secreting endocrine hormones and exocrine enzymes)61, nervous system (functionally dependent on ion channels and synaptic vesicles)62,63, muscles (relying on calcium release from ER to contract)64,65, and kidney (harboring several important ion channels, involved in metabolite re-absorption)66. Notably, these are among the most affected organs in Perlman syndrome patients with defective DMD. Thus, it would be of interest to study ER dysfunction in patient samples and DIS3L2 animal models of Perlman syndrome to determine the physiological relevance of defective DMD for 7SL RNA in tissue function and homeostasis21,26.We demonstrate that DIS3L2 functions in the quality control of truncated 7SL RNA. Several other DIS3L2 targets are also directly involved in the mRNA translation, including tRNAs2,10,12 and rRNAs9,13. However, our ribosome footprinting results point to a specific requirement for DIS3L2 in ER-targeted translation via elimination of aberrant 7SL RNAs with imprecise 3′ ends. Re-introducing WT (full-length) 7SL RNA into DIS3L2 knockout cells partially rescued ER-targeted translation. Moreover, we did not observe any codon usage differences upon DIS3L2 loss. Accordingly, we conclude that, at least in mESCs, defective ER translation is not caused by aberrant rRNAs or tRNAs. Instead, in the absence of DIS3L2, pathological truncated 7SL RNA incorporates into SRP particles, associates with ribosomes, and adversely affects homeostatic translation of several proteins, including membrane-targeted and secretory proteins. Cryo-electron microscopy and biochemical approaches have revealed that, at the nascent chain exit site, SRP contacts RPL23 in the large ribosomal subunit to bind the signal sequence40,42. Moreover, SRP perturbation leads to ER-translation defects and mistargeting ER proteins to the mitochondria39. The translation stall at the 5′ end of respective mRNAs that we observe in DIS3L2 knockout cells likely explains the increased ER calcium leakage and consequently elevated cytosolic calcium ion levels.The quality control function of DIS3L2 on 7SL RNA suggests that only a small fraction of the 7SL RNA pool in DIS3L2 knockout cells is truncated and uridylated. The remaining question is how this small portion of 7SL RNA species, although aberrant, negatively impact ER-targeted translation? Also, it needs to be further investigated why only a proportion of mRNAs predicted to be translated through ER is affected by DIS3L2 loss. Truncated and/or uridylated 7SL RNA species may have altered affinity to protein subunits comprising the SRP complex. This, even at a low concentration, may cause a dominant-negative effect on SRP function in recruitment of mRNA/translating ribosome machinery to ER, on SRP complex turnover, or on the robustness and specificity of mRNA substrate recognition by SRP complex36. It is worth noting that not all the ER-translated mRNAs are SRP clients. In fact, a considerable amount of ER-targeted translation is SRP independent39, possibly explaining why in DIS3L2-deficient cells only a subset of ER-associated mRNAs are affected. Furthermore, elevated translation of a subset of mitochondrial transcripts upon DIS3L2 loss as well as the pathological roles of other DIS3L2 targets, including rRNAs and miRNAs, should be investigated in future.The physiological connection between intracellular calcium homeostasis and cell proliferation, as well as differentiation, has been extensively studied (reviewed in ref. 67). For example, in lateral mesoderm retinoic acid induces the translocation of calcium channel TRPP2 from ER to the cell membrane, which in turn potentiates calcium influx to the cytosol, a signaling pathway that leads to the transcriptional activation of renal progenitor factor Pax868–70. Accordingly, temporal and spatial regulation of calcium signaling is in particular critical for renal lineage development, in which Ca2+ stimulates lateral mesoderm to differentiate into renal progenitors69,70. The altered differentiation propensity of DIS3L2-depleted cells toward the renal lineage could therefore be explained by their elevated intracellular Ca2+ levels as well as their impaired intracellular storage mechanisms. Moreover, regulation of intracellular calcium homeostasis has been recently shown to be critical for exit from pluripotency and differentiation59. While LIF/2i-adapted DIS3L2-deficient ESCs remain undifferentiated, the greater competence of these cells to respond to differential stimuli after an in vitro, and perhaps in vivo, exit from pluripotency results in enhanced renal lineage differentiation program and eventually leads to kidney overgrowth, similar to that in the Perlman syndrome patients. This highlights the relevance of DIS3L2 loss-induced defects we observe and the significance of stem cells as a tool for in vitro disease modeling. Finally, our findings raise the possibility that defective SRP complex function caused by DIS3L2 deficiency might be responsible for the severe developmental phenotypes associated with Perlman syndrome20, as well as the perinatal lethality reported in DIS3L2 knockout mice26. We therefore propose that compromised DMD leads to a “SRPopathy” marked by defective ER-targeted mRNA translation (Fig. 6i). Further investigations are required to determine the extent by which the proposed SRPopathy may contribute to the phenotypes observed in Perlman syndrome patients.MethodsCell cultureTC1 mESCs were cultured without feeder on 0.2% gelatin (Sigma) as previously described71. DIS3L2 knockout ES cells were generated previously using CRISPR/Cas9 gene editing7. ESCs were cultured in LIF/serum medium [containing Dulbecco’s modified Eagle’s medium (DMEM) (Gibco), 1000 U/ml mouse LIF (Gemini), 15% Stem Cell Qualified fetal bovine serum (FBS) (Gemini), 2 mM HEPES (Gibco), 1 mM sodium pyruvate (Gibco), 1× non-essential amino acid (NEAA) (Gibco), 2 mM l-glutamine (Gibco), 50 µM 2-mercaptoethanol (Thermo Fisher), and 1% penicillin–streptomycin (Gibco)] on 0.2% gelatin-coated dishes without feeders. Min6 cells were cultured in DMEM high glucose (Gibco) supplemented with 15% FBS (Gemini), 0.5% β-mercaptoethanol (99%) (ACROS), and 1% penicillin–streptomycin (Gibco). All the other human cancer cell lines were cultured in DMEM (Gibco) supplemented with 10% FBS (Gemini) and 1% penicillin–streptomycin (Gibco). Mouse embryonic fibroblast cultures were prepared from embryonic tissues at E12.5 according to standard protocol as described previously71. To prepare ex vivo cultures of E18.5 embryonic kidney and liver, tissues were isolated under aseptic condition, rinsed twice in phosphate-buffered saline (PBS), dissected into ~1–5 mm3 pieces, and incubated for 10–15 min with collagenase type IV (Stem Cell Technologies; 1 mg/ml) with occasional pipetting. Later, the cells were centrifuged and washed twice with cold PBS to remove the trace of enzymatic digestion. Finally, single cells were cultured on 6-well dishes and fed by ex vivo culture medium [1:1 DMEM/F12 (Gibco) and Neurobasal medium (Gibco) supplemented with 1× N2 and B27 supplements (Gibco), 10% FBS (Gemini), 1 mM sodium pyruvate (Gibco), 1× NEAA (Gibco), 2 mM l-glutamine (Gibco), 50 µM 2-mercaptoethanol (Thermo Fisher), and 1% penicillin–streptomycin (Gibco)]. Alkaline phosphatase staining of the mESCs was performed as previously described72.Transfection and ER-reporter assaysFor transient knockdown experiments, following ON-TARGET plus siRNAs (small interfering RNAs) (all Dharmacon) were used: control siRNA pool (D-001810-10), siTUT7 (L-056770-01), and siTUT4 pool (L-065226-00). ESCs were reverse transfected using siRNAs and Lipofectamine RNAiMax (Invitrogen) complexes prepared in Opti-MEM (Gibco) for 48–72 h. For ER-reporter translation assays, 1 μg GLuc reporter plasmid was transfected into 106 ESCs in 6-well dishes. At 12 h after the transfection, cells were washed twice, and supplemented with 1 ml fresh ESC medium. After 2–6 h, the conditioned media were collected and cells were washed twice with PBS and then lyzed in 1× Passive Lysis Buffer (Promega). Some cultures were alternatively treated with 2 µM thapsigargin or equal volume dimethyl sulfoxide (DMSO) (mock) and incubated for 2 h before analysis. Relative luciferase activities in both supernatants and lysates were analyzed using BioLux Gaussia Kit. As negative controls, psiCHECK-2 (Promega) or pRL-TK (Addgene) vectors were transfected and analyzed similarly to GLuc reporter. Values were normalized to total protein and/or DNA content of the cultures. DIS3L2-stable knockdown ESCs (shDIS3L2 line) were generated previously1. Full-length (WT) and truncated 7SL RNAs were amplified from mouse genomic DNA and cloned into pGEM-T easy (Promega) or U6 promoter-less pLKO.1 (Sigma) vectors (see Supplementary Table 1 for the oligos used for cloning) and sequenced. For rescue experiment, 1 µg of these plasmids were transfected in mESCs for 48 h, after which the cells were re-transfected with pSV40-GLuc ER-reporter plasmid. For the negative control transfection, empty pGEM-T easy vector was ligated, and transformed into competent cells, from which maxiprep (Qiagen) preparation of plasmid was performed. Western blotting analysis was performed as previously described9. Following antibodies were used: anti-FLAG (Sigma; 1:10,000 dilution); anti-DIS3L2 (Novusbio; 1:1000 dilution); anti-β-ACTIN (Abcam; 1:5000 dilution); anti-SRP68 (Proteintech Group; 1:1000 dilution); anti-RPL23a (Proteintech Group; 1:1000 dilution); anti-SIX2 (Abcam; 1:500 dilution); anti-PAX2 (a kind gift from Dr. Kreidberg; 1:1000 dilution); anti-FGF9 (Abcam; 1:1000 dilution).RNA extraction and qRT-PCRCells were washed twice with PBS, lyzed in Trizol (Ambion), and RNAs were chloroform–isopropanol extracted and washed twice with 70% ethanol. Two micrograms of RNA was treated with RQ1 DNase for 30 min at 37 °C. Using random hexamers (to analyze relative expression) or oligo-dA12 oligonucleotides (to measure relative uridylation), cDNAs were made with SuperScript III reverse transcriptase (Invitrogen) and RNaseOUT (Invitrogen). List of primers and oligos are provided in Supplementary Table 1. All the qRT-PCR experiments were normalized to β-Actin levels in the respective cDNA samples. Northern blotting of 7SL RNA was performed as previously described7.ESC differentiationESCs were maintained in LIF/2i culture condition before differentiation as previously described71. For spontaneous differentiation, EBs were formed in hanging drops using 500 ESCs for 2 days, and then transformed to polyHEMA-coated (Sigma) (20 mg/ml in 75% ethanol) dishes and maintained in suspension for indicated timepoints before analysis. Differentiation was performed using ES medium depleted of LIF and 2i, and instead, supplemented with 15% FBS (Gemini). For renal differentiation, a previously described method60 was used with slight modification: at the first step of differentiation (epiblast differentiation), 1000 ESCs were induced for 2 days to form aggregates in hanging drops containing HI medium [75% Ham’s F12 medium and 25% Iscove’s modified Dulbecco’s medium (both from Gibco)] and then aggregates were transferred to polyHEMA-coated dishes and further induced for renal differentiation as described before60. At 8.5 days after differentiation (at metanephric mesenchyme stage), cell aggregates were harvested for qRT-PCR or western blot analysis. Alternatively, cell aggregates were dissociated with type 1 collagenase 1.5 mg/ml (Worthington Biochemical) for 10 min at room temperature with gentle shaking. Dissociated cells were centrifuged, washed with PBS (Invitrogen), and resuspended in 500 μl PBS and analyzed by flow cytometry (BD FACSAria). Undifferentiated mESCs (not expressing GFP) were used as the negative control.Metabolic labelingEqual numbers of control and DIS3L2 knockout mESCs were plated overnight and then washed with PBS and incubated in methionine- and cysteine-free DMEM (Gibco) medium for 2 h. Some cultures were alternatively treated with 2 µM thapsigargin or equal volume of DMSO (mock). Cells were then incubated for 1 h after supplementation with [35S]-methionine ([35S]-Met; 100 mCi/ml; PerkinElmer), after which they were washed with PBS to eliminate free radiolabeled amino acids. Total protein lysates were collected and the concentration of the proteins was measured by using Bradford assay. To measure radiolabeled secreted proteins, cells were starved in methionine- and cysteine-free ESC medium (Gibco) for 1 h, incubated for 1 h with [35S]-Met containing ESC medium, washed twice with PBS, and then supplemented with fresh ESC medium (without additional [35S]-Met) and the medium was collected immediately (0 min, to measure background, i.e., free amino acids) or after 1, 4, or 8 h (Supplementary Fig. 2a). For the quantitation of [35S]-Met-labeled proteins in the lysates or in the supernatant medium, [35S]-Met-labeled proteins were subjected to liquid scintillation analysis.RNA immunoprecipitationmESCs were transfected with FLAG-WT DIS3L2, FLAG-mutant DIS3L21, or empty pFLAG-CMV2 (as mock) vectors. At 48 h after transfection, mESCs were ultraviolet crosslinked, lyzed, and then RNA IP was performed using anti-FLAG M2 Affinity Gel beads (Sigma) as previously described7, and co-precipitated RNAs were isolated, purified, and analyzed by qRT-PCR. Thirty micrograms of FLAG-SRP68 vector (OriGene Technologies, MR204949) was transfected into 107 ESCs in 15-cm dishes of DIS3L2 heterozygote (control) or knockout ESCs in triplicates using Lipofectamine 2000 (Invitrogen) overnight before changing the medium. At 48 h after transfection, cells were harvested without crosslinking and FLAG-SRP68 was precipitated using anti-FLAG M2 Affinity Gel beads (Sigma). Rabbit anti-RPL23a (Proteintech Group; 10 µg antibody per IP) or normal rabbit IgG (Cell signaling) and protein A agarose beads (Roche) were used for IP of ribosomes in control and DIS3L2 knockout ESCs. Co-precipitated RNAs were isolated, purified, and analyzed by qRT-PCR.cRACEInput and FLAG-mutant DIS3L2 IP RNA samples were circularized with 1 μl T4 RNA ligase I, 10 mM ATP, and 10% PEG 8000 in 1× T4 RNA ligase buffer for 2 h at 37 °C and then the ligase was inactivated, as previously described8. After DNase I treatment, circularized RNAs were reverse transcribed with 7SL RNA-specific reverse primer (Supplementary Table 1) and SuperScript III. cDNAs were amplified by divergent internal primers (Supplementary Table 1) and AccuPrime GC-rich DNA Polymerase (Invitrogen) to generate chimeric PCR products corresponding to 5′ and 3′ ends of 7SL RNA transcripts. PCR products were size selected on 2% agarose gel, purified, and used in library preparation for MiSeq analysis using TruSeq Stranded mRNA Sample Preparation Kits (Illumina). To evaluate uridylation in RIP sample, sequencing reads containing ≥5 consecutive uridines (UUUUU) were considered as uridylated reads, which distinguishes them from reads corresponding to genomically extended 7SL RNA species with four uridines (Fig. 3a, b).mRNA sequencingPolyA+ mRNAs were isolated at indicated timepoints from differentiated EBs and cDNA libraries were prepared using Illumina TruSeq® Standard Total RNA Sample Preparation Guide and sequenced through Illumina NextSeq 500 Sequencing pipeline. Sequencing reads were aligned to a reference genome built by Subread v 1.6.373 using ENSEMBLE mouse genome v88 (GRCm38), and then the transcripts were quantified by featureCounts v 1.6.373. The normalization was performed using edgeR74 employing counts per million normalization with R version 3.6 and RStudio v 1.2.1335. GraphPad Prism software was used for data presentation.Ribosome profilingRibosome profiling was performed in two biological replicates of ESC cultures using TruSeq® Ribo Profile (Mammalian) Kit from Illumina. After adapter trimming and quality filtering, the sequences were further filtered to exclude rRNA and tRNA reads by aligning to mouse rRNA and tRNA sequence, which were retrieved from Ensembl (Release 91) (https://useast.ensembl.org/Mus_musculus/Info/Index) and GtRNAdb databases75, respectively. The cleaned RPFs were aligned to the mouse canonical known gene models (UCSC, mm10) using Bowtie with a maximum one mismatch allowed76. For codon occupancy analysis, the A site were inferred using an offset of 15 nucleotides (nt) from 5′ end of 28–31 nt fragments, which were uniquely mapped and translated in the zero frame of CDS77. The codon occupancy of an A site was further normalized by its basal occupancy, which is the average codon occupancy among +1, +2, and +3 downstream of the A site77. The TE was calculated by dividing RPF abundancy on CDS by its mRNA abundancy. We used 2-fold change of TE as the threshold to define the differential translation genes78,79. For the metagene analysis, we divided each CDS region into 50 equal bins and counted the RPF occupancy in each bin, which were then normalized by total number of RPFs of respective gene.Measurement of cytosolic calciumCytosolic calcium measurement was performed by using FLUOFORTE Calcium Assay Kit (Enzo). Cells in the 96-well plates were stained with 100 µl Hanks’ balanced salt solution (HBSS) buffer containing FLUOFORTE-AM and 4′,6-diamidino-2-phenylindole (DAPI) at room temperature for 1 h. After washing and changing fresh HBSS buffer, basal cytosolic calcium level was determined by measuring the fluorescence emitted at 525 nm after the cells were excited at 490 nm and normalized by DAPI signal. Later, cells were stimulated by different chemicals (1 mM EGTA, 2 mM ATP, 2 µM thapsigargin, 100 µM IP3) and monitored the change of the fluorescence every 20–30 s for 5–8 min.Measurement of ER calciumER calcium measurement was performed using Mag-Fluo4 acetoxymethyl ester (Mag-Fluo4-AM, Molecular Probes). Cells in the 96-well plate were stained with 5 μM of Mag-Fluo4-AM and DAPI at 37 °C for 40 min. After washing with the buffer (125 mM KCl, 25 mM NaCl, 10 mM HEPES, and 0.1 mM MgCl2, pH 7.2), the plasma membrane was then selectively permeabilized in the same buffer with 0.01% digitonin for 2 min at 37 °C. Basal ER calcium level was determined by measuring the fluorescence emitted at 510 nm after the cells were excited at 488 nm and normalized by DAPI signal. Later, cells were stimulated by different chemicals (1 mM EGTA, 100 µM IP3) and monitored the change of the fluorescence every 20–30 s for 5–8 min to determine the calcium leakage and release from ER.Insulin secretionMin6 cells were cultured in 15% FBS in DMEM high glucose medium (Gibco) supplemented with 0.5% β-mercaptoethanol and 1% penicillin–streptomycin (Gibco) and transfected with indicated siRNAs for 72 h. Cell were then incubated with indicated concentrations of glucose or KCl, and then supernatant samples were harvested. Insulin levels in the supernatant samples were assessed by Ultra Sensitive Mouse Insulin ELISA Kit (Crystal Chem, #90080), according to the manufacturer’s instructions. Corresponding cell lysates were centrifuged at 12,000 × g for 10 min at 4 °C to clear the lysates. One hundred microliters of the cleared lysates were dried at room temperature and the pellets were dissolved in 100 µl dH2O and used for DNA concentration measurement. The rest of 400 µl cleared lysates were diluted 1:1000 and were used for insulin ELISA assay. Finally, secreted insulin values were normalized to the insulin contents of the lysates.StatisticsAll the experiments were performed more than three times. Quantitative data are presented as mean ± standard error of means or standard deviation. Student’s t tests were used to analyze the significance of difference between different samples.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary
Description of Additional Supplementary Files
Supplementary Data 1
Supplementary Data 2
Supplementary Data 3 | nature communications | [
"Article"
] | [
"Mechanisms of disease",
"Endoplasmic reticulum",
"Non-coding RNAs",
"Translation",
"Embryonic stem cells"
] | surveillance protein-coding messenger RNAs nonsense-mediated decay (NMD investigated little quality control non-coding RNAs DIS3L2-mediated decay (DMD) identified surveillance pathway for ncRNAs ribosomal transfer small nuclear microRNAs mitochondrial RNA processing aberrant ncRNAs oligouridylated by terminal uridyl transferases TUT4 TUT7 degraded by 3′–5′ exoribonuclease DIS3L2. DMD implicated in cellular processes miRNA biogenesis rRNA mRNA degradation cell gametogenesis24 alternative splicing22 DIS3L2 binds to degrades RNA species tagged by poly-uridine 3′ modification catalyzed by cytoplasmic TUTases1,5 evidence tag marks structured ncRNAs with imprecise 3′ ends for degradation impair folding formation of multi-subunit ribonucleoproteinenriched DIS3L2 7SL RNA endoplasmic mRNA mutations DIS3L2 associated with Perlman biological significance impaired DMD for aberrant ncRNAs 7SL RNA obscure pathological RNAs not DMD contribution to mRNA translation addressed.Eukaryotic ER translating ribosome machineries ER main organelle biogenesis folding post-translational modification sorting membrane secretory extracellular ER unique cell intracellular storage reservoir for calcium ions responds to environmental cues developmental signals stress sensing biogenesis secreted growth factors hormones membrane-localized signaling receptors metabolites ion channels rely on ER mRNA translation SRP-dependent recruitment of ribosome mRNAs to ER translocons first step towards final destination encoded SRP conserved ribonucleoprotein complex RNA polymerase III-encoded 7SL RNA six protein subunits SRPs 72 68 54 19, 14 9 disruption SRP complex ER-associated mRNA translation secretory protein sorting39 intact SRP complex for normal secretory membrane proteinsER-targeted mRNA translation starts cytosolic ribosomes mRNAs translation signal peptide peptide recognition binding by SRP essential for stall recruitment mRNA to ER membrane Perturbation SRP abrogates mRNA translation protein sorting increased calcium leakage from ER study key role DMD quality control 7SL RNA DIS3L2 aberrant uridylated 7SL RNA inhibits SRP defective translation secreted transmembrane proteins ER calcium homeostasis embryonic stem cell (ESC) differentiation renal lineage perturbed renal abnormalities in Perlman syndrome required for ER-targeted mRNA mRNA expression translation efficiency in DIS3L2 knockout mouse ESCs ribosome profiling DIS3L2 loss affect global mRNA expression levels altered translation mRNAs detected by changes abundance ribosome-protected fragments TEs mRNAs changed 2-fold in DIS3L2 knockout cells compared control ESCsanalysis downregulated mRNAs showed enrichment membrane- ER-localized transcripts translation for protein synthesis upregulated transcripts associated mitochondria (Fig. connection between DIS3L2 function mitochondrial physiology on role DIS3L2 in ER protein synthesis downregulated mRNAs observed decreased RPFs regions accumulation RPFs at 5′ end CDS signal peptides RPFs upregulated mRNAs increased throughout CDS tools more transmembrane signal peptide-containing proteins detected among transcripts with downregulated translation in knockout cells results suggest in DIS3L2 knockout cells translation ER-associated mRNAs stalled at sequences signal peptide.Fig. 1Dysregulated ER-targeted mRNA translation upon DIS3L2 loss Scatter plot mRNA expression translation efficiency in DIS3L2 knockout cells heterozygote cells Translationally up- downregulated genes marked in red green Scatter plot perturbation translation efficiency upon DIS3L2 loss P values from two-sided Wald tests adjustments for multiple comparisonsGene ontology analysis dysregulated mRNAs DIS3L2 mESCs distribution RPFs downregulated mRNAs over-representation transmembrane signal sequences transcripts downregulated mRNAs values Fisher’s tests).Metabolic labeling synthesized proteins reduced levels secreted proteins culture media not cell lysates DIS3L2 knockout cells. inhibition sarco/ER Ca2+ ATPase thapsigargin suppressed protein synthesis DIS3L2 cells suggesting compromised ER-mediated calcium homeostasis DIS3L2 depletion ER-targeted mRNA translation used luciferase reporter) secreted accumulates culture Secreted luciferase produced lower level in DIS3L2-depleted mESCs Re-expression DIS3L2 not catalytic mutant protein levels secreted luciferase cells requirement for DIS3L2 expression secreted luciferase reporter detected in human cell lines (Fig. investigate signal peptide sequence requirement defective ER-targeted translation DIS3L2 engineered dual luciferase reporters Firefly Renilla luciferases without ER-specific signal peptide ER-retention signalAssessing luciferase activities parental in lysates supernatant samples control DIS3L2 knockout mESCs showed defective protein translation in DIS3L2 cells depends on signal peptide-coding sequence at 5′ end luciferase reporter removal KDEL signal from 3′ end luciferase reporters to supernatant diminishes detection in lysates underscoring robustness reporter systems (Fig. 2f). no tangible changes observed in mRNA expression of transfected luciferase reporters in DIS3L2 knockout cells Fig 2b–e). results highlight requirement DIS3L2 for normal ER-targeted translation mRNAs proteins.Fig. 2DIS3L2 required for ER-targeted mRNA translation Quantification of de novo protein synthesis in cell supernatants lysate samples metabolic labeling Bars mean P values de novo protein synthesis after TG treatment Normalized GLuc luciferase activity in supernatant lysates Left panel Western blot right panel normalized activity of secreted GLuc reporter after over-expression DIS3L2 proteinsBars mean SEM P values Student’s t test 0.95 confidence intervals 6 experiments Upper panel Western blot analysis DIS3L2 expression human cell lines lower panel normalized activities ER (GLuc cytosolic (Renilla luciferases after DIS3L2 knockdown 4 experiments (293T 5 (HCT116 cells cells Minimum maximum median boxes 25th to 75th percentiles represented P values (unpaired Student’s t test 0.95 confidence intervals Boxes mark minimum maximum values Upper panel dual luciferase reporters psiCHECK-2 5′-end signal peptide/3′-end KDEL-tagged 5′-end peptide lower panels normalized activity luciferase reporter supernatants lysate samples control DIS3L2 knockout mESCs Bars mean ± SEM P values (unpaired-tailed Student’s t test 0.95 confidence intervals 3 experiments indicated Source data.DIS3L2-mediated quality control 7SL ncRNADMD structured ncRNAs4,7,10 DIS3L2 implicated no change steady-state rRNAs levels rRNA unlikely account defect ER-targeted translationexplored defects tRNAs defective translation DIS3L2-deficient cells Ribo-seq data CLIP-seq results DIS3L2-targeted analysis codon usage frequencies in DIS3L2 mESCs no codon exclusion preference misprocessed tRNAs unlikely impaired ER-targeted translation DIS3L2 caused by defective tRNA incorporation involvement 7SL RNA ER-targeted defective ER translation due DIS3L2 exoribonuclease quality control 7SL RNA Rapid amplification of DNA) ends RNAs 3′ end 7SL RNA in DIS3L2-depleted cells extensive 3′-end uridylation in DIS3L2-bound transcripts truncated 7SL RNAs Logistic regression analysis positive correlation between 7SL RNA 3′-end truncation probability 3′-end uridylation quality control function of DIS3L2 for truncated 7SL RNAs 3DIS3L2-targeted quality control of 7SL cRACE analysis DIS3L2-bound 7SL RNAs cells extensive 3′-end uridylationcRACE analysis DIS3L2-bound 7SL RNAs canonical 3′ end reads end before end Nucleotides mature 7SL RNA extensions transcribed uridine marked magenta excluded Logistic regression analysis U-tailed occurrence 7SL RNA truncation extension Co-precipitation 7SL RNA ectopically FLAG-SRP68 protein analyzed qRT-PCR Western blot Equal enrichment 7SL RNA FLAG-SRP68 precipitates knockout control samples relative uridylation 7SL RNA samples Bars mean ± SEM P values = 4 Enrichment uridylated 7SL RNA RPL23a-IP samples DIS3L2 knockout samples Western blot qRT-PCR mean ± SEM P values Northern blot analysis 7SL RNA expression samples control DIS3L2 knockout cells Slow migrating uridylated 7SL RNA species marked repeated thrice similar results qRT-PCR analysis accumulation uridylated 7SL RNA subpolysome polysome fraction DIS3L2 knockout cells Bars mean ± SEM P valuesNormalized activity GLuc Renilla reporters DIS3L2 knockout mESCs effect TUTases depletion Bars mean ± SEM P values test 0.95 confidence intervals three KO DIS3L2 knockout mESCs CTRL heterozygous control mESCs Normalized secreted luciferase activity forced expression full-length truncated 7SL RNAs 6 Bars mean ± SEM P values test 0.95 confidence intervals Source data FLAG-SRP68 protein PCR uridylated 7SL RNA associated SRP ribosomal protein RNA accumulates ribosomes knockout mESCs Northern blot analysis accumulation uridylated 7SL RNA DIS3L2-deficient mESCs sucrose gradient centrifugation qRT-PCR presence uridylated 7SL RNA subpolysome polysomes DIS3L2-deficient ESCs defective ER-targeted translation DIS3L2 knockout cells rescued knockdown TUTases 7SL RNA uridylation. 3h forced expression truncated 7SL RNA. 3iinhibited ER translation luciferase mRNA over-expression 7SL RNA defective ER translation DIS3L2 cells supports DMD pathway surveillance 7SL RNA ER-targeted mRNA translation.Impaired intracellular calcium storage insulin release DIS3L2 defective ER translation decreased calcium levels ER DIS3L2 cells increased calcium ion (Ca2+) concentration cytosol dysregulation calcium homeostasis resulted increased calcium leakage ER Inhibition translation anisomycin treatment translocon Ca2+-impermeable anisomycin treatment alleviated ER calcium leakage DIS3L2 cells Stimulation ER calcium release supports defective ER calcium mobilization DIS3L2-depleted cells ER-targeted translation function interconnected DIS3L2 safeguards ER-targeted translation depletion defective/delayed ER translation calcium leakage imbalanced intracellular localization calcium 4Impaired intracellular calcium storage DIS3L2 lossLeft panel increased ER calcium leakage membrane translocon translation inhibition DIS3L2 Basal free ER cytosolic Ca2+ levels right panels = 12 Bars mean ± SEM P values t test 0.95 confidence intervals ER calcium leakage mESCs pretreated anisomycin ≥ overlapping values control cells treated anisomycin Normalized free cytosolic Ca2+ treatment thapsigargin ATP ER calcium release mESCs IP3 stimulation 10 Red arrows chemical treatments Normalized luciferase activity Min6 cells four insulin secretion levels Min6 cells glucose potassium stimulations 6 Basal free ER cytosolic Ca2+ levels Min6 cells transfected siRNAs 8 ER calcium leakage DIS3L2-depleted Min6 cells 10 red time EGTA treatment Cytosolic calcium levels high glucose stimulation siDIS3L2-transfected Min6 cells 10 glucose Bars represent mean ± SEM P values Student’s t test 0.95 confidence Source data.Insulin secretion complex calcium intracellular concentrationDIS3L2 loss hormone secretion used Min6 cells pancreatic β-cells55 cells sense increased glucose metabolize leads increased ATP/ADP potassium channel causes cell membrane depolarization calcium ion influx insulin secretion DIS3L2 depletion ER-targeted translation glucose-stimulated insulin secretion attenuated in Min6 cells bypassing glucose-sensing steps KCl administration DIS3L2-depleted cells failed secrete insulin DIS3L2 knockdown caused ER calcium depletion increased cytosolic calcium calcium leakage from ER defect regulation calcium homeostasis DIS3L2-depleted cells failed increase cytosolic calcium level glucose stimulation requirement DIS3L2 for endogenous protein secretion calcium leakage.DIS3L2 required for embryonic stem cell differentiationRegulation calcium signaling critical stem cell differentiation organ development perturbation leads malignancy analysis DIS3L2 knockout animal models differentiation failure renal lineage molecular mechanism obscureDIS3L2 loss calcium homeostasis stem cells renal differentiation utilized spontaneous lineage-directed in vitro differentiation DIS3L2-depleted mESC line 6a renal-specific Osr1-GFP stable DIS3L2 knockdown short hairpin RNA TUTase-dependent downregulation ER-targeted translation reporter DIS3L2 knockdown no effect pluripotency mESCs evident normal proliferation pluripotency marker expression clonogenicity control knockdown ESCs DIS3L2 deficiency larger embryoid bodies spontaneous ESC differentiation High-throughput sequencing PolyA+ hierarchical clustering revealed changes mRNA expression DIS3L2-deficient cells later stages EB differentiation 5c–e defective ESC differentiation dissociation replating cells EBs day 12 leukemia inhibitory factor) serum generated ESC colonies DIS3L2-depleted cultures not control knockdown ESCs analysis differentially expressed genes d10 d12 EB differentiation marked enrichment molecular functions calcium ion binding biological processes extracellular matrix organizationDEGs at d12 associated with appendicular skeleton glomerular disease abortion anterior fontanel” Perlman syndrome related to calcium homeostasis data suggest ESC pluripotency unaffected DIS3L2 deficiency causes late differentiation defects calcium homeostasis bone renal development Perlman syndrome. 5Impaired vitro differentiation DIS3L2 loss size of EBs during Minimum maximum median boxes 25th to 75th percentiles 30 EBs images of EBs at d12 repeated twice results number of differentially expressed genes during differentiation Hierarchical clustering PCA analysis mRNA sequencing data Phase contrast AP staining of d12 differentiated cells replated 5 days LIF ESC colonization repeated twice Heatmap of differentially expressed genes at d12 differentiation Gene ontology analysis of DEGs d12 DEGs.DIS3L2 loss perturbs calcium signaling pathways renal differentiationTime analysis gene expression differentiation increased expression of Ca2+ signaling target/sensor genes in DIS3L2 knockdown cellspolycystic kidney disease proteins (Pkd1 calcium/calmodulin-dependent protein kinases Regucalcin not randomly selected. 6a 3) elevated expression calcium signaling response perturbed calcium homeostasis DIS3L2-depleted cells 4) shDIS3L2 cells expression renal transcription factors dysregulated ESC differentiation. 6b Data 2 renal differentiation DIS3L2-deficient cells DIS3L2 knockdown renal differentiation increased renal progenitor cells increased Osr1-GFP-expressing cells. 6d increased expression renal-specific markers metanephric mesenchyme progenitor 6f observations kidney—overgrowth developmental defects Perlman syndrome patients dysregulated 7SL RNA decay DIS3L2-depleted cells over-expressed truncated 7SL RNA DIS3L2 induced in vitro renal differentiation-PCR analysis renal differentiation showed similar DIS3L2 cells cells express higher renal progenitor markers over-expression truncated 7SL RNA worsened defect elevated expression Six2 IGF2 calcium-sensing protein Camk1g (Fig. highlights function DIS3L2 renal differentiation connects role truncated 7SL RNA defective differentiation DIS3L2-deficient cells DIS3L2-depleted ESCs behave normally manifest differentiation defects show overgrowth renal lineage differentiation consistent with organomegaly kidney hypoplasia Perlman syndrome patients likely due cell-autonomous abnormalities recapitulated ex vivo.Fig. 6Altered calcium-sensing renal differentiation propensity DIS3L2 loss expressed genes calcium protocol in vitro differentiation mESCs renal progenitors Phase-contrast fluorescent images Osr1-GFPpositive renal progenitors d8.5 renal differentiation larger size differentiation increased intensity Osr1-GFP signal shDIS3L2 culture experiment repeated similar results Scale bar = 100 μmLeft panels Flow cytometry analysis Osr1-GFPpositive cells samples d right panel d8.5 renal differentiation Bars mean ± SEM P values t test 0.95 confidence intervals n 3 experiments Western blotting analysis samples d metanephric mesenchyme stage days after differentiation images two experiments similar results qRT-PCR analysis renal-specific markers day differentiation metanephric mesenchyme Bars mean ± SEM P values-tailed test 0.95 confidence intervals n 3 qRT-PCR analysis mRNAs day 8.5 differentiation Bars mean ± SEM P values test 0.95 confidence intervals n 3 Control DIS3L2 knockout cells transduced pLKO.1 empty vector truncated 7SL RNA endogenous promoter model biological function DIS3L2 ER-targeted translation calcium homeostasis stem cell differentiation Source data cellular models physiological requirement DMD molecular mechanism DIS3L2 aberrant 7SL ncRNA SRP safeguards ER-targeted mRNA translation membrane secretory proteinsDIS3L2 cellular calcium homeostasis ER function impaired ER mRNA translation calcium homeostasis critical for function organs communication pancreas enzymes nervous system ion channels synaptic vesicles muscles calcium release ER kidney ion channels metabolite re-absorption most affected organs in Perlman syndrome patients defective DMD study ER dysfunction in patient samples DIS3L2 animal models relevance defective DMD 7SL RNA tissue function homeostasis21 DIS3L2 functions truncated 7SL RNA other DIS3L2 targets involved in mRNA translation tRNAs2 rRNAs9,13 ribosome footprinting results requirement for DIS3L2 ER-targeted translation elimination aberrant 7SL RNAs Re-introducing 7SL RNA DIS3L2 rescued ER-targeted translation codon usage differences DIS3L2 loss defective ER translation not caused by aberrant rRNAs tRNAs absence DIS3L2 pathological truncated 7SL RNA incorporates SRP particles associates ribosomes affects homeostatic translation proteins secretory proteinsCryo-electron microscopy biochemical approaches chain exit site SRP contacts RPL23 ribosomal subunit signal sequence40 SRP perturbation leads to ER-translation defects mistargeting ER proteins mitochondria39 translation stall at 5′ end mRNAs in DIS3L2 cells explains increased ER calcium leakage elevated cytosolic calcium ion levels DIS3L2 7SL RNA small 7SL RNA pool cells truncated uridylated how impact ER-targeted translation why proportion mRNAs ER affected by DIS3L2 loss Truncated uridylated 7SL RNA species altered affinity to protein subunits SRP complex effect on SRP function mRNA ribosome ER SRP complex turnover robustness mRNA substrate recognition not all ER-translated mRNAs are SRP clients ER-targeted translation SRP DIS3L2-deficient cells subset ER-associated mRNAs affected elevated translation of mitochondrial transcripts upon DIS3L2 loss pathological roles of other DIS3L2 targets shouldconnection between intracellular calcium homeostasis cell proliferation differentiation studied lateral mesoderm retinoic acid induces translocation calcium channel TRPP2 from ER to cell membrane potentiates calcium influx to cytosol activation renal progenitor factor Pax868–70 regulation calcium signaling critical for renal lineage development Ca2+ stimulates mesoderm renal altered differentiation propensity of DIS3L2-depleted cells by elevated Ca2+ levels impaired storage mechanisms regulation calcium homeostasis critical for exit from pluripotency LIF/2i-adapted DIS3L2-deficient ESCs undifferentiated competence differential stimuli enhanced renal lineage differentiation leads to kidney overgrowth Perlman syndrome relevance of DIS3L2 loss-induced defects stem cells for in vitro disease modeling defective SRP complex function DIS3L2 deficiency for severe developmental phenotypes Perlman perinatal lethality in DIS3L2 knockout mice26 propose compromised DMD leads to “SRPopathy” defective ER-targeted mRNA translation (Fig. Further investigations required SRPopathy phenotypes Perlman syndrome patientsmESCs cultured without feeder 0.2% gelatin DIS3L2 ES cells generated CRISPR/Cas9 gene cultured LIF/serum medium medium 1000 U/ml mouse LIF 15% fetal bovine serum 2 mM HEPES 1 mM sodium pyruvate non amino acid 2 mM l 50 μM 2-mercaptoethanol 1% penicillin–streptomycin 0.2% gelatin-coated dishes without feeders Min6 cells cultured DMEM high glucose 15% FBS 0.5% β-mercaptoethanol 1% penicillin–streptomycin other human cancer cell lines cultured DMEM 10% FBS 1% penicillin–streptomycin embryonic fibroblast cultures embryonic tissues E12.5 vivo cultures E18.5 embryonic kidney tissues isolated rinsed-buffered saline dissected mm3 pieces incubated 10–15 min collagenase type IV cells centrifuged washed cold PBS enzymatic digestionsingle cells cultured 6-well dishes fed ex vivo medium [1:1 DMEM/F12 Neurobasal supplemented N2 B27 supplements 10% FBS 1 mM sodium pyruvate NEAA 2 mM l-glutamine 50 μM 2-mercaptoethanol 1% penicillin–streptomycin Alkaline staining mESCs performed ER-reporter ON-TARGET siRNAs control siRNA siTUT7 ESCs reverse transfected Lipofectamine RNAiMax complexes Opti-MEM 48–72 h 1 μg GLuc reporter plasmid transfected 106 ESCs 6-well dishes 12 h cells washed supplemented 1 ml fresh ESC medium After 2–6 h conditioned media collected cells washed lyzed 1× Passive Lysis Buffer cultures treated 2 μM thapsigargin dimethyl sulfoxide incubated 2 h before analysis Relative luciferase activities supernatants lysates analyzed BioLux Gaussia Kit controls psiCHECK-2 pRL-TK vectors transfected analyzed Values normalized protein DNA contentDIS3L2-stable knockdown ESCs generated Full-length truncated 7SL RNAs amplified mouse DNA cloned into pGEM-T (Promega U6 promoter-less pLKO.1 (Sigma vectors Supplementary Table 1 sequenced rescue experiment 1 μg plasmids transfected mESCs 48 h cells re-transfected with pSV40-GLuc ER-reporter plasmid negative control transfection empty pGEM-T easy vector ligated transformed into cells maxiprep preparation plasmid Western blotting analysis performed antibodies used anti-FLAG anti-DIS3L2 anti-β-ACTIN anti-SRP68-RPL23a anti-SIX2 anti-PAX2 anti-FGF9 extraction qRT-PCRCells washed PBS lyzed Trizol RNAs chloroform–isopropanol extracted washed 70% ethanol Two micrograms RNA treated RQ1 DNase 30 min at 37 °Crandom hexamers relative expression oligo-dA12 oligonucleotides cDNAs made with SuperScript III reverse transcriptase RNaseOUT primers oligos Supplementary Table 1. qRT-PCR experiments normalized to β-Actin levels cDNA Northern blotting of 7SL RNA performed differentiationESCs maintained in LIF/2i culture condition before differentiation spontaneous differentiation EBs formed in hanging drops 500 ESCs 2 days transformed to polyHEMA-coated 75% ethanol dishes maintained suspension analysis ES medium depleted LIF 2i supplemented with 15% FBS renal differentiation 1000 ESCs induced 2 days form aggregates drops 25% transferred to polyHEMA-coated dishes induced for renal differentiation 8.5 days after differentiation cell aggregates harvested for qRT-PCR western blot analysis dissociated with type 1 collagenase 1.5 mg/ml 10 min Dissociated cells centrifuged washed with PBS resuspended in 500 μl PBS analyzed by flow cytometry Undifferentiated mESCs GFP negative controlMetabolic labelingEqual control DIS3L2 knockout mESCs plated overnight washed PBS incubated methionine- cysteine-free DMEM medium 2 h treated 2 μM thapsigargin DMSO Cells incubated 1 h after supplementation [35S]-methionine/ml washed PBS eliminate free radiolabeled amino acids protein lysates collected concentration measured Bradford assay radiolabeled proteins cells starved methionine- cysteine-free ESC medium 1 h incubated 1 h [35S]-Met washed PBS supplemented fresh ESC]-Met medium collected immediately after 1 4 8 h [35S]-Met-labeled proteins liquid scintillation analysis.RNA immunoprecipitationmESCs transfected FLAG-WT DIS3L2 FLAG-mutant DIS3L21 empty pFLAG-CMV2 vectors 48 h after transfection ultraviolet crosslinked lyzed RNA IP anti-FLAG M2 Affinity Gel beads co-precipitated isolated purified analyzed qRT-PCRThirty micrograms FLAG-SRP68 vector Technologies transfected 107 ESCs 15-cm DIS3L2 Lipofectamine 2000 overnight 48 h cells harvested FLAG-SRP68 precipitated anti-FLAG M2 Affinity Gel beads Rabbit anti-RPL23a μg rabbit IgG protein A agarose beads ribosomes Co-precipitated RNAs isolated purified analyzed qRT-PCR FLAG-mutant DIS3L2 RNA samples circularized 1 μl T4 RNA ligase I 10 mM ATP 10% PEG 8000 T4 RNA ligase buffer 2 h 37 °C inactivated RNAs reverse transcribed 7SL RNA reverse primer SuperScript III divergent internal primers AccuPrime GC-rich DNA Polymerase chimeric PCR products 5′ 3′ ends 7SL RNA transcriptsPCR products 2% agarose gel purified used library preparation MiSeq analysis TruSeq Stranded mRNA Sample Preparation Kits uridylation RIP sample ≥5 consecutive uridines uridylated 7SL RNA species four uridines (Fig. 3a sequencingPolyA+ mRNAs isolated EBs cDNA libraries TruSeq® Standard Total RNA Sample Preparation Guide sequenced NextSeq 500 Sequencing pipeline reads aligned reference genome Subread v 1.6.373 ENSEMBLE mouse genome v88 transcripts quantified featureCounts v 1.6.373 normalization edgeR74 million R version 3.6 RStudio v 1.2.1335 GraphPad Prism software data presentation.Ribosome two replicates ESC cultures TruSeq® Ribo Profile (Mammalian) Kit sequences filtered exclude rRNA tRNA reads mouse rRNA tRNA sequence retrieved Ensembl cleaned RPFs aligned mouse gene models Bowtie maximum one mismatch codon occupancy analysis A site inferred 15 nucleotides 28–31 nt fragments mapped translated zero frame CDS77codon occupancy A site normalized by basal occupancy average +1 +2 +3 downstream TE calculated RPF abundancy CDS by mRNA abundancy used 2-fold change TE threshold differential translation analysis divided CDS region into 50 bins counted RPF occupancy normalized by RPFs gene cytosolic FLUOFORTE Calcium Assay Kit Cells 96-well plates stained 100 μl Hanks’ buffer FLUOFORTE-AM-phenylindole room temperature 1 h After basal cytosolic calcium level determined fluorescence 525 nm nm normalized DAPI signal cells stimulated by chemicals (1 EGTA 2 ATP 2 thapsigargin 100 μM IP3) monitored fluorescence every 20–30 s 5–8 min ER Mag-Fluo4 acetoxymethyl ester Cells 96-well stained with 5 μM Mag-Fluo4-AM DAPI 37 °C 40 min buffer KCl 25 NaCl 10 HEPES 0.1 MgCl2 plasma membrane permeabilized buffer with 0.01% digitonin 2 min at 37 °CER calcium level determined fluorescence 510 nm cells nm normalized by DAPI signal cells stimulated by chemicals (1 mM EGTA 100 μM IP3) monitored fluorescence every 20–30 s 5–8 min calcium leakage release ER.Insulin secretionMin6 cells cultured 15% FBS DMEM high glucose medium 0.5% β-mercaptoethanol 1% penicillin–streptomycin transfected siRNAs 72 h incubated with glucose KCl supernatant samples harvested Insulin levels assessed by Ultra Sensitive Mouse Insulin ELISA Kit cell lysates centrifuged at 12,000 × g 10 min 4 °C hundred microliters lysates dried room temperature dissolved in 100 μl dH2O for DNA concentration measurement 400 μl lysates diluted 1:1000 for insulin ELISA assay secreted insulin values normalized to contents experiments performed three times data presented mean standard error standard deviation Student’s t tests difference samples Nature Research Reporting Summary.Supplementary information | 48.4 | 0.894075 |
10.1038/s41467-021-20921-1 | PMC7846787 | Lima bean is an important crop for improving food security in Latin America and elsewhere. Here, the authors assemble its genome, conduct population genomics analysis using genotyping-by-sequencing data, and identify differentially expressed genes between two pod developmental stages. | Lima bean (Phaseolus lunatus L.), one of the five domesticated Phaseolus bean crops, shows a wide range of ecological adaptations along its distribution range from Mexico to Argentina. These adaptations make it a promising crop for improving food security under predicted scenarios of climate change in Latin America and elsewhere. In this work, we combine long and short read sequencing technologies with a dense genetic map from a biparental population to obtain the chromosome-level genome assembly for Lima bean. Annotation of 28,326 gene models show high diversity among 1917 genes with conserved domains related to disease resistance. Structural comparison across 22,180 orthologs with common bean reveals high genome synteny and five large intrachromosomal rearrangements. Population genomic analyses show that wild Lima bean is organized into six clusters with mostly non-overlapping distributions and that Mesomerican landraces can be further subdivided into three subclusters. RNA-seq data reveal 4275 differentially expressed genes, which can be related to pod dehiscence and seed development. We expect the resources presented here to serve as a solid basis to achieve a comprehensive view of the degree of convergent evolution of Phaseolus species under domestication and provide tools and information for breeding for climate change resiliency. | IntroductionThe Phaseolus genus comprises more than 70 species, of which five have been domesticated, namely P. acutifolius A. Gray (tepary bean), P. coccineus L. (ayocote or runner bean), P. dumosus Macfady (num, piloy, or year bean), P. lunatus L. (Lima bean), and P. vulgaris L. (common bean)1. Lima bean and common bean are the two agronomically and economically most significant species within the Phaseolus genus2–4. Lima bean provides a vital source of nutrients globally; its seeds contain at least 20% protein, more than 50% carbohydrates; it is a rich source of amino acids such as tryptophan, lysine, methionine, phenylalanine, threonine, valine, isoleucine, and leucine5,6.Wild and domesticated Lima beans are found in a wide variety of climatic conditions along with their range of distribution from northern Mexico to northern Argentina7–9. Wild Lima beans are structured into three major gene pools according to genomic data10: two Mesoamerican (MI and MII) and one Andean (AI) gene pool. While MI occurs mainly in central-western Mexico, MII is more widely distributed from southern Mexico and Central America to tropical South America. The Andean gene pool AI is restricted to southern Ecuador and northern Peru, where this species apparently originated11. The possible existence of another Andean gene pool, the AII gene pool restricted to the Andes in central Colombia has been proposed, although this requires further confirmation. At least two domestication processes took place in Lima bean, one in Mesoamerica and one in the Andes10,12–17. The Andean domestication occurred from gene pool AI and gave rise to the Andean varieties characterized by large and flat seeds (Big Lima cultivars). The second event occurred in central-western Mexico from gene pool MI and gave rise to the Mesoamerican varieties characterized by having small rounded or oval-shaped seeds (Potato and Sieva cultivars, respectively)18,19. Lima bean is a good example of convergent evolution since both Mesoamerican and Andean landraces evolved similar traits under domestication, mainly larger pods and seeds, reduction or loss of pod dehiscence, loss of seed dormancy, determinate growth habit and reduced content of antinutritional seed compounds, among others.Lima bean is a species of great interest for evolutionary research not only because it provides an opportunity to study the molecular basis of convergent phenotypic adaptation, but also because it shows adaptation to a wider range of ecological conditions compared to common bean9, especially to heat and drought stresses, traits that are key in scenarios of adaptation to climate change20. For these reasons, it is important to document the domestication process in Lima bean, since a good understanding of the phenotypic adaptations involved and their genetic control may lead to the identification of the related genes and alleles and efficient use of these alleles for future breeding activities. Lima bean genetic research has previously relied on the common bean reference genomes21–24. Lima and common bean are both diploid species (2n = 2x = 22 chromosomes; DNA: ~622 Mbp/1 C) with high levels of homozygosity throughout their genomes because of predominant autogamy25. Previous cytogenetic research confirms a high degree of synteny between the two species26. However, relying on a P. vulgaris genome alone may have consequences for downstream diversity analyses due to reference bias, produces loss of information and could even be misleading for predictions of genomic loci related to traits. The development of a whole-genome reference sequence for P. lunatus would provide a groundbreaking genetic resource for Lima bean research, while highlighting genomic differences among its domesticated relatives25–29.In this research, we describe a large collaborative effort to provide high-quality genomic resources for Lima bean genetics and breeding. These include a high-quality reference genome, gene expression information in different tissues, accessions and developmental stages, and the most comprehensive assessment of genomic variability in a sample of close to 500 wild and domesticated accessions. By combining approaches based on comparative genomics and population genetics, we reveal a complete view of the distribution of genomic variability across the species and its relationship with the common bean. Moreover, we provide gene functional annotations and genetic loci associations for different traits relevant to domestication processes and breeding in Lima beans.ResultsA chromosome-level high-quality assembly for Lima beanWe generated a chromosome-level assembly of the Lima bean genome from G27455, a domesticated accession from the Mesoamerican gene pool MI collected in northern Colombia. Data from the use of Pacific Biosciences and Illumina sequencing technologies and four experimental protocols, namely paired-end whole genome sequencing (WGS), 10x, genotyping-by-sequencing (GBS), and RNA sequencing, were combined to achieve the contiguity and base quality of this assembly. An initial backbone assembly of PacBio WGS reads polished using paired-end Illumina reads produced a draft assembly with 512 contigs adding up to 542 Mbp. A total of 206 of these contigs—totaling 512 Mbp—were further assembled in scaffolds, sorted, and oriented in 11 pseudomolecules based on an analysis of linked reads obtained from the G27455 accession following the 10x protocol and analysis of GBS data from the F8 generation of UC 92–UC Haskell recombinant inbred line population. A linkage map was developed for this population with 10,497 SNPs across 522 unique loci with an estimated genetic length of 1064 cM (Supplementary Fig. 1). Linkage groups were established for each of the 11 chromosomes (Supplementary Fig. 2 and Supplementary Table 1). This linkage map had an average genetic and physical spacing between loci of 2.18 cM and 1.10 Mbp, respectively. Genetic gaps larger than 20 cM were observed on three linkage groups: Pl01, Pl02, and Pl09, with 20.5, 24.4, and 32.8 cM gaps, respectively. Marker coverage varied across and within linkage groups with the densest marker coverage observed in the pericentromeric regions of Pl02, Pl05, Pl07, and Pl11, and the sparsest marker coverage observed in the pericentromeric regions of Pl01 and Pl09. Recombination rates varied within linkage groups, with the lowest rates of recombination in the centromeric and pericentromeric regions and the highest rates towards the telomeric ends (Fig. 1a and Supplementary Table 2). The Pl03 linkage group had the highest average recombination rate, with recombination events occurring every 662 kbp. The Pl10 linkage group had the lowest average recombination rate, with recombination events occurring on average every 2074 kbp, which may be influenced by the high degree of segregation distortion observed in the pericentromeric region towards the UC Haskell haplotype. The largest spans of the pericentromeric regions were on the Pl04, Pl10, and Pl11 linkage groups, and the shortest spans were on Pl03 and Pl06 (Fig. 1b and Supplementary Table 2). The Pl01 and Pl09 linkage groups had particularly sparse marker coverage across the pericentromeric regions, which likely reduced the accuracy of the recombination rates in these regions and the definition of the pericentromeric regions for these linkage groups.Fig. 1Chromosome-level genome assembly for Lima bean.a Genetic distance (cM) and recombination rate (cM/Mbp) by physical position (Mbp) on the Lima bean reference genome for the UC 92–UC Haskell RIL population. Chromosomes are labeled as Pl01-Pl11. b Chromosome lengths and pericentromeric regions. c Density of repetitive elements. d Density of gene models. e Density of SNPs. f–i. LOD scores of QTL for four different traits screened in the RIL population. Colored zones represent LOD scores greater than 3 for determinacy (green), flowering time (blue), hundred seed weight (red), and volatile cyanide (yellow). j Phenotypic distribution of traits in the RIL population with parental phenotypes represented by vertical lines. Source data are provided as a Source Data file.Base-pair-level quality was assessed by the mapping of Illumina reads to the assembly, which reached 99% of the raw reads after polishing. In addition, the identification of orthologs reached 98.8% of 1614 genes known to be conserved in a single copy across plant species (Supplementary Fig. 3). Annotation of repetitive elements was performed using repeat masker30 and based on a common bean library of 796 transposable elements31. A total of 656,928 events were identified covering 225 Mbp (41% of the assembly). More than half of these regions (174 Mbp) were covered by long terminal repeats (LTRs). Additional 8 Mbp were covered by other class I retrotransposons, namely LINE and SINE elements. DNA (Class II) transposons covered 25 Mbp of the assembly and other 6 Mbp was covered by other unclassified transposable elements (Supplementary Table 3). Repetitive elements are more abundant within pericentromeric regions of the genome (Fig. 1c).To perform automated structural and functional annotation of gene models for the Lima bean genome, repeat elements were masked and Illumina RNA-seq data from three tissues (leaf, flower, and pod), two developmental stages (initiation of pod elongation and before seed filling), and two accessions (G25230: wild from Manzanillo, Colima, Mexico; and G27455) were analyzed. In contrast to repetitive elements, gene density and observed number of SNPs is higher outside pericentromeric regions (Fig. 1d, e). By integrating publicly available RNA-seq data and gene models from the common bean genome, a total of 28,326 gene models and 35,881 transcripts were predicted with an average total length of 3.7 kbp, and an average protein length of 413 amino acids. Distributions of gene and protein length are consistent with the current gene annotation for P. vulgaris (Supplementary Fig. 4). Gene ontology functional annotations could be recognized for 21,642 (76%) of the gene models by ortholog identification from other plant species. Common ontology terms included response to stress, different metabolic processes, transport, anatomical structure development, signal transduction, cellular component assembly, and homeostasis (Supplementary Fig. 5). A total of 19,554 gene models were annotated with pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG). In total, functional annotations were assigned to 22,634 (80%) gene models. Gene expression was evidenced in at least one RNA-seq dataset for 26,295 (93%) gene models. Moreover, orthologs with P. vulgaris within synteny blocks were identified for 22,180 (78%) gene models (details in the next section). Considering only gene expression and orthology with P. vulgaris in synteny blocks, direct evidence could be identified for 27,029 (95%) of the annotated gene models. From the remaining gene models, 416 show at least 50% protein sequence similarity with P. vulgaris genes outside synteny blocks. The remaining 3% are either paralogs of genes with direct evidence or have functional annotations from other plant species.QTL mapping of agronomic-related traits in Lima beanQuantitative trait loci (QTL) were mapped in the biparental population UC Haskell–UC 92 used to build the genetic map for genome assembly (Fig. 1f–i and Supplementary Table 4). Determinacy and three quantitative traits (flowering time, FT; hundred-seed weight, HSW; and cyanide content) were screened in this population (Fig. 1j). Nine quantitative trait loci (QTL) were identified in the biparental population. One major QTL for determinacy was identified on the long arm of chromosome Pl01, explaining 78% of the phenotypic variation. The peak LOD score for determinacy was located at the first marker on the long arm, after a nearly 20 cM and 20 Mbp gap in the pericentromeric region of chromosome Pl01. A likely causative gene for this locus is an ortholog of the Arabidopsis gene TFL1. The common bean ortholog PvTFL1y was previously mapped at 45Mbp of chromosome Pv0132,33. We identified the ortholog PlTFL1 in Lima bean at 41Mbp of Pl01 (Supplementary Data 1). For flowering time, transgressive segregation was observed for both earlier and later progenies than the UC 92 and UC Haskell parents, respectively (Fig. 1j). A major QTL was also found on chromosome Pl01 and it is likely that the causative gene for this QTL is also PlTFL1. However, this QTL explained only 30% of the phenotypic variation, which suggests that other genes influence flowering time in this population.Regarding seed weight, transgressive segregation was observed for seed weights below the small-seeded parent UC Haskell, but not for larger seed weights. This observation is consistent with prior results showing a shift towards smaller-seeded segregants, observed in common bean34,35. Four additive minor QTL were identified collectively explaining 28% of the phenotypic variation, including one on the long arm of chromosome Pl10 that explained 11% of the phenotypic variation. Finally, a major QTL for cyanogenesis in floral bud tissue was found on the long arm of chromosome Pl05 and explained 93% of the phenotypic variation, and collectively with two other minor QTL explained 97.5% of the total phenotypic variation for cyanogenesis. The UC 92 variety did not show measurable cyanide content in contrast with UC Haskell. There was transgressive segregation for cyanide content above the levels observed in UC Haskell. The three QTL showed epistatic interactions: the UC 92 allele of the larger QTL on Pl05, causal of the absence of cyanogenic glucosidase, prevented the expression of the two QTLs on Pl08 and Pl10 (Supplementary Fig. 6). The significance interval of the QTL for cyanide content on Pl05 includes a sequence for a glucosidase with homology to a cyanogenic glucosidase in white clover36,37.Evolution of paralogs and orthologs and speciation eventsPredicted proteins for representative transcripts of all annotated genes were aligned to each other to build 3499 paralog clusters representing the gene families generated through different genome evolution processes. Classification of paralog relationships and interchromosomal synteny analysis revealed 1647 genes with paralogs generated by the ancient whole-genome duplication events documented in the history of Fabaceae38. Chromosome pairing inferred from these paralogs is consistent with that reported in the genome of P. vulgaris23 (Internal links in Fig. 1). Intrachromosomal duplication events were identified for a total of 7285 genes. Even after removing highly repeated genes (with more than ten paralogs), 5849 genes were involved in intrachromosomal duplication events. Figure 2a shows that the Ks values for these cases are significantly smaller than those of whole-genome duplication (WGD) paralogs (p-value < 10−15 for a Wilcoxon rank test), meaning that intrachromosomal duplications are more recent than WGD paralogs. Protein evolution among the two types of paralogs was assessed by calculating the Ka/Ks ratio between pairs of paralogs to identify patterns of selection. In contrast to Ks values, Ka/Ks values of intrachromosomal duplications are significantly larger than those of WGD paralogs (p-value < 10−15 for a Wilcoxon rank test), which means that these duplications are diverging faster than WGD paralogs (Fig. 2b). Moreover, 12% of the local duplications seem to experience a rapid sequence divergence showing Ka/Ks ratios above 1. Functional enrichment of 368 genes involved in recent duplications with high Ka/Ks ratios revealed a relatively large set of interconnected biological processes mostly related to immune response, including cell death, cell communication, and signaling (Supplementary Fig. 7). These findings are consistent with those reported by Qiao et al.38 for other plant species. Other processes enriched in evolving intrachromosomal duplications include lipid transport and metabolism of lignan.Fig. 2Comparative genomics between P. lunatus and P. vulgaris.a Ks and b Ka/Ks statistics for P. lunatus and G. max paralogs, as well as orthologs between P. lunatus and P. vulgaris and orthologs between P. lunatus and V. unguiculata. WGD: Whole Genome Duplication. Sample sizes (N) correspond to gene pairs. Middle lines are medians and box limits represent first (Q1) and third (Q3) quartiles. Lines are drawn from Q1 minus 1.5 of the interquartile range (IQR) to Q3 + 1.5*IQR. c Chromosome by chromosome synteny between P. lunatus and P. vulgaris for detailed visualization of structural rearrangements. P. lunatus chromosomes are labeled as Pl01-Pl11 and P. vulgaris chromosomes are labeled as Pv01-Pv11. d Number of homologs of resistance genes by chromosome. e NJ Radial tree diagram showing genetic variability among LRR type resistance genes. Light blue is proteins with domains NB-ARC and LRR, purple is proteins with domains TIR, NB-ARC, and LRR, orange is proteins with the TIR and LRR domains. Source data are provided as a Source Data file.We compared the genome of Lima bean assembled in this study with that of common bean v1.023 based on the identification of orthologs between the two species and synteny blocks. Orthologs could be identified for 25,564 (94%) of the P. vulgaris genes and 26,009 (92%) of the P. lunatus genes. As reported in previous studies26, a high collinearity was observed between the P. lunatus and P. vulgaris genomes (Fig. 2c). The most important structural events identified in this study are an inversion of the short arm of chromosome Pl10 and a large translocation of the pericentromeric region of Pv02 within the short arm of chromosome Pl02. Other large events include a 5Mbp inversion within the long arm of chromosome Pl03, a 10Mbp inversion within the long arm of chromosome Pl07, and a complex translocation within the short arm of chromosome Pl09. The rearrangements in Pl02 and Pl10 confirm previous works suggesting pericentromeric inversions in these chromosomes based on Fluorescence in situ hybridization (FISH) assays28. As seen in other species, some of these rearrangements could be related to the previously observed reproductive isolation between Lima bean and common bean39,40.Figure 2a shows that the Ks distribution of 22,180 orthologs between P. lunatus and P. vulgaris identified in synteny blocks is centered around 0.05. This is about half of the average obtained for paralogs generated by the recent (about 13 MYA) WGD event within the G. max genome41, suggesting that the speciation between P. vulgaris and P. lunatus occurred around 6 MYA. This date is close to the age of the Phaseolus crown clade B (that contains P. vulgaris and P. lunatus) estimated on the basis of evolutionary rates of the chloroplast trnK locus (5.0 ± 0.7 MYA) and older than the date estimated from ITS/5.8 S sequences (3.4 ± 0.4 MYA)2. Comparing the genomes of Lima bean and Vigna unguiculata42, the separation of Phaseolus from Vigna could be dated right before the WGD of G. max, around 15 MYA. This date is much older than that estimated by Delgado-Salinas et al.43 for the Vigna sensu lato crown clade (9.1 ± 1.0 MYA) from chloroplast trnK sequences. Protein evolution between orthologs was also assessed by calculating the Ka/Ks ratio, in this case, to identify patterns of selection after the speciation event separating P. vulgaris and P. lunatus. In line with previous studies in other species38, the distribution of Ka/Ks was centered far below 1, suggesting that most gene coding sequences evolve under purifying selection (Fig. 2b). Conversely, 656 gene pairs within the main synteny blocks showed rapid sequence divergence with Ka/Ks values larger than 1. Functional enrichment of these genes shows ontologies related to the metabolism of aminoglycan, chitin, and lignan (Supplementary Fig. 7). As described in the last section, genes related to these processes have increased expression values during the development of the pod.Orthologs of genes related to agronomic characteristicsGenes of agronomic interest were predicted by ortholog relationships of genes associated with agronomically relevant traits in other crops (Supplementary Data 1). We identified Lima bean orthologs of 30 genes having reported associations with traits in previous studies, 27 of which were reported in common bean. Traits included due to their importance in plant breeding are yield44, nutritional quality45, herbicide resistance46, plant architecture47, tolerance to abiotic stresses48, among others. Moreover, seed coat color and growth habit, important characteristics needed to meet consumer and farmer preferences in Lima bean and common bean breeding and marketing, were also included.In particular, 1917 genes distributed across the 11 chromosomes were related to resistance to biotic stresses predicted on the base of bioinformatics analysis and the presence of the LRR (leucine-rich repeat-containing) and other important domains for disease resistance such as toll/interleukin-1 receptor (TIR), leucine zipper (LZ), coiled-coil (CC), nucleotide-binding site (NBS/NB) shared by ARC (Apaf-1, R proteins, and CED-4) (NBS/NB-ARC) domain, serine–threonine kinase, and WRKY (Fig. 2d and Supplementary Fig. 8). Serine–threonine and other receptor-like protein kinases were one of the most abundant types among the selected disease resistance candidate genes (828). The WRKY domain was present in 91 genes and the leucine zipper domain in 74. The CC domain was only found in 11 genes. However, a low number of CNL (CC-NBS-LRR) genes has been previously observed and reported in dicots49. In contrast, 98 TNLs (TIR-NB-LRR) were identified in the annotated gene models. Furthermore, 631 genes contained the LRR domain, 151 the NB-ARC domain, and 91 both domains (Supplementary Data 2). Large numbers of predicted disease resistance genes were localized to chromosomes Pl02, Pl04, Pl08, Pl10, and Pl11.The subset of genes with the LRR domain tended to be clustered in discrete regions of the genome (Supplementary Fig. 9). This subset includes the following domain arrangements: LRR, NB-ARC-LRR, TIR-NB-ARC-LRR, and TIR-LRR. Neighbor joining clustering of this family showed some correspondence between chromosome clustering and sequence similarity (Fig. 2e). Proteins with the domains NB-ARC-LRR formed a large cluster, but five of them were nested into the TNL group that also contained the three proteins with the TIR and LRR domains only. Common bean orthologs were identified for most of the predicted genes related to biotic stress resistance. These orthologs were located on the same chromosomes and were mainly collinear. Also, motifs were mostly conserved between the gene sequences of both species. Furthermore, the genomic positions of the genes correspond with resistance loci associated by previous studies with some of the most important diseases affecting the common bean. For instance, we found clusters of genes within or close to resistance loci for anthracnose50–54, angular leaf spot51,52,55, halo blight52,56, bean golden yellow mosaic virus51; as well as other viral diseases, rust, and mildew23,52.Population structure analysis reveals genetic clusters in Lima beanWild Lima bean presumably originated in the northern Andes, during Pleistocene times, and from there it expanded to other areas in the Andes and Mesoamerica11. As a result, wild Lima bean reached a widespread distribution, from northern Mexico to northern Argentina, and became differentiated into three major gene pools (MI, MII, and AI) with mostly non-overlapping distributions, as documented by previous studies. Later, humans domesticated this species twice, once in Mesoamerica from gene pool MI and once in the Andes from gene pool AI10,15,16. To investigate in greater detail the genetic structure of Lima bean, we combined previously analyzed genotyping-by-sequencing (GBS) data from 270 Lima bean accessions10 with GBS data for 212 additional samples to increase the amount of variation (Supplementary Data 3). From a raw number of 116,030 biallelic SNVs, 12,398 were selected for diversity analysis (Supplementary Data 4).The samples sequenced in this study increased representation of the three major gene pools and the geographic sampling in countries such as the United States, Mexico, and Colombia (Supplementary Fig. 10). Different statistical and heuristic clustering analyses, including Neighbor Joining (NJ), discriminant analysis of principal components (DAPC), and Bayesian clustering (STRUCTURE) were applied. The optimal number of clusters according to the decrease of BIC is between K = 5 and K = 6, while STRUCTURE results suggest that the optimum K is 6 (Supplementary Fig. 11). The results not only supported the existence of the three major wild gene pools (MI, MII, and AI) but also showed evidence for two large novel clusters (Fig. 3a, Supplementary Data 5, and Supplementary Figs. 12–14). First, domesticated MI accessions were clearly separated from wild accessions in a single cluster, which is in agreement with a single domestication of Mesoamerican landraces (dark blue cluster in Fig. 3a from K = 5 onwards and Supplementary Fig. 12), and second, a more complete sampling of wild accessions in the central Andes of Colombia supported the presence and genetic differentiation of an Andean wild gene pool, the AII gene pool (yellow cluster in Fig. 2a, from K = 4 onwards and Supplementary Fig. 12). The wild AII group shows a very restricted distribution on the eastern slope of the Andes in central Colombia, specifically in the departments of Cundinamarca and Boyacá (although one accession occurs in Peru) (yellow dots in Supplementary Figs. 13, 14). It is worth noting that five wild accessions from Guatemala, one from Honduras, and one from Chiapas (Mexico) clustered within the Andean gene pool AI. When K = 6 is considered, a further subcluster is detected within wild gene pool MII (Fig. 3a. dark green cluster) that contains accessions from Peninsula of Yucatan in Mexico, northern Guatemala, Costa Rica, and northern Colombia.Fig. 3Lima bean genetic diversity.a STRUCTURE analysis of the genetic variability between 482 wild and domesticated Lima bean accessions collected across the Americas. Wild accessions are organized (from left to right) into a south-north geographic pattern. DOM domesticated, MEX Mexico, GUA Guatemala, CR Costa Rica, COL Colombia. Classification of accessions into gene pools MI, MII, AI, or AI is shown. b Linkage disequilibrium decay within different subgroups of wild and domesticated accessions. DOM domesticated, DOM-AND Andean landraces. c Radial clustering of the 482 accessions according to the analysis performed by fineSTRUCTURE. Major gene pools are shown by different colors (purple cluster: wild MI from northern-western Mexico (NORTH MEXICO); pink cluster: wild MI from southern-western Mexico (SOUTH MEXICO); medium blue cluster: domesticated MI from South America; dark blue cluster: domesticated MI from Mexico and Central America (MEXICO/CA); light blue cluster: domesticated MI from Yucatan Peninsula; yellow cluster: AII gene pool; red cluster: AI gene pool; green cluster: MII gene pool from Yucatan, Central America, and Colombia (YUCATAN COL-CR); light green cluster: MII gene pool from southern and central Mexico. d Distribution of chromosomal segments contributed by different gene pools in a set of 15 wild and domesticated accessions. Wild accessions are marked in bold. Lima bean chromosomes are labeled as Pl01-Pl11. Source data are provided as a Source Data file.Because Lima bean is an inbred species, heterozygosity rates per accession are generally low (details below), which facilitates the inference and analysis of haplotypes. Figure 3b shows the comparison of linkage disequilibrium (LD) decay between pairs of variants in the same chromosome. LD decays faster in wild accessions (dark gray line) than in domesticated accessions (light gray line). However, due to population structure, LD remains high (on average over 0.3 for the wild gene pool and over 0.4 for the domesticated gene pools) even at distances larger than 1 Mbp. LD analyses within wild and domesticated populations showed that in the wild MI and wild MII gene pools LD decays to basal levels at about 100 kbp, whereas in the domesticated MI, MII, and AI populations, it decays to background levels at about 500 kbp. LD decayed faster in MI and MII landraces than in the AI landraces, which agrees with the higher genetic diversity observed in the former (Supplementary Table 5).Taking advantage of the reconstructed haplotypes and relatively long blocks of linkage disequilibrium, we applied the software fineSTRUCTURE57 to exploit the information contained in genome-wide linkage disequilibrium patterns for dissecting fine population structure. This approach not only identified a much larger number of populations than STRUCTURE (in total 181) but also revealed the genetic relationships among the inferred populations (Supplementary Data 5). Examination of the radial tree shown in Fig. 3c, from the highest to the lowest level of clustering, allowed to view the whole population at multiple structuring levels. In general, major clusters in the tree (shown by different colors) corresponded to the major gene pools detected by STRUCTURE, DAPC, and NJ. Also, the subgroups observed within the wild MI, Dom MI, and wild MII gene pools are clearly related to geography, as described below (Supplementary Figs. 13, 14).Among the wild accessions, MI accessions were separated into two subgroups according to their geographical origin: one subgroup (purple cluster) included 55 accessions assigned to ten populations mainly distributed in northern-western Mexico, from the states of Sinaloa to Colima, and the other subgroup (pink cluster) included 34 accessions assigned to 13 populations mainly distributed in southern-western Mexico, from the state of Michoacan to Oaxaca. MII accessions were also separated into two subclusters: one of them (light green cluster) contained 66 accessions mainly distributed in Mexico and southern Guatemala, and the other one (dark green) contained 49 accessions from northern Guatemala, Costa Rica, northern Colombia, and nine accessions from the Peninsula of Yucatan. Among the domesticated accessions, MI landraces were separated into three subgroups according to their geographical origin: one subgroup (dark blue cluster) included 43 accessions assigned to 13 populations mainly distributed in Mexico and Central America, with only 12 accessions from other countries (United States, Colombia, and Ecuador), a second subgroup (medium blue cluster) included 51 accessions assigned to nine populations mainly distributed in South America, with only 8 accessions from the United States and one from Mexico, and the third subgroup (light blue cluster) contained 59 accessions assigned to five populations (numbers 19, 60, 80, 88, and 101. See Supplementary Data 5 and supplementary Fig. 15 for population numbers) distributed in the Yucatan Peninsula in Mexico. This third subgroup was also observed in the STRUCTURE results at K = 7 (Fig. 3a). By examining in more detail the landraces contained in each one of these five populations from the Yucatan Peninsula, there is some tendency to group accessions by variety or seed shape. According to the Mayan nomenclature registered by Martinez-Castillo et al.58, population 88 comprises 14 accessions that belong to four landrace varieties known as Bacalar-ib, Chak-saak, Chak-ib, and Bayo-ib, all of them with small and flattened or semi-flattened seeds (the typical morphotype of the Sieva cultigroup). Population 101 includes 24 accessions that belong to landrace varieties known as Putsica-Sutsuy, Mulicion blanco, Mulicion rojo, Box-ib, Pool-santo, Kan-ib, Morado-ib, Yete Boch ib, Kolbihi, and Chak ib, most of them characterized by having small rounded or semi-rounded seeds (the typical morphotype of the Potato cultigroup). Population 60 comprises 13 accessions that belong to varieties known as Sac-ib, Chak-chi, Bayo-ib, Mejen-ib, Bacalar, and Madza-kitam with seed morphology intermediate between both cultigroups. Population 80 contains four landrace accessions that belong to the variety known as Box-ib that carry small purple-black semi-flattened or rounded seeds. Population 19 contains 4 accessions of the landrace variety known as Chak-chi with white-red small seeds with seed morphology also intermediate. The intermediate forms may arise by the fact that some farmers in the Yucatan Peninsula may grow up to seven different landrace varieties together, which may promote opportunities for crossing58,59.Supplementary Table 5 shows basic diversity statistics for all major gene pools in wild and domesticated accessions. Observed heterozygosity was much lower than expected heterozygosity, as expected for an inbred species as Lima bean (Bartlett’s K-squared = 18257, df = 1, p value < 2.2e-16; paired-t test t = 165.28, df = 12397, p value < 2.2e-16). Among the wild clusters, the most diverse were the Mesoamerican MI and MII gene pools (HE = 0.128 and HE = 0.133, respectively), while the least diverse were the Andean AI (HE = 0.040) and AII gene pool (HE = 0.052). The genetic diversity of the seven wild accessions from Chiapas, Guatemala, and Honduras, clustered within the AI gene pool, was also low (HE = 0.053). Domestication brought a reduction in genetic diversity in landraces (the domestication founder effect). When measured in the whole sample, this reduction was about 25%, but when measured within major gene pools, the reduction was more drastic for the Mesoamerican domestication (MI gene pool, 55%) than for the Andean domestication, where no reduction was observed. It is worth noting that the genetic diversity of MI landraces from the Yucatan Peninsula (HE = 0.029) is significantly lower than other MI landraces (HE = 0.065). This result is in agreement with a late introduction of the crop in the Yucatan Peninsula from its area of origin in central-western Mexico, as well as a late development (or introduction) of agriculture in the Maya Lowlands60.Pairwise FST distances among gene pools showed that the wild gene pool most closely related to the Mesoamerican landraces is MI (FST = 0.33), distributed in central-western Mexico (Supplementary Table 6). As for the Andean landraces, the most closely related wild gene pool is AI (FST = 0.21), mostly distributed in the Andes of Ecuador-northern Peru. FST values also showed that the wild cluster, AII is most closely related to the Mesoamerican gene pools (FST values ranged from 0.56 to 0.65) than to the Andean gene pool AI (FST = 0.86). Such a close relationship was also shown by Caicedo et al.61 on the basis of AFLP polymorphisms and Toro et al.62 on the basis of electrophoretic patterns of phaseolin. STRUCTURE results also suggest a close relationship of gene pool AII with gene pool MII (Fig. 3a), however, the analysis of fineSTRUCTURE shows that gene pool AII is most closely related to gene pool AI (Fig. 3c).As stated above, FST values showed high genetic differentiation among gene pools, and consistent with this, haplotype introgression analyses clustered most of the accessions within their respective gene pools. However, there were 103 instances of chromosomal segments distributed in 35 accessions (9 wild and 26 domesticated) that could represent genetic contributions from different gene pools (Supplementary Data 6). The 103 chromosomal segments varied in size from 1 to 53 Mbp. In six accessions these chromosomal segments occupy more than 25% of their genome length, and concordantly these accessions were classified by STRUCTURE as admixed. Figure 3d shows the 58 chromosomal segments that were larger than 5 Mbp and that were observed in 15 accessions. Most of these segments represent genetic contributions between Mesoamerican gene pools (MI and MII) or between Andean gene pools (AI and AII), and more rarely between Mesoamerican and Andean gene pools.These genetic contributions may be the result of recent contact between wild and domesticated accessions, or between domesticated accessions of different origin. We found examples where introduced domesticated populations may have contributed chromosomal segments to the genetic makeup of wild accessions. For example, five MII wild accessions, distributed in the Mexican states of Chiapas and Campeche, in Guatemala, and in northern Colombia, all carry chromosomal segments that could have been potentially introgressed from introduced MI domesticated accessions. In all these geographical places, wild MII and MI domesticated accessions are distributed. Two AII wild accessions from central Colombia carry chromosomal segments potentially derived from domesticated AI accessions. We also observed 18 MI domesticated accessions carrying chromosomal segments potentially derived from the MII gene pool. These could represent cases of introgression from MII wild accessions into MI landraces in places where both kinds of accessions coexist such as the Mexican states of Veracruz and Chiapas, Costa Rica, El Salvador, and the Caribbean coast in northern Colombia. Interestingly, we detected four MI landraces from the United States and one from Colombia with chromosomal segments potentially derived from Andean AI landraces. At least in Colombia and Ecuador MI and AI landraces coexist. We also found two AI landraces, G25172 and G26184 collected in Ica, Peru, where the foreign haplotypes belong to the MI gene pool suggesting that these accessions might be the result of an interbreeding occurrence, maybe due to the presence of both Andean and Mesoamerican landraces in Ica, Peru. An early introduction of Mesoamerican landraces in coastal Peru has been suggested by the occurrence of pod remains (typical of small-seeded landraces) in association with the ceramics of the Guanape Period (from 1200 B.C. to 400 B.C.) and Cupisnique Period (from 1500 B.C. to 500 B.C.) in Huaca Prieta, Peru63, although more recent introductions cannot be excluded. Finally, an interesting pattern that we observed was that a single 2.6 Mbp MII segment located in chromosome Pl07 was shared by six MI landraces from different departments in Colombia, and that a second 2.2 Mbp MII segment located in chromosome Pl08 was shared by seven different MI landraces cultivated in different places of Central America and Colombia (Supplementary Data 6).Gene expression during pod developmentReduction or loss of pod dehiscence is one of the key domestication traits in Lima bean, and also a trait of agronomic importance64,65. To obtain information on the genetic regulation of the pod development, we carried out an RNA-seq experiment measuring expression levels at the initiation of pod elongation (T1) and before seed filling (T2) in one wild and one domesticated accession. At the developmental stage T1, pod valves become visible with the flower corolla still attached (or recently detached). At the developmental stage T2 pods reach their maximum length and weight (excluding the seeds) before the initiation of seed filling. Principal component analysis of expression values inferred from RNA-seq reads, including publicly available reads from a previous study66 consistently clustered replicates of each library with only one outlier which was removed for downstream analysis (Supplementary Fig. 16). Differential expression (DE) analyses revealed a total of 4275 genes with patterns of differential expression either across developmental stages or between the wild and the domesticated accession (Supplementary Data 7). Figure 4a shows a general heatmap of normalized expression values for these genes. Hierarchical clustering based on these values distinguished between five and seven gene clusters following different expression patterns.Fig. 4Gene expression at pod developmental stages T1 and T2.a Heatmap of normalized expression values within genes with differential expression. The left dendrogram corresponds to an unsupervised hierarchical clustering of the genes based on the normalized expression values. b Expression trajectories of the gene PlPDH1 in the domesticated accession (blue) and the wild accession (red) across two developmental times. TPM, transcripts per million. c Number of genes with differential expression (DEGs) between one wild and one domesticated accession and between two developmental times. d Concept map of functional categories enriched for genes more expressed in the second developmental time only in the wild accession. Source data are provided as a Source Data file.Looking at genes previously identified as related to pod dehiscence, the PDH1 gene significantly increased expression between T1 and T2 (Fig. 4b). The expression at the second stage was over 2-fold higher in the wild accession compared to the domesticated accession but the difference between accessions was not significant due to a relatively low expression value of one of the replicates. The PDH1 gene is involved in the formation of fibrous, strongly lignified cell layers between the inner and outer parenchyma of the pod, thus increasing the torsion force of pods in shattering genotypes67. Furthermore, it has been recently identified as a strong pod dehiscence QTL on chromosome Pv03 in common bean65.The ratio of gene expression between developmental stages or accessions showed that in general the domesticated accession had increased expression values for a larger number of genes than the wild accession (especially at T1) and that the number of genes with increased expression is larger than the number of genes with decreased expression between T1 and T2, especially in the wild accession (Fig. 4c). These patterns are collectively explained by about 1500 genes increasing expression between T1 and T2 in the wild accession. Genes with consistently increased expression between T1 and T2 in both the wild and the domesticated accession are enriched for terms that include cell wall biogenesis and organization, which is related to the biosynthesis of polysaccharides, particularly xylan and lignin, important components of the seed (Supplementary Fig. 17). Conversely, genes that only increase expression between T1 and T2 in the wild accession are enriched for metabolism-related genes of lignan, chitin, and aminoglycan and the fruit ripening process (Fig. 4d). Enrichment of chitin metabolism-related genes was mainly generated by a cluster of seven genes located at 38 Mbp of chromosome Pl09. Lignan metabolism-related genes were also enriched in this subset, mainly due to an array of nine genes with increased expression located at 9 Mbp of Pl11. Genes that remained more expressed in the domesticated accession only at T2 show enrichment of functions related to the development of the reproductive system, and particularly with the development of the fruit (Supplementary Fig. 18). Finally, genes with consistently reduced expression between T1 and T2 were enriched for developmental processes such as cuticle development and metabolism of different compounds, including cyanogenic glycoside. These findings are expected because the plant completed the formation of the pod and is about to start filling the seed.DiscussionIn this work, we conducted a large collaborative effort to provide a comprehensive set of genetic and genomic information for Lima bean. This effort resulted in information on the content, organization, and function of chromosomes and sequences, evolutionary relationships with close relatives at different taxonomic levels, population genomics of wild and cultivated accessions, and inheritance of agronomic traits. The analysis of this information revealed, at a greater detail, the genetic structure of wild and domesticated Lima beans across their large distribution range in the Americas, and provided insights on the genetic basis of variability in different agronomically relevant traits. Knowing the genes controlling these traits represents a great advantage for breeding programs and could potentially accelerate the development and release of improved Lima bean varieties in the future.The backbone of these achievements is the chromosome-level genome assembly for P. lunatus and a comprehensive genotyping of the available genetic variability within the species. The effort to build a high-quality reference genome sequence, both in terms of contiguity and base-pair quality, was rewarded by the assembly and functional prediction of large clusters of genes related to different traits. Although these predictions must be experimentally validated, the information provided is useful to prioritize genes to perform experimental validation as a community effort. Genes that confer resistance to biotic stresses on plants show large nucleotide diversity but good correspondence with common bean resistance genes based on orthology predictions and literature. As observed in previous studies in species with high-quality genomes such as rice, the variability of disease resistance genes is a key component of the defense mechanisms in plants68 and can contribute to functional redundancy favoring durability of resistance in the field69. Moreover, local gene duplication was also observed in genes performing functions such as metabolism of xyloglucan and chitin and regulation of flower development, which are directly related to yield traits such as seed weight and flowering time.The evolutionary history of plant genomes is shaped by several WGD and local duplication events, characterized by a wide range of evolutionary rates within and between gene families with important consequences in gene expression and function38. A full understanding of these processes is only possible through the development of high-quality genome sequences29. In the case of Phaseolus species, a synteny analysis of our assembly with that of P. vulgaris not only confirms the high degree of chromosome conservation between these genomes but also provides a detailed view of five major rearrangements between these genomes. Thus, this work makes a significant contribution to ongoing genome assembly and resequencing efforts to allow a full reconstruction of the evolution of genomes within the legume family, including the complete characterization of potential convergent evolution processes triggered by multiple domestication events.Previous phylogenetic analyses have placed the origin of wild Lima bean in the Andes of Ecuador-northern Peru during Pleistocene times and have suggested that, in this ancestral area (gene pool AI), wild populations experienced a range fragmentation event that reduced genetic diversity11. Accordingly, we found that Andean wild populations are relatively less diverse than the Mesoamerican ones, although more sampling is needed to have a more precise estimation of genetic diversity in this population. In wild common beans, a reduction of diversity was also documented in Andean populations but it was attributed to the effect of rare long-distance dispersal events from Mesoamerican populations to the Andes70. The identification of a wild Lima bean cluster (AII) in the Andes of central Colombia is of great significance because it supports the dispersal of the species from south to north and may represent a source of alleles not found elsewhere, as suggested by the characteristic phaseolin type (M8) that has been observed in this gene pool (see Supplementary Data 3). With regard to domestication, this study revealed that the domestication bottleneck in Mesoamerica seems to be more severe than previously thought10. The genetic diversity that was lost during domestication was apparently hardly re-gained by gene flow from wild to domesticated types, as suggested by the low number of landrace accessions showing signals of introgression. However, further studies including whole genome resequencing of larger numbers of landraces in a wider range of geographic locations are needed to completely assess the role of introgression in the current genetic diversity of domesticated Lima bean and avoid potential biases produced by low sampling or missing data in GBS experiments.The higher number of samples and genetic markers analyzed in the present study also provided us more in-depth information on the genetic substructure of the MI gene pool and its relation to geography, although we acknowledge that there could still be some sources of bias due to the relatively low SNP density and missing data obtained from GBS experiments. Within the MI wild cluster, two subclusters were observed: northern-western Mexico and southern-western Mexico. Also, within the cluster of Mesoamerican MI landraces, three subclusters were detected: Mexico-Central America, South America, and the Peninsula of Yucatan. The observation that landraces from the Peninsula of Yucatan form a separate subcluster and that their genetic diversity is relatively lower than other MI landraces, is in agreement with the hypothesis that after its domestication in central-western Mexico, Lima bean was a relatively late adoption of the Mayan communities of the Peninsula of Yucatan60. Unfortunately, in this region, Lima bean is under serious threat of genetic erosion58,71. The genetic differentiation observed among wild gene pools, the predicted domestication bottleneck in Mesoamerica and the threat of genetic erosion of landraces in the Yucatan Peninsula have important implications for conservation purposes.The development of the biparental population UC Haskell–UC 92 presented in this work not only guided the genome assembly but also provided a unique tool for further investigation on the drivers of domestication traits through QTL mapping. As in common bean72, the presence of the PlTFL1 gene is associated with a QTL for both determinacy and flowering time: determinacy can cause earlier flowering by converting the terminal meristems from a vegetative state into reproductive ones. In common bean, the determinacy trait is more common in the Andean gene pool73. In this experiment, the determinacy trait was contributed by the Andean parent UC 92. This observation raises several questions. Given the presence of determinacy in the Mesoamerican gene pool as well, what is the origin of this trait in the latter gene pool? Is it due to an independent mutation at the same or different loci or is due to introgression from the Andean gene pool? Whole genome resequencing data among a broader sample of Andean and Mesoamerican accessions is necessary to answer these and other questions.The three quantitative traits scored in the UC 92–UC Haskell population showed transgressive segregation to varying extents. However, seed weight only showed transgressive segregation for seed weights below the small-seeded parent UC Haskell, but not for larger seed weights. This observation points to the difficulty in developing large-seeded improved progenies in populations arising from selfing only. Possible solutions are the development of population based on backcrosses to a large-seeded parent. An alternative is suggested by the phenotypic distribution for HSW, discriminated by allelic combinations (Supplementary Fig. 6). The presence of three marker alleles of the large-seeded parent (UC 92) for three seed weight QTL corresponds to the heaviest seeds and vice-versa for the individuals with the three marker alleles of the small-seeded parent (UC Haskell). Thus, indirect selection using these markers in early generations may shift breeding populations towards heavier seeds. Regarding cyanogenesis, the combination of QTL alleles giving the highest levels of cyanogenic glucosidase was the Pl05 (UC Haskell)–Pl08 (UC 92)–Pl10 (UC Haskell). Whether or not this combination should be selected for will depend on considerations such as a concern for consumer safety or a potential role in insect resistance.We also used RNA-seq to follow the gene expression patterns at two pod developmental stages in one wild (shattering) and one domesticated (non shattering) Lima bean accession. Hundreds of differentially expressed genes between the two accessions were detected between two developmental stages of the pod (T1 at the initiation of pod development and T2 when pods reached their maximum length before seed filling). In particular, we identified differential expression in the PDH1 gene, known to be involved in the control of pod dehiscence in other crops74. The PDH1 gene is a ‘dirigent’-type gene involved in the polymerization of lignin monomers, consistent with the pod fiber role in dehiscent pods of wild types. In soybean, this gene is involved not only in lignification of pod walls but also in the increase in the torsion force of the pod67. Hence, the reduced expression of this gene is consistent with reduced pod dehiscence in domesticated types.In P. vulgaris, PvPDH1 is known to be only transcribed in the pod tissue75. Recently, this gene has been associated with pod dehiscence and with the adaptation of the domesticated ecogeographic race Durango of P. vulgaris to arid conditions in northern Mexico65. Anatomical and histological analyses have shown that shattering genotypes have an extensive lignified wall fiber layer (LFL) in pod walls while in non-shattering genotypes LFL deposition is reduced or absent. Our observation that the PlPDH1 gene increases expression in the wild accession of Lima bean at T2 is compatible with the results in common bean; therefore, PlPDH1 is a good candidate for further investigation. Moreover, the information obtained from differential expression was also useful to identify genes related to the regulation of flower development and metabolism of xylan and pectin, which in turn can be related to flowering time and seed weight. In addition, interesting patterns of expression were also observed for genes with annotated biological processes such as metabolism of chitin and cyanogenic glycoside, cuticle development, response to auxin, cell wall biogenesis and fruit ripening, which could be related to different physiological characteristics and even seed quality traits such as cookability or cyanide content.Given the importance of Lima bean, both as a current food security crop and as a potential protein source for different climate change scenarios, we expect that this work will provide a basis for many future studies in Phaseolus species with applications to breeding and will make an important contribution to the field of Phaseolus genomics across the five domesticated species of the genus.MethodsSequencingLima bean genomic DNA was extracted from young leaves of two-week-old seedlings of the domesticated accession G27455 according to the requirements for DNA concentration and integrity of each sequencing technology. High molecular weight (HMW) DNA for Pacific Bioscience and 10X Genomics sequencing was extracted with the Qiagen MagAttract HMW DNA Kit (QIAGEN, Germantown, MD, USA) following manufacturers’ instructions. For the construction of libraries for Pacific Bioscience sequencing, a SMRTbell Template Prep Kit was used according to the manufacturer’s instructions (Pacific Biosciences, Menlo Park, CA, USA). DNA was randomly fragmented by adding fragmentation buffer using Covaris g-TUBE devices and the purification was carried out with AMPurePB magnetic beads after concentration. Fragments greater than 3 kbp underwent a damage repair step combined with an end-repair, followed by a blunt end ligation with hairpin adapters. Libraries were sequenced on a PacBio Sequel platform at the sequencing provider Novogene Corporation Inc.For 10X Genomics DNA library construction, one μg of Lima bean DNA was prepared with Chromium Genome HT Library & Gel Bead Kit V2/10x Genomics according to the manufacturer’s protocol (10x Genomics, Pleasanton, CA, USA). The resulting 500–700 bp insert libraries were quantified using a Qubit 2.0 fluorometer (Thermo Fisher Scientific, Waltham, MA, USA) and quantitative PCR. The size distribution was analyzed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Suitable libraries were sequenced on an Illumina HiSeq Platform (Illumina, San Diego, CA, USA) using a paired-end 150 run (2 × 150 bases).For Illumina sequencing, young trifoliate leaves from two-week-old seedlings were collected and frozen with liquid nitrogen. DNA isolation was performed with the same extraction method used for genetic diversity and population structure analyses (see below). The Illumina library used 1.0 μg of DNA according to a NEBNext DNA Library Prep Kit following the manufacturer’s recommendations (New England BioLabs, Ipswich, MA, USA). Genomic DNA was fragmented to a size of 350 bp, fragments were ligated to NEBNext adapters and enriched by PCR. The library was analyzed for size distribution with an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and quantified using real-time PCR. Libraries were sequenced on an Illumina HiSeq Platform (Illumina, San Diego, CA, USA) using a paired-end 150 run (2 × 150 bases) and insert size of 450 bp.RNA sequencing and de novo transcriptome assemblyPlants from accessions G25230 (Mesoamerican wild) and G27455 (Mesoamerican landrace) were grown under greenhouse conditions at the Centro Internacional de Agricultura Tropical (CIAT; Palmira, Valle del Cauca, Colombia). Total RNA was extracted from leaves, pods, and flowers with a specific protocol for each tissue76,77. For the analysis of differential gene expression related to pod dehiscence, RNA was extracted for the wild and domesticated accession at two pod developmental stages (with three replicates each), at the initiation of pod elongation (T1) and before seed filling (T2). At T1 pod valves become visible with the flower corolla still attached (or recently detached) and at T2 pods reach their maximum length and weight before seed filling. Previous studies show that strong lignin deposition at the dehiscence zone is already observed at this stage of the pod development in wild samples78. RNA samples were quantified with Nanodrop 2000 and quality was assessed with Agilent Bioanalyzer 2100 system (Agilent Technologies, Santa Clara, CA, USA) and agarose gel electrophoresis. Samples showing a 260/280 ratio of absorbance between 1.9–2.0 and RIN (RNA Integrity Number) values above seven were selected. mRNA samples were enriched using oligo(dT) beads and randomly fragmented, then cDNA was synthesized using mRNA templates. RNA-seq library preparation was completed through size selection and PCR enrichment. Sequencing of the library was accomplished using a paired-end 150 run (2 × 150 bases) on an Illumina HiSeq Platform (Illumina, San Diego, CA, USA).The quality of the raw reads was evaluated with the fastQC v.0.11.2 program79 and low-quality reads and adapter sequences were filtered with Trimmomatic v.0.3680, removing approximately 3% of the reads. De novo transcriptome assemblies were obtained by using the software Trinity v.2.4.081. Each transcriptome assembly was compared with 1440 single-copy orthologs from the OrthoDB v.9.1 database using the BUSCO (Benchmarking Universal Single-Copy Orthologs) v.2 pipeline82.To increase the expression evidence for gene annotation, we also downloaded public RNA-seq reads sequenced in a previous study66 from the NCBI Sequence read archive (bioproject accession number PRJNA275266), using the fastq-dump facility of the SRA toolkit v.2.9.4 (https://github.com/ncbi/sra-tools/wiki).Sequencing and assembly into scaffolds and pseudomoleculesA chromosome-level assembly of the Lima bean genome was achieved by combining reads from four different sequencing protocols (Supplementary Table 7). The entire sequencing effort added up to a total of 97.6 Gbp of raw data, which represents about 157x of the haploid genome size, initially estimated to be around 622 Mbp27. The genome assembly pipeline included four main steps. First, a de novo assembly of the PacBio reads was performed using Canu v.1.6 with default parameters83. This step resulted in a draft assembly of 496 contigs with N50 of 5.5 Mbp and a total length of 542 Mbp (Supplementary Fig. 19). An initial analysis of the linked reads data and the GBS data from the biparental population (see details below) allowed us to identify and break 12 potential misassemblies. As a second step, polishing was performed to achieve high base pair quality integrating the paired-end Illumina data to correct base pair and indel errors. Reads were aligned to the 508 contigs using bowtie2 v.2.3.584 and variants were called using the command FindVariants of NGSEP 3.3.285 with the following parameters: -runRep, -runRD, -runRP, -minMQ 0, -maxBaseQS 30, -minQuality 40, -h 0.0001 and providing predictions of STRs performed using tandem repeats finder86. Assembly errors, called as homozygous alternative variants, were corrected using the command VCFIndividualGenomeBuilder of NGSEP. This process was repeated four times. The options -runRep and -runRD were executed to identify repetitive regions from multiple alignments and CNVs from read depth signal, respectively. Although 439 contigs (86%) were identified as repetitive because they included predicted repeats over at least 60% of their total length, the total length spanned by these contigs is 194 Mbp (36% of the assembled length). The remaining 69 scaffolds span 348 Mbp mostly because they are the longest contigs (Supplementary Fig. 20). In a third step, linked reads were aligned to the polished contigs using bwa v.0.7.1787 to perform scaffolding. In the 10x protocol, reads having the same barcode should have been sequenced from the same initial molecule. Hence, reads with the same barcode mapping to different contig ends provide evidence of linkage between such ends. We built an undirected graph having contig ends as vertices and evidence of linkage as edges to run a clustering algorithm similar to that implemented in the software Salsa for Hi-C data88. This procedure allowed us to combine 164 contigs in 40 scaffolds.Finally, in a fourth step, a dense genetic map including more than ten thousand selected variants was built by analyzing GBS data from an F8 biparental population (parents UC 92 and UC Haskell, details in the next section). Linkage disequilibrium was calculated for each pair of variants and pairs in different contigs having an r2 value greater than 0.8 were considered as evidence for linkage between contigs. Connected components on a graph with contigs as vertices and linkage evidence as edges allowed us to identify the expected 11 linkage groups. Combining this linkage information with the linkage provided by the scaffolding process, 206 contigs adding up to 512 Mbp were sorted and oriented to produce the final 11 pseudomolecules representing the chromosomes of the species. The GBS reads were reanalyzed against the assembled pseudomolecules to verify the structural consistency of the final assembly and to identify recombination breakpoints per sample.RNA-seq data-based genome annotation, comparisons, and expression analysisAnnotation of repetitive elements was performed using RepeatMasker v.4.0.530 using the library of repetitive elements for P. vulgaris kindly provided by the authors of Gao et al.31. The genome assembly was masked by replacing nucleotides within these regions with N characters to perform gene annotation. The genome annotation strategy combined ab initio predictions, cDNA sequences (pod, flower, and leaves), and homology-based approaches following the Maker pipeline v.2.31.989. Raw RNA-Seq data were filtered using Trimmomatic v.0.36 to remove adapter and low-quality sequences (>Q30). We used 57,742 de-novo transcript assemblies obtained following the Trinity pipeline81 and 36,995 putative protein sequences from common bean as evidence in the Maker annotation process. In parallel, RNA-seq clean reads were processed according to the Tuxedo pipeline90. HISAT2 v.2.1.091 was used to align the reads to the Lima bean reference genome with default parameters. StringTie v.1.3.592 was used to predict transcript annotations from aligned reads. The two annotations were merged using a custom script available with the NGSEP distribution (class ngsep.transcriptome.io.GFF3CombineAnnotations). Suspiciously long genes and genes with an abnormally high number and distribution of transcripts and covering other genes were curated according to sequence similarity with common bean. This annotation was initially used for validation of the base pair quality of the genome assembly searching 1440 single-copy orthologs from the OrthoDB v9.1 database using the BUSCO (Benchmarking Universal Single-Copy Orthologs) v.2 pipeline82. The Tuxedo pipeline was also followed to generate the count of expression levels for each gene and each sample. HISAT2 v.2.1.0 was used to align the reads to the Lima bean reference genome and StringTie v.1.3.5 was used to obtain the matrix of read counts. We carried out a differential expression analysis with DeSeq2 v.3.193 performing independent comparisons between developmental stages for each accession and between accessions within each developmental stage. Results of a linear model design combining developmental stages and accessions were also evaluated. Significant differential expression was predicted if the comparison p value was below 0.05 and the log-fold change was above two. To assess at which level the results were skewed by reference bias, expression levels for the transcripts assembled de-novo were estimated directly from raw reads using the software tool Salmon v.1.2.194. Differential expression analysis of these expression levels was also assessed with DeSeq2. The trends were consistent between expression levels obtained following the two pipelines (Supplementary Fig. 21).Functional annotation of the predicted gene models and transcripts was performed following the Trinotate pipeline v.3.1.195. In brief, blastx and blastp searches from NCBI Blast v.2.10.0 were performed against the UniProt database using as queries the cDNA and amino acid sequences of each transcript, respectively. The best hit of each search was retrieved if its corresponding e value was below 0.001. The hmmscan tool of HMMer v.3.3.196 was also used to identify conserved domains registered in the Pfam database. Results of these queries were combined into a Trinotate database to generate the final annotation report.To assess the quality of the gene models and perform automated curation, orthology with P. vulgaris, expression measured as having a transcript per million (TPM) measure greater than 0.5 in at least one tissue and functional annotation were considered evidence supporting the existence of each annotated gene. A total of 10,263 annotations were filtered keeping genes with at least one type of evidence or having both proteins length over 200 amino acids and a paralog with direct evidence.Paralog identification and comparison with other annotated genomes were performed running the GenomesAligner command of NGSEP. In brief, this command builds an FMIndex with the proteomes of each species and then performs amino acid searches of non-overlapping k-mers taken from each sequence on each FM-Index. A homology relationship is called if the percentage of matching query k-mers for a sequence is larger than a given parameter. To achieve identification of paralog relationships from the latest whole-genome duplication event, the GenomesAligner was executed with a k-mer length k = 5 and a minimum percentage of k-mers p = 20. A WGD paralog relationship is called if the two genes have at most ten paralogs in total, are located in different chromosomes and at least half of the neighbors in a window of up to ten genes share a syntenic homology relationship.Functional enrichment of gene ontology (GO) terms for gene sets selected from the different experiments was performed running the topGO package of Bioconductor v.2.36.097 using a Fisher exact test. Each set was compared against the genes in the Lima bean genome as background. Visualization of enriched GO terms was performed using REVIGO98 and visualization of GO knowledge graphs was performed in Cytoscape v.3.8.199.Identification of genes for biotic disease resistance was performed based on the functional annotation of the genome and orthology prediction. The complete list of gene models with known functions was manually screened for known domains associated with resistance (R)-genes as predicted or confirmed by at least one of the searched databases: NCBI (BLASTp), Pfam, and eggNOG/GO/KEGG. The domains included in the selection were: CC, NB-ARC, TIR, LRR, WRKY, LZ, and protein kinase. According to the different domain arrangements within the nucleotide-binding site (NBS)-Leucine-rich repeat (LRR) gene family different subfamilies were classified for each of the 11 Lima bean chromosomes. Genomic locations of these genes and their different protein sequences were derived from the genome assembly and structural annotation. A physical map of the 11 chromosomes with individual NBS-LRR genes was generated by MapGene2Chrom web v.2.1100 and a multiple sequence alignment of the corresponding proteins was created using muscle v.3.8.31101. The alignment was used to produce a tree topology for gene diversity analysis based on the Neighbor-joining approach by MEGA X v.10.8.1102, which was visualized and edited by iTOL v.4.4.2103.Genetic analyses in the UC 92 and UC Haskell recombinant inbred populationA biparental recombinant inbred line (RIL) population was developed at the University of California, Davis (38.542790° N. Lat.; 121.763049° W. Long.) from reciprocal crosses of two contrasting California Central Valley-adapted cultivars, UC 92 and UC Haskell, from distinct domestication gene pools of Lima bean. UC Haskell is a small-seeded, vine-type cultivar of Mesoamerican origin and UC 92 is a large-seeded, bush-type cultivar of Andean origin. Crosses were made in a greenhouse in 2012 and progeny seeds were grown for hybrid verification. F1 hybrids resulting from this cross were verified using the polymorphic codominant Pvctt001 simple sequence repeat (SSR) marker using a PCR protocol performed as recommended using Taq DNA Polymerase with ThermoPol Buffer (New England Biolabs). PCR products were analyzed on a 2% agarose gel using 1x TAE buffer. Subsequently, the population was advanced to the F8 stage by single-seed descent. Leaf tissue of the two parents and each RIL was collected from two-week-old seedlings; DNA was extracted using a DNEasy Plant Mini Kit (Qiagen). The resulting RIL population demonstrated segregation for many agronomic traits, including germination rate, flowering time, inflorescence position, plant height, plant habit, pod position, pod density, yield, and biotic stress tolerances21.Leaf tissue for DNA extraction was sampled from two-week-old seedlings of 238 RILs in the F8 generation. Leaf tissue samples were collected into 96-well plates from plants grown in a greenhouse, immediately put on ice, and lyophilized for 24 h. DNA was extracted using an adapted DNA extraction protocol for Lima and common bean104. The presence of DNA was confirmed for samples with 260/280 absorbance ratios above 1.8 using a NanoDrop Lite (Thermo Fisher Scientific). DNA was quantified using the Quant-iT PicoGreen dsDNA Assay Kit (Thermo Fisher Scientific) and 100 ng of DNA from each sample was transferred to a PCR plate. GBS barcode libraries and adapters for the CviAII restriction enzyme (New England Biolabs) were prepared using an adapted protocol for common bean104. The CviAII restriction enzyme104, CutSmart buffer (New England Biolabs), and unique barcode identifiers for each well were added to the plates, spun down and incubated in a PCR machine for 2 h at 25 °C. T4 (10x) buffer and T4 ligation were added to the wells and run in the PCR machine for 1 h at 22 °C followed by 30 min at 65 °C. Seven microliters of each sample was added to a petri dish, mixed, and transferred to Eppendorf tubes. Binding buffer and isopropanol were added to the tubes and mixed. Eight-hundred microliters were transferred to a GeneJET (Thermo Fisher Scientific) purification column, centrifuged for 60 s, and eluted with water. DNA was quantified on the QUBIT dsDNA HS Assay Kit (Thermo Fisher Scientific/Invitrogen) prior to GBS sequencing. Two genomic libraries were prepared, each containing 144 unique barcode identifiers for the CviAII restriction enzyme, and a total of 240 unique genetic lines were sequenced. The libraries were sequenced using the SR100 protocol on two lanes of an Illumina HiSeq flow cell at the University of California, Davis, Genome Center.Sequence data was de-multiplexed and adapter contamination was removed using the Demultiplex command of NGSEP. Single reads for each sample were aligned to the contigs using bowtie2 with default parameters. Then, the command MultisampleVariantsDetector of NGSEP was used to identify variants and call genotypes for each sample with parameters similar to those used to analyze the GBS diversity data. A total of 51,897 SNVs, 2927 biallelic indels, and 758 biallelic STRs were identified in 183 contigs after filtering calls with genotype quality less than 30, minor allele frequency less than 0.3, heterozygosity rate larger than 0.05, and variants for which less than 100 individuals were genotyped. Missing genotypes were imputed and parental assignment of individual haplotypes was obtained running the ImputeVCF command of NGSEP with the following parameters: ‘-p UCHaskell, UC92 -k 2 -c 0.003 -ip –is’. Recombination events between each pair of neighboring markers were identified and centimorgans were estimated assuming five generations of crossover and one expected crossover per individual chromosome per generation.Ninety-three RILs and 10,497 polymorphic SNP markers were used to construct the genetic map using the ASMap v.1.0.4 and R/qtl v.1.44 packages in R105,106. Markers with less than 20% missing genotypes were used for map construction, and individuals with excessive recombination rates and more than 50% missing genotypes were removed from map construction. Linkage groups were formed using the ‘mstmap’ function with p < e-8 and genetic distances were calculated using the ‘Kosambi’ mapping function. Linkage groups were merged when originating from the same chromosome and recombination frequencies were recalculated by chromosome with lower p-values. Recombination rates across the genetic map were calculated using the ‘MareyMap’ function using the ‘sliding window’ interpolation method every 200 kbp in a 1 Mbp sliding window107. Pericentromeric regions of the chromosomes were defined when recombination rates exceeded 2 Mbp/cM for a given locus. Chromosome numbering followed the numbering of P. vulgaris chromosomes to obtain a one-to-one correspondence, justified by the high level of synteny108.QTL mapping for plant habit, seed weight, days to first flower, and cyanogenesis was performed in the UC 92–UC Haskell RIL population using the R/qtl package105. A genome-wide scan for single QTLs was performed using the ‘scanone’ function with the extended Haley-Knott regression method105. For traits that were controlled by a single QTL, composite interval mapping was performed using the ‘cim’ function with the extended Haley-Knott regression method, the Kosambi mapping function, and 1000 permutations to identify the position and LOD score of QTLs above the 95% significance threshold. For traits that contained multiple significant QTL, the ‘makeqtl’ and ‘fitqtl’ functions were used to identify the optimal multiple QTL model, and to calculate LOD scores, percent of phenotypic variation explained by the QTL and QTL effects.Days to first flower were collected from an experiment grown at two locations in Davis, California in 2018. The date was recorded when a majority of plants in the plot had at least one floral bud open. An ANOVA of a generalized linear mixed model, including genotype and treatment as fixed effect factors and location and blocks as random effect factors, was performed for days to first flower. Phenotypic characterization for plant habit, hundred seed weight, and delayed senescence included in the QTL mapping study were performed on single plants that were grown within a greenhouse.Cyanide quantification of samples of floral bud tissue was collected in triplicate subsamples from an experiment grown at two locations in Davis, California in 2018. Tissue samples were kept on dry ice, weighed, and organized in 96-well plates and frozen at −80 °C. Cyanide quantification was performed using an adapted Fiegl-Anger method109. Tissue samples thawed at room temperature for 30 min before Fiegl-Anger paper was placed over each plate for 30 min. Standards of hydrogen cyanide were created with concentrations of 0, 25, 50, 75, 100, 250, 500, 750, and 1000 nM of potassium cyanide exposed to Fiegl-Anger paper for 30 min. Fiegl-Anger paper was immediately scanned, and quantification of the blue absorbance intensity was calculated using ImageJ software and the ‘readplate2’ plugin. Blue absorbance intensity was calculated as -log(Mean/255). A standard regression curve was calculated from the standard HCN concentrations and nanomolar concentrations for the samples were calculated based on this curve and calculated as an nM/min rate across the initial 30-min interval measured. Two separate linear models and ANOVAs were performed for the RIL population grouped by haplotype at the QTL peak on chromosome Pl05, since this QTL effect consistently produced bimodal distributions. The best fitting linear model for floral bud cyanogenesis included genotype, location, and treatment as fixed effect factors.Genetic diversity and population structure analysesFor GBS library construction, five biological samples of leaf tissue (three young trifoliate leaves of five different plants) were collected for each accession. The samples were frozen and stored at −20 °C until processing. DNA isolation was performed from frozen trifoliate leaves using the extraction method developed by Vega-Vela and Chacon Sanchez (2011)110. DNA quality was checked with a Nanodrop 2000 and analyzed in 1% agarose gel electrophoresis; DNA with no visible degradation and a ratio 260/280 above 1.8 was selected. DNA libraries (one per accession) were generated by mixing the DNA of five individuals in equal parts (with the help of a Nanodrop 2000 and electrophoresis calibration with lambda DNA at concentrations of 25, 50, and 100 ng/µL), then the DNA restriction for each sample was performed with the ApeKI enzyme and ligated to adapters containing one of 94 unique barcodes for each plate. The library was sequenced in paired end using an Illumina Hi-Seq2000. Both library preparation and sequencing were done at the Australian Genome Research Facility (Melbourne, Australia).A total of 482 Lima bean accessions from the International Center for Tropical Agriculture—CIAT (262 accessions) and Centro de Investigacion Cientifica de Yucatan—CICY (220 accessions) were analyzed in this study (see Supplementary Data 3). Of these accessions, 215 were domesticated and 267 wild. Wild accessions covered the known natural geographic range for this species and domesticated accessions were landraces from different countries in the Americas (Supplementary Fig. 10). These accessions complemented those analyzed in a previous study10 to increase the sampling of wild accessions from the MI gene pool (the putative ancestral gene pool of Mesoamerican landraces) and gene pool AII from central Colombia. Sequenced accessions from CIAT are available through the mechanisms established by the germplasm collection. Sequenced accessions from Centro de Investigacion Cientifica de Yucatan (CICY) are available directly upon request.The sequences obtained from each sample were de-multiplexed by their barcode using Next Generation Sequencing Experience Platform (NGSEP) v.3.3.285. Individual fastq files were mapped to the Lima bean reference genome by using default parameters of bowtie2 and adding an ID code for each accession84. Variants were identified and individuals were genotyped using the MultisampleVariantsDetector command of NGSEP. The parameters used for SNP calling were: 50 as a maximum number of alignments per start position, heterozygosity rate of 0.0001 (prior probability of finding in every position a heterozygous SNP), and a minimum genotype quality score of 40 (codified in Phred, where 40 means 0.9999 of posterior probability that each genotype call is correct). Initial filtering to obtain a raw set of reliable variants was obtained by filtering using the FilterVCF command of NGSEP with the following criteria: (1) Exclude variants from repetitive regions and in not-assembled scaffolds; (2) Exclude variants genotyped in less than 100 of the 482 accessions; and (3) Exclude variants with minor allele frequency (MAF) less than 0.01. This procedure generated an initial set of 116,030 biallelic SNVs, 7517 biallelic indels and STRs, and 22,138 multiallelic variants (96.6% of them triallelic SNVs) with approximately 40% of missing data. Further filtering was applied for the different downstream population genetics analyses (details below).Linkage disequilibrium (LD) was measured as r2, the correlation coefficient among pair of alleles across pairs of SNP loci, specifically as average r2 against 10 and 100 kbp physical intervals through the 11 chromosomes of Phaseolus lunatus using TASSEL v.5111. This analysis was conducted on a filtered dataset with the wild samples, as well as a filtered dataset with the domesticated samples. We also conducted the analysis for Mesoamerican and Andean gene pools. Results were plotted in R software by using an adapted code.To estimate the number of clusters in Lima bean useful to describe the genetic data, we used algorithmic approaches and model-based approaches. For this, we built a multilocus-genotype matrix containing the 482 accessions (267 wild and 215 domesticated) described earlier. The raw VCF file described in the previous section was filtered keeping only biallelic SNVs with a distance of more than 10 bp from other variants, genotyped with quality at least 40 in at least 450 individuals and with minor allele frequency larger than 0.03. A total of 12,398 SNVs were retained after these filters with only 5.3% of missing data. For each sample, genotype calls based on read data were present for at least 75% of the SNPs and this percentage is less than 90% for only six samples (Supplementary Data 4).The algorithmic approaches were carried out with the software Darwin v.6.0.021112 and the R package Adegenet v.2.1.2113. In Darwin, a pairwise distance matrix was calculated by estimating the proportion of matching alleles per locus for every pair of individuals. For the representation of the genetic structure, we used three methods: a principal coordinate analysis (PCoA in Darwin), a tree method (neighbor-joining -NJ- algorithm in Darwin), and a discriminant analysis of principal components114 (DAPC in Adegenet), that unlike PCoA, optimizes the variance between groups while minimizing the variance within groups. In Adegenet, the function ‘find.clusters’ was used to identify the optimal number of clusters based on the Bayesian information criterion (BIC), 500 principal components were retained, four discriminant functions were used and the proportion of conserved variance was 0.95. For statistical support of NJ tree structure, 1000 bootstrapped matrices were obtained. The tree was visualized using iTol103.The software STRUCTURE v.2.3.4115 was used as an alternative approach to infer the population structure of the sequenced accessions. The filtered VCF file was converted to the input format of STRUCTURE using the ConvertVCF command of NGSEP. A total of 200 STRUCTURE runs were executed varying the number of clusters (K parameter) from 1 to 10 and performing 20 repetitions for each K value. For each run, 10,000 burn-in iterations and 30,000 sampling iterations were executed. The main and extra parameters were left with default values with the exception of the option ‘ONEROWPERIND’ which was set to 1 to make it consistent with the format provided by the VCF converter of NGSEP. Stability of the Markov chain was assessed looking at the variance of the likelihood of the clustering achieved by each repetition. The solution with the best likelihood was selected for each K value between 2 and 8 and cluster numbers were reorganized to visually assess the cluster stability across different K values (Fig. 3a). The Evanno test was performed to predict the optimal number of clusters116.Haplotype based population structure was assessed with the software fineSTRUCTURE v.4.1.057. In brief, this software builds a sample relationship matrix capturing the haplotype similarity across the genome. This matrix is then used to run a Markov chain to sample group configurations and tree topologies consistent with the similarity matrix. For this analysis, the VCF file with population information was filtered keeping only biallelic SNVs with a distance of more than 10 bp from other variants, genotyped with quality at least 40 in at least 300 individuals, MAF larger than 0.02, and observed heterozygosity (OH) below 0.5. A total of 28,753 SNVs were retained after these filters with 11% missing data and 332,817 (2.4%) heterozygous data points. Imputation of missing data and statistical haplotyping was performed using Beagle 5.0117 setting the phase states to 50, the imputation states to 400, and the effective population size to 100,000. For each chromosome, the VCF was converted to the input format of fineSTRUCTURE using the command VCFConverter of NGSEP. The four steps pipeline of fineSTRUCTURE was executed using default parameters following the user manual and providing genetic distances inferred from the genetic map developed from the biparental population.Basic diversity statistics for populations, namely observed heterozygosity (HO), expected heterozygosity (HE), endogamy (FIS), global FST, and pairwise FST118 were estimated using the R packages Adegenet113 and Hierfstat v.0.5119 from the same dataset used to run STRUCTURE. Difference between HO and HE was assessed by means of the Bartlett test and paired t test. The significance of population differentiation was tested by means of a likelihood ratio G-statistic120.Introgression analysis was performed running the ‘IntrogressionAnalysis’ command of NGSEP. The input VCF for this analysis was the same used to run STRUCTURE. The input file with population assignments contained genetic populations inferred by STRUCTURE at K = 4, which separates the MI, MII, AI, and AII populations. Only 448 samples with clear assignments were included in this file. The analysis was executed with default parameters (non-overlapping windows of 50 SNPs discriminating at least one pair of populations). The raw set of predicted introgression events called with this procedure was filtered keeping only events that span more than one contiguous window or, for a single window of 50 SNPs, if the similarity score between the haplotype of the sample and the consensus haplotype of the sample background population was below 15. Then, the D-statistic implemented in the ANGSD package v.0.93121 was calculated to assess significance of the introgression events. GBS reads from common bean were aligned to the P. lunatus genome to use them as an outgroup for this analysis. Events with Z-scores larger than 2 were retained. Additionally, the introgression called for the MII sample JMC_1097 was retained because it spanned 53Mbp over four chromosomes. Finally, two introgressions of about 2Mbp called at 36 Mbp of Pl07 and 4.6Mbp of Pl08 were retained because they were consistently called in six and seven accessions, respectively. Although the D-statistic probably did not have enough power to assess significance of these introgressions, they were validated by construction of neighbor-joining clusters within the specific regions (Supplementary Fig. 22).Reporting SummaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting SummaryDescription of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2Supplementary Data 3Supplementary Data 4Supplementary Data 5Supplementary Data 6Supplementary Data 7 | nature communications | [
"Article"
] | [
"Genetic variation",
"Agricultural genetics",
"Genome evolution",
"Plant domestication"
] | Phaseolus genus 70 species five domesticated P. acutifolius (tepary P. coccineus. dumosus P. lunatus (Lima P. vulgaris (common bean Lima bean common bean agronomically economically significant species Lima bean nutrients seeds contain 20% protein 50% carbohydrates rich source amino acids tryptophan lysine methionine phenylalanine threonine valine isoleucine leucine5,6 domesticated Lima beans climatic conditions northern Mexico to northern Wild Lima beans three gene pools two Mesoamerican (MI MII one Andean (AI) gene pool MI central-western Mexico MII southern Mexico Central America to South America Andean gene pool AI restricted to southern Ecuador northern Peru another Andean pool AII Andes central Colombia proposed confirmation two domestication processes in Lima bean Mesoamerica Andes10 Andean domestication from gene pool AI varieties large flat seeds Lima second event central-western Mexico from gene pool MI Mesoamerican varieties small rounded oval-shaped seeds (Potato Sieva cultivarsLima bean convergent evolution Mesoamerican Andean landraces similar traits domestication larger pods seeds pod dehiscence seed dormancy growth habit reduced antinutritional seed compounds.Lima bean interest for evolutionary research convergent phenotypic adaptation shows adaptation to ecological conditions heat drought stresses key climate important document domestication process Lima bean understanding phenotypic adaptations genetic control identification related genes alleles for future breeding Lima bean genetic research relied on common bean reference Lima common bean diploid species (2n 2x 22 chromosomes DNA ~622 Mbp/1 C high homozygosity predominant autogamy25 cytogenetic research confirms synteny between relying on P. vulgaris genome diversity reference bias loss information misleading predictions genomic loci whole-genome reference sequence for P. lunatus groundbreaking genetic resource for Lima bean research genomic differences among domesticated research collaborative effort high-quality genomic resources for Lima bean genetics breeding high-quality reference genome gene expression information tissues accessions developmental stages comprehensive assessment genomic variability in 500 wild domesticated accessionscombining comparative genomics population genetics genomic variability relationship common bean provide gene functional annotations genetic loci associations traits domestication breeding Lima beans chromosome-level-quality assembly Lima generated Lima bean genome from G27455 accession Mesoamerican gene pool northern Colombia Data Pacific Biosciences Illumina sequencing technologies protocols paired-end genome sequencing 10x genotyping-by-sequencing RNA sequencing combined contiguity base quality initial assembly PacBio WGS Illumina assembly 512 contigs 542 Mbp 206 512 sorted oriented in 11 pseudomolecules linked reads G27455 accession 10x GBS data F8 generation UC 92–UC Haskell recombinant line population linkage map developed 10,497 SNPs 522 unique loci estimated genetic length 1064 cM Linkage groups established 11 chromosomes average genetic physical spacing between loci 2.18 cM 1.10 Mbp Genetic gaps larger than 20 cM observed on three linkage groups Pl01 Pl02 Pl09 20.5 24.4 32.8 cM gapsMarker coverage varied linkage groups densest in regions Pl02 Pl05 Pl07 Pl11 sparsest Pl01 Pl09. Recombination rates varied groups lowest rates in centromeric pericentromeric regions highest towards telomeric ends (Fig. 1a 2) Pl03 group highest average recombination rate every kbp Pl10 lowest recombination rate every 2074 kbp influenced by high segregation distortion UC Haskell haplotype largest spans on Pl04 Pl10 Pl11 linkage groups shortest spans Pl03 Pl06 (Fig 1b Pl01 Pl09 linkage groups sparse marker coverage reduced accuracy recombination rates definition.Fig. 1Chromosome-level genome assembly for Lima bean Genetic distance recombination rate by position Lima bean reference genome UC 92–UC Haskell RIL population Chromosomes labeled Pl01-Pl11 Chromosome lengths pericentromeric regions Density of repetitive elements gene models SNPs LOD scores of QTL for four traits screened RIL populationColored zones represent LOD scores 3 determinacy flowering time seed weight volatile cyanide Phenotypic distribution traits in RIL population parental phenotypes vertical lines Source data file-pair-level quality assessed by mapping Illumina reads to assembly reached 99% raw reads after polishing identification orthologs reached 98.8% of 1614 genes single copy plant species Annotation repetitive elements using repeat common bean library 796 transposable 656,928 events identified 225 Mbp (41% half regions (174 Mbp covered by long terminal repeats 8 Mbp covered by class I retrotransposons LINE SINE DNA (Class II) transposons covered 25 Mbp 6 Mbp unclassified transposable elements Repetitive elements abundant pericentromeric regions automated structural functional annotation Lima bean genome repeat elements masked Illumina RNA-seq data from three tissues developmental stages accessions analyzed gene density SNPs higher outside pericentromeric regionsRNA-seq data gene models common bean genome 28,326 gene models 35,881 transcripts predicted average total length kbp protein length 413 amino acids protein consistent with current gene annotation P. vulgaris Gene functional annotations recognized for 21,642 (76%) identification species terms response stress metabolic processes transport anatomical structure development signal transduction cellular component assembly homeostasis 19,554 gene models annotated pathways Kyoto Encyclopedia Genes functional annotations assigned 22,634 (80%) gene models Gene expression evidenced RNA-seq dataset 26,295 (93%) gene models orthologs with P. vulgaris synteny blocks identified 22,180 (78%) gene models orthology direct evidence 27,029 (95%) annotated models remaining 416 show 50% protein sequence similarity with P. vulgaris genes outside synteny blocks remaining 3% paralogs functional annotations from other species.QTL mapping agronomic traits Lima biparental population UC Haskell–UC 92 Determinacy three quantitative traits (flowering time hundred-seed weight cyanide content screenedNine trait loci identified biparental population One major QTL long chromosome Pl01 explaining 78% phenotypic variation peak LOD score first marker 20 cM 20 Mbp gap pericentromeric Pl01 likely causative gene ortholog Arabidopsis gene TFL1 common bean ortholog PvTFL1y mapped at 45Mbp chromosome Pv0132 identified ortholog PlTFL1 Lima bean at 41Mbp Pl01 flowering time transgressive segregation observed earlier later progenies UC 92 UC Haskell parents major QTL on chromosome Pl01 likely causative gene PlTFL1 QTL 30% phenotypic variation other genes influence flowering time transgressive segregation observed seed weights below small-seeded parent UC Haskell not larger consistent with results shift towards smaller-seeded segregants Four minor QTL 28% phenotypic variation one chromosome Pl10 11% variation major QTL cyanogenesis floral bud tissue chromosome Pl05 explained 93% phenotypic variation two 97.5% total phenotypic variation UC 92 variety show measurable cyanide content UC Haskelltransgressive segregation cyanide content above levels UC Haskell three QTL showed epistatic interactions UC 92 allele QTL Pl05 cyanogenic glucosidase prevented expression QTLs Pl08 Pl10 Fig. 6) significance interval QTL cyanide Pl05 includes sequence glucosidase homology cyanogenic glucosidase white clover36.Evolution paralogs orthologs speciation proteins genes aligned 3499 paralog clusters gene families genome evolution Classification paralog relationships interchromosomal synteny analysis revealed 1647 genes paralogs ancient whole-genome duplication events Chromosome pairing paralogs consistent genome P. vulgaris23 Fig. 1) Intrachromosomal duplication events identified 7285 genes removing highly repeated genes 5849 genes involved duplication Figure 2a Ks values smaller than whole-genome duplication (WGD) paralogs (p-value < 10−15 duplications more recent Protein evolution among paralogs assessed Ka/Ks ratio Ka/Ks values intrachromosomal duplications larger than WGD paralogs-value < 10−15 duplications diverging faster than WGD paralogs12% local duplications rapid sequence divergence Ka/Ks ratios above 1. enrichment genes high Ka/Ks ratios interconnected processes immune response cell death communication signaling findings consistent Qiao et al plant species processes duplications include lipid transport metabolism lignan. 2Comparative genomics P. lunatus P. vulgaris Ka/Ks statistics P. lunatus G. max paralogs orthologs V. unguiculata Whole Genome Duplication Sample sizes gene pairs Middle lines medians box limits first third quartiles Lines Q1 minus 1.5 to Q3 + 1.5*IQR Chromosome synteny P. lunatus P. vulgaris structural rearrangements P. lunatus labeled Pl01-Pl11 P. vulgaris Pv01-Pv11 Number homologs resistance genes by chromosome Radial tree diagram genetic variability LRR resistance genes blue domains NB-ARC LRR purple TIR NB-ARC LRR orange TIR LRR Source data filecompared genome Lima bean with common bean v1.023 orthologs Orthologs identified for 25,564 (94%) P. vulgaris genes 26,009 (92%) P. lunatus genes high collinearity observed between P. lunatus P. vulgaris genomes (Fig. 2c). important structural events inversion chromosome Pl10 large translocation pericentromeric region Pv02 Pl02. events include 5Mbp inversion Pl03 10Mbp inversion Pl07 complex translocation Pl09. rearrangements in Pl02 Pl10 confirm pericentromeric inversions rearrangements could related to reproductive between Lima bean common bean39.Figure 2a shows Ks distribution of 22,180 orthologs between P. lunatus P. vulgaris centered around 0.05. half average paralogs recent 13 MYA) WGD event G. max genome41 speciation between P. vulgaris P. lunatus occurred around 6 MYA date close to age Phaseolus crown clade B vulgarislunatus estimated evolutionary rates chloroplast trnK locus (5.0 ± 0.7 MYA older than date ITS/5.8 S sequences (3.4 ± 0.4 MYA genomes Lima bean Vigna separation Phaseolus Vigna before WGD G. max 15 MYA date older than Delgado-Salinas et al Vigna sensu lato crown clade (9.1 ± 1.0 MYA chloroplast trnK sequences Protein evolution between orthologs assessed Ka/Ks ratio speciation separating P. vulgaris P. lunatus distribution Ka/Ks below 1 gene coding sequences evolve under purifying selection (Fig. 656 gene pairs synteny blocks rapid sequence divergence Ka/Ks values larger than 1. Functional enrichment genes shows ontologies metabolism aminoglycan chitin lignan genes increased expression values development pod.Orthologs genes agronomic characteristicsGenes predicted ortholog relationships traits crops identified Lima bean orthologs of 30 genes 27 common bean Traits nutritional herbicide plant abioticseed coat color growth habit consumer preferences Lima bean common bean breeding marketing included 1917 genes 11 chromosomes related to resistance biotic stresses bioinformatics analysis LRR domains disease resistance toll/interleukin-1 receptor leucine zipper coiled-coil nucleotide-binding site ARC serine–threonine kinase WRKY (Fig. 2d Serine–threonine protein kinases abundant disease resistance genes domain in 91 genes leucine zipper domain 74 CC domain 11 genes low number CNL (CC-NBS-LRR) genes dicots49 98 TNLs (TIR-NB-LRR) identified in annotated gene models 631 genes contained LRR 151 NB-ARC domain 91 both domains disease resistance genes localized to chromosomes Pl02 Pl04 Pl08 Pl10 Pl11 genes with LRR domain clustered discrete regions genome Fig. 9) includes domain arrangements LRR NB-ARC-LRR TIR-NB-ARC-LRR TIR-LRR clustering correspondence between chromosome clustering sequence similarityProteins with domains NB-ARC-LRR formed large cluster five nested into TNL group three proteins with TIR and LRR domains Common bean orthologs identified for predicted genes related to biotic stress resistance located on same chromosomes mainly collinear motifs conserved between gene sequences species genomic positions genes correspond with resistance loci important diseases found clusters of genes resistance loci for angular leaf halo blight52 bean golden yellow mosaic other viral diseases rust mildew23.Population structure analysis reveals genetic clusters in Lima beanWild Lima bean originated in northern Andes expanded to Andes Mesoamerica11 reached distribution northern Mexico to Argentina differentiated into three gene pools (MI MII AI non-overlapping distributions domesticated species twice in Mesoamerica Andes genetic structure combined data from 270 Lima bean with data for 212 additional samples to increase variation From 116,030 biallelic SNVs, 12,398 selected for diversity analysissamples sequenced study increased three major gene pools geographic sampling United States Mexico Colombia Fig statistical heuristic clustering analyses Neighbor Joining discriminant analysis Bayesian clustering applied optimal clusters BIC between K = 5 and K = 6 STRUCTURE results suggest optimum K 6 results supported three major wild gene pools (MI MII AI showed evidence two large novel clusters (Fig. 3a Data 5 Figs 12–14) domesticated MI accessions separated from wild single cluster agreement with Mesoamerican landraces complete sampling wild accessions central Andes Colombia supported genetic differentiation Andean wild gene pool AII gene pool AII group restricted distribution on eastern slope Andes central Colombia Cundinamarca Boyacá one accession in Peru 13, five wild accessions from Guatemala one Honduras one Chiapas (Mexico clustered within Andean gene pool AI subcluster detected within wild gene pool MII accessions from Peninsula of Yucatan Mexico northern Guatemala Costa Rica northern Colombia.Fig. 3Lima bean genetic diversitySTRUCTURE analysis genetic variability 482 wild domesticated Lima bean accessions Americas organized south-north geographic pattern DOM domesticated MEX Mexico GUA Guatemala CR Costa Rica COL Colombia Classification gene pools MI MII AI AI Linkage disequilibrium decay subgroups wild domesticated accessions DOM domesticated DOM-AND Andean landraces Radial clustering 482 accessions fineSTRUCTURE gene pools colors wild northern Mexico pink southern-western South America dark Mexico Central America light Yucatan Peninsula yellow AII red AI green MII Yucatan Central America Colombia light green MII southern central Mexico Distribution chromosomal segments gene pools 15 wild domesticated accessions marked Lima bean chromosomes labeled Pl01-Pl11 Source data Lima bean inbred species heterozygosity rates per accession low facilitates inference analysis haplotypes Figure 3b comparison linkage disequilibrium (LD) decay variants same chromosomeLD decays faster in wild than domesticated due to population structure LD remains high over 0.3 for wild over 0.4 domesticated even at distances than 1 Mbp LD analyses wild MI MII gene pools decays to basal levels at 100 kbp domesticated MI MII AI decays to background levels at 500 kbp LD decayed faster in MI MII landraces than AI landraces with higher genetic diversity in Table 5) reconstructed haplotypes long blocks linkage disequilibrium applied software fineSTRUCTURE57 linkage disequilibrium for population structure identified larger number populations 181 revealed genetic relationships among inferred populations radial tree Fig. 3c population at multiple structuring levels major clusters corresponded to major gene pools detected by STRUCTURE subgroups within wild MI MI wild MII gene pools related to geographywild MI accessions separated two subgroups geographical origin (purple cluster 55 accessions ten populations northern-western Mexico Sinaloa to Colima (pink cluster 34 accessions 13 populations southern-western Mexico Michoacan to Oaxaca accessions separated two subclusters one (light green cluster 66 accessions Mexico southern Guatemala (dark green 49 northern Guatemala Costa Rica Colombia nine Peninsula Yucatan domesticated accessions MI landraces separated three subgroups geographical origin (dark blue cluster 43 accessions 13 populations Mexico Central America 12 from States Colombia (medium blue cluster 51 accessions nine populations South America 8 United States one Mexico third (light blue cluster 59 accessions five populations 19, 60 80 88 101 Yucatan Peninsula Mexico third subgroup observed STRUCTURE results at K = 7 (Fig. 3a). landraces five populations Yucatan Peninsula tendency group accessions by variety seed shape. Mayan nomenclature Martinez-Castillo et alpopulation 88 14 accessions Bacalar-ib Chak-saak Chak-ib Bayo-ib small flattened seeds Sieva Population 101 24 accessions Putsica-Sutsuy Mulicion blanco rojo Box-ib Pool-santo Kan-ib Morado-ib Yete Boch ib Kolbihi Chak ib small rounded seeds Potato cultigroup). Population 60 13 accessions Sac-ib Chak-chi Bayo-ib Mejen-ib Bacalar Madza-kitam seed intermediate Population 80 four accessions Box-ib small purple-black semi-flattened seeds Population 19 4 accessions Chak-chi white-red small seeds intermediate intermediate forms farmers Yucatan Peninsula grow seven landrace varieties Table 5 diversity statistics gene pools wild domesticated accessions Observed heterozygosity lower expected Lima bean (Bartlett’s K-squared = 18257 df = 1 p value < 2.2e-16 paired-t test t = 165.df = 12397 p value < 2.2e-16). wild clusters most diverse Mesoamerican MI MII gene pools (HE = 0.128 0.133 least diverse Andean AI = 0.040) AII = 0.052) genetic diversity seven wild accessions from Chiapas Guatemala Honduras AI gene pool low (HE = 0.053) Domestication genetic diversity landraces sample reduction 25% drastic Mesoamerican domestication (MI gene pool 55% Andean domestication no reduction genetic diversity MI landraces Yucatan Peninsula (HE = 0.029) lower other MI landraces (HE = 0.065) late introduction crop Yucatan Peninsula late Maya FST distances wild gene pool related Mesoamerican landraces MI (FST = 0.33) central-western Mexico Andean landraces most closely related AI (FST = 0.21) Andes Ecuador-northern Peru FST values AII closely related to Mesoamerican gene pools 0.56 to 0.65) Andean gene pool AI (FST = 0.86) close relationship shown by Caicedo et al.61 Toro et al.62 electrophoreticSTRUCTURE results suggest relationship gene AII with MII (Fig. analysis fineSTRUCTURE shows AII related to AI (Fig. FST values showed high genetic differentiation among gene pools haplotype introgression analyses clustered accessions within gene pools 103 instances chromosomal segments in 35 accessions (9 wild 26 domesticated represent contributions different gene pools 103 segments varied 1 to 53 Mbp six accessions more than 25% genome length classified as admixed Figure 3d shows 58 chromosomal segments larger than 5 Mbp in 15 accessions segments represent contributions between Mesoamerican gene pools (MI MII or Andean gene pools rarely Andean contributions may contact wild domesticated different origin introduced domesticated populations contributed to wild five MII wild accessions Chiapas Campeche Guatemala northern Colombia carry segments from MI domesticated accessions wild MII accessions Two AII wild accessions central Colombia carry segments derived from AI accessions 18 MI domesticated accessions segments derived from MII gene poolintrogression from MII wild accessions into MI landraces Mexican Veracruz Chiapas Costa Rica El Salvador Caribbean coast northern Colombia detected four MI landraces from United States one Colombia with chromosomal segments derived from Andean AI landraces in Colombia Ecuador MI AI landraces coexist found two AI landraces G25172 G26184 in Ica, Peru foreign haplotypes belong to MI gene pool interbreeding Andean Mesoamerican landraces early introduction of Mesoamerican landraces in coastal Peru suggested by pod remains ceramics Guanape Period Cupisnique Period in Huaca Prieta, recent introductions single 2.6 Mbp MII segment in chromosome Pl07 shared by six MI landraces Colombia second 2.2 Mbp MII segment chromosome Pl08 shared by seven MI landraces Central America Colombia.Gene expression during pod developmentReduction of pod dehiscence key domestication in Lima bean agronomic genetic regulation-seq experiment expression levels pod elongation before seed filling in one wild one domesticated accessiondevelopmental stage T1 pod valves visible corolla attached detached). stage T2 pods reach maximum length weight seeds before seed filling analysis expression values from RNA-seq reads previous study66 clustered replicates one outlier removed for downstream analysis Fig. 16). Differential expression) analyses revealed 4275 genes differential expression across developmental stages wild domesticated accession Data 7) Figure 4a heatmap normalized expression values genes Hierarchical clustering five seven gene clusters different expression patterns.Fig. 4Gene expression pod developmental stages T1 T2.a expression values left dendrogram unsupervised hierarchical clustering Expression trajectories gene PlPDH1 domesticated wild accession across two developmental times Number genes differential expression between wild domesticated accession times Concept map functional categories for genes expressed second developmental time wild accession Source data provided Source Data file PDH1 gene increased expression between T1 T2 (Fig. 4b). expression second stage 2-fold higher wild accession domesticated accession difference not significant low expression valuePDH1 gene formation fibrous lignified cell layers inner outer parenchyma pod torsion force pods shattering genotypes67 identified strong pod dehiscence QTL on chromosome Pv03 in common bean65 ratio gene expression developmental stages domesticated accession increased expression larger genes wild T1) increased expression larger decreased expression T1 T2 (Fig. explained by 1500 genes increasing expression between T1 T2 wild accession Genes increased expression T1 T2 enriched for cell wall biogenesis organization biosynthesis polysaccharides xylan lignin seed genes expression between T1 T2 wild enriched for metabolism-related genes lignan chitin aminoglycan fruit ripening process (Fig Enrichment chitin metabolism genes generated by seven genes at 38 Mbp chromosome Pl09. Lignan metabolism genes enriched nine genes increased expression at 9 Mbp Pl11 Genes more expressed domesticated accession at T2 show enrichment functions reproductive system development fruit genes reduced expression between T1 T2 enriched for developmental processes cuticle development metabolism compounds cyanogenic glycosidefindings expected plant completed formation pod filling seed conducted large collaborative effort comprehensive genetic genomic information for Lima bean resulted in information content organization function chromosomes sequences evolutionary relationships population genomics wild cultivated accessions inheritance of agronomic traits analysis revealed genetic structure wild domesticated Lima beans large distribution range Americas insights genetic variability in agronomically relevant traits Knowing genes controlling traits advantage for breeding programs accelerate development release improved Lima bean varieties achievements chromosome-level genome assembly for P. lunatus comprehensive genotyping genetic variability species effort high-quality reference genome sequence rewarded assembly functional prediction of large clusters genes related traits predictions experimentally validated information useful prioritize genes experimental validation Genes resistance to biotic stresses show large nucleotide diversity good correspondence with common bean resistance genes variability of disease resistance genes key defense mechanisms to functional redundancy favoring durability resistance local gene duplication observed in genes metabolism xyloglucan chitin regulation development related to yield traits seed weight flowering timeevolutionary history plant genomes shaped by WGD local duplication events evolutionary rates gene families consequences in gene expression function38 full understanding possible through high-quality genome sequences29 Phaseolus species synteny analysis with P. vulgaris confirms chromosome conservation provides detailed view of five major rearrangements work to genome assembly resequencing full reconstruction evolution legume family characterization potential convergent evolution processes domestication events phylogenetic analyses origin of wild Lima bean in Andes Ecuador-northern Peru during Pleistocene populations experienced range fragmentation reduced genetic diversity11 Andean wild populations less diverse than Mesoamerican more sampling needed diversity wild common beans reduction of diversity documented in Andean populations attributed to rare long-distance dispersal events from Mesoamerican to Andes70 identification of wild Lima bean cluster (AII) in Andes central Colombia supports dispersal species from south to north may source of alleles not found characteristic phaseolin type (M8) observed in gene pool study bottleneck in Mesoamerica more severegenetic diversity lost during domestication re-gained by gene flow from wild to domesticated low landrace accessions further studies genome resequencing needed to assess role introgression in genetic diversity domesticated Lima bean avoid biases low sampling missing data GBS experiments higher number samples genetic markers in-depth information on genetic MI gene pool relation to geography sources bias due to low SNP density missing data GBS MI wild cluster two subclusters northern-western southern-western Mexico Mesoamerican MI landraces three subclusters Mexico-Central America South America Peninsula of Yucatan landraces Peninsula of Yucatan form separate subcluster genetic diversity lower than landraces hypothesis Lima bean late of Mayan communities Peninsula of Yucatan60 Lima bean threat of genetic genetic differentiation among wild gene pools domestication bottleneck in Mesoamerica threat of genetic erosion Yucatan Peninsula for conservation biparental population UC Haskell–UC 92 guided genome assembly tool for investigation drivers domestication traits through QTL mappingcommon bean72 PlTFL1 gene associated with QTL determinacy flowering time determinacy earlier flowering terminal meristems into reproductive determinacy trait common in Andean gene contributed by Andean parent UC 92 raises questions Mesoamerican gene pool origin trait? due to independent mutation or introgression from Andean? genome resequencing Andean Mesoamerican necessary three quantitative traits in UC 92–UC Haskell population showed transgressive segregation varying seed weight showed transgressive segregation for weights below small-seeded parent UC Haskell not larger weights difficulty developing large-seeded improved progenies selfing solutions backcrosses to large-seeded parent alternative phenotypic distribution for HSW discriminated by allelic combinations three marker alleles of large-seeded parent (UC 92) for seed weight QTL corresponds to heaviest seeds vice-versa for small-seeded parent (UC Haskell). indirect selection markers may shift breeding populations towards heavier seedscyanogenesis combination QTL alleles highest levels cyanogenic glucosidase Pl05 (UC Haskell)–Pl08 (UC 92)–Pl10 (UC Haskell). combination consumer safety insect resistance used RNA-seq gene expression two pod developmental stages one wild (shattering one domesticated (non shattering Lima bean accession differentially expressed genes accessions detected stages (T1 T2 maximum seed filling). identified differential expression in PDH1 gene pod dehiscence ‘dirigent’ polymerization lignin monomers soybean lignification pod walls torsion force reduced expression reduced pod dehiscence in domesticated types P. vulgaris PvPDH1 transcribed in pod tissue75 associated with pod dehiscence domesticated race. to arid conditions northern Mexico65 Anatomical histological analyses shattering genotypes have extensive lignified wall fiber layer (LFL) in pod walls non-shattering genotypes LFL deposition reduced or absent PlPDH1 gene increases expression in wild accession Lima bean at T2 compatible with results common bean PlPDH1 good candidate for further investigationinformation from differential expression genes development metabolism of xylan pectin to flowering time seed weight patterns expression observed for genes with processes metabolism chitin cyanogenic glycoside cuticle development response to auxin cell wall biogenesis fruit ripening related to physiological characteristics seed quality traits cyanide content importance of Lima bean food security crop potential protein source for climate change work basis for future studies in Phaseolus species breeding contribution to Phaseolus genomics five domesticated species.MethodsSequencingLima bean genomic DNA extracted from young leaves two-week-old seedlings G27455 concentration integrity High molecular weight (HMW) DNA for Genomics extracted with Qiagen MagAttract HMW DNA Kit SMRTbell Template Prep Kit used DNA randomly fragmented buffer Covaris g-TUBE devices purification with AMPurePB magnetic beads after concentration Fragments greater than 3 kbp damage repair end-repair blunt end ligation with hairpin adapters Libraries sequenced on PacBio Sequel platform at Novogene Corporation Inc10X Genomics DNA library one μg Lima bean DNA prepared Chromium Genome HT Library Gel Bead Kit V2/10x Genomics Pleasanton CA 500–700 bp libraries quantified Qubit 2.0 fluorometer Scientific Waltham MA quantitative PCR size distribution analyzed Agilent 2100 Bioanalyzer Santa Clara libraries sequenced Illumina HiSeq Platform San Diego paired-end 150 run trifoliate leaves two-old seedlings collected frozen liquid nitrogen DNA extraction method genetic diversity population structure library 1.0 μg DNA NEBNext DNA Library Prep Kit BioLabs Ipswich DNA fragmented 350 bp ligated NEBNext adapters enriched PCR library analyzed size distribution Agilent 2100 Bioanalyzer quantified real-time PCR Libraries sequenced Illumina HiSeq Platform paired-end 150 run (2 × 150 bases insert size 450 bp.RNA sequencing novo transcriptome assemblyPlants G25230 (Mesoamerican G27455 grown greenhouse Centro Internacional Agricultura Tropical Cauca RNA extracted leaves pods flowers specific protocoldifferential gene expression pod dehiscence RNA extracted wild domesticated accession two pod stages three replicates pod elongation seed filling T1 pod valves visible flower corolla T2 pods maximum length weight before seed filling strong lignin deposition dehiscence zone wild RNA samples quantified Nanodrop 2000 quality assessed Agilent Bioanalyzer 2100 agarose gel electrophoresis Samples 260/280 ratio absorbance 1.9–2.0 RIN Integrity Number) values above seven selected mRNA enriched oligo(dT) beads randomly fragmented cDNA synthesized mRNA templates RNA-seq library preparation size selection PCR enrichment Sequencing paired-end 150 run Illumina HiSeq Platform quality raw reads evaluated fastQC v.0.11.2 low-quality filtered Trimmomatic v.0.3680 removing 3% De novo transcriptome assemblies obtained software Trinity v.2.4.081 compared 1440 single-copy orthologs OrthoDB v.9.1 v.2expression evidence gene annotation downloaded RNA-seq reads previous NCBI Sequence read archive accession PRJNA275266) fastq-dump SRA toolkit v.2.9.4-tools/wiki).Sequencing assembly scaffolds chromosome-level assembly Lima bean genome combining reads four sequencing protocols Table 7) 97.6 Gbp raw data 157x haploid genome size estimated 622 Mbp27 genome assembly four steps de novo assembly PacBio reads Canu v.1.6 default draft assembly 496 contigs N50 5.5 Mbp total length 542 Mbp Fig. initial analysis reads data 12 potential misassemblies polishing high base pair quality paired-end Illumina data correct base pair errors Reads aligned 508 contigs v.2.3.584 variants called command FindVariants NGSEP 3.3.285 parameters -runRep -minMQ 0 -maxBaseQS 30 -minQuality 40 -h 0.0001 predictions STRs tandem repeats Assembly errors homozygous alternative variants corrected command VCFIndividualGenomeBuilder NGSEP process repeated four timesoptions -runRep -runRD repetitive regions alignments CNVs read depth signal 439 contigs (86%) identified repetitive predicted repeats over 60% total length total length 194 Mbp (36% assembled length). remaining 69 scaffolds span 348 Mbp longest contigs (Supplementary Fig. 20). linked reads aligned to polished contigs using bwa v.0.7.1787 scaffolding 10x protocol reads same barcode sequenced from same initial molecule barcode provide evidence linkage built undirected graph contig ends vertices evidence linkage edges clustering algorithm Salsa Hi-C 164 contigs in 40 scaffolds dense genetic map ten thousand variants built analyzing GBS data from F8 biparental population Linkage disequilibrium calculated r2 value greater than 0.8 evidence linkage Connected components graph contigs vertices linkage evidence edges expected 11 linkage groups scaffolding process 206 contigs to 512 Mbp sorted oriented produce final 11 pseudomolecules chromosomes species GBS reads reanalyzed against assembled pseudomolecules verify structural consistency assembly identify recombination breakpoints sampleRNA-seq genome annotation comparisons expression repetitive elements RepeatMasker v.4.0.530 library P. vulgaris Gao et al genome assembly masked replacing nucleotides with N characters annotation annotation initio predictions cDNA sequences homology approaches Maker pipeline v.2.31.989 RNA-Seq data filtered Trimmomatic v.0.36 low sequences used 57,742 de-novo transcript assemblies Trinity pipeline81 36,995 putative protein sequences from common bean Maker annotation RNA-seq clean reads processed Tuxedo pipeline90 HISAT2 v.2.1.091 align reads Lima bean genome StringTie v.1.3.592 predict transcript annotations aligned reads annotations merged script NGSEP Suspiciously long genes abnormally high number distribution curated sequence similarity with common bean validation base pair quality genome assembly 1440 single-copy orthologs OrthoDB v9.1 database) v.2 Tuxedo pipeline expression levels each gene sample HISAT2 v.2.1.0 reads Lima bean genome StringTie v.1.3.5 matrix read counts differential expression analysis DeSeq2 v.3.193 comparisons developmental stages linear model design evaluated Significant differential expression predicted if comparison p value below 0.05 log-fold change above two reference bias expression levels transcripts estimated from raw reads Salmon v.1.2.194 Differential expression analysis assessed DeSeq2. trends consistent levels Fig. 21).Functional annotation gene models transcripts Trinotate pipeline v.3.1.195 blastx blastp searches NCBI Blast v.2.10.0 performed against UniProt database cDNA amino acid sequences best hit retrieved if e value below 0.001 hmmscan tool HMMer v.3.3.196 identify conserved domains Pfam database Results combined Trinotate database final annotation report quality gene models automated curation orthology with P. vulgaris expression transcript per million (TPM) measure greater than 0.5 tissue functional annotation evidence gene 10,263 annotations filtered genes one type evidence proteins length over 200 amino acids paralog direct evidence.Paralog identification comparison with annotated genomes GenomesAligner NGSEPcommand builds FMIndex proteomes performs amino acid searches non-overlapping k-mers each sequence homology relationship called if percentage matching query k-mers larger than parameter GenomesAligner executed with k-mer length k = 5 minimum percentage p = 20. WGD paralog relationship called if two genes have ten paralogs in different chromosomes half neighbors share syntenic homology relationship.Functional enrichment of gene ontology (GO) terms for performed Bioconductor v.2.36.097 Fisher exact test set compared against Lima bean genome Visualization GO terms using REVIGO98 GO Cytoscape v.3.8.199.Identification of genes for biotic disease resistance based functional annotation orthology prediction complete list gene models screened for domains resistance-genes NCBI (BLASTp), Pfam eggNOG/GO/KEGG domains were CC, NB-ARC TIR LRR WRKY LZ protein kinase domain arrangements-Leucine-rich repeat) gene family subfamilies classified for 11 Lima bean chromosomes Genomic locations genes protein sequences derived from genome assembly structural annotationmap 11 chromosomes NBS-LRR genes generated MapGene2Chrom v.2.1100 sequence alignment proteins v.3.8.31101 tree topology gene diversity analysis-joining approach MEGA X v.10.8.1102 iTOL v.4.4.2103 analyses UC 92 UC Haskell recombinant inbred biparental recombinant population University of California Davis 121.763049° crosses UC 92 UC Haskell UC Haskell small-seeded Mesoamerican UC 92 large-seeded Andean Crosses made greenhouse 2012 progeny seeds grown hybrid verification F1 hybrids verified polymorphic codominant Pvctt001 sequence repeat) marker PCR protocol Taq DNA Polymerase ThermoPol Buffer PCR analyzed 2% agarose gel 1x TAE buffer population advanced F8 stage single-seed descent Leaf tissue parents RIL collected two-old seedlings DNA extracted DNEasy Plant Mini Kit RIL population segregation agronomic traits germination rate flowering time inflorescence position plant height habit pod position density yield biotic stressLeaf tissue DNA extraction sampled from two-old seedlings 238 RILs F8 generation collected 96-well plates ice lyophilized 24 h DNA extracted protocol Lima common bean104 DNA confirmed samples 260/280 absorbance ratios above 1.8 NanoDrop Lite DNA quantified Quant-iT PicoGreen dsDNA Assay Kit 100 ng transferred PCR plate GBS barcode libraries adapters CviAII enzyme prepared bean104 CutSmart buffer unique barcode added plates incubated PCR machine 2 h 25 °C T4 (10x) buffer T4 ligation added PCR machine 1 h 22 °C 30 min 65 °C Seven microliters sample added petri dish mixed transferred Eppendorf tubes Binding buffer isopropanol added mixed Eight-hundred microliters transferred GeneJET purification column centrifuged 60 s eluted water DNA quantified QUBIT dsDNA HS Assay Kit GBS sequencing Two genomic libraries prepared 144 barcode CviAII enzyme 240 unique genetic lines sequenced sequenced SR100 protocol Illumina HiSeq flow cell University of California, Davis Genome Centerdata de-multiplexed contamination removed Demultiplex command NGSEP reads aligned contigs default parameters MultisampleVariantsDetector variants genotypes GBS diversity 51,897 SNVs 2927 biallelic indels 758 biallelic STRs identified in 183 contigs genotype quality less than 30 minor allele frequency less than 0.3 heterozygosity rate larger 0.05 variants less than 100 genotyped Missing genotypes imputed parental assignment obtained ImputeVCF NGSEP UC92 -k -c 0.003 -ip Recombination events between markers identified centimorgans estimated five generations crossover one crossover per chromosome.Ninety-three RILs 10,497 polymorphic SNP markers genetic map ASMap v.1.0.4 R/qtl v.1.44 less than 20% missing genotypes excessive recombination rates than 50% missing genotypes removed Linkage groups formed ‘mstmap’ p < e-8 genetic distances calculated ‘Kosambi’ Linkage groups merged chromosome recombination frequencies recalculated chromosome lower p-values Recombination rates calculated ‘MareyMap’ 200Pericentromeric regions defined when recombination rates exceeded 2 Mbp/cM Chromosome numbering followed P. vulgaris chromosomes one-to-one correspondence justified high mapping for plant habit seed weight days to first cyanogenesis UC 92–UC Haskell RIL population R/qtl genome-wide scan for single QTLs ‘scanone’ function extended Haley-Knott regression traits controlled single QTL composite interval mapping ‘cim’ function Kosambi mapping function 1000 permutations position LOD score QTLs above 95% significance threshold multiple QTL ‘makeqtl’ ‘fitqtl’ functions optimal multiple QTL model calculate LOD scores phenotypic variation QTL to first collected experiment two Davis California 2018. recorded plants floral bud open ANOVA linear mixed model genotype performed for days first flower Phenotypic characterization plant habit seed weight delayed senescence on single plants greenhouse.Cyanide quantification floral bud tissue collected triplicate subsamples Davis California 2018. samples kept on dry ice weighed 96-well plates frozen at −80 °CCyanide quantification Fiegl-Anger Tissue samples thawed room 30 min before Fiegl-Anger paper 30 min Standards hydrogen cyanide created 0 25 50 75 100 250 500 750 1000 nM potassium cyanide exposed Fiegl-Anger paper 30 min paper scanned blue absorbance intensity calculated ImageJ software ‘readplate2’ plugin -log(Mean/255) standard regression curve calculated HCN concentrations nanomolar concentrations nM/min rate 30-min interval linear models ANOVAs RIL population haplotype QTL peak chromosome Pl05 linear model floral bud cyanogenesis genotype location treatment fixed effect factors diversity population structure GBS library construction five samples leaf tissue leaves plants collected samples frozen stored −20 °C processing DNA Vega-Vela Chacon Sanchez DNA quality checked Nanodrop 2000 analyzed 1% agarose gel electrophoresis DNA no degradation ratio 260/280 above 1.8 selectedDNA libraries per accession generated mixing DNA five individuals Nanodrop 2000 electrophoresis calibration lambda DNA concentrations 25 50 100 ng/μL), DNA restriction ApeKI enzyme ligated to adapters 94 unique barcodes library sequenced Illumina Hi-Seq2000 preparation sequencing Australian Genome Research Facility (Melbourne 482 Lima bean accessions from International Center for Tropical Centro de Investigacion Cientifica de Yucatan—CICY (220 analyzed Data 215 domesticated 267 wild Wild accessions natural geographic range domesticated countries Americas sampling wild accessions MI gene pool gene pool AII central Colombia Sequenced accessions from CIAT available germplasm collection Centro de Investigacion Cientifica de Yucatan upon request sequences de-multiplexed by barcode Next Generation Sequencing Experience Platform (NGSEP) v.3.3.285 fastq files mapped to Lima bean reference genome default parameters ID code each accession84 Variants identified genotyped MultisampleVariantsDetector command NGSEPparameters for SNP calling 50 maximum per start position heterozygosity rate 0.0001 minimum genotype quality score 40 Phred 0.9999 probability genotype Initial filtering FilterVCF command NGSEP Exclude variants from repetitive regions not-assembled scaffolds variants less than 100 of 482 accessions variants with minor allele frequency less than 0.01. generated 116,030 biallelic SNVs 7517 biallelic indels STRs 22,138 multiallelic variants (96.6% triallelic SNVs 40% missing data Further filtering for downstream population genetics analyses disequilibrium (LD) measured as r2 correlation among alleles SNP loci average r2 against 10 100 kbp intervals 11 chromosomes Phaseolus lunatus TASSEL v.5111 analysis conducted filtered wild domesticated samples for Mesoamerican Andean gene pools Results plotted in R software adapted code clusters in Lima bean used algorithmic model-based approaches built multilocus-genotype matrix 482 accessions (267 wild 215 domesticatedraw VCF file filtered biallelic SNVs distance more than 10 bp from variants genotyped quality 40 in 450 individuals minor allele frequency larger than 0.03. 12,398 SNVs retained after filters 5.3% missing data genotype calls read data present for 75% SNPs less than 90% for six samples (Supplementary Data algorithmic approaches with software Darwin v.6.0.021112 R package Adegenet v.2.1.2113 Darwin pairwise distance matrix calculated matching alleles per locus genetic structure used three methods principal coordinate analysis tree method discriminant analysis of principal variance function ‘find.clusters’ optimal number clusters 500 principal components retained four discriminant functions used conserved variance 0.95 NJ tree structure 1000 bootstrapped matrices obtained tree visualized using iTol103 software STRUCTURE v.2.3.4115 infer population structure filtered VCF file converted to STRUCTURE ConvertVCF command NGSEP 200 STRUCTURE runs executed varying number clusters (K) from 1 to 10 20 repetitions for each K value 10,000 burn-in sampling iterations executedmain extra parameters left default exception option ‘ONEROWPERIND’ set to 1 VCF converter NGSEP Stability Markov chain assessed likelihood clustering solution best likelihood selected each K value between 2 8 cluster numbers reorganized stability values (Fig. 3a). Evanno test optimal population structure assessed software fineSTRUCTURE v.4.1.057 builds relationship matrix haplotype similarity Markov chain group configurations tree topologies consistent VCF file filtered biallelic SNVs distance more than 10 bp genotyped quality 40 in 300 individuals MAF larger than 0.02 observed heterozygosity) below 0.5 28,753 SNVs retained filters 11% missing data 332,817 (2.4%) heterozygous data Imputation missing data statistical haplotyping Beagle 5.0117 phase states 50 imputation states 400 effective population size 100,000 VCF converted to format fineSTRUCTURE VCFConverter NGSEP four steps pipeline fineSTRUCTURE executed default parameters genetic distances inferred from map biparental population.Basic diversity statistics observed heterozygosity expected endogamy global FST pairwise FST118 estimated R packages Adegenet113 Hierfstat v.0.5119 dataset STRUCTURE Difference between HO HE assessed Bartlett test paired t test significance population differentiation tested likelihood ratio G-statistic120.Introgression analysis ‘IntrogressionAnalysis’ command NGSEP input VCF same STRUCTURE input file population assignments genetic populations inferred STRUCTURE K = 4 MI MII AI AII populations samples clear assignments included analysis executed default parameters (non-overlapping windows 50 SNPs introgression events filtered events more than one contiguous window similarity score between haplotype consensus haplotype background population below 15. D-statistic ANGSD package v.0.93121 significance introgression events GBS reads common bean aligned to P. lunatus genome outgroup Events Z-scores larger than 2 retained introgression MII sample JMC_1097 retained spanned 53Mbp over four chromosomes two introgressions 2Mbp 36 Mbp Pl07 4.6Mbp Pl08 retained consistently in six seven accessions D-statistic significance validated by construction neighbor-joining clusters regions (Supplementary Fig. 22).Reporting SummaryFurther information research design Nature Research Reporting SummarySupplementary informationSupplementary InformationPeer Review FileReporting SummaryDescription Additional Supplementary FilesSupplementary Data 7 | 50.6 | 1.567908 |
10.1038/s41467-020-16203-x | PMC7210270 | Self-grooming is a frequently observed repetitive behaviour in rodents that is believed to contribute to post-stress de-arousal. The authors identified a previously unknown limbic circuit that includes the ventral lateral septum in rats and is involved in regulating stress-induced self-grooming. | Prolonged exposure to negative stressors could be harmful if a subject cannot respond appropriately. Strategies evolved to respond to stress, including repetitive displacement behaviours, are important in maintaining behavioural homoeostasis. In rodents, self-grooming is a frequently observed repetitive behaviour believed to contribute to post-stress de-arousal with adaptive value. Here we identified a rat limbic di-synaptic circuit that regulates stress-induced self-grooming with positive affective valence. This circuit links hippocampal ventral subiculum to ventral lateral septum (LSv) and then lateral hypothalamus tuberal nucleus. Optogenetic activation of this circuit triggers delayed but robust excessive grooming with patterns closely resembling those evoked by emotional stress. Consistently, the neural activity of LSv reaches a peak before emotional stress-induced grooming while inhibition of this circuit significantly suppresses grooming triggered by emotional stress. Our results uncover a previously unknown limbic circuitry involved in regulating stress-induced self-grooming and pinpoint a critical role of LSv in this ethologically important behaviour. | IntroductionStress in the form of emotional and physiological challenges is ubiquitous in everyday lives. Physical and emotional stressors can upset body homoeostasis that pertains to a steady milieu of physiological parameters as well as states of mind1–3. Generation of adaptive responses to stress involves the evaluation of real or perceived stressors, and the homoeostatic resolution via optimization of emotional and physiological adaptations1,4,5. Behavioural adaptations could include increased arousal, attention and vigilance1. On the other hand, maladaptive response to stress has been linked to the aetiology of anxiety, depression and a variety of other neuropsychiatric conditions6–8. Therefore, strategies evolved to cope with stress are essential for health and survival.Stress in animals, including human, often results in grooming and other repetitive behaviours such as circling and rocking9–12. These displacement activities are believed to have adaptive values. Indeed, self-grooming is a frequently observed repetitive behaviour in rodents that serve functions more than hygiene maintenance and thermoregulation. This behaviour may represent adaptive response to stress, or restraining force that prevents over-response to stress, such as post-stress de-arousal11–14. Thus, unravelling the mechanism of stress-induced self-grooming is highly valuable towards understanding the neurobiological basis of stress management.In the mammalian brain, a number of brain areas have been implicated in the generation of grooming behaviour. These include the basal ganglia13,15, brain stem13,16 and cerebellum13,17 that represent the downstream, mechanical motor pathways. On the other hand, some components of the limbic system, including the hypothalamus, amygdala and orbitofrontal cortex, participate in the regulation of grooming. For example, focal activation of specific hypothalamic nuclei could evoke robust grooming18–20, and control of self-grooming versus social behaviour by distinct amygdala neuronal subpopulations had been demonstrated21. Also, repeated stimulation of the orbitofrontal-striatal pathway could generate compulsive grooming22. Many previous studies that specifically link stress to grooming focused on the neuroendocrine system, namely the hypothalamic–pituitary–adrenocortical axis12,13,23–25, in accordance with the observation that central administration of some stress-related neuropeptides like corticotropin-releasing hormone (CRH), adrenocorticotrophic hormone and melanocyte-stimulating hormone could elicit grooming26,27, and CRH neurons orchestrate post-stress behaviours including grooming12. More recently, the neural pathways involved in different stress-related responses including grooming have begun to be revealed12,14,24,25. For example, a hypothalamic-septal pathway that mediates the influence of emotional states on grooming, escape and feeding behaviour was identified24. However, the relationship between stress and self-grooming is likely to be complex13,14,28, and the neural circuit basis of stress-induced grooming is still unresolved. In particular, the complete neural circuitry that conveys perceived stress, especially those with strong emotional component, that lead to grooming is yet to be uncovered.In this study, by identifying and dissecting stress-related neural circuit, we revealed a previously unknown limbic circuit linking the hippocampal ventral subiculum (VS), ventral division of lateral septum (LSv) and lateral hypothalamus that regulates stress-induced self-grooming. Optogenetic activation of this di-synaptic circuitry triggered delayed but robust grooming with patterns closely resembling those evoked by emotional stress. A distinct feature of activation of this circuit is its association with a clear positive valence, unlike many other studies on grooming behaviour. In addition, we found that the neural activity of LSv precedes and is necessary for emotional stress-induced grooming while targeted functional inhibition of this circuitry suppressed grooming caused specifically by stressful paradigms. Our results thus advance our understanding of the neural circuit basis of repetitive behaviour particularly relevant to the adaptation to emotional stress.ResultsActivation of ventral subdivision of LS triggers robust grooming behaviourWe began by examining the c-Fos expression in the rat brain limbic system after imposing body restraint, a well-known stressor29–33, to the animals for an extended period of time, viz 20 min (n = 4). This protocol induced significantly increased time spent in grooming by the animals (Fig. 1a). A robust c-Fos expression was found in the LSv. Notably, the c-Fos expression was mainly confined to the ventral LSv but not in the dorsal lateral septum (LSd) (Fig. 1b, left panel). Significantly increased c-Fos signals was found in LSv compared with the control animals not receiving body restraint treatment (Fig. 1b, right panel). To probe the consequence of activating LSv neurons, after micro-injection of AAV9-Syn-ChR2-eYFP into LSv (Fig. 1c), delivery of blue light onto LSv for 5 min could trigger robust grooming (Fig. 1d; Supplementary Movie 1). At the same time, although grooming behaviour dominated during light stimulation, it exhibited a characteristic delay of several tens-of-seconds while rearing-like arousal behaviour was frequently observed prior to the occurrence of grooming (Fig. 1e). We also confirmed that self-grooming rather than social grooming was induced by LSv activation by placing a littermate in the same arena during optogenetic stimulation (n = 3, Fig. 1f and Supplementary Movie 2).Fig. 1Activation of ventral subdivision of LS triggers delayed but robust grooming behaviour.a Body restraint imposed on rats for 20 min induced significantly increased grooming behaviour within 10 min compared with control, before sacrífice for c-Fos staining. n = 4; **P = 0.0024; Student’s unpaired two-tailed t-test. b Representative c-Fos staining revealing activation of neurons in LSv but not LSd following body restraint stress. Magnified LSd and LSv regions (red squares) are shown on the right. Scale bar, 500 μm (left), 200 μm (right). c Optogenetic activation of LSv by targeting an optic fibre at unilateral LSv that expressed ChR2-eYFP after injection with AAV9-hSyn-hChR2(H134R)-eYFP. Scale bar 500 μm. d A 5-min off +5-min on +5-min off optogenetic activation paradigm showed that activation of LSv neurons induced increased time spent in grooming behaviour (n = 8, One-way repeated measures ANOVA with Tukey post-hoc test, pre-off vs. on, **P = 0.0019; on vs. post-off, ***P = 0.0008). e Comparison of the delay and time spent in grooming and arousal behaviours, including rearing and heading, in the initial 2 min of LSv stimulation. Delay time, ***P < 0.0001; Time spent, **P = 0.0041; Student’s paired two-tailed t-test. f LSv optogenetic stimulation only induced self-grooming but not social grooming in the rat (comparison during light on period between the two behaviours: n = 3, Student’s paired two-tailed t-test, ***P < 0.0001). g Left panel: implantation of optic fibres targeting LSd and LSv in contralateral sides of the brain that expressed ChR2-eYFP. Right panel: only blue light stimulation of LSv (n = 4) but not LSd (n = 4) resulted in increased time spent in grooming. Scale bar, 1000 μm. (comparison during light on period between the two groups: Student’s paired two-tailed t-test, ***P < 0.0001). h No significant (ns) difference in the time spent in grooming between unilateral (n = 5) and bilateral (n = 4) LSv stimulation (comparison between the two groups during light on period: P = 0.0553, Student’s two-tailed t-test). All data are presented as mean ± SEM. See also Supplementary Table 1 for further statistical information. Source data are provided as a Source Data file.Consistent with the c-Fos finding, by injecting AAV9-Syn-ChR2-eYFP in LSv and LSd separately in the same animal, only optogenetic activation of LSv but not LSd induced grooming (Fig. 1g; Supplementary Movie 1), confirming the specific involvement of LSv. Furthermore, when comparing the effects of unilateral vs. bilateral stimulation of LSv, the latter induced more time spent in grooming but to a modest extent only (Fig. 1h), indicating that activities of unilateral LSv is sufficient to generate the full expression of self-grooming. Taken together, these results suggest that LSv is involved in the manifestation of repetitive self-grooming that is closely related to stress.Upstream and downstream nuclei of LSv contributing to self-groomingTo dissect the circuit underlying LSv-regulated grooming, we mapped the upstream and downstream areas of LSv. Injection of the retrograde tracer chlorea toxin b (CTB) 488 into LSv resulted in fluorescent signals in a number of brain regions, but a highly discrete and prominent labelling of the VS of the ipsilateral hippocampus was found (Fig. 2a). In agreement, anterograde labelling confirmed that VS is an upstream region of LSv (Fig. 2b). We also micro-injected AAV9-Syn-ChR2-eYFP into LSv and observed conspicuous fluorescent nerve terminals in the lateral hypothalamus tuberal nucleus (Tu, Fig. 2c). Consistently, CTB-488 injection in Tulabelled cells in LSv (Fig. 2d). Based on these observations, we performed experiments to specifically manipulate these LSv-connected upstream and downstream pathways. First, we injected AAV9-CaMKIIα-ChR2-mCherry into VS and a light cannula was implanted onto LSv (Fig. 2e). Optogenetic stimulation of the terminals of the excitatory VS → LSv pathway located in LSv at a frequency of 25 Hz triggered self-grooming behaviour similar to those of stimulating LSv (Fig. 2f), including arousal behaviour (Fig. 2g). In brain slices obtained from these animals showing grooming response to optogenetic stimulation in vivo, activation of the ChR2-expressing VS neurons generated depolarizing response leading to firing (Supplementary Fig. 1a, b). In voltage-clamp recording, the corresponding light-evoked excitatory postsynaptic currents (oEPSCs) was sensitive to 10 μM CNQX (Supplementary Fig. 1c). While tetrodotoxin (TTX, 1 μM) completely eliminated the oEPSC, addition of 4-amino-pyridine (4-AP, 1 mM) restored them (Supplementary Fig. 1d), indicating that this connection is monosynaptic in nature.Fig. 2Mapping upstream and downstream nuclei of LSv contributing to self-grooming.a Retrograde labelling by CTB-488 microinjection in the LSv resulted in prominent expressions in the ventral subiculum (VS) of the hippocampal formation in the ipsilateral brain. Scale bar, 1000 μm. Inset shows enlarged ventral subiculum. Scale bar, 200 μm. b Anterograde tracing by injection of AAV9-CamIIKα-ChR2-mCherry confirmed the VS-LSv projection. Left, scale bar, 1000 μm; Right, scale bar, 200 μm. c Anterograde tracing by AAV9-hSyn-hChR2(H134R)-eYFP microinjection in the LSv resulted in prominent expressions in the tuberal nucleus (Tu) of the lateral hypothalamus. Left, scale bar, 1000 μm; Right, scale bar, 200 μm. d Injection of CTB-488 into Tu confirmed the LSv-Tu projection. Left, scale bar, 1000 μm; middle, scale bar 500 μm; Right, scale bar 200 μm. e and f Optogenetic activation of the VS → LSv pathway for 5 min at 25 Hz induced increased grooming behaviour that was significantly higher than those in the 5-min pre-light and post-light-off periods (n = 5, One-way repeated measures ANOVA with Tukey post-hoc test, F (1.489, 5.954) = 13.05, P = 0.0083; off vs. on, *P = 0.0189; on vs. post-off, **P = 0.0096). g Comparison of the delay and time spent in grooming and arousal behaviours, including rearing and heading, in the initial 2 min of stimulation. n = 5; delay time, ***P = 0.0004; time spent, ***P = 0.0009; Student’s paired two-tailed t-test. h and i Optogenetic activation of the LSv → Tu pathway for 5 min induced increased grooming behaviour that was significantly higher than those in the 5-min pre-light-off and post-light-off periods. (n = 7, one-way repeated measures ANOVA with Tukey post-hoc test, F (1.104, 6.622) = 39.73, P = 0.0004; pre-off vs. on, **P = 0.0025; on vs. post-off, ***P = 0.0010). j Comparison of the delay and time spent in grooming and arousal behaviours, including rearing and heading, in the initial 2 min of stimulation. (n = 7; delay time, ***P = 0.0002; time spent, **P = 0.0147; Student’s paired two-tailed t-test). All data are are presented as mean ± SEM. Source data are provided as a Source Data file.In a separate set of experiments, after injection of AAV9-Syn-ChR2-eYFP into the LSv (Fig. 2h), optogenetic stimulation of the LSv → Tu pathway by targeting the LSv terminals in Tu induced self-grooming (Fig. 2i). Again, the delay and the time spent in grooming and arousal were similar to that of stimulating LSv (Fig. 2j). In brain slice experiments, the functionality of ChR2 in LSv neurons was validated (Supplementary Fig. 1e, f). Furthermore, as LSv is populated mainly by GABAergic neurons34, we also confirmed in brain slice the GABAergic, monosynaptic connection from LSv to Tu in contributing to grooming (Supplementary Fig. 1g, h).Positive valence of LSv-associated grooming behaviourOne important question is whether the grooming modulated by LSv and its output pathway signifies a positive or negative emotional state. When we assessed the intrinsic desirability vs. averseness of LSv-modulated grooming by real-time place preference (RTPP), we found significant preference for the animals to stay in the compartment associated with stimulation of the LSv (Fig. 3a, b) or LSv → Tu pathway (Fig. 3a, c), after subtracting the time spent in grooming. To reduce the influence of grooming itself in the assessment, we also conducted a conditioning place preference (CPP) test. Similarly, after 2 × 3 min of association per day between photo-stimulation and one chamber for 3 days, the animals spent significantly higher amount of time in the stimulation-associated chamber in the preference test (Fig. 3d–f). Therefore, LSv → Tu triggered grooming is clearly associated with a positive affective valence.Fig. 3Positive valence of LSv-associated grooming behaviour.a–c Schematics and results of real-time place preference (RTPP) test based on a dual chamber setup showing increased time spent by an animal in the chamber that was associated with optogenetic stimulation of the LSv (b, Interaction; F(1, 6) = 8.908, P = 0.0245; Virus main effect; F(1, 6) = 1.589e−031, P > 0.9999; optostimulation main effect; F(1, 6) = 34.35, P = 0.0011; ***P = 0.0005) or LSv → Tu pathway (c, Interaction; F(1, 6) = 4.789, P = 0.0712; virus main effect; F(1, 6) = 2.139e−030, P > 0.9999; optostimulation main effect; F(1, 6) = 10.40, P = 0.0180; **P = 0.0043), confirmed by statistical analysis shown in the right panel in which the time spent in grooming was subtracted. Control animals received injection of AAV5-hSyn-eGFP. n = 4; ns: not significant; Two-way repeated measures ANOVA with Sidak post-hoc test. d–f Schematics and results of conditioning place preference (CPP) test based on a dual chamber setup. Animals spent significantly higher amount of time in the stimulation-associated chamber in the preference test for LSv (e interaction; F(1,6) = 19.86, P = 0.0043; virus main effect; F(1,6) = 7.353, P = 0.0350; optostimulation main effect; F(1, 6) = 25.24, P = 0.0024; **P = 0.0011) or LSv → Tu pathway (f interaction; F(1, 6) = 15.84, P = 0.0073; virus main effect; F(1, 6) = 12.81, P = 0.0116; optostimulation main effect; F(1, 6) = 7.159, P = 0.0368; **P = 0.0073), confirmed by statistical analysis shown in the right panel. Control animals received injection of AAV5-hSyn-eGFP. n = 4; ns: not significant; two-way repeated measures ANOVA with Sidak post-hoc test. All data are presented as mean ± SEM. Source data are provided as a Source Data file.The VS → LSv → Tu di-synaptic limbic circuit modulates self-groomingHaving established the role of the VS → LSv and LSv → Tu projections in triggering self-grooming behaviour, we asked whether the neurons in LSv that receive command from the VS are the same neurons conveying grooming-related signals to the downstream Tu. To address this question, we first injected AAV9-Syn-ChR2-eYFP into VS and CTB-555 into Tu (Fig. 4a, upper panel) and prepared brain slices for whole-cell recordings targeting LSv neurons that were tagged with biocytin during recording. It was found that many neurons that generated EPSCs in response to photo-stimulation of VS terminals also expressed CTB signal (Fig. 4a, lower panel), confirming the presence of a di-synaptic VS → LSv → Tu pathway. To confirm its function, we adopted a strategy that allowed us to manipulate this specific pathway in vivo. By injecting anterograde, synapse-crossing AAV1-Syn-Cre35 in VS and the retrograde retroAAV-EF1α-DIO-ChR2-mCherry in Tu (Fig. 4b upper panel), only LSv neurons that were innervated by VS and also projected directly to Tu would express ChR2-mCherry (Fig. 4b, lower panel). In these animals, light delivered to LSv (Fig. 4c) induced robust self-grooming (Fig. 4d), with delays dependent on the frequency of light stimulation (Fig. 4e). Interestingly, rearing and heading behaviours were observed much less frequently (Fig. 4f; Supplementary Movie 3), suggesting that the exploratory behaviour per se is not mediated by the VS → LSv → Tu circuitry. In brain slices obtained from these animals, photo-stimulation evoked membrane depolarization of ChR2-mCherry-positive LSv neurons leading to firing (Fig. 4g). Furthermore, single-cell RT-PCR of cytoplasmic content extracted from LSv neurons was positive for GAD67 mRNA but not the glutamate neuron marker vGluT2, confirming their GABAergic nature (Fig. 4h).Fig. 4The VS → LSv → Tu disynaptic limbic circuit modulates grooming behaviour.a Upper panel: injection paradigm to confirm that neurons in LSv innervated by VS are neurons that send signals to Tu. AAV9-Syn-ChR2-eYFP was injected into VS and CTB-555 was injected into Tu. Lower panel: An example of a biocytin-tagged LSv neuron that generated EPSCs in response to stimulation of VS → LSv terminals was also labelled by CTB-555. Scale bar, 10 μm. b Experimental strategy to elucidate the role of VS → LSv → Tu circuitry in grooming. Upper panel: AAV1-Syn-Cre was injected into VS and retroAAV-EF1α-DIO-ChR2-mCherry was injected into Tu. An optical fibre was implanted onto LSv. Lower panel: as a result, only neurons in LSv that simultaneously received innervation from VS and projected to Tu could express mCherry as confirmed by confocal microscopy. Left, scale bar, 500 μm; Right, scale bar, 200 μm. c–e In these rats, in vivo light delivery onto LSv activated the VS → LSv → Tu pathway specifically in vivo c and significantly increased grooming behaviour (d, n = 8, F(1.260, 8.822) = 64.89, P < 0.0001, one-way repeated measures ANOVA with Tukey post-hoc test; pre-off vs. on, ***P < 0.0001; on vs. post-off, ***P = 0.0006) in a frequency-dependent manner (e, n = 4, F(1.517, 4.551) = 63.70, P = 0.0006, one-way repeated measures ANOVA with Tukey post-hoc test; 0 vs. 20 Hz, *P = 0.0134; 0 vs. 30 Hz, ***P < 0.0001; 0 vs. 40 Hz, **P = 0.0013; 0 vs. 50 Hz, **P = 0.0020). f Time spent in arousal and grooming behaviours in the initial 2 min of stimulation of the VS → LSv → Tu circuitry. n = 6; Time spent, ***P < 0.0001; Student’s paired two-tailed t-test. g Validation of functional ChR2 was confirmed by in vitro optogenetic stimulation in whole-cell recording following in vivo experiments. h Single-cell RT-qPCR result of cytoplasmic content of patched LSv neurons was positive to GAD67 but negative to vGluT2. All data are presented as mean ± SEM. Source data are provided as a Source Data file.VS → LSv → Tu-induced grooming resembles those caused by emotional stressNext, we asked whether this VS → LSv → Tu pathway really is involved in stress-induced self-grooming. We exploited the fact that grooming behaviour are context-sensitive and could be reflected in their sequence patterns or microstructure36,37. To do so, we generated four additional grooming models that are related to physical and emotional challenges in different degrees, including those following free swimming38, water spray25, bright light exposure37 and body restraint29,31,32 (Fig. 5a and see “Methods” section). The former two models provoke more physical stress via moistening of the fur while the latter two models are often employed as models implicating emotional stress30–33.Fig. 5Grooming induced by optogenetic activation of the VS → LSv → Tu circuitry resembles emotional stress-induced grooming.a The six grooming models in this study, including optogenetic activation of LSv (OS), body restriant (RS), light exposure (LS), swimming (SM), water spray (WS) and spontaneously occurring (ST). b The OS-induced (n = 9), RS-induced (n = 10) and LS-induced (n = 8) grooming models spent similar average time in this behaviour but were significantly lower than that in SM-induced (n = 8) and WS-induced (n = 9) models. n = 14 for ST model (one-way ANOVA with Tukey post-hoc test; ns: not significant; OS vs. ST, ***P < 0.0001; OS vs. WS, *P = 0.0387; OS vs. SM, ***P < 0.0001). c–e The grooming frequency (bouts per min, c), single bout duration d and transitions per bout e were variable among the different models. One-way ANOVA with Tukey post-hoc test. *P < 0.05; **P < 0.01; ***P < 0.001. f Optogenetics-induced grooming, the body restraint and light exposure models are highly similar in term of bout frequency, bout duration and transitions per bout. The dimension of the symbol along an axis is defined by the SEM of the corresponding parameter. g The average number of times spent on grooming different body parts in the six experimental models of this study, expressed as % of total. h 3-D plot of the number of times spent in grooming different body parts revealed higher similarity of the optogenetics model and the restraint and light models. All data are presented as mean ± SEM. See also Supplementary Table 2 for further statistical information. Source data are provided as a Source Data file.In agreement to previous reports13,14,36,37 and our prediction, compared with spontaneous grooming, the four induced grooming models exhibited higher percentages of incorrect phase transition and interrupted bouts implicating elevated stress levels (Supplementary Fig. 2a, b). Interesting, for some parameters, our LSv-optogenetics stimulation model shares similarity with the light exposure and body restraint models. These include total time spent in grooming (Fig. 5b), and also the combination of bout frequency (Fig. 5c), duration of individual bouts (Fig. 5d) and the transitions per bout (Fig. 5e). Combination of these parameters together suggested a higher similarity of the optogenetics model with the body restraint and light exposure model (Fig. 5f and Supplementary Fig. 2c–e). In addition, rats in these three models also spent a higher frequency in paw licking while in the two fur moistening models they spent a higher frequency in grooming the head and body instead (Fig. 5g and Supplementary Fig. 2f–h). A plot of the number of times spent in different body parts also implicate the similarity among these models (Fig. 5h and Supplementary Fig. 2i–k).To validate a higher level of stress associated with the body restraint and light exposure models, we conducted a series of tests to measure stress levels in the different models, including a new latency to nest in open field test (Fig. 6a), a light–dark box test (Fig. 6b), and the elevated plus maze (EPM) test (Fig. 6c). These were conducted after stress induction but before full expression of grooming behaviours. Combining the results of the tests (Fig. 6d) confirms that the body restraint paradigm is associated with the highest level of stress, which is followed by light exposure and water spray. Least stress is associated with free swimming and the control group. Interestingly, although water sprayed to the head or body alone rather than the whole body resulted in more time spent in grooming the head or body as expected (Supplementary Fig. 3a), these two models cluster well with the whole body water spray model in terms of bout frequency/transitions per bout/single bout duration analysis (Supplementary Fig. 3b).Fig. 6Assessment of stress levels associated with different grooming models.a–c The latency to nest in open field test (OFT) (a, F(4, 45) = 5.525, P = 0.0010; Control vs. RS, *P = 0.0451), the time in light box during the light–dark box test (b, F(4, 45) = 5.252, P = 0.0015; Control vs. RS, **P = 0.0011; Control vs. LS, *P = 0.0137) and the time in the open arms during the elevated plus maze test (c, F(4, 45) = 6.001, P = 0.0006; Control vs. RS, *P = 0.0113; Control vs. LS, *P = 0.0404) recorded in different models indicate significant differences mainly between the body restraint and light exposure models with the control group. n = 10 rats for each group; One-way ANOVA with Tukey post-hoc test. d 3-D plot of these parameters confirms that the body restraint paradigm was associated with the highest level of stress, which was followed by light exposure and water spray. Least stress was associated with free swimming and the control group. All data are presented as mean ± SEM. Source data are provided as a Source Data file.Microstructures of grooming triggered by stressors and the VS → LSv → Tu circuitIt has been suggested that the patterns, or microstructure, of grooming is variable and may reflect differences in the context and underlying neural mechanisms13,36,37. But despite increased emphasis on the significance of analyzing grooming structures13,36, sensitive methods that can distinguish different context-dependent grooming are still not available. Here, we developed an alternative approach to address this question. Instead of focusing on arbitrarily defined ‘incorrect’ transition of different grooming phases, we analysed the frequencies of all possible transitions among different phases (Supplementary Fig. 4a) and then calculated the percentage of each transition and expressed as a phase transition probability matrix (Fig. 7a). It is obvious that the phase transition matrices derived from different models are by no means uniform. To quantitatively determine the similarity among the matrices, two quantitative indices, namely, the cross-correlations and the Euclidean distances were derived. The pairwise comparisons indicate that the optogenetics model and the light exposure and body restraint models have the highest cross-correlations (Fig. 7b) and the least Euclidean distances (Fig. 7c). Hierarchical clustering analyses also confirmed their higher resemblance. In fact, when these matrices were transposed into binary-coded format (Fig. 7d) or presented as phase connection graphs highlighting the most probable transitions and prominent phase (Supplementary Fig. 4b) the similarities among the optogenetics, light exposure and restraint models are obvious.Fig. 7Microstructure analysis of grooming under different contexts.a The transition probability matrices summarizing the likelihood from one grooming phase to another phase in the six grooming models. The different phases of no grooming/grooming activities were defined as no grooming (phase 0), paw licking (phase 1), nose/face/head grooming (phase 2), body grooming (phase 3), leg grooming (phase 4) and tail/genital grooming (phase 5). The numbers show the occurrence in % and are colour-coded. b, c The correlation coefficients b, and Euclidean distances c between each pair of transition probability matrices were derived. Hierarchical clustering analysis supports a higher similarity of the optogenetics model and the restraint and light models. RS body restraint stress, OS optogenetics stimulation of LSv, LS bright light exposure, ST spontaneous grooming, WS water spray, SM swimming. d The patterns and microstructures of six grooming models summarized by the phase transition probability matrix shown in a were transformed into two-colour binary format using a threshold of 25% illustrating higher similarity among the optogenetics model, the restraint model and the light model.Increased activity of LSv neurons precedes emotional stress-induced groomingTo gain further insight into the role of LSv in stress-mediated self-grooming behaviour, we determined the population dynamics of LSv neurons in freely behaving rats in the different grooming models by fibre photometry based on the GCaMP6s reporter (Fig. 8a). After stress induction, the calcium activities of the neurons were monitored and aligned to the start of detected grooming bouts. In the body restraint (Fig. 8b) and light exposure model (Fig. 8c), we detected on average a clear rise in calcium signals of LSv neurons shortly before the start of a grooming event. Significant differences in ΔF/F in the pre-grooming and post-grooming periods were found (Fig. 8b, c). In contrast, no discernible changes in calcium signals were found in the swimming (Fig. 8d) and water spray (Fig. 8e) models as well as in spontaneous grooming (Fig. 8f). In addition, the fluorescent signals in control animals expressing eGFP in the LSv neurons showed no change during the grooming behaviour.Fig. 8Activation of LSv neurons precedes emotional stress-induced grooming.a Setup of fibre photometry to record calcium activity from LSv neurons infected with AAV9-Syn-GCaMP6s or AAV5-hSyn-eGFP virus. The correct expression of GCaMP6s and placement of the fibre optics were verified post-mortem. Scale bar 500 μm. b–f Typical results of calcium activity in LSv around the start of grooming in multiple bouts (left panel) and the averaged results (red traces, middle panel) in the body restraint model (b; n = 40 trails from five rats; ***P < 0.0001), light exposure model (c; n = 40 trails from four rats; ***P < 0.0001), swimming model (d; n = 40 trails from four rats), water spray model (e; n = 40 trails from six rats) and spontaneous grooming (f; n = 40 trails from four rats). There was significant rise in calcium activity (ΔF/F) prior to the start of the grooming in the body restraint and light exposure models but not the other models as shown in the right panels. ns non-significant; ns: not significant; Student’s paired two-tailed t-test. The blue traces represent signals from animals which expressed only eGFP but not GCaMP. All data are presented as mean ± SEM. Source data are provided as a Source Data file.Inhibition of the VS → LSv → Tu circuit suppressed grooming caused by emotional stressFinally, to establish the causal relationship between activation of VS → LSv → Tu and self-grooming, we applied optogenetic to achieve targeted inhibition of this circuitry. We injected AAV1-Syn-Cre into VS and AAV9-EF1α-DIO-eNpHR-eYFP into LSv, which allowed the VS → LSv → Tu pathway to be inhibited specifically by targeting the NpHR-expressing terminals in Tu (Fig. 9a). The expression and function of eNpHR in LSv neurons were verified in brain slices (Fig. 9b). When yellow light was delivered into Tu, there was still increased self-grooming observed in body restraint and light exposure models but the level was suppressed (Fig. 9c, d). Nevertheless, the level of grooming was significantly weaker than when the light was off, confirming a major contribution of this pathway in post-stress grooming. In contrast, light delivery did not affect the time spent in grooming in both the swimming and water spray models (Fig. 9e, f). Spontaneous grooming was also unaffected (Fig. 9g). We performed control study in which non-functional AAV5-hSyn-eGFP was injected into VS, LSv, and Tu. No abnormality was found in these animals, including grooming activities, when light was delivered (Supplementary Fig. 5). To validate the effect of optogenetic inhibition of LSv → Tu terminals on activity of Tu neurons, AAV-expressing eNpHR were injected into LSv as before. At the same time, we also injected AAV8-hSyn-hM3Dq-mCherry into LSv (Fig. 9h). In subsequent brain slice experiments, we patched Tu neurons and held them at a membrane potential whereas spontaneous firing occurred. Under this condition, CNO superfusion applied to activate LSv neurons suppressed the firing rate of Tu neurons, which was accompanied by membrane hyperpolarization, in agreement with increased GABA release from LSv terminals (Fig. 9h). Under this condition, when yellow light is shone on the LSv terminals aiming to inhibit GABA release, the decreased firing of Tu neurons was rectified (Fig. 9h). Parallel experiments were performed based on chemogenetic inhibition of LSv neurons in the VS → LSv → Tu pathway (Supplementary Fig. 6a), with results supporting the same conclusion (Supplementary Fig. 6b–f).Fig. 9Activation of LSv neurons is necessary for emotional stress-induced grooming.a Schematics for optogenetics manipulation. Targeted functional inhibition of the VS → LSv → Tu circuitry was achieved by injection of AAV1-Syn-Cre into VS, AAV9-EF1a-DIO-eNpHR-eYFP into LSv and implantation of optical fibre onto Tu. Scale bar 500 μm. b Whole-cell recordings from LSv neurons in brain slices obtained from these animals validated eNpHR-mediated inhibition via prolonged (5 min) yellow light delivery. c–f While significant increases in grooming time were found in light-off control group following body restraint (c, n = 9) and light exposure (d, n = 9), the increases in grooming time were significantly smaller when light was delivered. In contrast, the increases in grooming time after swimming (e, n = 6) and water spray (f, n = 6) were similar in both light-off and light-on trials. ns not significant; *P < 0.05; **P < 0.01; ***P < 0.001, two-way repeated measures ANOVA with Sidak post-hoc test. g Optogenetic inhibition did not affect the time that the animal spent in spontaneous grooming. n = 6, P = 0.4861; Student’s paired two-tailed t-test. h Targeted functional activation and inhibition LSv terminals in Tu by injection of AAV9-EF1a-eNpHR-eYFP and AAV8-hSyn-hM3Dq-mCherry into LSv and recording in Tu neurons. The spontaneous firing of Tu neurons induced by −50 mV holding level can be partially inhibited by applying CNO and recovered by yellow light stimulation. (n = 11, one-way repeated measures ANOVA with Tukey post-hoc test; **P < 0.01). All data are presented as mean ± SEM. See also Supplementary Table 3 for further statistical information. Source data are provided as a Source Data file.DiscussionGrooming in mammals represents an important adaptive response to stress, and provides a valuable model for elucidating the brain mechanism of stress management. In this study, we have identified a di-synaptic hippocampal–septal–hypothalamus circuit in the limbic system of rodent brain that regulates self-grooming but not social-grooming behaviour. Through a combination of tract tracing, optogenetic and electrophysiological experiments, we showed that in this circuitry, LSv neurons that receive monosynaptic excitatory innervation from VS in turn send monosynaptic GABAergic projection to Tu neurons in the hypothalamus. There were different studies in the past that implicate the involvement of the subiculum, the LSv and various hypothalamus regions in processing stress or mediating stress-induced responses39–43. A recent study integrating gene expression and brain-wide connectivity in the mouse also revealed a hippocampal-septo-hypothalamic network pertinent to cognitive-limbic integration44. However, our study reveals the precise connections among the specific sub-divisions of these three nuclei that plays a critical role in modulating stress-induced response. Also, the grooming behaviour we observed distinguishes from those of many previous studies in its strong association with emotional stress, positive rather than negative valence and a delayed instead of immediate response following stimulation.Among the different stress-induced grooming models studied, we showed by fibre photometry that only the body restraint and light exposure protocols activate LSv neurons with a peak activity preceding grooming, and that optogenetic and chemogenetic inhibition of the discrete VS → LSv → Tu circuit suppressed grooming triggered specifically by these two paradigms. A body of evidence suggests that these two paradigms evoke mainly emotional stress30–33 rather than the physical stress as in the swimming and water spray models30,45. In agreement, our own assessment confirmed that the body restraint and light exposure paradigms are associated with a higher level of stress when compared with the water spray and swimming models. Our findings therefore strongly implicate that the di-synaptic circuit from VS to Tu regulates grooming particularly relevant to emotional stress. It is noted that optogenetic and chemogenetic inhibition of this pathway could not completely suppress the post-stress grooming. It is likely that additional pathway exists to effect similar response.It has been suggested that analysis of grooming activity and its microstructure may serve as a useful index of stress and anxiety13,36,37. In this study, we found that activation of the VS → LSv → Tu circuit triggered excessive grooming with microstructures closely resembling those evoked by the restraint and light exposure paradigms. We analysed the probabilities of all possible transitions among different stages of grooming, and revealed subtle but discernible differences in the microstructures of grooming driven by different types of stressors. This finding affirms that the grooming microstructure is highly related to the specific context that triggers the behaviour13,14,28,37. It is noteworthy that, according to our analysis result, the microstructure of spontaneous grooming is more similar to restraint-induced and light exposure-induced grooming rather than the two water-associated models. This finding may reflect that the physical stress of body wetting generated relatively unique patterns of grooming different from other types of grooming. At the same time, although spontaneous grooming occurs at a much lower frequency, this behaviour may encompass those that are generated by internal drive and motivational elements related to psychological need akin to emotional stress-induced grooming. On the whole, our approach represents a valuable addition to the currently available methods in analyzing context-dependent microstructure of grooming. However, caution should be taken in interpreting the results as the microstructure could also be affected by other factors like the method that induces grooming.If the circuitry we identified is involved in the manifestation of emotional stress-induced response, it seems counter-intuitive that the grooming response is associated with a positive valence, as demonstrated by our RTPP and CPP results. Indeed, Xu et al. 24 had demonstrated negative emotional state associated grooming response following activation of a hypothalamic-septal pathway. Behavioural response to stress often manifests as heightened arousal, attention and alertness supported by concomitant physiological adaptations, and believed to be associated with a negative emotional state. However, these responses are often time-limited due to restraining forces that prevent over-response that could be harmful4,12. The desirability of the LSv → Tu pathway-modulated post-stress grooming behaviour implicates a calming effect, which is consistent with an adaptive displacement activity to stress, or suppression of over-response to stress. In fact, this pathway may not be responsible for generating stress-induced emotional state itself, but only for initiating grooming action that helps to maintain a sense of well-being.As the limbic component of the hippocampus, the subiculum has been suggested to exert a generalized up-stream influence on integration of limbic functions46,47 probably via its innervation to various limbic forebrain structures44,48. Indeed, the ventral division of subiculum, VS, may play a role in stressor selection with respect to regulation of neuroendocrine response to stress39,40,49,50. However, whether VS exerts an excitatory or inhibitory influence on the hypothalamo-pituitary-adrenocortical axis is not clear39,40. In relation to this, our findings not only extend the understanding of the role of VS in processing emotional stress-related information but that via its downstream LSv → Tu pathway, behavioural adaptation like grooming could be effected. On the other hand, the hypothalamus is well-known to be a primary output node for the limbic system49,51, mediating many innate behaviours. Although the Tu has not been studied in great detail before, consistent with our finding, the lateral hypothalamic area including the Tu has recently been implicated in grooming behaviour, including the development of pathological grooming12,25,52. Since a number of other hypothalamic areas are known to be involved in grooming behaviour19,20,53, the relationship of Tu with respect to these other areas remains to be investigated.The LSv is a limbic structure long known to be associated with a variety of cognitive and emotional processes41,51. For example, it has been implicated for the modulation of anxiety54, expression of fear-conditioning to context55, and regulation of autonomic responses to aversive stimuli32. The impact of the LSv on stress-related behaviour is however controversial56–60, partly due to that previous studies relied on studying the effects of lesion or electrical stimulation and therefore were non-specific. Many of these studies also considered the LSv as a single nucleus, despite evidence for the presence of heterogenous neuronal populations constituting different sub-regions61,62. Our results revealed that only the ventral subdivision of the LSv, and more specifically its GABAergic neuronal population, conveys emotional stress relevant to the generation of repetitive grooming behaviour. Based on our findings, we propose that LSv receives emotional stress-related information from VS and in turn regulates down-stream Tu in triggering grooming.After stimulation, the typical tens-of-seconds of delay before grooming takes place is puzzling. This is in sharp contrast to immediate grooming behaviours triggered by stimulating some other brain areas12,24,25,52,53, e.g. the dorsomedial hypothalamus and the orbitofrontal-striatal projection22 that are reminiscent of compulsive-like behaviour. One interpretation is that a build-up time is needed for the manifestation of the response, which is consistent with an integration or decision-making role of LSv. At the same time, we found that stimulation of LSv neurons and related pathways could trigger other behaviours, notably rearing-like arousal behaviour. In contrast, when the VS → LSv →Tu pathway is activated specifically, the grooming is almost free of preceding arousal behaviour. This dissociation between arousal and grooming strongly suggests that while LSv is involved in arousal/exploratory behaviour, this is mediated by microcircuitries other than the VS → LS →Tu pathway that controls grooming per se. Thus, we speculate that as an integration hub, LSv is composed of heterogenous neuronal subpopulations that might map different inputs to different innate behaviours. Indeed, Xu et al.24 found that LSv also receives emotional state-related signals from the PVN, and triggers responses including grooming, escape behaviour and suppression of feeding, with negative rather than positive valence. Our findings therefore enrich the central role of LSv in the fine regulation and coordination of different innate behaviors. As the output pathway of LSv to Tu is GABAergic, the nature of the interaction between these inhibitory neurons with neurons in Tu should be clarified in future studies.As aberrant response to stress is regarded as a factor driving compulsive repetitive behaviour in some neuropsychiatric disorders6,7,10, notably autism63,64 and obsessive-compulsive disorders22,65,66, our findings therefore not only uncover a limbic circuit that plays a significant role in emotional stress response but also provide a basis for deciphering the complete circuit of emotional processing and their malfunctions that could lead to abnormal repetitive behaviours in different brain disorders.MethodsAnimalsAdult male Sprague-Dawley (SD) rats weighing 300–320 g were used in this study. The animals were bred and maintained by the Laboratory Animal Service Centre of The Chinese University of Hong Kong (CUHK). The animal room was controlled at a temperature of 23 °C on a 12-h light/dark cycle. All animals were handled in strict accordance with the CUHK guidelines and the procedures were approved by the Animal Experimentations and Ethics Committee. All experiments were performed during the light phase (09:00–19:00).Viral constructsAdeno-associated viruses (AAV) including AAV5-hSyn-eGFP, AAV9-hSyn-hChR2(H134R)-eYFP, AAV9-CaMKIIα-hChR2(E123A)-mCherry, AAV8-hSyn-hM3Dq-mCherry, RetroAAV-EF1α-double floxed-hChR2(H134R)-mCherry-WPRE-HGHpA, AAV9-Ef1α-DIO-eNpHR3.0-eYFP, RetroAAV-hSyn-DIO-hM4Di-mCherry, AAV9-Syn-GCaMP6s-WPRE-SV40 were all purchased from Addgene (Watertown, USA). All viral titre were >5 × 1012 particles per ml.Stereotaxic surgeries for in vivo studiesSD rats were anesthetized with ketamine (75 mg/kg, i.p.) and xylazine (6 mg/kg, i.p.), and placed gently in a stereotaxic frame (Narashige, Tokyo). For micro-injections, a Hamilton syringe (33-gauge) filled with AAV virus or tracer was unilaterally or bilaterally placed into the target area according to the corresponding coordinates: LSv (−0.12 mm A/P, ±1.0 mm M/L, 6.0 mm D/V), VS (−5.40 mm A/P, ±4.0 mm M/L, 9.2 mm D/V), and Tu (−3.24 mm A/P, ±1.4 mm M/L, 9.2 mm D/V) from dura. 0.1–0.5 μl virus or tracer were injected at speed of 10–50 nl/min. The needle was left in place for an additional 10 min before retraction. The scalp incision was sutured, and postinjection analgesics were given for 3 days to aid recovery. The animals were allowed at least 3 weeks to recover and express a virus before optical fibre implantation or behavioural test. For fibre photometry and optogenetics experiments, an optical fibre (200-μm core, NA = 0.48 for photometry, NA = 0.37 for optogenetics) or fluid-injection cannula (26-gauge guide cannula possessing a 32-gauge dummy cannula) was implanted directly above. The fibre or guide cannula, together with two stainless steel screws was secured to the skull using dental cement. The rats were allowed at least 1 week to recover before behavioural test. Correct location of implanted fibre/cannula was confirmed postmortem.Optogenetic and chemogenetic manipulationsFor ChR2 photostimulation, 473-nm light laser (10 ms, 25 Hz, unless otherwise indicated, Newdoon Technology) was delivered via an optic cable (200-μm core, 0.37 NA, Doric Lenses) and the stimulation duration varied in different experiments, as described in the relevant results. Laser power was 5 mW measured at the tip of the fibre, which was implanted 0.3 mm above the targeted nucleus. For eNpHR photoinhibition, two optical fibres were attached to the double cannula for constant illumination of the targeted site (589 nm, 10 mW from the tip of 200 μm fibre) of the targeted site. A 589-nm laser (10 mW, Newdoon Technology) was continuously turned on throughout the post-stress behaviour sessions. For chemogenetic manipulation, clozapine N-oxide (CNO) (5 μM/0.5 μl, Sigma), or vehicle (saline) was administered via the implanted cannula 40 mins before grooming was assessed. For all behavioural experiments, the animals were videotaped, and the behaviours were evaluated offline.Retrograde and anterograde tracingFor retrograde tracing, recombinant cholera toxin-b conjugated to AlexaFluor-488 (CTB-488, ThermoFisher) in PBS was injected to the target site at LSv. At least 7 days were allowed for complete retrograde transport before sacrifice. For anterograde tracing, AAV9-Syn-ChR2-eYFP was injected into LSv and allowed at least 4 weeks for full expression. Brain sections (30 μm) were prepared and examined under confocal laser scanning microscope (C1, Nikon). To establish the disynaptic circuit of VS → LSv → Tu, AAV9-Syn-ChR2-eYFP was injected into VS for at least 4 weeks for expression. One week prior to sacrifice, recombinant cholera toxin-b conjugated to AlexaFluor-555 (CTB-555, ThermoFisher) in PBS was injected into Tu. Coronal slices of 300 μm were prepared by using a vibrotome (Campden 5100MZ-PLUS Vibrotome) for whole-cell patch recordings. Neurons in LSv that responded to blue light stimulation with EPSC (oSPSC) were filled with biocytin. Biocytin was revealed with Alexa FluorTM 405-conjugated streptavidin–biotin complex (ThermoFisher). The expressions of biocytin and CTB-555 were examined under confocal laser scanning microscope (C1, Nikon).Behavioural studiesFor all behavioural tests, animals were habituated for 30 min/day consecutively for 3 days before testing. The behaviours of the animals were recorded in a test chamber (30 cm length, 30 cm width and 60 cm height) by a video camera (Logitech, C922). Each animal was tested at least three times on different days to verify reproducibility.For spontaneous grooming model, the animal was placed in the test chamber, and the spontaneous activities were recorded for 20 min.For restraint stress-induced grooming29,31,32, the animal was restrained in a black tube (5 cm diameter, 25 cm length) for 20 min and then put immediately into the test chamber for 20 min of video recording.For bright light-induced grooming, animals were transported to the dimly lit laboratory and left undisturbed for 30 min prior to exposure to the light box on the day of the experiment37. The animal was exposed to the bright light (60–80 W) from a lamp, 10 cm from above the home cage, for 20 min. Immediately afterward, the animal was placed in the test chamber, and its behaviour was recorded for 20 min.For optogenetically induced grooming, the behaviour was monitored for a total of 15 min, following a 5-5-5 min protocol in which blue light pulses (473 nm, 5 mW, 10 ms each at 25 Hz) was turned on in the middle 5 min. Grooming and other behaviours were video-taped. To test for social grooming behaviour, two animals were placed in the same chamber, and the same 5-5-5 min protocol was applied.For water spray-induced grooming25, the animal was sprayed with water squirts directed to the face, belly, and back (four squirts per area) with a spray bottle pre-filled with sterile water (25 °C) as the “whole-body” water spray model. For “head-only” model, water was directly sprayed to the left and right face (four squirts per side). For “body-only” model, water was sprayed to the belly and back (eight squirts per area). It was then placed in the test chamber for video recording of the behaviour for 10 min. Before testing, the animals were habituated once per day for 3 consecutive days.For swimming-induced grooming30,38, the animal was placed in an open swimming pool (100 cm × 50 cm and 50 cm high) that was filled with water at 25 °C. Water depth was set at 15 cm to ensure that the rat could stand freely and allowed them to freely swim for 5 min. After at least 3 days of adaptation, animals were allowed to swim for 2 min on the day of grooming testing, and excess water was then removed and allowed to drain away before the animals were placed immediately into the test chamber for 10 min of video recording.Based on the natural aversion of rodents towards brightly lit areas and at the same time a tendency towards exploratory behaviour67, the following behaviour tests were performed to evaluate the emotional status associated with different grooming-inducing paradigms.For open-field with a nest test, a dark shelter nest (20 cm × 20 cm and 40 cm high) was put in the corner of an open field (100 cm × 100 cm and 40 cm high). Animals were handled and habituated for 10-min to the open-field 2 days before the test. On the testing day, each rat was allowed to first freely explore the box for 5 min. Then, immediately after going though the induction protocol of a grooming model, the rat was placed in the box at the corner diagonal to the nest. The latency of escape to the nest was recorded as an index of stress-like behaviour. A shorter latency indicates a higher stress level.For light–dark box test67, the test box consists of two boxes (light and dark, each was 50 cm × 40 cm and 40 cm high) and connected by an open door. When testing, the animal was first placed in light box and the total time spent in light box was recorded for a period of 5 min. The less time spent in the light box indicates a higher stress level.For EPM test, it was used to measure anxiety-related behaviour in rodents67. The maze was consisted of a central square (15 cm × 15 cm), two open arms without walls (15 cm × 40 cm) and two closed arms with walls (15 cm × 40 cm and 40 cm high). Each rat was placed in the centre of the maze, after which the rat behaviour was recorded for a period of 5 min. The time spent in the open arms was used as an index of stress-like behaviour.Behavioural analysis: The grooming and other behaviours were quantified manually by three observers with the aid of a video editing software (iJianJi, Guangzhou Quying Technology). The observers were blind to the experimental conditions. Social grooming was assessed when a rat licked and chewed the fur of the conspecific, while placing its forepaws on the back or the neck of the other rat. Arousal behaviours including rearing and heading were also examined. For gross analysis of self-grooming, the number of grooming bouts, the duration of individual bout and therefore the total grooming time in the test period, were evaluated. Self-grooming was defined as when the animal licked, or used the forelimb to stroke, its own body parts including the paws, nose, eyes, head, body, legs, tail and genital. An interruption of 6 s or more separated two individual bouts36,37.Grooming patterns and microstructuresTo analyse the patterns and microstructures of grooming behaviour under different conditions, different phases, or stages, of grooming activities were defined, including paw licking (phase 1), nose/face/head grooming (phase 2), body grooming (phase 3), leg grooming (phase 4), tail and genital grooming (phase 5), according to conventional protocol36,37. In addition, no grooming was defined as phase 0. To facilitate the identification of these patterns by the observers who were blind to the treatment, the recorded videos were replayed at 1/4 of the actual speed. By noting the time points of the start and end of these different phases, the exact sequences of the different phases within a grooming bout was determined. A grooming bout was considered interrupted if at least one pause in action (<6 s) was recorded within its transitions of phases. In addition, the transitions between different phases of grooming pattern were used to evaluate the ‘correct’ transitions based on the conventional cephalocaudal progression stereotypy: (0 → 1), (1 → 2), (2 → 3), (3 → 4), (4 → 5), and (5 → 0) and the otherwise ‘incorrect’ transitions represented by other patterns. From these data, the total number of transitions and the percentages of incorrect transitions were calculated.To analyse the microstructures, all grooming bouts under each of the six experimental conditions were collected and pooled. Then, the transition probabilities from one phase to another phase were calculated, and a transition probability matrix was obtained. As an example, in spontaneous grooming, the animals together spent a total of 112 times in phase 1 (paw-licking). The transitions from paw-licking to other phases took place at the following frequencies: 9 (1 → 0), 96 (1 → 2), 4 (1 → 3), 2 (1 → 4) and 1 (1 → 5). The transition probabilities were therefore 8.04%, 85.71%, 3.57%, 1.79% and 0.89%, respectively. The overall patterns of phase transitions could be visually captured by the transposed transition matrix and connectivity graphs shown in Fig. 7d and Supplementary Fig. 4b.To be able to compare quantitatively the similarity among the transition probabilities matrices, we computed two indices, namely the Euclidean distance (D) and Pearson’s correlation coefficient (CC) for each pair of matrices, whereas1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D = \sqrt {\mathop {\sum}\nolimits_{i = 1}^n {(y_i - x_i)^2} }$$\end{document}D= ∑i=1n(yi−xi)2and2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{{CC}}} = \frac{{\mathop {\sum }\nolimits_{i = 1}^n (x_i - x)(y_i - \overline y )}}{{\sqrt {\mathop {\sum }\nolimits_{i = 1}^n (x_i - \overline x )^2} \sqrt {\mathop {\sum }\nolimits_{i = 1}^n (y_i - \overline y )^2} }}$$\end{document}CC=∑i=1n(xi−x)(yi−y¯)∑i=1n(xi−x¯)2 ∑i=1n(yi−y¯)2in which {xi} and {yi} is the corresponding elements in the two transition probability matrices and n is the number of elements in the matrix. Thus, D computes the overall disparity of the two matrices derived from the distance between the corresponding elements, so a smaller D indicates a higher similarity. On the other hand, a higher CC indicates that the pair of transition matrices bear a higher similarity. Then the dendrograms of hierarchical clustering by D and by CC were obtained using the MATLAB Statistics Toolbox functions linkage () and dendrogram (). The height of each inverted-U-shape in the dendrogram is proportional to the dissimilarity between the two nodes being connected. More closely related nodes were connected at lower levels.Real-time place preferenceThe animal was put in a box (120 × 50 cm, 40 cm high), which was divided into two equal chambers without any contextual cues68. One chamber was paired with a 25 Hz photostimulation and the other identical chamber was without photostimulation. The behaviour of the animal was monitored and subsequently analysed by the ANY-Maze tracking software (Version 4.7, Stoelting CO). The percentages of time that the animal spent on the stimulated and unstimulated sides of the chamber in a period of 15 min were quantified.Conditioned place preference (CPP)The animals were habituated to handling for 3 days prior to the beginning of the procedure. Experiments were conducted in two interconnected chambers (60 × 40 and 40 cm high each) that could be separated by a sliding door. The two chambers were decorated with different forms of stripes on the wall. Animal movements in each chamber were recorded and analysed with the Anymaze software69. The procedure consisted of three stages: preconditioning (baseline), conditioning, and testing phases. On the first day, the sliding door was retracted, and rats could explore the entire apparatus freely for 15 min (3 × 5 min). Animals that spent >70% of time in either of the compartments were excluded from further analysis. Immediately following the preconditioning phase, the rats underwent conditioning in both sides, respectively. During conditioning, one of the two chambers was paired with a photo-stimulation (25 Hz of 10 ms laser pulses, 473 nm) for 2 × 3-min/day for 3 days. During the test phase, the animals did not receive any treatment and had free access to both compartments for a total time of 10 min. Animal movements in each of the chambers were recorded, and the time spent in each chamber was analysed by the Anymaze software.Brain slice electrophysiology and optogeneticsAdult rat brain slices were prepared as following procedures70,71. The rats having completed in vivo optogenetic stimulation experiments were deeply anesthetized by isoflurane, and then perfused transcardially with 100 ml room temperature NMDG artificial cerebrospinal fluid (aCSF) containing (in mM): 92 NaCl, 2.5 KCl, 1.25 NaH2PO4, 20 NaHCO3, 10 HEPES, 25 Glucose, 5 Na-ascorbate, 2 thiourea, 3 Na-pyruvate, 10 MgSO4, 0.5 CaCl2, 12 N-acetyl-l-cysteine, which was saturated with carbogen (95% O2/5% CO2) prior to use. After perfusion the rats were decapitated and the brains were gently extracted from the skull and placed into the cutting solution (NMDG aCSF) for an additional 30 s. Coronal sections at 300 μm were cut with a vibrotome (Campden 5100MZ-PLUS Vibrotome) and transferred into a holding chamber containing 34 °C NMDG aCSF to recover for 10 min. After the initial recovery, the slices were transferred into a new holding chamber containing HEPES holding aCSF containing (in mM): 92 NaCl, 20 NaHCO3, 25 Glucose, 2.5 KCl, 1.25 NaH2PO4, 10 HEPES, 5 Na-ascorbate, 2 thiourea, 3 Na-pyruvate, 2 CaCl2, 2 MgSO4, 12 N-acetyl-l-cysteine. After another hour of recover, the slices could then be transferred to the recording chamber with normal aCSF containing (in mM): 125 NaCl, 2.5 KCl, 11 Glucose, 26 NaHCO3, 1.25 NaH2PO4, 2 CaCl2, and 2 MgCl2.For recording of optically evoked action potential, excitatory postsynaptic currents (oEPSCs) and spontaneous firing, we used internal solution containing (in mM): 130 mM K-gluconate, 10 mM KCl, 10 mM HEPES, 1 EGTA, 2 mM MgCl2, 2 mM Na2-ATP, 0.4 mM Na3-GTP. For optically evoked inhibitory postsynaptic currents (oIPSCs), the internal solution was (in mM): 50 mM KCl, 10 mM HEPES, 1 EGTA, 2 mM MgCl2, 2 mM Na2-ATP, 2 mM Tris GTP. Light pulses at 470 nm were delivered through a patterned light stimulator (Polygon400 DSI-E-0470-0590-NK1 Dynamic Spatial Illuminator) to activate ChR2 while recorded neurons were voltage-clamped at −70 mV. Access resistance (<20 MΩ) was not compensated and monitored continually throughout each experiment. Recordings were terminated whenever the input resistance increased >30% or access resistance exceeded 20 MΩ. Signals were acquired using a MultiClamp 700B amplifier controlled by Clampex 10.4 software via a Digidata 1550 interface (Molecular Devices). Responses were filtered at 3 kHz, digitized at 10 kHz, and analyzed using Clampfit 10.7 (Molecular Devices). In a subset of experiments, biocytin was included in the internal solution. After recording, slices were fixed overnight in 4% paraformaldehyde. Filled neurons were identified by staining using streptavidin-conjugated Alex 405 (ThermoFisher).Fibre photometryTo monitor the neuronal activity in LSv, the rats were injected with 100 nl of AAV9-Syn-GCaMP6s-WPRE-SV40 or AAV5-hSyn-eGFP into the right LSv. Two weeks after the virus injection, an optical fibre (NA 0.48, Newdoon Technology) was implanted to target the right LSv. With at least 1-week recovery, the optical fibre was connected to FibreOptoMeter (Plexon’s Multi-Fibre Photometry) through an optical fibre patch cord (200-μm, 0.53 NA, Doric) for photometry recording. To record fluorescence signals, a beam from a 470 LED was reflected with a dichroic mirror, focused with a lens coupled to a PMT. The LED power at the tip of the patch cord was <50 μW. Fluorescent signals were collected and synchronized with behaviour of the animal via the CineLyzer system (Plexon, Dallas, USA).For analysis of photometry data, smoothed ΔF/F values obtained from CineLyzer system were exported to Matlab for further analysis. The data were represented by heatmaps or averaged response (shaded area indicate SEM) in a time window between 9 s before and 3 s after grooming onset. Smoothed ΔF/F values were calculated through three different steps: To begin with, the raw fluorescence was averaged to get FAVG (n) over a sliding time window (0.75 s) in a symmetrical pattern. The baseline for a particular frame (FBASELINE (n)) was the minimum of the FAVG (n) in a time window (3 s) preceding this frame. Next, ΔF was calculated by subtracting FBASELINE (n) from FRAW (n) and then dividing by FBASELINE (n). Finally, ΔF was smoothed with an exponentially weighted moving average. The exponential weighting was described by a time constant \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau_0$$\end{document}τ0 and the averaging window was set to 5*\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau_0$$\end{document}τ0. The frame rate of system is 30.3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{\Delta }}F/{F}({n})_{{\mathrm{unsmoothed}}} = \frac{{{F}_{{\mathrm{RAW}}}({n}) - {F}_{{\mathrm{BASELINE}}}({n})}}{{{F}_{{\mathrm{BASELINE}}}\left( {n} \right)}}$$\end{document}ΔF/F(n)unsmoothed=FRAW(n)−FBASELINE(n)FBASELINEn4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{\tau 0} = {\mathrm{floor}}(5 \ast \tau _0 \ast {\mathrm{frame}}\,{\mathrm{rate}})$$\end{document}Nτ0=floor(5*τ0*framerate)5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {F}/{F}({n})_{{\mathrm{smoothed}}} = \frac{{\mathop {\sum }\nolimits_0^{{N}_{\tau 0}} {\mathrm{\Delta }}F/{F}({n} - x)_{{\mathrm{unsmoothed}}}{\mathrm{{e}}}^{\frac{{ - \left| x \right|}}{{{N}_{\tau 0}}}}}}{{\mathop {\sum }\nolimits_0^{{N}_{\tau 0}} {\mathrm{{e}}}^{\frac{{ - \left| x \right|}}{{{N}_{\tau 0}}}}}}$$\end{document}ΔF/F(n)smoothed=∑0Nτ0ΔF/F(n−x)unsmoothede−xNτ0∑0Nτ0e−xNτ0HistologyRats were anaesthetized and perfused with PBS followed by 4% PFA in PBS. The brain was fixed in 4% PFA at 4 °C overnight. Fixed samples were sectioned into 30-μm coronal sections using a cryostat (ThermoFisher). For c-Fos staining, the brain samples were prepared 60 min after behavioural test followed by fixation. The brain sections were blocked in 10% normal goat serum in PBS with 0.3% Triton X-100 in PBS for 2 h and then incubated with anti-c-Fos antibody (1:2000; Cell Signaling Technology) at 4 °C overnight. After thorough rinsing in PBS, the slices were incubated with goat anti-rabbit lgG secondary antibody (1:1000, Invitrogen) in block solution. The sections were rinsed in PBS and counterstained with DAPI and mounted. Images were taken under a confocal laser scanning microscope (C1, Nikon).Single cell RT-qPCRPatch pipettes were pulled and filled with 3.0–5.0 μl of nominally RNAse-free internal solution using a microloader. The tip diameter of the pipette was around 1/3 of the size of cell body. Under visual inspection, the cell was harvested into the patch pipette by applying a negative pressure. The cell mixture was put into a tube on wet ice for at least 1 min. The tube was frozen in liquid nitrogen and stored at −80 °C until the RT reaction was carried out. Synthesis of first-strand cDNA from 3 μl (for qPCR) of total RNA (in 10 μl) was carried out with the Prime Script RT reagent Kit (DRR037A, TaKaRa). qPCR was run on iCycler (Bio-Rad) by using SYBR Green Supermix (Bio-Rad). A multiplex two-round single-cell qPCR was carried out for simultaneous detection of vesicular glutamate transporters (vGluT2), glutamic acid decarboxylase 67 (GAD67). The first PCR conditions included an initial denaturation for 5 min at 94 °C, 30 cycles with 45 s denaturation at 94 °C, 45 s annealing at 61 °C and 70 s extension at 72 °C, final elongation 7 min at 72 °C. An anliquot (5 μl) of first round PCR product used as a template for the second round PCR (40 cycles) using nested primers. The conditions were the same as described for the first round. Products were identified with gel electrophoresis with GelRed nucleic acid gel stain (Biotium). The oligonucleotide primers used for single-cell RT-PCR are listed in Supplementary Table 4.ReproducibilityExperiments were repeated independently with similar results at least three times. Micrographic images presented in figures are representative ones from experiments repeated independently: Fig. 1b (four times), Fig. 1c (eight times), Fig. 1g (four times), Fig. 2a (three times), Fig. 2b (five times), Fig. 2c (four times), Fig. 2d (three times), Fig. 3 (four times), Fig. 4a (three times), Fig. 4b (five times), Fig. 4h (nine times), Fig. 8a (six times) and Fig. 9a (five times).StatisticsStatistical analysis was performed using GraphPad Prism 7.02, using unpaired two-tailed t-test, paired two-tailed t-test, one-way ANOVA with Tukey post-hoc test, two-way ANOVA unless otherwise indicated. Values are reported as mean ± standard error of the mean (SEM). The cutoff value of significance was P = 0.05. Electrophysiology data was analyzed using Clampfit 10.7 (Molecular Devices). All figures were prepared with Photoshop CC 2017, Adobe Illustrator CS6 (version 16), Microsoft Excel 365.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Description of Additional Supplementary Files
Supplementary Movie 1
Supplementary Movie 2
Supplementary Movie 3
Reporting Summary | nature communications | [
"Article"
] | [
"Neural circuits",
"Stress and resilience"
] | IntroductionStress emotional physiological challenges ubiquitous in lives stressors upset body homoeostasis adaptive responses to stress involves evaluation of stressors resolution via optimization emotional physiological adaptations1 Behavioural adaptations include increased arousal attention vigilance1 maladaptive response to stress linked to anxiety depression neuropsychiatric conditions6–8 strategies cope with stress essential for health survival.Stress in animals results in grooming repetitive behaviours circling rocking9–12 displacement activities adaptive values self-grooming behaviour in rodents hygiene adaptive response to stress restraining over-response unravelling mechanism of stress-induced self-grooming valuable understanding neurobiological stress management mammalian brain brain areas implicated in grooming behaviour include basal ganglia13 brain stem13 cerebellum13 motor pathways components limbic system hypothalamus amygdala orbitofrontal cortex grooming focal activation of hypothalamic nuclei evoke grooming18–20 control of self-grooming social behaviour by amygdala neuronal subpopulations repeated stimulation orbitofrontal-striatal pathway generate compulsive grooming22previous studies stress to grooming focused on neuroendocrine system hypothalamic–pituitary–adrenocortical axis12 central administration of stress-related neuropeptides like corticotropin melanocyte-stimulating hormone elicit grooming26 CRH neurons orchestrate post-stress behaviours including grooming12 neural pathways stress-related responses including grooming revealed12 hypothalamic-septal pathway emotional states on grooming escape feeding behaviour identified24. relationship between stress self-grooming complex13 neural circuit basis of stress-induced grooming unresolved complete neural circuitry perceived stress grooming yet uncovered study stress circuit revealed unknown limbic circuit linking hippocampal ventral subiculum division lateral septum lateral hypothalamus regulates stress-induced self-grooming activation triggered delayed grooming resembling emotional stress with positive valence grooming neural activity of LSv necessary for emotional stress-induced grooming functional inhibition suppressed grooming stressful paradigms results advance understanding of neural circuit basis of repetitive behaviour to emotional stressResultsActivation ventral subdivision LS triggers grooming c-Fos expression rat brain limbic system body restraint extended 20 min (n = 4) protocol induced increased time grooming (Fig. robust c-Fos expression LSv confined ventral LSv not dorsal lateral septum (Fig. 1b increased c-Fos signals control animals not restraint (Fig. 1b LSv neurons micro-injection AAV9-Syn-ChR2-eYFP LSv blue light 5 min grooming grooming behaviour dominated light stimulation delay tens-of-seconds rearing-like arousal behaviour observed prior grooming (Fig. 1e). confirmed self-grooming induced by LSv activation placing littermate same arena during optogenetic stimulation (n = 3 1Activation ventral subdivision LS triggers delayed grooming behaviour Body restraint rats 20 min induced increased grooming behaviour within 10 min compared control n = 4; **P = 0.0024 Student’s unpaired two-tailed t-testc-Fos staining activation LSv not LSd restraint stress Magnified LSd LSv regions squares right Scale bar 500 μm 200 μm Optogenetic activation LSv optic fibre unilateral LSv ChR2-eYFP injection AAV9-hSyn-hChR2(H134R)-eYFP Scale bar 500 μm 5-min optogenetic activation paradigm LSv increased grooming behaviour (n = 8 **P = 0.0019 0.0008) Comparison delay time grooming arousal behaviours initial 2 min LSv stimulation Delay ***P < 0.0001 Time spent **P = 0.0041 LSv stimulation self-grooming not social grooming rat light on period n = 3 < 0.0001) Left panel implantation optic fibres targeting LSd LSv contralateral sides brain ChR2-eYFP Right panel blue light stimulation LSv (n = 4) not LSd 4) increased time grooming Scale bar 1000 μmNo significant difference in time grooming between unilateral bilateral 4) LSv stimulation P = 0.0553 two t-test). data mean ± SEM See Supplementary Table 1 information Source data file c-Fos injecting AAV9-Syn-ChR2-eYFP in LSv LSd optogenetic activation LSv not LSd induced grooming (Fig. 1g involvement LSv unilateral bilateral stimulation LSv latter induced more time grooming modest extent (Fig. unilateral LSv sufficient full self-grooming suggest LSv repetitive self-grooming related to stress.Upstream downstream nuclei LSv self-groomingTo mapped upstream downstream areas LSv Injection retrograde chlorea toxin b) 488 into LSv fluorescent signals brain regions discrete labelling VS ipsilateral hippocampus (Fig. anterograde labelling confirmed VS upstream region of LSv (Fig. micro-injected AAV9-Syn-ChR2-eYFP into LSv observed fluorescent nerve terminals in lateral hypothalamus tuberal nucleusCTB-488 injection Tulabelled cells LSv (Fig. 2d). performed experiments manipulate LSv-connected upstream downstream pathways injected AAV9-CaMKIIα-ChR2-mCherry VS light cannula implanted LSv (Fig. Optogenetic stimulation terminals VS → LSv pathway 25 Hz triggered self-grooming behaviour arousal behaviour brain slices activation ChR2-expressing VS neurons generated depolarizing response firing 1a voltage-clamp recording light-evoked excitatory postsynaptic currents (oEPSCs sensitive to 10 μM CNQX tetrodotoxin 1 μM) eliminated oEPSC 4-amino-pyridine (4, 1 mM) restored connection monosynaptic.Fig. 2Mapping upstream downstream nuclei LSv self Retrograde labelling CTB-488 microinjection LSv expressions ventral subiculum (VS) hippocampal formation ipsilateral brain enlarged Anterograde tracing injection AAV9-CamIIKα-ChR2-mCherry confirmed VS-LSv projectionAnterograde tracing AAV9-hSyn-hChR2(H134R)-eYFP microinjection LSv expressions tuberal nucleus lateral hypothalamus 1000 μm 200 μm Injection CTB-488 Tu confirmed LSv-Tu projection 1000 μm 500 μm 200 μm Optogenetic activation VS → LSv pathway 5 min 25 Hz increased grooming behaviour higher 5-min pre-light post-light-off (n = 5 ANOVA F (1.489 = 13.05 P = 0.0083 off on = 0.0189 post-off Comparison delay time grooming arousal behaviours initial 2 min stimulation n = 5 delay time 0.0004 time spent 0.0009 Optogenetic activation LSv → Tu pathway 5 min increased grooming behaviour higher 5-min pre post-off = 7 ANOVA (1.104) = 39.73 P = 0.0004 pre-off on = 0.0025 post-off 0.0010) Comparison delay time grooming arousal behaviours initial 2 min stimulation(n = 7 delay time = 0.0002 time spent = 0.0147 Student’s paired two-tailed t-test). data presented mean ± SEM Source data file injection AAV9-Syn-ChR2-eYFP into LSv optogenetic stimulation LSv → Tu pathway induced self-grooming delay time spent grooming arousal similar to stimulating LSv brain slice experiments functionality ChR2 in LSv neurons validated Fig. 1e f). LSv populated by GABAergic neurons34 confirmed GABAergic connection from LSv to Tu grooming Fig. 1g, h).Positive valence of LSv-associated grooming grooming modulated LSv positive or negative emotional state assessed desirability averseness LSv-modulated grooming preference animals stay in compartment stimulation LSv Tu pathway after time grooming conducted conditioning place preference) test after 2 × 3 min association per day between photo-stimulation for 3 days animals spent higher time in stimulation-associated chamber testLSv → Tu grooming positive affective valence. 3Positive valence LSv grooming behaviour results real-time place preference test dual chamber setup increased time optogenetic stimulation LSv F(1 6) = 8.908 P 0.0245 1.589e−031 P > 0.9999 34.35 P 0.0011 LSv → Tu pathway F(1 4.789 P 0.0712 2.139e−030 P > 0.9999 10.40 P = 0.0180 0.0043) confirmed statistical analysis time grooming subtracted Control animals injection AAV5-hSyn-eGFP n = 4 ns not significant Two-way measures ANOVA Sidak post-hoc test d–f conditioning place preference test dual chamber setupAnimals spent time stimulation chamber test LSv F(1,6) = 19.86 P = 0.0043 virus 7.353 P = 0.0350 optostimulation F(1 = 25.24 P = 0.0024 LSv → Tu pathway F(1 6) = 15.84 P = 0.0073 virus F(1 = 12.81 P = 0.0116 optostimulation 7.159 P = 0.0368 confirmed statistical analysis Control animals received AAV5-hSyn-eGFP. n = 4 ns not significant two-way repeated measures ANOVA Sidak post-hoc test data mean ± SEM Source data file VS → LSv → Tu di-synaptic limbic circuit modulates self VS → LSv Tu asked neurons LSv VS conveying grooming signals Tu injected AAV9-Syn-ChR2-eYFP VS CTB-555 Tu prepared brain slices recordings targeting LSv neurons biocytin neurons EPSCs photo-stimulation expressed CTB signal di-synaptic VS → LSv → Tu pathway pathway in vivoinjecting AAV1-Syn-Cre35 in VS retrograde retroAAV-EF1α-DIO-ChR2-mCherry in Tu LSv neurons innervated by VS projected Tu express ChR2-mCherry light LSv induced self-grooming delays frequency light stimulation rearing heading behaviours less exploratory behaviour not mediated by VS → LSv → Tu circuitry photo-stimulation evoked membrane depolarization of ChR2-mCherry-positive LSv neurons firing single-cell RT-PCR LSv neurons positive for GAD67 mRNA not glutamate neuron marker vGluT2 GABAergic nature VS → LSv → Tu circuit modulates grooming behaviour injection paradigm neurons LSv innervated VS signals to Tu AAV9-Syn-ChR2-eYFP injected into VS CTB-555 Tu biocytin-tagged LSv neuron EPSCs labelled by CTB-555 Experimental strategy role VS → LSv → Tu circuitry in groomingUpper panel AAV1-Syn-Cre injected VS retroAAV-EF1α-DIO-ChR2-mCherry Tu optical fibre implanted LSv Lower panel neurons LSv innervation VS Tu express mCherry confirmed confocal microscopy Left scale bar 500 μm Right 200 μm vivo light delivery activated VS → LSv Tu pathway increased grooming behaviour n = 8 F(1.260, = 64.89 P < 0.0001 pre-off frequency-dependent n = 4 F(1.517, 4.551) = 63.70 P = 0.0006 20 Hz = 0.0134 30 Hz 40 Hz 0.0013 50 Hz 0.0020) Time arousal grooming behaviours initial 2 min stimulation VS → LSv → Tu circuitry n = 6 < 0.0001-test Validation functional ChR2 confirmed in vitro optogenetic stimulation whole-cell recording Single-cell RT-qPCR cytoplasmic patched LSv neurons positive GAD67 negative vGluT2. data mean ± SEM Source data fileVS → LSv → Tu-induced grooming resembles emotional asked VS → LSv → Tu pathway stress-induced self-grooming exploited grooming behaviour context-sensitive reflected sequence patterns generated four grooming models related to physical emotional challenges free water bright light body restraint29 (Fig. 5a former provoke physical stress moistening fur latter emotional 5Grooming induced optogenetic activation VS → LSv → Tu circuitry resembles emotional stress-induced grooming six grooming models optogenetic activation LSv body light exposure swimming water spray spontaneously occurring OS-induced RS-induced LS-induced 8) grooming models similar average time lower SM-induced WS-induced n = 14 for ST model ANOVA OS vs. ST ***P < 0.0001 WS *P = 0.0387 SM ***P < 0.0001) grooming frequency bout duration transitions per bout variable among models One-way ANOVA *P < 0.05 **P < 0.01 ***P < 0.001Optogenetics-induced grooming body restraint light exposure models similar bout frequency duration transitions per bout symbol defined by SEM parameter average number of times spent on grooming body parts in six models as % of total 3-D plot times grooming higher similarity of optogenetics restraint light models data presented as mean ± SEM See Supplementary Table 2 for statistical information Source data provided as file previous four induced grooming models exhibited higher percentages incorrect phase transition interrupted bouts elevated stress levels (Supplementary Fig. 2a LSv-optogenetics stimulation model shares similarity with light exposure body restraint models include total time spent grooming bout frequency duration transitions per bout higher similarity of optogenetics model with body restraint light exposure model rats in models spent higher frequency in paw licking two fur moistening models higher grooming head body plot of number of times spent in different body parts similarity (Fig. 5hvalidate higher stress body restraint light exposure models conducted tests stress including latency to nest open field test light–dark box test elevated plus maze test conducted after stress induction before expression grooming behaviours results. 6d confirms body restraint paradigm highest stress followed by light exposure water spray Least stress free swimming control group water sprayed head body resulted more time grooming models cluster with whole body water spray model bout frequency/transitions per bout duration analysis. 6Assessment stress levels grooming models latency to nest in open field test = 5.525 P = = 0.0451) time in light box light–dark box test 5.252 P = 0.0015 0.0011 0.0137) time in open arms during elevated plus maze test 6.001 P = 0.0006 = 0.0113 = 0.0404) indicate significant differences between body restraint light exposure models control group n = 10 rats each group One-way ANOVA with Tukey post-hoc test3-D plot confirms body restraint paradigm associated with highest stress followed by light exposure water spray Least stress with free swimming control group. data presented as mean ± SEM Source data provided as file.Microstructures grooming triggered by stressors VS → LSv → Tu suggested patterns grooming variable reflect differences context neural mechanisms13 despite emphasis analyzing grooming structures13 methods distinguish context-dependent grooming not available developed alternative approach transition analysed frequencies of transitions calculated percentage each transition expressed as phase transition probability matrix (Fig. phase transition matrices from models uniform similarity two indices cross-correlations Euclidean distances derived pairwise comparisons indicate optogenetics model light exposure body restraint models have highest cross-correlations least Euclidean distances Hierarchical clustering analyses confirmed higher resemblance matrices transposed into binary-coded format presented as phase connection graphs similarities among optogenetics light exposure restraint models obvious.Fig. 7Microstructure analysis of grooming under contextstransition probability matrices from grooming phase to in six grooming models phases defined as paw licking 1) nose/face/head grooming 2) body 3) leg 4)/genital grooming 5) numbers show occurrence in % colour-coded correlation coefficients Euclidean distances between matrices derived Hierarchical clustering analysis supports higher similarity of optogenetics restraint light models body restraint stress optogenetics stimulation LSv bright light exposure spontaneous grooming water spray SM swimming patterns microstructures of six grooming models transformed into two-colour binary format threshold of 25% higher similarity among optogenetics restraint light model.Increased activity of LSv neurons precedes emotional stress-induced determined population dynamics in freely behaving rats by fibre photometry After stress induction calcium activities monitored aligned to start grooming bouts restraint light exposure rise in calcium signals LSv neurons before start grooming event differences in ΔF/F in pre post-grooming periods8b no changes in calcium signals in swimming 8d water spray 8e models spontaneous grooming 8f). fluorescent signals in control animals eGFP LSv neurons no change during grooming. 8Activation LSv neurons precedes emotional stress-induced grooming fibre photometry calcium activity from LSv neurons infected with AAV9-Syn-GCaMP6s or AAV5-hSyn-eGFP virus correct expression GCaMP6s placement fibre optics verified post-mortem Scale bar 500 μm results calcium activity in LSv start grooming averaged results in body restraint model light exposure swimming water spray spontaneous grooming significant in calcium activity (ΔF/F prior start grooming in body restraint light exposure models not other models non-significant not significant Student’s paired two-tailed t-test blue traces represent signals from animals eGFP not GCaMP data presented as mean ± SEM Source data fileInhibition VS → LSv → Tu circuit grooming emotional causal relationship activation VS → LSv Tu self-grooming applied optogenetic inhibition injected AAV1-Syn-Cre VS AAV9-EF1α-DIO-eNpHR-eYFP LSv VS → LSv → Tu pathway NpHR-expressing terminals Tu expression function eNpHR LSv neurons verified brain slices yellow light delivered Tu increased self-grooming restraint light exposure models level suppressed level grooming weaker light off contribution post-stress grooming light delivery affect time grooming swimming water spray models Spontaneous grooming unaffected control study non-functional AAV5-hSyn-eGFP injected VS LSv Tu No abnormality grooming light delivered validate effect optogenetic inhibition LSv → Tu terminals Tu neurons AAV-expressing eNpHR injected LSv injected AAV8-hSyn-hM3Dq-mCherry into LSv experiments patched Tu neurons held membrane potential spontaneous firing occurredCNO superfusion LSv neurons suppressed firing rate Tu neurons membrane hyperpolarization increased GABA release LSv terminals (Fig. yellow light LSv terminals GABA release decreased firing Tu neurons rectified experiments chemogenetic inhibition LSv neurons VS → → Tu pathway results same conclusion 9Activation LSv neurons necessary emotional stress-induced grooming Schematics optogenetics manipulation inhibition VS → LSv → Tu circuitry injection AAV1-Syn-Cre VS AAV9-EF1a-DIO-eNpHR-eYFP LSv implantation optical fibre Tu Scale 500 μm-cell recordings LSv neurons validated eNpHR inhibition prolonged (5 min yellow light delivery increases grooming time light-off control group body restraint light exposure smaller light delivered increases after swimming water spray similar light-off light-on trials *P < 0.05 < 0.01 < 0.001 ANOVA test Optogenetic inhibition affect time spontaneous grooming n = 6 P = 0.4861 t-testTargeted functional activation inhibition LSv terminals Tu injection AAV9-EF1a-eNpHR-eYFP AAV8-hSyn-hM3Dq-mCherry LSv recording Tu neurons spontaneous firing Tu neurons −50 mV holding level inhibited CNO recovered yellow light stimulation (n = 11 one-way repeated measures ANOVA Tukey post-hoc test **P < 0.01). data mean ± SEM Supplementary Table 3 statistical information Source data file mammals adaptive response stress model brain mechanism stress management identified di-synaptic hippocampal–septal–hypothalamus circuit rodent brain regulates self-grooming not social-grooming electrophysiological experiments LSv neurons monosynaptic excitatory innervation VS send monosynaptic GABAergic projection Tu neurons hypothalamus studies implicate involvement subiculum LSv hypothalamus regions processing stress study revealed hippocampal-septo-hypothalamic network cognitive-limbic study reveals precise connections sub-divisions nuclei modulating stress-induced response grooming behaviour association emotional stress positive valence delayed response stimulationstress-induced grooming models showed body restraint light exposure protocols activate LSv neurons peak activity optogenetic chemogenetic inhibition of VS → LSv → Tu circuit suppressed grooming triggered by paradigms evidence suggests paradigms evoke emotional physical stress swimming water spray assessment confirmed body restraint light exposure paradigms with higher stress water spray swimming findings implicate di-synaptic circuit from VS to Tu regulates grooming emotional stress optogenetic chemogenetic inhibition pathway suppress post-stress grooming likely additional pathway exists response analysis of grooming activity microstructure may index of stress anxiety13 found activation of VS → LSv → Tu circuit triggered excessive grooming microstructures resembling restraint light exposure paradigms analysed probabilities transitions among stages grooming revealed subtle differences in microstructures grooming driven stressors affirms grooming microstructure related to context microstructure of spontaneous grooming similar to restraint-induced light exposure-induced grooming water-associated models physical stress of body wetting unique patterns groomingspontaneous grooming lower frequency may internal drive motivational elements psychological need emotional stress-induced grooming our approach valuable addition to analyzing context-dependent microstructure grooming caution interpreting results microstructure could affected by factors method grooming circuitry involved in emotional stress-induced response counter-intuitive grooming response with positive valence RTPP CPP results Xu et al. 24 demonstrated negative emotional state grooming response following activation hypothalamic-septal pathway Behavioural response to stress manifests as heightened arousal attention alertness physiological associated with negative emotional state responses time-limited restraining forces over-response LSv → Tu pathway-modulated post-stress grooming implicates calming effect consistent with adaptive displacement activity to stress suppression over-response pathway may not responsible for generating stress-induced emotional state initiating grooming action well-being limbic component hippocampus subiculum on integration limbic functions46 limbic forebrain ventral division subiculum may stressor selection regulation neuroendocrine response to stress39VS excitatory or inhibitory on hypothalamo-pituitary-adrenocortical axis not clear39 findings extend understanding role VS in processing emotional stress information LSv → Tu pathway behavioural adaptation like grooming could be effected hypothalamus primary output node for limbic system49 mediating innate behaviours Tu lateral hypothalamic area including Tu implicated in grooming behaviour pathological grooming12 other hypothalamic areas involved in grooming behaviour19 relationship of Tu remains investigated LSv limbic structure associated with cognitive emotional processes41 implicated for modulation of anxiety54 fear-conditioning regulation autonomic responses to aversive stimuli32 impact LSv on stress-related behaviour controversial56–60 previous studies non-specific considered LSv as single nucleus heterogenous neuronal populations results revealed only ventral subdivision of LSv GABAergic neuronal population conveys emotional stress relevant to repetitive grooming behaviour propose LSv receives stress information from VS regulates Tu triggering grooming.After stimulation typical tens-of-seconds delay before grooming puzzlingcontrast to grooming behaviours stimulating brain dorsomedial hypothalamus orbitofrontal-striatal projection22 compulsive-like behaviour build-up time needed for manifestation response consistent with role LSv stimulation of LSv neurons could trigger other behaviours rearing-like arousal behaviour VS → LSv →Tu pathway activated grooming arousal behaviour dissociation between arousal suggests LSv involved in arousal behaviour mediated by microcircuitries other than pathway integration hub LSv heterogenous neuronal subpopulations map inputs to behaviours Xu et LSv receives emotional state signals from PVN triggers grooming escape behaviour suppression feeding with negative valence findings enrich role of LSv in regulation innate behaviors output pathway LSv to Tu GABAergic interaction between inhibitory neurons with Tu future studies aberrant response to stress compulsive repetitive behaviour in neuropsychiatric disorders6,7 obsessive-compulsive disorders22 findings uncover limbic circuit in emotional stress response basis for deciphering circuit emotional processing malfunctions to abnormal repetitive behaviours in brain.Sprague-Dawley) rats 300–320 g study bred maintained Laboratory Animal Service Centre Chinese University Hong Kong room controlled 23 °C 12-h light/dark cycle handled CUHK guidelines approved Animal Experimentations Ethics Committee experiments light phase (09:00–19:00).Viral constructsAdeno-associated viruses) AAV5-hSyn-eGFP AAV9-hSyn-hChR2(H134R)-eYFP-CaMKIIα-hChR2 AAV8-hSyn-hM3Dq-mCherry RetroAAV-EF1α-hChR2(H134R) AAV9-Ef1α-DIO-eNpHR3.0-eYFP RetroAAV-hSyn-DIO-hM4Di-mCherry AAV9-Syn-GCaMP6s-WPRE-SV40 purchased from Addgene (Watertown viral >5 × 1012 particles per ml rats anesthetized ketamine xylazine (6 stereotaxic frame micro-injections Hamilton syringe) AAV virus tracer placed target area LSv (−0.12 mm/P 6.0 VS (−5 Tu (−324 mm A/P ±1.4 mm M/L mm D/V dura 0.1–0.5 μl virus injected 10–50 nl/min needle left 10 min retraction scalp incision sutured postinjection analgesics 3 days animals 3 weeks recover virus before optical fibre implantation behavioural test fibre photometry optogenetics experiments optical fibre (200-μm core NA 0.48 0.37 fluid-injection cannula implanted two stainless steel screws secured skull dental cement 1 week recover before behavioural test location implanted fibre confirmed postmortem.Optogenetic chemogenetic ChR2 photostimulation 473-nm light laser (10 25 Newdoon optic cable (200-μm core 0.37 NA duration varied Laser power 5 mW tip fibre implanted 0.3 mm above targeted nucleus eNpHR photoinhibition two optical fibres attached cannula illumination targeted site (589 nm 10-nm laser turned on chemogenetic manipulation clozapine N-oxide) (5 μM/0.5 μl vehicle (saline) administered implanted cannula 40 mins before groomingbehavioural experiments animals videotaped behaviours evaluated offline.Retrograde anterograde recombinant cholera toxin-b AlexaFluor-488 injected target site LSv 7 days retrograde transport before sacrifice anterograde tracing AAV9-Syn-ChR2-eYFP injected LSv 4 weeks full expression Brain sections (30 μm prepared examined laser scanning microscope disynaptic circuit VS → → Tu injected VS 4 weeks expression One week sacrifice recombinant cholera toxin-b AlexaFluor-555 injected Tu Coronal slices 300 μm prepared vibrotome whole-cell patch recordings Neurons LSv blue light stimulation filled biocytin revealed Alexa FluorTM 405-conjugated streptavidin–biotin complex expressions biocytin CTB-555 examined laser scanning microscope Nikon).Behavioural animals habituated 30 min/day 3 days before testing behaviours recorded test chamber video camera Each animal tested three times verify reproducibility spontaneous grooming model animal test chamber activities recorded 20 min.restraint stress-induced grooming29 animal restrained in black tube (5 cm diameter 25 cm length 20 min test chamber 20 min video recording bright light-induced grooming animals transported dimly lit laboratory left 30 min exposure light box animal exposed bright light (60–80 W) lamp 10 cm home cage 20 min test chamber behaviour recorded 20 min optogenetically induced grooming behaviour monitored 15 min 5-5-5 min protocol blue light pulses (473 nm, 5 mW 10 ms each 25 Hz turned on 5 min Grooming behaviours video-taped social grooming two animals same chamber 5-5-5 min protocol water spray-induced grooming25 animal sprayed water squirts face belly back sterile water (25 °C “head-only” model sprayed left right face “body-only” belly back (eight squirts placed test chamber video recording 10 min animals habituated once per day 3 consecutive days swimming-induced grooming30 animal open swimming pool (100 cm × 50 cm 50 cm high filled water 25 °C Water 15 cm 5 min3 days adaptation animals swim 2 min grooming testing excess water removed before test chamber for 10 min video recording aversion rodents brightly lit areas tendency exploratory behaviour67 behaviour tests emotional status grooming-inducing paradigms open-field nest test dark shelter nest (20 cm × 20 cm 40 cm high) corner open field (100 cm × 100 cm 40 cm high). Animals habituated 10-min open-field 2 days before test testing day each rat explore box 5 min rat placed in box corner diagonal nest latency escape recorded index stress-like behaviour shorter latency higher stress level light–dark box test67 two boxes 50 cm × 40 cm 40 cm high connected by open door animal first placed in light box total time light box recorded 5 min less time light box indicates higher stress level EPM test anxiety-related behaviour in rodents67 maze central (15 cm × 15 two open arms without walls 40 two closed arms with walls Each rat in centre maze rat behaviour recorded 5 min time spent in open arms index stress-like behaviour.Behavioural analysis grooming behaviours quantified manually by three observers video editing software observers blind to experimental conditionsSocial grooming assessed when rat licked chewed fur conspecific forepaws on back neck other rat Arousal behaviours rearing heading examined analysis self-grooming number grooming bouts duration total grooming time test evaluated Self-grooming defined animal licked body parts including paws nose eyes head body legs genital interruption 6 s or more separated two bouts36.Grooming patterns microstructuresTo patterns grooming behaviour different conditions phases grooming activities defined paw licking 1) nose/face/head grooming 2) body 3) leg grooming 4) genital grooming (phase 5) no grooming defined as phase 0 recorded videos replayed at 1/4 speed noting time points start end phases sequences phases grooming bout determined grooming bout interrupted if one pause action (<6 s) recorded within transitions phases transitions between phases used evaluate ‘correct’ transitions progression stereotypy → → (3 → 4) (4 → 5) (5 → 0 ‘incorrect’ transitions total number of transitions percentages of incorrect transitions calculatedmicrostructures grooming bouts six conditions collected pooled transition probabilities calculated matrix obtained spontaneous grooming animals 112 times phase 1 (paw transitions phases frequencies 9 (1 → 96 (1 → 2) 4 (1 → 3) 2 (1 → 4) 1 (1 → 5) probabilities 8.04%, 85.71% 3.57% 1.79% 0.89% patterns phase transitions captured transition matrix connectivity Fig. 7d Fig. 4bcompare similarity transition probabilities matrices computed two indices Euclidean distance Pearson’s correlation coefficient each pair matrices\documentclass[12pt]{minimal}{amsmath\oddsidemargin-69pt}{document}$D =\mathop\sum\nolimits_{i = 1}^n {(y_i - x_i)^2}\end{document}D= ∑i=1n(yi−xi)2and2\documentclass[12pt]{minimal}{amsmath}\oddsidemargin}-69pt}{document}\mathrm{{CC\mathop\sum\nolimits{i = 1}^n (x_i - x)(y_i - \overline y{i = 1}^n (x_i - \overline x )^2{i = 1}^n (y_i - \overline y )^2\end{document}CC=∑i=1n(xi−x)(yi−y ∑i=1n(yi−y{xi} {yi} elements in transition probability matrices n number elements D computes disparity matrices from distance smaller D indicates higher similarity higher CC matrices higher similarity dendrograms of hierarchical clustering by D CC obtained using MATLAB Statistics Toolbox functions linkage dendrogram height inverted-U-shape proportional to dissimilarity between nodes closely related nodes connected at lower levels.Real-time place animal in box (120 × 50 cm, 40 cm divided into two chambers One paired with 25 Hz photostimulation other without photostimulation behaviour monitored analysed by-Maze tracking software 4.7 percentages time spent on stimulated unstimulated sides chamber in 15 min quantified.Conditioned place preference (CPP animals habituated to handling for 3 days Experiments in two interconnected chambers (60 × 40 40 cm high each separated by sliding door decorated with stripes Animal movements recorded analysed with Anymaze procedure preconditioning conditioning testing first day sliding door retracted rats explore apparatus for 15 min Animals >70% time in excluded from analysispreconditioning rats underwent conditioning both sides paired photo-stimulation (25 10 ms laser pulses 473 nm 2 × 3-min/day 3 days test phase animals treatment free access compartments 10 min movements recorded time analysed Anymaze software slice electrophysiology optogeneticsAdult rat brain slices prepared rats anesthetized isoflurane perfused 100 ml temperature artificial cerebrospinal fluid) 92 NaCl 2.5 KCl 1.25 NaH2PO4 20 NaHCO3 10 HEPES 25 Glucose 5 Na-ascorbate 2 thiourea 3 Na-pyruvate 10 MgSO4 0.5 CaCl2 12 N-acetyl-l-cysteine saturated carbogen (95% O2/5% CO2) rats decapitated brains extracted placed cutting solution 30 s Coronal sections 300 μm cut transferred chamber 34 °C NMDG aCSF recover 10 minslices transferred chamber 92 NaCl 20 NaHCO3 25 Glucose 2.5 KCl 1.25 NaH2PO4 10 HEPES 5 Na-ascorbate 2 thiourea 3 Na-pyruvate 2 CaCl2 2 MgSO4 12 N-acetyl-l-cysteine slices transferred chamber aCSF 125 NaCl 2.5 KCl 11 Glucose 26 NaHCO3 1.25 NaH2PO4 2 CaCl2 2 optically evoked action potential spontaneous firing solution 130 mM K-gluconate 10 KCl HEPES 1 EGTA 2 MgCl2 2 Na2-ATP mM Na3-GTP currents 50 mM KCl 10 HEPES 1 EGTA 2 MgCl2 Na2-ATP GTP Light pulses 470 nm stimulator ChR2 neurons voltage-clamped −70 mV Access resistance<20 MΩ compensated monitored Recordings terminated input resistance >30% 20 MΩSignals acquired MultiClamp 700B amplifier Clampex 10.4 software Digidata 1550 interface Responses filtered 3 kHz digitized 10 kHz analyzed Clampfit 10.7 biocytin included internal solution slices fixed overnight 4% paraformaldehyde Filled neurons identified streptavidin-conjugated Alex 405 (ThermoFisher).Fibre neuronal activity rats injected 100 nl AAV9-Syn-GCaMP6s-WPRE-SV40 AAV5-hSyn-eGFP right LSv Two weeks optical fibre (NA 0.48 Newdoon Technology implanted right LSv 1-week recovery fibre connected FibreOptoMeter) optical fibre patch cord (200-μm, 0.53 NA recording fluorescence signals beam 470 LED reflected dichroic mirror lens PMT LED power <50 μW Fluorescent signals collected synchronized behaviour CineLyzer system smoothed ΔF/F values exported Matlab represented heatmaps averaged response 9 s before 3 s after grooming onset values calculated raw fluorescence averaged FAVG (n) sliding time window (0.75 s)baseline for frame (FBASELINE (n)) was minimum of FAVG (n) in time window (3 s) preceding ΔF calculated by subtracting FBASELINE (n) from FRAW (n) dividing by FBASELINE (n). ΔF smoothed with exponentially weighted moving average exponential weighting described by time constant\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts{upgreek\oddsidemargin-69pt} averaging window set to 5*[12pt]{minimal}{amsmath-69pt frame rate system is 30.[12pt]{minimal}{amsmath}{wasysym}{upgreek}\oddsidemargin{-69pt}{document}\mathrm{\Delta }}F/{F}({n}\mathrm{unsmoothed}}} =\frac{{{F}\mathrm{RAW}}}({n) - {F\mathrm{BASELINE}}}({n}\mathrm{BASELINE}}}\left( {n \right{document}ΔF/F(n)unsmoothed=FRAW(n)−FBASELINE(n[12pt]{minimal}{amsmath}{wasysym}}}}}{upgreek}\oddsidemargin{-69pt}{document}${N}{\tau 0} =\mathrm{floor}}(5\mathrm{frame}}\mathrm{rate}}{document}Nτ0=floor(5*τ0*framerate[12pt]{minimal}{amsmath}{wasysym}{mathrsfs}{upgreek}\setlengthsidemargin-69pt\Delta/(}\mathrm{smoothed\nolimits\tau/( - x)\mathrm{unsmoothed\left\right\tau\nolimits\tau\left\right\end}ΔF/F(n)smoothed=∑0Nτ0ΔF/F(n−x)unsmoothede−xNτ0∑0Nτ0e−xNτ0HistologyRats anaesthetized perfused PBS 4% PFA PBS brain fixed 4% PFA 4 °C overnight Fixed samples sectioned 30-μm coronal sections cryostat (ThermoFisher). c-Fos staining brain samples prepared 60 min test fixation blocked 10% normal goat serum 0.3% Triton X-100 2 h incubated anti-c-Fos antibody (1:2000 4 °C overnight incubated goat anti-rabbit lgG secondary antibody (1:1000 Invitrogen solution sections rinsed PBS counterstained DAPIImages confocal laser microscope Nikon).Single cell RT-qPCRPatch pipettes filled 3.0–5.0 μl RNAse-free solution microloader tip diameter 1/3 cell body cell harvested into pipette negative pressure cell mixture tube wet ice 1 min frozen liquid nitrogen stored −80 °C until RT reaction Synthesis first-strand cDNA from 3 μl Prime Script RT reagent Kit (DRR037A qPCR iCycler (Bio-Rad) SYBR Green Supermix multiplex two-round single-cell qPCR detection vesicular glutamate transporters glutamic acid decarboxylase 67 first PCR conditions initial denaturation 5 min 94 °C 30 cycles 45 s denaturation 94 °C 45 annealing 61 °C 70 s extension 72 °C final elongation 7 min 72 °C (5 μl) first round PCR template second round PCR (40 cycles nested primers conditions same Products identified gel electrophoresis GelRed nucleic acid gel stain oligonucleotide primers single-cell RT-PCR Supplementary Table 4.ReproducibilityExperiments repeated similar results three times images 1b 1c 1g 2a2c 2d 3 4a 4b 4h 8a 9a (five analysis GraphPad Prism 7.02 unpaired two-tailed t-test paired one-way ANOVA Tukey post-hoc test two-way ANOVA Values reported mean standard error cutoff value significance P = 0.05. Electrophysiology data analyzed Clampfit 10.7 (Molecular Devices). figures prepared Photoshop CC 2017 Adobe Illustrator CS6 Microsoft Excel 365.Reporting Nature Research Reporting Summary.Supplementary information Additional Supplementary Files Movie 1 2 3 Reporting Summary | 46.5 | 1.29421 |
10.1038/s41467-020-17864-4 | PMC7445303 | Active colloidal systems can serve as an enabling platform to study complex out-of-equilibrium physical phenomena. Using a magnetic control with a feedback loop, here the authors program the dynamics of active Brownian particles by updating their rotational diffusion coefficient depending on their locations. | The non-thermal nature of self-propelling colloids offers new insights into non-equilibrium physics. The central mathematical model to describe their trajectories is active Brownian motion, where a particle moves with a constant speed, while randomly changing direction due to rotational diffusion. While several feedback strategies exist to achieve position-dependent velocity, the possibility of spatial and temporal control over rotational diffusion, which is inherently dictated by thermal fluctuations, remains untapped. Here, we decouple rotational diffusion from thermal fluctuations. Using external magnetic fields and discrete-time feedback loops, we tune the rotational diffusivity of active colloids above and below its thermal value at will and explore a rich range of phenomena including anomalous diffusion, directed transport, and localization. These findings add a new dimension to the control of active matter, with implications for a broad range of disciplines, from optimal transport to smart materials. | IntroductionThe behavior of self-propelling colloidal particles sheds light on far-from-equilibrium physics and offers tantalizing opportunities to perform tasks beyond the reach of other micro- and nanoscale systems1. Many of these functions are inspired by the striking similarity that synthetic active matter exhibits with living systems such as motile bacteria. This analogy therefore also provides an ideal opportunity to understand the motion and (self-)organization of living systems through synthetic models2. The fundamental mathematical description of how active colloids move is given by active Brownian motion1: a microscopic particle of radius R in a fluid of viscosity η moves with constant speed v, while its orientation diffuses at a rate set by the rotational diffusivity DR = (kBT)/(8πηR3), with kBT being the characteristic thermal energy at absolute temperature T and kB the Boltzmann constant. This minimal model has been successfully employed to describe a wealth of phenomena, from the motion of active particles in complex structures3 to the optimization of search strategies4.Recently, there has been a growing interest in pushing the control of synthetic active matter beyond the standard active Brownian particle (ABP) model to mimic more complex behaviors, including directed transport and pattern formation. These phenomena typically arise when the particle velocity or the environmental fluctuations vary in space and time. For example, a position-dependent translational diffusivity has been proposed as a fundamental biological mechanism leading to anomalous diffusion and localization of biomolecules in cellular membranes5, while the temporal control of random walks enables the emergence of collective motion in active colloids6. The effect of a position-dependent velocity has also been investigated as a means to control the organization and the area explored by active particles (artificial and biological) as well as their interactions7–16. Beyond their fundamental relevance, these mechanisms can also be exploited for applications ranging from environmental remediation to targeted drug delivery1.Because translation and rotation in ABPs are coupled, introducing a feedback between rotational dynamics and position also provides a means to control active Brownian motion. Biological swimmers, such as chemotactic bacteria17, are in fact known to tune their rotational dynamics to climb up or down chemical gradients in order to localize food sources or to escape harmful chemicals. However, while biological swimmers can do so by varying their reorientation frequency (or tumbling rate)18, which is an internal degree of freedom, the rotational dynamics of a synthetic active particle is usually dictated by thermal fluctuations.Here, we control the rotational diffusivity of individual ABPs by decoupling the amplitude of the rotational fluctuations from the thermal bath. Through randomly-oriented magnetic fields and a discrete-time feedback loop, we spatially and temporally modulate the effective rotational temperature, above and below the environmental temperature. This allows us to study the effect of a position-dependent rotational diffusivity on the statistics of active Brownian motion. In analogy with biological and artificial sensor-actuator systems that rely on temporal sampling17,19, we also consider that the feedback between DR and the particle’s position is not instantaneous, but mediated by a discrete sampling of position that results in a finite sensorial delay. We find that periodic space-time modulations of the rotational dynamics bring about a broad range of exotic phenomena, ranging from anomalous diffusion reminiscent of glassy dynamics to directed transport and localization. We support our results with numerical simulations, which also indicate new directions for future developments.ResultsControlling rotational dynamicsOur model ABPs self-propel due to induced-charge electro-phoresis20–22. They consist of 4 μm-diameter silica Janus colloids, half-coated with a 120 nm-thick nickel cap, which is magnetized in the direction perpendicular to the Janus boundary (see inset in Fig. 1a). In this way, the propulsion’s direction dictated by the compositional asymmetry is aligned with the caps’ magnetic moment and can be externally controlled by a magnetic field. We let the particles sediment at the bottom of a liquid cell enclosed by two planar transparent electrodes, which are separated by a vertical gap h = 120 μm, and record the colloids’ position and cap orientation at a frame rate of 10 fps by video microscopy (see Methods for more details). The colloids swim over the bottom substrate due to locally unbalanced electrohydrodynamic flows generated by a spatially uniform 1 kHz AC electric field applied across the electrodes (Fig. 1a). The swimming velocity v is proportional to the square of the electric field \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\propto {({V}_{{\rm{pp}}}/h)}^{2}$$\end{document}∝(Vpp/h)2, where Vpp is the peak-to-peak voltage, which is varied in the range 1–10 V.Fig. 1Controlling rotational dynamics through randomly-oriented magnetic fields.a Schematic of the experimental setup. A Janus silica particle, half-coated with a magnetized Ni cap, undergoes self-propulsion at the bottom of a liquid cell enclosed by two transparent electrodes. Propulsion stems from induced-charge electro-phoresis generated by an AC electric field \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{E}$$\end{document}E→ perpendicular to the plane in which the particle moves (the curved yellow arrows depict local unbalanced electrohydrodynamic flows). A uniform magnetic field \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{B}$$\end{document}B→ (blue arrow), produced by four coils, controls the in-plane orientation of the particle, while its motion is observed with an optical microscope. The inset shows an optical micrograph of a 4 μm particle, with the Ni cap in black. The white arrow shows the particle’s orientation angle θ, which is aligned with the cap’s magnetic moment and thus with the magnetic field. Scale bar: 5 μm. b Imposed θ as a function of time for three values of DR. The colored bands delimit a 2π range. Data are shifted along the y-axis for clarity. c Probability distributions of angular displacements G(Δθ, Δt) for different DR and lag time Δt = 0.1 s (dashed lines: imposed Δθ, solid lines: measured Δθ) with the same colors as in b. The gray symbols show the measured G(Δθ, Δt) without magnetic field, corresponding to the thermal \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{th}}}$$\end{document}DRth (gray line: Gaussian fit, see Supplementary Table 1). d Trajectory of an ABP with v = 5.5 μm s−1 and an imposed DR varying over time. e MSD of an ABP with v = 5.5 μm s−1 for different imposed DR: 0.1 (diamonds), 1 (triangles), 5 (circles) and 10 (squares) rad2 s−1 (colors as in d). f Persistence length (LP) as a function of rotational relaxation time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{-1}$$\end{document}DR−1 for different values of v: 8.2 (circles), 6.7 (triangles), and 2.7 (squares) μm s−1. The inset shows LP as a function of v for the thermal \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{th}}}$$\end{document}DRth = 0.014 rad2 s−1 (gray squares) and for an imposed DR of 0.07 and 0.144 rad2 s−1 (black triangles and circles, respectively). Error bars correspond to the data standard deviation.We control the orientation angle θ of the colloids’ cap (see microscopy image in Fig. 1a) by two pairs of independent Helmholtz coils generating spatially uniform magnetic fields of any in-plane orientation (Fig. 1a, Supplementary Fig. 1, and Supplementary Movie 1). In contrast to previous works, where magnetic fields are used to remote-control active colloids23–27, we randomize the direction of the magnetic field to endow the colloids with an externally controlled rotational diffusivity, which is decoupled from the thermal bath and the propulsion scheme (Supplementary Movie 2). We vary the orientation of the magnetic field at f = 1 kHz by random angular displacements Δθ drawn from a Gaussian distribution with zero mean and variance σ2 = 2DR/f, where DR is the imposed rotational diffusivity. Fig. 1b shows the orientation angle of the magnetic field as a function of time for values of imposed DR ranging from 10−2 to 10 rad2 s−1. The corresponding distributions of the particles’ angular displacements G(Δθ, Δt), measured from the cap orientation, attest that θ and the direction of the magnetic field diffuse according to the same Gaussian process (Fig. 1c). By letting θ diffuse over the entire 2π range, we can therefore enforce effective values of DR that are above and below the thermal rotational diffusivity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{th}}}$$\end{document}DRth (1.4 × 10−2 rad2 s−1 at room temperature). In particular, we can achieve rotational dynamics that are orders of magnitude faster than what the thermal bath would otherwise dictate (rotational cooling is also shown in the SI for 2 μm-diameter colloids in Supplementary Figs. 2 and 3).External, independent control on DR and v enables us to adjust the persistence of particle trajectories in real-time. For example, as demonstrated in Fig. 1d, we can gradually increase the propensity of an ABP to move along straight paths by decreasing DR over time in a step-wise fashion from 10 to 10−1 rad2 s−1, while keeping v constant (Supplementary Movie 3, Supplementary Fig. 4 and Supplementary Movie 4 show the complementary case in which DR is kept constant and v is varied in a step-wise fashion. See also Supplementary Supplementary Fig. 2 and Supplementary Fig. 3 for 2 μm colloids). An analysis of the mean squared displacements (MSD), calculated for each segment at a fixed DR (Fig. 1e), shows that the timescale at which the ABP’s motion goes from being ballistic to being diffusive is proportional to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim\! {D}_{{\rm{R}}}^{-1}$$\end{document}~DR−1, as expected. As a result, the MSD at long times is larger for lower values of DR, suggesting the possibility of controlling the area explored by an ABP by varying DR on demand. By systematically varying the imposed DR while fixing the AC voltage, and thus v, we can extract both DR and v from the MSD to calculate the persistence length as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${L}_{{\rm{P}}}=v{D}_{{\rm{R}}}^{-1}$$\end{document}LP=vDR−1. Fig. 1f and its inset show that LP can be varied over a wide range of values, displaying the expected linear scaling on both control parameters. These data thus attest to our ability to engineer the motion of ABPs by independently controlling v and DR using electric and magnetic fields.A feedback loop for position-dependent rotational dynamicsTo study the impact of space-time modulations of the rotational dynamics on the statistics of active Brownian motion, we implemented a discrete-time feedback loop that updates DR based on the ABP’s position r(t) (Fig. 2a, Methods, and Supplementary Figs. 5 and 6). Similarly to the case of the non-instantaneous response of motile microorganisms to environmental cues17, we update DR at regular intervals based on the past ABP’s position r(t − τ), where t = nτ, τ is the sampling period and n is the number of samples. This is realized in the experiments by holding DR constant between consecutive sampling periods, and in the Langevin dynamics simulations by letting the rotational friction vary according to a zero-order hold (ZOH) model19, as described in the Methods section (see Fig. 2a). Moreover, in analogy with Brownian motion in periodic potentials, which is a paradigmatic model for the description of anomalous diffusion28–31, here we let DR vary according to a checkerboard pattern of alternating square regions of size L. In each square, DR takes on either a high (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH) or a low (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL) value (Fig. 2b). Specifically, we consider scenarios where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}=10$$\end{document}DRH=10 rad2 s−1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}=0.01$$\end{document}DRL=0.01 rad2 s−1, and L/v such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{1}{{D}_{{\rm{R}}}^{{\rm{L}}}}> \frac{L}{v}> > \frac{1}{{D}_{{\rm{R}}}^{{\rm{H}}}}$$\end{document}1DRL>Lv>>1DRH (Fig. 2c, d and Supplementary Movie 5 and Supplementary Fig. 7). These choices imply that the motion is predominantly diffusive in one region and ballistic in the other, with two well-separated relaxation timescales.Fig. 2Position-dependent rotational dynamics: sampling a checkerboard pattern with a sensorial delay.a (top) Block diagram illustrating the feedback loop enforced both in experiments and simulations to achieve space-time modulations of the rotational dynamics. The feedback function that couples DR with the particle’s position r(t) consists of four blocks in series: sampling, delay, zero-order hold (ZOH), and DR = f(r). a (bottom) Schematic representation of how r(t) is sampled at discrete time points t = nτ (corresponding to r[n]) and reconstructed via a ZOH model (r*(t)) after a delay of τ, i.e., using the position r[n-1] as input. b Schematic of an ABP moving over a checkerboard pattern of rotational diffusivity. c–d Side-by-side comparison of experimental (black) and simulated (red) trajectories of ABPs with DR varying according to the checkerboard pattern. (v = 3.5 μm s−1, L = 32 μm, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH = 10 rad2 s−1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}=0.01$$\end{document}DRL=0.01 rad2 s−1, and τ = 0.4 s).Given the existence of multiple parameters that affect the dynamics and the spatial organization of the ABPs, in the following subsections we examine them, and highlight their main contributions, separately. We begin by determining the role of the characteristic timescale L/v and continue with the effect of the sampling period τ, before concluding with an analysis on particle localization.Non-Gaussianity: L/v and exponential tailsWe find that our feedback scheme coupled with modulations of DR brings about different types of anomalous diffusion at different time and length scales, depending on the sampling period τ and the timescale L/v.We begin by examining the statistics of particle motion for nonzero but small sampling periods τ << L/v. In this regime, while the trajectories within each region are either ballistic or diffusive (Fig. 2c–d), the overall dynamics presents unique features. The distribution of one-dimensional displacements, rescaled by L, G(∣x∣/L, Δt) (Fig. 3a–d) presents different types of non-Gaussianity at different lag times Δt depending on the ratio Δt(L/v)−1. We quantify the departure from Gaussian behavior in terms of the excess kurtosis (Fig. 3e): \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma =\frac{\langle {x}^{4}\rangle }{{\langle {x}^{2}\rangle }^{2}}-3$$\end{document}γ=⟨x4⟩⟨x2⟩2−3, where γ = 0 for a normal distribution. For Δt << L/v, G(∣x∣/L, Δt) is broader and flatter than a normal distribution (γ < 0), which is consistent with the broadening of G(x, Δt) reported for ABPs at intermediate timescales, when the motion is dominated by ballistic segments32. However, as Δt approaches L/v, G(∣x∣/L, Δt) develops into a leptokurtic distribution (γ > 0) characterized by a Gaussian peak at small ∣x∣ followed by an exponential tail up to ∣x∣ ≃ L, after which it rapidly decays.Fig. 3Non-Gaussian statistics and exponential tails due to DR varying according to a checkerboard pattern.Experimental and simulated data are plotted as circles and red lines, respectively. a–d Probability distributions of rescaled one-dimensional displacements G(∣x∣/L, Δt), for Δt(L/v)−1 = 0.02 (a), 0.2 (b), 1 (c), and 5 (d). One-dimensional displacements in x and y are cumulated to increase statistics. The black lines in the third panel (Δt(L/v)−1 = 1) are a Gaussian and an exponential fit to G(∣x∣/L, Δt) in the range ∣x∣/L ≃ 0−0.2 and ≃0.4−1, respectively. e Excess kurtosis γ of G(x/L, Δt) as a function of Δt(L/v)−1. f Rescaled mean squared displacement MSD/L2 as a function of Δt(L/v)−1. The yellow and blue straight lines are the theoretical MSDs of ABPs with constant DR, equal to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH, respectively (v = 4 μm s−1, L = 32 μm, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH = 10 rad2 s−1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}=0.01$$\end{document}DRL=0.01 rad2 s−1, and τ = 0.2 s).Exponential tails in G(x, Δt) have already been observed in glassy systems and for the diffusion of colloids in macromolecular environments33–35. They are often explained in terms of dynamic heterogeneity, which is the coexistence of faster and slower particles, which explains their appearance also in our system. At timescales Δt ≃ L/v, we also have two distinct populations of particles, which travel over very different length scales depending on where they reside on the checkerboard: diffusive in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions and ballistic in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-regions. Because \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t\simeq L/v\;> > \;{D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}Δt≃L/v>>DRH, ABPs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions are effectively diffusive, with a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{e}}ff}\simeq {v}^{2}/{D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}Deff≃v2/DRH1, and thus their displacements are normally distributed. They are responsible for the Gaussian peak, and travel distances much smaller than L. Conversely, ABPs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-regions move in a ballistic fashion because \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t\simeq L/v\;<<\;1/{D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}Δt≃L/v<<1/DRL, and thus their displacements scale as vΔt, which are of order L. Interestingly, while dynamic heterogeneity in glassy colloidal systems arises from hindered translational diffusion due to steric interactions with other particles36, here it is the result of a spatially heterogeneous rotational dynamics. As we will show later, the parallel with glassy systems goes even further.As we examine dynamics at longer Δt, the excess kurtosis γ attains a positive maximum at Δt ≃ L/v (Fig. 3e and Fig. 4a–b) and later decays to zero, indicating that Gaussian statistics is eventually recovered at long times, as expected. Interestingly, the MSD exhibits a super-diffusive scaling ~Δt1.7 up to the same critical timescale, after which the diffusive regime (~Δt1) is gradually recovered in the limit of long time scales (see Fig. 3f). This in stark contrast to the dynamics of ABPs with constant DR. In the latter case, γ is negative at intermediate timescales and zero on short and long timescales, and a transition from a ballistic (MSD ~ t2) to a diffusive (MSD ~ t1) scaling of the MSD is found at Δt ≃ 1/DR32.Fig. 4Role of the timescale L/v and sampling period τ on the statistics of particle motion.a Simulated evolution of the excess kurtosis γ of the distribution of one-dimensional displacements G(x/L, Δt) with Δt for different values of L/v at a constant τ = 0.4 s. b Experimental and simulated Δt at which γ attains its maximum as a function of L/v (τ = 0.2–0.4 s). Error bars denote 95% confidence intervals. c–d Simulated evolution of γ with Δt(L/v)−1 for τ < L/v (c) and τ > L/v (d), with L = 32 μm and v = 3.5 μm s−1. The black circles corresponds to τ = 0. e–h Simulated evolution of the rescaled mean squared displacement MSD/L2 with Δt(L/v)−1 for different values of L/v and τ.Finally, at a given τ << L/v, the timescale L/v dictates not only the Δt at which the maximum degree of non-Gaussianity is attained, but also its extent (Fig. 4a, b). This is again a direct consequence of the motion being the combination of distinct ballistic and diffusive segments. The excess kurtosis γ grows with Δt because the extent of the exponential tail grows faster with Δt (~Δt) than the broadness of the Gaussian peak (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim\!\sqrt{\Delta t}$$\end{document}~Δt). Given that ABPs can travel in a ballistic fashion only up to ~L, the tail of G(∣x∣/L, Δt), and thus γ, keeps growing up to Δt ~ L/v. Hence, the greater L/v, the greater the maximum γ, which is correspondingly attained at larger Δt. It is worth noting that, in the limiting case of small \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L/v\;\lesssim\; 1/{D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}L/v≲1/DRH, an exponential tail cannot develop and as a result γ remains negative, approaching zero in the limit of long times, as in the case of ABPs with constant DR (Fig. 4a).Non-Gaussianity: τ and subdiffusionThe sampling period τ has a qualitative impact on the statistics of motion depending on its value relative to the timescale L/v (Fig. 4c–h). In particular, we identify three regimes: τ = 0, which corresponds to an instantaneous update of DR based on the ABP’s position, τ < L/v and τ > L/v.For τ = 0 (Fig. 4c, black circles and Supplementary Fig. 8) γ attains multiple local maxima at Δt(L/v)−1 = 1, 3, and 5, with the global maximum being at Δt(L/v)−1 = 3. Such maxima in γ also mark changes in the scaling of the MSD (Fig. 4e), which exhibits a superdiffusive scaling ~Δt1.7 up to Δt ≃ L/v, after which it starts to saturate and attain a subdiffusive scaling (MSD ~ Δta, a < 1) up to Δt ≃ 3–5L/v, with higher L/v leading to smaller a. For Δt >> L/v, the MSD eventually recovers the diffusive scaling (a = 1).The emergence of subdiffusion reinforces the previously mentioned analogy with glassy dynamics. Nonetheless, in our system, there is no physical caging by neighboring particles37. Subdiffusion is instead the result of an “effective dynamical caging”, arising from the randomization of the direction of motion of ballistic ABPs as these enter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions. In fact, up to Δt ≃ L/v, the MSD is dominated by ballistic segments of length ~vΔt in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-regions. Conversely, over the same timescale, diffusive ABPs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions only travel comparatively negligible distances \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim\! v\sqrt{\Delta t/{D}_{{\rm{R}}}^{{\rm{H}}}}$$\end{document}~vΔt/DRH. However, over timescales ≳L/v, ballistic ABPs cannot travel distances greater than ≃L without crossing into a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-region. As they do so, there is a finite probability that they diffuse back and cross the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-region from which they came in the opposite direction (Supplementary Movies 6 and 7). Such reflection events give virtually null displacements over timescales up to ~2L/v, the time ballistic ABPs take to cross a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-region and travel back. The MSD then grows only as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\simeq\! {v}^{2}/{D}_{{\rm{R}}}^{{\rm{H}}}\Delta t$$\end{document}≃v2/DRHΔt for those particles that cross a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-region an odd number of times and remain in a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-region. Finally, such a dynamical caging, and the subsequent subdiffusive motion, become less prominent with decreasing L/v, as the difference between diffusive and ballistic displacements diminishes (Fig. 4e).Moving to nonzero values of τ brings about qualitative changes to the characteristics of γ depending on whether τ < L/v or τ > L/v. Up to τ ~ L/v, higher values of τ translate into higher values of γ, into the increased prominence of the maximum at Δt ≃ L/v, and into the gradual disappearance of local minima (Fig. 4c). Because nonzero values of τ introduce a finite delay in the update of the rotational dynamics, ballistic ABPs can penetrate into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions up to lengths ~v(2τ) before adjusting their rotational dynamics (see Fig. 2a). This not only increases the effective distance that ABPs can travel in a ballistic fashion, but can also allow them to cross entire regions without adjusting their rotational dynamics. The increase in the maximum value of γ with increasing values of τ is therefore due to a larger disparity between the relative contributions of ballistic and diffusive displacements.Increasing τ beyond ≃L/v leads to the gradual decrease of γ across intermediate timescales and the disappearance of a maximum in the limit of τ ≫ L/v (Fig. 4d). For τ > L/v, the modulations of the rotational dynamics start to depart considerably from the inherent periodicity of the checkerboard pattern. In the limit of large τ, the rotational dynamics of the ABPs is determined by their initial position rather than the region over which they move. This implies the existence, at all times, of two different populations of particles moving in an either ballistic or diffusive fashion, whose DR remains constant for a period of time equal to τ. In this case, the overall dynamics is the mere superposition of the dynamics of ABPs with different DR. Therefore, in the limit of large τ, γ does not present a maximum and remains negative or close to zero (Fig. 4d).Values of τ > 0 also lead to the disappearance of subdiffusion at Δt ≃ L/v (Fig. 4f–h). Nonetheless, the MSD still displays a cross-over between a superdiffusive and a diffusive scaling on short and long timescales, respectively. The disappearance of subdiffusion is again due to the fact that a finite τ allows ballistic ABPs to penetrate into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions up to greater lengths before updating their DR. This fact not only minimizes the probability that ABPs diffuse back into the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-region from which they entered, but also increases the chances that ballistic ABPs cross entire \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{R}^{H}$$\end{document}DRH-regions without updating their DR, thus contributing to higher MSDs. For analogous reasons, increasing τ also causes the superdiffusive-to-diffusive transition to take place at Δt > L/v because ballistic ABPs can travel distances greater than L.LocalizationWhile a position-dependent rotational dynamics alone cannot sustain pattern formation38,39, the finite sensorial delay introduced by the sampling (τ > 0) in the feedback loop leads to the localization of ABPs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions (Fig. 5). Interestingly, the degree of localization is a nonmonotonic function of τ(L/v)−1. For instantaneous updates of DR (τ = 0), the time-averaged steady-state spatial distribution ρ(x, y) is homogeneous, with no manifestations of the underlying DR pattern (Fig. 5a, numerical simulations). As τ is increased, ρ(x, y) increases in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions at the expense of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-regions up to a maximum degree for a critical τ ≃ 0.1−0.3L/v, before returning to a homogeneous distribution in the limit of large τ(L/v)−1 (Fig. 5a).Fig. 5Emergence of localization for finite sampling period τ.a Simulated particle density distribution ρ(x, y) for different values of τ(L/v)−1, rescaled by the uniform distribution ρeq = 1/A, where A is the area of the simulation box. ρeq is recovered for a spatially homogeneous DR or τ = 0. The simulation parameters are v = 3.5 μm s−1, L = 32 μm, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH= 10 rad2 s−1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL = 0.01 rad2 s−1. The distributions are obtained by binning the positions of 2.5 × 104 particles for a simulated time of 900 s, after letting the particles move from their initially random positions for 100 s. Periodic boundary conditions are enforced over a square simulation box of size 10L × 10L. b Simulated ratio between the average particle densities in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-regions, ρH/ρL, as a function of τ for different values of L/v. ρH/ρL is calculated as 〈TH/TL〉, where TH and TL are the residence times in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-regions, respectively. c Simulated maximum ρH/ρL (stars) as a function of L/v, fitted to a linear model (black line). d–e Simulated and experimental distribution of d
TH and e
TL, normalized by the effective diffusive timescale \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{L}^{2}{D}_{{\rm{R}}}^{H}}{2{v}^{2}}$$\end{document}L2DRH2v2 and the ballistic timescale L/v, respectively. f Experimental and simulated ρH/ρL as a function of τ(L/v)−1 for different L/v, rescaled by the corresponding maximum. Lines and circles in d–f are color-coded based on L/v according to colormap of b. Error bars denote 95% confidence intervals.We further quantify the degree of departure from the homogeneous distribution by studying the evolution of the ratio between the simulated average ρ(x, y) in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-regions: ρH and ρL, respectively (Fig. 5b, c). In agreement with the density maps shown in Fig. 5a, the ρH/ρL ratio is 1 in the limit of small or large τ and presents a maximum at intermediate values of τ (Fig. 5b). In particular, the maximum ρH/ρL ratio is a linearly increasing function of L/v (Fig. 5c).The existence of a maximum degree of localization and its dependence with L/v can be explained with a simple transport argument based on the dynamic asymmetry introduced by finite values of τ. As previously mentioned, ballistic ABPs can in fact penetrate into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions up to lengths ≃v(2τ) before updating their DR, whereas diffusive ABPs keep diffusing up to lengths \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim\! v\sqrt{{D}_{{\rm{R}}}^{-1}\tau }$$\end{document}~vDR−1τ in between updates. The deeper ballistic ABPs can penetrate into diffusive regions, the longer it takes them to diffuse out. Therefore, nonzero values of τ imply that, overall, particles end up spending more time in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions. This picture is in qualitative agreement with the higher degree of localization in the center of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-regions for τ ≃ 0.3L/v (Fig. 5a).More quantitatively, because at steady state the net flux between regions must be zero, we can write:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{\rho }^{{\rm{H}}}}{{T}^{{\rm{H}}}}=\frac{{\rho }^{{\rm{L}}}}{{T}^{{\rm{L}}}},$$\end{document}ρHTH=ρLTL,where TH and TL are the average residence times of the ABPs in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH- and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL-regions, respectively. Therefore, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{m}}ax\left({\rho }^{{\rm{H}}}/{\rho }^{{\rm{L}}}\right)$$\end{document}maxρH/ρL is equal to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\max \left({T}^{{\rm{H}}}/{T}^{{\rm{L}}}\right)$$\end{document}maxTH/TL. For 0 < τ << L/v, we expect TH and TL to scale as ~(L/v)2 and ~L/v, respectively, because the residence time in a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH-region depends on the time that an ABP takes to penetrate into it (~L/v) and to diffuse out of it (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim \frac{{L}^{2}}{2{v}^{2}/{D}_{{\rm{R}}}^{{\rm{H}}}}$$\end{document}~L22v2/DRH). Hence, the maximum of the ratio ρH/ρL = TH/TL scales linearly with L/v, as shown in Fig. 5c.The different scaling of TH and TL is also confirmed by the experimental and simulated residence time distributions at different L/v (Fig. 5d–e). In particular, the distributions of TH and TL collapse onto the same master curves, when rescaled by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{L}^{2}}{2{v}^{2}/{D}_{{\rm{R}}}^{{\rm{H}}}}$$\end{document}L22v2/DRH and L/v, respectively, and quickly drop to zero for rescaled values of TH and TL greater than 1. Moreover, by renormalizing the ρH/ρL ratio by its respective maximum for a given L/v, and plotting it as a function of τ(L/v)−1, we find that both simulated and experimental data collapse onto a single master curve (Fig. 5f).Finally, in the limit of τ > L/v the degree of localization starts to drop because at such large sampling periods the rotational dynamics is decoupled from the inherent periodicity of the underlying checkerboard pattern. This cross-over can be viewed in terms of the Nyquist–Shannon’s theorem19. When τ ~ L/v, the frequency at which the ABPs sample their environment is comparable to the highest frequency at which the ABPs can cross a region of the checkerboard pattern. This means that for τ > L/v the ABPs cannot sense changes of DR happening on a length scale L. At all times, the ABPs will retain a given DR for a period equal to the sampling period. This leads to two different populations moving in an either ballistic or diffusive fashion depending on their respective initial positions, which are updated every τ.Interestingly, for τ > L/v, the disconnect between the ABP’s sampling resolution and the spatial periodicity of the underlying pattern brings about a degree of localization that oscillates over time (Fig. 6). Nonetheless, such oscillations are damped for τ >> L/v due to lingering correlations of particle motion based on past positions introduced by the delay component in the feedback loop. By removing the delay, while retaining the discrete sampling, such that DR is updated every t = nτ according to the ABP’s current position r(t) rather than the past one r(t − τ), we can obtain instead persistent oscillations with a period equal to τ (Fig. 6a, b).Fig. 6Oscillating localization for τ > L/v with and without delay in the feedback loop.a Simulated time evolution of ρH/ρL, directly calculated from the density distribution ρ(x, y) at each time step. DR is updated every t = nτ based on the past particle’s position r(t − τ). b Same as in a except for that DR is updated every t = nτ based on the current position r(t = nτ). The simulation parameters are v = 3.5 μm s−1, L = 32 μm, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH = 10 rad2 s−1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL = 0.01 rad2 s−1.DiscussionOur findings illustrate that engineering the feedback between the internal dynamics (e.g., DR, v) and the state (r(t)) of ABPs allows tailoring their response, both in terms of the statistics of motion at the single-particle level and in relation to their global spatio-temporal organization. In particular, we show that the response is defined by the balance of timescales in the system, the ones characteristic of active Brownian motion. i.e., set by v and DR, and the externally imposed ones, set by the modulation of dynamical landscapes and the feedback clock. Our experiments show that, by decoupling rotational fluctuations from the thermal bath, we now have full control on each of these timescales, which enables us to begin exploring new directions, where numerical simulations play an essential role in providing guidelines and large statistics for the validation of the results.Looking toward the future, our findings open up new interesting avenues to direct the dynamics and organization of ABPs. Local control over the rotational dynamics offers an alternative means to control the persistence of active trajectories at a given velocity, which can be harnessed to optimize the navigation of ABPs in complex environments40,41. The introduction of periodic modulations in the rotational dynamics of ABPs also defines a new framework to study a variety of anomalous diffusion phenomena28–31, beyond the cases presented here, with the emergence of interesting analogies with glassy dynamics, as discussed above. Since ABPs enable directed transport and pattern formation, even in the absence of particle interactions and external forces38,42,43, introducing various forms of feedback, communication2, delay44, information flows11 and sensory ability12 defines new opportunities. Borrowing ideas and tools from signal processing and control systems, we envisage the engineering of more complex dynamical responses. We could, for instance, adapt ideas developed for nuclear detectors (dead-time analysis) or robotic systems (feed-forward responses or negative delays39,45) to devise new signal reconstruction strategies between discrete sampling events for ABPs, or design higher-order or integral responses to mimic the way in which biological microswimmers sense and adapt to chemical signals46. Finally, a last challenge, which also constitutes an exciting opportunity, is to translate the capabilities provided by external feedback and control into internal responses, e.g., through adaptation and reconfiguration46, in order to develop fully autonomous artificial microswimmers.MethodsFabrication of Janus particlesIn order to fabricate our active Janus particles, silica colloids with 4.28 μm diameter (5% w/v, microParticles GmbH, Germany) are diluted to 1:6 in MilliQ water and spread on a glass slide, previously made hydrophilic by a 2-minute air plasma treatment. Upon drying of the suspension, a close-packed particle monolayer is formed. The monolayer is then sputter-coated with 120 nm of nickel (Safematic CCU-010, Switzerland) to create the Janus surface. After coating, the glass slide is placed overnight above a neodymium magnet (50 × 50 × 12.5 mm3, 1.2 T) to align the magnetic moments of all particles in the direction of the compositional asymmetry. The particles are retrieved from the glass slide by pipetting and withdrawing a droplet of water on top of the monolayer. An identical procedure is followed for 2 μm silica particles, for which data are shown in the SI.Cell preparationThe transparent electrodes are fabricated from 24 mm × 24 mm No. 0 coverglasses (85–115 μm-thick, Menzel Gläser, Germany) covered with 3 nm of chromium and 10 nm of Au deposited by metal evaporation (Evatec BAK501 LL, Switzerland), followed by 10 nm of SiO2, deposited by plasma enhanced chemical vapor deposition (STS Multiplex CVD, UK) to minimize the adhesion of particles to the substrate. A water droplet containing the particles is deposited on the bottom electrode within the 9 mm-circular opening of a 0.12 mm-thick sealing spacer (Grace Bio-Labs SecureSeal, USA). The top and bottom electrodes are connected to a signal generator (National Instruments Agilent 3352X, USA) that applies the AC electric field, with a fixed frequency of 1 kHz and varying the Vpp voltage between 1 and 10 V. For 5 V, the applied electric field is 42 V mm−1.ExperimentsThe magnetic moment of the Janus particles is confined to the electrode plane and freely rotated within it through a custom-built setup fitted with two pairs of independent Helmholtz coils47. The magnetic field is constant within a few percent over a 1 mm2 area in the center of the cell and the maximum applicable magnetic field is 65 mT. In the experiments, we use a field of 30–40 mT to orient the particles. In order to impose an effective rotational diffusivity to the particles, the magnetic field angle at step n + 1 (θn+1) is obtained by adding to θn a random angular displacement Δθ, which is given by Eq. (2), where DR is the target rotational diffusivity, Δt is the time step (1 ms in our experiments), and η is a random number sampled from a normal Gaussian distribution.2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\theta }_{t+1}={\theta }_{t}+\sqrt{2{D}_{{\rm{R}}}\Delta t}\eta$$\end{document}θt+1=θt+2DRΔtηThe Janus particles are imaged with a home-built bright-field microscope in transmission and image sequences are taken with a sCMOS camera (Andor Zyla) at 10 fps with a 512 × 512 pixels2 field of view. The image series to measure the thermal and imposed rotational diffusivity are acquired using a 50× objective (Thorlabs). The positions of the center of the JPs and of the metal cap are located using Labview routines. Then, a vector connecting both centers is used to determine the orientation of the particle at each frame for different DR, from which the angular displacement distribution in Fig. 1c was calculated. The image series of ABPs actuated by the magnetic and the AC electric fields, are acquired with a 10× objective (Thorlabs). In this case, only the particle center of mass is located, and all the dynamical information is extracted from the final particle trajectory.For the experiments with space-dependent DR, single particles are located in real time by a custom LabView software. Series of 1024 × 1024 pixels2 images are recorded at 5.88 fps. The coordinates are used to update the particle DR based on a predefined landscape. For the data presented in the main text, the field of view is divided into checkerboard patterns with 5 × 5, 10 × 10 and 20 × 20 squares, respectively, with alternating regions of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}$$\end{document}DRH = 10 rad2 s−1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}$$\end{document}DRL = 0.01 rad2 s−1. DR is updated every τ, using values ranging from the smallest possible delay of 170 ms to τ = 17 s, based on the particle coordinates at t − τ. We vary the ballistic timescale L/v by varying L between 16 and 64 μm um and v in the range 3–12 μm s−1.The particle thermal translational (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{T}}}^{{\rm{th}}}$$\end{document}DTth) and rotational (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\rm{R}}}^{{\rm{th}}}$$\end{document}DRth) diffusion coefficients at room temperature (24 °C) are extracted from their 2D trajectories in the absence of magnetic and electric fields.Numerical simulationsWe simulate the dynamics of the ABPs by solving the equations of motion:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m\ddot{x}= \; {f}_{x}(\theta )-{\gamma }_{{\rm{T}}} \dot{x}+\sqrt{2{k}_{{\rm{B}}}T{\gamma }_{{\rm{T}}} }{\eta }_{x}(t)\\ m\ddot{y} = \; {f}_{y}(\theta )-{\gamma }_{{\rm{T}}} \dot{y}+\sqrt{2{k}_{{\rm{B}}}T{\gamma }_{{\rm{T}}} }{\eta }_{y}(t)\\ I\ddot{\theta }= \; {\gamma }_{{\rm{R}}}(x,y,\tau )\dot{\theta }+\sqrt{2{k}_{{\rm{B}}}T{\gamma }_{{\rm{R}}}(x,y,\tau )}{\eta }_{\theta }(t),$$\end{document}mx¨=fx(θ)−γTx˙+2kBTγTηx(t)my¨=fy(θ)−γTy˙+2kBTγTηy(t)Iθ¨=γR(x,y,τ)θ˙+2kBTγR(x,y,τ)ηθ(t),where m and I are the mass and the moment of inertia of the colloid, respectively, fx(θ) and fy(θ) are the x and y components of the active force acting on the colloid, γT is the friction coefficient associated with translational motion, γR(r, τ) is the state-dependent friction coefficient associated with rotational motion, and ηx(t), ηy(t), and ηθ(t) are uncorrelated random numbers satisfying:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle {\eta }_{x}\rangle =\langle {\eta }_{y}\rangle =\langle {\eta }_{\theta }\rangle =0;\,\langle {\eta }_{x}^{2}\rangle =\langle {\eta }_{y}^{2}\rangle =\langle {\eta }_{\theta }^{2}\rangle =1.$$\end{document}⟨ηx⟩=⟨ηy⟩=⟨ηθ⟩=0;⟨ηx2⟩=⟨ηy2⟩=⟨ηθ2⟩=1.The active forces fx(θ) and fy(θ) are set equal to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{{\rm{T}}} v\cos (\theta )$$\end{document}γTvcos(θ) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{{\rm{T}}} v\sin (\theta )$$\end{document}γTvsin(θ), respectively, such that in the absence of thermal noise and in the limit of long times the particles move at a constant velocity equal to v. We solved Eq. (3) in the underdamped limit through a Verlet-type integration scheme proposed by Gronbech-Jensen and Farago using the Itô convention48. Although Eq. (3) could also be solved in the overdamped limit, this approach allowed us to achieve a faster convergence to the homogeneous distribution for τ = 0 using a relatively small integration step dt = 0.001 s.For a position-dependent DR, we let the rotational friction γR(r) = kBT/DR(r) vary as a function of the ABP’s position r = [x(t), y(t)] according to a checkerboard pattern as follows:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{{\rm{R}}}({\bf{r}})=\frac{{\gamma }_{{\rm{R}}}^{{\rm{H}}}-{\gamma }_{{\rm{R}}}^{{\rm{L}}}}{2}\left\{1+{\rm{sgn}}\left[\sin \left(\frac{\pi x}{L}\right)\sin \left(\frac{\pi y}{L}\right)\right]\right\}+{\gamma }_{{\rm{R}}}^{{\rm{L}}},$$\end{document}γR(r)=γRH−γRL21+sgnsinπxLsinπyL+γRL,where:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{sgn}}(x)=\left\{\begin{array}{l}1,\,x\ge 0\\ -1,\,x\;<\;0\end{array}\right.,$$\end{document}sgn(x)=1,x≥0−1,x<0,and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{{\rm{R}}}^{{\rm{H}}}$$\end{document}γRH and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{{\rm{R}}}^{{\rm{L}}}$$\end{document}γRL correspond to the regions of high and low DR, respectively.For the implementation of the discrete time-feedback loop using the ZOH model, we update γR every t = nτ, where n is the number of samples. Since the rotational diffusivity is a physical quantity that must be continuous in time, we reconstruct it from discrete-time inputs using the following function:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{{\rm{R}}}({\bf{r}},\tau )=\mathop{\sum }\limits_{j = -\infty }^{+\infty }{\gamma }_{{\rm{R}}}[j]\Pi \left({\rm{t}}-{\rm{n}}\tau \right),$$\end{document}γR(r,τ)=∑j=−∞+∞γR[j]Πt−nτ,where γR[j] is γR(r) evaluated at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bf{r}}\left({\rm{t}}=j\tau \right)$$\end{document}rt=jτ:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{{\rm{R}}}[j]= \int_{-\infty }^{+\infty }{\gamma }_{{\rm{R}}}\left({\bf{r}}\right)\delta \left({\rm{t}}-j\tau \right)d{\rm{t}},$$\end{document}γR[j]=∫−∞+∞γRrδt−jτdt,Π is a rectangular function defined as:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Pi =\left\{\begin{array}{l}1,\,0\le {\rm{t}}\;<\;\tau \\ 0,\,\,\text{otherwise}\,.\end{array}\right.$$\end{document}Π=1,0≤t<τ0, otherwise.and j is an integer number which is set equal to n − 1 in the simulations for a delay equal to τ, and to n for the those without delay.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Movie 1Supplementary Movie 2Supplementary Movie 3Supplementary Movie 4Supplementary Movie 5Supplementary Movie 6Supplementary Movie 7 | nature communications | [
"Article"
] | [
"Colloids",
"Statistical physics, thermodynamics and nonlinear dynamics"
] | behavior self-propelling colloidal particles physics opportunities tasks beyond micro nanoscale functions inspired by similarity synthetic active matter with motile bacteria analogy understand motion-)organization living systems through synthetic description active colloids active Brownian motion1: microscopic particle radius R in fluid viscosity η moves constant speed v diffuses rotational diffusivity DR = (kBT)/(8πηR3) kBT characteristic thermal energy kB Boltzmann constant model employed phenomena motion active particles in complex optimization of search strategies4 growing interest in control synthetic active matter beyond model complex behaviors directed transport pattern formation phenomena arise when particle velocity environmental fluctuations vary in space time position-dependent translational diffusivity proposed biological mechanism anomalous diffusion localization biomolecules in cellular membranes5 temporal control of random walks collective motion in active colloids6 position-dependent velocity investigated control organization area explored by active particles mechanisms exploited for applications environmental remediation to targeted drug delivery1 translation rotation in ABPs coupled feedback between rotational dynamics position control active Brownian motionBiological swimmers chemotactic tune rotational dynamics chemical gradients localize food sources escape harmful chemicals reorientation frequency rotational dynamics of synthetic active particle dictated by thermal fluctuations control rotational diffusivity of ABPs decoupling amplitude from thermal bath randomly-oriented magnetic fields discrete-time feedback loop modulate effective rotational temperature study effect position-dependent rotational diffusivity on statistics active Brownian motion feedback between DR particle’s position not instantaneous mediated by discrete sampling position finite sensorial delay periodic space-time modulations rotational dynamics bring about exotic phenomena anomalous diffusion to directed transport localization support results with numerical simulations indicate new directions for future developments rotational model ABPs self-propel due to induced-charge electro-phoresis20–22 consist of 4 μm-diameter silica Janus colloids half-coated with 120 nm-thick nickel cap magnetized perpendicular to Janus propulsion’s direction compositional asymmetry aligned with caps’ magnetic moment controlled by magnetic fieldparticles sediment bottom liquid cell two electrodes separated vertical gap h = 120 μm record colloids’ position orientation frame rate 10 fps video microscopy Methods colloids swim over bottom substrate unbalanced electrohydrodynamic flows uniform 1 kHz AC electric field across electrodes (Fig. 1a). swimming velocity v proportional to electric field\documentclass[12pt{minimal{amsmath-69pt}{document}}/h(Vpp/h)2 Vpp peak-to-peak voltage varied 1–10 V.Fig. 1Controlling rotational dynamics randomly-oriented magnetic fields Schematic experimental setup Janus silica particle half-coated magnetized Ni cap self-propulsion bottom liquid cell two transparent electrodesPropulsion induced-charge electro-phoresis AC electric field[12pt]{minimal}{amsmath{wasysym-69pt} perpendicular to plane particle moves curved yellow arrows depict unbalanced electrohydrodynamic flows). uniform magnetic field[12pt]{minimal}{amsmath{wasysym{upgreek-69pt} (blue arrow), four controls in-plane orientation particle motion observed with optical microscope optical micrograph 4 μm particle Ni cap in black white arrow shows orientation angle θ aligned with cap’s magnetic moment magnetic field Scale bar: 5 μm Imposed θ function of time for three values DR colored bands delimit 2π range Data shifted y-axis for clarity Probability distributions of angular displacements G(Δθ, Δt) for different DR lag time Δt = 0.1 s lines Δθ same colors b.gray symbols show measured G(Δθ, Δt) without magnetic field thermal[12pt]{minimal{amsmath-69pt (gray line: Gaussian fit Supplementary Table 1) Trajectory of ABP with v = 5.5 μm s−1 imposed DR varying over time MSD of ABP with v = 5.5 μm s−1 for different imposed DR: 0.1 (diamonds), 1 (triangles), 5 (circles) 10 (squares) s−1 Persistence length (LP) function of rotational relaxation time[12pt{minimal}-69pt}DR−1 different values of v: 8.2 (circles), 6.7 2.7 (squares) μm s−1inset shows LP function v thermal[12pt{minimal\usepackage{amsmath\oddsidemargin-69pt}DRth 0.014 rad2 s−1 (gray imposed DR 0.07 0.144 rad2 s−1 (black triangles circles Error bars data standard deviation control angle θ colloids’ cap image Fig. 1a independent Helmholtz coils generating uniform magnetic fields-plane orientation (Fig. 1a Supplementary 1 1) contrast randomize direction magnetic field colloids controlled rotational diffusivity decoupled thermal bath propulsion scheme 2) vary orientation magnetic field f = 1 kHz random angular displacements Δθ Gaussian distribution zero mean variance σ2 = 2DR/f DR imposed rotational diffusivity Fig. 1b shows orientation angle magnetic field function time imposed DR 10−2 to 10 rad2 s−1 distributions particles’ angular displacements G(Δθ cap orientation attest θ direction magnetic field diffuse same Gaussian process (Fig.letting θ diffuse over 2π range enforce values DR above below thermal rotational diffusivity\documentclass[12pt{minimal}\usepackage{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}}DRth (1.4 × 10−2 rad2 s−1 at room temperature). achieve rotational dynamics faster than thermal bath (rotational cooling shown SI 2 μm-diameter colloids Supplementary Figs. 2 3) control on DR v adjust particle trajectories real-time 1d increase propensity ABP move straight paths decreasing DR time from 10 to 10−1 rad2 s−1 keeping v constant (Supplementary Movie 3 Fig. 4 Movie 4 DR constant v varied step-wise See Supplementary Fig. 2 Fig 3 for 2 μm colloids). analysis mean displacements each segment at fixed DR (Fig.1e), shows timescale ABP’s motion ballistic to diffusive proportional to\documentclass[12pt]{minimal}{amsmath{wasysym-69pt{document$ {D}{R{-1}{document~DR−1 MSD at long times larger for lower values DR controlling area explored ABP by varying DR on demand varying DR fixing AC voltage extract DR and v from MSD to calculate persistence length[12pt]{minimal}{amsmath-69pt}{document}{L}{P=v{D}{R{-1}{document}LP=vDR−1. Fig. 1f show LP be varied over wide range of values displaying expected linear scaling on both control parameters data attest to ability to engineer motion of ABPs by independently controlling v and DR using electric and magnetic fieldsfeedback loop position-dependent rotational study impact space-time modulations active Brownian motion implemented discrete-time feedback loop updates DR ABP’s position r(t) (Fig. 2a Methods Supplementary Figs. 5 6) response motile microorganisms environmental update DR regular intervals ABP’s position r(t − t = nτ τ sampling period n number samples realized experiments holding DR constant between sampling periods Langevin dynamics simulations rotational friction vary zero-order hold (ZOH) Methods Fig. analogy Brownian motion periodic potentials anomalous DR vary checkerboard pattern alternating regions size L.square DR takes high\documentclass[12pt]{minimal}{amsmath}{wasysym}{upgreek\oddsidemargin{-69pt}\begin{document}${D}_{{\rm{R}}}\rm{H}}}\end{document}DRH low[12pt]{minimal}{amsmath{wasysym{upgreek\setlength{\oddsidemargin}{-69pt}\begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}\end{document}DRL value (Fig.consider scenarios[12pt]{minimal}{amsmath{wasysym\oddsidemargin}{-69pt}{document}${D}_{{\rm{R}}}^{{\rm{H}}}=10{document}DRH=10 rad2 s−1[12pt{minimal{amsmath{wasysym}{-69pt}{document}{D}_{{\rm{R}}}\rm{L}}}=0.01{document}DRL=0.01 rad2 s−1[12pt]{minimal}{amsmath}{wasysym}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\begin{document}\frac{1}{{D}_{{\rm{R}}}^{{\rm{L}}}}> \frac{L}{v}>\frac{1}{{D}_{{\rm{R}}}^{{\rm{H}}}}\end{document}1DRL>Lv>>1DRH (Fig.d Supplementary Movie 5 Fig. 7) choices imply motion diffusive one ballistic other two relaxation timescales. 2Position-dependent rotational dynamics sampling checkerboard pattern sensorial delay Block diagram feedback loop experiments simulations space-time modulations rotational dynamics feedback function couples DR position r(t) four blocks sampling delay zero-order hold DR f Schematic r(t) sampled time points t nτ reconstructed ZOH model(t delay τ position r[n-1] input Schematic ABP moving checkerboard pattern rotational diffusivity comparison experimental simulated trajectories ABPs DR varying checkerboard pattern(v = 3.5 μm s−1 L = 32 μm\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin{-69pt}}{D}\rm{R}}}{H}}}}DRH = 10 rad2 s−1[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}{\oddsidemargin}{-69pt}{D}_{{\rm{R}}}{L}}}=0.01{document}DRL=0.01 rad2 s−1 τ = 0.4 multiple parameters dynamics spatial organization ABPs examine contributions separately role characteristic timescale L/v effect sampling period τ analysis on particle localization.Non-Gaussianity: L/v exponential feedback scheme modulations of DR different anomalous diffusion time length scales sampling period τ timescale L/v statistics particle motion for nonzero small sampling periods τ << L/vregime trajectories ballistic or diffusive (Fig. 2c–d), dynamics unique features distribution one-dimensional displacements rescaled by L G(∣x∣/L Δt) (Fig. 3a–d non-Gaussianity lag times Δt ratio Δt(L/v)−1 departure from Gaussian behavior excess kurtosis (Fig. 3e):=⟨x4⟩⟨x2⟩2−3 γ = 0 normal distribution Δt << L/v G(∣x∣/L, Δt) broader flatter than normal distribution (γ < consistent broadening G(x, Δt) ABPs intermediate timescales ballistic Δt approaches L/v G/L develops leptokurtic distribution (γ > 0) Gaussian peak at ∣x∣ exponential tail up to ∣x∣ L decays.Fig. 3Non-Gaussian statistics exponential tails due DR checkerboard pattern.Experimental simulated data plotted as circles red linesProbability distributions rescaled-dimensional displacements G Δt(L/v)−1 0.02 0.2 1 5 displacements x y increase statistics black lines third panel (Δt(L/v)−1 = 1) Gaussian exponential fit G 0−0.2 0.4−1 Excess kurtosis γ G(x/L function Δt(L/v)−1 Rescaled displacement MSD/L2 function Δt(L/v)−1yellow blue lines theoretical MSDs ABPs constant DR equal\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{amssymb{mathrsfs{upgreek\oddsidemargin-69pt}{document}${D}\rm{R}}}\rm{L}}}\end{document}DRL[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin}-69pt}${D}\rm{R}}}\rm{H}}}\end{document}DRH (v = 4 μm s−1 L = 32 μm[12pt]{minimal}{amsmath}{wasysym{amsfonts{amsbsy{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}${D}_{{\rm{R}}}\rm{H}}}\end{document}DRH = 10 rad2 s−1[12pt]{minimal}{amsmath{wasysymage{amsbsy}\usepackage{mathrsfs}{upgreek}\setlength-69pt}{document}{D}_{{\rm{R}}}\rm{L}}}=0.01{document}DRL=0.01 rad2 s−1 τ = 0.2 s).Exponential tails in G(x, Δt) observed in glassy systems diffusion colloids macromolecular explained dynamic heterogeneity coexistence faster slower particles explains appearance system timescales Δt L/v two distinct populations particles different length scales checkerboard diffusive[12pt]{minimal}{amsmath}{wasysym}}{upgreek-69pt}}${D}_{{\rm{R}}}^{{\rm{H}}}\end{document}DRH-regions ballistic[12pt]{minimal}{amsmath}{wasysym}}}{upgreek}-69pt}}{D}_{{\rm{R}}}^{{\rm{L}}}{document}DRL-regions.\documentclass[12pt]{minimal}{amsmath\oddsidemargin}{-69pt}{document}$$\Delta t\simeq L/v\;> >{D}_{{\rm{R}}}^{{\rm{H}}}\end{document}ΔtL/v>>DRH[12pt]{minimal}{amsmath{wasysym\oddsidemargin}{-69pt}{document}$${D}_{{\rm{R}}}^{{\rm{H}}}\end{document}DRH-regions diffusive[12pt]{minimal}{amsmath{wasysym}{upgreek}\oddsidemargin}{-69pt}{document}$${D}_{{\rm{e}}ff}\simeq {v}^{2}/{D}_{{\rm{R}}}^{{\rm{H}}}\end{document}Deffv2/DRH1 displacements normally distributed responsible for Gaussian peak travel distances smaller than L.ABPs in\documentclass[12pt{minimal}{amsmath{wasysym{upgreek-69pt}{R{L{document}DRL-regions move ballistic[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}\Delta t\simeq L/v<<;1/{D}_{{\rm{R{L\end{document}ΔtL/v<<1/DRL displacements scale as vΔt order L dynamic heterogeneity in glassy colloidal systems arises from translational diffusion to steric interactions result of spatially heterogeneous rotational dynamics parallel with glassy systems further dynamics at longer Δt excess kurtosis γ positive maximum at Δt L/v (Fig. 3e and Fig. 4a–b decays to zero Gaussian statistics recovered at long timesMSD super-diffusive scaling ~Δt1.7 critical diffusive regime~Δt1) recovered long scales Fig. contrast dynamics ABPs constant DR γ negative intermediate zero short long transition from ballistic to diffusive scaling at Δt 1/DR32.Fig. 4Role timescale L/v sampling period τ on statistics particle motion Simulated evolution excess kurtosis γ one-dimensional displacements values L/v constant τ = 0.4 s Experimental γ maximum function L/v (τ = 0.2–0.4 Error bars 95% confidence intervals Simulated evolution γ with Δt(L/v)−1 τ < L/v τ > L/v L = 32 μm v = 3.5 μm s−1 black circles τ = 0. Simulated evolution rescaled mean displacement MSD/L2 with Δt(L/v)−1 different values L/v τ timescale L/v dictates maximum non-Gaussianity extent (Fig. 4a consequence of motion ballistic diffusive segmentsexcess kurtosis γ grows with Δt because exponential tail grows faster with Δt than Gaussian peak (\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts{upgreek-69pt{document ABPs travel up to ~L tail of G(∣x∣/L, Δt), and γ keeps growing up to Δt ~ L/v. greater L/v greater maximum γ attained at larger Δt. in limiting case of small \documentclass[12pt]{minimal}{amsmath{upgreek-69pt exponential tail cannot develop γ remains negative approaching zero in limit of long times as in ABPs with constant DR (Fig. 4a).Non-Gaussianity: τ and subdiffusionThe sampling period τ on statistics of motion depending on value relative to timescale L/v (Fig.4c–h). identify three regimes: τ = 0 instantaneous update DR ABP’s position τ < L/v τ > L/v τ = 0 (Fig. 4c Supplementary Fig. 8) γ attains local maxima at Δt(L/v)−1 = 1 3 5 global maximum Δt(L/v)−1 = 3. maxima γ mark changes scaling MSD (Fig. superdiffusive scaling ~Δt1.7 up to Δt L/v subdiffusive scaling < 1) up to Δt 3–5L/v higher L/v smaller a Δt >> L/v MSD recovers diffusive scaling (a = 1) emergence subdiffusion reinforces analogy glassy dynamics no physical caging by neighboring particles37 Subdiffusion result “effective dynamical randomization direction motion ballistic ABPs}DRH-regionsup to Δt L/v MSD dominated by ballistic segments length\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{upgreek}{-69pt}\begin{document}$${D}_{{\rm{R}}}^{{\rm{L}}}\end{document}DRL-regions same timescale diffusive ABPs in\documentclass[12pt]{minimal}{amsmath}{-69pt}{document}$${D}_{{\rm{R}}}\rm{H}}}{document}DRH-regions travel negligible distances\documentclass[12pt]{minimal}{amsmath}{wasysym{upgreek{\oddsidemargin}{-69pt}\begin{document}$$\sim v\sqrt{\Delta t/{D}_{{\rm{R}}}^{{\rm{H}}}}\end{document}~vΔt/DRH.over timescales L/v ballistic ABPs travel distances greater than L without crossing into\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{upgreek}\oddsidemargin{-69pt}{document{D}\rm{R}}}{H\end{document}DRH-region finite probability they diffuse back cross\documentclass[12pt]{minimal}{amsmath}{wasysym}{upgreek}\oddsidemargin}{-69pt}{document}{D}\rm{R}}}{L}}}\end{document}DRL-region opposite direction (Supplementary Movies 6 and 7).reflection events give null displacements over timescales ~2L/v ballistic ABPs cross[12pt]{minimal}{amsmath}{wasysym{upgreek}{\oddsidemargin}{-69pt}{document}${D}\rm{R}}}{L}}}\end{document}DRL-region travel backMSD grows as[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin}{-69pt}{document}\simeq {v^{2}/{D}_{{\rm{R}}}\rm{H}}}\Delta\end{document}v2/DRHΔt particles cross[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin}{-69pt}{document}$${D}_{{\rm{R}}}^{{\rm{L}}}\end{document}DRL-region odd number times[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek{\oddsidemargin}{-69pt}\begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}\end{document}DRH-region dynamical caging subdiffusive motion less prominent decreasing L/v difference between diffusive ballistic displacements diminishes (Fig.to nonzero values τ brings changes γ depending τ < L or > L/v to τ ~ L/v higher values τ translate into higher values γ increased prominence maximum at Δt L/v disappearance of local minima (Fig. 4c). nonzero values τ introduce finite delay rotational dynamics ballistic ABPs penetrate into up to lengths ~v(2τ) before adjusting rotational dynamics Fig. 2a). increases effective distance ABPs cross regions without adjusting dynamics increase in maximum value γ with increasing values τ due to larger disparity between ballistic diffusive displacements.Increasing τ beyond L/v leads to decrease γ disappearance maximum in limit τ ≫ L/v (Fig. 4d). For τ > L/v modulations rotational dynamics depart from periodicity checkerboard pattern In limit large τ rotational dynamics ABPs determined by initial position regionimplies existence two populations particles moving ballistic or diffusive DR constant time equal to τ dynamics superposition of ABPs with different DR in limit of large τ, γ maximum remains negative close to zero (Fig. 4d).Values τ > 0 lead to disappearance subdiffusion at Δt L/v (Fig. 4f–h). MSD displays cross-over between superdiffusive and diffusive scaling short long timescales disappearance subdiffusion due to finite τ allows ballistic ABPs penetrate\documentclass[12pt]{minimal}\usepackage{amsmath{-69pt}\begin{document}}DRH-regions greater lengths before updating DRminimizes probability ABPs diffuse into\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{amssymb{amsbsy{mathrsfs{upgreek\oddsidemargin-69pt{L{document}DRL-region increases chances ballistic ABPs cross entire\documentclass[12pt{minimal}{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}{document{D}_{R}{H{document}DRH-regions without updating DR contributing to higher MSDs increasing τ causes superdiffusive-to-diffusive transition at Δt > L/v ballistic ABPs travel distances greater than L.position-dependent rotational dynamics pattern finite sensorial delay sampling (τ > 0) leads localization of ABPs[12pt{minimal{amsmath\oddsidemargin-69pt}{document}{R\end}DRH-regions (Fig. 5) degree localization nonmonotonic function of τ(L/v)−1 instantaneous updates DR (τ = time-averaged steady-state spatial distribution ρ(x, y) homogeneous no manifestations of underlying DR pattern (Fig.5a numerical simulations). τ increased ρ(x, y) increases\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek-69pt}{D{R{H}DRH-regions expense[12pt]{minimal}{amsmath{upgreek-69pt{D{R{L}DRL-regions maximum degree critical τ 0.1−0.3L/v returning homogeneous distribution limit large τ(L/v)−1 (Fig. 5a).Fig. 5Emergence localization finite sampling period τ Simulated particle density distribution ρ(x, y) different values τ(L/v)−1 rescaled uniform distribution ρeq = 1/A A area simulation box ρeq recovered for spatially homogeneous DR or τ = 0.simulation parameters v = 3.5 μm s−1 L = 32 μm[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin{-69pt}}\rm{R}}}{H}DRH= 10 rad2 s−1[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin}{-69pt}{R{L{document}DRL = 0.01 rad2 s−1 distributions obtained binning positions 2.5 × 104 particles simulated time 900 s particles move random positions 100 s Periodic boundary conditions enforced simulation box 10L × 10L.Simulated ratio average particle densities\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{upgreek\oddsidemargin}{-69pt}\begin{document}{D}\rm{R}}}\rm{H\end{document}DRH-[12pt]{minimal}{amsmath{wasysym}}}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document{D}_{{\rm{R}}}\rm{L}}}\end{document}DRL-regions ρH/ρL function τ different values L/v.ρH/ρL calculated 〈TH/TL〉 TH TL residence times\documentclass[12pt]{minimal\usepackage{amsmath}{wasysym{upgreek\oddsidemargin}{-69pt}\begin{document}$${D}_{{\rm{R}}}\rm{H}}}\end{document}DRH-[12pt]{minimal}{amsmath}{wasysym}}{upgreek}{\oddsidemargin}{-69pt}{document$${D}_{{\rm{R}}}\rm{L}}}\end{document}DRL-regionsSimulated maximum ρH/ρL function L/v linear model line). d–e Simulated distribution d TH e TL normalized diffusive timescale[12pt{minimal}{amsmath\oddsidemargin-69pt}{document\frac{{L}^{2{D}{R{H}}{2{v\end{document}L2DRH2v2 ballistic timescale L/v Experimental simulated ρH/ρL function τ(L/v)−1 different L/v rescaled maximum Lines circles d–f color-coded L/v colormap b Error bars 95% confidence intervalsquantify departure from homogeneous distribution studying evolution ratio simulated average ρ(x, y)\documentclass[12pt]{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}{D}_{{\rm{R}}}\rm{H}}}\end{document}DRH-[12pt]{minimal}{amsmath{mathrsfs{upgreek\oddsidemargin-69pt}{D}\rm{R}}}{L}}}{document}DRL-regions: ρH and ρL (Fig. 5b, c). density maps Fig. 5a ρH/ρL ratio 1 in limit small large τ maximum at intermediate values τ (Fig. maximum ρH/ρL ratio linearly increasing function of L/v (Fig. maximum degree localization dependence with L/v explained with transport argument dynamic asymmetry by finite values τballistic ABPs penetrate into[12pt]{minimal}{amsmath{wasysym\oddsidemargin}{-69pt}{document}{D}\rm{R}}}{H\end{document}DRH-regions up to lengths v(2τ) before updating DR diffusive ABPs diffusing up to lengths[12pt]{minimal}{amsmath{wasysym{upgreek{\oddsidemargin}{-69pt}\begin{document}\sim\sqrt{{D}\rm{R}}}{-1}\end{document}~vDR−1τ between updates deeper ballistic ABPs penetrate diffusive regions longer to diffuse outnonzero values τ imply particles more time in\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{upgreek\oddsidemargin{-69pt}\begin{document}{D}_{{\rm{R}}}{H\end{document}DRH-regions picture agreement with higher localization center\documentclass[12pt]{minimal}{amsmath}{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}\begin{document}{D}_{{\rm{R}}}{H\end{document}DRH-regions for τ 0.3L/v (Fig.quantitatively steady state net flux between regions zero write:1\documentclass[12pt]{minimal}{amsmath}{wasysym}{mathrsfs{upgreek}\oddsidemargin{-69pt}\begin{document}\frac{{\rho\rm{H}}}}{{T=\end{document}ρHTH=ρLTL TH TL average residence times ABPs[12pt]{minimal}{amsmath}{wasysym}{mathrsfs{upgreek}\oddsidemargin}{-69pt}\begin{document}$${D}_{{\rm{R}}}^{{\rm{H}}}\end{document}DRH-[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts}{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}{D}_{{\rm{R}}}\rm{L}}}\end{document}DRL-regions.\documentclass[12pt]{minimal{amsmath{wasysym{upgreek\oddsidemargin}{-69pt}{document}{\rm{m}}ax\left({\rho\rm{H}}}\rm{L}}}\right\end{document}maxρH/ρL equal to[12pt]{minimal}{amsmath}{wasysym}}{upgreek}{\oddsidemargin}{-69pt}{document}$$\max \left({T}^{{\rm{H}}}/{T}^{{\rm{L}}}\right\end{document}maxTH/TL.0 < τ << L/v expect TH TL scale as ~(L/v)2 ~L/v residence time\documentclass[12pt]{minimal}{amsmath{wasysym-69pt}}{D}{R}DRH-region depends on time ABP penetrate~L/v diffuse\documentclass[12pt]{minimal}{amsmath{upgreek\oddsidemargin}{-69pt}{document}$$\sim\frac{{L}^{2}}{2{v}{2}/{D}_{{\rm{R}}}{H}}}}\end{document}~L22v2/DRH). maximum ratio ρH/ρL = TH/TL scales linearly with L/v Fig. 5c different scaling of TH TL confirmed by experimental simulated residence time distributions at different L/v (Fig.5d–e). distributions TH and TL collapse same master curves rescaled by\documentclass[12pt{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt}{document}}L22v2/DRH L/v drop to zero for rescaled values TH and TL greater than 1. renormalizing ρH/ρL ratio by maximum for L/v plotting as function of τ(L/v)−1 simulated and experimental data collapse single master curve (Fig. limit of τ > L/v degree localization large sampling periods rotational dynamics decoupled from periodicity checkerboard pattern cross-over Nyquist–Shannon’s theorem19. τ ~ L/v frequency ABPs sample environment comparable to highest frequency region checkerboard pattern for τ > L/v ABPs sense changes DR on length scale L retain DR for period equal to sampling period two populations moving ballistic or diffusive initial positions updated every τ.τ > L/v disconnect sampling resolution spatial periodicity pattern localization over time (Fig. 6) oscillations damped τ >> L/v correlations particle motion delay removing delay retaining discrete sampling DR updated every t current position r persistent oscillations period equal to τ (Fig. 6a b).Fig. 6Oscillating localization τ > L/v with without delay feedback loop Simulated time evolution of ρH/ρL calculated from density distribution ρ(x, y) stepDR updated every t = nτ past particle’s position r(t − τ). Same as a DR updated every t = nτ current position r(t = nτ). simulation parameters v = 3.5 μm s−1 L = 32 μm\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek-69pt}DRH = 10 rad2 s−1[12pt{minimal{amsmath{upgreek-69pt}DRL = 0.01 rad2 s−1 findings illustrate engineering feedback between internal dynamics DR, v) state (r(t)) of ABPs allows tailoring response statistics of motion single-particle level global spatio-temporal organization response defined by balance of timescales system active Brownian motion set by v and DR externally imposed set by modulation of dynamical landscapes feedback clockexperiments show decoupling rotational fluctuations from thermal bath full control on timescales exploring new directions numerical simulations essential guidelines statistics for validation results findings open new avenues to direct dynamics organization of ABPs Local control over rotational dynamics offers alternative control persistence active trajectories at velocity optimize navigation ABPs in complex environments40 introduction of periodic modulations in rotational dynamics defines new framework to study anomalous diffusion phenomena28–31 analogies with glassy dynamics ABPs enable directed transport pattern formation particle interactions external introducing feedback communication2 delay44 information flows11 sensory ability12 defines new opportunities Borrowing ideas from signal processing control systems envisage engineering of complex dynamical responses adapt ideas for nuclear detectors robotic systems devise new signal reconstruction strategies for ABPs design-order responses mimic biological microswimmers chemical signals46 challenge translate capabilities external feedback control into internal responses develop autonomous artificial microswimmersJanus silica colloids 4.28 μm diameter (5% w/v microParticles GmbH diluted 1:6 MilliQ water spread glass slide hydrophilic 2-minute plasma treatment close-packed particle monolayer formed sputter-coated 120 nm nickel (Safematic CCU-010 Janus surface glass slide placed overnight above neodymium magnet (50 × 50 × 12.5 mm3 1.2 magnetic moments asymmetry particles retrieved pipetting water identical procedure 2 μm silica particles data.Cell transparent electrodes fabricated 24 mm × 24 mm No. 0 coverglasses (85–115 μm-thick Menzel Gläser Germany covered 3 nm chromium 10 nm Au metal evaporation (Evatec BAK501 10 nm SiO2 vapor deposition water droplet deposited bottom electrode 9 mm-circular opening 0.12 mm-thick sealing spacer (Grace Bio-Labs SecureSeal electrodes connected signal generator (National Instruments Agilent 3352X AC electric field fixed frequency 1 kHz voltage 1 10 V 5 V field 42 V mm−1.magnetic moment Janus particles confined to electrode plane rotated through custom-built setup two Helmholtz coils47 magnetic field constant over 1 mm2 area center cell maximum field 65 mT field 30–40 mT orient particles effective rotational diffusivity magnetic field angle at step n + 1 (θn+1) obtained by adding θn random angular displacement Δθ given by Eq. (2) DR target rotational diffusivity Δt time step (1 ms η random number normal Gaussian distribution\documentclass[12pt{minimal\usepackage{amsmath-69pt\theta}θt+1=θt+2DRΔtηThe Janus particles imaged with home-built bright-field microscope image sequences taken sCMOS camera) at 10 fps 512 × 512 pixels2 field of view image series thermal rotational diffusivity acquired using 50× objective (Thorlabs). positions center JPs metal cap located using Labview routinesvector connecting centers used orientation particle at each frame for different DR angular displacement distribution in Fig. 1c calculated image series of ABPs actuated by magnetic and AC electric fields acquired with 10× objective (Thorlabs). only particle center of mass located dynamical information extracted from final particle trajectory.For experiments with space-dependent DR single particles located in real time by custom LabView software. Series of 1024 × 1024 pixels2 images recorded at 5.88 fps coordinates update particle DR based on predefined landscape data main text field of view divided into checkerboard patterns with 5 × 5, 10 × 10 20 × 20 squares alternating regions of[12pt{minimal}{amsmath}DRH = 10 rad2 s−1}DRL = 0.01 rad2 s−1.DR updated every τ using values delay 170 ms to τ = 17 s based on particle coordinates at t − τ vary ballistic timescale L/v by varying L between 16 64 μm um v 3–12 μm s−1 particle thermal translational (\documentclass[12pt]{minimal{amsmath-69pt}{document{D}_{{\rm{T}}}{th\end{document}DTth) rotational ([12pt]{minimal-69pt}{R}}}{th\end{document}DRth diffusion coefficients at room temperature (24 °C) extracted from 2D trajectories absence of magnetic electric fieldsNumerical simulate dynamics ABPs solving equations motion[12pt{minimal{amsmath{wasysym\oddsidemargin-69pt}{document}\ddot{x} {f}{x}(\theta )-\gamma\rm{T}}}{x}+\sqrt{2{k}\rm{B{x}(t{y} = {f}_{y}(\theta )-\gamma{T}}}{y}+\sqrt{2{k}\rm{B\eta{y}(t\theta }\gamma\rm{R}}}(x,y,\tau )\theta }+\sqrt{2{k}\rm{B}}}T{R(x,y{\eta\theta }{document}mx ̈=fx(θ)−γTx ̇+2kBTγTηx(t=fy(θ)−γTy ̇+2kBTγTηy(t=γR(x,y,τ)θ ̇+2kBTγR(x,y,τ m I mass momentcolloid fx(θ) fy(θ) x y components active force colloid γT friction translational motion γR(r τ) state-dependent friction rotational motion ηx(t), ηy(t), ηθ(t) uncorrelated random numbers\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{amsfonts{mathrsfs}{upgreek}\oddsidemargin{-69pt}\begin{document}\langle {\eta_{x}\rangle ={y\theta =0{y{2\theta =1\end{document}⟨ηx⟩=⟨ηy⟩=⟨ηθ⟩=0;⟨ηx2⟩=⟨ηy2⟩=⟨ηθ2⟩=1active forces fx(θ) fy(θ) equal to\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document}\gamma{T v\cos (\theta{document}γTvcos(θ)[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin-69pt}{document\gamma{T v\sin (\theta{document}γTvsin(θ), absence thermal noise long times particles move constant velocity equal to v solved Eq. (3) underdamped limit Verlet-type integration scheme Gronbech-Jensen Farago Itô convention48 Eq. (3) solved overdamped limit approach faster convergence to homogeneous distribution for τ = 0 small integration step dt = 0.001 s.position-dependent DR rotational friction γR(r) = kBT/DR(r) vary ABP’s position r = [x(t), y(t)] checkerboard pattern\documentclass[12pt]{minimal}\usepackage{amsmath}}\oddsidemargin-69pt}{document}\gamma\rm{R}}}\bf{r}})=\frac{{\gamma\rm{R}}}{H}}}\left\{1+\rm{sgn}}\left x} y\gamma\rm{R}}}\rm{L}}}\end{document}γR(r)=γRH−γRL21+sgnsinπxLsinπyL+γRL\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}}{mathrsfs}{upgreek}\setlength\oddsidemargin-69pt}\begin{document}\rm{sgn}}(x)=\left\\begin{array}{l}1,\,x0\\ -1,,x\;<;0\end{array}{document}sgn(x)=1,x≥0−1,x<0[12pt]{minimal}{amsmath}{wasysym}}{\oddsidemargin}{-69pt}{document}\gamma\rm{R}}}^{{\rm{H}}}\end{document}γRH[12pt]{minimal}{amsmath}{wasysym}}}}{upgreek}{\oddsidemargin}{-69pt}{document}\gamma\rm{R}}}^{{\rm{L}}}\end{document}γRL regions high and low DR discrete time-feedback loop ZOH model update γR every t = nτ n number samples.rotational diffusivity physical quantity continuous time reconstruct from inputs function[12pt]{minimal}{amsmath{wasysym}\oddsidemargin}-69pt}{document}\gamma\rm{R}}}\bf{r}}=\mathop{\sum }\limits{j = -\infty+\infty\gamma\rm{R}}}[j]\Pi\left({\rm{t}}-{\rm{n}}\tau \right),\end{document}γR(r,τ)=∑j=−∞+∞γR[j]Πt−nτ γR[j] is γR(r) evaluated[12pt]{minimal}{amsmath}{wasysym}}}{upgreek}\setlength{\oddsidemargin}{-69pt}{document}\bf{r}}\left(\rm{t}}=j\tau \right\end{document}rt=jτ:8\documentclass[12pt]{minimal}{amsmath}{wasysym}setlength-69pt}}\gamma\rm{R}}}[j]=-\infty+\infty{R\left\bf{r}}\right)\delta\left\rm{t}}-j\tau\right\rm{t}}\end{document}γR[j]=∫−∞+∞γRrδt−jτdt,Π rectangular function\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym}{upgreek}\setlength\oddsidemargin-69pt}$\Pi =\left{array}1,\rm{t}}\tau 0,\text{otherwise\end{array}\right\end{document}Π=1,0≤t<τ0 j integer number set equal to n − 1 simulations delay equal to τ n without delay.Supplementary Review FileDescription Additional Supplementary FilesSupplementary Movie 7 | 47.8 | 1.820267 |
10.1038/s41467-021-22083-6 | PMC7979926 | Nascent non-coding RNA can mediate chromatin silencing, however mechanistically this process is poorly understood. Here the authors show that resolution of an R-loop during 3'-end processing of a plant antisense transcript recruits chromatin modifiers to promote chromatin silencing. | RNA-mediated chromatin silencing is central to genome regulation in many organisms. However, how nascent non-coding transcripts regulate chromatin is poorly understood. Here, through analysis of Arabidopsis FLC, we show that resolution of a nascent-transcript-induced R-loop promotes chromatin silencing. Stabilization of an antisense-induced R-loop at the 3′ end of FLC enables an RNA binding protein FCA, with its direct partner FY/WDR33 and other 3′-end processing factors, to polyadenylate the nascent antisense transcript. This clears the R-loop and recruits the chromatin modifiers demethylating H3K4me1. FCA immunoprecipitates with components of the m6A writer complex, and m6A modification affects dynamics of FCA nuclear condensates, and promotes FLC chromatin silencing. This mechanism also targets other loci in the Arabidopsis genome, and consistent with this fca and fy are hypersensitive to a DNA damage-inducing drug. These results show how modulation of R-loop stability by co-transcriptional RNA processing can trigger chromatin silencing. | IntroductionAntisense transcription is pervasive in many genomes, although the transcripts are rarely detected due to their rapid turnover1. This has made the analysis of antisense transcript function difficult to elaborate. A classic example of antisense-mediated chromatin silencing is at the floral repressor locus, Arabidopsis FLC. The antisense transcripts, named COOLAIR through their induction by cold temperatures2, are also functionally important at warm temperatures as part of the autonomous floral pathway3. The autonomous pathway transcriptionally silences FLC, and thus accelerates flowering. Genetic screens have identified RNA-binding proteins (FCA, FPA, and FLK), conserved 3′-end processing factors (FY/WDR33, CstF64, CstF77), the PRP8 splicing factor, and FLD a histone demethylase homolog, as components necessary for this transcriptional silencing3–6. These co-transcriptional regulators promote proximal polyadenylation of COOLAIR7–9, which leads to the recruitment of a set of physically interacting chromatin modifiers FLD/LD/SDG2610. These inhibit H3K4me1 and H3K36me3 accumulation and by antagonizing transcription, promote H3K27me3 accumulation, reducing FLC transcriptional initiation and elongation rates. Previously, we have reported an R-loop (three-stranded nucleic acid structure, RNA hybridized to DNA with a displaced single-stranded DNA) is generated by the COOLAIR transcript at the 3′-end of FLC11. Genomic analysis indicates a link between R-loop accumulation and certain histone modification12,13, with links to chromatin silencing14–17, however, how R-loops influence chromatin silencing is still unclear. Here, we describe a mechanism in which modulation of R-loop stability by co-transcriptional RNA processing involving m6A modification can trigger chromatin silencing.ResultsStabilization of the COOLAIR R-loop promotes FLC chromatin silencingThe COOLAIR R-loop at the 3′-end of FLC is stabilized by the homeodomain protein AtNDX, which inhibits further antisense transcription11. The R-loop corresponds in length to the nascent short proximal COOLAIR (~626 nucleotides) (Fig. 1a), whose formation is promoted by the RNA-binding proteins FCA and FPA and the canonical 3′-end processing factors FY, CstF64, and CstF777,18. To address how NDX and FCA function together mechanistically at FLC, we analyzed the genetic interaction between fca and ndx. FLC levels are slightly de-repressed in ndx1-4, significantly de-repressed in fca-9, and similar to fca-9 in the double mutant (Fig. 1b and Supplementary Fig. 1), consistent with NDX facilitating FLC epigenetic silencing via FCA. FCA associates along the length of the nascent COOLAIR transcript (Fig. 1c) and promotes proximal polyadenylation of COOLAIR19. However, this association was found to be reduced in ndx1-4 (Fig. 1c), which together with reduced levels of the R-loop11, suggests that R-loop stability influences FCA–COOLAIR association. We, therefore, asked whether the reduced FCA–COOLAIR association would affect FLC chromatin silencing and found that H3K4me1, a modification we have recently shown needs to be removed to establish a H3K27me3 domain at FLC19, accumulates over FLC in ndx1-4, although to a lesser extent than fca-9 (Fig. 1d). Consistent with FLC expression level, the double mutant did not show an additive effect (Fig. 1d). These data imply that the NDX stabilized R-loop structure enhances FCA–COOLAIR association to trigger FLC chromatin silencing. R-loop stabilization, potentially with RNA Pol II stalling, may provide an extended time window for FCA to function.Fig. 1Co-transcriptionally formed R-loop promotes FLC chromatin silencing.a S9.6-DNA/RNA immunoprecipitation (DRIP)-qPCR analyzing the R-loop over 3′-end of FLC in wild-type Col-0, with and without RNase H treatment. The number on x axis is the distance to FLC transcription start site (TSS = 0), and x axis is corresponding to the schematic on the top. TTS transcription termination site. Data are mean ± s.e.m. from three independent experiments. b Expression of spliced FLC relative to UBC in various genetic backgrounds. Data are normalized to wild-type Col-0. Data are mean ± s.e.m. from three to five biological replicates. Two-tailed P value from multiple t test corrected by Holm–Sidak method. c FCA–RIP–qPCR analyzing FCA enrichment on COOLAIR transcript in Col-0, ndx1-4 and fca-9 (negative control). The number on x axis is the distance to COOLAIR 5′-end. Data are mean ± s.e.m. from three biological replicates. d ChIP analysis of H3K4me1 level at FLC in various genetic backgrounds. The number on x axis is the distance to FLC TSS. Data are mean ± s.d. from three biological replicates. e DRIP–qPCR analyzing the R-loop in Fig. 1a (COOLAIR R-loop) in Col-0 and mutants fca-9 and fld-4. Data are mean ± s.e.m. from three biological replicates. f DRIP–qPCR analyzing the COOLAIR R-loop in Col-0 and fy-2. Data are mean ± s.e.m. from three biological replicates. Source data are provided as a Source Data file.FCA and FY promote R-loop resolution via efficient 3′-end processingEmerging evidence supports the idea that R-loops play regulatory roles in many processes16,20,21, but accumulated R-loops are harmful to genome stability22–24. Therefore, what determines R-loop homeostasis at FLC and how FCA might participate in this process were important questions to answer. We found the R-loop level over the COOLAIR proximal polyadenylation site was increased significantly in fca-9 (Fig. 1e). An unrelated high GC region on a COPIA transposon was used as a positive control for R-loop formation and showed no R-loop difference (Supplementary Fig. 2). The increased R-loop abundance was not just a feature of enhanced transcription, as no increase was found in fld-4 (Fig. 1e), a mutant of the histone demethylase FLD shown to be required for removing H3K4me1 at FLC19, with a similar FLC expression level as in fca-9. These data also suggest R-loop regulation acts upstream of chromatin modification at FLC. Factors in RNA metabolism have previously been shown to prevent R-loop formation25,26, but our genetic analysis suggests NDX stabilization of R-loop is necessary for FCA action, arguing against FCA preventing R-loop formation. Instead, we considered that FCA, which physically interacts with RNA 3′-end processing factors, might promote efficient 3′-end processing to facilitate R-loop resolution. To test this possibility, we tested the role of FY/WDR33, a canonical 3′-end RNA-processing factor that is a direct interactor of FCA, and is required for FCA function and COOLAIR proximal polyadenylation18. R-loop levels increased significantly in fy-2 (Fig. 1f), suggesting that FCA and 3′-end processing factors, which localize into dynamic liquid-like nuclear condensates19, mediate efficient RNA 3′-end processing to facilitate R-loop resolution.m6A modification on COOLAIR promotes FCA-mediated FLC repressionFactors co-immunoprecipitated with FCA after formaldehyde cross-linking19 may function in this FCA/FY-mediated R-loop resolution mechanism. DNA/RNA helicases were identified (Supplementary Table 1), and these might facilitate the unwinding of the RNA/DNA hybrid. Several subunits of the m6A methyltransferase writer complex MTA (homolog of human METTL3), MTB (METLL14), and FIP37(WTAP) were also identified (Supplementary Table 1). The m6A methyltransferase writer complex delivers reversible and dynamic m6A modification to eukaryotic mRNA and lncRNA and is thought to have multiple roles in RNA metabolism and processing27–31. To test the functional relevance of m6A methylation in this FCA-mediated mechanism, we carried out genetic analysis by introducing m6A writer mutations into a transgenic line (C2) carrying 35 S::FCAγ (FCA overexpressor) and FRIGIDA (encodes an activator of FLC) transgenes, a line we had used for suppressor screens to identify components required for FCA-mediated FLC repression7–9,32. FLC is strongly repressed in this sensitized background, which enables us to identify relatively weak-effect mutations that suppress FCA-mediated FLC repression. Null mutations of m6A methyltransferases are embryonic lethal, so we used transgenic lines containing embryo-rescued genotypes (mta carrying the pABI3::MTA transgene33 and fip37 carrying the pLEC1::FIP37 transgene34), where m6A modification levels in the seedlings are largely diminished compared to wild-type. Lack of this m6A methyltransferase activity strongly suppressed FCA function resulting in enhanced FLC expression (Fig. 2a, b). Consistent with this, mta in the C2 background showed a strong late-flowering phenotype (Supplementary Fig. 3a). Importantly, neither FCAγ or endogenous FCA protein levels were affected by the mta or fip37 mutation, respectively (Supplementary Fig. 3b, c). These results were reminiscent of the studies in mammals elucidating the roles of RNA-binding protein and m6A writer complex in m6A-modified XIST RNA-mediated gene silencing35,36. m6A has been reported to be deposited co-transcriptionally37, we, therefore, asked whether nascent COOLAIR is coated with m6A. m6A was enriched over the region of COOLAIR that coincides with the proximal form (Fig. 2c). This suggests that FCA does not directly bind to m6A because FCA associates with most segments of COOLAIR nascent transcript (Fig. 1c). To confirm this, we undertook in vivo RNA immunoprecipitation (RIP) to analyze m6A enrichment in RNA fractions bound to FCA. The isolated RNAs were digested into nucleosides and subsequently injected into liquid chromatography–tandem mass spectrometry (LC-MS/MS) for m6A detection (a cartoon description in Supplementary Fig. 4). Surprisingly, we found m6A-modified RNAs were significantly enriched in the FCA–RIP fraction (Fig. 2d), indicating FCA binds to RNAs enriched for m6A. However, after RNase T1 digestion, which leaves only FCA-directly bound regions, the m6A enrichment was lost (Fig. 2d), indicating FCA indeed does not bind m6A directly. Interestingly, we have found the m6A to A ratio of total mRNA is reduced in fca mutants (Supplementary Fig. 5a) and the m6A level on COOLAIR is also reduced in fca-9 (Supplementary Fig. 5b), suggesting FCA facilitates the deposition of m6A on some transcripts, including COOLAIR. We also found m6A is enriched over the 3′-end of the FLC sense transcript, but the level did not change in fca-9 (Supplementary Fig. 5b), suggesting a parallel FCA-independent pathway. Our data collectively show the interplay between m6A and FCA is not linear; FCA facilitates the addition of m6A on COOLAIR but m6A modification is required for FCA-mediated FLC silencing.Fig. 2m6A methylation on COOLAIR is required for FCA-mediated FLC silencing.a Expression of unspliced FLC relative to UBC in C2 and fip37 C2 2-week-old seedlings, both carrying transgene pLEC1::FIP37. Data are normalized to C2. Data are mean ± s.e.m. from three biological replicates. P value from two-tailed t test. b Expression of unspliced and spliced FLC relative to UBC in C2 and mta C2 (carrying transgene pABI3::MTA) plants. Data are normalized to C2. Data are mean ± s.e.m. from three biological replicates. P value from two-tailed t test. c m6A-IP-qPCR analyzing m6A enrichment on chromatin-associated COOLAIR transcript after immunoprecipitated by m6A antibody. a–g represent amplicons over COOLAIR in qPCR analysis, with positions indicated on the top schematic. Data are mean ± s.d. from five to six biological replicates. d LC-MS/MS determined m6A-to-A ratio on nuclear RNA immunoprecipitated by FCA, with and without RNase T1 digestion. Data are mean ± s.d. from six biological replicates. Two-tailed P value from multiple t test corrected by Holm–Sidak method. Source data are provided as a Source Data file.We next asked whether m6A affects FCA association with COOLAIR using in vitro and in vivo FCA–RIP assays. The 5′-end of FCA containing the RRM domains was incubated with total nuclear RNA, which had been treated with the m6A demethylase FTO38, removing about 70% of the m6A modification (Supplementary Fig. 6a). Loss of m6A reduced the ability of FCA (RRM) to immunoprecipitate COOLAIR (Supplementary Fig. 6b). Full-length FCA showed the same binding to COOLAIR as an FCA-RRM domain, and again this was dependent on m6A (Supplementary Fig. 6c). In vivo FCA–RIP analysis in mta consistently showed less occupancy of FCA on COOLAIR at the regions deposited with m6A (Fig. 3a). Nevertheless, we noted FCA largely remained associated with COOLAIR in the mta mutant (Fig. 3a). This again argues against m6A being part of the direct linear FCA recruitment mechanism, as then we would have expected a reduction in the overall FCA occupancy on COOLAIR. Rather it suggests the interaction of FCA with COOLAIR is enhanced by the RNA m6A modification, possibly through an m6A influenced COOLAIR structure change, as COOLAIR has been shown to be strongly structured39.Fig. 3m6A enhances FCA binding to COOLAIR and FCA condensates formation.a FCA–RIP-qPCR analyzing FCA binding to COOLAIR transcript in mta mutant and the corresponding wild-type control (CTL). x axis represents the amplicons in qPCR (positions refer to the schematic in Fig. 2d). Data are mean ± s.d. from three biological replicates. Two-tailed P value from multiple t test corrected by Holm–Sidak method. b Representative images of root tip nuclei expressing pFCA::FCA-mTurquoise2 in plants with and without mta mutation. Maximum intensity projections of Z-stack spanning the entire width of a nucleus were applied. Scale bars, 5 μm. c Quantification of FCA-mTurquoise2 condensates number in root cells in plants with and without mta mutation. Data were plotted from minima to maxima. The box extends from the 25th to 75th percentiles. The line inside the box marks the median. The whiskers go down to the minima and up to the maxima. P value from two-tailed t test. d, The distribution of FCA-mTurquoise2 condensates of different sizes (in two groups, size between 0.01 and 0.02 μm2 and bigger than 0.02 μm2) in plants with (n = 59 nuclei) and without (n = 60 nuclei) mta mutation. P = 0.0237, two-sided Fisher’s exact test. e The ratio of proximal-to-distal isoforms of COOLAIR transcripts (refer to the schematic in Fig. 2d) in mta C2 relative to corresponding wild-type C2. Data are mean ± s.d. from three biological replicates. f DRIP–qPCR analyzing COOLAIR R-loop in mta mutant and corresponding wild-type control (CTL). Data are mean ± s.d. from three biological replicates. Source data are provided as a Source Data file. Raw and processed images for Fig. 3c, d are available from Figshare (10.6084/m9.figshare.13645730.v1).m6A influences the dynamics of FCA nuclear condensates in vivoThe secondary structure of RNA can influence the properties of liquid compartments40, so it is pertinent that FCA compartmentalizes with 3′-end processing factors in dynamic liquid-like nuclear condensates both in vitro and in vivo19. The phase separation processes involved in the formation of stress granules are enhanced by RNAs containing multiple m6A modifications41. Both MTA and FIP37 are targeted to nuclear speckles in a transient assay42. Arabidopsis MTA protein and its orthologs in other plant species (i.e., G. max, S. tuberosum) are predicted to be highly disordered and contain a prion-like domain (PrLD) (Supplementary Fig. 7a, b). Interestingly, MTA orthologs in other organisms (i.e., H. sapiens, D. melanogaster, D. reiro) are also predicted to be highly disordered but do not contain any PrLDs (Supplementary Fig. 7b). FCA contains two predicted PrLDs19, so we hypothesized MTA and m6A deposition might affect the dynamics of FCA nuclear condensates, enhancing the association of FCA and COOLAIR. Functionally this would prolong occupancy time of FCA and 3′-end processing factors at sites of nascent COOLAIR production and promote R-loop resolution. To test this hypothesis, we undertook an in vivo analysis of the nuclear condensates by imaging root nuclei in transgenic plants. We generated plants carrying an pFCA::FCA-mTurquoise2 fusion (in a Col genotype), in which the transgene fully complemented the fca-9 mutation, and crossed to the embryo-rescued genotype pABI3::MTA in mta, reduced in m6A levels in seedlings. In wild-type plant roots, the FCA-mTurquoise2 was localized to nuclear condensates of various sizes (Fig. 3b), similar to the FCA-eGFP fusion which we had reported previously (although this had been in a Ler genotype)19. After the introduction of the mta mutation, the number of FCA-mTurquoise2 foci reduced from a mean of 5.3 per nucleus to 3.4, and the number of larger condensates reduced from ~67% of the total to ~60% of the total (compared to segregants from the same cross, guaranteed to carry the same amount of FCA-mTurquoise2 transgene, see Genotyping in “Methods”) (Fig. 3b–d). These data support that the m6A modification affects the equilibrium influencing the dynamic lifetime of the FCA nuclear condensates, promoting component association, albeit probably over short timescales in vivo.We speculate the change in the dynamics of FCA nuclear condensates affects COOLAIR 3′-end processing, which underlies the de-repression of FLC when m6A is diminished. Indeed, we found a reduced use of the proximal COOLAIR polyadenylation site in mta C2, like in fca-9 and fy-2 (Fig. 3e). Consistently, the COOLAIR R-loop was increased in mta (Fig. 3f). Collectively, our data show that stabilization of the R-loop by NDX1 enables FCA-mediated RNA 3′-end processing to result in COOLAIR proximal polyadenylation and thus resolve the co-transcriptionally formed R-loop. FCA and RNA 3′-end processing factors condense into nuclear speckles and m6A modification enhances this condensation. Such a dynamic interplay helps explain the relationship between m6A and R-loop stability, which has been controversial43,44. We did not find global R-loop changes in mta (Supplementary Fig. 8), which further argues against a simple linear regulatory relationship between m6A and R-loop regulation, with the context being important.FCA and FY promote R-loop resolution at a subset of targets in the genomeFCA had been originally identified as a regulator of flowering but was subsequently shown to have widespread roles in the Arabidopsis genome45,46. Thus, we speculated that FCA-promoted R-loop resolution might not be specific to FLC. Indeed, we observed R-loop accumulation globally in fca-9 using a dot blot analysis (Supplementary Fig. 9). We had previously shown that FCA bound and promoted proximal polyadenylation of many Arabidopsis transcripts. The read-through transcription found in fca mutants generated chimeric, spliced transcripts (referred to as unannotated segments -UA loci)46 (Supplementary Fig. 10). In order to directly test for a general association of FCA-mediated 3′-end processing with R-loop regulation, we analyzed R-loops at several UA loci. We found R-loops formed near proximal polyadenylation sites over UA2, UA10B, and XRCC4, which increased in fca-9 (Fig. 4a). We also tested these R-loops in fy-2. The R-loops strongly increased at UA10B and XRCC4 (Fig. 4b), and slightly increased at UA2 (Fig. 4b). These data suggest that FCA-bound transcripts are frequently associated with R-loop formation and FCA/FY promotion of efficient 3′-end processing to trigger R-loop resolution occurs at least at a subset of FCA targets in the Arabidopsis genome.Fig. 4FCA and FY promote R-loop resolution at a subset of targets, so protecting the genome from DNA damage during replication.a DRIP–qPCR analyzing R-loops at UA2, UA10B, and XRCC4 (UA R-loops) in Col-0 and fca-9, with and without RNase H treatment. p1 to p4 are amplicons in the qPCR, which are indicated in the schematic of each locus. Data are mean ± s.d. from three biological replicates. b DRIP–qPCR analyzing UA R-loops in Col-0 and fy-2. Data are mean of two biological replicates. c seedlings were treated with and without bleomycin. After treatment and recovery (see “Methods”), plants with or without true leaves were scored. Data are presented as the number of seedlings in two categories. d “Detangle model”. FCA associates with co-transcriptionally formed R-loops, potentially with Pol II stalling. FCA facilitates the deposition of m6A onto the nascent transcript by the m6A writer complex. This enhances the multivalent interactions promoting the formation of dynamic nuclear condensates. Consequently, this promotes polyadenylation at noncanonical polyadenylation (p(A)) sites, concomitantly resolving the R-loop. At FLC, this process is linked to chromatin silencing via interaction of the 3′-end processing factors with a histone K4 demethylase FLD. Without FCA or FY or the m6A writer complex, R-loops accumulate, and transcripts polyadenylate at efficient p(A) sites. Source data for Fig. 4a–c are provided as a Source Data file.DiscussionIn summary, our analysis identifies the importance of RNA 3′-end processing in the resolution of a COOLAIR-generated R-loop to trigger chromatin silencing at the Arabidopsis floral repressor locus, FLC. A simple recruitment model of the RNA-binding protein FCA to the COOLAIR nascent transcript is unlikely given our data. Instead, the mechanism involves dynamic nuclear condensates promoted by multivalent interactions with positive feedbacks involving m6A methylation. These condensates produce locally high concentrations of RNA-processing factors that efficiently terminate transcription, thus resolving the R-loop and producing the proximally polyadenylated COOLAIR transcript. The local chromatin environment eg. negative supercoiling, stalled RNA Pol II may be the trigger generating the R-loop structure, with the initial biochemical unwinding involving DNA/RNA helicases (Supplementary Table 1 and Fig. 4d). At FLC, the 3′-end processing factors in this noncanonical polyadenylation mechanism physically link to the recruitment of chromatin modifiers that silence the locus19. However, FCA has been shown to have widespread roles in RNA-mediated chromatin silencing in the Arabidopsis genome, with parallels to co-transcriptional silencing mechanisms in S. pombe47,48. This noncanonical polyadenylation mechanism could therefore be viewed as a clearing tool for chromatin “tangles” (e.g., R-loops) (Fig. 4d), frequently generated but extremely damaging to the genome, especially during DNA replication12,49,50. This more widespread genomic function is supported by the hypersensitivity of fca and fy mutants to the DNA damaging drug Bleomycin, as evidenced by the significant reduction in cell division measured through true leaf production in mutant versus wild-type seedlings after drug treatment (Fig. 4c and Supplementary Fig. 11a, b). Our work provides mechanistic insights on the dynamic interplay between co-transcriptionally formed R-loops, RNA-processing factors, and m6A RNA modification, and how that interplay connects to patterns of chromatin modification.MethodsPlant materials and growth conditionsMutant alleles ndx1-411, fca-951, fld-48, and fy-251 were described previously. C2 line was described previously8. mta (pABI3::MTA in mta) was provided by Dr Kamil Ruzicka (Institute of Experimental Botany of the Czech Academy of Sciences) along with Dr Rupert G. Fray (University of Nottingham) and was described previously33. fip37 (pLEC1::FIP37 in fip37-4) was provided by Dr. Hao Yu (National University of Singapore) and was described previously34.To generate the pFCA:: FCA-mTurquoise2 transgenic line, FCA genomic DNA was amplified and inserted into the pCambia1300 vector. The FCA ApaI fragment (two ApaI sites on FCA genomic DNA) was swapped by FCA ApaI fragment fused with mTurquoise2 sequence, which was inserted before the stop codon via the HindIII site. The construct was transformed into the fca-9 mutant.Seedlings were surface sterilized and sown on standard half-strength Murashige and Skoog (½ MS) medium plate without glucose and stratified at 4 °C for 3 days before transferred to long-day conditions (16-h light at 20 °C, 8-h darkness at 16 °C).DRIPIn all, 2-g 10-day-old seedlings were harvested and grounded into a fine powder. The powder was suspended in 30 mL of Honda buffer (20 mM HEPES, 0.44 M sucrose, 1.25% Ficoll, 2.5% dextran T40, 10 mM MgCl2, 0.5% Triton x-100, 5 mM DTT, 1x protease inhibitor cocktail (Roche)), filtered through two layers of Miracloth, and centrifuged at 3500×g for 15 min. Nuclear pellets were resuspended in 1 mL Honda buffer and centrifuged at 8000×g for 1 min. Pellets were then resuspended in the lysis buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA, 1% SDS) supplied with 0.1 mg/mL proteinase K (AM2546, Invitrogen) and digested at 55 °C overnight with gentle rotation. The mixture was phenol/chloroform extracted, followed by DNA precipitation with NaOAc and isopropanol. The DNA pellet was dissolved gently in water and treated with Proteinase K for another 2 h, followed by phenol/chloroform extraction and DNA precipitation. The DNA pellet was dissolved in water and quantified with Qubit DNA quantification kit (Invitrogen). In total, 1 μg of DNA was dissolved in 50 μl lysis buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA, 0.5% SDS), sonicated with Diagenode Bioruptor® for 15 times, 30 s on/30 s off at high setting. DNA was then diluted ten times with dilution buffer (16.7 mM Tris pH 7.5, 167 mM NaCl, 2.2 mM EDTA, 0.1% Triton X-100) and 1% was stored at −20 °C as input. In all, 5 μg of S9.6 antibody (1:100 dilution, ENH001, Kerafast) was added, then incubated overnight at 4 °C. The next day, 50 μl Protein G Agarose (Invitrogen) was added and incubated for another 2 h. The immunoprecipitants were washed five times with dilution buffer and twice with TE buffer, then were eluted in 200 μl elution buffer (10 mM Tris pH 7.5, 2 mM EDTA, 0.2% SDS, 100 ng/μl tRNA) at 55 °C for 1 h, together with input samples. The nucleic acids were precipitated with NaOAc, isopropanol, and glycogen, dissolved in water, and subjected to qPCR analysis via LightCycler480 II (Roche). The data were normalized to 1% of input. Primers were listed in Supplementary Table 2. For RNase H-treated sample, DNA was treated overnight with RNase H at 37 °C after RNase A treatment for 1 h at 37 °C.S9.6 dot blotGenomic DNA was extracted and sonicated as described in DRIP (see above). RNase H treatment was also described in DRIP. After quantification by Qubit DNA quantification kit, 2 μl DNA was blotted on Hybond™-N + membrane (Amersham). Before totally dry, the membrane was cross-linked with UV (1200 mJ/cm2). The membrane was blocked in 5% (w/v) milk in TBST buffer for 1 h at room temperature, then 0.5 μg/ml S9.6 antibody (1:2000 dilution, ENH001, Kerafast) was added and incubated overnight at 4 °C overnight. After several washes in TBST buffer, the membrane was incubated in mouse IgG horseradish peroxidase linked whole antibody (1:20,000 dilution, NA931, GE Healthcare) before using chemiluminescence (Thermo Scientific) for detection. Loading was either stained in 0.1% (w/v) Methylene Blue or by SYBR™ Safe DNA Gel Stain (Thermo Scientific).In vivo RIPIn all, 2-g 10-day-old seedlings were harvested and cross-linked with 1% formaldehyde. After being ground into a fine powder, the material was suspended in 30 mL of Honda buffer in the presence of 50 ng/μl tRNA, 20 U/mL RNase inhibitor (SUPERase•In™, Invitrogen) and 1x cOmplete Protease inhibitor (Roche), filtered through two layers of Miracloth, and centrifuged at 3500 × g for 5 min. Nuclear pellets were resuspended in 2.5 volumes of Nuclei Lysis Buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA, 1% SDS, 1× protease inhibitor cocktail, 50 ng/μl tRNA) and sonicated with Diagenode Bioruptor® for five times, 30 s on/30 s off at high setting, followed by another ten times at low setting. Immunoprecipitation was performed by incubating 30 μl Dynabeads™ protein A (Invitrogen), FCA antibody (1:500 dilution, homemade52), and 1.2 mL of diluted chromatin (containing 100 μL of sonicated chromatin) at 4 °C for 1.5 h. After IP, the beads were washed four times in the washing buffer (167 mM NaCl, 16.7 mM Tris pH 7.5, 1.2 mM EDTA, 0.8% Triton X-100, 1× protease inhibitor, 50 ng/μl tRNA, 20 U/mL RNase inhibitor). Reverse cross-linking and elution were done by adding 200 μl elution buffer (2 mM EDTA, 0.2% SDS, 100 ng/μl tRNA, 0.4 U/μl RNaseOUT (Invitrogen)) to the washed beads and incubating at 55 °C overnight. The RNA from the supernatant was precipitated with isopropanol and glycogen, dissolved, DNase treated, and then used as a template for reverse transcription (RT) with gene-specific primers. Data were presented as IP/1% of input (RNA). Minus RT controls were set up to ensure the values reflect the level of RNA and not DNA contamination. UBC was used as a negative control for the experiment. Primers are listed in Supplementary Table 2.ChIPHistone ChIP was performed as previously described53. Protein A magnetic beads (Invitrogen, 10002D), anti-H3 (3 μg, ab1791, Abcam), and anti-H3K4me1 (6 μg, ab8895, Abcam) were used. After immunoprecipitation, recovered DNA was quantified by qPCR with primers listed in Supplementary Table 2. Data were normalized to input, and values were shown as the ratio of H3K4me1 to H3.Expression analysisTen-day-old seedlings were harvested, and RNA was extracted. For fip37, seedlings were harvested 14 days after germination, seedlings without true leaves were taken as homozygous mutants. After treated with TURBO DNase (Ambion) to remove DNA contamination, RNA was reverse-transcribed by SuperScript IV Reverse Transcriptase (Invitrogen) using gene-specific primers. qPCR analysis was performed and data were normalized to UBC. Primers are described in Supplementary Table 2.Western blot analysisTotal protein extracts were separated on NuPAGE 4–12% Bis-Tris Gels (Invitrogen) and transferred to 0.45-μm PVDF membrane (GE Healthcare). The membrane was blocked in 5% (w/v) milk in phosphate-buffered saline (with 0.1% Tween-20) (PBST) for 1 h at room temperature. Anti-FCA (1:8000 dilution, homemade) or anti-H3 (1:1500 dilution, ab1791, Abcam) antibody was added, and the incubation was carried out at 4 °C overnight. After several washes in PBST buffer, the membrane was incubated in rabbit IgG horseradish peroxidase linked whole antibody (1:10,000 dilution, NA934, GE Healthcare) before using chemiluminescence for detection. Protein loading was visualized after Coomassie Blue staining.Cloning, expression, and purification of recombinant proteinsThe sequence corresponding to the N-terminal of FCA (ATG to 948 bp) containing both RRM domains or FCA full length was amplified from cDNA and inserted into the pGEX-6P-1 vector (GE Healthcare). Freshly transformed cells (E. coli BL21DE3) were grown in terrific broth medium at 37 °C for 6 h, followed by induction of protein expression for 3 h at 30 °C with 1 mM IPTG. The GST-tagged protein was purified from the cells by following a protocol provided with Glutathione Sepharose® 4 Fast Flow (GE Healthcare).In vitro RIPGST-tagged FCA protein (full length or N-terminal, including both RRM domains) was expressed and purified as described before. Human FTO protein for m6A demethylation reaction was expressed and purified as previously described. For RNA extraction, nuclei were prepared as in RIP assay and digested with protease K at 55 °C for 2 h. The mixture was extracted with phenol/chloroform (pH < 5.0) followed by DNase I digestion at 37 °C for 2 h and additional phenol/chloroform extraction. For m6A erased sample, 3 μg of resulting RNA were treated with 1 nmol purified FTO protein in reaction buffer (50 mM HEPES pH 7.0, 300 μM Fe2+, 500 μM α-ketoglutaric acid, 2 mM L-ascorbic acid, 0.1 U/μl SUPERase•In, 0.5 U/μl RiboLock (Thermo Scientific™)) at 37 °C for 1 h. For the control sample, 3 μg RNA and 1 nmol purified FTO were mixed in buffer with 50 mM HEPES pH 7.0, 5 mM EDTA, 0.1 U/μl SUPERase•In, 0.5 U/μl RiboLock (without any cofactor so that the demethylation reaction never happens) incubating at 37 °C for 1 h. The m6A/A concentration ratio was detected with LC-MS/MS. For each in vitro RIP sample, 150 pM protein (GST-tagged FCA-RRM or only GST) and 1.5 μg RNA (m6A erased or control) were mixed in binding buffer with 20 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.1% NP-40, 0.5 U/μl RiboLock at 4 °C for 2 h. Glutathione Sepharose® 4 Fast Flow (GE Healthcare) was washed and blocked with 1% BSA, 10 μg/ml yeast tRNA, and 10 μg/ml salmon sperm DNA for 2 h. For each sample, 100 μl of pre-blocked sepharose was added to the mixture incubating at 4 °C for another 2 h. The sepharose was washed with a binding buffer five times. Immunoprecipitated RNA in each sample was extracted with Trizol and reverse-transcribed using Superscript III reverse transcriptase (Invitrogen) with random hexamer. The pulldown assay using GST-tagged full-length FCA was performed under the same condition, except for using Dynabeads Protein A coated with FCA antibody instead of Glutathione Sepharose.Quantification of RNA modification (LC/MS-MS)Isolated RNA samples were digested with 0.5 U nuclease P1 in 50 μl 10 mM ammonium acetate (pH 5.3) at 42 °C for 3 h, followed by the addition of 5 μl of 1 M fresh NH4HCO3 and 0.5 U shrimp alkaline phosphatase (NEB). The mixture was incubated at 37 °C overnight. The resulting nucleosides were separated by UPLC and detected by Triple Quad™ 5500 (AB SCIEX) mass spectrometer. Nucleosides were quantified using the nucleoside-to-base ion mass transitions of m/z 268.0 to 136.0 (A), m/z 282.0 to 150.1 (m6A). Concentrations of nucleosides in samples were calculated by fitting the signal intensities to the standard curves, and the m6A/A ratios were calculated accordingly.FCA-binding region RNA-modification detectionRNA immunoprecipitation was performed as previously described in RIP assay (see above), except using salmon sperm DNA instead of yeast tRNA. In total, 1% sonicated chromatin was saved as an input sample. After IP procedure, half of the washed beads were eluted and the resulting RNA was regarded as FCA–RIP sample. The other half of the beads was treated with RNase T1 (1 U/μl, Thermo Scientific) in 20 μl buffer with 20 mM Tris-HCl pH 7.5, 2.5 mM EDTA, 1× protease inhibitor cocktail at 37 °C for 10 min. The beads were then washed and eluted as described in the RIP assay to obtain the FCA-binding region sample, while the supernatant was regarded as RNase T1 digested flow-through the sample. The m6A/A concentration ratio was detected with LC-MS/MS for all samples as described above.m6A immunoprecipitationNuclear pellets were isolated and sonicated as described in the RIP assay (without cross-linking). RNA in the supernatant was extracted with TRIzol reagent and treated with TURBO DNase twice to remove any DNA contamination. In total, 5% extracted RNA was saved as an input sample. Each 10 μg RNA was incubated with 5 μg m6A antibody (202003, Synaptic Systems) in 400 μl IP buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 5 mM EDTA, 0.1% NP-40, 1 × Ribolock RNase inhibitor) at 4 °C for 2 h. In all, 35 μl Dynabeads Protein A was washed three times with washing buffer (IP buffer without RNase inhibitor) and incubated with the mixture for an additional 2 h. The beads were washed five times in a new DNA LoBind tube with washing buffer. The immunoprecipitated RNA was eluted with 100 μl elution buffer (IP buffer with 10 mM N6-methyladenosine) twice at 4 °C for 1 h, then precipitated with isopropanol and glycogen. The immunoprecipitated RNA and input RNA were reverse-transcribed and quantified by qPCR as described in the RIP assay. Primers are listed in Supplementary Table 2.Microscopy and image analysisSeedlings were grown on ½ MS plate with 1% (w/v) sucrose and 0.5% (w/v) phytagel (Sigma-Aldrich) for 7 days. Fresh roots were mounted in water and FCA-mTurquoise2 signal was captured by Zeiss LSM780 confocal microscope using a 40 × /1.2 water objective through the GaAsP spectral detector. mTurquoise2 was excited at 458 nm and detected at 463–579 nm. z-stack projection was applied for every image to capture 3D data of a nucleus. “StackReg” Plugin was applied for correcting drifted signal54 when using z-stack maximum projection. The number of condensates in each nucleus was obtained manually (each counted cell was labeled and the information was available from Figshare (10.6084/m9.figshare.13645730.v1). Nuclei were extracted from the whole root for quantification (as high auto-fluorescence outside the nuclei would influence the quantification) by ImageJ software, which was used for analyzing the sizes of the condensates. “Analyze Particles” tool was applied to obtain “area” data for each condensate after manual thresholding. The same settings were used on all images). Raw data and processed images are available from Figshare (10.6084/m9.figshare.13645730.v1).GenotypingC219, mta33, and fip3734 plants were genotyped, as described previously. To genotype the fca-9 mutation, genomic DNA was amplified with primers fca-9-dCAPS_F + R, respectively. PCR products were digested with StyI, followed by 3% agarose gel electrophoresis. ndx1-4 was genotyped with the transfer-DNA primer p745 and ndx1-4_GT_R to detect the presence of the insertion. PCR using ndx1-4_GT_F + R was performed to check whether the transfer-DNA insertion was homozygous. Primers are listed in Supplementary Table 2.To obtain pFCA::FCA-mTurquoise2 in mta and appropriate control, the seeds of pFCA::FCA-mTurquoise2 homozygous plants in mta/ + background were first obtained. After germination, thirty 7-day-old seedlings were picked up, and only the root tips were mounted in water for imaging and later image analysis. After imaging, seedlings were grown for another 7 days for DNA extraction and genotyping. Among these individuals, only seven were in mta−/− background. Another seven individuals with the same genetic background (MTA + / + ) were analyzed.Bleomycin treatmentFour-day-old seedlings of different genotypes were transferred to small Petri dishes containing 10 ml of the liquid growth medium, either without (control) or with 1 μg/ml of bleomycin. After 5 days of incubation (with seedlings floating in liquid, but no shaking) in the illuminated growth chamber (9-day-old seedlings), remove the medium, wash extensively by flooding the plate five times with 20 ml of liquid media. Transfer seedlings to solid medium plates. Allow seedlings to recover for 24 h before analysis. Plants were scored for true leaf emergence (shown in Supplementary Fig. 11a, b).Statistical analysisStatistical analyses were performed using software GraphPad Prism version 8.4.3. P value, sample number, and adjusted P value (after multiple comparison correction) are included in Figures or Figure legends.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Long non-coding RNAs",
"RNA modification",
"Plant molecular biology"
] | transcription pervasive in genomes rarely detected rapid turnover1 analysis difficult example antisense chromatin silencing at floral Arabidopsis FLC antisense transcripts COOLAIR important at warm temperatures autonomous floral pathway3 silences FLC accelerates flowering Genetic screens identified RNA-binding proteins (FCA FPA FLK), conserved 3′-end processing factors PRP8 splicing factor FLD histone demethylase necessary for transcriptional-transcriptional regulators promote proximal polyadenylation of COOLAIR7–9 recruitment chromatin modifiers FLD/LD/SDG2610 inhibit H3K4me1 H3K36me3 accumulation promote H3K27me3 accumulation FLC transcriptional initiation elongation rates R-loop generated by COOLAIR transcript at 3′-end of FLC11 Genomic analysis indicates link between R-loop accumulation histone modification12 chromatin R unclear mechanism modulation R-loop stability by co-transcriptional RNA processing m6A modification chromatin silencingCOOLAIR R-loop promotes FLC chromatin R-loop 3′-end FLC stabilized by protein AtNDX inhibits antisense R-loop corresponds to nascent short proximal COOLAIR (~626 nucleotides (Fig. formation promoted by RNA-binding proteins FCA FPA 3′-end processing factors FY CstF64 CstF777 NDX FCA analyzed genetic interaction fca ndx FLC levels slightly de-repressed in de-repressed in fca-9 similar double mutant NDX facilitating FLC silencing via FCA FCA associates COOLAIR transcript promotes proximal polyadenylation COOLAIR19 association reduced in ndx1-4 reduced levels R R-loop stability influences FCA–COOLAIR association reduced FCA–COOLAIR association FLC chromatin silencing H3K4me1 accumulates over FLC in ndx1-4 lesser than fca-9 double mutant additive effect NDX stabilized R-loop structure enhances FCA–COOLAIR association FLC chromatin silencing R-loop stabilization RNA stalling may extended time window for FCA functionR-loop FLC silencing S9.6-DNA/RNA immunoprecipitation-qPCR R-loop-end FLC wild-type Col-0 RNase H treatment distance FLC transcription start site transcription termination site mean ± s.m. three experiments Expression spliced FLC UBC genetic backgrounds normalized wild-type Col-0 mean ± three five replicates Two-tailed P value t test corrected Holm–Sidak method FCA–RIP–qPCR FCA enrichment COOLAIR transcript Col-0 fca-9 x distance COOLAIR 5′-end mean ± s.e.m. three replicates ChIP H3K4me1 level FLC genetic backgrounds x distance FLC TSS mean three replicates DRIP–qPCR R-loop Fig. 1a Col-0 mutants fca-9 fld-4 mean s three replicates DRIP–qPCR COOLAIR R-loop Col-0 fy-2 mean ± s.e.m. three replicates Source data fileFCA FY promote R-loop resolution 3′-end evidence R-loops regulatory accumulated harmful genome determines R-loop homeostasis at FLC FCA important R-loop level COOLAIR proximal polyadenylation site increased in fca-9 unrelated high GC region COPIA transposon R-loop no R difference increased R-loop abundance transcription no increase in fld-4 mutant histone demethylase FLD required removing H3K4me1 similar FLC expression level fca-9 suggest R-loop regulation acts upstream chromatin modification FLC Factors metabolism prevent R-loop genetic analysis suggests NDX stabilization R-loop necessary for FCA action FCA preventing-loop FCA interacts with RNA 3′-end processing factors might promote efficient 3′-end processing R-loop resolution tested role FY/WDR33 3′-end RNA-processing factor interactor FCA required for FCA function COOLAIR proximal polyadenylation18 R-loop levels increased in fy-2 FCA 3′-end processing factors mediate efficient RNA 3′-end processing R-loop resolutionm6A modification COOLAIR promotes FCA FLC repressionFactors co-immunoprecipitated with FCA after formaldehyde cross function FCA/FY R-loop resolution mechanism DNA/RNA helicases identified facilitate unwinding RNA/DNA hybrid subunits m6A methyltransferase writer complex MTA METTL3) MTB FIP37(WTAP) identified m6A delivers reversible modification to eukaryotic mRNA lncRNA multiple roles in RNA metabolism test relevance m6A methylation FCA mechanism genetic analysis m6A mutations into transgenic line 35 S::FCAγ FRIGIDA transgenes FLC repressed sensitized background identify weak-effect mutations suppress FCA FLC repression Null mutations m6A embryonic lethal used transgenic lines embryo-rescued genotypes (mta fip37 transgene34) m6A modification levels seedlings diminished wild-type Lack m6A methyltransferase activity FCA function enhanced FLC expressionmta C2 background showed late-flowering phenotype Fig. neither FCAγ or endogenous FCA protein levels affected by mta or fip37 mutation Fig 3b results studies RNA-binding protein m6A writer complex in m6A-modified XIST RNA gene m6A deposited co-transcriptionally37 asked nascent COOLAIR coated with m6A enriched over region COOLAIR proximal form (Fig. FCA directly bind to m6A FCA associates with COOLAIR nascent transcript in vivo RNA immunoprecipitation) m6A enrichment isolated RNAs digested into nucleosides injected into liquid spectrometry) for m6A detection m6A-modified RNAs enriched in FCA–RIP fraction FCA binds to RNAs enriched m6A after RNase T1 digestion m6A enrichment lost FCA bind m6A directly m6A to A ratio reduced in fca mutants m6A level on COOLAIR reduced in fca-9FCA facilitates deposition m6A transcripts COOLAIR m6A enriched 3′-end FLC transcript level fca-9 parallel FCA-independent pathway interplay m6A FCA linear FCA addition m6A COOLAIR modification required FCA FLC silencing 2m6A methylation COOLAIR required FCA FLC silencing unspliced FLC UBC C2 2-week-old seedlings transgene pLEC1::FIP37 normalized C2. mean ± s.m. three replicates two-tailed t test unspliced spliced FLC UBC C2 C2 plants normalized C2. mean ± s.e.m. three replicates two-tailed t test m6A-IP-qPCR m6A enrichment COOLAIR transcript immunoprecipitated m6A antibody amplicons COOLAIR qPCR mean ± s.d. five six replicates LC-MS m6A-to-A ratio nuclear RNA immunoprecipitated FCA T1 digestion mean ± s.d. six replicates test corrected Holm–Sidak method Source dataasked m6A FCA association COOLAIR vitro vivo FCA–RIP assays 5′-end FCA RRM incubated nuclear RNA treated m6A demethylase FTO38 70% m6A modification Loss m6A reduced FCA immunoprecipitate COOLAIR Full-length FCA COOLAIR FCA-RRM domain dependent m6A vivo FCA–RIP analysis less occupancy FCA COOLAIR regions m6A FCA associated COOLAIR mta mutant argues against m6A FCA recruitment mechanism FCA occupancy COOLAIR suggests interaction FCA COOLAIR enhanced RNA m6A modification influenced COOLAIR structure change 3m6A enhances FCA binding COOLAIR FCA condensates formation FCA–RIP-qPCR FCA binding COOLAIR transcript mta mutant wild-type control axis amplicons qPCR Fig Data mean ± s.d. three biological replicates Two-tailed P value multiple t test corrected Holm–Sidak method images root tip nuclei expressing pFCA:FCA-mTurquoise2 plants without mta mutationintensity projections Z-stack nucleus Scale bars 5 μm Quantification FCA-mTurquoise2 condensates root cells plants without mta mutation Data plotted minima to maxima box extends 25th to 75th percentiles line marks median whiskers down minima maxima P value two-tailed t test distribution FCA-mTurquoise2 condensates sizes 0.01 0.02 μm2 plants 59 without 60 mta mutation P = 0.0237 two-sided Fisher’s test ratio proximal-to-distal isoforms COOLAIR transcripts 2d mta C2 wild-type C2. Data mean ± s.d. three replicates DRIP–qPCR COOLAIR R-loop mta mutant wild-type control Data mean ± s.d. three replicates Source data file processed images Fig. 3c d Figshare.m6A influences FCA nuclear condensates liquid FCA compartmentalizes 3′-end processing factors liquid nuclear condensates vitro phase separation processes stress granules enhanced RNAs m6A MTA FIP37 targeted nuclear speckles transient Arabidopsis MTA protein orthologs other plant speciesG. max S. tuberosum disordered contain prion-like domain (PrLD) Fig. 7a MTA orthologs H. sapiens D. melanogaster D. reiro disordered contain PrLDs FCA contains two PrLDs19 hypothesized MTA m6A deposition affect dynamics FCA nuclear condensates association FCA COOLAIR occupancy time FCA 3′-end processing factors COOLAIR production R-loop resolution test in vivo analysis nuclear condensates root nuclei transgenic plants generated plants carrying pFCA::FCA-mTurquoise2 fusion transgene complemented fca-9 mutation crossed embryo-rescued genotype pABI3::MTA reduced m6A levels seedlings wild-type plant roots FCA-mTurquoise2 localized nuclear condensates various sizes (Fig. similar FCA-eGFP fusion After introduction mta mutation FCA-mTurquoise2 foci reduced 5.3 per nucleus to 3.4 larger condensates reduced ~67% to ~60% segregants same cross (Fig.data support m6A modification affects lifetime FCA nuclear condensates promoting component association short timescales vivo change dynamics FCA affects COOLAIR 3′-end processing de-repression FLC m6A diminished reduced use proximal COOLAIR polyadenylation site in mta C2 fca-9 fy-2 COOLAIR R-loop increased in mta data show stabilization R-loop by NDX1 enables FCA RNA processing COOLAIR proximal polyadenylation co-transcriptionally formed R-loop FCA RNA processing condense into nuclear speckles m6A modification enhances condensation interplay relationship m6A R-loop stability global R-loop changes in mta argues against linear regulatory relationship m6A R-loop context important.FCA FY promote R-loop resolution targets genomeFCA regulator flowering widespread roles in Arabidopsis speculated FCA-promoted R-loop resolution not specific FLC observed R-loop accumulation globally in fca-9 analysis FCA proximal polyadenylation Arabidopsis transcriptsread-through transcription fca mutants generated spliced transcripts unannotated segments -UA loci FCA 3′-end processing R-loop regulation analyzed R-loops at UA loci R-loops formed near polyadenylation sites UA2 UA10B XRCC4 increased in fca-9 tested in fy-2 increased at UA10B XRCC4 slightly at UA2 suggest FCA-bound transcripts associated with R-loop formation FCA/FY promotion 3′-end processing R-loop resolution FCA targets Arabidopsis genome FY promote R-loop resolution genome from DNA damage replication DRIP–qPCR R-loops at UA2 UA10B XRCC4 in Col-0 fca-9 with without RNase H treatment Data mean from three replicates DRIP–qPCR UA R-loops in Col-0 fy-2 mean two replicates seedlings treated with without bleomycin plants with without leaves scored Data number seedlings two categories FCA associates with co-transcriptionally formed R-loops Pol II stallingFCA deposition m6A transcript enhances interactions nuclear condensates promotes polyadenylation at noncanonical sites R-loop FLC linked to chromatin silencing 3′-end processing factors K4 demethylase FLD Without FCA FY m6A R-loops accumulate transcripts polyadenylate at efficient p(A) sites Source data Fig. 4a–c analysis importance RNA 3′-end processing COOLAIR-generated R-loop chromatin silencing Arabidopsis floral repressor locus FLC simple recruitment RNA FCA COOLAIR transcript unlikely mechanism involves dynamic nuclear condensates interactions feedbacks m6A methylation condensates produce high concentrations RNA-processing factors terminate transcription R-loop proximally polyadenylated COOLAIR transcript local chromatin environment stalled RNA R-loop biochemical unwinding DNA/RNA helicases Table 1 Fig. FLC 3′-end processing factors noncanonical polyadenylation link to recruitment chromatin modifiers FCA roles in RNA-mediated chromatin silencing Arabidopsis genome parallels-transcriptional silencing mechanisms S.noncanonical polyadenylation mechanism tool for chromatin “tangles” R-loops (Fig. damaging to genome DNA replication12 function hypersensitivity fca fy mutants to drug Bleomycin evidenced reduction cell division in mutant wild-type seedlings after drug treatment (Fig. 4c 11a work provides insights interplay between co-transcriptionally R-loops-processing factors m6A RNA modification chromatin modification.MethodsPlant materials growth conditionsMutant alleles ndx1-411 fca-951 fld-48 fy-251 described C2 line mta Dr Kamil Ruzicka Experimental Botany Dr Rupert G. Fray Nottingham fip37 Dr. Hao Yu (National University of Singapore pFCA: FCA-mTurquoise2 transgenic line FCA genomic DNA amplified inserted into pCambia1300 vector FCA ApaI fragment swapped by FCA ApaI fragment fused with mTurquoise2 sequence inserted before stop codon construct transformed into fca-9 mutantSeedlings sterilized sown stratified 4 °C 3 days-day (16-h 20 8-h 16 2-g 10-day seedlings harvested grounded powder suspended 30 mL Honda buffer (20 HEPES 0.44 sucrose 1.25% Ficoll 2.5% dextran 10 MgCl2 0.5% Triton x-100 5 DTT protease inhibitor filtered Miracloth centrifuged 3500×g 15 min pellets resuspended 1 mL Honda buffer centrifuged 8000×g 1 min resuspended lysis buffer (50 mM Tris-HCl pH 8.0 10 EDTA 1% SDS 0.1 mg/mL proteinase K digested 55 °C overnight/chloroform extracted DNA precipitation NaOAc isopropanol pellet dissolved water Proteinase K 2 h quantified Qubit DNA quantification kit 1 μg DNA dissolved 50 μl buffer-HCl EDTA 0.5% sonicated Diagenode Bioruptor® 15 times diluted ten times buffer (16.7 mM Tris pH 7.5 167 NaCl 2.2 mM EDTA 0.1% Triton X-100 stored −20 °C5 μg S9.6 antibody (1:100 dilution Kerafast added incubated overnight 4 °C 50 μl Protein G Agarose incubated 2 h immunoprecipitants washed five times buffer eluted 200 μl buffer (10 pH 7.5 2 mM EDTA 0.2% SDS 100 ng/μl tRNA 55 °C 1 h samples nucleic acids precipitated NaOAc isopropanol glycogen dissolved water qPCR analysis LightCycler480 data normalized 1% input Primers Supplementary Table 2. H-treated sample DNA treated overnight 37 °C 1 h DNA extracted sonicated RNase H 2 μl DNA blotted HybondTM-N + membrane cross-linked UV mJ/cm2) blocked 5% milk TBST buffer 1 h 0.5 μg/ml S9.6 antibody (1:2000 dilution ENH001 Kerafast incubated overnight 4 °C mouse IgG horseradish peroxidase antibody (1:20,000 dilution NA931 Healthcare chemiluminescence detectionstained 0.1% Methylene Blue Safe DNA Gel Stain 2-g 10-day seedlings harvested cross-linked 1% formaldehyde suspended 30 mL Honda buffer 50 ng/μl tRNA 20 U RNase filtered Miracloth centrifuged 3500 5 min pellets resuspended 2.5 volumes Nuclei Lysis Buffer (50-HCl 8.0 10 mM EDTA 1% SDS 1× protease inhibitor 50 ng/μl tRNA sonicated Diagenode Bioruptor® five times Immunoprecipitation 30 μl DynabeadsTM protein A FCA antibody 1.2 mL chromatin 4 °C 1.5 h beads washed four times buffer mM NaCl 16.7 Tris pH 7.5 1.2 mM EDTA 0.8% Triton X-100 1× protease inhibitor 50/μl tRNA Reverse cross-linking elution 200 μl buffer (2 mM EDTA 0.2% SDS 100 ng/μl tRNA U/μl beads 55 °C overnightRNA supernatant precipitated isopropanol glycogen dissolved DNase treated template reverse transcription gene-specific primers Data presented IP/1% input RT controls RNA DNA contamination UBC negative control Primers Supplementary Table performed Protein A magnetic beads anti-H3 (3 anti-H3K4me1 (6 used immunoprecipitation recovered DNA quantified qPCR primers Data normalized to input values ratio H3K4me1 to H3 analysisTen-day-old seedlings harvested RNA extracted fip37 harvested 14 days after germination homozygous mutants treated TURBO DNase reverse-transcribed SuperScript IV Reverse Transcriptase (Invitrogen) gene-specific primers qPCR analysis data normalized to UBC Primers Supplementary Table 2.Western blot protein extracts separated NuPAGE 4–12% Bis-Tris Gels transferred to 0.45-μm PVDF membrane membrane blocked in 5% milk-buffered saline 0.1% Tween-20 1 h room temperature Anti-FCA anti-H3 antibody added incubation 4 °C overnightwashes membrane incubated IgG horseradish peroxidase antibody (1:10,000 dilution chemiluminescence detection Protein loading Coomassie Blue staining expression purification recombinant sequence N-terminal FCA amplified inserted pGEX-6P-1 vector transformed cells. BL21DE3) grown broth 37 °C 6 h protein expression 3 h 30 °C 1 mM IPTG GST-tagged protein purified Glutathione Sepharose® Fast Flow vitro RIPGST-tagged FCA protein expressed purified FTO protein m6A demethylation expressed purified extraction nuclei digested protease K 55 °C 2 h extracted/chloroform digestion 37 °C 2 h phenol extraction m6A 3 μg RNA treated 1 nmol purified FTO protein buffer (50 mM HEPES pH 7.0 300 μM Fe2+ 500 μM α-ketoglutaric acid 2 mM L-ascorbic acid 0.1 U/μl 0.5 U/μl RiboLock 37 °C 1 hcontrol sample 3 μg RNA 1 nmol FTO 50 mM HEPES pH 7.0 5 mM EDTA 0.1 0.5 RiboLock 37 °C 1 h m6A/A concentration detected LC-MS vitro RIP 150 pM protein 1.5 μg RNA mixed 20 mM Tris-HCl pH 7.5 150 mM NaCl 0.1% NP-40 0.5 RiboLock 4 °C 2 h Glutathione Sepharose® blocked 1% BSA 10 μg/ml yeast tRNA salmon sperm DNA 2 h 100 μl pre-blocked sepharose 4 °C 2 h washed buffer five times Immunoprecipitated RNA extracted Trizol reverse-transcribed pulldown assay FCA Dynabeads Protein A FCA antibody Glutathione Sepharose RNA RNA samples digested 0.5 U nuclease P1 50 10 mM ammonium acetate 5.3 42 °C 3 h 5 μl NH4HCO3 0.5 U shrimp alkaline phosphatase incubated 37 °C overnightnucleosides separated UPLC detected Triple QuadTM 5500 mass spectrometer quantified mass transitions m/z 268.0 to 136.0 (A), 282.0 to 150.1 Concentrations calculated signal intensities curves m6A/A ratios-binding region RNA-modification immunoprecipitation salmon sperm DNA yeast tRNA 1% sonicated chromatin saved input sample half washed beads eluted RNA FCA–RIP sample half treated with RNase T1 (1 U/μl 20 μl buffer 20 mM Tris-HCl pH 7.5 2.5 mM EDTA 1× protease inhibitor cocktail 37 °C 10 min beads washed eluted FCA-binding region sample supernatant RNase T1 digested sample m6A/A concentration ratio detected LC-MS/MS immunoprecipitationNuclear pellets isolated sonicated RNA supernatant extracted with TRIzol reagent treated TURBO DNase 5% extracted RNA saved input sample10 μg RNA incubated 5 μg m6A antibody (202003 Synaptic Systems 400 μl IP buffer (20 mM Tris-HCl pH 7.5 150 mM NaCl 5 mM EDTA 0.1% NP-40 1 × Ribolock RNase inhibitor 4 °C 2 h 35 μl Dynabeads Protein A washed three times incubated 2 h washed five times DNA LoBind tube immunoprecipitated RNA eluted 100 μl buffer 10 mM N6-methyladenosine 4 °C 1 h precipitated isopropanol glycogen reverse-transcribed quantified qPCR Primers Supplementary Table 2.Microscopy analysisSeedlings grown 1⁄2 MS plate 1% sucrose 0.5% 7 days roots water FCA-mTurquoise2 signal captured Zeiss LSM780 confocal microscope 40 × /1.2 mTurquoise2 458 nm detected 463–579 nm z-stack projection “StackReg” Plugin drifted condensates nucleus obtained manually FigshareNuclei extracted from root for quantification auto-fluorescence by ImageJ software condensates “Analyze Particles” tool data condensate settings data processed images from Figshare.GenotypingC219 mta33 fip3734 plants genotyped fca-9 mutation DNA amplified with primers fca-9-dCAPS_F + R PCR products digested with StyI 3% agarose gel electrophoresis ndx1-4 genotyped with transfer-DNA primer p745 ndx1-4_GT_R PCR using ndx1-4_GT_F + R-DNA insertion Primers in Supplementary Table pFCA::FCA-mTurquoise2 seeds of homozygous plants + background obtained After 7-day-old seedlings root tips mounted in water for imaging analysis seedlings grown 7 days for DNA extraction genotyping seven in mta−/− background seven same genetic background (MTA + / + ) analyzed.Bleomycin treatmentFour-day-old seedlings transferred to Petri dishes 10 ml liquid growth medium 1 μg/ml bleomycin5 days incubation seedlings floating liquid no shaking illuminated growth chamber (9-day-old remove medium wash plate five times 20 ml liquid media Transfer seedlings solid plates recover 24 h analysis Plants scored leaf emergence Supplementary Fig. 11a b).Statistical GraphPad Prism 8.4.3. P value sample number adjusted P value Figures legends Nature Research Reporting Summary.Supplementary Review Summary | 49.8 | 0.733357 |
10.1038/s41467-020-18001-x | PMC7452890 | Backbone extended monomers are poorly compatible with the natural ribosomes, impeding their polymerization into polypeptides. Here the authors design non-canonical amino acid analogs with cyclic structures or extended carbon chains and used an engineered ribosome to improve tRNA-charging and incorporation into peptides. | Ribosome-mediated polymerization of backbone-extended monomers into polypeptides is challenging due to their poor compatibility with the translation apparatus, which evolved to use α-L-amino acids. Moreover, mechanisms to acylate (or charge) these monomers to transfer RNAs (tRNAs) to make aminoacyl-tRNA substrates is a bottleneck. Here, we rationally design non-canonical amino acid analogs with extended carbon chains (γ-, δ-, ε-, and ζ-) or cyclic structures (cyclobutane, cyclopentane, and cyclohexane) to improve tRNA charging. We then demonstrate site-specific incorporation of these non-canonical, backbone-extended monomers at the N- and C- terminus of peptides using wild-type and engineered ribosomes. This work expands the scope of ribosome-mediated polymerization, setting the stage for new medicines and materials. | IntroductionThe cellular translation system (the ribosome and associated factors for protein biosynthesis) catalyzes the synthesis of sequence-defined polymers (polypeptides) using a set of amino-acylated transfer RNA (tRNA) substrates and a defined coding template (messenger RNA). In nature, only a limited set of α-l-amino acid monomers are utilized by this system, thereby limiting the potential diversity of polymers that can be synthesized. Over the past two decades, however, efforts to expand the genetic code have shown that the natural translation system is capable of selectively incorporating a wide range of non-canonical monomers1–5. These monomers include α-6, β-7–9, γ-10–12, D-13,14, aromatic15–17, aliphatic15,18, malonyl16, N-alkylated19, and oligomeric amino acid analogs10,20,21, among others (Fig. 1a).Fig. 1Expanding the chemical substrate scope of the translation apparatus to include long chain carbon and cyclic amino acids.a Substrates for translation compatible with the flexizyme (Fx) and cell-free protein synthesis (CFPS) platforms. Long chain carbon (lcc) amino acid incorporation into peptides has proved challenging. b Examples of prominent polyamide polymers that possess significantly different properties, such as tensile strength (TS), based on backbone length, monomer functionality, and/or monomer sequence. c tRNA charging of lcc amino acids by the Fx system has remained challenging due to the resulting intramolecular lactam formation. d Strategy for incorporation of long chain carbon amino acids via Fx and in vitro translation.Site-specific incorporation of such diverse chemistries into peptides and proteins has led to a wave of exciting applications. For example, foldamers incorporated into the N-terminus of a peptide have created macrocyclic foldamer–peptide hybrids with unique bioactivity22. In addition, benzoic acids and 1,3-dicarbonyl substrates have been incorporated into diverse aramid–peptide and polyketide–peptide hybrid molecules15,16, which may enable new classes of functional materials and polyketide natural products. Furthermore, β-amino acid peptides have made possible new protease resistant, peptidomimetic drugs23–27.Having access to an even broader repertoire of monomers for ribosome-mediated polymerization holds promise to further increase the number of polymers that could be synthesized in a sequence-defined manner, which has been called the next “Holy Grail” of polymer science28. For example, polyamides (outside of polypeptides) make use of a key set of privileged molecular architectures to obtain exceptional polymer properties, such as improved thermal stability, elastic modulus, and tensile strength, based on polymer backbone and chain microstructure (i.e., Nylon-6 versus Kevlar29,30, Fig. 1b). The ability to introduce these architectures into polypeptides and modulate their properties could open new opportunities at the intersection of materials science and synthetic biology. However, direct incorporation of these monomers—such as long chain carbon amino acids (≥γ-)—has proved challenging for two key reasons. First, natural ribosomes have been evolutionarily optimized to polymerize α-l-amino acids, leading to poor compatibility with backbone-extended monomers. Second, acylating (or charging) these monomers to tRNAs to make aminoacyl-tRNA substrates is difficult. Chemical aminoacylation is technically difficult and laborious, aminoacyl-tRNA synthetases have not been evolved for these long chain carbon monomers, and efforts to use the flexizyme system (Fx, an aminoacyl tRNA synthetase-like ribozyme)23,31 have been unsuccessful, due to intramolecular lactam formation after the tRNA charging reaction (Fig. 1c)10,12,25,32,33. Taken together, these limitations have restricted the scope of long chain carbon (or backbone-extended) amino acid monomers incorporated into sequence-defined polyamides by the ribosome.Here, we set out to address these limitations by investigating the Fx-catalyzed tRNA charging of γ-, δ-, ε-, and ζ-amino acids containing long chain carbon structures and demonstrating subsequent in vitro incorporation of such amino acid derivatives into peptides by the ribosome. This stands distinct from our recent work to study flexizyme design rules associated with four chemically diverse scaffolds (phenylalanine, benzoic acid, heteroaromatic, and aliphatic monomers) with different electronic and steric factors15. Here, we consider how to avoid intramolecular nucleophilic attack of the monomer amino group of backbone-extended monomers to facilitate tRNA charging. In addition, we focus on long chain carbon and cyclic monomers, which is unique from many works showing the incorporation of a variety of non-canonical α-6 and β-amino acids7,8,25,34. We first confirm through NMR and LC–MS analysis that tRNA charging of linear γ-amino acids via flexizyme fails due to deleterious lactam formation (Fig. 1c and Supplementary Fig. 1). Next, we circumvent this limitation of Fx-catalyzed tRNA-charging by designing amino acid substrate architectures that control the intramolecular reaction kinetics of the tRNA:substrate complex by lengthening the carbon chain and/or introducing a rigid central architecture (Fig. 1d, top panel) such that lactam formation is reduced or avoided altogether. Then, we demonstrate incorporation of backbone-extended monomers into the N-terminus of peptides using wild-type ribosomes. Finally, we use a previously engineered ribosome24,27,34 with mutations in the peptidyl transferase center (PTC) to enable C-terminal incorporation of these non-canonical amino acids into a peptide (Fig. 1d, bottom panel).ResultsLong chain carbon and cyclic amino acid flexizyme chargingTo gain insights about possible constraints for using Fx to charge long chain carbon amino acid substrates onto tRNAs, 10 substrates (1–5 in Fig. 2a and 2i–2v in the characterization section in Supplementary Information) were examined with increasing numbers of carbons in the monomer backbone. Dinitrobenzyl ester (DNB)-derivatized or amino-derivatized benzyl thioester (ABT)-activated forms of 3-aminopropanoic acid (1, β-alanine) and 4-aminobutyric acid (2 and 2i) were synthesized for Fx-mediated charging. We used a tRNA mimic, microhelix tRNA (mihx), to determine the yields of the Fx-mediated acylation reaction using the conventional Fx reaction condition20. Aminoacylation efficiency was estimated by acid-denaturing polyacrylamide gel electrophoresis (PAGE, Supplementary Fig. 1). We found that 1 was successfully charged, while 2 was not as previously reported10,25 (Fig. 2a and Supplementary Fig. 1). We tested four additional γ-amino acid substrates (4-methylaminobutyric acid (2ii) and 2,2-dimethylaminobutyric acid (2iii), cis- (2iv), and trans-2-aminocyclopropane-1-carboxylic acid (2v)) for Fx-mediated tRNA charging, but no γ-amino acid substrates (2 and 2i–v, see characterization section in Supplementary Information) were found to be charged (Supplementary Fig. 1), indicating our results are consistent with previous literature and that Fx-mediated charging of γ-amino acid analogs with a linear carbon chain is indeed challenging.Fig. 2Systematic design of long chain carbon and cyclic amino acids.a The range of amino acids bearing a linear carbon chain was extended to γ-, δ-, ε-, and ζ-amino acids. Higher acylation yields by Fx were observed as the amino acid chain length increased, presumably because larger (>5-membered) ring formation via lactamization is kinetically less favorable than 5-membered ring formation. b Introducing cyclic and rigid bonds into substrates helps increase Fx acylation yields. c An increased acylation yield (from ~6% for 7 up to ~95% for 12) was obtained for the γ-amino acids with a rigid bond (7) or cyclic structure (11–15). These data suggest the rigid carbon scaffold efficiently inhibits the intramolecular 5-membered lactam formation reaction. The acylation yield of each substrate represents the percent yield of a microhelix tRNA observed at 24 h/120 h (see Supplementary Information for gels). Data are representative of three independent experiments.To confirm the hypothesis that lactam formation is the cause of poor tRNA charging results, we next investigated whether a lactam is observed in the Fx-catalyzed reaction. A Fx-catalyzed acylation reaction of 4-methylaminobutyric acid (2ii) with mihx was set up and monitored over 24 h. Notably, analysis by LC–MS of the reaction mixture incubated for 24 h yielded a single new peak (2.3 min, light green, Fig. 3a). The ESI-MS generated by combining mass spectra obtained across the peak at 2.3 min showed an accurate mass corresponding to the theoretical mass of the lactam, 1-methylpyrrolidin-2-one (Fig. 3b). Furthermore, a lactam is only observed when both Fx and mihx are present in the reaction mixture, suggesting that lactam formation is catalyzed by these species.Fig. 3Observation of lactam in Fx-mediated acylation of γ-amino acid.The lactam produced (light green) in the Fx-mediated acylation of substrate 2ii is observed. The extracted ion chromatogram a for the mixture of Fx reaction incubated for 24 h on ice showed a new peak corresponding to a theoretical mass of a lactam b. Data are representative of three independent experiments.Next, we synthesized long chain carbon derivatives 5-aminopentanoic acid (3), 6-aminohexanoic acid (4), and 7-aminoheptanoic acid (5), to further support our hypothesis that acylation yields of tRNA would increase because the formation of larger rings (>5-membered) is less kinetically favorable than 5-membered ring formation. As expected, we observed higher acylation yields for increasing lengths of the carbon chain in the amino acid derivatives (Fig. 2a and Supplementary Fig. 1), further suggesting the deficiency of linear γ-amino acids in the genetic code reprogramming is due to the propensity for lactam formation amongst these substrates using Fx-mediated catalysis. Of note, this result is in a good agreement with a general rule for ring closure reactions35,36 that shows the rate constant for the 5-membered ring self-cyclization is the largest. The rate constant decreases by 1–2 orders of magnitude (i.e., self-cyclization slows) as the ring size increases from 5-members to 10-members35.Based on these results, we sought to design molecular architectures that would circumvent intramolecular lactam formation by steric restriction of the amino and activated ester functionalities. We synthesized five substrates (6–10 in Fig. 2b) containing a rigid spacer (cyclic, aryl, or vinyl) and tested acylation. Notably, all of the substrates (6–10), which are γ-amino acid and δ-amino acid, were successfully charged to tRNA using flexizymes. To further expand the range of the monomers for diverse polyamides, we synthesized five additional amino acids (11–15 in Fig. 2c) containing a cyclic structure in the central region of amino acid. When these substrates were charged to tRNAs, we found the acylation yield was dramatically increased compared to the other γ-type amino acids, suggesting that the rigid cyclic carbon scaffold efficiently prevents the intramolecular 5-membered lactam formation reaction. This observation is consistent with our recently described design rules for flexizyme-catalyzed acylation15, as well as another recent report that showed incorporation of cyclic-gamma amino acids into peptides12. In short, the cyclic structures contain less steric hindrance about the carbonyl relative to structures (1–5) and increased electrophilicity relative to the conjugated structures (6–8) allowing for efficient tRNA attack15. Overall, we found that 13 non-canonical monomers were charged, with efficiencies of 6–95%, with (E)-4-aminobut-2-enoic acid (7) as the lowest and trans-3-aminocyclobutane-1-carboxylic acid (12) as the highest yield, respectively.Ribosomal polymerization of backbone-extended monomersNext, we investigated whether the newly found flexizyme substrates charged to tRNAs are accepted by the natural protein translation machinery. The goal was to demonstrate that the ribosome was compatible with these substrates, rather than focus on a specific application. We performed the Fx-catalyzed acylation reaction for tRNAs under the same reaction conditions obtained from the acylation reaction of mihx (Supplementary Fig. 1). Previous works have shown that the acylation reaction yield and kinetics between in vitro-transcribed tRNA mimics (e.g., mihx or microhelix) and tRNAs are comparable37–41. After the Fx-mediated tRNA acylation, unreacted monomers were separated from the tRNAs using ethanol precipitation20 and the resulting tRNA fraction that includes the tRNA-substrates was supplemented as a mixture into a cell-free protein synthesis42 reaction containing a minimal set of components required for protein translation (PURExpressTM)43. We then determined incorporation of the non-canonical substrates into either the N- or C-terminus of a small model Streptavidin tag by MALDI mass spectrometry.As the initiator tRNA, tRNAfMet was selected for N-terminal incorporation studies. For C-terminal incorporation, we assessed several tRNAs (fMet, Pro1E2, GluE2, and AsnE2)44 previously engineered to efficiently incorporate non-canonical amino acids into polypeptides by the ribosome. We observed no significant difference in incorporation efficiency, depending on the codon variations. As such, Pro1E244 was selected because it has an engineered D-arm and T-stem interacting with other protein translation factors such as EF-Tu and EF-P that can be additionally supplemented into the cell-free translation reaction when it is necessary to promote the incorporation of charged substrate8,25,45. For the codons, we used AUG (CAU anticodon), as it is the canonical start codon for N-terminal incorporation. For C-terminal incorporation, we selected the ACC codon (GGU anticodon), which decodes the Thr(ACC) codon on mRNA. This was selected because threonine is excluded from the polypeptide Streptavidin tag (WSHPQFEK) that was used for our study. This prevented corresponding endogenous tRNAs in the PURExpressTM reaction from being aminoacylated and used in the translation reaction.We charged all 14 substrates onto tRNAfMet (CAU) and tRNAPro1E2 (GGU) to yield a set of acylated tRNAs, which were subsequently used in the PURExpressTM translation reaction. The PURExpressTM reaction was carried out in the presence of all Escherichia coli (>46) endogenous tRNAs, but only nine amino acids encoding the polypeptide Streptavidin tag (WSHPQFEK) and the non-canonical aminoacyl-tRNA substrate were used. Two different sets of amino acids (X + WSHPQFEK + T and M + WSHPQFEK + X) were used for the N- and C-terminus incorporation, respectively, where X indicates the position to which a Fx-charged backbone extended monomer is incorporated (Fig. 2a, see Supplementary Information in detail). Following translation (Fig. 4a), we found that every substrate that could be charged onto tRNAs was successfully incorporated into a peptide at the N-terminus, confirmed by a peak corresponding to a theoretical mass of peptide in MALDI spectra (Fig. 4b–n). However, attempts to produce a peptide containing these amino acids at the C-terminus were unsuccessful (Fig. 5a–c and Supplementary Fig. 2b, e). This is presumably because the C-terminal incorporation forming an amide bond with a nascent peptide requires more precise alignment of substrate in the PTC46 and the wild-type ribosome is not efficient at incorporation of non-canonical, backbone-extended substrates into polypeptides.Fig. 4Ribosomal synthesis of N-terminal functionalized peptides with backbone-extended monomers.a All backbone-extended amino acids (3–15) charged to tRNAfMet(CAU) by Fx were incorporated into the N-terminus of a peptide by ribosome-mediated polymerization in the PURExpressTM system. The peptides were purified via the Streptavidin tag (WSHPQFEK) and characterized by MALDI mass spectrometry. The observed mass of each peptide corresponds to the theoretical mass, which is b [M + H]+ = 1345; [M + Na]+ = 1367, c [M + H]+ = 1359; [M + Na]+ = 1381, d [M + H]+ = 1373; [M + Na]+ = 1395, e [M + H]+ = 1369; [M + Na]+ = 1391, f [M + Na]+ = 1351, g [M + H]+ = 1379; [M + Na]+ = 1401, h [M + H]+ = 1371; [M + Na]+ = 1393, i [M + H]+ = 1372; [M + Na]+ = 1394, j [M + H]+ = 1343; [M + Na]+ = 1365, k [M + Na]+ = 1365, l [M + H]+ = 1357; [M + Na]+ = 1379, m [M + H]+ = 1371; [M + Na]+ = 1393, n [M + H]+ = 1371; [M + Na]+ = 1393. The peaks denoted with an asterisk are a truncated peptide not bearing the target substrate at the N-terminus ([M + H]+ = 1246; [M + Na]+ = 1268). Data are representative of three independent experiments.Fig. 5Ribosomal synthesis of peptides with aminocyclobutane-carboxylic acid (ACB).a Peptides were synthesized in the PURExpressTM system using Fx-mediated tRNAPro1E2(GGU), purified via the Streptavidin tag, and characterized by MALDI mass spectrometry. b and c
cis-ACB and trans-ACB are not incorporated into the C-terminus of a peptide by the wild-type ribosome. d Engineered ribosomes facilitate C-terminal and mid-chain incorporation of cis/trans-ACB into peptides. e and f
cis-ACB and trans-ACB. Peptides containing cis/trans-ACB at the C-terminus were observed when an engineered ribosome, developed by Maini et al.24, 58, was added into the protein translation reaction in vitro. g and h
cis and trans-ACB. Additional amino acid residues (Ile and Ala) were elongated after the incorporation of cis/trans-ACB, demonstrating that the engineered ribosome enabled site-specific incorporation of ACB. Data are representative of three independent experiments. See Supplementary Fig. 2 for full spectrum.Engineered ribosomes enhance incorporation of novel monomersRecently, advances by the Hecht group showed that an engineered ribosome (termed 040329) enabled incorporation of dipeptides into a growing polymer chain by the ribosome24,27 in vivo and in vitro, where the ribosome forms an amide bond with the nascent peptide using the far-distance amine of a substrate. We hypothesized that this engineered ribosome would also be more permissive towards the backbone-extended monomers described here. To test this, we co-expressed the mutant ribosomes in cells using previously established protocols47 (see Supplementary Information for details). From these cells, we lysed and purified ribosomes through ultracentrifugation on a sucrose cushion (see Supplementary Information for details). The resulting ribosome sample contained a mixture of wild type and 040329 ribosomes, which were subsequently used in translation assays to determine their activity towards elongated backbone monomers. Based on previous literature, we expected the 040329 ribosomes to constitute around 25% of the purified ribosome population. To test the feasibility of incorporating long chain carbon amino acids into peptides with engineered ribosomes, we added the ribosome mixtures (Fig. 5d) into the PURExpressTM system containing the substrates charged to tRNAPro1E2(GGU) by Fx. In our MALDI mass spectrum, we observed a peak corresponding to the theoretical mass of the target peptide containing cis- and trans-3-aminocyclobutane-1-carboxylic acids (ACB, 11 and 12, from Fig. 2, respectively) at the C-terminus (fMWSHPQFEKS11/12 in Fig. 5e, f, and Supplementary Fig. 2c, f), which was not observed in the experiments with the wild-type ribosome alone (Fig. 5b, c, and Supplementary Fig. 2b, e). The relative percent yields of the target peptide containing cis and trans-ACB at the C-terminus were approximately 11% and 15%, respectively, based on the total of full-length and truncated peptide products (fMWSHPQFE, fMWSHPQFEK, and fMWSHPQFEKS, Supplementary Fig. 2).We finally investigated whether additional amino acids can be elongated after the incorporation of cis-ACB and trans-ACB (11 and 12, Fig. 5g, h, and Supplementary Fig. 2d, g) at the C-terminus. We designed a new plasmid that encodes two additional amino acid residues, Ile (AUC) and Ala (GCC), and performed a PURExpressTM reaction under the same reaction conditions, using a new set of 11 amino acids (M + WSHPQFEK + X + IA). While inefficient, we observed peaks corresponding to the theoretical mass of the target peptides (fMWSHPQFEKS11/12IA), demonstrating the engineered ribosome is capable of continuing to elongate following insertion of cis-ACB and trans-ACB.DiscussionIn this work, we expanded the range of backbone-extended amino acid substrates for molecular translation. To do so, we investigated mechanistic aspects that limit the acylation step of γ-amino acids onto tRNAs by Fx. Then, through systematic and rational substrate design, we showed that a diverse repertoire of 15 amino acids with long chain carbon and cyclic structures could be acylated to tRNA by the Fx system in yields of 6–95%. Next, we demonstrated that these charged acylated tRNA-monomers could be used in ribosome-mediated polymerization, expanding the diversity of polyamides that can be produced by ribosomal synthesis.While the field of genetic code expansion has incorporated hundreds of α-based non-canonical amino acids, until now, it was not known if the ribosome was capable of incorporating the backbone-extended (γ-, δ-, ε-, and ζ-) and cyclic (cyclobutane, cyclopentane, and cyclohexane) amino acid-based structures presented here. Our work shows that the ribosome is capable of polymerizing such structures using the genetic code reprogramming approach. Not surprisingly, the efficiency of incorporation, especially at the C-terminus or mid-chain, is low. This is likely because the shape, physiochemical, and dynamic properties of the ribosome have been evolved to work with canonical α-amino acids, or in the case of the modified ribosome 040329, β-amino acids34. It is likely that wild-type and 040329 ribosomes still discriminate against the backbone-extended stereoisomer monomers introduced here. Looking forward, the incorporation efficiency of such substrates could be improved by supplementing the combination of EF-P and engineered tRNAs8,12,48. In addition, in vitro ribosome assembly49 and selection50 platforms could evolve ribosomes with altered properties that increase incorporation efficiency of the backbone-extended monomers into peptides (i.e., form less truncated products) and facilitate the synthesis of polymers comprised solely of such monomers. Finally, extension to cellular systems with orthogonal engineered tethered, or stapled, ribosomes51–55 offers another exciting direction. However, the lack of aminoacyl tRNA-synthetases (aaRS) that charge the monomers into tRNA in the cell will need to be addressed.By expanding the scope of long chain carbon and cyclic amino acids available for use in ribosome-mediated polymerization, we expect this work to motivate new directions in efforts to synthesize non-canonical sequence-defined polymers. For example, the monomers shown here could be directly used with in vitro screening and selection methods like mRNA or ribosome display to discover innovative peptide drugs56. In addition, future works could enable unique functional materials and polymers of defined atomic sequence, exact monodisperse length, and programmed stereochemistry.MethodsGeneral Fx-mediated acylation reactionMicrohelix acylation: 1 μL of 0.5 M HEPES (pH 7.5) or bicine (pH 8.8), 1 μL of 10 μM microhelix, and 3 μL of nuclease-free water were mixed in a PCR tube with 1 μL of 10 μM eFx, dFx, and aFx, respectively. The mixture was heated for 2 min at 95 °C and cooled down to room temperature over 5 min. 2 μL of 300 mM MgCl2 was added to the cooled mixture and incubated for 5 min at room temperature. Followed by the incubation of the reaction mixture on ice for 2 min, 2 μL of 25 mM activated ester substrate in DMSO was then added to the reaction mixture. The reaction mixture was further incubated for 16–120 h on ice in cold room.tRNA acylation: 2 μL of 0.5 M HEPES (pH 7.5) or bicine (pH 8.8), 2 μL of 250 μM tRNA, 2 μL of 250 μM of a Fx selected on the microhelix experiment and 6 μL of nuclease-free water were mixed in a PCR tube. The mixture was heated for 2 min at 95 °C and cooled down to room temperature over 5 min. 4 μL of 300 mM MgCl2 was added to the cooled mixture and incubated for 5 min at room temperature. Followed by the incubation of the reaction mixture on ice for 2 min, 4 μL of 25 mM activated ester substrate in DMSO was then added to the reaction mixture. The reaction mixture was further incubated under the optimal reaction conditions determined by the microhelix experiment.In vitro synthesis of polyamidesN-terminus incorporation: As a reporter peptide, a T7 promoter-controlled DNA template (pJL1_StrepII) was designed to encode a streptavidin (Strep) tag and additional Ser and Thr codons (XWSHPQFEKST (Strep tag), where X indicates the position of the non-canonical amino acid substrate). The translation initiation codon AUG was used for N-terminal incorporation of the non-canonical amino acid substrate, X. Peptide synthesis was performed using only the 9 amino acids that decode the initiation codon AUG and the purification tag in the absence of the other 11 amino acids to prevent corresponding endogenous tRNAs from being aminoacylated and used in translation. The PURExpressTM Δ (aa, tRNA) kit (NEB, E6840S) was used for polyamide synthesis reaction and the reaction mixtures were incubated at 37 °C for 3 h. The synthesized peptides were then purified using Strep-Tactin®-coated magnetic beads (IBA), denatured with SDS, and characterized by MALDI-TOF mass spectroscopy.C-terminus incorporation: The same plasmid (pJL1-StrepII) encoding the same amino acids (MWSHPQFEKSX, where X indicates the position of the cyclic amino acid) was used for C-terminal incorporation and the cyclic amino acid was incorporated into the Thr codon (ACC) using a custom-made PURExpress® Δ (aa, tRNA, ribosome) kit (NEB, E3315Z). For C-terminal incorporation, the wildtype ribosome provided in the kit was not used. 15 μM (final concentration) of the engineered ribosome was added to the reaction mixture only containing the 9 amino acids that decode the Strep tag and incubated at 37 °C for 3 h.Central-position incorporation: A plasmid (pJL1-StrepII_TIA) designed to encode additional Ile and Ala downstream of Thr was used (see plasmid map for details) for incorporation of cyclic amino acid into the middle position of polyamide (MWSHPQFEKSXIA, where X indicates the position of the cyclic amino acid). The polyamide was produced using 11 amino acids in the PURExpressTM Δ (aa, tRNA, ribosome) kit under the same reaction conditions used for C-terminal incorporation.Purification and characterization of polyamidesThe polyamides containing a non-canonical amino acid were purified using an affinity tag purification technique and characterized by MALDI spectrometry as previously described15. For sample preparation, 1.5 μL of the purified peptide (0.1% SDS in water) was dried with 0.5 μL of the matrix (α-cyano-4-hydroxycinnamic acid in THF, 10 mg/mL). The dried sample was characterized on a Bruker rapifleX MALDI-TOF and processed using FlexControl v2.0 software (Bruker).Preparation of the cells containing 040329 ribosomesA plasmid containing the rrnB operon under the pL promoter (pAM552) was used as the template for generating a modified rrnB gene with mutations 2057AGCGTGA2063 and 2502TGGCAG2507 in the 23S rDNA, referred to as the 040329 mutation. Plasmids harboring either the wild type (WT) or modified (040329) rrnB genes were transformed into POP2136 using electroporation and plated on LB-agar with 100 μg/mL of carbenicillin. The plates were incubated for 16–18 h at 30 °C (POP2136 harbors the cI repressor and thus represses expression of rRNA when grown at 30 °C). A single colony from the plate was used to inoculate 25 mL of LB-Miller containing 100 μg/mL of carbenicillin and the culture was grown for 16–18 h at 30 °C. When the culture had reached saturation, a 2 L culture of 2X YTP with 100 μg/mL of carbenicillin was pre-warmed to 42 °C, and inoculated with 20 mL of the overnight culture. Growth at 42 °C disrupts repression of the pL promoter and thus induces expression of the rrnB operon, which encodes for the 040329 mutant rRNA. Previous studies suggest the resulting ribosome population contains up to 20% of plasmid-encoded ribosomes. Optical density was measured regularly (every hour, then 15–30 min when close to the target OD) until the culture reached an OD between 0.4 and 0.6. Then, the cultures were pelleted via centrifugation at 8000 × g for 10 min. The resulting cell pellet was resuspended in Buffer A (see below for composition), and centrifuged again at 8000 × g for 10 min. Resuspension and centrifugation were repeated two more times for a total of three washes. After the final centrifugation, the cell pellet was flash frozen in liquid nitrogen and stored at −80 °C until further processing.Purification of ribosome mixturesFrozen cell pellets were resuspended in Buffer A at a specified ratio (5 mL of Buffer A per 1 g of cell pellet) and lysed using homogenization at 20,000–25,000 psi. The resulting solution was centrifuged at 12,000 × g for 10 min to obtain clarified lysate. The clarified lysate was then layered onto a sucrose cushion at an even volumetric ratio (1 mL of cell lysate per 1 mL of Buffer B (see below for composition)) and ultracentrifuged at 90,000 × g for 18 h. This yielded a pellet on the bottom of the ultracentrifuge tube that contained ribosomes. The ribosome mixture was resuspended with Buffer C (see below for composition) with gentle shaking at 4 °C for 4–8 h, then diluted to obtain a concentration of 20–25 μM of ribosomes measured by absorbance at 260 nm on a spectrophotometer (1 A260 unit = 4.17 × 10−5 μM ribosomes). After complete resuspension and dilution, samples were aliquoted and flash frozen with liquid nitrogen, and stored at −80 °C until use in PURE reactions. Although further purification methods such as sucrose gradients could have been performed, the decision was made to use the crude mixture to maximize the absolute number of mutant ribosomes present in the ribosome mixture. *Reagents used—Buffer A: 20 mM Tris–HCl (pH 7.2), 100 mM NH4Cl, 10 mM MgCl2, 0.5 mM EDTA, 2 mM DTT; Buffer B: 20 mM Tris–HCl (pH 7.2), 500 mM NH4Cl, 10 mM MgCl2, 0.5 mM EDTA, 2 mM DTT, 37.7% (v/v) sucrose; Buffer C: 10 mM Tris–OAc, (pH 7.5), 500 mM NH4Cl, 7.5 mM Mg(OAc)2, 0.5 mM EDTA, 2 mM DTT. Oligos used for construction of 040329 ribosome plasmid: (1) To generate insert: 5′-AGTGTACCCGCGGCAAGACGAGCGTGACCCGTGAACCTTTACTATAGCTTGA-3′ and 5′-GCCCCAGGATGTGATGAGCCCTGCCAGAGGTGCCAAACACCGCCGTC-3′, 2) To generate backbone: 5′-GGCTCATCACATCCTGGGGCTG-3′ and 5′-CGTCTTGCCGCGGGTACACT-3′. Resulting PCR products were assembled together using isothermal DNA assembly57.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationReporting Summary | nature communications | [
"Article"
] | [
"Synthetic biology",
"Peptides",
"Protein design"
] | cellular translation system catalyzes synthesis sequence-defined polymers amino-acylated transfer RNA substrates coding template limited α-l-amino acid monomers diversity polymers genetic code translation system non-canonical include α-6 β-7–9 γ-10–12 D-13,14 aromatic15–17 aliphatic15 malonyl16 N-alkylated19 oligomeric amino acid analogs10 (Fig. 1Expanding chemical substrate scope long chain carbon amino acids Substrates translation compatible flexizyme) cell-free protein synthesis) platforms Long chain carbon) amino acid incorporation into peptides challenging polyamide polymers different properties strength backbone length monomer functionality sequence tRNA charging lcc amino acids Fx system challenging intramolecular lactam formation Strategy incorporation long chain carbon amino acids via Fx in vitro translation incorporation diverse chemistries into peptides proteins led exciting applications foldamers into N-terminus peptide created macrocyclic foldamer–peptide hybrids unique bioactivity22benzoic acids 1,3-dicarbonyl substrates incorporated into aramid–peptide polyketide–peptide hybrid molecules15 new functional materials polyketide products β-amino acid peptides new protease resistant peptidomimetic access to broader repertoire monomers for ribosome-mediated polymerization polymers next “Holy Grail” of polymer polyamides privileged molecular architectures exceptional polymer properties improved thermal stability elastic modulus tensile strength Nylon-6 versus Kevlar29 introduce into polypeptides modulate properties opportunities materials science synthetic biology direct incorporation of long chain carbon amino acids challenging natural ribosomes optimized to polymerize α-l-amino acids poor compatibility with backbone-extended monomers acylating monomers to tRNAs aminoacyl-tRNA substrates difficult Chemical aminoacylation difficult laborious aminoacyl-tRNA synthetases not evolved for carbon monomers efforts flexizyme system unsuccessful due to intramolecular lactam formation after tRNA charging reactionlimitations restricted long chain carbon amino acid monomers into polyamides ribosome limitations investigating Fx-catalyzed tRNA charging of γ δ ε- ζ-amino acids long chain carbon structures in vitro incorporation amino derivatives into peptides distinct from work flexizyme design rules diverse scaffolds (phenylalanine benzoic acid heteroaromatic aliphatic monomers avoid intramolecular nucleophilic attack amino group backbone-extended monomers tRNA charging focus on long chain carbon cyclic monomers unique non-canonical α-6 β-amino acids7,8 confirm tRNA charging of linear γ-amino acids via flexizyme fails due to deleterious lactam formation circumvent limitation designing amino acid substrate architectures lengthening carbon chain introducing rigid central architecture lactam formation demonstrate incorporation of backbone-extended monomers into N-terminus peptides using wild-type ribosomes use engineered ribosome24 with mutations peptidyl transferase center C-terminal incorporation non-canonical amino acids into peptidebottom panel).ResultsLong chain carbon amino acid Fx carbon amino acid substrates tRNAs 10 substrates (1–5 Fig 2a 2i–2v examined increasing carbons monomer backbone Dinitrobenzyl ester benzyl 3-aminopropanoic acid β-alanine 4-aminobutyric acid (2 2i synthesized Fx charging used tRNA microhelix tRNA yields Fx-mediated acylation reaction Aminoacylation efficiency estimated acid-denaturing polyacrylamide gel electrophoresis 1 charged 2 not tested four γ-amino acid substrates (4-methylaminobutyric acid 2,2-dimethylaminobutyric acid cis- trans-2-aminocyclopropane-1-carboxylic acid Fx-mediated tRNA charging no γ-amino acid substrates (2 2i–v charged results consistent literature Fx-mediated charging γ-amino acid analogs linear carbon chain challenging. 2Systematic design long chain carbon cyclic amino acids range amino acids linear carbon chain extended γ- δ- ε- ζ-amino acidsHigher acylation yields Fx amino acid chain length larger>5-membered ring formation lactamization less favorable rigid bonds Fx acylation yields increased acylation yield ~6% 7 up to ~95% for 12 γ-amino acids with rigid bond (7) cyclic structure (11–15) rigid carbon scaffold inhibits 5-membered lactam formation acylation yield substrate represents percent yield microhelix tRNA at 24 h/120 h Data three experiments lactam formation poor tRNA charging results investigated lactam Fx-catalyzed reaction acylation reaction 4-methylaminobutyric acid with mihx monitored over 24 h analysis LC–MS 24 yielded single new peak (2.3 min light green Fig ESI-MS mass spectra 2.3 mass theoretical mass lactam 1-methylpyrrolidin-2-one lactam observed when Fx mihx present lactam formation catalyzed by species lactam Fx-mediated acylation γ-amino acid lactam Fx acylation substrate 2iiextracted ion chromatogram Fx reaction 24 h ice showed new peak theoretical mass lactam b Data three experiments synthesized long chain carbon derivatives 5-aminopentanoic acid 6-aminohexanoic 7-aminoheptanoic acid (5) support hypothesis acylation yields tRNA increase larger rings>5-membered less favorable observed higher acylation yields increasing lengths carbon chain amino acid derivatives (Fig. 2a Fig 1) suggesting deficiency γ-amino acids due lactam formation substrates Fx-mediated catalysis result rule ring closure rate constant 5-membered ring self-cyclization largest rate constant decreases 1–2 orders slows ring size increases to design molecular architectures circumvent lactam formation restriction amino activated ester functionalities synthesized five substrates (6–10 Fig. 2b) rigid spacer (cyclic aryl vinyl tested acylation substrates γ-amino acid δ-amino acid charged tRNA flexizymes expand synthesized five additional amino acids (11–15 Fig. 2c) structure central region amino acidsubstrates charged to tRNAs acylation yield increased γ-type amino acids rigid cyclic carbon scaffold prevents intramolecular 5-membered lactam formation reaction consistent with design rules for flexizyme-catalyzed acylation15 report cyclic-gamma amino acids into peptides12 cyclic structures contain less steric hindrance increased electrophilicity efficient tRNA 13 non-canonical monomers charged efficiencies 6–95% (E)-4-aminobut-2-enoic acid (7) lowest trans-3-aminocyclobutane-1-carboxylic acid (12) highest yield.Ribosomal polymerization of backbone-extended investigated flexizyme substrates charged to tRNAs by natural protein translation machinery goal demonstrate ribosome compatible with substrates performed Fx-catalyzed acylation reaction for tRNAs same conditions mihx acylation reaction yield kinetics between vitro-transcribed tRNA mimics tRNAs After acylation unreacted monomers separated from tRNAs using ethanol precipitation20 resulting tRNA fraction supplemented into cell-free protein synthesis42 reaction components protein translationdetermined incorporation non-canonical substrates into N or C-terminus Streptavidin tag by MALDI mass spectrometry initiator tRNAfMet selected for N-terminal incorporation C-terminal incorporation assessed tRNAs (fMet Pro1E2 GluE2 engineered incorporate non-canonical amino acids polypeptides observed no significant difference incorporation efficiency codon variations Pro1E244 selected engineered D-arm T-stem interacting protein translation factors EF-Tu-P supplemented cell-free translation reaction used AUG (CAU N-terminal incorporation C-terminal incorporation selected ACC codon (GGU decodes Thr(ACC) codon mRNA threonine excluded from polypeptide Streptavidin tag (WSHPQFEK) prevented endogenous tRNAs aminoacylated charged 14 substrates tRNAfMet tRNAPro1E2 (GGU) acylated tRNAs used PURExpressTM translation reaction Escherichia coli (>46) endogenous tRNAs nine amino acids encoding Streptavidin tag) non-canonical aminoacyl-tRNA substrate usedamino acids (X + WSHPQFEK + T M + WSHPQFEK + X N- C-terminus incorporation X Fx-charged backbone monomer (Fig. 2a. substrate charged tRNAs incorporated into peptide N-terminus peak theoretical mass MALDI spectra (Fig. attempts peptide amino acids C-terminus unsuccessful (Fig. 5a–c C-terminal incorporation requires precise PTC46 wild-type ribosome-extended substrates. 4Ribosomal synthesis N-terminal peptides with backbone-extended monomers backbone-extended amino acids (3–15) charged to tRNAfMet(CAU by Fx incorporated into N-terminus peptide ribosome polymerization PURExpressTM system peptides purified via Streptavidin tag (WSHPQFEK) characterized MALDI mass spectrometryobserved mass peptide theoretical mass + H]+ = 1345 + Na+ = 1367 H+ = 1359 Na+ = 1381 H+ = 1373 + Na+ = 1395 H+ = 1369 Na+ = 1391 1351 H+ = 1379 Na+ = 1401 H+ = 1371 Na 1393 H+ = 1372 Na 1394 H+ = 1343 Na 1365 Na 1365 H+ = 1357 Na = 1379 H+ = 1371 Na+ = 1393 H = 1371 Na+ 1393 peaks truncated peptide not target substrate N-terminus[M + H]+ = 1246; [M + Na]+ = 1268) three experiments. 5Ribosomal synthesis peptides aminocyclobutane-carboxylic acid Peptides synthesized system Fx-mediated tRNAPro1E2 purified Streptavidin tag characterized MALDI mass spectrometry cis trans-ACB not incorporated C-terminus wild-type ribosome Engineered ribosomes facilitate incorporation cis/trans-ACBPeptides cis/trans-ACB at C-terminus observed engineered ribosome Maini et al added protein translation reaction vitro trans-ACB amino acid residues (Ile Ala elongated after incorporation cis/trans-ACB engineered ribosome site-specific incorporation ACB Data three experiments Supplementary Fig. 2.Engineered ribosomes enhance incorporation novel Hecht group engineered ribosome 040329) incorporation dipeptides into growing polymer chain vivo vitro amide bond nascent peptide far-distance amine hypothesized ribosome permissive towards backbone-extended monomers co-expressed mutant ribosomes cells lysed purified ribosomes through ultracentrifugation sucrose cushion ribosome sample wild type 040329 ribosomes used translation assays activity elongated backbone monomers 040329 ribosomes 25% purified ribosome population test incorporating long chain carbon amino acids into peptides with engineered ribosomes added ribosome mixtures (Fig. 5d) into system substrates charged tRNAPro1E2(GGU) FxMALDI mass spectrum observed peak theoretical mass target peptide cis- trans-3-aminocyclobutane-1-carboxylic acids 11 12 Fig. 2 C-terminus (fMWSHPQFEKS11/12 Fig. 5e f 2c not observed experiments wild-type ribosome 5b 2b relative percent yields peptide cis trans-ACB 11% 15% full-length truncated peptide products (fMWSHPQFE fMWSHPQFEK Fig. 2) investigated additional amino acids elongated after incorporation cis-ACB trans-ACB (11 12 Fig. 5g h 2d g C-terminus designed new plasmid additional amino acid residues Ile (AUC) Ala performed PURExpressTM reaction new set 11 amino acids observed peaks theoretical mass target peptides (fMWSHPQFEKS11/12IA), engineered ribosome cis-ACB trans-ACB expanded range backbone-extended amino acid substrates molecular translation investigated acylation step γ-amino acids tRNAs Fxsubstrate design showed 15 amino acids long chain carbon cyclic structures acylated to tRNA by Fx system yields 6–95% demonstrated charged acylated tRNA-monomers in ribosome-mediated polymerization diversity polyamides ribosomal synthesis genetic code expansion incorporated α-based non amino acids known ribosome backbone-extended (γ- δ- ε- ζ- (cyclobutane cyclopentane cyclohexane amino acid structures work shows ribosome polymerizing structures genetic code reprogramming efficiency incorporation C-terminus mid-chain low shape physiochemical dynamic properties ribosome evolved work with canonical α acids β-amino wild-type 040329 ribosomes discriminate against backbone-extended stereoisomer monomers incorporation efficiency supplementing EF-P engineered in vitro ribosome assembly49 platforms evolve ribosomes properties increase incorporation efficiency backbone-extended monomers peptides less truncated products facilitate synthesis polymers monomers extension to cellular systems with orthogonal engineered tethered ribosomes51–55 offers exciting directionlack aminoacyl tRNA-synthetases monomers tRNA expanding long chain carbon cyclic amino acids ribosome-mediated polymerization non-canonical sequence-defined polymers monomers used with vitro screening mRNA ribosome innovative peptide future works enable unique functional materials polymers defined atomic sequence monodisperse length programmed stereochemistry Fx-mediated acylation reactionMicrohelix acylation 1 μL 0.5 M HEPES bicine 1 μL 10 μM microhelix 3 μL nuclease-free water mixed PCR tube 1 μL 10 μM eFx dFx aFx heated 2 min 95 °C cooled room temperature 5 min 2 μL 300 mM MgCl2 added incubated 5 min incubation ice 2 min 2 μL 25 mM activated ester substrate DMSO added incubated 16–120 h ice.tRNA acylation 2 μL 0.5 M HEPES) or bicine 2 μL 250 μM tRNA 2 μM Fx 6 μL nuclease-free water PCR tube heated 2 min 95 cooled room temperature 5 min4 μL 300 mM MgCl2 added cooled mixture incubated 5 min room temperature incubation ice 2 min 4 μL 25 mM activated ester substrate DMSO added incubated conditions microhelix experiment vitro synthesis polyamidesN-terminus incorporation peptide T7 promoter-controlled DNA template (pJL1_StrepII streptavidin tag Ser Thr codons non-canonical amino acid initiation codon AUG N-terminal incorporation non amino acid substrate Peptide synthesis 9 amino acids codon AUG purification tag other 11 amino acids prevent endogenous tRNAs PURExpressTM Δ (aa tRNA kit polyamide synthesis incubated 37 °C 3 h synthesized peptides purified Strep-Tactin®-coated magnetic beads denatured SDS characterized MALDI-TOF mass spectroscopyC-terminus incorporation plasmid (pJL1-StrepII amino acids used for C-terminal incorporation cyclic amino acid incorporated into Thr codon PURExpress® Δ (aa tRNA ribosome kit wildtype ribosome not used 15 μM engineered ribosome added reaction mixture 9 amino acids Strep incubated at 37 °C 3 h.Central-position incorporation plasmid (pJL1-StrepII_TIA Ile Ala Thr cyclic amino acid into middle position polyamide (MWSHPQFEKSXIA polyamide produced using 11 amino acids PURExpressTM Δ kit same reaction conditions C-terminal incorporation.Purification characterization polyamides non amino acid purified affinity tag purification characterized by MALDI spectrometry sample preparation 1.5 μL purified peptide (0.1% SDS water dried 0.5 μL matrix (α-cyano-4-hydroxycinnamic acid THF 10 mg sample characterized Bruker rapifleX MALDI-TOF processed FlexControl v2.0 cells 040329 plasmid rrnB operon pL promoter modified rrnB gene mutations 2057AGCGTGA2063 2502TGGCAG2507 23S rDNA 040329 mutation Plasmids rrnB genes transformed POP2136 plated LB-agar 100 μg/mL carbenicillin plates incubated 16–18 h 30 °C rRNA single colony 25 mL LB-Miller 100 μg/mL carbenicillin culture grown 16–18 h 30 °C 2 L culture 2X YTP 100 μg/mL carbenicillin pre-warmed 42 °C inoculated 20 mL overnight culture Growth 42 °C disrupts repression pL promoter induces expression rrnB operon 040329 mutant rRNA ribosome population 20% plasmid-encoded ribosomes Optical density measured until culture OD 0.4 0.6 cultures pelleted centrifugation 8000 × g 10 min.cell pellet resuspended in Buffer A centrifuged at 8000 × g 10 min repeated two times three washes flash frozen in liquid nitrogen stored at −80 °C until processing.Purification ribosome mixturesFrozen cell pellets resuspended in Buffer A (5 mL per 1 g cell pellet lysed at 20,000–25,000 psi solution centrifuged at 12,000 × g 10 min clarified lysate layered sucrose cushion (1 mL lysate per 1 mL Buffer B ultracentrifuged at 90,000 × g 18 h yielded pellet ribosomes ribosome mixture resuspended with Buffer C 4 °C 4–8 h diluted 20–25 μM ribosomes 260 nm (1 = 4.17 × 10−5 μM ribosomes). After resuspension dilution samples aliquoted flash frozen with liquid nitrogen stored at −80 °C until PURE reactions purification crude mixture maximize mutant ribosomesA 20 mM Tris–HCl 100 mM NH4Cl 10 mM MgCl2 0.5 mM EDTA 2 mM DTT Buffer B 20 mM Tris–HCl 500 mM NH4Cl 10 mM MgCl2 0.5 mM EDTA 2 mM DTT 37.7% sucrose Buffer C 10 mM Tris–OAc 500 mM NH4Cl 7.5 mM Mg(OAc)2 0.5 mM EDTA 2 mM DTT Oligos 040329 ribosome plasmid insert 5′ backbone PCR products isothermal DNA Nature Research Reporting Summary | 48.4 | 0.870512 |
10.1038/s41467-020-20275-0 | PMC7791071 | Artificial enzymes with reprogrammed and augmented catalytic activity and substrate selectivity have emerged to tackle limitations of noble metals or transition metal oxides. Here, the authors report Au25 clusterzymes which are endowed with high catalytic activity and selectivity in a range of enzyme-mimicking reactions. | Emerging artificial enzymes with reprogrammed and augmented catalytic activity and substrate selectivity have long been pursued with sustained efforts. The majority of current candidates have rather poor catalytic activity compared with natural molecules. To tackle this limitation, we design artificial enzymes based on a structurally well-defined Au25 cluster, namely clusterzymes, which are endowed with intrinsic high catalytic activity and selectivity driven by single-atom substitutions with modulated bond lengths. Au24Cu1 and Au24Cd1 clusterzymes exhibit 137 and 160 times higher antioxidant capacities than natural trolox, respectively. Meanwhile, the clusterzymes demonstrate preferential enzyme-mimicking catalytic activities, with Au25, Au24Cu1 and Au24Cd1 displaying compelling selectivity in glutathione peroxidase-like (GPx-like), catalase-like (CAT-like) and superoxide dismutase-like (SOD-like) activities, respectively. Au24Cu1 decreases peroxide in injured brain via catalytic reactions, while Au24Cd1 preferentially uses superoxide and nitrogenous signal molecules as substrates, and significantly decreases inflammation factors, indicative of an important role in mitigating neuroinflammation. | IntroductionDue to their exclusive catalytic activity and selectivity, artificial enzymes are exploited as promising tools for wide-reaching biomedical implications1–8, particularly as advanced diagnostics9,10 and therapeutics11–16 of diseases. Earlier studies shed light on the oxidase- and peroxidase-like activities of noble metals17. Gold-based materials were unraveled to possess versatile enzyme-like activities, such as nuclease, glucose oxidase, peroxidase (POD), catalase (CAT), and superoxide dismutase (SOD)17,18. The Michaelis–Menten constant (Km) to the H2O2 substrate of gold nanoparticles toward the POD enzymatic reaction is below 1 mM, but the catalytic activity is weak19. In contrast, Pt-based materials generally confer a high overall catalytic activity but it can only show a good H2O2 substrate affinity when Km is up to 16.7 mM14,20, and modulation of selective catalysis often needs to be purposely realized through rationally designed combination with other catalysts21. Meanwhile, metal oxides have also revealed great potentials as enzyme mimetics22. Typically, Fe3O4 nanoparticles display the POD-like activity23,24 but are limited by their affinity to the H2O2 substrate (Km at ~154 mM) and a maximal reaction rate (5.9 µM/min) that do not meet expectations. Mn3O4 nanoparticles concurrently exhibit SOD-, CAT-, and glutathione peroxidase (GPx)-like activities via the redox switch between Mn3+ and Mn4+ with a maximum reaction rate reaching 6–125 mM/min at nanomolar levels, which is unfortunately still inferior to natural enzymes25. Thus the development of catalytic artificial enzymes with exceptional activity, adequate selectivity, and satisfactory stability remains a major challenge for any foreseeable practical applications.As is well known to all, most brain injuries involve enzyme-related catalytic processes and continuous neuroinflammation26–28. However, it is largely unclear yet which specific catalytic route(s) can be selectively targeted to inhibit neuroinflammatory responses, primarily because brain injuries simultaneously trigger various kinds of multi-enzymatic reactions between free radicals and numerous bioactive molecules29. Therefore, exploration of versatile artificial enzymes with different catalytic routes and desirable selectivity is beneficial to establish the relationship between oxidative stress and inflammation and to reveal the underlying molecular pathways of catalysis30–32. Atomic-level catalysts suffice a viable solution for the unmet need of improved catalytic activity and precisely modulated selectivity in a controllable manner, with lots of Fe- and Pt-based single-atom nanozymes developed33–40. In particular, Au contains excessive transition metal electronic states and rich electronic energy levels, which provide a solid basis for designing atomic-scale enzyme. Nevertheless, hindered by uncontrollable syntheses and complicated spatial coordination, it is difficult to reveal their electronic structures accurately, which can further influence the catalytic activity and prevent researchers from understanding exact catalytic mechanisms at atomic levels41.In this work, an exemplified Au-based clusterzyme is rationally designed at atomic precision with ultrahigh catalytic activity and superiority over natural antioxidants, and favorable enzymatic selectivity can be achieved via exquisite single-atom substitution by modulating single Cu or Cd active site, consequently serving as a promising artificial enzyme with tuned catalytic selectivity for treatment of neuroinflammation in the brain.ResultsStructural properties of clusterzymesExceedingly different from most previously reported nanozymes1,4,5, the as-developed 3-mercaptopropionic acid (MPA)-protected Au25 clusterzyme is stringently defined by its unambiguous atomic configuration and geometry structure (Fig. 1a). The hydrodynamic size of Au25 is determined to be 2.0 nm by dynamic light scattering (DLS), and the zeta potentials of all clusterzymes are around −35 mV, suggesting the ultrasmall size and good colloid stability (Supplementary Fig. 1). The characteristic absorption at 450 and 670 nm of Au25 is attributed to its unique interband transitions42,43, while a single-atom substitution of Cu or Cd induces a 2–3-nm minor shift, showing insignificant influence on optical properties (Fig. 1b and Supplementary Fig. 2). Electrospray ionization–mass spectra (ESI-MS) reveal a distinct m/z peak at ~2271, assigned to [Au25MPA18−3H]3– (Fig. 1c). After one atom substitution by Cu and Cd, the characteristic m/z peak shifts to ~2226.6 and ~2243, respectively. The inductively coupled plasma-mass spectrometry (ICP-MS) confirms that the ratios of Cd and Cu to the total metal are 5 and 4%, respectively, further validating the successful introduction of single atoms (Supplementary Fig. 3). X-ray photoelectron spectroscopy (XPS) further confirms that Au (0) is the dominant state in all clusterzymes (Supplementary Fig. 4). To identify the precise spatial atomic configuration, extended X-ray absorption fine structure (EXAFS) spectra at the Au, Cu, and Cd edges were recorded (Fig. 1d, e and Supplementary Figs. 5 and 6). The L3 edges of Au in all clusterzymes have higher white-line intensities than the bulk standard Au foil. This is ascribed to larger surface area and alloying effects from partial oxidation with more d-band vacancies from nanoscale sizes and surface molecule-like interactions (Au(I)–thiolate). The characteristic absorption edges of Au clusterzymes were found at ~11,920 eV, which is assigned to the 2p → 5d electronic transition of Au suggesting a reduced population of unoccupied valence d-states. The increase of intensity in MPA-protected Au24Cu1 and Au24Cd1 indicates that the density of 5d electrons of Au is decreased by the one atom substitutions of Cu and Cd through the transfer of their 4d electrons (Supplementary Fig. 5)44,45. The k-space oscillations of Au25 clusters and the Au foil are shown in Supplementary Fig. 5. The k-space of the Au foil exists in typical fcc oscillation patterns, which are apparently absent in all Au clusterzymes due to their small core sizes. Besides, we also investigated the X-ray absorption near edge structure (XANES) spectra of Cu and Cd foils as well as the corresponding atomic counterparts within clusterzymes, clearly displaying differences between single atoms and bulk metals (Supplementary Fig. 6). To further pinpoint the doping sites of Cu and Cd atoms, we performed fitting analysis on the EXAFS data of Cu and Cd. Figure 1d shows the R space of the EXAFS data of the Cu K edge in Au24Cu1. It can be seen that there is only one major peak in the range of 1.6–5.0 Å. This peak roughly corresponds to the scattering path of photoelectron waves from the X-ray absorbing Cu atom to the neighboring S atoms of different shells, and IFEFFIT program is used to fit this peak. The EXAFS parameters obtained after fitting are shown in Supplementary Table 1. The Cu-S coordination number (CN) obtained from the fitting is 1.9 ± 0.2 Å. This value is close to 2 Å, which may indicate that the replacement of Au25 by a Cu atom occurs at the oligomer site, consistent with previous work44. Similarly, the R space of EXAFS data on Cd L3 edge in Au24Cd1 shows a peak in the range of 1.6–5.0 Å, and the fitted Cd-S CN is 2.3 ± 1.7 Å, which is close to the CN of the bond with S at the oligomer site of Au25, indicating that Cd atom substitution may occur at the oligomer site (Fig. 1e)46.Fig. 1Structural characterization of Au25 clusterzymes.a Structure illustration of Au25 and Cu- and Cd-substituted Au24Cu1 and Au24Cd1. b UV-vis absorption spectra and c electrospray ionization mass spectra (in negative ion mode) of Au25 (black curve) before and after Cu (green curve) and Cd (red curve) substitutions. The inset in b is a magnification of the absorption spectrum at ~670 nm. It can be seen that the characteristic absorption is slightly redshifted after introduction of Cu and Cd, indicating decreased band gaps. The red line of the insets in c represents the simulated isotope distribution of [Au25MPA18-3H]3−, [Au24Cu1MPA18-3H]3−, and [Au24Cd1MPA18-3H]3−, respectively. Exp. experiment, Calc. calculated. d Cu K-edge and e Cd L3-edge FT-EXAFS spectra and associated fitting in R space of Au24Cu1 and Au24Cd1, showing the surrounding atoms adjacent to the Cu and Cd atoms.Antioxidant properties of clusterzymesWe tested the general antioxidative properties of all clusterzymes using the ABTS method (Fig. 2a) with reference to standard antioxidants trolox and anthocyanin. Negligible antioxidant activity was observed for pure MPA-protected Au25 clusterzyme. Single-atom substitutions with Cu or Cd in the structure, however, induce a dramatic increase in antioxidant activity with increasing concentrations (Fig. 2b). Among a variety of metals including Ag, Cu, Zn, Er, Pt, and Cd, single-atom substituted candidates Au24Cu1 and Au24Cd1 show the highest activity, representing the optimal substituting elements and ratios (Supplementary Fig. 7). Time-course kinetics of Au24Cu1 and Au24Cd1 exhibit rapid responses to the substrate in seconds with high reaction rates (Fig. 2c). The quantitative results show Au24Cu1 and Au24Cd1 are 41 and 48 times higher in antioxidant activity than Au25, respectively (Fig. 2c). Compared with standard natural antioxidant controls, Au24Cu1 and Au24Cd1 are 137 and 160 times higher in activities than trolox and 7.5 and 9 times higher than anthocyanin, respectively. The reaction rates of Au24Cu1 and Au24Cd1 at 10 and 14 μM/s are 8–11 times higher than Au25 or 38 and 51 fold higher than trolox, respectively. In addition, in a parallel comparison with other elements, substitution with exactly one atom of Cu or Cd presents the foremost activity amidst all substituents (Supplementary Fig. 7d). Preceding studies have evidenced that atomically precise gold clusters, such as Au25 and Au38, are endowed with the oxidation catalytic activities47–50, but their antioxidant activities are rarely reported. Herein, we discovered its ultrahigh antioxidant activity with fast kinetics via atom substitution.Fig. 2The total antioxidant capacity of clusterzymes.a Schematic illustration of the reaction mechanism of the ABTS assay. b Concentration-dependent (0–5 ng/µL) and c time-dependent investigation of ABTS•+ in the presence of Au25, Au24Cu1, and Au24Cd1. d Comparison of the antioxidant capacities and e reaction rates of Au25, Au24Cu1, and Au24Cd1 and natural antioxidants show that the antioxidant performance is greatly improved after one atom substitution with Cu or Cd (n = 3 independent experiments, data are presented as mean ± SD). Compared with natural antioxidants, Au24Cu1 and Au24Cd1 show 137 and 160 times higher activity than trolox, and 7.5 and 9 times higher activity than anthocyanin, respectively.Enzyme-like properties of clusterzymesThe general catalytic profile of clusterzymes and the schematic diagram showing catalytic processes are displayed as in Fig. 3a, b. To pinpoint catalytic selectivity of these clusterzymes, we first investigated the GPx-like activity of Au25, Au24Cu1 and Au24Cd1 at the concentration of 10 ng/μL. Surprisingly, Au25 shows the strongest tendency toward GPx-like activity with a maximum reaction rate of 0.47 mM/min, higher than 0.34 mM/min for Au24Cu1 and 0.10 mM/min for Au24Cd1 (Fig. 3c), and also significantly higher than those of previously reported Mn3O4 nanoflowers (0.056 mM/min)51 and Co/PMCS (0.013 mM/min)52. The turnover frequency (TOF) value of Au25 calculated by the Michaelis–Menten equation is 320 min−1, 4.7 times higher than Au24Cd1 (Supplementary Fig. 8). This result is interesting because metals are generally considered to have low GPx-like activity, but the high GPx-like activity of Au25 can be exploited to eliminate lipid peroxides and oxidative damages. The CAT-like activity of clusterzymes were studied at the concentration of 20 ng/μL as in Fig. 3d. The maximum reaction rate of Au25 is 0.074 mM/min, whereas the introduction of a Cu single atom gives rise to a 4.7-fold increase to 0.35 mM/min, suggesting its CAT-like catalytic preference. The calculated TOF value of Au24Cu1 for CAT-like activity is 116.7 min−1 (Supplementary Fig. 9), which is significantly higher than that of Pd octahedrons (1.51 min−1)53. The SOD-like activity of pure Au25 can only inhibit 31% of the substrate, while one Cd atom substitution considerably increases the inhibition rate to 89%, empowering SOD-like selectivity (Fig. 3e). The aforementioned results suggest enzyme-mimicking preferences of each individual clusterzyme: Au25 as GPx and Au24Cu1 and Au24Cd1 as CAT and SOD, respectively. The structures of clusterzymes before and after reaction with H2O2 suggest unchanged structures of the clusterzymes (Supplementary Figs. 10 and 11)54,55. Previous work mainly focused on the atomic substitutions of Au25 using noble metals for catalytic reactions of hydrogen and CO/CO256–59. Our work herein constructively hypothesized and demonstrated that Au25 can possess various unique enzyme-like activity modulated by single-atom substitution with non-precious metals like Cu and Cd, instead of Pt, in the geometric structure.Fig. 3Enzyme-mimetic properties and ROS scavenging activity of Au25 clusterzymes.a The radar map of enzymatic activities and free radical scavenging abilities of various clusterzymes. b Schematic illustration of catalytic selectivity of the clusterzyme system. Au25 exhibits significant superiority in GPx-like activity; Au24Cu1 shows advantages in the CAT-like activity through the Cu single active site; Au24Cd1 preferably exhibits the SOD-like activity via the Cd single active site, each demonstrating a unique catalytic selectivity. c GPx-, d CAT-, and e SOD-like activities of Au25, Au24Cu1, and Au24Cd1. ROS scavenging activities of Au25, Au24Cu1, and Au24Cd1 clusterzymes for f •OH and g O2•− studied by the ESR spectroscopy. BMPO is used as the ROS-capturing agent and the sources of •OH and O2•− are H2O2 and KO2, respectively. h Corresponding quantifications of the scavenging efficiencies (n = 3 independent experiments, data are presented as mean ± SD).The corresponding specific scavenging of free radicals by the clusterzymes was further investigated. The scavenging of •OH free radical was investigated using electron spin resonance (ESR) by employing 5-tert-butoxycarbonyl-5-methyl-1-pyrroline N-oxide (BMPO) as the trapping agent. The ESR signal of •OH is strong for the BMPO control, suggesting the presence of excessive •OH, while there is only a minor decrease after adding Au25, indicative of a weak scavenging efficiency for •OH (Fig. 3f). However, Au24Cu1 almost completely diminishes all ESR signals (~100%), consistent with the observed best CAT-like activity as in Fig. 3d. Similarly, the scavenging of O2•− by clusterzymes was also investigated (Fig. 3g). The ESR signal stays strong for the control and slightly decreases after addition of Au25 and Au24Cu1, with surplus remaining residues. In contrast, the ESR signal of O2•− almost disappears in the presence of Au24Cd1 further validating its superior specialized SOD-like activity (Fig. 3e). Besides, we also tested the free-radical scavenging capability of the clusterzymes toward reactive nitrogen species (RNS) including •NO, ONOO–, and DPPH•. Au24Cd1 shows the most robust overall scavenging efficiency against DPPH• (Supplementary Fig. 12). The ESR reveals that Au24Cd1 has the best scavenging capability toward •NO at a low concentration of 2.7 ng/μL, whereas Au25 presents ignorable activity (Supplementary Fig. 13). Likewise, both Au24Cd1 and Au24Cu1 also manifest significantly higher scavenging efficiency toward ONOO− than Au25 (Supplementary Figs. 14 and 15). Au24Cd1 is more selective against RNS than Au24Cu1, while Au25 has insignificant catalytic activity. Thus it is rational to conclude that the high selectivity for enzymes and radicals originates from the single-atom substitutions of Cu and Cd, which induce redistribution of surface electrons and exert influence on electronic structures and states.DFT calculations and the mechanism of catalytic selectivityTo reveal the catalytic mechanism, the density functional theory (DFT) was employed to investigate the catalytic selectivity and quantum properties. By exploring possible structures in the literature, we adopt the gold core of the well-known Au25 clusters60 protected by ligands, which are still connected to the core via S atoms. To evaluate the catalytic behavior and the intermediate states during the chemical reactions, each ligand unit -SCH2CH2COOH is simplified to -SCH3. DFT optimization confirms the stability of the modeled cluster.Due to the symmetry of the gold cluster, all possible replacements of the guest metallic atoms fall into three categories as follows: oligomer, the surface of core, and core replacements. Figure 4a demonstrates the surface sites of the Au13 core and the oligomer site replacement. The oligomer replacement is the common form that is extensively discussed in the literature, but DFT simulations indicate that the surface replacement may be another possibility. However, based on the coordination analysis in the experiment, the oligomer replacement matches the EXAFS results, which yield a significantly lower CN than the surface replacement. With the optimized structure of the clusters, the associated CNs can be theoretically generated even within the clusters involved in the intermediate structures during the catalytic process. The averaged simulated CN values agree with the experimental values that confirm the oligomer replacement (Supplementary Table 2). Although more theoretical investigations on the surface replacements61 can be found in the appendix (Supplementary Figs. 16–19 and Supplementary Tables 3 and 4), we focus on the oligomer replacement and the associated catalytic efficiency.Fig. 4DFT calculations and the mechanism of catalytic selectivity.a Demonstration of atomic doping: surface and oligomer replacement. b Doped atom caused change in bonds: the more rigid Cu-S bond and the more flexible Cd-S bond. The flexibility of Cd allows the angular motion. c Mechanism and d, e energies profiles of catalytic process of the SOD and CAT processes. The black dotted line indicates that the catalytic products may be connected to other processes. Energy profiles and geometry structures of the intermediate states of f SOD and g CAT process in the lower panel. Align optimized intermediate structures (normal color) and transition structures (black).Unlike the surface replacement that may cause the expansion of the core, the oligomer replacement causes the oligomer bending. It is different from the normal S-Au-S chain, which aligns in a (nearly) straight line (Supplementary Fig. 20). The doped Cu shrinks the S-X-S chain while the doped Cd extends it. Compared with the typical bond length of S-Au at 2.3 Å, S-Cu and S-Cd bonds are 2.2 and 2.55 Å, respectively, as shown in Fig. 4b and Supplementary Fig. 21. With the bent chain, the distances between Cu/Cd atoms to the surface of the core are comparable, around 3.1 Å. The similarity between the Cu and Au atoms guarantees that the binding of S-Cu-S is so “firm” that the relative positions of Cu to S atoms can be hardly changed by the dynamics during the catalytic procedures, which are discussed extensively below. In contrast, the relative position of the doped Cd atom may be significantly affected by the local environment such as the adhesion of small chemical units (Supplementary Figs. 22 and 23).We observed the excellent performance of the clusterzymes in both CAT and SOD reactions with the reaction pathways summarized in Fig. 4c. The CAT reaction usually refers to the catalytic degradation of hydrogen peroxide, and the decomposition mechanism of H2O2 may involve multiple chemical stages (Fig. 4d, e). For the process of SOD, we assumed that the clusters were involved in similar mechanisms to the general catalytic scheme of SOD reaction. It is worth noting that the release of oxygen completes the CAT process, while the SOD process occurs simultaneously, and the two processes are mutually permeated. The reduced cluster, Cluster(I), may also be involved in both CAT and SOD processes, which depends on the concentration of different components.Inspired by the Arrhenius equation, we performed the search of transition states and ground states of various types of molecules and ions to estimate the activation energies and evaluated the catalytic efficiencies. The energy profiles in Fig. 4f, g agree with the behaviors of the clusterzymes in our experiments. In a series of reactions with multiple steps, the reaction rate is dominated by the slowest step, i.e., the transition with the largest activation energy. Such a feature can be seen in the first part of the catalytic SOD process by Au24Cu1. The ground state of the electrons in the corresponding intermediate structure is a triplet state suggested by DFT simulations. It indicates that the high activation energy of 131.2 kJ/mol is related to the spin matching issues, which can be selective to the spin of superoxide ions. In the CAT processes, the critical step is related to the decomposition of (cluster…OOH)2+, in which the Au24Cd1 exhibits higher activation energy (71.3 kJ/mol) that reduces the efficiency. The simulations clearly explain the SOD–CAT-selective behaviors of the doped clusterzymes. The details of the reaction pathways are provided in “Methods.”Our results of DFT simulations show some insights of the catalytic mechanisms. Assume in typical clusters, a substituted atom may turn into an active site itself to be involved in the catalytic process, which may be accompanied by changes in the geometry. Herein, we named two mechanisms as SA (simple adhesion) and MA (bond modulated adhesion) correspondingly. The distances between the adsorbed molecule/ion and metal atoms designate the roles of the substituted atom.The SA mechanism is mainly seen in Au24Cu1. Due to the firmness of the S-Cu-S oligomer, the Cu atom is relatively rigid (Fig. 4f, g). The SA mechanism is also seen in the first step of SOD process catalyzed by Au24Cd1. The catalytic process includes the distance change and orientation change of small units. Significant changes in the distance between the active site (doped atom) and the small units are seen in most of the SOD processes. In contrast, the orientation change is the main character in most of the CAT processes.The MA mechanism is seen in Au24Cd1 in the second stage (Cluster(I)) of the SOD processes and most of the CAT processes. The bond modulation refers to the position change of the Cd atom, which may deviate from the oligomer plane until a third S atom from another oligomer stops it. Thus the S-Cd bonds are changed significantly (Fig. 4f, g). The characteristics of transition and intermediate states involve the rotation of the superoxide ion (or oxygen molecule) and the position adjustment of the doped Cd atom. To be more precise, the motion of the Cd atom is along the perpendicular direction of the oligomer plane. Once a small unit joins the doped cluster to form an intermediate structure, the Cd atom sometimes leaves the oligomer plane. Therefore, the CN value of the Cd atom is larger when the Cd atom becomes the neighbor of three S atoms. When the Cd atom starts from its original state (S-Cd = 2.55 and 2.55 Å), passes its transition state (S-Cd = 2.63 and 2.85 Å), and arrives at the intermediate state (S-Cd = 2.57 and 2.60 Å), the angular motion is terminated (Supplementary Fig. 24). During such a process, the distance between the attached oxygen atoms is slightly expanded toward the normal distance of oxygen molecules, which indicates the completion of the entire catalytic procedure (Supplementary Tables 5 and 6). A similar procedure for the Au24Cd1 can be observed at the adhesion of OOH- at the first stage of CAT process. Such a unique process allows the doped Cd atom to be an active site that can be self-modulated in a wide spatial range compared to the firm Cu atom. This may explain its good performance in the SOD process.Modulation of neuroinflammationTo reveal the biological activity of clusterzymes, the cell toxicity for different nerve cell lines (HT22, BV2, and MA-c) were measured by the 3-[4,5-dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide (MTT) assay (Fig. 5a and Supplementary Fig. 25), showing that Au25, Au24Cu1, and Au24Cd1 present acceptable biocompatibility. Cell survival of H2O2-stimulated neuron cells was performed with the incubation of Au25, Au24Cd1, or Au24Cu1. As shown in Fig. 5b, the clusterzyme treatment could improve the viability of neuron cells. To explore the correlation between the oxidative stress and the neuron viability, reactive oxygen species (ROS), especially •OH and O2•−, were quantified and detected by a FACS flow cytometer and a fluorescence microscope using hydroxyphenyl fluorescein (HPF) and dihydroethidium (DHE) fluorescence probes, respectively. The H2O2 stimulation significantly elevates the fluorescence signal, indicating the presence of excessive amount of •OH and O2•− (Fig. 5c–h). All clusterzymes decrease the ROS signals, with Au24Cu1 showing the best clearance efficiency against •OH (Fig. 5c, e, g) and Au24Cd1 displaying the best clearance capability for O2•−, suggesting their individual selectivity (Fig. 5d, f, h). Meanwhile, mouse models of traumatic brain injury (TBI) were used to examine the in vivo effects of clusterzymes. As shown in Fig. 5i–l, the indicators of malondialdehyde (MDA), H2O2, SOD, and GSH/GSSG in the TBI group are relatively severe at day 1 post injury but are slightly alleviated 3 days post injury and are further improved slightly 7 days post injury. Therefore, the decrease in SOD and GSH/GSSG levels from TBI can be well rescued by clusterzymes with prominent recoveries 7 days after treatment (Fig. 5i, j). Comparatively, Au24Cd1 induce a better recovery in SOD than Au24Cu1, which correlates well with their in vitro SOD-like activity (Fig. 3). As the by-products of the oxidative stress, lipid peroxides and H2O2 show higher accumulations in the brain following TBI, resulting in severe oxidative damage (Fig. 5k, l). Both Au24Cu1 and Au24Cd1 significantly inhibit the production of these harmful molecules, while Au25 barely alters the TBI-induced increase. These results are conceivable because O2•− is known to be continuously produced by immediate injuries at the early stage, followed with subsequent production of lipid peroxides and H2O2. With regard to Au24Cd1, it can recover the diminished SOD in the first place due to its high catalytic selectivity for O2•− and then sustain the continuous decrease of lipid peroxides and H2O2 as the secondary catalytic options. In contrast, Au24Cu1 is primarily prone to increase the levels of lipid peroxides and H2O2 at the early stage due to its preference for CAT-like activity and •OH, but these molecules are intermediate products at relatively low concentrations after TBI onset, and consequently it accounts for the increasing clearance capability in the long term.Fig. 5Oxidative stress levels in vitro and in vivo before and after treatment of clusterzymes.a HT22 cell viability of clusterzymes (n = 5 per group, data are presented as mean ± SD). b HT22 cell viability in the presence of H2O2 with or without treatment of clusterzymes as determined by MTT assays (n = 6 per group, data are presented as mean ± SD). Fluorescence quantification of cell staining for c, e •OH and d, f O2•− by flow cytometry (n = 3 per group). Data are presented as mean ± SD and compared with the Con and H2O2 groups, analyzed by one-way ANOVA with two-sided LSD test (adjusted p values are shown). Fluorescence microscopic images of intracellular g •OH (green) and h O2•– (red) levels induced by 100 μM H2O2 with or without clusterzymes treatment, stained by HPF and DHE probes, respectively. It can be seen that Au24Cu1 has a better scavenging ability for •OH and Au24Cd1 shows better specificity for O2•−, suggesting their individual selectivity for •OH and O2•− respectively. i–l Indicators for oxidative stress, including SOD, GSH/GSSG, MDA, and H2O2, of TBI mice with or without treatment of clusterzymes 1, 3, and 7 days post injury (n = 5 per group). Data are presented as mean ± SEM and compared with the Sham and TBI groups, analyzed by one-way ANOVA with two-sided LSD test (adjusted p values are shown). Experiments were repeated independently g, h three times with similar results.Finally, the effects of clusterzymes on neuroinflammation were examined. From the western blots and the relevant quantification analysis (Fig. 6a, d), the expression levels of interleukin (IL)-1β and IL-6 are significantly upregulated following TBI 1 day post injury, indicative of strong local inflammations. Au24Cd1 sharply downregulates IL-1β and IL-6 levels, suggesting the anti-neuroinflammation effect. In comparison, Au25 only shows minor downregulation. Similarly, TBI leads to significant upregulation of tumor necrosis factor-α (TNFα) 1 day post injury, but Au24Cu1 can significantly downregulate the expression of TNFα, presenting superior efficacy over Au25 and Au24Cd1 (Fig. 6a). Although the expression levels of these inflammation cytokines induced by TBI are gradually suppressed by autoimmunity in the vehicle control group over time, especially 7 days post injury, there are still significant differences in IL-1β and IL-6 levels between the Sham and TBI groups (Fig. 6b–d). However, the clusterzyme treatment results in cytokines close to the normal level, indicating a better suppression effect on neuroinflammation. Enzyme-linked immunosorbent assay (ELISA) further validated the immunoblotting results that Au24Cd1 and Au24Cu1 are capable of decreasing the inflammatory cytokines in brain tissues such as IL-1β, IL-6, and TNFα, while Au25 does not significantly alter the inflammatory cytokine patterns (Fig. 6e–g). Au24Cd1 can eliminate IL-1β- and IL-6-associated inflammatory responses, while Au24Cu1 has a better effect on reduction of TNFα, indicating their relevant selectivity toward modulation of neuroinflammation. Finally, immunofluorescence staining of cerebral cortex harvested from mice also shows that Au24Cd1 and Au24Cu1 can remarkably decrease the TBI-elevated expression levels of IL-1β, IL-6, and TNFα (Fig. 6h, i and Supplementary Figs. 26 and 27), therefore alleviating neuroinflammation. Colocalization studies with markers for neurons (NeuN), microglia (Iba-1), or astrocytes (glial fibrillary acidic protein (GFAP)) were performed in injured cortex on day 3 post injury. Figure 6h, i reveal that IL-1β is mainly produced by microglia after TBI, similar with IL-6 and TNFα (Supplementary Figs. 26 and 27). In addition, quantitative analyses of the number of positive cells show that massive microglia and astrocytes are activated, and many neurons are depleted after TBI (Fig. 6j). With the clusterzyme treatment, most of these nerve cells are rescued. Meanwhile, the morphology of TBI-activated astrocytes can be recovered to near normal levels after treatment with clusterzymes (Supplementary Fig. 28), and the neuroinflammatory responses are also prevented likewise by verified histology (Supplementary Figs. 29–32). The clusterzymes can also restore the TBI-induced body weight loss (Supplementary Fig. 33). Moreover, behavioral tests were studied by the Morris water maze (MWM). As shown in Supplementary Fig. 34a, b, all the mice apparently learned the task during the acquisition phase of days 13–17 and 28–31, while the distance traveled and latency to hidden platform with Au24Cu1 and Au24Cd1 treatment obviously decreased. For the probe trial on days 18 and 32 (Supplementary Fig. 34c, d), the percentage time in the missing platform quadrant and the number of platform crossings were significantly reduced in the TBI group but almost return to the normal level after Au24Cu1 and Au24Cd1 treatment. These results reveal trends in the improvements of learning ability and spatial memory with Au24Cu1 and Au24Cd1 treatment. In addition, we systematically studied the pharmacokinetics and toxicology of clusterzymes. It can be seen that the clusterzymes accumulated in major organs can be removed by the kidney (urine) and liver (feces). After 48 h, ~80% of the total dose can be excreted, and most of it is excreted through the kidney (>70%) (Supplementary Fig. 35). No significant changes in organs or blood chemistry or hematology are found, suggesting that renal-clearable clusterzymes do not cause significant biological toxicity in vivo (Supplementary Figs. 36–38). Artificial enzymes have persistently been shown to exhibit multiple enzyme-like catalytic activities with a diversified class of materials15. Low catalytic activity as compared to natural enzymes, however, is one of the most noticeable disadvantages due to limited electron transfers at atomic levels15. The rationally designed clusterzymes with single-atom substitutions overcome such barriers with antioxidant activity nine times higher than that of anthocyanin, which is known to be one of the most reactive antioxidant molecules in nature. Besides, unlike the structurally ambiguous traditional artificial enzymes, the definitive molecular structures of clusterzymes are accurately elucidated, allowing us to distinguish the catalytically active sites and scrutinize the electronic structures and reaction energies62–65. As a result, the substituting single atoms can be arranged into a specific spatial location of the clusterzyme freely, thus tuning electronic structures and affecting the catalytic activity66–70. Meanwhile, the interactions between host atoms (i.e., Au) and the introduced substituting atoms (i.e., Cu or Cd) can induce coupled electron states and in turn influence the catalytic selectivity71. In our work, the GPx-, SOD-, and CAT-like catalytic selectivity were assigned to Au25, Au24Cd1, and Au24Cu1, respectively, via modulated bond lengths to the active center, and thus it is conceived that such a platform of clusterzymes will generate various selectivity against different molecules. By employing the three catalytically selective clusterzymes, we successfully established the relationship between oxidative stress and neuroinflammation, demonstrating the importance of O2•− and long-term benefits in TBI. Specifically, Au24Cd1 can significantly mitigate the neuroinflammation via inhibiting IL-1β and IL-629,72, while Au24Cu1 differentially reduces neuroinflammation by inhibiting TNFα, showing selectivity against anti-neuroinflammation. Meanwhile, due to the innate ultrasmall size of clusterzymes, it can penetrate the kidney barriers and be excreted by renal, avoiding long-term hepatotoxicity and multi-organ injuries. Therefore, the clusterzymes are presumably influential as a biomedicine, especially in the field of neuroscience.Fig. 6Inflammation levels in brain tissues after clusterzyme treatment.a–c Western blotting for IL-1β, IL-6, and TNFα in the brain tissues 1, 3, and 7 days post TBI after treatment (n = 3 per group), respectively. d Western blotting quantitative analysis of inflammatory factors at different time points (n = 3 per group). All the samples were derived from the same experiment and blots were processed in parallel. Data are presented as mean ± SEM and compared with the Sham and TBI groups, analyzed by one-way ANOVA with two-sided LSD test (adjusted p values are shown). It can be seen that Au24Cd1 can rapidly and significantly reduce the upregulated inflammatory cytokines of IL-1β and IL-6 after brain injury, while Au24Cu1 has a better ability to reduce the expression of TNFα. e–g ELISA quantitative analysis of IL-1β, IL-6, and TNFα levels in brain tissues on days 1, 3, and 7 with or without clusterzymes treatment (n = 5 per group), respectively. Data are presented as mean ± SEM and compared with the Sham and TBI groups, analyzed by one-way ANOVA with two-sided LSD test (adjusted p values are shown). h Immunofluorescence co-staining of IL-1β and microglia (Iba-1), astrocytes (GFAP), or neurons (NeuN) in injured cortex 3 days post injury with or without clusterzyme treatment. Quantitative analysis of i the number of IL-1β+ expression in different positive cells and j the pixels density of Iba-1/NeuN/GFAP cells in the injured cortex with or without clusterzyme treatment (n = 3 per group). Data are presented as mean ± SEM and compared with the Sham and TBI groups, analyzed by one-way ANOVA with two-sided LSD test (adjusted p values are shown). Experiments were repeated independently a–c twice and h three times with similar results.DiscussionIn summary, we report a systemic single-atom substitution approach to fabricate artificial enzymes on the basis of MPA-protected Au25 clusters, namely clusterzymes. The clusterzymes show the ultrahigh antioxidant activity up to 137–160 times higher than the natural trolox. Moreover, the catalytic selectivity toward GPx, CAT, SOD, and nitrogen-related signaling molecules can be fine-tuned by single-atom substitutions. DFT calculations conclude that reaction pathways are modulated by the single active site of Au24Cd1 and Au24Cu1 at bond lengths. The biological results show that Au24Cd1 preferentially decreases IL-1β and IL-6, while Au24Cu1 tends to decrease TNFα, indicative of their different selectivity for modulating alleviation of neuroinflammation.MethodsMaterialsAll chemicals are commercially available with the highest purity and used without further treatment. Gold chloride (HAuCl4·3H2O) was purchased from Sigma-Aldrich; sodium hydroxide (NaOH), sodium borohydride (NaBH4), copper nitrate (Cu(NO3)2), cadmium nitrate (Cd(NO3)2), and MPA were purchased from Aladdin. Ultrapure water (18.2 MΩ·cm) was used for all the experiments.Materials’ preparationThe gold nanoclusters were synthesized according to the previous literature73. In detail, aqueous solutions of HAuCl4 (20 mM, 0.25 mL) and MPA (5 mM, 2 mL) were added to water (2.35 mL) and stirred at room temperature for 5 min. Then an aqueous NaOH solution (1 M, 0.3 mL) was added to the reaction solution, followed by the addition of 0.1 mL of an NaBH4 solution (prepared by dissolving 43 mg of NaBH4 powder in 10 mL of 0.2 M NaOH solution). The whole reaction was carried out in the dark, and Au25MPA18 was collected after stirring at room temperature for 3 h, and the final reaction solution aged at 4 °C for 12 h. The syntheses of various metal-substituted AuxM25-xSG18 were also based on the same method. The only difference was that the Au atoms in HAuCl4 (20 mM, 0.25 mL) were replaced by various nitrate metal ions (Cu2+, Cd2+) at a 4% molar ratio (Au:M = 24:1). For further purification of nanoclusters, we used ultrafiltration tubes of 3 and 10 K at 3500 rpm/min for ultrafiltration to remove smaller organic ligands and larger-sized clusters and lyophilized to obtain the purified product for further testing and application.Materials’ characterizationUltraviolet–visible (UV-vis) absorption spectra were recorded on Shimadzu 3600 UV-vis-NIR spectrophotometer. ESI-MS were acquired on Bruker microTOF-Q system. XANES along with EXAFS analyses were tested and provided by Beijing Synchrotron Radiation Facility. The module ARTEMIS of programs of IFEFFIT were used for processing data of XANES and EXAFS74,75. Clusterzymes were pressed into pellets, and Au L3-edge, Cu K-edge, and Cd L3-edge were measured at room temperature. The Au/Cu/Cd foil was also measured for comparison. XPS of the metal element was performed on a K-Alpha spectrometer with a monochromatic Al Kα X-ray source operating at 300 W (ThermoFisher Scientific). The C 1s level of 284.8 eV was used as an internal standard to correct for peak drift, and the XPS peaks were fitted using the XPSPEAK41 software. Fourier transformed infrared spectra of various clusterzymes before and after the reactions were recorded on AVATR360 Spectrometer (Thermo Nicolet, US). The scan wavenumber ranges from 400 to 3800 cm−1, and the samples were determined by powder method. Raman spectrum was performed on INVIA Reflex spectrometer (Renishaw, UK) excited by a 633 nm He-Ne laser. A Malvern Zetasizer nano ZS90 (UK) was employed for measuring DLS to test the hydrodynamic size and zeta potential of clusterzyme. ICP-MS was tested with 7900 ICPMS (Agilent, US) to determine the content of metallic elements in clusterzymes. The scavenging process of •OH, •NO, and O2•– was determined by ESR spectrometer (Bruker EMX plus, Germany). The kinetic test of CAT-like was performed by Dissolved Oxygen Meter (HACH HQ40d, US) with LDO101 probe.Antioxidant and free radical scavenging testsTotal antioxidant capacity test (ABTS rapid method)The total antioxidant capacity (T-AOC) of clusterzyme and contrast (trolox and anthocyanin) was determined by the rapid ABTS method using the T-AOC Assay Kit (S0121, Beyotime). Please refer to the specification for specific sampling methods. The antioxidant capacity was evaluated by measuring the absorption value at 414 nm. In the process of reaction kinetic analysis, we adjusted the concentration of different concentration of ABTS•+ (the molar extinction coefficient of ABTS•+: ɛ414 nm = 3.6 × 104 mol−1 cm−1) by changing the concentration of H2O2. The reaction kinetic analysis process was reflected by the change of absorbance at 414 nm monitored by the UV-vis spectrophotometer under kinetic mode. The steady-state kinetic parameters were determined by varying the concentration of ABTS•+ in the presence of clusterzymes (5 ng/µL). The maximum reaction velocity (Vmax) and Michaelis–Menten constant (Km) were calculated using the Lineweaver–Burk equation.RNS scavenging testThe RNS scavenging capacity of clusterzyme was performed for 1,1-diphenyl-2-picrylhydrazyl radical (DPPH•). The scavenging capacity of free radicals was evaluated by measuring the absorption wavelength at 510 nm. Briefly, 50 µM DPPH• and 5 ng/µL clusterzyme were dissolved in a mixture of dimethyl sulfoxide and water (1:40). The changes of absorption spectra with time in the range of 300–1000 nm were determined.ONOO– scavenging testThe preparation of ONOO− was reported by methods in the literature4. Specifically, the aqueous solution of NaNO2 (5 mL, 50 mM) and H2O2 (5 mL, 50 mM) was rapidly stirred and mixed in an ice bath. Then HCl (2.5 mL, 1 M) and NaOH (2.5 mL, 1.5 mM) were quickly added and continued to stir for 5 min to get the yellowish ONOO− reserve solution. The ONOO− reserve was diluted to about 1.3 mM (ɛ302 nm = 1670 ± 50 M−1 cm−1), and 5 ng/µL of clusterzyme was added. The scavenging capacity of ONOO− was evaluated by measuring the change of absorption spectra in the range of 250–400 nm over time.•OH scavenging testFirst, a 5 mM H2O2 solution was prepared with a 10 mM phosphate-buffered saline (PBS) buffer, and then the hydroxyl radical was generated by UV-laser irradiation for 5 min. We used an ESR spectrometer (Bruker EMX plus, Germany) to determine the scavenging process of hydroxyl radicals. BMPO (50 mM) was used as the capturing agent for hydroxyl radical, and the spin adduct (BMPO/•OH) generated with •OH presented four peaks under ESR spectrometer. The removal process of •OH was determined by testing the change of its peak strength before and after the addition of clusterzymes (2.7 ng/µL).O2•– scavenging testKO2 of 2.5 mM and 18-crown-6 of 3.5 mM were used as the generation source and stabilizer of O2•–. The 25 mM 5-(diethoxyphosphoryl)−5-methyl-1-pyrroline-N-oxide (DEPMPO) was used as a spin capturing agent, and its spin adduct (DEPMPO/O2•–) presented six peaks under ESR spectrometer. Monitoring the signal intensity change of peaks before and after the addition of 2.7 ng/µL clusterzyme could verify the scavenging capacity to O2•–.•NO scavenging testCarboxy-PTIO of 10 µM with 5 peaks under ESR spectrometer was used as the capturing agent, and S-nitroso-N-acetylpenicillamine of 250 µM was used as the •NO contributor. The spin adduct (Carboxy-PTI) generated presented seven peaks under ESR spectrometer. The scavenging capacity for •NO was evaluated by monitoring the changes of peaks intensity and types before and after the addition of 2.7 ng/µL clusterzymes.Enzyme-like activity testCAT-like testCAT-like activity of clusterzymes was determined by two methods. First, according to the unique absorption peak of H2O2 at 240 nm, the optical density decreases with the decomposition of H2O2, and the extinction coefficient of H2O2 (43.6 mM−1 cm−1 at 240 nm) was used to calculate its activity. The reaction solutions contained 53 µM H2O2 and 10 ng/µL of clusterzyme in 200 µL PBS. H2O2 10 M was treated with or without 50 ng/µL clusterzymes for 30 min to obtain the photos of H2O2 decomposition in the centrifuge tube. Another method was to measure the kinetics of the CAT-like using the Dissolved Oxygen Meter (HACH HQ40d, US) with LDO101 probe. First, the solubility of O2 in solution was reduced to 0.6 mg/L by continuous infusion of Ar into 5 mL PBS solution. Then, different concentrations of H2O2 (50–1000 mM) and 20 ng/µL of clusterzymes were added to the system to monitor the solubility change of O2 every 10 s. The interference deducted from each set of experiments is the effect of H2O2 self-decomposition under the same conditions. The maximum reaction velocity (Vmax) and Michaelis–Menten constant (Km) were calculated using the Lineweaver–Burk equation by the Origin 9.0 software.GPx-like testThe GPx-like activity of clusterzymes was determined by the method in the literature. Briefly, 200 μM H2O2, 2 mM GSH, 200 μM NADPH, 1.7 units/mL GR, and 10 ng/μL clusterzymes were added to 200 μL PBS neutral buffer. The activity of GPx-like was evaluated by monitoring the changes in absorbance at 340 nm. The absorbance at 340 nm represents the concentration of NADPH (ε340 nm = 6.22 mM–1 cm–1). GPx-like activity = (ACon. − AClusterzyme)/ACon. × 100%. The reaction kinetic analysis process was reflected by the change of absorbance at 340 nm monitored by the UV-vis spectrophotometer under kinetic mode. The formation concentration of the substrate was changed by adjusting the concentration of H2O2 (100–700 μM), and the other conditions remained the same. The maximum reaction velocity (Vmax) and Michaelis–Menten constant (Km) were calculated using the Lineweaver–Burk equation by the Origin 9.0 software.SOD-like testThe SOD-like activity of clusterzymes was tested according to the description in the SOD Activity Assay Kit. After adding clusterzymes (0–10 ng/μL) of different concentrations, the absorbance changes at 560 nm were monitored with UV-vis spectrometer to further evaluate SOD-like activity.DFT calculationsTo investigate the catalytic effect, we focus on the energy profile of the clusterzymes in their intermediate structures. The adsorption energy that describes the energy change of an adsorbate when being attracted by a cluster is calculated by the following equation:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{ads}}} = E_{{\mathrm{cata}} + {\mathrm{mol}}} - (E_{{\mathrm{cata}}} + E_{{\mathrm{mol}}}).$$\end{document}Eads=Ecata+mol−(Ecata+Emol).The Ecata and Emol denote the energy of the cluster and the adsorbate, respectively, while the Ecata+mol is the energy of the intermediate structure. In addition, we also simulate the activation energies, which are the differences of the energies between the transition states and energetically stable states.2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\mathrm{act}}} = E_{{\mathrm{cata}} + {\mathrm{mol}}}^{{\mathrm{TS}}} - E_{{\mathrm{cata}} + {\mathrm{mol}}}.$$\end{document}Eact=Ecata+molTS−Ecata+mol.The transition states, which are used to estimate the activation energy barriers that corresponds to the saddle points on an energy surfaces (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_{{\rm{cata}} + {\rm{mol}}}^{{\rm{TS}}}$$\end{document}Ecata+molTS), are searched using Berny algorithm.The CAT processA CAT reaction usually refers to the catalytic degradation of hydrogen peroxide. The total reaction is3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2{\mathrm{H}}_2{\mathrm{O}}_2 \to 2{\mathrm{H}}_2{\mathrm{O}} + {\mathrm{O}}_2.$$\end{document}2H2O2→2H2O+O2.In our experiment, we observed outstanding performance of the clusterzymes as the CATs. The mechanism of the decomposition of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{H}}_2{\mathrm{O}}_2$$\end{document}H2O2 may involve multiple steps and multiple paths. Considering the catalytical mechanism of copper (II), we propose the clusterzyme initiate the reaction in the following steps:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{cluster}}^{2 + } + {\mathrm{H}}_2{\mathrm{O}}_2 \to \left( {{\mathrm{cluster}} \cdots {\mathrm{OOH}}} \right)^ + + \,\,{\mathrm{H}}^ +.$$\end{document}cluster2++H2O2→cluster⋯OOH++H+.Then the intermediate structure, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{cluster}} \cdots {\mathrm{OOH}}^ +$$\end{document}cluster⋯OOH+, decomposes into three pieces: a cation, a superoxide ion, and the original cluster with an extra electron, Cluster (I):5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( {{\mathrm{cluster}} \cdots {\mathrm{OOH}}} \right)^ + \to {\mathrm{cluster}}^ + + {\mathrm{O}}_2^{ \cdot - } + {\mathrm{H}}^ +,$$\end{document}cluster⋯OOH+→cluster++O2⋅−+H+,6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\cdot {\mathrm{OH}} + {\mathrm{O}}_2^{ \cdot - } \to {\mathrm{O}}_2 + {\mathrm{OH}}^ -.$$\end{document}⋅OH+O2⋅−→O2+OH−.The superoxide ion may be involved in a process of hydroxyl scavenging (Eq. 6) or follow another SOD reaction path. The reduced cluster can cause a disproportionation-like process in which the hydrogen peroxide is not equally divided:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{cluster}}^ + + {\mathrm{H}}_2{\mathrm{O}}_2 \to ({\mathrm{cluster}} \cdots {\mathrm{OH}})^{2 + } + {\mathrm{OH}}^ -,$$\end{document}cluster++H2O2→(cluster⋯OH)2++OH−,8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( {{\mathrm{cluster}} \cdots {\mathrm{OH}}} \right)^{2 + } \to {\mathrm{cluster}}^{2 + } + \cdot {\mathrm{OH}}.$$\end{document}cluster⋯OH2+→cluster2++⋅OH.The hydroxyl radicals produced (Eq. 8) may be cleaned by the process in Eq. (6). The superoxide ions produced in Eq. (5) are partially involved in the SOD reaction as we shall discuss in the next part.The SOD processFor the process of SOD, we assume that the clusters are involved in similar mechanisms to the general catalytic scheme of SOD reaction [Eq. (6)]:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{O}}_2^{ \cdot - } + {\mathrm{Cluster}}^{2 + } \to {\mathrm{O}}_2 + {\mathrm{Cluster}}^ +,$$\end{document}O2⋅−+Cluster2+→O2+Cluster+,10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{O}}_2^{ \cdot - } + 2{\mathrm{H}}^ + + {\mathrm{Cluster}}^ + \to {\mathrm{H}}_2{\mathrm{O}}_2 + {\mathrm{Cluster}}^{2 + }.$$\end{document}O2⋅−+2H++Cluster+→H2O2+Cluster2+.Equation (9) shows the release of oxygen, which complete the CAT process, and it is also the initialization of SOD process. The reduced cluster, Cluster (I), may also be involved in both CAT and SOD process, which depends on the concentration of different components. Equations (9) and (10) are combined to the total reaction:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2{\mathrm{O}}_2^{ \cdot - } + 2{\mathrm{H}}^ + \to {\mathrm{H}}_2{\mathrm{O}}_2 + {\mathrm{O}}_2.$$\end{document}2O2⋅−+2H+→H2O2+O2.Please note that Eq. (10) is not the end of the reaction, but the end of SOD. The hydrogen peroxide can be decomposed by the CAT process mentioned above. Furthermore, the superoxide ions may react with the hydrogen peroxide as the following reaction:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{O}}_2^{ \cdot - } + {\mathrm{H}}_2{\mathrm{O}}_2 \to \cdot {\mathrm{OH}} + {\mathrm{OH}}^ - + {\mathrm{O}}_2.$$\end{document}O2⋅−+H2O2→⋅OH+OH−+O2.The hydroxyl radicals are produced in such reaction, but the reaction is slow to be biologically significant. So in this work, we do not consider such reaction and the production of hydroxyl radicals.In vitro experimentsMouse hippocampal neuronal HT22 cells were obtained from the Institute of Radiation Medicine, Chinese Academy of Medical Sciences, and Peking Union Medical College and employed in all the cellular experiments. Mouse microglia BV2 cells and mouse astrocytes-cerebellar MA-c cells were obtained from Tianjin Huanhu Hospital. Cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) (Gibco), supplemented with 10% fetal bovine serum (FBS, BI) at 37 °C with 5% CO2. In all, 100 U/mL penicillin and 100 mg/mL streptomycin sulfate (Solomo) were applied according to the growth state.Cytotoxicity assayHT22 cells (2 × 103), BV2 cells (3 × 103), and MA-c cells (4 × 103) were seeded in 96-well plates filled with 0.01 M PBS (Gibco) at the border in 100 µL medium overnight. The culture medium was replaced by different doses of Au25, Au24Cu1, or Au24Cd1 dissolved in the DMEM, and then cells were incubated for another 24 h. Wells were washed with 0.01 M PBS once, and the medium were replaced by fresh culture medium with serum-free DMEM. Cell cytotoxicity was determined by SPSS 19 MTT assay at the MTT concentration of 5 mg/mL for 2.5 h and detected at optical density (OD) 490 nm.Cell viabilityHT22 cells (2 × 103) were cultured in the 96-well plate in 100 μL culture media. When reaching roughly 60% confluence in each well, cells were stimulated by 100 µM H2O2 for 6 h. Then the culture media were substituted by fresh media containing Au25, Au24Cu1, or Au24Cd1 at different doses, and cells were incubated overnight. The plates were washed with PBS and incubated with 5 mg/mL MTT for 2.5 h. Cell viability was determined by MTT assay and analyzed at OD 490 nm.Measurement of intracellular oxidative stressHT22 cells (2 × 105) were cultured into 6-well plate in 2 mL culture medium. HT22 cells were grown to 60% confluence and treated for 6 h under 100 µM H2O2 conditions. The solution was replaced by fresh culture medium with 10% FBS containing 6 ng/µL Au25, Au24Cu1, or Au24Cd1 clusterzymes and cells were incubated for another 18 h. Then cells were incubated with 25 µM DHE (Beyotime, S0063) for 25 min at 37 °C in the dark to determine O2•– level. After 25 min, the culture medium containing DHE was removed, and the wells were washed with 0.01 M PBS. For •OH levels, the cells were incubated with 50 µM HPF solution (Sigma-Aldrich, H4290) for 25 min at 37 °C in the dark. Intracellular oxidative stress was captured by using a fluorescence microscope (EVOS, AMG), collected data by a FACS flow cytometer (BD AccuriTM C6), and used Flowjo 10.6.2 for quantitative analysis of free radical. Cells were gated based on size and granularity by forward and side scatter (SSC-A versus FCS-A). Then cell gate is analyzed for fluorescence intensity to determine the scavenging ability of clusterzymes. Among samples, control is recognized as the negative group, and H2O2 is referred as the positive group.In vivo treatmentAll animal procedures were approved by the Institute of Radiation Medicine, Chinese Academy of Medical Sciences, and Peking Union Medical College. Procedures were applied to minimize the number of animals used and the pain mice suffered.Animal modelsMale C57BL/6J mice were purchased from SPF (Beijing) Biotechnology Co., Ltd. Mice were housed in a constant temperature (21–23 °C) and animal humidity environment (45–60%) with a 12-h light–dark cycle. Food and water were available ad libitum. Surgery was performed after 1 week of transportation to adapt to the environment. Controlled cortical impact (CCI) models were conducted on adult male C57BL/6J mice (7–9 weeks, 21–23 g). C57BL/6 mice were assigned to the Sham (n = 37 per group), TBI, TBI+Au25, TBI+Au24Cu1, and TBI+Au24Cd1 groups (n = 49 per group) randomly. Mice were anesthetized with 10% chloral hydrate (10 mg/kg) by intraperitoneal injection. The period of anesthesia was judged through skin pinching reaction and toe stimulation reaction. The surgery was conducted when the mice were in the deep anesthesia stage. Mice were fixed in a stereotaxic frame and the scalp was cut to expose the skull. The craniotomy was performed by drilling the skull above the right side of the parietal–temporal cortex in the circle of 2 mm in diameter. The circular lesion (coordinates; between bregma and lambda in the parietal bone centered at 2 mm lateral from the sagittal suture) was produced by exposing the cortex through removing the bone flap. A controlled cortex impact driven by an electromagnetically CCI injury device (eCCI-6.3, Custom Design & Fabrication, Inc.) was made with an impactor of 5 m/s velocity, 0.61 mm depth, 150 ms duration, and 20° angle of dura mater on the vertical axis. The scalp was sutured together carefully. Then the mice were kept on a heated blanket under control until being recovered from anesthesia. The clusterzymes were intravenously injected into the CCI mice at the dose of 50 mg/kg dissolved in 0.01 M PBS. An injection volume of 200 µL was used for the Sham, TBI, and other groups. All mice were marked, classified, and put into the cages (5 mice/cage) under an specific pathogen-free-level environment. Sham-injured groups (control) received the same craniotomy and intravenously injected 0.01 M PBS without CCI injury. CCI-induced TBI groups received the same injury and intravenously injected 0.01 M PBS as the vehicle group. All animals fully recovered from surgical procedures and gradually gained weight after surgery. Mice were treated with cervical dislocation after finishing related animal experiments. Brain tissues and other organs (heart, liver, spleen, lung, kidney, bladder, and testicular) were taken out on days 1, 3, and 7 post injury. MWM tests were conducted on days 13–18 and 28–32 post injury.Ex vivo treatmentBlood–brain barrier (BBB) penetrationBrain tissues were harvested from mice with CCI injury by flushing blood from blood vessels through the heart with cold PBS at the time points of 1, 4, 12, and 24 h post injection (n = 3 per group per time point). Brain tissues were weighed and detected for the Au element by ICP-MS to evaluate the BBB penetration.Oxidative stress and inflammatory levelsAt days 1, 3, and 7 after the TBI or Sham operation (n = 5 per group), mice were cleaned with 10 mL PBS perfusion, and brain samples were rapidly harvested. Homogenates were centrifuged at 10,000 × g for 10 min, and the supernatants were saved at −80 °C for preparation. Supernatant protein concentration was measured and used for oxidative stress- and inflammation-related cytokine quantification with enhanced BCA Protein Assay Kit (Beyotime, P0010). Oxidative stress-related factors MDA, SOD, GSH/GSSG, and H2O2 were detected with lipid peroxidation MDA Assay Kit (Beyotime, S0131S), total SOD Assay Kit with WST-8 (Beyotime, S0101M), GSH and GSSG Assay Kit (Beyotime, S0053), and Hydrogen Peroxide Assay Kit (Beyotime, S0038). ELISA kits for IL-6 (Abcam, ab100712), IL-1β (Abcam, ab197742), and TNFα (Abcam, ab208348) were used to detect inflammation levels. These assays were carried out according to the instructions provided by the manufacturer. Each sample was detected twice at least and analyzed using the Microsoft Excel 2010 software.ImmunostainingMice brain samples were fixed in 4% paraformaldehyde (PFA) for 24–48 h, embedded in paraffin, and mounted on slides (4 µm coronal sections). The tissue slices were dewaxed twice in xylene for 10 and 5 min and dehydrated in a gradient ethanol solution (100, 95, 80, and 70%). Slices were rinsed three times and 2 min each with PBS. Antigen retrieval was performed in citrate antigen retrieval solution (C1032, Solarbio, China) at 95 °C for 10 min in a pressure cooker. After cooling naturally in the retrieval solution, slices were rinsed with PBS, followed by blocking with 5% bovine serum albumin at room temperature for 2 h, and the excess liquid was shaken off. The primary antibodies of target cells and cytokines were added at 4 °C overnight. The primary antibody information is as follows: anti-TNFα antibody (1:150, Abcam, ab183218), anti-IL-6 antibody (1:200, Bioss, bs-0782R), anti-IL-1β (1:200, Bioss, bs-0812R), antibody anti-NeuN antibody (1:800, GeneTex, GTX00837), anti-Iba1 antibody (1:300, Abcam, ab48004), and anti-GFAP (1:400, Abcam, ab90601). The primary antibody was removed and slices were rinsed with PBS three times 3 min each. Indicated fluorescence-labeled secondary antibodies were added and incubated at room temperature for 1 h in dark. The secondary antibody information is as follows: CoraLite488-conjugated Affinipure Donkey Anti-Rabbit IgG (H+L) (1:500, Proteintech, SA00013-6), Goat Anti-Chicken IgY H&L (Alexa Fluor® 647) (1:1000, Abcam, ab150171), Donkey Anti-Sheep IgG H&L (Alexa Fluor® 647) (1:1000, Abcam, ab150179), and Donkey Anti-Goat IgG H&L (Alexa Fluor® 647) (1:1000, Abcam, ab150131). Finally, slices were mounted with anti-fade mounting medium with 4,6-diamidino-2-phenylindole (S2110, Solarbio, China) and photographed with fluorescence microscope (EVOS, AMG). For immunohistochemistry staining, the primary antibody information utilized is as follows: anti-TNFα antibody (1:200, Abbkine, ABP0127), anti-IL-6 antibody (1:200, Proteintech, 66146-1-lg), and anti-IL-1β (1:200, Abbkine, ABP52932). After rinsing with PBS, the biotinylated secondary antibody was initially applied for 30 min, after reaction enhancer in Universal Two-step Detection Kit (ZSGB-BIO, pv9000) was used for additional 30 min. 3,3’-Diaminobenzidine tetrahydrochloride hydrate was utilized for detection. Slides were then counterstained with hematoxylin to stain nuclei. Samples were captured by microscopy.MWM testsMWM performance was assessed on days 13–18 and 28–32 following injury (n = 7 per group). The hidden platform test was used to investigate spatial learning and memory. Water in the pool was maintained at 25 °C (±1 °C), made opaque by milk. Before spatial learning, visual discrimination learning was performed to determine whether the vision of mice was normal. In this procedure, each animal performed one trial where the platform was placed above the water level without recording. Spatial learning was assessed across repeated trials for 5 days approximately at the same time each day between 13:00 and 18:00. A circular stainless steel tank 122 cm in diameter and 51 cm in height on both sides with non-reflective interior surfaces was used. The water maze was divided into four quadrants (I–IV), and the platform was set in the center of quadrant I. Mice were given four trials a day with an inter-trial interval (ITI) of 60 min using a starting position randomly within four positions. The procedures of MWM followed a previously published protocol76. Each trial was limited within 60 s with an ITI of 15 s. The mouse failing to find the platform hidden behind the water within this time limit was allowed for 15 s to be placed on the platform or guided to learn. The trials were repeated for 5 days, and the probe trials to evaluate long-term spatial memory were administered for 60 s on day 6 at a new starting position with the platform removed. Latency to locate and rest on the hidden platform and the distance traveled (path length) to the hidden platform were recorded for spatial learning trials. During the probe trials, the percentage of time spent in the missing platform quadrant and the number of platform location crossings were recorded to analyze the search strategy of mice in each group.Western blottingTissue samples (n = 3 per group per day) were lysed, operated on ice, and extracted protein from brain homogenates in radio-immunoprecipitation assay lysis buffer (strong) (CWBIO, CW2333) containing protease inhibitors at 95 °C for 5 min. Tissue extract supernatant protein concentrations were determined by a BCA assay (Beyotime, P0010). Sodium dodecyl sulfate-polyacrylamide gel electrophoresis was performed to resolve protein lysates (50 µg) before protein lysates were transferred onto nitrocellulose membranes (CWBIO, CW0022S). Antibodies specific for TNFα, IL-6, and IL-1β were used: anti-TNFα (1:1000, Abcam, ab34674), anti-IL-6 (1:1000, Abcam, ab7737), anti-IL-1β (1:1000, Abcam, ab234437), and β-actin antibody (1: 2000, Sigma-Aldrich, A5441). Full scans of all the blots used in this work are provided in Source data.Pharmacokinetic and toxicological studiesThe pharmacokinetic parameters of clusterzymes were measured on 7–9 weeks (21–23 g) male C57BL/6J mice. Mice were subject to intravenous injection at a dose of 50 mg/kg in 200 µL volume to evaluate the biodistribution (n = 3 per group), blood half-life (n = 3 per group), and excretion (n = 3 per group) of different clusterzymes. Mouse organs (heart, lung, liver, spleen, kidney, muscle, bladder, testicles, intestine, and brain) were collected, washed with PBS, and weighed 24 h post injection. To determine the half-life of clusterzymes, blood was collected from the retro-orbital sinus at 2 min, 12 min, 1 h, 1.5 h, 3 h, 5 h, 8 h, 24 h, and 48 h, and the volume was 50 µL. The elemental Au in organs and blood was quantified with ICP-MS. Standards were prepared and counted along with tissue samples to calculate the percentage-injected dose per gram of tissue (%ID/g). Drug excretions of clusterzymes were determined as well. Stool and urine were collected within 48 h and detected for Au element. Hematology and blood biochemistry panels were detected on the day 7 post injection. Blood samples were obtained from retro-orbital sinus and saved in tubes with K2EDTA for testing. Blood samples for biochemistry analysis were left to stand for 30 min and then centrifuged twice at 3500 rpm for 15 min. All organs were collected and fixed in 4% PFA for 24–48 h, embedded in paraffin, and mounted on slides (4-µm coronal sections). Slides were stained through hematoxylin–eosin staining to observe the toxicity of clusterzymes in major organs, including the heart, liver, spleen, lung, kidney, and brain.Quantitative analysis of immunostainingQuantitative image analysis of the immunofluorescence for GFAP, Iba-1, and NeuN cells were performed on five cerebral cortex areas of each 4 brain slices taken with the ×40 objective (n = 3 mice per group). Immunofluorescence intensity was calculated using the threshold method and defined as the average number of pixels per slice by the ImageJ software, then divided by the area (mm2) in the imaged field with the average background subtracted77,78. For quantification of the immunofluorescence double staining, the co-expressed cells in the five regions of the cortex of each 4 brain slices were counted under a microscope (EVOS, AMG) at ×400 magnification (n = 3 mice per group). The results are expressed as an average number of positive cells per unit area (mm2) of each slice79,80. For the quantitative analysis of inflammatory factors in immunohistochemistry, the investigators who were blinded to the experimental groups randomly collected five high-power field images at ×400 magnification in cerebral cortex areas under a microscope (EVOS, AMG) of each animal (n = 3 mice per group)80. The cytoplasmic staining areas that showed light yellow or brownish yellow were selected as positive cells, and the expression of inflammatory factors was quantified by the average count of positive staining cells per animal.Statistic methodsData are presented as mean ± standard deviation (SD) or standard error of the mean (SEM). For multiple comparisons, one-way analysis of variance (ANOVA) was performed using the SPSS 19 software to assess difference in means among groups and compared with the Sham and TBI groups, analyzed by ANOVA.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileReporting Summary | nature communications | [
"Article"
] | [
"Enzyme mechanisms",
"Enzymes",
"Neuroimmunology",
"Biocatalysis",
"Organometallic chemistry"
] | catalytic activity selectivity artificial enzymes for biomedical advanced diagnostics9 studies oxidase peroxidase-like activities of noble metals17 Gold-based materials versatile enzyme-like activities nuclease glucose oxidase peroxidase catalase superoxide dismutase Michaelis–Menten constant (Km) to H2O2 substrate gold nanoparticles below 1 mM catalytic activity weak19 Pt-based materials high catalytic activity good H2O2 substrate affinity when Km up to 16.7 mM14 modulation selective catalysis combination other catalysts21 metal oxides potentials as enzyme mimetics22 Fe3O4 nanoparticles display POD-like limited by affinity to H2O2 substrate ~154 mM maximal reaction rate (5.9 μM/min Mn3O4 nanoparticles exhibit SOD- CAT- glutathione peroxidase-like activities maximum reaction rate 6–125 mM/min inferior to natural enzymes25 development of catalytic artificial enzymes with exceptional activity selectivity stability challenge brain injuries involve enzyme-related catalytic processesunclear catalytic route(s neuroinflammatory responses brain injuries trigger multi-enzymatic reactions free radicals bioactive exploration artificial enzymes catalytic routes selectivity relationship between oxidative stress molecular pathways catalysis30–32 Atomic-level catalysts for improved catalytic activity modulated selectivity Fe- Pt-based single-atom nanozymes Au contains transition metal electronic states rich electronic energy levels basis for designing atomic-scale enzyme uncontrollable syntheses complicated spatial coordination difficult to reveal electronic structures catalytic activity prevent understanding catalytic mechanisms atomic Au-based clusterzyme designed atomic precision ultrahigh catalytic activity superiority over natural antioxidants enzymatic selectivity achieved via single-atom substitution modulating Cu or Cd active site promising artificial enzyme for treatment neuroinflammation properties of different from 3-mercaptopropionic acid (MPA)-protected Au25 clusterzyme defined by unambiguous atomic configuration geometry structurehydrodynamic size Au25 2.0 nm scattering zeta potentials clusterzymes −35 mV ultrasmall size good colloid stability absorption at 450 nm Au25 unique interband single-atom substitution Cu or Cd 2–3-nm shift insignificant influence optical properties 1b Electrospray ionization–mass spectra m/z peak at ~2271 [Au25MPA18−3H]3– substitution Cu Cd m/z peak shifts to ~2226.6 ~2243 coupled plasma-mass spectrometry ratios Cd Cu to total metal 5 and 4% introduction single atoms X-ray spectroscopy confirms Au (0) dominant state in clusterzymes 4) atomic configuration extended X-ray absorption fine structure) spectra at Au, Cu Cd edges recorded. 1d 5 6) L3 edges Au clusterzymes higher white-line intensities than Au foilascribed larger surface area alloying effects partial oxidation more d-band vacancies nanoscale surface molecule interactions absorption edges Au clusterzymes at ~11,920 eV 2p → 5d electronic transition reduced unoccupied valence d-states increase intensity MPA-protected Au24Cu1 Au24Cd1 density 5d electrons decreased atom substitutions Cu Cd 4d electrons k-space oscillations Au25 clusters Au foil Supplementary Fig. 5. k-space Au foil fcc oscillation patterns absent Au clusterzymes small core sizes investigated X-ray absorption spectra Cu Cd foils atomic counterparts differences single atoms bulk metals doping sites Cu Cd atoms fitting analysis EXAFS data Cu Cd Figure 1d R space EXAFS data Cu K edge Au24Cu1 one major peak 1.6–5.0 Å corresponds scattering path photoelectron waves X-ray absorbing Cu atom neighboring S atoms IFEFFIT program peak EXAFS parameters fitting Supplementary Table 1. Cu-S coordination number) 1.9 ± 0.2 Å.value close to 2 Å replacement Au25 by Cu atom oligomer site EXAFS data Cd L3 edge Au24Cd1 peak 1.6–5.0 Å Cd-S CN 2.3 ± 1.7 Å close to CN bond S oligomer site Au25 Cd atom substitution (Fig. 1e. 1Structural Au25 clusterzymes Structure Au25 Cu- Cd-substituted Au24Cu1 Au24Cd1 UV-vis absorption spectra electrospray ionization mass spectra Au25 before after Cu Cd substitutions magnification absorption spectrum ~670 nm absorption redshifted after introduction Cu Cd decreased band gaps red line simulated isotope distribution of [Au25MPA18-3H]3− [Au24Cu1MPA18-3H [Au24Cd1MPA18-3H Cu K-edge Cd L3-edge-EXAFS spectra fitting R space Au24Cu1 Au24Cd1 surrounding atoms adjacent Cu Cd atoms properties tested antioxidative properties clusterzymes ABTS method Negligible antioxidant activity for pure MPA-protected Au25 clusterzymeSingle-atom substitutions with Cu or Cd induce antioxidant activity concentrations (Fig. 2b). Ag Cu Zn Er Pt Cd single-atom Au24Cu1 Au24Cd1 show highest activity optimal Au24Cu1 Au24Cd1 rapid responses high reaction rates (Fig. Au24Cu1 Au24Cd1 41 48 times higher antioxidant activity than Au25 Au24Cu1 Au24Cd1 137 160 times higher than trolox 7.5 9 times higher than anthocyanin reaction rates Au24Cu1 at 10 14 μM/s 8–11 times higher than Au25 or 38 51 higher than trolox substitution with one atom Cu or Cd foremost activity Fig atomically precise gold clusters Au25 Au38 oxidation catalytic antioxidant activities rarely reported ultrahigh antioxidant activity fast kinetics via atom substitution.Fig. total antioxidant capacity of clusterzymes reaction mechanism ABTS assay Concentration-dependent time-dependent investigation of ABTS• Au25, Au24Cu1 Au24Cd1Comparison antioxidant capacities reaction rates Au25 Au24Cu1 Au24Cd1 natural antioxidants performance improved after substitution with Cu or Cd = 3 experiments mean ± Au24Cu1 Au24Cd1 show 137 160 times higher activity than trolox 7.5 9 times anthocyanin-like properties catalytic profile Fig. 3a investigated GPx-like activity Au25 Au24Cu1 Au24Cd1 at concentration 10 ng/μL Au25 tendency GPx-like activity maximum reaction rate 0.47 mM/min higher than 0.34 mM Au24Cu1 0.10 mM/min Au24Cd1 higher than Mn3O4 nanoflowers (0.056 mM/min Co/PMCS (0.013 mM/min turnover frequency) value Au25 320 min−1 4.7 times higher than Au24Cd1 high GPx-like activity Au25 lipid peroxides oxidative damages CAT-like activity studied at concentration 20 ng/μL Fig. 3d maximum reaction rate Au25 0.074 mM/min introduction Cu single atom 4.7-fold increase 0.35 mM/min CAT-like catalytic preferenceTOF value Au24Cu1 CAT activity 116.7 min−1 higher than Pd octahedrons (1.51 SOD activity Au25 31% substrate Cd atom substitution increases inhibition 89% SOD selectivity (Fig. results suggest enzyme-mimicking preferences clusterzyme Au25 GPx Au24Cu1 Au24Cd1 CAT SOD structures before after reaction H2O2 suggest unchanged Figs 10 11 work focused atomic substitutions Au25 noble metals reactions hydrogen CO work hypothesized Au25 unique enzyme activity single-atom substitution with non-precious metals Cu Cd. 3Enzyme-mimetic properties ROS scavenging activity Au25 clusterzymes map enzymatic activities radical scavenging abilities catalytic selectivity Au25 GPx activity Au24Cu1 CAT-like activity Cu Au24Cd1 SOD-like activity Cd active site unique catalytic selectivity GPx CAT SOD-like activities Au25 Au24Cu1 Au24Cd1 scavenging activities Au25 Au24Cu1 Au24Cd1 clusterzymes for •OH O2•− studied ESR spectroscopyBMPO ROS-capturing agent sources •OH O2•− H2O2 KO2 scavenging efficiencies (n = 3 experiments mean scavenging free radicals clusterzymes investigated scavenging •OH electron spin resonance (ESR) 5-tert-butoxycarbonyl-5-methyl-1-pyrroline N-oxide (BMPO) trapping agent ESR signal •OH strong BMPO control excessive •OH minor decrease Au25 weak scavenging efficiency (Fig. Au24Cu1 diminishes ESR signals (-like 3d scavenging O2•− clusterzymes investigated ESR signal strong decreases after Au25 Au24Cu1 surplus residues ESR signal O2•− disappears Au24Cd1 superior SOD-like activity tested free-radical scavenging capability clusterzymes reactive nitrogen species) •NO ONOO– DPPH• Au24Cd1 robust scavenging efficiency against DPPH• ESR Au24Cd1 best scavenging capability toward •NO low concentration 2.7 ng/μL Au25 ignorable activity Au24Cd1 Au24Cu1 higher scavenging efficiency toward ONOO− than Au25Au24Cd1 selective against RNS than Au24Cu1 Au25 insignificant catalytic activity high selectivity for enzymes radicals from single-atom substitutions of Cu Cd redistribution surface electrons electronic structures states.DFT calculations catalytic density functional theory (DFT) catalytic selectivity quantum properties adopt gold core Au25 protected by ligands connected core via S atoms catalytic behavior intermediate states chemical ligand unit -SCH2CH2COOH simplified to -SCH3 DFT optimization confirms stability modeled cluster symmetry gold cluster replacements metallic atoms into three categories oligomer surface core core replacements Figure 4a surface sites Au13 core oligomer site replacement oligomer replacement common form DFT simulations surface replacement possibility oligomer replacement matches EXAFS results lower CN than surface replacement optimized structure clusters associated CNs generated within during catalytic process averaged simulated CN values agree with experimental values confirm oligomer replacement (Supplementary Table 2) theoretical investigations on surface in appendix Figs. 16–19 Tables 3 4) focus on oligomer replacement catalytic efficiency4DFT calculations catalytic selectivity atomic doping surface oligomer replacement Doped atom change bonds rigid Cu-S flexible Cd-S bond flexibility Cd allows angular motion Mechanism energies profiles catalytic process SOD CAT processes black dotted line catalytic products connected other processes Energy profiles geometry structures intermediate states SOD CAT process lower panel intermediate structures transition structures surface replacement expansion oligomer replacement causes bending different normal S-Au-S chain aligns straight line Fig doped Cu shrinks S-X-S chain doped Cd extends typical bond length S-Au 2.3 Å S-Cu S-Cd bonds 2.2 2.55 Å Fig. 4b Fig. 21. bent chain distances Cu/Cd atoms surface core comparable 3.1 Å similarity Cu Au atoms S-Cu-S relative positions Cu S atoms changed dynamics catalytic procedures position doped Cd atom affected local environment small chemical units Figs. 22 excellent performance clusterzymes CAT SOD reactions reaction pathways Fig. 4cCAT reaction degradation hydrogen peroxide decomposition H2O2 multiple chemical stages (Fig. 4d, e). SOD assumed clusters involved similar mechanisms scheme SOD release oxygen completes CAT SOD occurs simultaneously mutually permeated reduced cluster Cluster(I), involved in CAT SOD processes depends on concentration components Arrhenius equation search transition states ground states molecules estimate activation energies evaluated catalytic efficiencies energy profiles in Fig. 4f, g agree with behaviors clusterzymes reactions multiple steps rate dominated by slowest step largest activation energy SOD process Au24Cu1 ground state electrons intermediate structure triplet state DFT simulations high activation energy 131.2 kJ/mol related to spin matching issues selective superoxide ions CAT processes critical step decomposition of (cluster...OOH)2+ Au24Cd1 higher activation energy (71.3 kJ/mol) reduces efficiency simulations explain SOD–CAT-selective behaviors doped clusterzymes details reaction pathways in results DFT simulations show insights catalytic mechanismsclusters substituted atom active site catalytic process changes geometry two mechanisms SA (simple adhesion) MA (bond modulated adhesion) distances between adsorbed molecule/ion metal atoms designate roles substituted atom SA mechanism seen in Au24Cu1 firmness S-Cu-S oligomer Cu atom rigid (Fig. 4f, first step SOD process catalyzed by Au24Cd1 catalytic process includes distance change of small units changes distance between active site small units SOD processes orientation change main CAT processes MA mechanism in Au24Cd1 second stage (Cluster(I) SOD CAT processes bond modulation position change Cd atom from oligomer plane until third S atom stops S-Cd bonds changed (Fig. 4f, characteristics transition states involve rotation of superoxide ion position adjustment of doped Cd atom motion Cd atom perpendicular direction oligomer plane small unit joins doped cluster Cd atom leaves oligomer plane CN value Cd atom larger when neighbor of three S atomsCd atom starts original state = 2.55 2.55 passes transition 2.63 2.85 arrives intermediate state 2.57 2.60 angular motion terminated Fig. 24). distance oxygen atoms expanded normal completion catalytic procedure Tables 5 6) similar procedure Au24Cd1 adhesion OOH first stage CAT process unique process allows doped Cd atom active self-modulated wide spatial range firm Cu atom good performance SOD process.Modulation clusterzymes cell nerve cell lines (HT22 BV2 MA-c) measured 3-[4,5-dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide) assay (Fig. 5a Fig. Au25 Au24Cu1 Au24Cd1 acceptable biocompatibility Cell survival H2O2-stimulated neuron cells performed incubation Au25 Au24Cd1 Au24Cu1 Fig. 5b clusterzyme treatment improve viability neuron cells correlation oxidative stress neuron viability reactive oxygen species (ROS), •OH O2•− quantified detected FACS flow cytometer fluorescence microscope H2O2 stimulation elevates fluorescence signal excessive •OH O2•−clusterzymes decrease ROS signals Au24Cu1 best clearance against •OH. 5c Au24Cd1 best clearance for O2•− selectivity (Fig. 5d f mouse models traumatic brain injury (TBI effects clusterzymes Fig. 5i–l indicators malondialdehyde H2O2 SOD GSH/GSSG TBI severe day 1 post injury alleviated 3 days improved 7 days post decrease SOD GSH/GSSG levels TBI by clusterzymes recoveries 7 days after treatment (Fig. 5i Au24Cd1 better recovery in SOD in vitro SOD-like activity (Fig. 3) by-products oxidative stress lipid peroxides H2O2 following TBI severe oxidative damage (Fig. 5k Au24Cu1 Au24Cd1 inhibit production harmful molecules Au25 alters TBI-induced increase O2•− produced by injuries production lipid peroxides H2O2. Au24Cd1 diminished SOD high catalytic selectivity for O2•− decrease lipid peroxides H2O2Au24Cu1 lipid peroxides H2O2 early stage CAT-like activity •OH intermediate low after TBI onset accounts increasing clearance capability long term. 5Oxidative stress levels vitro vivo before after treatment clusterzymes HT22 cell viability clusterzymes = 5 mean ± cell viability H2O2 without treatment MTT assays = 6 per group mean Fluorescence quantification cell staining •OH O2•− flow cytometry = 3 per Data mean ± SD compared Con H2O2 groups one-way ANOVA two-sided LSD test p values images intracellular •OH O2•–) levels 100 μM H2O2 without treatment HPF DHE probes Au24Cu1 better scavenging •OH Au24Cd1 specificity O2•− selectivity Indicators oxidative stress SOD GSH/GSSG MDA H2O2 TBI mice without treatment clusterzymes 1 3 7 days post injury (n = 5 per Data mean ± SEM compared Sham TBI groups one-way ANOVA LSD test p values Experiments repeated three times similar resultseffects clusterzymes on neuroinflammation examined western blots analysis. 6a levels interleukin)-1β IL-6 upregulated following TBI 1 day post injury strong Au24Cd1 downregulates IL-1β IL-6 anti-neuroinflammation effect Au25 minor downregulation TBI upregulation tumor necrosis factor-α (TNFα) 1 day post injury Au24Cu1 expression TNFα superior efficacy over Au25 Au24Cd1 (Fig. cytokines TBI suppressed autoimmunity 7 days post injury significant differences in IL-1β IL-6 levels between Sham TBI groups (Fig. clusterzyme treatment cytokines close to normal level better suppression neuroinflammation Enzyme assay) Au24Cd1 Au24Cu1 inflammatory cytokines IL-1β Au25 alter inflammatory cytokine patterns. Au24Cd1 IL-1β IL-6 inflammatory responses Au24Cu1 TNFα modulation neuroinflammation immunofluorescence staining cerebral cortex Au24Cd1 Au24Cu1 decrease TBI-elevated expression levels IL-1β IL-6 TNFα (Fig. 6halleviating neuroinflammation Colocalization studies with markers for neurons microglia astrocytes performed in injured cortex day 3 post injury Figure 6h IL-1β produced by microglia after TBI similar with IL-6 TNFα Figs 26 analyses show microglia astrocytes activated neurons depleted after TBI. clusterzyme treatment nerve cells TBI-activated astrocytes recovered normal after treatment neuroinflammatory responses prevented 29–32) clusterzymes restore TBI-induced body weight loss behavioral tests studied by Morris water maze mice learned task during acquisition phase days 13–17 28–31 distance traveled latency to hidden platform with Au24Cu1 Au24Cd1 treatment decreased trial days 18 32 percentage time missing platform quadrant platform crossings reduced in TBI group to normal after Au24Cu1 Au24Cd1 treatment results reveal improvements learning ability spatial memory with Au24Cu1 Au24Cd1 treatment studied pharmacokinetics toxicology of clusterzymesclusterzymes in organs removed by kidney liver After 48 h ~80% total dose excreted most through kidney (>70% Fig 35). No significant changes in organs blood chemistry hematology renal-clearable clusterzymes cause biological toxicity in vivo 36–38) Artificial enzymes exhibit multiple enzyme-like catalytic activities with diversified Low catalytic activity due to limited electron transfers at atomic clusterzymes with single-atom substitutions overcome barriers with antioxidant activity nine times higher than anthocyanin reactive antioxidant molecular structures of clusterzymes elucidated distinguish catalytically active sites electronic structures reaction substituting atoms arranged into spatial location tuning electronic structures affecting catalytic interactions between host atoms Au substituting atoms Cu Cd induce coupled electron states influence catalytic selectivity71 GPx- SOD- CAT-like catalytic selectivity assigned to Au25 Au24Cd1 Au24Cu1 via modulated bond lengths to active center platform of clusterzymes generate selectivity against different moleculesemploying three selective clusterzymes established relationship oxidative stress neuroinflammation importance benefits TBI Au24Cd1 neuroinflammation IL-1β IL-629,72 Au24Cu1 reduces neuroinflammation TNFα-neuroinflammation ultrasmall size clusterzymes penetrate kidney barriers excreted renal hepatotoxicity multi-organ injuries clusterzymes influential biomedicine neuroscience.Fig. 6Inflammation levels brain after clusterzyme treatment Western blotting IL-1β IL-6 TNFα 1 3 7 days post TBI treatment = 3 per Western blotting analysis inflammatory factors points = 3 per samples experiment blots processed parallel Data mean SEM compared with Sham TBI groups analyzed one-way ANOVA two-sided LSD test Au24Cd1 inflammatory cytokines IL-1β IL-6 after injury Au24Cu1 expression TNFα ELISA analysis IL-1β IL-6 TNFα levels days 1 3 7 with without clusterzymes treatment (n = 5 perData mean ± SEM compared Sham TBI groups one-way ANOVA two-sided LSD test p values Immunofluorescence co-staining IL-1β microglia astrocytes neurons injured cortex 3 days post injury clusterzyme treatment analysis IL-1β+ expression positive cells pixels density Iba-1/NeuN cells cortex = 3 per Data mean SEM compared Sham TBI groups one-way ANOVA two-sided LSD test p values Experiments repeated three times similar results systemic single-atom substitution approach artificial enzymes MPA-protected Au25 clusters clusterzymes clusterzymes ultrahigh antioxidant activity 137–160 times higher natural trolox catalytic selectivity GPx CAT SOD nitrogen signaling molecules fine-tuned single-atom substitutions reaction pathways modulated Au24Cd1 Au24Cu1 lengths results Au24Cd1 decreases IL-1β IL-6 Au24Cu1 TNFα neuroinflammation chemicals available highest purity without treatmentGold chloride (HAuCl4·3H2O purchased from Sigma-Aldrich sodium hydroxide sodium borohydride (NaBH4) copper nitrate cadmium nitrate MPA from Aladdin Ultrapure water (18.2 MΩ·cm used experiments gold nanoclusters synthesized HAuCl4 (20 mM 0.25 mL MPA (5 mM 2 mL added water (2.35 mL stirred 5 min NaOH solution (1 M 0.3 mL added 0.1 mL NaBH4 solution 43 mg 10 mL 0.2 M NaOH dark Au25MPA18 collected 3 h aged 4 °C 12 h metal-substituted AuxM25-xSG18 Au atoms HAuCl4 (20 replaced by nitrate metal ions (Cu2+ Cd2+) 4% molar ratio 24:1) purification used ultrafiltration tubes 3 10 K 3500/min ligands lyophilized purified product absorption spectra recorded Shimadzu 3600 UV-vis-NIR spectrophotometer ESI-MS Bruker microTOF-Q system XANES EXAFS analyses tested Beijing Synchrotron Radiation Facilitymodule ARTEMIS IFEFFIT data XANES EXAFS74,75 Clusterzymes pressed pellets Au Cu Cd measured room temperature Au/Cu/Cd foil measured XPS metal element K-Alpha spectrometer monochromatic Al Kα X-ray source 300 W (ThermoFisher C 1s level 284.8 eV peak drift XPS peaks fitted XPSPEAK41 software transformed infrared spectra clusterzymes reactions recorded AVATR360 Spectrometer scan wavenumber 400 to 3800 cm−1 samples determined powder method Raman spectrum INVIA Reflex spectrometer 633 nm He-Ne laser Malvern Zetasizer nano ZS90 DLS hydrodynamic size zeta potential clusterzyme ICP-MS tested 7900 ICPMS (Agilent metallic elements clusterzymes scavenging process •OH •NO O2•– ESR spectrometer kinetic test CAT-like Dissolved Oxygen Meter LDO101 probe.Antioxidant free radical scavenging testsTotal capacity test clusterzyme contrast anthocyanin determined T-AOC Assay Kit (S0121 specification sampling methodsantioxidant capacity evaluated absorption at 414 nm adjusted concentration ABTS•+ extinction coefficient ɛ414 nm = 3.6 × 104 mol−1 cm−1) changing concentration H2O2. reflected by change absorbance at 414 nm monitored UV-vis spectrophotometer steady-state kinetic parameters determined varying concentration ABTS•+ in clusterzymes (5 ng/μL). maximum reaction velocity Michaelis–Menten constant) calculated Lineweaver–Burk equation.RNS scavenging capacity clusterzyme for 1,1-diphenyl-2-picrylhydrazyl radical (DPPH•). absorption wavelength at 510 nm 50 μM DPPH• 5 ng/μL clusterzyme dissolved in dimethyl sulfoxide water changes of absorption spectra with time 300–1000 nm determined.ONOO– scavenging preparation of ONOO− aqueous solution NaNO2 (5) H2O2 (5) stirred HCl NaOH 1.5) added 5 min yellowish ONOO− reserve solution diluted to 1.3 mM 50 5 ng/μL clusterzyme addedscavenging capacity ONOO− evaluated absorption spectra 250–400 nm scavenging 5 mM H2O2 solution prepared 10 mM-buffered saline) buffer hydroxyl radical generated UV-laser irradiation 5 min ESR spectrometer scavenging BMPO (50 mM) capturing agent spin adduct/•OH four peaks removal process •OH peak strength before after addition clusterzymes (2.7 ng/μL).O2•– scavenging testKO2 2.5 mM 18-crown-6 3.5 mM generation source stabilizer 25 mM 5-(diethoxyphosphoryl)−5-methyl-1-pyrroline-N-oxide (DEPMPO) spin capturing adduct/O2•– six peaks signal intensity before 2.7 ng/μL clusterzyme scavenging capacity O2•– scavenging testCarboxy-PTIO 10 μM 5 peaks capturing agent S-nitroso-N-acetylpenicillamine 250 μM •NO contributor spin adduct-PTI seven peaks scavenging capacity •NO evaluated peaks intensity types before after addition 2.7 ng/μL clusterzymes-like activity determined methodsabsorption peak H2O2 240 nm optical density decreases decomposition extinction coefficient (43.6 mM−1 cm−1 240 nm activity reaction solutions 53 μM H2O2 10 ng/μL clusterzyme 200 μL PBS H2O2 treated without 50 ng/μL clusterzymes 30 min photos decomposition kinetics Dissolved Oxygen Meter LDO101 probe solubility O2 reduced mg/L infusion Ar 5 mL PBS solution concentrations H2O2 (50–1000 mM) 20 ng/μL clusterzymes added solubility 10 s interference deducted H2O2 self-decomposition maximum reaction velocity Michaelis–Menten constant) calculated Lineweaver–Burk equation Origin 9.0 software.GPx-like activity clusterzymes determined 200 μM H2O2 2 mM GSH 200 μM NADPH 1.7 units/mL GR 10 ng/μL clusterzymes added 200 μL PBS neutral buffer activity evaluated absorbance 340 nm NADPH 6.22 mM–1 GPx-like activity = AClusterzyme)/ACon × 100%reaction kinetic analysis reflected by change absorbance at 340 nm monitored UV-vis spectrophotometer kinetic mode formation concentration substrate changed by adjusting H2O2 (100–700 μM), other conditions remained same maximum reaction velocity Michaelis–Menten constant) calculated using Lineweaver–Burk equation Origin 9.0 software.SOD-like activity of clusterzymes tested SOD Activity Assay Kit adding clusterzymes (0–10 ng/μL) different concentrations absorbance changes at 560 nm monitored with UV-vis spectrometer SOD-like activity.DFT calculationsTo catalytic effect on energy profile clusterzymes in intermediate structures adsorption energy change adsorbate attracted by cluster calculated by equation:1\mathrm = E{cata}} +{mol -=Ecata+mol− Ecata Emol denote energy cluster adsorbate Ecata+mol energy intermediate structure simulate activation energies differences between transition states energetically stable states.[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin-69pt}{document}\mathrm{act}}} =\mathrm{cata}} +\mathrm{mol{TS - E{cata}} +\mathrm{mol}}}{document}Eact=Ecata+molTS−Ecata+mol transition states estimate activation energy barriers saddle points energy surfaces[12pt]{minimal}{amsmath}{wasysym}}}}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document_{{\rm{cata}} +\rm{mol}}}{TS}}}\end{document}Ecata+molTS), searched using Berny algorithm CAT catalytic degradation hydrogen peroxidetotal reaction is3[12pt]{minimal{amsmath{wasysym{amsfonts{mathrsfs{upgreek}\oddsidemargin-69pt}{document}{\mathrm{H}}_2{\mathrm{O}}_2{H}}\end{document}2H2O2→2H2O+O2.In experiment observed performance clusterzymes CATs mechanism decomposition[12pt]{minimal}{amsmath}{wasysym{amsfonts{amssymb{amsbsy{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}\mathrm{H}}_2{\mathrm{O}}_2\end{document}H2O2 multiple steps pathscatalytical mechanism copper propose clusterzyme initiate reaction[12pt{minimal}\usepackage{amsmath}{wasysym{mathrsfs{upgreek}\oddsidemargin{-69pt}{document}\mathrm{cluster}}{2 +\mathrm{H}}_2\mathrm{O}}_2\left\mathrm{cluster}} {\mathrm{OOH}}}\right + +\mathrm{H}}{document}cluster2++H2O2→cluster⋯OOH++H+intermediate structure[12pt]{minimal}\usepackage{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin-69pt}{document}\mathrm{cluster}}\mathrm{OOH}}}cluster⋯OOH decomposes into three pieces cation superoxide ion original cluster extra electron Cluster (I):5\documentclass[12pt]{minimal}{amsmath}{wasysym}{amsfonts}{amssymb}{mathrsfs}{upgreek}\oddsidemargin}-69pt}{document}\left\mathrm{cluster}} {\mathrm{OOH}}}\mathrm{cluster}}\mathrm{O}}_2\mathrm{H}}{document}cluster⋯OOH+→cluster++O2⋅−+H+\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}{\oddsidemargin}{-69pt}{document}$\cdot {\mathrm{OH}} +\mathrm{O}}_2 - }_2 +{OH}}^ -{document}⋅OH+O2⋅−→O2+OH− superoxide ion hydroxyl scavenging (Eq. 6) or SOD reaction path reduced cluster disproportionation process hydrogen peroxide not divided\documentclass[12pt]{minimal}{amsmath}{wasysym}}-69pt}{document}\mathrm{cluster}}^ + +\mathrm{H}}_2\mathrm{O}}_2\mathrm{cluster}}\mathrm{OH}})^{2 + } +{OH}} -{document}cluster++H2O2→)2++OH−[12pt]{minimal}{amsmath}{upgreek}-69pt}{document}$\left\mathrm{cluster}}\mathrm{OH\right)^{2 + } {\mathrm{cluster}}^{2 + } + \cdot {\mathrm{OH}}.{document}cluster⋯OH2+→cluster2++⋅OH hydroxyl radicals (Eq. 8) cleaned by process Eq. (6) superoxide ions Eq. (5) partially involved in SOD reaction next part SOD clusters involved similar mechanisms catalytic scheme SOD reaction [Eq. (6)]\documentclass[12pt{minimal\usepackage{amsmath-69pt}\mathrm{O}}_2\cdot - } +\mathrm{Cluster}}^{2 +\mathrm{O}}_2 +\mathrm{Cluster}}^ +{document}O2⋅−+Cluster2+→O2+Cluster+\documentclass[12pt]{minimal}{amsmath{upgreek-69pt{document}\mathrm{O}}_2 - } +\mathrm{H}}^ +\mathrm{Cluster}}^ +\mathrm{H}}_2\mathrm{O}}_2 + {\mathrm{Cluster}}^{2 + }{document}O2⋅−+2H++Cluster+→H2O2+Cluster2+.Equation (9) shows release oxygen CAT process initialization SOD reduced cluster Cluster (I), involved in CAT SOD depends concentration components Equations (9) (10) combined total reaction:11\documentclass[12pt]{minimal}\usepackage{amsmath}{wasysym{upgreek}{\oddsidemargin}{-69pt}\begin{document}{\mathrm{O}}_2 +\mathrm{H}}{document}2O2⋅−+2H+→H2O2+O2.Please Eq. (10) not end reaction end of SOD hydrogen peroxide decomposed by CAT processsuperoxide ions react hydrogen peroxide\documentclass[12pt{minimal{amsmath{upgreek-69pt\mathrm{O}}_2}O2⋅−+H2O2→⋅OH+OH− hydroxyl radicals produced slow biologically significant consider reaction.In vitro experimentsMouse hippocampal neuronal HT22 cells Institute of Radiation Medicine Chinese Academy of Medical Sciences Peking Union Medical College experiments Mouse microglia BV2 cells mouse astrocytes-cerebellar MA-c cells Tianjin Huanhu Hospital Cells cultured Dulbecco’s Modified Eagle Medium) supplemented 10% fetal bovine serum) 37 °C 5% CO2. 100 U/mL penicillin 100 mg/mL streptomycin sulfate) applied growth state.Cytotoxicity assayHT22 cells (2 BV2 cells (3 × MA-c cells (4 × 103) seeded in 96-well plates 0.01 M PBS (Gibco) 100 μL medium overnightculture medium replaced by Au25 Au24Cu1 Au24Cd1 DMEM cells incubated 24 h Wells washed 0.01 M PBS replaced fresh serum-free DMEM cytotoxicity determined SPSS 19 MTT assay 5 mg/mL 2.5 h 490 nm viabilityHT22 cells (2 cultured 96-well plate 100 μL 60% confluence stimulated 100 μM H2O2 6 h substituted fresh Au25 Au24Cu1 Au24Cd1 incubated overnight plates washed PBS incubated 5 mg/mL MTT 2.5 h viability determined MTT assay analyzed nm intracellular oxidative stressHT22 cells (2 105 cultured 6-well plate 2 mL medium grown 60% confluence treated 6 h 100 μM H2O2 replaced fresh 10% FBS 6 ng/μL Au25 Au24Cu1 Au24Cd1 clusterzymes incubated 18 h incubated 25 μM DHE 25 min 37 °C O2•– level wells washed 0.01 M PBS incubated 50 μM HPF solution 25 min 37 °CIntracellular oxidative stress captured fluorescence microscope FACS flow cytometer Flowjo 10.6.2 analysis free radical Cells gated size granularity forward (SSC-A versus FCS cell gate analyzed fluorescence intensity scavenging ability clusterzymes control negative H2O2 positive vivo animal procedures approved Institute of Radiation Medicine Chinese Academy of Medical Sciences Peking Union Medical College animals pain modelsMale C57BL/6J mice purchased from SPF (Beijing) Biotechnology Co. Ltd housed constant temperature (21–23 °C) animal humidity (45–60%) 12-h light–dark cycle Food water ad libitum Surgery after 1 week transportation Controlled cortical impact) models on adult male C57BL/6J mice (7–9 weeks 21–23 assigned Sham TBI+Au25 groups randomly anesthetized 10% chloral hydrate (10 mg/kg) intraperitoneal injection anesthesia judged skin pinching toe stimulation surgery deep anesthesia stage fixed in stereotaxic frame scalp cut skull craniotomy skull parietal–temporal cortex 2 mmcircular lesion between bregma lambda parietal bone 2 mm lateral sagittal suture produced exposing cortex removing bone flap controlled cortex impact CCI injury device 5 m/s velocity 0.61 mm depth 150 ms duration 20° angle vertical scalp sutured mice kept on heated blanket until anesthesia clusterzymes intravenously injected CCI mice 50 mg/kg 0.01 M PBS injection volume 200 μL for Sham TBI other groups mice marked classified cages (5 pathogen-free-level environment Sham-injured groups craniotomy injected 0.01 M PBS without CCI injury CCI-induced TBI groups same injury injected 0.01 M PBS animals recovered gained weight after treated cervical dislocation Brain tissues organs spleen lung kidney bladder testicular taken days 1 3 7 post injury MWM tests days 13–18 28–32 vivo treatmentBlood–brain barrier (BBB) penetrationBrain tissues harvested CCI flushing blood cold PBS 1 4 12 24 h post injection 3 tissues weighed detected for Au element ICP-MSOxidative stress inflammatory days 1 3 7 after TBI operation 5 mice cleaned 10 mL PBS perfusion brain samples harvested Homogenates centrifuged at 10,000 × g 10 min supernatants saved at −80 °C Supernatant protein concentration measured oxidative stress BCA Protein Assay Kit Oxidative stress factors MDA SOD GSH/GSSG H2O2 detected lipid peroxidation MDA Assay Kit SOD GSH GSSG Hydrogen Peroxide ELISA kits IL-6 IL-1β TNFα inflammation assays Each sample detected twice analyzed Microsoft Excel 2010 brain samples fixed in 4% paraformaldehyde 24–48 h embedded in paraffin on slides (4 tissue slices dewaxed in xylene 10 5 min dehydrated ethanol solution (100 95 80 rinsed three times 2 min PBS Antigen retrieval citrate antigen retrieval solution 95 °C 10 min pressure cookercooling retrieval slices rinsed PBS 5% bovine serum albumin room temperature 2 h excess liquid shaken primary antibodies cells cytokines added 4 °C overnight antibody information anti-TNFα:150 Abcam anti-IL-6 Bioss anti-IL-1β-NeuN GeneTex anti-Iba1:300 Abcam ab48004) anti-GFAP (1:400 primary antibody removed slices rinsed PBS three times 3 min fluorescence-labeled secondary antibodies added incubated room temperature 1 hsecondary antibody CoraLite488 Affinipure Donkey Anti-Rabbit IgG) (1:500 Proteintech SA00013-6), Goat Anti-Chicken IgY H&L (Alexa Fluor® 647) Donkey Anti-Sheep IgG H&L Donkey Anti-Goat IgG H slices anti-fade medium-diamidino-2-phenylindole (S2110 Solarbio China photographed fluorescence microscope immunohistochemistry primary antibody anti-TNFα:200 Abbkine ABP0127) anti-IL-6 anti-IL-1β ABP52932) rinsing PBS biotinylated secondary antibody applied 30 min reaction enhancer Universal Two-step Detection Kit 30 min 3,3’-Diaminobenzidine tetrahydrochloride hydrate detection Slides counterstained hematoxylin Samples captured microscopy.MWM performance assessed days 13–18 28–32 injury (n = 7 per hidden platform test spatial learning memory Water pool 25 °C (±1 opaque milk visual discrimination learning vision mice animal trial platform above water level without recordingSpatial learning assessed trials 5 days same time between 13:00 18:00 circular stainless steel tank 122 cm diameter 51 cm height non-reflective surfaces used water maze divided four quadrants platform center quadrant I Mice given four trials day inter-trial interval 60 min starting position randomly four procedures followed protocol76 Each trial limited within 60 s ITI 15 s mouse failing find platform water allowed 15 s platform trials repeated 5 days probe trials spatial memory 60 s day 6 new starting position platform removed Latency locate hidden platform distance traveled recorded time spent missing platform quadrant platform location crossings recorded search strategy blottingTissue samples (n = 3 per group day lysed operated on ice extracted protein from brain homogenates radio-immunoprecipitation assay protease inhibitors 95 °C 5 min Tissue extract supernatant protein concentrations determined BCA assaySodium dodecyl sulfate-polyacrylamide gel electrophoresis protein lysates before nitrocellulose membranes Antibodies for TNFα IL-6 IL-1β used anti-TNFα anti-IL-6-IL-1β β-actin antibody Full scans blots Source data.Pharmacokinetic parameters clusterzymes measured 7–9 weeks (21–23 g male C57BL/6J mice intravenous injection 50 mg/kg 200 μL volume biodistribution blood half-life excretion clusterzymes Mouse organs lung liver spleen kidney muscle bladder testicles intestine brain collected washed PBS weighed 24 h post injection half blood collected retro-orbital sinus 2 12 1 h 1.5 3 5 8 24 48 h volume 50 μL elemental Au organs blood quantified ICP-MS Standards counted tissue samples percentage-injected dose per Drug excretions clusterzymes determined Stool urine collected within 48 h detected for Au element Hematology blood biochemistry panels detected day 7 post injectionBlood samples from retro-orbital sinus saved in tubes with K2EDTA for testing left 30 min centrifuged at 3500 rpm 15 min organs collected fixed in 4% PFA 24–48 h embedded in paraffin mounted on slides (4-μm coronal Slides stained hematoxylin–eosin staining toxicity clusterzymes in organs heart liver spleen lung kidney brain analysis immunofluorescence GFAP Iba-1 NeuN cells on five cerebral cortex areas 4 brain slices ×40 objective = 3 mice Immunofluorescence intensity calculated threshold method defined average pixels per slice divided by area (mm2) immunofluorescence double staining co-expressed cells five regions counted ×400 magnification results expressed average positive cells per unit area (mm2) inflammatory factors immunohistochemistry collected five high-power field images at ×400 magnification cerebral cortex cytoplasmic staining areas light yellow brownish yellow selected positive cells inflammatory quantified by average count positive staining cells per animalStatistic methodsData presented as mean ± standard deviation standard error multiple comparisons one-way analysis variance (ANOVA) performed SPSS 19 software difference compared with Sham TBI groups ANOVA.Reporting Nature Research Reporting Summary.Supplementary Review FileReporting Summary | 49.8 | 1.382308 |
10.1038/s41467-021-21043-4 | PMC7854640 | Nasopharyngeal carcinoma is a diverse cancer characterised by a heterogeneous microenvironment. Here, the authors use single cell sequencing to analyse the tumour microenvironment in 10 nasopharyngeal carcinoma tumours and identify different cell types including immune-suppressive T regulatory, tolerogenic dendritic, and exhausted CD8 T cells. | The heterogeneous nature of tumour microenvironment (TME) underlying diverse treatment responses remains unclear in nasopharyngeal carcinoma (NPC). Here, we profile 176,447 cells from 10 NPC tumour-blood pairs, using single-cell transcriptome coupled with T cell receptor sequencing. Our analyses reveal 53 cell subtypes, including tumour-infiltrating CD8+ T, regulatory T (Treg), and dendritic cells (DCs), as well as malignant cells with different Epstein-Barr virus infection status. Trajectory analyses reveal exhausted CD8+ T and immune-suppressive TNFRSF4+ Treg cells in tumours might derive from peripheral CX3CR1+CD8+ T and naïve Treg cells, respectively. Moreover, we identify immune-regulatory and tolerogenic LAMP3+ DCs. Noteworthily, we observe intensive inter-cell interactions among LAMP3+ DCs, Treg, exhausted CD8+ T, and malignant cells, suggesting potential cross-talks to foster an immune-suppressive niche for the TME. Collectively, our study uncovers the heterogeneity and interacting molecules of the TME in NPC at single-cell resolution, which provide insights into the mechanisms underlying NPC progression and the development of precise therapies for NPC. | IntroductionNasopharyngeal carcinoma (NPC) is a distinct type of head and neck cancer, which has been closely linked with the infection of Epstein-Barr virus (EBV)1. NPC has a remarkable ethnic and geographic prevalence, where high incidence rate of 15–50 cases per 100,000 people was reported in Southern China and Southeast Asia, as compared to 0.4 per 100,000 in western populations2. In general, patients with NPC are diagnosed with advanced stages largely due to non-specific symptoms. Radiotherapy is the primary treatment modality for patients with NPC because of the radiosensitive nature of its tumour cells2. Survival outcomes of patients with NPC improve substantially, reaching a 5-year overall survival rate of 85.6%3,4, mainly benefited from the evolution of radiotherapy techniques and the addition of platinum-based chemotherapy in patients with loco-regionally advanced disease. Nevertheless, more than 10% of patients develop recurrent and metastatic NPC2, for whom recent studies showed an overall response rate of 11.7% and 25.9–34% to targeted therapy with the addition of an inhibitor for epidermal growth factor receptor5 and immunotherapies using the immune checkpoint blockade6,7, respectively. These variations in treatment responses and survival outcomes indicate the heterogeneous nature of NPC.Individual genetic makeup and genomic instability foster genetic diversity of cancer cells that contribute to tumour heterogeneity. Genome sequencing analyses have revealed diverse profiles of somatic alterations in NPC tumours, with high mutational frequencies at CYLD, NFKBIA, TP53, and CDKN2A/B, as well as accumulated mutations in MHC class I genes and chromatin modification genes, which were associated with poor overall survival of the patients8,9. Apart from heterogeneous cancer cells, tumours exhibit another dimension of heterogeneity, which contain diverse normal cells creating the tumour microenvironment (TME) for the maintenance of cancer hallmarks10. Heterogeneous immune cells and stromal cells have been characterised using transcriptional profiling at single-cell resolution in several cancers, revealing that certain subtypes of immune cells and gene signatures in TME are important for tumour progression and sustained treatment responses11,12. Profound infiltration of lymphocytes has been observed in histological biopsies of NPC, amid other stromal cells and tumour cells of different morphology13,14. Moreover, high density of tumour infiltrating lymphocytes (TILs) was associated with favourable survival outcomes of patients with NPC15,16. However, the composition of diverse cell populations in the TME has not been well illustrated in NPC. Two recent studies demonstrated the presence of T cells with various functional states and different immune cells in NPC tumours, using single-cell transcriptome analysis17,18.In this work, we aim to provide a comprehensive global view of tumour heterogeneity of NPC, by analysing the single-cell transcriptional profiles of 176,447 cells from 10 treatment-naïve patients with NPC. With the combination of T cell receptor (TCR) repertoire information and individual tumour-blood sample pairs, we further characterise the clonality and migrations of T cells. In addition, we generate a potential cellular interaction network of cell populations in the TME of NPC.ResultsLandscape view of cell composition in tumour biopsy and PBMC in patients with NPCTo shed light on the complexity of tumour microenvironment (TME) in NPC, we performed single-cell RNA sequencing in combination of TCR repertoire sequencing on viable cells derived from tumour biopsies and matched peripheral blood mononuclear cells (PBMC) for 10 patients with EBV-positive NPC prior to any anti-cancer treatment (Fig. 1a, Supplementary Fig. 1a, b, and Supplementary Table 1). On average, we obtained more than 380 million sequencing reads for each sample, with the median of sequencing saturation (covering the fraction of library complexity) at 90.75% (75.90%–94.50%; Supplementary Table 2). After strict quality control filters (low expression of representative genes and inferred doublets; see Methods; Supplementary Fig. 1c and Supplementary Table 3), a total of 176,447 cells were identified from the 10 patients (including 82,622 and 93,825 for tumours and PBMC, respectively; Supplementary Data 1). We obtained about 1500 genes and 4950 unique molecular identifiers (UMIs) on average for each cell, indicating sufficient coverage and representative of transcripts (Supplementary Fig. 1d and Supplementary Data 1).Fig. 1The landscape profiling of single cells in NPC tumours and matching PBMC.a An experimental scheme diagram highlights the overall study design. Single viable cells were collected using flow cytometry sorting (FACS) and subjected for cell barcoding. The cDNA libraries of 5’-mRNA expression and TCR were constructed independently, followed by high throughput sequencing and downstream analyses. b UMAP plot of 176,447 single cells grouped into six major cell types (left panel) and the normalised expression of marker genes for each cell type (right panel). Each dot represents one single cell, coloured according to cell type (left panel), and the depth of colour from grey to blue represents low to high expression (right panel). c UMAP plot of the above single cells coloured according to their origins from peripheral blood or tumour (left panel), and the fraction of cell types originating from each patient (right panel). Each dot represents one single cell, coloured according to sample origin.Next, to define groups of cells with similar expression profiles, we performed unsupervised clustering analysis implemented in Seurat software19. The distribution of cell clusters for each patient matched well with that of other patients, suggesting that the potential variation of expression due to batch effect of sample processing was minimal (see Methods; Supplementary Fig. 1e). Each cluster was further identified as a specific cell subpopulation according to the expression of the most variable genes and the canonical markers, including CD4+ T cells (gene markers: PTPRC, CD3D, and CD4), CD8+ T cells (PTPRC, CD3D, and CD8A), myeloid cells (CD14, ITGAX for CD11C), malignant cells (EPCAM and KRT5), B cells (CD19 and MS4A1), and NK cells (FCGR3A and NCAM1; Fig. 1b). Besides, we detected 56 fibroblasts and seven endothelial cells with sparse distribution among TME cells (Supplementary Fig. 1f), which might reflect their intrinsic nature in NPC or their low representation due to technical limitations. All these cell types were widespread in tumour samples, indicating the heterogeneous cell composition of TME in NPC (Fig. 1b), consistent with a recent single-cell transcriptome study of NPC17. We observed that the proportions of CD8+ T and B cells were increased with 1.34 and 2.33 times on average, respectively, while the NK cells were decreased in the tumours compared to the PBMC, suggesting two distinct immune landscapes between tumour and peripheral blood (Fig. 1c). Moreover, we compared cell compositions between NPC and other types of cancers with single-cell data publicly available, including non-small-cell lung cancer (NSCLC), colorectal cancer (CRC), and pancreatic ductal adenocarcinoma (PDAC). We observed the common occurrence of infiltrating immune cells amid individual heterogeneity of cell composition in all types of cancers (Supplementary Fig. 1g). Notably, we observed a significantly higher proportion of T cells in NPC compared to any other cancers (Supplementary Fig. 1h), which is consistent with a previous finding that the tumour infiltrating leucocytes was the main characteristic of NPC stroma20.Heterogeneity of T cells and the diversity of TCR repertoireConsidering the abundance of T and NK cells in NPC tumour samples and their anti-tumour capabilities, we explored the intrinsic structure and potential functional subtypes of the overall T and NK cell populations. We grouped all 141,875 T and NK cells into 32 subgroups using clustering analysis, of which the majority were CD4+ and CD8+ T cells (Fig. 2a, Supplementary Fig. 2a, Supplementary Data 1 and 2). To identify any gene with a specific expression on a cell type, we performed differential gene expression (DEG) analysis of T cell clusters. We observed that CD4+ and CD8+ T cell clusters in tumour samples had widespread overexpression of exhaustion markers (LAG3, TIGIT, PDCD1, HAVCR2, and CTLA4) and effector molecules (GZMB, GZMK, INFG, NKG7, GNLY, and IL2; Supplementary Fig. 2b), with remarkably high expression of the proliferative signatures for CD8_C10_MKI67 and Treg_C3_MKI67 (Supplementary Fig. 2c and Supplementary Data 3). The co-expression of exhaustion and effector genes in tumour infiltrating T cells has been also demonstrated recently in NPC17. Together, these observations suggest that T cells exhibited anti-tumour effects against antigens, but their effector functions were somehow suppressed in the TME of NPC. By contrast, we observed naïve gene signatures (high expression of TCF7, SELL, CCR7 and LEF1) in the resting T cells in PBMC, including CD4_C1_LEF1, CD8_C1_LEF1, CD8_C2_TCF7, and DN_LEF1 (CD4−CD8−) cell clusters, and partially in Treg_C1_SELL and DP_TCF7 (CD4+CD8+; Supplementary Fig. 2b). Besides, we identified two clusters of CD4+ Th1-like cells in tumours, including Th1_like_C1_CCR7 and Th1_like_C2_TNF, with specific expression of naïve T cell markers and pro-inflammatory cytokines, respectively, as well as a common expression of Th1-like cell markers21 (CXCL13, BHLHE40, and CXCR3; Supplementary Fig. 2d).Fig. 2Expression profile and development of CD8+ T cells.a UMAP plot of 141,875 T and NK cells grouped into 32 cell types. Each dot represents a cell, coloured according to the cell types indicated at the right legend. b Violin plots showed the normalised expression of CD8+ T cell markers (rows) in each CD8+ T cell cluster (columns). Cell clusters and the expression level of a gene are indicated at the x- and y-axis, respectively. c Pseudotime trajectory analysis of selected CD8+ T cells (CD8_C5, CD8_C7, CD8_C8, CD8_C9, CD8_C10, and CD8_C11; n = 10,000) with high variable genes. Each dot represents one single cell, coloured according to its cluster label. The inlet plot showed each cell with a pseudotime score from dark blue to yellow, indicating early and terminal states, respectively. For CD8+ T cell clusters, 10,000 cells were randomly selected for the analysis. d Box plots showed the transition-index scores of exhausted CD8+ T cells (CD8_C11_PDCD1) and other CD8+ T cells (n = 10). Comparison was made using a two-sided Wilcoxon test. Cell clusters and transition-index scores are indicated at the x- and y-axis, respectively. Endpoints depict minimum and maximum values; centre lines denote median values; whiskers denote 1.5× the interquartile range; coloured dots denote each patient. e Box plots showed the expansion- (top panel) and PBMC-Tumour migration-index (bottom panel) scores of each CD8+ T cell cluster (n = 10). Each comparison was made using either a two-sided Wilcoxon test (top panel) or Kruskal–Wallis test (bottom panel). Cell clusters are indicated at the x-axis, and the y-axis shows the expansion- and PBMC-Tumour migration-index scores at the top and bottom panel, respectively. Endpoints depict minimum and maximum values; centre lines denote median values; whiskers denote 1.5× the interquartile range; coloured dots denote each patient.Next, we performed T cell receptor (TCR) repertoire analysis based on the sequences of α and β chains of TCR, which revealed 38,720 (32.97%; out of 117,447) T cells with detectable TCR α-β pairs or clonotypes after strict quality control (Supplementary Fig. 2e and Supplementary Data 4). We observed no sharing of any identical TCR clonotype among different patients with NPC, although they had certain preferences of V and J fragments as well as V–J pairs (Supplementary Fig. 3a, b). Interestingly, we observed the sharing of the most variable CDR3 sequences across the patient samples (Supplementary Fig. 3b), among which CAVRGTGTASKLTF and CASSFSGANVLTF have been associated with the recognition of MLANA and EBV antigens in the VDJ database22, respectively (Supplementary Table 4). Moreover, we observed that both CD4+ and CD8+ T cells have more clonal T cells, which are derived from identical TCR clonotypes and consistent with a previous study17, in the tumours compared to the PBMC, suggesting the clonal expansion of certain dominant clones of tumour infiltrating T cells upon continuous stimulations by tumour antigens (Supplementary Fig. 3c).Diversity of CD8+ T cells and the development of exhausted intratumoral CD8+ T cellsWe identified a total of 62,244 CD8+ T cells in all NPC samples, which were grouped into 11 clusters based on their expression of canonical markers, including two naïve (CD8_C1_LEF1 and CD8_C2_TCF7), blood central memory (CD8_C3_KLRB1), blood effector memory (CD8_C4_KLRG1), high migration (CD8_C5_CX3CR1), tumour central memory (CD8_C6_IL7R), tumour effector memory (CD8_C7_GZMK and CD8_C8_MHC), tissue resident memory (CD8_C9_XCL1), high proliferating (CD8_C10_MKI67), and exhausted (CD8_C11_PDCD1) T cells (Fig. 2a, b, Supplementary Data 1 and 2). The majority (>97.68%) of CD8_C6_IL7R, CD8_C7_GZMK, CD8_C8_MHC, CD8_C9_XCL, CD8_C10_MKI67, and CD8_C11_PDCD1 were found in NPC tumours, whereas the majority (>94.85%) of CD8_C1_LEF1, CD8_C2_TCF7, CD8_C3_KLRB1, CD8_C4_KLRG1, and CD8_C5_CX3CR1 were in the peripheral blood (Supplementary Data 1).To evaluate the functional status of CD8+ T cells, we calculated cytotoxicity, proliferation, and exhaustion scores for all CD8+ T cell clusters. We observed the highest cytotoxicity score for CD8_C5_CX3CR1, the highest proliferation score for CD8_C10_MKI67, and the highest exhaustion score for CD8_C11_PDCD1 (Supplementary Fig. 2c and 4a), suggesting their potential cytotoxic, proliferation, and exhausted functions, respectively. Next, the DEG analysis revealed high expression of chemokine receptors (CX3CR1, CXCR1, and CXCR2), S1P receptors (S1PR1, S1PR4, and S1PR5), and integrins (ITGB2, ITGA4, ITGAL, and ITGB7) in CD8_C5_CX3CR1, which were responsible for the regulation of CD8+ T cell migration23 (Supplementary Fig. 4b). Moreover, signalling pathway enrichment analyses of the genes with differential expression revealed that tumour cytotoxic CD8+ T cell clusters (CD8_C7_GZMK, CD8_C8_MHC, and CD8_C9_XCL1) were enriched with the pathways related to cytokine production and lymphocyte activation; and CD8_C5_CX3CR1 was enriched with the pathways related to leukocyte trans-endothelial migration and leukocyte migration (Supplementary Fig. 4c), which are consistent with their capability in peripheral circulation and infiltrating to tumour24.To further explore the development of CD8+ T cells in NPC, we first performed pseudotime trajectory analysis using Monocle2 to order each CD8+ T cell along trajectories according to their expression and transition profiles. We observed the developmental trajectories from CD8_C5_CX3CR1 cells at the initial state or CD8_C10_MKI67 cells at the intermediate state to CD8_C11_PDCD1 cells at the terminal state (Fig. 2c). Supportively, compared to the exhausted CD8_C11_PDCD1, CD8_C10_MKI67 had an intermediate exhaustion score (Supplementary Fig. 4a) and lower expression of known exhaustion markers including PDCD1, LAG3, and HAVCR2 (Supplementary Fig. 2b). TCR repertoire sequencing revealed 21,099 CD8+ T cells (out of 62,244) with TCR clonotypes (Supplementary Fig. 4d and Supplementary Data 4). We observed that CD8_C11_PDCD1 shared considerable proportions of identical TCRs with other CD8+ T cell clusters, ranging from 17.68% to 41.67% for infiltrating T cell clusters and 5.31% for peripheral CD8_C5_CX3CR1 (Supplementary Fig. 4e). To track the dynamic relationships among T cell clusters from NPC tumour and peripheral blood, we quantitated the expansion (exp, clonal expansion), migration (migr) and transition (tran, developmental transition or differentiation) of T cells using gene expression and TCR information with STARTRAC method24. Consistently, we observed the highest transition mobility of CD8_C11_PDCD1 with CD8_C10_MKI67, followed by CD8_C7_GZMK, CD8_C8_MHC, and CD8_C9_XCL1 (Fig. 2d). These observations strongly suggest that CD8_C11_PDCD1 cells were mainly expanded by proliferating pre-exhausted intratumoral CD8+ T cells. Moreover, we observed that CD8_C5_CX3CR1 had the largest number of clonal T cells (Supplementary Fig 4d) and the highest expansion mobility in CD8+ T cell clusters (Fig. 2e). Furthermore, CD8_C5_CX3CR1 had the highest proportion of shared TCR between peripheral blood and tumour (Fig. 2e). The proportions of TCRs shared with peripheral CD8_C5_CX3CR1 ranged from 4.76% to 12.77% for infiltrating CD8+ T cell clusters (Supplementary Fig. 4e). These data suggest a common origin of the intratumoral CD8+ T cells in NPC tumour from CD8+ T cells in peripheral blood including CD8_C5_CX3CR1 cells.The diversity and trajectory of Treg cells in NPCTreg cells are potent suppressors of immune cells and are essential to maintaining immunological tolerance and homoeostasis. We identified a total of 11,631 Treg cells based on their transcription of canonical markers (CD4, IL2RA, and FOXP3), which were grouped into four cell clusters, including Treg_C1_SELL, Treg_C2_HSPA1A, Treg_C3_MKI67, and Treg_C4_TNFRSF4 (Fig. 2a, Supplementary Fig. 5a, Supplementary Data 1 and 2). We observed that the proportion of Treg cells among CD4+ T cells in the tumours was much higher than that in the PBMC (Supplementary Fig. 5b). All Treg_C4_TNFRSF4 cells and the majority of Treg_C2_HSPA1A (99.5%; 4,762 out of 4,786) and Treg_C3_MKI67 (90.0%; 1,187 out of 1,319) cells were found in the tumours, while Treg_C1_SELL cells were in the PBMC (Supplementary Data 1). To explore the immune-regulatory functions of Treg cells, we first calculated the IL2R scores for each cell based on their expression level of CD25 (IL2RA), CD122 (IL2RB), and CD132 (IL2RG) using the AddModuleScore function implemented in Seurat software. The three genes encode transmembrane proteins that form a receptor complex competitively binding IL2 (the T cell growth factor) with high affinity so as to inhibit effector T cells25. We observed the highest IL2R score for Treg_C4_TNFRSF4 among all Treg clusters (Fig. 3a), suggesting the strongest IL-2 binding potential of Treg_C4_TNFRSF4 cells. Similarly, we also observed the highest inhibitory and co-stimulatory scores for Treg_C4_TNFRSF4 cells based on their expression levels of genes with immune-inhibitory functions and co-stimulatory receptors, respectively (Fig. 3a and Supplementary Fig. 5a), suggesting that Treg_C4_TNFRSF4 cells had a stronger suppression potential on immune response and were much activated than the other Treg cells. Consistently, such a subset of Treg cells were also identified in CRC, NSCLC, and hepatocellular carcinoma (HCC), with high activation and immune-suppressive potential as indicated by the high IL2R, inhibitory, and co-stimulatory scores (Supplementary Fig. 5c). Besides, we observed elevated expression levels of chemokine receptors in Treg_C4_TNFRSF4 cells, including CXCR3, CXCR6, and CCR8 that have been implicated in several cancers26 (Supplementary Fig. 5a).Fig. 3Expression profile and development of Treg cells.a Violin plots showed the IL2R (left panel), inhibitory (middle panel), and co-stimulatory (right panel) scores for each Treg cell cluster (n = 11,631). Box plots inside the violins indicated the quartiles of corresponding score levels. Endpoints depict minimum and maximum values; centre lines denote median values; whiskers denote 1.5× the interquartile range; black dots denote each cell. Violin plots are coloured according to cell types, and signature scores are indicated at the y-axis. b Heatmap showed the selected signalling pathways (rows) that were significantly enriched in GO and KEGG analyses for each Treg cell cluster (columns). Filled colours from blue to red represent scaled expression levels (normalised −log10P values) from low to high. P values were calculated by one-sided hypergeometric test and adjusted for multiple comparisons. Orange and purple squares on the left column represent the results derived from GO and KEGG signalling pathways analysis, respectively. c Pseudotime trajectory analysis of Treg cells (Treg_C1, Treg_C2, Treg_C3, and Treg_C4; n = 11,631) with high variable genes. Each dot represents one single cell, coloured according to its cluster label. The inlet plot showed each cell with a pseudotime score from dark blue to yellow, indicating early and terminal states, respectively. d Box plots showed the expansion- (top panel) and migration-index (bottom panel) scores of each CD4+ T cell cluster (n = 10). Comparison was made using two-sided Wilcoxon test. Cell clusters are indicated at the x-axis, and the y-axis shows the expansion- or migration-index at the top or bottom panel, respectively. Endpoints depict minimum and maximum values; centre lines denote median values; whiskers denote 1.5× the interquartile range; coloured dots denote each patient. e Box plots showed the transition-index scores of Treg_C4_TNFRSF4 (left panel) and Treg_C2_HSPA1A (right panel) with other Treg cells (n = 10). Comparison was made using two-sided Kruskal-Wallis test. Cell clusters and the transition-index scores are indicated at the x- and y-axis, respectively. Endpoints depict minimum and maximum values; centre lines denote median values; whiskers denote 1.5× the interquartile range; coloured dots denote each patient.To characterise the potential functions of Treg cells, we first performed signalling pathway enrichment analyses for each Treg cluster based on the expression levels of genes implicated in each pathway. We observed a distinct pattern of pathway enrichment for each Treg cluster, suggesting their various functions. Particularly, the ‘cytokine-cytokine receptor interaction’ was highly enriched in Treg_C4_TNFRSF4 (Fig. 3b), consistent with their chemotactic potentials as mentioned earlier. Moreover, the ‘interleukin-10 production’, ‘TNF signalling pathway’ and ‘NF-κB signalling pathway’ were enriched in both Treg_C4_TNFRSF4 and Treg_C2_HSPA1A (Fig. 3b). As such, we further compared the major pathways between Treg_C4_TNFRSF4 and Treg_C2_HSPA1A using Gene Set Enrichment Analysis (GSEA), which revealed a higher enrichment of pathways related to cell cycle, chemokine, TGF-β, and negative regulation of T cell proliferation, as well as transcription factors activating NF-κB and STAT pathways in Treg_C4_TNFRSF4 (Supplementary Fig. 5d). Since Treg_C4_TNFRSF4 expressed CCR8 specifically (Supplementary Fig. 5a), we used the normalised mRNA ratio of CCR8/FOXP3 to estimate the fraction of Treg_C4_TNFRSF4 cells in Treg cells (FOXP3+). Survival analysis showed that the higher ratio of CCR8/FOXP3 was associated with the decreased progression-free survival (PFS; Supplementary Fig. 5e), suggesting that a higher fraction of Treg_C4_TNFRSF4 cells with activated potential in Treg cells had a strong immune-suppressive function in NPC.To trace the origin of intratumoral Treg cells, we first performed pseudotime trajectory analysis using Monocle2, which revealed the most terminal status with the highest pseudotime scores for Treg_C4_TNFRSF4 cells and two developmental trajectories of Treg_C4_TNFRSF4 cells from Treg_C1_SELL cells in PBMC and Treg_C3_MKI67 cells in tumours (Fig. 3c). We next examined the expression of tissue resident markers (CD69, ITGAE, and BHLHE40; Supplementary Table 5) in Treg cells and observed much higher expression of ITGAE and BHLHE40 in Treg_C4_TNFRSF4 and Treg_C3_MKI67 cells than Treg_C2_HSPA1A and Treg_C1_SELL cells (Supplementary Fig. 6a). Given that the majority of Treg_C2_HSPA1A cells was in tumours and originated from Treg_C1_SELL cells according to the pseudotime trajectory analysis (Fig. 3c), the scarce expression of the resident markers might suggest the most recent recruitment of Treg_C2_HSPA1A cells from peripheral blood.TCR repertoire analysis revealed 17,621 (out of 47,384) CD4+ T cells assigned with clonotypes, among which Treg_C2_HSPA1A and Treg_C4_TNFRSF4 had intermediate numbers of clonotypes (Supplementary Fig. 6b and Supplementary Data 4). Noteworthily, Treg_C4_TNFRSF4 had the largest proportion of clonal cells, meaning the highest clonality, among all CD4+ T cells (Supplementary Fig. 6b). Consistently, we observed that Treg_C4_TNFRSF4 had the highest expansion score, meaning the highest degree of clonal expansion, among the Treg cell clusters (Fig. 3d). We also observed the highest migration score, meaning the highest mobility, for Treg_C1_SELL derived from the PBMC (Fig. 3d). DEG analysis revealed that Treg_C1_SELL had high expression of chemokine receptors CCR4, which are chemotactic counterparts for CCL5, CCL17, and CCL22 produced by intratumoral CD8+ T, NK, and myeloid cells in NPC (Supplementary Figs. 5a and 6c). These observations suggest that the migration capability and chemotactic interaction potentials with intratumoral cells make the movement of peripheral Treg_C1_SELL cells to tumour site possible. Indeed, we observed a small number of shared TCRs between Treg cells from tumour and peripheral blood (Supplementary Fig. 6d), which is consistent with a previous finding that intratumoral Treg cells were partially recruited from peripheral blood27. We further examined the transition mobility of Treg_C2_HSPA1A and Treg_C4_TNFRSF4 with other Treg cells. We observed that Treg_C4_TNFRSF4 cells had the highest transition mobility with Treg_C3_MKI67 cells, followed by Treg_C2_HSPA1A and Treg_C1_SELL cells; and Treg_C2_HSPA1A cells had high transition mobility with Treg_C4_TNFRSF4 and Treg_C3_MKI67 cells (Fig. 3e). These observations again supported the developmental trajectory of intratumoral Treg_C4_TNFRSF4 cells from naïve Treg_C1_SELL cells through intermediate Treg_C2_HSPA1A or Treg_C3_MKI67 cells (Fig. 3c).Diversity of B cells in NPCWe identified a total of 22,892 B cells, which were grouped into nine clusters (Supplementary Fig. 7a, Supplementary Data 1 and 2). Among them, B_C1_TCL1A, B_C2_FCRL3, and Plamsa_C1_IgA clusters were derived from PBMC and the other six clusters were from tumour samples (Supplementary Fig. 7a). DEG analysis revealed unique gene signatures for B cell clusters in tumour samples, including B_C5_ISG15 with interferon induced genes, B_C6_HSPA1A with stressful gene expression, and Plasma_C2_IgG with elevated expression level of IgH genes (Supplementary Fig. 7b). We further identified two B cell clusters (B_C1_TCL1A and B_C4_IFITM3) before the class switch recombination based on the expression of IGHM and IGHD (Supplementary Fig. 7b and Supplementary Table 5). Moreover, correlation analysis revealed that the expression of TCL1A was highly correlated with that of IGHM and IGHD in the two clusters, which could be a sufficient marker to classify B cells before the class switch recombination (Supplementary Fig. 7c). Signalling pathway enrichment analyses of the genes with differential expression revealed that B cell clusters were enriched with various pathways related to immune regulation (Supplementary Fig. 7d). Particularly, B_C4_IFITM3, B_C5_ISG15, and B_C6_HSPA1A had the enrichment of ‘EBV infection’, ‘defence response to virus’, ‘viral carcinogenesis’, and ‘response to interferon-gamma’ pathways, suggesting that the three groups of cells might be responsible for the immune response against EBV infection (Supplementary Fig. 7d).Tumour-associated LAMP3+ DCs display a tolerogenic phenotype in NPCA total of 8,893 myeloid cells were identified and clustered into 10 subsets including one for mast cells, five for monocyte or macrophage cells, three for conventional dendritic cells, and one for plasmacytoid dendritic cells (Fig. 4a, Supplementary Fig. 8a and 8b, Supplementary Data 1 and 2). Among the four clusters of dendritic cells, DC_C2_CD1C, DC_C3_LAMP3, and DC_C4_JCHAIN were derived from tumours, and DC_C1_FCER1A was derived from peripheral blood, which was assigned as monocyte-like DC because of the expression of monocyte marker S100A8 (Supplementary Fig. 8a, b). Noteworthily, we identified DC_C3_LAMP3 cells as a group of DCs with high maturation, activation, and migration potentials in NPC, based on the expression levels of the signature genes related to maturation (LAMP3, MARCKSL1, IDO1, and UBD), activation (CD80, CD83, and CD40), and migration (CCR7, FSCN1, and SLCO5A1; Fig. 4b, Supplementary Fig. 8c and Supplementary Table 5), respectively. Moreover, DC_C3_LAMP3 cells had high expression of special chemokine ligands (CCL17, CCL19, and CCL22), which are known to recruit immune cells expressing chemokine receptors CCR4, CCR7, and CXCR3 (Fig. 4b). We also observed significant correlations of expression between the marker gene LAMP3 and other functional genes related to maturation, migration, activation, and chemokine ligands in DC_C3_LAMP3 (Supplementary Fig. 8d). These observations suggest that DC_C3_LAMP3 cells might be LAMP3+ DCs, which are featured with high migration, activation, and maturation in several cancers as reported previously28,29.Fig. 4Expression and development of dendritic cells.a UMAP plot of 8,893 myeloid cells grouped into 10 cell types. Each dot represents a cell, coloured according to cell types. b Heatmap showed the normalised mean expression of genes associated with maturation, activation, migration, and chemokine ligand (rows) in three dendritic cell clusters (DC_C1, DC_C2, and DC_C3; columns). Filled colours from black to yellow represent scaled gene expression levels from low to high. c Heatmap showed the selected signalling pathways (rows) with significant enrichment of GO and KEGG terms for three dendritic cell clusters (DC_C1, DC_C2, and DC_C3; columns). Filled colours from blue to red represent scaled expression levels (normalised −log10P values) from low to high. P-values were calculated by one-sided hypergeometric test and adjusted for multiple comparisons. Orange and purple squares on the left column represent the results derived from GO and KEGG signalling pathways analysis, respectively. d Violin plots showed the differentiation, apoptosis, antigen presentation, and dysfunction scores of three dendritic cell cluster (DC_C1, DC_C2, and DC_C3; n = 1134). Box plots inside the violins indicated the quartiles of corresponding score levels. Endpoints depict minimum and maximum values; centre lines denote median values; whiskers denote 1.5 × the interquartile range; black dots denote each cell. Cell clusters and the signature scores are indicated at the x- and y-axis, respectively. e Pseudotime trajectory analysis of three dendritic cell clusters (DC_C1, DC_C2, and DC_C3; n = 1134) with high variable genes. Each dot represents one single cell, coloured according to its cluster label. The inlet plot showed each cell with a pseudotime score from dark blue to yellow, indicating early and terminal states, respectively. f Venn diagram showed overlapped transcription factors regulating LAMP3 gene, immune-suppressive molecules, and HLA-II in DC_C3_LAMP3 cells.Signalling pathway enrichment analyses using GO and KEGG revealed a specific pattern of enriched pathways among the three conventional DC cell clusters, where the ‘antigen processing and presentation’ was significantly upregulated in DC_C2_CD1C but downregulated in DC_C3_LAMP3 (Fig. 4c). Moreover, apoptosis, NF-κB, and MAPK signalling pathways as well as myeloid cell differentiation were also upregulated in DC_C3_LAMP3 compared to other two clusters (Fig. 4c). These observations are consistent with the GSEA analyses (Supplementary Fig. 8e). As such, we scored the expression levels of genes related to these pathways in each cluster (Supplementary Data 3), which revealed the highest levels of differentiation and apoptosis but the lowest antigen presentation for DC_C3_LAMP3 (Fig. 4d). Furthermore, we observed that the gene signatures corresponding to the activation of immune response was reduced in DC_C3_LAMP3 (Fig. 4c), which was consistent with the highest immune-regulatory score and the increased expression of a subset of immune-suppressive genes, including CD274 (PD-L1), PDCD1LG2 (PD-L2), CD200, EBI3, IDO1, IL4I1, SOCS1, SOCS2, and SOCS3 (Supplementary Fig. 9a and Supplementary Data 3). Besides, we observed a similar expression profile of LAMP3+ DC among NPC, HCC, and NSCLC (Supplementary Fig. 9a). These observations suggest that DC_C3_LAMP3 cells could be considered as a group of regulatory and tolerogenic DCs, which restrain the activation of T cells30.Next, we performed pseudotime trajectory analysis and observed that DC_C1_FCER1A cells developed into two branches including DC_C2_CD1C and DC_C3_LAMP3 cells, and DC_C3_LAMP3 cells had the highest pseudotime score meaning the most differentiated and matured DC (Fig. 4e). Combined with their immune-regulatory and antigen-presenting scores (Fig. 4d), these data suggest that DC_C1_FCER1A cells in peripheral blood might infiltrate to tumour, convert to DC_C2_CD1C cells with increased antigen-presenting capacity and to immune-suppressive DC_C3_LAMP3 cells (Fig. 4e). Consistently, we observed the similar pattern of changes in the expression of transcription factors (TFs) specific for the genes with differential expression from DC_C1_FCER1A to DC_C2_CD1C and then DC_C3_LAMP3 cells (Supplementary Fig. 9b). We further constructed an accurate cellular network to infer the regulons associated with transcription factors and signalling molecules in DC_C3_LAMP3 using ARACNe (Supplementary Data 5). We observed that the upregulation of LAMP3 was linked with multiple TFs, including ETV3, ETV6, HMGN3, GPBP1, TRAFD1, ATF3, KDM2B, JUN, HIVEP1, KLF6, ZBTB10, and NFKB1 that have been related to the maturation of DC in mouse31, while the downregulation of LAMP3 was linked with CREM (Supplementary Fig. 9c). We also observed that among these TFs KDM2B, KLF6, ETV6, JUN, HMGN3, and TRAFD1, as well as NFKB1, REL, and RELB in the NF-κB pathway were linked with the upregulated expression of immune-suppressive molecules like CD274, PDCD1LG2, CD200, and IDO1, but the downregulated expression of HLA-class II genes (Fig. 4f and Supplementary Fig. 9c). By contrast, SOX4 and CREM were associated with the downregulation of CD274, CD200, and IDO1 (Supplementary Fig. 9c). These observations suggest that multiple TFs regulate the immune-suppressive function, antigen-presenting capacity, and maturation of DC_C3_LAMP3 in NPC.Heterogeneity of malignant cells with different EBV infection statusWe identified a total of 2,787 malignant epithelial cells in NPC tumours based on their presence of large-scale chromosomal copy number variation (CNV) compared to a reference data of normal epithelial cells32 (see Methods; Fig. 5a). Given that EBV is a known factor responsible for the malignant transformation and tumorigenesis of NPC33, we examined the expression of EBV molecules in the malignant cells and divided them into EBV+ (EP_C1_LMP1) and EBV− (EP_C2_EPCAM) malignant cells according to their detectable or not EBV transcripts (LMP-1/BNLF2a/b, RPMS1/A73, LMP-2A/B, and BNRF1; Fig. 5b, c, Supplementary Fig. 10a, Supplementary Data 1 and 2). We observed higher expression of EPHA2 and EGFR in EP_C1_LMP1 cells (Fig. 5d), which have been related to the susceptibility of EBV infection34. Moreover, immunofluorescence staining of an EBV-encoded protein (LMP1) confirmed the presence of EBV+ malignant cells (LMP1+EPCAM+) and EBV− malignant cells (LMP1−EPCAM+) in NPC (Fig. 5e). We also observed specifically high activations of the major genes involved in NF-κB and Notch pathways as well as chemokines including CX3CL1 in EP_C1_LMP1 compared to EP_C2_EPCAM cells (Fig. 5d). Consistently, we observed high expression of CX3CL1 in an independent collection of NPC tumours (n = 113) compared to non-cancerous samples (rhinitis, n = 10; Supplementary Fig. 10b). Interestingly, we noted that the overexpression of CX3CR1, the receptor of CX3CL1, in multiple types of immune cells in peripheral blood (Supplementary Fig. 10c). Signalling pathway enrichment analyses revealed that EP_C1_LMP1 were enriched with cytokine-mediated, regulation of cell death, apoptosis, and cancer-related pathways (Fig. 5f). Taken together, these observations suggest that malignant NPC cells exhibit different susceptibility to EBV infection, leading to distinct expression profiles.Fig. 5Heterogeneity of malignant cells with distinct EBV infection in tumour tissues.a Heatmap showed the large-scale CNVs for epithelial cells (rows along y-axis) from 10 NPC tumours. CNVs were inferred according to the average expression of 100 genes spanning each chromosomal position (x-axis). Red: gains; blue: losses. Malignant NPC cells from different patients and the range of different chromosomes are indicated as different colour bars on the left and top to the heatmap, respectively. b UMAP plot of 2,787 malignant cells grouped into two cell clusters (EP_C1_LAMP1 and EP_C2_EPCAM). Each dot represents a cell, coloured according to a cell cluster. c UMAP plots showed the expression of EBV-encoded genes (LMP-1/BNLF2a/b, RPMS1/A73, LMP-2A/B, and BNRF1) in malignant cells. Each dot represents a single cell, and the depth of colour from grey to red represents low to high expression. d Violin plots showed the normalised expression of cluster markers, chemokines, and genes associated with NF-κB and Notch pathways in each cluster. In each plot, cell clusters and the expression level of a gene as the chart tile are indicated at the x- and y-axis, respectively. e Representative images of multiplex immunofluorescence staining of malignant cells in NPC tissues. Proteins detected using respective antibodies in the assays are indicated on top. The red, green, and orange arrows indicated the representative cells positive for EPCAM, LMP1, and co-expression of EPCAM and LMP1 proteins in malignant cells, respectively. Images are representative of three biological replicates. Scale bars, 50 µm. f Bar plots showed the selected signalling pathways with significant enrichment of GO (top panel) and KEGG (bottom panel) terms for EBV+ malignant cells (EP_C1_LMP1) compared to EBV- malignant cells (EP_C2_EPCAM), coloured from light to dark according to their −log10(P-values) from low to high. P-values were calculated by one-sided hypergeometric test and adjusted for multiple comparisons.To further explore the intra-tumour and inter-tumour heterogeneity of malignant NPC cells, we first divided them into five prominent cell subgroups (C1–C5; Supplementary Fig. 10d, e), using clustering analysis without EBV information. Next, we deciphered the variations of gene expression in malignant cells for different clusters using gene set variation analysis (GSVA), which revealed a distinct enrichment of signalling pathway for each cluster (Supplementary Fig. 10f). We observed variable proportions of cell subtypes among different tumour samples (Supplementary Fig. 10e), which might contribute to the inter-tumour heterogeneous expression profiles. Indeed, the GSVA data showed different enrichments of signalling pathways among tumour samples (Supplementary Fig. 10f). Noteworthily, the P02, P08, P11, and P15 samples with a higher proportion of C4 cluster compared to other samples showed enrichment of cell cycle (E2F, MYC, and G2M checkpoint) related pathways (Supplementary Fig. 10e–g), although C4 cluster had the highest proliferation scores but overall low content in NPC tumours compared to the other three clusters. These observations suggest the intra- and inter-tumour heterogeneity of the malignant cells in NPC.Intercellular interaction network in NPCTo explore the cellular communication network in NPC, we examined potential ligand-receptor binding among different cell clusters derived from NPC tumours and PBMC, using CellPhoneDB software35 (Supplementary Data 6). We observed intensive cellular interactions among the DC_C3_LAMP3 cells, Treg cells, and exhausted CD8+ T cells (CD8_C11_PDCD1) via inhibitory, co-stimulatory molecules, or chemokines (Fig. 6a, b). Among them, DC_C3_LAMP3 cells were predicted to interact with Treg_C1_SELL cells in peripheral blood through CCL17-CCR4 and CCL22-CCR4, which are known for recruiting Treg cells into tumour tissue36 (Fig. 6a). Treg_C4_TNFRSF4 cells had high expression of CTLA4, ENTPD1, and CSF1, which showed ligand-receptor bindings to CD80/CD86, ADORA2A, and SIRPA on DC_C3_LAMP3 cells, suggesting the potential interaction between Treg_C4_TNFRSF4 and DC_C3_LAMP3 cells (Fig. 6a). DC_C3_LAMP3 cells were also predicted to interact with CD8_C11_PDCD1 cells through CD200-CD200R signalling, a non-classical immune-suppressive pathway involved in the suppression of anti-tumour responses37 (Fig. 6b). Potential ligand-receptor interactions were observed between Treg_C4_TNFRSF4 and CD8_C11_PDCD1 cells, including those of chemokines (CCL4-CCR8), adhesive connection (ITGAL-ICAM1 and SELPLG-SELL), and immune regulation (HAVCR2-LGALS9; Fig. 6b and Supplementary Data 6), which are well-known in the TME of tumour and promote the immune-suppressive activity of Treg cells and CD8+ T cells exhaustion26,38. Notably, these potential interactions were commonly observed in our NPC cohort (Supplementary Fig. 11a). Consistently, in another independent NPC sample collection (n = 113), we observed strong correlations of expression among the gene signatures for DC_C3_LAMP3 cells, Treg cells, and exhausted CD8+ T cells (CD8_C11_PDCD1; r > 0.8, P < 2.2 × 10−16; Supplementary Fig. 11b). These observations suggest the widespread occurrence of the immune-regulatory interactions among DC_C3_LAMP3, Treg_C4_TNFRSF4, and CD8_C11_PDCD1 cells in NPC tumours. We further performed multiplex immunohistochemistry (IHC) staining of NPC biopsies and confirmed the physical juxtapositions of CD80-expressing DC_C3_LAMP3 cells (CD80+) and CTLA4-expressing Treg cells (CD3+CD4+FOXP3+), as well as PD-L1-expressing DC_C3_LAMP3 cells (CD80+) and PD-1-expressing CD8+ T cells (CD3+CD8+; Fig. 6c, d).Fig. 6Intercellular interactions among immune and malignant cells in NPC.a, b Dot plots showed selected ligand-receptor interactions between two cell clusters, for Treg and DC_C3_LAMP3 cells (a) and for exhausted CD8+ T (CD8_C11_PDCD1) and DC_C3_LAMP3 cells (b). The ligand-receptor interactions and cell-cell interactions are indicated at columns and rows, respectively. The means of the average expression levels of two interacting molecules are indicated by colour heatmap (right panel), with blue to red representing low to high expression. The log10(P-values) were indicated by circle size in one-sided permutation test. Different colour boxes at the bottom represent different function modules of receptor-ligand interactions. c Representative images of multiplex IHC staining for the juxtaposition of CTLA4-expressing Treg cells (CD3+CD4+FOXP3+) and CD80-expressing DC_C3_LAMP3 cells in NPC tissue samples. Proteins detected using respective antibodies are indicated on top. The green, red, magenta, cyan, and orange arrows indicated positive cells with the expression of CD3, CD4, FOXP3, CTLA4, and CD80 proteins in NPC tissue, respectively (bottom panel). Images are representative of three biological replicates. Scale bars, 100 µm and 20 µm for top and bottom panels, respectively. d Representative images of multiplex IHC staining for the juxtaposition of PD1-expressing CD8+ T cells (CD3+CD8+) and PD-L1-expressing DC_C3_LAMP3 cells (CD80+) in NPC tissue samples. Proteins detected using respective antibodies are indicated on top. The yellow, cyan, magenta, red, and green arrows indicated positive cells with the expression of CD3, CD8, CD80, PD1, and PD-L1 proteins in NPC tissue, respectively (bottom panel). Images are representative of three biological replicates. Scale bars, 100 and 20 µm for top and bottom panels, respectively.Between malignant NPC cells and immune cells, we observed that EBV+ EP_C1_LMP1 cells had significantly more receptor-ligand interactions than EBV−EP_C2_EPCAM cells in each NPC patient (Supplementary Fig. 12a). We noted that EP_C1_LMP1 cells uniquely expressed CX3CL1 in tumour, which was predicted to interact with CX3CR1 expressed on peripheral immune cells including CD8_C5_CX3CR1 cells, DC_C1_FCER1A cells, NK cells, and monocytes (Supplementary Fig. 12b), suggesting the chemotactic potential of EP_C1_LMP1 cells to immune cells from peripheral blood. Moreover, we observed that EGFR on EP_C1_LMP1 cells was predicated to bind TGFB1 on multiple cell types, which has been reported to regulate the EBV life cycle39 (Supplementary Fig. 12c). We also observed potential interacting pairs between EP_C1_LMP1 cells with activated Notch pathway and multiple cell types through NOTCH1-TNF and NOTCH2-JAG2, which have been related to radiation sensitivity40 and cancer stem-like side population cells41 in NPC (Supplementary Fig. 12d). In addition, the above-mentioned interactions were commonly determined but with individual variable intensity among the patients (Supplementary Data 6), suggesting that the interactions are widespread phenomena and heterogeneous in NPC (Supplementary Fig. 12b–d).DiscussionThrough the comprehensive single-cell transcriptome study on NPC, we provided a landscape view of the heterogeneous cell composition and complex interacting network in the tumour microenvironment and peripheral circulating blood of NPC at single-cell resolution. Transcriptome analyses of more than 176,000 individual cells of 53 subtypes revealed two distinct microenvironments between tumour and peripheral blood in NPC. With such large-scale single-cell data, we identified novel cell populations with specific gene signatures in NPC. Moreover, in combination of TCR repertoire sequencing, we delineated the potential developmental trajectories of intratumoral immune cells. Furthermore, we dissected a multiple intercellular network in NPC using ligand-receptor paring analyses (Fig. 7).Fig. 7Schematic diagram of cross-talks among multiple immune cells in the TME of NPC.EBV infects nasopharyngeal epithelial cells and participates in the tumorigenic process of NPC. EBV-positive malignant NPC cells secret a variety of chemokines (CX3CL1, etc.) and initiate the recruitment and tumoral infiltration of multiple immune cells with the chemokines receptors from the peripheral blood. Multiple tumour infiltrating immune cells activate EGFR and Notch pathway in EBV-positive malignant NPC cells. Naive CD8+ cells infiltrate to the lesion and develop to effector and further exhausted CD8+ cells. Peripheral DCs infiltrate to the tumour and differentiate into LAMP3+ DCs. The mature LAMP3+ DCs with the expression of PD-L1/PD-L2 interact with PD1 on CD8+ T cells whereby the signalling restrains the activation of CD8+ T cells and promotes their exhaustion. Treg cells interact with LAMP3+ DCs through CTLA4-CD80/CD86, which might limit the antigen presentation process of DCs and promote the secretion of IDO1 to induce the proliferation of Treg cells. The intensive cell-cell interactions among LAMP3+ DCs, Treg cells, exhausted CD8+ T cells, and malignant cells foster an immune-suppressive niche for the tumour microenvironment of NPC.CD8+ T cells are the key effector of anti-tumour immunity with cytotoxicity to kill tumour cells42. We observed an abundance of tumour infiltrating CD8+ T cells in NPC, which exhibited clonal expansion, effector, proliferation, and exhausted status, suggesting that the CD8+ T cells were largely suppressed amid being stimulated by the tumour neoantigens in the TME of NPC. This observation is consistent with the previous findings in other cancer types11,43, suggesting a common immunosuppressed state of CD8+ T cells in tumours. Combining TCR repertoire and transcriptome analyses for each T cell, we revealed, for the first time to our knowledge, the differentiation trajectory of CD8+ T cells in NPC, by which CX3CR1+CD8+ T cells (CD8_C5_CX3CR1) in the peripheral blood infiltrated and transformed to the exhausted CD8+ T cells (CD8_C11_PDCD1) in tumours. Consistently, the CX3CR1+CD8+ T cells of highly cytotoxic potential shared TCR clonotypes with tumour infiltrating CD8+ T cells, which are responsible for recognising antigens. In light of the previous findings that CX3CR1+CD8+ T cells are essential for viral control and infiltrating to tumour site to reduce tumour growth44,45, our findings raise possible T cell therapies for NPC by infusing the CX3CR1+CD8+ T cells from peripheral blood after ex vivo expansion or engaging chimeric TCR of T cells with CX3CR1 chemotactic potential towards tumour site.As cancer cells grow with high immunogenicity, inflammatory cells that are actively immunosuppressive, including regulatory T cells are recruited to help evade immune destruction by suppressing cytotoxic lymphocytes46. We observed three heterogeneous Treg cell clusters in NPC tumours, each with specific gene signatures and functions. Treg_C4_TNFRSF4 cells featured with high expression of TNFRSF family genes (TNFRSF4, TNFRSF9, and TNFRSF18) as well as CCR8 and exhibited the strongest immune-suppressive function compared to other Treg cell clusters in NPC. Consistently, it has been reported previously that Treg cells with the expression of TNFRSF family genes facilitate tumour immune evasion and promote cancer development36,47. Moreover, high expression of CCR8 has been demonstrated as a signature of Treg cells that restrain immunity, of which their amount in tumours is significantly associated with poor prognosis in several cancers26,48. Indeed, our present study revealed that the fraction of Treg_C4_TNFRSF4 in Treg cells might contribute to the poor survival of NPC. Given that CCR8 can promote immune-suppressive Treg cells26, a blockade of CCR8 signalling in Treg cells might abolish their specific suppressive effect on cytotoxic lymphocytes and thus inhibit tumour growth. While the presence of tumoral Treg cells in NPC has been confirmed using IHC staining assays in our present study and previously by other group15, the origin of tumour infiltrating Treg cells in NPC remains exclusive. Pseudotime trajectory and TCR repertoire analyses revealed that Treg_C4_TNFRSF4 cells were differentiated from naive Treg_C1_SELL cells in peripheral blood through intermediate Treg_C2_HSPA1A or Treg_C3_MKI67 cells. We suspected that reducing the migration of Treg cells in the peripheral blood to tumour sites might attenuate the immune suppressions endowed by Treg cells. With this, we note that the high expression of CCR4 in Treg_C1_SELL cells might be a potential therapeutic target, since it has been considered as the key molecule for Treg cells migrating into tumour49.Tumour-associated myeloid cells are heterogeneous as reported previously50. We observed multiple clusters of myeloid cells in NPC tumours, among which DC_C3_LAMP3 could be considered as a group of regulatory and tolerogenic DC, showing high expression of the migration (CCR7) and maturation (LAMP3) related genes. Both are the signatures of LAMP3+ DCs that have been recently demonstrated in multiple cancers28,29. Moreover, DC_C3_LAMP3 cells had the elevated expression of CD274, PDCD1LG2, CD200, IDO1, EBI3, SOCS1, SOCS2, and SOCS3, which are immune-suppressive related genes with similar expression pattern as LAMP3+ DCs in lung cancer51. These observations suggest that DC_C3_LAMP3 cells are a group of LAMP3+ DCs in NPC, which are a ubiquitous cell population in tumours and exert immune-regulatory function and the control of T cell activation52. Our data also revealed the developmental trajectory of the LAMP3+ DCs in NPC and the master transcription factors that are potentially crucial for the promoting maturation, decreased antigen presentation capability, and increased immune regulatory capability of LAMP3+ DCs. These transcription factors are connected to each other, forming a robust regulatory gene expression network in LAMP3+ DCs. We suspect that targeting these transcription factors might reshape the dendritic cells towards a normal antigen presentation phenotype, leading to potential therapeutic benefits for NPC.EBV infection is the key feature of NPC in the endemic regions, and its role in malignant transformation and tumorigenesis has been implicated in NPC33. We identified two groups of malignant cells with (EP_C1_LMP1) or without (EP_C2_EPCAM) EBV infection in NPC tumours. The EBV+ NPC cells showed a distinct transcriptional state compared to the EBV− NPC cells, with the specific activation of major genes implicated in EBV entry and cancer-related pathways, such as NF-κB53 and Notch pathways54. Moreover, the EBV+ NPC cells had more chemotactic interactions with immune cells derived from tumours and peripheral blood, which are important for the regulation of EBV life cycle and the shaping of tumour behaviours including radiation sensitivity and stemness40,41. The potential chemotactic interactions might also explain the abundant infiltration of immune cells in NPC tumour stroma55. Given that all endemic NPC are EBV positive, our findings are of particular interests for future studies on the role of EBV infection in NPC development and the persistence of EBV in body cells compared to its gradual loss during in vitro culture. The specific presence of EBV molecules in NPC cells has been harnessed as an adoptive immunotherapy for NPC using cytotoxic T lymphocyte recognising EBV in clinical trials56,57. We note that the presence of TCRs recognising EBV antigen among multiple patients with NPC may engage T cells targeting NPC cells with higher specificity. However, it’s noteworthy to explore whether and how the heterogeneous malignant NPC cells especially the EBV negative cells and other subtypes of cells with different proliferative capabilities contribute to the variable treatment outcomes. In addition, we didn’t observe EBV transcripts in any B cells, although B cells have been known as a primary host for EBV58. It might be explained by the low expression level of EBV genes in infected cells and the small fraction of B cells captured in our study.Our study revealed an intercellular network among LAMP3+ DCs (DC_C3_LAMP3), Treg cells, and exhausted CD8+ T cells (CD8_C11_PDCD1) in NPC, suggesting potential cross-talks among multiple immune cells to foster an immune-suppressive niche for the TME of NPC (Fig. 7). Indeed, the link between LAMP3+ DCs and Treg cells through the interaction between CCL17-CCR4 and CCL22-CCR4 has also been demonstrated in other cancers36, by which LAMP3+ DCs potentiate the chemotactic recruitment of peripheral Treg cells and promote their infiltration in tumours59. Moreover, LAMP3+ DCs also had abundant expression of IDO1 in NPC, which could induce the proliferation of tumour infiltrating Treg cells as reported previously60. On the other hand, our observation of the interaction between tumoral Treg cells and LAMP3+ DCs through CTLA4 and CD80/CD86 has been consistently reported in other cancers61, whereby Treg cells regulate the maturation of tolerogenic LAMP3+ DCs. By contrast, the cross-talks between cytotoxic CD8+ T cells and either DCs or Treg cells have been widely addressed in multiple cancers28,62,63. The mutual interactions between LAMP3+ DCs and Treg cells may enhance immune-suppressive effects on the exhausted CD8+ T cells in NPC. Together with the observation of chemotactic potential of EBV+ NPC cells to recruit peripheral immune cells, these findings suggest that these cross-talks among diverse cell types play important roles in maintaining the homoeostasis of TME in NPC. It would be plausible that disrupting the interactions might break the balance of TME and thus cure the tumour. Besides the promising results of PD-1/PD-L1 blockade immunotherapy in NPC, antibodies targeting EGFR have been reported to enhance the current treatment paradigms for locoregionally advanced NPC64. We suspect that immune-suppressive interaction of CD200-CD200R1 and LGALS9-HAVCR2 among LAMP3+ DCs, Treg cells, and exhausted CD8+ T cells might be also potential immunotherapeutic targets for NPC.Taken together, through uncovering the heterogeneous tumour microenvironment of NPC at a high resolution, we identified the essential cells and molecules with potential contributions to NPC tumorigenesis, and thus provide insights into the mechanisms underlying NPC progression and the development of potential therapeutic strategies for NPC. We acknowledge that our study has several limitations. First, we observed some trends of the associations of cell compositions with clinical characteristics (Supplementary Fig. 13). Particularly, patients with more advanced NPC had a higher proportion of peripheral CD8_C10_MKI67 and more intensive cellular interactions in DC_C4_JCHAIN and NK_C2_FCER1G cells (Supplementary Fig. 13c, e). However, we are not yet to draw a conclusion with such a limited cohort size that any components of the TME could be associated with the clinical outcomes. Second, we believe that NPC is one of the malignancies that are shaped together with immunity and neoantigens endowed by the somatic alterations in malignant cells. However, we are not able to predict any neoantigens, because we have no DNA level (whole genome or exome sequencing) data with our patient samples for HLA estimations and somatic mutations callings. Lastly, although our findings of the key interaction network and molecules of the TME in NPC were obtained using robust bioinformatic analyses and additional immunostaining assays, further functional experiments are awaited to explore the biological consequences and underlying mechanisms.MethodsPatient recruitment and sample collectionTen male individuals with nasopharyngeal carcinoma (NPC) were recruited from a local hospital in Guangzhou, China, an endemic region with high prevalence of NPC, between June 2018 and September 2018. The patients were histopathologically diagnosed with primary NPC by at least two pathologists according to the World Health Organization (WHO) classification. No history of cancer and any anti-tumour therapy prior to the primary diagnosis was self-reported. Clinical staging of NPC was determined according to the 8th edition of the International Union against Cancer (UICC) and American Joint Committee on Cancer (AJCC) staging system. Fresh tumour sample was obtained using endoscopic nasopharyngeal biopsy and matching peripheral blood sample was collected for each patient, followed immediately by single cell preparation as described below. All patients were EBV positive as confirmed using in situ hybridisation of EBV encoded small RNAs (EBERs) in tumour tissue. The average age was 50.6 and the patient’s characteristics were listed in the supplementary (Supplementary Data 1). For immunostaining assays, additional NPC biopsies were collected. The specimens were collected within 30 min after the tumour resection and fixed in formalin for 48 h. Written informed consent was obtained from all participants, and the study was approved by the Institutional Review Boards at the Sun Yat-sen University Cancer Center (SYSUCC).Preparation of single cell suspensionsFresh tumour samples were processed independently with enzymatic digestion and mechanical dissociation immediately after collection to generate single cell suspensions. Briefly, each tumour was cut into small pieces with approximately 1-mm3 in a D10 resuspension buffer, containing culture medium (DMEM medium; Gibco™, USA; Cat. no. 11965092) with 10% foetal bovine serum (FBS; Gibco; cat. no. 10099141), followed by enzymatic type II (Thermo Fisher, USA; cat. no. 17101015) and IV (Thermo Fisher; cat. no. 17104019) digestion for 30 min on a rotator at 37 °C. The digested mixture was passed through a 40 μm cell-strainer (BD Biosciences, USA; Cat. no. 352340) to obtain dissociated cells. The filtered mixture was centrifuged at 400 g for 5 min, and after removal of the supernatant, the pelleted cells were resuspended in 0.8% NH4Cl red blood cell lysis buffer and incubated on ice for 10 min. After washing twice with DPBS (Gibco; cat. no. 14190250), the dissociated cells from tumour were resuspended in a sorting buffer, consisting of 1X DPBS supplemented with 0.04% BSA (Sigma-Aldrich, USA; cat. no. 9048468). Viable cells were collected using fluorescence activated cell sorting (FACS; BD FACSAria III; BD Biosciences) with negative staining of propidium iodide (PI; Thermo Fisher, cat. no. P1304MP). At least 300,000 cells were collected for each tissue sample.From blood sample, PBMCs were isolated using a leukocyte separation solution, following the manufacture’s instruction (HISTOPAQUE-1077; Sigma-Aldrich; cat. no. 10771). Briefly, 5-ml of fresh peripheral blood was collected in EDTA anticoagulant tubes (BD; Cat. no. 366643) and subsequently transferred onto the solution. After density gradient centrifugation for 20 min at 750 X g, PBMCs settled at the interphase were carefully collected and washed twice with DPBS. Residual red blood cells were lysed using the same procedure abovementioned. Viable cells were collected using FACS with PI staining.Library construction for single cell gene expression and TCR profilingImmune repertoire measurement and gene expression at single cell resolution were conducted using Chromium Single Cell V(D)J Reagent Kit (10x Genomics, USA) following the manufacturer’s instructions. Briefly, the sorted cells were washed twice with the sorting buffer. Cell viability and number were determined using Trypan Blue (Thermo Fisher; Cat. no. 15250061) exclusion assay. Appropriate volume of cell suspension with a concentration of 700–1200 cells/µl were loaded in each channel, targeting a capture of 8,000 cells per sample, which were further mixed with barcoded gel beads on a Chromium Controller (10x Genomics). After reverse transcription reaction, cDNA amplification for 14 cycles was conducted on a thermal cycler (C1000; Bio-Rad, USA). The post-amplification cDNA was used as template to further enrich TCR fragments. Sequencing libraries for cDNA and TCR were separately constructed according to the instructions. The average fragment size of a library was quantitated using Qseq100 (Bioptic; Taiwan).Next generation sequencing and data processingEach DNA library was loaded into a sequencing lane on a HiSeq X system (Illumina, USA) and was sequenced with pair-end reads of 150 bp. Raw data of Binary Base Call (BCL) format was converted to FASTQ files using bcl2fastq (version v2.19.0.316, Illumina). Next, Cell Ranger pipelines (version 3.0.1; 10x Genomics) were used to align sequencing reads in the FASTQ files to reference genomes of interest and generate feature-barcode matrices. Single-cell 5'-gene expression data and TCR enriched data from the same cDNA library were processed using Cell Ranger count and Cell Ranger vdj implemented in the pipelines, respectively. The gene expression data was mapped to human genome reference sequence (GRCh38; http://cf.10Xgenomics.com/supp/cell-exp/refdata-cellranger-GRCh38-1.2.0.tar.gz) and EBV reference sequence65 (Akata; https://github.com/flemingtonlab/public/tree/master/annotation) for cDNA sequencing reads. The TCR enriched data were mapped to the VDJ reference sequence (http://cf.10Xgenomics.com/supp/cell-vdj/refdata-cellranger-vdj-GRCh38-alts-ensembl-2.0.0.tar.gz) for TCR sequencing reads.Single-cell gene expression quantification and determination of cell typesDoublets are artefactual libraries generated from two cells arising due to errors in droplet encapsulation of cells, and thus commonly affect the quality of single-cell sequencing data. The R package “DoubletFinder” (https://github.com/chris-mcginnis-ucsf/DoubletFinder) was applied to predict doublets in our data. Basically, a doublet is defined as a single-cell library representing more than one cell, and a closer examination of some known markers would suggest that the offending cluster consists of doublets of more than one cell type, while no cell type is known to strongly express both markers at the same time. We removed doublets in each sample individually, with an expected doublet rate of 0.05 and default parameters used otherwise (Supplementary Table 3). The remaining cells survived from the filtering criteria were single cells. Then the gene expression matrices for all remaining PBMC and tumour cells were combined and converted to a Seurat object using the R package Seurat (version 2.3.4, https://satijalab.org/seurat). Next, any cells were removed for which had either less than 101 UMIs, or expression of less than 501 genes, or over 15% UMIs linked to mitochondrial genes. From the remaining cells, gene expression matrices were generated with log normalisation and linear regression using the NormalizeData and ScaleData function of the Seurat package.Because the samples were processing independently and high-dimensional variables are common in single-cell sequencing data, which might introduce potential batch effect, we used canonical correlation analysis (CCA) and RunUMAP function implemented in Seurat to reduce dimensionality and remove batch effect. Cell clusters were identified using the FindClusters function in Seurat, with a K parameter of 20 and default parameters used otherwise. We annotated the clusters as different major cell types based on their average gene expression of well-known markers, including CD4+ T cell (PTPRC, CD3D, and CD4), CD8+ T cell (PTPRC, CD3D, and CD8A), myeloid cell (CD14 and ITGAX encoding CD11C), malignant cell (EPCAM and KRT family genes), B cell (CD19 and MS4A1), and NK cell (FCGR3A and NCAM1).Repeating the abovementioned steps (normalisation, dimensionality reduction, and clustering), we further identified sub-clusters and annotated them as different specific cell subtypes by the average expression of respective gene sets in each major cell type. To identify marker genes for each sub-cluster within the major cell types (CD4+ T, CD8+ T, NK, B, myeloid, and malignant cells), the expression profiles of the sub-cluster were contrasted with those of the other sub-clusters using the Seurat FindAllMarkers function. Differential expression analysis implemented in the function compared all the genes in the two datasets using the default two-sided non-parametric Wilcoxon rank sum test. A significant differentially expressed gene was determined if it had the Bonferroin-adjusted P value lower 0.05 and an average natural logarithm (ln) fold-change of expression of at least 0.1 and 0.25 for malignant cells and other cells, respectively. The cluster with multiple well-defined marker genes of different cell types and an elevated number of UMI was considered cell contamination and removed in downstream analysis. For each cluster (like C1) of a major cell type (like CD4+ T cells), we assigned a cluster identifier with a marker gene (like LEF1) as “CD4_C1_LEF1”. The selection criteria for the marker gene included (1) with top ranking at the differential gene expression analysis for the corresponding cell cluster, (2) with strong specificity of gene expression meaning high expression ratio within the corresponding cell cluster but low in other clusters, and (3) with literature supports that it’s either a marker gene or functional relevant to the type of cell.Collection of public single-cell datasetsTo compare the features of tumour microenvironment including cell compositions between NPC and other types of cancers, we collected the single-cell data publicly available for multiple cancers, including NSCLC12 (downloaded from https://gbiomed.kuleuven.be/scRNAseq-NSCLC), CRC66 (Gene Expression Omnibus; https://www.ncbi.nlm.nih.gov/geo/; GEO accession number: GSE132465), and PDAC67 (Genome Sequence Achieve; https://bigd.big.ac.cn/gsa/; GSA accession number: CRA001160), as well as that for specific cell types, including CD4+ T cells in CRC68 (GEO accession number: GSE146771), NSCLC11 (GEO accession number: GSE99254) and HCC28 (GEO accession number: GSE140228) and DCs in NSCLC29 (GEO accession number: GSE127465) and HCC28 (GEO accession number: GSE140228).Calculation of functional module scoresTo evaluate the potential functions of a cell cluster of interest, we calculated the scores of functional modules for the cell cluster, using the AddModuleScore function in Seurat at single cell level. The average expression levels of the corresponding cluster were subtracted by the aggregated expression of control feature sets. All analysed genes were binned based on averaged expression, and the control features were randomly selected from each bin69. The functional modules including proliferation score for T and malignant cells, cytotoxicity and exhausted scores for CD8+ T cells, IL2R, inhibitory, and co-stimulatory scores for Treg cells, as well as maturation, activation, migration, differentiation, apoptosis, antigen presentation, and immune regulatory scores for dendritic cells. The involved genes were listed in the supplementary material (Supplementary Data 3).Pathway enrichment analysisTo gain functional and mechanistic insights of a cell cluster, we performed Gene Ontology (GO) and KEGG Pathway enrichment analyses using Metascape (http://metascape.org/) to identify biological pathways that were enriched in a certain gene list more than that would be expected by chance. For malignant cells, the gene list imported to Metascape included the top 100 differentially expressed genes (DEGs) with a natural logarithm of fold changes of expression (lnFC) > 0.1 in clusters. For non-malignant cells, the gene list included the top 100 DEGs with lnFC > 0.25 in clusters. P value <0.05 was considered to be a significant enrichment. To compare the difference of signalling pathway enrichment between two clusters (Treg_C2_HSPA1A versus Treg_C4_TNFRSF4 and DC_C2_CD1C versus DC_C3_LAMP3), we performed the gene set enrichment analysis (GSEA; version 3.0) using the selected molecular signatures database v7.070. To explore the heterogeneous expression of malignant cells, we performed gene set variation analysis (GSVA, version 1.34.0), using 18 hallmark pathways described in the molecular signature database.Developmental trajectory inferenceTo characterise the potential process of immune cell functional changes and determine the potential lineage differentiation among diverse immune cells, we performed trajectories analyses for Treg, CD8+ T, and dendritic cells, using Monocle2 (version 2.8.0; http://cole-trapnell-lab.github.io/monocle-release/monocle2/). The data of the indicated clusters calculated in Seurat was fed directly into Monocle2. Next, we carried out density peak clustering (Monocle2 dpFeature procedure) to order cells based on the genes with differential expression between clusters, using the differentialGeneTest function in Monocle2. The top 1,000–2,000 significant genes (ordered by q value) were used for ordering in all instances. Then the immune cell differentiation trajectory was inferred after dimension reduction and cell ordering with the default parameters of Monocle2.TCR repertoire analysisThe outputs of CellRanger vdj included the assembled nucleotide sequences for both α and β chains, the coding potential of the nucleotide sequences (that is productive or not), the translated amino acid sequence, the CDR3 sequences, and the estimated UMI value of α or β chains. Only cells with UMI values larger than 1 for α and β chains were kept. The dominant TCR of a single cell was defined based on an in-frame TCR α-β pair. If one clonotype defined as a unique in-frame TCR α-β pair was present in at least two cells, this clonotype would be considered clonal, and the number of cells with such dominant α-β pair indicated the degree of clonality of the clonotype.R package STARTRAC (version 0.1.0) was used to assess the enrichment of TCR in various T cell clusters. The degree of clonal expansion, tissue migration, and state transition of T cell clusters upon TCR tracking were determined using three STARTRAC indices, STARTRAC-expa, STARTRAC-migr, and STARTRAC-tran, respectively. For detailed pipeline, please referred to the website (https://github.com/Japrin/STARTRAC).Bulk RNA sequencing and data analysisTo identify the genes responsible for EBV infection and carcinogenesis, we compared the expression profiles between NPC and non-cancerous control cohorts. The data for the NPC cohort was retrieved from the public GEO database (Accession number: GSE102349), including 113 NPC tissue samples profiled by RNA-seq71. The control cohort was in-house RNA-seq data of 10 rhinitis samples that had been published recently72. For data analysis, pair-end reads with high quality were aligned to ribosome RNAs using Bowtie273, and reads after removal of those being aligned as ribosome RNAs were realigned to the human genome (GRCh38) and EBV (Akata) reference sequence using HISAT2 with default settings74. HTseq was used to quantitate the read counts of each gene75. The expression levels of genes were normalised as Transcripts Per Kilobase Million (TPM), to minimise the potential effect of tumour purity.We next assessed the associations of the immune signatures and the survival of NPC. Receiver operating characteristic (ROC) was used to determine the optimal cut-off value of gene expression for patient stratification. Kaplan-Meier analysis was conducted to reveal the prognostic ability of normalised mRNA ratio of CCR8/FOXP3 in 88 NPC samples with prognostic information from the public cohort, and a log-rank test was performed to compare the survival between high and low normalised mRNA ratios of CCR8/FOXP3. For the correlation analyses of the expression of immune signatures genes among LAMP3+ DC (DC_C3_LAMP3), Treg cell, and exhausted CD8+ T cell (CD8_C11_PDCD1), we first selected the signature genes based on the top 200 differentially expressed genes among all cell subsets (Supplementary Data 7) and then calculated the mean of the expression (TPM) for signature genes as signature scores. Pearson correlation between signatures was calculated by cor() function in R programme.Assembly of context specific regulatory models and master regulator analysis of LAMP3+ dendritic cell (DC)A LAMP3+ DC context-specific network model of transcriptional regulation was assembled with the ARACNe76 (https://github.com/califano-lab/ARACNe-AP), based on 357 LAMP3+ DC (DC_C3_LAMP3) single cell RNA-sequencing expression profiles. ARACNe was run with 100 bootstraps, a P value threshold of 10−8, and 0 data processing inequality (DPI) tolerance, generating a network of 38 TFs associated with 2,495 target genes by 10,759 interactions. Pearson correlation between TFs and target genes was calculated by cor() function in R. We considered the correlation coefficient between TF and downstream target genes greater than 0 as positive regulation, otherwise negative regulation.inferCNV analysisTo identify malignant cells, we identified evidence for somatic alterations of large-scale chromosomal copy number variants, either gains or losses, in a single cell using inferCNV (https://github.com/broadinstitute/inferCNV), in addition to the expression of EPCAM. The raw single-cell gene expression data was extracted from the Seurat object according to the software recommendation. A public single-cell data derived from normal epithelium cells was included as a control reference32 (GEO accession number: GSE121600). We preformed inferCNV analysis with the default parameters.Cellular communication analysisTo investigate the potential cell-cell communications between any two different cell types in NPC, we performed ligand-receptor analyses using CellPhoneDB software (version 2.0.6; https://github.com/Teichlab/cellphonedb). CellPhoneDB applies an algorithm that considers only receptors and ligands with broad expression among the tested cell types, followed by calculating the likelihood of cell-type specificity of a given receptor-ligand complex with a sufficient number of permutations. The gene expression matrixes of CD8+ T, CD4+ T, NK, B, myeloid, and malignant cells were selected as input for the CellPhoneDB analysis. We identified the most relevant cell-type specific ligand-receptor interactions and considered only ligands and receptors with expression in more than 20% of the cells in the corresponding sub-clusters. Moreover, we permuted the change of cell type label for each cell at 1000 times to calculate the significance of each pair. The P value was calculated using the proportion of the mean value for specific receptor–ligand pairs compared to a randomly permuted mean distribution. Finally, we prioritised the interactions with a P-value greater than 0.05 and selected the interaction pairs with biologically relevance.Immunostaining assaysFor tissue sample stored in the formalin, dehydration and embedding in paraffin were performed according to routine methods. The paraffin blocks were cut into 5 µm slides and adhered on the glass slides. The paraffin-embedded sections were dewaxed, rehydrated, subjected to the blockade of endogenous peroxidase activity, and antigen retrieval at high-temperature. Subsequently, the sections were processed further for either multiplex immunofluorescence (IF) or immunohistochemistry (IHC) staining assays.Multiplex IF staining assays were conducted to determine the presence of EBV+ and EBV− NPC cells. The sections were permeabilized in PBS with 0.5% Triton X-100 (Sigma-Aldrich; Cat. no. T8787) and incubated for overnight at 4 °C with the following primary antibodies: anti-EPCAM (rabbit; Abcam, USA; Cat. no. ab71916; 1:100) and anti-LMP1 (mouse; Abcam; Cat. no. ab78113; 1:500). Subsequently, the sections were incubated with Cy3 conjugated Goat Anti-Rabbit IgG and FITC conjugated Goat Anti-Mouse IgG secondary antibodies (Servicebio, China; Cat. no. GB21303/GB22301; 1:300). Nuclei were counterstained with 4’-6’-diamidino-2-phenylindole (DAPI; Sigma-Aldrich; Cat. no. D9542). Images were captured using a confocal laser-scanning microscope (LSM880; Zeiss, Germany).To determine the spatial contact of LAMP3+ DCs, Treg cells, and CD8+ T cells, we performed multiplex IHC staining assays using the PANO 7-plex IHC kit (Panovue, China) according to the manufacturer’s instructions. The slides were incubated with blocking antibody diluent at room temperature for 10 min, and then incubated overnight at 4 °C with primary antibodies. The slides were then incubated with the secondary antibody (HRP polymer, anti-mouse/rabbit IgG) at room temperature for 10 min. Subsequently, fluorophore (tyramide signal amplification or TSA plus working solution) was applied to the sections, followed by heat-treatment with microwave. The primary antibodies were applied sequentially, followed by incubation with the secondary antibody and TSA treatment. Nuclei were stained with DAPI after all the antigens had been labelled. Multispectral images for each stained slide were captured using the Mantra System (PerkinElmer, USA). Primary antibodies included anti-CD3 (rabbit; Abcam; Cat. no. ab135372; 1:50), anti-CD4 (rabbit; Abcam; Cat. no. ab133616; 1:100), anti-CD8A (mouse; CST, USA; Cat. no. CST7030; 1:100), anti-FOXP3 (mouse; Abcam; Cat. no. ab22510; 1:100), anti-CD80 (mouse; R&D Systems, USA; Cat. no. MAB140; 1:80), anti-PD1 (mouse; CST; Cat. no. CST43248; 1:50), anti-PD-L1 (rabbit; CST; Cat. no. CST13684; 1:50), and anti-CTLA4 (rabbit; Abcam; Cat. no. ab237712; 1:100).Statistics analysisAll statistical analyses were performed using R (http://www.r-project.org), including two-sided paired student t-test, two-sided Wilcoxon test, two-sided Pearson correlation test, and two-sided Kruskal–Wallis test. P < 0.05 was considered as statistical significance. Multiplex IF and IHC staining assays were confirmed in at least three biological replicates.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary informationSupplementary InformationPeer Review FileDescription of Additional Supplementary FilesSupplementary Data 1Supplementary Data 2Supplementary Data 3Supplementary Data 4Supplementary Data 5Supplementary Data 6Supplementary Data 7Reporting Summary | nature communications | [
"Article"
] | [
"Cancer microenvironment",
"Head and neck cancer",
"Tumour heterogeneity",
"Tumour immunology"
] | IntroductionNasopharyngeal carcinoma (NPC) head neck cancer linked with Epstein-Barr virus (EBV ethnic geographic prevalence high incidence rate 15–50 cases per 100,000 Southern China Southeast Asia to 0.4 per 100,000 western diagnosed advanced stages non-specific symptoms Radiotherapy primary treatment radiosensitive tumour Survival outcomes improve 5-year survival rate 85.6%3,4 benefited from radiotherapy platinum-based chemotherapy 10% patients develop recurrent metastatic NPC2 studies response rate 11.7% 25.9–34% to targeted inhibitor epidermal growth factor immunotherapies immune checkpoint variations treatment responses survival outcomes indicate heterogeneous NPC genetic makeup genomic instability foster genetic diversity cancer tumour heterogeneity Genome sequencing analyses diverse somatic alterations in NPC tumours high mutational frequencies at CYLD NFKBIA TP53 CDKN2A/B accumulated mutations in MHC class I genes chromatin modification genes poor survival diverse normal cells tumour microenvironment) maintenance cancerHeterogeneous immune cells stromal cells characterised transcriptional profiling cancers subtypes gene signatures TME important for tumour progression treatment responses11 Profound infiltration lymphocytes observed in biopsies NPC different high density tumour infiltrating lymphocytes favourable survival outcomes diverse cell populations TME not illustrated in NPC studies demonstrated T cells various functional states different immune cells in NPC tumours single-cell transcriptome work aim comprehensive view tumour heterogeneity NPC analysing single-cell transcriptional profiles 176,447 cells from 10 treatment-naïve patients NPC T cell receptor) repertoire information tumour-blood sample pairs characterise clonality migrations T cells generate potential cellular interaction network TME NPC view cell composition tumour biopsy PBMC performed single-cell RNA sequencing TCR repertoire sequencing on viable cells biopsies peripheral blood mononuclear cells) for 10 patients EBV-positive NPC prior anti-cancer treatment (Fig obtained 380 million sequencing reads each sample median sequencing saturation.(75.90%–94.50% Supplementary Table 2) quality control filters (low expression genes inferred doublets 176,447 cells identified from 10 patients (including 82,622 93,825 for tumours PBMC obtained 1500 genes 4950 unique molecular identifiers (UMIs) each cell sufficient coverage representative transcripts Fig. 1d Data 1).Fig. landscape profiling single cells in NPC tumours PBMC experimental scheme diagram study design cells collected flow cytometry sorting) subjected cell barcoding cDNA libraries 5’-mRNA expression TCR constructed independently high throughput sequencing downstream analyses UMAP plot of 176,447 single cells six cell types normalised expression marker genes dot cell coloured type colour grey to blue low to high expression UMAP plot single cells coloured origins peripheral blood tumour fraction cell types each patient origin similar expression profiles unsupervised clustering analysis Seurat software19 distribution cell clusters patient matched other patients potential variation expression batch effect sample processing minimalcluster identified as cell subpopulation variable genes markers including CD4+ T cells PTPRC CD3D CD4) CD8+ T cells (PTPRC CD3D CD8A), myeloid cells (CD14 ITGAX CD11C), malignant cells (EPCAM KRT5) B cells (CD19 MS4A1) NK cells (FCGR3A NCAM1 detected 56 fibroblasts seven endothelial cells sparse distribution among TME cells cell types widespread in tumour samples heterogeneous cell composition TME in NPC consistent with-cell study proportions CD8+ T B cells increased 1.34 2.33 times NK cells decreased in tumours distinct immune landscapes between tumour peripheral blood compared cell compositions between NPC other cancers single-cell data-cell lung cancer colorectal cancer pancreatic ductal adenocarcinoma common occurrence of infiltrating immune cells heterogeneity higher proportion of T cells in NPC cancers consistent tumour infiltrating leucocytes main characteristic NPC stroma20Heterogeneity T cells diversity TCR abundance T NK NPC tumour samples anti-tumour capabilities explored structure subtypes T NK cell populations grouped 141,875 T NK cells 32 subgroups majority CD4+ CD8+ T cells. 2a performed differential gene expression) analysis T cell clusters CD4+ CD8+ T cell clusters overexpression exhaustion markers (LAG3 TIGIT PDCD1 HAVCR2 CTLA4) effector molecules (GZMB GZMK INFG NKG7 GNLY IL2 high expression proliferative signatures CD8_C10_MKI67 Treg_C3_MKI67 co-expression exhaustion effector genes tumour T cells demonstrated NPC17 T cells anti-tumour effects antigens effector functions suppressed TME NPC observed naïve gene signatures expression TCF7 SELL CCR7 LEF1) resting T cells CD4_C1_LEF1 CD8_C1_LEF1 CD8_C2_TCF7 DN_LEF1 cell clusters Treg_C1_SELL DP_TCF7identified two clusters CD4+ Th1-like cells tumours_C1_CCR7_C2_TNF naïve T cell markers pro-inflammatory cytokines common expression Th1-like cell (CXCL13 BHLHE40 CXCR3 Fig. 2Expression profile development CD8+ T cells UMAP plot 141,875 T cells 32 cell types dot represents cell coloured Violin plots expression CD8+ T cell markers cluster Cell clusters expression level gene x- y-axis Pseudotime trajectory analysis CD8+ T cells_C5_C7_C11 n = 10,000 high variable genes dot cell coloured cluster label inlet plot cell pseudotime score dark blue to yellow early terminal states CD8+ T cell 10,000 cells randomly selected Box plots transition-index scores exhausted CD8+ T cells (CD8_C11_PDCD1) T cells Comparison two-sided Wilcoxon test Cell clusters transition-index scores x- y-axis Endpoints minimum maximum values median values whiskers interquartile range coloured dots patientBox plots showed expansion PBMC-Tumour migration-index scores CD8+ T cell cluster (n = 10). comparison using Wilcoxon test or Kruskal–Wallis test Cell clusters indicated at x-axis y-axis shows expansion PBMC-Tumour migration-index scores Endpoints minimum maximum values centre lines median values whiskers 1.5× interquartile range coloured dots patient performed T cell receptor repertoire analysis revealed 38,720 (32.97% out of 117,447) T cells with detectable TCR α-β pairs after quality control 2e no sharing identical TCR clonotype among patients with NPC preferences V J fragments pairs variable CDR3 sequences across samples CAVRGTGTASKLTF CASSFSGANVLTF associated with recognition MLANA EBV antigens in VDJ CD4+ CD8+ T cells have more clonal T cells derived from identical TCR clonotypes suggesting clonal expansion of dominant clones tumour T cells upon stimulations antigensCD8+ T cells exhausted intratumoral identified 62,244 cells NPC samples grouped 11 clusters markers naïve_C1_LEF1_C2_TCF7) blood memory_C3_KLRB1) memory_C4_KLRG1) high migration_C5_CX3CR1) tumour memory_C6_C7_GZMK_C8 tissue resident memory_C9_XCL1) high proliferating_C10_MKI67) exhausted (CD8_C11_PDCD1) cells majority>97.68%) CD8_C6_IL7R CD8_C7_GZMK_C8_MHC_C9_XCL_C10_MKI67_C11_PDCD1 NPC tumours majority>94.85%) CD8_C1_LEF1 CD8_C2_TCF7 CD8_C3_KLRB1 CD8_C4_KLRG1 CD8_C5_CX3CR1 peripheral blood functional status CD8+ cells calculated cytotoxicity proliferation exhaustion scores clusters highest cytotoxicity CD8_C5_CX3CR1 proliferation CD8_C10_MKI67 exhaustion CD8_C11_PDCD12c 4a), potential cytotoxic proliferation exhausted functions DEG analysis revealed high expression chemokine receptors (CX3CR1 S1P receptors (S1PR1 integrins (ITGB2 ITGA4 ITGAL ITGB7) in CD8_C5_CX3CR1 CD8+ T cell Fig 4b). signalling pathway enrichment analyses tumour cytotoxic CD8+ T cell clusters (CD8_C7_GZMK_C8_MHC_C9_XCL1) enriched pathways cytokine production lymphocyte activation CD8_C5_CX3CR1 enriched pathways leukocyte trans-endothelial migration migration Fig consistent peripheral circulation infiltrating development CD8+ T cells NPC performed pseudotime trajectory analysis Monocle2 expression transition profiles observed developmental trajectories from CD8_C5_CX3CR1_C10_MKI67 to CD8_C11_PDCD1 (Fig. 2c). CD8_C10_MKI67 intermediate exhaustion score lower expression exhaustion markers PDCD1 HAVCR2TCR sequencing revealed 21,099 CD8+ T cells 62,244) clonotypes Fig 4d CD8_C11_PDCD1 shared identical TCRs CD8+ T cell clusters 17.68% to 41.67% infiltrating 5.31% peripheral CD8_C5_CX3CR1 relationships clusters quantitated expansion migration transition gene TCR STARTRAC method24. highest transition mobility CD8_C11_PDCD1 CD8_C10_MKI67 followed CD8_C7_GZMK CD8_C8_MHC CD8_C9_XCL1 CD8_C11_PDCD1 cells expanded pre-exhausted intratumoral CD8+ T cells CD8_C5_CX3CR1 largest clonal T cells highest expansion mobility CD8+ T cell clusters CD8_C5_CX3CR1 highest proportion shared TCR between peripheral blood tumour TCRs shared CD8_C5_CX3CR1 ranged 4.76% to 12.77% infiltrating CD8+ T cell clustersdata suggest origin intratumoral CD8+ T cells NPC tumour from peripheral blood CD8_C5_CX3CR1 diversity Treg cells suppressors immune essential immunological homoeostasis identified 11,631 Treg cells (CD4 IL2RA FOXP3) grouped four cell clusters Treg_C1_SELL_C2_HSPA1A_C3_MKI67_C4_TNFRSF4 (Fig. 2a proportion Treg cells among CD4+ T cells tumours higher than PBMC Treg_C4_TNFRSF4 majority Treg_C2_HSPA1A (99.5%; 4,762 out of 4,786)_C3_MKI67 (90.0%; 1,187 out of 1,319) in tumours Treg_C1_SELL cells in PBMC immune-regulatory functions Treg cells calculated IL2R scores CD25 (IL2RA), CD122 CD132 (IL2RG) AddModuleScore function Seurat software genes encode transmembrane proteins receptor complex IL2 inhibit effector T highest IL2R score Treg_C4_TNFRSF4 among Treg clustersstrongest IL-2 potential Treg_C4_TNFRSF4 cells observed highest inhibitory co-stimulatory scores cells expression genes immune-inhibitory (Fig. 3a Treg cells stronger suppression potential immune response activated cells cells identified CRC NSCLC hepatocellular carcinoma high activation immune-suppressive potential high IL2R inhibitory co scores Fig. observed elevated expression levels chemokine receptors Treg_C4_TNFRSF4 cells CXCR3 CXCR6 CCR8 implicated Fig. 5a).Fig. 3Expression profile development Treg cells Violin plots IL2R inhibitory co-stimulatory) scores each Treg cell cluster (n = 11,631) Box plots quartiles score levels Endpoints minimum maximum values median values whiskers interquartile range black dots each cell plots coloured cell types signature scores y-axis Heatmap signalling pathways enriched GO KEGG analyses each Treg cell cluster Filled colours blue to red represent scaled expression levels low to highP values calculated hypergeometric test adjusted comparisons Orange purple results GO KEGG signalling pathways analysis Pseudotime trajectory analysis Treg cells_C1 n = 11,631) high variable genes dot represents cell coloured cluster label inlet plot cell pseudotime score dark blue to yellow early terminal states Box plots expansion migration-index scores CD4+ T cell cluster (n = 10). Comparison two-sided Wilcoxon test Cell clusters x-axis y-axis expansion migration-index Endpoints minimum maximum values lines median values whiskers interquartile range coloured dots denote each patient Box plots transition-index scores Treg_C4_TNFRSF4 Treg_C2_HSPA1A Treg cells (n = 10). Comparison two Kruskal-Wallis test Cell clusters transition-index scores x y-axis Endpoints minimum maximum median values whiskers interquartile range coloured dots each patient potential functions Treg cells performed signalling pathway enrichment analyses cluster expression distinct pattern pathway enrichment cluster functions‘cytokine receptor enriched in Treg_C4_TNFRSF4 consistent chemotactic potentials ‘interleukin-10-κB signalling enriched in Treg_C4_TNFRSF4 Treg_C2_HSPA1A compared pathways between Treg_C4_TNFRSF4_C2_HSPA1A Gene Set Enrichment Analysis higher enrichment pathways cell cycle chemokine TGF-β regulation T cell proliferation transcription factors activating NF-κB STAT pathways in Treg_C4_TNFRSF4 Treg_C4_TNFRSF4 expressed CCR8 used normalised mRNA ratio CCR8/FOXP3 Treg_C4_TNFRSF4 cells Treg cells higher ratio CCR8/FOXP3 decreased progression-free survival higher fraction Treg_C4_TNFRSF4 cells strong immune-suppressive function Treg cells pseudotime trajectory analysis Monocle2 terminal status highest pseudotime scores for Treg_C4_TNFRSF4 cells developmental trajectories from Treg_C1_SELL PBMC Treg_C3_MKI67 tumoursexamined expression tissue resident markers (CD69 ITGAE BHLHE40 in Treg cells higher expression ITGAE BHLHE40 Treg_C4_TNFRSF4 Treg_C3_MKI67 cells than Treg_C2_HSPA1A_C1_SELL majority Treg_C2_HSPA1A cells in tumours originated from Treg_C1_SELL cells scarce expression resident markers suggest recent recruitment cells from peripheral blood.TCR analysis revealed 17,621 (out of 47,384) CD4+ T cells clonotypes Treg_C2_HSPA1A Treg_C4_TNFRSF4 intermediate numbers clonotypes Treg_C4_TNFRSF4 largest proportion clonal cells highest clonality among CD4+ T cells highest expansion score among Treg cell clusters highest migration score for Treg_C1_SELL DEG analysis Treg_C1_SELL high expression chemokine receptors CCR4 counterparts CCL5 CCL17 CCL22 CD8+ T NK myeloid cellsobservations suggest migration chemotactic interaction intratumoral movement Treg_C1_SELL cells tumour observed shared TCRs between Treg cells tumour peripheral blood consistent Treg cells recruited from peripheral examined transition mobility Treg_C2_HSPA1A Treg_C4_TNFRSF4 Treg cells Treg_C4_TNFRSF4 cells highest Treg_C3_MKI67 cells followed Treg_C2_HSPA1A_C1_SELL cells_C2_HSPA1A high Treg_C4_TNFRSF4_C3_MKI67 cells. observations supported developmental trajectory intratumoral Treg_C4_TNFRSF4 cells from_C1_SELL cells Treg_C2_HSPA1A Treg_C3_MKI67 cells. B cells identified 22,892 B cells grouped nine clusters 7a B_C1_TCL1A B_C2_FCRL3 Plamsa_C1_IgA clusters derived from PBMC six tumour samplesDEG analysis revealed gene signatures B cell clusters tumour samples B_C5_ISG15 interferon induced genes B_C6_HSPA1A stressful gene expression Plasma_C2_IgG elevated IgH genes identified two B cell clusters (B_C1_TCL1A B_C4_IFITM3) before class recombination expression IGHM IGHD Table 5) correlation analysis expression TCL1A IGHM IGHD classify B cells before recombination Signalling pathway enrichment analyses B cell clusters enriched pathways immune regulation B_C4_IFITM3 B_C5_ISG15 B_C6_HSPA1A ‘EBV response interferon-gamma’ pathways responsible immune response EBV infection-associated LAMP3+ DCs display tolerogenic phenotype 8,893 myeloid cells identified clustered 10 subsets one mast cells five monocyte macrophage cells three conventional dendritic cells one plasmacytoid dendritic cells (Fig. 4a Supplementary Fig. 8a 8bfour clusters dendritic cells DC_C2_CD1C DC_C3_LAMP3 DC_C4_JCHAIN derived from tumours DC_C1_FCER1A peripheral blood monocyte-like DC monocyte marker S100A8 8a identified DC_C3_LAMP3 cells high maturation activation migration potentials NPC expression signature genes (LAMP3 MARCKSL1 activation (CD80 CD83 CD40) migration (CCR7 FSCN1 SLCO5A1 DC_LAMP3 cells high expression chemokine ligands (CCL17 CCL19 CCL22) recruit immune cells CCR4 CCR7 CXCR3 correlations marker gene LAMP3 functional genes maturation migration activation chemokine ligands in DC_C3_LAMP3 suggest DC_C3_LAMP3 cells be LAMP3+ DCs high migration activation maturation cancers.Fig. 4Expression development dendritic cells UMAP plot 8,893 myeloid cells 10 cell types Each dot cell colouredHeatmap showed expression genes maturation activation migration chemokine ligand three cell clusters colours black to yellow represent gene expression low to high Heatmap signalling pathways GO KEGG terms three clusters colours blue to red represent expression levels low to high P-values calculated one hypergeometric test adjusted multiple comparisons Orange purple represent results GO KEGG signalling pathways analysis Violin plots showed differentiation apoptosis antigen presentation dysfunction scores three Box plots quartiles score levels Endpoints minimum maximum values lines median values whiskers interquartile range black dots denote each cell Cell clusters signature scores indicated x- y-axis Pseudotime trajectory analysis three clusters high variable genes Each dot represents cell coloured cluster label inlet plot cell score dark blue to yellow early terminal states Venn diagram showed overlapped transcription factors regulating LAMP3 gene immune-suppressive molecules HLA-II in DC_C3_LAMP3 cellsSignalling pathway enrichment analyses GO KEGG revealed enriched pathways three DC cell clusters ‘antigen processing upregulated in DC_C2_CD1C downregulated in DC_C3_LAMP3 apoptosis-κB MAPK signalling pathways myeloid cell differentiation upregulated in DC_C3_LAMP3 consistent with GSEA analyses Fig scored expression levels genes pathways each cluster 3) highest differentiation apoptosis lowest antigen presentation for DC_C3_LAMP3 gene signatures activation immune response reduced in DC_C3_LAMP3 consistent with highest immune-regulatory score increased expression immune-suppressive genes CD274 PDCD1LG2 CD200 EBI3 IDO1 IL4I1 SOCS1 SOCS3 9a 3) similar expression profile LAMP3+ DC among NPC, HCC NSCLC suggest DC_C3_LAMP3 cells regulatory tolerogenic DCs restrain activation Tperformed pseudotime trajectory analysis DC_C1_FCER1A cells developed into branches DC_C2_CD1C DC_C3_LAMP3 cells_C3_LAMP3 highest pseudotime score most differentiated matured DC immune-regulatory antigen-presenting scores suggest DC_C1_FCER1A cells infiltrate tumour convert DC_C2_CD1C immune-suppressive DC_C3_LAMP3 cells observed changes expression transcription factors genes DC_C1_FCER1A to DC_C2_CD1C DC_C3_LAMP3 cells constructed cellular network regulons transcription factors signalling molecules DC_C3_LAMP3 ARACNe upregulation LAMP3 linked with TFs ETV3 ETV6 HMGN3 GPBP1 TRAFD1 ATF3 KDM2B JUN HIVEP1 KLF6 ZBTB10 maturation DC downregulation LAMP3 linked with CREMobserved TFs KDM2B KLF6 ETV6 JUN HMGN3 TRAFD1 NFKB1 REL RELB NF-κB pathway linked expression immune-suppressive molecules CD274 PDCD1LG2 CD200 IDO1 downregulated expression HLA-class II genes (Fig. 4f SOX4 CREM associated downregulation CD274 CD200 IDO1 suggest regulate immune-suppressive function antigen-presenting capacity maturation DC_C3_LAMP3 in NPC.Heterogeneity malignant cells EBV infection identified 2,787 malignant epithelial cells NPC tumours large-scale chromosomal copy number variation) normal EBV factor malignant transformation tumorigenesis examined expression EBV molecules malignant cells divided into EBV+_LMP1) EBV− (EP_C2) cells detectable EBV transcripts (LMP-1/BNLF2a/b RPMS1/A73 LMP-2A observed higher expression EPHA2 EGFR in EP_C1_LMP1 cells related susceptibility EBVimmunofluorescence staining EBV-encoded protein (LMP1) confirmed EBV+ NPC observed high activations genes-κB Notch pathways chemokines CX3CL1 EP_C1_LMP1 compared EP_C2_EPCAM cells high expression CX3CL1 NPC tumours (n = 113) compared non-cancerous samples 10 overexpression CX3CR1 receptor CX3CL1 immune cells peripheral blood Signalling pathway enrichment analyses EP_C1_LMP1 enriched cytokine-mediated regulation cell death apoptosis cancer-related pathways (Fig. 5f). observations suggest malignant NPC cells different susceptibility EBV infection distinct expression profiles.Fig. 5Heterogeneity malignant cells EBV infection tumour tissues Heatmap large-scale CNVs epithelial cells 10 NPC tumours average expression 100 genes chromosomal position Red gains blue losses Malignant NPC cells different patients chromosomes indicated different colour barsUMAP plot 2,787 malignant cells clusters (EP_C1_LAMP1 EP_C2 Each dot represents cell coloured UMAP plots expression EBV-encoded genes (LMP-1/BNLF2a/b RPMS1/A73 LMP-2A/B BNRF1) dot represents cell colour grey to red low to high expression Violin plots expression cluster markers chemokines genes NF-κB Notch pathways cell clusters expression level indicated x- y-axis images multiplex immunofluorescence staining malignant cells NPC Proteins detected antibodies red green orange arrows cells positive EPCAM LMP1 co-expression proteins Images three biological replicates Scale bars 50 μm Bar plots signalling pathways KEGG EBV+ malignant cells (EP_C1_LMP1) EBV- cells (EP_C2 coloured light to dark −log10(P-values) low high P-values calculated one-sided hypergeometric test adjusted multiple comparisons heterogeneity malignant NPC cells divided five cell subgroups (C1–C5 clustering analysis without EBV informationdeciphered variations gene expression malignant cells analysis revealed distinct enrichment signalling pathway each cluster observed variable proportions cell subtypes tumour samples inter-tumour heterogeneous expression profiles GSVA data showed different enrichments signalling pathways samples P02 P08 P11 P15 samples higher proportion C4 cluster showed enrichment cell cycle (E2F MYC G2M checkpoint) pathways C4 cluster highest proliferation scores low content in NPC tumours suggest intra- inter-tumour heterogeneity malignant cells NPC.Intercellular interaction network examined ligand-receptor among cell clusters NPC tumours CellPhoneDB observed cellular interactions among DC_C3_LAMP3 cells Treg cells CD8+ T cells via inhibitory chemokines (Fig 6a DC_C3_LAMP3 cells interact with Treg_C1_SELL cells through CCL17-CCR4 CCL22-CCR4 recruiting Treg cells tumourTreg_C4_TNFRSF4 cells high expression CTLA4 ENTPD1 CSF1 ligand-receptor bindings CD80/CD86 ADORA2A SIRPA DC_C3_LAMP3 cells potential interaction DC_C3_LAMP3 cells cells interact with CD8_C11_PDCD1 cells CD200-CD200R signalling non-classical immune-suppressive pathway anti-tumour Potential ligand-receptor interactions observed between Treg_C4_TNFRSF4 CD8_C11_PDCD1 cells chemokines (CCL4-CCR8) adhesive connection (ITGAL-ICAM1 SELPLG immune regulation (HAVCR2-LGALS9 immune-suppressive activity CD8+ T cells interactions observed NPC cohort NPC sample collection correlations gene signatures DC_C3_LAMP3 cells Treg cells exhausted CD8+ T cells suggest widespread occurrence immune-regulatory interactions DC_C3_LAMP3 Treg_C4_TNFRSF4 CD8_C11_PDCD1 cells in NPC tumoursperformed multiplex immunohistochemistry staining NPC biopsies confirmed juxtapositions CD80 DC_C3_LAMP3 cells CTLA4-expressing Treg cells PD-L1-expressing DC_C3_LAMP3 PD-1 CD8+ T Fig. 6Intercellular interactions immune malignant cells NPC ligand-receptor interactions cell clusters Treg DC_C3_LAMP3 cells CD8+ T_C3_LAMP3 ligand-receptor interactions cell-cell interactions indicated columns average expression levels colour heatmap blue red low high expression log10(P-values) circle size colour boxes represent function modules receptor-ligand interactions images multiplex IHC staining juxtaposition CTLA4-expressing Treg cells+CD4+FOXP3+) CD80 DC_C3_LAMP3 cells NPC tissue samples Proteins detected antibodies top green red magenta cyan orange arrows positive cells CD3 CD4 FOXP3 CTLA4 CD80 proteins tissue three biological replicates Scale bars 100 μm 20 μmimages multiplex IHC staining CD8+ T cells PD DC_C3_LAMP3 cells NPC tissue samples Proteins antibodies yellow cyan magenta red green positive cells CD3 CD8 CD80 PD1 PD-L1 proteins three biological replicates Scale bars 100 20 μm top bottom panels malignant NPC immune cells EBV+ EP_C1_LMP1 cells receptor-ligand interactions EBV−EP_C2_EPCAM cells NPC patient EP_C1_LMP1 cells expressed CX3CL1 tumour interact CX3CR1 peripheral immune cells CD8_C5_CX3CR1 cells DC_C1_FCER1A cells NK cells monocytes chemotactic potential EP_C1_LMP1 cells immune cells peripheral EGFR EP_C1_LMP1 cells bind TGFB1 multiple cell types EBV life interacting pairs EP_C1_LMP1 cells activated Notch pathway multiple cell types NOTCH1-TNF NOTCH2-JAG2 related radiation cancer stem-like population NPCinteractions determined variable intensity patients Data 6) widespread heterogeneous in NPC Fig. single-cell transcriptome study NPC view heterogeneous cell composition complex interacting network tumour microenvironment peripheral blood Transcriptome analyses 176,000 cells 53 subtypes two microenvironments tumour peripheral blood identified novel cell populations gene signatures NPC TCR repertoire sequencing delineated potential developmental trajectories intratumoral immune cells dissected multiple intercellular network NPC ligand-receptor paring analyses (Fig. 7) diagram cross-talks immune cells TME NPC infects nasopharyngeal epithelial cells tumorigenic process NPC EBV-positive malignant NPC cells secret chemokines initiate recruitment tumoral infiltration immune cells tumour immune cells activate EGFR Notch pathway cells Naive CD8+ cells infiltrate develop CD8+ cells DCs infiltrate differentiate LAMP3+ DCsLAMP3+ DCs PD-L1/PD-L2 interact with PD1 CD8+ T cells activation promotes exhaustion Treg cells interact LAMP3+ DCs CTLA4-CD80/CD86 limit antigen presentation promote secretion IDO1 proliferation interactions among LAMP3+ DCs Treg cells exhausted CD8+ T cells malignant cells foster immune-suppressive niche tumour microenvironment.CD8+ T cells key anti-tumour immunity tumour observed abundance tumour infiltrating CD8+ T cells NPC clonal expansion proliferation exhausted status suppressed tumour neoantigens consistent findings cancer immunosuppressed state CD8+ T cells tumours TCR repertoire transcriptome analyses revealed differentiation trajectory CD8+ T cells CX3CR1+CD8+ T cells transformed to exhausted CD8+ T cells tumours CX3CR1+CD8+ T cells shared TCR clonotypes with tumour infiltrating CD8+ T cells recognising antigensprevious findings CX3CR1+CD8+ T cells essential for viral control tumour site findings raise possible T cell therapies for NPC infusing CX3CR1+CD8+ T cells peripheral blood ex vivo expansion T cells CX3CR1 potential tumour site cancer cells grow high immunogenicity inflammatory cells immunosuppressive regulatory T cells evade immune destruction cytotoxic lymphocytes46 observed three Treg cell clusters in NPC tumours specific gene signatures functions Treg_C4_TNFRSF4 cells high expression TNFRSF family genes CCR8 strongest immune-suppressive function Treg cells TNFRSF genes facilitate tumour immune evasion promote cancer high expression CCR8 Treg cells immunity associated poor prognosis study revealed Treg_C4_TNFRSF4 cells poor survival NPC CCR8 immune-suppressive Treg cells26 blockade CCR8 signalling might abolish suppressive effect inhibit tumour growth presence tumoral Treg cells in NPC confirmed IHC staining assays origin tumour infiltrating Treg cells remains exclusiveTCR analyses Treg_C4_TNFRSF4 cells differentiated from Treg_C1_SELL cells through Treg_C2_HSPA1A Treg_C3_MKI67 cells reducing migration Treg cells to tumour immune suppressions high expression CCR4 in Treg_C1_SELL cells potential target key molecule Treg cells migrating-associated myeloid cells heterogeneous observed clusters in NPC tumours DC_C3_LAMP3 regulatory tolerogenic DC high expression migration (CCR7) maturation (LAMP3) genes signatures LAMP3+ DCs DC_C3_LAMP3 cells elevated expression CD274 PDCD1LG2 CD200 IDO1 EBI3 SOCS1 SOCS2 SOCS3 immune-suppressive genes similar LAMP3+ DCs lung suggest DC_C3_LAMP3 cells group LAMP3+ DCs NPC ubiquitous in tumours immune-regulatory function T cell data revealed developmental trajectory LAMP3+ DCs master transcription factors crucial for maturation decreased antigen presentation increased immune regulatory capability LAMP3+transcription factors connected gene expression network LAMP3+ DCs targeting reshape dendritic cells normal antigen presentation therapeutic benefits NPC.EBV infection key feature NPC endemic regions role malignant transformation tumorigenesis identified two groups malignant cells with (EP_C1_LMP1) or without (EP_C2_EPCAM) EBV infection in NPC tumours EBV+ NPC cells distinct transcriptional state EBV− activation genes EBV entry cancer pathways NF-κB53 EBV+ NPC cells more chemotactic interactions with immune cells tumours peripheral blood important EBV life cycle tumour behaviours radiation chemotactic interactions explain infiltration immune cells in NPC tumour stroma55 all endemic NPC EBV positive findings future studies role EBV infection NPC development persistence EBV presence EBV molecules in NPC cells harnessed adoptive immunotherapy for NPC T lymphocyte TCRs recognising EBV antigen patients NPC may engage T cells targeting NPC cells explore heterogeneous malignant NPC cells EBV negative cells proliferative variable treatment outcomesdidn’t observe EBV transcripts B cells primary host low expression EBV genes infected cells small B cells captured study study revealed intercellular network among LAMP3+ DCs Treg cells exhausted CD8+ T cells NPC potential cross-talks immune-suppressive niche TME NPC (Fig. link between LAMP3+ DCs Treg cells CCL17-CCR4-CCR4 demonstrated in other potentiate chemotactic recruitment Treg cells promote infiltration DCs expression IDO1 NPC induce proliferation tumour Treg cells interaction between tumoral Treg cells LAMP3+ DCs CTLA4 CD80/CD86 reported in other Treg cells regulate maturation tolerogenic LAMP3+ DCs cross-talks between cytotoxic CD8+ T cells DCs Treg cells addressed in multiple interactions LAMP3+ DCs Treg cells enhance immune-suppressive effects exhausted CD8+ T cells NPC chemotactic potential EBV+ NPC cells recruit peripheral immune cells suggest cross-talks cell homoeostasis TME NPCplausible disrupting interactions might break balance TME cure tumour promising results PD-1/PD-L1 immunotherapy NPC antibodies targeting EGFR enhance treatment paradigms advanced NPC64 suspect immune-suppressive interaction CD200-CD200R1 LGALS9-HAVCR2 LAMP3+ DCs Treg cells exhausted CD8+ T cells potential immunotherapeutic targets NPC uncovering heterogeneous tumour microenvironment NPC identified essential cells molecules potential contributions NPC tumorigenesis insights mechanisms NPC progression development potential strategies study limitations observed trends associations cell compositions clinical characteristics patients advanced NPC higher proportion peripheral CD8_C10_MKI67 intensive cellular interactions DC_C4_JCHAIN NK_C2_FCER1G cells not conclusion limited cohort size components TME associated clinical outcomes believe NPC shaped with immunity neoantigens somatic alterations malignant cells predict neoantigens no DNA level data patient samples HLA estimations somatic mutationsfindings interaction network molecules TME in NPC obtained bioinformatic analyses immunostaining assays further experiments awaited biological consequences mechanisms recruitment sample collectionTen male individuals nasopharyngeal carcinoma recruited hospital Guangzhou China high prevalence June 2018 September 2018. histopathologically diagnosed primary NPC by two pathologists World No history cancer anti-tumour self-reported Clinical staging NPC determined International Union against Cancer American Joint Committee on Cancer staging system Fresh tumour sample obtained endoscopic nasopharyngeal biopsy peripheral blood sample single cell preparation patients EBV positive hybridisation RNAs average age 50.6 characteristics listed supplementary immunostaining additional NPC biopsies collected specimens collected 30 min after tumour resection fixed in formalin 48 h Written informed consent obtained study approved by Institutional Review Boards Sun Yat-sen University Cancer Center single cell tumour samples processed enzymatic digestion mechanical dissociation single cell suspensions tumour cut into pieces D10 resuspension buffer culture medium10% foetal bovine serum Gibco enzymatic type II Fisher 17101015) IV 17104019) digestion 30 min 37 °C digested mixture passed 40 μm cell-strainer Biosciences 352340) dissociated cells filtered mixture centrifuged 400 g 5 min pelleted cells resuspended 0.8% NH4Cl red blood cell lysis buffer incubated 10 min washing DPBS dissociated cells resuspended sorting buffer 1X DPBS 0.04% BSA (Sigma-Aldrich USA 9048468) Viable cells collected fluorescence activated cell sorting negative staining propidium iodide 300,000 cells collected each tissue sample blood PBMCs isolated leukocyte separation solution Sigma-Aldrich 10771) 5-ml fresh peripheral blood collected EDTA anticoagulant tubes 366643) transferred solution density gradient centrifugation 20 min 750 X g PBMCs settled collected washed twice DPBS Residual red blood cells lysed procedure Viable cells collected FACS PI stainingLibrary construction single cell gene expression TCR profilingImmune repertoire measurement gene expression single cell Chromium Single Cell V(D)J Reagent Kit (10x Genomics USA sorted cells washed twice buffer Cell viability number determined Trypan Blue exclusion assay cell suspension 700–1200 cells/μl loaded each channel cells per sample mixed barcoded gel beads Chromium Controller (10x cDNA amplification 14 cycles thermal cycler (C1000 Bio-Rad post-amplification cDNA enrich TCR fragments Sequencing libraries cDNA TCR separately constructed average fragment size quantitated Qseq100 (Bioptic Taiwan).Next generation sequencing data DNA library loaded sequencing lane HiSeq X system (Illumina USA sequenced pair-end reads 150 bp data converted to FASTQ files Cell Ranger pipelines 3.0.1 sequencing reads genomes generate feature-barcode matrices Single-cell 5'-gene expression data TCR enriched data library processed Cell Ranger count vdj gene expression data mapped human genome reference sequence (GRCh38gz EBV sequence65 (Akata/flemingtonlab/public/tree/master/annotation cDNA sequencing TCR enriched data mapped VDJ sequence.10Xgenomics/cell-vdj-cellranger-vdj-GRCh38-alts-ensembl-2.0.0.tar.gz TCR sequencing-cell gene expression quantification determination cell typesDoublets artefactual libraries cells errors droplet encapsulation affect quality single-cell sequencing data R package “DoubletFinder” predict doublets doublet single-cell library more than one cell cluster doublets more no cell type both markers removed doublets individually expected doublet rate 0.05 default parameters Table 3) remaining cells single cells gene expression matrices PBMC tumour cells combined converted to Seurat object R package Seurat (version 2.3.4 cells removed less than 101 UMIs less than 501 genes over 15% UMIs linked mitochondrial genes remaining cells gene expression matrices generated log normalisation linear regression NormalizeData ScaleData Seurat packagesamples processing independently high-dimensional variables common single-cell sequencing potential batch effect used canonical correlation analysis RunUMAP function Seurat reduce remove batch effect Cell clusters identified FindClusters function Seurat K parameter 20 default parameters annotated clusters major cell types average gene expression CD4+ T cell (PTPRC CD3D CD4) CD8+ T cell (PTPRC CD3D myeloid cell (CD14 malignant cell (EPCAM KRT B cell (CD19 MS4A1) NK cell (FCGR3A NCAM1) identified sub-clusters annotated cell subtypes by average expression gene sets cell expression profiles contrasted other Seurat FindAllMarkers function Differential expression analysis compared genes datasets default two-sided non-parametric Wilcoxon rank sum test significant differentially expressed gene Bonferroin-adjusted P value lower 0.05 average natural logarithm) fold-change expression least 0.1 0.25 for malignant cells other cells cluster with multiple marker genes cell types elevated UMI considered cell contamination removedeach cluster C1) major cell type CD4+ T cells), assigned cluster identifier marker gene LEF1) “CD4_C1_LEF1”. selection criteria marker gene top ranking differential gene expression analysis cell cluster strong specificity gene expression high expression ratio low other clusters literature supports marker gene or functional relevant type cell.Collection public single-cell datasetsTo compare tumour microenvironment cell compositions between NPC other cancers collected single-cell data multiple cancers including NSCLC12 CRC66 PDAC67 specific cell types CD4+ T cells in CRC68 GSE146771), NSCLC11 GSE99254) HCC28 GSE140228) DCs in NSCLC29 GSE127465 HCC28 GSE140228).Calculation functional module scoresTo evaluate potential functions cell cluster calculated scores functional modules AddModuleScore function in Seurat single cell level average expression levels cluster subtracted by aggregated expression control feature sets analysed genes binned based on averaged expression control features randomly selected each bin69.functional modules proliferation T malignant cells cytotoxicity exhausted scores CD8+ T IL2R inhibitory co-stimulatory scores Treg cells maturation activation migration differentiation apoptosis antigen presentation immune regulatory scores dendritic cells genes listed supplementary material 3).Pathway enrichment cell cluster performed Gene Ontology KEGG Pathway enrichment analyses Metascape pathways enriched gene list malignant cells gene list top 100 differentially expressed genes > 0.1 non-malignant cells top 100 DEGs lnFC > 0.25 P value <0.05 significant enrichment difference signalling pathway enrichment clusters_C2_HSPA1A Treg_C4_TNFRSF4 DC_C2_CD1C DC_C3_LAMP3) performed gene set enrichment analysis molecular signatures database v7.070 heterogeneous expression malignant cells performed gene set variation analysis 18 hallmark pathways.Developmental trajectory potential immune cell functional changes lineage differentiation cells performed trajectories analyses Treg CD8+ T dendritic cells Monocle2/monocle-release data clusters Seurat fed into Monocle2. carried density peak clustering cells genes differential expression differentialGeneTest function Monocle2. top 1,000–2,000 significant genes by q value used for ordering immune cell differentiation trajectory inferred after reduction cell ordering default parameters Monocle2.TCR repertoire outputs CellRanger vdj included nucleotide sequences α β chains coding potential translated amino acid sequence CDR3 sequences estimated UMI value α β chains cells UMI values larger than 1 α β chains kept dominant TCR defined in-frame TCR α-β pair unique TCR α-β pair present in two cells clonal number cells dominant α-β pair indicated clonality STARTRAC (version 0.1.0) enrichment TCR T cell clusters clonal expansion tissue migration state transition determined using STARTRAC indices-expa-migr-tran website/Japrin/STARTRAC).Bulk RNA sequencing data genes EBV infection carcinogenesis compared expression profiles between NPC non-cancerous control cohortsdata NPC cohort retrieved from public GEO database 113 NPC tissue samples profiled by RNA-seq71 control cohort in-house RNA-seq data 10 rhinitis samples published pair-end reads quality aligned to ribosome RNAs Bowtie273 after realigned to human genome EBV) reference sequence HISAT2 HTseq read counts expression levels normalised as Transcripts Per Kilobase Million tumour purity assessed associations immune signatures survival NPC Receiver operating characteristic (ROC) optimal cut-off value gene expression patient stratification Kaplan-Meier analysis prognostic ability normalised mRNA ratio CCR8/FOXP3 in 88 NPC samples log-rank test survival between high low mRNA ratios correlation analyses expression immune signatures genes among LAMP3+ DC Treg cell CD8+ T cell selected signature genes top 200 expressed calculated mean expression (TPM) Pearson correlation between signatures calculated by cor() function R programmecontext regulatory models analysis LAMP3+ dendritic cell network model transcriptional regulation ARACNe76 357 LAMP3+ DC single cell RNA-sequencing expression profiles ARACNe run 100 bootstraps P value threshold 10−8 0 data processing inequality (DPI) tolerance 38 TFs 2,495 target genes 10,759 interactions Pearson correlation TFs target genes calculated cor() function correlation coefficient TF target genes greater 0 positive regulation negative regulation.inferCNV malignant cells identified evidence alterations large-scale chromosomal copy number variants single cell inferCNV raw single-cell gene expression data extracted Seurat object public single-cell data normal epithelium cells control reference32 (GEO accession number GSE121600). inferCNV analysis default parameters.Cellular communication potential cell-cell communications cell types ligand-receptor analyses CellPhoneDB software 2.0.6 receptors ligands broad expression cell types likelihood cell-type specificity receptor-ligand complex permutationsgene expression matrixes CD8+ T CD4+ T NK B myeloid malignant cells selected for CellPhoneDB analysis identified relevant cell-type ligand-receptor interactions considered ligands receptors 20% cells permuted cell type label 1000 times significance P value calculated mean receptor–ligand pairs prioritised interactions P-value greater than 0.05 selected pairs relevance.Immunostaining assaysFor tissue sample formalin dehydration paraffin performed paraffin blocks cut into 5 μm slides adhered glass slides-embedded sections dewaxed rehydrated endogenous peroxidase activity antigen retrieval high-temperature sections processed for multiplex immunofluorescence immunohistochemistry) staining assays IF staining assays presence EBV+ EBV− NPC cells sections permeabilized PBS with 0.5% Triton X-100 incubated overnight at 4 °C primary antibodies anti-EPCAM anti-LMP1 sections incubated with Cy3 conjugated Goat Anti-Rabbit IgG FITC conjugated Goat Anti-Mouse IgG secondary antibodiesGB21303/GB22301 1:300). Nuclei counterstained 4’-6’-diamidino-2-phenylindole (DAPI Sigma-Aldrich Cat.. D9542). Images captured confocal laser-scanning microscope (LSM880 Zeiss contact LAMP3+ DCs Treg cells CD8+ T cells performed multiplex IHC staining assays PANO 7-plex IHC kit (Panovue China slides incubated blocking antibody diluent room temperature 10 min incubated overnight 4 °C primary antibodies incubated secondary antibody (HRP polymer anti/rabbit IgG) room temperature 10 min fluorophore applied heat-treatment microwave primary antibodies applied sequentially incubation secondary antibody TSA treatment Nuclei stained DAPI after antigens labelled Multispectral images each stained slide captured Mantra System (PerkinElmer Primary antibodies anti-CD3-CD4 anti-CD8A anti-FOXP3 anti-CD80 anti-PD1 anti-PD-L1 (rabbitCST13684 1:50), anti-CTLA4 (rabbit Abcam Cat.. ab237712 1:100).Statistics analyses performed using R.r-project-sided student t-test Wilcoxon test Pearson correlation Kruskal–Wallis test P < 0.05 statistical significance Multiplex IF IHC staining assays confirmed three biological replicates Nature Research Reporting Summary.Supplementary Review Additional Supplementary FilesSupplementary Data Summary | 50.3 | 1.52948 |
10.1038/s41467-021-21158-8 | PMC7870970 | Here, the authors report intrinsic donor bound dark exciton states with associated phonon replicas in monolayer WSe2, and defect control crystal synthesis for the deterministic creation of these states. | The monolayer transition metal dichalcogenides are an emergent semiconductor platform exhibiting rich excitonic physics with coupled spin-valley degree of freedom and optical addressability. Here, we report a new series of low energy excitonic emission lines in the photoluminescence spectrum of ultraclean monolayer WSe2. These excitonic satellites are composed of three major peaks with energy separations matching known phonons, and appear only with electron doping. They possess homogenous spatial and spectral distribution, strong power saturation, and anomalously long population (>6 µs) and polarization lifetimes (>100 ns). Resonant excitation of the free inter- and intravalley bright trions leads to opposite optical orientation of the satellites, while excitation of the free dark trion resonance suppresses the satellitesʼ photoluminescence. Defect-controlled crystal synthesis and scanning tunneling microscopy measurements provide corroboration that these features are dark excitons bound to dilute donors, along with associated phonon replicas. Our work opens opportunities to engineer homogenous single emitters and explore collective quantum optical phenomena using intrinsic donor-bound excitons in ultraclean 2D semiconductors. | IntroductionA promising route for optical encoding of matter is to employ excitons, Coulomb-bound electron-hole pairs, which are elementary optical excitations in semiconductors. Monolayer transition metal dichalcogenides (TMDs) are an emergent platform for exploring excitonic physics at the two-dimensional (2D) limit. This is largely due to their strong light-matter interactions, easy access to electric and magnetic control, and unique combination of spin-valley coupling and valley contrasting circular dichroism1–4. Encapsulation of monolayer TMDs within hexagonal boron nitride (hBN) has led to drastically improved sample quality5,6, allowing the identification and detailed studies of a variety of optically bright and dark valley excitonic states7–21, and their phonon replicas22,23 assisted by both zone center24,25 and zone edge phonons26,27.Despite the rapid progress in sample quality and understanding of their excitonic physics, these monolayer semiconductors are still far from perfect, and questions remain. For example, while localized single-photon emitters have been observed in these materials28–31, they exhibit random emission energies over a broad spectral range (>100 meV), and commonly lose the desirable valley optical selection rules due to the random anisotropy. Moreover, they generally appear in low quality samples, where inhomogeneous broadening obscures the underlying rich excitonic manifold that has been observed in clean samples. While the precise nature of these quantum light sources remains unclear, O interstitials32,33 and extrinsic confinement potentials such as strain appear to play a role in the localized emission34–38. However, even the cleanest samples reported to date inevitably contain intrinsic defects (e.g. self-flux growth WSe2 crystal with defect density of ~1 × 1011 cm−2)39. Several of these native defects have been identified and their electronic structure probed using scanning tunneling microscopy (STM) and spectroscopy40–43. A natural question arises as to if new spectral features may arise from excitons bound to such dilute intrinsic defects in these ultraclean monolayer crystals44.In this work, we report the observation of donor bound dark excitons and their phonon replicas in ultraclean monolayers of WSe2 encapsulated in hBN. The samples in our study (N = 10) were fabricated from different sources of WSe2 and hBN bulk crystals and yielded consistent and reproducible results. The data presented in the main text are mainly from two devices. For electrostatic control of the charge carrier density in the WSe2, a local graphite bottom gate is used. Fig. 1a, b show an optical microscope image of Device 1 and its schematic, respectively. The following experiments were performed at a temperature of either 1.6 or 4 K, and with excitation laser of energy 1.96 eV, unless otherwise specified.Fig. 1Emergence of deeply bound exciton satellites.a Optical image of electrostatic gating device composed of exfoliated monolayer WSe2 (red area) encapsulated in hBN with graphite backgate (outlined in black). Scale bar is 10 µm. b Schematic of sample side view. c Photoluminescence as a function of back gate voltage with the laser spot indicated by the red dot in a. Three satellites peaks near 1.60 eV appear when the sample is n-doped. d Spatial map of the integrated PL from the satellite peaks at Vb = 0.5 V. Integration spectral region shown by arrows on top of c, and spatial region outlined by dashed white lines in a. e Spatial map of the peak energy of the highest-energy satellite at Vb = 0.5 V. f Waterfall plot of PL spectra from three different samples, showing homogeneous satellite binding energies and robust three peak spectral features. The energy axis is scaled relative to the free neutral exciton (X0 at ~1.735 eV).ResultsObservation of robust excitonic satellitesThe PL intensity of Device 1 as a function of back gate voltage (Vb) and emission energy is shown in Fig. 1c. The excitation power is 5 µW. All reported excitonic features such as excitons, trions, dark states, and various phonon replicas in the spectral range of 1.65 to 1.75 eV are well-resolved (Supplementary Fig. 1). The sample is also nearly intrinsic, with charge neutrality occurring at Vb ≈ −0.25 V. All these factors attest to the high sample quality. An observation is the emergence of low energy excitonic satellites occurring between 1.58 and 1.63 eV under n-type doping. The three main features, indicated as S1, S2, and S3 from high to low energy, are the focus of this paper.The excitonic satellites have homogeneous spatial distribution and energy across the entire sample. Figure 1d displays the spatial map of the satellite emission intensity (spectral range shown on top of Fig. 1c at Vb = 0.5 V), while Fig. 1e shows the spatial distribution of S1 peak energy. Its peak position relative to neutral exciton is unchanged across the whole sample (See Supplementary Fig. 2). Remarkably, the spectral structure and energetic positions of the excitonic satellites are also consistent across many samples. The normalized PL spectrum of three exemplary samples under similar experimental conditions is shown in Fig. 1f. The satellite emission energies, relative to the neutral exciton, between total 10 different samples fabricated over a 3-year span with different crystal sources are well aligned (Supplementary Fig. 3). From the spatial and energetic homogeneity of the satellite PL, we can safely rule out their origin from contamination, cracks, or other extrinsic confinement potentials.Evidence of intrinsic defect-bound excitonsThe satellite emission exhibits distinct power dependence compared to the known free 2D excitons. Figure 2a presents the normalized PL spectra of Device 2 with powers ranging over three orders of magnitude (20 nW to 60 µW). The satellite emission dominates the entire spectrum at low powers, but quickly saturates and the free 2D excitonic manifold appears as the power increases. Above 10 µW excitation power, the satellite emission is overwhelmed by the linear response of the higher energy 2D exciton species (see Supplementary Fig. 4 for additional samples). The strong power saturation is a hallmark of defect-localized excitons45, and the low saturation threshold implies low defect density and long exciton lifetime. Note that the integrated photoluminescence saturation count rate is comparable to that of single-photon emitters31.Fig. 2Power, polarization, and time-resolved PL from satellites.a Waterfall plot of normalized PL spectra at selected excitation powers. The exciton satellites dominate the spectrum at low powers. b Polarization-resolved spectrum of satellite peaks under σ+circularly polarized excitation (1 µW). Inset: power dependence of η, the degree of circular polarization, which grows with increasing power up to η ≈ 0.6. c Time-resolved PL of S1 reveals a population lifetime of 6.95 ± 0.05 μs and a polarization lifetime of 116 ± 4 ns (red line is single exponential fit). All data are from Device 2 at Vb = 0 V. Top arrows in b indicate the spectral region of integration.Circularly polarized optical pumping reveals non-trivial polarization of the satellite PL. We observe that the two highest-energy satellites, S1 and S2, are co-circularly polarized with the excitation laser, while the lowest energy satellite S3 is nearly unpolarized, as shown in Fig. 2b for Device 2. The degree of circular polarization is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\eta \equiv (I_ + - I_ - )/(I_ + + I_ - )$$\end{document}η≡(I+−I−)/(I++I−), where I±denotes the intensity of the σ± polarized components of the PL. For S1, we observe that |η| ~ 0 at low excitation power (<100 nW), but increases with the excitation power to a value of |η| ≈ 0.6 at powers above ~10 μW (inset of Fig. 2b, integration over S1). Optical orientation of exciton spins is often observed in excitons bound to defects, where the localized electrons can become spin polarized, e.g. via efficient exchange interactions with photoexcited free electron spins46. The strong power dependence of circular polarization is thus another signature of localized excitonic spin states. We note that under linearly polarized excitation, the satellite emission is not polarized in the linear basis, regardless of the axis of excitation (Supplementary Fig. 5). The lack of linear polarization and preservation of the circular optical selection rules imply underlying rotational symmetry (C3) of the defect. This behavior is different from all reported quantum emitters found in WSe2, which emit with linear polarization with strong spatial and spectral inhomogeneity28–31.Time-resolved PL reveals anomalously long lifetime of the satellite states, as shown in Fig. 2c. A bi-exponential fit of the long PL component yields a population lifetime of 6.95 ± 0.05 μs. This is 3–5 orders of magnitude larger than that of both 2D bright and dark excitonic species13,47,48, and consistent with the observed strong saturation of the satellite PL. We further extract the circular polarization lifetime of the satellite emission to be 116 ± 4 ns, as determined by fitting η with a single exponential (inset of Fig. 2c). The polarization lifetime is also ~2–4 orders of magnitude longer than that of 2D bright48 and dark excitons41,42. Such an enormously long population lifetime implies a dramatic reduction in the non-radiative lifetime of the satellite state. Moreover, the long polarization lifetime suggests small intervalley scattering rate, and weak transverse-longitudinal splitting, which further supports high symmetry (C3) of the underlying confinement center.All the above experimental features confirm that the excitonic satellites have distinct origin from both localized quantum emitters and free excitonic species previously reported in monolayer WSe2. The spatial and spectral homogeneity across many samples, unusually long population and polarization lifetimes, and strong pump power dependence in both emission intensity and polarization, imply their origin as excitonic states bound to dilute intrinsic defects. The strong dependence of the observed PL features on electron doping is the first indication of the defect type as donors.Controllable synthesis of crystals with intrinsic defect-bound excitonsFigure 3a–c shows the wide field STM topographic images of WSe2 with three different growth parameters (see Methods, labeled as F1, F2, and F3). Two main types of defects, bright and dark features in these images, are observed. Scan tunneling spectroscopy reveals in gap states, implying the bright and dark defects are donor and acceptor in nature, respectively (see Supplementary Fig. 6 and Supplementary Note 1 for details). Across samples, F1, F2, and F3, the total defect density increases from 2 × 1011 to 4 × 1011, and to 8 × 1011 cm−2, while the donor density decreases from 1.5 × 1011 cm−2, to 1.3 × 1011 cm−2, to 4 × 1010 cm−2. Figures 3d, e show the corresponding gate dependent photoluminescence of monolayers exfoliated from these three types of crystal. The excitonic satellites are clear in sample F1 with abundant donors but lowest total defect density, are barely visible in F2, and are not seen in sample F3. The satellites peak intensity thus track the donor density, supporting assignment of the observed photoluminescence satellites to excitonic states bound to dilute donor defects.Fig. 3Characterization of defects and defect-bound excitons in WSe2.a–c Scanning Tunneling Microscopy (STM) topographic images of 50 x 50 nm2 area of WSe2 with different growth parameters (F1, F2, and F3). Scale bar is 10nm. Imaging conditions for the STM topographic images were a tunneling bias of 1.4 V and current of 400 pA for F1 and a tunneling bias of 1.4 V and current of 150 pA for F2 and F3. The WSe2 crystals were cleaved in ultra-high vacuum STM chamber (base pressure <2.0 × 10−10 torr) to obtain a clean surface before imaging. White and black dashed circles highlight the donor and acceptor defects, respectively. See Supplementary Fig. 6 details. d–f Photoluminescence (PL) color maps as a function of the back-gate bias for d, F1, e, F2, and f, F3. The white dashed boxes in the PL color maps indicate the energy and the back-gate bias ranges for defect-bound excitons. For the PL color map, the samples were excited using a continuous-wave (CW) laser with an excitation wavelength of 532 nm with a fluence of 650 W/cm2 (10 μW) at 4 K.Donor bound dark excitons and phonon replicasWe examine the energy difference between the observed satellites peaks. The extracted energies of S2 and S3, relative to S1, at zero gate voltage are (12.5 ± 0.7) meV and (23.1 ± 0.5) meV, respectively, where the error bar is the standard deviation of the peak positions from 10 samples. These energy differences are similar to the calculated phonon energy of K2 (12 meV) and Γ5 (22 meV), respectively26. We thereby attribute S1 to the zero-phonon line of donor bound excitonic emission, while S2 and S3 as its K2 and Γ5 (or E′) phonon replicas. The assignment of S2 and S3 resembles the observed phonon replicas for free 2D excitons, pointing to the strong phonon-exciton interactions in monolayer WSe224–27. The E″ phonon replica from a localized quantum emitter have also been observed in monolayer WSe2,49 albeit with different emitter properties from those reported here.Since in WSe2 the ground state configuration of a free exciton is the spin-forbidden dark exciton Xd,26,27 we infer that the zero-phonon line S1 arises from a donor bound dark exciton DXd. The dark exciton long lifetime makes possible efficient formation of DXd, even at very low exciton density. This state should be composed by a positive charged ion, a donor bound electron, and an electron–hole pair. For the lowest energy configuration, the two electrons should have opposite spins to form spin singlet state.Photoluminescence excitation spectroscopyThe above understanding of the charge configuration of DXd is further supported by polarization-resolved PL excitation spectroscopy (PLE). We tuned the energy of a continuous wave laser from 1.65 to 1.75 eV while collecting the satellite emission. As a reference, the PL spectrum obtained with 1.96 eV excitation is provided in Fig. 4a. The polarization-resolved PL response of the satellites under σ+ excitation is shown in the left (σ+ detection) and middle (σ− detection) panels of Fig. 4b. The extracted PL polarization η is shown in the right panel with the satellite PL under 1.96 eV excitation overlaid in black. Comparing Fig. 4a, b, we can see that satellite emission is greatly enhanced when the laser is in resonance with free neutral exciton, which produces large population of dark excitons and thus DXd. The satellite emission is suppressed when resonant with the dark trion. This can be understood as the creation of dark trion competes with the formation of dark exciton, and thus donor bound dark exciton states.Fig. 4Excitation energy dependent spin-valley depletion and PL polarization reversal.All data are from Device #1 at Vb = 0.5 V. a PL spectra with a HeNe laser excitation. b Polarization-resolved PL spectra by sweeping σ+polarized excitation from 1.75 to 1.65 eV. The σ+ and σ− components of the PL are shown in the left and middle panels, respectively, while the right panel shows the extracted degree of polarization η. c Polarization-dependent PL with the excitation in resonance with the free intervalley (top, 1.706 eV) and intravalley (bottom, 1.699 eV) trions, showing polarization reversal of the exciton satellite PL.A notable feature is the polarization-dependent response when the laser is in resonance with free bright trions. As shown in the η panel in Fig. 4b, the blue regions represent PL polarization reversal when the laser is resonant with the intervalley trion (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T^ -$$\end{document}XT−), opposite to that under excitation at the intravalley resonance (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_S^ -$$\end{document}XS−). This contrast is highlighted by polarization-dependent spectra in Fig. 4c. The satellite emission is co- and cross-circularly polarized with laser in resonance with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_S^ -$$\end{document}XS− (E = 1.699 eV, bottom panel) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T^ -$$\end{document}XT− (E = 1.706 eV, top panel), respectively. Note that the vertical white stripe in η at ~1.59 eV indicates that S3 has negligible circular polarization.DiscussionDonor spin state initializationFigure 5 explains the above trion excitation dependent PL polarization via donor spin state initialization process (see Supplementary Fig. 7 and Supplementary Note 2 for details). Figure 5a depicts the valley-spin coupled band edges, with initially unpolarized electrons on shallow donor levels50. Using resonant excitation of intervalley trion \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T^ -$$\end{document}XT− in Fig. 5b as an example, σ+ polarized excitation creates an electron and hole pair in the K valley, which pairs with a spin-up electron in the K′ valley and forms \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T^ -$$\end{document}XT−. This optical pumping process depletes the spin-up electrons associated with donors in K′, eventually resulting in a net population of spin down electrons in the K valley bound to the shallow donors (denoted as DK,↓). DK,↓ then selectively binds to a dark exciton in the K′ valley (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_d^{K\prime }$$\end{document}XdK′), because of the Pauli exclusion. The donor bound state emission can happen via either defect assisted direct electron-hole recombination yielding S1 PL peak, or phonon assisted stokes emission (either K2 for S2 peak or Γ5 for S3 peak), via an intermediate bright trion state (Fig. 5c). In all cases, the emission is σ− polarized as determined by the hole valley configuration, cross-polarized to the excitation26.Fig. 5Schematic of donor bound dark excitons and spin state initialization.a Top is a schematic of the spin-valley coupled band edges. Green and brown lines denote electron (hole) spin pointing up (down) and down (up), respectively. Dashed lines indicate a shallow donor level, where carrier’s spin-valley locked index is preserved in the relatively smooth trapping potential. Bottom cartoon illustrates the donor configuration D where an electron (blue circle) is trapped by a positive charge center. Under σ+ polarized resonant excitation of free intervalley trion \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T^ -$$\end{document}XT−, b the formation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T^ -$$\end{document}XT− (top cartoon) depletes the donor electron spins in K′ valley. This results in optical orientation of spin and valley polarized electron coupled to the shallow donor (bottom cartoon, DK,↓). DK,↓ can then capture a neutral dark exciton in the opposite valley, c, forming a donor bound dark exciton \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$DX_d^{K\prime }$$\end{document}DXdK′. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$DX_d^{K\prime }$$\end{document}DXdK′ can emit light via defect assisted direct electron-hole recombination (peak S1), or via coupling to the bright trion assisted by phonons, which are by either emitting valley conserved but electron spin flipped Γ5 phonon (peak S3), or emitting electron spin conserved but valley flipped K2 phonon (peak S2). The opposite hole valley configuration of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T^ -$$\end{document}XT− and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$DX_d^{K\prime }$$\end{document}DXdK′ dictates that the emitted light of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$DX_d^{K\prime }$$\end{document}DXdK′ is σ- polarized and opposite to that of excitation. d, e illustrates the scenario of resonant excitation of free intravalley \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_S^ -$$\end{document}XS−, which depletes spin down electrons of donor in the same valley. The donor bound dark exciton formed then has the same hole valley configuration, and emit light of the same circular polarization as the excitation. See text for details. Note that the schematics used here are based on single particle picture, for the convenience of explaining spin, valley, and charge degrees of freedom of the quasiparticles, and the stokes process with different phonons.When σ+ polarized excitation is in resonance with intravalley trion \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_S^ -$$\end{document}XS−, it depletes instead the spin down electrons in K valley, and leaves a net population of spin-up electron in K′ valley bound to the shallow donor (DK′,↑), as shown in Fig. 5d. This leads to donor bound dark exciton states (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$DX_d^K$$\end{document}DXdK) with spin-valley polarization opposite to that in resonant excitation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X_T^ -$$\end{document}XT−. The donor bound state then emits σ+ polarized light, co-polarized with the excitation (Fig. 5e). We note that, although pump is Iσ+ polarized, supply of free neutral dark exciton is expected in both valleys because of the ultrafast valley depolarization through electron–hole exchange in the formation of neutral excitons, so the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$DX_d^K$$\end{document}DXdK spin-valley polarization is determined by the optical orientation of donor electron. We would also like to point it out that although S2 behaves similarly to S1, which both have strong circular polarization, S3 is unpolarized. This suggests S3 maybe not be as simple as a Γ5 phonon replica of S1, which needs further investigation.Origin of intrinsic defectOur results demonstrated the observation of excitons bound to intrinsic defects. Extrinsic defects (e.g. O at Se site) and confinement potential, which can be responsible for localized single-photon emitters with strong optical anisotropy32–38, are thus not candidates. There are four types of intrinsic defects: W vacancy, Se vacancy, W replace Se site, and Se replace W site. Although Se vacancy has been suggested to be responsible for observed broad low energy PL features51, our atomically resolution STM measurements (Supplementary Fig. 6e) rules out vacancy as donor candidates. In addition, previous study on similar crystals with low defect density found the chalcogenide vacancy to be very rare39. Calculation also suggested Se at W antisite (SeW) is a deep defect with multiple in gap states32, and the exciton bound to SeW is expected to be about 300 meV below the WSe2 A exciton.32 All these are distinct from our experimental observation: our STS on donor (Supplementary Fig. 6a) only shows one shallow defect band near the CBM, while the donor bound exciton is about 120 meV below the A exciton. The leaves W at Se antisite (Wse) as a possible candidate. However, calculation suggests that Wse is an acceptor, which has multiple charge levels with a band near CBM52. This is distinct from our results that the defect is donor type with only one charge level within the gap. The current study cannot resolve the exact type of the defect, which requires further experimental and theoretical efforts (see discussion).OutlookOur results reveal light emission from dark exciton bound to intrinsic dilute donors and possible phonon replicas in the ultraclean monolayer WSe2. Similar behavior is observed in III–V semiconductors53, where defect density must be low before individual defect-bound exciton peaks and charging states are resolvable. It remains to be seen whether the defect-bound excitons exist in multilayer samples and other transition metal dichalcogenides. The observed long population and polarization lifetimes are advantageous for exploiting the spin-valley functionalities using monolayer WSe2. The possibility of probing single-donor emitting sites by further reducing the native defect density, or by using near-field techniques, is an interesting direction towards realizing optical spin quantum memory in 2D materials. The PLE results are promising in this respect, since they demonstrate optical initialization of the donor-bound electron spin state by selective excitation of different trion states. On the other hand, interactions between neighboring donor sites may lead to interesting many-body behaviors, such as long-range correlation. Non-classical photon statistics are expected to emerge in both regimes, although weak oscillator strength may pose significant challenges. Our work prompts research efforts to identify and engineer the underlying defect and its electronic configuration, and we expect that further improvements in sample quality will enable detailed studies of the satellite fine structure, such as vibrational and rotational spectrum, hyperfine interactions, excited states, and optical orientation by magnetic fields control of donor spin states.MethodsCrystal growthFor growth of WSe2 crystals with varying defect densities (Fig. 3), WSe2 crystals were synthesized by reacting W with Se flux. W powder (99.999%) and Se shot (99.999%) mixed in 1:100 (F1), 1:15 (F2), and 1:5 (F3) atomic ratios, were loaded into quartz ampules separately which were then evacuated and sealed at ~10−6 Torr. The ampules were vertically placed into a box furnace and heated to 1080 °C over 48 h. After a dwelling time of 1 week at 1080 °C, it was slowly cooled down to 300 °C at a rate of 0.6 °C/h. The obtained WSe2 crystals were subsequently filtered from the Se flux by quartz wool and annealed at 275 °C for 24 h in a vacuum quartz ampule.Sample fabricationThe van der Waals heterostructure samples used in this study were fabricated by polycarbonate-based viscoelastic dry-transfer techniques. The polymer residues were cleaned by baths in chloroform and isopropyl alcohol. The individual layers were obtained from the mechanical exfoliation of bulk crystals onto 285 nm of thermally-grown SiO2 on p+ doped Si wafers. The thickness of hBN layers was determined by atomic force microscopy. Monolayers of WSe2 were identified by their optical contrast, which was later confirmed by their low energy PL spectrum.Electrostatic dopingStandard electron beam lithography was used to produce PMMA masks for subsequent electron beam evaporation of V/Au (nominally, 5/50 nm) electrodes to the graphite backgates, as well as to a small portion of the monolayer WSe2 protruding from underneath the top gate dielectric. For fully-encapsulated monolayer WSe2 samples, electrodes were deposited to a second piece of thin graphite that was both in contact with the monolayer WSe2 and also protruding from underneath the top gate dielectric. The applied backgate voltage was controlled by analog output from a National Instruments USB I/O DAQ board using mxdaq drivers in Matlab environment. The gate leakage current was actively monitored using a combination of transimpedance amplifier and analog input on the DAQ board.Photoluminescence spectroscopyPhotoluminescence measurements were performed in a home-built confocal microscope, in reflection geometry, normal to the plane defined by the monolayer WSe2. The samples were either (1) mounted on the cold head of a closed-cycle He cryostat at a temperature of 5 K and studied using an IR-enhanced achromatic 50X objective lens (0.65 NA) or (2) mounted inside a He-exchange-gas cooled cryostat (attocube attoDRY 2100) at a temperature of 1.6 K and studied using an IR-enhanced achromatic 100X objective (0.81 NA) with non-magnetic Ti housing. In both cases, the samples were illuminated by power-stabilized HeNe laser light (λ = 1.96 eV) focused to a beam waist of ~1 µm. Circularly polarized excitation and detection was achieved by an appropriate combination of fixed linear polarizers and λ/4-waveplates, with achromatic λ/2-waveplates mounted in stepper-motor-controlled rotation stages. Linearly polarized excitation and detection was achieved by an appropriate combination of fixed linear polarizers and λ/2-waveplates mounted in stepper-motor controlled rotation stages. The collected PL was directed into a 0.5-m spectrometer, where it was dispersed by a 600-line/mm grating with 750 nm blaze before being detected by Si charge-coupled-device. The polarization of the light entering the spectrometer was S-polarized for all polarization-resolved measurements.Photoluminescence excitation spectroscopyThe excitation source was a narrowband (<20 kHz linewidth) and frequency tunable Ti:Sapphire continuous-wave laser (M2 SolsTiS). The laser line was filtered from the collection path using a combination of bandpass filter and tunable long- and short-pass filters (Semrock VersaChrome). The optical power was stabilized by servo control of the power in the first order diffracted beam using feedback control of the voltage applied to the acoustic-optic modulator.Time resolved photoluminescence spectroscopyTime-resolved PL measurements were performed by directing the collected photons onto an IR-enhanced single-photon avalanche photodiode (Excelitas) connected to a time-correlated single photon counting system (PicoHarp 400). Spectral filtering of the signal was achieved by either (1) a combination of bandpass filter and tunable long- and short-pass filters (Semrock VersaChrome), or (2) using the 0.5-m monochromator to disperse the signal, which was then filtered at the exit port by an adjustable slit assembly. The excitation was provided by spectrally filtered (~2 nm bandwidth at FWHM) output from a supercontinuum fiber laser with 100 kHz repetition rate and ~10 ps pulse duration.Scanning tunneling microscopySTM measurements were performed using a Scienta Omicron STM system at room temperature under an ultra-high vacuum (base pressure <1.0 × 10-10 torr). WSe2 bulk crystals were mounted onto a metallic sample holder with silver epoxy, and then cleaved in situ in the UHV STM chamber to obtain a clean surface. The tungsten tip was cleaned and calibrated against an Au(111) surface before all the measurements. For each STM image, the defect density was calculated through the number of defects divided by the scanning size. To avoid the localized counting, each value was obtained from the average of 30 (50 × 50 nm2) STM images in different area.Supplementary informationSupplementary InformationPeer Review File | nature communications | [
"Article"
] | [
"Two-dimensional materials",
"Two-dimensional materials"
] | route optical encoding excitons Coulomb-bound electron-hole pairs elementary optical excitations semiconductors Monolayer transition metal dichalcogenides (TMDs) platform exploring excitonic physics two-dimensional due to strong light-matter interactions electric magnetic control spin-valley coupling contrasting circular Encapsulation TMDs hexagonal boron nitride improved sample detailed studies bright dark valley excitonic phonon assisted zone edge phonons26 progress sample quality understanding excitonic physics monolayer semiconductors far from perfect questions remain localized single-photon emitters observed exhibit random emission energies>100 lose valley optical selection rules random anisotropy appear in low quality samples inhomogeneous broadening obscures excitonic manifold nature quantum light sources unclear O interstitials32 extrinsic confinement potentials role localized cleanest samples contain intrinsic defects self-flux growth WSe2 crystal defect density ~1 × 1011 cm−2)39 native defects identified electronic structure probed using scanning tunneling microscopy spectroscopy40–43question new spectral features from defects ultraclean monolayer donor bound dark excitons phonon replicas in ultraclean monolayers WSe2 hBN samples (N = 10 from different WSe2 hBN crystals consistent results data from two devices electrostatic control WSe2 local bottom gate used Fig. 1a b optical microscope image Device 1 schematic experiments temperature 1.6 or 4 K excitation laser 1.96 eV. 1Emergence deeply bound exciton satellites image electrostatic gating device exfoliated monolayer WSe2 hBN backgate Scale bar 10 μm Schematic sample side view Photoluminescence back gate voltage laser spot red dot Three satellites peaks near 1.60 eV sample n-doped map integrated PL satellite peaks at Vb = 0.5 V spectral region arrows peak energy highest-energy satellite at Vb = 0.5 V Waterfall plot PL spectra from three samples homogeneous satellite binding energies peak spectral featuresenergy axis scaled to neutral exciton (X0 at ~1.735 robust excitonic PL intensity Device 1 back gate voltage emission energy in Fig. 1c excitation power 5 μW excitonic features excitons trions dark states phonon replicas 1.65 to 1.75 eV-resolved sample nearly intrinsic charge neutrality at Vb ≈ −0.25 V high sample quality low energy excitonic satellites between 1.58 and 1.63 eV under n-type doping three main features S1, S2 S3 high to low energy focus excitonic satellites homogeneous spatial distribution energy across sample Figure 1d satellite emission intensity Vb = 0.5 Fig. 1e S1 peak energy position relative to neutral exciton unchanged across spectral structure energetic positions excitonic satellites consistent across samples normalized PL spectrum of three samples Fig. 1f satellite emission energies relative neutral exciton between 10 samples aligned 3) out origin from contamination cracks extrinsic confinement potentials defect-bound satellite emission distinct power free 2D excitonsFigure 2a PL spectra Device 2 powers (20 to 60 satellite emission dominates low powers saturates 2D excitonic manifold power increases Above 10 μW overwhelmed higher energy 2D exciton Supplementary Fig. 4 strong power saturation defect-localized low saturation threshold low defect density long exciton lifetime integrated photoluminescence saturation comparable single-photon. 2Power polarization time-resolved PL satellites Waterfall plot normalized PL spectra excitation powers satellites dominate low powers Polarization-resolved spectrum satellite peaks+circularly polarized excitation (1 power polarization grows power Time-resolved PL S1 population lifetime 6.95 ± 0.05 μs polarization lifetime 116 ± 4 ns data from Device 2 = 0 V arrows spectral region integration polarized optical pumping non-trivial polarization satellite PL highest-energy satellites S1 S2 co-circularly polarized laser lowest S3 nearly unpolarizedDevice 2. circular polarization defined as[12pt{minimal{amsmath{wasysym\oddsidemargin-69pt}{document (I_ + -\end(I+−I−)/(I++I−), I±denotes intensity σ± polarized components PL S1 |η| ~ 0 at low excitation power<100 increases to ≈ 0.6 powers above ~10 μW Fig. 2b S1) Optical exciton spins observed in bound defects localized electrons spin polarized strong power dependence circular polarization signature of localized excitonic spin states linearly polarized excitation satellite emission not polarized linear axis excitation Fig. 5) lack of linear polarization preservation circular optical selection rules imply rotational symmetry (C3) defect different from quantum emitters WSe2 linear polarization strong spatial spectral inhomogeneity28–31.Time-resolved PL reveals long lifetime satellite states Fig. 2c.bi-exponential fit PL population lifetime 6.95 ± 0.05 μs 3–5 orders larger 2D bright dark excitonic consistent strong saturation satellite PL circular polarization lifetime satellite emission 116 ± 4 ns fitting η single exponential polarization lifetime ~2–4 orders longer 2D dark excitons41 long population lifetime implies reduction non lifetime satellite long polarization lifetime suggests small intervalley scattering weak transverse-longitudinal splitting supports high symmetry (C3) confinement center confirm excitonic satellites distinct origin localized quantum emitters free excitonic species spatial spectral homogeneity long population polarization lifetimes strong pump power dependence emission intensity polarization imply origin excitonic states intrinsic defects PL features electron doping indication defect type synthesis crystals defect-bound excitonsFigure 3a–c wide field STM topographic images WSe2 three growth parameters Two types defects bright dark observed tunneling spectroscopy reveals gap states bright dark defects donor acceptorSupplementary Note 1 samples F1 F2 F3 defect density increases 2 × 1011 to 4 × 1011 8 × 1011 cm−2 donor density decreases 1.5 1.3 4 × 1010 Figures 3d show gate dependent photoluminescence monolayers crystal excitonic satellites clear F1 lowest defect density barely visible F2 not F3 satellites track donor density photoluminescence satellites excitonic states donor defects. defects-bound excitons Scanning Microscopy images 50 x 50 nm2 WSe2 growth parameters (F1 F2 F3) Scale bar 10nm conditions tunneling bias 1.4 V current 400 pA F1 1.4 V 150 pA F2 F3 WSe2 crystals cleaved ultra vacuum STM chamber White black dashed circles highlight donor acceptor defects Supplementary Fig. 6 Photoluminescence (PL) color maps back-gate bias F1 F2 F3 white dashed boxes indicate energy back-gate bias ranges defect-bound excitonsPL color map samples excited laser 532 nm fluence 650 W/cm2 (10) 4 K.Donor bound dark excitons phonon energy difference satellites peaks extracted energies S2 S3 zero gate voltage (12.5 ± 0.7) meV (23.1 ± 0.5) meV error bar standard deviation peak positions samples differences similar phonon energy K2 (12 meV Γ5 (22 S1 zero-phonon line donor bound excitonic emission S2 S3 K2 Γ5 phonon replicas S2 S3 phonon replicas 2D excitons strong phonon-exciton interactions monolayer WSe224–27 E′′ phonon replica quantum emitter observed monolayer WSe2,49 different properties ground exciton spin-forbidden dark exciton Xd zero-phonon line S1 donor bound dark exciton DXd long lifetime formation DXd low exciton density positive charged ion donor bound electron electron–hole pair electrons opposite spins statePhotoluminescence excitation charge configuration DXd supported by polarization-resolved PL excitation spectroscopy tuned energy continuous wave laser from 1.65 to 1.75 eV collecting satellite emission PL spectrum 1.96 eV excitation Fig. 4a polarization-resolved PL response σ+ excitation left middle panels Fig. extracted PL polarization η right panel satellite PL under 1.96 eV excitation black satellite emission enhanced laser with neutral exciton dark excitons DXd satellite emission suppressed resonant with dark trion dark trion competes exciton exciton.Fig. 4Excitation energy dependent spin-valley depletion PL polarization reversal data from Device #1 Vb = 0.5 V spectra HeNe laser excitation Polarization-resolved PL spectra σ excitation from 1.75 to 1.65 eV σ+ σ− components left middle panels right panel extracted polarization η Polarization-dependent PL resonance free intervalley intravalley trions polarization reversal satellite PL polarization-dependent response laser with free bright trions panelblue regions represent PL polarization reversal when laser resonant with intervalley trion\documentclass[12pt]{minimal}{amsmath{wasysym{upgreek-69pt}{document$$X_T^\end{document}XT−), opposite under excitation at intravalley resonance\documentclass[12pt{minimal}{amsmath{upgreek\oddsidemargin}{-69pt}{document}$$X_S^\end{document}XS−). contrast highlighted by polarization-dependent spectra in Fig. 4c.satellite emission co cross-circularly polarized with laser\documentclass[12pt]{minimal{amsmath-69pt$X_S^}XS− (E = 1.699 eV bottom panel[12pt]{minimal-69pt$X_T^}XT− (E = 1.706 eV top panel), vertical white stripe in η at ~1.59 eV S3 negligible circular polarization spin state initializationFigure 5 explains trion excitation dependent PL polarization donor spin state initialization process Supplementary Fig. 7 Supplementary Note 2 Figure 5a valley-spin coupled band edges initially unpolarized electrons on shallow donor levels50resonant excitation intervalley trion\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}{document}$$X_T^\end{document}XT− Fig. 5b σ+ polarized excitation creates electron hole pair K valley pairs spin-up electron K′ valley[12pt]{minimal}{amsmath}{wasysym{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$$X_T^\end{document}XT− optical pumping depletes spin-up electrons donors K′ net population spin down electrons K valley shallow donors DK,↓). DK,↓ binds dark exciton K′ valley[12pt]{minimal}{amsmath}{wasysym}{mathrsfs}{upgreek}\oddsidemargin}{-69pt}{document}$$X_d^{K\prime\end{document}XdK′), Pauli exclusiondonor bound state emission defect assisted electron-hole recombination phonon assisted stokes emission K2 S2 Γ5 S3 intermediate bright trion state. emission σ− polarized hole valley configuration cross-polarized. 5Schematic donor bound dark excitons spin state initialization spin-valley band edges Green brown lines electron spin up Dashed lines shallow donor level spin-valley locked index preserved smooth trapping potential donor configuration D electron trapped positive charge centerσ+ polarized resonant excitation free intervalley trion[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin}{-69pt}{document}$$X_T^\end{document}XT− formation[12pt]{minimal}{amsmath}{upgreek}\oddsidemargin}{-69pt}$$X_T^{document}XT− cartoon depletes donor electron spins K′ valley optical orientation spin valley polarized electron shallow donor (bottom cartoon DK,↓). DK,↓ capture neutral dark exciton opposite valley c donor bound dark exciton[12pt]{minimal}{amsmath}{wasysym}}{upgreek}\setlength{\oddsidemargin}{-69pt}{document}$$DX_d^{K\prime\end{document}DXdK′.[12pt{minimal{amsmath{wasysym\oddsidemargin-69pt{document$DX_d{K\prime}DXdK′ emit light defect assisted electron-hole recombination (peak S1) coupling bright trion assisted emitting valley conserved electron spin flipped Γ5 phonon (peak S3) or valley flipped K2 phonon (peak S2)opposite hole valley configuration\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin}{-69pt}\begin{document}$$X_T^\end{document}XT−[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}{\oddsidemargin}{-69pt}{document}$$DX_d^{K\prime\end{document}DXdK′ dictates emitted light[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs}{upgreek}\setlength{\oddsidemargin}{-69pt}{document}$$DX_d^{K\prime\end{document}DXdK′ σ- polarized opposite excitation.d, e illustrates resonant excitation free intravalley\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{upgreek-69pt}$X_S^}XS− depletes spin down electrons donor same valley donor bound dark exciton same hole valley configuration light same circular polarization excitation See text details schematics based single particle picture explaining spin valley charge degrees freedom quasiparticles stokes process different phonons σ+ polarized excitation resonance intravalley trion[12pt]{minimal}{amsmath{wasysym{upgreek}-69pt}$X_S^}XS− depletes spin down electrons K valley leaves net population spin-up electron K′ valley bound shallow donor (DK′,↑), Fig. 5d.leads to donor bound dark exciton states\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek-69pt}$DX_d^K}DXdK with spin-valley polarization opposite resonant excitation[12pt{minimal{amsmath{wasysym{upgreek-69pt}$$X_T}XT− donor bound state emits σ+ polarized light co-polarized with excitation (Fig. 5e). pump Iσ+ polarized supply free neutral dark exciton expected in both valleys ultrafast valley depolarization through electron–hole exchange formation neutral excitons\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}-69pt}$DX_d^K}DXdK spin-valley polarization determined by optical orientation of donor electronS2 to S1 strong circular polarization S3 unpolarized suggests S3 Γ5 phonon replica S1 needs investigation.Origin intrinsic results excitons bound to intrinsic defects Extrinsic defects O at Se site confinement potential single-photon emitters strong optical not candidates four types of intrinsic defects W vacancy Se vacancy W replace Se site Se W site Se vacancy responsible for low energy PL STM measurements rules out vacancy donor candidates study chalcogenide vacancy Se at W antisite deep defect multiple gap exciton 300 meV below WSe2 A exciton distinct experimental observation STS on donor shows one shallow defect band near CBM donor bound exciton 120 meV below A exciton W at Se antisite (Wse) possible candidate suggests Wse acceptor multiple charge levels band near CBM52 distinct from results defect donor type one charge level within gapstudy defect requires further experimental efforts results reveal light emission dark exciton dilute donors phonon replicas in ultraclean monolayer WSe2. Similar behavior in III–V defect density low before exciton peaks charging states defect-bound excitons in multilayer samples transition metal dichalcogenides long population polarization lifetimes advantageous spin-valley functionalities monolayer WSe2. probing single-donor emitting sites reducing defect density near-field techniques optical spin quantum memory 2D materials PLE results promising demonstrate optical initialization donor-bound electron spin state excitation trion states interactions between donor sites lead many-body behaviors long-range correlation Non-classical photon statistics expected regimes weak oscillator strength challenges work prompts identify defect electronic configuration improvements sample quality studies satellite fine structure vibrational rotational spectrum hyperfine interactions excited states optical magnetic donor spin states varying defect densities synthesized reacting W with Se flux mixed 1:100 1:15 1:5 ratios loaded into quartz ampules evacuated sealed at ~10−6 Torr.ampules placed box furnace heated to 1080 °C 48 h After 1 week cooled to 300 °C 0.6 °C/h WSe2 crystals filtered by quartz wool annealed at 275 °C 24 h vacuum quartz ampule van der Waals heterostructure samples fabricated polycarbonate viscoelastic dry-transfer techniques polymer residues cleaned chloroform isopropyl alcohol layers exfoliation bulk crystals 285 nm thermally-grown SiO2 on p+ doped Si wafers thickness determined by atomic force microscopy Monolayers WSe2 identified optical contrast low energy PL spectrum.Electrostatic electron beam lithography PMMA masks evaporation V/Au 5/50 nm) electrodes to backgates monolayer WSe2 dielectric fully-encapsulated monolayer WSe2 samples electrodes deposited second graphite backgate voltage controlled by analog output National Instruments USB I/O DAQ board gate leakage current monitored transimpedance amplifier analog inputPhotoluminescence measurements home-built confocal microscope reflection geometry monolayer WSe2. samples cold head closed-cycle He cryostat 5 K IR-enhanced achromatic 50X objective lens (0.65 NA He-exchange-gas cooled cryostat 1.6 K IR-enhanced achromatic 100X objective (0.81 NA) non-magnetic Ti housing illuminated power-stabilized HeNe laser light (λ = 1.96 eV beam waist ~1 μm Circularly polarized excitation detection fixed linear polarizers λ/4-waveplates achromatic λ/2-waveplates stepper-motor-controlled rotation stages Linearly polarized polarizers collected PL directed into 0.5-m spectrometer dispersed 600-line/mm grating 750 nm blaze detected Si charge-coupled-device light entering spectrometer S-polarized measurementsPhotoluminescence excitation source narrowband<20 kHz frequency tunable Ti:Sapphire continuous-wave laser laser line filtered bandpass filter tunable long- short-pass filters optical power stabilized by servo control feedback control acoustic-optic modulator resolved photoluminescence measurements collected photons IR-enhanced single avalanche photodiode (Excelitas time-correlated photon counting system (PicoHarp Spectral filtering bandpass filter long short-pass filters 0.5-m monochromator filtered exit port adjustable slit assembly excitation spectrally filtered (~2 nm bandwidth output supercontinuum fiber laser 100 kHz repetition rate ~10 ps pulse duration.Scanning tunneling measurements Scienta Omicron STM system room temperature ultra-high vacuum <1.0 × 10-10 WSe2 bulk crystals metallic sample holder silver epoxy cleaved situ UHV STM chamber clean surface tungsten tip cleaned calibrated against Au(111) surface before measurements defect density calculated defects divided by scanning size from 30 (50 × 50 nm2) STM imagesReview File | 47.9 | 0.698912 |
10.1038/s41467-020-17457-1 | PMC7385628 | Transcription-repair coupling factors (TRCFs) are large ATPases that mediate the preferential repair of the transcribed DNA strand. Here the authors reveal the cryo-EM structure of DNA-bound Mfd, the bacterial TRCF, and provide molecular insights into its mode of action. | Mfd couples transcription to nucleotide excision repair, and acts on RNA polymerases when elongation is impeded. Depending on impediment severity, this action results in either transcription termination or elongation rescue, which rely on ATP-dependent Mfd translocation on DNA. Due to its role in antibiotic resistance, Mfd is also emerging as a prime target for developing anti-evolution drugs. Here we report the structure of DNA-bound Mfd, which reveals large DNA-induced structural changes that are linked to the active site via ATPase motif VI. These changes relieve autoinhibitory contacts between the N- and C-termini and unmask UvrA recognition determinants. We also demonstrate that translocation relies on a threonine in motif Ic, widely conserved in translocases, and a family-specific histidine near motif IVa, reminiscent of the “arginine clamp” of RNA helicases. Thus, Mfd employs a mode of DNA recognition that at its core is common to ss/ds translocases that act on DNA or RNA. | IntroductionGiven that DNA serves as a track for multiple essential molecular machines, including those carrying out replication, transcription, and repair, the movement of nucleic-acid translocases along single-stranded (ss) or double-stranded (ds) DNA is highly regulated spatiotemporally1. This control has a central role in development, neurodegeneration, aging, and the diseased state2, and despite advances, still remains insufficiently understood. Translocation on dsDNA has been challenging to study structurally because many DNA motors have little sequence specificity, are difficult to trap within a crystal, and translocation on dsDNA, as opposed to on ssDNA, leads to no detectable product. These challenges are compounded by the finding that translocases often function while coupled to macromolecular machines. They also display varied processivity and coupling to ATP hydrolysis, and while featuring conserved sequence motifs, have mechanisms of action that are modulated by accessory domains in family-specific ways, evading facile generalizations.Transcription-repair coupling factors (TRCFs) are large superfamily 2 ATPases that mediate the preferential repair of the transcribed DNA strand (aka transcription-coupled DNA repair, TCR) in organisms ranging from bacteria to humans3–5. These factors provide a complex but poorly understood example of regulated translocation on dsDNA. At its core, TCR appears universally conserved, and relies on (1) the ability of transcription-repair coupling ATPases to release damage-stalled RNA polymerases (RNAPs) off the nucleic-acid template owing to ATP-driven translocation on dsDNA upstream of the transcription bubble, and (2) on their ability to recruit nucleotide excision repair (NER) machinery6. In bacteria, a single protein, Mfd (aka TRCF) is necessary and sufficient for the coupling process7,8. Although additional pathways have been implicated in TCR9, Mfd remains the only bacterial factor for which both RNAP release and repair enzyme recruitment functions have been demonstrated3,8,10–14. Critically, Mfd associates with RNAP in cells even in the absence of exogenous DNA damage15, it decreases class II transcriptional pausing16, promotes strand-specific repair “at a distance” downstream of a transcriptional pause site17, and dissociates transcription elongation complexes (TECs) stalled not only by DNA damage, but also by protein roadblocks18, including replication forks colliding head-on with the transcription machinery19. Thus, given these different contexts in which it acts on TECs (Fig. 1), Mfd is more appropriately viewed as a general transcription factor. Paradoxically, under certain conditions, Mfd acts as an evolvability factor, promoting hypermutation, the accelerated evolution of lagging-strand genes20,21 and the rapid development of resistance to multiple, unrelated classes of antibiotics22–25. This makes Mfd an attractive target for the development of a broad-spectrum anti-evolution drug, which could be administered in combination with well-characterized antibiotics to curtail the worldwide crisis of antimicrobial resistance.Fig. 1All Mfd functions are dependent on translocation on DNA or action upstream of transcription elongation complexes.Shown at the top is a schematic of an elongating RNAP (gray) with labeled nucleic-acid moieties (RNA, red; NTS, non-template strand, green; TS, template strand, blue) that can either become stalled or temporarily paused as indicated in a–d. a In Mfd-dependent rescue of class II transcriptional pausing, Mfd binds to backtracked RNAPs and promotes their forward translocation thereby rescuing transcript elongation16,26. b In canonical TCR, RNAP becomes stalled at a lesion (orange hexagon) on the TS and recruits Mfd (colored by domain, with D1a and D1b in blue, D2 in cyan, D3 in orange, D4 (RID) in magenta, D5 in yellow, D6 in green and D7 in red), either through 3D-diffusion or 1D-diffusion, enabled by a catch-up-then release mechanism dependent on ATP hydrolysis16, disruption of the D2–D7 clamp, RNA release, and UvrAB recruitment29,31. c In repair “at a distance”, Mfd is recruited to RNAP paused at class II pause signals (red line), which is first released off the nucleic-acid chains and acts as a processivity factor for Mfd to translocate toward downstream lesions in the TS, and initiate strand-specific repair17. d When RNAPs collides with protein roadblocks head-on, including replisomes, Mfd will release RNAP off the nucleic-acid chains, thereby freeing the DNA for other DNA-based processes18,19.Mfd translocation on dsDNA is central to all Mfd functions. This is an ATP-dependent13,26–29 and regulatable process16, which it shares with chromatin remodelers belonging to the same superfamily 2 (SF2) of ATPases. Like chromatin remodelers, Mfd binds and remodels a large macromolecular assembly—the TEC, composed of core RNAP, DNA scaffold, and nascent RNA. TECs activate the Mfd translocase to a level at which processive translocation can readily be detected using classic biochemistry26,30. The structural changes underlying this activation are likely complex and may involve multiple steps27 and interlocked structural elements29,31. Mfd is composed of an ATP-dependent motor core composed of domains D5 and D6 as well as six ancillary domains, including D2 and D7 that pack against each other as an inhibitory “clamp” to restrain the ATPase (Fig. 2a–c) and mask binding determinants for recruitment of UvrA, a component of early NER29,31. Conformational changes in full-length Escherichia coli Mfd occurring during its functional cycle have remained speculative since only a single structure—that of nucleotide-free E. coli Mfd31—has been reported in peer-reviewed literature. Clamp opening was originally proposed to be prerequisite for translocation and dependent on the interaction with RNAP30 (Fig. 1). Recent and more-sensitive single-molecule studies have demonstrated that in fact Mfd can also translocate on naked DNA, albeit with limited processivity16, suggesting that Mfd exists in a dynamic conformational equilibrium that can be shifted by TEC binding. However, provided that ATP is supplied, Mfd can make excursions to the translocation-competent form even in the absence of RNAP16, suggesting that Mfd locates its targets not only by 3D diffusion, but also a more efficient 1-D search along dsDNA, possibly colliding with stalled/paused TECs ahead of it16. This scenario can lead to two outcomes: RNAP rescue by forward translocation (Fig. 1a) or dissociation (Fig. 1b–d), occurring when RNAP encounters severe hindrances to forward movement16. It is important to note that Mfd processivity on naked DNA is substantially lower than that of Mfd bound to a RNAP that has been released off the DNA chains, but remains associated with Mfd via interactions with domain D4, and possibly other unknown binding sites16.Fig. 2DNA binding by Mfd becomes tightest in the presence of transition state analog ADP•AlFx.a Domain organization of E. coli Mfd with annotated conserved sequence motifs and key residues mutated in this study. b Structure-based sequence alignment of the ATPase motifs of Mfd and other dsDNA and ssDNA translocases used for structural modeling purposes. Mutated residues are indicated by asterisks and sequence conservation, based on an alignment of Mfd proteins is color-coded from dark green (strongly conserved) to light green (variable). ATPase domain motifs are shown underneath for Escherichia coli RecG, Sulfolobus solfataricus Swi2/Snf2, Drosophila melanogaster Vasa, HCV NS3, and Anas platyrhynchos RIG-1. The arginine clamp R393 of HCV NS3 is shown in red. c Structural overview of nucleotide-free E. coli Mfd (PDB ID 2EYQ) colored by domain and with annotated ATPase motifs and key structural elements. Shown as CPK models are residues implicated in UvrA and RNAP binding29,31. d Fluorescence anisotropy DNA-binding curves for Mfd in distinct nucleotide states: nucleotide-free (black), with 2 mm ATPγS (red), ADP (blue), or ADP•AlFx (green). Error bars represent S.D.M. (n = 3) and are often smaller than symbols. Curves obtained in the absence/presence of ATPγS and ADP are replotted for convenience from the study by Le et al.16 Source data are provided as a Data Source file.Despite intense research in the last decade, the mechanochemistry of TRCFs remains poorly defined. How do TRCFs recognize DNA and what are the conformational changes associated with translocation on dsDNA? Here, we shed light on these questions by defining the path of DNA across Mfd using electron cryo-microscopy (cryo-EM), structure-guided mutagenesis, and functional assays. We demonstrate that in the absence of RNAP, independent, single substitutions within the two lobes of the Mfd motor core lead to severe DNA-binding defects, and that translocation relies on the existence of a moderate and a strong binding state that correspond to the ATP- and ADP•AlFx-bound states, respectively. We pinpoint conserved residues within translocase motifs Ic, IV, V, and Vb as contributing to DNA loading, and identify an evolutionary conserved threonine in motif Ic and a family-specific histidine near motif IVa as critical for translocation. We also show that DNA binding to Mfd in the presence of transition state analog ADP•AlFx leads to large-scale rotation of the UvrB homology module, resulting in disruption of the D2–D7 clamp as well as repositioning of domain D3 of hitherto unknown function. This leads to the breaking of contacts of D3 with D5 and D6 and the establishment of new contacts with D7. Our study provides the first glimpse of a substrate-bound bacterial TRCF, and suggests that these ancient ATPases utilize a mode of DNA recognition that at its core, is common to ss/ds nucleic-acid translocases.ResultsMfd binds dsDNA tightly with transition state analog ADP•AlFxFor processive autonomous movement on dsDNA (i.e., not dependent on tethering to RNAP) to occur, Mfd must cycle through several states, and must possess at least two DNA contact points, utilized differentially during the ATPase cycle. Mfd binding to DNA with non-hydrolysable ground ATP mimics, ATPγS and AMPPNP has been well documented14,32. ATP hydrolysis leads to DNA dissociation, e.g., Mfd translocation towards the end of the fragment3,26,33.Importantly, although ATP binding is thought to reposition D5/D6 and align the ATPase motifs for catalysis31, there are no reports of a second state tightly associating with DNA. This would be critical because it would provide the necessary powerstroke for unidirectional movement. We have thus asked whether trapping such an alternate state might be possible using ADP and ADP•AlFx. During ATP hydrolysis, the γ-phosphate of ATP passes through a trigonal planar configuration with three equatorial oxygen atoms and two axial ligands, an oxygen from the β-phosphate and the attacking water molecule. To mimic this state, we used Mg2+•ADP•AlFx, the most extensively employed analog of the ATP hydrolysis transition state. This binds as an AlF4− species to most ATPases and adopts an octahedral geometry when complexed to the β-phosphate and the nucleophilic water moiety34. To quantify DNA binding under equilibrium conditions, we used a fluorescence anisotropy assay with a hexachlorofluorescein (HEX)-labeled 40mer dsDNA fragment. In the presence of ATPγS, the equilibrium dissociation constant, Kd, was 151 ± 3 nm (Fig. 2d, Supplementary Table 1), in agreement with previous determinations29. We detected much reduced binding with ADP and enhanced binding with ADP•AlFx (KdADP•AlFx = 24 ± 2 nm). In agreement with several reports in E. coli29, but inconsistent with reports of oligomeric Mycobacterium tuberculosis Mfd35, we observed no oligomerization by gel filtration (Supplementary Fig. 1a), suggesting that the increase in affinity in the presence of the transition state is owing to intramolecular rearrangements rather than oligomerization, as in other SF2 nucleic-acid translocases36.The primary DNA-binding site in Mfd is its motor coreBoth our study and previous work have not detected stable binding of Mfd to dsDNA in the absence of nucleotide14. Modeling of a duplex DNA-Mfd complex based on similarity with other SF2 proteins such as Sulfolobus solfataricus Snf2 is consistent with this finding. In the nucleotide-free form, the motor core of Mfd is more open. dsDNA modeled onto nucleotide-free Mfd contacts a highly conserved and positively charged D5 surface patch centered around invariant K712, but is nevertheless positioned too far away from D6 for direct contacts with this domain (Supplementary Fig. 1b–d), supporting that affinity is dictated by nucleotide status and suggesting that D6 swings down to clamp the DNA (Supplementary Fig. 1b–d). However, the conformation of the ATPase motor module in many translocases/helicases, and by extension, in Mfd, is modulated not only by nucleotide status but also interactions with family-specific accessory domains37, suggesting that changes that are more global rather than restricted to the motor module.Our observation that ADP•AlFx enhances the affinity for DNA allowed us to reconstitute stable Mfd-DNA complexes, and prepare specimens suitable for single-particle cryo-EM for structure determination. For our DNA substrate, we chose a 21mer blunt dsDNA fragment, consistent with previously determined Mfd footprints of ~26bp16,26,29. This allowed us to restrict binding to one Mfd molecule per DNA fragment. Initial reconstructions obtained from untilted specimens were severely affected by the preferred orientation of the particles at the air-water interface, limiting resolution. We thus acquired images at 30° tilts, which allowed us to obtain a more isotropic map resolved to 5.5 Å (Methods, Supplementary Fig. 2, and Supplementary Table 2). This enabled us to readily dock all the individual domains of Mfd31 into the map and perform real-space refinement to obtain the final model. At this resolution, we do not observe density that can be assigned to the transition state analog moiety. The model (Fig. 3) affords several critical observations.Fig. 3Cryo-EM reconstruction of dsDNA-bound Mfd reveals a mobile D3 module and the interaction of dsDNA with the motor lobes.a, b Orthogonal views of nucleotide-free Mfd (PDB ID 2EYQ). Scale bar is equivalent to 10 Å. c, d Orthogonal views of the cryo-EM reconstruction (gray surface) with the fitted Mfd model shown as a cartoon and colored as in a. DNA is shown in gray. e, f Cartoons of Mfd structures in the nucleotide free e and DNA-bound form f after superposition on the Cα trace of D5 and highlighting the relative rotation of Mfd domains in the free and tight DNA-bound state. g, h Chord plots highlighting intramolecular rearrangements in Mfd before g and after h DNA binding. In all panels, Mfd is colored by domain as in Fig. 2a.First, our reconstruction reveals an unexpected large-scale swiveling motion of the UvrB homology module (D1a-D2-D1b), which results in a ~49 Å displacement and an 82° rotation of D2. Importantly, D1a-D2-D1b appears to move in concert, as a rigid body, which is consistent with previous structures of D1aD2D1b and D1aD2D1bD3 truncations, which assumed the same structure in isolation as in the context of the full-length, nucleotide-free protein (r.m.s.d. of 0.8 Å between PDB ID 2EYQ and PDB ID 2B2N and 0.6 Å between PDB ID 2EYQ and PDB ID 3HJH)32,38. Limited proteolysis of Mfd in the absence/presence of nucleotides and Mg•ADP•AlFx did not reveal differences in protease susceptibility (Supplementary Fig. 3), suggesting that the conformational changes we observe are induced by DNA binding and not by the progression from the ATP ground to the transition state. Recruitment of UvrA might thus preferentially occur in the ADP-Pi state and might promote product release and dissociation of Mfd off the DNA, leaving behind an excinuclease loaded in a strand-specific manner.Second, particularly striking is the repositioning of domain of unknown function D3, which swings from the ventral side laterally toward D7 upon DNA binding (Fig. 3 and Supplementary Movie 1), establishing new bridging interactions with the UvrB homology module and D7. This represents a large scale ~47 Å displacement of D3 enabled by its flanking flexible linkers connecting D3 to D1b and D4 (Figs. 2a–b, 3). The distance between the termini of the fitted D3 and D4 (36 Å) is fully compatible with the length of this partially disordered D3–D4 linker. Consistent with our data, the sequence corresponding to D3 in the structure of D1aD2D1bD3 was completely disordered32, pointing to it forming a dynamic and separate structural unit. Recent structures of mycobacterial Mfd in the presence of ADP and ADP•Pi also revealed movement of D339, but this is distinct from what we observe. In mycobacterial Mfd, sharing ~33% identity with E. coli Mfd, D3 is repositioned when ADP and Pi are bound such that it contacts D1a and D5 (Fig. 4a–c). These contacts are presumably important for nucleotide binding, as they are accompanied by a flip-flop action of the D3–D4 linker (Fig. 4 and Supplemental Fig. 5)39. Movement of this linker relieves steric occlusion of the conserved phenylalanine of motif Q (F641 mycobacterial Mfd, F599 in Eco Mfd), which becomes free to stack against the adenine base of ADP. In contrast, when Mfd is bound to DNA, D3 makes contacts with D1a and D7 (Fig. 3), swapping position with D2. The UvrA-binding surface on D2, including residues critical for binding such as R165 and R18129 (Fig. 2d–f) becomes fully exposed. Contacts of D2 with conserved residues in D7, D0148, and F1050 are lost.Fig. 4DNA binding unmasks the UvrA recruitment surface within domain D2 of Mfd.a–c Top view of nucleotide-free E. coli Mfd a, E. coli Mfd complexed to DNA and ADP-AlFx
b and M. smegmatis Mfd bound to ADP and Pi c (PDB ID 6AXC). Insets show the D3–D4 linker, and C-terminal residues in D3 as positional markers. The D3–D4 linker is not visible in our map of DNA-bound Mfd b or in M. smegmatis Mfd c. d–f Overall views of Mfd with insets highlighting the disruption of the D2–D7 interaction upon DNA binding. Amino acids involved in domain–domain interactions (R165, R181, D1048, F1050) are shown as CPK with stick models. Middle inset highlights the unmasking of R165 and R181 (spheres), important for UvrA recruitment. In all panels, Mfd is colored by domain as in Fig. 2a.Last, but not least, our reconstruction confirms the long-hypothesized role of the TRG motif in DNA translocation. The TRG (translocation in RecG) motif is common to both the RecG and Mfd subfamilies, and consists of a helical hairpin engaging a helical “hook” C-terminal to it (Fig. 2c). This hook was partially disordered in substrate-bound RecG, but helical in Mfd and wrapped around yet another helix—the relay helix, connecting D4 to the motor core (Fig. 2c). In our reconstruction, the two helices of the TRG snap together and are likely stabilized by ADP•AlFx. binding. In contrast, in nucleotide-free Mfd, the TRG helices are splayed open by motif VI helix ATP sensor31. The closed TRG motif engages part of the hook structure but releases motif VI, which undergoes a 46 Å displacement upon rotation of D6, and indirectly modulates the conformation of the relay helix, which becomes slightly bent and rotates about a hinge located in the loop connecting it to D5 (Fig. 5 and Supplementary Movie 1). Thus, we conclude that domain D3 of hitherto unknown function plays a critical role in unmasking of the UvrA binding site and that the interlocking of motif VI, TRG, hook and relay helices enable the complex and long-range conformational changes associated with DNA binding.Fig. 5Structural interlocking of the TRG motif, motif VI, and relay helix.a View of motif VI (red), TRG motif (dark green), hook helices (green), and relay helix (yellow) in the context of full-length nucleotide-free E. coli Mfd (left) and, for emphasis, as close-ups. b Similar views of DNA-bound E. coli Mfd highlighting conformational changes in this region The TRG helical hairpin closes and a bend is introduced into the relay helix.Linchpin residues for DNA loading span both lobes of the motor coreFor translocation to occur, Mfd must first load the DNA. Our reconstruction suggests that this involves D5 and D6 exclusively, and no other contacts (Fig. 3a–f), although secondary binding determinants might be unmasked under very specific conditions.We have thus asked if isolated Mfd domains might contribute to ss or dsDNA binding. We have purified Mfd functional modules as defined by the crystal structure of E. coli Mfd (UvrB homology module MfdD1aD2D1b, domain of unknown function MfdD3, RNAP interaction domain MfdD4, and clamp domain MfdD7 (Fig. 2a and Supplementary Fig. 4) with the exception of MfdD5D6. We then assessed DNA binding with either a 40mer HEX-labeled dsDNA or a shorter Cy3-labeled double- and single-stranded 27mer. As expected, we observed no stable binding to any of the purified modules (Supplementary Fig. 4). However, we cannot rule out lower affinity, secondary binding sites that might require initial contacts with MfdD5D6, a DNA fragment longer than what was used for our EM reconstruction, or a particular DNA structure, akin to an open or partially open bubble. On longer cellular DNA substrates, multiple domains may be required for DNA binding and contacts with individual modules, as defined here, may be too weak to be detected.Although the EM reconstruction allowed us to identify residues proximal to DNA, specific contacts with the nucleic acid are not resolved with confidence at this resolution. To identify residues critical for DNA loading in both the ATP- and ADP•AlFx states, we used our EM reconstruction but also took advantage of evolutionary relationships and structural superpositions with well-characterized nucleic acid bound ATPases. Structural alignments with RecG, the helicase most closely related to Mfd, proved unfruitful, as crystallized RecG only contains a short fork DNA that does not reach across the helicase domains40. Instead, we extended our analyses to other ss/ds nucleic-acid translocases of known structure. We focused on the similarity with Snf2, an SF2 chromatin remodeler for which several structural models in different functional states are available. These structures include models of Snf2 bound to dsDNA41, akin to the DNA moiety found in our specimen, but also curved DNA, as found in the structure of Snf2 bound to nucleosomes42,43. We combined the information derived from our EM reconstruction and these superpositions (Supplementary Figure 6), and selected a set of well-conserved residues, highlighted in Fig. 6a and the Supplementary Movie 1, as likely key for DNA recognition. We constructed Mfd variants carrying substitutions in either D5 (MfdR685A, MfdK690A, MfdT710A, MfdH711, MfdK712A) or D6 (MfdY816A, MfdN817A, MfdH842A, MfdT868A, MfdR887E) (Fig. 2a). All were active in ATPase assays at levels comparable to wild type (Fig. 6a), indicating they are competent for nucleotide binding and hydrolysis, and consistent with proper folding, which was assessed using circular dichroism (Supplementary Fig. 7b). We then assessed binding to the same 40mer HEX-labeled dsDNA used above. The variants displayed various degrees of DNA-binding defects in the presence of both ATPγS (Fig. 6b, Supplementary Fig. 7c and Supplementary Table 1) and ADP•AlFx (Fig. 6c, Supplementary Fig. 7d and Supplementary Table 1). Interactions with DNA were more sensitive to salt in the ATPγS than the ADP•AlFx state (Supplementary Fig. 7e, f), pointing to the formation of non-electrostatic contacts as partially accounting for the gain in affinity in the transition state.Fig. 6Key roles for R685, N817, and R868 in DNA loading.a Steady-state ATPase activities of single Mfd variants. ATP hydrolysis rates were measured using an ATP/NADH-coupled ATPase assay. Data shown are the means ± SD (n = 3). Variants are color-coded by domain as in Fig. 2a (D5 variants, yellow hues; D6 variants, green hues). Data for MfdR685A and MfdN817A are replotted for convenience from Le et al.16. b–c Equilibrium dissociation constants for the Mfd-dsDNA interaction derived from fluorescence anisotropy binding curves. D5 and D6 variants are color-coded by domain as in Fig. 2a. Data represent mean values ± SD (n = 3). Error bars are often smaller than symbols. Arrows indicate data sets that could not be reliably fit to a binding model owing to compromised affinity for DNA, and likely represent an overestimation of the true affinity. Binding curves are shown in Supplementary Fig. 7. d–j Structural comparison between double-stranded nucleic-acid translocases and Escherichia coli (Eco) Mfd (d) using superpositions restricted to the Cα trace of D5. Shown are Sulfolobus solfataricus (Sso) Snf2 (r.m.s.d. of 1.1 Å, e), Saccharomyces cerevisiae (Sce) Snf2 in its nucleotide-free (r.m.s.d of 1.4 Å, f) and ADP•BeFx -bound state (r.m.s.d. of 1.4, g), RIG-1 (r.m.s.d. of 1.0 Å, h) and ssRNA translocases, Drosophila melanogaster (Dme) Vasa (r.m.s.d. of 0.9 Å, i) and HCV NS3 (r.m.s.d. of 1.1 Å, j). Bound Mg2+ is shown as a dark green sphere, analog ADP•AlF3− as an aquamarine CPK model, bound nucleotide as magenta sticks, ADP as a red CPK model, BeFx as a dark red CPK model and nucleic-acid strands are colored red and orange. The two lobes of the ATPase core are in yellow (N-terminal lobe) and green (C-terminal lobe), except for DNA-binding residues, which are highlighted as slate CPK models. Other domains are shown as a gray ribbon with annotated ATPase motifs. Those motifs contacting DNA are color-coded in slate font and some are not visible in these views. Indicated in parenthesis is the degree of conservation between the various motor cores and D5 and D6 of E. coli Mfd. Source data are provided as a Source Data file.MfdR685A, MfdN817A, and MfdT868A were severely impaired with both ATPγS and ADP•AlFx, indicating a critical role in loading DNA. T868 is particularly intriguing, as it is located in conserved helicase motif V (Fig. 2a–b), which in several cases has been shown to contribute to either ATP or oligonucleotide binding44. In the case of Mfd, substitution of T868 affects DNA binding severely, and less so ATP binding or hydrolysis (Fig. 6a–c). This is not without precedent. The structure of duck RIG-I RNA helicase bound to 19mer dsRNA and ADP•AlFx, shows this highly conserved threonine (T698 in duck RIG-I) in direct contact with a phosphate of the RNA backbone (Fig. 6h), making a hydrogen bond with the phosphoryl oxygen atom45. However, mutation of this conserved residue to alanine in human RIG-I eliminates dsRNA binding while retaining ATP hydrolysis activity46, suggesting that this residue confers affinity that is uncoupled from the splitting of the phosphoanhydride bond. In Thermotoga maritima RecG, the mutation of T478 reduces fork reversal (i.e., translocation) without affecting ATP hydrolysis47. A structurally equivalent threonine, T546, also contacts the RNA phosphate backbone in the structure of Vasa RNA helicase bound to ssRNA and AMPPNP (PDB ID 2DB3, Fig. 6i)48. At last, the SF2 helicase domain of virally encoded DExH helicase NS3 also has a similar conserved threonine, and this was seen to contact oligonucleotide in multiple structural studies49,50. As with Mfd, mutation of this threonine (T411 in NS3h, Fig. 6j) does not affect its basal ATPase activity, but abolishes oligonucleotide binding and duplex unwinding activity51.Mfd mechanism of translocation is reminiscent of that used by RNA helicasesIn the presence of ADP•AlFx, some of our variants (MfdK690A, MfdT710A, MfdK712A, MfdY816A, MfdH842A) showed only a slight impairment in function, or no DNA-binding defect at all, as was the case for MfdH711A (Fig. 6 and Supplementary Table 1). Most striking are the substitutions of T710A and H842A (Fig. 6a), which lead to severe loss of affinity in the ATPγS state, but had a comparatively minor effect on DNA affinity in the presence of ADP•AlFx. Using optical trapping, we have previously shown that in the context of a nucleotide-starved TEC, Mfd binds to the TEC such that D5 is located in front (adjacent to the TEC), followed by D6, and that upon ATPγS binding D6 moves 12 bp toward a stationary D5, approaching the RNAP footprint16. As ATP hydrolysis is required for processive translocation, this suggested to us that upon hydrolysis and/or product release, D5 may invade the RNAP footprint. We thus reasoned that the residues with nucleotide-dependent roles might be critical for cycling through all the states required for translocation. To probe translocation, we developed a fluorescent triplex displacement assay, based on the ability of dsDNA translocases to displace a fluorescently labeled triplex forming oligonucleotide (TFO) off a DNA duplex. Triplex displacement assays have been well established for constitutively hyperactive Mfd variants, such as MfdD7−AAA, carrying the E1045A, D1048A, and R1049A substitutions in D7 to disrupt the inhibitory clamp interaction (Fig. 7a)52. Given the recently determined short processivity of Mfd of ~200 bp16, we employed a 70 bp substrate, shorter than the Mfd processivity length, in conjunction with the hyperactive MfdD7−AAA variant. We included in our reaction non-labeled ssDNA, for which Mfd has a weaker affinity14, but acts as an efficient trap to inhibit a contaminating endonuclease activity that initially severely reduced the signal corresponding to full-length TFO (not shown). In the presence of MfdD7−AAA, free TFO accumulated over time, and this activity was inhibited by non-hydrolysable analog, AMPPNP and the absence of nucleotide (Fig. 7b–c and Supplementary Fig. 8). Both mutants, MfdD7−AAA;T710A and particularly MfdD7−AAA;H842A displayed defects in translocation and loading, but not ATP hydrolysis (Fig. 7f), and featured circular dichroism spectra similar to MfdD7−AAA (Supplementary Fig. 8b). The loading defects of these complex variants are somewhat different than the defects on the single T710A and H842A variants in the wild-type background (Figs. 6a–c, 7), and suggest that artificial disruption of the clamp interaction partially uncouples DNA binding from nucleotide status, consistent with a now derepressed enzyme30. As such, these variants display severe translocation defects, with MfdH842A and MfdT710 displaying no detectable translocation activity (Fig. 7b, c). At higher protein concentrations used in an attempt to overcome the loss in affinity of the D7-AAA mutants, both in the presence of ATPγS and ADP•AlFx, the apparent translocation activity of both MfdD7−AAA and its two variants decreases (Supplementary Fig. 8), likely owing to multiple loading molecules interfering with translocation, which complicates the analysis. Interestingly, although wild-type Mfd is not stimulated by the addition of herring sperm DNA, MfdD7−AAA and to a lesser extent MfdD7−AAA;T710A, and MfdD7−AAA;H842 show an increase in ATPase activity (Fig. 7f and Supplementary Fig. 8f). Thus, whereas deregulation via clamp disruption hyperactivates Mfd, mutations within motifs Ic and IVa may repress Mfd activity.Fig. 7T710 and H842 are critical regulators of translocation by Mfd.a EM-based model of DNA-bound Mfd with boxed region highlighting the structural environment of T710 and H842 residues. The resolution of our reconstruction precludes precise determination of the conformation of motif IVa loop and of amino-acid rotameric forms. Crucial residues are indicated as Cα spheres, H842 in green, T710A in yellow and E1045, D1048, and R1049, mutated in Mfd D7AAA in red. b–c TFO displacement assays carried out with MfdD7−AAA
b and Mfd c variants. Percentage of remaining TFO-Triplex is plotted vs time to show translocation of Mfd variants. Plot shows means ± SD (n = 3), with error bars often smaller than symbols. d–e Electrophoretic separation of TFO displacement assays carried out with MfdD7−AAA. Representative gels for MfdD7−AAA
d and Mfd e +/-ATP and +AMPPNP are shown. Example gels for Mfd, MfdD7−AAA;T710A and MfdD7AAA;H842A variants are shown in Supplementary Fig. 8 (n = 3). f Steady-state ATPase activities of Mfd variants normalized to wild-type Mfd activity. Data represent means ± SD (n ≥ 3). **denotes p < 0.01, ****p < 0.0001 and ns, p > 0.05 (unpaired two-tailed t test). The p value for MfdD7AAA;T710 with/without DNA was 0.0012. g, h Equilibrium dissociation constants for the Mfd-dsDNA interaction derived from fluorescence anisotropy binding curves obtained by titrating increasing amounts of protein into a mixture of 10 nm HEX-labeled 40mer DNA and 2.0 mm ATPγS e and ADP•AlFx. Data represent mean values ± SD (n = 3). KD values are tabulated in Supplementary Table 1. Source data are provided as a Source Data file.H842 maps to a loop N-terminal to motif IVa and is conserved within the Mfd family, but not across translocases (Fig. 2a). However, based on a structural alignment, it is located close to where R393 in HCV NS3 maps (Supplementary Fig. 9). This has been coined the arginine clamp and is key for nucleic-acid binding and translocation53. In agreement with our findings, a set of structures of HCV translocase NS3h bound to DNA alone and in complex with ground-state ATP and transition state mimics revealed contacts of nucleic acid with R393 as well as a series of nucleotide-dependent structural transitions involving the equivalent of T710 (T269 in NS3h, Fig. 6j) as well as NS3h T411, the equivalent of T86850. T269 is in helicase motif Ic (aka TxGx motif), which while not originally identified as a conserved sequence, has been shown to be involved in nucleotide-dependent oligonucleotide binding in a variety of helicases, including NS344,54. The NS3 structure obtained in the absence of ATP shows these threonine residues three nucleotides apart, however they are only two nucleotides apart in the structure obtained with the ATP transition state analog (Supplementary Fig. 9b–d)49,50. T411 maintains contact with the same nucleotide throughout the ATP hydrolysis cycle, whereas T269 releases the DNA in the ATP-bound state and makes a new contact with the DNA one nucleotide away when NS3h is in complex with the ATP transition-state analog. Conservation of these two threonine residues in Mfd, the nucleotide-dependent role of T710 and the confirmation that T868 is critical for DNA binding by Mfd regardless of the status of the phosphoanhydride bond point to common features of translocation in NS3h and Mfd, despite Mfd lacking helicase activity and binding preferentially to DNA rather than RNA.DiscussionMfd-dependent processes rely heavily on its translocation on dsDNA. This is important not only for early events required to locate RNAP targets on the chromosome, but also for destabilizing TECs halted by DNA damage, pause signals, or various roadblocks6. Mfd action on transcription complexes effectively builds up torque to reanneal the upstream edge of the transcription bubble, shorten the bubble, and release the RNA transcript26,55. As a first step in shedding light on these events, here we describe the first structure of DNA-bound E. coli Mfd as well as complementary functional analyses. Our data allow for the first model for how nucleotide status and DNA binding control the breaking and formation of interdomain contacts and also reveal the molecular basis for DNA recognition by Mfd proteins.We observe that DNA is bound by the two lobes of the motor core, consistent with a previous truncation analysis, which determined that in the presence of ATPγS, the DNA interaction determinants are restricted to D5 and most of D614. In this study, domain boundary definitions were not based on structural information, and notably truncated ATPase motif VI and the TRG motif (Fig. 2a–c), with crucial roles in coupling nucleotide status to translocation33. However, even with our structure-based truncations, we detect no binding of ds/ssDNA to domains flanking the motor core. Previous studies noted a striking similarity between D1aD2D1b and a region in UvrB that contacts DNA during NER, and also locally melts it for damage recognition using a highly conserved beta hairpin at the interface between D1a and D1b38. This beta hairpin extension is absent in Mfd, is replaced by a short loop, which lacks the aromatics involved in nucleotide flipping in UvrB. This loop is located ~50 Å away from the D5/D6 binding site (Fig. 2c), further supporting that the similarity might be strictly architectural. Our reconstruction suggests that on DNA substrates longer than used here, the DNA fragment would likely approach the C-terminal region of the relay helix, where previous studies have identified one substitution (MfdW550A) that enhances DNA binding and motor activity56. This could involve bending of the DNA across the protein surface, and/or modest rearrangements of the hook and TRG motifs, located in close proximity to the DNA.We thus conceptualize DNA binding by Mfd in terms of two clusters of residues located at the tip of the D5 and D6 pincers that establish contacts with the substrate, primarily the tracking strand, and cycle between a high-affinity state (nucleotide-bound) and a low affinity (nucleotide-free) state. This shares features with findings from the HCV NS3 system, but bears a distinction in that NS3 cycles between a low affinity state when nucleotide-bound to a high-affinity state when nucleotide-free. Cluster I is comprised of R685 and T710, whereas cluster II is comprised of R865, N817, T868, and H842. Within these clusters, we distinguish between residues key for loading (R685, T868) and for translocation (T710 and H842), the latter having markedly nucleotide-dependent roles as also observed in the HCV NS3 system. It will be interesting to determine if Mfd operates via a Brownian ratchet that involves translocation in 1 bp steps (like in NS3), and whether and how the binding of RNAP might rectify this ratchet to achieve high processivity on the order of hundreds of base pairs, detected using single-molecule techniques16,28. We demonstrate that although there is no biochemical evidence for the TRG helices being directly involved in DNA binding, this structural element in direct communication with ATPase motif VI and the hook is mobile and likely used as a pawl during translocation. Here, we present the first structural evidence for the hypothesized interlocking of nucleotide-responsive elements (the TRG, the hook and relay helices, seen in Fig. 2c, 5) that is driven by movement of ATPase motif VI.Translocases that move along dsDNA often make more extensive interactions with one DNA strand, termed the tracking strand. In the structure of S. solfataricus Snf2, more extensive interactions exist between the tracking strand and motifs I, Ia, Ib, Ic, and IIa41, located in the N-terminal lobe (Fig. 2a, c), whereas recent studies of Snf2 bound to a nucleosome suggest tighter interactions with the motifs in the C-terminal lobe42,43. It is impossible for us to determine at this stage which of the strand represents the “tracking” strand. This remains an important outstanding question since it underlies strand-specific probing of damage, for whose recognition UvrA, UvrB and Mfd are required52.Our identification of an Mfd family-specific residue important for translocation (H842, located in motif IVa) provides an attractive avenue to pursue in the development of an Mfd inhibitor of high selectivity. Mfd deletion limits the acquisition of antimicrobial resistance in multiple bacteria, including E. coli25, Salmonella tyiphimurium25, Bacillus subtilis25, Helicobacter pylori23, Campylobacter jejuni22, and difficult to treat pathogens such as M. tuberculosis25. Furthermore, in Bacillus cereus and Shigella flexneri, recent experiments have implicated Mfd in resistance to the nitric oxide response generated by the immune response during infection24. Thus, an Mfd inhibitor would provide a novel strategy for antibiotic development. To block function while achieving selectivity, this could target highly variable motif IVa, rather than one of the other, more conserved ATPase motifs. In the NS3 system, the arginine clamp in motif IVa has been shown to be druggable57. We note that at the current resolution of our EM reconstruction, establishing whether H842 directly contacts DNA or may exert its effect indirectly by modulating the conformation of the motif IVa loop remains challenging.Finally, our work provides the first evidence that domain D3 of unknown function might be key for TCR regulation in E. coli and related species. While D3 is sandwiched between D1a and D5/D6 in nucleotide-free Mfd, it is repositioned in the presence of DNA and transition state analog to make bridging interactions with D7 (Supplementary Movie 1). Motion of D3 is coupled to a peppermill-like motion of the UvrB homology module relative to the C-terminal region of Mfd (D4D5D6D7), which in turn undergoes local rearrangements to close the D5-D7 interdomain cleft, whereas D4–D7 interactions remain largely unperturbed and the UvrA recruitment surface is unmasked. Our work establishes that these conformational changes are coupled to the active site, primarily through motif VI, but also movements of the TRG, hook and relay helices, which remain unresolved at the residue level and await higher resolution structural and mechanistic analyses.MethodsCloning, protein expression, and purificationConstructs pMS1-pMS8 were obtained via site-directed mutagenesis of pAD6 (encoding N-terminally His-tagged E. coli Mfd58) using the method of Edelheit et al.59. All constructs were verified by DNA sequencing, and in select cases, purified protein was verified using mass spectrometry. Overexpression and purification was performed according to published protocols by a succession of Ni2+-affinity chromatography, His-tag cleavage with Prescission protease, heparin chromatography and gel filtration59. For constructs pMS2 and pMS5, HiTrap Q-sepharose column (GE Healthcare) was used instead of heparin affinity as these variants failed to bind to heparin affinity matrix. Size-exclusion chromatography was carried out on a Superdex 200 10/30 Increase column (GE Healthcare) using a buffer consisting of 20 mm Tris pH 8.0, 0.10 m NaCl, 5 mm DTT. Mfd truncations were expressed as N-terminal fusions to polyhistidine tags (Supplementary Table 3), which were cleaved using Prescission protease after the initial immobilized metal affinity chromatography step.Triplex displacement assaysTo generate the TFO-Triplex, linear ds 70mer DNA (50 nm) was mixed with 5’ Alexa647-labeled 21mer TFO oligo (25 nm) in an opaque tube in triplex reconstitution buffer (10 mm MES pH 5.5, 12.5 mm MgCl2) and heated to 57 °C for 15 min. The reaction was then placed at 4 °C and allowed to thermally equilibrate overnight. Protein samples were dialyzed against 10 mm Tris-HCl pH 8, 100 mm NaCl, 10 mm MgCl2, 2 mm β-mercaptoethanol. Reactions were assembled by mixing: 10 nm TFO-triplex, 50 nm 44 bp ssDNA trap in reaction buffer (50 mm Tris-HCl pH 8, 10 mm MgCl2, 1 mm DTT) and protein (450 nm for constructs used in Fig. 7b–c; 4.5 μμ for the experiments of Supplementary Fig. 8d). Triplex displacement was initiated by addition of 2 mm ATP. Samples were withdrawn at indicated timepoints, stopped by addition of GSM buffer (15% (w/v) glucose, 3% (w/v) SDS, 250 mm MOPS pH 5.5), and loaded on 3–8% Tris-Acetate gels (Invitrogen, Carlsbad, CA), which were pre-run for 60 min at 75 V using Tris-Acetate Running buffer (40 mm Tris-Acetate pH 5.5, 5 mm sodium acetate, 1 mm MgCl2) at 4 °C. Samples were electrophoresed for 180 min at 75 V and imaged using a ChemiDoc system (Biorad, Hercules, CA). The disappearance of TFO-Triplex over time was quantified using BioRad ImageLab and plotted using the Graphpad Prism software package.Specimen preparation for cryo-EMSize-exclusion chromatography (Superdex 200 10/30, GE Healthcare) was performed prior to complex reconstitution in a buffer containing 20 mm Tris-HCl pH 8, 100 mm NaCl, 20 mm MgCl2, 2 mm TCEP. For complex formation, Mfd was incubated with 21mer dsDNA in a 1:1.5 molar excess in the presence of 2 mm ADP, 10 mm NaF, and 1 mm AlCl3. 2.5 µl Mfd-DNA complex at 1.0 mg/mL was applied to UltrAuFoil R1.2/1.3 300 mesh grids (Quantifoil) that were previously plasma-cleaned using a Gatan Solarus (75% argon/25% oxygen atmosphere, 15 W for 7 s), then manually blotted with a Whatman No. 1 filter paper in a cold room with >80% humidity, and plunged into liquid ethane using a manual plunger. The dsDNA used for reconstitution had the following sequence: 5′-ATAGGATACTTACAGCCATCG-3′.Cryo-EM data collectionAutomated EM image acquisition was performed with Leginon60. Data collection was carried out on a Talos Arctica microscope (FEI) operating at 200 kV and equipped with a K2 Summit direct electron detector (Gatan) at a nominal magnification of ×36,000 and a defocus range from 2.0 μm to 4.0 μm giving a pixel size of 1.15 Å at the specimen level. To ameliorate preferential specimen orientation, images were collected using 30° tilts, as previously described61. Other than setting the nominal tilt angle during data acquisition, standard procedures were employed. Details are in Supplementary Table 2.Cryo-EM image processingMovies were aligned and dose-weighted using MotionCor262. CTF estimation and particle picking was performed in Warp63, which resulted in a stack of 69,721 particles that were imported into Relion (version 3.06)64 for 2D classification. One round of 2D classification was performed, and bad particles were excluded. Subsequently, a round of 3D classification, using five classes was run in Relion (Supplementary Fig. 2). The initial model used for 3D classification was generated from a small subset of the data, during data collection using cisTEM65. The best class was selected from 3D classification for moving forward. The selected particle stack from Relion contained 19,083 particles and was then input into cryoSPARC for iterative 2D and 3D classification. This cleaned stack containing 9822 particles was then imported into cisTEM and refined using the Auto-Refine procedure, whereby only frequencies up to 80% of Nyquist were included. This resulted in a final resolution of 5.47 Å, assessed by the conventional Fourier Shell Correlation (FSC) criteria. The refined map was sharpened in cisTEM by flattening the amplitude spectrum between 8 Å and the nominal resolution of 5.5 Å. Directional resolution volumes were generated using the 3D FSC tool61, whereas the local resolution was calculated using sxlocres.py, which is implemented within the Sparx processing package66. Details on model building and refinement are in Supplementary Methods.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Description of Additional Supplementary Files
Supplementary Movie 1
Reporting Summary | nature communications | [
"Article"
] | [
"DNA repair enzymes",
"Cryoelectron microscopy"
] | DNA for molecular machines replication transcription repair movement of nucleic-acid translocases along or double DNA regulated spatiotemporally1 control in development neurodegeneration aging diseased state2 insufficiently understood Translocation on dsDNA challenging motors little sequence specificity difficult to trap within crystal to no detectable product challenges translocases function coupled to macromolecular machines display varied processivity coupling to ATP hydrolysis mechanisms action modulated by accessory domains-specific.Transcription-repair coupling factors (TRCFs) large superfamily 2 ATPases mediate repair of transcribed DNA strand in organisms bacteria to complex poorly understood regulated translocation on dsDNA TCR conserved relies on transcription-repair ATPases to release damage-stalled RNA polymerases (RNAPs) off nucleic-acid template recruit nucleotide excision repair (NER) machinery6 In bacteria single protein, Mfd TRCF) necessary for coupling process7,8 additional pathways in Mfd only bacterial factor for RNAP release repair enzyme recruitment functions demonstrated3Mfd associates with RNAP exogenous DNA decreases class II transcriptional pausing16 promotes strand-specific repair transcriptional pause dissociates transcription elongation complexes stalled by DNA damage protein roadblocks18 replication with transcription Mfd as general transcription factor under conditions acts as evolvability factor hypermutation accelerated evolution of lagging-strand genes20 rapid development resistance to Mfd attractive target for development broad-spectrum anti-evolution drug administered with antibiotics curtail crisis antimicrobial resistance Mfd functions dependent on translocation on DNA action upstream of transcription elongation complexes schematic of elongating RNAP) with labeled nucleic-acid moieties become stalled or paused Mfd-dependent rescue of class II transcriptional pausing binds to backtracked RNAPs promotes forward translocation transcript elongation16canonical TCR RNAP stalled at lesion (orange hexagon TS recruits Mfd (colored domain D1a D1b blue D2 cyan D3 orange D4 magenta D5 yellow D6 green D7 3D or 1D-diffusion catch-up release mechanism ATP hydrolysis16 disruption D2–D7 clamp release UvrAB recruitment29 repair Mfd recruited to RNAP paused at class II pause signals (red released off nucleic-acid chains processivity factor for Mfd downstream lesions strand-specific with protein roadblocks Mfd RNAP freeing DNA for processes18.Mfd translocation on dsDNA central to Mfd functions ATP-dependent13 regulatable shares with chromatin remodelers superfamily Mfd binds remodels macromolecular TEC core RNAP DNA scaffold nascent RNA TECs activate Mfd processive translocation detected biochemistry26 structural changes complex multiple interlocked structural elements29Mfd ATP-dependent motor core domains D5 D6 six ancillary domains D2 D7 inhibitory ATPase (Fig. 2a–c mask binding determinants for recruitment UvrA early NER29 Conformational changes in Escherichia coli Mfd functional cycle speculative single nucleotide-free E. coli reported Clamp opening prerequisite for translocation dependent on interaction with RNAP30 (Fig. 1) studies Mfd can translocate on naked DNA limited processivity16 dynamic conformational equilibrium shifted by TEC binding ATP supplied Mfd to translocation-competent form absence RNAP16 Mfd locates targets by 3D diffusion 1-D search dsDNA colliding with stalled/paused TECs outcomes RNAP rescue by forward translocation or dissociation when RNAP hindrances Mfd processivity on naked DNA lower than bound to RNAP associated with Mfd via interactions domain D4 unknown binding.Fig. 2DNA binding by Mfd tightest transition state analog ADP•AlFx.a Domain organization E.Mfd sequence motifs residues mutated Structure sequence alignment ATPase motifs Mfd dsDNA ssDNA translocases structural modeling Mutated residues asterisks sequence conservation color-coded dark green to light green ATPase motifs Escherichia coli RecG Sulfolobus solfataricus Swi2/Snf2 Drosophila melanogaster Vasa HCV NS3 Anas platyrhynchos RIG-1 arginine clamp R393 HCV NS3 red Structural overview nucleotide-free E. coli Mfd ID 2EYQ colored domain ATPase motifs structural elements models residues UvrA RNAP Fluorescence anisotropy DNA-binding curves Mfd states nucleotide-free 2 mm ATPγS ADP ADP•AlFx Error bars S.D.M. smaller symbols Curves absence ATPγS ADP replotted Le et al Source data file mechanochemistry TRCFs poorly defined DNA conformational changes translocation dsDNA path DNA Mfd cryo-microscopy structure-guided mutagenesis functional assaysRNAP substitutions Mfd motor core lead to DNA-binding defects translocation relies moderate strong binding state ATP ADP•AlFx-bound states conserved residues translocase motifs Ic IV V Vb DNA loading evolutionary conserved threonine in motif Ic family-specific histidine near motif IVa critical for translocation DNA binding to Mfd transition state analog ADP•AlFx leads to rotation UvrB homology module disruption D2–D7 clamp repositioning domain D3 breaking contacts D3 with D5 D6 establishment new contacts D7 study provides glimpse substrate-bound bacterial TRCF suggests ancient ATPases utilize DNA recognition common nucleic-acid translocases binds dsDNA with transition state analog ADP•AlFxFor autonomous movement Mfd cycle through states two DNA contact points cycle Mfd binding to DNA with non-hydrolysable ATP ATP hydrolysis leads to DNA dissociation Mfd translocation ATP binding D5/D6 ATPase motifs for catalysis31 no reports second state associating with DNAcritical powerstroke for unidirectional movement asked trapping alternate state using ADP ADP•AlFx ATP hydrolysis γ-phosphate passes trigonal planar configuration three equatorial oxygen atoms two axial ligands oxygen from β-phosphate water molecule used Mg2+•ADP•AlFx analog ATP hydrolysis transition state binds as AlF4− species to ATPases adopts octahedral geometry complexed to β-phosphate nucleophilic water DNA binding equilibrium used fluorescence anisotropy assay-labeled 40mer dsDNA fragment ATPγS equilibrium dissociation constant Kd 151 ± 3 nm (Fig. 2d agreement reduced binding with ADP enhanced binding with ADP•AlFx = 24 ± 2 E. inconsistent Mycobacterium tuberculosis no oligomerization by gel filtration increase affinity transition state intramolecular rearrangements oligomerization primary DNA-binding site in Mfd motor detected stable binding Mfd to dsDNA absence nucleotide14Modeling duplex DNA-Mfd complex similarity SF2 proteins Sulfolobus solfataricus consistent finding nucleotide-free form motor core Mfd open dsDNA nucleotide-free Mfd contacts conserved charged D5 surface patch invariant K712 positioned far from D6 for direct contacts affinity dictated by nucleotide status D6 clamp DNA conformation ATPase motor module in translocases/helicases Mfd modulated by nucleotide status interactions family-specific accessory domains37 changes global ADP•AlFx enhances affinity for DNA reconstitute stable Mfd-DNA complexes specimens for single-particle cryo-EM structure determination DNA substrate chose 21mer blunt dsDNA fragment consistent with Mfd footprints ~26bp16,26,29 binding to one Mfd molecule per DNA fragment Initial reconstructions affected particles air-water interface limiting resolution acquired images at 30° tilts isotropic map resolved to 5.5 Å dock domains Mfd31 into map perform real-space refinement final model resolution density transition state analog moiety model (Fig. 3) affords critical observationsFig. 3Cryo-EM reconstruction dsDNA-bound Mfd mobile D3 module interaction dsDNA motor lobes views nucleotide-free Mfd (PDB ID Scale bar 10 Å cryo-EM reconstruction fitted Mfd model cartoon colored DNA gray Mfd structures nucleotide free DNA-bound form superposition Cα trace D5 rotation domains free DNA-bound state h Chord plots intramolecular rearrangements Mfd before DNA binding Mfd colored domain Fig. 2a reconstruction reveals large-scale swiveling motion UvrB homology module (D1a ~49 Å displacement 82° rotation D2. D1a-D2-D1b rigid body consistent previous structures D1aD2D1b D1aD2D1bD3 truncations structure nucleotide-free protein 0.8 Å PDB ID 2EYQ PDB 2B2N 0.6 Å 2EYQ 3HJH Limited proteolysis Mfd absence nucleotides Mg•ADP•AlFx differences protease susceptibility conformational changes induced DNA binding not progression ATP ground transition stateRecruitment UvrA ADP-Pi state promote product release dissociation Mfd DNA excinuclease loaded strand-specific repositioning function D3 swings toward D7 DNA binding (Fig. 3 1) bridging interactions with UvrB module D7 large scale ~47 Å displacement D3 linkers connecting D3 to D1b D4 (Figs. 2a–b 3) distance between D3 D4 (36 Å) compatible with length disordered D3–D4 linker sequence D3 structure D1aD2D1bD3 disordered32 separate structural unit structures mycobacterial Mfd ADP ADP•Pi revealed movement D339 distinct mycobacterial Mfd. Mfd D3 repositioned when ADP Pi bound contacts D1a D5 (Fig. 4a–c). contacts important for nucleotide binding flip-flop action D3–D4 linker (Fig. 4 Movement linker relieves steric occlusion conserved phenylalanine motif Q (F641 stack against adenine base ADP Mfd bound DNA D3 contacts with D1a D7 (Fig.swapping position with D2. UvrA-binding surface D2 residues R165 R18129 exposed Contacts D2 with residues D7 D0148 F1050 lost 4DNA binding unmasks UvrA recruitment surface domain D2 Mfd-free E. coli Mfd complexed to DNA ADP-AlFx. smegmatis bound to ADP Pi c show D3–D4 linker C-terminal residues D3 positional markers D3–D4 linker not visible in DNA-bound Mfd M. smegmatis Mfd Mfd disruption D2–D7 interaction DNA binding Amino acids interactions (R165 R181 D1048 F1050 as CPK unmasking R165 R181 for UvrA recruitment Mfd colored by domain Fig. 2a reconstruction confirms TRG motif in DNA translocation common to Mfd subfamilies helical hairpin “hook” C-terminal. hook disordered in RecG helical in Mfd relay connecting D4 to motor corehelices TRG snap stabilized by ADP•AlFx. binding nucleotide-free Mfd TRG helices splayed open by motif VI helix ATP closed TRG motif engages hook structure releases motif VI 46 Å displacement rotation D6 modulates conformation relay helix bent rotates D5 (Fig. 5 domain D3 unmasking UvrA binding site interlocking motif VI TRG hook relay helices enable complex conformational changes DNA binding. interlocking TRG motif VI relay helix motif VI TRG hook helices relay helix nucleotide-free E. coli Mfd DNA-bound E. coli Mfd conformational changes TRG helical hairpin closes bend into relay helix residues for DNA loading span both lobes motor translocation Mfd load DNA involves D5 D6 no other contacts (Fig. secondary binding determinants might unmasked under specific conditions if isolated Mfd domains contribute to ss dsDNA binding purified Mfd functional modules crystal structure E.coli Mfd homology module MfdD1aD2D1b unknown function MfdD3 RNAP interaction domain MfdD4 clamp domain MfdD7 (Fig. 2a Fig. 4) exception MfdD5D6 assessed DNA binding 40mer HEX-labeled dsDNA shorter Cy3-labeled 27mer no stable binding purified modules Fig. 4) lower affinity secondary binding sites contacts MfdD5D6 DNA fragment longer DNA structure open bubble longer cellular DNA substrates multiple domains DNA binding contacts modules weak EM reconstruction residues proximal DNA specific contacts nucleic acid not resolved residues DNA loading ATP- ADP•AlFx states used EM reconstruction evolutionary relationships structural superpositions nucleic acid bound ATPases Structural alignments with RecG related Mfd unfruitful short fork DNA reach helicase extended analyses other ss/ds nucleic-acid translocases focused similarity with Snf2 SF2 chromatin remodeler structural models functional statesstructures include Snf2 bound to dsDNA41 curved DNA Snf2 to nucleosomes42,43 combined information from EM reconstruction superpositions selected well-conserved residues in Fig. 6a 1 key for DNA recognition constructed Mfd variants substitutions in D5) or D6) All active in ATPase assays comparable to wild type for nucleotide binding hydrolysis consistent with proper folding assessed using circular dichroism 7b). assessed binding to 40mer HEX-labeled dsDNA variants displayed DNA-binding defects in ATPγS and ADP•AlFx Interactions with DNA more sensitive to salt in ATPγS than ADP•AlFx state formation non-electrostatic contacts for gain in affinity in transition state.Fig. 6Key roles for R685 N817 R868 in DNA loading Steady-state ATPase activities of single Mfd variants ATP hydrolysis rates measured using ATP/NADH-coupled ATPase assayData means ± SD (n = 3) Variants color-coded domain Fig. 2a (D5 yellow D6 green Data MfdR685A MfdN817A replotted from Le et al.16 Equilibrium dissociation constants Mfd-dsDNA interaction fluorescence anisotropy binding curves D5 D6 variants color-coded domain Fig. 2a mean values ± SD (n = 3) Error bars smaller Arrows indicate data sets model compromised affinity DNA overestimation affinity Binding curves Supplementary Fig. 7. comparison double-stranded nucleic-acid translocases Escherichia coli superpositions Cα trace D5 Sulfolobus solfataricus Snf2 1.1 Saccharomyces cerevisiae) Snf2 1.4 ADP•BeFx -bound RIG-1 ssRNA Drosophila melanogaster HCV NS31.1 Bound Mg2+ dark green sphere ADP•AlF3− aquamarine CPK model nucleotide magenta sticks ADP red CPK BeFx dark red CPK nucleic-acid strands red orange two lobes ATPase core yellow green (C DNA-binding residues highlighted slate CPK models domains gray ribbon annotated ATPase motifs motifs contacting DNA color-coded slate font not visible parenthesis conservation motor cores D5 D6 E. coli Mfd Source data Source Data file.MfdR685A MfdN817A MfdT868A impaired ATPγS ADP•AlFx critical role loading DNA T868 conserved helicase motif V (Fig. ATP oligonucleotide Mfd substitution T868 affects DNA binding less ATP binding hydrolysis (Fig. duck RIG-I RNA helicase 19mer dsRNA ADP•AlFx conserved threonine (T698 contact phosphate RNA backbone (Fig. hydrogen bond phosphoryl oxygenmutation residue to alanine in human RIG-I eliminates dsRNA binding ATP hydrolysis confers affinity uncoupled from splitting phosphoanhydride bond In Thermotoga maritima RecG mutation T478 reduces fork reversal translocation without ATP hydrolysis47 equivalent threonine T546 contacts RNA phosphate backbone Vasa RNA helicase bound ssRNA AMPPNP SF2 helicase domain DExH helicase NS3 has similar threonine oligonucleotide in mutation threonine (T411 in NS3h affect basal ATPase activity abolishes oligonucleotide binding duplex unwinding.Mfd mechanism translocation reminiscent RNA helicasesIn ADP•AlFx variants (MfdK690A showed slight impairment function no DNA-binding defect MfdH711A substitutions of T710A H842A loss affinity ATPγS state minor effect on DNA affinity ADP•AlFxoptical trapping shown nucleotide-starved TEC Mfd binds to TEC D5 front followed by D6 ATPγS binding D6 moves 12 bp toward D5 approaching RNAP ATP hydrolysis required for translocation D5 may invade RNAP footprint residues with nucleotide-dependent roles critical for translocation developed fluorescent triplex displacement assay dsDNA translocases labeled triplex oligonucleotide off DNA duplex Triplex displacement assays established for hyperactive Mfd variants MfdD7−AAA E1045A D1048A R1049A substitutions D7 disrupt inhibitory clamp interaction (Fig. 7a short processivity of Mfd ~200 employed 70 bp substrate shorter hyperactive MfdD7−AAA variant included non-labeled ssDNA Mfd weaker trap contaminating endonuclease activity signal full-length TFO MfdD7−AAA free TFO accumulated inhibited by non-hydrolysable analog AMPPNP absence of nucleotide (Fig. 7b–c Supplementary Figmutants MfdD7−AAA;T710A;H842A defects translocation loading not ATP hydrolysis (Fig. circular dichroism spectra similar MfdD7−AAA Fig loading defects variants different than T710A H842A variants wild-type background. 6a–c suggest artificial disruption clamp uncouples DNA nucleotide status derepressed variants display severe translocation defects MfdH842A MfdT710 no translocation activity (Fig. 7b higher protein concentrations ATPγS ADP•AlFx translocation activity MfdD7−AAA variants decreases Fig. 8) multiple loading molecules interfering analysis wild-type Mfd not stimulated herring sperm DNA MfdD7−AAA MfdD7−AAA;T710A;H842 ATPase activity (Fig. 7f 8f). deregulation clamp disruption hyperactivates Mfd mutations motifs Ic IVa repress Mfd activity. 7T710 H842 critical regulators translocation Mfd model DNA-bound Mfd T710 H842 residuesconformation motif IVa loop amino-acid rotameric forms residues Cα spheres H842 T710A yellow E1045 D1048 R1049 Mfd D7AAA TFO displacement assays MfdD7−AAA variants remaining TFO-Triplex plotted time translocation ± SD 3) error bars smaller Electrophoretic separation TFO MfdD7−AAA gels MfdD7−AAA +/-ATP +AMPPNP gels variants Supplementary Fig. 8 3) Steady-state ATPase activities Mfd variants wild-type Mfd activity Data means ± SD 3) p < 0.01 < 0.0001 > 0.05 p value MfdD7AAA;T710 DNA 0.0012 Equilibrium dissociation constants Mfd-dsDNA interaction fluorescence anisotropy binding curves titrating protein 10 nm HEX-labeled 40mer DNA 2.0 mm ATPγS ADP•AlFx Data mean values ± SD (n 3) KD values Supplementary Table 1. Source data fileH842 maps-terminal motif IVa conserved within Mfd family not across translocases (Fig. 2a). close to R393 in HCV NS3 maps Fig. 9) coined arginine clamp key for nucleic-acid binding structures HCV translocase NS3h bound to DNA ATP transition state mimics contacts nucleic acid with R393 nucleotide-dependent structural transitions T710 (T269 NS3h T411 T86850 T269 in helicase motif Ic TxGx involved in nucleotide-dependent oligonucleotide binding helicases including NS344 NS3 structure absence ATP shows threonine residues three nucleotides apart only two apart in ATP transition state analog Fig. 9b–d T411 maintains contact same nucleotide ATP hydrolysis T269 releases DNA ATP-bound contact nucleotide NS3h complex ATP transition-state analog Conservation two threonine residues in Mfd nucleotide-dependent role of T710 T868 critical for DNA binding features translocation in NS3h Mfd Mfd lacking helicase activity binding DNA RNA-dependent processes rely on translocation dsDNA important for early events RNAP targets destabilizing TECs halted by DNA damage signals roadblocks6 Mfd action on transcription complexes torque transcription bubble release RNA transcript26 first structure of DNA-bound E. coli Mfd complementary functional analyses data allow first model nucleotide status DNA binding control interdomain contacts reveal molecular basis for DNA recognition by Mfd proteins DNA bound by two lobes motor core with previous analysis ATPγS DNA interaction determinants restricted to D5 D614 domain boundary definitions not based on structural information truncated ATPase motif VI TRG motif (Fig. nucleotide status translocation33 no binding of ds/ssDNA to domains flanking motor core studies similarity between D1aD2D1b region in UvrB contacts DNA during NER melts for damage recognition beta hairpin beta hairpin absent in Mfd replaced by short loop lacks aromatics nucleotide flipping loop ~50 Å away from D5/D6 binding site similarity architecturalreconstruction suggests DNA substrates DNA fragment approach C-terminal region relay helix studies identified substitution (MfdW550A) enhances DNA binding motor could involve bending DNA across protein surface rearrangements hook TRG motifs DNA conceptualize DNA binding by Mfd two clusters residues at tip D5 D6 pincers establish contacts with substrate tracking strand cycle between high-affinity-bound low-free shares features with HCV NS3 system cycles between low nucleotide-bound to high-free Cluster I R685 T710 cluster II R865, N817, T868 H842 distinguish residues key for loading (R685 T868 translocation (T710 nucleotide-dependent roles HCV NS3 system interesting determine if Mfd operates via Brownian ratchet translocation 1 bp steps NS3) RNAP rectify ratchet high processivity base pairs-molecule no biochemical evidence for TRG helices involved in DNA binding element with ATPase motif VI hook mobile likely used as pawl during translocation first structural evidence for hypothesized interlocking of nucleotide-responsive elements TRG, hook relay helices Fig2c, driven by ATPase motif VI.Translocases along dsDNA extensive interactions with DNA strand tracking strand structure S. solfataricus Snf2 extensive interactions between tracking strand motifs I Ia Ib Ic IIa41 N-terminal lobe (Fig. 2a, studies Snf2 suggest tighter interactions with motifs C-terminal lobe42 impossible to determine which strand represents “tracking” strand question underlies strand-specific probing damage UvrA UvrB Mfd required52 identification of Mfd family-specific residue for translocation (H842 in motif IVa provides avenue Mfd inhibitor high selectivity Mfd deletion limits antimicrobial resistance in bacteria E. Salmonella Bacillus Helicobacter Campylobacter jejuni22 M. tuberculosis25 in Bacillus cereus Shigella flexneri experiments implicated Mfd in resistance to nitric oxide response Mfd inhibitor novel strategy for antibiotic development could target variable motif IVa arginine clamp in motif IVa druggable57 establishing H842 contacts DNA or conformation motif IVa loop challengingwork evidence D3 key TCR regulation E. coli D3 between D1a D5/D6 nucleotide-free Mfd repositioned interactions D7 Motion D3 coupled peppermill-like motion UvrB homology module C-terminal region Mfd (D4D5D6D7) rearrangements close D5-D7 interdomain cleft D4–D7 interactions unperturbed UvrA recruitment surface unmasked conformational changes coupled active site motif VI movements TRG hook relay helices unresolved await higher resolution analyses.MethodsCloning protein expression purificationConstructs pMS1-pMS8 obtained site-directed mutagenesis pAD6-tagged E. coli Mfd58) Edelheit et al constructs verified DNA sequencing purified protein mass spectrometry Overexpression purification Ni2+-affinity chromatography His-tag cleavage heparin chromatography gel constructs pMS2 pMS5 HiTrap Q-sepharose column (GE Healthcare) used heparin affinity Size-exclusion chromatography Superdex 200 10/30 Increase column (GE Healthcare) buffer 20 mm Tris pH 8.0,10 NaCl 5 mm DTT truncations N-terminal fusions polyhistidine tags Table 3) cleaved Prescission protease affinity chromatography displacement 70mer DNA (50 nm mixed 21mer TFO oligo (25 nm (10 pH 5.5 12.5 mm MgCl2) heated 57 °C 15 min 4 °C overnight Protein samples dialyzed 10 mm Tris-HCl pH 8 100 mm NaCl 10 mm MgCl2 2 mm β-mercaptoethanol Reactions 10 nm TFO-triplex 50 nm 44 bp ssDNA trap 10 MgCl2 1 mm DTT protein nm 4.5 Triplex displacement 2 mm ATP Samples withdrawn GSM buffer glucose 3% SDS 250 pH loaded 3–8% Tris-Acetate gels pre-run 60 min 75 V-Acetate buffer pH 5.5 5 sodium acetate 1 mm MgCl2) 4 °C Samples electrophoresed 180 min 75 V imaged ChemiDoc system disappearance TFO-Triplex quantified BioRad ImageLab plotted Graphpad Prismpreparation cryo-exclusion chromatography 200 10/30 GE Healthcare reconstitution buffer 20 mm Tris-HCl pH 8 100 mm NaCl 20 MgCl2 2 mm TCEP Mfd incubated 21mer dsDNA 1:1.5 molar excess 2 mm ADP 10 mm NaF 1 mm AlCl3 2.5 μl Mfd-DNA complex 1.0 mg/mL UltrAuFoil R1.2/1.3 300 mesh grids plasma-cleaned Gatan Solarus argon/25% oxygen 7 blotted Whatman No 1 filter paper cold room >80% humidity plunged liquid ethane dsDNA reconstitution sequence 5′-ATAGGATACTTACAGCCATCG-3′-EM data image Leginon60 Talos Arctica microscope 200 kV K2 Summit electron detector magnification ×36,000 defocus range 2.0 μm to 4.0 μm pixel size 1.15 Å level 30° tilts standard procedures Supplementary Table-EM image processingMovies aligned dose-weighted MotionCor262. CTF estimation particle picking Warp63 69,721 particles imported Relion 2D classification round bad particles excluded3D classification five classes run in Relion Fig 2) initial model generated from small subset data best class selected from 3D particle stack Relion contained 19,083 particles input into cryoSPARC for 2D 3D classification cleaned stack 9822 particles imported into cisTEM refined Auto-Refine frequencies up to 80% Nyquist included final resolution 5.47 Å Fourier Shell Correlation (FSC) criteria refined map sharpened in cisTEM amplitude spectrum between 8 Å and resolution 5.5 Å Directional resolution volumes generated 3D local resolution calculated sxlocres.py Sparx processing Details model building refinement in Supplementary Methods Nature Research Reporting Summary.Supplementary information Files | 50 | 0.805411 |
10.1038/s41467-020-19773-y | PMC7683529 | There is a long-standing experimental effort to observe field-induced correlated states in three-dimensional materials. Here, the authors observe an unconventional Hall response in the quantum limit of the bulk semimetal HfTe5 with a plateau-like feature in the Hall conductivity. | Interacting electrons confined to their lowest Landau level in a high magnetic field can form a variety of correlated states, some of which manifest themselves in a Hall effect. Although such states have been predicted to occur in three-dimensional semimetals, a corresponding Hall response has not yet been experimentally observed. Here, we report the observation of an unconventional Hall response in the quantum limit of the bulk semimetal HfTe5, adjacent to the three-dimensional quantum Hall effect of a single electron band at low magnetic fields. The additional plateau-like feature in the Hall conductivity of the lowest Landau level is accompanied by a Shubnikov-de Haas minimum in the longitudinal electrical resistivity and its magnitude relates as 3/5 to the height of the last plateau of the three-dimensional quantum Hall effect. Our findings are consistent with strong electron-electron interactions, stabilizing an unconventional variant of the Hall effect in a three-dimensional material in the quantum limit. | IntroductionApplying a strong magnetic field to an electron gas confines the electrons motion in cyclotron orbits with a set of discrete eigenenergies—the Landau levels. In two-dimensional (2D) systems, this quantization leads to a fully gapped energy spectrum and to the emergence of the quantum Hall effect (QHE)1. In the limit where only the lowest Landau level (LLL) is occupied (the so-called quantum limit), electron–electron interactions can play a significant role, leading to the appearance of correlated states, such as the fractional quantum Hall effect2. In contrast, the Landau level spectrum of a three-dimensional (3D) electron gas is not fully gapped and becomes like that of a one-dimensional system. As a consequence, the electrons can still move along the field direction, which in turn destroys the quantization of the Hall effect3–5. Nevertheless, it has been predicted that a generalized version of the QHE can emerge in 3D electron systems that exhibit a periodically modulated superstructure6–8. Analogous to as in two dimensions, in the vicinity of the quantum limit, 3D electron systems are also prone to form a variety of correlated electron states, including Luttinger liquids; charge, spin and valley density waves; excitonic insulators; Hall and Wigner crystals; or staging transitions in the case of highly anisotropic layered systems3,4,6,9–13. It has been theoretically pointed out that some of these states are related to quantum Hall physics in three dimensions and likewise could manifest themselves in a Hall response that should be observable in the quantum limit of 3D semimetals10,13,14.Inspired by these ideas, the possibility of finding a 3D QHE has been explored in several material systems. For example, signatures of the integer quantum Hall effect (IQHE) have been found in quasi-2D semiconducting multilayer lattices15, Bechgaards salts16,17, η-Mo4O1118, n-doped Bi2Te319, and EuMnBi220, in which the layered crystal structure itself supplies the stack of 2D systems. Very recently, the QHE has also been observed in 3D graphite films5, bulk ZrTe521, and HfTe5 samples22. In graphite, the imposed periodic superstructure has been attributed to the formation of standing electron waves. In ZrTe5 and HfTe5, the IQHE was originally believed to arise from a charge density wave (CDW), due to the scaling of plateau height with the Fermi wavevector. This scenario is, however, in contrast with thermodynamic and thermoelectric measurements on ZrTe5 that did not reveal any signatures of a field-induced CDW transition. Instead, it was proposed that ZrTe5 should be considered a stack of weakly interacting Dirac 2DEGs with the plateau height scaling originating from the interplay of small carrier density and the peculiarities of Landau quantization of the Dirac dispersion23. In parallel to the search for the 3D QHE, there has been a long-standing experimental effort to observe field-induced correlated states in the quantum limit of three-dimensional materials. Although those studies have provided signatures of field-induced states in the longitudinal electrical resistivity of Bi24,25, ZrTe521,26 and graphite5,27–30, correlated states with Hall responses have yet to be observed.In this work, we present measurements of the low-temperature longitudinal and Hall resistivities of the 3D semimetals HfTe5 and ZrTe5. Previous studies have shown that HfTe5 is an isostructural counterpart of ZrTe531. Both materials share an orthorhombic crystal structure and a single elliptical 3D Fermi surface, comprising less than 1% of the Brillouin zone and hosting massive Dirac Fermions with almost linearly dispersing bands in the vicinity of the Fermi level (see Supplementary Information for details). These specific properties have been considered essential for the observation of the 3D IQHE in both materials21–23. Moreover, recent progress in ZrTe521 and HfTe532 single-crystal growth has enabled a Hall mobility μ that exceeds 100,000 cm2 V−1 s−1 at low temperatures (<4 K) (see “Methods” section and Supplementary Information S1). The quality of these crystals is comparable to that of graphene samples, which have previously proven appropriate for observing the FQHE in two dimensions33. While the 3D band structure of ZrTe5 and HfTe5 is very similar21, hafnium has a higher atomic number than zirconium and hence naturally introduces stronger spin–orbit coupling (SOC)31, which has been previously shown to stabilize correlated states in the quantum limit of 2D electron systems34. Therefore, HfTe5 is the more promising candidate for the observation of unconventional Hall responses of correlated ground states in its quantum limit.ResultsHfTe5 and ZrTe5 typically grow as millimeter-long ribbons with an aspect ratio of approximately 1:3:10, reflecting their crystalline anisotropy. Details of the growth process, crystal structure, and first transport characterization of our samples can be found in refs. 21,32. We have measured the longitudinal electrical resistivity ρxx and Hall resistivity ρxy (see “Methods” section) of three HfTe5 samples (A, B, C) and three ZrTe5 samples (D, E, F) as a function of magnetic field B and temperature T, with the electrical current applied along the a-axis of the crystals.At T = 300 K, ρxx is around 0.5 mΩ cm (see Fig.1a, Supplementary Fig. S1 and ref. 32) with an electron density of n = 1.3 × 1019 cm−3 and μ = 10,000 cm2 V−1 s−1 21,23,32. Upon cooling in zero magnetic field, ρxx increases with decreasing T until it reaches a maximum at TL = 70 K (Fig.1a). Such a maximum has previously been observed in HfTe532 and ZrTe521, and it is attributed to a Lifshitz transition, here inducing a change in charge-carrier type. Consistently, the slope of ρxy(B) changes sign at TL, indicating electron-type transport for T < TL32. At 3 K, we find n = 8.7 × 1016 cm−3 and μ = 250,000 cm2V−1s−1 (see Supplementary Note 1 and ref. 21). All investigated samples show similar electrical transport properties. In the main text, we focus on data obtained from HfTe5 sample A and ZrTe5 sample D. Additional data of samples B, C, E, and F can be found in the Supplementary Information and ref. 23.Fig. 1Three-dimensional morphology of the Fermi surface in HfTe5.a Longitudinal electrical resistivity ρxx as a function of temperature T at zero magnetic field. Inset: Sketch of the measurement configuration in the three spatial directions x, y, and z. The bias current I is applied along x and the magnetic field B along y. The corresponding voltage responses are measured in x- (Vxx) and in y direction (Vxy). b
ρxx as a function of B at 3 K with B applied along x, c along y and d along the z direction. e Variation of the longitudinal electrical resistivity Δρxx as a function of B at 3 K with B applied along x, f, along y, and g along the z direction. h Landau-index fan diagram for the integer Landau levels N for different angles α of B in the z–x plane (see inset; α is positive from the z to the x direction) as a function of B−1. Data are obtained from the minima of ρxx in extended data Fig. 4a. i, Shubnikov-de Haas frequency as a function of angle α and j. β.
β is the rotation angle of B in the z–y plane (see inset; β is positive from the z to the y direction). The black dots represent the measurement data. The red fitting curve represents a planar 2D Fermi surface, the blue fitting curve corresponds to an ellipsoidal 3D Fermi surface. k The Fermi surface of HfTe5 in momentum space along the kx, ky, and kz direction.To characterize the Fermi-surface (FS) morphology of our pentatelluride samples, we have measured Shubnikov-de Haas (SdH) oscillations with respect to the main crystal axes at 3 K. For this purpose, we followed the analysis of ref. 21 and rotated B in the z–y and z–x planes, while measuring ρxx (B) at a series of different angles (Fig. 1, Supplementary Fig. S3 and ref. 23). The SdH frequency BF,i is directly related to the extremal cross-section of the Fermi surface SF,i, normal to the applied B direction via the Onsager relation BF,i = SF,i(ℏ/2πe). Examples of the SdH oscillations for which the magnetic field was aligned along the three principal crystallographic directions (x, y, and z axes) are shown in Fig. 1b–d (upper panels). In each field direction, we have observed maxima in ρxx that are periodic in 1/B, each of which corresponds to the onset of a Landau level. In the associated minima, ρxx(B) does not vanish, which is a consequence of the remaining dispersion in z direction in 3D systems and Landau level-broadening due to disorder (see Supplementary Note 4, Supplementary Fig. S4–S7 and ref. 23 for details). To determine the SdH oscillation frequency, we have subtracted the smooth high-temperature (50 K)—ρxx(B) from the low-T-data, obtaining the oscillating part of the longitudinal resistivity Δρxx(B). Employing a standard Landau-index fan diagram analysis to Δρxx(B) (Fig. 1h, Supplementary Figs. S3, S8, S9, Supplementary Note 2 and ref. 23), we have found only a single oscillation frequency for all rotation angles measured, consistent with a single electron pocket at the Fermi energy. The extracted BF,i of sample A for B along the three principal directions are BF,x = (9.9 ± 0.1) T, BF,y = (14.5 ± 0.5) T, and BF,z = (1.3 ± 0.1) T. Here, the errors denote the standard deviation of the corresponding fits.In contrast to 2D materials, HfTe5 and ZrTe5 show in-plane SdH oscillations when B is aligned with x and y, indicating a 3D Fermi-surface pocket. The shape of the FS is further determined by the analysis of the rotation angle-dependence of BF. As shown in Fig. 1i, j, the angle-dependent SdH frequency is well represented by a 3D ellipsoidal equation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B_{{\mathrm{F}},3D} = B_{{\mathrm{F}},z}B_{{\mathrm{F}},i}/\sqrt {(B_{{\mathrm{F}},z}\sin \theta )^2 + (B_{{\mathrm{F}},i}\cos \theta )^2}$$\end{document}BF,3D=BF,zBF,i/(BF,zsinθ)2+(BF,icosθ)2, where θ is the rotation angle in the z–i plane. As a cross-check, we also plot the formula of a 2D cylindrical Fermi surface BF,2D = BF,z/cosθ, which deviates significantly from the experimental data for θ > 80°. Hence, the ellipsoid equations can be used to obtain the Fermi wave vectors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_{F,x} = \sqrt {S_{{\mathrm{F}},y}S_{{\mathrm{F}},z}} /\sqrt {{\uppi}S_{{\mathrm{F}},x}}$$\end{document}kF,x=SF,ySF,z/πSF,x = (0.005 ± 0.001) Å−1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_{{\mathrm{F}},y} = \sqrt {S_{{\mathrm{F}},x}S_{{\mathrm{F}},z}} /\sqrt {{\uppi}S_{{\mathrm{F}},y}}$$\end{document}kF,y=SF,xSF,z/πSF,y = (0.008 ± 0.001) Å−1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k_{{\mathrm{F}},z} = \sqrt {S_{{\mathrm{F}},x}S_{{\mathrm{F}},y}} /\sqrt {{\uppi}S_{{\mathrm{F}},z}}$$\end{document}kF,z=SF,xSF,y/πSF,z = (0.058 ± 0.006) Å−1 that span the 3D FS of HfTe5 sample A in the x, y, and z direction, respectively (Fig. 1k). The errors in kF,i originate from the errors of the BF,i. The preceding analysis indicates that for our HfTe5 and ZrTe5 samples, the quantum limit with the field along the z is achieved already for the field of BC = 1.8 and 1.2 T21,23, respectively. Further details of our band-structure analysis can be found in Supplementary Fig. S6, Supplementary Table S1, refs. 21,23. Above 6 T, we find that for both materials, ρxx(B) steeply increases with the magnetic field (Fig. 2b). Such a steep increase has previously been observed in ZrTe5 and has been attributed to a field-induced metal-insulator transition21.Fig. 2Three-dimensional integer quantum Hall effect in HfTe5.a Longitudinal electrical resistivity ρxx (blue, left axis) and Hall resistivity ρxy (red, right axis) and as a function of B at T = 3 K with B applied in z. The blue arrows mark the onset of a Landau level (LL). The blue numbers denote the index N of the corresponding LL. The plateau in ρxy scales with (h/e2) π/kF,z, with the Planck constant h, the electron charge e, and the Fermi wavevector in z direction kF,z. b
ρxx and c
ρxy as a function of B for various temperatures T ≥ 3 K with B applied in z. d
ρxy as a function of |B|cos(β) for magnetic fields along the direction with angles β at 3 K.For the field-aligned with the z axis (B||z), we additionally observed in both studied compounds pronounced plateaus in Hall resistance ρxy(B) that appear at the minima of the SdH oscillations in ρxx(B)—features commonly related to the QHE (Fig. 2a and ref. 23). The height of the last integer plateau is given by (h/e2) π/kF,z, similar to as reported in the literature for the 3D IQHE21,22. The plateaus are most pronounced at low temperatures, but still visible up to T = 30 K (Fig. 2b, c and ref. 23).We note that the observed quantization of ρxy is not immediately obvious from the predicted quantization in σxy. The Hall resistivity tensor is given by ρxy = σxy/(σxxσyy + σxy2) with a magnetic field in z direction, where σxx and σxx are the longitudinal component of the conductivity tensor in x and y direction, respectively. Vice versa, the Hall conductivity tensor element is given by σxy = ρxy/(ρxx
ρyy + ρxy2) with a magnetic field in z direction. However, in our samples at low temperatures σxx < σxy (Supplementary Figs. S11 and S12) and thus σxy−1 ≈ ρxy, enabling the direct observation of the quantization. Due to the geometry of the HfTe5 crystals (elongated needles) and their mechanical fragility, performing reliable measurements of ρyy is not possible. Instead, we estimate the error of the σxy using the ratio of Drude resistivities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _{yy}/\rho _{xx} = (n_{xy}e^2t_x/m_x^ \ast )/({\it{n}}_{xy}e^2t_y/m_y^ \ast )$$\end{document}ρyy/ρxx=(nxye2tx/mx*)/(nxye2ty/my*) given by the quantum lifetimes and effective masses obtained from Shubnikov-de Haas oscillations on sample A (Supplementary Table S1). nxy is the charge-carrier concentration in the x–y plane. Based on this analysis, we find ρyy/ρxx ≈ 0.4, which results in an error of below 8 % in the estimated σxy for the investigated field range owing to ρxx (B) < ρxy (B). Both these errors lay within the estimated error of kF,z of 10 %.Figure 2d shows the angular-dependence of the Hall plateaus, which we find to scale with the rotation angle. This behavior is very similar to the sister compound ZrTe5.21 In both materials, the height of the Hall plateaus and its position in |B| depends only on the field component that is perpendicular to the x–y plane B⊥ = |B|cosθ21,23.So far, our analysis focused on similarities between the Hall effects observed in ZrTe5 and HfTe5. Upon cooling the samples to 50 mK, an obvious difference emerges, as shown in Fig. 3 and Supplementary Figs. S11–S15. At low fields below the quantum limit (B < BC), both compounds exhibit signatures of new peaks and plateaus in ρxx(B) and ρxy(B). Such features have been observed in the past and are related to spin splitting of the Landau levels23,35. However, at high fields (B > BC)—in the quantum limit, HfTe5 exhibits an additional peak in ρxx(B), accompanied by a plateau-like feature in ρxy(B). This is in sharp contrast to ZrTe5, in which ρxx(B) and ρxy(B) smoothly increase. Using the Landau-index fan diagram obtained at 3 K and gauging the indexing of Landau bands with respect to the N = 1 band, we find that the additional peak in ρxx(B), in the quantum limit of HfTe5 is situated at N = 3/5. This indexing is confirmed by corresponding maxima in ρxx(B) and/or σxx(B) of all three HfTe5 samples investigated (compare Supplementary Figs. S11 and S12), despite being less pronounced in some of them.Fig. 3Low-temperature longitudinal magnetoresistivity and Hall resistivity in isostructural HfTe5 and ZrTe5 at 50 mK.a Longitudinal electrical resistivity ρxx (blue, left axis) and Hall resistivity ρxy (red, right axis) of ZrTe5 as a function of B/BQL at T = 50 mK with the magnetic field B applied in z for 0 T ≤ B ≤ 3 T (upper panel) and 0 T ≤ B ≤ 9 T (lower panel). The blue arrows mark the onset of the Landau levels. BQL denotes the magnetic field of the onset of the N = 1 Landau level. The blue numbers label the index N of the Landau level and the red numbers label the corresponding value of ρxy with respect to (h/e2)π/kF,z. b Longitudinal electrical resistivity ρxx (blue, left axis) and Hall conductivity σxy (red, right axis) of ZrTe5 as a function of B at T = 50 mK with B applied in z for 0 T ≤ B ≤ 3 T (upper panel) and 0 T ≤ B ≤ 9 T (lower panel).Although the relation between the magnitude of the plateau-like feature and its corresponding Landau index is not obvious from ρxy(B), a comparison of the respectively calculated conductivity reveals that the magnitude of the plateau-like feature in the quantum limit scales as 3/5-times with respect to the plateau related to the LLL. A close investigation of the plateau height in conductivity (Supplementary Fig. S16) reveals that both the plateau-like features at N = 1 and N = 3/5 are well developed in the conductivity, within 1 and 2% of N·(e2/h)kF,z/π in the range of 0.5 T around the plateau center.In order to verify whether the observed features can be explained by invoking the presence of a second pocket at the Fermi energy, we have performed additional magneto-transport measurements up to 70 T (Supplementary Fig. S16 and ref. 23). The measurements did not reveal any additional quantum oscillations, which is consistent with band-structure calculations31 and a previous ARPES study on our samples36: The Fermi level, obtained from the analysis of the Shubnikov-de Haas oscillations is (9 ± 2) meV (Supplementary Information), which is in agreement with the ARPES experiment. According to the ARPES data, at 15 K, the nearest additional band is located ~5 meV above the Fermi level (lowest temperature measured in the ARPES study) as compared to the Fermi function broadening of kB·15 K ≈ 1 meV. Below 15 K, the Fermi level stays constant with respect to the band edges, as indicated by the temperature-independent Shubnikov-de Haas frequency in our experiments. Hence, the next nearest band in our samples is ~kB·60 K away from the Fermi level and does not contribute to the low-temperature transport experiments. Our data can, therefore, be analyzed in terms of a single electron-type Dirac pocket.Further insight into the possible origin of the N = 3/5 state in the quantum limit can be obtained from the line shape of Δρxx(B), which resembles the line shape of σxx(B), (Fig. 4e, f) a common feature of canonical 2D QHE systems37 A related empirical observation is that in both fractional and IQHE in 2D systems the longitudinal resistance is ρxx(B) is connected to ρxy(B) via ρxx(Bz) = γB·dρxy(B)/dB, where γ is a dimensionless parameter of the order of 0.01–0.05, which measures the local electron concentration fluctuations38,39. Comparison of σxx(B) (Fig. 4d) and γB·dρxy(B)/dB (Fig. 4e, upper panel) as a function of B−1 reveals that both quantities show maxima and minima at the same field positions as Δρxx(B). In particular, the derivative relation is well fulfilled with γ = 0.04, which is in the expected range reported for 2DESs. These results suggest that the observed plateau-like feature observed in the quantum limit in HfTe5 is related to quantum Hall physics.Fig. 4Three-dimensional Hall response in the quantum limit of HfTe5.a Longitudinal electrical resistivity ρxx (blue, left axis) and Hall resistivity ρxy (red, right axis) and as a function of B at T = 50 mK with B applied in z for 0 T ≤ B ≤ 3 T (upper panel) and 0 T ≤ B ≤ 9 T (lower panel). The blue arrows mark the onset of a Landau level. The blue numbers label the index N of the Landau level and the red numbers label the corresponding value of σxy with respect to (e2/h)kF,z/π. b Variation of the longitudinal electrical resistivity Δρxx as a function of B (upper panel, left axis), γ
B·dρxy/dB (right axis, upper panel) (γ = 0.04) and Landau-index fan diagram (lower panel) as a function of B−1 at T = 50 mK with B applied in z. c
ρxx and d
ρxy as a function of B for various temperatures 4 K ≥ T ≥ 50 mK with B applied in z.To gain quantitative insights into the states that cause the features in the Hall effect in HfTe5, we have estimated the gap energies of the x–y plane ΔN associated with N = 1 and N = 2 below and N = 3/5 in the quantum limit. We have fitted the T-dependence of the ρxx(B) minima (Fig. 4f) in the thermally activated regime ρxx(B) α exp(−ΔN/2kBT), where kB is the Boltzmann constant (Supplementary Fig. S17). For integer N, we find Δ1 = (40 ± 2) K at N = 1 and Δ2 = (9 ± 1) K at N = 2. The gap energy of the N = 3/5 state in the quantum limit is two orders of magnitude lower: Δ3/5 = (0.49 ± 0.09) K. The deviations given for the gaps are the errors obtained from the thermally activated fits in Supplementary Fig. S13. In spite of considerable Landau level-broadening, both the size of the gaps of the integer and the N = 3/5 states compare well with the gaps obtained for integer and correlated quantum Hall states in 2DESs37,40. The different ΔN are also in agreement with the T-dependence of the corresponding Hall features (Fig. 4g). While the integer plateaus are observable up to tens of Kelvin, the plateau-like feature in the quantum limit vanishes at around 0.5 K.Those considerations suggest that the Hall feature observed in the quantum limit of HfTe5 is associated with physics in the LLL only (as pointed out above, this Landau level is non-degenerate since HfTe5 is like ZrTe5 a gapped Dirac semimetal). The finite value of ρxx at N = 3/5 implies the absence of a fully established bulk gap, which in turn means that a truly quantized Hall effect as in 2D systems without the kF,z-scaling cannot be expected. Nevertheless, the emergence of the plateau-like feature in the quantum limit of a single-band system at low temperatures calls for an explanation beyond a simple single-particle picture: The Hall conductivity of a non-interacting single band in which the chemical potential adjusts to keep the particle number fixed simply decreases as 1/B, as observed in the isostructural ZrTe5 (Fig.3b).DiscussionAlthough possible scenarios for the emergence of a plateau in the Hall resistance in the quantum limit of electron plasma include the formation of a CDW, Luttinger liquid, Wigner crystallization, or the so-called Hall crystal41,42, a favorable scenario builds on the notion that ZrTe5 and HfTe5 can be thought of as a stack of interacting 2DEGs. Based on a Hartree–Fock analysis, it was proposed13 that in a layered structure the gain in exchange energy can exceed the energy cost for distributing electrons unequally between layers. The electrons then undergo spontaneous staging transitions in which only every i-th layers is occupied, while all other layers are emptied (the number i depends on the average electron density and the state formed)—some of which are only stabilized due to the interplay of electron interaction and spin–orbit coupling43. Depending on layer separation, electron density, and the strength of electron–electron interactions, various types of layered Laughlin states or Halperin states can then be formed3,13,14. These states are naturally associated with Hall responses. While staging transitions are unlikely in isotropic three-dimensional materials at high electron densities, HfTe5 has a very anisotropic band structure with small tunneling amplitudes along z, and hosts only a relatively small number of electrons in its Dirac pocket. Our data are thus consistent with strong interactions stabilizing a correlated state that gives a Hall response in HfTe5 in the quantum limit.In conclusion, our measurements reveal an unconventional correlated electron state manifested in the Hall conductivity of the bulk semimetal HfTe5 in the quantum limit, adjacent to the 3D IQHE at lower magnetic fields. The observed plateau-like feature is accompanied by a Shubnikov-de Haas minimum in the longitudinal electrical resistivity and its magnitude is approximately given by 3/5(e2/h)kF,z/π. Analysis of derivative relations and estimation of the gap energies suggest that this feature is related to quantum Hall physics. The absence of this unconventional feature in the quantum limit of isostructural single-band ZrTe5 samples with similar electron mobility and Fermi wavevector indicates the presence of a correlated state that may be stabilized by spin–orbit coupling. However, further experimental and theoretical efforts in determining the real interactions and texture of the field-induced correlated states in HfTe5 are necessary to settle the puzzle of the unconventional Hall response in the quantum limit. In particular, experiments directly probing the density of states and the real space charge distribution such as Scanning Tunneling Spectroscopy and in-field X-ray diffraction could shed additional light on the nature of the observed feature.MethodsSingle-crystal sample growth and pre-characterizationSingle crystals of HfTe5 were obtained via a chemical vapor transport method. Stoichiometric amounts of Hf (powder, 3 N) and Te (powder, 5 N) were sealed in a quartz ampoule with iodine (7 mg ml−1) and placed in a two-zone furnace. A temperature gradient in the range of 400–500 °C was applied. After ca. 1-month, long ribbon-shaped HfTe5 single crystals were extracted from the ampule with a typical size of the single crystals 1 mm × 0.5 mm × 3 mm (width × height × length). High-quality needle-shaped (about 0.1 × 0.3 × 20 mm3) single crystals of ZrTe5 were synthesized using the tellurium flux method and high-purity elements (99.9999% zirconium and 99.9999% tellurium). The lattice parameters of the crystals were confirmed by single-crystal X-ray diffraction. The samples used in this work are of the same batch as the ones reported in refs. 21,23,32,36 and have similar Fermi level positions. As shown in these papers, in our HfTe5 and ZrTe5 samples a three-dimensional topological Dirac semimetal state emerges only at around Tp ≈ 65 K (at which the resistivity shows a pronounced peak), manifested by a large negative magnetoresistance. This Dirac semimetal is a critical point between two distinct topological insulator phases: weak (T > Tp) and strong (T < Tp). At high temperatures, the extracted band gap is around 30 meV (185 K), and at low temperatures 10 meV (15 K)36. However, we note that the Fermi level at these temperatures is not located in the gap, but several meV in the valance band for T > Tp and in the conduction band for T < Tp. Hence, our HfTe5 and ZrTe5 samples are metallic at both high and low temperatures.Electrical transport measurementsElectrical contacts to the HfTe5 and ZrTe5 single crystals were defined with an Al hard mask. Ar sputter etching was performed to clean the sample surface prior to the sputter deposition of Ti (20 nm) and Pt (200 nm) with a BESTEC UHV sputtering system. Subsequently, Pt wires were glued to the sputtered pads using silver epoxy. All electrical transport measurements up to ±9 T were performed in a temperature-variable cryostat (PPMS Dynacool, Quantum Design), equipped with a dilution refrigerator inset. To avoid contact-resistance effects, only four-terminal measurements were carried out. The longitudinal ρxx and Hall resistivity ρxy were measured in a Hall-bar geometry with standard lock-in technique (Zurich instruments MFLI and Stanford Research SR 830), applying a current of 10 μA with a frequency of f = 1 kHz across a 100 kΩ shunt resistor. The electrical current is always applied along the a-axis of the crystal.The pulsed magnetic field experiments up to 70 T were carried out at the Dresden High Magnetic Field Laboratory (HLD) at HZDR, a member of the European Magnetic Field Laboratory (EMFL).Supplementary informationSupplementary InformationPeer Review File | nature communications | [
"Article"
] | [
"Quantum Hall",
"Phase transitions and critical phenomena",
"Quantum fluids and solids"
] | strong magnetic field to electron gas confines motion in cyclotron orbits discrete Landau levels two-dimensional (2D systems quantization leads to gapped energy spectrum emergence quantum Hall effect (QHE lowest Landau level occupied quantum electron–electron interactions correlated states quantum Hall Landau level spectrum three-dimensional (3D) electron gas not fully gapped like one-dimensional system electrons move field direction destroys quantization Hall predicted generalized QHE can emerge in 3D electron systems periodically modulated quantum limit 3D electron systems form correlated electron states Luttinger liquids charge spin valley density waves excitonic insulators Hall Wigner crystals staging transitions anisotropic layered theoretically states related to quantum Hall physics three dimensions could manifest in Hall response in quantum limit 3D semimetals10 possibility 3D QHE explored in material systems signatures integer quantum Hall effect found in quasi-2D semiconducting multilayer lattices15 Bechgaards salts16 η-Mo4O1118 n-doped Bi2Te319 EuMnBi220 layered crystal structure supplies 2D systemsQHE observed in 3D graphite ZrTe521 HfTe5 samples22 graphite periodic superstructure attributed to standing electron waves In ZrTe5 HfTe5 IQHE believed from charge density wave scaling plateau height Fermi wavevector with thermodynamic measurements on ZrTe5 field-induced CDW transition proposed ZrTe5 weakly interacting Dirac 2DEGs plateau height scaling small carrier density Landau quantization Dirac search 3D QHE effort field-induced correlated states in quantum limit three-dimensional materials studies states longitudinal electrical resistivity of Bi24 ZrTe521 correlated states with Hall responses yet observed measurements low-temperature longitudinal Hall resistivities of 3D semimetals HfTe5 ZrTe5 HfTe5 isostructural counterpart of ZrTe531 materials share orthorhombic crystal structure single elliptical 3D Fermi surface less than 1% of Brillouin zone massive Dirac Fermions linearly dispersing bands Fermi level essential for 3D IQHE inprogress in ZrTe521 HfTe532 growth enabled Hall mobility 100,000 cm2 V−1 at low temperatures<4 K quality crystals comparable to samples for observing FQHE 3D band structure ZrTe5 HfTe5 hafnium higher atomic number zirconium introduces stronger spin–orbit coupling correlated states 2D HfTe5 promising candidate for unconventional Hall responses ZrTe5 grow as millimeter-long ribbons aspect ratio 1:3:10 crystalline anisotropy growth process crystal structure first transport characterization in refs. 21,32 measured longitudinal electrical resistivity ρxx Hall resistivity ρxy of three HfTe5 samples ZrTe5 function of magnetic field B temperature T electrical current along a-axis T = 300 K ρxx around 0.5 mΩ cm electron density n = 1.3 × 1019 cm−3 μ = 10,000 cm2 V−1 s−1 cooling in zero magnetic field ρxx increases T maximum at TL = 70 Kmaximum observed in HfTe532 ZrTe521 attributed to Lifshitz transition change charge-carrier type slope ρxy(B) changes at TL electron-type transport T < TL32 At 3 K n = 8.7 × 1016 cm−3 μ = 250,000 cm2V−1s−1 Supplementary Note 1 ref. samples show similar electrical transport properties data from HfTe5 A ZrTe5 D Additional data samples B C E F in Supplementary Information ref. 23.Fig. Fermi surface in HfTe5 Longitudinal electrical resistivity ρxx function temperature T at zero magnetic field measurement configuration x y z bias current I along x magnetic field B y voltage responses measured x- y ρxx function B at 3 K x y Variation longitudinal electrical resistivity function B at 3 K x f y g Landau-index fan diagram levels angles α B z–x plane function B−1 Data from minima ρxx extended data Fig. Shubnikov-de Haas frequency function of angle α j rotation angle of B z–y plane black dots represent measurement datared curve planar 2D blue ellipsoidal 3D Fermi surface HfTe5 momentum space kx ky kz direction Fermi-surface measured Shubnikov-de Haas (SdH) oscillations crystal axes at 3 K followed analysis ref. 21 rotated B z–y z–x planes ρxx (B) angles (Fig. 1 Fig. S3 ref SdH frequency BF,i related extremal cross-section Fermi surface SF B direction Onsager relation BF,i = SF,i(ħ/2πe). SdH oscillations magnetic crystallographic directions (x y z axes Fig. 1b–d maxima ρxx periodic 1/B onset Landau level ρxx(B) vanish dispersion z direction 3D systems Landau level-broadening Supplementary Note 4 Fig. S4–S7 ref. 23 SdH oscillation frequency subtracted high-temperature (50 K)—ρxx(B) low-T-data oscillating longitudinal resistivity Δρxx(B). standard Landau-index fan diagram analysis Δρxx(B) (Fig. 1h Figs. S3 S8 S9 Note 2found single oscillation frequency all rotation angles consistent single electron pocket Fermi energy extracted BF,i sample A B three directions are,x = (9.9 ± 0.1) T,y = (14.5 ± 0.5) T,z = (1.3 ± 0.1) T errors denote standard deviation fits contrast 2D HfTe5 ZrTe5 show in-plane SdH oscillations when B aligned with x y 3D Fermi-surface pocket shape FS determined by analysis rotation angle-dependence BF Fig. 1i, j angle-dependent SdH frequency represented by 3D ellipsoidal equation{amsmath-69pt\mathrm{F}} =^2 +}BF,3D=BF,zBF,i/(BF,zsinθ)2+(BF,icosθ)2 θ rotation angle z–i plane-check plot formula 2D cylindrical Fermi surface BF/cosθ deviates experimental data θ > 80°ellipsoid equations Fermi wave vectors[12pt]{minimal}{amsmath}{wasysym}{upgreek}\oddsidemargin}-69pt}{document}_{F,x} =\mathrm{F}},y}S\mathrm{F}},z}}{document}kF,x=SF,ySF,z/πSF,x = (0.005 ± 0.001) Å−1[12pt]{minimal}{amsmath}{wasysym}}}}}{upgreek}\oddsidemargin}{-69pt}{document}$$k_{{\mathrm{F}},y} =\sqrt {S_{{\mathrm{F}},x}S{F}},z}}}S_{{\mathrm{F}},y}}\end{document}kF,y=SF,xSF,z/πSF,y = (0.008 ± 0.001) Å−1[12pt]{minimal}{amsmath}{wasysym}{mathrsfs}-69pt\mathrm{F}},z,z=SF,xSF,y/πSF = (0.058 ± 0.006) Å−1 3D FS HfTe5 sample A x y z direction (Fig. 1k). errors kF,i errors,i analysis HfTe5 ZrTe5 samples quantum limit field z achieved BC = 1.8 1.2 T21,23 details band-structure analysis Supplementary Fig. S6 Table S1 refs. 21,23 materials ρxx(B) increases with magnetic field (Fig. 2b). increase ZrTe5 field-induced metal-insulator transition21.Fig. 2Three-dimensional integer quantum Hall effect HfTe5 Longitudinal electrical resistivity ρxx Hall resistivity ρxy function of B at T = 3 K B z blue onset Landau level blue numbers index N LL plateau ρxy scales with (h/e2) π/kF,z Planck constant h electron charge e Fermi wavevector z direction ρxx c ρxy function of B temperatures T ≥ 3 K Bρxy function |B|cos(β) magnetic fields angles β 3 K field-aligned z axis observed plateaus Hall resistance ρxy(B) minima SdH oscillations ρxx related QHE (Fig. 2a ref. 23). height last plateau (h/e2) π/kF,z 3D IQHE21,22 plateaus pronounced low temperatures visible up to T = 30 K (Fig. 2b c ref. observed quantization ρxy not obvious predicted quantization σxy Hall resistivity tensor ρxy = σxy/(σxxσyy + σxy2) magnetic field z direction σxx Hall conductivity tensor σxy = ρxy/(ρxx ρyy + ρxy2) magnetic field z direction low temperatures σxx < σxy Figs. S11 S12) σxy−1 ρxy observation quantization geometry HfTe5 crystals mechanical fragility reliable measurements ρyy possibleestimate error σxy using Drude resistivities[12pt{amsmath\oddsidemargin-69pt} ={xy}e^2t_x/m_x^^2t_y}ρyy/ρxx=(nxye2tx/mx*)/(nxye2ty/my*) quantum lifetimes effective masses from Shubnikov-de Haas oscillations on sample A nxy charge-carrier concentration in x–y plane ρyy/ρxx ≈ 0.4 error below 8 % in estimated σxy ρxx (B) < ρxy (B). errors within estimated error 10 %.Figure 2d shows angular-dependence Hall plateaus with rotation angle similar to sister compound ZrTe5.21 height Hall plateaus position depends on field component perpendicular to x–y plane B⊥ = |cosθ21,23 analysis focused on similarities Hall effects in ZrTe5 HfTe5 cooling samples to 50 mK difference emerges in Fig. 3 Supplementary Figs. S11–S15.low fields below quantum limit < compounds peaks plateaus in ρxx ρxy related to spin splitting Landau high fields > BC HfTe5 additional peak ρxx plateau-like feature ρxy contrast to ZrTe5 ρxx ρxy increase Landau-index fan diagram 3 K indexing Landau bands N = 1 band additional peak ρxx at N = 3/5 confirmed maxima ρxx σxx(B) three HfTe5 samples S11 S12). 3Low-temperature longitudinal magnetoresistivity Hall resistivity in HfTe5 ZrTe5 at 50 mK Longitudinal electrical resistivity ρxx Hall resistivity ρxy ZrTe5 function of B/BQL at T = 50 mK magnetic field B applied z 0 T ≤ B ≤ 3 T 0 T ≤ 9 T blue arrows mark onset Landau levels BQL magnetic field onset N = 1 Landau level blue numbers label index N Landau level red numbers value ρxyLongitudinal electrical resistivity ρxx left Hall conductivity σxy of ZrTe5 function of B at T = 50 mK B applied z for 0 T ≤ B ≤ 3 T (upper panel 0 T ≤ B ≤ 9 T (lower relation magnitude plateau-like feature Landau index not obvious from ρxy comparison conductivity magnitude feature quantum limit scales 3/5-times plateau LLL plateau height conductivity plateau-like features at N = 1 N = 3/5 developed conductivity within 1 2% of N·(e2/h)kF,z/π range 0.5 T around plateau center verify second pocket Fermi energy additional magneto-transport measurements up to 70 T additional quantum oscillations consistent with band-structure calculations31 previous ARPES study Fermi level (9 ± 2) meV agreement with ARPES experiment at 15 K nearest additional band ~5 meV above Fermi level function broadening K ≈ 1 meV Below 15 K Fermi level constant band edges temperature-independent Shubnikov-de Haas frequencynearest band ~kB·60 K from Fermi level contribute low-temperature transport experiments data analyzed single electron-type Dirac pocket origin N = 3/5 state quantum limit line shape Δρxx resembles σxx 2D QHE IQHE 2D systems longitudinal resistance ρxx(B) connected to ρxy(B) via ρxx(Bz) = γB·dρxy(B)/dB γ dimensionless parameter 0.01–0.05 measures local electron concentration Comparison of σxx(B) γB·dρxy(B)/dB show maxima minima at same field positions as Δρxx derivative relation fulfilled with γ = 0.04 expected range for 2DESs plateau-like feature quantum limit HfTe5 related to quantum Hall physics. 4Three-dimensional Hall response quantum limit HfTe5 Longitudinal electrical resistivity ρxx Hall resistivity ρxy function of B at T = 50 mK 0 T ≤ ≤ 3 T ≤ 9 T blue arrows mark onset Landau level blue numbers label index N Landau level red numbers value σxy (e2/h)kF,z/πlongitudinal electrical resistivity Δρxx function B γ B·dρxy/dB 0.04) Landau-index fan diagram B−1 T = 50 mK B z ρxx ρxy function B temperatures 4 K T 50 mK B z Hall effect HfTe5 estimated gap energies x–y plane ΔN N = 1 N = 2 N 3/5 quantum limit fitted T-dependence ρxx(B) minima. 4f thermally activated regime α exp kB Boltzmann constant integer N Δ1 = (40 ± 2) K N = 1 Δ2 (9 ± 1) K N = 2. gap energy N = 3/5 state quantum limit lower Δ3/5 = (0.49 ± 0.09) K deviations gaps errors thermally activated fits Fig S13 gaps integer N = 3/5 states compare gaps integer quantum Hall states 2DESs37,40 ΔN T-dependence Hall features (Fig. integer plateaus observable tens Kelvin plateau-like feature quantum limit vanishes 0.5 Kconsiderations suggest Hall feature in quantum limit HfTe5 associated with LLL Landau level non-degenerate HfTe5 like ZrTe5 gapped Dirac semimetal). finite value ρxx at N = 3/5 implies absence bulk gap quantized Hall effect 2D systems without kF,z-scaling expected emergence plateau-like feature in limit single-band system at low temperatures calls explanation beyond Hall conductivity of non-interacting single band decreases as 1/B in isostructural ZrTe5 (Fig possible scenarios for plateau in Hall resistance electron plasma include formation CDW Luttinger liquid Wigner crystallization Hall favorable scenario ZrTe5 HfTe5 interacting 2DEGs Hartree–Fock analysis layered structure gain in exchange energy exceed cost distributing electrons unequally between electrons undergo staging transitions i-th layers occupied other layers emptied depends on electron density state stabilized due to electron interaction spin–orbit Depending on layer separation electron density strength interactions layered Laughlin states or Halperin states states associated with Hall responsesstaging transitions unlikely in isotropic materials high electron densities HfTe5 anisotropic band structure small tunneling amplitudes hosts small electrons in Dirac pocket data consistent with strong interactions stabilizing correlated state Hall response in HfTe5 measurements reveal unconventional correlated electron state in Hall conductivity bulk semimetal HfTe5 adjacent 3D IQHE lower magnetic fields plateau-like feature Shubnikov-de Haas minimum longitudinal electrical resistivity magnitude 3/5(e2/h)kF,z/π derivative relations gap energies suggest related to quantum Hall physics unconventional feature in isostructural single-band ZrTe5 samples indicates correlated state stabilized by spin–orbit coupling further experimental efforts interactions field-induced correlated states HfTe5 necessary Hall response experiments density states space charge distribution Scanning Tunneling Spectroscopy in-field X-ray diffraction feature-crystal sample growth pre-characterizationSingle crystals HfTe5 obtained via chemical vapor transport amounts Hf Te sealed in quartz ampoule iodine (7 in two-zone furnace temperature gradient 400–500 °C applied1-month ribbon-shaped HfTe5 crystals extracted ampule typical size 1 mm × 0.5 mm 3 mm High-quality needle-shaped 0.1 × 0.3 20 mm3) crystals ZrTe5 synthesized tellurium flux method high-purity elements (99.9999% zirconium 99.9999% tellurium). lattice parameters confirmed-crystal X-ray diffraction samples same batch refs. 21,23,32,36 similar Fermi level positions HfTe5 ZrTe5 three-dimensional Dirac semimetal state emerges Tp 65 K resistivity large negative magnetoresistance critical insulator phases weak > strong < high temperatures extracted band gap 30 meV (185 low 10 meV (15 K Fermi level not gap valance band T > Tp conduction band T < Tp HfTe5 ZrTe5 samples metallic high low temperatures.Electrical transport contacts HfTe5 ZrTe5 crystals defined Al hard mask sputter etching sample surface deposition Ti Pt UHV system Pt wires glued to sputtered pads silver epoxyelectrical transport measurements ±9 T temperature-variable cryostat Dynacool Quantum dilution refrigerator-resistance four-terminal measurements longitudinal ρxx Hall resistivity ρxy measured Hall-bar geometry lock-in technique instruments Stanford Research SR 830) current 10 μA frequency f = 1 kHz 100 kΩ shunt resistor current a-axis crystal pulsed magnetic field experiments 70 T Dresden High Magnetic Field Laboratory HZDR European Magnetic Field Laboratory | 48.3 | 0.802256 |
10.1038/s41467-020-20244-7 | PMC7822848 | The primary stages of carrier relaxation in atomically thin transition metal dichalcogenides are hardly accessible due to their fast timescales. Here, the authors measure the first stages of carrier–phonon interaction in monolayer WSe2 via a series of periodic maxima in the hot photoluminescence intensity, assigned to phonon cascades. | Energy relaxation of photo-excited charge carriers is of significant fundamental interest and crucial for the performance of monolayer transition metal dichalcogenides in optoelectronics. The primary stages of carrier relaxation affect a plethora of subsequent physical mechanisms. Here we measure light scattering and emission in tungsten diselenide monolayers close to the laser excitation energy (down to ~0.6 meV). We reveal a series of periodic maxima in the hot photoluminescence intensity, stemming from energy states higher than the A-exciton state. We find a period ~15 meV for 7 peaks below (Stokes) and 5 peaks above (anti-Stokes) the laser excitation energy, with a strong temperature dependence. These are assigned to phonon cascades, whereby carriers undergo phonon-induced transitions between real states above the free-carrier gap with a probability of radiative recombination at each step. We infer that intermediate states in the conduction band at the Λ-valley of the Brillouin zone participate in the cascade process of tungsten diselenide monolayers. This provides a fundamental understanding of the first stages of carrier–phonon interaction, useful for optoelectronic applications of layered semiconductors. | IntroductionThe optical properties of group VI transition metal dichalcogenide monolayer (1L-TMD) semiconductors are dominated by excitons (bound electron-hole, e-h, pairs) with binding energies of hundreds of meV1, with spin and valley properties (such as valley-selective circular dichroism2) highly beneficial for optoelectronics2, valleytronics3 and spintronics3–12. Following optical excitation of a semiconductor above the band gap, the subsequent energy relaxation pathways play an important role in optics13–15 and charge carrier transport16,17. These processes are related to hot (i.e. not in thermal equilibrium) charge carriers and excitons1, and determine electron mobility18, optical absorption in indirect band gap semiconductors19, and intervalley scattering of hot electrons19. Photoluminescence (PL) and Raman scattering can be used to probe the interactions of carriers with phonons20. Different types of phonons with different energies can participate in the relaxation process of excited carriers. However, in some materials one type of phonon plays a dominant role and leads to high-order processes, e.g. up to nine longitudinal optical (LO) phonon replicas were reported in the hot PL of CdS and CdSe20–22. Multiphonon processes are important in defining the optoelectronic performance of ZnO23–26, GaN27 and bulk MoS228. The optical oscillator strength in 1L-TMDs, i.e. the probability of optical transitions between valence and conduction states, is higher than in III-V quantum wells19, resulting in short (~1ps29) exciton lifetimes. This favors hot PL emission, as excitons relax between several real states30,31. Examination of phonon-induced cascade-like relaxation processes in 1L-TMDs has been proposed for future pump-probe experiments32. However, observation of direct optical signatures in the early stages of carrier relaxation still remains a significant challenge, because of the ultrafast timescale (~100fs33) of these processes. Understanding the relaxation pathways in tungsten diselenide monolayers (1L-WSe2) is important for optoelectronic applications, such as photodetectors34 and lasers35, because it determines the recovery rate (i.e. the population of carriers relaxing to the ground state over time) and, as a result, the devices’ speed and efficiency.Here, we use ultra-low (~5 cm−1 ~0.6 meV) cut-off frequency (ULF) Raman spectroscopy to investigate the light scattered and emitted by 1L-WSe2 on SiO2, hBN and Au, as well as suspended 1L-WSe2. We observe phonon-assisted emission of hot PL, periodic in energy both in the Stokes (S) and anti-Stokes (AS) spectral range, and we extract a phonon energy ~15 meV. The S signal shows 7 maxima in the range of temperatures (T) from 78 to 295 K. We also detect up to 5 maxima in the AS signal ~75 meV above the laser excitation energy, increasing in intensity as T is raised. We assign these to phonon cascades36. We include finite T effects to compare S and AS signals and to understand carrier relaxation at room temperature (RT). By analyzing the T and excitation energy dependence, we conclude that a continuum of states (in the free-carrier gap) is involved in e-h relaxation in 1L-WSe2. Intermediate states in the conduction band around the Λ-valley of the Brillouin zone (BZ) participate in the cascade process. Hot PL so close in energy to the excitation laser gives access to the initial stages of carrier relaxation. These processes are ultrafast (e.g. ~100fs in GaAs33) and it is therefore challenging to access them in time-resolved experiments. Our approach can be extended to all layered materials (LMs) and their heterostructures (LMHs), as well as to other materials systems, such as perovskites37,38.Results1L-WSe2 flakes are exfoliated from bulk 2H-WSe2 crystals (2D Semiconductors) by micromechanical cleavage on Nitto Denko tape39, then exfoliated again on a polydimethylsiloxane (PDMS) stamp placed on a glass slide for inspection under optical microscope. Optical contrast is used to identify 1L prior to transfer40. Before transfer, 85 nm (for optimum contrast40) SiO2/Si substrates are wet cleaned41 (60s long ultrasonication in acetone and isopropanol) and subsequently exposed to oxygen-assisted plasma at 10W for 60s. The 1L-WSe2 flakes are then stamped on the substrate with a micro-manipulator at 40 °C, before increasing T up to 60 °C to release 1L-WSe242. The same procedure is followed for transfer of 1L-WSe2 on hBN, Au and Si, with 2 μm Au trenches made by lithography, to suspend the samples.The Raman and hot PL spectra are recorded in a back-reflection geometry with a ×50 objective (NA = 0.45) and a spot size ~1 μm. A liquid nitrogen cryostat (Linkam Scientific) placed on a XY translational stage is used to control T between 78 K and 295 K and excitation area. Imaging of the sample and monitoring of the excitation spot position are achieved using a set of beam splitters, aligned to a charge-coupled device (CCD) camera. The PL and Raman signals collected in the backward direction are filtered by 3 notch volume Bragg filters with a total optical density (OD) = 9. The cut-off frequency is ~5 cm−1 (~0.6 meV). The filtered signals are then focused on the spectrometer slit and dispersed by a 1800l/mm grating before being collected by the detector.A typical RT Raman spectrum for 1L-WSe2 on SiO2/Si measured at 532 nm is shown in Fig. 1a. The degenerate in-plane, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E^{\prime}$$\end{document}E′, and out-of-plane, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}^{\prime}$$\end{document}A1′, modes of 1L-WSe243 dominate the spectrum at ~−250 cm−1 (−31 meV) and ~+250 cm−1 (+31 meV) in the AS and S range, while weaker Raman peaks are also observed between 90 cm−1 (11 meV) and 500 cm−1 (62 meV) (see Methods and Supplementary Note 1) as discussed in refs. 44–46. Rescaling the intensity within the region marked in yellow in Fig. 1a reveals an underlying periodic pattern, Fig. 1b. Hereafter, for the energy scale we will use meV instead of cm−1. We fit all the peaks between −120 meV and +120 meV using Lorentzians, as shown in red in Fig. 1b. The fitting process is described in Methods. There are 7 S peaks and 5 AS at 295 K. The peak ~120 meV (~970 cm−1) originates from a combination of the Si substrate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Gamma }_{1},{\Gamma }_{12},{\Gamma }_{2{5}^{\prime}}$$\end{document}Γ1,Γ12,Γ25′ phonons47. Although the energy separation between two consecutive peaks is constant, the intensity decreases as a function of energy with respect to the excitation energy (here fixed at 0). To exclude other contributions, such as thin-film interference effects48, we measure 1L-WSe2 transferred on Au, placed on top of few-layer (FL) (~10 nm) hBN and also suspended, Fig. 1d (see Methods and Supplementary Note 1 for optical microscope images and PL characterisation). The intensity of the hot PL is comparable among the same steps of the cascade, and the position of the peaks is the same. Therefore, the cascade is linked to intrinsic relaxation mechanisms of 1L-WSe2, not to substrate-induced interference. Henceforth we will focus on 1L-WSe2 on SiO2/Si.Fig. 1Raman and hot PL spectra of 1L-WSe2.a Emission and scattering spectrum of 1L-WSe2 at 295K as a function of energy shift with respect to the excitation laser (532 nm~2.33eV). The degenerate in-plane (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E^{\prime}$$\end{document}E′) and out-of-plane (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}^{\prime}$$\end{document}A1′) Raman mode ~250 cm−1
43, as well as the Si Raman peak ~521 cm−1
47, are prominent in both S and AS. b Magnified portion of the spectrum in yellow in a. This reveals 7 periodic S peaks and 5 AS. Their intensity decreases as a function of the energy shift for both S and AS. c Raman spectra of 1L-WSe2 on SiO2/Si at 488, 514, 532, 633 nm and 295 K, shifted vertically for clarity. d Raman spectra of 1L-WSe2 on different substrates (Au, hBN and suspended) at 295K and 514 nm excitation. Black points: experimental data. Red lines: fitted cascades. Orange line: sum of fitted Lorentzians.To exclude the possibility that our laser is in resonance with a specific transition, we perform variable excitation wavelength experiments at 295K. Figure 1c plots the spectra measured at 488 nm (~2.54 eV), 514 nm (~2.41 eV), 532 nm (~2.33 eV) and 633 nm (~1.96 eV). We observe the same high-order features with identical energy separations in both S and AS. All these excitation energies lie above the free-carrier gap of 1L-WSe2 ~ 1.89 eV49–51. By comparing results for 1L-WSe2 on different substrates and for different excitation energies, we deduce that phonon-assisted hot PL is the dominant mechanism, whereas contributions from other excitations, such as plasmons52, are negligible, otherwise intensity and/or energy variations would be expected between Au and SiO2/Si, hBN, suspended cases.Figure 2a plots the energy offset with respect to the excitation laser (here 532 nm) of each emission feature as a function of the number of steps in the cascade at 295 K. Applying a linear fit, we extract ~15.42 ± 0.08 meV, regardless of substrate and excitation energy. This periodic modulation of the detected light intensity suggests that the scattering of photoexcited carriers is dominated by one prominent phonon mode. Since we excite above the free-carrier gap of 1L-WSe249, the intermediate states of the transitions are real. The e-h pair representation is in Fig. 3a.Fig. 2Energy separation and T dependence.a Emission energies as a function of number of steps in the cascade, extracted from the RT spectrum in Fig. 1b. The dashed black line is a linear fit, giving a step energy ~15.42 ± 0.08meV. b–d 532 nm Hot PL spectra of 1L-WSe2 at b 78 K, c 200 K, d 250 K.Fig. 3Comparison between experiments and theory.a Scheme of phonon-assisted hot PL. The incident, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hslash$$\end{document}ℏωi, and outgoing, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hslash$$\end{document}ℏωf, photons are shown by dotted magenta vertical arrows. The phonons participating in the cascade are indicated by the green arrows. The e–h pair dispersion curve is the blue parabola. The light cone is shown by red dashed lines. b Calculated S/AS spectra at different T. c IS/IAS for different numbers of cascade steps as a function of T. Filled circles are experimental data at 532 nm. The fit with Eq. (9) is indicated by dot-dashed lines. d Extended BZ of 1L-WSe2. Corresponding valleys are marked as Γ, K, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K^{\prime}$$\end{document}K′ and Λi, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Lambda }_{i}^{\prime}$$\end{document}Λi′ (i = 1, …, 3). e Experimental data at 160K (black line) compared to the calculated spectrum from Eq. (2) for no = 0.5. f Ratio of measured intensities of j = 1 to j = 2 peaks and corresponding fit with Eq. (10).The lattice T could affect the peaks intensity, as phonon occupation increases with T53,54. We thus perform T dependent measurements from 78 to 295 K, while keeping the excitation power constant ~26 μW. No emission of AS features is observed at 78 K, Fig. 2b, with the exception of two sharp lines ~−30 and ~−60 meV, originating from 1L-WSe2 and Si Raman modes, respectively. The hot PL peaks are seen at 200 K, Fig. 2c, and a further increase in intensity is observed at 250 K, Fig. 2d. Additional measurements at 120, 160, and 295 K are performed and used in the fits in Fig. 3c. Thermal effects are expected to modify the phonon energies55. However, in the 78–298 K range we do not observe any measurable shifts in the position of the hot PL peaks, because the shifts induced by acoustic phonons are smaller compared to our experimental error, as discussed in Methods.DiscussionAt low T (78 K), phonon absorption processes are suppressed because of the insufficient lattice thermal energy53. Optical excitation results in free e-h pair formation56,57 or virtual formation of an exciton with small in-plane wavevector (k ≲ ωi/c with ωi the excitation laser frequency)1. With the subsequent phonon emission, the e-h pair reaches a real final state (blue parabola in Fig. 3a), for which radiative recombination is forbidden by momentum conservation19. This triggers a cascade relaxation process36, whereby at each step a phonon is emitted (or absorbed for a T whereby the thermal energy is equal or higher than the phonon one energy)19. If the interaction with one phonon mode with energy ℏΩ dominates overall other inelastic scattering processes, the exciton loses energy by integer multiples of ℏΩ19,36. After emission of several (≥2) phonons, the exciton recombines and emits a photon with frequency ωf in a two-step process via an intermediate state with a small (k ≲ ωi/c) wavevector, for which radiative recombination is momentum allowed. Thus, we have secondary emission or scattering of light with S shift ωi − ωf = jΩ, where j = 2, 3, …., while j = ±1 are impossible as we scatter out of the light cone (i.e. the region of small wavevectors) with the first event. At finite T, in addition to phonon emission, absorption also comes into play, and AS emission is observed at ωf − ωi = jΩ.Multiphonon processes that do not involve real states require higher order exciton–phonon interactions58, and are therefore less probable. In contrast, the process in Fig. 1c is resonant, since excitation in the free-carrier gap means that all intermediate states are real. This allows us to describe the phonon emission cascade via the kinetic equation for the exciton distribution function f(ε), where ε is the exciton energy, as derived in Supplementary Notes 2, 3. Since the energy of the exciton changes in each scattering event by ±ℏΩ, the distribution function can be written as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f(\varepsilon )=\mathop{\sum }\limits_{j = -\infty }^{\infty }{f}_{j}\delta ({\varepsilon }_{0}-j\hslash \Omega )$$\end{document}f(ε)=∑j=−∞∞fjδ(ε0−jℏΩ)where ε0 is the excitation energy, δ(ε) is the Dirac δ-distribution (phonon dispersion and damping result in the broadening of the δ-distribution, as detailed in Supplementary Notes 2, 3), fj describes the peaks intensity. At steady state (partial derivative with respect to time equals zero) these obey a set of coupled equations describing the interplay of in- and out-scattering processes:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{l}\gamma {f}_{j}={\gamma }_{o}\left[{f}_{j-1}({n}_{o}+1)+{f}_{j+1}{n}_{o}\right]+g{\delta }_{j,0},\\ \quad\qquad \qquad \qquad j=\ldots ,-2,-1,0,1,2,\ldots .\end{array}$$\end{document}γfj=γofj−1(no+1)+fj+1no+gδj,0,j=…,−2,−1,0,1,2,….where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${n}_{o}={\left[\exp \left(\hslash \Omega /{k}_{B}T\right)-1\right]}^{-1}$$\end{document}no=expℏΩ/kBT−1−1 is the phonon mode occupancy at T, γo is the rate of the spontaneous phonon emission, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma ={\gamma }_{o}(2{n}_{o}+1)+\gamma ^{\prime}$$\end{document}γ=γo(2no+1)+γ′, is the total damping rate of the exciton, which includes recombination and inelastic scattering processes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma ^{\prime}$$\end{document}γ′. The last term in Eq. (2), gδj,0, describes the exciton generation at energy ε0, and is proportional to the exciton generation rate. Eq. (2) has the boundary conditions:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lim\limits_{j\to -\infty }{f}_{j}=0,\quad {f}_{K+1}=0,$$\end{document}limj→−∞fj=0,fK+1=0,where K is the maximum number of steps in the cascade:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K=\left\lfloor \frac{\hslash {\omega }_{i}-{E}_{1}}{\hslash \Omega }\right\rfloor ,$$\end{document}K=ℏωi−E1ℏΩ,with E1 the energy of the exciton band bottom. Eq. (2) is derived assuming γo and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma ^{\prime}$$\end{document}γ′ independent of ε. This assumption is needed to get an analytical solution of Eq. (2), but can be relaxed (see Supplementary Notes 2, 3).The general solution of Eq. (2) is:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{j}=\left\{\begin{array}{l}A{x}_{+}^{j},\quad j\;> \; 0,\\ B{x}_{+}^{j}+C{x}_{-}^{j},\quad j\le 0,\end{array}\right.$$\end{document}fj=Ax+j,j>0,Bx+j+Cx−j,j≤0,where6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{\pm }=\frac{\gamma \pm \sqrt{{\gamma }^{2}-4{n}_{o}({n}_{o}+1){\gamma }_{o}^{2}}}{2{\gamma }_{o}{n}_{o}},$$\end{document}x±=γ±γ2−4no(no+1)γo22γono,and x+ > 1 and x− < 1, A, B and C are the coefficients. For cascades with K ≫ 1 we can set B = 0 and:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{l}A=C=\frac{g}{\sqrt{{\gamma }_{o}^{2}+{\gamma }^{^{\prime} 2}+2{\gamma }_{o}\gamma ^{\prime} (1+2{n}_{o})}}.\end{array}$$\end{document}A=C=gγo2+γ′2+2γoγ′(1+2no).In this model, the spectrum of the scattered light consists of peaks with I ∝ fj, and scattering cross-section:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{l}\sigma ({\omega }_{i},{\omega }_{f})= \;{\sigma }_{0}({\omega }_{i},{\omega }_{f})\\ \; \times \mathop{\sum }\limits_{j = 2}^{\infty ^{\prime} }\frac{1}{\pi }\frac{2\Gamma }{4{\Gamma }^{2}+{(j\Omega -{\omega }_{i}+{\omega }_{f})}^{2}}{f}_{j}.\end{array}$$\end{document}σ(ωi,ωf)=σ0(ωi,ωf)×∑j=2∞′1π2Γ4Γ2+(jΩ−ωi+ωf)2fj.Here σ0(ωi, ωf) is a smooth function of frequency, Γ is the phonon damping. This description is valid for peaks with ∣j∣ > 1, the prime at the summation denotes that the terms with j = 0, ±1 are excluded. Accordingly, the peaks with Raman shift ±ℏΩ are suppressed. At no → 0 (limit of low T), x+ ≫ 1 and Ij with negative j (AS components) are negligible. At the same time, x− → (γo/γ) and the S peak intensities, IS, scale as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${({\gamma }_{o}/\gamma )}^{j}$$\end{document}(γo/γ)j. This scaling is natural for cascade processes19,59,60, since the probability of phonon emission relative to all other inelastic processes is given by γo/γ, thus IS decays in geometric progression. At finite T, the AS peaks appear with IAS proportional to the thermal occupation of the phonon modes. Thus, the S/AS intensity ratio, IS/IAS, with j steps in the cascade, can be written as:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{\rm{I}}}_{S}(j)}{{{\rm{I}}}_{AS}(j)}=\frac{{f}_{j}}{{f}_{-j}}={\left(1+\frac{1}{{n}_{o}}\right)}^{j},$$\end{document}IS(j)IAS(j)=fjf−j=1+1noj,and corresponds to the ratio of phonon emission and absorption rate to the power of j.The calculated I distribution and spectra at various T (corresponding to different no) are presented in Fig. 3b. Figure 3c plots IS/IAS as a function of T from Eq. (9). The experimental points collected from the fitted I of each step in the cascade at 532 nm excitation are displayed with circles. The absence of data at 78 K indicates no detection of IAS at this T. Applying Eq. (9) to the steps 2–5 in the cascade, with a phonon energy ~15.4 meV extracted from Fig. 2a, gives the dashed lines in Fig. 3c, in good agreement with experiments.Our model captures the main experimental observations well. The periodic pattern of hot PL intensity is reproduced by the calculations, Fig. 3b, and IS/IAS closely follows Eqs. (9), Fig. 3c. There is good agreement between our data and the calculated spectra from Eq. (2). An example for no = 0.5 at 160 K is in Fig. 3e. In our model, the peaks with j = ±1 are absent because N ≥ 2 phonons are needed for the first step of the cascade process, as for Fig. 3a. However, Fig. 1 shows that j = ±1 peaks are smaller than j = ±2 ones, but still detectable. We consider IS(1)/IS(2) as plotted in Fig. 3f. The possible mechanisms of j = 1 peak formation are as follows. (i) Elastic disorder or acoustic phonon-induced scattering, which provides a transfer between states within the light cone and states at the dispersion. (ii) Combination of phonon emission and absorption, where the j = 1 peak appears as a result of two phonon emission, followed by one phonon absorption. In (i) IS(1)/IS(2) does not depend on T. In (ii):10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{\rm{I}}}_{{\rm{S}}}(1)}{{{\rm{I}}}_{{\rm{S}}}(2)}=\exp \left(-\frac{\hslash \Omega }{{k}_{B}T}\right),$$\end{document}IS(1)IS(2)=exp−ℏΩkBT,strongly depends on T. This is indeed the case in our experiment, see Fig. 3f. This additional channel is also based on the interaction with the same phonon energy ~15 meV. Elastic processes could be the origin of a small offset between the experiment and the fitted curve.To get a better understanding of the relaxation pathways, we consider different scattering mechanisms. Scattering within the same valley is not plausible, due to the mismatch of BZ centre phonon energies61. The energy ~15 meV could correspond to either Γ − K or Γ − Λ phonons. The phonon dispersion in 1L-WSe2 shows acoustic phonons with energies ~15 meV46,61. These have a flat dispersion, necessary to observe the high number of oscillations we report, and are compatible with the model in Fig. 3a.Another option involves K-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K^{\prime}$$\end{document}K′ scattering of e (h) or, equivalently, Γ-K scattering of excitons. This would result in intensity oscillations as a function of the step in the cascade, due to the suppression of the process \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma \to K\to K^{\prime} \to \Gamma$$\end{document}Γ→K→K′→Γ compared to, Γ → K → Γ (see Supplementary Notes 4, 5). However, we do not observe intensity oscillations for different cascade steps in our spectra. As a result, we exclude this scenario. Therefore, the excitonic states in the Λ valleys play a role as intermediate states, Fig. 3d. The conduction band minima in these valleys are relatively close (~35 meV) to K, and play a crucial role in exciton formation and relaxation62–65. In this case, h remain in K (or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K^{\prime}$$\end{document}K′), but e scatter to any of the 6 available Λ valleys, and then scatter between these Λ valleys, before going back to K (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K^{\prime}$$\end{document}K′). This can be described taking into account all pathways, as:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{photon}}\to \Gamma \mathop{\longrightarrow }\limits^{\hslash \Omega }{\underbrace{{\Lambda }_{i}{\mathop{\longrightarrow }\limits^{\hslash \Omega }\ldots \mathop{\longrightarrow }\limits^{\hslash \Omega }{\Lambda }_{j}^{\prime}}}_{j}}\mathop{\longrightarrow }\limits^{\hslash \Omega }{\rm{photon}},$$\end{document}photon→Γ→ℏΩΛi→ℏΩ…→ℏΩΛj′⏟j→ℏΩphoton,with arbitrary number of steps j (both odd and even). The matrix elements of the processes are similar. We do not observe any noticeable periodic emission for 1L-MoS2 and 1L-WS2. This supports our interpretation, as the phonon scattering mechanism is linked to the particular bandstructure of 1L-WSe262.Similar oscillations can appear for free e and h36, see Supplementary Note 4. The basic description of the effect is similar to what we observe here, and our model can be extended to take into account e/h distribution functions. The spectra of scattered light and IS/IAS are similar to those calculated above. We cannot distinguish between exciton and the free-carrier cascades directly in our experiments. The excitonic description, however, seems straightforward due to enhanced (with respect to bulk materials) Coulomb effects in 1L-TMDs1.In conclusion, we investigated the light scattered and emitted by 1L-WSe2 excited above the free-carrier gap. We detected a periodic modulation of phonon-assisted hot PL with a period ~15 meV both in S and AS. We measured the S and AS intensity evolution from 78 to 295 K. We explained these high-order processes using a cascade model where electrons (holes) make successive transitions between real states with a finite probability of radiative recombination at each step. The electron states in the Λ valleys are intermediate states for efficient exciton relaxation. Our findings provide fundamental understanding of the initial steps of exciton relaxation in 1L-WSe2, and can be used to design optoelectronic devices based on this material. Our approach can be extended also to other layered materials and their heterostructures, as well as to perovskites.MethodsRaman and PL spectra fittingSupplementary Fig. 1a–c presents optical microscopy images of representative samples: (a) 1L-WSe2 on Au and suspended 1L-WSe2; (b) 1L-WSe2 on SiO2/Si; (c) 1L-WSe2 on hBN. Representative PL spectra collected 295 K at 514 nm excitation are in Supplementary Fig. 1d, showing a peak ~1.65 eV related to the A-exciton resonance1,12. Supplementary Fig. 2a shows representative data fits. The spectrum, at 295 K for 532 nm excitation, is shown with black dots. Blue lorentzian functions are used to fit the Raman peaks (FWHM ~ 1–10 cm−1). The residual spectral weight is also fitted with Lorentzians and results into the broader (FWHM ~ 50–80 cm−1) peaks of the hot PL (red). A flat baseline is taken into account for the whole energy scale, since the background in the S spectral range increases due to the higher intensity of the S cascades compared to the AS ones. A fit is shown in Supplementary Fig. 2b. The Lorentzians overlap, creating an asymmetric broad background (indicated by yellow dashed lines in Supplementary Fig. 2b).We now consider the thermally induced shift in the hot PL cascades of 1L-WSe2 in the 78–295 K range. We analyze Pos(E\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{\prime}$$\end{document}′, A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\,}_{1}^{\prime}$$\end{document}1′), as shown in the normalized intensity spectra in Supplementary Fig. 2c. Pos(E\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{\prime}$$\end{document}′, A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\,}_{1}^{\prime}$$\end{document}1′) red shift as a function of T, Supplementary Fig. 2d. Although the overall T dependence is not linear55, in the 78–295 K range we get:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=(-0.00755\pm 0.00083)\,{{\rm{cm}}}^{-1}{{\rm{K}}}^{-1},$$\end{document}k=(−0.00755±0.00083)cm−1K−1,where k is the slope. This corresponds to a shift of ~0.2 meV in the 78–295 K range. However, acoustic modes participate in the hot PL phonon cascades and their T dependence is weaker compared to optical phonons66. Thus ~0.2 meV is an upper limit of the expected shift of the hot PL cascades in this T range. The period ~15.42 ± 0.08 meV is quantified at 295 K by applying a linear fit in the position of the steps in the cascade in Fig. 2a, while the error bar corresponds to the standard error of the linear fit. This does not take into account other sources, such as the error in the dispersion of the grating, the fitting accuracy, etc, therefore the actual error bar is expected to be larger than 0.08 meV. Thus, although a shift of the order of less than one tenth of meV induced by acoustic phonons would be expected in this T range, it is very challenging to experimentally observe it in hot PL.Supplementary informationSupplementary InformationPeer Review File | nature communications | [
"Article"
] | [
"Two-dimensional materials",
"Electronic properties and materials"
] | optical properties group VI transition metal dichalcogenide monolayer (1L-TMD) semiconductors dominated by excitons electron pairs energies hundreds meV1 spin valley properties circular beneficial for optoelectronics2 valleytronics3 optical excitation above band gap energy relaxation pathways charge carrier related to hot charge carriers determine electron optical absorption band gap intervalley scattering of hot Photoluminescence) Raman scattering interactions carriers with Different phonons energies relaxation one phonon leads to high-order processes nine longitudinal optical phonon replicas in hot PL CdS Multiphonon processes optoelectronic performance of ZnO23–26 GaN27 MoS228 optical oscillator strength in 1L-TMDs states higher than III-V quantum wells19 short exciton lifetimes favors hot PL emission excitons relax between Examination phonon-induced cascade-like relaxation processes in 1L-TMDs proposed for future pump-probeobservation optical signatures early stages carrier relaxation challenge ultrafast timescale~100fs33) Understanding relaxation pathways tungsten diselenide monolayers (1L-WSe2) important for optoelectronic applications determines recovery rate carriers speed efficiency ultra-low~5 cm−1 ~0.6 meV) cut-off frequency) Raman spectroscopy light emitted 1L-WSe2 on SiO2 hBN Au suspended 1L-WSe2. observe phonon-assisted emission hot PL Stokes anti-Stokes spectral range extract phonon energy ~15 meV S signal shows 7 maxima 78 to 295 K 5 maxima AS signal ~75 meV above laser excitation energy increasing intensity T raised assign phonon include finite T effects compare S AS signals understand carrier relaxation at room temperature continuum of states involved in relaxation 1L-WSe2. Intermediate states conduction band Λ-valley Brillouin zone participate cascade process Hot PL close excitation laser initial stages carrier relaxation processes ultrafast ~100fs challenging access time-resolved experimentsapproach layered materials heterostructures systems-WSe2 flakes exfoliated from 2H-WSe2 crystals cleavage Nitto Denko exfoliated polydimethylsiloxane) stamp glass slide inspection optical microscope Optical contrast 1L 85 nm SiO2/Si substrates wet acetone isopropanol exposed oxygen-assisted plasma 10W 60s 1L-WSe2 flakes stamped substrate micro-manipulator 40 °C increasing T 60 °C release 1L-WSe242. procedure transfer 1L-WSe2 on hBN Au Si 2 μm Au trenches lithography Raman PL spectra recorded back-reflection geometry ×50 objective (NA = 0.45) spot size ~1 μm liquid nitrogen cryostat Scientific translational stage control T between 78 K 295 K excitation area Imaging monitoring excitation spot beam splitters camera PL Raman signals filtered by 3 notch Bragg filters optical density) = 9. cut-off frequency ~5 cm−1 (~0.6 filtered signals focused spectrometer slit dispersed by 1800l/mm grating before collected detectortypical RT Raman spectrum 1L-WSe2 on SiO2/Si measured at 532 nm Fig. 1a degenerate in-plane out-of-plane modes 1L-WSe243 dominate spectrum at ~−250 cm−1 (−31 meV) and ~+250 cm−1 (+31 meV) in AS and S range weaker Raman peaks between 90 cm−1 (11 meV) and 500 cm−1 (62 meV) Methods Supplementary Note 1). Rescaling intensity region marked yellow Fig. 1a reveals underlying periodic pattern Fig. 1b energy scale use meV instead of cm−1 fit peaks between −120 meV and +120 meV using Lorentzians shown in red Fig. 1b fitting process described in Methods. 7 S peaks 5 AS at 295 K.peak ~120 meV (~970 cm−1) from Si substrate[12pt{minimal{amsmath{wasysym{upgreek\oddsidemargin-69pt}{document\Gamma{1{5{document}Γ1,Γ12,Γ25′ energy separation between peaks constant intensity decreases excitation energy fixed at 0). exclude thin-film interference measure 1L-WSe2 transferred on Au few-layer (FL) (~10 nm) hBN suspended Fig. 1d Methods Supplementary Note 1 optical microscope images PL characterisation). intensity hot PL comparable cascade position peaks same cascade linked to intrinsic relaxation mechanisms 1L-WSe2 not substrate-induced interference focus on 1L-WSe2 SiO2/Si.Fig1Raman hot PL spectra 1L-WSe2.a Emission scattering spectrum at 295K energy shift excitation laser (532 nm~2.33eV). degenerate in-plane\documentclass[12pt{minimal}{amsmath out-of-plane[12pt{minimal-69pt Raman mode ~250 cm−1 43 Si Raman peak ~521 cm−1 47 prominent in S AS. Magnified portion spectrum in yellow reveals 7 periodic S peaks 5 AS intensity decreases energy shift S AS Raman spectra 1L-WSe2 on SiO2/Si at 488, 514 532 633 nm 295 K shifted vertically for clarity. Raman spectra-WSe2 on different substrates (Au suspended at 295K 514 nm excitation Black points experimental data Red lines: fitted cascades Orange line sum fitted Lorentzians.exclude laser transition variable excitation wavelength experiments at 295K Figure 1c plots spectra 488 nm~2.54 514 532 633~1.96 high-order features identical energy separations S AS excitation energies above free-carrier gap 1L-WSe2 ~ 1.89 comparing results substrates energies phonon-assisted hot PL dominant contributions other excitations negligible variations between Au SiO2/Si cases.Figure 2a plots energy offset excitation laser 532 nm each emission feature steps cascade at 295 K linear fit extract ~15.42 ± 0.08 meV substrate excitation modulation scattering photoexcited carriers dominated phonon mode excite above free-carrier gap 1L-WSe249 intermediate states transitions real e-h pair Fig. 3a. 2Energy separation T dependence Emission energies steps cascade spectrum Fig. 1b dashed black line linear fit step energy ~15.42 ± 0.08meV 532 nm Hot PL spectra 1L-WSe2 at 78 200 K 250 K.Fig. 3Comparison experiments theory Scheme phonon-assisted hot PLincident\documentclass[12pt]{minimal}\usepackage{amsmath{wasysym{amsfonts{mathrsfs{upgreek\oddsidemargin-69pt} outgoing\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts{amsbsy{mathrsfs}{upgreek}\oddsidemargin{-69pt} photons dotted magenta vertical arrows cascade green arrows e–h pair dispersion curve blue parabola light cone red dashed lines Calculated S/AS spectra at different T. IS/IAS different cascade steps function T Filled circles experimental data at 532 nm fit with Eq. (9) indicated dot-dashed linesExtended BZ 1L-WSe2. valleys marked Γ, K,[12pt]{minimal}\usepackage{amsmath{wasysym{upgreek-69pt}}K′ Λi{minimal-69pt}Λi′ (i = 1, ... 3) Experimental data at 160K (black line compared calculated spectrum Eq. (2) no = 0.5. Ratio measured intensities j = 1 to j = 2 peaks fit with Eq. (10) lattice T affect intensity phonon occupation increases with T53,54 T dependent measurements from 78 to 295 K excitation power constant ~26 μW No emission AS features at 78 K two sharp lines ~−30 and ~−60 meV from 1L-WSe2 Si Raman modes hot PL peaks at 200 K increase intensity at 250 K Additional measurements at 120, 160 295 K fits Fig. 3c.Thermal effects modify phonon energies55 78–298 K range measurable shifts position hot PL peaks shifts acoustic phonons smaller experimental error low T (78 phonon absorption suppressed insufficient lattice thermal Optical excitation results e-h pair virtual formation exciton small in-plane wavevector phonon emission e-h pair reaches final state (blue parabola Fig. radiative recombination forbidden by momentum conservation19 triggers cascade relaxation each phonon emitted absorbed thermal energy equal or higher than phonon interaction phonon mode ħΩ dominates exciton loses energy multiples ħΩ19 After emission (≥2) exciton recombines emits photon frequency ωf two-step process intermediate state small wavevector radiative recombination allowed secondary emission scattering light with S shift ωi − ωf = jΩ j = 2 3 j = ±1 impossible scatter out light cone first At finite T absorption AS emission observed at ωf − ωi = jΩMultiphonon processes states require higher exciton–phonon less probable process Fig. 1c resonant excitation in free-carrier gap intermediate states real phonon emission cascade kinetic equation exciton distribution function f(ε), ε exciton energy derived Supplementary Notes 2, 3. energy changes scattering event by ±ħΩ distribution function written\documentclass[12pt{minimal}{amsmath\oddsidemargin-69pt}\begin{document}\varepsilon=\mathop{\sum -\infty\delta-j\end{document}f(ε)=∑j=−∞∞fjδ(ε0−jħΩ ε0 excitation energy δ(ε) Dirac δ-distribution (phonon dispersion damping broadening δ-distribution Supplementary Notes 2, 3) fj describes peaks intensitysteady state derivative time equals zero obey coupled equations in out-scattering processes[12pt{minimal{amsmath\oddsidemargin-69pt}{document}{array}\gamma {f{j}=\gamma{o}\left{f}{j-1}{o}+1)+{f}_{j+1}\right]+g\delta{j,0}\qquad j=\ldots,-1,0,1,2,\ldots\end{array}}γfj=γofj−1(no+1)+fj+1no+gδj,0,j=...,−2,−1,0,1,2\documentclass[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek}\oddsidemargin}{-69pt}\begin{document}$${n}{o}\left[\exp\Omega\right)-1]\end{document}no=expħΩ/kBT−1−1 phonon mode occupancy T γo rate spontaneous phonon emission[12pt]{minimal}{amsmath}{wasysym}{upgreek}{\oddsidemargin}{-69pt}{document}$$\gamma ={o}(2{n}}+1)+\gamma\prime\end{document}γ=γo(2no+1)+γ′ total damping rate exciton recombination inelastic scattering processes[12pt]{minimal}{amsmath}{wasysym}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$$\gamma\prime\end{document}γ′ last term in Eq.(2) gδj,0 describes exciton generation at energy ε0 proportional to exciton generation rate Eq. (2) boundary conditions\documentclass[12pt]{minimal}\usepackage{amsmath\oddsidemargin-69pt}{document}\lim\limits{j -\infty{f}{j}=0{K+1}=0\end{document}limj→−∞fj=0,fK+1=0 K maximum steps cascade[12pt]{minimal}{amsmath{wasysym\oddsidemargin}{-69pt}{document}$$K=\left\lfloor\omega-{E}_{1}}\right\rfloor\end{document}K=ħωi−E1ħΩ E1 energy exciton band bottom. Eq.(2) derived assuming γo[12pt]{minimal\usepackage{amsmath\oddsidemargin}{-69pt}\begin{document}\gamma\end{document}γ′ independent ε assumption needed solution Eq. (2) relaxed Supplementary Notes 2, 3) general solution Eq. (2)\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{upgreek}{\oddsidemargin}{-69pt}\begin{document}${f}_{j}=\left\begin{array}{l}A{x}_{+}^{j} j 0{x}_{+}^{j}{x}_{-}^{j} 0\end{array}\right.\end{document}fj=Ax+j,j>0,Bx+j+Cx−j,j≤0[12pt]{minimal{amsmath{wasysym-69pt}\begin{document}${x}\pm=\frac{\gamma \gamma }^{2}-4{n}_{o}(}+1){\gamma{2\end{document}x±=γ±γ2−4no(no+1)γo22γono x+ > 1 x− < 1, A B C coefficients cascades with K ≫ 1 set B = 0:7[12pt]{minimal}{amsmath}{upgreek}}{-69pt}{document}{array}{l}A=C=\frac{g}{\sqrt{{\gamma{o}{2}+{\gamma\prime 2}+2{\gamma{o}\prime} (1+2{n}_{o}{array}}A=C=gγo2+γ′2+2γoγ′(1+2no).In spectrum scattered light peaks I fj scattering cross-section:8\documentclass[12pt]{minimal}\usepackage{amsmath\oddsidemargin{-69pt}\begin{document}{array}{l\sigma ({\omega_{i}_{f})={\sum\limits_{j = 2}\prime{1{\pi{2\Gamma+\Omega\omega{i_{f}{j}{array}}σ(ωi,ωf)=σ0(ωi,ωf)×∑j=2∞′1π2Γ4Γ2+(jΩ−ωi+ωf)2fj σ0(ωi ωf) function frequency Γ phonon damping valid for peaks with ∣j∣ > 1 prime terms j = 0, ±1 excluded peaks with Raman shift ±ħΩ suppressed.no → 0 low T), x+ ≫ 1 Ij negative j (AS components negligible x− → (γo/γ) S peak intensities scale\documentclass[12pt]{minimal}{amsmath\oddsidemargin{-69pt}{document}\gamma/\gamma{j\end{document}(γo/γ)j scaling natural cascade processes19,59,60 probability phonon emission given γo/γ IS decays geometric progression finite T AS peaks appear IAS proportional thermal occupation phonon modesS/AS intensity ratio IS/IAS j steps cascade written as[12pt{minimal{amsmath-69pt(j)IAS(j)=fjf−j=1+1noj corresponds ratio phonon emission absorption rate to power j calculated I distribution spectra at T no in Fig. 3b Figure 3c plots IS/IAS function of T from Eq. (9) experimental points step at 532 nm excitation displayed with circles data at 78 K no detection IAS T Eq. (9) to steps 2–5 phonon energy ~15.4 meV Fig. 2a dashed lines in Fig. 3c agreement experiments model captures experimental observations periodic pattern hot PL intensity reproduced calculations IS/IAS follows Eqs. (9) 3c good agreement between data calculated spectra from Eq. (2) example for no = 0.5 at 160 K Fig. 3e.model peaks with j = ±1 absent N ≥ 2 phonons needed for first step cascade process Fig. 3a. Fig. 1 shows j = ±1 peaks smaller than j = ±2 still detectable consider IS(1)/IS(2) in Fig. 3f possible mechanisms of j = 1 peak formation Elastic disorder or acoustic phonon-induced scattering transfer between states light cone dispersion Combination of phonon emission and absorption j = 1 peak two phonon emission one absorption IS(1)/IS(2) depend on T\documentclass{amsmath}IS(1)IS(2)=exp−ħΩkBT depends on T. in experiment see Fig. 3f additional channel based on interaction with same phonon energy ~15 meV Elastic processes could small offset between experiment fitted curve relaxation pathways consider different scattering mechanisms.Scattering within same valley plausible due to mismatch of BZ centre phonon energies61 energy ~15 meV could correspond to Γ − K or Γ − Λ phonons phonon dispersion in 1L-WSe2 shows acoustic phonons with energies ~15 meV46 flat dispersion high oscillations compatible with model Fig. 3a option involves K-\documentclass[12pt{amsmath-69pt}K′ scattering of e (h) or Γ-K scattering of excitons in intensity oscillations step cascade due to suppression process{amsmath-69pt}Γ→K→K′→Γ Γ → K → Γ (see Supplementary Notes 4, 5) observe intensity oscillations for different cascade steps in spectra exclude this scenario excitonic states in Λ valleys play as intermediate states, Fig. 3d.conduction band valleys close~35 meV K crucial exciton formation relaxation62–65.h remain in K[12pt]{minimal}{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}{-69pt}\begin{document}$$K^{\prime\end{document}K′), e scatter to 6 Λ valleys between Λ valleys back to K[12pt]{minimal}{amsmath{wasysym{upgreek\oddsidemargin}{-69pt}\begin{document}$$K^{\prime\end{document}K′). described pathways\documentclass[12pt]{minimal}{amsmath}{wasysym{amsfonts{amsbsy{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}\begin{document}$${\rm{photon}} \Gamma \mathop{\longrightarrow }\limits^{\hslash \Omega }\underbrace{{\Lambda }\longrightarrow\limits\hslash \Omega\longrightarrow\limits\hslash \Omega\Lambda }{j}\mathop\limits{photon}}}photon→Γ→ħΩΛi→ħΩ...→ħΩΛj′j→ħΩphoton arbitrary steps odd matrix elements similar noticeable periodic emission 1L-MoS2 1L-WS2. supports interpretation phonon scattering mechanism linked bandstructure 1L-WSe262.Similar oscillations free e h36 Supplementary Note 4. description effect similar model extended e/h distribution functions spectra light IS/IAS similar distinguish exciton free-carrier cascades excitonic description straightforward enhanced Coulomb effects 1L-TMDs1 investigated light emitted 1L-WSe2 above free-carrier gap detected periodic modulation phonon-assisted hot PL period ~15 meV S AS measured S AS intensity evolution 78 to 295 K explained processes cascade model electrons transitions states finite probability radiative recombination electron states Λ valleys intermediate states for exciton relaxation findings provide understanding initial steps exciton relaxation 1L-WSe2 design optoelectronic devices approach extended other layered materials heterostructures perovskitesPL spectra Fig. 1a–c microscopy images samples 1L-WSe2 Au 1L-WSe2 SiO2/Si hBN PL spectra 295 K 514 nm excitation Fig. 1d peak ~1.65 eV A-exciton Fig 2a data fits spectrum 295 K 532 nm excitation black dots Blue lorentzian functions fit Raman peaks 1–10 cm−1) residual spectral weight fitted Lorentzians broader 50–80 cm−1) peaks hot PL flat baseline energy scale background S spectral range increases higher intensity S cascades fit Fig. 2b Lorentzians overlap asymmetric broad background yellow dashed lines thermally induced shift hot PL cascades 1L-WSe2 78–295 K rangeanalyze Pos\documentclass[12pt]{minimal{amsmath{wasysym}{amsfonts}{upgreek}\oddsidemargin}{-69pt}{document}${\prime\end{document[12pt]{minimal}{amsmath}{wasysym}{amsfonts}}{amsbsy}{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}{1}{\prime\end{document}1′), normalized intensity spectra Supplementary Fig. 2c.\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs{upgreek}\oddsidemargin}{-69pt}{document}${\prime\end{document[12pt{minimal{amsmath{wasysym{mathrsfs{upgreek\oddsidemargin}-69pt}{document}\prime\end{document}1′ red shift function of T Supplementary Fig. 2d. T dependence not linear55 78–295 K range\documentclass[12pt]{minimal}{amsmath{wasysym{amsfonts{mathrsfs}{upgreek}{\oddsidemargin}{-69pt}{document}$k=(-0.00755\pm 0.00083)\rm{cm}}}{-1{K\end{document}k=(−0.00755±0.00083)cm−1K−1 k slope shift of ~0.2 meV in 78–295 K rangeacoustic modes participate hot PL phonon cascades T dependence weaker to optical ~0.2 meV upper limit expected shift hot PL cascades range period ~15.42 ± 0.08 meV quantified at 295 K linear fit steps cascade Fig. 2a error bar corresponds standard error linear fit other sources fitting accuracy actual error bar larger than 0.08 meV shift less than one tenth of meV acoustic phonons expected range challenging to observe in hot PL.Supplementary Review File | 47.1 | 0.820759 |
10.1038/s41467-020-16819-z | PMC7305203 | Necroptosis is a regulated form of inflammatory cell death driven by activated MLKL. Here, the authors identify a mutation in the brace region that confers constitutive activation, leading to lethal inflammation in homozygous mutant mice and providing insight into human mutations in this region. | MLKL is the essential effector of necroptosis, a form of programmed lytic cell death. We have isolated a mouse strain with a single missense mutation, MlklD139V, that alters the two-helix ‘brace’ that connects the killer four-helix bundle and regulatory pseudokinase domains. This confers constitutive, RIPK3 independent killing activity to MLKL. Homozygous mutant mice develop lethal postnatal inflammation of the salivary glands and mediastinum. The normal embryonic development of MlklD139V homozygotes until birth, and the absence of any overt phenotype in heterozygotes provides important in vivo precedent for the capacity of cells to clear activated MLKL. These observations offer an important insight into the potential disease-modulating roles of three common human MLKL polymorphisms that encode amino acid substitutions within or adjacent to the brace region. Compound heterozygosity of these variants is found at up to 12-fold the expected frequency in patients that suffer from a pediatric autoinflammatory disease, chronic recurrent multifocal osteomyelitis (CRMO). | IntroductionNecroptosis is a lytic form of programmed cell death associated with the production of pro-inflammatory cytokines, the destruction of biological membranes and the release of intracellular damage associated molecular patterns (DAMPs)1. Necroptosis depends on the activation of the mixed lineage kinase domain-like (MLKL) pseudokinase by receptor interacting protein kinase 3 (RIPK3)2–4. RIPK3-mediated phosphorylation of MLKL triggers a conformational change4,5 that facilitates the translocation to, and eventual irreversible disruption of, cellular membranes. While the precise biophysical mechanism of membrane disruption is still a matter of debate, common features of contemporary models are the formation of an MLKL oligomer and the direct association of the executioner four-helix bundle domain (4HB) of MLKL with biological membranes6–10. In mouse cells, the expression of the murine MLKL 4HB domain alone (residues 1–125), 4HB plus brace helices (1–180), or the expression of phosphomimetic or other single site pseudokinase domain (PsKD) mutants is sufficient to induce membrane translocation, oligomerization and membrane destruction4,9. While capable of disrupting synthetic liposomes when produced recombinantly, similarly truncated and equivalent single site (PsKD) mutant forms of human MLKL do not robustly induce membrane-associated oligomerization and cell death without forced dimerization11–13. Furthermore, both mouse and human MLKL mutants have been reported that have the capacity to form membrane-associated oligomers, but fail to cause irreversible membrane disruption and cell death9,13. Recent studies have revealed that necroptosis downstream of MLKL phosphorylation and membrane association can be modulated by processes that engage the endosomal sorting complex required for transport (ESCRT) family of proteins. One model proposes a role for ESCRT in limiting necroptosis via plasma membrane excision and repair14 while other models limit plasma membrane disruption by ESCRT-mediated release of phosphorylated MLKL in extracellular vesicles15–17 and/or the internalization of phosphorylated MLKL for lysosomal degradation17.In mice, the absence of MLKL does not appear to have obvious deleterious developmental or homeostatic effects4,18. However, genetic deletion of Fadd, Casp8 or Ripk1, leads to inappropriate activation of MLKL and ensuing necroptosis during embryogenesis that is incompatible with life beyond embryonic day (E)10.5, E10.5 and 1–3 days post-natally, respectively19–25. Exploring the precise physiological consequences of inappropriate MLKL activation in these scenarios is complicated by the fact that FADD, Caspase-8 and RIPK1 also play important roles in cellular processes other than modulation of MLKL-induced necroptotic cell death23,26–30.Aberrant levels of MLKL-dependent cell death contribute to disease in several genetic and experimental mouse models23,31–35. In humans, MLKL mRNA and protein levels are positively correlated with survival of patients with pancreatic adenocarcinoma, cervical-, gastric-, ovarian- and colon- cancers (reviewed by ref. 36). Interestingly, high levels of phosphorylated MLKL are associated with reduced survival in esophageal and colon cancer patients37. Two missense MLKL somatic mutations identified in human cancer tissue have been found to confer a reduction in necroptotic function in cell-based assays4,13. Very recently, siblings suffering from a novel neurodegenerative disorder were reported as homozygous for a rare haplotype involving a frameshift variant in MLKL, as well as an in-frame deletion of one amino acid in the adjacent fatty acid 2-hydroxylase (FA2H) gene38. The significant enrichment of an ultra-rare MLKL stop-gain gene variant p.Q48X has been reported in Hong Kong Chinese patients suffering from a form of Alzheimer’s disease39, however more common germline MLKL gene variants are only weakly associated with human disease in GWAS databases.We have identified a single base pair germline mutation of mouse Mlkl that encodes a missense substitution within the MLKL brace region and confers constitutive activation independent of upstream necroptotic stimuli. Given this mutant Mlkl allele is subject to the same developmental and environmental controls on gene expression as wild-type Mlkl, the postnatal lethality in these mice provides insight into the physiological and pathological consequences of dysregulated necroptosis. In parallel, these findings inform the potential functional significance of three common human MLKL polymorphisms that encode non-conservative amino acid substitutions within, or in close proximity to, the brace helix that is mutated in the MlklD139V mouse.ResultsGeneration of a constitutively active form of MLKLMpl−/− mice, owing to genetic deletion of the major receptor for thrombopoietin, have only 10% the wild-type number of peripheral platelets. An ENU mutagenesis screen was performed to identify mutations that ameliorate thrombocytopenia in Mpl−/− mice via thrombopoietin independent platelet production40. A G1 founder, designated Plt15, had a modestly elevated platelet count of 189 × 106 per mL compared with the mean for Mpl−/− animals (113 ± 57 × 106 per mL) and yielded 19 Mpl−/− progeny. Ten of these mice had platelet counts over 200 × 106 per mL, consistent with segregation of a dominantly acting mutation (Fig. 1a). Linkage analysis and sequencing identified an A to T transversion in Mlkl that was heterozygous in all mice with an elevated platelet count (Fig. 1b). The MlklPlt15 mutation results in a non-conservative aspartic acid-to-valine substitution at position 139 within the first brace helix. In the full-length mMLKL structure, D139 forms a salt bridge with an arginine residue at position 30 (α2 helix) of the MLKL four-helix bundle (4HB) domain4 (Fig. 1c). This salt bridge represents one of a series of electrostatic interactions between residues in helix α2 of the MLKL 4HB domain and the two-helix ‘brace’ region. D139 of mouse MLKL is conserved in all MLKL orthologues in vertebrata reported to date (Fig. 1d). We have shown that the exogenous expression of the 4HB domain of murine MLKL alone is sufficient to kill mouse fibroblasts whereas exogenous expression of full-length MLKL does not, suggesting an important role for this ‘electrostatic zipper’ in suppressing the killing activity of the MLKL 4HB9. To determine if MLKLD139V exhibited altered ability to induce necroptotic cell death relative to MLKLWT, we stably expressed these full-length proteins under the control of a doxycycline-inducible promoter in immortalized mouse dermal fibroblasts (MDF) isolated from Wt, Mlkl−/−, Ripk3−/− or Ripk3−/−;Casp8−/− mice. While expressed at comparable levels, MLKLD139V induced markedly more death than MLKLWt, on each of the genetic backgrounds tested (Fig. 1e–f, Supplementary Fig. 1a), and formed a high molecular weight complex observable by BN-PAGE in the absence of exogenous necroptotic stimuli (Supplementary Fig. 1b). This indicates that MLKLD139V is a constitutively active form of MLKL, capable of inducing necroptotic cell death independent of upstream signaling and phosphorylation by its activator RIPK3. Consistent with this interpretation, exogenous expression of MLKLD139V in Ripk3−/−;Casp8−/− MDFs was sufficient to induce the organelle swelling and plasma membrane rupture characteristic of TNF-induced necroptosis when examined by Transmission Electron Microscopy (Fig. 1g).Fig. 1Murine MLKLD139V is a constitutively active form of MLKL.a Platelet counts from Mpl−/− mice (open circles, n = 80, 60) and offspring from matings between Plt15 mice and Mpl−/− mice (closed orange circles, n = 19, 113) on a C57BL/6 or mixed C57BL/6:129/Sv background used for linkage analysis (Mixed N2). b A missense mutation (D139V) in the second exon of Mlkl was identified in Plt15 mutant mice. DNA sequence shown for wild type (top), a heterozygous mutant (middle), and a homozygous mutant (bottom). c Aspartate 139 contributes to an ‘electrostatic zipper’ joining brace helix 1 and the 4HB α2 helix of mouse MLKL (PDB code 4BTF)4. d Sequence logo of MLKL brace domain generated from multiple sequence alignment of all Vertebrata MLKL sequences (257) available on OrthoDB. e Mouse dermal fibroblasts (MDFs) of indicated genotypes were stably transduced with MlklWt and MlklD139V and expression induced with doxycycline (dox, white bars) or not induced (black bars) for 21 h. PI-positive cells were quantified by flow cytometry. Means ± SEM are plotted for n = 4–8 experiments (a combination of biological repeats and independent experiments) for each genotype with the exception of R3−/−C8−/− + MlklWt (n = 2, ±range). f Western blot analysis of whole cell lysates taken 6 h post doxycycline induction. g Transmission electron micrographs of MDFs stimulated as indicated. Images selected for (f) and (g) are representative of 2–3 independent analyses with similar results. TBZ; TNF + Birinapant + Z-VAD-fmk.MlklD139V causes a lethal perinatal inflammatory syndromeTo define the phenotypic consequences of constitutively active MLKL in the absence of any confounding effects resulting from Mpl-deficiency, all subsequent studies were performed on a Mpl+/+ background. Homozygous MlklD139V/D139V pups were born at expected Mendelian frequencies (Supplementary Table 1) and were ostensibly normal macroscopically and histologically at E19.5 (Supplementary Fig. 2a–d). However, by 3 days of age, although outwardly indistinguishable from littermates (Fig. 2a), they exhibited reduced body weight (Supplementary Fig. 2b) and failed to thrive, with a maximum observed lifespan of 6 days under conventional clean housing conditions. Like MlklWt/D139V mice, Mlklnull/D139V compound heterozygotes were present at the expected frequency at P21 and developed normally to adulthood (Supplementary Table 2). Thus, the constitutive activity of MLKLD139V was not affected by the presence of normal MLKL protein suggesting it is the absolute allelic dose of MlklD139V that determines perinatal lethality. To confirm that the phenotype of the ENU derived MlklD139V mice was due to the MlklD139V missense mutation, we independently generated MlklD139V mice using CRISPR-Cas9 genomic editing. Homozygote CRISPR-MlklD139V/D139V mice also died soon after birth (Supplementary Table 3).Fig. 2Homozygous MlklD139V neonates exhibit dispersed upper body inflammation.a Macroscopic appearance of MlklWt/Wt, MlklWt/D139V and MlklD139V/D139V mice at postnatal day 3. b Coronal section of mouth and neck region of postnatal day 2 litter mates stained with hematoxylin and eosin (H&E). Dilated blood vessels and edema are indicated by arrows. c Serial mandible sections from postnatal day 3 litter mates stained with H&E and anti-CD45. Inset black boxes are magnified in right panel. SL, sublingual gland. SM, submandibular gland. Images representative of n = 3–4 P3 pups per genotype. d H&E stained sections from mediastinum of postnatal day 2 litter mates. Thymic cortical thinning and pericardial infiltration are indicated by arrows. For full anatomical annotations for (b) and (d) see Supplementary Fig. 2h. (b) and (d) representative of n = 5–6 P2 pups examined with similar characteristics. Scale bars for (b–d) range from 50 to 1000 μm as indicated. Multiplex measurement of plasma cytokine levels at E19.5 (e) and postnatal day 3 (f). Each symbol represents one independent pup sampled; MlklWt/Wt – blue circles, MlklWt/D139V- red squares, MlklD139V/D139V- green triangles, with bar height and error bars representing mean ± SD respectively for n = 3–to 19 pups as indicated. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.005 calculated using an unpaired, two-tailed t-test.Hematoxylin-Eosin stained-sections from both P2 and P3 MlklD139V/D139V pups revealed multifocal acute inflammation characterized by neutrophilic infiltration, dilated blood vessels and edema (Fig. 2b) in the dermis and subcutis of the head and neck. These inflammatory features were not observed in MlklWt/Wt or MlklWt/D139V littermates, nor in Mlkl−/− mice of the same age (Supplementary Fig. 2i). Cells of hematopoietic origin, revealed by immunohistochemical staining for CD45, were sparsely distributed throughout the lower head and neck and confined predominantly to a clearly delineated developing lymph node in MlklWt/Wt and MlklWt/D139V littermates (Fig. 2c). In contrast, CD45+ cells were more numerous and distributed throughout the cutis, subcutis and salivary glands of MlklD139V/D139V pups (Fig. 2c). A mixture of diffuse and focal inflammatory infiltration was also observed within the mediastinum and pericardial space of all P2/P3 MlklD139V/D139V pups examined, as was a paucity of thymic cortical lymphocytes (Fig. 2d, Supplementary Fig. 2e), phenotypes not evident in E19.5 embryos (Supplementary Fig. 2d). No other consistent lesions were observed by histopathology. Consistent with this inflammatory phenotype, significantly elevated levels of several pro-inflammatory cytokines and chemokines were evident in the plasma of both E19.5 and P3 MlklD139V/D139V pups (Fig. 2e, f). Blood glucose levels were normal (Supplementary Fig. 2f, g).Hematopoietic defects in MlklD139V miceAlthough blood cell numbers were unchanged in MlklD139V/D139V pups at E19.5 relative to MlklWt/Wt and MlklWt/D139V littermates, by P3 significant deficits were evident in total white blood cell count (due predominantly to reductions in lymphocyte numbers) and platelet numbers (Fig. 3a–c, Supplementary Fig. 3a). Similarly, the numbers of hematopoietic stem and progenitor cells were present at normal proportions in fetal livers of E18.5 MlklD139V/D139V pups, although increased levels of intracellular ROS were uniformly evident in live cells, (Fig. 3d, e, Supplementary Fig. 3b,c). By P2, deficits in CD150+CD48+ and CD150+CD48− populations were present (Fig. 3f), accompanied by increased AnnexinV binding in live cells (which indicates phosphatidyl serine exposure) of all lineages (Fig. 3g). In adult MlklWt/D139V mice, numbers of hematopoietic stem and progenitor cells were unaffected (Fig. 3h); however, upon myelosuppressive irradiation, recovery of hematopoietic cell numbers was delayed and characterized by increased expression of ROS and Annexin V (Supplementary Fig. 3d, e). When challenged with the cytotoxic drug 5-fluorouracil (5-FU), blood cell recovery in MlklWt/D139V mice was similarly delayed (Fig. 3i). In competitive transplants in which test MlklWt/D139V or MlklWt/Wt marrow was co-injected with wild-type competitor marrow in 10:1 excess, as expected, MlklWt/Wt marrow contributed to 90% of recipient blood cells 8 weeks after transplantation and maintained that level of contribution for 6 months (Fig. 3j). In contrast, MlklWt/D139V marrow performed poorly, contributing to 25% and 51% of recipient blood cells at these times (Fig. 3j). Similarly, while wild-type fetal liver cells contributed to the vast majority of blood cells in irradiated recipients up to 6 months after transplantation, cells from MlklD139V/D139V embryos failed to compete effectively during this period (Fig. 3k). Heterozygote MlklWt/D139V fetal liver cells contributed poorly in the first month following the graft but recovered to contribute more after six months (Fig. 3k). Thus, while tolerated under steady-state conditions, heterozygosity of MlklD139V is deleterious under conditions of hematopoietic stress. Bone marrow- derived HSCs from MlklWt/D139V adults and fetal liver- derived HSCs from MlklWt/D139V and
MlklD139V/D139V pups also formed fewer and smaller colonies in the spleens of lethally irradiated recipient mice after 8 days (Supplementary Fig. 3f).Fig. 3Alterations in hematopoietic cells and defective emergency hematopoiesis in MlklD139V mice.a–c Absolute white blood cell (WBCB), lymphocyte and platelet numbers in peripheral blood of E19.5 and P3 pups, n = 6, 27, 44, 41, 10, and 11 as indicated. d Proportions of HSC (Lineage-Sca-1+c-kit+ (LSK) CD150+ CD48−), MPP (LSK CD150− CD48−), HPC-1 (LSK CD150− CD48+) and HPC-2 (LSK CD150+ CD48+)82, n = 5 per genotype and (e) relative levels of ROS (n = 4, 9, 5) (f) P2 bone marrow LSK populations (n = 9, 18, and 11) and (g) relative AnnexinV binding (n = 2, 11, 7). (h) HSC subtypes in adult bone marrow, n = 9 per genotype. a–h Each symbol represents one independent animal; MlklWt/Wt – blue circles, MlklWt/D139V- red squares, MlklD139V/D139V- green triangles, with bar height and error bars representing mean ± SD respectively, or range when n = 2. Red and white blood cells and platelets in MlklWt/Wt (blue circles) and MlklWt/D139V (red squares) mice after treatment with 150 mg per kg 5FU or saline. Means ± SEM from one experiment in which three mice were sampled at each time point for each treatment group, similar results were obtained in an independent cohort. j Bone marrow from MlklWt/Wt or MlklWt/D139V mice on CD45Ly5.2 background was mixed with wild-type CD45Ly5.1 competitor bone marrow and transplanted into irradiated CD45Ly5.1/Ly5.2 recipients. Peripheral blood mononuclear cells (PBMCs) quantified after 56 and 180 days. Mean ± SEM are shown (3 donors per genotype, 3–5 recipients per donor). k Fetal liver cells (CD45Ly5.2; MlklWt/Wt, MlklWt/D139V or MlkD139V/D139V) were transplanted into lethally irradiated recipients (CD45Ly5.1/Ly5.2) together with competitor bone marrow (CD45Ly5.1). Contribution to PBMCs 28 days and 180 days after transplantation. Mean ± SEM are shown (2–10 donors per genotype, 2–6 recipients per donor). Host contribution (CD45Ly5.1/Ly5.2) is depicted in gray, competitor (CD45Ly5.1) in white, and test (CD45Ly5.2) in black for (j) and (k). *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.005 calculated using an unpaired, two-tailed t-test.MlklD139V fibroblasts are less sensitive to necroptosisTo examine if the constitutive activity of exogenously expressed MLKLD139V results in an enhanced propensity for necroptosis in cells that express MLKLD139V under the control of its endogenous promoter, we immortalized MDFs from MlklWt/Wt, MlklWt/D139V and MlklD139V/D139V littermates and from Mlkl−/− E19.5 pups. We observed no significant differences in basal cell death levels, nor any differences in the sensitivity of these cells to an apoptotic stimulus such as TNF plus Smac mimetic (Fig. 4a, Supplementary Fig. 4a). Surprisingly and in apparent contradiction to our initial observations using exogenous expression systems, endogenous expression of this Mlkl mutant revealed a significant and consistent decrease in sensitivity to TNF-induced necroptosis using three different pan-caspase inhibitors Q-VD-OPh, zVAD-fmk and IDN-6556/emricasan in a MlklD139V dose-dependent manner (Fig. 4a, Supplementary Fig. 4a). MDFs isolated from MlklD139V/D139V homozygotes were up to 60% less sensitive to TNF-induced necroptosis compared with MlklWt/Wt MDFs, but were not as resistant as Mlkl−/− MDFs (Fig. 4a).Fig. 4MLKLD139V undergoes constitutive post-translation turn-over.MDFs were isolated from MlklWt/Wt, MlklWt/D139V, MlkD139V/D139V or Mlkl−/−
pups, immortalized and stimulated as indicated for 21 h for quantification of PI-positive cells using flow cytometry (n = 4, 4, 4, and 6) (a), or for 4 h for western blot analysis (b). Mlkl−/− MDFs were stably transduced with doxycycline-inducible FLAG-MLKLWT and FLAG-MLKLD139V constructs to examine MLKL protein stability after doxycycline withdrawal (c) and in the presence of indicated compounds (FLAG-MLKLD139V) (d). e Immortalized MDFs from (a) were stimulated as indicated for 21 h for quantification of PI-positive cells using flow cytometry (n = 2–3, 3–4, 4, 2–3). f E14.5 fetal liver cells from MlklWt/Wt, MlkD139V/D139V or Mlkl−/− embryos were plated in the presence of indicated dose of IFN-β and colonies enumerated after 7 days (n = 4–6). (a, e and f) represent mean ± SEM (A,E) or ±SD (f). b–e Representative images of at least three similar experiments. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.005 calculated using an unpaired, two-tailed t-test.While there were no obvious differences in the levels of MLKLWt and MLKLD139V protein following doxycycline induced exogenous expression (Fig. 1f), MLKL was virtually undetectable by Western blot in MlklD139V/D139V pup-derived fibroblasts immortalized and cultured ex vivo (Fig. 4b). There was, however, no significant reduction in Mlkl gene transcript levels in these cells (Supplementary Fig. 4b) suggesting that this reduction was post-transcriptional. A reduction in MLKLD139V protein levels was also evident in whole E14.5 embryo protein lysates and in single cell clones derived from HOXA9 factor dependent myeloid cell lines derived from MlklD139V/D139V E14.5 embryos (Supplementary Figs. 4c, j). Lysates from E14.5 embryos also clearly show that MlklWt/D139V heterozygotes have intermediate levels of MLKL, reflecting the intermediate sensitivity of MlklWt/D139V MDFs to necroptotic stimuli (Supplementary Fig. 4c and Fig. 4a).MLKLD139V protein turnover requires proteasome activityMeasuring the half-life of endogenously expressed MLKLD139V is not possible using conventional ‘pulse chase’ methods because this mutant protein induces necroptotic cell death, so we capitalized on our previous observation that an N-terminally FLAG-tagged MLKL 4HB forms a high molecular weight membrane-associated complex just like the untagged form, but, unlike the untagged version, does not kill cells9. Consistent with this observation, N-FLAG full-length mouse MLKLD139V did not induce cell death when inducibly expressed in Mlkl−/− MDFs (Supplementary Fig. 4f).Using this system, we were able to measure the cellular turn over of MLKL by inducing N-FLAG-MLKLWT or N-FLAG-MLKLD139V expression in Mlkl−/− MDFs for 15 h using doxycycline then washing and culturing them in the absence of doxycycline for a further 2–24 h. In the absence of a stimulus (UT), the levels of N-FLAG-MLKLWT remained consistent over the 24-h period (Fig. 4c), indicating that non-activated wild-type MLKL is a stable protein in MDFs. However, when these cells were treated with a necroptotic stimulus (TSI) to activate MLKL, the levels of wild-type MLKL rapidly declined even though these cells were unable to undergo a necroptotic cell death. Consistent with the fact that untagged MLKLD139V behaves as an auto-activated form of MLKL (Fig. 1e), the half-life of N-FLAG-MLKLD139V (4–6 h) was similar to the WT version stimulated with TSI (Fig. 4c). Thus, the absence of endogenously expressed MLKLD139V in E14.5 embryo lysates and cultured fibroblasts can be attributed to the reduced post-translational stability of this mutant auto-activated form of the protein.To determine which cellular mechanism(s) are required for the clearance of activated MLKLD139V, we included a series of proteasome, lysosome and specific protease inhibitors during the ‘chase’ period after doxycycline was withdrawn (schematic in Fig. 4d). The doses of all inhibitors were carefully titrated and combined with pan-caspase inhibitor IDN6556 to minimize any toxicity-associated apoptotic cell loss during the chase period. To exclude any confounding RIPK3-mediated activation of the necroptotic pathway by proteasome inhibitors41 (Supplementary Fig. 4f), the same experiment was also performed in Mlkl−/−, RIPK3−/− MDFs (Supplementary Fig. 4d). Even at the very low doses used, addition of the proteasome inhibitor PS341 was accompanied by reduced clearance of N-FLAG-MLKLD139V and the stabilization of higher molecular weight species that resemble mono- and poly-ubiquitinated MLKL (Fig. 4d, Supplementary Fig. 4d, i). This PS341 mediated protection of activated MLKL was also evident when the same assay was performed for phospho(p)S345-N-FLAG-MLKLWT (Supplementary Fig. 4e). The less potent proteasome inhibitor MG132 did not stabilize MLKLD139V to levels that could be resolved by western blotting of total MLKL in this assay but did facilitate some stabilization of (p)-N-FLAG-MLKLWT. Chloroquine, Bafilomycin and NH4Cl also partially protected against (p)-N-FLAG-MLKLWT clearance, supporting the potential role for lysosome mediated degradation of natively phosphorylated MLKLWT15,17, but this was not observed for constitutively activated N-FLAG-MLKLD139V using this approach (Fig. 4d, Supplementary Fig. 4d).Based on these findings we hypothesized that this MLKL-clearance mechanism limits the capacity of MLKLD139V to kill MlklWt/D139V and MlklD139V/D139V cells in culture and in vivo by maintaining protein levels below a critical threshold. To test whether this protective mechanism could be overwhelmed, we incubated MDFs with agents that have been shown to induce Mlkl expression (TNF, interferons (IFN) β and γ)42–44, or inhibit its turnover (proteasome and lysosome inhibitors). MLKLD139V protein in untreated MlklD139V/D139V MDFs was undetectable by Western blot but became faintly detectable following addition of these stimuli (Fig. 4b and Supplementary Fig. 4g). This correlates with moderate but statistically significant increases in cell death (particularly when compared with the lack of sensitivity to conventional necroptotic stimuli (Fig. 4a)), when exposed to IFN-β alone and in combination with proteasome or lysosome inhibitors (Fig. 4e). A similar allele-dose dependent sensitivity is also evident in primary MDFs (Supplementary Fig. 4h). To examine if this mechanism may explain the reduced capacity of MlklD139V/D139V fetal liver cells to reconstitute an irradiated host (Fig. 2k), ex vivo colony forming assays were performed on fetal liver cells derived from MlklWt/Wt and MlklD139V/D139V E14.5 littermates, alongside E14.5 livers taken from Mlkl−/− mice. MlklD139V/D139V cells showed significantly increased sensitivity to the inhibitory effects of IFN-β, with reduced colony formation at low doses of cytokine that affected MlklWt/Wt and Mlkl−/− colony formation only marginally (Fig. 4f). Factor dependent myeloid cells generated through HOXA9 immortalization of E14.5 liver HSCs also demonstrated high rates of cell death under conventional FDM culture conditions when derived from MlklWt/D139V or MlklD139V/D139V embryos (Supplementary Fig. 4k). Together, these experiments provide further evidence for the existence of steady-state MLKL surveillance and turnover mechanisms that suppress cell death by lowering the abundance of activated MLKL below a killer threshold at the cellular level6,14–16 and provide an in vivo precedent for both the existence of this phenomenon and the lethal consequences of its dysregulation in the form of the MlklD139V mouse.To test whether the lethal inflammation in MlklD139V/D139V neonates was mediated by the direct or indirect activation of the inflammasome by active MLKL we crossed this line with the Caspase 1/11 null mouse strain45–47. This did not enhance the lifespan of MlklD139V/D139V pups (Table 1). The combined genetic deletion of Casp8 and Ripk3 did not rescue or extend the life of MlklD139V/D139V mice, indicating that postnatal lethality is not mediated by bystander extrinsic apoptotic cell death that may occur secondary to initial waves of MLKLD139V-mediated necroptosis (Table 1). The genetic deletion of Tnfr1, Myd88 or Ifnar individually did not provide any extension to the lifespan of MlklD139V homozygote pups (Table 1). These data indicate that the removal of any one of these routes to NF-κB- and interferon-mediated gene upregulation, inflammation or apoptotic cell death is not sufficient to protect mouse pups against a double allelic dose of MlklD139V.Table 1Postnatal lethality in MlklD139V/D139V homozygotes is independent of Tnfr1, Myd88, Ripk3, Casp8, Casp1, and Casp11.Stage genotypedE14E18P21P21P21P21P21P21P21C57BL/6 genetic backgroundWtWtWtTnfr1−/−Myd88−/−Ripk3−/−, C8+/−Ripk3−/−, C8−/−Ifnar−/−C1−/−, C11−/−MlklWt/Wt58 (39)7 (9)15 (11)19 (15)3 (2)10 (6)2 (2)15 (11)10 (8)MlklWt/D139V70 (78)17 (18)30 (22)41 (30)6 (4)14 (12)5 (4)30 (22)21 (16)MlklD139V/D139V28 (39)13 (9)0 (11)0 (15)0 (2)0 (6)0 (2)0 (11)0 (8)Total # genotyped15637456092474531() - number of pups expected from Mendelian segregation, calculated from total number of pups genotyped and rounded to the nearest whole number. Gene names italicized.E, embryonic day; P, days postnatal.Common human missense MLKL variants map to the brace regionGiven the severe inflammatory phenotype of murine MlklD139V/D139V neonates and the significant defects in stress hematopoiesis observed in murine MlklWt/D139V adults, we explored the prevalence of brace region variation in human MLKL. Examination of the gnomAD database48, which contains human MLKL exome or genome sequence data from a total of over 140,000 individuals revealed that the second and third highest frequency human MLKL missense coding variants; rs34515646 (R146Q) and rs35589326 (S132P), alter the same brace helix (Table 2, Fig. 5a). The 4th most common human MLKL polymorphism, rs144526386 (G202*V) is a missense polymorphism identified exclusively in the context of a shorter splice isoform of MLKL (*) named ‘MLKL2’49 (Table 2, Fig. 5b). The full-length canonical transcript of MLKL encodes a 471 amino acid protein, while MLKL2 is an alternatively spliced isoform of MLKL that is 263 amino acids long. MLKL2 lacks a large portion of the pseudokinase domain which functions to repress the killing potential of the 4HB domain6–9 and recruit co-effectors like RIPK3 and HSP9013,50–52. Glycine202* is encoded by an extension to exon 9 that is unique to the MLKL2 splice isoform (Fig. 5a, b).Table 2Human MLKL brace helix polymorphism frequency human MLKL SNP.FeatureR146Q – rs34515646S132P – rs35589326G202aV – rs144526386CADD Score (phred-scaled)0.4076.3813.825UK Biobank – Total MAF (n)0.0253 (487,658)0.0161 (487,625)0.0147 (487,488)gnomAD – Total MAF (n)0.0152 (141,339)0.0138 (141,442)0.01228 (141,400)gnomAD – Highest MAF (n) population0.0252 (64,541) European (Non-Finnish)0.0311 (5185) Ashkenazi Jewish0.0245 (5184) Ashkenazi Jewish1000 genomes – Total MAF (n)0.0052 (2504)0.0088 (2504)0.0102 (2504)1000 genomes – Highest MAF (n) population0.018 (503) European0.024 (489) South Asian0.021 (503) EuropeanGene names italicized.N number of individuals sequenced, MAF Minor Allele Frequency – count.aAlternative transcript.Fig. 5Three of the four highest frequency missense human MLKL SNPs encode non-conservative amino acid substitutions within or adjacent to the brace helix region.a S132 and R146 (magenta) are located on either side of D140 (yellow—equivalent to mouse D139) in the first human MLKL brace helix. Alternate amino acids encoded by human polymorphisms indicated in parentheses. b G202 is predicted to be on an α helix unique to MLKL2 isoform and to form an interface along with S132 and R146. The mouse equivalent of human rs35589326 (hMLKLS132P), mMLKLS131P, spontaneously forms membrane-associated high molecular weight complexes following Blue Native (BN) PAGE (c) and kills MDFs (d) in the absence of extrinsic necroptotic stimuli when expressed in mouse dermal fibroblasts for 6 (c) and 21 hrs respectively (d). C; cytoplasmic fraction, M; crude membrane fraction, TSI; TNF, Smac-mimetic and IDN6556, Chlor: Chloroquine. c Representative of two independent experiments with similar results. Error bars in (d) indicate the mean ± SEM of 4–5 independent experiments. e Schematic showing brace helix variant combinations identified as alleles in trans in three CRMO patients. f MTRs are mapped onto the structure of MLKL to show regions that have low tolerance to missense variation in the human population (red) and regions that have increased tolerance to missense variation (blue), normalized to the gene’s MTR distribution. g Multiple sequence alignment (MSA) conservation scores are mapped onto the structure of MLKL to show regions that are highly conserved through evolution (red) and regions that are less conserved through evolution (blue).While the amino acid substitution MLKLR146Q is classified as ‘tolerated’ and ‘benign’ by SIFT/POLYPHEN 2 algorithms53,54 (Supplementary Table 1), R146 of human MLKL shows NMR chemical shift perturbations in the presence of the negatively charged IP3 and IP6 phospholipid head groups, indicating a possible role in membrane association and disruption11,55. Ser-132 lies before the first structured residue of the first brace helix in human MLKL (Fig. 5a)13,56,57. A Serine-to-Proline substitution at this position is predicted to significantly impact the conformation of the immediately adjacent W133 (brace helix) and in turn, the proximal W109 within the 4HB domain (Supplementary Fig. 5a). When mapped to a model of MLKL splice-isoform 249 Glycine 202* is predicted to be on an isoform 2-specific helix and to form an interface along with S132 and R146 of brace helix 1. While the precise structural consequence of these three brace polymorphisms is unknown, modeling of human MLKL predicts that disruption in the brace region favors adoption of an activated conformation13. Consistent with this prediction, the murine equivalent of the human S132P variant, mMLKLS131P, formed high molecular weight membrane-associated complexes and killed MDFs in the absence of a necroptotic stimulus (Fig. 5c, d) when expressed at close to endogenous levels (Supplementary Fig. 5b). Similarly to mMLKLD139V, unstimulated mouse dermal fibroblasts generated from the first generation of heterozygote and homozygote mutant pups of a recently generated mMlklS131P CRISPR modified mouse line demonstrated a clear reduction in MLKL protein levels relative to those prepared from wild-type littermates (Supplementary Fig. 5c), though the cellular clearance is not as complete as observed for mMLKLD139V. Together, these data indicate that constitutive activation and reduced protein stability is not a unique, idiosyncratic feature of the mMLKLD139V, but also a feature of a closely situated MLKL brace mutant, mMLKLS131P.MLKL brace variants occur in trans more frequently in CRMOTo investigate if human MLKL brace region polymorphisms play a role in human autoinflammatory disease we examined their frequency in cohorts suffering from ankylosing spondylitis (AS), chronic recurrent multifocal osteomyelitis (CRMO), Guillain Barré Syndrome (GBS) and Synovitis, Acne, Pustulosis, Hyperostosis and Osteitis (SAPHO) Syndrome. The individual minor allele frequencies of R146Q, S132P, and G*202V are not enriched in these disease cohorts relative to healthy controls when population distribution is accounted for (Supplementary Tables 4 and 5). However, these alleles occur in trans (making ‘compound heterozygotes’—schematic in Fig. 5e) in 3 out of 128 CRMO patients. This is 29 times the frequency that these combinations are observed in healthy NIH 1000 genomes samples (where there are only two compound heterozygotes for these polymorphisms out of 2504 healthy individuals sequenced), or at 10–12 times the frequency when only European CRMO patients and two separate healthy European control populations were compared (Table 3).Table 3Human MLKL brace helix compound heterozygotes in CRMO vs healthy controls.PopulationFrequency of relevant compound Hetsa in CRMOFrequency of relevant compound Hetsa in healthy controlsCRMO:Healthyb2 tailed P value (Fisher’s exact)2 tailed P value (chi square with Yate’s)Globalc0.023 (3/128)0.0008 (2/2504) NIH 1KG29:10.0010.0001Europeand0.02 (2/101)0.002 (1/503) NIH 1KG10:10.0740.1215Europeand0.02 (2/101)0.0017 (25/14,542) QUT controls12:1n/a0.0022aCombinations of R146Q – rs34515646, S132P – rs35589326 and G202*V – rs144526386 (Fig. 5e).bFrequency ratio rounded to nearest whole number.cCRMO patients and healthy controls of all ancestries included.dCRMO patients and healthy controls of only European descent included.DiscussionIn contrast to apoptosis, necroptosis is widely considered to be an inflammatory form of cell death. However, definitive evidence for this proposition has yet to emerge. Because MLKL is activated by inflammatory stimuli such as TNF it is very difficult to separate cause from effect. The serendipitous identification of an auto-activating mutant of MLKL (MlklD139V) in mice has allowed us to explore the consequences of inappropriate necroptosis in the absence of such confounding factors. Furthermore, it has led to significant insights into the critical adult hematopoietic and perinatal developmental processes that are most sensitive to excessive MLKL activation, and into physiological mechanisms that have evolved to neutralize activated MLKL.In the absence of a robust immunohistochemical marker for RIPK3-independent necroptosis, it is not possible to pinpoint exactly which cell type/s undergo necroptosis in MlklD139V mice. Nevertheless, the presence of high levels of circulating pro-inflammatory cytokines in MlklD139V/D139V pups at E19.5 relative to MlklWt/Wt and MlklWt/D139V littermates suggests that necroptosis and ensuing inflammation begins in the sterile in utero environment. This is not enough to overtly retard prenatal development or affect hematopoietic cell populations. However, upon birth and/or exposure to the outside environment the capacity of homozygous MlklD139V/D139V pups to suppress MLKLD139V activity is overwhelmed and they die within days of birth. This is clearly a dose-dependent effect because both MlklD139V/Wt and MlklD139V/null heterozygous mice are viable. Postnatal death cannot be prevented by combined deficiencies in Ripk3 and Casp8 nor by deficiency of other important inflammatory genes including Tnfr1, Myd88 or Ifnar. In light of the elevated levels of circulating G-CSF, IL-6 and IL-5 observed, the role of these key mediators in the initiation or potentiation of pre- and perinatal inflammation in MlklD139V/D139V pups will be the subject of future investigations.The MlklD139V mutation was initially identified for its capacity to moderately increase platelet production independent of the thrombopoietin receptor Mpl. While the mechanism underlying this observation remains unclear, it follows observations by others that another member of the necroptotic pathway, RIPK3, plays a role in platelet activation58. The reduced platelet levels observed in MlklD139V/D139V pups is unlikely to be the sole cause of death given much more severe thrombocytopenia is not lethal in Mpl−/− mice40. Difficulty with suckling due to inflammatory infiltration of the head and neck and resulting failure to thrive is one possible explanation for the lethality in MlklD139V/D139V pups. However, the narrow window of mortality for these pups and marked pericardial immune infiltration make heart failure another potential cause of sudden neonatal death.The MlklD139V mouse reveals that maintaining MLKL levels below a threshold can prevent necroptotic activation. This strain is a potential tool for the mechanistic and physiological examination of MLKL-mediated extracellular vesicle generation or other cell death-independent roles related to inflammation that is unconfounded by RIPK3 activation. While others have recently shown that an ESCRT dependent repair or extracellular vesicle extrusion can help protect membranes from limited MLKL damage14–16, and that p-MLKL can be internalized and degraded by the lysosome17 our data also suggest a role for the proteasome in the disposal of activated MLKL, be it directly, or in its capacity to generate free ubiquitin. This creates the possibility that these mechanisms or the previously described ESCRT mechanisms intersect in some way. Finally, the ability of these mechanisms to hold single gene-dose levels of active MLKL in check without deleterious consequences in vivo supports the idea that direct inhibition of activated MLKL may be an effective means to therapeutically prevent unwanted necroptotic cell death. Similarly, the MlklD139V mouse and assorted relevant crosses may prove to be a useful tool for the further examination of whether ROS production is co-incident with-, causative of- or consequential to- necroptotic plasma membrane disruption in varied tissue types and under highly physiologically relevant contexts (recently reviewed by59).While any mouse MLKL-human MLKL comparisons must be made cautiously in light of the species-specific structural and mechanistic differences5,12,13, it is notable that out of over 140,000 individuals surveyed, there is only one recorded case of a human carrying a substitution equivalent to the mMlklD139V mouse (hMLKLD140V; rs747627247) in the gnomAD database, and this individual is heterozygous for this variant. To our surprise, 3,841 individuals in gnomAD (55 of which are homozygotes) carry a very closely situated MLKL brace region variant –MLKLS132P. Our CRISPR-generated MlklS131P mouse equivalent supports the connection between constitutive MLKL activation and decreased MLKL protein stability. Preliminary observations show that this variant manifests in a much milder and context specific phenotype in mice than mMlklD139V, which is consistent with its high frequency presence in the human population.Overlaid with structural, biochemical, cell and animal-based evidence of function, it is tempting to speculate that these human MLKL brace region variants lead to altered MLKL function and/or regulation in what is most likely a highly tissue-, context- or even pathogen specific way60–62. While increased numbers and examination of independent cohorts will be required to confirm the statistical enrichment of human MLKL brace variants occurring in trans in the autoinflammatory disease CRMO, this patient cohort offers a tantalizing clue into their potential as modifiers of complex, polygenic inflammatory disease in present day humans.MethodsMiceAll mice were backcrossed to C57BL/6 mice for >10 generations or generated on a C57BL/6J background. Mlkl−/−, Tnfr1−/−, Myd88−/−, IFNAR1−/−, Ripk3−/−, Casp8−/−, and Casp1/Casp11−/−
mice were generated as described4,45,46,63–67. Mice designated as E19.5 were obtained by Caesarean section from mothers that received progesterone injections at E17.5 and E18.5. Independent mouse strains that carry the D139V or S131P mutation in the Mlkl gene (MLKLD139V CRISPR) were generated using CRISPR/Cas9 as previously described68. For D139V, one sgRNA of the sequence GGAAGATCGACAGGATGCAG (10 ng per μL), an oligo donor of the sequence ATTGGAATACCGTTTCAGATGTCAGCCAGCCAGCATCCTGGCAGCAGGAAGATCGACAGGTTGCAGAAGAAGACGGgtgagtctcccaaagactgggaaagagtaggccagggttgggggtagggtgg (10 ng per μL) and Cas9 mRNA (5 ng per μL) were injected into the cytosol of C57BL/6J zygotes. Mice were sequenced across the mutated region to confirm incorporation of the altered codon and analysis was performed after at least 2 back-crosses to C57BL/6. The same procedure was followed for the generation of MLKLS131P CRISPR mice, using sgRNA (CTGTCGATCTTCCTGCTGCC) and oligo donor (CTGTTGCTGCTGCTTCAGGTTTATCATTGGAATACCGTTTCAGATGTCAGCCAGCCAGCACCATGGCAGCAGGAAGATCGACAGGATGCAGAGGAAGACGGgtgagtctcccaaagactggga). Sex was not recorded for mice that were sampled at E19.5, P2 and P3. Experiments using adult mice were performed with a combination of both males and females between 8 and 12 weeks of age. Mice were housed in a temperature and humidity controlled specific pathogen free facility with a 12 h:12 h day night cycle. The WEHI Animal Ethics Committee approved all experiments in accordance with the NHMRC Australian code for the care and use of animals for scientific purposes.Linkage analysisWe mapped the chromosomal location of the Plt15 mutation by mating affected mice to 129/Sv Mpl−/− mice to produce N2 (backcross) and F2 (intercross) generations. A genome-wide scan using 20 N2 mice with the highest platelet counts (287 ± 74 × 106 per mL, compared with 133 ± 75 × 106 per mL for the overall population, Fig. 1a) localized the mutation to a region of chromosome 8 between D8Mit242 and D8Mit139 and linkage to this region was then refined. Analysis of the F2 population revealed a significant reduction in the frequency of mice homozygous for C57BL/6 alleles in this interval (e.g., D8Mit200 3/81 F2 mice homozygous C57BL/6, p = 2.2 × 10−5
χ2-test), suggesting the Plt15 mutation results in recessive lethality. The refined 2.01 Mb interval contained 31 annotated genes, only five of which appeared to be expressed both in the hematopoietic system and during embryogenesis (http://biogps.gnf.org/): Dead box proteins 19a and 19b (Ddx19a and Ddx19b), Ring finger and WD repeat domain 3 (Rfwd3), Mixed lineage kinase domain like (Mlkl), and WD40 repeat domain 59 (Wdr59). Sequencing identified a single mutation, an A to T transversion in Mlkl that was heterozygous in all mice with an elevated platelet count.ReagentsAntibodies; Rat-anti mRIPK3 and rat anti-mMLKL 8F6 (selected for affinity to residues 1–30 of mouse MLKL) and rat anti-MLKL 3H14 (MLKL brace region) were produced in-house. Anti-Pro Caspase 8 (#4927) and GAPDH (#2113) were purchased from Cell Signaling Technology. Anti-mouse MLKL pS345 (ab196436) and anti-Actin (ab5694) were purchased from Abcam. Anti-VDAC (AB10527) was purchased from Millipore. Fc-hTNF was produced in house and used at a final concentration of 100 ng per mL. Recombinant mouse IFN-γ and β were purchased from R&D Systems (Minneapolis, MN, USA) Q-VD-OPh and zVAD-fmk were purchased from MP Biomedicals (Seven Hills, NSW, Australia). Smac mimetic also known as Compound A, and the caspase inhibitor IDN-6556 were a gift from TetraLogic (Malvern, PA, USA). Propidium iodide, doxycycline, and bafilomycin were purchased from Sigma-Aldrich (Castle Hill, NSW, Australia).Cell line generation and culturePrimary mouse dermal fibroblasts were prepared from skin taken from the head and body of E19.5 pups delivered by C-section or from the tails of adult mice69. Primary MDFs were immortalized by stable lentiviral transduction with SV40 large T antigen. Immortalized MDFs were stably transduced with exogenous mouse MLKL cloned into the pFTRE 3 G vector, which was generated by Toru Okamoto, and allows doxycycline- inducible expression as described4. The following oligonucleotides were used for the assembly of constructs;mMlkl fwd; 5′-CGCGGATCCGCGCCACCatggataaattgggacagatcatcaag-3′,mMlkl rev; 5′-CGGAATTCttacaccttcttgtccgtggattc-3′,N-FLAG mMlkl fwd; 5′-CGCGGATCCAA gccacc atg gcg cgc cag gac-3′N-FLAG mMlkl rev; 5′-CGCGGATCC tta cac ctt ctt gtc cgt gga ttc-3′mMlkl D139V fwd; 5′-gaagatcgacaggTtgcagaggaagac-3′mMlkl D139V rev; 5′-gtcttcctctgcaAcctgtcgatcttc-3′mMlkl S131P fwd; 5′-gccagcctgcaCcctggcagcag-3′mMlkl S131P rev; 5′-ctgctgccaggGtgcaggctggc-3′Cells were maintained in culture as previously described44. 4-hydroxy-tamoxifen regulated HOXA9 immortalised Factor Dependent Myeloid cells were generated from mouse E14.5 fetal liver cells and cultured as described previously70.Cell death assaysFlow Cytometry based cell death assays were performed using 5 × 104 MDFs per well in 24 or 48-well tissue culture plates4. Doxycycline (20 ng per mL) was added together with death stimuli. Fc-hTNF was produced in house and used at 100 ng per mL, Compound A Smac mimetic and IDN6556 were used at 500 nM and 5 μM respectively. zVAD-fmk and QVD-OPh were used at 25 and 10 μM respectively. Mouse and human interferons γ and β were used at 30 ng per mL, PS341 and MG132 at 2 and 200 nM respectively and Bafilomycin at 300 nM. For Incucyte automated imaging, MDFs were plated at a density of 8 × 103 cells per well of a 96-well plate and permitted to attach for 3 h. FDMs were plated at a density of 5 and 10 × 103 cells per well of a 48-well plate. 0.2 μg per mL propidium iodide was in media alongside stimuli as indicated. Images were recorded at intervals of 1 and 2 h using an IncuCyte S3 and numbers of PI positive cells per mm2 at each time point quantified and plotted using IncuCyte S3 software.MLKL turn-over assays5 × 104 MDFs per well were plated in 24-well tissue culture plates and allowed to settle. Doxycycline (20 ng per mL) +/− TNF, Smac Mimetic and IDN6556 was added. After 15 h, ‘no dox’ and ‘0’ wells were harvested. Media was removed from remaining wells and cells were washed with PBS and fresh media containing IDN6556 was re-added. Wells were then harvested 2, 4, 6, 8, and 24 h from this point. Cells were harvested by direct lysis in reducing SDS-PAGE loading buffer.MLKL protection assays5 × 104 MDFs per well were plated in 24-well tissue culture plates and allowed to settle. Doxycycline (20 ng per mL) was added. After 18 hrs, ‘no dox’ and ‘T0’ samples were harvested. Media was removed and cells washed before addition of fresh media containing TSI or IDN alone for 3 h. Cells were washed again and media restored with IDN6556 alone (UT), or IDN6556 + inhibitor (MG132 (200 nM), PS341 (10–40 nM), Chloroquine (50 μM), Bafilomycin (300 nM), Ca-074 Me (20 μM), TLCK (100 μM) and AEBSF (100 μM)) for a further 21 h. Cells were harvested by direct lysis in reducing SDS-PAGE loading buffer.UBA pull downs2 × 106 MDFs stably transduced with doxycycline inducible N-FLAG-mMLKLWT or N-FLAG-mMLKLD139V expressing constructs were seeded and settled O/N before stimulation with 1 μg per mL doxycycline +/− TSI for 5 hrs. Cells were lysed in Urea-based UBA-pull down buffer, ubiquitylated proteins enriched and Usp21-treated as described previously71.Transmission electron microscopyMurine dermal fibroblasts prepared from mice of the indicated genotypes were untreated or stimulated with the indicated agents for the indicated hours. Then, cells were fixed with 2% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4, postfixed with 2% OsO4, dehydrated in ethanol, and embedded in Epok 812 (Okenshoji Co.). Ultrathin sections were cut with an ultramicrotome (ultracut N or UC6: Leica), stained with uranyl acetate and lead citrate, and examined with a JEOL JEM-1400 electron microscope. The viability of a portion of these cells was determined by measuring LDH release as described previously72.Mouse histopathologyCaesarian-sectioned E19.5 and Day P2/3 pups were euthanized by decapitation and fixed in 10% buffered formalin. Five-micrometers coronal sections were taken at 200-μm intervals for the full thickness of the head, 5-μm sagittal sections were taken at 300-μm intervals for the full thickness of the body. A thorough examination of these sections was performed by histopathologists Aira Nuguid and Tina Cardamome at the Australian Phenomics Network, Melbourne. Findings were confirmed by Veterinary Pathologist Prof. John W. Finney, SA Pathology, Adelaide and clinical Pathologist Prof. Catriona McLean, Alfred Hospital, Melbourne.Measurement of relative thymic cortical thicknessRepresentative images of thymus sections were analysed to determine relative cortical thickness using ImageJ. Briefly, medullary areas were identified on the basis of H and E staining and removed from the larger thymus structure using the Image J Image Calculator function to isolate the cortical region. The thickness of the cortical region, defined by the radius of the largest disk that can fit at a pixel position, was determined using the Local Thickness plugin in ImageJ (http://www.optinav.info/Local_Thickness.htm).ImmunohistochemistryFollowing terminal blood collection, P0 and P3 pups were fixed for at least 24 h in 10% buffered formalin and paraffin embedded before microtomy. Immunohistochemical detection of cleaved caspase 3 (Cell Signaling Technology #9661) and CD45 (BD) was performed as described previously23.Cytokine quantificationAll plasma was stored at −80 °C prior to cytokine analyses. Cytokines were measured by Bioplex Pro mouse cytokine 23-plex assay (Bio- Rad #M60009RDPD) according to manufacturer’s instructions. When samples were designated ‘<OOR’ (below reference range) for a particular cytokine, they were assigned the lowest value recorded for that cohort (as opposed to complete exclusion or inclusion as ‘zero’ which would artificially inflate or conflate group averages respectively). Values are plotted as fold change relative to the mean value for the Wt/Wt samples, and p values were calculated in Microsoft Excel using a two-tailed TTEST, assuming unequal variance. Data is only shown for cytokines that displayed statistically significant differences between genotypes at either of or both day E19.5 and day P3.Hematological analysisBlood was collected from P0 and P3 pups into EDTA coated tubes using heparinized glass capillary tubes from the neck cavity immediately after decapitation. After centrifugation at 500 G for 5 min, 5–15 μL of plasma was carefully removed and this volume was replaced with PBS. Blood cells were resuspended and diluted between 8–20-fold in DPBS for automated blood cell quantification using an ADVIA 2120 hematological analyzer within 6 h of harvest. Blood was collected from adult mice retro-orbitally into tubes containing EDTA and analyzed using an ADVIA120 automated hematological analyzer (Bayer).Transplantation studiesDonor bone marrow or fetal liver cells were injected intravenously into recipient C57BL/6-CD45Ly5.1/Ly5.2 mice following 11 Gy of gamma-irradiation split over two equal doses. Recipient mice received neomycin (2 mg per mL) in the drinking water for 4 weeks. Long term capacity of stem cells was assessed by flow cytometric analysis of donor contribution to recipient mouse peripheral blood and/or hematological organs up to 6 months following engraftment. Recovery from cytotoxic insult was assessed by automated peripheral blood analysis at regular times following treatment of mice with 150 mg per kg 5-fluorouracil (5-FU).Flow cytometryTo analyze the contribution of donor and competitor cells in transplanted recipients, blood cells were incubated with a combination of the following antibodies: Ly5.1-PE, Ly5.2-FITC, Ly5.2-biotin or Ly5.2 PerCPCy5.5 (antibodies from Becton Dickenson, Ca). If necessary, cells were incubated with a streptavidin PECy5.5 (BD), mixed with propidium iodide (Sigma) and analysed on a LSRI (BD Biosciences) flow cytometer. To analyse the stem- and progenitor cell compartment, bone marrow cells were incubated with biotinylated or Alexa700 conjugated antibodies against the lineage markers CD2, CD3, CD4, CD8, CD34, B220, CD19, Gr-1, and Ter-119. For stem and progenitor cell detection antibodies against cKit, Sca-1, CD48, AnnexinV, CD105, FcγRII/III or CD135 in different combinations (see antibody list for details). Finally FluoroGold (AAT Bioquest Cat#17514) was added for dead cell detection. Cells were then analysed on LSRII or Fortessa1 (BD Biosciences) flow cytometers.Reactive oxygen species (ROS) detectionROS was detected by using Chloromethyl-H2DCFDA dye according to the manufacturer’s instructions (Invitrogen Cat#C6827). In brief, bone marrow cells were loaded with 1 μM Chloromethyl-H2DCFDA for 30 min at 37 °C. Loading buffer was then removed, and cells were placed into 37 °C StemPro-34 serum free medium (ThermoFisher Cat#10639011) for a 15-min chase period. After incubation cells were placed on ice and stained with surface antibodies suitable for FACS analysis. Cells were analysed using a LSRII flow cytometer (Becton Dickinson).E14.5 fetal liver colony forming assays1 × 104 fetal liver cells were plated as 1 mL cultures in 35 mm Petri dishes in DMEM containing 10% FCS, 0.3% agar and 104 U per mL GM-CSF. IFN-β was added to the cultures in increasing concentrations from 0 to 30 ng per mL. Colony formation was scored after 7 days of incubation at 37 °C, fully humidified with 10% CO2.Quantitative PCRRNA was prepared using Trizol (Invitrogen) according to the manufacturer’s instructions and 10 μg was used for first strand cDNA synthesis using SuperScript II (Life Technologies). 0.5 μg of cDNA was then used in a TaqMan PCR reaction with Universal PCR mastermix and murine Mlkl (Mm1244222_n1) and GAPDH (Mm99999915_m1) Taqman probes (ThermoFisher) on an ABI 7900 Fast Real-Time PCR instrument (Applied Biosystems). Mlkl expression relative to GAPDH control was determined using SDS version 2.3 program (Applied Biosystems) and expressed as ΔCT values.Statistics (mouse and cell-based assays)Please consult figure legends for description of error bars used. All data points signify independent experimental repeats, and/or biologically independent repeats. All p values were calculated in Microsoft Excel or Prism using an unpaired, two-tailed t-test, assuming unequal variance and not adjusted for multiple comparison. Asterices signify that p ≤ 0.05 (*), p ≤ 0.01(**) or p ≤ 0.005 (***). All comparisons were made between MlkWt/Wt and MlkD139V/D139V groups only (with the exception of data derived from adult mice, which were comparisons between MlkWt/Wt and MlkWt/D139V groups only.Whole-exome sequencingDNA from CRMO probands and their family members (when available) was purified from saliva or blood and prepared for whole-exome sequencing (WES). The samples underwent WES at several different times, enriched using the Agilent SureSelect Human All Exon V4, V5 or V6 + UTR (Agilent Technologies) before sequencing at either Otogenetics, Inc (Atlanta, GA), Beckman Coulter Genomics (Danvers, MA), or at the University of Iowa Genomics Core (Iowa City, IA). The fastq files were quality-checked and processed to vcf format as described73. Variants for all samples were called together using GATK’s Haplotype Caller74 and were recalibrated and hard-filtered in GATK as described73. Variants were annotated with minor allele frequencies (MAFs) from 1000 genomes75, ExAC and gnomAD48 and with information regarding the effect of each variant using SNPSift/SNPEff76. The databases used for annotation were dbNSFP2.977 (for MAFs) and GRCh37.75 for protein effect prediction.Ancestry determinationAncestry was determined for each CRMO proband using the LASER software package78. A vcf file including ten probands at a time was uploaded to the LASER server and the TRACE analysis was selected using the Worldwide panel. For probands with indeterminate ancestry using the Worldwide panel, the European and Asian panels were used. Principal component values for each proband were plotted using R Statistical Software and the code provided in the LASER package.MLKL variant quantification1000 Genomes: Vcf files from 1000 genomes were annotated and filtered as described previously79. Values for MLKL variants rs35589326 (S132P), rs34515646 (R146Q), and rs144526386 (G202V) as well as all MLKL coding variants were queried and tabulated for allele and genotype count for participants of all ancestry (n = 2504), and for those of European ancestry (n = 503). Compound heterozygous variants were evident due to the phasing of all variants in the 1000 genomes dataset. CRMO: Allele and genotype counts for all MLKL coding variants were tabulated in probands of European ancestry (n = 101) and for all probands (n = 128). Compound heterozygous variants were identified using parental sequence data. AS: DNA from all subjects in AS cohort were genotyped using the Illumina CoreExome chip following standard protocols at the Australian Translational Genomics Centre, Princess Alexandra Hospital, Brisbane. Bead intensity data was processed and normalized for each sample and genotypes called using the Illumina Genome Studio software. All the samples listed in the table have passed quality control process80. GB: Genotyping was performed in an ISO15189-accredited clinical genomics facility, Australian Translational Genomics Centre (ATGC), Queensland University of Technology. All samples were genotyped by Illumina HumanOmniExpress (OmniExpress) BeadChip81. QUT controls: A collection of healthy control data of verified European ancestry from various cohort studies, complied by the Translational Genomics Group, QUT and typed on an Illumina CoreExome microarray. Includes data from The UK Household Longitudinal Study, led by the Institute for Social and Economic Research at the University of Essex and funded by the Economic and Social Research Council. The survey was conducted by NatCen and the genome-wide scan data were analysed and deposited by the Wellcome Trust Sanger Institute. University of Essex. Institute for Social and Economic Research, N. S. R., Kantar Public. Understanding Society: Waves 1–8, 2009–2017 and Harmonised BHPS: Waves 1–18, 1991–2009. [data collection]. 11th Edition. UK Data Service., (2018).Patient recruitmentAll genomic data was derived from patients recruited with consent as described previously48,80,81, and with the approval of human ethics review boards of all Institutes that participated in human genetics studies; University of Iowa Carver College of Medicine, Queensland University of Technology, Australian National University, Shanghai Renji Hospital, JiaoTong University of Shanghai, The Hospital for Sick Children and the University of Toronto, University of Sydney, Australian Institute of Sport, University of Freiburg, Princess Alexandra Hospital, Memorial Hermann Texas Medical Centre, The University of Queensland, Oregon Health and Science University), Groupe Française d’Etude Génétique des Spondylarthrites (GFEGS) and the University of Oxford.Statistical analysis (human data)Statistical comparisons were performed at the level of allele frequency or the level of compound heterozygote sample frequency using either a Fisher’s exact test or a Chi-Squared test with Yates correction as specified under each table. Compound heterozygous variants were quantified and compared at the individual rather than the allelic level, where individuals with and without qualifying variants were compared at the allelic level.Web resourcesgnomAD – https://gnomad.broadinstitute.org/
http://asia.ensembl.org
OrthoDB - https://www.orthodb.orgCADD - https://cadd.gs.washington.edu/Clustal Omega - https://www.ebi.ac.uk/Tools/msa/clustalo/WEBLOGO - https://weblogo.berkeley.edu/logo.cgiMissense Tolerance Ratio (MTR) Gene Viewer - http://biosig.unimelb.edu.au/mtr-viewerUK biobank - https://www.ukbiobank.ac.ukUnderstanding Society - https://www.understandingsociety.ac.uk/Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Reporting Summary | nature communications | [
"Article"
] | [
"Necroptosis",
"Inflammatory diseases",
"Haematopoietic stem cells"
] | IntroductionNecroptosis is programmed cell death production pro-inflammatory cytokines destruction membranes release intracellular damage patterns (DAMPs depends on activation mixed lineage kinase (MLKL pseudokinase by protein kinase 3 phosphorylation MLKL triggers conformational translocation to irreversible disruption cellular membranes mechanism membrane disruption features formation MLKL oligomer association executioner four-helix bundle domain (4HB) MLKL with biological In mouse cells expression murine MLKL 4HB domain phosphomimetic single site pseudokinase domain (PsKD) mutants induce membrane translocation oligomerization destruction4,9 synthetic liposomes single mutant forms human MLKL induce membrane oligomerization cell death without forced dimerization11–13 mouse human MLKL mutants form membrane oligomers cause irreversible membrane disruption cell death9 studies necroptosis MLKL phosphorylation membrane association modulated by endosomal sorting complex transport (ESCRT) family proteinsmodel proposes ESCRT necroptosis via plasma membrane excision other models limit disruption by ESCRT release phosphorylated MLKL extracellular internalization for lysosomal mice absence MLKL deleterious genetic deletion of Fadd Casp8 Ripk1 leads to activation MLKL necroptosis during embryogenesis incompatible life beyond embryonic day)10.5 E10.5 1–3 days post-natally physiological consequences inappropriate MLKL activation complicated FADD Caspase-8 RIPK1 play roles processes MLKL-dependent cell death contribute disease in mouse humans MLKL mRNA protein levels correlated with survival pancreatic adenocarcinoma cervical gastric ovarian colon- cancers high levels phosphorylated MLKL reduced survival in esophageal colon cancer MLKL mutations in human cancer tissue reduction necroptotic function siblings neurodegenerative disorder reported homozygous for rare haplotype frameshift variant MLKL-frame deletion of acid fatty acid 2-hydroxylase)enrichment MLKL gene variant p.Q48X reported Hong Kong Chinese patients Alzheimer’s common MLKL variants weakly associated human disease GWAS databases identified single base pair germline mutation mouse Mlkl encodes missense substitution MLKL brace region constitutive activation independent necroptotic stimuli mutant allele subject developmental environmental controls wild-type Mlkl postnatal lethality physiological pathological consequences dysregulated necroptosis findings inform functional significance three human MLKL polymorphisms non amino acid substitutions brace helix mutated MlklD139V mouse active form MLKLMpl−/− mice genetic deletion 10% wild-type peripheral platelets ENU mutagenesis screen mutations thrombocytopenia G1 founder Plt15 elevated platelet count 189 × 106 per mL yielded 19 Mpl−/− progeny Ten mice platelet counts over 200 × 106 per mL dominantly acting mutation Linkage analysis sequencing identified A to T transversion Mlkl heterozygous mice elevated platelet countMlklPlt15 mutation aspartic acid-to-valine substitution at position 139 first brace helix full-length mMLKL structure D139 forms salt bridge with arginine residue at position 30 (α2 helix) MLKL four-helix bundle (4HB) (Fig. salt bridge represents electrostatic interactions between residues helix α2 MLKL 4HB two-helix ‘brace’ region D139 mouse MLKL conserved in all MLKL orthologues vertebrata exogenous expression 4HB domain murine MLKL mouse fibroblasts exogenous expression full-length MLKL not ‘electrostatic zipper’ suppressing killing activity MLKL 4HB9 MLKLD139V necroptotic cell death expressed full-length proteins doxycycline-inducible promoter in immortalized mouse dermal fibroblasts MLKLD139V induced more death than MLKLWt formed high molecular weight complex observable-PAGE absence exogenous necroptotic stimuli MLKLD139V active form MLKL necroptotic cell death independent signaling activator RIPK3expression MLKLD139V Ripk3−;Casp8− MDFs organelle swelling plasma membrane rupture TNF-induced necroptosis Transmission Electron Microscopy (Fig. MLKLD139V active form Platelet counts Mpl−/− mice 80 60 offspring matings Plt15 Mpl−/− C57BL/6 C57BL/6:129/Sv background missense mutation (D139V) second exon Mlkl identified Plt15 mutant mice DNA sequence wild heterozygous homozygous Aspartate 139 contributes ‘electrostatic zipper’ brace helix 1 4HB α2 helix mouse MLKL Sequence logo MLKL brace domain alignment Vertebrata MLKL sequences Mouse dermal fibroblasts (MDFs transduced with MlklWt MlklD139V induced doxycycline 21 h PI-positive cells quantified flow cytometry Means ± SEM plotted 4–8 experiments each genotype exception R3−/−C8− + MlklWt 2 Western blot analysis whole cell lysates 6 h post doxycycline inductionTransmission electron micrographs MDFs stimulated Images (f) (g) 2–3 analyses similar results TBZ TNF Birinapant Z-VAD-fmk.MlklD139V causes lethal perinatal inflammatory consequences active MLKL studies Mpl+/+ background Homozygous MlklD139V/D139V pups born Mendelian frequencies normal macroscopically histologically E19.5 3 days reduced body weight failed maximum lifespan 6 days housing Mlklnull/D139V heterozygotes present frequency P21 developed normally adulthood constitutive activity MLKLD139V not affected normal MLKL protein absolute allelic dose MlklD139V determines perinatal lethality confirm MlklD139V mice due MlklD139V mutation generated MlklD139V mice CRISPR-Cas9 genomic editing Homozygote CRISPR-MlklD139V/D139V mice died after birth 2Homozygous MlklD139V neonates exhibit dispersed upper body inflammationMacroscopic MlklWt/D139V mice postnatal day 3. Coronal section neck 2 stained hematoxylin eosin Dilated blood vessels edema mandible sections H&E anti-CD45 black boxes magnified SL sublingual gland SM submandibular gland 3–4 P3 pups per genotype H&E sections mediastinum Thymic cortical thinning pericardial infiltration arrows anatomical annotations Supplementary Fig. 2h 5–6 P2 pups Scale bars 50 to 1000 μm plasma cytokine levels E19.5 postnatal day 3 symbol MlklWt/Wt blue circles/D139V red squares green triangles bar height error bars mean ± SD 3–to 19 pups *p ≤ 0.05 ≤ 0.01 ***p ≤ 0.005 unpaired two-tailed t-test-Eosin stained-sections P2 MlklD139V pups multifocal acute inflammation neutrophilic infiltration dilated blood vessels edema 2b dermis subcutis head neckinflammatory features not observed in MlklWt/Wt/D139V littermates Mlkl− mice age Cells hematopoietic origin CD45 sparsely distributed lower head neck confined lymph node in MlklWt/Wt/D139V littermates CD45+ cells numerous distributed cutis subcutis salivary glands MlklD139V/D139V pups diffuse focal inflammatory infiltration mediastinum pericardial P2/P3 MlklD139V pups paucity thymic cortical lymphocytes not in E19.5 embryos No other consistent lesions observed elevated levels pro cytokines chemokines in plasma E19.5 P3 MlklD139V pups 2e Blood glucose levels normal 2f g).Hematopoietic defects in MlklD139V blood cell numbers unchanged in MlklD139V/D139V pups E19.5 deficits in white blood cell count reductions lymphocyte numbers platelet numbers (Fig. 3a–chematopoietic stem progenitor cells normal in fetal livers E18.5 MlklD139V pups increased intracellular ROS evident live cells (Fig. 3d 3b deficits in CD150+CD48++CD48− populations. increased AnnexinV binding live cells exposure. 3g). adult MlklWt/D139V mice hematopoietic stem progenitor cells unaffected (Fig. myelosuppressive irradiation recovery delayed increased ROS Annexin V cytotoxic drug 5-fluorouracil blood cell recovery MlklWt/D139V mice delayed (Fig. competitive transplants MlklWt/D139V marrow co-injected wild-type competitor marrow marrow contributed 90% blood cells 8 weeks after transplantation 6 months MlklWt/D139V marrow poorly 25% 51% blood cells wild-type fetal liver cells contributed majority blood cells irradiated recipients 6 months after transplantation cells MlklD139V/D139V embryos failed competeHeterozygote MlklWt/D139V fetal cells contributed poorly first month recovered more after six months tolerated steady-state heterozygosity MlklD139V deleterious hematopoietic stress Bone marrow HSCs/D139V adults fetal liver pups formed fewer smaller colonies spleens irradiated mice after 8 days 3Alterations hematopoietic cells defective emergency hematopoiesis MlklD139V mice white blood cell lymphocyte platelet numbers peripheral blood E19.5 P3 pups 6 27, 44 41 10 11 Proportions HSC MPP HPC-1 HPC-2 5 per genotype relative levels ROS 4 9 P2 bone marrow LSK populations 9 18 AnnexinV binding 2 11 HSC subtypes adult bone marrow 9 per genotypesymbol represents animal MlklWt/Wt blue circles/D139V red squares green triangles bar height error bars mean ± SD 2. Red white blood cells platelets MlklWt/D139V mice after treatment 150 mg per kg 5FU saline SEM similar results cohort Bone marrow MlklWt/Wt mixed with wild CD45Ly5.1 bone marrow transplanted irradiated CD45Ly5.1/Ly5.2 recipients Peripheral blood mononuclear cells quantified after 56 180 days Mean ± SEM (3 donors per genotype 3–5 recipients per Fetal liver cells (CD45Ly5.2 transplanted into irradiated recipients/Ly5.2 competitor bone marrow Contribution PBMCs 28 180 days after transplantation Mean ± SEM (2–10 donors per genotype 2–6 recipients per Host contribution gray competitor white test black *p ≤ 0.05 0.01 0.005 unpaired two-tailed t-testMlklD139V fibroblasts less sensitive necroptosisTo activity MLKLD139V necroptosis cells immortalized MDFs from MlklWt/Wt/D139V littermates Mlkl−/− E19.5 pups observed no significant differences basal cell death levels apoptotic stimulus TNF (Fig. 4a endogenous expression Mlkl mutant revealed decrease TNF-induced necroptosis using pan-caspase inhibitors Q-VD-OPh zVAD-fmk IDN-6556/emricasan MlklD139V dose-dependent MDFs isolated from MlklD139V homozygotes 60% less sensitive TNF-induced necroptosis MlklWt/Wt MDFs not as resistant as Mlkl−/− MDFs (Fig.. 4MLKLD139V post-translation turn-overMDFs isolated from MlklWt/Wt/D139V/− pups immortalized stimulated 21 h for PI-positive cells cytometry 4 h western blot analysis MDFs transduced with doxycycline FLAG-MLKLWT-MLKLD139V constructs MLKL protein stability after doxycycline withdrawal compounds Immortalized MDFs stimulated 21 h PI-positive cells = 2–3 3–4 E14.5 fetal liver cells from MlklWt/Wt/− embryos plated IFN-β colonies enumerated after 7 days (n = 4–6) represent mean ± SEM ±SD three similar experiments *p ≤ 0.05 ≤ 0.01 ≤ 0.005 two-tailed t-test no differences in MLKLWt MLKLD139V protein following doxycycline exogenous expression MLKL undetectable by Western blot in MlklD139V pup-derived fibroblasts immortalized cultured ex vivo no significant reduction in Mlkl gene transcript levels post-transcriptionalreduction MLKLD139V protein levels in E14.5 embryo protein lysates single cell clones from HOXA9 factor myeloid cell lines MlklD139V E14.5 embryos Figs 4c Lysates E14.5 embryos show MlklWt/D139V heterozygotes intermediate levels MLKL sensitivity to necroptotic stimuli Fig. 4c 4a).MLKLD139V protein turnover requires proteasome activityMeasuring half-life of expressed MLKLD139V protein induces necroptotic cell death N-terminally FLAG-tagged MLKL 4HB forms high complex kill N-FLAG MLKLD139V induce cell death expressed in Mlkl−/− MDFs Fig. cellular turn over MLKL inducing N-FLAG-MLKLWT-MLKLD139V expression in Mlkl−/− MDFs for 15 h doxycycline 2–24 h levels N-FLAG-MLKLWT consistent 24-h (Fig. non-activated-type MLKL stable in MDFscells treated with necroptotic stimulus (TSI) MLKL levels wild-type MLKL declined necroptotic death untagged MLKLD139V auto-activated form MLKL half-life N-FLAG-MLKLD139V (4–6 h) similar to WT version stimulated TSI (Fig. of endogenously expressed MLKLD139V in E14.5 embryo lysates cultured fibroblasts reduced post-translational stability mutant auto-activated form clearance activated MLKLD139V included proteasome lysosome specific protease inhibitors during period after doxycycline withdrawn Fig. doses inhibitors titrated combined with pan-caspase inhibitor IDN6556 minimize apoptotic cell loss activation necroptotic experiment in Mlkl−/− RIPK3−/− MDFs low doses proteasome inhibitor PS341 reduced clearance of N-FLAG-MLKLD139V stabilization of higher molecular weight species mono- poly-ubiquitinated MLKL PS341 protection of activated MLKL evident for phospho(p)S345-N-FLAG-MLKLWTless potent proteasome inhibitor MG132 stabilize MLKLD139V levels western blotting stabilization-N-FLAG-MLKLWT Chloroquine Bafilomycin NH4Cl protected against-N-FLAG-MLKLWT clearance potential lysosome degradation of phosphorylated MLKLWT15 not observed for activated N-FLAG-MLKLD139V (Fig. 4d hypothesized MLKL-clearance mechanism limits capacity MLKLD139V to kill/D139V cells culture vivo maintaining protein levels below critical threshold incubated MDFs with agents induce Mlkl expression (TNF β inhibit turnover lysosome MLKLD139V protein in untreated MDFs undetectable by Western blot faintly detectable following addition stimuli 4b correlates with moderate increases in cell death exposed to IFN-β alone combination with proteasome lysosome inhibitors similar allele-dose dependent sensitivity in primary MDFs Fig. examine mechanism reduced capacity of MlklD139V/D139V fetal liver cells to reconstitute irradiated hostex vivo colony assays on fetal liver cells from MlklWt/Wt MlklD139V/D139V E14.5 littermates livers Mlkl−/− mice MlklD139V/D139V cells increased sensitivity to IFN-β reduced colony formation at low doses cytokine (Fig. Factor dependent myeloid cells HOXA9 immortalization E14.5 liver high rates cell death FDM culture MlklWt/D139V embryos experiments evidence steady-state MLKL surveillance turnover mechanisms cell death activated MLKL below killer threshold vivo precedent for existence phenomenon lethal consequences dysregulation MlklD139V mouse lethal inflammation in MlklD139V/D139V neonates MLKL crossed line with Caspase 1/11 null mouse enhance lifespan of MlklD139V pups genetic deletion of Casp8 Ripk3 extend life MlklD139V mice postnatal lethality not mediated by apoptotic cell death MLKLD139V-mediated necroptosisgenetic deletion Tnfr1 Myd88 Ifnar extension lifespan MlklD139V homozygote pups removal routes NF-κB interferon gene upregulation apoptotic cell death pups against double allelic dose MlklD139V 1Postnatal lethality MlklD139V homozygotes independent of Tnfr1 Myd88 Ripk3 Casp8 Casp1 Casp11 genetic C8 C11−/−MlklWt/Wt58 (39)7 (9)15 (11)19 (15)3 (8)MlklWt/D139V70 (78)17 (18)30 (22)41 (30)6 (4)14 (22)21 (16)MlklD139V (39)13 genotyped15637456092474531 number pups Mendelian segregation calculated rounded Gene names italicized embryonic day P days postnatal MLKL variants map brace severe inflammatory phenotype MlklD139V neonates defects stress hematopoiesis adults explored prevalence brace region variation human MLKLgnomAD MLKL data 140,000 individuals second third frequency MLKL missense coding variants rs34515646 (R146Q) rs35589326 alter same brace helix (Table 2 4th common MLKL polymorphism rs144526386 (G202*V) missense shorter splice isoform MLKL ‘MLKL2’49 full MLKL encodes 471 amino acid protein MLKL2 263 amino acids long MLKL2 lacks pseudokinase domain killing potential 4HB co-effectors RIPK3 HSP9013 Glycine202* encoded extension exon 9 unique MLKL2 splice isoform (Fig. 5a b).Table 2Human MLKL brace helix polymorphism frequencyrs34515646S132P rs35589326G202aV rs144526386CADD Score Biobank Total MAF (487,658)0.0161 (487,625)0.0147 (487,488 (141,339)0.0138 (141,442)0.01228 (141,400 Highest MAF population0.0252 (64,541) European-Finnish)0.0311 (5185) Ashkenazi Jewish0.0245 (5184) genomes Total MAF (2504)0.0088 Highest MAF (503) European0.024 (489) South Asian0.021 (503) sequenced MAF Allele Frequency highest frequency MLKL SNPs encode non amino acid substitutions brace helix S132 R146 D140 MLKL brace helix Alternate amino acids polymorphisms parentheses G202 α helix isoform interface S132 R146mouse equivalent human rs35589326 (hMLKLS132P), forms membrane high molecular weight complexes Blue Native PAGE kills MDFs necroptotic stimuli mouse dermal fibroblasts for 6 21 hrs cytoplasmic fraction crude membrane fraction TSI TNF Smac-mimetic IDN6556 Chlor Chloroquine two experiments similar results Error bars indicate mean ± SEM 4–5 experiments Schematic brace helix variant combinations three CRMO patients MTRs mapped MLKL show low tolerance missense variation increased Multiple sequence alignment (MSA) conservation scores mapped MLKL regions highly conserved less conserved amino acid substitution MLKLR146Q ‘tolerated’ ‘benign’ SIFT/POLYPHEN 2 R146 human MLKL shows NMR chemical shift perturbations negatively charged IP3 IP6 phospholipid head groups possible role membrane association disruption11 Ser-132 before first structured residue brace helix human MLKL Serine-to-Proline substitution conformation adjacent W133 (brace helix) proximal W109 4HB domainmapped MLKL splice-isoform 249 Glycine 202* predicted isoform 2-specific helix interface with S132 R146 brace helix 1. structural consequence three brace polymorphisms unknown human MLKL predicts disruption brace region favors activated conformation13 murine equivalent human S132P variant mMLKLS131P formed high molecular weight membrane complexes killed MDFs necroptotic stimulus (Fig. 5c d close endogenous levels mMLKLD139V unstimulated mouse dermal fibroblasts mutant pups mMlklS131P CRISPR modified mouse line reduction MLKL protein levels wild-type littermates cellular clearance not complete mMLKLD139V data indicate constitutive activation reduced protein stability not unique mMLKLD139V MLKL brace mutant mMLKLS131P.MLKL brace variants occur human MLKL brace region polymorphisms human autoinflammatory disease examined frequency cohorts ankylosing spondylitis chronic multifocal osteomyelitis Guillain Barré Syndromeminor allele frequencies R146Q S132P G*202V not enriched in disease cohorts healthy controls Tables 4 5) alleles occur in trans 5e in 3 out of 128 CRMO patients 29 times frequency in healthy NIH 1000 genomes samples two compound heterozygotes out of 2504 healthy individuals 10–12 times frequency European CRMO patients healthy European control populations (Table 3) 3Human MLKL brace helix compound heterozygotes in CRMO vs healthy controls.PopulationFrequency relevant compound Hetsa in P value)Globalc0.023 (3/128)0.0008 NIH 1KG29:10.0010.0001Europeand0.02 NIH of R146Q S132P rs35589326 G202*V – rs144526386 (Fig. 5e).bFrequency ratio rounded to nearest whole number patients healthy controls all included European descent necroptosis inflammatory form cell death definitive evidence MLKL activated by inflammatory stimuli difficult to separate cause from effectidentification auto-activating mutant MLKL (MlklD139V) in mice consequences inappropriate necroptosis led to insights into adult hematopoietic perinatal developmental processes sensitive excessive MLKL activation physiological mechanisms activated MLKL immunohistochemical marker for RIPK3-independent necroptosis pinpoint cell type necroptosis in MlklD139V mice high levels pro-inflammatory cytokines in MlklD139V/D139V pups at E19.5 suggests necroptosis inflammation begins sterile utero environment not enough retard prenatal development hematopoietic cell populations birth exposure homozygous MlklD139V pups suppress activity overwhelmed die within days dose-dependent effect both MlklD139V/Wt heterozygous mice viable Postnatal death prevented by deficiencies Ripk3 Casp8 inflammatory genes Myd88 elevated levels G-CSF IL-6 IL-5 role pre perinatal inflammation in MlklD139V pups subject future investigationsMlklD139V mutation platelet production independent thrombopoietin receptor Mpl mechanism unclear follows RIPK3 platelet activation58 reduced platelet levels in MlklD139V/D139V pups unlikely sole cause death severe thrombocytopenia not lethal in Mpl−/− mice40 Difficulty suckling inflammatory infiltration failure explanation lethality in MlklD139V/D139V pups narrow window mortality pericardial immune infiltration make heart failure potential cause sudden neonatal death MlklD139V mouse maintaining MLKL levels below threshold prevent necroptotic activation potential tool for examination MLKL-mediated extracellular vesicle generation death roles inflammation unconfounded RIPK3 activation ESCRT dependent repair extracellular vesicle extrusion protect membranes from MLKL p-MLKL internalized degraded by lysosome17 data suggest role proteasome in disposal activated MLKL free ubiquitin possibility mechanisms ESCRT mechanisms intersect mechanisms hold gene-dose levels active MLKL without deleterious consequences inhibition activated MLKL prevent necroptotic cell deathMlklD139V mouse crosses useful tool for ROS production necroptotic plasma membrane disruption in varied tissue types relevant contexts reviewed mouse MLKL-human MLKL comparisons species-specific structural mechanistic differences5 140,000 individuals surveyed one case human carrying substitution equivalent to mMlklD139V mouse (hMLKLD140V; rs747627247) in gnomAD database heterozygous for variant 3,841 individuals in gnomAD (55 homozygotes) carry MLKL variant –MLKLS132P CRISPR-generated MlklS131P mouse equivalent supports connection between MLKL activation decreased MLKL protein stability variant manifests milder context specific phenotype in mice consistent with high frequency presence in human population speculate human MLKL variants lead to altered MLKL function regulation tissue pathogen specific way60–62 increased numbers cohorts required to confirm human MLKL variants in autoinflammatory disease CRMO patient cohort offers potential modifiers of complex polygenic inflammatory disease in humansmice backcrossed to C57BL/6 >10 generations generated C57BL/6J background Mlkl− Tnfr1− Myd88− IFNAR1− Ripk3− Casp8− Casp1/Casp11−/− mice generated,46,63–67 Mice E19.5 obtained Caesarean section from mothers progesterone injections E17.5 E18.5. mouse strains D139V S131P mutation Mlkl gene CRISPR generated using CRISPR/Cas9 D139V sgRNA GGAAGATCGACAGGATGCAG oligo donor Cas9 mRNA (5 ng μL injected into cytosol C57BL/6J zygotes Mice sequenced across mutated region confirm altered codon analysis after 2 back-crosses to C57BL/6 procedure generation MLKLS131P CRISPR mice sgRNA oligo donor Sex not recorded mice sampled at E19.5, P2 P3 Experiments adult mice performed males females between 8 12 weeks ageMice housed temperature humidity controlled pathogen free facility 12 cycle WEHI Animal Ethics Committee approved experiments NHMRC mapped chromosomal location Plt15 mutation mating 129/Sv Mpl− mice N2 F2 generations genome-wide scan 20 N2 mice highest platelet counts (287 ± 74 × 106 per mL 133 ± 75 × 106 localized mutation chromosome 8 between D8Mit242 D8Mit139 linkage refined Analysis F2 population reduction frequency homozygous C57BL/6 alleles D8Mit200 3/81 F2 mice p = 2.2 × 10−5 Plt15 mutation recessive lethality refined 2.01 Mb interval 31 annotated genes five expressed hematopoietic embryogenesis Dead box proteins 19a 19b Ring finger WD repeat domain 3 Mixed lineage kinase domain WD40 repeat domain 59 Sequencing identified single mutation A to T transversion Mlkl heterozygous mice elevated platelet count Rat-anti mRIPK3-mMLKL 8F6 anti-MLKL 3H14 produced in-houseAnti-Pro Caspase 8 GAPDH#2113) Cell Signaling Technology Anti-mouse MLKL pS345 anti-Actin Abcam Anti-VDAC Millipore Fc-hTNF 100 ng per mL Recombinant mouse IFN-γ β R&D Systems Q-VD-OPh zVAD-fmk MP Biomedicals Hills Smac mimetic Compound A caspase inhibitor IDN-6556 TetraLogic PA Propidium iodide doxycycline bafilomycin Sigma-Aldrich Hill NSW line generation culturePrimary mouse dermal fibroblasts skin E19.5 pups tails adult MDFs immortalized lentiviral transduction SV40 T antigen Immortalized MDFs transduced exogenous mouse MLKL pFTRE 3 G vector Toru Okamoto doxycycline expressionoligonucleotides used for assembly constructs;mMlkl 5′,N-FLAG 5′-CGCGGATCCAA gac-3′N-FLAG 5′-gtcttcctctgcaAcctgtcgatcttc-3′mMlkl-ctgctgccaggGtgcaggctggc-3′Cells maintained in culture 4-hydroxy-tamoxifen regulated HOXA9 Factor Dependent Myeloid cells generated from mouse E14.5 fetal liver cells cultured.Cell death assaysFlow Cytometry 5 × 104 MDFs per well in 24 or 48-well tissue culture Doxycycline (20 ng per mL added death stimuli Fc-hTNF used 100 ng per mL Compound A Smac mimetic IDN6556 500 nM 5 μM zVAD-fmk QVD-OPh 25 10 μM Mouse human interferons γ β 30 ng per mL PS341 MG132 2 200 nM Bafilomycin 300 nMIncucyte imaging MDFs plated 8 103 cells 96-well plate 3 h FDMs plated 5 10 × 103 cells 48-well plate 0.2 μg per mL propidium iodide media stimuli Images recorded 1 2 h IncuCyte S3 PI positive cells per mm2 quantified plotted IncuCyte S3 turn-over assays5 104 MDFs plated 24-well plates settle Doxycycline (20 ng mL TNF Smac Mimetic IDN6556 added 15 h dox’ ‘0’ wells harvested Media removed cells washed PBS fresh media IDN6556 re-added Wells harvested 2 4 6 8 24 h Cells direct lysis SDS-PAGE loading buffer protection assays5 104 MDFs plated 24-well settle Doxycycline (20 ng mL added 18 hrs ‘no dox’ ‘T0’ samples harvested Media removed cells washed fresh media 3 h restored IDN6556 + inhibitor PS341 Chloroquine Bafilomycin Ca-074 TLCK AEBSF 21 h Cells harvested direct lysis bufferpull 106 MDFs transduced doxycycline N-FLAG settled stimulation 1 μg per mL doxycycline 5 hrs Cells lysed Urea UBA buffer ubiquitylated proteins enriched Usp21-treated electron microscopyMurine dermal fibroblasts mice untreated stimulated agents cells fixed 2% glutaraldehyde 0.1 M phosphate buffer pH 7.4 postfixed 2% OsO4 dehydrated ethanol Epok 812 sections cut ultramicrotome stained uranyl acetate lead citrate examined JEOL JEM-1400 electron microscope viability LDH release histopathologyCaesarian-sectioned E19.5 P2/3 pups euthanized fixed 10% buffered formalin Five-micrometers coronal sections 200-μm intervals 5-μm sagittal sections 300-μm intervals examination histopathologists Aira Nuguid Tina Cardamome Australian Phenomics Network Melbourne confirmed Veterinary Pathologist Prof. John W Finney Pathologist Catriona McLean Alfred Hospitalrelative thymic cortical images thymus sections analysed cortical thickness ImageJ medullary areas identified H E staining removed thymus structure Image J Calculator cortical region thickness defined radius largest disk pixel position determined Local Thickness plugin ImageJ terminal blood collection P0 P3 pups fixed 24 h in 10% buffered formalin paraffin before microtomy Immunohistochemical detection cleaved caspase 3 CD45 (BD) performed.Cytokine plasma stored at −80 °C cytokine analyses Cytokines measured Bioplex Pro mouse cytokine 23-plex assay) samples designated ‘<OOR’ cytokine assigned lowest value cohort Values plotted change mean value samples p values calculated Microsoft Excel two-tailed TTEST unequal variance Data shown cytokines significant differences between genotypes day E19.5 P3.Hematological analysisBlood collected from P0 P3 pups into EDTA coated tubes heparinized glass capillary tubes neck cavity after decapitationcentrifugation 500 G 5 min 5–15 μL plasma removed replaced PBS Blood cells resuspended diluted 8–20-fold DPBS quantification ADVIA 2120 analyzer 6 h Blood collected adult mice tubes EDTA analyzed ADVIA120 analyzer (Bayer).Transplantation studiesDonor bone marrow fetal cells injected C57BL/6-CD45Ly5.1/Ly5.2 mice 11 Gy gamma-irradiation two doses neomycin (2 mg per mL 4 weeks Long term capacity stem cells assessed flow cytometric analysis 6 months engraftment Recovery cytotoxic insult blood analysis 150 mg per kg 5-fluorouracil (5-FU).Flow incubated antibodies Ly5.1-PE Ly5.2-FITC Ly5.2-biotin Ly5.2 PerCPCy5.5 streptavidin PECy5.5 propidium iodide analysed LSRI) flow cytometer bone marrow cells incubated biotinylated Alexa700 conjugated antibodies lineage markers CD2 CD3 CD4 CD8 CD34 B220 CD19 Gr-1 Ter-119.stem progenitor cell detection antibodies cKit Sca-1 CD48 AnnexinV CD105 FcγRII/III CD135 FluoroGold Bioquest#17514) dead cell detection Cells analysed LSRII Fortessa1 Biosciences cytometers oxygen species Chloromethyl-H2DCFDA dye bone marrow cells loaded 1 μM Chloromethyl-H2DCFDA 30 min 37 °C 37 °C StemPro-34 serum free medium 15-min cells stained surface antibodies FACS analysis analysed LSRII cytometer fetal colony 104 liver cells 1 mL cultures 35 mm Petri dishes 10% FCS 0.3% agar 104 U per mL GM-CSF IFN-β added concentrations 0 30 ng per mLColony formation 7 days incubation 37 °C humidified 10% PCRRNA prepared Trizol 10 μg first strand cDNA synthesis SuperScript II 0.5 μg cDNA TaqMan PCR reaction Universal PCR mastermix Mlkl (Mm1244222_n1) GAPDH (Mm99999915_m1) probes ABI 7900 Fast Real-Time PCR instrument Mlkl expression GAPDH determined SDS 2.3 expressed ΔCT values cell-based assays figure legends error data points independent p values calculated Microsoft Excel Prism unpaired two-tailed t-test unequal variance not adjusted multiple comparison Asterices p ≤ 0.05 0.01 0.005 comparisons between MlkWt/Wt MlkD139V groups exception adult mice MlkWt/Wt-exome sequencingDNA from CRMO probands family members purified from saliva prepared for sequencing samples underwent WES enriched Agilent SureSelect Human All Exon V4 V5 V6 + UTR before sequencing Otogenetics, Inc Beckman Coulter Genomics University of Iowa Genomics Core fastq files quality-checked processed to vcf format Variants called Haplotype recalibrated hard-filtered in GATK Variants annotated with minor allele frequencies from 1000 genomes75 information effect SNPSift/SNPEff76 databases annotation dbNSFP2.977 GRCh37.75.Ancestry determined CRMO proband LASER software vcf file probands uploaded LASER TRACE analysis selected Worldwide panel indeterminate ancestry European Asian panels used Principal component values plotted R Statistical Software LASER package.MLKL variant quantification1000 Genomes Vcf files annotated filtered Values for MLKL variants rs35589326 rs34515646 rs144526386 all MLKL coding variants queried tabulated for allele genotype count for ancestry 2504) European ancestry = 503)heterozygous variants phasing 1000 genomes dataset Allele genotype counts MLKL variants tabulated European = 101 variants identified parental sequence data DNA subjects cohort genotyped Illumina CoreExome chip Australian Translational Genomics Centre Princess Alexandra Hospital Brisbane Bead intensity data processed normalized Illumina Genome Studio software samples passed quality control Genotyping ISO15189-accredited Australian Translational Genomics Centre Queensland University Technology samples genotyped Illumina HumanOmniExpress BeadChip81 healthy control data European cohort studies Translational Genomics Group Illumina CoreExome microarray data UK Household Longitudinal Study Institute Social Economic Research University Essex funded Economic Social Research Council survey conducted NatCen genome-wide scan data analysed Wellcome Trust Sanger Institute University Essex Kantar Public Understanding Society Waves 1–8 2009–2017 Harmonised BHPS Waves 1–18 1991–2009 11th Edition UK Data Servicegenomic data derived from patients recruited consent approval ethics review boards Institutes University of Iowa Carver College Medicine Queensland University of Technology Australian National University Shanghai Renji Hospital JiaoTong University Shanghai Hospital Sick Children University of Toronto University of Sydney Australian Institute of Sport University of Freiburg Princess Alexandra Hospital Memorial Hermann Texas Medical Centre University of Queensland Oregon Health Science Groupe Française d’Etude Génétique des Spondylarthrites University of Oxford analysis comparisons allele frequency compound heterozygote sample frequency Fisher’s exact test Chi-Squared test Yates correction Compound heterozygous variants quantified compared individual allelic level without qualifying variants allelic level resourcesgnomAD OrthoDB.washington.edu Omega.berkeley.edu Tolerance Ratio) Gene Viewer.unimelb.edu biobank.ukUnderstanding Society research Nature Research Reporting Summary information | 49.3 | 1.11084 |
10.1038/s41467-021-21543-3 | PMC7910584 | It is currently challenging to identify protein structures at low concentrations. Here the authors report optical tweezers-coupled Raman spectroscopy to generate tunable and reproducible SERS enhancements with single-molecule level sensitivity and use the method to detect protein structural features. | Surface-enhanced Raman spectroscopy (SERS) has emerged as a powerful tool to detect biomolecules in aqueous environments. However, it is challenging to identify protein structures at low concentrations, especially for the proteins existing in an equilibrium mixture of various conformations. Here, we develop an in situ optical tweezers-coupled Raman spectroscopy to visualize and control the hotspot between two Ag nanoparticle-coated silica beads, generating tunable and reproducible SERS enhancements with single-molecule level sensitivity. This dynamic SERS detection window is placed in a microfluidic flow chamber to detect the passing-by proteins, which precisely characterizes the structures of three globular proteins without perturbation to their native states. Moreover, it directly identifies the structural features of the transient species of alpha-synuclein among its predominant monomers at physiological concentration of 1 μM by reducing the ensemble averaging. Hence, this SERS platform holds the promise to resolve the structural details of dynamic, heterogeneous, and complex biological systems. | IntroductionRaman spectroscopy probes the endogenous vibrations of molecules upon irradiation to delineate their chemical structures and surrounding environments1, feasible for the label-free characterization of biomolecules in aqueous environments2,3. Inheriting these advantages, surface-enhanced Raman spectroscopy (SERS) boosts up the sensitivity to detect proteins at low concentrations, even at the single-molecule level, to mimic physiological conditions4–6. It is also a powerful tool to characterize the dynamic ensembles of variable conformations of intrinsically disordered proteins (IDPs)7–10. IDPs lack stable secondary and tertiary structures as monomers in aqueous environments, but sometimes self-assemble into oligomers with various structures and further grow into amyloid fibrils11, which are associated with the incurable neurodegenerative diseases12. The dynamic conversion from monomers to oligomers is the key step in the early pathological development13. However, the transient nature and the low population of oligomers make it challenging to characterize their structural features, which are responsible for the cellular toxicity and the on-going amyloid aggregation14. In particular, the SERS study on alpha-synuclein, an IDP closely related to Parkinson’s disease15–17, is scarce18. Besides, practical difficulties emerge when using conventional nanoparticle-based SERS substrates generated from salt-induced random aggregations to probe proteins in dilute solutions, such as low efficiency, poor reproducibility, and potential structural perturbation during long incubation time in SERS measurements19,20. Hence, there is an urgent need to customize the SERS approach to better characterize protein structures at low concentration for greater biological significance.The SERS activity arises from the induced strong electromagnetic field confined in the junctions between plasmonic nanostructures, known as “hotspots”21,22. The salt-induced aggregation of silver nanoparticle (AgNP) colloids is the most accessible method to form SERS active gaps23–25. Yet such random aggregates lead to unequal enhancement factors and fluctuating SERS signals26. Thus, one major challenge in the field is to improve the reproducibility. Advancements have been achieved mainly on fabricating SERS substrates with sharp size distribution27–29, in order to provide consistent SERS enhancements. Nevertheless, when measuring analytes at low concentrations, it is inconvenient to locate and focus on the nanometer-size hotspots from the conventional aggregated nanoparticles in colloidal suspension for in situ detection, thus lowering the feasibility and efficiency of the technique. To address this dilemma, mechanical manipulations are desirable alternatives to create hotspots dynamically at the designated location7,30,31. Recently, optical tweezers-assisted SERS has been developed to increase the efficiency for the SERS analysis in aqueous conditions32–41, by introducing the additional trapping laser beams to manipulate SERS substrates42,43. Kall and co-workers utilized optical tweezers to bring two AgNPs together to form a SERS active dimer44, but the hotspot size and the precise location was uncertain due to the diffraction limit of optical microscope44. Smith and co-workers trapped one partially Ag-coated silica microparticle as a mobile and visible SERS probe45, which was further developed by Petrov and co-workers to accomplish interfacial detection of living cells46. However, there is still lack of an explicit control on the hotspot and the subsequent SERS performance. The optical tweezers-based SERS platform awaits further developments to exploit its mechanical control to improve the efficiency, sensitivity, and reproducibility for the investigation of various biomolecules in dilute environments.In this work, we introduce a convenient approach to visualize and control hotspots to provide consistently high SERS enhancements for the characterization of protein structures and conformational fluctuations at physiological concentration, by developing the optical tweezers-coupled Raman spectroscopy. Specifically, two AgNP-coated micrometer-size silica beads are trapped and approached to form the SERS active interparticle gap under the precise manipulation and real-time visualization for in situ spectroscopic measurements, even offering single-molecule level sensitivity. This dynamic and tunable SERS window is placed in a microfluidic flow chamber to detect the passing-by proteins. The flow rate of the protein solution is fine-tuned to minimize the interaction between proteins and AgNP-coated beads, in order to preserve their native states and conformations. The structural features of three typical globular proteins (hemoglobin, lysozyme, and bovine serum albumin) and an IDP (alpha-synuclein) are analyzed in dilute solutions. In particular, the SERS characterization of alpha-synuclein at physiological concentration (1 μM)47 is reported, showing the structural variations arisen from its transient species. This sensitive, reliable, and convenient SERS platform enables flash spectroscopic snapshots of proteins at low concentration to reduce the ensemble averaging in time and in quantity, thus has great potential to investigate IDP oligomers to provide perspectives on the initiation of amyloid protein aggregation and tackle the problems in complex biological systems.ResultsThe optical tweezers-coupled Raman microscope and the trappable SERS substrateThe home-built SERS platform combines dual-trap optical tweezers and Raman microscope. As illustrated in Fig. 1a, two 1064 nm trapping laser beams (red) and one 532 nm Raman probe beam (green) propagate inside the inverted microscope through stereo double-layer-pathways, which are then combined by a dichroic mirror to enter the objective from the bottom. Spatially separated, the beams are focused inside a flow chamber connected to the microfluidic system. The two trapping laser beams manipulate the flowing AgNP-coated beads in 3D while the Raman probe beam irradiates the gap between two trapped AgNP-coated beads as shown in Fig. 1b. The backscattered light is collected for spectroscopic measurements. In the forward direction, the two trapping laser beams are collimated by a condenser for force detections and camera imaging. (More instrumental details are illustrated in Supplementary Fig. 1). Figure 1c shows the real-time camera image of the two trapped AgNP-coated beads for direct distance adjustments. This instrumental set-up enables the optical manipulation of SERS substrates under microscopic visualization and the in situ SERS measurements at the same time.Fig. 1Illustration of the controllable SERS probe experiments.a Schematic diagram of the optical tweezers-coupled Raman spectroscopic platform with microfluidic set-up. b Two trapping laser beams (red) to manipulate two AgNP-coated beads and one Raman probe beam (green) to detect signals from the gap between the two AgNP-coated beads. c The real-time camera image of two trapped AgNP-coated beads from the microscope. d SEM image of the gap between two AgNP-coated beads. The scale bar is 0.1 μm. e SEM image of AgNP-coated beads to show the uniform AgNP coating. The scale bar is 1 μm. All micrographs are representative images of three independent measurements.The AgNP-coated beads were chemically fabricated for manipulation and visualization as the dynamic SERS probe45,46,48. First, we modified the surface of micrometer-size silica beads with amino groups in anhydrous ethanol, then coated them with AgNP colloids in continuous agitation. The AgNPs were prepared from the reduction of silver nitrate by trisodium citrate and washed three times to remove the reducing agent before mixing with the modified silica beads. The parameters such as interaction time, stirring rate, beads size, and the reactant ratio were fine-tuned to generate the uniform AgNP coating at the density of 17.6% on the beads (Fig. 1e), in order to ensure the reproducible and stable optical trapping performance. (Details are exhibited in Supplementary Fig. 2.) Figure 1d displays the SEM image of the interparticle gap when two AgNP-coated beads are in close proximity, which could determine the SERS activity. Compared to the direct trapping of AgNPs, the development of micrometer-size AgNP-coated beads enables the precise position control with sub-nanometer spatial resolution while minimizes their Brownian motions for more effective, reproducible, and stable spectroscopic measurements. (Details are demonstrated in Supplementary Figs. 3 and 4.) Moreover, the visualization of the SERS hotspot between the AgNP-coated beads on the brightfield camera or the eyepiece of Raman microscopes makes it easy to adjust the Raman excitation focus, superior to the dispersed and invisible AgNP colloids with non-specific aggregations in dilute solutions. Thus, the strategy to create the hotspots between micrometer-size AgNP-coated beads improves the operational efficiency and simplicity for sensitive SERS detections.The creation of the real-time controllable hotspotAs a proof of concept, the dynamic SERS probe was created and adjusted between the two AgNP-coated beads on the optical tweezers-coupled Raman microscope to detect the SERS signal of 1% ethanol aqueous solution. Two AgNP-coated beads were trapped at different distance reversibly, meanwhile the real-time camera images (Fig. 2a) and the corresponding SERS spectra with 1 s acquisition time (Figs. 2b and d) were recorded. At the beginning, the two beads were separated away by 900 nm on the two sides of the Raman excitation spot to acquire the blank spectrum in Fig. 2b (black), which is the same as spontaneous Raman of 1% ethanol in Supplementary Fig. 5 showing no apparent signal. The two beads were approached to decrease the distance below 100 nm, while the Raman signal gradually emerged and enhanced, indicating the formation of the SERS active gap at the Raman excitation spot. The characteristic peaks at 886 cm−1 (C–C stretching), 1054 cm−1 (C–O stretching), 1094 cm−1 (CH3 rocking), 1280 cm−1 (CH2 deformation) and 1458 cm−1 (CH3 asymmetric bending)49 are consistent with the spontaneous Raman of pure ethanol shown in Supplementary Fig. 5. To confirm the SERS enhancement effect from the hotspot between the two approached AgNP-coated beads, the entire surface of individual AgNP-coated beads was scanned in time-series measurements. The blank spectra of the single AgNP-coated bead trapped at the Raman excitation spot before and after the hotspot creation in Supplementary Fig. 6 imply no intra-bead hotspots. It is worth noting that the spectral features were boosted up during beads approaching in a 10 nm step-size from 30 nm to 0 nm, indicating that the SERS active interparticle gap was under precise control and the slight adjustment resulted in the dramatic signal enhancement. In the reverse moving direction, the SERS signal gradually decreased when the two AgNP-coated beads were separated further from 20 nm to 100 nm, shown in Fig. 2d. At the same beads distance, the signal intensities obtained in beads separating are comparable to those obtained in beads approaching. Figure 2c demonstrates the intensity of the ethanol characteristic peak at 1458 cm−1 as a function of the distance between the two trapped AgNP-coated beads, obtained from the trapping laser position and the brightfield image analysis. It is clear that the vibrational features of ethanol were emerged, enhanced, decreased, and vanished in inversely proportional relationship to the beads distance, under the precise control by optical tweezers with excellent flexibility and reproducibility. Furthermore, the two trapped beads were pushed at a distance of 0 nm to maximize the SERS enhancement and generate the detectable force between them to confirm the contact of two beads (Supplementary Fig. 7) and ensure the stability of the SERS active interparticle gap (Supplementary Fig. 6). It was utilized to probe 10−9 M and 10−11 M Rhodamine B, showing the Raman signatures of the aromatic C–C stretching of Rhodamine B at ~1360 cm−1, ~1565 cm−1, and ~1650 cm−1 in Supplementary Fig. 850. The theoretical simulation indicated its SERS enhancement factor up to 109, which is sufficient to empower the single-molecule level detection.51 (Details are explained in Supplementary Fig. 8.) Overall, the creation and the adjustment of hotspot could be achieved efficiently and reversibly at the Raman excitation spot, which offers a convenient platform to generate tunable and reproducible SERS enhancements for various analytes.Fig. 2Creation and adjustment of the dynamic hotspot with in situ SERS measurements of 1% ethanol aqueous solution.a The real-time camera images of the two AgNP-coated beads trapped at different distance. b SERS spectra of 1% ethanol aqueous solution with 1 s acquisition time when the two AgNP-coated beads approaching. c The intensity of the ethanol characteristic peak at 1458 cm−1 as a function of the distance between the two AgNP-coated beads from beads approaching to beads separating. The reversible bead positions are illustrated as inset. d SERS spectra of 1% ethanol aqueous solution with 1 s acquisition time when the two AgNP-coated beads separating. Source data are provided as a Source data file.The spectroscopic characterizations of native hemoglobinWe utilized the controllable SERS probe as a detection window to analyze the flowing hemoglobin in dilute aqueous solution without incubation or aggregation with AgNPs. As illustrated in Supplementary Fig. 9, two AgNP-coated beads were trapped in the microfluidic bead channel then moved to the hemoglobin channel for SERS measurements. The flow rate of the hemoglobin solution was fine-tuned to minimize the interaction between proteins and AgNP-coated beads. The spectroscopic scans of the individual AgNP-coated beads at the Raman excitation spot show the blank spectra in Supplementary Fig. 10 to confirm neither intra-bead hotspots existence nor hemoglobin attachments. Under the visualization and manipulation by optical tweezers, the two AgNP-coated beads were approached at incremental distance and gradually adjusted Raman excitation power to optimize the SERS signal of 100 nM hemoglobin in aqueous solution in Fig. 3a and b. When the beads distance was smaller than 20 nm and the laser power was larger than 10% (2.5 mW), the subtle spectral features were overwhelmed by the intense characteristic peaks arisen from amorphous carbon (~1370 cm−1 and ~1580 cm−1)52, suggesting molecular damages. Thus, the SERS detection window was set under the bead distance at 20 nm and the Raman excitation power at 10% with 1 s acquisition time. It is showcasing the great adjustability of the optical tweezers-coupled Raman spectroscopy to preserve the protein native states.Fig. 3The spectroscopic characterizations of hemoglobin at its native states.a SERS spectra of 100 nM hemoglobin in aqueous solution when the two AgNP-coated beads were trapped at different distance (50–10 nm). b SERS spectra of 100 nM hemoglobin solution under different Raman excitation power (5–100%). c The comparison between SERS spectra of 100 nM hemoglobin solution with 1 s acquisition time (blue) and the spontaneous Raman spectrum of 250 μM hemoglobin solution with 60 s acquisition time (black). d 3D stacking plot of SERS spectra of 100 nM hemoglobin solution obtained from AgNP-coated beads trapped at 20 nm with 1 s acquisition time. e, f The histograms of the intensities of the hemoglobin characteristic peaks at 1376 cm−1 (mean = 4929.15 in arb. units, RSD = 14.63%) and 1594 cm−1 (mean = 5391.41 in arb. units, RSD = 16.35%) across the 50 SERS spectra in (d), respectively. Source data are provided as a Source data file.To examine if the hemoglobin retains native, five SERS spectra of 100 nM hemoglobin with 1 s acquisition time (blue) were compared to the spontaneous Raman spectrum of 250 μM hemoglobin with 60 s acquisition time (black) in Fig. 3c. The identical frequencies between the SERS spectra and the Raman spectrum of hemoglobin indicates the similar protein states in measurements3. As the oxidation state marker bands, the subtle peaks at 1346 cm−1 and 1376 cm−1 among these spectra are distinctly similar, attributed to the ferrous state and the ferric state53, respectively. Vibrational bands at 1568 cm−1 and 1594 cm−1 corresponding to 6-coordinated high-spin heme and 6-coordinated low-spin heme are apparent at matching positions53,54. Bands appearing at 1089 cm−1, 1311 cm−1 and 1440 cm−1, assigned to the vinyl group deformation in the porphyrin ring of the heme center, are in good agreements.55 Whereas, the marker bands of 5-coordinated high-spin heme at 1494 cm−1 and 1572 cm−153,54, representing the non-native state from the perturbation of metal surface, were not observed. All peak assignments of hemoglobin are listed in Supplementary Table 153,55,56–58. It is clear that the SERS probe created by two AgNP-coated beads well preserved the oxidation state, the coordination number, and the spin state of hemes in hemoglobin, which are closely linked to the native structure and function of hemoglobin54. Furthermore, the high similarity of these SERS spectra demonstrates the reproducibility of this SERS platform.To further preserve the protein native states, the two AgNP-coated beads were replaced freshly in parallel SERS measurements to minimize the interaction time with proteins19,59. Figure 3d displays SERS spectra of 100 nM hemoglobin solution obtained from 50 parallel measurements. Apparently, the vibrational signatures of the heme center of hemoglobin at 1346, 1376, 1568, and 1594 cm−1 among the 50 SERS spectra are approximately the same. Figure 3e and f demonstrate the histograms of the peak intensities at 1376 cm−1 and 1594 cm−1 across the 50 SERS spectra with relative standard deviation (RSD) as 14.63% and 16.35%, respectively. Histogram analysis of other spectral features is presented in Supplementary Fig. 11. These reproducible and stable spectra prove the consistent SERS enhancements in the parallel measurements when two AgNP-coated beads were trapped at a constant distance. Hence, the controllable SERS probe inside the microfluidic flow chamber could preserve the flowing proteins at their native states and generate reproducible SERS spectra.The spectroscopic characterizations of the compact globular structure of lysozymeTo demonstrate the ability to detect protein conformations with high accuracy, we employed the dynamic SERS probe as a detection window to characterize two well-known globular proteins in solutions: lysozyme and bovine serum albumin (BSA). With the experimental protocol analogous to the previous section, two fresh AgNP-coated beads were trapped at 20 nm to analyze the flowing proteins in dilute aqueous solutions, which would be replaced freshly in parallel experiments. Figure 4a displays the SERS spectra of 1 μM lysozyme solution acquired from 50 parallel experiments, showing the vibrational frequencies of lysozyme identical to its spontaneous Raman spectrum. Specifically, the peaks at 766, 1015, 1337, and 1557 cm−1 are assigned to the aromatic residues (tryptophan) and the peak at 1450 cm−1 is attributed to aliphatic residues (CH2)60. The amide I band at 1655 cm−1 and the amide III band at 1250 cm−1 imply the existence of α-helix in the folded globular structure of lysozyme. The spectral contributions from different secondary structures were deconvolved to α-helix (45.2%), β-sheet (11.3%), and random coil (43.5%) in Supplementary Fig. 12, which is consistent with the previous investigations61,62. Figure 4b demonstrates the Amide I band distribution of the 50 SERS spectra of 1 μM lysozyme solution with 1655 ± 2 cm−1 (0.1% RSD), indicating its structural stability and homogeneity as a typical globular protein. Moreover, the SERS spectra of 1 μM BSA in Supplementary Figs. 13 and 14 also provide the structural component assessment (66.5% α-helix, 9.5% β-sheet and 24.1% random coil) supported by the previous studies63. It is worth noting that the 50 SERS spectra of 1 μM lysozyme in Fig. 4a demonstrate high similarities, due to the stability and reproducibility of this SERS probe as well as the nature of the stable, compact, and globular conformation of lysozyme. Since the sizes of these globular proteins are smaller than the bead distance at 20 nm62, these SERS spectra reflect the whole protein structures. As illustrated in Fig. 4c, the small-size sampling in the parallel SERS measurements unveil the protein structural fluctuation to complement the ensemble averaging in the spontaneous Raman measurement.Fig. 4The spectroscopic characterizations of lysozyme in its compact globular structure.a The comparison between 50 SERS spectra of 1 μM lysozyme solution with 1 s acquisition time (bottom) and the spontaneous Raman spectrum of 1 mM lysozyme solution with 5 min acquisition time (top). b Histogram of the Amide I band distribution of the 50 SERS spectra of 1 μM lysozyme solution in (a), indicating the mean as 1655 cm−1 with 0.1% RSD. c Illustration of the ensemble averaging from the spontaneous Raman measurement of lysozyme in the concentrated solution and the small-size sampling from the SERS measurements of lysozyme in the dilute solution. Source data are provided as a Source data file.The spectroscopic investigation of the transient species of alpha-synuclein at physiological concentrationTo exploit the advantages of the controllable SERS probe, we characterized the structural features of alpha-synuclein in aqueous solutions. As a typical IDP, alpha-synuclein undergoes intrinsic conformational conversions to form transient species in a low population at physiological concentration, existing in a dynamic equilibrium mixture to determine its amyloid aggregation at the early stage64. The CD spectrum of 200 μM alpha-synuclein aqueous solution in Fig. 5a presents a negative peak at around 200 nm, indicating the ensemble conformations as random coils13. In Fig. 5b, the spontaneous Raman spectrum of 2 mM alpha-synuclein solution shows the vibrational features in amide III region at 1249 cm−1 and amide I region at 1673 cm−1, which are attributed to the majority population of disordered conformations16. The protein signal of 250 µM alpha-synuclein solution in Fig. 5b is too weak to analyze, due to its low Raman cross-section and the gentler laser power (25 mW) in our setup, lower than the 800 mW laser power used in the previous Raman studies16,17. With the large sample quantity and long detection time in the bulk spectroscopic measurements, the structural features of alpha-synuclein transient species are overwhelmed by the conformational ensemble.Fig. 5The spectroscopic characterizations of an intrinsically disordered protein: alpha-synuclein.a CD spectrum of 200 μM alpha-synuclein in aqueous solution. b Spontaneous Raman spectra of 2 mM (green) and 250 μM (blue) alpha-synuclein solution with 10 min acquisition time. c The comparison among three representative types of SERS spectra of 1 μM alpha-synuclein solution with 1 s acquisition time (blue, red, and black) and the SERS spectrum of 250 μM alpha-synuclein solution with 5 min acquisition time (purple). d Mapping of 200 SERS spectra of 1 μM alpha-synuclein solution obtained from two AgNP-coated beads trapped at 20 nm with 1 s acquisition time. The color bar shows the normalized intensities from low (dark blue) to high (red). e Illustration of the ensemble averaging from the measurement of alpha-synuclein at high concentration with long accumulation time and the small-size sampling from the measurements of alpha-synuclein at low concentration with short accumulation time. Source data are provided as a Source data file.The physiological concentration of alpha-synuclein at non-aggregated states is 1 μM47, below the detection threshold of Raman spectroscopy. Whereas the experimental limit of detection (LOD) of alpha-synuclein on our SERS platform is 100 nM, thus this sensitive SERS approach is feasible to characterize the transient species of alpha-synuclein in dilute solutions. (Details are shown in Supplementary Figs. 15 and 20.) Similar to the experimental protocol in the previous section, two fresh AgNP-coated beads were trapped at 20 nm as the dynamic SERS window to characterize the flowing alpha-synuclein on our platform. Figure 5c demonstrates three representative types of SERS spectra of 1 μM alpha-synuclein solution with 1 s acquisition time among the 200 parallel experiments shown in Fig. 5d. Strikingly, the amide I and the amide III bands of the SERS spectra of 1 μM alpha-synuclein exhibit prominent variations in comparison to the uniform spectral patterns of 1 μM lysozyme in Fig. 4, indicating the co-existence of different alpha-synuclein species with various structures at physiological concentration7. In Fig. 5c, the spectral characteristics of the blue SERS spectrum of 1 μM alpha-synuclein fall in the intervals of 1650–1657 cm−1 (amide I bands) and 1270–1300 cm−1 (amide III bands), indicating the alpha-synuclein species in α-helix structure16,65. The red SERS spectrum of 1 μM alpha-synuclein shows the amide I band at around 1662–1665 cm−1 and the amide III band at 1230–1240 cm−1, which are associated with β-sheet structure16,17,65 This assignment is further confirmed by the SERS spectrum of 1 μM alpha-synuclein at pH 3 in Supplementary Fig. 16, since alpha-synuclein folds into ordered β-sheet conformations in acidic conditions. The amide I band at around 1671 cm−1 and the amide III band at 1240–1250 cm−1 from the black SERS spectrum of 1 μM alpha-synuclein are attributed to random coil structure16,65,66, consistent with its spontaneous Raman spectrum. While the vibrational fingerprints from Phe (1006 cm−1) and deformation from aliphatic residues CH2 and CH3 (1450 cm−1) of alpha-synuclein16 are still uniform across these SERS spectra of 1 μM alpha-synuclein, since they are insensitive to the change of protein conformations. All the peak assignments of the 1 μM alpha-synuclein SERS spectra are summarized in Supplementary Table 216, since the subtle spectral features of alpha-synuclein acquired at physiological concentration could reveal the structural details of its transient species with great biological significance. In particular, the direct characterization of the β-sheet containing oligomers among the unstructured monomers of alpha-synuclein might provide further insight to the pathological aggregation of alpha-synuclein at the very early stage as this conformation is involved prior to fibrillation67,68.Owing to high sensitivity and stability, our SERS platform resolved the structural variations of alpha-synuclein arisen from its transient species at physiological concentration. In the statistics of the 200 parallel SERS measurements of 1 μM alpha-synuclein solution in Fig. 5d, the probability to observe monomers in random coil structures was very high, since they are the predominant species. Occasionally, the transient species showing α-helix or β-sheet structures were observed. It is consistent with the spectral deconvolution of the 2 mM alpha-synuclein Raman spectrum which shows 13.3% α-helix, 8.4% β-sheet and 78.3% random coil in Supplementary Fig. 18 and the 250 μM alpha-synuclein SERS spectrum which presents 15.1% α-helix, 9.5% β-sheet and 75.4% random coil in Supplementary Fig. 19. The overall population of alpha-synuclein transient species is low at physiological concentration, resulting in a lag phase prior to the growth of fibrils at macroscopic-level. As illustrated in Fig. 5e, the dynamic SERS probe enables the small-size sampling in the measurements of alpha-synuclein in low concentration solution with short accumulation time to directly characterize the structural features of different alpha-synuclein species. The statistics of the structural variations of alpha-synuclein from its transient species provides the distribution of different secondary structure components, comparable and complementary to the ensemble averaging in the measurement of alpha-synuclein in high concentration solution with long accumulation time. Such direct identification of the structural variation of alpha-synuclein verifies that our sensitive SERS platform could reduce the ensemble averaging to reveal more structural information on IDP transient species, providing perspective to investigate the behaviors and functions of IDPs during complex biological processes.DiscussionIn summary, we mechanically controlled a dynamic SERS probe to characterize typical globular proteins and intrinsically disordered proteins in dilute solutions, using in situ optical tweezers-coupled Raman spectroscopy. Under microscopic visualization and precise manipulation, two AgNP-coated beads were approached by optical tweezers to create tunable hotspots for efficient, reproducible, and convenient SERS measurements with single-molecule level sensitivity. This dynamic SERS detection window was utilized in the microfluidic flow chamber to detect the flowing proteins at their native states and confirmations, verified by the spectral analysis of hemoglobin, lysozyme, and BSA. With high sensitivity and stability, it resolved the structural variations of alpha-synuclein arisen from its transient species in the low population at physiological concentration, which are buried under the averaging signals in the conventional bulk measurements but crucial for the understanding of the initiation of its amyloid aggregation. Hence, the controllable SERS probe on the optical tweezers-coupled Raman platform is feasible to reveal the unperturbed structural information of various proteins in dilute solutions.Our strategy enables the precise control of the hotspot between the two trapped micrometer-size AgNP-coated beads to improve the SERS efficiency and reproducibility in the aqueous detections. Except for the tunable SERS enhancement, the coupled optical tweezers also offer sub-nanometer spatial resolution and sub-piconewton force sensitivity to monitor light-matter interactions in the plasmonic hotspot for extra physical insight. More importantly, our method opens a door to characterize the structural variations of IDPs in dilute solutions, which remains a significant challenge in the biophysics community. This dynamic SERS probe has great potential to investigate the oligomeric state of amyloidogenic proteins and resolve the structural details as well as the conformational population of these transient species prior to the amyloid aggregation, providing profound molecular insights to understand the onset of neurodegenerative diseases. Ultimately, it will be exciting to fully exploit the precise force manipulation of the integrated optical tweezers to unfold a single protein inside the controllable hotspot to resolve its structural dynamics.MethodsChemicalsSilica beads were purchased from Spherotech Inc. Silver nitrate (≥99.0%), trisodium citrate (≥99.0%), Tris (≥99.9%), sodium chloride (≥99.0%), (3-aminopropyl) triethoxysilane (≥98.0%), lysozyme (≥95.0%) and bovine serum albumin (≥96.0%) were purchased from Sigma-Aldrich. Yeast extract was purchased from Fisher BioReagentsTM. Tryptone was purchased from OxoidTM. Hemoglobin was purchased from Worthington Biochemical Corporation.Expression and purification of proteinsRecombinant human wild type alpha-synuclein was overexpressed by E. coli BL21(DE3) with plasmid pET28a. Cells were grown in LB medium in the presence of 50 µg/mL Kanamycin and protein expression was induced by 0.3 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). The cell pellet was resuspended in Tris buffer (25 mM Tris-HCl, pH 7.4) and lysed by sonication. After centrifugation at 30,000 × g for 45 min at 4 °C, the supernatant was boiled to remove most E. coli proteins. After centrifugation at 30,000 × g for another 60 min at 4 °C, the supernatant was loaded onto HiPrep DEAE FF 16/10 column (GE Healthcare). A gradual sodium chloride gradient was chosen and applied to elute the target protein. After SDS-PAGE gel analysis, fractions containing the target protein were desalted by HiPrep™ 26/10 Desalting column. The desalted solution was loaded onto HiPrep 16/60 Sephacryl S-100 column (GE Healthcare) for further purification, with 25 mM Tris-HCl, pH 7.4 buffer as the running buffer. Fractions were analyzed by SDS-PAGE gel, and targeted protein were collected, concentrated and stored at −20 °C. The concentration of alpha-synuclein was determined by UV-1800 spectrophotometer (SHIMADZU) with the extinction coefficient of 5960 cm−1 M−1 at 276 nm.Fabrication of AgNP-coated beadsA step-by-step protocol describing the fabrication of AgNP-coated beads can be found at Protocol Exchange69. AgNP-coated beads were prepared by coating the AgNPs on the surface of silica beads (R = 0.63 µm)45,46. First, the silver nanoparticles were synthesized based on the conventional method with optimized parameters3,50. 50 mL 1 mM AgNO3 aqueous solution was heated to boiling, followed by the drop by drop addition of 1.0 mL 0.1 M trisodium citrate solution. The mixture was kept boiling for 16 min under constant stirring, then was cooled down to room temperature showing yellowish gray color. The AgNPs colloid was washed with Milli-Q ultrapure water for three times, in order to remove the excess reducing agent. Meanwhile, the surface of the micrometer-size silica beads was modified for the AgNP coating. Silica beads stock (5.0% w/v) was dried at 60 °C before dispersed in anhydrous ethanol. Then, a 0.2% ethanol solution of (3-aminopropyl)triethoxysilane (APTES) was added into beads suspension (final concentration 0.1%) for 24 h shaking incubation at room temperature. The solution was purified by centrifugation with distilled ethanol at 1055 × g for three times to discard the supernatant. After drying at 60 °C to remove the ethanol and adding double-distilled water, the silica beads with the amino groups modified surface were obtained. Lastly, the AgNP-coated beads were prepared in 1 mL volumes by the continuous agitation of the AgNP colloids and the surface-modified silica beads. The spectral scan of the individual AgNP-coated beads was performed to ensure the complete removal of the excess reducing agent trisodium citrate.Instrumental set-upThe experimental platform integrates dual-trap optical tweezers manipulations and Raman spectroscopic measurements into an inverted microscope with the detailed instrumental layout shown in Supplementary Information (Supplementary Fig. 1). The dual-trap optical tweezers, force detection module, and bright field imaging are built-in the optical tweezers microscopy system (m-trap, LUMICKS, Netherlands). The 1064 nm laser source is splitted into two via a polarizing beamsplitter and focused inside the sample cell by a ×60 water immersion objective with a 1.2 numerical aperture (N.A.) for dual trapping, which are collected in the forward direction by position sensitive detector (PSD) for force analysis. A 780 nm light-emitting diode (LED) and a complementary metal oxide semiconductor (CMOS) camera are integrated into the beam path stereoscopically by a pair of 830 nm long-pass dichroic mirrors for bright field illumination and imaging, respectively. In addition, the 532 nm Raman excitation source (MLL-III-532-50 mW, CNI, China) is reflected by a notch filter (NF533-17, Thorlab, United States) before entering the flex port of the optical tweezers microscope into its stereo double-layer-pathways with a 750 nm long-pass dichroic mirror for the recombination with the original two trapping laser beams. Adjusted by ND filters in the beam path to generate 25 mW input to the microscope, the power density at the Raman excitation focus is estimated to be 3.2 × 106 W/cm2. The backscattered light is reflected by the 750 nm long-pass dichroic mirror and transmitted through the notch filter to enter a spectrometer (IsoPlane SCT-320, 1200 lines/mm, Teledyn Princeton Instrument, United States) with a liquid nitrogen-cooled charge-coupled device (CCD) camera (400B eXcelon, Teledyn Princeton Instrument, United States) at a spectral resolution of 2 cm−1 for spectroscopic measurements. Besides, the flow cell at the sample stage is connected to a microfluidic system driven by compressed air to combine several microfluidic channels into several adjacent laminar fluidic streams at the center chamber during the experimental operations.Manipulation of two AgNP-coated beads and the in situ SERS measurementThe experiments were conducted inside the microfluidic flow chamber with three adjacent laminar fluidic streams from AgNP-coated beads channel, buffer channel, and sample channel, shown in Supplementary Information (Supplementary Fig. 9). Moving the microfluidic flow cell by the equipped two-axis motorized translation micro-stage, different parts in the three adjacent laminar fluidic streams can be set at the place of the trapping laser beams and Raman excitation beam quickly. In the first step, the micro-stage moved the flow cell to a position that the focus of the trapping laser beams fall into the AgNP-coated beads stream to trap AgNP-coated beads. In the next step, the two AgNP-coated beads were trapped at the trapping laser beams and placed in the sample stream by the movement of the flow cell on the micro-stage. Prior to SERS measurement, spectra of solution in the sample stream were recorded as spectroscopic background. Spectra of individual trapped AgNP-coated beads under the Raman excitation spot were scanned to ensure neither intra-beads hotspots nor analyte attachments on the surface of AgNP-coated beads. Then, the two trapped AgNP-coated beads were approached in close proximity and the interparticle gap was placed at the spot for Raman excitation to correlate the relative bead distance to the spectroscopic detection. The motorized trapping laser positions, the camera-based bead distance estimations, and the trapping force detections were utilized to ensure the precise control of the two trapped AgNP-coated beads to create and adjust the SERS active structure under the microscopic visualization. When the two trapped AgNP-coated beads were set at constant separated distance and placed at the Raman excitation spot in the sample steam, the SERS spectra were recorded. All presented spectra were obtained upon the subtraction of the background accordingly and smoothed by Savitzky–Golay filter. Protein samples were dissolved and prepared in 1× PBS buffer, pH 7.4 for SERS measurements. The pH was measured before and after the SERS characterizations to ensure that the protein solutions maintained at physiological pH in the experiments. Ethanol was only used as the analyte in the first experiment whereas not involved in other experiments. The microfluidic flow chamber was cleaned thoroughly between experiments.Reporting summaryFurther information on experimental design is available in the Nature Research Reporting Summary linked to this paper.Supplementary informationSupplementary InformationReporting SummaryPeer Review File | nature communications | [
"Article"
] | [
"Raman spectroscopy",
"Intrinsically disordered proteins",
"Bioanalytical chemistry",
"Nanoparticles",
"Optical manipulation and tweezers"
] | spectroscopy probes vibrations molecules irradiation chemical structures environments1 label-free characterization aqueous surface-enhanced Raman spectroscopy (SERS) detect proteins low concentrations physiological variable conformations intrinsically disordered proteins (IDPs IDPs lack stable secondary tertiary structures self-assemble into oligomers grow into amyloid fibrils11 associated with incurable neurodegenerative diseases12 conversion from monomers to oligomers key early pathological transient nature low population oligomers challenging characterize structural features responsible for cellular toxicity amyloid aggregation14 SERS study on alpha-synuclein related to Parkinson’s scarce18 difficulties conventional nanoparticle-based SERS substrates low efficiency poor reproducibility potential structural perturbation long incubation urgent need to customize SERS approach characterize protein structures low concentration biological significance SERS activity arises from strong electromagnetic field junctions between plasmonic nanostructures “hotspots”21 salt-induced aggregation of silver nanoparticle (AgNP) colloids method form SERS activerandom aggregates lead to unequal enhancement factors fluctuating SERS signals26 challenge improve reproducibility Advancements achieved fabricating SERS substrates with sharp size consistent enhancements analytes low concentrations inconvenient to locate nanometer-size hotspots lowering feasibility efficiency mechanical manipulations create hotspots optical tweezers-assisted SERS developed increase efficiency in aqueous additional trapping laser beams Kall co-workers utilized optical tweezers AgNPs SERS active dimer44 hotspot size location uncertain due to diffraction limit optical Smith co-workers trapped partially Ag-coated silica microparticle as mobile visible SERS probe45 developed by Petrov co-workers interfacial detection of living cells46 explicit control on hotspot SERS performance optical tweezers-based SERS platform awaits developments improve efficiency reproducibility investigation biomolecules dilute environments work approach to visualize control hotspots high SERS enhancements for protein structures conformational fluctuations physiological concentration developing optical tweezers-coupled Raman spectroscopytwo AgNP-coated silica beads trapped SERS interparticle gap manipulation real-time visualization spectroscopic measurements single-molecule SERS window microfluidic flow chamber passing proteins flow rate protein solution-tuned interaction proteins beads native states conformations structural features three proteins (hemoglobin lysozyme bovine serum albumin IDP (alpha-synuclein analyzed dilute solutions SERS characterization alpha-synuclein physiological concentration (1 μM)47 structural variations transient species sensitive SERS platform enables flash spectroscopic snapshots proteins low concentration ensemble averaging quantity investigate IDP oligomers amyloid protein complex biological systems optical tweezers Raman microscope trappable SERS SERS platform combines dual-trap optical tweezers Raman microscope two 1064 nm laser beams one 532 nm Raman probe beam propagate microscope double-layer-pathways combined dichroic mirror enter objective beams focused flow chamber microfluidic systemtwo laser beams manipulate AgNP beads 3D Raman probe beam irradiates gap between beads Fig. 1b backscattered light collected for spectroscopic measurements beams collimated by condenser for force detections camera imaging details Supplementary Fig. 1) Figure 1c real camera image two trapped beads for distance adjustments enables optical manipulation SERS substrates microscopic in situ SERS measurements. controllable SERS probe experiments Raman spectroscopic platform microfluidic set-up Two laser beams manipulate beads one Raman probe beam signals gap real-time camera image beads SEM image gap scale bar 0.1 μm SEM image uniform coating scale bar 1 μm micrographs three measurements AgNP-coated beads chemically fabricated for manipulation visualization SERS modified micrometer-size silica beads with amino groups anhydrous ethanol coated with AgNP colloids AgNPs prepared silver nitrate trisodium citrate washed three times before mixing with silica beadsparameters interaction time stirring rate beads size reactant ratio fine-tuned uniform AgNP coating density 17.6% beads (Fig. stable optical trapping performance Supplementary Fig. 2. Figure 1d SEM image interparticle gap two AgNP-coated beads close SERS activity micrometer-size AgNP-coated beads precise position control sub-nanometer resolution minimizes Brownian motions effective stable spectroscopic measurements Supplementary Figs. 3 4. visualization SERS hotspot between beads easy adjust Raman excitation focus superior dispersed AgNP colloids strategy hotspots between micrometer-size AgNP beads improves operational efficiency simplicity SERS detections real-time controllable dynamic SERS probe created adjusted between two AgNP-coated beads Raman microscope detect SERS signal 1% ethanol aqueous solution Two beads trapped different distance reversibly real-time camera images (Fig. 2a SERS spectra 1 s acquisition time (Figs. 2b d recorded two beads separated by 900 nm Raman excitation spot blank spectrum Fig2b same as spontaneous Raman 1% ethanol Supplementary Fig. 5 no signal two beads approached distance below 100 nm Raman signal enhanced SERS active gap Raman excitation spot peaks at (C–C 1054 (C–O 1094 (CH3 1280 deformation 1458 cm−1 (CH3 asymmetric bending consistent with spontaneous Raman pure ethanol Fig. 5. confirm SERS enhancement hotspot between AgNP-coated beads surface scanned time-series blank spectra single bead Raman spot Fig 6 no intra-bead hotspots spectral features boosted during beads approaching 10 nm step-size from 30 nm to 0 nm SERS active interparticle gap control signal enhancement reverse SERS signal decreased when beads separated 20 nm to 100 nm Fig. 2d distance signal intensities beads separating comparable to approaching Figure 2c intensity ethanol peak at 1458 cm−1 function distance between two trapped beads trapping laser position brightfield image analysisvibrational features ethanol emerged enhanced decreased vanished to beads distance control by optical tweezers two trapped beads pushed at 0 nm SERS enhancement detectable force contact 7) stability SERS interparticle gap 6) 10−9 M 10−11 M Rhodamine B Raman signatures aromatic C–C stretching at ~1360 cm−1 ~1565 cm−1 ~1650 cm−1 Supplementary Fig. 850 theoretical simulation SERS enhancement factor up to 109 single-molecule level detection Supplementary Fig. 8. creation adjustment hotspot achieved at Raman excitation spot platform tunable reproducible SERS enhancements for analytes.Fig. 2Creation adjustment dynamic hotspot SERS measurements 1% ethanol aqueous solution real-time camera images of two AgNP-coated beads trapped different distance SERS spectra of 1% ethanol solution 1 s time intensity ethanol characteristic peak at 1458 cm−1 distance between beads reversible bead positions illustrated SERS spectra of 1% ethanol solution 1 s acquisition time Source data filespectroscopic characterizations native utilized SERS probe flowing hemoglobin aqueous solution without incubation aggregation AgNPs two AgNP-coated beads trapped microfluidic bead channel moved to hemoglobin channel for SERS measurements flow rate fine-tuned minimize interaction between proteins AgNP beads spectroscopic scans show blank spectra intra-bead hotspots hemoglobin attachments two beads approached incremental distance adjusted Raman excitation power optimize SERS signal 100 nM hemoglobin aqueous solution beads distance smaller than 20 nm laser power larger than 10% (2.5 spectral features overwhelmed by peaks amorphous carbon suggesting molecular damages SERS detection window set bead distance 20 nm Raman excitation power 10% 1 s time adjustability optical tweezers Raman spectroscopy preserve protein native states spectroscopic characterizations hemoglobin native statesSERS spectra 100 nM hemoglobin aqueous solution AgNP-coated beads different distance (50–10 Raman excitation power (5–100%). comparison SERS spectra 1 s acquisition time spontaneous Raman spectrum 250 μM hemoglobin 60 s acquisition time 3D stacking plot SERS spectra 100 AgNP-coated beads 20 nm 1 s time hemoglobin peaks 1376 cm−1 4929.15 14.63%) 1594 cm−1 5391.41 16.35%) 50 SERS spectra Source data hemoglobin retains native five SERS spectra 100 nM 1 s acquisition time compared spontaneous Raman spectrum 250 μM 60 s acquisition time Fig. 3c identical frequencies SERS Raman similar protein states oxidation state marker bands peaks 1346 cm−1 1376 cm−1 similar ferrous state ferric Vibrational bands 1568 cm−1 1594 cm−1 6-coordinated high-spin low-spin heme matching positions53 Bands 1089 cm−1 1311 cm−1 1440 cm−1 vinyl group deformation porphyrin ring heme center good.marker bands 5-coordinated high-spin heme at 1494 1572 cm−153 non-native state perturbation metal observed peak assignments hemoglobin Supplementary Table 153,55,56–58 SERS probe two AgNP-coated beads preserved oxidation state coordination number spin state hemes hemoglobin linked native structure function high similarity SERS spectra reproducibility platform two AgNP beads replaced parallel measurements minimize interaction time Figure 3d SERS spectra 100 nM hemoglobin solution 50 measurements vibrational signatures heme center at 1346 1376 1568 1594 cm−1 same Figure 3e f peak intensities 1376 1594 cm−1 relative standard deviation (RSD) 14.63% 16.35%, analysis features Supplementary Fig 11. reproducible spectra prove consistent SERS enhancements parallel measurements two AgNP-coated beads trapped constant distance controllable SERS probe preserve proteins native states generate reproducible SERS spectraspectroscopic characterizations globular structure employed dynamic SERS probe proteins lysozyme bovine serum albumin two AgNP-coated beads trapped 20 nm proteins aqueous solutions replaced experiments Figure 4a SERS spectra 1 μM lysozyme solution 50 experiments vibrational frequencies Raman spectrum peaks 766 1015 1337 1557 cm−1 aromatic residues (tryptophan peak 1450 cm−1 aliphatic residues amide I band 1655 amide III band 1250 cm−1 imply α-helix globular structure lysozyme spectral contributions structures deconvolved α-helix (45.2%) β-sheet (11 random coil (43.5%) Fig. 12 consistent Figure 4b Amide I band distribution 50 SERS spectra 1 μM lysozyme solution 1655 ± 2 cm−1 (0.1% structural stability homogeneity SERS spectra 1 μM BSA Figs. 13 14 structural component assessment (66.5% α-helix 9.5% β-sheet 24.1% random coil supported 50 SERS spectra 1 μM lysozyme Fig.similarities stability reproducibility SERS probe stable compact globular conformation lysozyme sizes globular proteins smaller than bead distance 20 SERS spectra reflect protein structures Fig. 4c small-size sampling SERS measurements unveil protein structural fluctuation Raman. spectroscopic characterizations lysozyme compact globular structure comparison 50 SERS spectra 1 μM lysozyme solution 1 s time Raman 1 mM 5 min acquisition time Amide I band distribution 50 SERS spectra 1 μM mean 1655 cm−1 0.1% RSD ensemble averaging spontaneous Raman measurement lysozyme concentrated solution small-size sampling SERS dilute solution Source data spectroscopic investigation transient species alpha-synuclein physiological SERS characterized structural features alpha-synuclein aqueous solutions alpha-synuclein conformational conversions transient species low population physiological concentration dynamic equilibrium mixture amyloid aggregation CD spectrum 200 μM alpha-synuclein aqueous solution Fig. 5a negative peak 200 nm ensemble conformations random Figspontaneous Raman spectrum 2 mM alpha-synuclein solution shows vibrational features amide III 1249 amide I 1673 cm−1 disordered protein signal 250 μM alpha-synuclein solution Fig. 5b weak low Raman cross-section gentler laser power (25 mW 800 mW previous Raman large sample quantity long detection time structural features alpha-synuclein species overwhelmed conformational ensemble. spectroscopic characterizations disordered protein alpha-synuclein spectrum 200 μM alpha-synuclein aqueous solution Spontaneous Raman spectra 2 mM 250 μM alpha-synuclein solution 10 min acquisition time comparison SERS spectra 1 μM alpha-synuclein solution 1 250 μM 5 min acquisition time Mapping 200 SERS spectra 1 μM alpha-synuclein solution AgNP-coated beads 20 nm 1 s time color bar normalized intensities low high ensemble averaging alpha-synuclein high concentration long low short Source data physiological concentration alpha-synuclein non-aggregated states 1 μM47 below detection threshold Raman spectroscopyexperimental limit detection alpha-synuclein SERS platform 100 nM sensitive SERS approach transient species dilute solutions Supplementary Figs. 15 20. two AgNP-coated beads trapped 20 nm flowing alpha-synuclein Figure 5c three SERS spectra 1 μM alpha-synuclein solution 1 s time 200 experiments Fig. 5d amide I III bands 1 μM alpha-synuclein variations patterns 1 μM lysozyme Fig. 4 co-existence alpha-synuclein species Fig. 5c spectral characteristics blue SERS spectrum 1 μM alpha-synuclein 1650–1657 cm−1 (amide I 1270–1300 cm−1 (amide III α-helix red SERS spectrum 1 μM-synuclein amide I band 1662–1665 cm−1 amide III band 1230–1240 cm−1 associated β-sheet confirmed SERS spectrum 1 μM alpha-synuclein pH 3 Supplementary Fig. 16 β-sheet conformations acidic conditionsamide I band 1671 amide III band 1240–1250 cm−1 SERS spectrum 1 μM alpha-synuclein random coil Raman spectrum vibrational fingerprints Phe (1006 cm−1) deformation aliphatic residues CH2 CH3 (1450 cm−1) alpha uniform SERS spectra 1 μM insensitive protein conformations peak assignments 1 μM alpha-synuclein spectra summarized Supplementary Table 216 subtle features structural details transient species β-sheet oligomers monomers pathological early stage high stability SERS platform resolved structural variations alpha-synuclein transient species physiological concentration 200 parallel SERS measurements 1 μM alpha-synuclein Fig. 5d probability monomers random coil structures high predominant Occasionally transient species α-helix β-sheet structures observed consistent 2 mM alpha-synuclein Raman spectrum 13.3% α-helix 8.4% β-sheet 78.3% random coil Supplementary Fig. 18 250 μM alpha-synuclein SERS spectrum 15.1% α-helix 9.5% β-sheet 75.4% random coil Figpopulation alpha-synuclein transient species low concentration lag phase growth fibrils macroscopic-level Fig. 5e dynamic SERS probe enables small-size sampling alpha-synuclein low concentration solution short accumulation time characterize structural features species statistics structural variations alpha-synuclein distribution secondary structure components comparable ensemble averaging high concentration solution structural variation alpha-synuclein sensitive SERS platform reduce ensemble averaging reveal more structural information IDP transient species investigate behaviors functions IDPs complex biological processes controlled dynamic SERS probe characterize globular proteins disordered proteins dilute solutions optical tweezers-coupled Raman spectroscopy two AgNP-coated beads approached optical tweezers tunable hotspots efficient SERS measurements single-molecule level sensitivity dynamic SERS detection window microfluidic flow chamber flowing proteins native states confirmations verified spectral analysis hemoglobin lysozyme BSA high sensitivity resolved structural variations alpha-synuclein transient species low population concentration crucial understanding initiation amyloid aggregationcontrollable SERS probe optical Raman platform structural information proteins dilute solutions strategy enables control hotspot between micrometer-size AgNP-coated beads SERS efficiency reproducibility aqueous detections optical tweezers offer sub-nanometer spatial resolution sub-piconewton force light-matter interactions plasmonic hotspot method opens structural variations IDPs in dilute solutions challenge biophysics SERS probe oligomeric state amyloidogenic proteins structural details conformational population species amyloid aggregation molecular insights onset neurodegenerative diseases exploit precise force manipulation optical tweezers unfold single protein hotspot resolve structural dynamics beads from Spherotech Silver nitrate trisodium citrate Tris sodium chloride-aminopropyl triethoxysilane lysozyme bovine serum albumin from Sigma-Aldrich Yeast extract Fisher BioReagentsTM Tryptone OxoidTM Hemoglobin Worthington Biochemical Corporation purification human wild type alpha-synuclein overexpressed by E. coli BL21(DE3) with plasmid pET28aCells grown LB medium 50 μg/mL Kanamycin protein expression induced 0.3 mM isopropyl β-D-1-thiogalactopyranoside cell pellet resuspended Tris buffer (25 mM-HCl pH 7.4) lysed sonication centrifugation 30,000 45 min supernatant boiled E. coli proteins 60 min loaded HiPrep DEAE FF 16/10 column sodium chloride gradient target protein protein desalted HiPrepTM 26/10 Desalting column desalted solution loaded HiPrep 16/60 Sephacryl S-100 column purification 25 mM Tris-HCl pH 7.4 buffer Fractions analyzed SDS-PAGE gel targeted protein collected concentrated stored −20 °C concentration alpha-synuclein determined UV-1800 spectrophotometer extinction coefficient 5960 cm−1 M−1 at 276 nm AgNP-coated protocol Protocol Exchange69 silica beads silver nanoparticles synthesized 50 mL 1 mM AgNO3 aqueous solution heated boiling 1.0 mL 0.1 M trisodium citrate solutionmixture boiling 16 min cooled room temperature yellowish gray color AgNPs colloid washed Milli-Q ultrapure water three times remove excess agent surface micrometer-size silica beads modified AgNP coating Silica beads stock (5.0% dried 60 °C dispersed anhydrous ethanol 0.2% ethanol solution (3-aminopropyl)triethoxysilane added beads suspension concentration 0.1%) 24 h incubation solution purified centrifugation distilled ethanol 1055 × g three times discard supernatant drying 60 °C double-distilled water silica beads modified surface obtained AgNP-coated beads prepared 1 mL spectral scan beads removal excess reducing agent trisodium citrate platform dual-trap optical tweezers Raman spectroscopic measurements inverted microscope Supplementary Information dual-trap optical tweezers force detection module bright field imaging built microscopy system 1064 nm laser source splitted two focused sample cell ×60 water immersion objective 1.2 numerical aperture dual trapping collected detector force analysis780 nm diode (LED) metal oxide semiconductor (CMOS camera integrated beam path by 830 nm-pass dichroic mirrors for illumination imaging 532 nm Raman excitation source (MLL-III-532-50 mW CNI China reflected by notch filter (NF533-17 Thorlab before port stereo-layer-pathways 750 nm mirror recombination laser beams Adjusted ND filters 25 mW input power density Raman focus 3.2 × 106 W/cm2. backscattered light reflected by 750 nm mirror transmitted through notch filter spectrometer (IsoPlane SCT-320 1200 Teledyn Princeton Instrument liquid nitrogen-cooled camera (400B eXcelon spectral resolution 2 cm−1 for spectroscopic measurements flow cell sample stage connected to microfluidic system compressed air channels streams center chamber.Manipulation two AgNP-coated beads situ SERS experiments conducted inside microfluidic flow chamber with three fluidic streams Supplementary Information Fig. 9)microfluidic flow cell two-axis motorized translation micro-stage parts in three fluidic streams set at trapping laser beams Raman excitation beam first micro-stage moved flow cell trapping laser beams into AgNP-coated beads stream next two beads trapped at laser beams placed in sample stream measurement spectra solution recorded as spectroscopic background Spectra trapped beads under Raman scanned intra-beads hotspots analyte two trapped beads approached interparticle gap placed at spot for Raman excitation bead distance detection motorized trapping laser positions camera-based bead distance estimations trapping force detections control of two trapped beads SERS active structure two trapped beads set at constant distance placed at Raman excitation spot SERS spectra recorded spectra obtained smoothed by Savitzky–Golay filter Protein samples dissolved prepared in 1× PBS buffer, pH 7.4 for SERS measurements pH measured before after SERS characterizations protein solutions physiological pH Ethanol used as analyte in first experiment not involved in other experimentsmicrofluidic flow chamber cleaned between experiments experimental design Nature Research Reporting Summary.Supplementary SummaryPeer Review File | 49.3 | 0.809741 |
10.1038/s41467-020-18337-4 | PMC7493956 | Trough mouth fans are created via repeated glacigenic sediment transport from ice sheets. Here the authors use 3D seismic reflection data to present a formation model for the North Sea Fan and find that exceptionally large volumes of meltwater may mean that freshwater supply is underestimated during glacial cycles. | Trough mouth fans comprise the largest sediment deposits along glaciated margins, and record Pleistocene climate changes on a multi-decadal time scale. Here we present a model for the formation of the North Sea Fan derived from detailed horizon and attribute interpretations of high-resolution processed 3D seismic reflection data. The interpretation shows that stacked channel-levee systems form up to 400 m thick sedimentary sequences. The channels are elongated and can be traced from the shelf edge towards the deep basin for distances of >150 km, and document long-distance sediment transport in completely disintegrated water-rich turbidite flows. Downslope sediment transport was a continuous process during shelf-edge glaciations, reaching accumulation rates of 100 m/kyr. Our data highlight that exceptionally large volumes of meltwater may discharge to the slopes of trough mouth fans and trigger erosive turbidite flows. We conclude that freshwater supply is likely an underestimated factor for sedimentary processes during glacial cycles. | IntroductionTrough mouth fans are products of repeated glacigenic sediment delivery from former fast-flowing outlets of ice sheets, and act as high-resolution paleoclimate and ice-sheet monitors1–3. The fans have highest sedimentation rates and maximum periods of growth during glacial maxima, whereas they become ice-distal glacimarine environments with low sedimentation rates during interglacials4.The large sediment volumes building these fans are dominated by two sediment types accumulated during very short time periods: The first type are fans characterized by rapidly deposited glacigenic debris flows (GDFs), which indicate Pleistocene periods when eroding ice-streams reached the shelf edge and released the eroded sediment to the upper slopes (Fig. 1a)1,5–11. During shelf-edge glaciations, rapidly deposited glacial sediments are thought to be temporarily stored on the upper slopes, and eventually become unstable and generate GDFs with maximum runouts of >250 km5,6,12,13. These GDFs have been studied using 2D seismic data, which led to the conclusion that they have a lens-shaped geometry in profile view and a lobe-shaped expression in planar view14,15. GDF deposits are further documented to have transparent, generally incoherent acoustic facies with convex tops and pinch-out edges13. A temporary sediment storage at the upper slope with upcoming failure every 34–170 years during the last glacial maximum, and wedges and scars yet to be identified, was suggested for the Bear Island Trough Mouth Fan5,12,16. Sediment remobilization in the form of GDFs was further suggested as a relatively slow and non-disintegrating sediment transport process in very low-viscosity debris flows6,13. The deposits of these flows are poorly-sorted, matrix-supported diamicts with a sand content of up to 40% and higher shear strengths than glacimarine sediments6,9,17. Sediment cores showed that GDFs along Arctic margins have a finer grain-size composition than their Antarctic counterparts3.Fig. 1Two types of sedimentary systems forming trough mouth fans.a Glacigenic debris flow dominated model implying temporarily stored sediment (glacial wedge) and non-disintegrating sediment transport. b Meltwater dominated model implying continuous channelized sediment transport in water-rich flows and surface plumites. Arrows indicating glacial meltwater are conceptual and not absolute values.The second type are meltwater-dominated fans, which previously have been documented on mid-latitude, glacier-influenced margins (Fig. 1b). Large-volume meltwater delivery forms hyperpycnal flows, which result in the deposition of turbiditic sequences2,18–21. Turbidites detected on glacial fans are thus used as a proxy for meltwater delivery13,22. The relative importance of meltwater appears greater at lower than at higher latitudes23. Additional to the turbidite flows, deglacial plumites are released as the increased meltwater generated during the ice-sheet decay generates sediment plumes that also deposit with high sedimentation rates on the upper slopes2,9,14,24–30.Due to glacial erosion and the lack of distinct imprints of ice sheets on the paleo-shelves, trough mouth fan deposits are especially important when reconstructing pre-Weichselian glaciations and understanding glacial-interglacial cycles31. Sedimentological characterization of the uppermost meters of GDF and turbidite deposits are well established for a large variety of trough mouth fans2,3,11,13. However, ice-stream-dominated marine sedimentary systems are lacking extensive, high-resolution data. Depositional processes are thus still relatively poorly understood. The relevance of these marine depositional systems is growing due to economic activities on the seabed and its subsurface, especially in the Arctic region. Similarly, paleo-environmental and paleo-climatic reconstructions are gaining interest, specifically during past episodes of climatic transitions. Three-dimensional (3D) seismic reflection data offer new insights into the geometries, internal architecture and flow mechanisms of sediment remobilization processes31,32. Here, we test if trough mouth fan models derived from 2D seismic data are applicable in a 3D framework. This study aims to understand the nature of sediment delivery across the North Sea Fan during the last glaciation in three dimensions, to relate the sediments to the glacial history of the NE Atlantic margins (Fig. 2), and to discuss the implications for the formation of trough mouth fans.Fig. 2Oblique view of the bathymetry of the North Sea Fan.a The extent of the high-resolution processed 3D seismic data (red line) and the North Sea Fan (white line) are outlined. Locations of the piston cores (red dots), Troll 8903 borehole (yellow dot), seismic profile of Fig. 3 (dashed line) and maps of Figs. 4–6 (white boxes) are shown. Vertical exaggeration is 25× for offshore and 2.5× for onshore domains. Scale bar approximate for central part of figure. b Piston cores used for age correlation (modified after39). Peaks in sand content (grain size >63 μm) indicate iceberg disintegration. IRD ice-rafted debris counts in thousands.The southeast–northwest-oriented North Sea Fan covers an area of c. 110,000 km2 extending from water depths of 300 m at the shelf break into depths of 3500 m in the Norwegian Sea (Fig. 2a). Compared to other trough mouth fans, the North Sea Fan is a clear outlier due to its large fan area3. Sediment transport related to GDFs may have been operating on the North Sea Fan for the last 1.1 Myr, i.e., since the first ice-stream evidence in the Norwegian Channel33. The North Sea Fan received terrigenous sediment from hinterland-to-deep-sea sediment-routing systems with a catchment of c. 215,000 km2, and comprises a sediment volume of c. 32,000 km334,37. Vertical erosion has been estimated to 164 m based on volume backstripping to the catchment34, and has been modelled from 200 to 600 m for the route of the Norwegian Channel Ice Stream35. With a total annual output of 1.1 Gt of sediment (equivalent to 8000 m3/yr per meter width of ice stream front), the Norwegian Channel Ice Stream was an extremely powerful sediment transport agent in the Late Quaternary36,37. Rapid sediment deposition associated with active ice streams resulted in the initiation of GDFs along a gently-dipping seabed (<1°)6, which are deposited within massive clinoform packages consisting of low-amplitude seismic reflections (Fig. 2a). Nygård et al.7 suggested six units of GDFs deposited in the Late Quaternary, and some of these thick units have been remobilized by megaslides38. Becker et al.39 documented several sedimentation pulses characterized by coarse-grained sediment input to the Atlantic Margin during the last glaciation (Fig. 2b). The grounding line of the Norwegian Channel Ice Stream started to retreat from the continental shelf edge by c. 19 ka with an average retreat rate of 450 m/a and the channel was completely deglaciated by c. 17.5 ka40.Results and discussionSeismic stratigraphyThe sediment package related to the last glaciation (Weichselian) is defined by a continuous positive-amplitude reflection at the base (Horizon Base MIS 2) and a continuous negative-amplitude reflection at the top (Horizon 51) (Fig. 3). The up to 450 m thick sediment package outcrops at the seabed of the deeper slopes, and is overlain by weakly layered deglacial and (glaci)marine sediments, up to 70 m toward the shelf break and c. 20 m on the shelf. For the 16,000 km2 of this study, the sediment package comprises a volume of c. 6400 km3. The homogenous seismic facies of the sediment package is interrupted by seven continuous high-amplitude reflections, separating eight sub-units with thicknesses of 20–80 m (Fig. 3). Four intercalating horizons have a negative-amplitude reflection (54, 56–58), and three horizons (52, 53, 55) have a positive-amplitude reflection (Supplementary Fig. 2). Horizons 51 and 54 can be traced into the North Sea, where they are characterized by lower seismic amplitudes.Fig. 3Seismic stratigraphy of the deposits related to the last glaciation (Weichselian, MIS 2) of the North Sea Fan.Eight glacial sub-units are colored in yellow to red, and are indicated by arrows. The top reflections of the sub-units can have a negative-amplitude reflection (54, 56–58), or a positive amplitude reflection (52, 53, and 55). Deep, V-shaped depressions are recognized both at the top of the sub-units and along reflections within the sub-units. Contourites (light grey), Tampen Slide MTD (dark grey), GDFs related to the Saalian glaciation (MIS 6, light yellow to light red), and paleo-shelf break positions (black triangles) are shown. MIS marine isotope stage, MTD mass transport deposit. Profile located in Fig. 2a. V.E. vertical exaggeration. For uninterpreted seismic profile see Supplementary Fig. 2.The sediment packages related to pre-Weichselian glaciations (e.g., GDF II, Saalian) have a lower seismic amplitude response than the sediment sequence related to the last glaciation. GDFs of marine isotope stage 6 (GDF II, Fig. 3) have failed during the Tampen Slide7,38, whose mass transport deposits can still be recognized as a package of deformed sediments onlapping a striking headwall (Fig. 3). The sedimentation of the glacial sediments is associated with a total paleo-shelf break migration of 5 km for the first five sub-units, and 16 km for the last three sub-units. The slope gradients of the paleo-seabeds have been reduced from 1.9° for the first five sub-units to 0.6° for the last three sub-units. Channels crosscut both reflections defining the borders of the sub-units and the reflections within the sub-units.Seismic geomorphologyThree-dimensional seismic data have given rise to the discipline of seismic geomorphology, which is described by Posamentier et al.41 as “the application of analytical techniques pertaining to the study of landforms and to the analysis of ancient, buried geomorphological surfaces as imaged by 3D seismic data”. The mapped horizons of this study reveal multiple sharp, 5–50 m deep and 100–1000 m wide landform systems (Figs. 4 and 5), which locally truncate underlying reflections (Fig. 4a, b). The southeast–northwest oriented landform systems of the North Sea Fan are characterized by the highest seismic amplitudes, and can be traced from the shelf break to the deeper slopes over distances >150 km (Fig. 6, Supplementary Fig. 2). We interpret these landforms as channels, as the morphologies can be traced over large distances along the gently dipping seabed with typical channel geometries. The high-amplitude seismic response of the channels is characteristic for channel infill (Fig. 6). The structure maps show flat terrains between the channels and local wedge-shaped deposits on the channel banks (Figs. 4 and 5a). Seismic attribute maps, however, show well-developed low-amplitude bars between the channels (Figs. 4c and 5). These bars have elongated geometries and a homogenous seismic facies (Fig. 5b). There is no correlation between seafloor relief (e.g., bathymetric lows) and channel occurrence on the evenly dipping paleo-seabeds of the North Sea Fan (Fig. 6). The channels on the North Sea Fan occur with a lateral spacing of 1–20 km, diverge and converge within short distances, and are characterized by a rather low than pronounced sinuosity (Fig. 4).Fig. 4Examples showing the detailed morphology of channels, using zoomed seismic profiles, structure, and horizon attribute maps.a Deeply eroded, wide channels on Horizon 54 (red line). Channel 1 crosscuts underling Horizons 55 and 56. b Channel-levee system on Horizon 55 (red line). The overlying channel (Horizon 54) is eroding into Horizon 55. Channels have harder amplitudes than levees. c Link between channels (infill) and levees on Horizon 57. The channels have high negative amplitudes (very soft), whereas the elongated levees have low negative amplitudes (soft). The location of the maps is shown in Fig. 2a and the stratigraphical position of the different horizons in Fig. 3. V.E. vertical exaggeration. For detailed interpretation of seismic profiles, see Supplementary Fig. 4. Seismic data courtesy of TGS.Fig. 5Seismic geomorphology of the uppermost channel-levee system.a Minimum amplitude extraction of Horizon 51 showing seismic response of channel-levee system. b Seismic profile across channel-levee system highlighting levee geometry and levee facies. Horizon 51 (red line) and Horizon 52 (yellow line) are shown. Seismic data courtesy of TGS.Fig. 6North Sea Fan at the beginning of the last glaciation.3D view of the Horizon Base MIS 2 (Fig. 2) draped by the minimum amplitude extraction in a window of 30 ms. The very soft bands (blue) are interpreted as channels of seismically distinct turbidite flows at the initiation of the last shelf-edge glaciation (t = 23 kyr). The Norwegian Channel Ice Stream, located at the shelf edge, forms two sediment sources at that time (indicated by A and B), from where meltwater turbidites fill the escarpment shaped by the Tampen Slide.Channels with similar dimensions are identified along the seabed and in the subsurface in trough mouth fans of both hemispheres8,42. Glacial gullies, in contrast, are mainly expressed on the upper slopes of fans and have a rather straight expression and V-shaped incisions2. We interpret the reflections characterizing the elongated bars neighboring the channels as submarine levees of two types: a first type are asymmetric, wedge-shaped levees flanking submarine channels, similar to what has been described by Deptuck and Sylvester43 for river-fed submarine fans (Fig. 5). The second, and more common type, are flat-topped levees deposited as uniform blankets on the pre-existing topography (Fig. 4). This levee geometry is different to levee geometries from fluvially derived systems, and has been observed on levees characterizing mud-rich turbidite systems20,44. The seismic character of the levee deposits is similar to the weakly stratified and transparent channel-levees described from the Northwest Atlantic Mid-Ocean Channel45.The high-amplitude reflections defining the channels (Fig. 4c) represent a strong impedance contrast to the underlying homogenous reflections. These contrasts in density and/or velocity most likely indicate coarser-grained sediment in active channels overlying fine-grained sediments. Similar conclusions were drawn on backscatter data on the Belgica Fan, where high backscatter returns from channel beds suggest a hard, eroded surface and/or a relatively coarse-grained component to the downslope flows that cut them8. Levees are characterized by low-amplitude reflections and lack the strong impedance contrast (Fig. 4c). Thus, levees rather indicate fine-grained sediment deposition originating from the suspensive load associated with hyperpycnal flows. Coarse-grained sediment commonly accumulates on the floors or at the mouths of submarine channels, whereas finer-grained sediment preferentially accumulates on channel banks and on adjacent aggradational levees43,44,46. Axial channel deposits have been documented to produce high-amplitude reflections in different fans globally47. In the case of the western Niger Delta slope47, these reflections indicate predominantly sandy channel infill of turbiditic origin, whereas the levees consist of clay-grade sediment48.Based on detailed seismic interpretation, we conclude that the several 100-m-thick sediment sequence related to the last glaciation is dominated by channels and not, as previously suggested, by debris lobes. Extensive 3D seismic data are thus fundamental to correctly interpret glacial processes and its deposits.Implications on sedimentation and ice-stream activityThe sediments of the last-glacial package of the North Sea Fan derive from subglacially transported sediments, which were deposited when the Norwegian Channel Ice Stream reached the shelf edge6,36,37. The eight sub-units within the uppermost sediment package indicate that the ice stream oscillated eight times during the last glaciation (Fig. 3). Sediment has meanwhile continuously been transported downslope within commonly observed channel systems (Fig. 4), and fills the slide escarpment formed by the Tampen Slide (Fig. 6), resulting in a shelf-break migration of c. 5 km/kyr (Fig. 3). The data show that the ice stream delivered sediment from multiple sources and that the northern part of the fan was active first during the last glaciation (Fig. 6).The growth and decay of the Norwegian Channel Ice Stream resulted in highly variable rates of sediment delivery to the continental margin39. The geometry and occurrence of channels identified at multiple levels within the sediment package related to the last glaciation (Fig. 3) document that the sediment delivery from the Norwegian Channel Ice Stream had a continuous pattern. The orientation of the channels does not significantly shift during the last glaciation (Fig. 6, Supplementary Fig. 3). Our data show a rather uniform sedimentation pattern within the same glaciation, while a shift of sediment depocenters has been observed for different glaciations on the North Sea Fan49,50. A relatively continuous subglacial release of material has previously been suggested as the origin for GDFs of this package15. Although straight or sinuous gullies related to the last glaciation have been described using side-scan sonar of the North Sea Fan15, the seabed reflection of the 3D seismic data is rather mirroring deglacial than glacial processes (Figs. 2a and 3).The Norwegian Channel Ice Stream was located at or close to the shelf break allowing dense sediment flows to develop from meltwater. The ice-stream-fed channels have a minimum age of c. 17.5 ka, as by that point the Norwegian Channel was completely deglaciated40. Most channels were formed before 18.7 ± 0.2 kyr, marking the retreat of the Norwegian Channel Ice Stream from the shelf edge37,39. The correlation of the five horizons with sediment cores from the distal part of the North Sea Fan indicate an age of 19–23 kyr for the sediment package related to the last glaciation39 (Fig. 2b). Deposited within 4 kyrs, the up to 450 m thick package correlates with an average sedimentation rate of c. 100 m/kyr and a sediment flux of 1500 km3/kyr for the study area. Sedimentation rates in the most proximal core, just 2 km southwest of the sequence pinch-out, range from 0.5 to 1.5 m/kyr during the last glacial cycle39. The sedimentation rates within the North Sea Fan are thus 100 times higher compared to the rates outside the areas affected by direct ice-stream sedimentation. In line with previous studies (e.g., ref. 6), the sediment supply associated with the channels outshines simultaneous glacimarine sedimentation. We further suggest higher deglacial sedimentation rates directly on the North Sea Fan, where the deglacial sediment package is up to 70 m thick (18 m/kyr), compared to cores from areas outside the fan (1 m/kyr). Similarly, Lucchi et al.27 calculated extreme sedimentation rates of 34 m/kyr for the deglacial plumites from the upper-slope area of the Storfjorden Trough Mouth Fan in a period of less than 150 years.Meltwater turbiditesThe low-sinuous channels, whose downslope terminations expand over the extent of our data, indicate long-distance down-slope sediment bypassing. Long runout distances on low-gradient slopes were previously explained by excess pore fluid pressures51 or the incorporation of a thin layer of ambient water underneath a subaqueous debris flow52. Muddy turbidites with long runout distances and feeding deep-sea fans have been documented in turbidite systems all over the world53–56. Based on the channel morphology and extent, we suggest sediment disintegration in water-rich flows as the dominating flow mechanism (Fig. 6). As channels and closely associated overbank deposits dominate the stratigraphy of the North Sea Fan, we suggest the fan to be maintained generally by glacial meltwater-sourced flows. We conclude that meltwater is an underestimated factor for the formation of trough mouth fans.Sediment-carrying meltwater events can lead to the generation of erosive hyperpycnal flows56,57. Erosive sediment transport is shown by deep channels crosscutting underlying reflections (Fig. 4a, b). The flows have been more erosive at the uppermost slopes, where the channels are deepest. A lower degree of erosion observed by discontinuous channels in deeper waters of this study area could be supported by subaqueous turbidites running over antecedent turbidite deposits with no detectable remobilization, as shown for debris flows in experiments by Mohrig et al.52.Meltwater can transport large quantities of lithogenic particles derived from glacial erosion58, and large turbidity currents are documented to have lost their freshwater after distances of up to 300 km59. An increased meltwater input from the Norwegian Channel Ice Stream could activate sediment-downslope transport in turbidity currents, and thereby explaining the long runout distances. The rapidly deposited sediment sequences along glaciated margins originally consist of poorly sorted and unstable glacial material (e.g.,9,23), but long-distance channelized sediment transport in turbidites might favor grain-size fractionation of these sediments45. The submarine channel network controls sediment distribution in the deep-water depositional system, and depending on transport distance and channel proximity, grain size might considerably vary along glacial reflections. Sediment cores collected from turbidite channels on the Squamish Delta contain multiple units of massive sands, with thicknesses of 1–2 m60. Such sand beds are resolvable by the seismic data used in this study, and high-amplitude reflections characterizing the channels can reflect sandy deposits. However, the acoustically transparent character of the sediment package between the high-amplitude reflections excludes the existence of thick sand beds, and indicates mainly silty and clayey deposits. A transparent seismic signature related to muddy material has been suggested for glacial fans in the Norwegian Channel61.Model for the North Sea Fan during the last glaciationThe presence of frequent well-defined stacked channel-levees in the proximal part of the North Sea Fan demonstrates that sediment has been transported downslope within commonly observed channel systems throughout the last glaciation (Fig. 4). We propose that the Norwegian Channel Ice Stream rapidly delivered eroded sediment (Fig. 6), resulting in multi-sourced turbidite systems on the fan. Turbidity currents linked to downslope flow of sediment-laden meltwater from the shelf edge could have directly formed the channels, which then functioned as conduits for focused turbidity current flow to the deep basin (Fig. 7)11,45. The bedload of such glacial turbidites can consist of medium sand and coarser sediments45. Additional to the turbidites, the sedimentation on the North Sea Fan was influenced by suspension settling from turbid-surface plumes released at the grounding line, which accumulated sediment in a clayey grain-size fraction (Fig. 7). Settling plume events are documented to trigger long-runout turbidity currents themselves on the Squamish Fan62.Fig. 7Conceptual model for sedimentation during the last glaciation (MIS2).Meltwater turbidites and turbid-surface plumites are the dominating processes, and result in channel-levee systems on the North Sea Fan. Continuous sediment transport in water-rich flows build an up to 450 m thick sediment sequence in the time period of 23 to 19 ka. Dimensions are approximate, and given in km.Different studies show that the period around the last glacial maximum was characterized by major input of meltwater events22,26,63, and that trough mouth fans have highest growth in these periods4. Based on the glaciation history of the Norwegian Channel39,40,64, we suggest that the high-density, sand-rich turbidity currents originate from the Norwegian Channel Ice Stream in the time period between 23 and 19 ka. These turbidity currents occurred during major meltwater discharges at the beginning of slope sedimentation. The last glacial maximum period is thus characterized by major input of meltwater events. As the turbidites are related to massive meltwater delivery, the turbidite intervals could correspond to short warmer periods, and a Norwegian Channel Ice Stream undergoing several smaller collapses within or after the last glacial maximum. Freshwater could largely be provided by seasonal meltwater discharge and iceberg calving, processes previously suggested by sedimentary records in other glaciated margins65.The new model suggests turbid glacial meltwater driven channelized sediment supply to be the dominating process shaping the mid-latitude North Sea Fan, and the Norwegian Channel to have acted as a major outlet for meltwater (Fig. 7). We suggest that coarse-grained turbidites at the beginning of a shelf-edge glaciation (Fig. 6), and delivery of coarser-grained material during a glaciation (Fig. 3), are the causes for the observed high-amplitude reflections. The data show that rapid glacial sedimentation is a continuous process during glaciations, with sediment accumulation two magnitudes higher in areas affected by channels. Voluminous meltwater production in periods with active ice streams at or close to the shelf break could increase turbidity current activity on the fan. Long-distance channelized sediment transport along gently dipping seabeds could explain kilometer thick glacial sequences hundreds of kilometers away from the shelf break.Implications for trough mouth fansThe use of high-quality 3D seismic data allowed new interpretations of the style and variation of ice-proximal sedimentation on the North Sea Fan. Our study shows that sedimentation related to glacial meltwater played a fundamental role in the construction of the mid-latitude North Sea Fan compared to previously suggested sediment deposition by GDFs6,7. The stratigraphy of the mid-latitude North Sea Fan records a strong meltwater signal for the last glaciation, which distinguishes the fan from high-latitude fans with predominating low-water-content GDF deposition5. A strong meltwater delivery has also been suggested for mid-latitude depocenters of the Laurentian Fan and the Disko Fan2,20, and for deglaciation in the high-latitude Storfjorden Fan27. The timing of the increased meltwater discharge on the North Sea Fan correlates with turbidite deposits observed in both the Notre Dame and the Hawke Fans, which are associated to periods of major meltwater supply from 29 to 17 ka22. The volume and abundance of subglacial meltwater is largely controlled by strain heating and the geothermal heat flux beneath an ice sheet66,67. Freshwater fluxes are challenging to quantify and not necessarily correlated with fluxes of iceberg rafted debris65. Sediment core analysis of Becker et al.39 proves that pulses of iceberg rafted debris not exclusively occurred between 23 and 19 ka, and we conclude that meltwater and iceberg supply to the NW Atlantic are asynchronous processes. A strong meltwater signal suggested by the glacial turbidite systems indicates that the North Sea Fan was probably a warmer environment during full-glacial and deglacial conditions compared with the more northerly glacial depocenters.Gales et al.3 suggested that ice-sheet drainage basin size influences the abundance and volume of subglacial meltwater released from beneath an ice sheet, and that turbidity current activity would increase in areas of greater meltwater. Turbiditic sedimentation is dominating the North Sea Fan, which is characterized by a large drainage basin. Similar process could dominate fan evolution of the Prydz Channel Fan and the Crary Fan in Antarctica, which both have ice-stream drainage basin areas of >1,000,000 km2. High-resolution 2D seismic profiles and sediment cores often only cover the very uppermost meters on trough mouth fans2,3. The 3D seismic data of the North Sea Fan evidence that sedimentation of the complete last-glacial package took place through an overbank relationship to the channels. Glacial turbidites as a dominating trough mouth fan process have been documented for thinner sediment packages from other glaciated margins (e.g.,2), but never for a thickness comparable to the late glacial sequence of the North Sea Fan. In contrast to other settings2,11,68, we do not observe any association between downslope process and slope gradient. A strong meltwater contribution could further imply that the Norwegian Channel Ice Stream had not to be positioned exactly at the shelf break during the deposition of the thick sediment package. The absence of shelf-edge glaciations is thus not excluding high sediment accumulation rates. Glacial sediment could instead be delivered by subglacial meltwater from an ice margin that was no longer at the shelf edge, as suggested for the Donegal Barra Fan24,69.The detailed interpretation of the 3D seismic data shows that high-energetic hyperpycnites deposit up to 450 m of glacial turbidites, whereas low-water content GDFs could not form. Implied sedimentation rates of 100 m/ka outpace turbidite sedimentation rates of 1–3 m/ka in other fans22. Large volumes of rhythmic turbidites along glaciated margins are partly related to subglacial outbursts (e.g.,20). Therefore, the North Sea Fan might record multiple large outburst events during the last glacial maximum. The data further indicate that rapid meltwater-driven sedimentation dominate all of the last glacial sequence. Such a sedimentation pattern is in contrast to observations from other ice sheets, where sedimentation is changing from low-energy at the beginning to increased discharges at a later stage of the glacial cycle13.The extensive 3D seismic data set presented here allows better assessments of the significance of meltwater pulses during glaciations, and is thus relevant to more strongly constrain glacial and deglacial ice-sheet evolution. Strong sediment-laden turbidity current systems dominating glacial sedimentation are applicable for glacial settings with potential of high meltwater delivery and catchments with sediment available for erosion. Differences in the mode of sediment delivery to the continental slope and deep-sea basin strongly affected the evolution of the North Sea Fan. These differences most likely result in a distinct morphology of mid-latitude fans and their high-latitude counterparts.MethodsSeismic dataThe study is based on seismic interpretation of 16,000 km2 of high-resolution, industry-standard processed 3D seismic reflection data collected in 2017 from the proximal North Sea Fan (Fig. 2a). The data were collected by TGS using a triple-sourced airgun with a volume of 3000 in3 and a shot point interval of 12.5 m. The acquisition consisted of twelve 8100 m long streamers, which were separated by 112.5 m. A high-resolution volume at 2 ms sample rate and 6.25 × 18.75 m binning was designed to increase the resolution of the shallow stratigraphy. The data for this volume have been cut at minimum of twice seabed time and 5000 ms two-way time before zero-phasing. The 3D seismic reflection data allow to image the buried sediment packages in a resolution of 2 m vertically and in a bin size of 20 × 5 m horizontally.Seismic interpretationSix seismic horizons within the last glacial sediment package were picked with an in-line spacing of 150 m, followed by gridding, horizon attribute extraction, sediment volume calculations, and seismic geomorphological interpretation. Seismic attributes, such as the minimum and maximum amplitudes of prominent reflections, provide additional geological information that cannot be extracted from structure maps and allow an improved geological process interpretation. The interpreted surfaces are characterized by both hard and soft reflections (Supplementary Fig. 1). P-wave velocities of 1500 and 1700 m/s were used for time-to-depth conversion of the water column and the last glacial sediment package, respectively7. We merged interpretations of regional 2D and 3D seismic data with bathymetric data from GEBCO 2014 to image the bathymetry of the study area (Fig. 2a).ChronostratigraphyThis study focuses on the uppermost 500 m below the seabed, and follows the chronostratigraphy previously established for this sediment package, which is dominated by GDFs related to the last glacial cycle7. Glacial chronologies previously established by piston cores next to the North Sea Fan cover the last glacial cycle39, and core ties allow constraining the ages of the sediment package of this study (Fig. 2b).Supplementary informationSupplementary InformationPeer Review File | nature communications | [
"Article"
] | [
"Hydrology",
"Physical oceanography"
] | mouth fans products glacigenic sediment delivery from high-resolution paleoclimate ice-sheet highest sedimentation rates growth during glacial maxima low sedimentation during interglacials4 large sediment volumes dominated by two sediment types short first rapidly deposited glacigenic debris flows Pleistocene eroding ice-streams released sediment upper slopes shelf-edge glaciations rapidly deposited sediments stored upper slopes unstable generate GDFs maximum runouts >250 km5,6 GDFs studied 2D seismic data lens-shaped geometry profile lobe-shaped expression planar GDF deposits transparent incoherent acoustic facies convex tops pinch-out temporary sediment storage at upper slope failure every 34–170 years during last glacial maximum wedges scars for Bear Island Trough Mouth Sediment remobilization GDFs slow non-disintegrating sediment transport in low-viscosity debris flows6 deposits poorly-sorted matrix-supported diamicts sand content up 40% higher shear strengths than glacimarine sediments6,9GDFs Arctic margins finer grain composition Antarctic. 1Two sedimentary systems trough mouth fans Glacigenic debris flow stored sediment non-disintegrating transport Meltwater continuous channelized sediment transport water-rich flows surface plumites Arrows glacial meltwater conceptual not second meltwater-dominated fans documented mid-latitude glacier-influenced margins Large-volume meltwater delivery forms hyperpycnal flows deposition turbiditic Turbidites glacial fans proxy meltwater importance meltwater lower higher deglacial plumites released increased ice-sheet decay sediment plumes high sedimentation rates upper glacial erosion lack imprints ice paleo-shelves trough mouth fan deposits important reconstructing pre-Weichselian glaciations glacial-interglacial Sedimentological characterization uppermost meters GDF turbidite deposits established trough mouth ice-stream-dominated marine sedimentary systems lacking high-resolution data Depositional processes poorly understood relevance marine systems economic activities seabed subsurface Arctic regionpaleo-environmental-climatic reconstructions gaining interest climatic transitions Three-dimensional (3D seismic reflection data offer insights geometries mechanisms sediment remobilization test trough mouth fan models 2D 3D framework study aims understand sediment delivery North Sea Fan last glaciation relate sediments to glacial history NE Atlantic margins discuss implications formation trough mouth fans. bathymetry North Sea Fan high-resolution 3D seismic data North Sea Fan Locations piston cores Troll 8903 borehole seismic profile Fig. 3 maps Figs. 4–6 Vertical exaggeration 25× offshore 2.5× onshore Scale bar central part Piston cores age correlation Peaks sand content size >63 μm indicate iceberg disintegration ice-rafted debris counts in thousands southeast–northwest-oriented North Sea Fan covers 110,000 km2 300 m into 3500 m Norwegian Sea North Sea Fan outlier large fan Sediment transport GDFs North Sea Fan since first ice-stream evidence Norwegian North Sea Fan received terrigenous sediment from hinterland-deep-sea sediment-routing systems catchment215,000 km2 sediment volume 32,000 km334 Vertical erosion estimated 164 m volume backstripping modelled 200 to 600 m Norwegian Channel Ice annual output 1.1 Gt sediment 8000 m3/yr per meter width ice stream Norwegian Channel Ice Stream powerful sediment transport agent Late Quaternary36 Rapid sediment deposition GDFs-dipping seabed deposited clinoform packages low seismic reflections (Fig. Nygård et al six units GDFs Late Quaternary remobilized by megaslides38 Becker et al documented sedimentation pulses coarse-grained sediment Atlantic Margin last glaciation. grounding line Norwegian Channel Ice Stream shelf 19 ka average retreat rate 450 m/a deglaciated 17.5 sediment package last glaciation continuous positive-amplitude reflection base negative-amplitude reflection top 51) (Fig. 3) up to 450 m thick sediment package outcrops seabed slopes overlain layered deglacial (glaci)marine sediments 70 m toward shelf break 20 m on shelf 16,000 km2 sediment package 6400 km3homogenous seismic sediment interrupted by seven high-amplitude reflections separating eight sub-units thicknesses 20–80 m (Fig. 3) Four negative-amplitude reflection (54 56–58) three (52 53 55) positive-amplitude Horizons 51 54 traced North Sea lower seismic amplitudes 3Seismic stratigraphy deposits last glaciation (Weichselian MIS 2) North Sea Fan.Eight glacial sub-units colored yellow to red indicated by top reflections negative (54 56–58) or positive (52 53 55). Deep V-shaped depressions recognized top Contourites Tampen Slide MTD GDFs Saalian glaciation paleo-shelf break positions shown marine isotope stage MTD mass transport deposit Profile Fig. 2a profile Supplementary Fig. sediment packages pre-Weichselian glaciations II lower seismic amplitude response last glaciation GDFs marine isotope stage 6 failed during Tampen Slide7,38 deposits deformed sediments headwall (Fig.sedimentation glacial sediments associated paleo-shelf break migration 5 km first five 16 km last three slope gradients paleo-seabeds reduced 1.9° first five to 0.6° last three Channels crosscut reflections.Seismic geomorphologyThree-dimensional seismic data seismic geomorphology techniques study landforms analysis ancient geomorphological surfaces 3D seismic mapped horizons reveal sharp 5–50 m deep 100–1000 m wide landform systems (Figs. 4 5) truncate reflections southeast–northwest oriented landform systems North Sea Fan highest seismic amplitudes traced shelf break to deeper slopes distances >150 km (Fig. 6 landforms as channels traced distances seabed channel geometries high-amplitude seismic response characteristic channel infill (Fig. 6) structure maps show flat terrains between channels wedge-shaped deposits channel banks (Figs. 4 Seismic attribute maps show-developed low-amplitude bars between channels elongated geometries homogenous seismic facies no correlation between seafloor reliefbathymetric lows channel occurrence paleo-seabeds North Sea Fan (Fig. 6) channels lateral spacing 1–20 km diverge converge short distances low sinuosity (Fig. 4) channels zoomed seismic profiles structure horizon attribute maps eroded channels Horizon 54 1 crosscuts Horizons 55 56 Channel-levee system Horizon 55 overlying channel 54 eroding Horizon 55 Channels harder amplitudes levees Link channels levees Horizon 57 channels high negative amplitudes elongated levees low negative amplitudes location maps Fig. 2a stratigraphical position Fig. 3. Fig. 4. Seismic data TGS.Fig. 5Seismic geomorphology uppermost channel-levee system Minimum amplitude extraction Horizon 51 response Seismic profile channel-levee levee geometry facies Horizon 51 Horizon 52 TGS 6North Sea Fan last glaciation view Horizon Base MIS 2 (Fig. 2) minimum amplitude extraction 30 mssoft bands (blue as channels seismically distinct turbidite flows last shelf-edge glaciation (t = 23 kyr). Norwegian Channel Ice Stream at shelf edge forms two sediment sources A turbidites fill escarpment Tampen Slide.Channels similar dimensions along seabed subsurface mouth fans hemispheres8 Glacial gullies upper slopes straight expression V-shaped reflections elongated bars channels as submarine levees first asymmetric wedge-shaped levees Deptuck second flat-topped levees uniform blankets pre-existing topography (Fig. 4) levee geometry different to fluvially systems observed on mud-rich turbidite seismic character levee deposits similar to weakly stratified transparent channel-levees Northwest Atlantic Mid-Ocean high-amplitude reflections channels (Fig. 4c represent strong impedance contrast to homogenous reflections contrasts indicate coarser-grained sediment active channels fine-grained sediments Similar conclusions backscatter data Belgica Fan high backscatter returns suggest hard eroded surface coarse-grained component downslope flowsLevees low-amplitude reflections lack strong impedance contrast (Fig. indicate fine-grained sediment deposition load hyperpycnal flows Coarse-grained sediment accumulates floors mouths submarine channels finer-grained channel banks aggradational Axial channel deposits produce high-amplitude reflections in fans western Niger Delta reflections indicate sandy channel infill turbiditic origin levees clay-grade seismic interpretation 100-m-thick sediment sequence last glaciation dominated by channels not debris lobes Extensive 3D seismic data fundamental glacial processes deposits.Implications on sedimentation-stream sediments last-glacial package North Sea Fan from subglacially transported sediments deposited Norwegian Channel Ice Stream reached shelf eight sub-units ice stream oscillated eight times last glaciation Sediment transported downslope fills slide escarpment Tampen Slide shelf-break migration c. 5 km/kyr ice stream delivered sediment from multiple sources northern part fan active last glaciation growth decay Norwegian Channel Ice Stream variable rates sediment delivery to continentalgeometry occurrence channels sediment package last glaciation (Fig. 3) sediment delivery Norwegian Channel Ice Stream continuous pattern orientation shift glaciation (Fig. 6 data show uniform sedimentation pattern glaciation shift sediment depocenters observed for different glaciations North Sea continuous subglacial release suggested origin for GDFs straight sinuous gullies glaciation described side-scan sonar seabed reflection 3D seismic deglacial glacial processes (Figs. 2a 3) Norwegian Channel Ice Stream shelf break dense sediment flows ice-stream-fed channels minimum age 17.5 ka deglaciated40 channels formed before 18.7 ± 0.2 kyr retreat from shelf correlation five horizons with sediment cores North Sea Fan age 19–23 kyr sediment package last (Fig. within 4 kyrs 450 m thick package correlates average sedimentation rate 100 m/kyr sediment flux 1500 km3/kyr study area Sedimentation rates proximal core 2 km southwest sequence-out range 0.5 to 1.5 m/kyr last glacial sedimentation rates North Sea Fan 100 times higher compared outside areas ice-stream sedimentation previous studiessediment supply outshines glacimarine sedimentation higher deglacial sedimentation rates North Sea Fan sediment 70 m thick (18 m compared outside (1 m/kyr). Lucchi et al calculated extreme sedimentation rates 34 m/kyr deglacial plumites upper-slope Storfjorden Trough Mouth Fan less than 150 years.Meltwater low-sinuous channels indicate long-distance sediment bypassing Long runout distances explained excess pore fluid thin layer ambient water subaqueous debris Muddy turbidites long runout distances feeding deep-sea fans documented in turbidite systems sediment disintegration in water-rich flows dominating flow mechanism channels overbank deposits dominate North Sea Fan fan maintained by glacial meltwater flows underestimated factor formation trough mouth fans.Sediment-carrying meltwater erosive hyperpycnal flows56 Erosive sediment transport deep channels crosscutting reflections (Fig. 4a flows more erosive uppermost slopes lower degree erosion discontinuous channels deeper waters subaqueous turbidites over turbidite deposits no remobilization Mohrig et alMeltwater lithogenic particles from glacial turbidity currents freshwater after 300 increased from Norwegian Channel Ice Stream activate sediment-downslope transport long runout distances rapidly deposited sediment sequences along glaciated margins poorly sorted unstable glacial material long-distance sediment transport grain-size fractionation submarine channel network controls sediment distribution deep-water transport distance channel proximity grain size vary glacial reflections Sediment cores from turbidite channels on Squamish Delta contain massive sands thicknesses 1–2 sand beds resolvable by seismic data high-amplitude reflections reflect sandy deposits acoustically transparent sediment reflections excludes thick sand beds indicates silty clayey deposits transparent seismic signature to muddy material suggested for glacial fans in Norwegian North Sea Fan last stacked channel-levees in North Sea Fan sediment transported downslope last glaciation Norwegian Channel Ice Stream delivered eroded sediment multi-sourced turbidite systems on fan Turbidity currents to sediment-laden meltwater formed channels conduits for turbidity flow to deep basinbedload glacial turbidites medium sand coarser sediments45 sedimentation North Sea Fan influenced by suspension from turbid-surface plumes grounding line sediment (Fig. 7) Settling plume events trigger long-runout turbidity currents Squamish. model sedimentation last glaciation.Meltwater turbidites turbid-surface plumites dominating result channel-levee systems North Sea Fan sediment transport water-rich flows build 450 m thick sediment sequence 23 to 19 ka Dimensions approximate in km last glacial maximum major meltwater trough mouth fans highest growth high-density sand-rich turbidity currents originate from Norwegian Channel Ice Stream between 23 19 ka during major meltwater discharges slope sedimentation last glacial maximum major meltwater turbidites related to meltwater delivery intervals short warmer periods Norwegian Channel Ice Stream smaller collapses Freshwater provided by seasonal meltwater discharge iceberg calving new model suggests turbid glacial meltwater driven channelized sediment supply dominating mid-latitude North Sea Fan Norwegian Channel major outlet for meltwater (Fig. 7)coarse-grained turbidites shelf-edge glaciation (Fig. 6) material during glaciation (Fig. 3) causes high-amplitude reflections data show rapid glacial sedimentation continuous during glaciations sediment accumulation higher in areas affected channels Voluminous meltwater production active ice streams break increase turbidity fan Long-distance channelized sediment transport seabeds explain kilometer thick glacial sequences.Implications for trough mouth high-quality 3D seismic data interpretations ice-proximal sedimentation North Sea Fan study shows sedimentation related glacial meltwater construction mid-latitude North Sea Fan stratigraphy records strong meltwater signal last glaciation distinguishes from low strong meltwater delivery suggested for Laurentian Fan Disko deglaciation Storfjorden increased meltwater discharge correlates with turbidite deposits Notre Dame Hawke Fans major meltwater supply 29 to 17 ka22 volume abundance subglacial meltwater controlled by strain heating geothermal heat flux beneath ice Freshwater fluxes not correlated with iceberg rafted debris65 Sediment core analysis Becker et al.iceberg debris between 23 19 ka meltwater iceberg supply NW Atlantic asynchronous strong meltwater signal glacial systems North Sea Fan warmer during full deglacial northerly depocenters ice-sheet drainage basin size influences subglacial meltwater turbidity activity in greater meltwater Turbiditic sedimentation North Sea Fan large drainage basin Similar process Prydz Channel Fan Crary Fan in Antarctica ice-stream drainage basin areas >1,000,000 km2. 2D seismic profiles sediment cover uppermost meters on trough mouth 3D seismic data North Sea Fan sedimentation last-glacial package overbank relationship channels Glacial turbidites documented for thinner sediment packages from other glaciated margins never thickness late glacial sequence North Sea Fan association between downslope process slope gradient strong meltwater contribution Norwegian Channel Ice Stream not at shelf break during deposition thick sediment package absence of shelf-edge glaciations not high sediment accumulation rates Glacial sediment could delivered by subglacial meltwater from ice margin shelf edge Donegal Barra3D seismic data shows high-energetic hyperpycnites deposit 450 m glacial turbidites low-water GDFs form sedimentation rates 100 m/ka outpace turbidite sedimentation rates 1–3 m/ka other rhythmic turbidites along glaciated margins related to subglacial outbursts North Sea Fan might record large outburst events last glacial maximum rapid meltwater-driven sedimentation last glacial sequence other ice sheets extensive 3D seismic data allows meltwater pulses during glaciations relevant glacial deglacial ice-sheet evolution Strong sediment-laden turbidity systems applicable for settings high meltwater delivery erosion Differences sediment delivery continental slope deep-sea basin evolution North Sea Fan result in distinct morphology mid-latitude fans high-latitude counterparts study 16,000 km2 high-resolution 3D seismic reflection data collected 2017 North Sea Fan data collected TGS triple-sourced airgun 3000 in3 shot point interval 12.5 m acquisition twelve 8100 m long streamers separated by 112.5 mhigh-resolution volume 2 ms sample rate 6.25 × 18.75 m binning shallow stratigraphy data cut seabed time 5000 ms two-way time zero-phasing 3D seismic reflection data buried sediment packages 2 m vertically size 20 × 5 m horizontally interpretationSix horizons last glacial sediment package-line spacing 150 m gridding attribute extraction sediment volume calculations geomorphological interpretation attributes minimum maximum amplitudes reflections geological information geological process interpretation interpreted surfaces hard soft reflections P-wave velocities 1500 1700 m/s time-to-depth conversion water column last glacial sediment package merged regional 2D 3D seismic data bathymetric data GEBCO 2014 bathymetry study area. focuses uppermost 500 m below seabed follows chronostratigraphy sediment dominated GDFs last glacial Glacial chronologies piston cores North Sea Fan last glacial constraining ages sediment package (Fig. 2b).Supplementary Review | 49.4 | 0.535723 |
10.1038/s41467-020-16874-6 | PMC7303143 | The properties of a polymer are known to be intrinsically related to its molecular weight distribution. Here the authors use a design to synthesis protocol for producing a targeted molecular weight distribution with a computer controlled tubular flow reactor. | The properties of a polymer are known to be intrinsically related to its molecular weight distribution (MWD); however, previous methodologies of MWD control do not use a design and result in arbitrary shaped MWDs. Here we report a precise design to synthesis protocol for producing a targeted MWD design with a simple to use, and chemistry agnostic computer-controlled tubular flow reactor. To support the development of this protocol, we constructed general reactor design rules by combining fluid mechanical principles, polymerization kinetics, and experiments. The ring opening polymerization of lactide, the anionic polymerization of styrene, and the ring opening metathesis polymerization are used as model polymerizations to develop the reactor design rules and synthesize MWD profiles. The derivation of a mathematical model enables the quantitative prediction of the experimental results, and this model provides a tool to explore the limits of any MWD design protocol. | IntroductionA polymer’s molecular weight distribution (MWD) impacts material properties such as processability, mechanical strength, and morphological phase behavior1–6. This correlation is general across all polymers and has motivated the development of many synthetic and process techniques. Much of the work regarding MWDs, has primarily focused on the development of polymerization methods to access polymers with narrow MWDs (referred to as controlled polymerization)7,8. Controlled polymerizations have revolutionized the synthesis of advance materials, however, the control they offer does not directly provide tunability for broad MWDs which are advantageous for many applications9. In fact, broad distributions remain a staple in industry2,10. For example, polyethylene produced by the Philipps catalyst has a dispersity >10, as this provides the ideal balance of fast processability and high mechanical strength2,11. Moreover, the latest improvements to polyolefins have been the tailoring MWDs for specific applications2. In addition, emerging areas of applications for thermoplastics (e.g., 3d printing) demand high mechanical performance while maintaining ease of processing, which can be achieved through tuning of the MWD12,13. While coarse MWD tuning is common practice, independent control over MW, MWD breadth, and MWD shape remains challenging. Thus, achieving high precision tailored MWDs remains a topic of high interest for studying material properties, as well as tailoring materials for specific applications.Early approaches to engineered MWDs relied on blending distinct batches of polymers with known MW1,2,5,14–18. This simple approach requires synthesizing a large number of batches to reduce the multimodality of the final MWD (Fig. 1a). More recently, this has been streamlined by performing the polymerization with multisite catalysis or with cascade reactors2. However, multimodal MWD has been shown not to exhibit the same properties as a smooth MWD, making them unsuitable for several applications, e.g., macrophase separation1,16. This limitation motivated the development of synthetic methodologies for accessing polymers with tunable and smooth MWDs. Initial work in this area has focused on reducing the precision of controlled polymerizations which results in the smooth broadening of the MWD19–27. This strategy provides only little control over MW and no control over MWD shape. More recent work has sought to expand beyond broadening the MWD by utilizing reactor engineering strategies. The first approach enables the skewing MWD profiles through the metered addition of a discrete initiating species into a controlled polymerization3,28–30. While successful, it is difficult to a-priori design the MWD of the polymer due to the complex polymerization kinetics caused by the constant variation in concentration of the growing chain and monomer31. The second approach implements flow reactors to synthesize narrow MWD polymer that accumulates into a reception vessel to build up a MWD profile32–35. However, the complex fluid mechanics and transport involved with performing a controlled polymerization in flow has limited the chemistry, molecular weight, and precision achievable33,36–39. The lack of reactor design rules necessitated significant empirical optimization of the process. Ultimately, the key limitation of prior MWD methodologies is that they do not use a design and result in arbitrary shaped MWDs.Fig. 1Overview of prior literature and this manuscripts methodology.a Prior literature on MWD design. b Direct design to syntheisis methodology. MW molecular weight.Herein, we report a chemistry agnostic protocol to synthesize any MWD profile directly from a targeted design (Fig. 1b). The protocol consists of implementing a computer-controlled flow reactor to produce narrow MWDs (Fig. 2). Polymers with narrow MWDs accumulate in a collection vessel to construct any targeted MWD. The key feature of this approach is the ability to a-priori calculate the reactor flow rates needed to turn a MWD design into an actual polymer sample, which we describe as a design to synthesis protocol. The development of this protocol is supported by in-depth fluid mechanics and polymerization kinetics studies. These fundamental studies led us to establish the guiding principles for the design and operation of tubular flow reactors. We apply these principles for the synthesis of polymers with tailored MWDs. In addition, we developed a mathematical model that quantitatively predicts the experimental results, and enables us to explore the limit of any MWD design protocol that implements controlled polymerizations.Fig. 2Overview of the flow reactor setup and methodology for MWD control.The computer-controlled flow reactor generates narrowly dispersed polymer samples which accumulate in the collection vessel to build the target MWD. Reprinted (adapted) with permission from Macromolecules 2019, 52, 13, 4847–4857. Copyright (2019) American Chemical Society.First, we describe the key reactor engineering concepts that enable the design of flow reactors. Second, we report on the implementation of this reactor design to the synthesis of MW sweep and unique MWDs. Lastly, we discuss the theoretical limits of MWD control via the development of a mathematical model.ResultsFluid mechanics and reactor design rulesThe vast majority of laboratory-scale flow chemistry is performed in the laminar flow regime (low flow rates, small volume)37,38,40,41. Under these conditions, a parabolic flow velocity profile results in a wide distribution of residence times, and limited mixing (smooth streamlines) at the reactor inlet leads to inhomogeneity early in the flow process. For polymerizations, the distribution in residence times and/or poor mixing leads to the undesired broadening of MWDs40,42–49. Several strategies to circumvent these undesired effects have been developed over the years. The first one consists of performing the polymerization at very high flow rates50,51. This high flow rate consumes a significant amount of reagents making it expensive to operate and is only amenable to extremely fast polymerizations52,53. Another approach involves performing the polymerization in droplet flow, where plugs of non-reactive solvents or gases are placed in between plugs of polymerization mixtures35,47,49,54. The need to introduce pulses of an inert fluid into the reactor complicates the operation. Given the challenges of the previous methods, we elected to implement a simple tubular reactor that operates under laminar flow and achieves the essential plug flow via Taylor dispersion55.Taylor dispersion achieves a plug flow like behavior through diffusion and dispersion56–59. Specifically, the laminar flow will initially cause a solute pulse to stretch into a parabola, however, radial diffusion in conjunction with the radial velocity gradient will cause the homogenization of the concentration profile to yield a plug-like flow (Fig. 3c)58,59. This plug behavior is the enabling feature that makes it possible to achieve a narrow polymer MWD in laminar flow, as broad polymer MWD would be obtained otherwise due to initiators having different residence times. Taylor dispersion can also be leveraged more generally across flow chemistry where plugs are desired, such as high throughput synthesis60,61.Fig. 3Reactor design, chemistry, fluid flow diagram, and data for polymer tracer experiments.a Flow reactor configuration for the tracer experiments. b DBU catalyzed ROP of lactide with an octanol initiator. A UV detectable (266 nm) pyrene initiator will periodically be pulsed into the flow reactor as the tracer. c Depiction of laminar flow with Taylor dispersion. d Experimental data of tracer pulses for different flow rates in a single run.To confirm Taylor dispersion’s ability to generate plugs for polymerizations, we explored Talyor’s original mathematical derivation in the context of our flow reactor55,58,59. The concentration profile that emerges from the derivation is functionally a normal distribution (Eq. 1) where the volume of the plug will depend on reactor radius (R), length (L), and flow rate (Q) (see Supplementary Discussion for derivation). The derivation reveals that the plug volume will have a 2nd order dependency on reactor radius, 0.5 order dependency reactor length, and 0.5 order dependency on the flow rate (Eq. 2). In addition, the conditions under which Taylor dispersion is expected to apply for any tubular flow reactor set-up can be determined using in Supplementary Equation 41.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c\left( t \right) = \frac{M}{{2\pi ^{3/2}R^2\sqrt {D_{app}t} }}e^{ - \frac{{\left( {t - t_r} \right)^2}}{{4D_{app}t_r/v_{z,avg}^2}}};\;D_{app} = \frac{{R^2v_{z,avg}^2}}{{48D_{ab}}}$$\end{document}ct=M2π3/2R2Dappte−t−tr24Dapptr/vz,avg2;Dapp=R2vz,avg248Dab2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Plug\;volume \propto Q\sigma \propto R^2\sqrt {LQ} $$\end{document}Plugvolume∝Qσ∝R2LQTo validate this theoretical prediction, we performed several pulse tracer experiments by periodically introducing a UV absorbing initiator as a tracer to the ring-opening polymerization (ROP) of lactide (Fig. 3)62. We systematically varied the radius (0.0889–0.254 mm), length (7.6–15.2 m), and flowrate (63.4–267.5 μL/min) to determine their effect on the tracer pulse width. The tracer pulse width was determined by analyzing the polymerization mixture exiting the end of the tubular reactor using a GPC equipped with a RI and UV detector. The MW of the polymer produced during the tracer experiments was set to Mn = 4500 g/mol, and RI GPC data confirmed the stability of the process with a polymer Mn = 4400 ± 200 g/mol and Mw/Mn = 1.066 ± 0.004 throughout the entire experiment (see Supplementary Fig. 2). It is worth noting that the narrow MWD achieved in flow is very similar to batch data reported in the literature (Mn = 4300 g/mol, Mw/Mn = 1.05)63. Processing the UV tracer data gave rise to the predicted normal distribution of the concentration as a function of time. The standard deviation (σ) of these distributions, which is proportional to plug volume (Eq. 2), was calculated to have a second-order dependence on reactor radius, and half order dependency on length, matching the theoretical derivation (Fig. 4). However, a −0.86 order dependency on flow rate was measured (compared with the theoretical −0.5 dependency). Further tracer studies reviled, a −0.5 order dependency on flow rate when a small molecule tracer was used, and thus the deviation appears to be polymerization specific which will be the focus of a later study (see Supplementary Figs. 12–14). Validation of the theory enables Eq. (2) to be a quick and simple design rule for flow reactor design.Fig. 4Processed tracer experimental results.Green line is the best fit for σ with different reactor radius (R). Tan line is the best fit for σ with different reactor lengths (L). Red line is the best fit for σ with different reactor flow rates (Q) while a polymerization is occuring. Purple line is the best fit for σ with different reactor flow rates (Q) while no polymerization is occuring. Blue line is the best fit for σ with different reactor flow rates (Q) using a small molecule tracer while no polymerization is occuring. (σ: standard deviation of the normal distribution of tracer exiting the tubular reactor).Controlled polymerizations, in general, require simultaneous initiation of all initiators in order to produce a polymer with narrow MWDs. However, under laminar flow, mixing of the monomer and initiator at the reactor’s inlet is difficult due to the smooth streamlines40,43–46. To achieve mixing in this regime, custom static mixers are often implemented37,64. These mixers tend to be expensive and cause detrimental pressure drops for polymer synthesis. Thus, we elected to simply implement a Tee and rely on diffusion and dispersion to mix the solutions (Fig. 5a). For the initiation of the polymerization to occur homogeneously using diffusion and dispersion, the polymerization rate must be slow enough for mass transfer to homogenize the solutions. Therefore, experiments to determine the fastest maximum polymerization rate for our set-up was undertaken, while ensuring that we remain in a Taylor dispersion regime. We implemented the ring opening metathesis polymerization (ROMP) of an exo-norbornene type monomer in the flow reactor to probe the mixing efficiency65–68. ROMP was an ideal choice as it is extremely fast (90% conversion in 5s.), and the rate of polymerization can be tuned with the addition of pyridine over several orders of magnitude (see Supplementary Fig. 42).Fig. 5Diagram and experimental data for polymerization mixing experiments.a Representation of the heterogeneity of the fluid at the entrance of the tubular reactor. b GPC traces for mixing experiment. c Dependancy of polymer dispersity on the rate of reaction.The mixing efficiency was quantitatively probed by systematically increasing the 3-bromopyridne (Br-Py) to 3rd generation Grubbs catalyst (G3) ratio in a flow experiment, and analyzing the polymer produced with GPC (Fig. 5b, c). As expected, with very fast polymerization rates the heterogeneity of the reaction during the early stages of the polymerization caused a broadening of the MWD. However, with decreasing polymerization rates it was observed that the dispersity approaches the batch dispersity (Mw/Mn = 1.03) leveling off around 0.01 M/min63. Also, no additional effect on polymer dispersity from mixing was observed when the total flow rate or ratio of flow rates were changed (see Supplementary Figs. 19–25)65,66. This data provides a guide for designing flow reactors such that the rate of polymerization, or more generally, the rate of reaction does not exceed the mixing efficiency of the flow reactor implemented if precision is desired.Overall, the above section provides the guiding principles to design flow reactors, which are broadly applicable to the field of flow chemistry. To leverage these reactor design principles for the engineering of MWD, we consider the MWD protocol consists of accumulating polymers with low dispersity. It then makes it advantageous to operate our reactor in a way that provides the smallest plug volume, since smaller plug volumes will result in a finer design resolution. We found that a reactor length of 762 cm and a reactor radius of 0.0889 mm or 0.127 mm achieves small plug while not generating too much backpressure for our syringe pumps. Additionally, the operation parameters (flow rates) should be on the order of 10’s μL/min to achieve the appropriate residence time, given that the highest precision is achieved when the rate of polymerization is around 0.01 M/min. Using 10 mL syringes with this setup enables the synthesis of up to 500 mg of polymer with a single run. If larger quantities of sample are desired, multiple syringes pumps or continuous pumps can be used without altering the reactor design to produce multiple grams of material per day at high design resolution. If even larger amounts of polymer are needed, the radius of the reactor can be increased, however, flow rate, reactor length, and polymerization rate may need to be adjusted to maintain the highest MWD design resolutions.MW sweeps in flowWith the flow reactor design rules established, we sought to demonstrate the flow reactor’s ability to produce precise quantities of polymer over a wide range of different MWs. We first targeted 6 different MWs (4, 16, 50, 300, and 1000 kg/mol) covering four orders of magnitude in a single flow experiment with ROMP. This MW sweep was achieved with a constant monomer flow rate, and by decreasing the G3 loading step-wise (Fig. 6). Br-Py (250 eq to G3) was added to the G3 solution to ensure that the rate of polymerization enabled sufficient time for mixing. The maximum flow rate was set so that the residence time allowed the reaction to reach complete conversion. It is interesting to note that by mixing the pyridine with the ruthenium, the polymerization becomes effectively zeroth order in initiator, and full monomer consumption is achieved in the same amount of time for all MWs65. Upon performing and analyzing the MW sweep, 6 peaks were detected in the GPC chromatogram. The molecular weight of each peak matched the targeted MW confirming that the polymerizations were performed to completion, and the dispersities were low (Mw/Mn = 1.02–1.11). In addition, normal distributions were fitted to each of the 6 peaks and the area under each peak, which is proportional to the quantity of polymer formed, was measured to be 16.7 ± 0.7% (targeted value is 16.6%). Combining these two elements demonstrates the fine precision in molecular weight control achievable with our flow reactor.Fig. 6Chemistry and experimental data for MW sweep in flow reactor.a ROMP of a norbornene type monomer with MW sweep chromatogram below it. b Anionic polymerization of styrene with MW sweep chromatogram below it. (dashed vertical lines are the target MWs).We then demonstrated the generality of our reactor engineering strategy by performing the sec-BuLi initiated anionic polymerization of styrene69. The rate of polymerization was tuned with the addition of THF to the monomer mixture70. Repeating a MW sweep in a similar fashion as ROMP, establishes the ability of the flow reactor to achieve low MW dispersities (Mw/Mn = 1.02–1.11). Since we had use ROP to establish the design rule for the reactor we did not perform a MW sweep for the ROP of lactide. In contrast to ROMP and anionic polymerization, the ROP employs a catalyst and an initiator. This different composition led us to operate our reactor so that the degree of polymerization is altered by varying the conversion of monomer through the variation of the catalyst loading (kept below 70% to maintain low dispersity) while maintaining a constant residence time55. This approach of varying the degree of polymerization through variation of the monomer conversion is thought to be more compatible with other controlled polymerization (e.g. ATRP, anionic ROP, Coordinative Chain Transfer Polymerization (CCTP)).MWD design and modelingHaving established the ability of the flow reactor to produce precise quantities of polymers with MWs spanning multiple orders of magnitudes, we sought to establish a design to synthesis protocol capable of converting any MWD profile into an actual sample. Building on the precision of the flow reactor, we developed a general mathematical procedure (see SI Figure 37) to calculate the process flow rates directly from the targeted shape; i.e. we can convert any design to synthesis. This straight forward design to synthesis workflow enables the a-priori design capability. To demonstrate this methodology we first targeted square MWDs (Fig. 7a) using the anionic polymerization of styrene. We produced four different MWDs with widths ranging from 1.5–5 kg/mol to 1.5–45 kg/mol. The MWD design is defined to be weight fraction on a log scale, as this is representative of a GPC chromatogram (see Supplementary Methods for more discussion). Figure 7a contains the four produced chromatograms, which qualitatively matches the targeted square MWDs (Table 1).Fig. 7Designs, experimental data, and predictions for targeted MWD.a (left) Design for the four targeted square MWDs. (right) GPC traces and predictions for the four square MWDs. b (left) Design for the targeted complex shape. (right) GPC trace and prediction for complex MWD.Table 1Data for the synthesis of square MWD by anionic polymerization of styrene.MW range (kg/mol)Mn (theory) (g/mol)aMw/Mn (theory)aMn (g/mol)bMw/Mnb1.5–525001.1623701.161.5–1032501.3829201.371.5–2543601.9140301.931.5–4551302.4946302.50aValues are from MATLAB code.bData obtained from GPC(THF) with respect to PS standards.To further demonstrate the versatility of this methodology, we also synthesized two triangle MWDs using the anionic polymerization of styrene (see Supplementary Fig. 36). The two triangles span a MW range from 1.5 to 25 kg/mol with one design sloping right, and the other sloping left. Once again a qualitative match is observed with the targeted design and GPC data (see Supplementary Table 15 for tabulated data). Expanding beyond the anionic polymerization of styrene, we implemented ROMP for the synthesis of a target consisting of two discrete square distributions of different height and a triangle. This design is complex and covers a MW range from 2 to 400 kg/mol, challenging the methodology. As seen in Fig. 7b, the GPC trace shows the targeted MWD design features. An additional benefit of synthesizing non-normal distributions is the ability to easily generate large numbers of additional MWDs through blending (see Supplementary Fig. 37). Since the individual distributions are shaped and broad, there isn’t a need to synthesize a large number of samples to achieve a smooth profile for the desired MWD design.To further our analysis of the synthesized MWDs, a mathematical model was generated to predict the MWD. The mathematical model inputs the MWD design and generates a large number of polymer distributions, which are summed to produce the predicted MWD (Figs. 7 and 8). Polymer distributions are modeled by log-normal distributions, as is was observed that a normal distribution describes the MW sweeps GPC data well, and GPC data are plot on a log(MW) scale. Log-normal distributions are defined by two parameters: mean (μLN) and standard deviation (σLN). The mean is defined by ln(MW), in which MW was determined from the design to ensure a linear spacing based on weight fraction. The standard deviation of the distribution was set by experimentally defined dispersities. As seen in Fig. 7 the model (dashed lines) produces a very close match to the experimental results (solid lines). Furthermore, the Mn and dispersity of the sample can be calculated using the model and compared against the experimental values. The predicted Mn and dispersity are in good agreement with experimental values (see Table 1 and Supplementary Table 15). Overall the quantitative match between the model and experiments establishes the precision of the synthesis and the validity of the model.Fig. 8Overview of mathematical model for the prediction of MWD.The mathematical model inputs the MWD design and generates a large number of log-normal polymer distributions with different MW which are summed to produce the predicted MWD. The is used to study the effect of dispersity on the final MWD. (P(x): log-normal probability density function, x: MW scale; μ: mean of log-normal distribution; σ: standard deviation of log-normal distribution; MW*: linear spaced by weight fraction MW from targeted MWD design).Finally, we use this mathematical model to investigate the effects of the dispersity of the polymer produced by the flow reactor on the final MWD. It becomes apparent that the resolution of the MWD is sensitive to the dispersity of the polymer produced. This is noted by the deviation of the predicted MWDs (colored lines) in comparison with the targeted MWD (black line) in Fig. 8. This result should be interpreted as the theoretical limit of the precision for any MWD design technique that relies on controlled polymerizations. In the context of flow reactors, this emphasizes the need for both chemistries that can produce very narrow MWD and a carefully designed reactor to ensure fluid mechanics does not erode the control of the polymerization.DiscussionWe have developed a simple and robust procedure for converting any MWD profile directly into an actual sample by using a computer-controlled tubular flow reactor. This methodology consists of performing any controlled polymerization to produce low dispersity polymers in flow which accumulate into a collection vessel to generate the targeted MWD. Using this protocol, we prepared square, triangle, and a complex MWD profile, and quantitatively matched the experimental results to mathematical predictions. Fluid mechanics and polymerization kinetics provided the guiding principles for the construction and operation of the tubular flow reactor. The development of these design principles will significantly aid in the design and operation of flow reactors, which is broadly beneficial to the field of flow chemistry. We also anticipate that this simple, versatile, and precise methodology to directly synthesize any MWD will aid in fundamental material property studies, as well as, aid in the tuning of materials for specific applications.MethodsGeneral informationAll reactions were performed in oven-dried glassware under an argon atmosphere in an argon-filled glovebox (O2 < 2 ppm, H2O < 0.5 ppm) at room temperature unless otherwise specified. All solvents were dried using a solvent purification system. All commercially obtained reagents were used as received: 3-Bromopyridine {Aldrich, 99%}, Ethyl Vinyl Ether {Aldrich, 99%, store at −20 °C}, 1,8-Diazabicyclo[5.4.0]undec-7-ene (DBU) {Aldrich, 98%, store at −20 °C}, Boric Acid {Aldrich, 99.5%}, D,L-lactide{Aldrich, 99%}, 1-Octanol{Aldrich, 99%}, SecBuLi {1.3 M sol. in cyclohexane/hexane (92/8) ACROS}. [(H2IMes)(3-Br-py)2(Cl)2Ru=CHPh], G3 was synthesized according to literature67. nor1 were synthesized according to literature63. Styrene {Aldrich} was pushed through an alumina column and distilled under vacuum prior to use.NMR experimentsNuclear Magnetic Resonance (NMR) spectra were recorded on a Bruker AVANCE III 500 MHz. Spectra are reported in ppm and referenced to the residual solvent signal: CDCl3 (1H 7.26 ppm, 13C 77.16 ppm), C6D6 (1H 7.16 ppm, 13C 128 ppm), CD2Cl2 (1H 5.32 ppm).Gel permeation chromatographyGel permeation chromatography (GPC) was performed using a Tosoh Ecosec HLC-8320GPC at 40 °C fitted with a reference column (6.0 mm ID × 15 cm), a guard column (6.0 mm ID × 4.0 cm × 5 μm), and two analytical columns (7.8 mm ID × 30 cm × 5 μm). The reference flow rate is 0.5 mL min−1 while the analytical column is at 1.0 mL min−1. THF (HPLC grade) was used as the eluent, and polystyrene standards (15 points ranging from 500 Mw to 8.42 million Mw) were used as the general calibration. An additional calibration was created for specifically for linear polylactic acid and only used for linear polylactic acid (10 points ranging from 500 Mw to 10,000 Mw).General procedure for MWD designIn a typical experiment, the flow reactor is setup by first filling one syringe with a solution containing the monomer and a second syringe is filled with a initiator (an activator or inhibitor may be added to either syringe to adjust the rate of polymerizations). The syringes are then attached to the flow reactor and placed into the computer-controlled syringe pumps. The exit of the reactor was feed into a pot with a quenching reactant. A preprogram flow rate sequence then started. Upon completion, the final reaction mixture is analyzed by GPC and the final polymer is isolated by precipitation. A more detailed procedure can be found in the supporting information for each of the specific chemistry used in this manuscript.Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Code 1 | nature communications | [
"Article"
] | [
"Polymer characterization",
"Polymer synthesis",
"Process chemistry",
"Chemical engineering"
] | polymer’s molecular weight distribution (MWD) impacts material properties processability mechanical strength phase correlation general across polymers motivated synthetic process techniques work focused on polymerization methods polymers narrow MWDs controlled polymerization Controlled polymerizations revolutionized synthesis materials control provide tunability for broad MWDs advantageous for many broad distributions staple in industry2 polyethylene Philipps catalyst has dispersity >10 balance fast processability high mechanical strength2 latest improvements to polyolefins MWDs for specific applications2. emerging applications 3d printing demand high mechanical performance ease of processing through tuning MWD12 coarse MWD tuning common independent control over MW breadth shape challenging high precision tailored MWDs interest material applications.Early approaches to engineered MWDs blending batches polymers with MW1 large batches reduce multimodality final MWD streamlined polymerization with multisite catalysis cascade reactors2. multimodal MWD same properties as smooth MWD unsuitable for several applications macrophase separation1limitation motivated synthetic methodologies accessing polymers tunable smooth MWDs Initial work focused precision controlled polymerizations smooth broadening MWD19–27 strategy provides little control over MW no control MWD shape recent work beyond broadening MWD reactor engineering strategies first approach enables skewing MWD profiles addition discrete initiating species difficult to a-priori design MWD due complex polymerization kinetics variation concentration second approach implements flow reactors synthesize narrow MWD polymer reception vessel complex fluid mechanics transport limited chemistry molecular weight precision lack of reactor design rules necessitated empirical optimization limitation prior MWD methodologies design result in arbitrary shaped MWDs 1Overview prior literature methodology MWD design Direct design syntheisis methodology weight chemistry agnostic protocol to synthesize MWD profile from targeted design protocol computer-controlled flow reactor produce narrow MWDs Polymers MWDs accumulate in collection vessel construct targeted MWD key a-priori calculate reactor flow rates turn MWD design into polymer sample design synthesis protocol supported by fluid mechanics polymerization kinetics studiesfundamental studies principles for design operation tubular flow reactors apply principles for synthesis polymers with tailored MWDs developed mathematical model predicts experimental results MWD design protocol controlled polymerizations.Fig. 2Overview flow reactor setup methodology for MWD control computer-controlled reactor generates dispersed polymer samples accumulate collection vessel target MWD Reprinted Macromolecules 2019, 52, 13, 4847–4857 Copyright American Chemical Society describe key reactor engineering concepts report implementation reactor design synthesis unique MWDs discuss theoretical limits MWD control mathematical model.ResultsFluid mechanics reactor design majority laboratory-scale flow chemistry in laminar flow regime (low flow rates small volume parabolic flow velocity profile wide distribution residence times limited mixing inhomogeneity polymerizations distribution poor mixing undesired broadening MWDs40 strategies circumvent undesired effects developed polymerization at high flow rates50 flow consumes reagents expensive amenable to fast polymerizations52approach polymerization in droplet flow plugs non-reactive solvents gases between polymerization mixtures35 inert fluid complicates operation simple tubular reactor under laminar flow plug flow via Taylor dispersion55 dispersion achieves plug flow like behavior through diffusion dispersion56–59 laminar flow solute pulse into parabola radial diffusion velocity gradient homogenization concentration plug-like flow (Fig. 3c plug behavior narrow polymer MWD in laminar flow Taylor dispersion leveraged across flow chemistry plugs desired high throughput synthesis60.Fig. 3Reactor design chemistry fluid flow diagram data for polymer tracer experiments Flow reactor configuration DBU catalyzed ROP of lactide octanol initiator UV detectable (266 nm pyrene initiator pulsed into Depiction laminar flow with Taylor dispersion Experimental data of tracer pulses for different flow rates Taylor dispersion’s generate plugs for polymerizations explored Talyor’s original mathematical derivation flow reactor55 concentration profile normal distributionvolume plug on reactor radius length flow rate Supplementary Discussion for derivation). derivation reveals plug volume 2nd order on reactor radius 0.5 length flow rate (Eq. 2)conditions Taylor dispersion tubular flow reactor set-up determined Supplementary Equation 41.1[12pt]{amsmath-69pt}}\left t\right) = \frac{M}{{2\pi ^{3/2}R^2\sqrt {D_{app}t}\frac{{\left\right^2}}{{4D_{app}t_r/v_{z,avg}^2}}}_{app} = \frac{{R^2v_{z,avg}^2}}{{48D_{ab}}}\end{document}ct=M2π3/2R2Dappte−t−tr24Dapptr/vz,avg2=R2vz,avg248Dab2[12pt]{minimal}{amsmath{upgreek{-69pt}{document}$Plug;volume Q\sigma R^2\sqrt {LQ}{document}PlugvolumeQσR2LQTo validate prediction performed pulse tracer experiments introducing UV absorbing initiator ring-opening polymerizationlactide (Fig. varied radius (0.0889–0.254 length (7.6–15.2 flowrate (63.4–267.5 μL/min) effect tracer pulse width determined polymerization mixture tubular reactor GPC RI UV detector MW polymer Mn = 4500 g/mol GPC data confirmed stability polymer Mn = 4400 ± 200 g/mol Mw/Mn = 1.066 ± 0.004 Supplementary Fig. 2) narrow MWD flow similar to batch data (Mn = 4300 g/mol Mw/Mn = Processing UV tracer data predicted normal distribution concentration function time standard deviation (σ) proportional to plug volume (Eq. 2) second-order dependence reactor radius half order length theoretical derivation (Fig. 4) −0.86 order dependency on flow rate measured −0.5 −0.5 order dependency flow rate small molecule tracer deviation polymerization specific focus later study Supplementary Figs. 12–14) Validation theory enables Eq. (2) design rule flow reactor design.Fig. 4Processed tracer experimental results.Green line reactor radius Tan line reactor lengths Red line reactor flow rates polymerizationPurple line best fit for σ different reactor flow rates no polymerization Blue line best fit for σ different reactor flow rates small molecule tracer no polymerization (σ standard deviation normal distribution tracer exiting tubular reactor).Controlled polymerizations require simultaneous initiation initiators polymer narrow MWDs laminar flow mixing monomer initiator inlet difficult smooth streamlines40 custom mixers expensive cause pressure drops polymer synthesis elected implement Tee rely diffusion dispersion mix solutions (Fig. polymerization polymerization rate slow mass transfer experiments fastest maximum polymerization rate undertaken Taylor dispersion regime implemented ring opening metathesis polymerization (ROMP) exo-norbornene monomer flow reactor mixing ROMP ideal fast (90% conversion in 5s rate polymerization tuned addition pyridine Supplementary Fig. 42).Fig. 5Diagram experimental data polymerization mixing experiments Representation heterogeneity fluid at entrance tubular reactor GPC traces mixing experiment Dependancy polymer dispersity on rate reactionmixing efficiency probed increasing 3-bromopyridne (Br-Py) to 3rd generation Grubbs catalyst (G3) ratio experiment analyzing polymer with GPC (Fig. 5b, c). fast polymerization rates heterogeneity early caused broadening MWD decreasing polymerization rates dispersity approaches batch dispersity (Mw/Mn = 1.03) leveling off around M/min63 no additional effect on polymer dispersity from mixing total flow rate or changed Supplementary Figs. 19–25)65 data guide for designing flow reactors polymerization exceed mixing efficiency precision section guiding principles design flow reactors applicable to flow chemistry MWD MWD protocol accumulating polymers with low dispersity operate reactor smallest plug volume finer design resolution reactor length 762 cm radius 0.0889 mm or 0.127 mm achieves small plug not backpressure for syringe pumps operation parameters (flow rates) 10’s μL/min appropriate residence time highest precision polymerization around 0.01 M/min 10 mL syringes synthesis of to 500 mg polymer single runlarger quantities sample desired multiple syringes continuous used without reactor design multiple grams material per day high resolution larger amounts polymer needed radius reactor increased flow rate reactor length polymerization rate highest MWD resolutions.MW sweeps reactor design rules precise quantities polymer MWs targeted 6 MWs (4 16 50 300 1000 kg/mol) four orders magnitude single flow experiment ROMP MW sweep achieved constant monomer flow rate decreasing G3 loading step-wise Br-Py (250 eq to G3) added G3 solution time mixing maximum flow rate set residence time complete conversion mixing pyridine with ruthenium polymerization zeroth order initiator full monomer consumption same time all MWs65 6 peaks detected GPC chromatogram molecular weight matched targeted MW polymerizations performed completion dispersities low (Mw/Mn = 1.02–1.11) normal distributions fitted 6 peaks area under each peak proportional to polymer measured 16.7 ± 0.7% (targeted 16.6%). precision molecular weight control flow reactor.Fig. 6Chemistry experimental data MW sweep flow reactorROMP norbornene monomer MW sweep chromatogram Anionic polymerization styrene MW sweep chromatogram (dashed vertical lines target demonstrated reactor engineering strategy sec-BuLi initiated anionic polymerization styrene69 rate polymerization tuned addition THF monomer mixture70 Repeating MW sweep ROMP flow reactor low MW dispersities (Mw/Mn = 1.02–1.11) ROP perform MW sweep ROP lactide contrast ROMP anionic polymerization ROP employs catalyst initiator different composition reactor polymerization altered varying conversion monomer variation catalyst loading below 70% low dispersity constant residence time55 approach polymerization compatible with other controlled polymerization ATRP anionic ROP Coordinative Chain Transfer Polymerization (CCTP)).MWD design established flow reactor produce precise quantities polymers MWs multiple orders magnitudes design to synthesis protocol converting MWD profile sample developed mathematical procedure Figure 37 calculate process flow rates targeted shape convert design to synthesis design synthesis workflow enables a-priori design capability targeted MWDs (Fig. 7a) anionic polymerization styreneproduced four MWDs widths 1.5–5 kg/mol to 1.5–45 kg/mol MWD design weight fraction log scale representative GPC chromatogram Figure 7a four chromatograms matches targeted square MWDs (Table. 7Designs experimental data predictions targeted MWD Design four square MWDs GPC traces predictions Design complex shape GPC trace prediction complex MWD.Table 1Data synthesis MWD anionic polymerization styrene.MW range (kg/mol)Mn (g/mol.161.5–1032501.3829201.371.5–2543601.9140301 from MATLAB code GPC(THF) PS standards synthesized two triangle MWDs anionic polymerization styrene Fig. triangles MW range 1.5 to 25 kg/mol one sloping right other left qualitative match targeted design GPC data Supplementary Table 15 implemented ROMP synthesis target two square distributions different height triangle design complex covers MW range 2 to 400 kg/mol challenging methodology Fig. 7b GPC trace shows targeted MWD design featuresbenefit synthesizing non-normal distributions MWDs blending Supplementary Fig. distributions shaped broad need synthesize large samples smooth profile for MWD design synthesized MWDs mathematical model generated predict MWD inputs MWD design generates polymer distributions produce predicted MWD (Figs. 7 8). Polymer distributions modeled by log-normal distributions normal distribution describes MW sweeps GPC data log Log-normal distributions defined by mean (μLN) standard deviation (σLN). mean defined by ln determined standard deviation set by experimentally defined dispersities Fig. 7 model produces close match to experimental results Mn dispersity sample calculated model compared against experimental values predicted Mn dispersity with experimental values (see Table 1 Supplementary Table 15). quantitative match between model experiments establishes precision synthesis validity model.Fig. 8Overview mathematical model for prediction MWD model inputs MWD design generates log-normal polymer distributions with different MW produce predicted MWD used study effect dispersity on final MWD(P(x): log-normal probability density function x: MW scale μ: mean distribution σ: standard deviation MW*: linear spaced weight fraction from targeted MWD mathematical model effects dispersity polymer flow reactor on final MWD resolution MWD sensitive to dispersity polymer noted by deviation of predicted MWDs with targeted MWD in Fig. 8. theoretical limit precision for MWD design technique controlled polymerizations emphasizes need for chemistries narrow MWD carefully designed reactor fluid mechanics control polymerization developed procedure for converting MWD profile into sample computer-controlled tubular flow reactor methodology controlled polymerization low dispersity polymers accumulate into collection vessel targeted MWD prepared square triangle complex MWD profile matched experimental results to mathematical predictions Fluid mechanics polymerization kinetics guiding principles for construction operation tubular flow reactor design principles aid design operation of flow reactors beneficial to flow chemistry methodology MWD material property studies tuning of materials for specific applicationsreactions performed oven-dried glassware argon atmosphere-filled glovebox (O2 < 2 ppm H2O < 0.5 ppm room temperature solvents dried purification system commercially reagents used 3-Bromopyridine Ethyl Vinyl Ether −20 1,8-Diazabicyclo[5.4.0]undec-7-ene Boric Acid 99.5% D,L-lactide 1-Octanol SecBuLi M [(H2IMes)(3-Br-py)2(Cl)2Ru=CHPh] G3 synthesized Styrene pushed through alumina column distilled under vacuum.NMR spectra recorded Bruker AVANCE III 500 MHz ppm solvent signal CDCl3 (1H 7.26 ppm 13C 77.16 ppm), C6D6 7.16 ppm 13C 128 CD2Cl2 (1H 5.32 ppm).Gel permeation Tosoh Ecosec HLC-8320GPC at 40 °C reference column (6 15 guard columnmm ID 4.0 cm × 5 two columns (7.8 mm ID 30 cm × 5 μm). reference flow rate 0.5 mL min−1 analytical column 1.0 mL min−1 THF (HPLC grade eluent polystyrene standards (15 points 500 Mw to 8.42 million Mw general calibration additional calibration linear polylactic acid (10 points 500 Mw to 10,000 procedure MWD flow reactor syringe solution monomer second syringe initiator activator inhibitor syringes attached reactor computer-controlled syringe pumps exit pot quenching reactant preprogram flow rate sequence final reaction mixture analyzed by GPC final polymer isolated by precipitation detailed procedure supporting information chemistry.Supplementary information Peer Review File Description Additional Supplementary Files Code 1 | 48.8 | 0.48173 |
10.1038/s41467-020-16329-y | PMC7248072 | Plasmonic-enhanced nanosensors are limited in practical applications, as it remains challenging to detect molecules at low concentrations. Here, the authors introduce a buoyant particulate strategy in order to enrich analytes in the plasmonic hot spots. | Detecting matter at a single-molecule level is the ultimate target in many branches of study. Nanosensors based on plasmonics have garnered significant interest owing to their ultrahigh sensitivity even at single-molecule level. However, currently, plasmonic-enhanced nanosensors have not achieved excellent performances in practical applications and their detection at femtomolar or attomolar concentrations remains highly challenging. Here we show a plasmonic sensing strategy, called buoyant plasmonic-particulate-based few-to-single particle-nanosensors. Large-sized floating particles combined with a slippery surface may prevent the coffee-ring effect and enhance the spatial enrichment capability of the analyte in plasmonic sensitive sites via the aggregation and lifting effect. Dimer and single particle-nanosensors demonstrate an enhanced surface-enhanced Raman spectroscopy (SERS) and a high fluorescence sensitivity with an enrichment factor up to an order of ∼104 and the limit of detection of CV molecules down to femto- or attomolar levels. The current buoyant particulate strategy can be exploited in a wide range of plasmonic enhanced sensing applications for a cost-effective, simple, fast, flexible, and portable detection. | IntroductionA plasmonic-enhanced nanosensing process involves complicated coupled three-body interactions among photons, molecules, and nanostructures1–3. In previous decades, fundamental issues regarding the interaction between light and nanostructures have been investigated intensively, and many types of plasmonic hot spots have been fabricated, e.g., nanogaps, nanotips, and nanopores, for enhanced surface-enhanced Raman spectroscopy (SERS) and fluorescence sensitivity4–8. However, SERS detection for many small cross-sections or weakly adsorbed molecules is still difficult owing to the deficiency of related investigations on the interaction between molecules and nanostructural surfaces. For example, in the commercial SERS protocol based on a colloidal aggregation route9, weakly adsorbed molecules cannot effectively adsorb onto a metallic surface during fast aggregation. Therefore, this natural defect makes them impossible to exhibit a remarkable sensitivity. In the form of SERS solid surface with precise nanopatterns10, dipping the SERS substrate in a solution containing analytes may yield a homogeneous molecule adsorption. However, the adsorption time (e.g., a few hours) is far beyond practical timescales. Instead, by drying a droplet containing analytes on a substrate, the molecule distribution on a substrate may encounter uniformity issue owing to the ubiquitous coffee-ring effect11, particularly in weakly adsorbed molecules. More importantly, relative to the large area of patterned surfaces, the hot spot sites (with areas of only a few square nanometers) possess only a small portion of the entire substrate, thereby resulting in poor molecular accessibility when only a few molecules are investigated. Currently, localizing analytes toward plasmonic hot spot sites with high-efficiency is paramount in improving the sensitivity of plasmonic-enhanced nanosensors.The coffee-ring effect is a very common phenomenon, and its nature is that the capillary flow outward from the center of the drop carries dispersed particulates to the edge as evaporation proceeds12. In many detections based on plasmonic nanosensors, the formation of a coffee-ring may result in a completely uncontrolled distribution of both colloidal nanoparticles (NPs) and target molecules, resulting in deteriorated signal uniformity and sensitivity13.Herein, we report a plasmonic-enhanced sensing strategy based on buoyant plasmonic particulates that are designed to thoroughly avoid the coffee-ring effect and guide target molecules into a spatially highly localized plasmonic hot spot region. The resulted dense-packed pattern, particularly for dimer or single particles, may significantly increase the sensitivity of plasmonic sensors with an enrichment factor of up to ∼104. Combined with the developed procedure using a superhydrophobic surface to accurately sort single particles, the current buoyant particulate strategy is believed to be applicable to a wider range of sensing devices, such as fluorescent, Raman, and infrared spectroscopes for a cost-effective, simple, fast, flexible, and portable detection. We call this new technique as the buoyant particulate-based few-to-single particles-plasmonic nanosensor.ResultsExact adsorption position of analyteThe strategy for the buoyant plasmonic particulates sensor is shown in Fig. 1a, b. Large-sized (30–100 μm), light-weight floatable particles, i.e., hollow SiO2 coated with Au nanoparticles, were synthesized by a seed-mediated growth route (Fig. 1c), which is a modified method described by Westcott et al., Shao et al., and Liu et al.14–16 (see Supplementary Figs. 1–3 for details). During the drying process, the buoyant particles may float on the top of the solvent, thereby significantly reducing their chances of being pinned on the slippery surface17–19 (Fig. 1d). The large size of the floating particles may increase the capillary force to drive these particles inward the droplet and aggregate as a dense-packed structure. In particular, at the final stage of evaporation, the landed particles on the substrate may serve as “post” to lift the droplet from the slippery surface, thereby enforcing the solvent and target molecule to dry in the vicinity of particle–particle junctions (Fig. 1d)20. Consequently, the obtained condensed pattern can significantly increase the sensitivity of the plasmonic nanosensors.Fig. 1Schematic illustrations of the buoyant particulate strategy.a The proposed buoyant particulate protocol consisting of slippery surface, solvent, and floating particles. b After drying, the containing probe molecules and plasmonic floating particles may aggregate and condense together, thus enhance the sensitivity of plasmon-type nanosensors. c The schematic processes of synthesizing light-weight hollow silica-coated Au shell particles by a seed-mediated growth route. d The evaporation processes of suspended hollow silica-coated Au shell particles within a droplet and the final aggregated pattern. The scale bar represents 20 μm.The most advantageous aspect of the current strategy is the aggregation effect of buoyant-particulates on a slippery surface (Fig. 1d and Supplementary Movie 1), which contributes to the enrichment of the solvent in the vicinity of the particle–particle interface. Consequently, target molecules that are highly localized toward plasmonic hot spot sites were achieved. Supplementary Figures 4 and 5 and Supplementary Movies 2 and 3 demonstrate clear aggregation and enrichment processes. To investigate the exact adsorbed position of the analyte, we firstly performed the fluorescent characterizations for the obtained aggregates using crystal violet (CV) molecules as a fluorescent probe. Figure 2a(i–v) shows the fluorescent images of buoyant particulates with different particle amounts dried on the slippery surface. A strong fluorescent signal can be observed clearly in the vicinity of the particle–particle interface for the configurations of more than two particulates. In the case of a single buoyant particulate, after the evaporation of the droplet, the analyte finally dries and adsorbs on/near the surface of the buoyant particulate according to the fluorescent image, as shown in Fig. 2a(v).Fig. 2Enriching effect, lifting effect, and fluorescent characterizations.a Fluorescent images of floating particle aggregate with various CV concentrations. A strong fluorescent signal can be observed clearly in the vicinity of particle–particle interface. b Fluorescent images with CV concentration of 10−8 M. The red circle showing that after we moved the aggregate, the original site on the substrate does not display any fluorescent signal. c The schematic illustrations of the solvent ring finally adsorbs on the particle–particle interface using the present buoyant particulate protocol. d FDTD simulation of the plasmonic coupling sites of dimer buoyant particulates. e Raman spectra collected from the interface and non-interface regions for a dimer floating particles with CV concentration of 10−9 M. The insets are the fluorescent images of a dimer buoyant particulate, demonstrating a 10–20 times of signal difference between interface and non-interface regions. The scale bar in (a) is 20 μm and in (b) is 100 μm.Next, we conducted an in-situ observation by moving the aggregate of landed particles (Supplementary Movie 4). It is interesting that after we moved the aggregate, the original site of the substrate did not display a network-shaped fluorescent signal (Fig. 2b(i and ii) and Supplementary Fig. 6). This phenomenon indicates that, after the drying of the solvent, the CV molecules do not adsorb on the substrate but in the vicinity of the particle–particle interface (Fig. 2a). To further verify this statement, two controlled experiments were conducted by only replacing the buoyant plasmonic particulates. Similar to the results shown in Fig. 2b, when we employed hollow bare SiO2 particles to enrich the CV molecules, the strong network-shaped fluorescent signal from the formed aggregate can still be observed (Supplementary Fig. 7). After moving the formed aggregate of hollow SiO2 particles, the fluorescent signal from the original site fully disappeared. However, when we evaporated a droplet containing only CV molecules on the slippery substrate, the fluorescent imaging of the CV molecules (Supplementary Fig. 8) was still detectable at concentrations down to the level of 10−8–10−9 M. Therefore, these comparison experiments revealed that, using the current buoyant plasmonic particulate strategy, after the drying of the droplet, the analytes mainly enrich and localize into the particle–particle interface, i.e., the hottest spot region.The fluorescent characterizations of drying a droplet containing buoyant particulates support the fact that a lifting effect may occur at the final stage of evaporation. Once the buoyant particulates have landed on the substrate, these particles may serve as a “post” to enforce the solvent and analyte to interact with the post. Consequently, the solvent and probe molecules are driven and enriched into the particle aggregate. The occurrence of the lifting phenomenon requests the capillary force in the vicinity of the particle–particle interface should be larger than the gravitation of the solvent and the adhesion force between the solvent and the substrate. In fact, this is possible because the Teflon substrate used and the perfluorinated lubricant acting as a superhydrophobic and slippery surface contribute to a low surface energy, thereby reducing the adhesion force significantly.Obviously, the aggregation and lifting effects are crucial in enforcing the solvent and analyte toward the particle–particle interface (Fig. 2c), i.e., additional plasmonic coupling site (Fig. 2d), which may significantly increase the sensitivity in many types of sensing. Figure 2e shows the Raman spectra for a dimer of buoyant particulate using the CV molecules as a probe. The interface regions display a remarkably enhanced signal intensity, e.g., 10–20 times, compared with that in non-interface regions.Droplet drying processes in buoyant particulate systemTo analyze the aggregation effect of the buoyant particulate, and the lifting and enriching effects of the solvent and probe molecules, we evaluated the influences of various suspended particulates on the drying process of a droplet on the slippery surface. We compared three types of particulates, i.e., Au NPs, solid SiO2 particles, and hollow SiO2− coated by Au nanoparticles. During the evaporation process, the initial contact angles of droplets with different suspended particulates were not significantly different (Fig. 3a). As the evaporation proceeded, the main differences were the changes in the contact angle and contact line (Fig. 3a). When a droplet contained solid silica particulates, the contact line and contact angle gradually decreased (Fig. 3a and Supplementary Fig. 9), finally forms a coffee-ring pattern. In the situation of Au-NP suspended droplet (Supplementary Fig. 10), after a rapid decrease in the contact angle at the early stage, a constant contact angle (CCA) of approximately 46°, was formed. However, after the diameter of the droplet shrinks to less than 1.0 mm, the contact angle gradually reduced to approximately 12° in the end, and the droplet diameter decreased continuously (Fig. 3a). Finally, the resulting aggregated patterns (Supplementary Fig. 11) exhibited the character of a micrometer-sized coffee-ring.Fig. 3Evaporation processes and mechanism.a The relationship between contact angle and droplet diameter for various particulates. The value of droplet diameter marked “A” is estimated owing to the limit of observation. b The final drying stage of the evaporation process for a droplet containing buoyant-particulates on slippery surface. The scale bar is 50 μm. c Schematics illustration of the close-packed particle aggregation process on a slippery substrate by evaporating a droplet containing floating particles. The inset figure displays the light-weight particle in the vicinity of the leading edge of the thin film.The aggregation processes of the buoyant-particulate suspended droplets on the slippery surface were observed through video microscopy and contact angle measurements (Supplementary Movie 1 and Supplementary Figs. 12 and 3b). As shown in Fig. 3a, after a rapid decrease, a CCA of approximately 58° appeared. More interestingly, the contact angle even increased at the final stage (Fig. 3a, b), which could be critical for the aggregation effect. From the images of final aggregates shown in Supplementary Fig. 13, we obtained direct evidence that the coffee-ring phenomenon had been completely prevented and that a dense-packed pattern had been finally achieved.To understand the aggregation mechanism of buoyant-particulate, and the lifting and enriching effects of the solvent and containing molecules, an analytical model based on the force analysis of suspended particulates at three-phase (air–water–substrate) interfaces (Fig. 3c) and the Young–Laplace equation has been developed (see details in Supplementary Information). According to the theoretical results, an enrichment and spatial localization mechanism of target molecules is proposed and schematically illustrated in Fig. 3c. Both the driving force to enforce the aggregation of buoyant-particulates and to lift the final solvent from the slippery substrate can be significantly affected by the particle size. A larger floating-particle size is preferable to obtain a dense-packed pattern, lift the solvent, and localize the analyte into the particle–particle interface. This theoretical prediction has been confirmed by a proof-of-concept experiment as shown in Supplementary Fig. 14, in which an aggregation state cannot be formed when the size of floating-particle less than 20 μm. In fact, previous studies have shown that suspended particles may be aggregated in a more close-packed manner on a hydrophobic surface and with increased θR21.Plasmonic sensing propertiesThe current buoyant particulate displays a highly remarkable enrichment capacity of the probe molecules. To exploit the practical applications of this technique, we developed a procedure to trap the buoyant particulates with an accurate sorting of single particles with ∼80% probability (Fig. 4a and Supplementary Fig. 15). Therefore, single, dimer, few particles, and more particle aggregates have been facilely manipulated (Fig. 4b).Fig. 4The slippery surface, the plasmonic sensing properties and applications.a The droplet with a certain volume on superhydrophobic slippery surface to sort single or dimer particles. The scale bar is 0.5 mm. b The optical images of sorting single and dimer buoyant particulate on the top of droplet. The scale bar is 50 μm. c SERS spectra of CV molecules obtained from aqueous solutions at different concentrations from 10 nM to 1 aM. d Probability of obtaining observable SERS signals at different concentrations for four types of SERS detection protocols, i.e., a droplet containing the Au NPs dries on PTFE porous surface and buoyant-particulate suspended droplet on slippery surface with single, dimer, and few particles. e The current SRES strategy for the portable detection of persistent organic pollutants (POPs). f Fluorescent spectroscopes and limit of detection of CV probe molecule using the present buoyant particulate protocol.The current detection protocol demonstrates an ultrahigh sensitivity for SERS detection. Figure 4c demonstrates SERS spectra using CV molecules as probe molecules, displaying typical Raman peaks of the CV molecules at 1172 and 1616 cm−1 bands22. The limit of detection (LOD) of the CV molecules using the current SERS protocol can be down to 1 aM (10−18 M). Furthermore, using the buoyant particulate-based dimer particle or single-particle-SERS, the 100% probability of collecting observable SERS signal reach the 10 fM or 0.1 pM level. These results represent a remarkable progress compared with other techniques reported in the literature23 and are at least 3 or 4 orders of magnitude higher than those using Au NPs as suspended particles (Fig. 4d and Supplementary Fig. 16). According to the calculation of surface area, the enrichment factor is on the scale of ∼104 (Supplementary Fig. 17), owing to the aggregation, enriching, and lifting effects.As potential applications of the current SERS strategy, we examined the LOD in various dye molecules, e.g., rhodamine 6G, rhodamine B, and malachite green, which are typical illegal additives in realistic food. Using the buoyant particulate-based dimer-particle-SERS protocol, an ultralow LOD, i.e., at 10–100 fM, can be obtained for these molecules (Supplementary Fig. 18). Furthermore, we evaluated the inspection in persistent organic pollutants (POPs), which are typically weakly adsorbed molecules and extensively emerging contaminants in many ground and surface waters. Bisphenol A (BPA), 2,4-dichlorophenol, and naphthalene were selected as target molecules (Supplementary Fig. 18). Figure 4e shows the Raman spectra of the BPA molecule. The LODs for the BPA molecules could reach to an ultralow concentration of ∼0.1 ppb. This sensitivity is much better than those of other SERS protocols from the literature for the molecule detection of POPs24. Therefore, owing to the availability of portable Raman spectrometers, the present SERS strategy will be applicable to the rapid in-situ analysis of safety in food manufacturing, environment pollutants, etc., as only a few minutes are required in an optimized process (Supplementary Table 1). Furthermore, we can improve the sensing performance by optimizing the materials and methodology (Supplementary Figs. 19–24). For example, the signal repeatability can be improved by optimizing the sizes of the laser spot and laser power (Supplementary Fig. 23). Furthermore, the sensitivity can also be improved by changing the irradiation laser wavelength and the size of Au NPs (Supplementary Fig. 24).The advantage of the current buoyant particulate protocol can be also found in other plasmonic nanosensors, e.g., fluorescent sensing. In Fig. 4f, the fluorescent signals can be distinguished even when the CV concentrations are decreased from 10−11 to 10−13 M. The LOD is determined to be as low as 10−12 M. By contrast, the fluorescent images of Au NPs as suspended particulates were difficult to identify even for the CV concentration of 10−9 M (Supplementary Fig. 25).DiscussionIn conclusion, we demonstrated a new buoyant particulate-based few-to-single particle-plasmonic sensing strategy, consisting of light-weight buoyant particulates, a solvent droplet containing analytes, and a slippery surface. During drying, the buoyant particles floated on the top of the solvent, thereby significantly reducing their chances of being pinned on a slippery surface. A large buoyant particulate can be favorable to enrich the solvent and analyte through the aggregation and lifting effects. Therefore, the resulted dense-packed pattern, particularly for dimer or single particles, may significantly increase the sensitivity of plasmonic sensors with an enrichment factor of up to ∼104. Combined with the developed procedure using a superhydrophobic surface to accurately manipulate single particles, the current buoyant particulate strategy is believed to be applicable to a wider range of sensing devices, such as fluorescent, Raman, and infrared spectroscopes for a cost-effective, simple, fast, flexible, and portable detection. Additionally, the strategy is opening new possibilities for a wide range of applications not only in plasmonic enhanced spectroscopes, but also in biological sensors, printing, photonic crystals, complex assemblies, and other devices25.MethodsSynthesisAs the schematic shown in Supplementary Fig. 1, the silica-coated Au shell particles were prepared by a seed-mediated growth method, and the detailed procedure could be regarded as the following steps. (i) The chemical modification of hollow SiO2 microspheres: 0.1 wt% aqueous hollow SiO2 microspheres were prepared by mixing 0.1 g of silica powder with 100 mL of ultrapure water. Next, 50 mL hollow SiO2 microspheres, 0.2 mL 0.4% (3-Aminopropyl)trimethoxysilane (APTMS), and 50 mL ultrapure water were mixed under stirring for 24 h, and was further heated at 80 °C. Excess APTMS was removed by centrifuging and re-dispersing in water for 3 times, and the obtained sample was termed as SiO2–NH2. (ii) The adsorption of Au NPs on hollow SiO2 microspheres: Spherical gold NPs with 23 nm in diameter, served as the nanoseeds, were synthesized based on a modified citrate reduction approach26, characterized by the scanning electron microscope (SEM) image and UV–vis spectra in Supplementary Fig. 2b. Then, an appropriate amount of SiO2–NH2 was injected dropwise to the gold seeds solutions. Under vigorous magnetic stirring, almost all of hollow silica microspheres could be covered with Au nanoseeds after 6 h, denoted as SiO2–Au nanoseeds. The excess gold seeds were removed by centrifuging in 7000 rpm. SEM images in Supplementary Fig. 2c demonstrated a uniform distribution of Au seeds on SiO2 surface. (iii) The synthesis of hollow silica-coated Au shell particles: Firstly, gold hydroxide was obtained by mixing 100 mL H2O, 4 mL HAuCl4 (10 mM), and 50 mg K2CO3 powder in dark for 24 h. Then, 2 mL aqueous SiO2–Au nanoseeds was added into 5 mL gold hydroxide solutions under vigorous stirring to prevent the up-floating of microsphere. After 5 min, 0.15 mL NH2OH·HCl (0.1 M) solution was added dropwise, and the Au shell coating would be finished 3 h later. In order to obtain more uniform coverage of Au shell, a spot of AgNO3 solution was brought into the reaction system. As shown in Supplementary Fig. 3, homogeneous and closely packed gold shell was prepared. The gold NPs were ∼57 nm with the particle–particle gaps for only several nanometers.Preparation of SERS substratesWe used the hollow silica-coated Au shell particles as SERS-active materials, and a hydrophobic slippery Teflon membrane as support surface. The fabrication of slippery membrane was as following: the Teflon membrane was fixed onto a flat glass slide (5 cm × 5 cm) by double-sided adhesive. Then, 0.5 mL perfluorinated lubricant (Dupont, GPL 105) was dispersed on the membrane by spin coating, and thus was heated at 90 °C for 30 min. For analytes detection, such as crystal violet, 50 μL aqueous solution of probe molecules and 10 μL silica-coated Au shell particles were dropped onto the hydrophobic slippery surface at 150 °C. During the drying process, we observed that the particles floated on the surface of droplets like a “boat” due to the ultra-low density of hollow SiO2, as shown in Fig. 1. The closed packaged few-particles aggregate would be obtained after solvent evaporation (shown in Supplementary Fig. 13), and then was used as SERS substrates.Raman and fluorescence spectroscopyThe Raman spectra, fluorescence spectra, and fluorescence imaging measurements were performed in a confocal microscope-based Raman spectrometer on a home-made optical testing platform in our lab. For SERS test, the samples were excited by a 633 nm laser with ∼0.2 mW, and the acquisition time was 30 s. For fluorescence test, the signal was collected by a 532 nm laser with the laser power of ∼5 mW. The acquisition time was 60 s. In addition, the fluorescence imaging was obtained by the excitation of a mercury lamp.The characterizations of buoyant particulatesThe morphology and structure of the products were characterized using a scanning electron microscope (JEOL, JSM-7000F). The optical properties were characterized using ultraviolet–visible spectroscopy (Agilent, Carry 60). The contact angle was measured on an optical contact angle instrument (KRUSS, DSA 100).Supplementary information
Supplementary Information
Peer Review File
Description of Additional Supplementary Files
Supplementary Movie 1
Supplementary Movie 2
Supplementary Movie 3
Supplementary Movie 4 | nature communications | [
"Article"
] | [
"Raman spectroscopy",
"Nanophotonics and plasmonics",
"Nanosensors",
"Imaging and sensing"
] | plasmonic-enhanced nanosensing involves complicated interactions among photons molecules nanostructures1–3 issues interaction between light nanostructures investigated plasmonic hot spots fabricated nanogaps nanotips for enhanced Raman spectroscopy (SERS) fluorescence sensitivity4–8 SERS detection for small cross-sections weakly adsorbed molecules difficult deficiency investigations interaction surfaces commercial SERS protocol weakly adsorbed molecules adsorb onto metallic surface during fast aggregation defect makes remarkable SERS dipping SERS substrate in solution analytes homogeneous molecule adsorption adsorption time few hours beyond practical timescales drying droplet analytes on substrate molecule distribution may uniformity coffee-ring in weakly adsorbed molecules hot spot sites possess small portion of substrate poor molecular accessibility when few molecules investigated localizing analytes toward plasmonic hot spot sites with high-efficiency paramount improving sensitivity of plasmonic-enhanced nanosensorscoffee-ring effect common capillary flow carries particulates to edge evaporation plasmonic nanosensors coffee-ring uncontrolled distribution colloidal nanoparticles target molecules signal uniformity plasmonic-enhanced sensing strategy buoyant plasmonic particulates avoid coffee-ring effect guide molecules into localized plasmonic hot spot region dense-packed pattern dimer single particles increase plasmonic sensors enrichment factor ∼104 superhydrophobic surface sort single particles buoyant particulate strategy applicable sensing devices fluorescent Raman infrared spectroscopes cost-effective simple fast flexible portable detection technique buoyant particulate-based few-to-single particles-plasmonic nanosensor adsorption position strategy buoyant plasmonic particulates sensor in Fig. 1a, b Large-sized (30–100 light-weight floatable particles hollow SiO2 coated Au nanoparticles synthesized by seed-mediated growth route (Fig. 1c), modified method Westcott Shao Liu et Supplementary Figs. 1–3 drying process buoyant particles float on top solvent reducing chances pinned slippery (Fig.large size floating particles capillary force aggregate structure final stage evaporation landed particles substrate lift droplet slippery surface enforcing solvent target molecule dry particle–particle junctions (Fig. 1d condensed pattern plasmonic nanosensors.Fig. buoyant particulate strategy proposed protocol slippery surface solvent floating particles After drying probe molecules plasmonic floating particles aggregate condense enhance sensitivity nanosensors synthesizing light-weight hollow silica-coated Au shell particles seed-mediated growth route evaporation processes suspended silica-coated Au shell particles droplet final aggregated pattern scale bar represents 20 μm advantageous aggregation effect buoyant-particulates on slippery surface (Fig. 1d Movie 1) contributes enrichment solvent particle–particle interface target molecules localized plasmonic hot spot sites achieved Figures 4 5 Movies 2 3 demonstrate aggregation enrichment processes adsorbed position analyte performed fluorescent characterizations aggregates crystal violet) moleculesFigure 2a(i–v shows fluorescent images buoyant particulates amounts dried slippery surface strong fluorescent signal observed particle–particle interface more than two particulates single buoyant particulate evaporation analyte dries adsorbs surface image Fig. 2a(v).Fig. 2Enriching effect effect fluorescent characterizations Fluorescent images floating particle aggregate various CV concentrations strong signal particle–particle interface images CV concentration 10−8 M moved aggregate original site fluorescent signal solvent ring adsorbs particle–particle interface buoyant particulate protocol FDTD simulation plasmonic coupling sites dimer buoyant particulates Raman spectra interface non-interface regions dimer floating particles CV concentration 10−9 M fluorescent images dimer buoyant particulate 10–20 times signal difference interface non-interface regions scale bar (a) 20 μm (b) 100 μm in-situ observation aggregate particles 4) after aggregate original site display-shaped fluorescent signal (Fig. 2b(i ii) Fig. 6)phenomenon indicates after drying solvent CV molecules adsorb on substrate vicinity particle–particle interface (Fig. 2a). two experiments conducted replacing buoyant plasmonic particulates employed hollow SiO2 particles enrich CV molecules strong network fluorescent signal from aggregate observed Fig. 7) After moving aggregate SiO2 fluorescent signal disappeared evaporated droplet containing CV molecules on slippery substrate fluorescent imaging detectable at concentrations to 10−8–10−9 M. comparison experiments revealed current buoyant plasmonic particulate strategy after drying analytes enrich localize into particle–particle interface hottest spot region fluorescent characterizations drying droplet buoyant particulates support lifting effect at final stage evaporation buoyant particulates substrate serve as solvent analyte solvent probe molecules driven enriched into particle aggregate lifting phenomenon requests capillary force vicinity interface larger than gravitation solvent adhesion force substrateTeflon substrate perfluorinated lubricant superhydrophobic slippery surface contribute low surface energy reducing adhesion force lifting effects crucial solvent analyte particle–particle interface (Fig. additional plasmonic coupling site increase sensitivity sensing Figure 2e shows Raman spectra dimer buoyant particulate CV molecules probe interface regions display enhanced signal intensity 10–20 times compared non-interface regions.Droplet drying processes in buoyant particulate aggregation lifting enriching effects solvent probe molecules evaluated influences suspended particulates drying process droplet slippery surface compared three types particulates Au NPs solid SiO2 particles hollow SiO2− coated Au nanoparticles evaporation initial contact angles droplets particulates not significantly different (Fig. main differences changes contact angle contact line droplet solid silica particulates contact line angle decreased coffee-ring pattern Au-NP suspended droplet decrease contact angle constant contact angle) approximately 46° formedafter diameter droplet to less than 1.0 mm contact angle reduced to 12° diameter decreased (Fig. 3a). aggregated patterns Fig. 11 micrometer-sized coffee-ring.Fig. 3Evaporation processes mechanism relationship between contact angle droplet diameter droplet diameter estimated limit observation final drying stage evaporation for droplet buoyant-particulates on slippery surface scale bar 50 μm close-packed particle aggregation on slippery substrate droplet particles light-weight particle edge thin film aggregation processes buoyant-particulate droplets slippery surface observed through video microscopy contact angle measurements Figs. 12 3a after rapid decrease CCA of 58° contact angle increased at final stage critical for aggregation effect final aggregates Fig. 13, coffee-ring phenomenon prevented dense-packed pattern achieved understand aggregation mechanism of buoyant-particulate lifting enriching effects solvent molecules analytical model force analysis of suspended particulates at three-phase interfaces (FigYoung–Laplace equation developed Supplementary results enrichment spatial localization mechanism molecules proposed illustrated in Fig. 3c driving force aggregation buoyant-particulates lift solvent substrate affected by particle size larger-particle size preferable dense-packed pattern lift solvent localize analyte interface confirmed proof-of-concept experiment Supplementary Fig. 14 state size floating-particle less than 20 μm previous studies suspended particles hydrophobic surface increased θR21 buoyant particulate enrichment capacity molecules developed procedure trap particulates accurate sorting single particles ∼80% probability (Fig. 4a Fig. single dimer more particle aggregates manipulated (Fig. 4b).Fig. slippery surface plasmonic sensing properties applications droplet volume superhydrophobic slippery surface sort single dimer particles scale bar 0.5 mm optical images sorting single dimer buoyant particulate droplet scale bar 50 μm SERS spectra CV molecules aqueous solutions concentrations 10 nM to 1 aM Probability SERS signals different concentrations for four SERS detection protocolsdroplet Au NPs dries PTFE surface buoyant-particulate suspended droplet slippery surface single particles current SRES strategy portable detection persistent organic pollutants Fluorescent spectroscopes limit detection CV probe molecule buoyant particulate protocol current detection protocol ultrahigh sensitivity SERS detection Figure 4c SERS spectra CV molecules probe Raman peaks 1172 1616 cm−1 limit detection) CV molecules 1 aM (10−18 M). buoyant particulate-based dimer particle single-particle-SERS 100% probability collecting SERS signal 10 fM or 0.1 pM level results remarkable progress 3 or 4 orders magnitude higher Au NPs suspended particles (Fig. 4d enrichment factor ∼104 Fig enriching lifting effects examined LOD dye molecules rhodamine 6G B malachite green illegal additives food buoyant particulate-based dimer-particle-SERS protocol ultralow LOD 10–100 fM molecules Fig. evaluated inspection persistent organic pollutants weakly adsorbed molecules emerging contaminants ground surface watersBisphenol A 2,4-dichlorophenol naphthalene selected target molecules Fig. 18). Figure 4e shows Raman spectra BPA molecule LODs BPA molecules ultralow concentration ∼0.1 ppb better other SERS protocols detection POPs24. portable Raman spectrometers SERS strategy applicable rapid-situ analysis safety food manufacturing environment pollutants few minutes required Table 1) improve sensing performance optimizing materials methodology Figs. 19–24) signal repeatability optimizing laser spot power changing irradiation laser wavelength size Au NPs advantage buoyant particulate protocol plasmonic nanosensors fluorescent sensing Fig. 4f fluorescent signals distinguished CV concentrations decreased 10−11 to 10−13 M LOD low 10−12 M fluorescent images Au NPs suspended particulates difficult identify CV concentration 10−9 M Fig. demonstrated new buoyant particulate-based particle-plasmonic sensing strategy light-weight buoyant particulates solvent droplet slippery surface buoyant particles floated solvent reducing chances pinned slippery surfacelarge buoyant particulate solvent analyte aggregation lifting dense-packed pattern single particles increase plasmonic sensors enrichment factor ∼104 superhydrophobic surface particles buoyant particulate strategy applicable sensing devices fluorescent Raman infrared spectroscopes cost-effective simple fast flexible portable detection strategy possibilities applications plasmonic spectroscopes biological sensors printing photonic crystals complex assemblies 1 silica-coated Au shell particles prepared seed-mediated growth method chemical modification hollow SiO2 microspheres 0.1% aqueous hollow SiO2 microspheres 0.1 g silica powder 100 mL ultrapure water 50 mL SiO2 microspheres 0.2 mL 0.4% (3-Aminopropyl)trimethoxysilane 50 mL ultrapure water mixed 24 h heated at 80 °C Excess APTMS removed centrifuging re-dispersing water 3 times sample SiO2–NH2. adsorption of Au NPs on hollow SiO2 microspheres Spherical gold NPs 23 nm diameter synthesized modified citrate reduction scanning microscope image UV–vis spectra Fig.SiO2–NH2 injected dropwise to gold seeds solutions silica microspheres covered with Au nanoseeds after 6 h excess gold seeds removed 7000 rpm SEM images uniform distribution Au seeds on SiO2 surface synthesis of silica-coated Au shell particles gold hydroxide 100 mL H2O 4 mL HAuCl4 50 mg K2CO3 powder for 24 h 2 mL aqueous SiO2–Au nanoseeds added into 5 mL gold hydroxide solutions After 5 min 0.15 mL NH2OH·HCl (0.1) solution added dropwise Au shell coating finished 3 h later AgNO3 solution homogeneous packed gold shell prepared gold NPs ∼57 nm gaps several nanometers.Preparation SERS silica-coated Au shell particles as SERS-active materials hydrophobic slippery Teflon membrane support surface fixed onto flat glass slide 0.5 mL perfluorinated lubricant) dispersed on membrane heated at 90 °C for 30 min.analytes detection crystal violet 50 μL aqueous solution probe molecules 10 μL silica-coated Au shell particles dropped hydrophobic surface 150 °C drying particles floated “boat”-low density hollow SiO2 Fig. 1. closed packaged few-particles aggregate obtained after solvent evaporation Fig used SERS substrates.Raman fluorescence Raman fluorescence imaging measurements confocal microscope Raman spectrometer home-made optical testing platform SERS test samples excited 633 nm laser ∼0.2 mW acquisition time 30 s fluorescence test signal collected 532 nm laser ∼5 mW acquisition time 60 s fluorescence imaging obtained mercury lamp characterizations buoyant morphology structure characterized scanning electron microscope optical properties ultraviolet–visible spectroscopy contact angle measured optical contact angle instrument 100).Supplementary information Files Movie 1 2 3 4 | 48.9 | 0.412615 |
10.1038/s41467-020-17304-3 | PMC7368010 | PRDM family members are transcriptional regulators involved in cell identity and fate determination. Here, the authors characterize PRDM10 and show that it functions to ensure global translation efficiency during early embryonic development. | Members of the PR/SET domain-containing (PRDM) family of zinc finger transcriptional regulators play diverse developmental roles. PRDM10 is a yet uncharacterized family member, and its function in vivo is unknown. Here, we report an essential requirement for PRDM10 in pre-implantation embryos and embryonic stem cells (mESCs), where loss of PRDM10 results in severe cell growth inhibition. Detailed genomic and biochemical analyses reveal that PRDM10 functions as a sequence-specific transcription factor. We identify Eif3b, which encodes a core component of the eukaryotic translation initiation factor 3 (eIF3) complex, as a key downstream target, and demonstrate that growth inhibition in PRDM10-deficient mESCs is in part mediated through EIF3B-dependent effects on global translation. Our work elucidates the molecular function of PRDM10 in maintaining global translation, establishes its essential role in early embryonic development and mESC homeostasis, and offers insights into the functional repertoire of PRDMs as well as the transcriptional mechanisms regulating translation. | IntroductionPRDM proteins are characterized by the presence of a conserved N-terminal PR (PRDI-BF1 and RIZ1) homology domain closely related to the lysine methyltransferase SET domain, followed by variable C2H2-type zinc finger repeats that typically mediate sequence-specific DNA binding. Several PRDMs have been shown to act as important transcriptional regulators controlling cell fate specification in various developmental contexts1–3. For example, Prdm1 is required for primordial germ cell specification and branchial arch patterning during embryonic development4, and also plays an important role in regulating hematopoietic lineage differentiation5. Prdm16 promotes brown fat adipogenesis6,7 and hematopoietic stem cell maintenance3. We and others have uncovered a critical and nonredundant role for Prdm14 and Prdm15 in maintaining naïve pluripotency of embryonic stem cells8,9.Prdm10, also known as tristanin10, is highly conserved in vertebrates and belongs to the same phylogenetic subfamily as Prdm1511. Prdm10 is expressed in various embryonic and adult tissues12,13. A large-scale phenotypic screen revealed that homozygous deletion of Prdm10 in mice is embryonic lethal14, and gene rearrangements involving PRDM10 have been described in some undifferentiated pleomorphic sarcomas15,16. Despite its potential biological significance, the molecular and functional properties of PRDM10 remain largely unknown, and its role in vivo has not been well-characterized.In this study, we establish a conditional Prdm10 knockout mouse model to uncover a critical role for PRDM10 during very early embryonic development, and utilize mouse embryonic stem cells (mESCs) to study PRDM10’s biochemical and molecular properties. We demonstrate that PRDM10 acts as a transcription factor that binds to the promoters of target genes and regulates their expression. Through direct transcriptional regulation of Eif3b, a key translation initiation factor, we show that PRDM10 plays a critical role in maintaining global translation essential for mESC survival.ResultsPRDM10 is essential for preimplantation embryogenesisPrdm10 encodes a protein containing an N-terminal PR domain, followed by ten C2H2 zinc fingers and a C-terminal glutamine (Q)-rich transactivation domain (Supplementary Fig. 1a) which is unique among the 17 PRDM family members. To explore the function of PRDM10 in vivo, we generated mice bearing a conditional allele (Prdm10F) in which exon 5 of Prdm10 is flanked by loxP sites (Supplementary Fig. 1b). Cre-mediated removal of exon 5 introduces a frameshift resulting in a nonfunctional truncated protein (Supplementary Fig. 1a), thus generating a null allele (Prdm10Δ) (Supplementary Fig. 1b). While Prdm10Δ/+ mice were viable and fertile with no gross morphological or behavioral abnormalities observed in daily husbandry, no Prdm10Δ/Δ live pups were recovered from heterozygous intercrosses.This prompted us to examine embryos from Prdm10Δ/+ intercrosses at pre- and post-implantation time-points to define the timing of embryonic lethality. While at embryonic day (E) 3.5, all genotypes were recovered at close to Mendelian ratios, Prdm10Δ/Δ embryos were slightly underrepresented at E4.5, and no viable Prdm10Δ/Δ embryos were recovered at E7.5 and E12.5 postimplantation stages (Fig. 1a, Supplementary Fig. 1c). At E3.5, Prdm10Δ/Δ embryos had an abnormal, morula-like appearance (86%), in contrast to control embryos being mostly expanding/expanded blastocysts (Fig. 1b, c and Supplementary Fig. 1d). Consistently, ex vivo cultured Prdm10Δ/Δ embryos developed normally from 2-cell to morula stage yet failed to form expanded blastocysts (Supplementary Fig. 1e). These data are consistent with the timing of embryonic death observed in utero and furthermore show that lethality is due to an embryo-intrinsic defect independent of implantation failure. Given these observations and the complete absence of Prdm10Δ/Δ embryos at postimplantation stages, we conclude that loss of PRDM10 causes developmental arrest before blastocyst formation and embryonic lethality peri-implantation.Fig. 1Prdm10 is essential for mouse preimplantation embryogenesis and mESC growth.a Frequency of embryo genotypes obtained from heterozygous intercrosses at each developmental stage. E3.5 embryos are recovered at the expected Mendelian distribution; no Prdm10Δ/Δ embryos are observed by E7.5. b Representative images of mutant (Prdm10Δ/Δ) and control (Prdm10+/+, Prdm10Δ/+) embryos isolated at E3.5. The inner cell mass (ICM) is labeled with an asterisk, and the blastocoel is defined by a red dashed line. Scale bar: 50 μm. c Scoring of E3.5 embryos into three phenotypic categories: morula, partially expanded blastocyst, or fully expanded/cavitated blastocyst. n = 27 (Prdm10+/+), n = 39 (Prdm10Δ/+), n = 21 (Prdm10Δ/Δ). d qRT-PCR analysis of Prdm10 exon 5 expression in OHT-treated Prdm10F/F; CreERT2 (Δ/Δ) mESCs compared with vehicle-treated (F/F) controls at indicated time-points post-induction. Expression normalized to Ubb; n = 3 biological replicates. e Western blot analysis of PRDM10 protein levels in Prdm10F/F; CreERT2 mESCs at indicated time-points (days) after exposure to EtOH (E) or OHT (O). Loading control, α-tubulin. f PRDM10-depleted mESCs exhibit an increasingly severe cell growth defect over time. Cells were passaged at constant density every 2 days and counted daily up to Day 7 post-induction; n = 4 samples. Y-axis: cumulative population doublings. g Representative brightfield images of Prdm10F/F and Prdm10Δ/Δ mESC colonies at Day 5 and 7 post-induction. Cells were plated at equal densities 2 days prior to image acquisition. Scale bar: 500 μm. h Representative images from colony formation assay; n = 6. Prdm10F/F and Prdm10Δ/Δ mESCs were seeded at Day 3 post-induction and fixed for analysis at Day 8. i Caspase 3/7 activity in Prdm10Δ/Δ mESCs relative to Prdm10F/F controls, at Day 4 to 7 post-induction; n = 4 technical replicates for each time-point. Data are presented as mean ± s.d. Representative data shown from one out of three independent experiments (f, g, and i). ***P < 0.001, ****P < 0.0001; two-tailed unpaired Student’s t test (f, h, and i).Remarkably, despite the fully penetrant preimplantation stage lethality phenotype, with evidence of increased cell apoptosis (Supplementary Fig. 2a), Prdm10Δ/Δ embryos still expressed lineage-specific markers such as OCT4 (inner cell mass; ICM), CDX2 (trophectoderm; TE), and NANOG (epiblast) at detectable levels (Supplementary Fig. 2a–c). This suggests that PRDM10 is not required for inducing lineage segregation, even though it is essential for embryo survival and developmental progression beyond preimplantation stages.Prdm10-null mESCs show reduced growth and increased apoptosisTo facilitate the investigation of PRDM10’s cellular and molecular functions in early development, we employed the strategy of using mESCs as an in vitro model, thus circumventing accessibility limitations to the embryo. We generated Prdm10F/F; ROSA26-CreERT2 mESCs in which Cre-mediated deletion is inducible by 4-hydroxytamoxifen (4-OHT) exposure. Recombination efficiency upon induction was verified at the genomic (Supplementary Fig. 3a), transcript (Fig. 1d, Supplementary Fig. 3b) and protein level (Fig. 1e, Supplementary Fig. 3c). Phenotypic characterization was performed on Prdm10F/F; ROSA26-CreERT2 mESCs maintained in serum-containing medium with leukemia inhibitory factor (serum/LIF).Following acute deletion of Prdm10, we observed a reduction in cell growth rates becoming detectable starting around 3–4 days post-deletion and increasing in severity thereafter (Fig. 1f). Consistent with the observed growth defect, Prdm10-null mESCs formed smaller colonies that expanded poorly compared with controls (Fig. 1g, h), although colony morphology was not significantly altered (Fig. 1g). Crucially, re-introduction of full-length Prdm10 (Supplementary Fig. 3d, e) was sufficient to rescue the growth defects in Prdm10Δ/Δ cells (Supplementary Fig. 3f). Also of importance, Prdm10-null mESCs cultured in defined serum-free 2i medium with LIF (2i/LIF) showed a similar defect in growth (Supplementary Fig. 3g), suggesting that the phenotypic consequences of PRDM10 deficiency are not rescued by MEK or GSK3 inhibition, and more generally, that the requirement for PRDM10 in mESCs is independent of specific culture conditions.To understand the basis for impaired cell growth in Prdm10-deficient mESCs, we evaluated possible impacts on cell survival and proliferation. For one, the loss of PRDM10 had no significant effect on cell cycle distribution (Supplementary Fig. 3h), suggesting that cell cycle progression is unaffected in Prdm10Δ/Δ mESCs. However, Prdm10Δ/Δ mESCs exhibited significantly higher levels of caspase 3/7 activity compared with controls (Fig. 1i), particularly at later time-points post-recombination (from Day 5 onwards), consistent with the kinetics of phenotypic onset observed in cell growth assays (Fig. 1f). These data demonstrate that PRDM10 deficiency leads to significantly increased apoptotic cell death in mESCs, and suggest that the Prdm10-null phenotype is independent of proliferation defects but instead can be mostly attributed to decreased cell survival.PRDM10 is dispensable for pluripotency and differentiationPrevious studies from our group and others have identified two essential members of the PRDM family, PRDM14 and PRDM15, as key regulators of naïve pluripotency in ESCs8,9,17, which raises the question of whether PRDM10 may also be required for maintenance of mESC pluripotency. To address this, we assessed the expression of several well-characterized pluripotency markers at multiple time-points (up to 8 days) after Prdm10 deletion. Global transcriptome analysis of Prdm10Δ/Δ mESCs compared with controls at days 2 and 4 post-deletion showed no significant downregulation of genes associated with mESC pluripotency and self-renewal; in particular, the transcription factors comprising the core pluripotency regulatory circuitry (Pou5f1, Klf4, Sox2, Nanog) were expressed at levels comparable to or slightly higher relative to controls (Supplementary Fig. 4a). As further validation, we examined selected pluripotency markers (Nanog, Pou5f1, Klf2, Klf4, Esrrb) by qRT-PCR at day 6 and 8 post-deletion, and confirmed that their expression was maintained even at time-points where Prdm10Δ/Δ mESCs exhibit significant growth and survival defects (Supplementary Fig. 4b).Similarly, we detected no reduction in SSEA-1 surface expression on Prdm10-null mESCs at day 4 and 6 post-deletion (Supplementary Fig. 4c). Prdm10Δ/Δ mESCs formed colonies smaller than that of controls, but nonetheless stained positive for alkaline phosphatase activity and showed a level of AP-positive colony formation ability comparable to that of controls, even at day 7 post-deletion (Supplementary Fig. 4d). Lastly, transcriptomic analysis of Prdm10Δ/Δ mESCs cultured under SL conditions revealed no significant misregulation of germ layer lineage markers (Supplementary Fig. 4e), confirming that loss of PRDM10 does not induce precocious differentiation. Taken together, our results indicate that PRDM10 promotes normal growth of mESCs and early embryos, but is dispensable for the maintenance of the pluripotent state.To understand if PRDM10 may play a role in mESC differentiation, we assessed the requirement for PRDM10 during embryoid body (EB) formation (Supplementary Fig. 5a). Following LIF withdrawal, Prdm10Δ/Δ mESCs formed EBs morphologically similar to that of controls (Supplementary Fig. 5b), and upregulated expression of various germ layer lineage markers (Supplementary Fig. 5c), indicating that they retained the ability to undergo EB differentiation. Moreover, regulators of the core pluripotency network (i.e., Pou5f1, Klf4, Nanog) were expressed at levels comparable to or slightly higher relative to controls (Supplementary Fig. 5d), consistent to what was observed in ES cells (Supplementary Fig. 4a, b), Importantly, visible deterioration of Prdm10Δ/Δ EBs was observed at day 6 post-Prdm10 deletion (Supplementary Fig. 5b), at a time-point similar to when Prdm10Δ/Δ mESCs cultured under pluripotency conditions also show a pronounced phenotype (Fig. 1f). Hence, our results are consistent with the essential role of PRDM10 in cell survival, and further suggest that PRDM10 is dispensable for the induction of EB differentiation.Genome-wide identification of PRDM10 binding sitesMost PRDM family members can act as transcriptional regulators2. We thus hypothesized that the requirement for PRDM10 in mESCs may also be mediated primarily through its predicted molecular function as a sequence-specific transcription factor. To test this, we performed genome-wide profiling of PRDM10 binding sites in mESCs by chromatin immunoprecipitation with sequencing (ChIP-seq). We validated three different polyclonal antibodies against PRDM10 and used them in ChIP experiments with Prdm10F/F; ROSA26-CreERT2 mESCs to generate three independent ChIP-seq datasets, from which we identified a set of 528 reproducible peaks (IDR < 0.05) (Supplementary Fig. 6a, Supplementary Data 1). To ascertain antibody specificity, parallel experiments were performed in mESCs depleted of PRDM10 protein after 4-OHT induction. Comparison of both sets of ChIP-seq data revealed that all peaks detected in wild-type (WT) cells were absent or strongly diminished in PRDM10-depleted cells due to PRDM10 protein reduction upon recombination (Supplementary Fig. 6a). Conversely, we did not find any peaks present in PRDM10-deficient cells but absent in WT mESCs, further validating the specificity of our approach.The PRDM10 binding sites revealed in our ChIP-seq data are strongly enriched at promoter regions, with 70.8% of peaks residing within 1 kb upstream from or overlapping with gene transcriptional start sites (TSSs) and only 9.8% mapping to intergenic regions (Fig. 2a, b). Consistent with its enrichment within gene promoters, PRDM10 binding is highly associated with regions of transcriptionally active chromatin marked by H3K4me3, H3K4me1, and H3K36me3 (Supplementary Fig. 6b).Fig. 2PRDM10 is a sequence-specific zinc finger transcription factor.a Genomic feature annotation of PRDM10 binding sites (n = 528) reveals strong enrichment in proximal gene promoters. b Distribution of PRDM10 peaks relative to gene transcriptional start sites (TSS) showing that most peaks are located within 500 bp of a TSS. c Sequence motif identified as highly enriched in PRDM10 binding sites (P = 10–257). d Density histogram showing localization of motif relative to PRDM10 peak centers. e Sequence conservation profiles for motifs detected within PRDM10 peaks (blue) vs. control regions (orange). Control: all genomic regions within ±1 kb of gene TSS. Y-axis: phyloP vertebrate conservation score. Shaded regions: 25–75% percentile of conservation scores. f PRDM10 motif validation by luciferase reporter assay. HEK293T cells were transfected with reporter constructs containing either the canonical (WT) or mutated (MUT) motif sequence, together with Prdm10 expression plasmid or vector control. PRDM10 stimulates transcriptional activation only in the presence of the canonical motif. n = 3 samples. g Binding of labeled probe (10 nM) containing WT vs. MUT motif in the presence of PRDM10441–880 protein (125–1000 nM), assessed by gel-shift assay. Open arrowheads: bound probe; solid arrowheads: free probe. h Competition assay showing specificity of PRDM10 interaction with its motif. PRDM10441−880 binding to 10 nM labeled WT probe is diminished in the presence of 20- to 320-fold molar excess of unlabeled WT probe; competition with the MUT probe has no effect. i Schematic of Prdm10 full-length and mutant expression constructs tested in reporter assays (left). Expression of PRDM10FL, PRDM10N-880, and PRDM10441-C selectively activates the WT motif reporter; n = 3 samples (right). j Western blot of FLAG-tagged PRDM10 constructs from whole cell lysates of transfected HEK293T cells. Arrowheads: proteins of interest. Loading control: β-actin. Reporter activity expressed as relative luminescence units (RLU) of firefly luminescence normalized to Renilla control (f and i). Representative data shown from one out of two (g and h) or three (f and i) independent experiments. Mean ± s.d. ***P < 0.001, ****P < 0.0001, n.s., not significant; two-tailed unpaired Student’s t test (f and i).PRDM10 is a sequence-specific transcription factorBy de novo motif discovery, we identified a consensus sequence highly enriched within PRDM10 binding sites (Fig. 2c). This motif showed central enrichment in PRDM10 peaks (Fig. 2d) and strong sequence conservation within PRDM10-bound sites compared with background genomic regions (Fig. 2e), leading us to hypothesize that it may be a functionally relevant candidate for DNA binding by PRDM10. To define the transcriptional impact of the sequence-specific recognition of this motif by PRDM10, we performed reporter assays in HEK293T cells transfected with constructs containing either the WT motif or a mutated version (MUT) cloned upstream of a minimal promoter to drive expression of a firefly luciferase gene (Fig. 2f). We observed strong activation of the WT motif reporter with PRDM10 overexpression; however, mutation of the consensus sequence fully abolished PRDM10-dependent reporter activation (Fig. 2f), showing that the presence of a specific cis-regulatory DNA sequence is required for PRDM10-mediated transcriptional activity.We performed gel shift assays to determine if PRDM10 binds directly to its putative DNA motif. A recombinant GST-fusion PRDM10 protein containing its central zinc finger array (GST-PRDM10441–880) was used for binding assays. We detected robust binding of PRDM10441–880 protein to labeled probe containing its cognate motif (Fig. 2g). This binding was specifically diminished by competition with excess unlabeled WT probe (Fig. 2h). In contrast, the MUT probe was less efficiently bound by PRDM10441–880 in direct binding (Fig. 2g) as well as competition assays (Fig. 2h). Our findings demonstrate specific, zinc finger-mediated binding of PRDM10 to the consensus sequence identified by our ChIP-seq analysis.Identification of PRDM10 transcriptional activation domainsTo elucidate the mechanisms by which PRDM10 regulates transcription, we sought to characterize the function of its N-terminal PR domain and C-terminal Q-rich region. Q-rich unstructured domains have been implicated in transcriptional activation18 and PRDM10 is the only family member harboring such domain. On the other hand, the biochemical function of the PR domain varies among PRDM family members19, and that of PRDM10 is currently unknown. We evaluated a series of PRDM10 deletion mutants (Fig. 2i, j) for their ability to activate a luciferase reporter construct containing the PRDM10 consensus motif. In line with our gel shift assay data (Fig. 2g, h), mutants lacking the central zinc finger DNA-binding domain (ZF-DBD) failed to elicit any activity, underscoring the importance of the ZF-DBD in recruiting PRDM10 to target DNA. The ZF-DBD alone was insufficient to activate transcription, indicating a requirement for additional effector domains (Fig. 2i). Expression of the ZF-DBD in combination with either the N-terminal PR domain (N-880) or C-terminal Q-rich domain (441-C) was sufficient for transcriptional activation, albeit at lower levels compared with that achieved by full-length PRDM10 (Fig. 2i). These results demonstrate that PRDM10 engages its targets via zinc finger-mediated sequence-specific DNA binding and drives target transcription by means of N- and C-terminal activation domains.Gene expression changes in Prdm10-null mESCs and embryosWe reasoned that the phenotypes observed in Prdm10Δ/Δ mESCs and embryos were the outcome of transcriptional misregulation following the loss of PRDM10. Therefore, we performed gene expression profiling by RNA-seq in both systems: (1) Prdm10Δ/Δ mESCs vs. vehicle-treated Prdm10F/F control mESCs; (2) Prdm10Δ/Δ vs. control single embryos obtained from Prdm10Δ/+ intercrosses. In order to predominantly capture transcriptional changes due to misregulation of primary target genes, we isolated 8-cell stage embryos at E2.5, one day before the onset of the phenotypic defects, within a developmental window where mutant embryos are morphologically indistinguishable from WT and heterozygous littermates (Supplementary Fig. 1e). For the same reason, we chose to analyze mESCs at Day 2 and 4 post-deletion (Supplementary Fig. 7a), prior to the onset of significant growth inhibition and cell death (Fig. 1f, i).Comparing PRDM10-deficient mESCs to controls at 2 days post-deletion, we found that 8.2% (n = 953) of genes were significantly up- and 8.7% (n = 1019) were downregulated (Padj < 0.05). More extensive transcriptional changes were observed at Day 4, with 20.7% (n = 2393) of genes being up- and 18.7% (n = 2157) being significantly downregulated (Padj < 0.05) in PRDM10-deficient mESCs (Supplementary Fig. 7a, Supplementary Data 2 and 3). Functional annotation of differentially expressed genes (Padj < 0.05) at Day 4 post-deletion revealed a gene expression signature significantly enriched in GO terms associated with ribosomal function, translation and peptide metabolism (Supplementary Fig. 7b, Supplementary Data 4), pointing to potential misregulation of protein synthesis processes in PRDM10-deficient mESCs.Analysis of RNA-seq data from early embryos revealed that 0.48% (n = 50) of genes were upregulated and 0.79% (n = 83) were downregulated (Padj < 0.05) in Prdm10Δ/Δ embryos compared with Prdm10+/+ and Prdm10Δ/+ controls (Supplementary Fig. 7c, Supplementary Data 5).PRDM10 binds and transcriptionally activates target genesTo further narrow the list of direct and relevant candidate PRDM10 targets, we integrated ChIP- and RNA-seq data. By ChIP-seq, we identified 633 unique genes associated with PRDM10 binding sites (Supplementary Data 6). Of these, 52 and 76 were differentialy expressed (Padj < 0.05, fold-change > 2) in mESCs at 2 and 4 days post-deletion, respectively (Fig. 3a, Supplementary Data 7). Notably, the majority of genes bound and regulated by PRDM10 (Padj < 0.05, fold-change > 2) showed decreased expression in Prdm10Δ/Δ mESCs (Fig. 3a). In contrast, genes that were not direct targets of PRDM10 displayed expression changes in both directions and are thus likely secondary targets (Fig. 3a). The same was true for embryos, in which loss of PRDM10 led to downregulation of all 28 differentially expressed direct targets (Padj < 0.05, fold-change > 2) (Fig. 3b, Supplementary Data 7). Taken together with our reporter assay data (Fig. 2f, i), these findings indicate that PRDM10 functions primarily as a transcriptional activator of its target genes.Fig. 3PRDM10 binding is associated with transcriptional activation of target genes.a Volcano plots of RNA-seq data from Prdm10Δ/Δ vs. Prdm10F/F mESCs at indicated time-points post-deletion, with Padj < 0.05, fold-change (FC) > 2 used as the cut off to define genes with significant changes in expression. Red: genes directly bound by PRDM10; green: genes not bound by PRDM10 (left panels). Of these genes, PRDM10 targets are predominantly downregulated (right). b Volcano plot of RNA-seq data from Prdm10Δ/Δ (KO) vs. Prdm10+/+ and Prdm10Δ/+ (CTL) 8-cell stage embryos (left). 28 genes bound by PRDM10 showed significant expression changes (Padj < 0.05, fold-change > 2); of these, all were downregulated in Prdm10-null embryos (right). c Venn diagram depicting the overlap of bound and downregulated genes in RNA-seq datasets for embryos (8-cell) and mESCs (Day 2 and/or Day 4 post-induction), with a total of 18 genes identified as top candidates. d Heatmap showing all 18 bound and downregulated genes from Fig. 3c, categorized by their respective gene knockout phenotypes. Asterisks indicate genes directly involved in translation, Eif3b and Eef1d. Color scale: Z-score for row-normalized expression values, scaled separately for Day 2 p.i mESCs, Day 4 p.i mESCs, and embryos. Pre-implant: preimplantation lethality; post-implant: postimplantation lethality; undefined: developmental timing of embryonic lethality unknown.Identification of Eif3b as a key downstream target of PRDM10To determine the mechanism underlying the phenotypes observed in PRDM10-deficient mESCs and early embryos, we attempted to identify direct targets of PRDM10 that might be functionally relevant in both systems. We compared PRDM10-bound genes that were significantly downregulated (Padj < 0.05, fold-change > 2) across all three RNA-seq datasets, and identified an overlapping set of 18 genes misregulated in both mESCs and embryos (Fig. 3c, d). Of these, seven genes are viable in the homozygous null condition, seven genes are embryonic lethal when deleted in mice, and the remaining four genes have no phenotype reported in the literature (Fig. 3d). We chose not to focus on homozygous viable genes as they were least likely to be relevant to the Prdm10-null embryonic lethal phenotype. Among the seven candidate genes associated with embryonic lethality, only Eif3b–the eukaryotic translation initiation factor 3 (eIF3) subunit B–has been implicated specifically in preimplantation development (Fig. 3d).Homozygous deletion of Eif3b results in early embryonic lethality by E3.520, within a developmental time-frame similar to that observed in Prdm10Δ/Δ embryos. Eif3b encodes a highly conserved core component of the multi-subunit eIF3 complex21, which promotes mRNA recruitment to the pre-initiation complex (PIC) as a necessary step in translation initiation22–25. The well-established role of Eif3b in translation is consistent with a gene expression signature in Prdm10Δ/Δ mESCs pointing to disrupted protein synthesis (Supplementary Fig. 7b). Furthermore, Eif3b has been identified as a positive hit in two genome-wide CRISPR screens for genes essential in mESCs26 (Supplementary Fig. 8a), suggesting it is likely to act as a critical mediator of mESC survival downstream of PRDM10.In addition to Eif3b, we identified two other PRDM10-regulated targets involved in translation and protein synthesis: Eef1d (eukaryotic translation elongation factor 1 delta) and Rpl19 (ribosomal protein L19). Eef1d was strongly downregulated in Prdm10-null embryos as well as mESCs (Fig. 3d, Supplementary Fig. 7d, e). However, published CRISPR screens in mESCs26 did not support an essential requirement for Eef1d (Supplementary Fig. 8a), and consistent with this, shRNA-mediated knockdown of Eef1d in mESCs had no observable effect on cell growth (Supplementary Fig. 8c, d, g). On the other hand, Rpl19 depletion led to significantly reduced cell growth (Supplementary Fig. 8b, e, f, g), suggesting that it may also be essential for mESC viability. However, because Rpl19 was only modestly downregulated (<2-fold) in PRDM10-deficient embryos, while Eif3b showed a <2-fold downregulation in vivo (Supplementary Fig. 7d), we prioritized Eif3b for functional validation.Effects of Eif3b depletion in embryos and mESCsIn an effort to obtain direct evidence supporting an essential role for Eif3b in preimplantation development, we performed knockdown experiments both in vivo and in vitro. WT mouse zygotes were microinjected with siRNAs, and allowed to develop to blastocyst stage over 4 days in culture (Fig. 4a). Strikingly, siRNA-mediated knockdown of Eif3b resulted in complete developmental arrest just prior to blastocyst formation (n = 40): no Eif3b-deficient embryos were observed to form mature, cavitated blastocysts (Fig. 4a) and most exhibited a morula-like morphology (Fig. 4b). This is in agreement with previous studies indicating a pre-E3.5 lethality phenotype for Eif3b-null embryos20. In contrast, siRNA-mediated depletion of other selected PRDM10-regulated candidate genes, including Selenow (n = 93), Ss18 (n = 87), and Ube2a (n = 66), showed no significant impact on embryo development (Fig. 4a).Fig. 4Loss of EIF3B causes lethality in preimplantation embryos and mESCs.a
Eif3b-deficient embryos arrest prior to blastocyst formation. Expression of candidate target genes was knocked down in wild-type zygotes by injection of siRNA: control (n = 38), Eif3b (n = 40), Selenow (n = 93), Ss18 (n = 87), and Ube2a (n = 66), and blastocyst morphology was scored after 4 days’ embryo culture. Y-axis: percentage of cavitated or non-cavitated blastocysts of total embryos analyzed in each experiment, for at least 3 independent experiments per target gene. b Representative images of embryos treated with control or Eif3b-targeting siRNA, acquired at 4.5 dpc. Scale bar: 100 μm. c Western blot analysis of E14 mESCs overexpressing Eif3b or vector control, transduced with indicated shRNA. Numbers in bottom row represent quantification of relative EIF3B protein levels after background subtraction and normalization to α-tubulin. d qRT-PCR validation of Eif3b knockdown in E14 mESCs transformed with vector control (vector) or Eif3b overexpression construct (Eif3b OE) and transduced with shRNA targeting the 3′UTR of Eif3b (shRNA-479). 3′UTR-specific primers were used to detect endogenous Eif3b transcript. n = 2 samples, data shown from one out of three independent experiments with similar results. e E14 mESCs transduced with shRNA-479 to deplete endogenous Eif3b exhibit slower growth (vector + 479; black dashed line), while expression of shRNA-resistant Eif3b restores normal growth (Eif3b OE + 479; red dashed line). Cells were plated at equal densities, passaged at Day 2 and Day 4, and counted at indicated time-points. n = 3 replicates, representative data shown (from same experiment as in Fig. 4d). Error bars denote mean ± s.d.; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, n.s., not significant; two-tailed unpaired Student’s t test (a and d), two-way ANOVA with Tukey’s multiple comparisons test (e).For in vitro validation, we transfected E14 mESCs with siRNAs to knock down Eif3b expression (Supplementary Fig. 9a), and observed severely impaired cell growth in Eif3b-depleted cells compared with controls up to 72 h post-transfection (Supplementary Fig. 9b, c). Similar results were obtained utilizing a lentiviral shRNA knockdown approach, in which a 3′UTR-specific hairpin sequence (Eif3b-479) was employed to disrupt endogenous Eif3b expression. A modest but consistent reduction in Eif3b mRNA and protein levels (Fig. 4c, d) was sufficient to significantly inhibit cell growth in Eif3b-479 shRNA-transduced cells (Fig. 4e). We verified that the observed phenotype was not due to off-target effects, as it was successfully rescued by expression of shRNA-resistant Eif3b lacking the targeted 3′UTR (Fig. 4c–e). Collectively, these results demonstrate that Eif3b functions as an essential gene in early embryos as well as mESCs.PRDM10 directly activates transcription of Eif3bBecause our data showed robust PRDM10 occupancy at the Eif3b promoter (Fig. 5a), as well as significant downregulation of Eif3b transcript in Prdm10Δ/Δ mESCs and embryos (Fig. 3d, Supplementary Fig. 7d, e, Supplementary Data 3 and 5), we next investigated the hypothesis that Eif3b is a direct target of PRDM10-mediated regulation. From ChIP-seq data, we identified two adjacent PRDM10-binding regions (P1 and P2) within the Eif3b promoter (Fig. 5b), each containing a consensus motif match (Supplementary Fig. 9d). Both P1 and P2 stimulated luciferase reporter activity driven by PRDM10 overexpression (Fig. 5b), supporting their role as cis-regulatory sequences that recruit PRDM10 to activate Eif3b transcription. We further verified that Prdm10Δ/Δ mESCs had significantly reduced expression of Eif3b mRNA (Fig. 5c) and protein (Fig. 5d), particularly at later time-points post-deletion. Moreover, Eif3b transcript levels in Prdm10Δ/Δ mESCs were fully restored upon re-expression of exogenous PRDM10 (Fig. 5e). Hence, PRDM10 is both necessary and sufficient to maintain normal Eif3b expression levels in mESCs.Fig. 5PRDM10 transcriptionally regulates Eif3b expression.a ChIP-seq signal tracks showing PRDM10 occupancy at the Eif3b promoter in EtOH- vs. OHT-treated Prdm10F/F; CreERT2 mESCs, analyzed at 2 days post-induction. Vertical axis: fold-change over input. b PRDM10 enrichment at the Eif3b promoter is detected as two subpeaks, denoted P1 and P2 (left). PRDM10-dependent activation of luciferase reporters containing Eif3b promoter sequences derived from subpeak P1 or P2 (right); n = 2 samples. Representative data shown from one out of three independent experiments. c qRT-PCR quantification of changes in Eif3b transcript at indicated time-points after Prdm10 deletion; n = 2–3 biological replicates. d Western blot analysis of EIF3B protein levels in Prdm10F/F; CreERT2 mESCs at indicated time-points after treatment with EtOH (E) or 4-OHT (O). e qRT-PCR analysis showing restoration of Eif3b transcript levels in Prdm10Δ/Δ mESCs by exogenous Prdm10. Data shown at Day 4 post-induction; n = 3. Gene expression shown relative to Ubb (c and e). Error bars denote mean ± s.d.; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, n.s., not significant; two-tailed unpaired Student’s t test (b, c and e).EIF3B acts downstream of PRDM10 to promote translationFinally, we examined the functional importance of EIF3B downstream of PRDM10, by stably overexpressing exogenous Eif3b in Prdm10F/F; ROSA26-CreERT2 mESCs (Fig. 6a, b), followed by 4-OHT treatment to induce Prdm10 deletion. Strikingly, restoring Eif3b expression was in itself sufficient to achieve a partial phenotypic rescue in Prdm10Δ/Δ mESCs, reducing doubling time by almost 40% compared with “vector-only” control Prdm10-null cells (Fig. 6c, d; Supplementary Fig. 9e). The partial nature of the phenotypic rescue by EIF3B points towards, unsurprisingly, additional contributions by other PRDM10-regulated targets (Fig. 3d, Supplementary Data 7). For example, we have identified Rpl19 (Supplementary Fig. 8b, e) as another PRDM10-regulated target with a key role in protein synthesis that may also contribute to the growth defect in Prdm10Δ/Δ mESCs. Nonetheless, our results strongly support a critical role for EIF3B as one of the major mediators of mESC survival downstream of PRDM10, and suggest the possibility that EIF3B may also function in a similar capacity in the context of PRDM10-deficient preimplantation embryos.Fig. 6EIF3B promotes global translation downstream of PRDM10.a Immunoblotting detection of FLAG-tagged EIF3B protein expression in Prdm10F/F; CreERT2 mESCs transformed with pJ549-Eif3b or vector. b qRT-PCR validation of Eif3b transcript in Prdm10F/F; CreERT2 mESCs transformed with pJ549-Eif3b or vector, analyzed 6 days post-induction. CDS-specific primers were used to detect both endogenous and overexpressed Eif3b transcript; n = 3 biological replicates. c Growth curve analysis of Prdm10F/F and Prdm10Δ/Δ mESCs in the presence of empty vector or exogenous Eif3b, showing partial phenotypic rescue by Eif3b overexpression. Cells were passaged at constant density every 2 days and counted daily up to Day 8 post-induction; n = 3. Representative data shown from one of three independent experiments. d Average population doubling times over the course of the growth assay for Prdm10F/F and Prdm10Δ/Δ mESCs expressing vector control or exogenous Eif3b. Each point represents data from one independent experiment; n = 3. e Representative polysome profiles of Prdm10F/F and Prdm10Δ/Δ mESCs analyzed 5 days post-induction (left). A significantly reduced polysome:monosome (P/M) ratio indicates impaired global translation in Prdm10Δ/Δ mESCs (right); n = 5 biological replicates, across two independent experiments. f Representative polysome profiles of Prdm10F/F and Prdm10Δ/Δ mESCs expressing Eif3b or vector, analyzed at Day 5 post-induction (left). Eif3b overexpression restores global translation in Prdm10Δ/Δ mESCs to levels comparable to Prdm10F/F controls, as quantified by P/M ratio (right); n = 5 biological replicates across two independent experiments. Gene expression presented relative to Ubb (b). Data presented as mean ± s.d.; **P < 0.01, ****P < 0.0001, n.s., not significant; one-way ANOVA with Tukey’s multiple comparisons test (b, d and f); two-way ANOVA with Tukey’s multiple comparisons test (c).Mammalian eIF3 is a large complex comprising 13 protein subunits, and is essential for stimulating multiple steps of the translation initiation pathway22. Given that EIF3B is one of the most highly conserved core subunits, with a critical role in the nucleation and function of the eIF3 complex21, we reasoned that loss of EIF3B would lead to decreased global translation in Prdm10-null mESCs. Consistent with this hypothesis, polysome profile analysis of Prdm10Δ/Δ mESCs revealed a dramatic reduction in translation rates, with a >3-fold decrease in the polysome-to-monosome (P/M) ratio compared with Prdm10F/F controls (Fig. 6e). Importantly, this effect was fully rescued by re-expression of exogenous EIF3B (Fig. 6f), demonstrating that the global translation defect in Prdm10-null cells was specifically caused by loss of EIF3B. Though not statistically significant, we noted a trend towards slightly higher P/M ratios in EIF3B-overexpressing cells compared with vector-transduced Prdm10F/F controls, suggestive of translation rates above baseline and consistent with previous reports that overexpression of the EIF3B subunit is sufficient to elevate levels of the entire eIF3 complex, thereby activating protein synthesis in cancer cell lines27. Taken together, our data strongly suggest that the growth defect observed in Prdm10Δ/Δ mESCs is a functional consequence of decreased translation efficiency arising from misregulation of Eif3b.DiscussionIn this study, we demonstrate an essential role for PRDM10 as a transcriptional regulator in early mammalian embryogenesis and mESC homeostasis. Our findings strongly support a model whereby PRDM10 supports cell growth and survival during early development by transcriptionally regulating Eif3b expression to sustain global translation. Although this work focuses on phenotypes related to early development, this does not imply that the function of PRDM10 is strictly specific to early embryogenesis, or that PRDM10 regulates processes unique to mESCs. Given that PRDM10 is expressed across multiple tissues and regulates a broad range of target genes, it is highly likely to have pleiotropic effects that may be revealed using different models for conditional deletion at later developmental stages or in specific tissues. Interestingly, PRDM10 gene fusions have been implicated in the pathogenesis of undifferentiated pleomorphic sarcoma15,16, raising the possibility of other yet-unknown roles in human disease. Hence, the data presented in this work provide a strong starting point for future studies to further extend our understanding of PRDM10 in development and disease.The PRDM family first appeared in metazoans, and Prdm10 is thought to have evolved during a family expansion just prior to vertebrate evolution. PRDM genes that emerged later in evolution tend to be specifically expressed in highly specialized cells to serve tissue-specific functions; examples include Prdm14 (germ cells and embryonic stem cells), Prdm7 (melanocytes), and Prdm9 (testis)28. Surprisingly, despite being one of fastest evolving paralogs, Prdm10 is expressed across a broad range of adult tissues28, and is the first example of a PRDM family member being implicated in early embryo development prior to mid-gestation. It is also interesting to note that while Prdm10 evolved fairly recently, the eIF3 complex arose before metazoan evolution, and EIF3B is one of five eIF3 subunits that are conserved in all eukaryotes. The significance and implications of PRDM10 regulating such an evolutionarily ancient gene remains an open question. Evolutionary expansion and divergence within zinc finger protein families drives diversification of DNA binding specificities and effector functions, and is highly correlated with increasing organism complexity during vertebrate evolution29. Recent findings have challenged the traditional view of eIF3 as a general translation initiation factor, suggesting that eIF3 may also exert selective translational control over specific mRNAs30. It is therefore intriguing to speculate that PRDM10 may contribute additional layers of regulatory control over Eif3b expression in a context-dependent or tissue-specific manner, or during different developmental stages.Although protein synthesis has traditionally been regarded as a basic cellular housekeeping function, emerging evidence suggests that it is dynamically regulated during development and plays a significant role in the maintenance of stem cells31–34. Furthermore, it has been reported that EIF3B, along with other proteins involved in translation initiation, is upregulated at the morula and blastocyst stages, suggesting a key role for protein synthesis in supporting embryonic growth during this developmental transition35. Notably, multiple studies have demonstrated the importance of global translational control in ESC pluripotency and differentiation36–40. However, despite major advances in our understanding of the transcriptional and epigenetic mechanisms governing ESC function41,42, much less is known about how regulation is achieved at the level of protein translation. As translation is mostly regulated at the initiation stage, eukaryotic initiation factors have been extensively characterized in terms of their biochemical and structural properties23; however, few studies have examined their potential roles in stem cells and development. By deciphering the molecular mechanisms underlying the requirement for PRDM10 in mESCs, our work provides insight into how precise transcriptional control of the translation machinery may modulate global translation to influence development.MethodsMicePrdm10 conditional knockout mice were generated on a C57BL/6 background using an ES cell clone (Prdm10tm1a(EUCOMM)Hmgu) from the EUCOMM consortium containing a knockout-first cassette targeting exon 5 of Prdm10. After successful germline transmission, Prdm10lacZ/+ mice were crossed to a FLPe recombinase transgenic line (C57BL/6-Tg(CAG-Flpe)2Arte, Taconic) to remove the neomycin selection cassette and generate the Prdm10 flox allele. Prdm10F/+ mice were then bred to an ACTB-Cre recombinase transgenic line to generate the Prdm10Δ null allele. 4-OHT-inducible knockouts were created by crossing Prdm10F/F mice with a ROSA26-CreERT2 transgenic strain. Mice were housed in specific pathogen-free conditions and maintained on a 12 h light-dark cycle with food and water available ad libitum. All procedures involving mice were performed in compliance with the Institutional Animal Care and Use Committee protocols #151042 and #181393, with the approval of the Biological Resource Centre (BRC), A*STAR.GenotypingFor genotyping of mice, crude DNA extracts were prepared from tail biopsies by overnight lysis at 55 °C in DirectPCR Lysis Reagent (Viagen) containing 0.2 mg/ml proteinase K. For genotyping of embryos, each embryo was lysed for 1 h at 55 °C in 10 μl lysis buffer (50 mM KCl, 10 mM Tris-HCl pH 8.3, 2 mM MgCl2, 0.45% NP-40, 0.45% Tween-20, 0.5 mg/ml proteinase K), followed by heat-inactivation at 95 °C for 10 min. 2 μl extract was used as template per 20 μl PCR reaction with 1X DreamTaq PCR master mix (Thermo Fisher Scientific). Primers used for genotyping are listed in Supplementary Data 8.Plasmid constructionFor overexpression studies, full-length Prdm10 was cloned by gene synthesis (AITBiotech) based on sequence information obtained from RNA-seq analysis and Ensembl transcript annotations. For stable expression in mESCs, Prdm10 was ligated into the NheI and BamHI sites of the pJ549 PiggyBac transposase expression vector (DNA 2.0), modified to contain an N-terminal FLAG tag. For transient transfections, Prdm10 was cloned into pcDNA3.1 (Invitrogen) downstream of an N-terminal FLAG tag via EcoRI and NotI restriction sites. For the construction of shRNA lentiviral vectors, gene-specific hairpin sequences were selected from the RNAi Consortium TRC lentiviral shRNA library and cloned into pLKO.1-Neo (Addgene, #13425) as annealed oligonucleotides (Supplementary Data 8). Endotoxin-free plasmids were prepared using the Nucleobond Xtra Midi EF Kit (Macherey-Nagel). All constructs were verified by Sanger sequencing.Embryo isolation and cultureNatural matings were set up and successful copulation (assumed to have occurred at 12 midnight) was ascertained by the presence of a vaginal plug. Pregnant females were sacrificed by cervical dislocation at 36 h post-copulation (E1.5) for 2-cell embryos and 84–90 h post-copulation (E3.5) for blastocysts. Preimplantation embryos were collected from the infundibulum (E1.5) or uterine horns (E3.5) by flushing with M2 medium (Millipore). Postimplantation embryos (E4.5–12.5) were dissected from the uteri. For in vitro studies, embryos were isolated at the 2-cell stage, cultured for 3 days in KSOM + AA medium (Millipore) at 37 °C with 5% CO2, and imaged at indicated time-points. Data were pooled from heterozygous intercrosses with Prdm10lacZ/+ and Prdm10Δ/+ mice as both mutant alleles yielded equivalent phenotypic outcomes.Immunofluorescence (IF) microscopyEmbryos were fixed in 4% paraformaldehyde for 30 min and permeabilized with 0.1% Triton X-100 for 30 min at room temperature. After blocking with 1% FBS, samples were incubated at 4 °C overnight using the following primary antibodies and dilutions: OCT4 (1:100; sc-5279, Santa Cruz), CDX2 (1:100; ab88129, Abcam), and NANOG (1:100; RCAB002P-F, ReproCELL). Samples were then washed and incubated in Alexa Fluor-conjugated secondary antibodies (1:500; Life Technologies) at room temperature for 1 h. Images were acquired on either an Olympus Fluorview 1000 (60× oil immersion objective) or Zeiss LSM800 (63× oil immersion objective) confocal laser-scanning microscope.Derivation of mouse ES cell linesBlastocysts were isolated at E3.5 and cultured in 2i/LIF medium on a feeder layer of mitotically inactivated mouse embryonic fibroblasts (MEFs). After 5 days, ICM outgrowths were disaggregated using a fine Pasteur pipette with trypsin-EDTA treatment, re-plated on MEF feeders in 2i/LIF medium, and gradually expanded over 4–5 passages. 2i/LIF medium comprised a 1:1 mix of DMEM/F12 (Gibco) and Neurobasal (Gibco) media, supplemented with GlutaMAX (Gibco), 0.05% bovine serum albumin (BSA), 100 U/ml penicillin/streptomycin, N2 (Gibco), B27 (Gibco), 10 μM 2-mercaptoethanol (Gibco), 1000 U/ml LIF (Millipore), 3 mM CHIR99021 (Axon), and 1 mM PD0325901 (Axon). To determine genotypes of newly-derived mESC lines, cells were plated on 0.1% gelatin for 1 h to allow feeder MEFs to preferentially attach, after which MEF-depleted ESCs were collected and genomic DNA purified using the DNeasy Blood & Tissue kit (Qiagen).Cell culture and transfectionAll cell cultures were maintained at 37 °C in a humidified incubator with 5% CO2. mESCs were cultured on 0.1% gelatin-coated plates in Dulbecco’s Modified Eagle’s Medium (DMEM, Hyclone) supplemented with 15% fetal bovine serum (FBS), GlutaMAX, nonessential amino acids, 1 mM sodium pyruvate, 100 U/ml penicillin/streptomycin, 5.5 μM 2-mercaptoethanol and 1000 U/ml mLIF (Millipore). For routine propagation, cells were trypsinized, resuspended in 10% FBS DMEM, and re-seeded at a ratio of 1:20–1:40 every 2–3 days. Plasmid and siRNA transfections into mESCs and HEK293T cells were performed using Lipofectamine 2000 (Thermo Fisher Scientific). All cell lines used in this study tested negative for mycoplasma contamination by a PCR-based assay (Biological Industries).Induction of Prdm10 deletionPrdm10F/F; ROSA26-CreERT2 mESCs were treated for 24 h with 50 nM 4-OHT (Sigma) or an equal volume of ethanol. Cells were then washed in dPBS and seeded for assays.Generation of cell lines with stable transgene expressionMouse ESCs were transfected with pJ549 PiggyBac expression plasmids (DNA 2.0) by reverse transfection with Lipofectamine 2000. Transfected E14 mESCs were selected in 0.75 μg/ml puromycin for 3 days. For stable transgenesis of Prdm10F/F; ROSA26-CreERT2 mESCs, GFP-positive cells were purified by fluorescence-activated cell sorting on a MoFlo XDP cell sorter (Beckman Coulter) and expanded in culture for assays.Apoptosis assayOne day before the indicated time-points, mESCs were seeded in a 96-well plate at a density of 2 × 104 cells per well in quadruplicate, and allowed to attach overnight. To detect caspase activity, an equal volume of Caspase-Glo 3/7 Assay reagent (Promega) was added to each well and cells were homogenized by gentle agitation. Samples were transferred to a 96-well white flat bottom plate (Corning), incubated for 30 min at room temperature, and luminescence readings acquired on a GloMax instrument (Promega). For each condition, background-subtracted Caspase-Glo signals in each well were normalized to cell numbers, which were measured in parallel using the CellTiter-Glo Cell Viability Assay (Promega).Cell growth analysismESCs were seeded in triplicate at a density of 1 × 105 cells per well in a 12-well plate. At indicated time-points, cells were harvested by trypsinization and total viable cell counts were measured by trypan blue exclusion on the Countess II Automated Cell Counter (Thermo Fisher). Brightfield images were acquired on a Nikon Eclipse TS100 inverted microscope with a 4× objective.Colony formation assaymESCs were seeded in 6-well plates at a density of 1 × 104/well and cultured for 5 days with regular medium changes. Cells were then washed twice in dPBS, fixed in ice-cold methanol for 10 min and stained with 0.5% crystal violet solution in 25% methanol. After washing and air-drying, colonies were imaged on a flatbed scanner and quantified using Fiji/ ImageJ.shRNA lentiviral production and transductionFor lentiviral production, every 1 μg of pLKO-Neo plasmid was co-transfected with 0.5 μg pMD2.G (Addgene) and 0.375 μg psPAX2 (Addgene) into HEK293T cells using Lipofectamine 2000. Viral supernatants were collected 48 and 72 h post-transfection, filtered through a 0.45 μm syringe filter, and centrifuged at 24,000 r.p.m. for 2 h at 4 °C. The viral pellet was reconstituted in Hank’s Balanced Salt Solution and stored in aliquots at −80 °C. E14 mESCs were infected with lentivirus supplemented with 8 μg/ml polybrene (Merck Millipore) and cultured in medium containing 400 μg/ml G418 (Invivogen) to select for stable transductants. After 96 h of selection, cells were re-seeded and maintained in 200 μg/ml G418 for the duration of the assays.siRNA-mediated knockdown in mESCs2.5 × 104 E14 mESCs were transfected in suspension with 20 pmol ON-TARGETplus SMARTpool siRNA (Dharmacon) and 1 μl Lipofectamine 2000 and seeded in 24-well plates. Cell viability was assessed at 48 and 72 h post-transfection.RNA extraction and qRT-PCRSamples were lysed in TRIzol (Invitrogen) and total RNA was purified using the PureLink RNA Mini Kit (Invitrogen) with on-column DNase treatment according to manufacturer’s instructions. Reverse transcription was performed using the Maxima First Strand cDNA Synthesis Kit (Thermo Scientific) with ~1 μg RNA per reaction. cDNA was diluted 10-fold with nuclease-free water for use in downstream assays. Quantitative real-time PCR was performed on a CFX96 Touch Real-Time PCR Detection System (Bio-Rad) using PowerUp SYBR Green Master Mix (Applied Biosystems). Target gene expression relative to an internal reference gene (Ubb) was calculated using 2−ΔCt. Primer sets were validated for specificity by melt-curve analysis and tested for linear amplification over four orders of magnitude. Primers used for qRT-PCR are listed in Supplementary Data 8.Western blot analysisWhole cell lysates were prepared in reducing sample buffer (32.9 mM Tris-HCl, 12.5% glycerol, 1% SDS, 2.5% 2-mercaptoethanol, 27 mg/ml DTT, 0.005% bromophenol blue) and heated at 98 °C for 10 min. Protein concentrations were measured using the RC DC Protein Assay (Bio-Rad). 20–40 μg total protein was loaded per well and samples were separated by SDS-PAGE gel electrophoresis in Tris-Glycine-SDS buffer (1st Base). Proteins were then transferred to Immun-Blot PVDF membranes (Bio-Rad) by wet electroblotting in Tris-glycine buffer containing 10% methanol. Membranes were blocked for 1 h at room temperature in TBS-T (Tris-buffered saline + 0.05% Tween-20) containing 5% milk or 3% BSA. Blots were incubated overnight at 4 °C with the following primary antibodies diluted in 3% BSA/TBS-T:PRDM10 (1:1000; A303-204A, Bethyl Laboratories), EIF3B (1:1000; VPA00380, Bio-Rad), FLAG M2 (1:1000, F1804, Sigma), alpha-tubulin (1:10,000; T5168, Sigma), beta-actin (1:1000; sc-47778, Santa Cruz). After three washes in TBS-T, blots were incubated with HRP-conjugated anti-mouse (1:10,000; sc-516102, Santa Cruz) or anti-rabbit (1:10,000; sc-2357, Santa Cruz) secondary antibody for 1 h at room temperature. SuperSignal West Femto Maximum Sensitivity Substrate or SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific) was used for chemiluminescent detection. Blots were imaged on a ChemiDoc Touch (Bio-Rad) and analyzed with Image Lab software (Bio-Rad).Luciferase reporter assaysTo characterize the PRDM10 binding motif, oligonucleotides containing the motif sequence were annealed and ligated into the NheI and BglII sites of pGL4.23 [luc2/minP] (Promega). Eif3b promoter sequences were amplified by PCR from mouse genomic DNA and cloned into pGL4.23 using KpnI and XhoI. pGL4.23 reporter constructs were co-transfected with pGL4.74 [hRluc/TK] and pcDNA3-Prdm10 expression plasmids into HEK293T cells cultured on 12-well plates. At 48 h post-transfection, firefly and Renilla luciferase activities were measured on a GloMax luminometer (Promega) using the Dual-Luciferase Reporter Assay System (Promega).EMSALabeled probes (Supplementary Data 8) were prepared by annealing 5′-biotinylated oligonucleotides (IDT). Binding reactions with purified recombinant protein and 10 nM labeled probe were performed at room temperature for 20 min in buffer containing 6 mM HEPES-KOH pH 7.9, 6% (v/v) glycerol, 100 μg/mL BSA, 0.4 μM ZnCl2 and 20 μg/mL poly(dI-dC). Samples were electrophoresed on 6% native polyacrylamide gels in 0.25× TBE buffer at 150 V, and transferred to Hybond N + nylon membranes (Amersham Pharmacia Biotech) in 0.5× TBE at 380 mA for 1 h. Biotinylated DNA was detected using the LightShift Chemiluminescent EMSA kit (Thermo Fisher Scientific). Blots were imaged on a ChemiDoc Touch (Bio-Rad) and analyzed using Image Lab (Bio-Rad).Flow cytometryFor cell cycle analysis, mESCs were seeded in six-well plates such that they were 40–60% confluent on the day of harvest. Cells were pulsed with 10 μM 5-ethynyl-2′-deoxyuridine (EdU) for 30 min, washed with PBS and dissociated into a single cell suspension by trypsin-EDTA. Fixation, permeabilization and EdU labeling was performed using the Click-iT Plus EdU Alexa Fluor 488 Flow Cytometry Assay Kit (Thermo Fisher Scientific) according to manufacturer’s instructions. Cells were incubated in 10 μg/ml DAPI on ice for >1 h to stain DNA and filtered through a 40 μm cell strainer immediately before data acquisition. To assess SSEA-1 surface expression, staining was carried out in FACS buffer (2% FBS in PBS) using mouse anti-SSEA-1 conjugated to Alexa Fluor 647 (BD Pharmingen, clone MC480) with 30 min incubation on ice. All flow cytometry data were acquired on a BD LSR II flow cytometer (BD Biosciences) and analyzed using FlowJo v10 (FlowJo LLC).Polysome fractionationAt Day 3 post-induction, mESCs were seeded at a density of 1 × 106 (Prdm10F/F) or 2.5 × 106 (Prdm10Δ/Δ) cells per 10-cm dish such that they were at similar confluence for analysis at Day 5. Cells were treated with 100 μg/ml cycloheximide (Sigma) for 10 min at 37 °C to arrest translation, then washed in ice-cold PBS and harvested on ice by scraping in 800 μl lysis buffer (10 mM Tris-HCl pH 7.4, 5 mM MgCl2, 100 mM KCl, 1% Triton X-100) supplemented with 2 mM DTT, 100 U/ml RNasin (Promega), protease inhibitor cocktail and 100 μg/ml cycloheximide. Lysates were sheared using a 26G needle and cleared by centrifuging at 1300 × g for 10 min at 4 °C. Clarified lysates were layered onto 10–50% sucrose gradients and centrifuged in an SW-41Ti rotor at 36,000 r.p.m. for 2 h. Gradients were fractionated using a BioComp Gradient Station fractionator, and absorbance at 254 nm was monitored to obtain the polysome profile. Polysome/monosome (P/M) ratios were derived by integrating the area under the respective peaks using Microsoft Excel.ChIP-seqFeeder-free mESCs were harvested by trypsinization at 48 h post-induction and resuspended to 5 × 106 cells/ml in 10% FBS DMEM. Cells were cross-linked with 1% formaldehyde for 15 min at room temperature, quenched with 125 mM glycine and washed twice in cold PBS. Chromatin extracts were obtained by successive rounds of lysis in LB1 (50 mM HEPES-KOH pH 7.5, 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% Nonidet-P40, 0.25% Triton X-100), LB2 (10 mM Tris-HCl pH 8.0, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA) and LB3 (10 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% sodium deoxycholate, 0.5% N-lauroylsarcosine), supplemented with 0.2 mM PMSF and protease inhibitor cocktail. Chromatin DNA was sheared to a size range of 100–500 bp with 5–6 cycles of sonication at 30% amplitude using a Branson Digital Sonifier (S540D). Triton X-100 was added to a final concentration of 1% and lysates were cleared by centrifugation. 5 μg antibody was added to 100 μg of sonicated chromatin and incubated overnight with rotation at 4 °C. 40 μl Protein A Dynabeads were added to each reaction and incubated at 4 °C for 4 h. Beads were then collected on a magnetic rack and washed in low salt buffer (10 mM Tris-HCl pH 8.0, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100), high salt buffer (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS), LiCl buffer (10 mM Tris-HCl pH 8.0, 250 mM LiCl, 1 mM EDTA, 0.5% sodium deoxycholate, 0.5% Nonidet-P40), and TE buffer with 50 mM NaCl, then incubated with elution buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA, 1% SDS) at 65 °C for 20 min with continuous agitation. Eluted protein/DNA complexes were reverse-crosslinked overnight at 65 °C, treated with RNase A (Sigma) and proteinase K, and ChIP DNA was column-purified using QIAquick PCR Purification Kit (Qiagen). DNA concentrations of input and immunoprecipitated samples were measured on a Qubit instrument with Qubit dsDNA HS Assay kit (Thermo Fisher Scientific). Libraries were prepared from 4 ng of ChIP DNA using the NEBNext Ultra II DNA Library Preparation Kit for Illumina (NEB), largely following manufacturer’s instructions. All libraries were amplified for 9 PCR cycles and final elution volumes were reduced. Libraries were quantified by High Sensitivity DNA Assay on a Bioanalyzer and quantitative real-time PCR (KAPA Library Quantification Kit for Illumina, Roche). Final ChIP-Seq libraries were pooled and sequenced on an Illumina NextSeq 500 using single-end 75 bp reads to generate ~20 M raw reads per library.ChIP-seq data analysisChIP-seq data were processed using the ENCODE Transcription Factor and Histone ChIP-Seq pipeline (https://github.com/ENCODE-DCC/chip-seq-pipeline2; v1.1.7, commit 2f567e6e). Briefly, raw fastq files were aligned with bwa (v0.7.13) then filtered to remove duplicates (with Picard v2.10.6), multi-mapping reads and low-quality alignments (with samtools v1.2). SPP (v1.13) was then used to call peaks from the filtered alignments; in addition to calling peaks on each individual sample, peaks were called on the pooled set of alignments across all samples, as well as two pseudo-replicates obtained from the pooled set by splitting alignments into two equal sets. Peaks in the ENCODE mm10 blacklist were filtered out. The filtered peak sets were then assessed for reproducibility using IDR43 (v2.0.4.2). The final peak set used was the “optimal” peak set from the pipeline, which was obtained from the pooled pseudo-replicates with IDR cut-off of 0.05. ChIP-seq peak annotation analyses were carried out using the R package ChIPpeakAnno44 (v3.16.1). Using the annoPeaks function with GRCm38.p6 annotations, peaks were annotated with genes if they were located within 5 kb upstream to 1 kb downstream of the gene body. Distribution of peaks over genomic features were summarized in peak-centric view using the function assignChromosomeRegion. ChIP-seq signal tracks were visualized using the Integrative Genomics Viewer45 (IGV, v2.4.14) and heatmaps were generated using SeqPlots46. Histone ChIP-seq datasets with the identifiers ENCFF043LTY (H3K4me1), ENCFF469DBC (H3K4me3), ENCFF289ATH (H3K9me3), ENCFF012GHA (H3K27me3), and ENCFF785WPG (H3K36me3) were downloaded from the ENCODE portal47.Motif analysisHomer48 (v4.10.4; findMotifsGenome.pl) was used to discover motifs from PRDM10 ChIP-seq peaks. Repeat-masked sequence from the mm10 assembly (mm10r) was extracted from peak regions (–size given) defined in the optimal peak set (pooled pseudo-replicates, IDR cutoff 0.05, blacklist filtered) and used to discover up to ten motifs (−S 10) for lengths from 8 to 20 bp with a step of 2 (−len 8,10,12,14,16,18,20), allowing up to four mismatches in the optimization (−mis 4). Homer (annotatePeaks.pl) was then used to scan the ChIP-seq peaks (−size given) as well as TSS ± 1 kb regions (tss mode, −size −1000,1000) for the discovered PRDM10 motif to obtain a set of hits for the conservation analysis. deeptools computeMatrix (v3.3.0) was used to extract phyloP scores 15 bp upstream and downstream of the motif hit sites; phyloP scores are for multiple alignments with 59 vertebrate genomes to the mouse genome and were obtained from the UCSC genome browser (http://hgdownload.cse.ucsc.edu/goldenpath/mm10/phyloP60way/mm10.60way.phyloP60way.bw). The scores were subsequently plotted with the Python package seaborn.mESC RNA-seqPrdm10F/F; ROSA26-CreERT2 mESCs were harvested in TRIzol at indicated time-points post-induction and total RNA purified as described above. Total RNA was quantified on a Nanodrop and RNA quality was assessed on an Agilent 2100 Bioanalyzer using the RNA 6000 Nano kit (Agilent Technologies). RNA integrity values were between RIN9.5–10, confirming high quality total RNA. RNA-Seq libraries were constructed from 1 mg total RNA using the TruSeq Stranded Total RNA Sample Preparation kit (Illumina) following manufacturer’s instructions with a few modifications. The resulting libraries were assessed by High Sensitivity DNA Assay on a Bioanalyzer and quantitative real-time PCR (KAPA Library Quantification Kit for Illumina, Roche). Final RNA-Seq libraries were pooled and sequenced on an Illumina NextSeq 500 instrument using paired-end 2 × 75 bp reads to generate >48 M raw reads per sample.Single embryo RNA-seqNaturally mated timed pregnant females from Prdm10Δ/+ intercrosses were sacrificed at E2.5. 8-cell stage single embryos were collected and individually snap frozen in 4 ul lysis buffer containing dNTP mix, oligo-dT primer and SUPERase-In RNase inhibitor (Invitrogen). RNA-seq library preparation was performed according to the Smart-seq2 protocol with slight modifications49. Embryos were isolated and processed in two separate batches. To pre-screen cDNA samples for downstream library preparation, mutant (Prdm10Δ/Δ) and control (Prdm10+/+, Prdm10Δ/+) embryos were identified by qRT-PCR detection of Prdm10 exon 5. 16 cycles of PCR pre-amplification were carried out for batch 1 and 13 cycles for batch 2. For both batches, 1 ng of cDNA was used for tagmentation with the Nextera XT DNA Library Prep Kit (Illumina) as described49. Final libraries were amplified and pooled for single-end sequencing on an Illumina NextSeq 500 instrument with 75 bp read length to yield ~4.5–10.6 M raw reads per embryo.RNA-seq data analysisRNA-seq data was processed using the ENCODE STAR-RSEM pipeline (https://github.com/ENCODE-DCC/long-rna-seq-pipeline/blob/master/DAC/STAR_RSEM.sh). In brief, sequence reads were mapped to the mouse genome build GRCm38.p6 using STAR50 (v2.6.0c), and gene-level transcript abundances were quantified by RSEM51 (v1.3.1) against the GENCODE mouse vM18 annotation set. DESeq252 (v1.22.2) was used for differential expression analysis; genes were considered to be significantly differentially expressed at Padj < 0.05, and a minimum expression threshold was applied to exclude low-abundance genes (mESCs: baseMean > 100; embryos: baseMean > 10). Volcano plots were generated using the R package EnhancedVolcano (v.1.0.1), expression heatmaps were generated using the R package pheatmap (v1.0.12), and GO enrichment analysis was performed using Metascape53 (http://metascape.org). Aligned reads with splice junctions were visualized on the IGV browser.StatisticsNo statistical methods were used to predetermine sample size. Results were represented as mean ± standard deviation (s.d.) and all experiments have at least three independent biological repeats unless otherwise noted in the figure legends. Differences between groups were examined for statistical significance using unpaired two-tailed Student’s t test (for two groups), one-way ANOVA or two-way ANOVA followed by Tukey’s multiple comparisons test (for more than two groups) in GraphPad Prism 8.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary
Description of Additional Supplementary Files
Supplementary Data 1
Supplementary Data 2
Supplementary Data 3
Supplementary Data 4
Supplementary Data 5
Supplementary Data 6
Supplementary Data 7
Supplementary Data 8 | nature communications | [
"Article"
] | [
"Developmental biology",
"Molecular biology"
] | proteins conserved N-terminal PR (PRDI-BF1 RIZ1) homology domain related to lysine methyltransferase SET domain variable C2H2-type zinc finger repeats sequence-specific DNA binding PRDMs transcriptional regulators cell fate developmental Prdm1 required for primordial germ cell specification branchial arch patterning embryonic hematopoietic lineage Prdm16 promotes brown fat adipogenesis6,7 hematopoietic stem cell role Prdm14 Prdm15 naïve pluripotency embryonic stem.Prdm10 tristanin10 conserved in vertebrates same subfamily as Prdm1511 expressed in embryonic adult tissues12 homozygous deletion of Prdm10 in mice embryonic gene rearrangements PRDM10 in pleomorphic sarcomas15 molecular functional properties PRDM10 unknown role in vivo study conditional Prdm10 knockout mouse model role early embryonic development mouse embryonic stem cells study biochemical molecular properties PRDM10 acts transcription factor target genes regulates expressionregulation Eif3b translation PRDM10 global translation mESC survival essential preimplantation encodes protein N-terminal PR domain ten C2H2 zinc fingers C-terminal glutamine-rich transactivation domain unique 17 PRDM family members PRDM10 generated mice conditional allele (Prdm10F) exon 5 Prdm10 by loxP sites removal exon 5 frameshift nonfunctional truncated protein null allele (Prdm10Δ) Prdm10Δ/+ mice viable fertile no abnormalities no Prdm10Δ/Δ pups recovered from heterozygous intercrosses embryos from Prdm10Δ/+ intercrosses pre- post embryonic lethality embryonic day 3.5 genotypes recovered Mendelian ratios Prdm10Δ/Δ embryos underrepresented at E4.5 no viable Prdm10Δ embryos recovered at E7.5 E12.5 postimplantation stages E3.5 Prdm10Δ/Δ embryos abnormal morula-like appearance control embryos expanding blastocystsvivo Prdm10Δ/Δ embryos developed 2-cell to morula failed form expanded blastocysts embryonic death lethality due embryo-intrinsic defect implantation failure Prdm10Δ/Δ embryos postimplantation loss PRDM10 causes developmental arrest before blastocyst embryonic lethality-implantation essential preimplantation embryogenesis mESC growth embryo genotypes heterozygous intercrosses E3.5 embryos recovered Mendelian distribution no Prdm10Δ/Δ embryos E7.5 images mutant (Prdm10Δ/Δ control+ embryos E3.5 cell mass) labeled asterisk blastocoel red line Scale 50 μm E3.5 embryos morula partially expanded fully expanded/cavitated blastocyst n 27 (Prdm10+/+) 39 (Prdm10Δ/+) 21 (Prdm10Δ qRT-PCR analysis Prdm10 expression OHT-treated mESCs-treated controls post-induction Expression normalized to Ubb n = 3 replicatesWestern blot analysis PRDM10 protein levels Prdm10F/F mESCs exposure EtOH OHT Loading control α-tubulin PRDM10-depleted mESCs severe cell growth defect Cells passaged constant density 2 days counted daily Day 7 n = 4 samples population doublings brightfield images Prdm10F/F Prdm10Δ/Δ mESC colonies Day 5 7 post-induction Cells plated equal densities 2 days Scale bar 500 μm images colony formation assay n = 6. Prdm10F/F/Δ mESCs seeded Day 3 fixed analysis Day 8. 3/7 activity Prdm10Δ/Δ mESCs Prdm10F/F Day 4 to 7 post-induction n 4 replicates Data mean ± s.d data three experiments ***P < 0.001 < 0.0001 two-tailed preimplantation lethality cell apoptosis Prdm10Δ/Δ embryos lineage-specific markers OCT4 CDX2 NANOG levels PRDM10 not required lineage segregation essential embryo survival developmental progressionPrdm10-null mESCs show reduced growth increased PRDM10’s functions early development mESCs in vitro model embryo generated Prdm10F/F; ROSA26-CreERT2 mESCs-mediated deletion by 4-hydroxytamoxifen exposure Recombination efficiency verified at genomic transcript protein level Phenotypic characterization on Prdm10F/F; ROSA26-CreERT2 mESCs in serum-containing medium with leukemia inhibitory factor acute deletion of Prdm10 reduction in cell growth rates 3–4 days post-deletion increasing Prdm10-null mESCs formed smaller colonies expanded poorly colony morphology not altered re-introduction of full-length Prdm10 growth defects in Prdm10Δ/Δ cells Prdm10-null mESCs cultured in serum-free 2i medium with LIF showed similar defect in growth PRDM10 deficiency not rescued by MEK or GSK3 inhibition requirement for PRDM10 in mESCs independent of culture conditionsimpaired cell growth in Prdm10-deficient mESCs evaluated impacts survival proliferation loss PRDM10 no effect cell cycle distribution cycle progression unaffected in Prdm10Δ/Δ mESCs Prdm10Δ/Δ mESCs exhibited higher caspase 3/7 activity controls later post-recombination Day 5 consistent onset cell growth PRDM10 deficiency leads increased apoptotic cell death Prdm10-null phenotype independent proliferation defects attributed decreased cell survival.PRDM10 dispensable for pluripotency differentiationPrevious studies PRDM PRDM14 PRDM15 key regulators pluripotency PRDM10 required for maintenance mESC pluripotency assessed expression pluripotency markers 8 days after Prdm10 deletion transcriptome analysis Prdm10Δ/Δ mESCs days 2 4 post-deletion showed no downregulation genes mESC pluripotency self-renewal transcription factors pluripotency (Pou5f1 Klf4 Sox2 Nanog expressed comparable or higher controlsexamined pluripotency markers (Nanog Pou5f1 Klf2 Klf4 Esrrb qRT-PCR day 6 8 post-deletion confirmed expression maintained Prdm10Δ/Δ mESCs growth survival defects no reduction SSEA-1 expression Prdm10-null mESCs day 4 6 post-deletion Prdm10Δ/Δ mESCs formed colonies smaller controls stained positive for alkaline activity AP-positive colony formation comparable controls day 7 post-deletion transcriptomic analysis Prdm10Δ/Δ mESCs no misregulation germ layer lineage markers loss PRDM10 induce precocious differentiation PRDM10 promotes normal growth mESCs embryos dispensable for maintenance pluripotent state assessed requirement during embryoid body formation LIF withdrawal Prdm10Δ/Δ mESCs formed EBs similar controls upregulated expression germ layer lineage markers retained ability EB differentiation regulators core pluripotency network Pou5f1 Klf4 Nanog expressed comparable to or higher controlsconsistent to ES cells Fig. 4a deterioration of Prdm10Δ/Δ EBs observed day 6 post-Prdm10 deletion Fig similar Prdm10Δ/Δ mESCs under pluripotency phenotype (Fig. 1f). results consistent with role PRDM10 in cell survival suggest PRDM10 dispensable for EB differentiation-wide identification PRDM10 binding PRDM family members transcriptional requirement for PRDM10 in mESCs mediated through sequence-specific transcription factor performed genome-wide profiling PRDM10 binding sites by validated three polyclonal antibodies against PRDM10 experiments with Prdm10F/F; ROSA26-CreERT2 mESCs three datasets identified 528 reproducible peaks (IDR < 0.05) Fig. 6a 1) parallel experiments in mESCs depleted PRDM10 after 4-OHT induction peaks in wild-type cells absent or diminished in-depleted cells reduction peaks in PRDM10-deficient cells absent in WT mESCs validating specificity approachPRDM10 binding sites ChIP enriched promoter regions 70.8% within 1 kb upstream gene transcriptional start sites 9.8% intergenic regions (Fig. 2a associated with active chromatin H3K4me3 H3K4me1 H3K36me3 2PRDM10 sequence-specific zinc finger transcription factor Genomic binding sites = 528) enrichment proximal gene promoters Distribution PRDM10 peaks most within 500 bp TSS Sequence motif enriched PRDM10 binding sites = 10–257) Density histogram motif PRDM10 peak centers Sequence conservation profiles PRDM10 peaks control regions regions within ±1 kb gene TSS Shaded regions 25–75% PRDM10 motif validation luciferase reporter assay HEK293T cells transfected canonical mutated motif sequence Prdm10 expression plasmid control PRDM10 stimulates transcriptional activation canonical motif n = 3 samples Binding labeled probe) WT vs MUT motif PRDM10441–880 protein (125–1000 assessed gel-shift assayOpen arrowheads bound probe solid free probe Competition assay PRDM10 motif PRDM10441−880 binding 10 nM WT probe diminished 20- 320-fold molar excess unlabeled WT probe MUT probe no effect Schematic Prdm10-length mutant expression constructs assays PRDM10FL-880-C activates WT motif reporter 3 samples Western blot FLAG-tagged PRDM10 constructs cell lysates transfected HEK293T cells Arrowheads proteins interest Loading control β-actin Reporter activity luminescence units firefly luminescence Renilla control Representative data three experiments Mean ± s. ***P < 0.001 ****P < 0.0001 not significant two-tailed unpaired test sequence-specific transcription consensus sequence enriched PRDM10 binding sites central enrichment PRDM10 peaks strong sequence conservation PRDM10-bound sites relevant candidate DNA binding PRDM10transcriptional impact recognition motif PRDM10 performed reporter assays in HEK293T cells transfected constructs WT motif or mutated version (MUT) firefly luciferase gene observed activation WT motif reporter with PRDM10 overexpression mutation consensus sequence abolished PRDM10-dependent reporter activation specific cis-regulatory DNA sequence required for PRDM10-mediated transcriptional activity performed gel shift assays PRDM10 binds to DNA motif recombinant GST-fusion PRDM10 protein zinc array used binding detected robust binding PRDM10441–880 to probe motif binding diminished by competition with unlabeled WT probe MUT probe less bound by PRDM10441–880 binding findings demonstrate specific zinc finger-mediated binding PRDM10 to consensus sequence PRDM10 transcriptional activation transcription function N-terminal PR domain C-terminal Q-rich region domains implicated in transcriptional PRDM10 only family member harboring domain biochemical function PR domain varies among PRDM family PRDM10 unknownevaluated PRDM10 deletion mutants (Fig. 2i j luciferase reporter construct consensus motif. 2g mutants lacking central zinc finger DNA-binding domain (ZF-DBD failed elicit activity importance PRDM10 ZF-DBD insufficient transcription additional effector domains (Fig. ZF-DBD N-terminal PR-880 or C-terminal Q-rich domain (441 sufficient transcriptional activation lower levels full-length PRDM10 (Fig. results PRDM10 engages targets zinc finger DNA binding drives transcription N- C-terminal activation domains expression changes in Prdm10-null mESCs phenotypes Prdm10Δ/Δ mESCs embryos transcriptional misregulation loss PRDM10 performed gene expression profiling RNA-seq Prdm10Δ/Δ mESCs Prdm10F/F control mESCs Prdm10Δ/Δ embryos Prdm10Δ/+ intercrosses isolated 8-cell stage embryos at E2.5 before onset phenotypic defects mutant embryos indistinguishable from WT heterozygous littermatesmESCs Day 2 4 post-deletion growth inhibition cell death PRDM10-deficient mESCs controls 2 days post-deletion 8.2% (n = 953) genes up- 8.7% 1019) downregulated (Padj < 0.05) extensive changes Day 4 20.7% (n = 2393) genes up- 18.7% 2157) downregulated (Padj < 0.05) PRDM10-deficient mESCs 7a annotation differentially expressed genes < 0.05) Day 4 post-deletion gene expression signature enriched ribosomal function translation peptide metabolism potential misregulation protein synthesis PRDM10-deficient mESCs RNA-seq data 0.48% (n = 50) genes upregulated 0.79% 83) downregulated (Padj < 0.05) Prdm10Δ/Δ embryos Prdm10+ controls.PRDM10 binds activates target integrated ChIP- RNA-seq data identified 633 unique genes PRDM10 binding sites 52 76 differentialy expressed (Padj < 0.05 fold-change > 2) mESCs 2 4 days post-deletionSupplementary Data majority genes PRDM10 (Padj < 0.05 fold-change > 2) showed decreased expression Prdm10Δ/Δ mESCs (Fig. genes not targets PRDM10 expression changes likely secondary targets embryos loss PRDM10 downregulation 28 expressed targets (Padj < 0.05-change > 2) assay data 2f PRDM10 transcriptional activator target genes. 3PRDM10 binding transcriptional activation target genes Volcano plots RNA-seq data Prdm10Δ/Δ vs Prdm10F/F mESCs post-deletion Padj < 0.05-change > 2 genes changes expression Red genes bound PRDM10 green genes not bound PRDM10 targets predominantly downregulated Volcano plot RNA-seq data Prdm10Δ/ΔPrdm10+ 8-cell embryos 28 genes bound PRDM10 expression changes (Padj < 0.05 fold-change > all downregulated in Prdm10-null embryos Venn diagram overlap bound downregulated genes RNA-seq embryos mESCs 4 18 genes top candidates Heatmap 18 bound downregulated genes Fig. 3c categorized gene knockout phenotypes Asterisks genes translation Eif3b Eef1d Color scale Z-score row-normalized expression values Day 2 Day 4 embryos Pre-implant post-implant lethality undefined timing embryonic lethality unknown Eif3b key target PRDM10-deficient mESCs early embryos targets PRDM10 compared PRDM10-bound genes downregulated (Padj < 0.05 fold-change > 2) RNA-seq datasets identified 18 genes misregulated in mESCs embryos (Fig. 3c seven viable in homozygous null condition seven embryonic lethal when deleted four genes no phenotypechose homozygous genes least likely relevant Prdm10-null embryonic lethal phenotype seven genes embryonic lethality eukaryotic translation initiation factor 3 subunit implicated preimplantation development (Fig. deletion Eif3b results early embryonic lethality by E3.520 time-frame similar Prdm10Δ/Δ embryos Eif3b encodes core component multi-subunit eIF3 promotes mRNA recruitment pre-initiation complex translation role Eif3b translation consistent gene expression signature Prdm10Δ/Δ mESCs disrupted protein synthesis Eif3b positive two genome-wide CRISPR screens for genes mESCs26 likely critical mediator mESC survival PRDM10 Eif3b identified two other PRDM10-regulated targets translation protein synthesis Eef1d Rpl19 (ribosomal protein L19) Eef1d downregulated in Prdm10-null embryos mESCs CRISPR screens mESCs26 support essential requirement Eef1d-mediated knockdown Eef1d mESCs no effect on cell growthRpl19 depletion reduced cell growth essential for mESC viability Rpl19 modestly downregulated<2-fold in PRDM10-deficient embryos Eif3b <2-fold downregulation in vivo prioritized Eif3b for functional validation.Effects Eif3b depletion in embryos Eif3b preimplantation development knockdown experiments in vivo vitro mouse zygotes microinjected with siRNAs blastocyst stage over 4 days siRNA-mediated knockdown of Eif3b developmental arrest blastocyst formation no Eif3b-deficient embryos mature blastocysts morula-like with studies pre-E3.5 lethality for Eif3b-null siRNA-mediated depletion of other PRDM10-regulated genes Selenow Ss18 Ube2a no impact on embryo development 4Loss of EIF3B causes lethality in preimplantation embryos mESCs Eif3b-deficient embryos arrest prior blastocyst formationgenes knocked wild-type zygotes injection siRNA control Eif3b Selenow Ss18 Ube2a blastocyst morphology scored after 4 embryo culture Y-axis percentage cavitated non blastocysts embryos 3 experiments per target gene images embryos treated control Eif3b siRNA 4.5 dpc Scale bar 100 μm Western blot analysis E14 mESCs overexpressing Eif3b transduced shRNA Numbers EIF3B protein levels after subtraction normalization α-tubulin qRT-PCR validation Eif3b knockdown E14 mESCs transformed control Eif3b transduced shRNA targeting 3′UTR Eif3b 3′UTR-specific primers detect endogenous Eif3b transcript n 2 samples one three experiments results E14 mESCs transduced shRNA-479 slower growth shRNA-resistant Eif3b restores normal growth Cells plated equal densities passaged Day 2 Day 4 counted time-points n = 3 replicates experiment Error bars mean ±*P < 0.05 **P < 0.01 ***P < 0.001 < 0.0001 not significant two-tailed unpaired Student’s t test two-way ANOVA Tukey’s comparisons test vitro transfected E14 mESCs siRNAs Eif3b expression observed impaired cell growth Eif3b-depleted cells controls 72 h post-transfection results lentiviral shRNA knockdown approach 3′UTR-specific hairpin sequence (Eif3b-479) Eif3b expression modest reduction Eif3b mRNA protein levels cell growth Eif3b-479 shRNA-transduced cells phenotype not off-target effects rescued shRNA-resistant Eif3b lacking 3′UTR results demonstrate Eif3b essential gene early embryos mESCs.PRDM10 activates transcription data PRDM10 occupancy Eif3b promoter downregulation Eif3b transcript Prdm10Δ/Δ mESCs embryos investigated hypothesis Eif3b target PRDM10 regulation identified two PRDM10-binding regions (P1 P2) Eif3b promoterconsensus motif match Fig P1 P2 stimulated luciferase reporter activity PRDM10 overexpression cis-regulatory PRDM10 Eif3b transcription Prdm10Δ/Δ mESCs reduced expression Eif3b mRNA protein post-deletion Eif3b transcript levels Prdm10Δ restored re-expression exogenous PRDM10 PRDM10 necessary normal Eif3b expression 5PRDM10 regulates Eif3b expression ChIP-seq signal PRDM10 occupancy Eif3b promoter EtOH OHT-treated Prdm10F mESCs 2 days post-induction PRDM10 enrichment Eif3b promoter two subpeaks P1 P2 PRDM10-dependent activation luciferase reporters Eif3b P1 P2 n 2 samples data three experiments qRT-PCR changes Eif3b transcript after Prdm10 deletion n = 2–3 replicatesWestern blot analysis EIF3B protein Prdm10F/F CreERT2 mESCs treatment EtOH 4-OHT qRT-PCR analysis restoration Eif3b transcript levels Prdm10Δ/Δ mESCs exogenous Prdm10 Data Day 4 post-induction n = 3. Gene expression Ubb Error bars mean ± *P < 0.05 < 0.01 < 0.001 < 0.0001 two-tailed unpaired Student’s t test PRDM10 functional importance EIF3B PRDM10 overexpressing exogenous Eif3b Prdm10F/F ROSA26-CreERT2 mESCs 4-OHT treatment Prdm10 deletion restoring Eif3b expression partial phenotypic rescue Prdm10Δ/Δ mESCs reducing doubling time 40% control Prdm10-null cells partial phenotypic rescue EIF3B contributions PRDM10-regulated targets. Rpl19 PRDM10-regulated target role protein synthesis growth defect Prdm10Δ/Δ mESCsresults support EIF3B mESC survival PRDM10 PRDM10-deficient preimplantation embryos 6EIF3B promotes translation PRDM10 Immunoblotting detection EIF3B protein Prdm10F/F mESCs transformed-Eif3b qRT-PCR validation Eif3b transcript Prdm10F/F analyzed 6 days post-induction CDS-specific primers endogenous overexpressed Eif3b transcript n = 3 replicates Growth curve analysis Prdm10F/F Prdm10Δ/Δ mESCs empty Eif3b partial rescue Eif3b overexpression Cells passaged 2 days counted daily Day 8 post-induction n = 3. data experiments population doubling times Prdm10F/F/Δ mESCs expressing exogenous Eif3b experiment = 3. polysome profiles Prdm10F/F Prdm10Δ/Δ mESCs analyzed 5 days post-induction reduced polysome:monosome) ratio impaired global translation Prdm10Δ/Δ n = 5 replicates two experimentspolysome profiles Prdm10F/F/Δ mESCs expressing Eif3b analyzed Day 5 post-induction Eif3b overexpression restores translation Prdm10Δ/Δ mESCs comparable Prdm10F/F controls quantified P/M ratio n = 5 replicates two experiments Gene expression relative Ubb Data mean ± s.d. **P < 0.01 ****P < 0.0001 not significant one-way ANOVA Tukey’s multiple comparisons test two-way ANOVA test eIF3 13 protein subunits essential translation initiation EIF3B conserved nucleation loss EIF3B decreased translation Prdm10-null mESCs polysome profile analysis Prdm10Δ/Δ mESCs reduction translation rates >3-fold decrease polysome-to-monosome (P/M) ratio compared Prdm10F/F controls re-expression exogenous EIF3B global translation defect Prdm10-null cells caused loss EIF3Bnot significant noted trend higher P/M ratios in EIF3B-overexpressing cells vector-transduced Prdm10F/F controls suggestive translation rates above baseline consistent reports overexpression EIF3B subunit eIF3 complex activating protein synthesis in cancer cell data suggest growth defect in Prdm10Δ/Δ mESCs decreased translation efficiency misregulation Eif3b study essential role PRDM10 transcriptional regulator in early mammalian embryogenesis mESC homeostasis findings support model PRDM10 supports cell growth survival early development regulating Eif3b expression imply function PRDM10 specific early embryogenesis regulates processes mESCs PRDM10 expressed across multiple tissues regulates target genes likely pleiotropic effects later stages PRDM10 gene fusions implicated in pathogenesis pleomorphic sarcoma15 possibility roles in human disease data provide starting point for future studies understanding PRDM10 development disease PRDM family appeared in metazoans Prdm10 evolved during family expansion prior vertebrate evolutionPRDM genes emerged later expressed in specialized cells tissue-specific functions examples include Prdm14 (germ cells embryonic stem Prdm7 (melanocytes), Prdm9 (testis fastest evolving Prdm10 expressed across adult first PRDM family member implicated in early embryo development Prdm10 evolved recently eIF3 complex arose before metazoan evolution EIF3B of five eIF3 subunits conserved in eukaryotes PRDM10 regulating ancient gene open question Evolutionary expansion divergence zinc protein families drives diversification DNA specificities functions correlated with increasing organism complexity vertebrate findings challenged eIF3 translation initiation factor selective translational control over specific mRNAs30 speculate PRDM10 may contribute regulatory control over Eif3b expression context-dependent tissue-specific developmental stages protein synthesis basic cellular housekeeping evidence dynamically regulated during development maintenance stem EIF3B upregulated at morula blastocyst stages key role for protein synthesis embryonic growth studies importance of global translational control in ESC pluripotencyadvances transcriptional epigenetic mechanisms ESC less regulation protein translation regulated initiation stage eukaryotic initiation factors characterized biochemical structural few studies examined roles stem cells development deciphering mechanisms PRDM10 mESCs transcriptional control translation development conditional knockout mice generated C57BL/6 background ES cell clone (Prdm10tm1a(EUCOMM knockout-first cassette exon 5 Prdm10 transmission Prdm10lacZ/+ mice crossed FLPe recombinase transgenic line (C57BL/6-Tg remove neomycin selection cassette generate Prdm10 flox allele Prdm10F/+ mice bred ACTB-Cre recombinase transgenic line generate Prdm10Δ null allele 4-OHT-inducible knockouts crossing Prdm10F/F mice ROSA26-CreERT2 transgenic strain housed pathogen-free conditions 12 h light-dark cycle food water libitum procedures Institutional Animal Care Use Committee protocols #151042 #181393 approval Biological Resource Centremice DNA extracts biopsies overnight lysis 55 °C DirectPCR Lysis Reagent 0.2 mg/ml proteinase K lysed 1 h 55 °C 10 μl lysis buffer (50 mM KCl Tris-HCl 2 MgCl2 0.45% NP-40 0.45% Tween-20 0.5 mg/ml proteinase heat-inactivation 95 °C 10 min 2 μl extract 20 μl PCR reaction DreamTaq PCR mix Primers Supplementary Data 8.Plasmid overexpression Prdm10 cloned gene synthesis RNA stable expression Prdm10 ligated NheI BamHI sites pJ549 PiggyBac transposase expression vector N-terminal FLAG tag transient transfections Prdm10 cloned pcDNA3.1 N-terminal FLAG shRNA lentiviral vectors gene-specific hairpin sequences RNAi Consortium cloned pLKO.1-Neo annealed oligonucleotides Endotoxin-free plasmids Nucleobond Xtra Midi EF Kit constructs verified Sanger sequencingEmbryo matings successful copulation 12 midnight vaginal plug Pregnant females sacrificed cervical dislocation 36 h post-copulation 2-cell embryos 84–90 h post blastocysts Preimplantation embryos collected infundibulum uterine horns M2 medium Postimplantation embryos dissected uteri in vitro studies embryos isolated 2-cell stage cultured 3 days KSOM + AA medium 37 °C 5% CO2 imaged time-points Data heterozygous intercrosses Prdm10lacZ/+ Prdm10Δ/+ mice outcomes.Immunofluorescence microscopyEmbryos fixed 4% paraformaldehyde 30 min permeabilized 0.1% Triton X-100 30 min 1% incubated 4 °C overnight antibodies OCT4 CDX2 NANOG washed incubated Alexa Fluor-conjugated secondary antibodies room temperature 1 h Images acquired Olympus Fluorview 1000 Zeiss LSM800 confocal laser-scanning microscopemouse cell linesBlastocysts isolated E3.5 cultured 2i/LIF medium embryonic fibroblasts 5 days outgrowths disaggregated re-plated MEF feeders expanded 4–5 passages 1:1 DMEM/F12 Neurobasal GlutaMAX 0.05% bovine serum albumin 100 U/ml penicillin/streptomycin B27 10 μM 2-mercaptoethanol 1000 U/ml LIF 3 mM CHIR99021 1 PD0325901 cells plated 0.1% gelatin 1 h-depleted ESCs collected DNA purified DNeasy Blood Tissue kit culture 37 °C humidified incubator 5% CO2. cultured 0.1% gelatin plates Dulbecco’s Modified Eagle’s Medium supplemented 15% fetal bovine serum GlutaMAX amino acids 1 mM sodium pyruvate 100 U/ml penicillin/streptomycin 5.5 μM 2-mercaptoethanol 1000 U/ml mLIF cells resuspended 10% FBS DMEM re-seeded 1:20–1:40 2–3 daysPlasmid siRNA transfections mESCs HEK293T cells Lipofectamine 2000 Fisher cell lines negative mycoplasma contamination PCR Prdm10 ROSA26-CreERT2 mESCs treated 24 h 50 nM 4-OHT ethanol Cells washed dPBS seeded assays stable transgene ESCs transfected pJ549 PiggyBac plasmids Lipofectamine 2000. Transfected E14 mESCs selected 0.75 μg/ml puromycin 3 days stable transgenesis GFP-positive cells purified fluorescence-activated cell sorting MoFlo XDP cell expanded culture assays.Apoptosis mESCs seeded 96-well plate 2 × 104 cells per well attach overnight activity Caspase-Glo 3/7 Assay reagent) added well cells homogenized agitationSamples transferred 96-well plate incubated 30 min luminescence readings GloMax Caspase-Glo signals normalized cell numbers measured CellTiter-Glo Cell Viability Assay growth seeded 105 cells 12-well plate cells harvested trypsinization counts measured trypan exclusion II Cell Counter Brightfield images Nikon Eclipse TS100 microscope 4× objective formation seeded 6-well plates 1 × 104/well cultured 5 days Cells washed fixed ice-cold methanol 10 min stained 0.5% crystal violet 25% methanol colonies imaged scanner quantified Fiji ImageJ lentiviral production 1 μg pLKO-Neo plasmid-transfected 0.5 μg pMD2.G 0.375 μg psPAX2 HEK293T cells Lipofectamine 2000. supernatants collected 48 72 h-transfection filtered 0.45 μm syringe filter centrifuged 24,000 r 2 h 4 °C pellet reconstituted Hank’s Balanced Salt Solution stored −80 °CE14 mESCs infected lentivirus 8 μg/ml polybrene cultured 400 μg/ml G418 transductants 96 h cells re-seeded 200 μg/ml G418 knockdown transfected 20 pmol ON-TARGETplus SMARTpool siRNA 1 μl Lipofectamine 2000 seeded 24-well plates viability assessed 48 72 h post extraction qRT lysed TRIzol RNA purified PureLink RNA Mini Kit-column treatment Reverse transcription Maxima First Strand cDNA Synthesis Kit ~1 μg RNA reaction diluted 10-fold nuclease-free water real-time PCR CFX96 Touch PowerUp SYBR Green Master Mix Target gene expression calculated Primer sets validated-curve tested linear amplification four Primers qRT-PCR Supplementary Data blot cell lysates buffer (32.9 mM Tris-HCl 12.5% glycerol 1% SDS 2.5% 2-mercaptoethanol 27 mg/ml DTT 0.005% bromophenol blue heated 98 °C 10 minProtein concentrations measured RC DC Protein Assay 20–40 μg protein loaded well samples separated SDS-PAGE gel electrophoresis Tris-Glycine-SDS buffer Proteins transferred Immun-Blot PVDF membranes wet electroblotting Tris-glycine buffer 10% methanol Membranes blocked 1 h TBS-T 0.05% Tween-20) 5% milk 3% BSA Blots incubated overnight 4 °C primary antibodies diluted 3% BSA/TBS-T:PRDM10 EIF3B FLAG M2 alpha-tubulin beta-actin incubated HRP-conjugated anti-mouse anti-rabbit secondary antibody 1 h room temperature SuperSignal West Femto Maximum Sensitivity Substrate Pico Chemiluminescent Substrate chemiluminescent detection Blots imaged ChemiDoc Touch analyzed Image Lab software reporter PRDM10 motif oligonucleotides annealed ligated NheI BglII sites pGL4.23 Eif3b promoter sequences amplified PCR mouse DNA cloned into pGL4.23pGL4.74 pcDNA3-Prdm10 HEK293T cells 12-well plates 48 h post firefly Renilla luciferase activities measured GloMax luminometer Dual probes 5′-biotinylated oligonucleotides reactions recombinant protein 10 nM probe temperature 20 min 6 mM HEPES-KOH pH 7.9 6% glycerol 100 μg/mL BSA μM ZnCl2 20 μg/mL poly Samples electrophoresed 6% polyacrylamide gels 0.25× TBE buffer 150 V transferred Hybond N + nylon membranes 0.5× TBE 380 mA 1 h Biotinylated DNA detected LightShift Chemiluminescent EMSA kit Blots imaged ChemiDoc Touch analyzed Image Lab seeded six-well plates 40–60% confluent Cells pulsed 10 μM 5-ethynyl-2′-deoxyuridine 30 min washed PBS dissociated trypsin-EDTA Fixation permeabilization EdU labeling 488 Flow Cytometry Assay KitCells incubated 10 μg/ml DAPI ice >1 h filtered 40 μm strainer acquisition SSEA-1 staining FACS buffer FBS anti-SSEA-1 Fluor 647 30 min incubation flow cytometry data BD LSR II cytometer analyzed FlowJo v10 Day 3 mESCs seeded 1 × 106 2.5 × 106 per 10-cm dish Day 5. treated 100 μg/ml cycloheximide 10 min 37 °C washed ice-cold PBS harvested ice 800 μl lysis buffer (10 Tris-HCl 5 mM MgCl2 100 mM KCl 1% Triton X-100 2 mM DTT 100 U/ml RNasin protease inhibitor 100 μg/ml cycloheximide Lysates sheared cleared 1300 × g 10 min 4 °C lysates layered 10–50% sucrose gradients centrifuged SW-41Ti rotor 36,000 2 h Gradients fractionated BioComp Gradient Station fractionator absorbance 254 nm monitored polysome profile ratios derivedmESCs harvested trypsinization 48 h resuspended 106 cells/ml 10% FBS DMEM cross-linked 1% formaldehyde 15 min quenched 125 mM glycine washed PBS Chromatin extracts lysis LB1 140 NaCl 1 EDTA 10% glycerol 0.5% Nonidet-P40 0.25% Triton 200 NaCl EDTA 0.5 EGTA LB3 100 NaCl EDTA 0.1% sodium deoxycholate 0.5% N supplemented 0.2 mM PMSF protease inhibitor Chromatin DNA sheared 100–500 bp 5–6 cycles sonication Sonifier Triton X-100 1% lysates cleared centrifugation 5 μg antibody 100 μg sonicated chromatin incubated overnight 4 °C 40 μl Protein A Dynabeads incubated 4 °C 4 hBeads collected washed low salt buffer (10 Tris-HCl 150 NaCl 1 mM EDTA 1% Triton high salt (20-HCl 500 mM NaCl 2 mM EDTA 1% Triton X-100 0.1% LiCl buffer (10-HCl 250 LiCl 1 mM EDTA 0.5% sodium deoxycholate 0.5% Nonidet-P40) buffer 50 mM NaCl incubated elution buffer (50-HCl 10 EDTA 1% SDS 65 °C 20 min Eluted protein/DNA reverse-crosslinked overnight 65 °C treated RNase A proteinase K DNA column-purified QIAquick PCR Purification Kit DNA concentrations measured Qubit Assay Libraries prepared 4 ng ChIP DNA DNA amplified 9 PCR cycles elution volumes reduced quantified High Sensitivity DNA Assay real PCR ChIP-Seq libraries pooled sequenced Illumina NextSeq 500 75 bp reads ~20 M raw reads per library processed ENCODE Transcription Factor ChIP-Seq pipelinefastq files aligned with bwa (v0.7.13) filtered duplicates Picard multi-mapping reads low-quality alignments samtools v1.2). SPP (v1.13) peaks filtered alignments pooled alignments two pseudo-replicates Peaks in ENCODE mm10 blacklist filtered out filtered peak sets assessed for reproducibility IDR43 (v2.0.4.2). final “optimal” pipeline obtained from pooled pseudo-replicates IDR cut-off 0.05. ChIP-seq peak annotation analyses R package ChIPpeakAnno44 (v3.16.1). peaks annotated with genes within 5 kb upstream to 1 kb downstream gene body Distribution peaks genomic features summarized peak-centric view assignChromosomeRegion ChIP-seq signal tracks visualized Integrative Genomics Viewer45.14) heatmaps generated SeqPlots46 ChIP-seq datasets ENCFF043LTY ENCFF469DBC downloaded from ENCODE portal47.Motif analysisHomer48 (v4.10.4 findMotifsGenome) discover motifs from PRDM10 ChIP-seq peaksRepeat-masked sequence mm10 assembly extracted from peak regions optimal set pseudo-replicates IDR cutoff 0.05 blacklist filtered ten motifs lengths 8 to 20 bp step 2 four mismatches optimization Homer ChIP-seq peaks TSS ± 1 kb regions for PRDM10 motif hits conservation analysis deeptools computeMatrix.0) phyloP scores 15 bp upstream downstream motif hit sites multiple alignments 59 vertebrate genomes mouse genome obtained UCSC genome browser scores plotted with Python package seaborn.mESC RNA-seqPrdm10F ROSA26-CreERT2 mESCs harvested in TRIzol post-induction total RNA purified RNA quantified Nanodrop RNA quality assessed Agilent 2100 Bioanalyzer RNA 6000 Nano kit RNA integrity values between RIN9.5–10 high quality total RNA RNA-Seq libraries constructed from 1 mg total RNA TruSeq Stranded Total RNA Sample Preparation kitlibraries assessed High Sensitivity DNA Assay real-time PCR RNA-Seq libraries pooled sequenced Illumina NextSeq 500 2 75 bp reads >48 M reads per sample embryo RNA females Prdm10Δ/+ intercrosses sacrificed E2.5 8-cell stage embryos collected frozen 4 lysis buffer dNTP oligo-dT primer RNase inhibitor-seq library preparation Smart-seq2 protocol Embryos isolated processed two batches mutant/Δ control embryos identified qRT-PCR detection Prdm10 exon 5. 16 cycles PCR pre-amplification batch 1 13 batch 2. 1 ng cDNA tagmentation Nextera XT DNA Library Prep Kit libraries amplified pooled single-end sequencing Illumina NextSeq 500 75 bp read length ~4.5–10.6 M reads per embryo-seq data processed ENCODE STAR-RSEM pipeline sequence reads mapped mouse genome GRCm38.p6 STAR50 gene-level transcript abundances quantified RSEM51 (v1.3GENCODE mouse vM18 annotation set DESeq252 (v1.22.2) for differential expression analysis genes expressed at Padj < 0.05 minimum expression threshold exclude low-abundance genes (mESCs baseMean > 100 embryos baseMean > 10). Volcano plots R EnhancedVolcano (v.1.0.1), expression heatmaps R pheatmap (v1.0.12) GO enrichment analysis Metascape53 Aligned reads splice junctions visualized IGV browser.StatisticsNo methods sample size Results represented mean ± standard deviation experiments three independent biological repeats Differences between groups examined unpaired two-tailed Student’s t test one-way ANOVA Tukey’s multiple comparisons test GraphPad Prism 8.Reporting Nature Research Reporting Summary.Supplementary information Peer Review File Additional Supplementary Files Supplementary Data | 49.4 | 1.147426 |
10.1038/s41467-020-15598-x | PMC7160132 | XPO5 mediates nuclear export of miRNA hairpin precursors (pre-miRNAs) through a RanGTP-dependent binding. Here the authors employ HITS-CLIP and biochemical analysis and show that XPO5 binds and promotes nuclear processing of clustered pri-miRNAs, with extensive double-stranded regions, independently of RanGTP. | XPO5 mediates nuclear export of miRNA precursors in a RanGTP-dependent manner. However, XPO5-associated RNA species have not been determined globally and it is unclear whether XPO5 has any additional functions other than nuclear export. Here we show XPO5 pervasively binds to double-stranded RNA regions found in some clustered primary miRNA precursors and many cellular RNAs. Surprisingly, the binding of XPO5 to pri-miRNAs such as mir-17~92 and mir-15b~16-2 and highly structured RNAs such as vault RNAs is RanGTP-independent. Importantly, XPO5 enhances the processing efficiency of pri-mir-19a and mir-15b~16-2 by the DROSHA/DGCR8 microprocessor. Genetic deletion of XPO5 compromises the biogenesis of most miRNAs and leads to severe defects during mouse embryonic development and skin morphogenesis. This study reveals an unexpected function of XPO5 for recognizing and facilitating the nuclear cleavage of clustered pri-miRNAs, identifies numerous cellular RNAs bound by XPO5, and demonstrates physiological functions of XPO5 in mouse development. | IntroductionXPO5 is a karyopherin protein1 that mediates nuclear export of miRNA hairpin precursors (pre-miRNAs)2–4. XPO5 binds to hairpin RNAs with 2–8 nt 3′ overhang including pre-miRNA and minihelix viral RNA in a RanGTP-dependent manner2–5. The formation of XPO5:pre-miRNA:RanGTP nuclear export complex was confirmed by X-ray crystallography6, in which RanGTP binding triggers a conformational change of XPO5 and several XPO5 residues in the HEAT repeats form hydrogen bonds with 5′ and 3′ ends including the 2-nt 3′ overhang of pre-miRNAs. Functional studies showed a requirement of XPO5 to transport pre-miRNAs from the nucleus to the cytoplasm2–4, providing the evidence that XPO5 is an essential component of miRNA biogenesis7. However, XPO5-independent pre-miRNA export pathway was also reported8,9. Notably, 7-methylguanosine-capped pre-miRNAs whose biogenesis is independent of the DROSHA/DGCR8 microprocessor are exported via the PHAX-XPO1 pathway9. In a recent study, deletion of XPO5 in human HCT116 cells only mildly affected the biogenesis of most miRNAs, suggesting that XPO5 is sufficient but not required for miRNA biogenesis10. However, in a CRISPR-Cas9 screen for genes that are critical for mir-19-mediated silencing also in human HCT116 cells, XPO5 was identified as essential11. In addition to these different reports, the lack of genetically engineered mouse models of XPO5, unlike other core components of the miRNA pathway such as Drosha, Dgcr8, Dicer1, or Ago genes, has hindered the understanding of XPO5 functions. Overall, the in vivo requirements of XPO5 for global miRNA biogenesis, mammalian development, and tissue formation remain to be determined.XPO5 is a highly expressed protein among core components of miRNA biogenesis in both mouse and human cells12. However, it is more sensitive to saturation caused by the expression of short hairpin RNA and highly structured viral RNA than other components such as Drosha and Dicer113,14. Although individual RNA species such as pre-miRNA2–4, some cellular tRNAs15 and minihelix viral RNAs such as adenovirus VA1 RNA5,14 have been identified as XPO5 cargoes for nuclear export, XPO5-associated cellular RNAs have not been determined at the genomic scale. It is unclear how many pre-miRNAs directly interact with XPO5 and require XPO5 for their biogenesis. Furthermore, since the discovery of XPO5 as the nuclear export factor for pre-miRNA hairpin2–4,6, much attention to XPO5 has been drawn to its nuclear export functions. However, XPO5 was originally identified as a binding protein of double-stranded RNA (dsRNA) binding proteins such as ILF3, PKR, and Staufen16. Although the direct binding of RNA species including pre-miRNAs, tRNA and VA1 RNA to XPO5 has since been well documented2–6,14,15, it remains unknown whether other cellular RNAs with double-stranded regions can bind to XPO5. In addition, all known RNA substrates of XPO5 bind to XPO5 in a RanGTP-dependent manner, which is a key feature of karyopherin-mediated nuclear export1. As a result, whether XPO5 plays any roles in cellular RNA metabolism other than nuclear export is an open question.To address these questions, we first perform HITS-CLIP analysis of XPO5-associated RNAs in human embryonic kidney cells 293T (HEK293T). We find that a large number of cellular RNAs including most pre-miRNAs and numerous noncoding, structural RNAs are bound by XPO5. Notably, some closely clustered primary miRNA precursors such as pri-mir-17~92 and pri-mir15b~16-2 show strong XPO5 HITS-CLIP signals outside of pre-miRNA hairpins. Surprisingly, purified XPO5 directly bind to pri-mir-17~92 and pri-mir15b~16-2 in a RanGTP-independent manner. In vitro processing assays reveal that XPO5 enhances the cleavage efficiency of pri-mir-19a and pri-mir-15b~16-2 by the DROSHA/DGCR8 microprocessor.To test the function of XPO5 in mouse development, we show that constitutive deletion of XPO5 leads to early embryonic lethality and failed gastrulation at approximately embryonic day 7.5. For tissue morphogenesis, we find that conditional knockout (cKO) of XPO5 in the epithelial cells of the skin shows similar but slightly different defects than those observed in Dicer1 or Dgcr8 cKO17–19. Quantitative measurement of miRNA biogenesis in the skin shows ~90% global loss for most canonical miRNAs except for a few Drosha/Dgcr8-independent miRNAs such as mir-320 and mir-48419. Taken together, this work provides insights into the function of XPO5 in miRNA biogenesis that is independent of RanGTP binding and nuclear export, and identifies numerous cellular RNAs as XPO5 binding partners. These findings establish a molecular basis to further investigate the regulatory roles of XPO5 in miRNA biogenesis and the metabolism of other cellular RNAs.ResultsXPO5 associates with primary and pre-miRNA precursorsUV crosslinking followed by immunoprecipitation and sequencing has been successfully used to identify highly structured miRNA precursors that are bound by DROSHA, DGCR8, and DICER120–22. To globally identify XPO5-associated RNA species in human cells, we applied a high-throughput sequencing of RNA isolated by crosslinking immunoprecipitation (HITS-CLIP) approach to map RNA fragments bound by endogenous XPO5 in HEK293T cells (Fig. 1a). In the absence of XPO5 antibody, we did not recover any RNA fragments upon immunoprecipitation (Fig. 1a). Five independent HITS-CLIP libraries were generated and they displayed generally consistent profiles of XPO5-associated RNAs. To increase stringency, we required the detection of any RNA fragments in at least three out of the five libraries for further analysis. In total, 64.2 million reads were generated and mapped to 83,169 genomic peaks (see “Methods”). Because XPO5 is known to interact with pre-miRNA hairpin and mediates their nuclear export2–4, we first determined whether pre-miRNA hairpins were recovered in our datasets. Indeed, all 37 pre-miRNAs of the most highly expressed miRNAs, judging from miRNA reads number (>1000 reads) from a quantitative smRNA-seq in the HEK293T cells (Supplementary Data 1), were detected in XPO5 HITS-CLIP (Fig. 1b). Overall, 96 out of the top 107 miRNAs, whose read number was >200 in the smRNA-seq, were also detected in our XPO5 HITS-CLIP. Furthermore, XPO5-associated miRNA sequences generally spanned the left arm, the top loop, and the right arm regions, reflecting the association of XPO5 with pre-miRNA hairpins prior to DICER1 cleavage. Only ~20% of XPO5 HITS-CLIP reads were mapped to either arm (Fig. 1c), which generally represents mature miRNAs. In comparison, >98% of mature miRNA reads recovered from smRNA-seq were mapped to the left or the right arm as expected (Fig. 1c).Fig. 1HITS-CLIP reveals association of XPO5 with primary and pre-miRNA precursors.a Autoradiogram of 32P-labelled XPO5-RNA complexes treated with different concentration of RNase is resolved on a PAGE gel. Original data are provided in the Source Data file. b Pre-miRNAs of most abundantly expressed miRNAs in HEK293T cells are detected in XPO5 HITS-CLIP. c The location of small RNA-seq reads and XPO5 HITS-CLIP reads relative to miRNA hairpin is shown in pie charts, respectively. Light blue indicates reads covering both the left arm and the loop region; blue indicates reads covering both the right arm and the loop region; dark blue indicates reads covering only the left arm; light gray indicates reads covering only the right arm; dark gray indicates reads covering the entire hairpin. d–h Metagene analysis of miRNA sequences associated by DROSHA, DGCR8, DICER1, AGO2/3, and XPO5 reveals the recognized regions by each core component and identifies pri-miRNA sequences associated by DROSHA, DGCR8, and XPO5 that are beyond pre-miRNA hairpin (red bracket). i, k IGV tracks of XPO5-CLIP, DROSHA-CLIP, DGCR8-CLIP, DICER1-PARCLIP, and AGO2/3-Seq reveal the specific association of XPO5 with pre-mir-31, pre-mir-3618, and pre-mir-1306. Negative reads number in IGV track indicates the mapping to the minus strand. j IGV tracks of XPO5-CLIP, DROSHA-CLIP, DGCR8-CLIP, DICER1-PARCLIP, and AGO2/3-Seq reveal the widespread association of XPO5 with pri-mir-17~92. Blue bars at the bottom and black lines together indicate the location of pre-miRNAs. Data range is shown on the left of each track.To compare the profile of XPO5-associated miRNA sequences with other essential components of the miRNA biogenesis pathway including DROSHA, DGCR8, DICER1, and AGO proteins, we downloaded previous published HITS-CLIP (DROSHA and DGCR8)20,21, PAR-CLIP (DICER1) and AGO2/3-Seq (mature miRNAs)22 datasets and examined the profile of miRNA sequences that are associated with each component (Fig. 1d–h). To facilitate the comparison of all miRNA profiles captured by each CLIP dataset, we aligned all detected miRNA sequences starting from 5′ end of 5p miRNAs as annotated in miRBase release 2123. Overall, DROSHA and DGCR8 HITS-CLIP showed similar profiles, which usually span not only the pre-miRNA hairpin regions but also the flanking regions. In agreement with the notion that DROSHA recognizes the lower stem from the basal junction that is outside of the pre-miRNA hairpin and DGCR8 recognizes the upper stem of the miRNA hairpin24, DROSHA-associated miRNA sequences showed more prominent signals in the sequences immediately upstream of pre-miRNA hairpin than DGCR8-associated miRNA sequences (Fig. 1d, e). DICER1 PAR-CLIP showed the recognition of pre-miRNA hairpin but not the lower stem regions outside of pre-miRNA hairpin (Fig. 1f), consistent with that DICER1 recognizes pre-miRNA hairpin after its release by DROSHA/DGCR8. AGO2/3-Seq showed the profile of mature miRNAs (Fig. 1g), which are derived from either the left or the right arm of miRNA hairpin. However, XPO5 HITS-CLIP showed not only the association of XPO5 with pre-miRNA hairpin as expected but also the flanking regions of the hairpin (Fig. 1h). Comparing the overall profile of XPO5-associated miRNA sequences to those of DICER1 and DGCR8, it was evident that XPO5-associated miRNA regions were similar to DICER1 within the annotated pre-miRNA hairpins but also resembled the profile of DGCR8-associated regions outside of the annotated pre-miRNAs (Fig. 1e, f, h).To understand how XPO5 associates with miRNA precursors outside of pre-miRNA hairpin, we next examined XPO5 HITS-CLIP data of individual miRNAs. To faithfully present the data, we used one IGV track to show all reads mapped from XPO5 HITS-CLIP data and another track to show all unique reads after collapsing potential duplicates (Fig. 1i–k and Supplementary Fig. 1a, b). While many miRNAs are encoded by a primary transcript individually, ~50% of miRNAs are also encoded by polycistronic transcription unit25,26. For all singly encoded miRNAs that we have examined such as mir-31, mir-21, and mir-34a, XPO5 HITS-CLIP signals were faithfully restricted within pre-miRNA hairpins as expected (Fig. 1i and Supplementary Fig. 1a, b). However, for many polycistronic miRNAs especially the ones that are closely clustered, XPO5 HITS-CLIP reads also covered primary miRNA transcripts. Among all XPO5-associated miRNAs, the mir-17~92 cluster harbored the most reads. Remarkably, XPO5 HITS-CLIP reads broadly covered the pri-miRNA, distinct from the profile of pre-miRNA precursors as detected in DICER1 PAR-CLIP (Fig. 1j). Inter-pre-miRNA regions between mir-17 and mir-19b1 all had abundant XPO5-associated RNA reads compared with those derived from pre-miRNA hairpins (Fig. 1j). Upon closer inspection, we noted two possible modes of XPO5 binding to pri-mir-17~92, judging by the distinct reads density covering the precursor regions. First, XPO5 bound to pre-miRNA hairpins of each of these six miRNAs. Second, XPO5 bound to inter-pre-miRNA regions with an even higher density than the pre-miRNA regions. For example, both mir-19a and mir-20a are abundantly expressed miRNAs. But XPO5-associated RNA reads were far more abundant in the inter-pre-miRNA regions than the pre-miRNA regions. Furthermore, many XPO5-associated RNA reads span the predicted DROSHA cleavage sites that mark the 5′ and 3′ ends of pre-miRNAs (Supplementary Fig. 1c), indicating the association of XPO5 with uncleaved pri-miRNAs. The only inter-pre-miRNA region that is depleted of XPO5-associated RNA reads was between mir-19b1 and mir-92a1. Notably, mir-17~19b1 pri-miRNA is released from the mir-17~92 pri-miRNA by an endonuclease CPSF3 and the spliceosome-associated ISY127. These data suggest a possibility that XPO5 recognizes pri-mir-17~92 cluster after the CPSF3/ISY1-mediated cleavage but prior to DROSHA/DGCR8-mediated processing.To confirm that the association of XPO5 to pri-miRNAs is specific and not simply due to the abundance of pri-mir17~92, we inspected other closely clustered pri-miRNAs such as both mir-15a~16-1 and mir-15b~16-2 clusters that are expressed at a much lower level than pri-mir-17~92. We still observed prominent inter-pre-miRNA reads in XPO5 HITS-CLIP data and the distinct reads density in the inter-pre-miRNA and pre-miRNA regions (Supplementary Fig. 1d). Interestingly, the widespread XPO5 reads coverage was not found in other miRNA clusters, in which pre-miRNA hairpins are located far away from each other such as the mir-200b cluster (>600 nt between hairpins) (Supplementary Fig. 1e). These data suggest that not all primary transcripts of polycistronic miRNAs bind to XPO5. In addition, two miRNAs, mir-3618 and mir-1306, are processed from the 5′ region of Dgcr8 mRNA28. However, the expression of mir-1306 is much higher than mir-3618 based on both the processing assay28 and the sequencing of AGO2/3 associated mature miRNAs22. Consistent with the pattern of mature miRNA expression, XPO5 HITS-CLIP reads on mir-1306 pre-miRNA were much more abundant than those on mir-3618 pre-miRNA hairpin (Fig. 1k), lending further support to the specificity of our XPO5 HITS-CLIP. Together, these data suggest two modes of XPO5 association with miRNA precursors: (1) for most miRNAs, XPO5 associates with pre-miRNA hairpins as expected; and (2) for some closely clustered polycistronic miRNAs, XPO5 associates with the primary miRNA transcripts with a high affinity in inter-pre-miRNA regions.XPO5 binds to pri-miRNA precursors independently of RanGTPTo study the binding of XPO5 to closely clustered primary miRNAs, we performed electrophoretic mobility shift assays in vitro. We purified human XPO5 and RanQ69L, a Ran mutant that is deficient in GTP hydrolysis and locks Ran in the RanGTP form29, from bacteria E.coli as described previously4. We first used in vitro transcribed pre-mir-30a hairpin RNA to confirm the binding activity of recombinant XPO5 and RanQ69L proteins to pre-miRNA. Indeed, XPO5 binds to pre-mir-30a hairpin in a RanGTP-dependent manner (Fig. 2a). The single-shifted band also confirmed the formation of XPO5:RanGTP:pre-mir-30a complex. Because pri-mir-17~92 is known to fold into a highly structured RNA conformation that has extensive base-pairing regions beyond pre-miRNA hairpins27 (Supplementary Fig. 2a), we in vitro transcribed and folded the pri-mir-17~92 precursor (Supplementary Fig. 2b) and tested its binding to XPO5. Surprisingly, XPO5 bound to pri-mir-17~92 in a RanGTP-independent manner (Fig. 2b). Inclusion of RanQ69L together with XPO5 and pri-mir-17~92 showed no interference to the binding of XPO5 and the pri-miRNA (Fig. 2c). Neither the size nor the intensity of the super-shifted bands containing XPO5 and pri-mir-17~92 changed in the presence or absence of RanQ69L (Fig. 2c). This result suggests that the complex does not contain RanQ69L. In addition, we observed continuously super-shifted bands with the increasing doses of XPO5, up to 80-fold of XPO5 in excess of pri-mir-17~92 RNA (Fig. 2b, c). These data suggest that multiple XPO5 molecules bind to pri-mir-17~92 precursor in a RanGTP-independent manner in vitro, corroborating with the extensive XPO5 HITS-CLIP signals detected on pri-mir-17~92 in vivo.Fig. 2XPO5 binds to primary miRNA precursors in a RanGTP-independent manner.a XPO5 binds to pre-mir-30a in a RanGTP-dependent manner. Black hairpin represents pre-mir-30a, XPO5 is coloured in orange, RanGTP is coloured in purple. b XPO5 binds to pri-mir-17~92 without RanGTP. Increasing amount of XPO5 results in super shifts. c XPO5 binding to pri-mir-17~92 is not affected by RanGTP. d XPO5 binds to pre-mir-19a in a RanGTP-dependent manner. Black hairpin represents pre-mir-19a, XPO5 is coloured in orange, RanGTP is coloured in purple. e XPO5 binds to pri-mir-19a in a RanGTP-independent manner. Increasing amount of XPO5 results in a super shift. Black hairpin represents pri-mir-19a, XPO5 is coloured in orange. f The dissociation constant (Kd) between XPO5 and pri-miR-19a is calculated based on the binding results in e. g XPO5 binds to truncated pri-mir-19a v1 (pri-mir-19a truncation v1) in a RanGTP-independent manner. Black hairpin represents pri-mir-19a truncation v1, XPO5 is coloured in orange. h The dissociation constant (Kd) between XPO5 and pri-miR-19a truncation v1 is calculated based on the binding results in g. i XPO5 does not bind to truncated pri-mir-19a v2 (pri-mir-19a truncation v2). For all binding assays, 1 nM substrates are used. Original data for a–e, g, and i are provided in the Source Data file.To further characterize how XPO5 binds to primary miRNA transcript, we focused on pri-mir-19a, whose flanking sequences are predicted to form extensive base-pairing outside of the 11 bp lower stem region that is characteristic of DROSHA/DGCR8 primary miRNA substrates7 (Supplementary Fig. 2c). The primary miRNA sequences flanking mir-19a also had very high XPO5 HITS-CLIP signals (Fig. 1j). We first confirmed that pre-mir-19a hairpin only bound to XPO5 in a RanGTP-dependent manner (Fig. 2d). In contrast, pri-mir-19a bound to XPO5 in a RanGTP-independent manner, similar to the pri-mir-17~92 results. Importantly, two distinct pri-mir-19a:XPO5 complexes, judging by the size of super-shifted pri-mir-19a, were detected with the increasing amount of XPO5 (Fig. 2e). Finally, the dissociation constant (Kd) of pri-mir-19a and XPO5 association was calculated at 14.56 nM (Fig. 2f).Because pre-mir-19a hairpin could not bind to XPO5 in the absence of RanGTP, these results suggested that XPO5 binds to the lower stem regions of pri-mir-19a. To test this hypothesis, we generated two truncated pri-mir-19a mutants. We deleted one basal stem region in pri-mir-19a truncation v1 and both basal stem regions in pri-mir-19a truncation v2 (Supplementary Fig. 2c). Indeed, pri-mir-19a truncation v1 bound to XPO5 in a RanGTP-independent manner with the Kd of 7.99 nM but only a singly shifted band was detected (Fig. 2g, h). In contrast, pri-mir-19a truncation v2 no longer bound to XPO5 (Fig. 2i) when both basal stem regions were deleted. These data suggest that the RanGTP-independent binding of pri-mir-19a by XPO5 requires the lower stem regions outside of the pre-miRNA hairpin.XPO5 facilitates the microprocessor cleavage of pri-miRNAsTo determine the function of XPO5 binding to pri-mir-19a precursors, we performed the DROSHA/DGCR8-mediated processing assay for pri-mir-19a and the two mutants in vitro. Flag-tagged DROSHA and DGCR8 expression plasmids were co-transfected into HEK293T cells to obtain the microprocessor complex using immunoprecipitation and purification as described previoulsy30. We also purified Flag-XPO5 protein from transfected HEK293T cells. Western blot and silver stain indicated that DROSHA/DGCR8 microprocessor does not copurify with XPO5 (Supplementary Fig. 3a, b). In the absence of XPO5, the microprocessor complex released pre-mir-19a hairpin from the primary transcript with a modest efficiency (Fig. 3a, lane 3), in line with previously studies30,31. Notably, preincubation of pri-mir-19a with an increasing amount of XPO5 drastically enhanced the processing efficiency. The improved processing was demonstrated by both the significant reduction of unprocessed pri-mir-19a and approximately twofold more accumulation of pre-mir-19a in an XPO5 dosage-dependent manner (Fig. 3a, lanes 4–6). XPO5 alone did not alter pri-mir-19a (Fig. 3a, lane 7), ruling out the possibility of microprocessor contamination.Fig. 3XPO5 facilitates the microprocessor cleavage of pri-mir-19a.a Preincubation of XPO5 enhances the microprocessor cleavage of pri-mir-19a. Incubation of increasing amount of XPO5 leads to increased processing efficiency of pri-mir-19a (lanes 3–6). Incubation of increasing amount of the XPO5 mutant also results in increased processing efficiency of pri-mir-19a (lanes 8–11). b Preincubation of XPO5 and XPO5 mutant enhances the microprocessor cleavage of truncated pri-mir-19a v1. c Preincubation of increasing amounts of XPO5 does not promote the cleavage efficiency of pri-mir-19a truncation v2 by the microprocessor. Representative images of three independent repeats are shown for each assay (a–c). (Quan. = quantification). Original data for all panels are provided in the Source Data file. .XPO5 has been well characterized for its binding to pre-miRNA hairpin that is dependent upon RanGTP and 2-nt 3′ overhang of pre-miRNA6,32. So far, we have showed the RanGTP-independent binding of pri-miRNA by XPO5 (Fig. 2b–e, g). We next asked whether XPO5 residues that mediate the interaction between XPO5 and the 5′ and 3′ ends of pre-miRNA are required for enhancing the microprocessor cleavage. To this end, we generated an XPO5 mutant with E445A, S606A, E711A, and R718A quadruple mutations. These residues were found to form hydrogen bonds between XPO5 HEAT repeats and the 5′ and 3′ ends of pre-mir-30a including the 2-nt 3′ overhang6. In support of the notion that XPO5 interaction with pri-miRNA is independent of its association with pre-miRNA, the XPO5 mutant retained the ability to potentiate the DROSHA/DGCR8 processing (Fig. 3a, lanes 8–11).To further confirm that the ability of XPO5 to promote pri-mir-19a processing requires RanGTP-independent binding, we performed the same assays using the two truncated mutants. Importantly, although both mutants can be cleaved by the microprocessor, only pri-mir-19a truncation v1, which is bound by XPO5 in the absence of RanGTP, was more efficiently cleaved by both the WT and mutant XPO5 (Fig. 3b). In contrast, the processing of pri-mir-19a truncation v2, which is not bound by XPO5, was not further enhanced by XPO5 (Fig. 3c). Together, these data reveal that RanGTP-independent binding of pri-mir-19a by XPO5 outside of the pre-miRNA hairpin promotes the microprocessor cleavage in intro.XPO5 promotes pri-miRNA processing of clustered miRNAsTo extend these findings beyond pri-mir-19a, we turned to pri-mir-15b~16-2, which encodes two miRNAs and also has extensive XPO5 association outside of pre-miRNA hairpin (Supplementary Fig. 1d). Notably, the predicted secondary structure of pri-mir-15b~16-2 also contains extensive base-pairing regions outside of the pre-miRNA hairpin (Fig. 4a). In support of the HITS-CLIP data, pri-mir-15b~16-2 bound to XPO5 in a RanGTP-independent manner. Super-shifted bands were also observed with the increasing dose of XPO5 (Fig. 4b), similar to the pattern of pri-mir-19a. In addition, an isolated pri-mir-15b containing the lower stem region also bound to XPO5 and formed a singly shifted complex with the Kd of 9.57 nM (Fig. 4c–e), similar to the pattern of pri-mir-19a truncation v1 (Fig. 2g, h). Together, these in vitro data reveal an unexpected RanGTP-independent binding of XPO5 to pri-mir-17~92 and pri-mir-15b~16-2 polycistronic miRNA, pri-mir-19a and pri-mir-15b precursors.Fig. 4XPO5 promotes primary miRNA processing of clustered miRNAs.a Secondary structure of pri-mir-15b~16-2 is predicted by RNAfold. Heat map indicates the probability of predicted structures (purple: 0—low; red: 1—high). b XPO5 binds to pri-mir-15b~16-2 in a RanGTP-independent manner. Increasing amount of XPO5 results in a super shift. Black hairpin represents pri-mir-15b~16-2. XPO5 is coloured in orange. c Secondary structure of pri-mir-15b is predicted by RNAfold. d XPO5 binds to pri-mir-15b in a RanGTP-independent manner. Black hairpin represents pri-mir-15b. XPO5 is coloured in orange. e The dissociation constant (Kd) between XPO5 and pri-miR-15b is calculated based on the binding results in d. f Preincubation of XPO5 enhances the microprocessor cleavage of pri-mir-15b~16-2. Incubation of increasing amount of XPO5 leads to increased processing efficiency of pri-mir-15b~16-2 (lanes 3–6). Incubation of increasing amount of the XPO5 mutant also results in increased processing efficiency of pri-mir-15b~16-2 (lanes 8–11). g Preincubation of increasing amounts of XPO5 promotes the cleavage efficiency of pri-mir-15b by the microprocessor. Representative images of three independent repeats are shown for each assay (f–g). (Quan. = quantification). Original data for b, d, f, and g are provided in the Source Data file.Because of the RanGTP-independent association of pri-mir-15b~16-2 and pri-mir-15b to XPO5 (Figs. 4b, d), we next tested these primary miRNAs with the processing assay. The processing efficiency of both pri-mir-15b~16-2 and pri-mir-15b by the microprocessor was also strongly enhanced (approximately twofold) by the presence of XPO5 (Fig. 4f, lanes 3–6 and Fig. 4g, lanes 3–5). Notably, the unprocessed pri-mir-15b~16-2 and pri-mir-15b was strongly reduced. We detected both singly cropped pri-mir-15b~16-2 intermediates and pre-mir-15b/16-2, and the accumulation of both was increased by XPO5. In addition, the quadruple XPO5 mutant also promoted the DROSHA/DGCR8 processing (Fig. 4f, lanes 8–11). Taken together, these data suggest that RanGTP-independent binding of pri-mir-19a, pri-mir-15b~16-2 and pri-mir-15b by XPO5 facilitates the DROSHA/DGCR8 processing. These findings reveal an additional function of XPO5 independent of nuclear export.XPO5 binds to cellular RNAs with double-stranded regionsIn addition to miRNA precursors, we identified numerous noncoding RNAs in the XPO5 HITS-CLIP datasets (Fig. 5a and Supplementary Data 2). Among these, noncoding RNAs were SINE and LINE elements (repeat), vault RNAs (vtRNAs), 7SL RNA, snRNA, snoRNA, 7SK RNA, tRNA, and hTR (Fig. 5b, g, i and Supplementary Fig. 4). vtRNAs were among the most abundant XPO5-associated RNAs. Notably, vtRNAs were abundantly bound by all core components of miRNA biogenesis and likely gave rise to small RNAs associated with AGO proteins (Fig. 5b). Similar to pri-mir-19a, vtRNAs were bound by XPO5 in a RanGTP-independent manner (Fig. 5c). Furthermore, although vtRNAs typically formed one major complex with XPO5, vtRNAs were also super-shifted by the increasing amount of XPO5 protein (Fig. 5c). The Kd of vtRNA and XPO5 association was determined at 36.27 nM (Fig. 4d), weaker than that of pri-mir-19a and pri-mir-15b. The weaker interaction between vtRNAs and XPO5 was correlated with the less prominent and shorter dsRNA structures of vtRNAs (Fig. 5e and Supplementary Fig. 5a).Fig. 5XPO5 associates with diverse cellular RNAs.a Peak annotation of XPO5-associated reads from XPO5 HITS-CLIP. b, g, i IGV tracks of vtRNA1-1, hTR, and 7SL RNA1 show the association of XPO5 with each noncoding RNA, compared with DROSHA, DGCR8, DICER1, and AGO2/3. Blue bar at the bottom indicates the coding region and the arrow indicates the direction of transcription. Data range is shown on the right of each track. c XPO5 binds to vtRNA1-1 without RanGTP. 1 nM vtRNA1-1 substrate was used. Red arrows indicate vtRNA1-1 RNA with (upper) and without (lower) XPO5 binding, red bracket indicates the super-shifted vtRNA1-1 RNA with more than one XPO5 molecule. d The dissociation constant (Kd) between XPO5 and vtRNA1-1 is calculated based on the binding results in c. e Predicted secondary structure of vtRNA1-1. The potential to form predicated structures is coloured (purple to red → low to high). f Detection of vtRNA and 5S RNA (loading control) from nuclear and cytoplasmic RNAs extracted from wild type and XPO5 KO MEF cells by northern blotting. h XPO5 HITS-CLIP reads align to the secondary structure of hTR. Red and green arrows indicate the 5′ and 3′ location of XPO5-associated RNA regions, respectively. Black lines outline the XPO5-associated hTR regions. j An RNA Folding analysis shows the folding energy of XPO5-assoicated repeat RNAs, XPO5-assocatied nonrepeat RNAs, XPO5-associated pre-miRNAs, miRbase pre-miRNAs, and randomly shuttled control sequences. For each boxplot, the middle line is the median, the vertical line spans the data range, and the hinges are the first and third quartiles. A two-sided Mann–Whitney test was used for statistical tests. Original data for c and f are provided in the Source Data file.To probe the potential effect of XPO5 on the biogenesis and localization of vtRNA, we generated XPO5 knockout (KO) MEF cells (Supplementary Fig. 5b, c, also see below). Interestingly, neither the level nor the cytoplasmic localization pattern of vtRNA was altered in the absence of XPO5 (Fig. 5f). These data suggest that the binding of vtRNA by XPO5 does not affect the maturation, accumulation, and localization of vtRNA.We noticed that many of XPO5-associated RNAs are predicted to form double-stranded regions. In general, XPO5 showed more widespread binding patterns than the other core components including DROSHA, DGCR8, and DICER1 in structured RNAs such as hTR and 7SL RNA (Fig. 5g–i). hTR interacts with XPO5 in three domains with the strongest interaction detected towards the 5′ end of hTR. Intriguingly, the number of XPO5 HITS-CLIP reads associated with each domain was different, despite of their origin from the same RNA (Fig. 5g). To probe whether XPO5 association may be correlated with the strength of base-pairing, we further examined XPO5-bound regions on hTR. Three highly structured regions of hTR including the template region, the CR4/CR5 domain and the scaRNA domain have been determined previously33,34. Among these regions, the template region forms the most extensive dsRNA structures, followed by the CR4/CR5 and scaRNA domains (Fig. 5h). Notably, XPO5-associated RNA reads were mostly enriched in the double-stranded regions (P2a-P2b-J2b/3) of the template region, followed by the reads over the double-stranded regions of the CR4/CR5 domain and the scaRNA domain, respectively (Fig. 5g). In addition, AGO interacts with small RNA fragments that are derived from the scaRNA domain located at the 3′ end of hTR but not the more prominent template region or the CR4/5 domain (Fig. 5g). These data suggest that XPO5 preferentially interacts with structured regions of hTR. Similar to hTR, 7SL RNA has well-defined secondary structures with extensive base-pairing regions that are conserved through evolution35,36. Interestingly, 7SL RNA strongly interacts only with XPO5 but not with other core components and AGO-associated small RNA fragments were not detected (Fig. 5i), indicating a possible function of XPO5 that is independent of miRNA biogenesis.To globally measure the preference of XPO5 to base-paired RNAs, we performed an RNA Folding energy analysis for all XPO5-associated RNAs based on HITS-CLIP. We classified these RNAs into repeat associated regions, nonrepeat associated regions and pre-miRNAs. As shown in Fig. 5j, the folding energy of XPO5-assoicated pre-miRNAs was significantly lower than that of randomly selected sequences and indistinguishable from that of all pre-miRNA hairpins. Interestingly, both repeated associated and nonrepeat associated regions bound by XPO5 had an even lower folding energy than that of pre-miRNA hairpins (Fig. 5j). These data suggest that XPO5 has a preference to double-stranded regions of cellular RNAs.XPO5 is required for mouse embryonic developmentHaving characterized XPO5-associated cellular RNAs, we next examined the function of XPO5 in mouse development. We identified a knock-in (KI) first, conditionally targeted embryonic stem cell clone for generating XPO5 KO mouse models37. Founder KI mice were produced and validated for germline transmission of the targeted allele that contains an IRES-LacZ and Neo cassette in the intron 7 and a floxed exon 8 of XPO5 (Fig. 6a). To generate the conditional allele, the founder KI mice were bred to Actb-Flp mice to remove the IRES-LacZ and Neo cassette. Finally, the conditional allele was bred to EIIA-Cre or Krt14-Cre mice to generate constitutive KO and skin cKO of XPO5, respectively (Fig. 6a).Fig. 6XPO5 is required for mouse embryonic development.a Schematics of the XPO5 allele design. b Loss of XPO5 results in embryonic lethality. XPO5 KO embryos show developmental defects as early as in E6.5 (c) and fails to go through gastrulation (d). Representative images of littermate embryos are shown for over ten control and KO embryos. e Whole-mount LacZ staining of XPO5KI/+ (Ctrl) and XPO5KI/KI (KO) shows universal expression of XPO5 throughout the embryo. f Depletion of XPO5 and mir-290 in XPO5 KO embryos at E7 is measured by qPCR. Dysregulation of selected genes in XPO5 KO embryos at E7 (g) and E8.5 (h) is measured by qPCR, respectively. Oct4 and Nanog are failed to be downregulated in XPO5 KO embryos at both E7 and E8.5. Scale bar 250 μm in c–e. Data shown are mean s.d. from three independent experiments. *P < 0.05; **P < 0.01; ***P < 0.001 determined by Student’s t test. Original data for f, g, and h are provided in the Source Data file.Neither XPO5KI/+ nor XPO5KO/+ breeding produced any viable XPO5 null (XPO5KI/KI or XPO5KO/KO) mice at birth, whereas XPO5 WT and heterozygous mice were obtained at the expected Mendelian ratio, indicating a requirement of XPO5 during mouse embryonic development (Fig. 6b). Closer inspection revealed that XPO5 null embryos showed signs of compromised development around embryonic day 6.5, a developmental stage that is correlated with the onset of gastrulation and embryonic germline layer formation. At this stage, XPO5 null embryos were much smaller, compared with their control littermates (Fig. 6c). Furthermore, XPO5 null embryos failed to go through gastrulation and initiate organogenesis by E8.5 (Fig. 6d). Judging by the LacZ signals in E7 XPO5KI/+ and XPO5KI/KI embryos, XPO5 was universally expressed in all embryonic cells (Fig. 6e).We then performed quantitative real-time PCR (qRT-PCR) analyses to validate the loss of XPO5 and mir-290, a representative miRNA that is highly expressed in mouse ESCs and early embryos, using total RNA isolated from E7 WT, het, and KO embryos (Fig. 6f). Because the miRNA pathway is required to promote the differentiation of ESCs to lineage specific cell types and the loss of Dicer1 and Dgcr8 blocks ESC differentiation38,39, we also quantified ESC marker genes such as Nanog and Oct4 as well as differentiation markers such as T and Tet1 by qRT-PCR. The results revealed a strong accumulation of Nanog and Oct4 mRNAs in the E7 XPO5 null embryos, and a relatively normal level of T and Tet1 (Fig. 6g). Furthermore, the high level of Oct4 persisted in E8.5 XPO5 null embryos (Fig. 6h), reflecting a compromised differentiation. Together, these data reveal that XPO5 is required for miRNA biogenesis and mouse embryonic development.XPO5 is required for skin morphogenesisWe have previously studied Dicer1, Dgcr8, as well as Ago1/2 in the skin using a K14-Cre line and observed unique defects in hair morphogenesis that are characteristic of the loss of the entire miRNA pathway in the skin12,18,19. Furthermore, both Dicer1 and Dgcr8 cKO skin typically loses >95% of individual miRNAs with a few exceptions of short hairpin miRNAs such as mir-320 and mir-484 that are dependent on Dicer1 but independent of Dgcr819. In comparison, Ago1/2 dKO skin loses ~80% of individual miRNAs and shows milder defects12. To compare the developmental defects of XPO5 cKO animals with these well-characterized cKO models of the miRNA biogenesis pathway in the skin, we generated K14-Cre/XPO5fl/fl mice (Fig. 7a).Fig. 7XPO5 is required for skin development.a Schematic of K14-Cre/XPO5fl/fl mouse model. b Whole-mount LacZ staining of XPO5KI/+ in P2 and P4 skin. c Neonatal mice of XPO5 cKO animals with littermate controls from P1.5 to P14. d, e
XPO5 cKO animals are neonatal lethal and fail to gain weight. f Depletion of mir-203 and mir-205 in XPO5 cKO epidermis is quantified by qPCR. Data shown are mean s.d. from three independent experiments. ***P < 0.001 by Student’s t test. g Depletion of XPO5 and AGO2 proteins in XPO5 cKO epidermis is confirmed by western blot. h H&E staining reveals reduced hair follicle formation and stunted hair follicle growth in XPO5 cKO skin at P2. i Stunted hair follicle down growth and evaginating hair germ (arrowhead) in XPO5 cKO skin are revealed by Lef1 and β4-integrin staining. j Reduced cell proliferation in XPO5 cKO skin is shown by Ki67 staining. k
XPO5 cKO skin shows normal epidermal differentiation as determined by Krt5 and Krt1 staining. l Normal epidermal differentiation is confirmed by qPCR detection of Krt1, Lor, Krt5, and Itgb4 (encodes β4-integrin). Data shown are mean s.d. from three independent experiments. *P < 0.05, n.s.—not significant, by Student’s t test. Scale bar: 1 cm in c, 100 μm in b, h, j, and k, 50 μm in i. Original data for d–g and l are provided in the Source Data file.We first documented the expression pattern of XPO5 in the skin by monitoring LacZ signals in XPO5KI/+ heterozygous mice. XPO5 was universally expressed in all skin cell types at postnatal day 2 and 4 (P2 and P4). By P4, when hair follicles gave rise to terminally differentiated inner root sheath and hair shaft, LacZ signals appeared to be elevated in these differentiated cells than their progenitor counterparts (Fig. 7b). In XPO5 skin cKO animals, we observed progressively compromised hair follicle development that was similar but slightly milder, compared with Dicer1 and Dgcr8 cKO (Fig. 7c). XPO5 cKO animals showed neonatal lethality between postnatal day 5 and 18 and failed to gain weight postnatally (Fig. 7c–e). This was in contrast to Ago1/2 dKO animals, which lose ~80% of total miRNAs and usually survive to adulthood despite the apparent loss of hair follicles12. Consistent with these observations, qRT-PCR quantification of mir-203 and mir-205, two of the most highly expressed miRNAs in the skin, showed >90% depletion for both miRNAs (Fig. 7f). In addition, AGO2 protein was also depleted in XPO5 null skin (Fig. 7g). Because mature miRNA accumulation is required to stabilize AGO proteins40, this result provided further support to the strong depletion of global miRNA expression in the absence of XPO5.To further document the role of XPO5 during skin morphogenesis, we examined neonatal skin. We observed reduced hair follicle formation and stunted hair follicle down growth in P2 skin (Fig. 7h). When examined specifically with Lef1 and β4-integrin staining for hair germ, we found many evaginating hair germs towards the epidermis that is characteristic of Dicer1 and Dgcr8 cKO skin17–19 (Fig. 7i). Furthermore, epidermal cell proliferation was strongly reduced (Fig. 7j). Finally, epidermal differentiation was not strongly affected as documented by the Krt5 and Krt1 staining that illuminated the basal and differentiated spinous layer, respectively (Fig. 7k) and by qRT-PCR analysis (Fig. 7l). Collectively, these studies provide further genetic evidence to support the requirement of XPO5 for miRNA biogenesis and skin development in mouse.XPO5 is required for miRNA biogenesisHaving demonstrated the requirement of XPO5 in mouse embryonic and tissue development, we further characterized the global miRNA expression levels in XPO5 KO skin using a quantitative small RNA sequencing method41 (Fig. 8a). Globally, small RNAs between 20 and 23 nt showed a strong depletion in the XPO5 KO samples (Fig. 8b). We then specifically examined miRNA depletion. Although it was somewhat weaker than the depletion observed in the Dicer1 and Dgcr8 cKO skin (Fig. 8c), the average level of depletion was ~90%. The depletion of mature miRNA was particularly evident for highly expressed miRNAs whose expression can be more robustly measured by small RNA sequencing (Fig. 8d and Supplementary Data 3). In addition to highly expressed mir-203 and mir-205 (Fig. 7f), we quantified the depletion of five miRNAs from the mir-17~92 and mir-15~16 clusters (Fig. 8e, f), which expressed at an intermediate level in the skin. We observed strong depletion for mir-17-5p, mir-18, mir-19a/b, and mir-20a except mir-92a as well as mir-15a/b and mir-16 (Fig. 8e, f). Interestingly, the loss of mir-19a/b was particularly strong, likely reflecting the requirement of XPO5 for its processing and nuclear export. To further validate the compromised miRNA biogenesis and distinguish the effect of XPO5 loss on different miRNA species, we performed northern blotting for mir-17 (Fig. 8g). We observed the loss of mature mir-17 but no strong changes of pre-mir-17 were detected (Fig. 8g). The lack of accumulation of pre-miRNAs is likely due to their unstable nature in the absence of XPO5, similar to the patterns reported in Dicer1 cKO18. In addition, mir-320 and mir-484, whose biogenesis is independent of Drosha/Dgcr8 processing19, were largely unchanged in XPO5 cKO skin (Fig. 8h), lending support to the notion that XPO5 is required to export pre-miRNA hairpin generated by the Drosha/Dgcr8 microprocessor. Taken together, these analyses reveal the requirement of XPO5 for the biogenesis of a majority of miRNAs in the skin, consistent with the strong defects observed in the cKO model. We also note, however, whether the phenotypes of XPO5 KO is entirely due to the loss of miRNAs should be further examined due to the binding of XPO5 to many cellular RNAs with double-stranded regions.Fig. 8XPO5 is required for miRNA biogenesis.a Bioinformatic pipeline of quantitative small RNA-seq and data analysis. b Small RNAs between 20 and 23 nt showed strong depletion in XPO5 cKO skin samples. Small RNA reads from 18 to 27 nt were charted from small RNA cDNA libraries. c The depletion of miRNAs in XPO5 cKO skin is weaker than that in Dicer1 and Dgcr8 cKO skin. For each boxplot, the middle line is the median, the vertical line spans the data range, and the hinges are the first and third quartiles. A two-sided Mann–Whitney test was used for statistical tests. d Depletion of mature miRNA reads in XPO5 cKO skin is evident for highly expressed miRNAs (red coloured dots). e Depletion of mir-17-5p, mir-18, mir-19a, and mir-20a except mir-92a in XPO5 cKO skin is measured by qPCR. f Depletion of mir-15a, mir-15b, and mir-16 in XPO5 cKO skin is measured by qPCR. g Depletion of mir-17-5p is confirmed by northern blotting. h Unchanged expression of mir-320 and mir-484 in XPO5 cKO skin is measured by qPCR. Data shown are mean s.d. from three independent experiments. **P < 0.01; ***P < 0.001, n.s.—not significant by Student’s t test. Original data for e–h are provided in the Source Data file.DiscussionIn this study, we have determined XPO5-associated RNA species at the genomic scale. As expected, XPO5 associates with pre-miRNA precursors for a vast majority of expressed miRNAs (Fig. 1b). Surprisingly, however, XPO5 strongly binds to primary miRNA precursors of some closely clustered polycistronic miRNAs such as mir-17~92 and mir-15b~16-2 (Fig. 1i–k and Supplementary Fig. 1). The in vitro binding assays provided evidence that XPO5 binds to these unprocessed primary miRNAs in a RanGTP-independent manner (Figs. 2 and 4). Studies of the mir-19a precursors further suggest that the binding site of XPO5 to these primary miRNAs is in the extensive base-pairing regions beyond pre-miRNA hairpin. Because RanGTP-dependent cargo association is characteristic of exportin-mediated nuclear export1, the RanGTP-independent association with these primary miRNAs suggests a hitherto unrecognized function of XPO5. Indeed, incubation of XPO5 with both pri-mir-19a, pri-mir-15b~16-2, and pri-mir-15b increased the processing efficiency of the DROSHA/DGCR8 microprocessor, providing evidence for the involvement of XPO5 in the nuclear cleavage of closely clustered, polycistronic miRNAs. Given the lack of association of XPO5 with monocistronic or sparsely clustered miRNAs, it is possible that the XPO5-mediated mechanism for efficient pri-miRNA cleavage is unique to closely clustered miRNAs with extensive dsRNA regions. It is tempting to speculate that closely clustered miRNAs may use multiple mechanisms including XPO5 binding to regulate their biogenesis, conferring more complex control to the production of these miRNAs posttranscriptionally. However, how many of these closely clustered miRNAs whose nuclear cleavage is regulated by XPO5 should be investigated in future studies in a cell type-specific manner.Second, XPO5 pervasively associates with many cellular RNAs in addition to miRNA precursors. Global RNA folding analyses and individual case examination suggest that XPO5 has a preference to dsRNA regions. In addition to miRNA precursors, XPO5 binds numerous cellular RNAs such as vtRNA, 7SL RNA, snRNA, snoRNA, 7SK RNA, tRNA, hTR, and SINE and LINE repeat elements, which have diverse cellular localization and functions42. Interestingly, the Kd of XPO5 to pri-mir-19a variants and vtRNA is similar to the Kd of MDA5 and dsRNA43. Although functional consequences of these binding events will be examined in future studies, it is possible that XPO5 has multiple functions beyond miRNA biogenesis1. We note mammalian XPO5 was originally identified as a binding protein to dsRNA binding proteins such as ILF3, PKR, and STAUFEN16. Our results now provide a molecular basis at the genomic scale for these earlier findings and establish a foundation to investigate the link between XPO5 and cellular dsRNAs and their binding proteins.Finally, we have unequivocally demonstrated the requirement of XPO5 for global miRNA biogenesis and mouse development by examining XPO5 KO in embryonic and skin development. In both systems examined, we observed strong developmental defects reminiscent of those observed in Dicer1 and Dgcr8 KO animals. In the skin, we have previously shown that the loss of Dicer1 or Dgcr8 leads to complete depletion of most miRNAs with a couple of exceptions such as mir-320 and mir-484, which are Dicer1-dependent but Dgcr8-independent18,19. These cKO animals were neonatal lethal and died before one-week old. In contrast, KO of Ago1 and Ago2 together (dKO) leads to 70–80% depletion of most miRNAs due to quantitative loss of AGO proteins and the dKO mice usually survived to adulthood with prominent defects in hair follicles12. Collectively, these models set useful boundaries to estimate the requirement of XPO5 for miRNA biogenesis if miRNAs are not completely depleted in the absence of XPO5. Our miR-seq, qPCR and northern blotting analyses suggested that KO of XPO5 leads to ~90% depletion of most miRNAs. Of note, Dgcr8-independent miRNAs, mir-320 and mir-484, were not affected by the loss of XPO5. This result is consistent with a previous report for XPO5-independent miRNA export9 and indicates an intrinsic link between DROSHA/DGCR8 produced pre-miRNA hairpin and XPO5-mediated export. In support of these molecular measurements, XPO5 skin cKO animals lived longer than their Dicer1 or Dgcr8 cKO counterparts but shorter than Ago1/2 dKO animals. Although other factors could weakly compensate for the loss of XPO5 for miRNA expression, they are unable to rescue the developmental defects. Therefore, we conclude that XPO5 is broadly required for miRNA biogenesis and mouse development. We note, however, because XPO5 also binds to many non-miRNA substrates the phenotypes of XPO5 KO may not be entirely due to the loss of miRNA biogenesis. This possibility warrants future investigations.MethodsMiceAll WT and transgenic animal breeding and operation procedures were approved by the Institutional Animal Care and Use Committees (IACUC) at the University of Colorado Boulder and in accordance with the guidelines and regulations for the care and use of laboratory animals, and compiled with all relevant ethical regulations for animal testing and research, and received relevant ethical approvals. The XPO5KI animal was purchased as an ES cell clone from the EU’s KO mouse consortium. EIIa-Cre mice were obtained from the Jackson laboratory (JAX #003724). The mouse was crossed with Flipase expressing mice to generate the XPO5fl/fl mice. XPO5KI, XPO5fl/fl, and K14-Cre; XPO5fl/fl mice were bred and housed in the University of Colorado Boulder. Embryos studies were timed by the presence of a plug, indicating gestational age E0.5. All mouse experiments were conducted in accordance with animal protocols approved by the IACUC.Cell cultureHEK293T cells, obtained from ATCC (ATCC CRL-3216), were cultured and maintained in DMEM (GIBCO) supplemented with 10% heat-inactivated fetal bovine serum and 1% penicillin/streptomycin (GIBCO) in the 5% CO2 incubator at 37 °C. XPO5fl/fl MEF cells were isolated from E14 XPO5fl/fl mice using the Pierce™ Mouse Embryonic Fibroblast Isolation Kit (#88279, Thermo Fisher Scientific). XPO5 KO MEF cells were generated by Adeno-cre virus infection. XPO5fl/fl and XPO5 KO MEF cells were cultured and maintained in DMEM for Pierce™ Primary Cell Isolation Kits (#88287, Thermo Fisher Scientific) supplemented with 10% heat-inactivated fetal bovine serum and 1% penicillin/streptomycin (GIBCO) in the 5% CO2 incubator at 37 °C.RNA purification and mRNA and miRNA qPCRTotal RNA was extracted using Trizol (Thermo Fisher Scientific). For mRNA analysis, 1 μg total RNA was used to synthesize cDNA by Superscript III Reverse Transcriptase (Thermo Fisher Scientific). For miRNA analysis, the miScript II RT Kit (Qiagen) was used to synthesize cDNA. Reactions were performed according to the manufacturer’s manual and on a CFX384 real-time system (Bio-Rad). Differences between samples and controls were calculated using the 2−ΔΔC(t) method. All the primers used for qPCR are listed in Supplementary Table 1.Immunostaining and imagesFor analysis of back skin phenotypes, OCT sections were fixed in 4% PFA for 10 min in phosphate-buffered saline (PBS) and washed three times for 5 min in PBS at room temperature. Block the sections with 2.5% NGS, 2.5% NDS in PBS. Primary antibodies against the following proteins were used: β4-integrin (β4, 1:100, BD Biosciences), Lef1 (1:500, Cell signaling), K5 (1:5000, Covance), K1 (1:2000, Covance), and Ki67 (1:500; Abcam). Imaging was performed on a Leica DM5500B microscope with an attached Hamamatsu C10600-10B camera and MetaMorph (version 7.7; MDS Analytical Technologies) software. For all images, single optical sections were used.X-gal histochemistry stainingThick sections (25 µm) of back skin samples were fixed in fix solution (0.5% glutaraldehyde, 1.25 mM EGTA pH 7.3, 2 mM MgCl2, 1xPBS) for 10 min at room temperature. For the embryos, whole-mount embryos were fixed in fix solution for 5 min at room temperature. Fixed sections and embryos were washed twice with detergent rinse (100 mM phosphate buffer pH 7.4, 2 mM MgCl2, 0.01% sodium deoxycholate, 0.02% NP-40), and then incubated in staining solution (100 mM phosphate buffer pH 7.4, 2 mM MgCl2, 0.02% NP-40, 5 mM potassium ferricyanide, 5 mM potassium ferrocyanide, 20 mM Tris pH 7.5, 1 mg/mL X-gal) at 37 °C for 48 h.HITS-CLIPXPO5 HITS-CLIP and data analyses were performed as described44 with slight modifications41. In Brief, HEK293T cells were UV-crosslinked, resuspended in lysis buffer. Supernatants were incubated with Anti-Exportin-5 antibody (Abcam) or control IgG for 1 h on ice with or without RNase. After incubation, dynabeads were added into the protein–RNA complexes solutions and incubate for 1 h at 4 °C, followed by labeling 5′ end of RNA with [γ-32P] ATP using PNK. Then 5′ Ligation Adapter-NN was linked to 5′ end of RNA using T4 RNA ligase 1 at room temperature for 2 h. Protein–RNA complexes were incubated in 1x Nu-PAGE loading buffer (Invitrogen) at 70 °C for 10 min, resolved on 8% Nu-PAGE Bis-Tris gel (Invitrogen), transferred into a nitrocellulose membrane and exposed to a Fuji film. Bands corresponding to specific protein–RNA complexes were excised, and then digested with proteinase K. RNAs were purified and adenylated linker was linked to 3′ end of purified RNAs. cDNAs were made using Superscript III(Invitrogen) with 3′ adenylated linker primers. cDNAs were separated on 15% denaturing urea polyacrylamide gel and one area around 100 bp were excised purified, followed by amplification by Phusion High Fidelity polymerase (NEB). PCR products were separated by 6% denaturing urea polyacrylamide gel and three bands around 100–150 were purified. Obtained CLIP libraries were subjected to high-throughput deep sequencing.Small RNA-seqTotal RNA samples from P2–P4 back skin epidermis were subject to high efficiency 3′ ligation as described previously41. Products were resolved on 15% PAGE-urea gels and stained with SybrGold (Molecular Probes S-11494 Carlsbad, CA, USA). The region corresponding to ligated miRNAs (46 nt) was excised, gel slices were thoroughly minced, and eluted in HSCB (400 mM NaCl, 25 mM Tris-HCl pH 7.5, 0.1% SDS) overnight at 4 °C. Nucleic acids were precipitated in the presence of glycogen carrier via 0.1 volumes sodium acetate and 2.5 volumes ethanol. Pellets were washed in 70% ethanol, and dissolved in the 5′ ligation mix without enzyme. Samples were briefly heated to 70 °C, snap chilled on ice, and then enzyme was added. Following ligation reactions cDNA was prepared using Superscript III RT (Invitrogen) according to the manufacturer′s recommendation using 3′ linker-specific RT primer. cDNA libraries served as templates for PCR amplification; amplicons were resolved on 8% native acrylamide gels. Bands of the correct molecular weight were isolated and used for high-throughput sequencing on the Illumina HiSeq2000 and HiSeq4000.Plasmids construction and transfectionConstructs coding for FLAG-tagged human XPO5, XPO5 mutant and RanQ69L were cloned in the pcDNA3.1 vector for transient expression. For the mir-17~92a pri-miRNA in vitro transcription plasmid pJMC101-T7-pri-miR17~92a, human mir-17~92a pri-miRNA was cloned using mir-17~92a pri-miRNA transcription forward and reverse primers (including T7 promotor) and T7:pri-mir-17~92a were inserted into pJMC101vector. Primers for plasmid construction are listed in Supplementary Table 1. HEK293FT transfection was performed using Mirus TransIT®-LT1 Transfection Reagent (# MIR 2300, Fisher) according to the manufacturer’s instructions.Protein expression and purificationFor XPO5-His6 and His6-RanQ69L proteins expression, pQE60-XPO5 and pQE32-RanQ69L plasmids are transformed separately into E.coli strain TG1 cells16. The XPO5 subcloned cells were grown in LB medium with 2% ethanol (vol/vol) at 37 °C, followed by induction with 400 µM IPTG at 18 °C for 16 h. XPO5-His6 was purified on NTA-Ni2 beads (QIAGEN). For His6-RanQ69L protein, subcloned cells were grown in LB medium at 37 °C, followed by induction with 100 μM IPTG at room temperature overnight. His6-RanQ69L was purified on NTA-Ni2 beads and loaded GTP on ice for 3 h29.Electrophoretic mobility shift assayCold mir-17~92a pri-miRNA, pre-mir-30a, pre-mir-19a, pri-mir-19a, pri-mir-19a truncation v1, pri-mir-19a truncation v2, pri-mir-15b~16-2, pri-mir-15b, and vtRNA1-1 transcripts were produced by the MEGAshortscript™ T7 Transcription Kit (#AM1354, Thermo Fisher Scientific) using pJMC101-T7-pri-mir-17~92 template (digested by HindIII), T7-pre-mir-30a, T7-pre-mir-19a, T7-pri-mir-19a, T7-pri-mir-19a truncation v1, T7-pri-mir-19a truncation v2, T7-pri-mir-15b~16-2, T7-pri-mir-15b, and T7-vtRNA1-1 PCR products separately as the templates (primers are listed in Supplementary Table 1), and purified by 5% PAGE gel. The 5′-phosphote of those cold substrates were removed by Antarctic Phosphatase (# M0289S, NEB). After dephosphorylation, those substrates were end-labeled by [r-32P]-ATP using T4 Polynucleotide Kinase (# M0201S, NEB). miR17~92a, miR19a, and miR15b-16-2 pri-miRNAs were refolded in refolding buffer {1XTHE (66 mM HEPES, 33 mM Tris, 1 mM EDTA), 100 mM NH4OAc, 5 mM MgOAc, 0.5% NP-40, 0.1 mM EDTA} at 60 °C for 10 min and slowly cooled down to 25 °C in 25 min. After refolding, mir-17~92a, mir-19a, mir-15b, mir-15b~16-2 pri-miRNAs, and vtRNA1-1 were separately incubated with XPO5 protein in a total volume of 10 μl at 25 °C for 45 min. Pre-mir-30a and pre-mir-19a were incubated with XPO5 protein and RanQ69L. The binding buffer contained 20 mM HEPES (pH 7.3), 150 mM potassium acetate, 2 mM magnesium acetate, 0.05% NP-40, 7 mM 2-mercaptoethanol, 1.5 μg/mL poly dIdC, and 0.2% BSA. After incubation, 0.3 μg of heparin was added to the reaction and incubated for another 5 min4. Samples were then analyzed by electrophoresis on a 0.7% nondenaturing agarose gel (for mir-17~92a pri-miRNA) at 4 °C or 5% nondenaturing PAGE gel (for pre-mir-30a, pre-mir-19a, pri-mir-19a, pri-mir-15b, mir-15b~16-2 pri-miRNAs, and vtRNA1-1) at room temperature. Gels were dried and detected by phosphorimager.In vitro analysis of pri-miRNA processing[α-32P]-CTP (#BLU008H250Uc, Perkin) labeled miR19a, miR15b, and miR15b-16-2 pri-miRNA transcripts were produced by the MEGAshortscript™ T7 Transcription Kit (#AM1354, Thermo Fisher Scientific) using T7 promotor-driven pri-MIRNA gene PCR products as the templates, and purified by 5% PAGE gel. pCK-Drosha-FLAG and pCK-FLAG-Dgcr8, pCDNA3-FLAG-XPO5, pCDNA3-FLAG-XPO5_mutant, and pCDNA3.1 plasmids were separately transfected into Human 293FT cells. After 48 h of culturing, cells were harvested and opened by hypotonic gentle lysis buffer (10 mM Tris-HCl pH 7.6, 10 mM NaCl, 2 mM EDTA, 0.5% Triton X-100, proteinase inhibitor). DROSHA and DGCR8 complex, XPO5 protein, and XPO5 mutant protein were purified by ANTI-FLAG®M2 Affinity Gel (#A2220, Sigma). XPO5 and XPO5 mutant proteins were eluted from ANTI-FLAG® M2 Affinity Gel by adding excess 3X FLAG® Peptide (#F4799, Millipore Sigma). [α-32P]-CTP labeled miR15b-16-2, miR15b, and miR19a pri-miRNA substrates were refolded in refolding buffer (1XTHE (66 mM HEPES, 33 mM Tris, 1 mM EDTA), 100 mM NH40Ac, 5 mM MgOAc, 0.5% NP-40, 0.1 mM EDTA) at 60 °C for 10 min and slowly cooled down to 25 °C in 25 min. After refolding, miR15b-16-2, miR15b, and miR19a pri-miRNA substrates were incubated with XPO5 or XPO5 mutant protein at 25 °C for 25 min. Then, DROSHA and DGCR8 complex was added into system and incubated at 37 °C for 1 h in processing buffer (100 mM Tris-HCl pH 7.6, 500 mM KCl, 1 mM EDTA, 64 mM MgCl2). Processing products were purified by phenol/chloroform and run on the 5% PAGE denaturing gel. Gels were dried and detected by phosphorimager.RNA northern blotNuclear and cytoplasmic RNAs were extracted from MEF cells using the Cytoplasmic & Nuclear RNA Purification Kit (# 21000, Norgen Biotek). Three micrograms of cytoplasmic RNA and two micrograms of nuclear RNA were separated by electrophoresis on 5% PAGE gels and electrically transferred into nylon N+ membrane. [γ-32P] ATP-labeled specific oligonucleotide probe sequences (vtRNA and 5S) were used and hybridized at 42 °C overnight. Total RNA was extracted from P4 back skin epidermis using Trizol (# 15596-018, Thermo Fisher Scientific). five micrograms of total RNA was separated by electrophoresis on 5% PAGE gels and electrically transferred into nylon N+ membrane. [γ-32P] ATP-labelled specific oligonucleotide LNA probe sequences (mir-17a) was used and hybridized at 42 °C overnight. Nonhybridized probes were washed away by wash buffer (2x SSC/0.2%SDS) for 20 min twice. Signals were detected by phosphorimager.CLIP analysisFASTQ files were trimmed using cutadapt (v1.8.3) (3′ adapter, TGGAATTCTCGGGTG CCAAG G, 5′ adapter CTACAGTCCGACGATC) to remove adapter sequences. The randomized 5′ and 3′ NN bases of the adapter were next appended to the FASTQ read ID. Reads were mapped to the human genome (hg19 build) using novoalign (v3.02.00). Reads > 25 nt after trimming were aligned with -l 25 -t 85 settings, whereas shorter reads were aligned with -l 20 -t 30 settings. PCR duplicates were removed using UMI-Tools (v. 0.5.3) with the “directional” option and alignments with MAPQ < 10 were discarded. Peaks were called by merging all overlapping alignments using bedtools merge (v2.26.0). Peaks were annotated based on overlaps with annotations downloaded from the UCSC table browser (hg19). Peaks were intersected with the following annotations in order requiring 1 bp overlap, miRNA, tRNA, snoRNA, lincRNA, utr3, utr5, cds, ncRNA, repeatMasker, introns, mitochondria, retroelements, pseudogenes, or intergenic. Peaks were classified based on the first intersecting annotation. Repeat associated regions/RNA annotations are peaks that intersect with repeatMasker annotations from UCSC genome browser. Peaks were filtered to only keep peaks present in three of five libraries, and with a minimum read count of 5. Libraries were also independently mapped to a database of pre-microRNA sequences (mirBase release 19) using blastn (v.2.2.28) to assess alignments with hairpin sequences. Publically available CLIP datasets were downloaded from GEO or ENCODE using the following accessions: DICER1 (GSE55333), DROSHA and DGCR8 (GSE61979), and DGCR8 (ENCFF491LUG).Small RNA analysisFASTQ files from DGCR8cko and DICER1cko experiments were trimmed to remove 3′ adapter sequence (TCGTATGCCGTCTTCTGCTTG)19. FASTQ files from XPO5cko experiments were trimmed to remove 3′ adapter sequence (TGGAATTCTCGG GTGCCAAGG), 5′ adapter (CTACAGTCCGACGATC), and randomized NN nucleotides introduced by the adapters from both the 5′ and 3′ end. Following trimming reads were then aligned with Bowtie2 (v.2.1.0) to the mouse genome (mm10) using local alignment. A custom annotation file was built containing GENCODE annotations, v19 for human, M8 for mouse with Mt_rRNA, lincRNA, misc_RNA, rRNA, sRNA, sm_RNA, snRNA, snoRNA, vaultRNA, and tRNA biotypes, and miRBase microRNA annotations (release 19). Reads overlapping annotations were counted using Htseq-count (v.0.6.0), requiring a minimum MAPQ of 10. FASTQ files from HEK293T cells were trimmed to remove 3′ adapter sequence (TGGAATTCTCGGGTGCCAAGG), 5′ adapter (CTACAGTCCGACG ATC), and randomized NN nucleotides introduced by the adapters from both the 5′ and 3′ end and aligned using Bowtie2 as described above. For comparing XPO5 HITS-CLIP to small RNA-Seq, small RNA-Seq reads were aligned with instead with novoalign (-l 18 -h 100 -t 60 -n 99) and reads were enumerated by counting intersections with a pre-miRNA database from miRbase.RNA secondary structure analysisRNA folding predictions for Fig. 5j were generated using RNAFold web server (v2.3.3). Peaks > 200 nt in length were excluded from the folding analysis. miRNA metagene plots were generated using the HTSeq python library (v0.6.0). The secondary structure of hTR in Fig. 5h was adapted from the telomerase database34,45. The secondary structures of noncoding RNAs in Figs. S2, S4, and S5 were generated using RNAFold web server (http://rna.tbi.univie.ac.at/cgi-bin/RNAWebSuite/RNAfold.cgi).Statistical informationFor all experiments with error bars, the standard deviation was calculated to indicate the variation within each experiment. Numbers of animals used for phenotype study has indicated in figures. Student’s t test was used for most experiments. A two-sided Mann–Whitney test was used for Fig. 5j.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article.Supplementary information
Supplementary Information
Peer Review File
Reporting Summary
Description of Additional Supplementary Files
Supplementary Data 1
Supplementary Data 2
Supplementary Data 3 | nature communications | [
"Article"
] | [
"Development",
"miRNAs"
] | karyopherin mediates nuclear export miRNA hairpin precursors binds to hairpin RNAs 2–8 nt 3′ overhang pre-miRNA minihelix viral RNA RanGTP-dependent XPO5:pre-miRNA:RanGTP nuclear export complex confirmed by X-ray crystallography6 RanGTP binding triggers conformational change XPO5 residues form hydrogen bonds 5′ 3′ ends 2-nt 3′ overhang pre-miRNAs studies XPO5 transport pre-miRNAs from nucleus to essential miRNA biogenesis7 XPO5-independent pre-miRNA export pathway 7-methylguanosine-capped pre-miRNAs DROSHA/DGCR8 microprocessor exported via PHAX-XPO1 pathway9 deletion XPO5 in human HCT116 cells mildly affected biogenesis miRNAs not required for miRNA biogenesis10 CRISPR-Cas9 screen XPO5 lack of genetically engineered mouse models of XPO5 hindered understanding XPO5 functions in vivo requirements XPO5 for global miRNA biogenesis mammalian development tissue formationXPO5 expressed protein miRNA biogenesis mouse human sensitive to saturation short hairpin RNA structured viral RNA Drosha Dicer113 RNA species pre-miRNA2–4 cellular minihelix viral RNAs adenovirus VA1 XPO5 cargoes for nuclear export XPO5-associated cellular RNAs not determined genomic scale unclear many pre-miRNAs interact with XPO5 require biogenesis XPO5 nuclear export factor pre-miRNA attention nuclear export functions XPO5 binding protein of double-stranded RNA binding proteins ILF3 PKR Staufen16 direct binding RNA pre-miRNAs tRNA VA1 RNA to XPO5 unknown other cellular RNAs double-stranded regions bind to XPO5 RNA substrates XPO5 bind RanGTP-dependent key karyopherin-mediated nuclear XPO5 cellular RNA metabolism nuclear export open question HITS-CLIP analysis XPO5-associated RNAs in human embryonic kidney cells large number cellular RNAs pre-miRNAs noncoding structural RNAs bound by XPO5clustered miRNA precursors pri-mir-17~92-mir15b~16-2 show XPO5 HITS-CLIP signals outside pre-miRNA hairpins purified XPO5 to-mir-17~92~16-2 RanGTP-independent In vitro XPO5 enhances cleavage efficiency pri-mir-19a~16-2 DROSHA/DGCR8 microprocessor mouse development deletion leads to early embryonic lethality failed gastrulation day 7.5 conditional knockout) XPO5 in epithelial cells skin shows defects Dicer1 Dgcr8 cKO17–19 miRNA biogenesis ~90% global loss miRNAs Drosha/Dgcr8-independent mir-320-48419 insights XPO5 in miRNA biogenesis independent RanGTP binding nuclear export cellular RNAs as XPO5 binding partners basis regulatory roles XPO5 in miRNA biogenesis metabolism associates with primary pre-miRNA crosslinking immunoprecipitation sequencing structured miRNA precursors bound by DROSHA DGCR8 DICER120–22.identify XPO5-associated RNA species in human cells applied high-throughput sequencing RNA (HITS-CLIP) approach map RNA fragments bound by XPO5 in HEK293T cells XPO5 antibody recover RNA fragments immunoprecipitation Five HITS-CLIP libraries generated displayed consistent profiles XPO5-associated RNAs required detection RNA fragments in three five libraries for analysis 64.2 million reads generated mapped to 83,169 genomic peaks XPO5 with pre-miRNA hairpin mediates nuclear determined pre-miRNA hairpins recovered all 37 pre-miRNAs highly expressed miRNAs detected in XPO5 HITS-CLIP 96 out of top 107 miRNAs read number >200 detected in XPO5 HITS-CLIP XPO5-associated miRNA sequences spanned left arm top loop right arm regions association with pre-miRNA hairpins ~20% of XPO5 HITS-CLIP reads mapped to either arm >98% mature miRNA reads recovered mapped to left or right arm-CLIP reveals association XPO5 with primary pre-miRNA precursors Autoradiogram of 32P-labelled XPO5-RNA complexes treated different RNase resolved on PAGE gel Original data in Source Data file Pre-miRNAs expressed miRNAs in HEK293T cells detected in XPO5 HITS-CLIP location of small RNA-seq reads XPO5 HITS-CLIP reads relative miRNA hairpin shown in pie charts Light blue left blue right dark blue left light gray right arm dark gray entire Metagene analysis of miRNA sequences DROSHA DGCR8 DICER1 AGO2/3 XPO5 reveals recognized regions pri-miRNA sequences beyond pre-miRNA hairpin IGV tracks XPO5-CLIP reveal association XPO5 with pre-mir-31 Negative reads in IGV track indicates mapping minus strand reveal widespread association XPO5 with pri-mir-17~92 Blue bars black lines indicate location pre-miRNAs Data range left each trackXPO5-associated miRNA sequences with biogenesis DROSHA DGCR8 DICER1 AGO proteins downloaded HITS-CLIP (DROSHA PAR-CLIP (DICER1) AGO2/3-Seq datasets examined miRNA sequences component (Fig. 1d–h). aligned miRNA sequences from 5′ end 5p miRNAs miRBase release 2123 DROSHA DGCR8 HITS-CLIP showed similar profiles pre-miRNA hairpin regions flanking regions DROSHA recognizes lower stem basal junction outside pre-miRNA hairpin DGCR8 recognizes upper stem DROSHA-associated miRNA sequences showed prominent signals sequences upstream pre-miRNA hairpin than DGCR8 (Fig. 1d DICER1 PAR-CLIP pre-miRNA hairpin not lower stem regions outside hairpin DICER1 DROSHA/DGCR8 AGO2/3-Seq showed profile mature miRNAs (Fig. derived from left right arm miRNA hairpin XPO5 HITS-CLIP showed association XPO5 with pre-miRNA hairpin regionsComparing XPO5-associated miRNA sequences to DICER1 DGCR8 regions similar to DICER1 within pre-miRNA hairpins resembled DGCR8 regions outside (Fig. 1e f XPO5 miRNA precursors outside pre-miRNA hairpin examined XPO5 HITS-CLIP data used one IGV track reads another unique reads after duplicates (Fig. 1i–k 1a encoded by primary transcript ~50% encoded by polycistronic transcription encoded miRNAs mir-31 mir-21 mir-34a XPO5 HITS-CLIP signals restricted within pre-miRNA hairpins (Fig. 1i 1a many polycistronic miRNAs XPO5 HITS-CLIP reads covered primary miRNA transcripts mir-17~92 cluster most reads XPO5-CLIP reads covered pri-miRNA distinct from pre-miRNA precursors in DICER1 PAR-CLIP (Fig. 1j). Inter-pre-miRNA regions between mir-17 and mir-19b1 had abundant XPO5-associated reads (Fignoted two modes XPO5 binding to pri-mir-17~92 reads density XPO5 to pre-miRNA hairpins inter-pre-miRNA regions higher density mir-19a mir-20a expressed XPO5-associated RNA reads abundant inter-pre-miRNA regions many RNA reads span DROSHA cleavage sites 5′ 3′ ends pre-miRNAs association XPO5 with uncleaved pri-miRNAs inter-pre-miRNA region depleted RNA reads between mir-19b1 and mir-92a1 mir-17~19b1 pri-miRNA released from mir-17~92 pri-miRNA by CPSF3 ISY127 suggest XPO5 recognizes pri-mir-17~92 cluster after cleavage prior DROSHA/DGCR8-mediated processing association XPO5 pri-miRNAs-mir17~92 inspected other clustered pri-miRNAs mir-15a~16-1 mir-15b~16-2 expressed lower level than pri-mir-17~92 observed inter-pre-miRNA reads in XPO5 HITS-CLIP data distinct reads density inter-pre regionsXPO5 coverage in miRNA clusters hairpins mir-200b cluster>600 data suggest not all primary transcripts polycistronic miRNAs bind to XPO5 two miRNAs mir-3618 mir-1306 processed from 5′ region Dgcr8 mRNA28 expression mir-1306 higher than mir-3618 processing sequencing AGO2/3 mature XPO5-CLIP reads on mir-1306 pre-miRNA more abundant than mir-3618 pre-miRNA (Fig. specificity XPO5 HITS-CLIP data suggest two modes XPO5 association with miRNA precursors most associates with pre-miRNA hairpins some associates with primary miRNA transcripts inter-pre-miRNA regions.XPO5 binds to pri-miRNA precursors independently performed electrophoretic mobility shift assays in vitro purified human XPO5 and RanQ69L mutant deficient GTP from bacteria E.coli used vitro transcribed pre-mir-30a hairpin RNA confirm binding activity XPO5 RanQ69L proteins pre-miRNA XPO5 binds to pre-mir-30a hairpin RanGTP-dependentsingle-shifted band confirmed formation XPO5:RanGTP:pre-mir-30a complex pri-mir-17~92 structured RNA conformation base-pairing regions beyond pre-miRNA transcribed folded pri-mir-17~92 precursor tested binding XPO5 XPO5 to pri-mir-17~92 RanGTP-independent RanQ69L XPO5 pri-mir-17 no interference binding XPO5 size intensity super-shifted bands XPO5 changed RanQ69L complex contain RanQ69L observed super-shifted bands increasing doses XPO5 80-fold pri-mir-17~92 RNA suggest XPO5 molecules bind to pri-mir-17~92 precursor RanGTP-independent XPO5 HITS-CLIP signals pri-mir-17~92 2XPO5 binds to miRNA precursors RanGTP-independent binds pre-mir-30a RanGTP-dependent Black hairpin represents pre-mir-30a XPO5 coloured orange RanGTP purple XPO5 binds to pri-mir-17~92 without RanGTP Increasing XPO5 results in super shiftsXPO5 pri-mir-17~92 not affected by RanGTP binds pre-mir-19a RanGTP-dependent Black hairpin coloured orange RanGTP purple XPO5 binds pri-mir-19a RanGTP-independent Increasing XPO5 super shift Black hairpin-mir-19a coloured orange dissociation constant) between XPO5 pri-miR-19a calculated results binds truncated pri-mir-19a v1 RanGTP-independent Black hairpin coloured orange dissociation constant (Kd) between XPO5 pri-miR-19a truncation v1 calculated binding results bind to truncated pri-mir-19a v2 binding assays 1 nM substrates used Original data for a–e g Source Data file XPO5 miRNA focused on pri-mir-19a flanking sequences form extensive base-pairing outside 11 bp lower stem region DROSHA/DGCR8 sequences flanking mir-19a high XPO5 HITS-CLIP signals pre-mir-19a hairpin to XPO5 RanGTP-dependentpri-mir-19a to XPO5 RanGTP-independent similar pri-mir-17~92 results two pri-mir-19a:XPO5 complexes detected with increasing XPO5 (Fig. dissociation constant) pri-mir-19a XPO5 calculated 14.56 nM (Fig pre-mir-19a hairpin bind XPO5 RanGTP XPO5 binds lower stem regions pri-mir-19a generated two truncated pri-mir-19a mutants deleted basal stem region v1 v2 truncation v1 bound to XPO5 RanGTP-independent Kd 7.99 nM singly shifted band detected truncation v2 bound to XPO5 basal stem regions deleted RanGTP-independent binding-mir-19a XPO5 requires lower stem regions outside pre-miRNA hairpin.XPO5 facilitates microprocessor cleavage XPO5 binding-mir-19a performed DROSHA/DGCR8-mediated processing assay for-mir-19a two mutants in vitro DROSHA DGCR8 expression plasmids co-transfected into HEK293T cells microprocessor complex immunoprecipitationpurified Flag-XPO5 protein from transfected HEK293T cells Western blot silver stain DROSHA/DGCR8 microprocessor copurify with XPO5 Fig. 3a XPO5 microprocessor released pre-mir-19a hairpin transcript modest efficiency (Fig. 3a lane 3) preincubation pri-mir-19a with XPO5 enhanced processing efficiency improved processing reduction unprocessed pri-mir-19a twofold more accumulation pre-mir-19a XPO5 dosage-dependent (Fig. 3a lanes 4–6) XPO5 alter pri-mir-19a (Fig. 3a lane 7) microprocessor contamination 3XPO5 facilitates cleavage pri-mir-19a Preincubation enhances cleavage processing efficiency 3–6) XPO5 mutant 8–11) Preincubation XPO5 mutant enhances cleavage truncated pri-mir-19a XPO5 promote cleavage efficiency pri-mir-19a truncation v2 images three repeats each assay Original data panels Source Data file .XPO5 to pre-miRNA hairpin dependent RanGTP 2-nt 3′ overhang pre-miRNA6showed RanGTP-independent binding pri-miRNA XPO5 (Fig. 2b–e asked XPO5 residues 5′ 3′ ends pre-miRNA enhancing microprocessor cleavage generated XPO5 mutant with E445A S606A E711A R718A mutations residues hydrogen bonds between XPO5 HEAT repeats 5′ 3′ ends pre-mir-30a 3′ XPO5 interaction-miRNA independent XPO5 mutant DROSHA/DGCR8 processing (Fig. 3a lanes 8–11) XPO5 pri-mir-19a processing RanGTP-independent binding performed assays two truncated mutants pri-mir-19a truncation v1 bound by XPO5 efficiently cleaved by XPO5 (Fig. processing pri-mir-19a truncation v2 not XPO5 not enhanced XPO5 (Fig. data reveal RanGTP-independent binding pri-mir-19a XPO5 pre-miRNA promotes microprocessor cleavage promotes pri-miRNA processing findings turned pri-mir-15b~16-2 encodes two miRNAs extensive XPO5 association outside pre-miRNA hairpin Figsecondary structure pri-mir-15b~16-2 contains base-pairing regions outside pre-miRNA hairpin (Fig. HITS-CLIP pri-mir-15b XPO5 RanGTP-independent Super-shifted bands observed increasing dose XPO5 similar pri-mir-19a pri-mir-15b lower stem region XPO5 singly shifted complex Kd 9.57 nM (Fig. similar pri-mir-19a truncation v1 in vitro data reveal RanGTP-independent binding XPO5 pri-mir-17~92 pri-mir-15b~16-2 polycistronic miRNA precursors. 4XPO5 promotes primary miRNA processing Secondary structure pri-mir-15b~16-2 predicted RNAfold XPO5 binds pri-mir-15b~16-2 RanGTP-independent Increasing XPO5 super shift Black hairpin coloured orange Secondary structure pri-mir-15b predicted RNAfold XPO5 binds pri-mir-15b RanGTP-independent dissociation constant) between XPO5 pri-miR-15b calculated binding results Preincubation XPO5 enhances microprocessor cleavage pri-mir-15b~16-2Incubation XPO5 processing efficiency pri-mir-15b~16-2 (lanes 3–6) XPO5 mutant pri-mir-15b~16-2 (lanes 8–11) Preincubation XPO5 promotes cleavage efficiency pri-mir-15b three repeats each assay Original data b d f g Source Data file RanGTP-independent association pri-mir-15b~16-2 XPO5. 4b tested miRNAs processing assay processing efficiency pri-mir-15b~16-2 enhanced XPO5 (Fig. 4f lanes 3–6 4g lanes 3–5) unprocessed pri-mir-15b~16-2 reduced detected cropped pri-mir-15b~16-2 intermediates-mir-15b/16-2 accumulation increased XPO5 quadruple XPO5 mutant DROSHA/DGCR8 processing (Fig. 4f lanes 8–11) RanGTP-independent pri-mir-19a-mir-15b~16-2 XPO5 facilitates DROSHA/DGCR8 processing function XPO5 nuclear export binds cellular RNAs double-stranded noncoding RNAs XPO5 HITS-CLIP datasets (Fig. 5a Supplementary Data 2)noncoding RNAs SINE LINE elements vault RNAs 7SL RNA snRNA 7SK RNA tRNA hTR (Fig. 5b Fig 4) vtRNAs abundant XPO5-associated bound by miRNA biogenesis small RNAs AGO proteins. pri-mir-19a by XPO5 RanGTP-independent (Fig. vtRNAs complex with XPO5 super-shifted by XPO5 protein vtRNA XPO5 association 36.27 nM (Fig. weaker than pri-mir-19a pri-mir-15b weaker interaction less shorter dsRNA structures (Fig. 5e Fig. 5XPO5 associates with diverse cellular RNAs XPO5-associated HITS-CLIP tracks vtRNA1-1 hTR 7SL RNA1 association XPO5 noncoding RNA compared DROSHA DGCR8 DICER1 AGO2/3 Blue bar coding region arrow direction transcription Data range XPO5 binds to vtRNA1-1 without RanGTP 1 nM vtRNA1-1 substrate usedRed arrows indicate vtRNA1-1 with without XPO5 binding red bracket super-shifted with XPO5 molecule dissociation constant between XPO5 vtRNA1-1 calculated binding results secondary structure vtRNA1-1 potential coloured red Detection vtRNA 5S RNA from nuclear cytoplasmic RNAs wild XPO5 KO MEF cells northern blotting XPO5 HITS-CLIP reads align secondary structure hTR Red green arrows indicate 5′ 3′ XPO5-associated RNA regions Black lines outline XPO5-associated hTR regions RNA Folding analysis shows folding energy of repeat nonrepeat pre-miRNAs randomly shuttled control sequences middle line median vertical line data range hinges first third quartiles two Mann–Whitney test Original data f Source Data file effect XPO5 on biogenesis localization vtRNA generated XPO5 knockout MEF cells level cytoplasmic localization vtRNA altered XPO5 binding by XPO5 affect maturation accumulation localization vtRNAXPO5-associated RNAs form double-stranded regions XPO5 showed binding patterns core components DROSHA DGCR8 DICER1 in hTR 7SL RNA hTR interacts with XPO5 in three domains strongest interaction 5′ end hTR XPO5 HITS-CLIP reads different origin same RNA base-pairing examined XPO5-bound regions on hTR Three structured regions hTR template CR4/CR5 scaRNA domain determined template region forms most extensive dsRNA structures followed CR4/CR5 scaRNA domains XPO5-associated RNA reads enriched in double-stranded regions (P2a-P2b-J2b/3 template region followed CR4/CR5 scaRNA domain AGO interacts with small RNA fragments scaRNA domain 3′ end hTR not template region CR4/5 domain suggest XPO5 interacts with structured regions hTR 7SL RNA has secondary structures extensive base-pairing regions conserved through 7SL RNA interacts with XPO5 not other core components AGO-associated small RNA fragments not detectedfunction XPO5 independent miRNA biogenesis preference XPO5 base-paired RNAs performed RNA Folding energy analysis XPO5-associated RNAs classified RNAs repeat nonrepeat pre-miRNAs Fig. folding energy XPO5-assoicated pre-miRNAs lower randomly selected sequences indistinguishable pre-miRNA hairpins repeated nonrepeat associated regions XPO5 lower folding energy suggest XPO5 double-stranded regions cellular RNAs required for mouse embryonic examined function XPO5 mouse development identified knock-in conditionally targeted embryonic stem cell clone for generating XPO5 KO mouse Founder KI mice validated germline transmission targeted allele IRES-LacZ Neo cassette intron 7 exon 8 XPO5 allele mice bred to Actb-Flp mice remove IRES-LacZ Neo cassette allele bred to EIIA-Cre Krt14-Cre mice generate skin cKO XPO5 6XPO5 required for mouse embryonic development Loss XPO5 embryonic lethalityXPO5 KO embryos show defects E6.5 gastrulation images littermate ten control KO embryos-mount LacZ staining XPO5KI/+ XPO5KI shows expression XPO5 Depletion XPO5 mir-290 KO at E7 measured by qPCR Dysregulation genes at E7 E8.5 measured qPCR Oct4 Nanog downregulated KO E7 E8.5 Scale bar 250 μm c–e Data mean from three experiments. *P < 0.05; **P < ***P < 0.001 Student’s t test Original data f g h Source Data file XPO5KI/+ XPO5KO/+ breeding produced viable XPO5 null mice at birth XPO5 WT heterozygous mice obtained expected Mendelian ratio requirement XPO5 embryonic development XPO5 null embryos compromised development embryonic day 6.5 gastrulation germline layer formation null embryos smaller control littermates embryos failed gastrulation organogenesis by E8.5LacZ signals E7 XPO5KI embryos XPO5 expressed in embryonic cells (Fig. performed-time PCR analyses loss XPO5 mir-290 miRNA expressed in mouse ESCs early embryos E7 WT KO embryos (Fig. miRNA pathway differentiation ESCs cell types loss of Dicer1 Dgcr8 blocks ESC quantified ESC marker genes Nanog Oct4 differentiation markers T Tet1 qRT-PCR results strong accumulation Nanog Oct4 mRNAs in E7 XPO5 null embryos normal level T Tet1 (Fig. high level Oct4 persisted in E8.5 XPO5 null embryos compromised differentiation XPO5 required for miRNA biogenesis mouse embryonic development.XPO5 required for skin studied Dicer1 Dgcr8 Ago1/2 skin observed defects in hair morphogenesis loss miRNA pathway Dicer1 Dgcr8 cKO skin loses >95% miRNAs exceptions miRNAs-320-484 dependent Ago1/2 dKO skin loses ~80% miRNAs shows milderdevelopmental defects XPO5 cKO animals models generated K14-Cre/XPO5fl/fl mice (Fig. required skin development K14-Cre/XPO5fl mouse model-mount LacZ staining XPO5KI+ P2 P4 skin Neonatal mice XPO5 cKO controls P1.5 to P14 lethal gain weight Depletion mir-203 mir-205 epidermis qPCR mean s. three experiments < 0.001 Student’s t test Depletion XPO5 proteins confirmed western blot H&E staining reduced hair follicle formation stunted growth P2. Stunted hair growth evaginating hair germ Lef1 β4-integrin staining Reduced cell proliferation Ki67 staining normal epidermal differentiation Krt5 Krt1 staining Normal epidermal differentiation qPCR detection Krt1 Lor Krt5 Itgb4 mean s.d. three experiments < 0.05 Student’s t test Scale bar 1 cm c 100 μm b h j k 50 μm i Original data d–g l Source Data filedocumented expression XPO5 skin LacZ signals in XPO5KI/+ heterozygous mice XPO5 expressed in skin cell types postnatal day 2 4 P4 hair follicles differentiated root sheath shaft LacZ signals elevated cells (Fig XPO5 skin cKO animals observed compromised hair follicle development milder compared with Dicer1 Dgcr8 cKO XPO5 cKO animals showed neonatal lethality postnatal day 5 18 failed gain weight postnatally contrast to Ago1/2 dKO animals lose ~80% miRNAs survive adulthood despite loss hair qRT-PCR mir-203 mir-205 showed >90% depletion AGO2 protein depleted in XPO5 null skin mature miRNA accumulation stabilize AGO depletion miRNA expression XPO5 XPO5 skin morphogenesis examined neonatal skin observed reduced hair follicle formation stunted hair follicle growth in P2 skin (Fig. found evaginating hair germs towards epidermis characteristic Dicer1 Dgcr8 cKO epidermal cell proliferation reducedepidermal differentiation affected by Krt5 Krt1 staining basal spinous layer. 7k qRT-PCR analysis (Fig. studies provide genetic evidence requirement XPO5 for miRNA biogenesis skin development in mouse.XPO5 required for miRNA demonstrated requirement mouse embryonic tissue development characterized global miRNA expression levels in XPO5 KO skin small RNA sequencing. small between 20 23 nt showed strong depletion in XPO5 KO samples examined miRNA depletion weaker than Dicer1 Dgcr8 cKO skin average depletion ~90% depletion mature miRNA evident for highly expressed miRNAs 8d quantified depletion of five miRNAs from mir-17~92 mir-15~16 clusters intermediate level in skin strong depletion for mir-17-5p mir-18 mir-19a/b mir-20a mir-15a/b mir-16 loss of mir-19a/b strong requirement XPO5 for processing nuclear export validate compromised miRNA biogenesis effect XPO5 loss miRNA species performed northern blotting for mir-17observed loss mature mir-17 no strong changes pre-mir-17 (Fig. lack accumulation pre-miRNAs due unstable absence XPO5 Dicer1 cKO18 mir-320 mir-484 biogenesis independent Drosha/Dgcr8 unchanged XPO5 cKO skin (Fig. XPO5 required export pre-miRNA hairpin Drosha/Dgcr8 analyses reveal requirement XPO5 biogenesis majority miRNAs consistent defects cKO model XPO5 due loss miRNAs binding XPO5 cellular RNAs double-stranded regions. 8XPO5 required for miRNA biogenesis Small RNAs between 20 23 nt strong depletion XPO5 cKO skin samples Small reads 18 to 27 nt charted depletion miRNAs XPO5 cKO skin weaker Dicer1 Dgcr8 cKO skin middle line median vertical line data range first third quartiles two-sided Mann–Whitney test used Depletion mature miRNA reads XPO5 cKO skin evident highly expressed miRNAs Depletion mir-17-5p mir-18 mir-19a mir-20a except mir-92a XPO5 measured by qPCR.Depletion mir-15a XPO5 skin measured qPCR Depletion mir-17-5p confirmed northern blotting Unchanged expression mir-320-484 XPO5 measured qPCR Data mean three experiments **P < 0.01 ***P < 0.001 significant Student’s t test Original data e–h Source Data file determined XPO5-associated RNA species genomic scale XPO5 associates pre-miRNA precursors majority miRNAs (Fig. XPO5 binds precursors closely clustered polycistronic miRNAs mir-17~92 mir-15b~16-2 in vitro binding assays XPO5 binds miRNAs RanGTP-independent 2 4) mir-19a precursors site XPO5 base-pairing regions beyond pre-miRNA hairpin-dependent association-mediated nuclear-independent association miRNAs suggests unrecognized function XPO5 incubation XPO5 with-mir-19a~16-2 increased processing efficiency DROSHA/DGCR8 microprocessor evidence involvement XPO5 nuclear cleavage closely clustered polycistronic miRNAslack association XPO5 with monocistronic sparsely clustered miRNAs cleavage unique to closely clustered miRNAs with extensive dsRNA regions closely clustered miRNAs may use multiple mechanisms XPO5 binding biogenesis complex control production posttranscriptionally cleavage regulated by XPO5 future studies cell XPO5 associates with many cellular RNAs analyses to dsRNA regions XPO5 binds cellular RNAs vtRNA 7SL RNA snRNA 7SK RNA tRNA hTR SINE LINE diverse cellular localization Kd of XPO5 to pri-mir-19a variants vtRNA similar to Kd MDA5 dsRNA43 possible XPO5 multiple functions beyond miRNA biogenesis1 mammalian XPO5 binding to dsRNA binding proteins ILF3 PKR STAUFEN16 results provide molecular basis for findings foundation to investigate link between XPO5 cellular dsRNAs binding proteins demonstrated requirement of XPO5 for global miRNA biogenesis mouse development XPO5 KO in embryonic skin developmentsystems observed developmental defects Dicer1 Dgcr8 KO animals loss of Dicer1 Dgcr8 depletion miRNAs exceptions mir-320 mir-484 Dicer1-dependent Dgcr8 animals neonatal lethal died before one-week old KO Ago1 Ago2 leads 70–80% depletion miRNAs AGO proteins mice survived adulthood defects hair models set boundaries requirement XPO5 for miRNA biogenesis not depleted XPO5 miR-seq qPCR analyses KO XPO5 leads ~90% depletion miRNAs Dgcr8-independent miRNAs mir-320 mir-484 not affected loss XPO5 consistent with report XPO5-independent miRNA indicates link between DROSHA/DGCR8 pre-miRNA hairpin XPO5-mediated export XPO5 skin cKO animals lived longer than Dicer1 Dgcr8 counterparts shorter than Ago1/2 dKO other factors compensate loss XPO5 rescue developmental defects XPO5 required for miRNA biogenesis mouse development XPO5 binds non-miRNA substrates XPO5 KO may not due to loss miRNA biogenesis future investigationstransgenic breeding procedures approved Committees University of Colorado Boulder ethical regulations received approvals XPO5KI animal purchased ES cell clone KO mouse consortium EIIa-Cre mice Jackson laboratory mouse crossed with Flipase mice XPO5fl/fl mice XPO5KI XPO5fl/fl K14-Cre XPO5fl mice bred housed University of Colorado Boulder Embryos timed plug gestational age E0.5 experiments protocols IACUC cultureHEK293T cells ATCC cultured maintained DMEM 10% heat fetal bovine serum 1% penicillin/streptomycin 5% CO2 incubator 37 °C XPO5fl/fl MEF cells isolated from E14 XPO5fl/fl mice PierceTM Mouse Embryonic Fibroblast Isolation Kit XPO5 KO MEF cells generated Adeno-cre virus infection XPO5fl/fl MEF cells cultured maintained DMEM Primary Cell Isolation Kits 10% fetal bovine serum 1% penicillin/streptomycin 5% CO2 incubator 37 °CRNA purification mRNA miRNA qPCRTotal RNA extracted Trizol (Thermo Fisher 1 μg RNA cDNA Superscript III Reverse Transcriptase miRNA miScript II RT Kit (Qiagen cDNA Reactions manufacturer’s manual CFX384 real-time system Differences samples controls calculated 2−ΔΔC(t) method primers qPCR Supplementary Table 1.Immunostaining back skin phenotypes OCT sections fixed 4% PFA 10 min washed three times 5 min sections 2.5% NGS 2.5% NDS PBS antibodies proteins β4-integrin Lef1 K5 Ki67 Imaging Leica DM5500B microscope Hamamatsu C10600-10B camera MetaMorph software single optical sections-gal histochemistry stainingThick sections (25 μm back skin samples fixed fix solution (0.5% glutaraldehyde 1.25 mM EGTA pH 7.3 2 mM MgCl2 1xPBS 10 min fixed fix solution 5 minsections embryos washed buffer pH 7.4 2 MgCl2 0.01% sodium deoxycholate 0.02% incubated staining solution NP-40 20 pH 7.5 1 mg/mL X-gal 37 °C 48 h data analyses performed HEK293T cells UV-crosslinked resuspended lysis buffer Supernatants incubated Anti-Exportin-5 antibody control IgG 1 h ice dynabeads added protein–RNA complexes 1 h 4 °C 5′ end RNA [γ-32P] ATP 5′ Ligation Adapter-NN linked 5′ end RNA room temperature 2 h incubated Nu-PAGE buffer 70 °C 10 min resolved 8% Nu-PAGE Bis-Tris gel transferred nitrocellulose membrane exposed Fuji film Bands protein–RNA excised digested proteinase K RNAs purified adenylated linker linked 3′ end purified RNAs cDNAs made Superscript III(Invitrogen 3′ adenylated linker primersseparated 15% urea polyacrylamide gel 100 purified Phusion High Fidelity polymerase PCR products separated 6% urea gel bands 100–150 purified CLIP libraries high sequencing RNA samples P2–P4 skin epidermis efficiency 3′ ligation resolved 15% PAGE-urea gels stained SybrGold region ligated miRNAs excised gel slices minced eluted HSCB (400 mM NaCl 25 mM Tris-HCl 7.5 0.1% SDS overnight 4 °C Nucleic acids precipitated 0.1 sodium acetate 2.5 volumes ethanol Pellets washed 70% ethanol dissolved 5′ ligation mix Samples heated 70 °C chilled enzyme added cDNA prepared Superscript III RT) 3′ linker-specific RT primer libraries PCR amplification resolved 8% native acrylamide gels Bands molecular weight isolated high-throughput sequencing Illumina HiSeq2000 HiSeq4000 XPO5 RanQ69L cloned pcDNA3.1 vectormir-17~92a pri-miRNA transcription plasmid-T7-pri cloned primers T7 promotor T7:pri-mir-17 pJMC101vector Primers plasmid construction Supplementary Table 1. HEK293FT transfection Mirus TransIT®-LT1 Transfection Reagent 2300 expression XPO5-His6-RanQ69L-XPO5-RanQ69L plasmids transformed E.coli strain TG1 XPO5 subcloned cells grown LB medium 2% ethanol 37 °C induction 400 μM IPTG 18 °C 16 h purified NTA-Ni2 beads His6-RanQ69L cells grown LB medium 37 °C induction 100 μM IPTG room temperature overnight purified NTA-Ni2 beads loaded 3Electrophoretic mobility shift assayCold mir-17~92a pri-miRNA-mir-30a v1~16-2 vtRNA1-1 transcripts MEGAshortscriptTM T7 Transcription Kit Thermo Fisher Scientific pJMC101-T7-pri-mir-17~92 template T7-pre-mir-30a PCR products purified by 5% PAGE gel 5′-phosphote removed Antarctic Phosphatase substrates end-labeled by [r-32P]-ATP T4 Polynucleotide Kinase miR17~92a miR15b-16-2 pri-miRNAs refolded buffer (66 mM HEPES 33 mM Tris 1 mM 100 mM NH4OAc 5 mM MgOAc 0.5% 0.1 mM EDTA at 60 °C 10 min cooled to 25 °C 25 min mir-17~92a mir-19a~16-2 pri-miRNAs vtRNA1-1 incubated with XPO5 protein 10 μl 25 °C 45 min Pre-mir-30a-mir-19a incubated with XPO5 protein RanQ69L.buffer 20 mM HEPES 7.3) 150 mM potassium acetate 2 mM magnesium acetate 0.05% NP-40 7 mM 2-mercaptoethanol 1.5 μg/mL poly dIdC 0.2% BSA 0.3 μg heparin added incubated 5 Samples analyzed electrophoresis 0.7% nondenaturing agarose gel~92a 4 °C 5% nondenaturing PAGE gel pre-mir-30a room temperature Gels dried detected vitro analysis pri-miRNA processing-CTP miR19a miR15b pri-miRNA transcripts MEGAshortscriptTM T7 Transcription Kit PCR products purified 5% PAGE gel-Drosha-FLAG-Dgcr8 pCDNA3-FLAG-XPO5_mutant plasmids transfected Human 293FT cells48 h culturing cells harvested opened hypotonic buffer (10 Tris-HCl NaCl 2 mM EDTA 0.5% Triton X-100 DGCR8 XPO5 mutant protein purified-FLAG®M2 Affinity Gel eluted 3X FLAG® Peptide miR15b-16-2 substrates refolded buffer mM HEPES 33 Tris 1 mM 100 mM NH40Ac 5 mM MgOAc 0.5% NP-40 0.1 mM EDTA 60 °C 10 min cooled 25 °C 25 min incubated XPO5 25 °C 25 min DROSHA DGCR8 complex incubated 37 °C 1 h buffer (100 mM Tris-HCl 7.6 500 mM KCl 1 mM EDTA 64 mM MgCl2) purified phenol/chloroform 5% PAGE denaturing gel Gels dried detected phosphorimagerblotNuclear cytoplasmic RNAs extracted MEF cells RNA Purification Kit 21000 Norgen Three micrograms cytoplasmic two nuclear RNA separated 5% transferred nylon N+ membrane ATP-labeled oligonucleotide probe sequences hybridized 42 °C overnight Total RNA extracted P4 back skin epidermis Trizol 15596-018 Thermo Fisher five micrograms total RNA separated 5% PAGE gels transferred nylon N+ membrane ATP-labelled probe hybridized 42 °C overnight Nonhybridized probes washed wash buffer SSC/0.2%SDS 20 min Signals detected analysisFASTQ files trimmed cutadapt (v1.8.3) sequences randomized 5′ 3′ NN bases appended FASTQ read ID Reads mapped human genome novoalign Reads > 25 nt aligned -l 25 -t 85 settings shorter reads aligned -l 20 -t 30 settings PCR duplicates removed UMI-Tools alignments MAPQ < 10 discarded overlapping alignments bedtoolsPeaks annotated overlaps UCSC table browser Peaks intersected with annotations 1 bp overlap miRNA tRNA snoRNA lincRNA utr3 utr5 cds ncRNA repeatMasker introns mitochondria retroelements pseudogenes intergenic Peaks classified first intersecting annotation Repeat regions peaks with repeatMasker annotations UCSC Peaks filtered in three of five libraries minimum read count 5. Libraries mapped to database pre-microRNA sequences (mirBase release 19 blastn.2.2 CLIP datasets downloaded from GEO ENCODE DICER1 DROSHA DGCR8 RNA analysisFASTQ files from DGCR8cko DICER1cko trimmed remove 3′ adapter sequence XPO5cko trimmed remove 3′ adapter sequence 5′ adapter randomized NN nucleotides reads aligned with Bowtie2 (v.2.1.0) to mouse genome local alignmentcustom annotation file built GENCODE v19 human M8 mouse Mt_rRNA lincRNA misc_RNA sm snRNA snoRNA vaultRNA tRNA miRBase microRNA annotations overlapping counted Htseq-count (v.0.6.0) minimum MAPQ 10. FASTQ files HEK293T cells trimmed 3′ adapter sequence 5′ adapter randomized NN nucleotides aligned Bowtie2 XPO5 HITS-CLIP small RNA-Seq reads aligned novoalign ( reads enumerated counting intersections pre-miRNA database miRbase.RNA secondary structure analysisRNA folding predictions Fig. 5j generated RNAFold web server Peaks > 200 nt excluded miRNA metagene plots generated HTSeq python library (v0.6.0) secondary structure hTR Fig. 5h adapted telomerase secondary structures noncoding RNAs Figs. S2 S4 S5 generated RNAFold server experiments standard deviation calculated Numbers animals phenotype study figures Student’s t test experiments two Mann–Whitney test Fig. 5jReporting research Nature Research Reporting Summary.Supplementary information Peer Review File Reporting Summary Supplementary Files Supplementary Data 1 2 3 | 50.8 | 1.265446 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.