Add files using upload-large-folder tool

.gitattributes

@@ -160,3 +160,21 @@ e083fae6e5e22d3e2a4ae06dbcb295df3700e37243dca63ca11dacfdb074619a/preprint/images
 e083fae6e5e22d3e2a4ae06dbcb295df3700e37243dca63ca11dacfdb074619a/preprint/images/Figure_3.tif filter=lfs diff=lfs merge=lfs -text
 e083fae6e5e22d3e2a4ae06dbcb295df3700e37243dca63ca11dacfdb074619a/preprint/images/Figure_6.tif filter=lfs diff=lfs merge=lfs -text
 ba3e5f15b7ab5215c8a0c3ad587f8294fe2d83ec05f39775525b7ce261c0cc99/preprint/images/Figure_5.tif filter=lfs diff=lfs merge=lfs -text
+ba3e5f15b7ab5215c8a0c3ad587f8294fe2d83ec05f39775525b7ce261c0cc99/preprint/images/Figure_3.tif filter=lfs diff=lfs merge=lfs -text
+ba3e5f15b7ab5215c8a0c3ad587f8294fe2d83ec05f39775525b7ce261c0cc99/preprint/images/Figure_2.tif filter=lfs diff=lfs merge=lfs -text
+ba3e5f15b7ab5215c8a0c3ad587f8294fe2d83ec05f39775525b7ce261c0cc99/preprint/images/Figure_7.tif filter=lfs diff=lfs merge=lfs -text
+ba3e5f15b7ab5215c8a0c3ad587f8294fe2d83ec05f39775525b7ce261c0cc99/preprint/images/Figure_6.tif filter=lfs diff=lfs merge=lfs -text
+ba3e5f15b7ab5215c8a0c3ad587f8294fe2d83ec05f39775525b7ce261c0cc99/preprint/images/Figure_4.tif filter=lfs diff=lfs merge=lfs -text
+ba3e5f15b7ab5215c8a0c3ad587f8294fe2d83ec05f39775525b7ce261c0cc99/preprint/images/Figure_1.tif filter=lfs diff=lfs merge=lfs -text
+fc0904622d1a9197d5f70db4aedec8e101d5b84440d49d5b4d038ce254db87d1/preprint/images/Figure_3.tif filter=lfs diff=lfs merge=lfs -text
+fc0904622d1a9197d5f70db4aedec8e101d5b84440d49d5b4d038ce254db87d1/preprint/images/Figure_1.tif filter=lfs diff=lfs merge=lfs -text
+fc0904622d1a9197d5f70db4aedec8e101d5b84440d49d5b4d038ce254db87d1/preprint/images/Figure_4.tif filter=lfs diff=lfs merge=lfs -text
+fc0904622d1a9197d5f70db4aedec8e101d5b84440d49d5b4d038ce254db87d1/preprint/images/Figure_5.tif filter=lfs diff=lfs merge=lfs -text
+fc0904622d1a9197d5f70db4aedec8e101d5b84440d49d5b4d038ce254db87d1/preprint/images/Figure_2.tif filter=lfs diff=lfs merge=lfs -text
+fc0904622d1a9197d5f70db4aedec8e101d5b84440d49d5b4d038ce254db87d1/preprint/images/Figure_7.tif filter=lfs diff=lfs merge=lfs -text
+fc0904622d1a9197d5f70db4aedec8e101d5b84440d49d5b4d038ce254db87d1/preprint/images/Figure_6.tif filter=lfs diff=lfs merge=lfs -text
+3b16c1a1cb0da25846c83bf4e800968ae2ecd44d56416ed9d713b378cf69173d/preprint/images/Figure_5.tif filter=lfs diff=lfs merge=lfs -text
+3b16c1a1cb0da25846c83bf4e800968ae2ecd44d56416ed9d713b378cf69173d/preprint/images/Figure_3.tif filter=lfs diff=lfs merge=lfs -text
+3b16c1a1cb0da25846c83bf4e800968ae2ecd44d56416ed9d713b378cf69173d/preprint/images/Figure_4.tif filter=lfs diff=lfs merge=lfs -text
+3b16c1a1cb0da25846c83bf4e800968ae2ecd44d56416ed9d713b378cf69173d/preprint/images/Figure_6.tif filter=lfs diff=lfs merge=lfs -text
+3b16c1a1cb0da25846c83bf4e800968ae2ecd44d56416ed9d713b378cf69173d/preprint/images/Figure_2.tif filter=lfs diff=lfs merge=lfs -text

00825d02408bba0138773a462d869d9c769f79db41ff235dc46dd168331a96e6/preprint/images_list.json

@@ -0,0 +1,66 @@
+[
+    {
+        "type": "image",
+        "img_path": "images/Figure_1.png",
+        "caption": "The workflow of the study. Four retrospective datasets were included in this study. Each whole slide image was preprocessed and tessellated into non-overlapping tiles of 512x512 pixels at 10x magnification. From these tiles, two random tiles of 224x224 pixels were extracted and inputted into the LGNet. The LGNet generated tile-level probabilities, which were then averaged to obtain slide-level probabilities for predicting glioma and PCNSL. The LGNet was constructed by assembling five well-trained individual classifiers at the output layer. The average probability outputs of these five individual classifiers were used as the prediction of the ensembled model LGNet. The performance of LGNet was developed and validated using an internal dataset, and further evaluated on two external retrospective datasets. Finally, the proof of concept dataset was used to assess the performance of LGNet in guiding neurosurgeons for decision-making during surgery.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_2.png",
+        "caption": "The diagnostic performance of the LGNet and pathologists on two external datasets. \u00a0A External cohort 1; B External cohort 2. LGNet, a deep learning model for distinguishing PCNSL from glioma; Pathologist 1, with no prior experience in intraoperative diagnosis; Pathologist 2, having five years of experience in intraoperative diagnosis; Pathologist 3, having up to ten years of experience in intraoperative diagnosis; Pathologist (unassisted), working without the aid of LGNet; Pathologist (assisted), assisted by LGNet; AUROC, the area under the receiver operating characteristic.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_3.png",
+        "caption": "The diagnostic performance of each pathologist was compared with and without the assistance of LGNet on the external cohort 1. The performance was measured using three parameters: Sensitivity (A), Specificity (B), AUROC (C). Pathologist 1, with no prior experience in intraoperative diagnosis; Pathologist 2, having five years of experience in intraoperative diagnosis; Pathologist 3, having up to ten years of experience in intraoperative diagnosis; Pathologist (unassisted), working without the aid of LGNet; Pathologist (assisted), assisted by LGNet; AUROC, the area under the receiver operating characteristic; NS, no significance.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_4.png",
+        "caption": "The workflow of the human-machine fusion strategies by combining pathologist\u2019s diagnosis (original diagnosis or modified diagnosis) with LGNet\u2019s prediction. Original diagnosis, the pathologist\u2019s diagnosis without the assistance of LGNet; Modified diagnosis, the pathologist\u2019s diagnosis with the assistance of LGNet; H-M fusion, human-machine fusion. Fusion strategy 1, the fusion from the LGNet\u2019s prediction and pathologist\u2019s original diagnosis; Fusion strategy 2, the fusion from the LGNet\u2019s prediction and pathologist\u2019s modified diagnosis",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_5.png",
+        "caption": "Fusion of the LGNet and pathologists on two external datasets. A External cohort 1; B External cohort 2. Pathologist 1, with no prior experience in intraoperative diagnosis; Pathologist 2, having five years of experience in intraoperative diagnosis; Pathologist 3, having up to ten years of experience in intraoperative diagnosis; Pathologist (unassisted), working without the aid of LGNet; Pathologist (assisted), assisted by LGNet; AUROC, the area under the receiver operating characteristic; Original diagnosis, the pathologist\u2019s diagnosis without the assistance of LGNet; Modified diagnosis, the pathologist\u2019s diagnosis with the aid of LGNet.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_6.png",
+        "caption": "The comparison of LGNet fusion prediction and pathologist prediction on the external cohort 1. The performance was measured using three parameters: Sensitivity (A), Specificity (B), AUROC (C). Pathologist 1, with no prior experience in intraoperative diagnosis; Pathologist 2, having five years of experience in intraoperative diagnosis; Pathologist 3, having up to ten years of experience in intraoperative diagnosis; Pathologist (unassisted), working without the aid of LGNet; Pathologist (assisted), assisted by LGNet; AUROC, the area under the receiver operating characteristic; Original diagnosis, the pathologist\u2019s diagnosis without the assistance of LGNet; Modified diagnosis, the pathologist\u2019s diagnosis with the aid of LGNet; NS, no significance.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_7.png",
+        "caption": "The LGNet model successfully predicted cases of lymphoma and glioma in Internal-cohort, External cohort 1, and External cohort 2. The histological images of the patients with PCNSL (A-C) and glioma (D-F) are shown in the left column. The heatmaps overlapped on the whole slide images (WSIs) in the middle column indicated the tissue tiles that LGNet predicted as PCNSL with a high score (reddish color) or as glioma with a low score (bluish color). The tiles with a high score for PCNSL were primarily localized in areas of perivascular cuffing of tumor cells, monomorphic nuclei, prominent nucleoli, scant cytoplasm, and poor cohesiveness (tiles at 10x magnification in the right column). Similarly, the tiles with a low score for glioma were more likely to be found in areas of fibrillary background, nuclear shape and size variation with hyperchromasia, and microvascular proliferation (tiles at 10x magnification in the right column). All results were reproducible and consistent, demonstrating the reliability and stability of the LGNet model.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_8.png",
+        "caption": "The proof-of-concept cohort was used to assess the performance of LGNet, pathologist, and the LGNet-pathologist combination. Pathologist A and B were involved in the evaluation process. Pathologist A was a pathologist without any experience in intraoperative diagnosis, whereas Pathologist B had up to 10 years of experience in this field. Two modes of evaluation were used: Pathologist (unassisted) and Pathologist (assisted), with the latter being aided by LGNet. The performance was measured using AUROC (area under the receiver operating characteristic). The LGNet-Pathologist A combination (L-PA) and LGNet-Pathologist B combination (L-PB) were also evaluated.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    }
+]

00825d02408bba0138773a462d869d9c769f79db41ff235dc46dd168331a96e6/preprint/preprint.md

@@ -0,0 +1,196 @@
+# Abstract
+
+Intraoperative differentiation of primary central nervous system lymphoma (PCNSL) and glioma is of great importance to decision-making for neurosurgeons. However, distinguishing these two diseases based on frozen sections presents a challenge for pathologists. Here, we aim to develop and validate a deep learning model (LGNet) that could accurately differentiate PCNSL from glioma on haematoxylin and eosin (H&E)-stained frozen whole-slide images. In this study, the LGNet was developed and validated to distinguish PCNSL from glioma on independent cohorts, and its performance was compared to that of three pathologists with varying levels of expertise. Additionally, a human-machine fusion approach was designed to consider the diagnostic results from both pathologist and LGNet, to improve the integrative diagnostic performance. A proof of concept study was further evaluated with an online pathological decision support platform. The LGNet achieved high area under the receiver operating characteristic curves (AUROCs) of 0·965 and 0·972 for discriminating PCNSL and glioma on the two external validation cohorts. Moreover, the LGNet outperformed the three pathologists, and assisted them in making the distinction. The diagnostic performance human-machine fusion was further improved using the human-machine fusion. Notably, the performance of LGNet was verified with the proof of concept cohort, and it was shown that the time-consumption of LGNet was significantly less than that of pathologists (<em>P</em> &lt; 0·001) in practical scenario. Also, the study demonstrated the association between histopathological characteristics and the LGNet’s prediction as derived from the logistic regression model. These findings suggest that the LGNet accurately and timely differentiates PCNSL from glioma based on frozen sections, and adds to the enhancement of pathologists’ diagnostic performance. Thus, our deep learning model LGNet has the application potential during intraoperative diagnosis.
+
+**Biological sciences/Cancer/Cancer imaging**  
+**Biological sciences/Cancer/CNS cancer**  
+deep learning  
+primary central nervous system lymphoma  
+glioma  
+human-machine fusion  
+proof-of concept study
+
+# Introduction
+
+Accurate intraoperative diagnosis is crucial for decision-making during tumor surgery. However, intraoperative discriminating between primary tumors of the central nervous system (CNS), such as primary CNS lymphoma (PCNSL) and glioma, has always been challenging<sup>1</sup>. These are two of the most common primary brain malignancies encountered during surgery, and establishing an accurate and timely diagnosis is a key principle in neuro-oncology because intraoperative treatment options for the two tumors are substantially different<sup>2–6</sup>.
+
+While certain distinctive histomorphology features can aid in differenatial diagnosis<sup>2–6</sup>, pathologists face challenges when attempting to distinguish between PCNSL and glioma, particularly on haematoxylin and eosin (H&E)-stained frozen sections<sup>7</sup>. Although neuroimaging provides valuable features on the distinction between the two tumors, but it cannot accurately distinguish them<sup>8</sup>. Pathologists lack access to immunohistochemical and molecular assays that can assist in differential diagnosis, and rely mainly on the interpretation of cytologic and histomorphological characteristics based on H&E frozen slides. Due to the time contraints of intraoperative diagnosis, pathologists are often pressured to make the diagnosis as quickly as possible<sup>9</sup>. This can result in an equivocal intraoperative diagnosis, which may impact the neurosurgeon's decision-making. Therefore, a tool that is both accessible and time-efficient is needed to accurately differentiate between PCNSL and glioma during operation.
+
+Deep learning has demonstrated potential in assisting with various aspects of tumor diagnosis, including histological classification<sup>10</sup>, molecular typing<sup>11</sup>, therapeutic efficacy assessment<sup>12</sup> and prognostic prediction<sup>13</sup> from H&E-stained formalin-fixed paraffin-embedded (FFPE) tissues slides. However, the application of deep learning to frozen samples is still limited and requires further exploration, with most studies focusing on H&E-stained FFPE samples. Histological artefacts present in frozen sections can hinder rapid diagnostic assessments during surgery<sup>14</sup>, but a deep-learning algorithm may improve the quality of whole-slide images (WSIs) from H&E-stained frozen sections, leading to more accurate tumor classification by pathologists<sup>15</sup>. Recent studies have also shown the ability of deep learning models to diagnose thyroid nodules<sup>16</sup> and determine the metastatic status of sentinel lymph nodes in breast cancer<sup>17</sup> from the conventional intraoperative frozen sections, highlighting the potential of frozen samples in developing deep learning models. Thus, we hypothesize a deep learning approach can facilitate the intraoperative diagnosis of brain tumors.
+
+Herein, we aimed to train and validate the LGNet deep learning model for accurately distinguishing between PCNSL and glioma using H&E-stained frozen WSIs. We also designed a human-machine fusion approach to improve the diagnostic performance by integrating the abilities of both LGNet and pathologists. Finally, we conducted a proof-of-concept study to simulate the real-world scenario of frozen diagnosis and assess the practicality of LGNet.
+
+# Results
+
+## 1. Study design and patient cohorts
+The overall workflow of the LGNet was described in Fig. 1. This study consisted of four retrospective cohorts, which are detailed in Supplementary Table 1.
+
+## 2. LGNet performance
+To develop a deep learning model that predicts PCNSL and glioma based on frozen tissue samples, we first trained ResNet50 on the internal dataset. We then assessed the performance of this model on both the internal and external cohorts. During five-fold cross-validation on the internal cohort, the AUROC ranged from 0.990 to 1.000, with a mean value of 0.992 at the patient level (Supplementary Table 2). On External cohort 1, LGNet achieved an AUROC of 0.965 (95% CI 0.95–0.98), a sensitivity of 0.857 (95% CI 0.73–0.94), and a specificity of 0.912 (95% CI 0.87–0.94) (Table 1 and Fig. 2). On External cohort 2, LGNet obtained an AUROC of 0.972 (95% CI 0.94-1.00), a sensitivity of 0.955 (95% CI 0.77-1.00), and a specificity of 0.868 (95% CI 0.83–0.90) (Table 1 and Fig. 2). Our results, as shown in Supplementary Table 3, revealed that there was no significant difference in AUROC with and without color normalization on External cohort 1 and External cohort 2. Furthermore, we evaluated the model's robustness on External cohort 2 by randomly selecting slides 10 times based on real-world PCNSL and glioma proportions¹⁸. The AUROCs were consistent, with an average of 0.972 (Supplementary Table 4). Taken together, our data suggest that the model is highly robust and can provide stable diagnostic performance even without color normalization or variations in slide selection.
+
+| Cohorts | Category | Predictive performance |  |  |  |  |  |
+| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
+|  |  | Sensitivity | *P* | Specificity | *P* | AUROC | *P* |
+| External cohort 1 | LGNet | 0.857<br>(0.73, 0.94) | NA | 0.912<br>(0.87, 0.94) | NA | 0.965<br>(0.95, 0.98) | NA |
+|  | Pathologist 1 | 0.510<br>(0.36, 0.66) | < 0.001 | 0.924<br>(0.88, 0.95) | 0·73 | 0.840<br>(0.78, 0.90) | < 0.001 |
+|  | Pathologist 2 | 0.612<br>(0.46, 0.75) | 0.008 | 0.984<br>(0.96, 1.00) | < 0·001 | 0.843<br>(0.78, 0.91) | < 0.001 |
+|  | Pathologist 3 | 0.735<br>(0.59, 0.85) | 0.18 | 0.980<br>(0.95, 0.99) | 0·001 | 0.922<br>(0.86, 0.98) | 0.09 |
+| External cohort 2 | LGNet | 0.955<br>(0.77, 1.00) | NA | 0.868<br>(0.83, 0.90) | NA | 0.972<br>(0.94, 1.00) | NA |
+|  | Pathologist 1 | 0.636<br>(0.41, 0.83) | 0.016 | 0.893<br>(0.86, 0.92) | 0·37 | 0.876<br>(0.81, 0.95) | 0.004 |
+|  | Pathologist 2 | 0.818<br>(0.60, 0.95) | 0.38 | 0.907<br>(0.87, 0.93) | 0·12 | 0.896<br>(0.82, 0.97) | 0.06 |
+|  | Pathologist 3 | 0.818<br>(0.60, 0.95) | 0.25 | 0.901<br>(0.87, 0.93) | 0·20 | 0.948<br>(0.90, 0.99) | 0.19 |
+
+Note: 95% confidence intervals are included in brackets; * indicates the comparison of the difference between each pathologist and the LGNet; AUROC, the area under the receiver operating characteristic; NA, not applicable; The data have been provided in the Source Data file.
+
+## 3. Performance comparison between LGNet and pathologist
+A pathologist reader study was conducted on the external cohorts to compare the diagnostic performance of LGNet and three pathologists with different years of experience in the intraoperative diagnosis of PCNSL and glioma. The pathologists reviewed H&E-stained frozen sections and made diagnoses based on the morphological features. As the pathologists increased in their years of experience, the diagnostic performance improved. On External cohort 1, LGNet had a significantly higher AUROC than pathologists 1 and 2 (P < 0.001) and showed a higher trend than pathologist 3 (P = 0.09) (Table 1). On External cohort 2, LGNet had a significantly better AUROC than pathologist 1 (P = 0.004) and also showed a higher trend than pathologists 2 (P = 0.06) and 3 (P = 0.19) (Table 1). On both external cohorts, LGNet had higher sensitivity than the three pathologists (Table 1).
+
+## 4. Performance comparison between LGNet-assisted pathologists and unassisted pathologists
+To evaluate the impact of LGNet on the diagnostic performance of pathologists, the pathologists re-reviewed H&E-stained frozen sections and re-diagnosed PCNSL and glioma after reviewing the predictive result of the LGNet. Our data showed that LGNet improved the sensitivity of each pathologist on both external cohorts (Fig. 3 A and Supplementary Fig. 1). The diagnostic performance of LGNet-assisted pathologists was found to be superior to that of pathologists alone. On External cohort 1, the AUROCs of LGNet-assisted pathologists 1, 2, and 3 were 0.900 (95% CI: 0.85–0.95), 0.919 (95% CI: 0.87–0.97), and 0.942 (95% CI: 0.89–0.99), respectively. LGNet-assisted pathologists had significantly higher AUROCs than their unassisted counterparts (P < 0.01) (Fig. 3). On External cohort 2, LGNet-assisted pathologists 1, 2, and 3 had AUROCs of 0.952 (95% CI: 0.92–0.98), 0.951 (95% CI: 0.92–0.99), and 0.977 (95% CI: 0.95-1.00), respectively. LGNet-assisted pathologists had higher AUROCs than pathologists alone (Supplementary Fig. 1). Notably, LGNet-assisted pathologist 1, who lacked intraoperative diagnostic expertise, had equivalent or higher AUROCs than pathologists 2 and 3, who have years of experience in diagnosing frozen slides.
+
+## 5. Human-machine fusion
+To investigate the effectiveness of LGNet in intraoperative diagnosis, we evaluated the human-machine fusion on two external cohorts. The original and modified fusion workflows are presented in Fig. 4, respectively. The original fusion of LGNet and pathologists 1, 2, and 3 on External cohort 1 yielded an AUROC of 0.960 (95% CI: 0.94–0.98; P = 0.47), 0.971 (95% CI: 0.95–0.99; P = 0.49), and 0.979 (95% CI: 0.96-1.00; P = 0.07), which was similar to or slightly higher than LGNet's (0.965) (Figs. 5 and 6). The modified fusion of LGNet and pathologists 1, 2, and 3 had AUROCs of 0.969 (95% CI: 0.95–0.99; P = 0.53), 0.973 (P = 0.25), and 0.978 (P = 0.06), exceeding LGNet (0.965). The modified fusion prediction outperformed the original fusion (0.969 vs. 0.960, P = 0.017; 0.973 vs. 0.971, P = 0.71; 0.978 vs. 0.979, P = 0.74) (Figs. 5 and 6). The similar results were observed in External cohort 2 (Fig. 5 B and Supplementary Fig. 2). Importantly, the modified fusion approach achieved a corresponding sensitivity of 0.909, 0.955, and 0.955 and specificity of 0.934, 0.959, and 0.953, respectively. Our data indicate that the modified human-machine fusion can further improve the performance of intraoperative differentiation between PCNSL and glioma.
+
+## 6. Association between histological characteristics and LGNet prediction of PCNSL
+To gain insights into how the deep learning model operates, we conducted a logistic regression analysis to examine the association between LGNet's prediction and the histopathological features. Our univariate analysis revealed that monomorphic nuclei, prominent nucleoli, scant cytoplasm and poorly cohesive were significantly associated with LGNet's PCNSL prediction on both external cohorts (All P < 0.001, Table 2 and Fig. 7).
+
+| Features | External cohort 1 |  | External cohort 2 |  |
+| :--- | :--- | :--- | :--- | :--- |
+|  | Univariate |  | Univariate |  |
+|  | OR (95% CI) | *P* | OR (95% CI) | *P* |
+| Perivascular cuffing of tumor cells | 16.71 (3.45, 80.88) | < 0.001 | 7.88E9 (0.00, +∞) | 1.00 |
+| Monomorphic nuclei | 22.60 (10.15, 50.34) | < 0.001 | 6.71 (3.27, 13.77) | < 0.001 |
+| Prominent nucleoli | 27.97 (10.82, 72.35) | < 0.001 | 11·66 (4.75, 28.60) | < 0.001 |
+| Scant cytoplasm | 11.80 (6.23, 22.36) | < 0.001 | 2.70 (1.54, 4.75) | 0.001 |
+| Poorly cohesive | 12.24 (6.47, 23.17) | < 0.001 | 12.63 (5.18, 30.76) | < 0.001 |
+| Apoptosis | 6.05E9 (0.00,+∞) | 1.00 | 8.13E9 (0.00, +∞) | 1.00 |
+| Fibrillary background | 0.22 (0.12, 0.40) | < 0.001 | 0.40 (0.22, 0.75) | 0.004 |
+| Variation in nuclear shape and size with accompanying hyperchromasia | 0.06 (0.03, 0.13) | < 0.001 | 0.18 (0.09, 0.37) | < 0.001 |
+| Microvascular proliferation | 0.19 (0.05, 0.82) | 0.026 | 0.43 (0.22, 0.84) | 0.013 |
+| Necrosis | 0.35 (0.08, 1.53) | 0.16 | 0.64 (0.26, 1.58) | 0.33 |
+
+Note: 95% confidence intervals are included in brackets. OR, odds ratio.
+
+## 7. Misdiagnosis from LGNet
+To gain a more comprehensive understanding of the deep learning model's performance, we analyzed the cases where LGNet misdiagnosed PCNSL and glioma. In External cohort 1, LGNet misdiagnosed a total of 29 slides, comprising 7 PCNSLs and 22 gliomas. In External cohort 2, LGNet misdiagnosed 49 slides, including 1 PCNSL and 48 gliomas (Supplementary Fig. 3). Among the 22 cases that LGNet misdiagnosed as PCNSL in External cohort 1, 4 (18.2%) and 3 (13.6%) had PCNSL-like features, including monomorphic nuclei and prominent nucleoli, respectively. In addition, 2 (4.2%) of the 48 cases misdiagnosed as PCNSL in External cohort 2 exhibited the feature of monomorphic nuclei (Supplementary Fig. 4 and Supplementary Table 5).
+
+## 8. Performance on the proof of concept study
+To assess the applicability of the LGNet in real-world clinical practice, a proof-of-concept study was conducted at our center. The process for predicting PCNSL and glioma using the deep learning model through the online pathological decision platform was demonstrated in the Video Record in the Supplement. The mean time taken by pathologists A and B to make the original diagnosis was 65.51 and 60.00 seconds, respectively, which was significantly longer than that taken by LGNet (20.06 seconds, P < 0.001, Supplementary Fig. 5). The LGNet demonstrated significantly higher AUROC than pathologist A (0.998 vs. 0.821, P = 0.005) and slightly higher AUROC than pathologist B (0.998 vs. 0.972, P = 0.251) (Table 3 and Fig. 8). When assisted by LGNet, pathologists A and B achieved an increased AUROC of 0.991 (95% CI: 0.98–1.01) and 0.991 (95% CI: 0.97–1.01), respectively, compared to their AUROC when working alone (P = 0.003 and P = 0.26) (Table 4 and Fig. 8). The combination of LGNet and pathologist (L-P) prediction had a slightly better AUROC than LGNet alone (1.000 vs. 0.998, P = 0.48; 1.000 vs. 0.998, P = 0.48) (Table 3 and Fig. 8), indicating that the L-P combination could further improve the performance of both the LGNet and pathologists working alone.
+
+| Category | Diagnostic metrics |  |  |  |  |  |
+| :--- | :--- | :--- | :--- | :--- | :--- | :--- |
+|  | Sensitivity (95% CI) | *P* | Specificity (95% CI) | *P* | AUROC | *P* |
+| LGNet | 0.857<br>(0.42,1.00) | NA | 0.984<br>(0.91,1.00) | NA | 0.998<br>(0.99,1.01) | NA |
+| Pathologist A | 0.571<br>(0.18,0.90) | 0.63 | 0.836<br>(0.72,0.92) | 0.012 | 0.821<br>(0.70,0.94) | 0.005 |
+| Pathologist B | 0.857<br>(0.42,1.00) | 1.00 | 0.967<br>(0.89,1.00) | 1.00 | 0.972<br>(0.92,1.02) | 0.25 |
+| L-PA Combination | 1.000<br>(0.59,1.00) | NA | 1.000<br>(0.94,1.00) | NA | 1.000<br>(1.00,1.00) | 0.48 |
+| L-PB Combination | 1.000<br>(0.59,1.00) | NA | 0.984<br>(0.91,1.00) | 1.00 | 1.000<br>(1.00,1.00) | 0.48 |
+
+Note: 95% confidence intervals are included in brackets. * indicates the comparison of the difference between the LGNet and the other categories (pathologists and L-P combination); L-P combination, the combination of LGNet and Pathologist with the assistance of LGNet; Pathologist A, the pathologist without any experience in intraoperative diagnosis; Pathologist B, the pathologist with up to 10 years of experience in intraoperative diagnosis; L-PA combination, the combination of LGNet and Pathologist A with the assistance of LGNet; L-PB combination, the combination of LGNet and Pathologist B with the assistance of LGNet; AUROC, the area under the receiver operating characteristic; NA, not applicable.
+
+| Performances | Pathologist A<br>(unassisted) | Pathologist A (assisted) | *P* | Pathologist B<br>(unassisted) | Pathologist B (assisted) | *P* |
+| :--- | :--- | :--- | :--- | :--- | :--- | :--- |
+| Sensitivity | 0.571<br>(0.18,0.90) | 1.000<br>(0.59,1.00) | 1.00 | 0.857<br>(0.42,1.00) | 1.000<br>(0.59,1.00) | NA |
+| Specificity | 0.836<br>(0.72,0.92) | 0.967<br>(0.89,1.00) | 1.00 | 0.967<br>(0.89,1.00) | 0.967<br>(0.89,1.00) | 1.00 |
+| AUROC | 0.821<br>(0.70,0.94) | 0.991<br>(0.98,1.01) | 0.003 | 0.972<br>(0.92,1.02) | 0.991<br>(0.97,1.01) | 0.26 |
+
+Note: 95% confidence intervals are included in brackets. * indicates the comparison of the difference between each pathologist both without and with assistance; Pathologist A, the pathologist without any experience in intraoperative diagnosis; Pathologist B, the pathologist with up to 10 years of experience in intraoperative diagnosis; Pathologist A (unassisted): Pathologist A without the assistance of LGNet; Pathologist A (assisted): Pathologist A with the assistance of LGNet; Pathologist B (unassisted): Pathologist B without the assistance of LGNet; Pathologist B (assisted): Pathologist B with the assistance of LGNet; AUROC, the area under the receiver operating characteristic; NA, not applicable.
+
+# Discussion
+
+In this study, we have developed an artificial intelligence approach to aid the intraoperative diagnosis of brain tumors using conventional frozen H&E slides. Our generated AI model, LGNet, provides considerable advantages in addressing challenges experienced by pathologists when differentiating PCNSL from glioma during operations. LGNet can predict PCNSL and glioma directly from conventional H&E-stained frozen WSIs, and its improved performance has been validated on two external cohorts. Our study offers a promising approach for the intraoperative diagnosis of brain tumors through the use of AI technology.
+
+Accurately detecting PCNSL in brain tumor patients can help prevent unnecessary surgical resection, thus reducing the risk of brain tissue damage and improving patient care. Surprisingly, our study found that model-assisted pathologists with no prior experience performed better than unassisted pathologists with five years of experience, and even performed as well as unassisted pathologists with ten years of experience. This indicates that LGNet can significantly enhance the performance of pathologists with no prior experience in diagnosing brain tumors during surgery. Thus, this approach can deliver expert-level intraoperative diagnosis where neuropathology resources are limited and improve diagnostic performance in resource-rich centers. Interestingly, our analysis of LGNet's misdiagnosed slides showed that 54 out of 78 slides were correctly diagnosed by all pathologists. Consistent with previous studies<sup>19</sup>, this suggests that combining deep learning algorithms with the expertise of pathologists can further improve diagnostic performance.
+
+To determine whether the deep learning model could improve diagnostic performance in time-limited intraoperative diagnosis, we utilized a human-machine fusion strategy that combined a deep learning model and a pathologist based on their prediction uncertainties, as described in our previously published study<sup>20</sup>. Our data showed that the modified fusion was a better strategy than the original fusion for distinguishing PCNSL from glioma during operation. The modified fusion, which combined LGNet and an inexperienced pathologist, significantly improved diagnostic performance. We further developed an online pathological decision support platform to assess the applicability of LGNet in real-world practice. The present study demonstrated that LGNet had robust performance in identifying PCNSL and could assist pathologists in timely and accurately improving their diagnostic performance. More importantly, we found that LGNet's performance without color normalization was comparable to that of LGNet with color normalization, saving time and maintaining the favorable diagnostic performance of the model. Overall, our study suggests that our model is suitable for application in intraoperative diagnosis within a limited time.
+
+Deep learning models are often considered as black boxes due to the lack of understanding of their underlying decision-making process<sup>21,22</sup>. To interpret the LGNet in our study, we first investigated the association between some well-known morphological characteristics and the LGNet’s prediction by constructing a logistic regression model. Although all ten characteristics we investigated have established connections with differentiating PCNSL from glioma, only three characteristics (the presence of monomorphic nuclei, prominent nucleoli, and scant cytoplasm) were significantly associated with the LGNet’s prediction for PCNSL through multivariable analysis. This method provides an indirect insight into the LGNet’s interpretability, thereby increasing pathologists’ confidence in the model's prediction. Our results could contribute to improving the transparency of deep neural networks before they are integrated into routine clinical workflows.
+
+Although our study provides promising findings, several limitations need to be addressed. Firstly, while the LGNet model achieved high AUROC, the model prediction process could still be accelerated. To achieve faster prediction, a light-weight version of LGNet could be developed based on model distillation techniques<sup>23,24</sup>. Secondly, the proof of concept and internal cohorts were both recruited from the same medical center, which may have led to biased results in the proof of concept study. Therefore, a prospective study from multiple medical centers will be conducted to further evaluate LGNet’s potential in real-world clinical scenarios. Thirdly, the online platform still needs to be refined before it can be deployed for real clinical use. Lastly, while the logistic regression model provides biological interpretability for the deep learning model, more advanced visual methods, such as high-resolution class activation mapping<sup>25</sup>, may be used to further understand the model.
+
+In summary, our study introduced the LGNet deep learning model for differentiating PCNSL from glioma on H&E-stained frozen WSIs, and demonstrated its outstanding performance on two external cohorts. The LGNet model outperformed board-certified pathologists, and could assist pathologists with varying experience in improving their diagnostic accuracy. More importantly, the human-machine fusion approach, particularly the modified fusion, could further enhance both LGNet and pathologists' performance. The results could be validated with the proof of concept cohort, providing guidance for neurosurgeons in making informed decisions for managing patients with different malignant brain tumors during surgery.
+
+# References
+
+1. Dolecek, T.A., Propp, J.M., Stroup, N.E. & Kruchko, C. CBTRUS statistical report: primary brain and central nervous system tumors diagnosed in the United States in 2005-2009. *Neuro Oncol* **14 Suppl 5**, v1-49 (2012).
+
+2. Calimeri, T., Steffanoni, S., Gagliardi, F., Chiara, A. & Ferreri, A.J.M. How we treat primary central nervous system lymphoma. *ESMO Open* **6**, 100213 (2021).
+
+3. Tom, M.C., et al. Management for Different Glioma Subtypes: Are All Low-Grade Gliomas Created Equal? *Am Soc Clin Oncol Educ Book* **39**, 133-145 (2019).
+
+4. Di Stefano, D., Scucchi, L.F., Cosentino, L., Bosman, C. & Vecchione, A. Intraoperative diagnosis of nervous system lesions. *Acta Cytol* **42**, 346-356 (1998).
+
+5. Yachnis, A.T. Intraoperative consultation for nervous system lesions. *Semin Diagn Pathol* **19**, 192-206 (2002).
+
+6. Plesec, T.P. & Prayson, R.A. Frozen section discrepancy in the evaluation of central nervous system tumors. *Arch Pathol Lab Med* **131**, 1532-1540 (2007).
+
+7. Tofte, K., Berger, C., Torp, S.H. & Solheim, O. The diagnostic properties of frozen sections in suspected intracranial tumors: A study of 578 consecutive cases. *Surg Neurol Int* **5**, 170 (2014).
+
+8. Toh, C.H., et al. Primary cerebral lymphoma and glioblastoma multiforme: differences in diffusion characteristics evaluated with diffusion tensor imaging. *AJNR Am J Neuroradiol* **29**, 471-475 (2008).
+
+9. Chen, Y., Anderson, K.R., Xu, J., Goldsmith, J.D. & Heher, Y.K. Frozen-Section Checklist Implementation Improves Quality and Patient Safety. *Am J Clin Pathol* **151**, 607-612 (2019).
+
+10. Kather, J.N., et al. Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer. *Nat Med* **25**, 1054-1056 (2019).
+
+11. Chen, C.L., et al. An annotation-free whole-slide training approach to pathological classification of lung cancer types using deep learning. *Nat Commun* **12**, 1193 (2021).
+
+12. Kuenzi, B.M., et al. Predicting Drug Response and Synergy Using a Deep Learning Model of Human Cancer Cells. *Cancer Cell* **38**, 672-684 e676 (2020).
+
+13. Courtiol, P., et al. Deep learning-based classification of mesothelioma improves prediction of patient outcome. *Nat Med* **25**, 1519-1525 (2019).
+
+14. Kang, M., et al. Intraoperative Frozen Cytology of Central Nervous System Neoplasms: An Ancillary Tool for Frozen Diagnosis. *J Pathol Transl Med* **53**, 104-111 (2019).
+
+15. Ozyoruk, K.B., et al. A deep-learning model for transforming the style of tissue images from cryosectioned to formalin-fixed and paraffin-embedded. *Nat Biomed Eng* **6**, 1407-1419 (2022).
+
+16. Li, Y., et al. Rule-based automatic diagnosis of thyroid nodules from intraoperative frozen sections using deep learning. *Artif Intell Med* **108**, 101918 (2020).
+
+17. Kim, Y.G., et al. Challenge for Diagnostic Assessment of Deep Learning Algorithm for Metastases Classification in Sentinel Lymph Nodes on Frozen Tissue Section Digital Slides in Women with Breast Cancer. *Cancer Res Treat* **52**, 1103-1111 (2020).
+
+18. Ostrom, Q.T., Cioffi, G., Waite, K., Kruchko, C. & Barnholtz-Sloan, J.S. CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States in 2014-2018. *Neuro Oncol* **23**, iii1-iii105 (2021).
+
+19. Steiner, D.F., et al. Impact of Deep Learning Assistance on the Histopathologic Review of Lymph Nodes for Metastatic Breast Cancer. *Am J Surg Pathol* **42**, 1636-1646 (2018).
+
+20. Zheng, X., et al. A deep learning model and human-machine fusion for prediction of EBV-associated gastric cancer from histopathology. *Nat Commun* **13**, 2790 (2022).
+
+21. Holzinger, A., Langs, G., Denk, H., Zatloukal, K. & Muller, H. Causability and explainability of artificial intelligence in medicine. *Wiley Interdiscip Rev Data Min Knowl Discov* **9**, e1312 (2019).
+
+22. van der Laak, J., Litjens, G. & Ciompi, F. Deep learning in histopathology: the path to the clinic. *Nat Med* **27**, 775-784 (2021).
+
+23. Xu, T.B. & Liu, C.L. Deep Neural Network Self-Distillation Exploiting Data Representation Invariance. *IEEE Trans Neural Netw Learn Syst* **33**, 257-269 (2022).
+
+24. Zhang, Y., et al. A Lightweight Fusion Distillation Network for Image Deblurring and Deraining. *Sensors (Basel)* **21** (2021).
+
+25. Shinde, S., Tupe-Waghmare, P., Chougule, T., Saini, J. & Ingalhalikar, M. Predictive and discriminative localization of pathology using high resolution class activation maps with CNNs. *PeerJ Comput Sci* **7**, e622 (2021).
+
+# Methods
+
+## Study participants
+To develop the deep learning model, we conducted a retrospective study using three independent cohorts of frozen section images from January 1, 2014, to August 31, 2021, including an internal cohort from Sun Yat-sen University Cancer Center and two external cohorts for Zhujiang Hosiptal and Nanfang Hospital of Southern Medical University and The First Affiliated Hospital of Sun Yat-sen University (External Cohorts 1 and 2). Furthermore, to evaluate the practicality of the LGNet in clinical settings, we recruited a proof of concept cohort from September 1, 2021, to March 1, 2022 in Sun Yat-sen University Cancer Center (see the eMethods for details). The criteria for all included and excluded patients are presented in the eMethods. All patient information was approved by the Institutional Ethics Committee, and informed consent was waived.
+
+## Slides scanning and WSIs preprocessing
+For each patient in the internal and external cohorts, we collected one or more representative H&E-stained frozen slides. The Aperio AT2 scanner (Leica Biosystems; Wetzlar, Germany) was used to obtain whole slide images (WSIs) at 40× magnification (0.25 μm/pixel) which were then stored in SVS format. The WSIs were tiled in non-overlapping 512x512 pixel windows using the openslide library. To generate two tiles of size 224×224 pixels as inputs to the model, we randomly selected two crops from each tile. The details regarding WSI processing were described in the previous study<sup>20</sup>.
+
+## LGNet development
+After the images were preprocessed, we trained the ensemble binary classifier LGNet on the internal cohort to accurately predict PCNSL and glioma. We randomly split the cohort into three sets: a training set, a validation set, and an internal test set. Importantly, there was no overlap of patients or slides between these sets. We utilized a five-fold cross-validation scheme to achieve slide-level or patient-level probability of LGNet. Subsequently, we dichotomized this probability into the final binary classification of patients as either PCNSL or glioma. More information about the development of the model can be found in the methods section of our previously published study<sup>20</sup>.
+
+## LGNet evaluation
+To assess the performance of our model LGNet, we conducted internal and external evaluations on the internal and external cohorts, respectively. For the internal evaluation, we divided the internal cohort into five folds, using four folds for training an ensemble classifier while the remaining fold served as the internal test set each time. To train an ensemble classifier using the four folds, the data was randomly divided into five new folds, with each fold being used to train an individual classifier as described above. The ensemble classifier model was then evaluated on the internal test set at both slide and patient levels. This process was repeated five times, with a different internal test set used each time, ensuring that each slide was only evaluated once on the internal dataset. For the external evaluation, we utilized the developed ensemble LGNet classifier, based on the entire internal dataset, to predict the classification probability of PCNSL or glioma at slide level. The predictive results from both internal and external datasets were then compared to the corresponding ground-truth tumor status.
+
+## Reader study
+To evaluate the effect of LGNet on the performance of pathologists, we recruited three pathologists with varying levels of experience in intraoperative diagnosis. Pathologist 1 had no prior experience, Pathologist 2 had approximately 5 years of experience, and Pathologist 3 had up to 10 years of experience. These pathologists were blinded to clinical information about the dataset, including the ratio of PCNSL and glioma, as well as the performance of LGNet. For each WSI from the external cohorts, the pathologists made a dichotomized prediction of either PCNSL or glioma for both the original diagnosis and the modified diagnosis. The former was defined as the initial evaluation by the pathologist alone, while the latter was defined as the pathologists' re-diagnosis after being provided with LGNet's prediction, including the predictive probability and binary classification as either PCNSL or glioma. Additionally, the pathologists provided 6-scale self-confidence scores for both the original and modified diagnoses, with scores ranging from ‘1’ to ‘6’ corresponding to ‘surely glioma’, ‘likely glioma’, ‘unsure, slightly suggestive of glioma’, ‘unsure, slightly suggestive of PCNSL’, ‘likely PCNSL’, and ‘surely PCNSL’. To better understand the association between specific histopathological characteristics and LGNet's predictions, we constructed a logistic regression model using frozen slides. The detailed description was shown in the eMethods.
+
+## Human-machine fusion
+To improve the diagnostic performance, we applied the human-machine fusion scheme, which is a simple extension of the fusion method originally developed in our previous study<sup>20</sup>. The description regarding human-machine fusion was presented in the eMethods. Human-machine fusion strategy is performed for all cases on external cohorts, but only cases in part would be selected to perform the human-machine fusion by pathologists on the proof of concept study. Therefore, we describe the human-machine fusion on the proof of concept study as LGNet-Pathologist combination (L-P combination) to differentiate it from human-machine fusion on external cohorts.
+
+## Evaluation of the model on the proof of concept study
+To simulate real-world frozen diagnosis scenarios for pathologists with varying levels of experience, we conducted a proof-of-concept study using 68 frozen slides suspected of either PCNSL or glioma from our center. Two pathologists participated in the study: Pathologist A, with no experience in intraoperative diagnosis, and Pathologist B, with 10 years of experience in intraoperative diagnosis. Both pathologists were blinded to the primary intraoperative diagnosis and final postoperative diagnosis. To facilitate their decision-making process and visually display their decisions, we designed and developed an online pathological decision support platform accessible only to intranet users. The pathologists viewed the original and unprocessed H&E slides from the proof-of-concept study and made diagnoses based on their selected strategies, such as human-machine fusion or non-human-machine fusion. We compared the time spent by each pathologist from opening the frozen section to making the original diagnosis with that of LGNet's prediction. Furthermore, we compared the performance of LGNet with that of the two pathologists, as well as compared the performance of the LGNet-Pathologist combination, or human-machine fusion, with that of LGNet.
+
+## Statistical analysis
+The clinicopathological data in the retrospective cohorts were analyzed using the T test, Chi-square test or variance analysis. To compare the area under the receiver operating characteristic curves (AUROCs) of LGNet and pathologists, Delong’s test was used. The cutoff threshold of LGNet’s ROC curve was determined by Youden’s J statistic to dichotomize LGNet’s probabilities into binary predictions. The McNemar test was used to compare the statistical differences in sensitivity and specificity. The Clopper-Pearson method was used to calculate 95% CIs. We considered a P value less than 0·05 as statistically significant. For statistical analysis, we used SPSS Statistics (version 20.0), Medcalc (version 15.2.2), and R (version 3.6.3). Python (version 3.9.6) and the deep learning platform PyTorch (version 1.9) were used for data pre-processing and model development.
+
+# Supplementary Files
+
+- [VideoRecorded.mp4](https://assets-eu.researchsquare.com/files/rs-2923081/v1/deacbdc155fc46fd65f83dd6.mp4)  
+  Supplementary Video Recorded
+
+- [Supplementaryinformation.docx](https://assets-eu.researchsquare.com/files/rs-2923081/v1/f5381d992e555ce25e5fcfe9.docx)

01373eaf0b10aef9dd59aac48cf8d9225a5fe60d718ad62b69a437df983b7125/preprint/images_list.json

@@ -0,0 +1,58 @@
+[
+    {
+        "type": "image",
+        "img_path": "images/Figure_1.png",
+        "caption": "Background and motivation for novel amide formation. A) Amides are ubiquitous in nature: autoimmune disease drug Avacopan and transcription factor hypoxia-inducible factors HIF-3\u03b1 (PDB-ID 7V7W) with natural ligand oleoylethanolamide. B) The classical approach to amide formation involves stoichiometric activation/coupling of carboxylic acid and amine. C) The novel multicomponent reaction amide synthesis involves an SN2 promoted C-C coupling reaction, followed by hydrolysis of the nitrilium ion. D) Frontier orbitals of the isocyanide to rationalize chameleonic behavior as a C-nucleophile and C-electrophile.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_2.png",
+        "caption": "Optimization of reaction parameter using high throughput experimentation. Scheme of the model reaction, parallel heating in a metal block, and stagged HPLC injections.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_3.png",
+        "caption": "Scope of the halide electrophile with adamantyl isocyanide as the fixed component. Several stick presentations of X-ray structures and their CCDC codes are given.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_4.png",
+        "caption": "Scope of the isocyanide nucleophile with methyl iodide as the fixed component.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_5.png",
+        "caption": "Mixed reaction examples of the isocyanide nucleophile with the alkyl halide.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_6.png",
+        "caption": "Late-stage diversification, scale-up and some follow-up chemistries.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_7.png",
+        "caption": "Proposed mechanism and evidence, RNC synthon, retrosynthetic valuation, and comparison with the classical amide coupling. A) Proposed mechanism and MS evidence of an imidoyl bromide\u00a0 intermediate (B). C) Most popular and efficient access to isocyanides by two different synthesis pathways. D) Comparison of the classical and the SN2-based amide formation and the amide carbanion isocyanide synthon. E) Venn diagram for chemical space analysis of the classical and SN2-based amide formation and its overlap.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    }
+]

01373eaf0b10aef9dd59aac48cf8d9225a5fe60d718ad62b69a437df983b7125/preprint/preprint.md

@@ -0,0 +1,124 @@
+# Abstract
+
+The nucleophilic substitution reaction (S<sub>N</sub>2) is one of the oldest, yet very useful organic transformations and has found widespread applications for the synthesis of drugs and natural products. Typically, cyanide, oxygen, nitrogen, sulfur, or phosphorous nucleophiles replace a halogen or sulfonyl ester leaving group to form a new bond between the nucleophile and the electrophile. Isocyanides display an unusual versatile chemistry based on their C-centered lone pair s and the C-centered p<sup>*</sup> frontier orbitals leading to radical, and multicomponent reactions. Surprisingly, the nucleophilic character of isocyanides has never been explored in S<sub>N</sub>2 reactions. We discovered that isocyanides react as versatile nucleophiles in S<sub>N</sub>2 reactions with alkyl halides in a general manner to afford highly substituted secondary amides by *in situ* hydrolysis of the intermediate nitrilium ion. The innovative 3-component reaction has a broad scope regarding the structures of the isocyanide and electrophile components, functional group compatibility, scalability, use for late-stage modification of a drug and synthesis of highly complex compounds otherwise not easily accessible from simple precursors. The chemical space of the new reaction is not only different but nearly doubled in size compared to the classical amid coupling. Significantly, the isocyanide nucleophile comprises an unusual Umpolung amide carbanion synthon R-NHC(=O), useful as an alternative to the classical amide coupling.
+
+[Physical sciences/Chemistry/Chemical synthesis/Synthetic chemistry methodology](/browse?subjectArea=Physical%20sciences%2FChemistry%2FChemical%20synthesis%2FSynthetic%20chemistry%20methodology)  
+[Physical sciences/Chemistry/Green chemistry/Sustainability](/browse?subjectArea=Physical%20sciences%2FChemistry%2FGreen%20chemistry%2FSustainability)
+
+# Introduction
+
+The amide bond is one of the most common functional and structural elements, as the backbones of all natural peptides and proteins and almost every second drug are composed of amide bonds. Thus, construction of amide bonds is fundamental to organic synthesis as it provides access to the backbone of pharmaceuticals, agrochemicals, natural products, peptides and proteins and functional materials (Fig.1A). Recent data mining charting the amine–acid cross-coupling space revealed that there are numerous opportunities for reaction discovery.  
+<sup>1</sup> Nonetheless, the great majority of amide groups are formed by coupling of the an nucleophilic amine and an electrophilic carboxylic acid building block. Notably, the direct coupling of amines and carboxylic acid is unfavorable, hence requires an activated esters which is formed by aggressive, expensive or waste-full reagents (Fig.1B).  
+<sup>2</sup> Far less common ways to form amides involve for example oxidative and radical-based methods or the alkylation of nitriles to yield nitrilium ions with <em>in situ</em> hydrolysis.  
+<sup>3-4</sup> Based on our longstanding interest in isocyanide chemistry we asked ourselves to which degree the known boundaries of isocyanide nucleophilicity can be pushed to exploit new and synthetically useful reactivities. The isocyanide has both, a s-type <em>C</em>-centered HOMO and a <em>C</em>-centered p-type LUMO which accounts for the unusual reactivity of isocyanides.  
+<sup>5</sup> An example for this is its ability to act both as a nucleophile and an electrophile in the a-addition of a nucleophile and an electrophile onto the same functional group atom, the isocyanide-<em>C</em>, which is a quite unusual feature in organic chemistry, and accounts for many reactions of isocyanides such as multicomponent reactions and heterocycle syntheses.  
+<sup>6-8</sup> Due to their unusual <em>C</em>-only centered reactivity, isocyanides were also coined ‘stereochemical chameleons’.  
+<sup>9</sup>
+
+Isocyanides in multicomponent reaction chemistry (IMCR) is probably the most famous application. It involves the nucleophilic attack to the electrophilic oxo and imine-<em>C</em> and the subsequent addition of an internal or external nucleophile onto the nitrilium-<em>C</em>, followed by a rearrangement and is well established in the Passerini and Ugi reactions.  
+<sup>7, 10</sup> Motivated by the powerful IMCR which is based on the unusual reactivity of the isocyanide-<em>C</em> we designed a novel multicomponent reaction (Fig.1C). Our design involves an initial nucleophilic attack of the isocyanide on an alkyl halide in the sense of a nucleophilic substitution reaction (S<sub>N</sub>2); the intermediate nitrilium ion then reacts with water to give the stable amide species. Conceptually and experimentally the reaction design is not obvious and potential issues preventing the planned reaction outcome could involve i.a. insufficient isocyanide nucleophilicity, premature hydrolysis of the isocyanide or the alkyl halide.
+
+# Results
+
+## Reaction optimizations
+
+The S<sub>N</sub>2 reaction is known to be very sensitive to the substrate structures, as well as reaction conditions. <sup>11-13</sup> While initial attempts to run the new reaction were promising by masspectrometry analysis, the yields were far from being synthetically useful. Taking into account the established condition knowledge of the S<sub>N</sub>2 reaction we first interrogated stoichiometry and ratio of the reactants, temperature, temperature source, and solvents employing high throughput experimentation (HTE). <sup>14-15</sup> HTE methods used were parallel reactions in 96, 48, and 24-well format, parallel heating in a metal block, parallel TLC analytics, and stacked injection into SFC. The methods are described in more detail in the SI. We chose the model reaction of p-chloro benzyl isocyanide with benzyl bromide, a good electrophile in S<sub>N</sub>2 reactions and good visibility of educts and product in TLC (Fig.2). Next, we investigated the result of additives in the S<sub>N</sub>2 reactions. Biphasic phase transfer catalyst (PTC) were often used in S<sub>N</sub>2 reactions to increase yields and conversion. <sup>16</sup> We screened 16 different common PTCs (SI). The addition of iodine salts is often described as advantageous in the S<sub>N</sub>2 reactions as it converts the less reactive chloride leaving groups into the more reactive iodo leaving group. After thorough optimization of all parameters the optimized conditions involved the microwave heating at 105<sup>o</sup>C for 3 h of 1:2 ratio of isocyanide, alkyl halide, 20 mol% KI catalyst, and 1 equivalent of water in acetonitrile in the presence of 2 equivalent of the inorganic base K<sub>2</sub>CO<sub>3</sub> (Fig.2).
+
+## Scope and limitations
+
+The substrate scope for this reaction is very broad (Fig. 3-5). With the optimized conditions in hand, we interrogated the scope of the halide with respect to the leaving group, sterical bulkiness, electronic nature and diversity. Amongst the halide leaving group, chloride, bromide and iodide reacted well according to the well-established leaving group trend I>Br>Cl. To test the functional group tolerance, we successfully reacted 21 different alkyl halides with adamantyl isocyanide on a mmol scale (Fig.3). Adamantyl isocyanide is a solid, non-smelling, bench stable powder which has been synthesized recently on a mol scale. <sup>17</sup> A variety of alkyl halides with different functionalities were well tolerated. The small methyl group can be easily introduced (<strong>21a</strong>), whereas bulky alkyl groups or alkyl groups with b-branching do not react. Long chain alky groups can be introduced (<strong>17a</strong>), also with a terminal phthalic amide amine protecting group (<strong>12a</strong>), whereas Boc-protecting groups were found to be not stable under the microwave conditions (SI). For several alkylation products, single crystals revealed X-ray structures which support the structural identity (<strong>4a, 8a, 12a, 16a</strong>). Allyl (<strong>5a</strong>), and benzyl (<strong>1a, 3a, 4a, 6a, 13a</strong>) groups react well due to the conjugated nature of the pentagonal bipyramidal transition state as suggested by the classical S<sub>N</sub>2 literature. Specifically, to mention is bis benzylchloride derived <strong>13a</strong> which can be mono alkylated in 32% yield, and can be potentially further reacted through the unreacted benzylchloride. Also, the nature of the heterocyclic structures which could be reacted is quite diverse, including benzimidazole (<strong>7a</strong>), pyrazole (<strong>8a</strong>), triazole (<strong>9a</strong>), phthalimide (<strong>12a</strong>), coumarin (<strong>11a</strong>), thiophene (<strong>15a</strong>), and quinoline (<strong>19a</strong>). Especially to mention are <strong>15a</strong> and <strong>18a</strong>, which are formed from bifunctional (hetero)aromatic benzylchloride benzaldehydes. The aldehyde functionality can be further derivatized as will be shown below. We also found substrates which did not give the expected products or gave very low yields (<30%, SI).
+
+The evaluation of the isocyanides also revealed a broad scope (Fig.4). We reacted 20 different isocyanides with methyl iodide in satisfactory to good yields. Benzylic (<strong>23a</strong>, <strong>24a</strong>, <strong>25a</strong>, <strong>29a</strong>, <strong>30a</strong>), aromatic (<strong>31a</strong>, <strong>33a</strong>, <strong>34a</strong>, <strong>35a</strong>, <strong>36a</strong>, <strong>37a</strong>), aliphatic (<strong>27a</strong>, <strong>42a</strong>) and heteroaromatic (<strong>26a</strong>, <strong>28a</strong>, <strong>32a</strong>) isocyanides all worked well. When isocyanides with a basic side chain were reacted, we observed the double alkylation and a quaternary amine salt formation (<strong>38a</strong>, <strong>39a</strong>). Noteworthy, also a-amino acid isocyanides (<strong>40a</strong>, <strong>41a</strong>) worked well.
+
+We performed a number of mixed examples to further elaborate the scope and usefulness of the reaction (Fig.5). Highly substituted <strong>47a</strong> is especially noteworthy, as it comprises a combination of a sterically hindered a,a-disubstituted cyclopropyl benzyl isocyanide with a bifunctional 4-formylbenzyl chloride. The new method is also applicable to the facile synthesis of diverse lipid derivatives (<strong>56a</strong>, <strong>60a</strong>, <strong>61a</strong>) which could be of interest in lipidomics applications. Bulky isocyanides (<strong>47a</strong>, <strong>50a</strong>, <strong>54a</strong>) and phenyl isocyanides with bulky o-substitutents (<strong>43a</strong>, <strong>51a</strong>, <strong>52a</strong>, <strong>53a</strong>, <strong>55a</strong>) reacted nicely. Amide <strong>55a</strong> is accessible with a free compatible benzylic hydroxyl group. 4-Methylpentenoic acid (pyroterebic acid) ester or amides are common in biologically active isoprenoid compounds from plants. Compound <strong>54a</strong> is a pyroterebic acid amide and it comprises an unprecedented synthesis. Another example of incorporation of an isoprenoid side chain (homo geranyl acid) is exemplified in <strong>60a</strong>. It is conceivable that this methodology can be used to incorporate isotope labeled carboxy-<em>C</em> via the isocyanide. In summary, complex structures can be accessed from simple available building blocks in one step.
+
+## Scaling and late-stage functionalization
+
+To further stress the reaction performance, we evaluated the robustness of this reaction towards pharmaceutical late-stage diversification on an actual drug. <sup>18</sup> Late-stage-functionalization is a drug discovery technique to selectively derivatize already complex ‘drug-like’ molecules and is used to further improve their properties. <sup>18</sup> Phenoxybenzamine (dibenzyline) is an alpha blocker used for the treatment of hypertension. To establish the usefulness of our novel S<sub>N</sub>2 reaction we reacted dibenzyline with adamantyl isocyanide and were able to isolate the expected amide product in 40% yield (Fig.6).
+
+Having demonstrated a robust substrate scope for this novel isocyanide to amide transformation, we considered a variety of applications. First, we performed the thiophene carbaldehyde on a gram scale in fair yields. For this we reacted 1.59 gram of the bifunctional 5-(chloromethyl)thiophene-2-carbaldehyde with 1.61 gram adamantyl isocyanide on a 10 mmol scale (Fig. 6). The product <strong>15a</strong> could be isolated in 51% yield (1.57 gram). We envisioned that the aldehyde group can be further functionalized to create molecule of high complexity in just a few steps. To increase the complexity of the products we used multicomponent reactions (MCR) for further derivatization of the thiophene carbaldehyde. <sup>7, 10, 19</sup> The carbaldehyde <strong>15a</strong> is of interest to test further reactivity due to its unprotected aldehyde group based on the functional group compatibility of the reaction. Thus, we used <strong>15a</strong>, each in a Ugi-4CR, a Groebke Blackburn Bienaymé (GBB‐3CR) reaction, and a Ugi tetrazole reaction to exemplify rapid increase of molecular complexity (Fig. 6). The Ugi-4CR product <strong>1b</strong> was obtained in 72% yield in one step from easily available building blocks. Noteworthy, an alkynyl amide is introduced in a straight forward mild manner. Electrophilic alkynyl amides are often used in covalent drug discovery targeting cysteines and an alkynylamide substructure can be found in the FDA approved Acalabrutinib Bruton's tyrosine kinase targeting drug. <sup>20</sup> Next, we investigated aldehyde <strong>15a</strong> as a substrate in the GBB‐3CR reaction. The GBB-3CR is a popular method to synthesize highly substituted bicyclic imidazo heterocycles which already have proven their value as drugs and candidates. <sup>21</sup> Thus, we reacted 2-aminopyridine with aldehyde <strong>15a</strong> and cyclohexyl isocyanide in a GBB-3CR, under microwave conditions in methanol to obtain complex heterocycle <strong>1c</strong> in 36% yield. Lastly, we performed a Ugi tetrazol reaction employing aldehyde <strong>15a</strong>. Tetrazoles are often used as advantageous carboxylic acid bioisosteres, and can be broadly obtained by multicomponent reaction chemistry. <sup>19</sup>
+
+In summary, the new S<sub>N</sub>2 reaction turned out to be scalable, useful in late-stage-functionalization, and can yield highly interesting intermediates for allowing further chemistries to increase structural diversity in a quasi-exponential complexity increase, in just three steps: isocyanide synthesis, S<sub>N</sub>2 reaction, further aldehyde reaction.
+
+## Mechanism and chemical space
+
+Preliminary observations support a S<sub>N</sub>2-type mechanism (Fig.7A). Accordingly, the nucleophile isocyanide attacks from the backside to form a trigonal bipyramidal transitions state <strong>I</strong> and kicks out the leaving halogen anion. The intermediately formed nitrilium ion <strong>II</strong> undergoes water attack on the isocyanide-C <strong>III</strong>, and through tautomerization reveals the final amide <strong>IV</strong> upon hydrolysis.
+
+Several lines of evidence support a S<sub>N</sub>2 mechanism: sterically hindered substrates such as neopentyl iodide or isobutylbromide do not give any reaction product; the reaction is strongly solvent dependent and runs well in the polar solvent DMF which are believed to stabilize the transitions state, but not in apolar toluene or protic methanol; the reaction rate depends on the nature of the nucleofuge as reported in the S<sub>N</sub>2 literature I>Br>Cl (SI). To exclude a possible radical mechanism, we performed the reaction in the presence of 2x stoichiometric amounts of the radical quencher TEMPO, and did not find any difference in the reactivity (SI). <sup>22-23</sup> While running the reaction in the absence of water and direct injection in the mass spectrometer we could observe a strong peak corresponding to the bromo nitrilium ion (Fig.7B). In conclusion, there is strong evidence that the reactions run according to a S<sub>N</sub>2 mechanism. In the classical amide coupling approach the carbonyl is a carbocation synthon, while in the S<sub>N</sub>2 approach the rare amide carbanion synthon is the result of an Umpolung (Fig.7D). The isocyanide is commonly synthesized from its primary amine precursor (Ugi method: formylation > dehydration or Hoffman reaction). <sup>10, 24</sup> Alternatively, the isocyanide can be produced from an aldehyde or ketone precursor through reductive amidation with formamide (Leukart Wallach) and dehydration (Fig.7C). <sup>25-26</sup> Phenomenologically, the overall transformation of this S<sub>N</sub>2 reaction corresponds to a coupling of a primary amine with a C1 synthon derived from chloroform (Hoffmann) or formic acid (Ugi) with an alkyl halide or coupling of an aldehyde/ketone through a NC synthon derived from formamide (Leukart-Wallach) with an alkyl halide (Fig 7.C). Due to the large number of commercially available primary amines and aldehydes and ketones as isocyanide precursors, and alkyl halides, the reaction can be of considerable synthetic utility. Noteworthy, in classical S<sub>N</sub>2 reactions mostly very simple nucleophiles are used (such as halides, CN<sup>-</sup>, thio- or alcoholates), whereas the herein described S<sub>N</sub>2 reaction can make use of the great structural diversity of isocyanides (Fig.4). This is leading to a strong increase in structural complexity upon coupling with alkyl halides (e.g. <strong>60a</strong>). Next, we asked the question whether the new reaction can access a chemical space different from the classical amide coupling. For this we investigated the commercial availability of the corresponding carboxylic acid needed to form the target amides and compared them with the corresponding halide (SI). Surprisingly, in 52% cases the corresponding carboxylic acids were not commercially available at all. Noteworthy, in the remaining 48% the carboxylic acid was on average 2.3 times more expensive than the corresponding chloride. It turned out that the chemical space accessible by the two orthogonal amide syntheses is very different and only 12% are overlapping (i.e. can be synthesized by both methods). In conclusion, our herein reported novel S<sub>N</sub>2 reaction is of high synthetic value as it allows to access a chemical space which otherwise can only accessed through time-consuming and lengthy multistep syntheses and leads to a strong increase in molecular complexity, otherwise uncommon in S<sub>N</sub>2 reactions.
+
+# Discussion
+
+Arguably, the amide bond formation is amongst the most important reactions in organic chemistry. The value of the amide group in organic chemistry cannot be overstated. It is on top of the most frequent functional groups occurring in bioactive molecules described in medicinal chemistry literature. <sup>27</sup> More than 1/2 of the marketed drugs contain at least one amide group. Thus, the amide bond formation is the most practiced reaction in medicinal chemistry and one of the most frequently used in process chemistry. <sup>2</sup> While the classical amide coupling is a powerful reaction, there are many more hypothetical ways in which amides can be formed, with each new transformation imprinting a unique accessibility fingerprint on the product. Discovery of novel reactivities is key to leverage untapped chemical space and to broaden the tool box used in medicinal and other chemistries, <sup>28</sup> exemplified by a recently described copper-catalyzed deaminative esterification with broad scope. <sup>29</sup> The use of isocyanides in S<sub>N</sub>2 reactions is such an example of unprecedented reactivity giving access to unusual otherwise difficult to synthesize amides. Classically the amide group is constructed from a carboxylic acid derivative and a primary or secondary amine using specific activation conditions and a plethora of aggressive, expensive, and waste-full coupling reagents have been described. <sup>2</sup> Therefore, sustainable and alternative amidations have emerged as an important synthetic strategy to exploit the commercial and natural prevalence of the amide functional group, <sup>30-31</sup> leading us to consider the transformation of an isocyanide into an amide by alkylation and hydrolysis. The reaction was specifically designed to complement the popular amide coupling reaction.
+
+Here, we show for the first time that isocyanides can be alkylated by a S<sub>N</sub>2 mechanism through a nitrilium ion with concomitant hydrolysis to the corresponding amide. In this novel 3-component reaction, the isocyanide can be described as a Umpolung-derived rare amide carbanion synthon. The use of isocyanides as acyl anion equivalents provides a conceptually innovative approach to amide synthesis. As isocyanides are most commonly derived from either primary amines or aldehydes or ketones, the new reaction connects a primary amine via a formyl-C to an alkyl halide or in the second case an aldehyde or ketone carbonyl is connected through the formamide-C to an alkyl halide. The position of the amide group in the classical amide coupling of amines and carboxylic acids and the herein described isocyanide/alkylation derived amides are different. By repurposing halide building blocks as amides, instead of the classic amide, a subtle change in synthetic accessibility emerges. Attempting to synthesize the same molecules by the two orthogonal methods, large scale data analysis of educts reveals a great disadvantage of the classical method, since the required carboxylic acid building blocks are more difficult to access, not available at all, and more expensive than the corresponding alkyl halides by an average factor of 2.5. We also performed a survey of commercial availability of the required building blocks from a commercial vendor catalog (Fig. 7E), which revealed that 3.8 million amides are accessible by the classical method and 4.2 million by the new method, with only 1 million matched molecular pairs between the two sets. A chemoinformatic analysis of commercial building blocks demonstrates that by utilizing halides and primary amine-, aldehyde-, or ketone-derived isocyanides, our method more than doubles the available amidation chemical space. There is minimal overlap of chemical space compared to the classical amide coupling, demonstrating that a halide-isocyanide amidation can provide broad access to new and complementary structures. Repurposing of halide and isocyanide building blocks provides an enormous opportunity to expand the accessible chemical space or amides, because halide and amine or aldehyde/ketone feedstocks are typically low cost and available in high diversity. A halide-isocyanide amidation would therefore leverage the abundance of one popular building block and easily to access isocyanides from other popular building blocks. Collectively, these analyses quantify the value that a halide–isocyanide amidation would provide as an addition to the synthetic toolbox. High-throughput experimentation was used to develop the reaction, along with classic scope studies, both of which demonstrated robust performance against many pairs of reactants. The new reaction can be carried out under practical, mild conditions with yields ranging from good to moderate to poor, depending on the structure of the reactants. The functional group compatibility of the reaction is high. Late-stage functionalization of a drug is exercised. Alkyl halides are very frequently encountered in pharmaceutical research, so harnessing this functional group would also provide plenty opportunities for late-stage diversification. Upscaling of the reaction to gram scale have been shown. Complex otherwise difficult to access compound classes such a lipids, isoprenoids or functionalized amino acids can be synthesized by this method. Electrophilic alkyl halides are a cheap, abundant feedstock and are commercially available in high diversity, making them a valuable starting material for the amide synthesis. Similarly, isocyanides are accessible in two steps from abundant primary amines or ketones or aldehydes. In conclusion, we have successfully developed an efficient new way to form amides, by reacting isocyanides, alkyl halides, and water in a three-component fashion. Our innovative amide synthesis will be of potential use in the synthesis and discovery of novel bioactive molecules. It is beyond current amide bond forming methods and with its unusual synthon features, exponential complexity increase, and its wide scope, it will allow to scout novel chemical space hitherto inaccessible. <sup>32</sup> It is conceivable that other nucleophiles than water can also react with the intermediate-formed nitrilium ion in this new multicomponent reaction, and work is currently underway in our laboratory to broaden the range of nucleophiles.
+
+# References
+
+1. Mahjour, B.; Shen, Y.; Liu, W.; Cernak, T., A map of the amine–carboxylic acid coupling system. *Nature* **2020**, *580* (7801), 71–75.
+
+2. Magano, J., Large-Scale Amidations in Process Chemistry: Practical Considerations for Reagent Selection and Reaction Execution. *Organic Process Research & Development* **2022**, *26* (6), 1562–1689.
+
+3. de Figueiredo, R. M.; Suppo, J.-S.; Campagne, J.-M., Nonclassical Routes for Amide Bond Formation. *Chemical Reviews* **2016**, *116* (19), 12029–12122.
+
+4. van Dijk, T.; Chris Slootweg, J.; Lammertsma, K., Nitrilium ions – synthesis and applications. *Organic & Biomolecular Chemistry* **2017**, *15* (48), 10134–10144.
+
+5. La Pierre, H. S.; Arnold, J.; Bergman, R. G.; Toste, F. D., Carbon Monoxide, Isocyanide, and Nitrile Complexes of Cationic, d⁰ Vanadium Bisimides: π-Back Bonding Derived from the π Symmetry, Bonding Metal Bisimido Ligand Orbitals. *Inorganic Chemistry* **2012**, *51* (24), 13334–13344.
+
+6. Nenajdenko Valentine, G., *Isocyanide Chemistry: Applications in Synthesis and Material Science*. Wiley-VCH: 2012.
+
+7. Dömling, A.; Ugi, I., Multicomponent Reactions with Isocyanides. *Angewandte Chemie International Edition* **2000**, *39* (18), 3168–3210.
+
+8. Zhu, J., Recent Developments in the Isonitrile-Based Multicomponent Synthesis of Heterocycles. *European Journal of Organic Chemistry* **2003**, *2003* (7), 1133–1144.
+
+9. Gomes, G. d. P.; Loginova, Y.; Vatsadze, S. Z.; Alabugin, I. V., Isonitriles as Stereoelectronic Chameleons: The Donor–Acceptor Dichotomy in Radical Additions. *Journal of the American Chemical Society* **2018**, *140* (43), 14272–14288.
+
+10. Dömling, A.; Wang, W.; Wang, K., Chemistry and Biology Of Multicomponent Reactions. *Chemical Reviews* **2012**, *112* (6), 3083–3135.
+
+11. Ingold, C. K., *Structure and mechanism in organic chemistry*. **1953**.
+
+12. March, J., *Advanced organic chemistry*. **1968**.
+
+13. Hamlin, T. A.; Swart, M.; Bickelhaupt, F. M., Nucleophilic Substitution (Sₙ2): Dependence on Nucleophile, Leaving Group, Central Atom, Substituents, and Solvent. *ChemPhysChem* **2018**, *19* (11), 1315–1330.
+
+14. Krska, S. W.; DiRocco, D. A.; Dreher, S. D.; Shevlin, M., The Evolution of Chemical High-Throughput Experimentation To Address Challenging Problems in Pharmaceutical Synthesis. *Accounts of Chemical Research* **2017**, *50* (12), 2976–2985.
+
+15. Shevlin, M., Practical High-Throughput Experimentation for Chemists. *ACS Medicinal Chemistry Letters* **2017**, *8* (6), 601–607.
+
+16. Makosza, M., Phase-transfer catalysis. A general green methodology in organic synthesis. **2000**, *72* (7), 1399–1403.
+
+17. Patil, P.; Ahmadian-Moghaddam, M.; Dömling, A., Isocyanide 2.0. *Green Chemistry* **2020**, *22* (20), 6902–6911.
+
+18. Cernak, T.; Dykstra, K. D.; Tyagarajan, S.; Vachal, P.; Krska, S. W., The medicinal chemist's toolbox for late stage functionalization of drug-like molecules. *Chemical Society Reviews* **2016**, *45* (3), 546–76.
+
+19. Neochoritis, C. G.; Zhao, T.; Dömling, A., Tetrazoles via Multicomponent Reactions. *Chemical Reviews* **2019**, *119* (3), 1970–2042.
+
+20. Boike, L.; Henning, N. J.; Nomura, D. K., Advances in covalent drug discovery. *Nature Reviews Drug Discovery* **2022**, *21* (12), 881–898.
+
+21. Boltjes, A.; Dömling, A., The Groebke-Blackburn-Bienaymé Reaction. *European Journal of Organic Chemistry* **2019**, *2019* (42), 7007–7049.
+
+22. Chen, D.; Li, J.; Wang, X.; Shan, Y.; Huang, K.; Yan, X.; Qiu, G., Catalytic metal-enabled story of isocyanides for use as “C1N1” synthons in cyclization: beyond radical chemistry. *Organic Chemistry Frontiers* **2022**, *9* (15), 4209–4220.
+
+23. Zhang, B.; Studer, A., Recent advances in the synthesis of nitrogen heterocycles via radical cascade reactions using isonitriles as radical acceptors. *Chemical Society Reviews* **2015**, *44* (11), 3505–3521.
+
+24. Gulevich, A. V.; Zhdanko, A. G.; Orru, R. V. A.; Nenajdenko, V. G., Isocyanoacetate Derivatives: Synthesis, Reactivity, and Application. *Chemical Reviews* **2010**, *110* (9), 5235–5331.
+
+25. Neochoritis, C. G.; Zarganes-Tzitzikas, T.; Stotani, S.; Dömling, A.; Herdtweck, E.; Khoury, K.; Dömling, A., Leuckart–Wallach Route Toward Isocyanides and Some Applications. *ACS Combinatorial Science* **2015**, *17* (9), 493–499.
+
+26. Neochoritis, C. G.; Zhang, J.; Dömling, A., Leuckart–Wallach Approach to Sugar Isocyanides and Its IMCRs. *Synthesis* **2015**, *47* (16), 2407–2413.
+
+27. Ertl, P.; Altmann, E.; McKenna, J. M., The Most Common Functional Groups in Bioactive Molecules and How Their Popularity Has Evolved over Time. *Journal of Medicinal Chemistry* **2020**, *63* (15), 8408–8418.
+
+28. Boström, J.; Brown, D. G.; Young, R. J.; Keserü, G. M., Expanding the medicinal chemistry synthetic toolbox. *Nature Reviews Drug Discovery* **2018**, *17* (10), 709–727.
+
+29. Shen, Y.; Mahjour, B.; Cernak, T., Development of copper-catalyzed deaminative esterification using high-throughput experimentation. *Communications Chemistry* **2022**, *5* (1), 83.
+
+30. Santos, A. S.; Silva, A. M. S.; Marques, M. M. B., Sustainable Amidation Reactions – Recent Advances. *European Journal of Organic Chemistry* **2020**, *2020* (17), 2501–2516.
+
+31. Massolo, E.; Pirola, M.; Benaglia, M., Amide Bond Formation Strategies: Latest Advances on a Dateless Transformation. *European Journal of Organic Chemistry* **2020**, *2020* (30), 4641–4651.
+
+32. Brown, D. G.; Boström, J., Analysis of Past and Present Synthetic Methodologies on Medicinal Chemistry: Where Have All the New Reactions Gone? *Journal of Medicinal Chemistry* **2016**, *59* (10), 4443–4458.
+
+# Supplementary Files
+
+- [FinalSN2SI.docx](https://assets-eu.researchsquare.com/files/rs-2735988/v1/0db2255c28f29534c6c2ec5b.docx)  
+  SI

050529daf1daf78ef8cedaf4674a0be73a105c7021e9dcbcfb5618448dc10399/preprint/images_list.json

@@ -0,0 +1,18 @@
+[
+    {
+        "type": "image",
+        "img_path": "images/Figure_1.png",
+        "caption": "(a) Time variation of the band gap along the PBE NPT-F trajectory at 300 K for different supercells. Spatial band gap variation on 64x 96-atoms subcells contained in the 6144-atoms FAPbI3 supercell (b) and on 8x 96-atoms subcells contained in a 768-atoms supercell (c).",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_2.png",
+        "caption": "Time correlation function characterizing the rotational dynamics of FA cations along the N-N (blue curve) and C-H (orange curve) axes, as well as the octahedral tilting (red curve), and band gap oscillations (green curve). The analysis was done on the 6144-atoms supercell over an equilibrated MD trajectory of 5 ps at 300 K.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    }
+]

050529daf1daf78ef8cedaf4674a0be73a105c7021e9dcbcfb5618448dc10399/preprint/preprint.md

@@ -0,0 +1,221 @@
+# Abstract
+
+Formamidinium-lead-iodide (FAPbI₃) has established itself as the state of the art for high solar-energy conversion efficiency in perovskite-based solar cells. FAPbI₃ has a rich phase diagram, and it has been noted that long-range correlation between organic and lattice dipoles can influence phase transitions and, consequently, optoelectronic properties. In this regard, system size effects can play a crucial role for an appropriate theoretical description of FAPbI₃. In this context, we perform a systematic study on the structural and electronic properties of the photoactive phase of FAPbI₃ (α-FAPbI₃) as a function of system size. Utilizing ab initio molecular dynamics at 300 K and first-principles calculations, we demonstrate that the selection of the computational system/setup must satisfy three criteria concurrently to ensure an accurate theoretical description: the (correct) value of the band gap, the extent (or the absence of) structural distortions, and the zeroing out of the total dipole moment. We demonstrate that the net dipole moment vanishes as the system size increases due to PbI₆ octahedra distortions rather than due to FA⁺ rotations. Additionally, we show that thermal band gap fluctuations are predominantly correlated with octahedral tilting. The optimal agreement between simulation results and experimental properties for FAPbI₃ is only achieved by system sizes approaching the nanoscale.
+
+Physical sciences/Chemistry/Theoretical chemistry/Computational chemistry  
+Physical sciences/Physics/Condensed-matter physics/Structure of solids and liquids
+
+# Introduction
+
+Metal halide perovskites (MHP) are one of the most promising classes for the photoactive layer (AL) of photovoltaic (PV) materials.¹,² Thanks to their remarkable optoelectronic properties such as a high absorption coefficient, tunable bandgap, high charge carrier mobility, and low exciton binding energy, perovskite-based solar cells (PSCs) lead to high photo conversion efficiencies (PCE) and solar cell performance. Perovskites have the chemical formula ABX₃, where in the case of MHPs, A is an organic or inorganic cation, B is a divalent metal cation, and X is an anion of the halogen group. Efficient and stable PSCs require uniform and defect-free perovskite thin films at large areas, which can improve charge transport, suppress non-radiative energy loss, and minimize device degradation pathways.³,⁴ So far, among all MHPs, formamidinum-lead-iodide (FAPbI₃) in the cubic symmetry phase (α-FAPbI₃) yields the best PV performance. The reported experimental band gaps for this phase are in the range of 1.45–1.51 eV,⁵–⁷ which is close to the ideal single-junction Shockley-Queisser band gap of 1.31 eV.⁸ Besides, compared with other MHPs such as methylammonium lead iodide (MAPbI₃), FAPbI₃ exhibits an improved thermal stability due to higher activation energies for thermal degradation.⁹ These two factors help to establish FAPbI₃ as the most promising AL among the MHPs for single-junction PSCs with PCE record of 26.7%.¹⁰
+
+However, the pursuit of enhanced stability and performance in FAPbI₃ has prompted experimentalists to seek theoretical support to elucidate the pivotal mechanisms involved in surface passivation and charge transport, which are critical to guide the design of new PSCs.¹¹–¹⁴ In response to this need, theorists have developed surface and bulk models of FAPbI₃ with an appropriate level of accuracy that can mimic the electronic and structural features of FAPbI₃ films. Nevertheless, numerous other salient issues persist in their theoretical intricacy. These include phenomena such as nucleation from solution, crystal growth, halide segregation in mixed halide MHPs, and defect formation, which still need to be resolved at the atomistic level. To address these collective mechanisms, simulations are required that employ an adequate level of accuracy while permitting a substantial increase in the number of atoms. Because of the system size requirements, the type of simulation techniques used to simulate FAPbI₃ phases have been mostly in the framework of classical molecular dynamics (CMD) and Monte Carlo (MC) simulations based on empirical force fields,¹⁵–¹⁸ or density functional tight binding approaches.¹⁷,¹⁸ In this regard, also force-matched force fields and machine-learning potentials trained on ab initio molecular dynamics (AIMD) data are promising solutions because they allow CMD to be performed with AIMD accuracy, provided the physics is correctly described in the training data set.¹⁹–²¹ For the generation of sufficiently accurate reference data, it is evident that an adequate and appropriate description of FAPbI₃ at the quantum level is of paramount importance.
+
+In this context, there is an important amount of computational literature on FAPbI₃- and MHPs in general - where different periodic models and levels of theory are used, sometimes leading to contradictory results for fairly fundamental properties such as band structure and band gap,²²–²⁴ suggesting a lack of predictive ability. Several computational methods have been adopted to determine the band gap of FAPbI₃, spanning from density functional theory (DFT) with different generalized gradient approximations, meta functionals or (range-separated) (meta) hybrid functionals, van der Waals functionals, to the GW approximation of many-body perturbation theory, or combinations of them. In addition, spin-orbit coupling (SOC) is usually implemented in presence of Pb/Sn atoms, which leads to a significant lowering of the band gap. In the context of lead perovskites, the combination with hybrid functionals has been empirically demonstrated to effectively mitigate the impact of SOC on the band gap.²⁵ A fairly comprehensive summary of the band gap calculated for FAPbI₃, depending on the level of theory used, is provided in this review.²⁶ Apart from the employed theoretical method, the band gap can also be influenced by the choice of system size. In fact, part of the variations in the computed FAPbI₃ properties with different levels of theories might also be due to different choices of system sizes, and in particular, the use of small supercells, which imposes an ordered molecular orientation of ferroelectric multidomains, which is not observed experimentally.²⁷–³⁰ There are some examples in literature in which larger systems sizes are adopted obtaining interesting results. Carignano et al.³¹ utilize fairly large FAPbI₃ supercells to analyze the cation dynamics with AIMD. As demonstrated by Ma et al.,³² the electronic structure of MAPbI₃ must be studied at the nanoscale to ensure an accurate capture of its properties. A similar approach was adopted by Wiktor et al.³³ for CsMX₃ (M = Sn, Pb; X = Cl, Br, I). Evidence has also been found for electron-phonon coupling in FAPbI₃ as a cause of long-range optical phonon modes, indicating large size effects.³⁴. Using CMD, Maheshwari et al.³⁵ demonstrated that phase transitions in hybrid halide perovskites are driven by a complex interplay between dipole–dipole interactions between organic cations and the metal halide lattice, resulting in the formation of large organized domains of organic cations. This has significant implications for the electronic structures of these materials.
+
+In this work, we want to address the key questions: what are the system size and level of theory needed to properly capture the structural and electronic properties of FAPbI₃ and, is the static zero Kelvin description sufficient or are simulations at finite temperature required? These are crucial questions from both theoretical and experimental perspectives.
+
+Using AIMD at 300 K and first-principles calculations, we characterized the structural and electronic properties of α-FAPbI₃ with increasing size of the simulation cell at the DFT level with both the PBE and PBE0 functionals. SOC was also considered when allowed by memory and/or computational resources. Our simulations show that it is necessary to use sufficiently large simulation cells (at least a 768-atom cell) in which all degrees of freedom (atomic positions and simulation cell) are fully relaxed in order to obtain an non-distorted α-FAPbI₃ structure, a converged band gap and a small overall dipole moment. For 0 K calculations, it is also essential to set up an initial system configuration in which the FAs cations are randomly oriented according to the 3-fold symmetry (Supplementary Fig. 1). Furthermore, with large simulation cells (from 2592 atoms upwards), PBE is able to describe the electronic band gap in good agreement with the experimental value when calculated as thermal average over finite-temperature equilibrated AIMD snapshots. We also show that the net dipole moment of the system goes to zero as the cell size increases and this is related to long-range effects due to the PbI₆ octahedral tilting. Finally, we demonstrate that thermal band gap oscillations are mainly related to the octahedral tilting.
+
+# Results
+
+Characterization of α-FAPbI₃ at 0 K
+
+The obtained band gaps for different theoretical methods and computational schemes are reported in Table 1 as a function of system size ranging from the minimal primitive cell (12-atoms) to larger supercells containing increasing even numbers of primitive cells. The periodicity imposed by the size of the supercell has to be compatible with the periodicity in the octahedral tilting, while the 12-atoms cell that is not compatible with this condition (but is still an often used setup in FAPbI₃ calculations) was included as a reference. At 0 K, we ran two series of calculations using as the initial structure α-FAPbI₃ from the material project database. In the first case (relax), only the atomic positions were optimized keeping the simulation cell fixed to cubic, while in the second (vc-relax), both the atomic positions and the lattice parameters were optimized. We also set up two initial configurations: (i) all-aligned: FA molecules are kept aligned with their dipole moments pointing all in the same direction in order to have a meaningful comparison between the 12-atoms cell and all the other supercells; (ii) pseudo-random: FAs are pseudo-randomly oriented (from the 96-atom supercell upwards), where the overall orientation of FAs preserves the 3-fold symmetry resulting into a null total dipole moment of the system (Supplementary Fig. 1). In all cases, k-point sampling was increased until convergence was achieved.
+
+For both the relax and vc-relax calculations, the band gap converges with increasing k-point grid but to different final values, indicating that the relaxation of all the cell degrees of freedom is necessary. For all the simulation cells with dipole-aligned FAs the relax band gap converges to a value around 1.48 eV. At first glance, this is in close agreement with the experimentally measured value for the α-FAPbI₃ band gap and can be justified by the fact that, by keeping the simulation cell cubic, we are, in a sense, simulating the average (pseudocubic) α-FAPbI₃; however, with a closer look at the structure, we show that this is actually an artifact. Indeed, in addition to the correct band gap, it is necessary to verify that the cell distortions from the α-FAPbI₃ structure upon vc-relax and the total dipole moment of the supercell have also physical meaning. The vc-relax band gap of the 96-atom cell converges to a slightly higher value with respect to the 12-atom one. This is related to the significant distortion of the crystal structure from the cubic symmetry after vc-relax, decreasing the overlap of the electronic orbitals and consequently opening the band gap. In principle, this finding aligns with experimental observations that indicate the orthorhombic perovskite phase as the most stable at low temperatures. To quantify the deviations from cubic symmetry, we have analyzed the mean squared error (MSE) between the perfect α-FAPbI₃ structure and the vc-relax one as well as the distribution of the octahedra tilting angles after vc-relax for the two cases - with and without pseudo-random FA orientation (Table 2). In the pseudo-random FA orientation case the initial total dipole moment is smaller then 10⁻³ Debye. The pseudo-random orientation of the FAs in the starting configuration allows cubic symmetry to be maintained almost perfectly, as the MSE is reduced by two orders of magnitude for all supercells compared to the case with fully aligned FAs. The distributions of the octahedral tilting angles of the structures optimized from the pseudo-randomly oriented FA configurations average 0 degrees as expected (Supplementary Fig. 2) and the maximum value for the tilting angle converge with increasing system size (Table 2). The 96-atoms cell has a wide spread distribution because of the significant lattice distortions after vc-relax. In contrast, the structures optimized with all-aligned FAs do not present a octahedral tilting pattern. Local distortions of the octahedra due to a collective upward or downward motion of I ions compensate for the strongly directional dipole due to the dipole-aligned FAs (Supplementary Fig. 3). A comparison of all-aligned and pseudo-random configurations reveals that the vc-relax band gap is only improved for the 2592-atoms (3×3×3) cell. This phenomenon may be attributed to the fact that only this particular supercell allows for the complete relaxation of all degrees of freedom. This is because both the 3-fold symmetry for FAs and the octahedral tilting pattern are satisfied. It has been observed that, for all supercells, the potential energy with pseudo-randomly oriented FAs is consistently lower than that of the all-aligned case. This finding suggests that the pseudo-random configuration is a more favorable option for the system. Indeed, all-aligned configuration has a net total dipole moment for the FA that induces structural distortions in the Pb-I cages to compensate the overall dipole moment that cost energy.
+
+By adding SOC, the band gaps of the 12- and 96-atoms cells converge to slightly different values −0.45 eV and 0.54 eV again indicating a potential problem in the description of the electronic structure related to the cell distortions for the 96-atoms cell. The band gap of the 768-atoms cell with SOC, also converges to the same value as the 12-atoms cell, and this may be related to the conservation of cubic symmetry by the 768-atoms cell. Remarkable is what is obtained with the PBE0 functional. It is known that for elements such as Pb, SOC and PBE0 contributions should cancel each other out. The results of our 0 K calculations show that this effect only comes into play at k-point convergence for the 96-atoms cell, whereas it is entirely absent for the 12-atoms cell; by moving from the 12-atoms cell to the 96-atoms cell, the electron charge localization decreases as the band gap correction due to the inclusion of PBE0 decreases by 0.65 eV, allowing for SOC compensation and convergence to a band gap of 1.60 eV. It is evident that the size of the system plays a crucial role in reproducing the empirical findings of canceling out SOC-PBE0 contributions in the band gap and ensuring the structural properties of the system converge well.
+
+Characterization of α-FAPbI₃ at 300 K
+
+The thermally-averaged FAPbI₃ band gap was also calculated at 300 K by performing AIMD in the isothermal-isobaric ensemble with flexible cell option (NPT-F) at the PBE and PBE0 levels of theory for a minimum of 7 ps and up to 11 ps (Table 1). The band gap was computed as the average along the equilibrated AIMD trajectory (Fig. 1-a). We run in NPT-F to avoid any kind of symmetry restriction to the system. Since we are operating at 300 K, the initial orientation of the FAs is no longer important, as the kinetic energy is sufficient to randomize the FA orientation after a few AIMD steps. The finite temperature PBE band gap for the 6144-atoms cell is 1.47±0.08 eV that matches very well the experimentally measured gap, illustrating again the efficient compensation of many-body and SOC effects. Indeed, for all simulation cells, PBE0 (without SOC) consistently leads to a strong overestimation of the band gap.
+
+We also exploit the larger system sizes studied by AIMD together with its statistical approach to study the instantaneous band gap dependence from the local lattice distortions. We have subdivided the 6144-atoms and 768-atoms cells into 64 and 8 96-atoms cells, respectively, and computed the local band gap by projecting the density of states locally along the NPT-F trajectory (Fig. 1-b, 1-c). The 3D arrangement of the supercells was projected into a 2D map for better visualization (Supplementary Fig. 4). There is a local band gap variation for both supercells of about 1 eV, but in the 6144-atoms cell there are more defined band gap domains, while for the 768-atoms cell the band gap is more homogeneous throughout the supercell. This again indicates the importance of the system size, in order to detect the possible formation of band gap domains. It further indicates that this is a phenomenon that has to be related to the process of long-range relaxation, stressing once again the importance of correctly describing the mechanisms that we have previously shown to be related to the size of the system, such as the octahedral tilting. To quantify the connection between band gap fluctuations and octahedral tilting, we have calculated the time correlation function of the band gap oscillations and the octahedral tilting for the 6144-atoms cell (Fig. 2), getting correlation times of about 25 fs and 30 fs, respectively, i.e. very similar time scales suggesting that the band gap variations are indeed closely connected to changes in octahedral distortions. The influence of the tilting angles on the band gap directly aligns with established literature, as the tilting angles impact the antibonding overlap of the orbitals involved in the band edges. Furthermore, the FA cations motion can be characterized by two characteristic rotational correlation times of 80 fs and 250 fs for the C-H and N-N vectors, respectively (Fig. 2), that are farer from the value calculated for the band gap oscillations. We have also computed the time correlation functions of the bottom of the conduction band and top of the valence band eigenvalues obtaining a trend comparable with that of the band gap (Supplementary Fig. 5). Local variations in the band gap might indicate that FAPbI₃ can potentially absorb photons with different wavelengths in different regions of the sample, which may be a further explanation of why this material performs as well as AL. The electronic charge distribution due to lattice motions in halide perovskites have implications on thermal fluctuations in the electronic structure. It has been shown that off-centering of Sn²⁺ and Br⁻ widens the band gap of CsSnBr₃. Additionally, the top of the valence band and the bottom of the conduction band of FAPbI₃ are dominated by I⁻ and Pb²⁺ contributions, respectively. For these reasons, we can expect that the fluctuations in the band gap occur on time scales similar to those characterizing the octahedral tilting fluctuations. The selection of an appropriate system size is imperative for the accurate description of the octahedral tilting pattern. Our findings demonstrate that PbI₆ tilting converges for a 768-atoms supercell. The last point needed to verify the proper description of FAPbI₃ is the dipole moment.
+
+| Simulation cell | NPT-F PBE 300 K (eV) | NPT-F PBE0 300 K (eV) | relax PBE 0 K (eV) | vc-relax PBE 0 K (eV) | vc-relax PBE+SOC 0 K (eV) | vc-relax PBE0 0 K (eV) | vc-relax PBE0+SOC 0 K (eV) | $n\times n\times n$ k-point grid |
+|-----------------|----------------------|------------------------|---------------------|------------------------|----------------------------|-------------------------|----------------------------|----------------------------------|
+| 12-atoms $(1\times 1\times 1)$ | 3.62±0.21 | 5.08±0.21 | 3.68 | 3.72 | 3.46 | 5.82 | 5.54 | 1 |
+| 12-atoms $(1\times 1\times 1)$ | - | - | 2.21 | 2.31 | 1.78 | 3.90 | 3.33 | 2 |
+| 12-atoms $(1\times 1\times 1)$ | - | - | 1.48 | 1.97 | 1.25 | 3.47 | 2.72 | 4 |
+| 12-atoms $(1\times 1\times 1)$ | - | - | 1.49 | 1.63 | 0.49 | 3.44 | 2.28 | 6 |
+| 12-atoms $(1\times 1\times 1)$ | - | - | 1.49 | 1.56 | 0.45 | 3.38 | 2.25 | 8 |
+| 12-atoms $(1\times 1\times 1)$ | - | - | 1.48 | 1.55 | 0.45 | 3.37 | 2.24 | 10 |
+| 96-atoms $(2\times 2\times 2)$ | 2.16±0.17 | 2.80±0.15 | 2.24 | 2.26 | 1.78 | 3.36 | 2.85 | 1 |
+| 96-atoms $(2\times 2\times 2)$ | - | - | 1.58 | 1.69 | 0.60 | 2.84 | (1.75) | 2 |
+| 96-atoms $(2\times 2\times 2)$ | - | - | 1.49 | 1.58 | 0.46 | 2.73 | (1.61) | 4 |
+| 96-atoms $(2\times 2\times 2)$ | - | - | 1.48 | 1.61 | 0.54 | 2.75 | (1.68) | 6 |
+| 96-atoms $(2\times 2\times 2)$ | - | - | 1.61* | 1.60* | 0.48* | 2.76* | (1.64)* | 6 |
+| 768-atoms $(4\times 4\times 4)$ | 1.76±0.04 | 2.28±0.03 | 1.54 | 1.60 | 0.47 | - | - | 1 |
+| 768-atoms $(4\times 4\times 4)$ | - | - | 1.74* | 1.74* | 0.63* | - | - | 1 |
+| 768-atoms $(4\times 4\times 4)$ | - | - | 1.48 | 1.56 | 0.44 | - | - | 2 |
+| 768-atoms $(4\times 4\times 4)$ | - | - | 1.69* | 1.69* | 0.56* | - | - | 2 |
+| 2592-atoms $(6\times 6\times 6)$ | 1.59±0.07 | 2.17±0.03 | 1.51 | 1.71 | - | - | - | 1 |
+| 2592-atoms $(6\times 6\times 6)$ | - | - | 1.50* | 1.50* | - | - | - | 1 |
+| 6144-atoms $(8\times 8\times 8)$ | 1.47±0.08 | 2.09±0.05 | 1.47 | 1.67 | - | - | - | 1 |
+| 6144-atoms $(8\times 8\times 8)$ | - | - | 1.68* | 1.69* | - | - | - | 1 |
+
+*FA pseudo-randomly oriented
+
+## Dipole moment
+
+The dipole moment was computed at 0 K and at 300 K (Table 2). In principle, the total dipole moment should be zero; however, for small-size simulation systems, it is not, due to the presence of residual, not fully compensated dipoles. Given the substantial decrease in dipole moment observed with increasing system size, it can be deduced that this phenomenon is associated with a long-range collective interaction, such as the long-range dipole-dipole correlations present between FA-FA and octahedral cage dipoles. FA has a non-zero permanent dipole moment, while the octahedral distortions, i.e. contraction or elongation along different I-Pb-I axes, might induce a significant dipole moment. Table 3 reports the dipole moment contributions obtained by subtracting the dipole moment of the FA cations in the initial configuration from the total dipole moment of the system. The result is similar for all configurations (Table 3), and comparable with the values in Table 2, concluding that the initial FA configuration does not affect the overall dipole moment of the cell. To estimate the actual contributions of FA and octahedral cage distortions to the dipole moment, two NPT-F simulations were performed with FA respectively PbI₆ octahedra frozen. The octahedral distortion compensates for the dipole moment of the system in cases the FA cation orientations create a non-zero dipole moment (Supplementary Fig. 6). The time correlation function associated to the dipole moment fluctuations shows that the correlation time (∼ 30 fs) is not affected by the system size (Supplementary Fig. 8) at contrast to the absolute value of the dipole moment (Table 2). Except for the 12-atoms and 96-atoms cells, which have already shown large size effects for structural as well as electronic properties, there is no major change in dipole moment when using PBE0. Finally, the values of the dipole moment obtained with PBE and PBE0 are statistically equivalent for the 6144-atoms cell (Table 2). This finding suggests that the residual dipole moment may not be directly associated with the degree of charge localization imposed by the functional.
+
+| Simulation cell | MSE 0 K | Octahedra tilting amplitude 0 K (deg) | Total Dipole PBE 0 K (Debye/ABX₃) | Total Dipole PBE 3000 K (Debye/ABX₃) | Total Dipole PBE0 3000 K (Debye/ABX₃) |
+|-----------------|---------|----------------------------------------|------------------------------------|----------------------------------------|----------------------------------------|
+| 12-atoms $(1\times 1\times 1)$ | 0.08 | 1.37 | 13.17 | 2.32±0.98 | 1.80±0.70 |
+| 96-atoms $(2\times 2\times 2)$ | 0.78 | 18.03 | 3.36 | 2.24±1.02 | 4.57±0.71 |
+| 96-atoms $(2\times 2\times 2)$ | 0.09* | 17.36* | 3.25* | 2.24±1.02 | 4.57±0.71 |
+| 768-atoms $(4\times 4\times 4)$ | 0.30 | 5.65 | 1.18 | 0.92±0.26 | 0.89±0.25 |
+| 768-atoms $(4\times 4\times 4)$ | $4\bullet{10}^{-3}$* | 6.46* | 1.10* | 0.92±0.26 | 0.89±0.25 |
+| 2592-atoms $(6\times 6\times 6)$ | 0.23 | 5.18 | 0.39 | 0.39±0.12 | 0.31±0.09 |
+| 2592-atoms $(6\times 6\times 6)$ | $8\bullet{10}^{-4}$* | 5.67* | 0.36* | 0.39±0.12 | 0.31±0.09 |
+| 6144-atoms $(8\times 8\times 8)$ | 0.25 | 4.11 | 0.13 | 0.24±0.07 | 0.24±0.07 |
+| 6144-atoms $(8\times 8\times 8)$ | $1\bullet{10}^{-4}$* | 5.52 | 0.18* | 0.24±0.07 | 0.24±0.07 |
+
+*FA pseudo-randomly oriented
+
+| FA orientation | Total dipole (Debye/ABX₃) | FA dipole (Debye/ABX₃) | Total dipole – FA dipole (Debye/ABX₃) |
+|----------------|----------------------------|--------------------------|----------------------------------------|
+| all-aligned | 0.82 | 0.30 | 1.02 |
+| random | 1.00 | 0.12 | 1.08 |
+| random_best | 0.94 | 0.06 | 1.02 |
+| smart_100 | 1.05 | 0.00 | 1.05 |
+| smart_quasi | 1.16 | 0.00 | 1.16 |
+
+# Discussion
+
+In conclusion, by means of large-scale first-principles calculations and ab initio molecular dynamics simulations at 300 K, we demonstrated that in order to get an accurate description of the structural and electronic properties of the α-phase of FAPbI3 the size of the simulated system needs to approach the nanoscale. In particular, we showed that three conditions have to be met simultaneously, namely a proper description of the band gap, minimization of structural distortions, and the zeroing out of the total dipole moment. For first-principles calculations, it is essential to start from an initial configuration where the FAs are pseudo-randomly oriented by preserving the 3-fold symmetry and minimizing the dipole moment. At 300 K, because of the finite temperature dynamics, the initial configuration of the FAs is not stringent and, from a 2592-atoms cell upwards, the PBE approximation is already able to describe the electronic band gap of α-FAPbI3. For the 6144-atoms cell, we have computed a band gap of 1.47 ± 0.08 eV which is in excellent agreement with the experimental values of 1.45–1.51 eV reported in literature (highlighting that PBE0 and SOC corrections only cancel out for this system size range). The same cell minimizes structural distortions with respect to the perfect α-FAPbI3 structure and has the lowest dipole moment among all the systems studied. A significant correlation was discovered between PbI6 octahedral tilting, band gap oscillations, and dipole moment. In particular, the dipole moment goes to zero only if the system size is large enough to properly relax the tilting pattern of the octahedra. Overall, an adequate size of the system (at least 6144-atoms cell) is needed to correctly describe its physics, as we have demonstrated with the identification of band gap domains related to a correct description of the octahedral tilting. Our work provides a detailed insight into the connection between structural and electronic properties of α-FAPbI3 - and MHPs in general - making an important contribution to the field of ab initio simulations dedicated to understanding fundamental physical principles, such as hole-electron transport, which is of paramount importance in the development of increasingly high-performance PSC devices.
+
+# Methods
+
+## First-principles calculations
+
+DFT simulations at 0 K were performed with the Quantum ESPRESSO (QE) suite of codes. All the calculations were run with DOJO fully relativistic norm-conserving PBE pseudopotentials and well-converged basis sets corresponding to an energy cutoff of 150 Ry for the wave functions and 600 Ry for the charge density. Different k-point Monkhorst-Pack grids were used, all centered on Γ-point. Semiempirical corrections accounting for the van der Waals interactions were included with the DFT-D3 approach. Different simulation cells were used, starting from a 1×1×1 α-FAPbI₃ (12-atoms) up to a 8×8×8 (6144-atoms). The electronic structure of fully relaxed structures (vc-relax) was also computed including spin-orbit coupling (SOC) and PBE0. The supercell distortion with respect to the perfect cubic α-FAPbI₃ has been estimated by the mean squared error (MSE) between the lattice vectors of the two systems. Because CP2K allows to run also DFT at 0 K, Γ-point simulations were performed with both the QE and CP2K software, obtaining equal band gap values to three decimal places, which means that the results achieved with the two software packages are comparable.
+
+### Ab initio molecular dynamics
+
+AIMD simulations were run in the DFT framework as implemented in the CP2K software. The PBE and PBE0 functional and the D3 dispersion correction were adopted together with Goedecker-Teter-Hutter pseudopotentials and a polarized double-ζ Gaussian basis set (DZVP) for valence electrons. The energy cut off for the expansion of the electron density was set to 400 Ry. Simulations were run with a time step of 0.5 fs in the NPT flexible ensemble using Born-Oppenheimer dynamics for 7–12 ps (PBE) and 2–5 ps (PBE0), while the temperature was controlled by the Bussi thermostat and the pressure by the Martyna barostat. Different simulation cells were used, starting from a 1×1×1 α-phase FAPbI₃ (12-atoms) up to a 8×8×8 (6144-atoms). The finite temperature band gap was computed as an average of different band gaps calculated from the projected density of states (PDOS) on several AIMD snapshots after the system equilibrated (∼ 2 ps). The spatial variation of the band gap within a supercell was calculated by grouping the PDOS of the atoms of interest. Real space positions of top of the valence and bottom of the conduction bands in FAPbI₃ were computed after quenching an equilibrated AIMD snapshot to 0 K. Different initial FA orientation were tested - completely ordered, random oriented, smart oriented (total FA dipole equal to zero) - to avoid any bias on the simulations. The total dipole moment was calculated at the quantum level as in the CP2K framework, while the contribution of FAs to the dipole in the initial configurations was estimated classically, assigning a + 1 charge to each FA.
+
+## Time correlation function analysis
+
+The rotational dynamics of FA and PbI₆ octahedra were characterized by the correlation function
+
+$$
+C_{\text{rot}}(t) = \frac{\langle \vec{\mu}(t) \cdot \vec{\mu}(0) \rangle}{|\vec{\mu}(0)|}
+$$
+
+Where $\vec{\mu}(t)$ is the C-H(N-N) vector for FA or the octahedra tilting function appropriate for the PbI₆ tilting. The timescale oscillations for the band gap were quantified by the correlation function
+
+$$
+C_{\text{gap}}(t) = \frac{\langle \Delta \epsilon_{cv}(t) \cdot \Delta \epsilon_{cv}(0) \rangle}{|\Delta \epsilon_{cv}(0)|}
+$$
+
+where ∆ε<sub>cv</sub>(t) is the difference between the eigenvalues of the bottom of the conduction band and the top of the valence band. The same time correlation function has been used to compute the correlations of the dipole moment fluctuations, where the quantity correlated in time was the value of the dipole moment.
+
+# References
+
+1. Zhang, M. et al. High-Efficiency Rubidium-Incorporated Perovskite Solar Cells by Gas Quenching. *ACS Energy Lett.* 2, 438–444, DOI: 10.1021/acsenergylett.6b00697 (2017).
+
+2. Jeong, J. et al. Pseudo-halide anion engineering for α-FAPbI3 perovskite solar cells. *Nature* 592, 381–385, DOI: 10.1038/s41586-021-03406-5 (2021). Publisher: Nature Publishing Group.
+
+3. Liu, Z. et al. A holistic approach to interface stabilization for efficient perovskite solar modules with over 2,000-hour operational stability. *Energy* 5, 596–604, DOI: 10.1038/s41560-020-0653-2 (2020). Number: 8 Publisher: Nature Publishing Group.
+
+4. Cao, Y. et al. Quantum Chemistry in the Age of Quantum Computing. *Rev.* 119, 10856–10915, DOI: 10.1021/acs.chemrev.8b00803 (2019). Publisher: American Chemical Society.
+
+5. Min, H. et al. Efficient, stable solar cells by using inherent bandgap of α-phase formamidinium lead iodide. *Science* 366, 749–753, DOI: 10.1126/science.aay7044 (2019). Publisher: American Association for the Advancement of Science.
+
+6. Li, Z. et al. Stabilizing Perovskite Structures by Tuning Tolerance Factor: Formation of Formamidinium and Cesium Lead Iodide Solid-State Alloys. *Mater.* 28, 284–292, DOI: 10.1021/acs.chemmater.5b04107 (2016). Publisher: American Chemical Society.
+
+7. Eperon, G. E. et al. Formamidinium lead trihalide: a broadly tunable perovskite for efficient planar heterojunction solar cells. *Energy & Environ. Sci.* 7, 982–988, DOI: 10.1039/C3EE43822H (2014). Publisher: The Royal Society of Chemistry.
+
+8. Shockley, W. & Queisser, H. J. Detailed Balance Limit of Efficiency of p-n Junction Solar Cells. *Appl. Phys.* 32, 510–519, DOI: 10.1063/1.1736034 (1961).
+
+9. Chen, H. et al. Advances to High-Performance Black-Phase FAPbI3 Perovskite for Efficient and Stable Photovoltaics. *Small Struct.* 2, 2000130, DOI: 10.1002/sstr.202000130 (2021). _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/sstr.202000130.
+
+10. USTC Set New Record in Perovskite Cell Efficiency-University of Science and Technology of China.
+
+11. Han, D. et al. Tautomeric mixture coordination enables efficient lead-free perovskite LEDs. *Nature* 622, 493–498, DOI: 10.1038/s41586-023-06514-6 (2023).
+
+12. Li, L. et al. Buried-Interface Engineering Enables Efficient and 1960-Hour ISOS-L-2I Stable Inverted Perovskite Solar Cells. *Mater.* 36, 2303869, DOI: 10.1002/adma.202303869 (2024). _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/adma.202303869.
+
+13. Park, S. M. et al. Low-loss contacts on textured substrates for inverted perovskite solar cells. *Nature* 1–3, DOI: 10.1038/s41586-023-06745-7 (2023). Publisher: Nature Publishing Group.
+
+14. Jeong, J. et al. Carbazole Treated Waterproof Perovskite Films with Improved Solar Cell Performance. *Energy Mater.* n/a, 2401965, DOI: 10.1002/aenm.202401965 (2024). _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/aenm.202401965.
+
+15. Ahlawat, P. et al. A combined molecular dynamics and experimental study of two-step process enabling low-temperature formation of phase-pure α-FAPbI3. *Adv.* 7, eabe3326, DOI: 10.1126/sciadv.abe3326 (2021). Publisher: American Association for the Advancement of Science.
+
+16. Ahlawat, P. et al. Atomistic Mechanism of the Nucleation of Methylammonium Lead Iodide Perovskite from Solution. *Mater.* 32, 529–536, DOI: 10.1021/acs.chemmater.9b04259 (2020). Publisher: American Chemical Society.
+
+17. Shu, J. et al. Monte-Carlo simulations of optical efficiency in luminescent solar concentrators based on all-inorganic perovskite quantum dots. *B: Condens. Matter* 548, 53–57, DOI: 10.1016/j.physb.2018.08.021 (2018).
+
+18. Labrim, H. et al. Magnetic proprieties of La2FeCoO6 double perovskite: Monte Carlo study. *Alloy. Compd.* 641, 37–42, DOI: 10.1016/j.jallcom.2015.04.068 (2015).
+
+19. Fransson, E., Wiktor, J. & Erhart, P. Phase Transitions in Inorganic Halide Perovskites from Machine-Learned Potentials. *The J. Phys. Chem. C* 127, 13773–13781, DOI: 10.1021/acs.jpcc.3c01542 (2023). Publisher: American Chemical Society.
+
+20. Jinnouchi, R., Lahnsteiner, J., Karsai, F., Kresse, G. & Bokdam, M. Phase Transitions of Hybrid Perovskites Simulated by Machine-Learning Force Fields Trained on the Fly with Bayesian Inference. *Rev. Lett.* 122, 225701, DOI:10.1103/PhysRevLett.122.225701 (2019). Publisher: American Physical Society.
+
+21. Liang, Y. et al. Toward stabilization of formamidinium lead iodide perovskites by defect control and composition engineering. *Commun.* 15, 1707, DOI: 10.1038/s41467-024-46044-x (2024). Publisher: Nature Publishing Group.
+
+22. Liu, S. et al. Role of organic cation orientation in formamidine based perovskite materials. *Reports* 11, 20433, DOI:10.1038/s41598-021-99621-1 (2021).
+
+23. Wang, S., Xiao, W.-b. & Wang, F. Structural, electronic, and optical properties of cubic formamidinium lead iodide perovskite: a first-principles investigation. *RSC Adv.* 10, 32364–32369, DOI: 10.1039/D0RA06028C (2020).
+
+24. Zhao, H. et al. Recent Advances on the Strategies to Stabilize the α-Phase of Formamidinium Based Perovskite Materials. *Crystals* 12, 573, DOI: 10.3390/cryst12050573 (2022). Number: 5 Publisher: Multidisciplinary Digital Publishing Institute.
+
+25. Borlido, P. et al. Large-Scale Benchmark of Exchange–Correlation Functionals for the Determination of Electronic Band Gaps of Solids. *Chem. Theory Comput.* 15, 5069–5079, DOI: 10.1021/acs.jctc.9b00322 (2019).
+
+26. RaeisianAsl, M., Panahi, S. F. K. S., Jamaati, M. & Tafreshi, S. S. A review on theoretical studies of structural and optoelectronic properties of FA-based perovskite materials with a focus on FAPbI3. *J. Energy Res.* 46, 13117–13151, DOI: 10.1002/er.8008 (2022). _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/er.8008.
+
+27. Amat, A. et al. Cation-Induced Band-Gap Tuning in Organohalide Perovskites: Interplay of Spin–Orbit Coupling and Octahedra Tilting. *Nano Lett.* 14, 3608–3616, DOI: 10.1021/nl5012992 (2014). Publisher: American Chemical Society.
+
+28. Bokdam, M. et al. Role of Polar Phonons in the Photo Excited State of Metal Halide Perovskites. *Reports* 6, 28618, DOI: 10.1038/srep28618 (2016).
+
+29. Jono, R. & Segawa, H. Theoretical Study of the Band-gap Differences among Lead Triiodide Perovskite Materials: CsPbI3, MAPbI3, and FAPbI3. *Lett.* 48, 877–880, DOI: 10.1246/cl.190256 (2019).
+
+30. Ummadisingu, A., Meloni, S., Mattoni, A., Tress, W. & Grätzel, M. Crystal-Size-Induced Band Gap Tuning in Perovskite Films. *Angewandte Chemie Int. Ed.* 60, 21368–21376, DOI: 10.1002/anie.202106394 (2021). _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/anie.202106394.
+
+31. Carignano, M. A. et al. A close examination of the structure and dynamics of HC(NH2)2PbI3 by MD simulations and group theory. *Chem. Chem. Phys.* 18, 27109–27118, DOI: 10.1039/C6CP02917E (2016). Publisher: The Royal Society of Chemistry.
+
+32. Ma, J. & Wang, L.-W. Nanoscale Charge Localization Induced by Random Orientations of Organic Molecules in Hybrid Perovskite CH3NH3PbI3. *Nano Lett.* 15, 248–253, DOI: 10.1021/nl503494y (2015).
+
+33. Wiktor, J., Fransson, E., Kubicki, D. & Erhart, P. Quantifying Dynamic Tilting in Halide Perovskites: Chemical Trends and Local Correlations. *Mater.* 35, 6737–6744, DOI: 10.1021/acs.chemmater.3c00933 (2023). Publisher: American Chemical Society.
+
+34. Wright, A. D. et al. Electron–phonon coupling in hybrid lead halide perovskites. *Commun.* 7, 11755, DOI: 10.1038/ncomms11755 (2016). Publisher: Nature Publishing Group.
+
+35. Maheshwari, S., Fridriksson, M. B., Seal, S., Meyer, J. & Grozema, F. C. The Relation between Rotational Dynamics of the Organic Cation and Phase Transitions in Hybrid Halide Perovskites. *The J. Phys. Chem. C* 123, 14652–14661, DOI: 10.1021/acs.jpcc.9b02736 (2019). Publisher: American Chemical Society.
+
+36. Woodward, P. M. Octahedral Tilting in Perovskites. I. Geometrical Considerations. *Acta Crystallogr. Sect. B Struct. Sci.* 53, 32–43, DOI: 10.1107/S0108768196010713 (1997).
+
+37. Jain, A. et al. Commentary: The Materials Project: A materials genome approach to accelerating materials innovation. *APL Mater.* 1, 011002, DOI: 10.1063/1.4812323 (2013).
+
+38. Meloni, S., Palermo, G., Ashari-Astani, N., Grätzel, M. & Rothlisberger, U. Valence and conduction band tuning in halide perovskites for solar cell applications. *Mater. Chem. A* 4, 15997–16002, DOI: 10.1039/C6TA04949D (2016).
+
+39. Jiang, S. et al. Phase Transitions of Formamidinium Lead Iodide Perovskite under Pressure. *Am. Chem. Soc.* 140, 13952–13957, DOI: 10.1021/jacs.8b09316 (2018). Publisher: American Chemical Society.
+
+40. Fabini, D. H. et al. Dynamic Stereochemical Activity of the Sn(2+) Lone Pair in Perovskite CsSnBr3. *Am. Chem. Soc.* 138, 11820–11832, DOI: 10.1021/jacs.6b06287 (2016).
+
+41. Giannozzi, P. et al. QUANTUM ESPRESSO: a modular and open-source software project for quantum simulations of materials. *Physics: Condens. Matter* 21, 395502, DOI: 10.1088/0953-8984/21/39/395502 (2009).
+
+42. van Setten, M. J. et al. The PseudoDojo: Training and grading a 85 element optimized norm-conserving pseudopotential table. *Phys. Commun.* 226, 39–54, DOI: 10.1016/j.cpc.2018.01.012 (2018).
+
+43. Hamann, D. R. Optimized norm-conserving Vanderbilt pseudopotentials. *Rev. B* 88, 085117, DOI: 10.1103/PhysRevB.88.085117 (2013). Publisher: American Physical Society.
+
+44. Monkhorst, H. J. & Pack, J. D. Special points for Brillouin-zone integrations. *Rev. B* 13, 5188–5192, DOI: 10.1103/PhysRevB.13.5188 (1976).
+
+45. Grimme, S. Density functional theory with London dispersion corrections. *WIREs Comput. Mol. Sci.* 1, 211–228, DOI: 10.1002/wcms.30 (2011).
+
+46. Kühne, T. D. et al. CP2K: An electronic structure and molecular dynamics software package - Quickstep: Efficient and accurate electronic structure calculations. *The J. Chem. Phys.* 152, 194103, DOI: 10.1063/5.0007045 (2020).
+
+47. Perdew, J. P. et al. Restoring the Density-Gradient Expansion for Exchange in Solids and Surfaces. *Rev. Lett.* 100, 136406, DOI: 10.1103/PhysRevLett.100.136406 (2008). Publisher: American Physical Society.
+
+48. VandeVondele, J. & Hutter, J. Gaussian basis sets for accurate calculations on molecular systems in gas and condensed phases. *The J. Chem. Phys.* 127, 114105, DOI: 10.1063/1.2770708 (2007).
+
+49. Bussi, G., Donadio, D. & Parrinello, M. Canonical sampling through velocity rescaling. *The J. Chem. Phys.* 126, 014101, DOI: 10.1063/1.2408420 (2007). Publisher: American Institute of Physics.
+
+50. Martyna, G. J., Tobias, D. J. & Klein, M. L. Constant pressure molecular dynamics algorithms. *The J. Chem. Phys.* 101, 4177–4189, DOI: 10.1063/1.467468 (1994).
+
+# Supplementary Files
+
+- [FAPbIbandgapSI.docx](https://assets-eu.researchsquare.com/files/rs-5730287/v1/ca27815f798ebd6f92981894.docx)  
+  Nanoscale size effects in α-FAPbI3 evinced by large-scale ab initio simulations

06b1630909e8e227d7da602a7a917bcb78d8a8e2df2080ead1f389e367c0558e/preprint/images_list.json

@@ -0,0 +1,50 @@
+[
+    {
+        "type": "image",
+        "img_path": "images/Figure_1.png",
+        "caption": "The stomatin-like protein StlP is important for morphogenesis in S. coelicolor. (A) Growth of S. coelicolor M145, its stlP mutant and the complemented strain (\u2206stlP+pXZ15) measured using a BioLector. Strains were grown in LPB medium, and biomass was measured in arbitrary unites in triplicates. (B) Colonies of S. coelicolor M145, its stlP mutant and the complemented strain (\u2206stlP+pXZ15) after 5 days of growth. (C) Quantitative measurement of average colony diameters. Error bars represent the standard error of the mean (***, p<0.0001). (D) Mycelial morphology of the strains grown on LPMA medium. 16 h-old mycelium was labeled with SYTO-9 (stains nucleic acids) and FM5-95 (lipid stain). CWD cells are indicated with the white arrowhead. (E) Stills of Supplementary Movie 1 showing growth of the stlPmutant on LPMA medium. Micrographs were taken every 10 min. (F) The absence of StlP causes hyperbranching in S. coelicolor. Hyphae were imaged after 16 h of growth on cellophane membranes overlaying LPMA plates. Note that reintroduction of stlP expressed from the constitutive gapAp promoter (pXZ15) restores normal branching in the stlP mutant. (G) Histograms showing the tip-to-branch distances in hyphae of M145, its stlP mutant and the complemented mutant (\u2206stlP+pXZ15). Strain were grown for 16 h. For the measurements, n was 188 for M145, 282 for the stlP mutant and 192 (complemented mutant). Scale bars represent 5 mm (B), 50 \u00b5m (D) and 5 \u00b5m (E and F), respectively.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_2.png",
+        "caption": "StlP localizes to hyphal tips and affects cell wall synthesis. (A) Localization of mCherry-StlP in the wild-type strain and the \u2206stlP mutant, both of which carry plasmid pXZ16. (B) Localization of DivIVA-mCherry in the wild-type strain and the \u2206stlP mutant, both of which carry plasmid pXZ17. Strains were grown in LPB medium for 16 h prior to imaging. The arrowheads indicate the location of DivIVA-mCherry. (C) Visualization of nascent peptidoglycan in S. coelicolor strains using fluorescent vancomycin (VanFL) staining. White arrows indicate PG synthesis sites. (D) Deletion of stlP increases hyphal diameter. Quantification was done by measuring the hyphal diameter after staining with FM5-95. (E) The absence of StlP affects proper deposition of the cellulose-like glycan. Foci are either splitting at hyphal tips or diffused along the filament in the \u2206stlP mutant. Strains were grown in LPB medium and fluorescent images were taken after 16 h of growth. Bars represent 20 \u03bcm. (F) Quantification of the cellulose-like glycan at hyphal tips using calcofluor white staining. For each tip, the total fluorescence in a square (1.5 \u03bcm by 1.5 \u03bcm) at the hyphal tip was measured using ImageJ software. (G) The absence of the cellulose-like glycan at hyphal tips increases lysozyme sensitivity of the \u2206stlP mutant. For each strain, ~1000 spores were plated on nutrient agar plates, and colony numbers were counted after 3 days. The percentage (number of colonies from plates with 0.25 mg ml-1 lysozyme divided by the number of colonies from plates without lysozyme) was used to evaluate the sensitivity of strains for lysozyme. (H) Cryo-ETs of sacculi of the wild-type and \u2206stlP mutant strain (top panel). The bottom panels indicate straightened apical regions of sacculi for the correspnding strain. White arrows indicate contractile injection systems (see Fig. S8 for more examples). (I) Thickness measurement of sacculi of the wild-type and \u2206stlP mutant strain. Error bars represent the standard error of the mean (***, P<0.001; ns, non-significance). Bars represent 20 \u03bcm (A), 5 \u03bcm (B), 1 \u03bcm (C),\u00a0 10 \u03bcm (E), 200 nm (H, top panel), and 50 nm (H, bottom panel), respectively. Error bars represent the standard error of the mean (****,P<0.0001; ***, P<0.001; **, P<0.01; ns, not significant).",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_3.png",
+        "caption": "StlP controls membrane fluidity at hyphal tips. (A) StlP assembles into oligomers, consistent with other stomatin proteins. The stomatin domain of StlP (StlPSD) was purified and 1 mg ml-1 of protein was cross-linked with glutaraldehyde (0%, 0.02%, 0.05% and 0.1%) for 10 or 60 min at room temperature. Lane BC and AC represent the protein before and after concentration, respectively. The ladder-like pattern on SDS-PAGE indicates oligomerization. (B) StlP decamer assembles into a 10-fold symmetric ring structure predicted using AlphaFold. The prediction utilized the amino acid sequence spanning two N-terminal transmembrane helices (TMHs) at positions 106-126 and 149-168, as well as the stomatin domain at positions 209-311 of StlP. (C) Visualization of fluid region at hyphal tips using DilC12 staining, in which the 16h-old culture of S. coelicolor M145, \u2206stlP mutant and the complemented mutant strain (\u2206stlP+pXZ15) were incubated with 100 \u03bcM DilC12 and grown for further 3 hours in LPB liquid medium prior to imaging. White arrows indicate fluid regions at hyphal tips. Please note that the extruded CWD cells are highly fluid. Qualitative (D) and quantitative (E) analysis of membrane fluidity using Laurdan staining. Mycelia from 16h-old cultures of different strains were collected and labeled with 1mM 100 \u03bcM Laurdan. The values in the graph (E) indicate the generalized polarization (GP), which ranges from -1 (more fluid) to +1 (less fluid). (F) Increased growth temperature restores membrane fluidity of the \u2206stlP mutant. Here, strains were grown in LPB liquid medium at 22, 30 or 37\u2070C for 16 h and the membrane fluidity is indicated as GP value. Approximately 45 hyphae were measured for each strain. Scale bar represents 5 \u03bcm (C), 10 \u03bcm (D, main images) and 0.5 \u03bcm (D, inlays), respectively. Error bars represent the standard error of the mean (***, P<0.001).",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_4.png",
+        "caption": "Phylogenetic analysis of the distribution of StlP proteins in Proteobacteria. Maximum-likelihood was used to construct the phylogenetic trees. PSI-BLAST and TMHMM prediction were used to identify StlP homologs in the dataset of 15,405 RefSeq representative bacteria and archaea (left panel). The right panel indicates the distribution of StlP homologs in the genus of actinobacteria. The inner strips indicates different phyla (left panel) or genera (right panel), represented by different colors, while the outside black strip indicates StlP homologs. The green and pink arrowheads represent Streptomyces coelicolor A3(2) and Kitasatospora viridifaciens DSM40239, respectively.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_5.png",
+        "caption": "Enhanced robustness of K. viridifaciens growth under hyperosmotic stress conditions through constitutive StlP expression. (A) Visualization of fluid membrane regions at hyphal tips of a K. viridifaciens derivative constitutively expressing StlP (K. viridifaciens+pXZ15) compared to the parental wild-type. Images were obtained after DilC12 staining of 16h-old cultures followed by growth for 3 hours in LPB liquid medium. White arrows indicate fluid regions at hyphal tips. Qualitative (B) and quantitative (C) analysis of membrane fluidity using Laurdan staining. Mycelia from 16h-old cultures were collected and labeled with 1mM Laurdan. The values in the graph (C) indicate the generalized polarization (GP), which ranges from -1 (more fluid) to +1 (less fluid). (D) Constitutive expression of stlP from the gapAp promoter increases the average colony size of K. viridifaciens. Colonies of the wild-type strain and a derivative constitutively expressing StlP (K. viridifaciens+pXZ15) are shown after 7 days of growth. (E) Quantitative assessment of the average colony diameter. (F) Constitutive expression of stlP from the gapAp promoter prevents hyperbranching. Individual hyphae were imaged after 16 h of growth on cellophane membranes overlaying LPMA plates. (G) Histograms indicating the distribution of tip-to-branch distances of K. viridifaciens with or without pXZ15. The number of filaments measured for each strain were 219 (K. viridifaciens) and 189 (K. viridifaciens+pXZ15). (H) Constitutive expression of StlP (from plasmid pXZ15) blocks the extrusion of wall-deficient cells in K. viridifaciens. (I) Quantification of cell wall-deficient cells was assessed after growing strains in LPB medium for 40 h. Scale bars represent 5 \u03bcm (A and F), 10 \u03bcm (B, main images), 0.5 \u03bcm (B, inlays), 5 mm (D) and 20 \u03bcm (H), respectively.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_6.png",
+        "caption": "Proposed model for tip growth under conditions of hyperosmotic stress. StlP, a stomatin-like protein, undergoes oligomerization, creating an assembly on the membrane that facilitates the formation of microdomains. These microdomains induce local membrane fluidization and serve as a platform for coordinated cell wall synthesis, ensuring normal polar growth. In the absence of StlP (right panel), membrane fluidity diminishes in the tip region, coinciding with the diffusion of apical cellulose-like glycan and peptidoglycan synthesis foci, resulting in the loss of spatially confined cell wall synthesis. This weakens the cell wall, leading to the extrusion of cells with deficient walls. The black arrowhead points to a CWD cell that rebuilds its cell wall.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    }
+]

06b1630909e8e227d7da602a7a917bcb78d8a8e2df2080ead1f389e367c0558e/preprint/preprint.md

@@ -0,0 +1,240 @@
+# Abstract
+
+The cell wall represents an essential structure conserved among most bacteria, playing a crucial role in growth and development. While extensively studied model bacteria have provided insights into cell wall synthesis coordination, the mechanism governing polar growth in actinobacteria remains enigmatic. Here we identify the stomatin-like protein StlP as a pivotal factor essential for orchestrating polar growth in filamentous actinobacteria under hyperosmotic stress. StlP facilitates the establishment of a membrane microdomain with increased membrane fluidity, a process crucial for maintaining proper growth. The absence of StlP leads to branching of filaments, aberrant cell wall synthesis, thinning of the cell wall, and the extrusion of cell wall-deficient cells at hyphal tips. StlP interacts with key components of the apical glycan synthesis machinery, providing protection to filaments during apical growth. Introduction of StlP in actinobacteria lacking this protein enhances polar growth and resilience under hyperosmotic stress, accompanied by the formation of a membrane microdomain. Our findings imply that stomatin-like proteins, exemplified by StlP, confer a competitive advantage to actinobacteria encountering hyperosmotic stress. Given the widespread conservation of StlP in filamentous actinobacteria, our results propose that the mediation of polar growth through membrane microdomain formation is a conserved phenomenon in these bacteria.
+
+**Biological sciences/Microbiology**  
+**Biological sciences/Biochemistry**  
+membrane microdomain  
+filamentous Actinobacteria  
+streptomyces  
+stomatin  
+StlP  
+polar growth  
+cell wall-deficient cells
+
+# Introduction
+
+The cell wall is considered an essential structure in bacteria that protects cells from environmental stresses<sup>1,2</sup>. To enable bacterial growth, the cell wall needs to be expanded, which involves inserting new cell wall material at the sites of growth. Elongation of rod-shaped cells typically occurs in two distinct manners<sup>3</sup>. Some rod-shaped bacteria, such as *Escherichia coli* and *Bacillus subtilis* elongate by incorporating new cell wall material in a rather diffuse manner in the cylindrical part of the cell<sup>4</sup>. By contrast, other bacteria grow by inserting new cell wall material at the cell poles, referred to as polar growth, which is widespread in actinobacteria<sup>5,6</sup>. This mode-of-growth has been well studied in actinomycetes, which are filamentous bacteria that form branched mycelial networks in soil environments. In their natural environment, actinomycetes are often confronted with suboptimal conditions, such as fluctuations in water availability causing dramatic osmotic imbalances<sup>7</sup>. Paradoxically, we recently showed how conditions of hyperosmotic stress causes shedding of the cell wall in several actinomycetes, implying that such conditions interfere with the process of cell wall growth<sup>8</sup>.
+
+Polar growth in actinomycetes is guided by the cytoskeletal protein DivIVA, which localizes in actively growing tips acting as a scaffold for other proteins involved in organizing tip growth, such as the coiled-coil protein Scy and the intermediate filament-like protein FilP<sup>9–11</sup>. Unlike Scy and FilP, DivIVA plays an essential function in polar growth, by directly interacting with the machinery involved in synthesis of peptidoglycan, a major constituent of the cell wall<sup>12,13</sup>. The partial depletion of DivIVA caused hyphal bulging and irregular branching<sup>9</sup>. Scy is an unusual long coiled-coil protein that co-localizes with DivIVA at hyphal tips. Scy was suggested to form higher order assemblies and thereby serves as a hub to stabilize the tip-organizing center, which also includes cell division proteins and proteins involved in chromosome segregation<sup>11,14,15</sup>. Collectively, these proteins make sure that cell wall synthesis is coordinated with chromosome segregation to ensure proper growth and development of the mycelium.
+
+DivIVA also interacts with the putative cellulose synthase CslA<sup>16</sup>. The glycan produced by CslA is thought to provide protection to the tips, which are constantly remodeled during growth, in particular under conditions of osmotic stress<sup>16</sup>. Synthesis of the cellulose-like glycan not only depends on the synthase CslA, but also on a range of other proteins that are all encoded in a conserved gene cluster (Supplementary Fig. 1A). Together with CslA, the radical copper oxidase GlxA are the key proteins responsible for synthesis and likely modification of the glycan<sup>17,18</sup>. Following synthesis of the glycan, the lytic polysaccharide monooxygenase LpmP and endoglucanase CslZ facilitate deposition of the glycan chain at the cell surface, possibly by creating a passage through the thick peptidoglycan (PG) layer<sup>19</sup>. The cooperation of CslA/GlxA and CslZ/LpmP implies that a multicomplex is established at the tip related to the proper synthesis and secretion of the cellulose-like glycan.
+
+One of the proteins in the cellulose biosynthetic gene cluster that has not been studied encodes a stomatin/prohibitin/flotillin/HflK/C (SPFH)-domain protein, hereinafter referred to as StlP (for stomatin-like protein). The SPFH-superfamily proteins facilitate formation of lipid rafts, which in eukaryotes often contain protein complexes that collectively carry out important biological processes, such as ion channel regulation and touch sensation<sup>20,21</sup>. Notably, prokaryotic SPFH proteins, such as flotillins, also form comparable structures known as functional membrane microdomains (FMMs)<sup>22</sup>. These FMMs were shown to be involved in cell wall biosynthesis in *B. subtilis*<sup>23</sup> and *Staphylococcus aureus*<sup>24</sup>. Additionally, it was revealed that, in *B. subtilis*, flotillins play a direct role in controlling membrane fluidity homeostasis<sup>23</sup>, with their expression intriguingly regulated by stress-specific signals<sup>25</sup>.
+
+In this study, we reveal the pivotal role of StlP in coordinating the creation of a microdomain crucial for sustaining tip growth in *Streptomyces coelicolor* during hyperosmotic stress conditions. The absence of StlP leads to anomalous hyphal shape alterations and, notably, the expulsion of cells lacking their cell wall. StlP undergoes polymerization into oligomers and localizes within the membrane, culminating in the formation of a membrane region characterized by heightened fluidity. This phenomenon is essential for harmonizing the growth of both membrane and cell wall during tip extension. Significantly, the ectopic expression of StlP in actinomycetes, known for naturally extruding wall-deficient cells, effectively prevents such extrusion. These findings collectively underscore the significance of StlP in establishing a membrane microdomain at hyphal tips, thereby playing a critical role in facilitating proper cell wall assembly under hyperosmotic stress conditions in filamentous actinobacteria.
+
+# Results
+
+StlP is a stomatin-like protein in *Streptomyces coelicolor*  
+StlP is encoded in the conserved cellulose biosynthesis gene cluster in streptomycetes and located downstream of *lpmP*<sup>19</sup> (Supplementary Fig. 1A). Analysis of protein domains with InterPro and structure prediction via AlphaFold reveal that StlP contains three domains: a disordered region (aa 1–100), followed by a transmembrane hairpin composed by two helices (aa 106–126 and 149–168), and a SPFH (stomatin, prohibitin, flotillin, HflK/C) domain (aa 209–311) (Supplementary Fig. 1B, Supplementary Fig. 1C).  
+
+Considering the diverse membrane topologies observed in members of the SFPH superfamily<sup>26</sup>, we examined the membrane topology of StlP to determine its classification within the family of SPFH-containing proteins. *In silico* analyses predicted that the C-terminal SPFH domain of StlP is in the cytoplasm (Supplementary Fig. 1D). To validate this prediction, we employed a β-lactamase assay in which this enzyme was fused to the C-terminus of StlP as described<sup>27</sup>. *E. coli* cells expressing StlP-BlaM<sub>NS</sub> fusions demonstrate sensitivity to ampicillin, in contrast to cells expressing BlaM<sub>FL</sub> with its original signal peptide (Supplementary Fig. 1E). This result confirms that the C-terminal SPFH domain of StlP is situated in the cytoplasm, consistent with the observed membrane topology in podocin/stomatin family proteins<sup>20</sup> and its N-terminal transmembrane hairpin. In contrast to StlP, other SPFH-like proteins from *S. coelicolor* have a significantly different predicted membrane topology (Supplementary Fig. 2). A phylogenetic analysis, utilizing amino acid sequences from well-known prokaryotic and eukaryotic SPFH proteins, revealed that *S. coelicolor* StlP and Mouse stomatin STOML-1 form a monophyletic clade, indicating that StlP resembles a stomatin-like protein (Supplementary Fig. 3A). Further sequence alignment of StlP with other stomatins indicates that *Streptomyces* StlP contains the signature proline<sup>28</sup> residue required for membrane hairpin formation and a comparable domain arrangement (Supplementary Fig. 3B). This alignment suggests that StlP shares more similarities with eukaryotic stomatins than prokaryotic stomatins. In summary, these findings collectively establish StlP as a stomatin-like protein within the prokaryotic family of SPFH proteins.
+
+## StlP is important for morphogenesis under osmotic stress conditions
+
+To investigate the function of StlP, a *stlP* mutant was created that contained a Tn5062 transposon insertion, positioned upstream of the stomatin domain. This constructed *stlP* mutant was used to study its phenotype in various conditions. The *stlP* mutant displayed significantly reduced growth rates in both TSBS and LPB liquid medium, which contain 10 and 22% sucrose, respectively (Fig. 1A and Supplementary Fig. 4A). On MS agar medium (without sucrose), no apparent differences in growth were observed between the parent and the *stlP* mutant (Supplementary Fig. 4B). By contrast, when grown on LPMA agar medium (containing 22% sucrose), the average diameter of individual colonies of the *stlP* mutant (1.4 ± 0.3 mm) was significantly reduced compared to the wild-type strain (3.8 ± 0.6 mm) (Fig. 1B, Fig. 1C), suggesting that growth was severely hampered. Furthermore, we noticed that excess membrane was extruded from hyphal tips of the *stlP* mutant, as evident in the FM5-95 panels comparing M145 and ∆*stlP* in Fig. 1C. Additionally, we observed many DNA-containing vesicles present in the medium (Fig. 1D) that were reminiscent of cell wall-deficient cells extruded by several filamentous actinobacteria. These vesicles were absent in the parental strain, consistent with earlier findings<sup>8</sup>. Time-lapse imaging revealed that the vesicles were extruded from hyphal tips (Fig. 1E, Supplementary Movies 2). Quantification revealed that the mutant formed 2.5x10<sup>5</sup> vesicles ml<sup>−1</sup>, while none were found in the parental strain (Supplementary Fig. 5). To confirm that these phenotypes were caused by the absence of StlP, we introduced plasmid pXZ15 in the mutant, in which *stlP* is expressed from the constitutive *gapAp* promoter. Reintroduction of this plasmid partially restored the growth speed in liquid media (Fig. 1A, Supplementary Fig. 4A) and the average colony diameter on LPMA medium (2.5 ± 0.5 mm) (Fig. 1B, Fig. 1C), while extrusion of membranes and DNA-containing vesicles was also reduced by 80% (Fig. 1D, Supplementary Fig. 5).
+
+Microscopy analysis also indicated that the hyphal branching pattern was affected by the deletion of *stlP* (Fig. 1F, Supplementary Fig. 4D, Supplementary Movies 1 and 2). More specifically, the number of branches was dramatically increased in the *stlP* mutant when grown under hyperosmotic stress conditions (LPMA medium) (Fig. 1F). To quantitatively compare branching between the strains, we measured the distance from the tip to the proximal branching point in hyphae (Supplementary Fig. 4D). Deletion of *stlP* changed the distribution of the tip-to-branch distances significantly. We found that more than 60% of all hyphae in ∆*stlP* had a proximal branch within the first 5 µm from the tip (Fig. 1G), as compared to approximately 15% for the parent and the complemented mutant. Also, no hyphae of the *stlP* mutant branched further than 35 µm from the tip. Additionally, when grown on MS medium, hyphae with a tip-to-branch distance less than 10 µm accounted for 3.6% and 8.9% of the parent and mutant strain, respectively (Supplementary Fig. 4E). This suggests a mild impact on the branching pattern due to the deletion of *stlP* under normal growth conditions. Taken together, these findings underscore that the absence of *stlP* significantly influences the growth and morphogenesis of *S. coelicolor*, particularly manifesting under conditions of hyperosmotic stress.
+
+## StlP is important for spatially confining cell wall synthesis to hyphal tips
+
+To investigate the localization of StlP, we introduced pXZ16 into *S. coelicolor* M145, thereby constitutively expressing a C-terminal mCherry fusion to StlP. Foci of StlP-mCherry mostly localized at growing tips and emerging branches (Fig. 2A). We also localized the polar growth determinant DivIVA in the presence and absence of StlP by expressing a DivIVA-mCherry fusion from the constitutive *gapAp* promoter. Interestingly, in the absence of StlP, DivIVA-mCherry was not only localized to hyphal tips but was also found in numerous foci along the cylindrical part of the filaments (Fig. 2B), suggesting the cell wall synthesis was no longer confined to the apex. In agreement, nascent PG was incorporated at multiple sites along the filament in the *stlP* mutant (Fig. 2C), coinciding with an increase in the average diameter of the hyphae (Fig. 2D). Furthermore, we noticed that multiple synthesis foci of the cellulose-like glycan by CslA appeared at established and emerging hyphal tips, which is contrary to the parental strain (Fig. 2E). Deposition of the cellulose-like glycan was affected in the *stlP* mutant and was no longer spatially confined to the hyphal tip (Fig. 2E). Quantitative analysis indicated that the absence of StlP let to some 60% reduced accumulation of glycans at hyphal tips, indicated by calcofluor white staining (Fig. 3F). Furthermore, the reduced glycan levels made the *stlP* mutant sensitive to lysozyme (Fig. 3G), which was previously observed in other mutants affected in apical glycan deposition. Glycan deposition and lysozyme resistance were restored when the complementation plasmid pXZ15 was introduced in the *stlP* mutant (Fig. 3E, Fig. 3F and Fig. 3G).
+
+To accurately evaluate the effect of the absence StlP on cell wall thickness of *S. coelicolor*, we conducted a thorough examination by preparing and imaging sacculi of M145 and the ∆*stlP* strains using cryo-electron tomography (cryo-ET). The analysis of cell wall thickness measurement revealed a significant reduction in the overall cell wall thickness upon *stlP* deletion compared to the parental strain. Specifically, the thickness in the apical region decreased from 35.5 ± 1.7 nm to 6.4 ± 0.6 nm, while the thickness in the subapical region was reduced from 37.1 ± 2.3 nm to 9.9 ± 0.6 nm (Fig. 2H and Fig. 2I). These results demonstrate the importance of StlP for cell wall synthesis and thickness of *S. coelicolor*.
+
+To establish how StlP contributes to delocalized cell wall synthesis, we tested interactions of StlP with the proteins involved in tip growth and cellulose biosynthesis. To this end, constructs were generated that produced C-terminal fusions of StlP, LpmP, SCO2835, CslA, GlxA, CslZ, DivIVA, Scy and FilP to either the T25 or T18 fragments of the adenylate cyclase, respectively. Co-transformation of these constructs in *E. coli* BTH101 revealed that StlP robustly interacts with LpmP, SCO2835, CslA and CslZ, but also with itself and weakly interacts with GlxA (Supplementary Fig. 6). Furthermore, StlP did not interact with DivIVA, Scy or FilP (Supplementary Fig. 6). Thus, StlP directly interacts with components of the cellulose biosynthesis complex and spatially confines cell wall synthesis to hyphal tips in *S. coelicolor*.
+
+### Deletion of *stlP* induces cell death in *S. coelicolor*
+
+We noticed a significant abundance of contractile injection system (CIS) structures within the sacculi of the *stlP* mutant cultivated for 16 h, in contrast to the absence of these structures in the sacculi of an equivalently aged parental strain (Fig. 2H). These CIS structures were recently found to be involved in programmed cell death regulation<sup>29,30</sup>. The observation of CIS led us to investigate if the deletion of *stlP* increased cell death of *S. coelicolor*. To investigate this, M145 and its *stlP* mutant were grown for 16 h on LPMA medium. Cell death in the mycelia was measured using a bacterial viability assay<sup>31</sup> (see Materials and Methods). In principle, when SYTO9 and propidium iodide (PI) nucleic acid stains are present simultaneously, stained viable mycelia and dead mycelia exhibited green, and red fluorescence, respectively. The live/dead (SYTO9/PI) ratio was notably reduced in the *stlP* mutant when compared to the parent strain (see Supplementary Fig. 8). This decrease suggests increased cell death in the *stlP* mutant, aligning with our findings that the deletion of *stlP* significantly impeded the growth of liquid-cultured *S. coelicolor* (Fig. 1A, supplementary Fig. 4A) Altogether, these results imply that deletion of *stlP* contributes to increased cell death of *S. coelicolor*.
+
+### StlP oligomerizes and forms membrane microdomain at hyphal tips of *S. coelicolor*
+
+Bacterial two-hybrid analysis suggested that StlP interacts with itself but also with components of the machinery involved in glycan synthesis (Supplementary Fig. 6). To verify if StlP interacts with itself via the stomatin domains, the corresponding domain (aa 204–326; referred to as StlP<sub>SD</sub>) was expressed in *E. coli* BL21(DE3). The StlP<sub>SD</sub> monomer was expressed, showing a predicted molecular mass of 18 kDa (lane BC, Fig. 3A). When samples were concentrated to 1 mg ml<sup>−1</sup>, a ladder of oligomers was found (lane AC), suggesting that monomers can self-assemble. Notably, less monomers and more crosslinked highly ordered polymers were observed with increasing concentrations of the cross-linker glutaraldehyde (Fig. 3A). These results suggest that StlP can assemble into oligomers via its stomatin domain at high concentrations, which in the full-length protein would probably result in formation of a membrane microdomain via its two N-terminal transmembrane helices. To corroborate this hypothesis, the partial structure of SltP (aa 106–326), which covers the two N-terminal TMHs (aa 106–126 and aa 149–168) and the stomatin domain (aa 209–311) was predicted using AlphaFold 2.0. Given that the putative stomatin domain of FliL from the marine bacterium *Vibrio alginolyticus*, the only structure known of a bacterial stomatin, assembles into a symmetric ring consisting of 10 monomers, 10 StlP monomers were used for prediction of the structure. This predicted that the cytoplasmic stomatin domain of StlP assembles into a highly ordered multimer via an end-to-end interaction pattern, in which the N-terminal transmembrane helical hairpins of the monomers contribute to forming a symmetric ring (Fig. 3B).
+
+In bacteria, SPFH proteins are involved in membrane fluidity homeostasis, evidenced by FloT and FloA that fluidize the membrane and absence of them leads to an overall rigidification<sup>23,32</sup>. We therefore predicted that the oligomerization of StlP leads to the formation of a local membrane region with increased fluidity (RIF) at hyphal tips. To verify this prediction, we stained mycelia of *S. coelicolor* with DilC<sub>12</sub>, a lipid dye with high specificity for fluid membranes due to its short hydrocarbon tail<sup>33,34</sup>. We observed bright DilC<sub>12</sub> staining RIFs at hyphal tips, and also along hyphae grown for 16 h in LPB medium (Fig. 3C). The signal corresponding to DilC<sub>12</sub> staining was absent from hyphal tips of the *stlP* mutant and restored in the complemented strain (Fig. 3C). These results suggest that hyphal tips possess a RIF, which is dependent on the presence of StlP. To substantiate the presence of RIFs at hyphal tips, we conducted a quantitative assessment of membrane fluidity using the membrane-intercalating dye Laurdan. This dye exhibits a shift in fluorescence emission wavelength based on membrane fluidity<sup>33,35</sup>. After culturing mycelia of M145 and its *stlP* mutant in LPB medium for 16 hours, we stained them with Laurdan and then calculated the general polarization (GP) value at the tip region of each hypha (see Materials and Methods). In the absence of *stlP*, the average GP value (-0.09 ± 0.08) increased compared to its parental strain (-0.12 ± 0.13), inferring a decreased membrane fluidity in the *stlP* mutant (Fig. 3D and Fig. 3E). Furthermore, the apical membrane fluidity of the *stlP* mutant could be restored by raising the growth temperature to 37 ⁰C (Fig. 4F), which aligns with the idea that elevated temperatures contribute to membrane phase transitions, leading to an increased fluidity<sup>36</sup>. Altogether, these results reveal that StlP oligomerizes and forms membrane domains with enhanced membrane fluidity at hyphal tips of *S. coelicolor*.
+
+## Morphogenesis controlled by StlP is conserved in filamentous actinobacteria
+
+Our results indicate that StlP contributes to proper cell wall synthesis and membrane organization under hyperosmotic stress, and thereby controls morphogenesis of *S. coelicolor*. To see how prevalent StlP is, we combined PSI-BLAST and TMHMM prediction to identify stomatin homologues of StlP in the dataset of 15045 RefSeq representative bacteria and archaea. This analysis showed that the majority of StlP homologs are present in actinobacteria, while a few are also present in *Rhizobium* (Fig. 4). Furthermore, inside the filamentous actinobacteria, StlP orthologs are present in genera including *Streptomyces, Kitasatospora* and *Streptacidiphillus* (Fig. 4). In some clades of these bacteria, all members have an orthologue of StlP, while in others StlP is less common or virtually absent. Notably, *Kitasatospora viridifaciens* DSM40239 is among the clades that lack StlP (Fig. 4).
+
+To see how widespread the function of StlP is in other actinobacteria, the construct pXZ15, wherein the *stlP* was expressed from the constitutive *gapAp* promoter, was introduced into *Kitasatospora viridifaciens* DSM40239 via conjugation, which is known to extrude wall-deficient cells under hyperosmotic stress<sup>8,37</sup>. Importantly, constitutive expression of *stlP* induced the formation of a fluid membrane microdomain at hyphal tips in *K. viridifaciens*, as evidenced by DilC<sub>12</sub> staining (Fig. 5A). Furthermore, constitutive expression of *stlP* significantly increased the average membrane fluidity of hyphae (Fig. 5B, Fig. 5C) and allowed colonies to cope much better with hyperosmotic stress, as shown by the strongly increased colony diameter of the strain expressing *stlP* (2.75 ± 0.6 mm) as compared to the parental strain (1.25 ± 0.9 mm) and by the reduced lateral branching of *K. viridifaciens* (Fig. 5D, Fig. 5E). Furthermore, constitutive expression of *stlP* largely abolished the extrusion of wall-deficient cells in *K. viridifaciens* (Fig. 5F). By contrast, the constitutive expression of the HflK/C-like protein BOQ63_030050 in *K. viridifaciens*, which also belongs to the SPFH superfamily of proteins and shares 21% identity with StlP had no effect on the extrusion of wall-deficient cells (Supplementary Fig. 9, 10). These results demonstrate that StlP controls growth of filamentous actinobacteria under hyperosmotic stress by the localized control of membrane fluidity.
+
+# Discussion
+
+Stomatin-like proteins are ubiquitous in all domains of life. A universal feature of these proteins is to form functional nanoscale microdomains in biological membranes. In turn, these microdomains serve as platforms to locate protein complexes involved in important biological processes, such as transducing mechanosensory signals in mice and the modulation of ion channels in mammalian cells<sup>38–40</sup>. The recent structural analysis of the stomatin-like protein FliL from *Vibrio alginolyticus* showed that FliL shares some structural elements with eukaryotic stomatins, suggesting that stomatin-like proteins are conserved from mammals to bacteria<sup>40</sup>. Here, we identified the stomatin-like protein StlP as a novel polar growth determinant in filamentous actinobacteria. We show that StlP is crucial for spatially organizing cell wall synthesis at hyphal tips in conditions of hyperosmotic stress, which likely provides competitive benefit to microbes frequently exposed to such conditions.
+
+Evolutionary investigation showed the group of stomatins are ancient and their evolution likely occurred at an early stage in the evolution of prokaryotes<sup>41</sup>. Most prokaryotic stomatins (so-called p-stomatins) have been found in archaeal species, for instance in isolates from environments with high temperatures (*Pyrococcus horikoshii*, *Archaeoglobus fulgidus* and *Methanothermobacter thermautotrophicus*) and isolates from sediments near the sea (*Aeropyrum pernix*)<sup>41,42</sup>. Some p-stomatins exist in bacterial species, such as in *Photobacterium aphoticum* isolated from coastal water<sup>43</sup> and the marine bacterium *Vibrio alginolyticus*<sup>40</sup>. However, their roles have not been characterized well. Excitingly, StlP, the p-stomatin of *S. coelicolor*, appears to play a pivotal role in regulating tip growth in conditions of hyperosmotic stress. Interestingly, *S. coelicolor* was originally isolated from the beach, which makes it plausible that *S. coelicolor*, contrary to *K. viridifaciens* isolated from mountain soil, is better adapted to salt-rich environments, for instance by having StlP.
+
+Stomatins oligomerize in cell membranes through interactions between their stomatin-domains, thus providing a scaffold for numerous other proteins. This functionality has been demonstrated for several stomatins, including mouse stomatin<sup>39</sup>, *Pyrococcus horikoshii* stomatin<sup>44</sup> and the stomatin FliL of *Vibrio alginolyticus*<sup>40</sup>. Cross-linking studies and structure predictions demonstrate that the stomatin domain of StlP oligomerizes and that StlP interacts with the machinery involved in synthesis of a cellulose-like glycan. One of these proteins is the cellulose synthase-like protein CslA, which in turn directly interacts with DivIVA<sup>16</sup>. Interestingly, PG synthesis of actinomycetes is directed at hyphal tips in a DivIVA-dependent manner. More specifically, DivIVA recruits penicillin-binding proteins (PBPs) during polarized growth, as deduced from the detected interaction between DivIVA and PBP3 in *M. tuberculosis*<sup>11,12</sup>. StlP seems to contribute to establishing a localized region in the cellular membrane carrying crucial components of the so-called tip organizing center, which controls cell wall synthesis during the polar growth in filamentous actinobacteria (Fig. <span class="InternalRef" refid="Fig6">6</span>). Indeed, the observation of diffused PG and surface cellulose synthesis foci and dramatically reduced cell wall thickness of the *stlP* mutant indicated that PG synthesis is significantly weakened upon the deletion of *stlP* (Fig. <span class="InternalRef" refid="Fig2">2</span>). We thus propose a model for the cell wall synthetic machinery at hyphal tips when filamentous actinobacteria grow in hyperosmotic stress conditions. In this model, DivIVA guides the machinery responsible for peptidoglycan (PG) and surface cellulose-like glycan synthesis to the poles through interactions with CslA and PBPs. At these poles, the systems are restricted by the formation of StlP rings, which, in turn, induce localized membrane fluidization (Fig. <span class="InternalRef" refid="Fig6">6</span>). The synthesis of the surface cellulose-like glycan involves substantial redox reactions facilitated by the presence of LpmP and GlxA. Here, the StlP ring, may serve to restrict oxidative stress within the ring, shielding proteins outside the StlP from this stress. Meanwhile, the StlP ring enables these filamentous actinobacteria to produce robust cell walls during polarized growth under conditions of hyperosmotic stress. Without StlP, the membrane fluidity in the tip region decreases. Combined with the loss of localized cell wall synthesis, this leads to the extrusion of wall-deficient cells. Reversely, constitutive expression of StlP in species that naturally lack this protein prevents the typical extrusion of wall-deficient cells. Taken together, this work provides a plausible mechanism how wall-deficient cells are extruded in polar-growing actinobacteria. By interrupting cell wall synthesis, by exposing the cells to hyperosmotic stress or antibiotics, the coordinated balance between membrane and cell wall synthesis is lost. In turn, this leads to shedding of excess membranes from the polar growth sites and is facilitated by the localized weakening of the cell wall.
+
+In summary, our work for the first time characterizes a stomatin-like protein that regulates tip growth in conditions of hyperosmotic stress. The strong phenotype associated with its absence make this also an interesting candidate to target in pathogenic bacteria that grow from the cell poles. In this context, it is noteworthy that mycobacteria have the ability to adopt a wall-deficient lifestyle<sup>45</sup>, and it’s worth highlighting that *Mycobacterium tuberculosis* possesses a StlP homolog.
+
+# Material and methods
+
+## Strains and growth conditions
+
+Strains used in this study are listed in Supplementary Table 1. Solid MS (Mannitol Soy flour) medium<sup>46</sup> was used for collection of *Streptomyces* spores and for conjugation experiments, while MYM medium<sup>47</sup> was used for obtaining *Kitasatospora* spores. To compare colony sizes and observe the release of cell wall-deficient (CWD) cells under hyperosmotic stress, solid LPMA medium<sup>8</sup> was used. TSBS<sup>46</sup> medium was used to grow *Streptomyces* in liquid medium without hyperosmotic stress. For quantification of the number of CWD cells, as well as measurement of hyphal diameters and membrane fluidity, liquid L-phase broth (LPB) was used<sup>8</sup>. Briefly, 10<sup>6</sup> CFU ml<sup>− 1</sup> spores were inoculated in 20 ml LPB in 50 ml flasks without coil while shaking at 100 rpm ml<sup>− 1</sup>. All *Streptomyces* and *Kitasatospora* strains were grown at 30°C. For hyphal branching detection, spores were inoculated onto cellophane membranes overlaying LPMA plates, which were then incubated at 30°C for 16 h prior to analysis. Lysozyme sensitive assay were performed essentially as described<sup>19</sup>.
+
+*E. coli* DH5α<sup>48</sup> was used for cloning and β-lactamase experiments. *E. coli* BL21 (DE3)<sup>49</sup> and BTH101<sup>50</sup> were used in protein expression and bacterial two-hybrid assays, respectively. For conjugation, *E. coli* ET12567 harboring pUZ8002 was used. All *E. coli* strains were grown in LB medium at 37°C with appropriate antibiotics, if needed.
+
+## Plasmid construction and transformation
+
+Plasmids and primers used in this study are listed in Supplementary Table 2 and Supplementary Table 3, respectively. For complementation of the ∆*stlP* mutant, the coding sequence of *stlP* (SCO2834) was amplified from genomic DNA of *S. coelicolor* with primers stlP-F/stlP-R1. The PCR product was ligated as an NdeI/BamHI fragment into plasmid hpXZ2<sup>19</sup> harboring the constitutive *gapAp* prompter of SCO1947, yielding plasmid pXZ15.
+
+For localization of StlP and DivlVA, the mCherry reporter was fused to the C-terminus of these two proteins. Coding sequences for *stlP* and *divlVA* were amplified from genomic DNA of *S. coelicolor* with primers stlP-F/stlP-R2 and 2077-F/2077-R, respectively. The gene encoding for mCherry was amplified from plasmid pGWS791<sup>27</sup> using primers mCh-F/mCh-R. After digestions with NdeI/HindIII (for *stlP* and *divlVA*) and HindIII/XbaI (for *mCherry*), combinations of *stlP/mCherry* and *divlVA/mCherry* were ligated with plasmid hpXZ2 that was cut with NdeI/XbaI, generating plasmid pXZ16 and pXZ17, respectively.
+
+To constitutively express the HflK/C-like protein BOQ63_030050 in *Kitasatospora viridifaciens* and the *S. coelicolor stlP* mutant, the coding sequence of BOQ63_030050 was amplified from genomic DNA of *K. viridifaciens* DSM40239 with primers 0050-F/0050-R. The PCR product was ligated as an NdeI/XbaI fragment into plasmid hpXZ2 harboring the constitutive *gapAp* prompter, yielding plasmid pXZ41.
+
+To express the stomatin-encoding domain of StlP (StlP<sub>SD</sub>) in *E. coli*, nucleotides 610–1107 of *stlP* were amplified from *S. coelicolor* genomic DNA with primers 2834<sub>sto</sub>-F/2834<sub>sto</sub>-R and ligated into pET28a plasmid as a NdeI and HindIII fragment, yielding plasmid pXZ18. Consequently, StlP<sub>SD</sub> was expressed carrying a N-terminal Histidine-tag (6xHis).
+
+All plasmids were introduced into *E. coli* and *Streptomyces* via heat-shock transformation<sup>51</sup> and conjugation<sup>46</sup>, respectively.
+
+## Inactivation of stlP in Streptomyces
+
+For inactivation of *stlP* in *S. coelicolor and S. lividans*, we used cosmid StE20 carrying the Tn5062 transposon inserted after nucleotide position 892 relative to the start site of *stlP* (kindly provided by Prof. Paul Dyson, see Supplementary Table 2). This cosmid was introduced into *S. coelicolor* via conjugation using ET12567/pUZ8002<sup>46</sup>, after which exconjugants were screened as described<sup>52</sup>. Colonies that were kanamycin-sensitive and apramycin-resistant were selected and used for further analysis. Mutants carrying the expected phenotype were verified by sequencing.
+
+## Growth curve generation
+
+For preparing germinated spores, spores of different strains were resuspended in double-strength germination medium<sup>46</sup> at a final concentration of 10<sup>6</sup> CFU ml<sup>− 1</sup> and incubated at 30 °C for 6 ~ 8 h while shaking at 200 rpm min<sup>− 1</sup>. For generating the biomass growth curve, the RoboLector L-4-BL-II equipped with a parallelized shaken cultivation device was used<sup>53</sup>. Briefly, the germinated spores were centrifuged and resuspended in either TSBS or LPB media before being distributed (1 ml per well) into a 48-well FlowerPlates (Basesweiler Germany). The temperature and humidity were set at 30 °C and 85%, respectively. The biomass was collected automatically and measured in arbitrary units. All measurements were performed in triplicate.
+
+## Bacterial 2-hybrid analysis
+
+To assess interactions of StlP with other tip-localizing proteins, the bacterial hybrid assay was used<sup>50</sup>. Therefore, *lpmP* (SCO2833), *stlP* (SCO2834), *sco2835*, *cslA* (SCO2836), *glxA* (SCO2837), *cslZ* (SCO2838), *divlVA* (SCO2077), *filP* (SCO5396) and *scy* (SCO5397) were amplified using primers Th2833-F/Th2833-R, Th2834-F/Th2834-R, Th2835-F/Th2835-R, Th2836-F/Th2836-R, Th2837-F/Th2837-R, Th2838-F/Th2838-R, divlVA-F/divlVA-R, filP-F/filP-R and scy-F/scy-R, respectively. All amplified DNA fragments were cloned into the pKT25 and pUT18C plasmids using EcoRI and XbaI, yielding plasmids pXZ19 (pUT18C + *lpmP*), pXZ20 (pKT25 + *stlP*), pXZ21 (pUT18C + *stlP*), pXZ22 (pKT25 + *SCO2835*), pXZ23 (pUT18C + *SCO2835*), pXZ24 (pKT25 + *cslA*), pXZ25 (pUT18C + *cslA*), pXZ26 (pKT25 + *glxA*), pXZ27 (pUT18C + *glxA*), pXZ28 (pKT25 + *cslZ*), pXZ29 (pKT25 + *divlVA*), pXZ30 (pUT18C + *divlVA*), pXZ31 (pKT25 + *filP*), pXZ32 (pUT18C + *filP*), pXZ33 (pKT25 + *scy*) and pXZ34 (pUT18C + *scy*) (see Supplementary Table 2).
+
+*E. coli* BT101 carrying combinations of these constructs were used in bacterial 2-hybrid experiments to evaluate protein interactions as described<sup>54</sup>.
+
+## Topology determination of StlP
+
+To study the transmembrane topology of StlP, the β-lactamase-encoding gene *blaM* without its signal sequence was fused to the 3’ end of *stlP*, as described previously<sup>27</sup>. Only if BlaM is secreted, cells will be resistant to ampicillin. To this end, a *blaM* variant lacking the region encoding the signal sequence for secretion (*blaM*<sub><em>NS</em></sub>) was amplified from plasmid pHJL401<sup>55</sup> with primers blaM-F/blaM-R. In parallel, full length *blaM* (including the region for the signal sequence, hereinafter referred to as *blaM*<sub><em>FL</em></sub>) was amplified from the same plasmid using the blaM<sub>FL</sub>-F/blaM-R primers. The *stlP* gene was amplified from *S. coelicolor* genomic DNA using primers stlP-F/*stlP*-R3. Subsequently, the amplified products were cut using restriction enzymes NdeI-HindIII (*stlP*), HindIII-EcoRI (*blaM*<sub><em>NS</em></sub>) and NdeI-EcoRI (*blaM*<sub>FL</sub>), and digested combinations of *stlP*/*blaM*<sub><em>NS</em></sub>, and *blaM*<sub>FL</sub> were separately ligated into pXZ2<sup>19</sup> that was cut with NdeI-EcoRI, yielding pXZ35 and pXZ36 (see Supplementary Table 2).
+
+*E. coli* DH5α harboring plasmid pSET152, pXZ35 and pXZ36 were used to assess the membrane topology of StlP, which were performed as described previously<sup>27</sup>. Full-length BlaM (BlaM<sub>FL</sub>) expressed from pXZ36 served as a control for β-lactamase activity in the medium.
+
+## Quantification of CWD cells
+
+Culturing and filtration of CWD cells of *Streptomyces* and *Kitasatospora* strains was essentially performed as described<sup>8</sup>, with the exception that *S. coelicolor* strains were grown for 16 h, while *K. viridifaciens* strains were grown for 40 h. Quantification of CWD cells was performed with a Bright-Line™ Hemocytometer (Merck), as described<sup>56</sup>. Briefly, 10 µl of filtered culture supernatant were loaded into the counting chamber, after which cells were quantified under a Zeiss Axio microscope equipped with an Axiocam 105 camera. CWD cell numbers were counted manually, and the density was calculated. For each strain, the measurements were performed in triplicate.
+
+## Protein expression and purification
+
+To purify the stomatin-encoding domain of StlP (StlP<sub>SD</sub>), *E. coli* BL21 (DE3) cells harboring plasmid pXZ17 were cultured at 37°C to an OD<sub>600</sub> of 0.8 in LB medium containing 50 mg ml<sup>− 1</sup> kanamycin. Then, 0.5 mM isopropyl β-D-thiogalactopyranoside was added to induce protein expression, after which cells were grown at 30°C for 18 h. The induced cells were subsequently lysed by sonication in binding buffer (50 mM Tris–HCl, 200 mM NaCl, pH 8.0), and after centrifugation, the lysate was loaded on a Ni<sup>2+</sup>-chelating column equilibrated with binding buffer. Ten column volumes of washing buffer (50 mM Tris–HCl, 200 mM NaCl, 0.1 mM imidazole, pH 8.0) and 5 mL of elution buffer (50 mM Tris–HCl, 200 mM NaCl, 10 mM imidazole, pH 8.0) were used to wash and elute StlP<sub>SD</sub>, respectively. Finally, the protein was purified by gel filtration using a Hiload 16/600 Superdex 200 pg column (GE Healthcare) equilibrated with buffer (50 mM Tris–HCl, 100 mM NaCl, pH 8.0). Sample fractions were analyzed on a 12.5% SDS-PAGE gel. Fractions were stored directly at -80°C or first concentrated to 1 mg ml<sup>− 1</sup> with the 3 kDa molecular weight cutoff concentrator (Millipore) and then stored.
+
+## Chemical cross-linking
+
+The cross-linking of StlP<sub>SD</sub> with glutaraldehyde (GA, 50% in H<sub>2</sub>O, Sigma) was performed as described<sup>44</sup> with the following modifications. Briefly, 1 mg ml<sup>− 1</sup> of StlP<sub>SD</sub> was treated with 0.02, 0.05 or 0.1% GA for 10 or 60 min at room temperature in 50 µl buffer (50 mM Tris–HCl, 100 mM NaCl, pH 8.0). Reactions were quenched with 0.2 M glycine-NaOH (pH 9.5) for 5 min at room temperature. Detection of cross-linked protein was performed by loading the samples on a 12.5% SDS-PAGE gel.
+
+## Microscopy
+
+To visualize the emergence of CWD cells, spores of the wild-type strain and the ∆*stlP* mutant were pre-germinated in double strength germination medium<sup>46</sup>. Then 10 µl of germlings were used for live-imaging, for which an ibiTreat 35 mm low imaging dish (ibidi) and a LPMA-pad was used as before<sup>8</sup>. Live imaging was carried out using a Zeiss LSM900 Airyscan 2 microscope. If necessary, Z-Stack acquisitions were used. For visualization of membrane and nucleic acids, 0.05 mg ml<sup>− 1</sup> FM5-95 and 0.5 µM SYTO-9 (Sigma) were used, respectively.
+
+The detection of nascent peptidoglycan was done by using BODIPY-FL vancomycin (Sigma), essentially as described<sup>57</sup>. Briefly, after growing *Streptomyces* strains in LPB medium for 16 h, mycelia were collected (3300 rpm min<sup>− 1</sup>, 10 min) and resuspended in 200 µl of fresh LPB medium containing 1 µg ml<sup>− 1</sup> BODIPY FL vancomycin and 1 µg ml<sup>− 1</sup> vancomycin. After 10 min incubation at 30°C, the mycelia were washed 3 times with PBS. 5 µl of the washed mycelia was used for microscopy analysis using a Zeiss LSM900 Airyscan 2 microscope.
+
+For visualizing fluid membrane microdomains, mycelia were stained with DilC<sub>12</sub><sup>33,34</sup>. Briefly, DilC<sub>12</sub> was dissolved in DMSO and 10 mg ml<sup>− 1</sup> stock was prepared. For sample preparation, spores of each strain were inoculated in LPB medium at a final concentration of 10<sup>6</sup> CFU ml<sup>− 1</sup>. After 16h of growth, mycelia were collected by centrifuge (3300 rpm min<sup>− 1</sup>, 10 min) and resuspended in prewarmed fresh LPB medium supplemented with 100 µg ml<sup>− 1</sup> DilC12, followed by growth for 3 h at 30°C while shaking at 100 rpm min<sup>− 1</sup>. Then, mycelia were collected, washed 3 times with prewarmed LPB supplemented with 1% DMSO, and resuspended in the same wash buffer. DilC<sub>12</sub> signal was detected via a Cy3 filter (535 nm excitation and 590 nm emission) using a Zeiss LSM900 Airyscan 2 microscope, wherein the cultivation chamber had been set 30°C to avoid temperature changes during imaging.
+
+For measurement of membrane fluidity, samples were prepared essentially as described<sup>32</sup>. Briefly, Laurdan (6-Dodecanoyl-N, N-dymethyl2-naphthylamine, Sigma) was dissolved in dimethylformamide (DMF) and a 10 mM stock was prepared. For preparation of mycelia for Laurdan staining, 20 ml 16 h-old mycelia were collected (3300 rpm min<sup>− 1</sup>, 10 min) and resuspended in 1 ml 30°C pre-warmed LPB medium containing 1 mM Laurdan. For sample preparation of the *stlP* mutant, the culture was filtrated through a 100 µm cut-off filter (Falcon Cell Strainer 100 µm Nylon) to remove the majority of the CWD cells. The filtered mycelium was resuspended in 1 ml 30°C pre-warmed LPB medium and used for Laurdan staining. After 10 min incubation in the dark at 30°C, stained mycelia were collected again and washed twice in 30°C pre-warmed PBS buffer supplemented with 20% sucrose and 1% DMF, and finally resuspended in 200 µl pre-warmed PBS buffer supplemented with 20% sucrose. Fluorescent intensities were measured at 435 and 490 nm, following excitation at 350 nm using a Zeiss LSM900 Airyscan 2 microscope, wherein the cultivation chamber had been set 30°C to avoid temperature changes during imaging. To calculate the membrane fluidity at hyphal tips, the tip region was cropped as a square (1 µm x 1 µm) from the image and the corresponding generalized polarization (GP) value was determined as described<sup>58</sup>.
+
+For measurement of hyphal diameters, 16 h-old mycelia were collected (3300 rpm min<sup>− 1</sup>, 10 min) and resuspended in fresh LPB medium containing 0.05 mg ml<sup>− 1</sup> FM5-95. The distance between the stained membranes was used to measure the hyphal diameters by averaging the diameter at 3 distinct spots in the hypha.
+
+The cellulose-like glycan at hyphal tips was visualized by calcofluor white (Sigma) staining and quantified as previous described<sup>19</sup>.
+
+For analysis of hyphal branching patterns, mycelium was grown from single spore and imaged using a Zeiss Axio microscope equipped with an Axiocam 105 camera as described previously<sup>19</sup>. The distance from the tip to the proximal branch point was measured. A proximal branch was defined as having a length of 1–4 µm as previous described<sup>59</sup>.
+
+For measurement of colony sizes, strain were grown on LPMA medium in petri dishes with a 9 cm diameter aiming for ± 100 colonies per plate. After growing them for 5 ~ 7 days at 30°C, plates were scanned with Epson Perfection V37 scanner and colony size was measured subsequently.
+
+Mycelial Live/dead staining was performed using the LIVE/DEAD<em>Bac</em> Light™ Bacterial viability kit (L7012; ThermoFisher) following the manufacturer’s instructions. Briefly, spores of the wild-type strain and the *stlP* mutant were streaked on LPMA medium. After growth for 16h, the excised ager pieces were inverted and positioned atop 10 µl mixture of SYTO-9 and propidium iodide (PI) nucleic acid stains from the kit with final concentration at 5 µM. Images were taken using a Zeiss LSM900 Airyscan 2 microscope after 10 mins incubation. The viability was calculated by dividing the integrated grey intensity of the fluorescence in the green channel by the integrated intensity of the fluorescence in the red channel. All measurement and images processing were executed with ImageJ software (version 2.0.0/1.53c/Java 1.8.0_172/64-bit).
+
+## Sacculus isolation and Cryo-electron tomography
+
+Isolation of sacculi of *S. coelicolor* M145 and the *stlP* mutant was essentially performed as described<sup>60</sup>, except that 16h-old liquid cultures were used and the step of removing teichoic acids was neglected.
+
+Sample preparation for cryo-electron tomography (cryo-ET) was performed as described<sup>60</sup>. Briefly, after adding the colloidal gold beads, sacculi solutions were vitrificated and applied on the EM grids. Grids were examined using a 120 kV Talos TEM (FEI/ThermoFisher) and cryo-ET data were collected using a Titan Krios instrument (ThermoFisher Scientific). The measurement of cell wall thickness was performed as described<sup>60</sup>.
+
+## Bioinformatic analysis
+
+Protein domains and protein structures were predicted by InterPro (<span class="ExternalRef"><span class="RefSource">https://www.ebi.ac.uk/interpro/</span></span>) and AlphaFold 2.0<sup>61</sup>. The prediction of protein membrane topology was performed by TMHMM (Version 2.0) (<span class="ExternalRef"><span class="RefSource">https://services.healthtech.dtu.dk/service.php?TMHMM-2.0</span></span>). Alignment of protein structures was done by PyMOL software (Version 2.5). Amino acids sequence alignment was done by ESPript 3.0 (<span class="ExternalRef"><span class="RefSource">https://espript.ibcp.fr/ESPript/cgi-bin/ESPript.cgi</span></span>).
+
+To phylogenetically compare StlP with other SPFH proteins, MEGA 7 was used. Amino acid sequences of all SPFH proteins were downloaded from the UniPort database. For phylogenetic analysis of the distribution of StlP, the amino acid sequence of StlP was used to run Position-Specific Iterative (PSI)-BLAST to find homologs in the dataset of 15405 RefSeq representative bacteria and archaea. The homologs with a bitscore &gt; 130 were chosen and subsequently each hit was subjected to membrane topology prediction using TMHMM server (Version 2.0). Only hits with an identical membrane topology were considered valid StlP homologs. The phylogenetic tree was annotated using iTOL (<span class="ExternalRef"><span class="RefSource">https://itol.embl.de/</span></span>).
+
+## Statistical analysis
+
+For statistical analyses, GraphPad Prism software (version 8.0.2) was used. Significance was determined using student’s t-test.
+
+# References
+
+1. Cabeen MT, Jacobs-Wagner C (2005) Bacterial cell shape. Nat Rev Microbiol 3:601–610
+2. Mueller EA, Levin PA (2020) Bacterial cell wall quality control during environmental stress. *mBio* 11, 10.1128/mbio. 02456–02420
+3. Daniel RA, Errington J (2003) Control of cell morphogenesis in bacteria: two distinct ways to make a rod-shaped cell. Cell 113:767–776
+4. Letek M et al (2008) DivIVA is required for polar growth in the MreB-lacking rod-shaped actinomycete *Corynebacterium glutamicum*. J Bacteriol 190:3283–3292. https://doi.org/10.1128/JB.01934-07
+5. Flärdh K (2003) Growth polarity and cell division in *Streptomyces*. Curr Opin Microbiol 6:564–571
+6. Flärdh K (2010) Cell polarity and the control of apical growth in *Streptomyces*. Curr Opin Microbiol 13:758–765. https://doi.org/10.1016/j.mib.2010.10.002
+7. Brown AD (1976) Microbial water stress. Bacteriol Rev 40:803–846
+8. Ramijan K et al (2018) Stress-induced formation of cell wall-deficient cells in filamentous actinomycetes. Nat Commun 9:5164. https://doi.org/https://doi.org/10.1101/094037
+9. Flärdh K (2003) Essential role of DivIVA in polar growth and morphogenesis in *Streptomyces coelicolor* A3(2). Mol Microbiol 49:1523–1536
+10. Fröjd MJ, Flärdh K (2019) Apical assemblies of intermediate filament-like protein FilP are highly dynamic and affect polar growth determinant DivIVA in *Streptomyces venezuelae*. Mol Microbiol 112:47–61
+11. Holmes NA et al (2013) Coiled-coil protein Scy is a key component of a multiprotein assembly controlling polarized growth in *Streptomyces*. *Proc. Natl. Acad. Sci. USA* 110, E397-406. https://doi.org/10.1073/pnas.1210657110
+12. Mukherjee P et al (2009) Novel role of Wag31 in protection of mycobacteria under oxidative stress. Mol Microbiol 73:103–119
+13. Meniche X et al (2014) Subpolar addition of new cell wall is directed by DivIVA in mycobacteria. Proc Natl Acad Sci U S A 111:E3243–3251. https://doi.org/10.1073/pnas.1402158111
+14. Walshaw J, Gillespie MD, Kelemen GH (2010) A novel coiled-coil repeat variant in a class of bacterial cytoskeletal proteins. J Struct Biol 170:202–215. https://doi.org/10.1016/j.jsb.2010.02.008
+15. Ditkowski B et al (2013) Dynamic interplay of ParA with the polarity protein, Scy, coordinates the growth with chromosome segregation in *Streptomyces coelicolor*. Open Biol 3:130006. https://doi.org/10.1098/rsob.130006
+16. Xu H, Chater KF, Deng Z, Tao M (2008) A cellulose synthase-like protein involved in hyphal tip growth and morphological differentiation in *Streptomyces*. J Bacteriol 190:4971–4978. https://doi.org/10.1128/JB.01849-07
+17. Zhong X et al (2023) CslA and GlxA from *Streptomyces lividans* form a functional cellulose synthase complex. *bioRxiv*, 2011. 2020.567928 (2023)
+18. Chaplin AK et al (2015) GlxA is a new structural member of the radical copper oxidase family and is required for glycan deposition at hyphal tips and morphogenesis of *Streptomyces lividans*. Biochem J 469:433–444. https://doi.org/10.1042/BJ20150190
+19. Zhong X, Zhang L, van Wezel GP, Vijgenboom E, Claessen D (2022) Role for a lytic polysaccharide monooxygenase in cell wall remodelling. *mBio* 13, e00456-00422. https://doi.org/https://doi.org/10.1128/mbio.00456-22
+20. Browman DT, Hoegg MB, Robbins SM (2007) The SPFH domain-containing proteins: more than lipid raft markers. Trends Cell Biol 17:394–402
+21. Langhorst MF, Reuter A, Stuermer CA (2005) O. Scaffolding microdomains and beyond: the function of reggie/flotillin proteins. Cell Mol Life Sci 62:2228–2240
+22. Lopez D, Koch G (2017) Exploring functional membrane microdomains in bacteria: an overview. Curr Opin Microbiol 36:76–84
+23. Zielińska A et al (2020) Flotillin-mediated membrane fluidity controls peptidoglycan synthesis and MreB movement. eLife 9:e57179
+24. García-Fernández E et al (2017) Membrane microdomain disassembly inhibits MRSA antibiotic resistance. Cell 171:1354–1367
+25. Huang X, Gaballa A, Cao M, Helmann JD (1999) Identification of target promoters for the *Bacillus subtilis* extracytoplasmic function σ factor, σW. Mol Microbiol 31:361–371
+26. Ma C et al (2022) Structural insights into the membrane microdomain organization by SPFH family proteins. Cell Res 32:176–189. https://doi.org/10.1038/s41422-021-00598-3
+27. Zhang L, Willemse J, Claessen D, van Wezel G (2016) P. SepG coordinates sporulation-specific cell division and nucleoid organization in *Streptomyces coelicolor*. Open Biol 6:150164. https://doi.org/10.1098/rsob.150164
+28. Lapatsina L, Brand J, Poole K, Daumke O, Lewin GR (2012) Stomatin-domain proteins. Eur J Cell Biol 91:240–245
+29. Casu B, Sallmen JW, Schlimpert S, Pilhofer M (2023) Cytoplasmic contractile injection systems mediate cell death in *Streptomyces*. Nat Microbiol 8:711–726. https://doi.org/10.1038/s41564-023-01341-x
+30. Vladimirov M, Zhang RX, Mak S, Nodwell JR, Davidson AR (2023) A contractile injection system is required for developmentally regulated cell death in *Streptomyces coelicolor*. Nat Commun 14:1469. https://doi.org/10.1038/s41467-023-37087-7
+31. Boulos L, Prevost M, Barbeau B, Coallier J, Desjardins R (1999) LIVE/DEAD® *Bac* Light™: application of a new rapid staining method for direct enumeration of viable and total bacteria in drinking water. *J. Microbiol. Methods*, 77 – 86
+32. Bach JN, Bramkamp M (2013) Flotillins functionally organize the bacterial membrane. Mol Microbiol 88:1205–1217
+33. Strahl H, Bürmann F, Hamoen LW (2014) The actin homologue MreB organizes the bacterial cell membrane. Nat Commun 5:3442
+34. Wenzel M, Vischer NOE, Strahl H, Hamoen LW (2018) Assessing membrane fluidity and visualizing fluid membrane domains in bacteria using fluorescent membrane dyes. Bio Protoc 8:e3063–e3063
+35. Müller A et al (2016) Daptomycin inhibits cell envelope synthesis by interfering with fluid membrane microdomains. *Proc. Natl. Acad. Sci. USA* 113, E7077-E7086
+36. Bramkamp M, Scheffers DJ (2023) Bacterial membrane dynamics: compartmentalization and repair. Mol Microbiol. https://doi.org/10.1111/mmi.15077
+37. Ultee E, Zhong X, Shitut S, Briegel A, Claessen D (2021) Formation of wall-less cells in *Kitasatospora viridifaciens* requires cytoskeletal protein FilP in oxygen-limiting conditions. Mol Microbiol 115:1181–1190. https://doi.org/10.1111/mmi.14662
+38. Wetzel C et al (2007) A stomatin-domain protein essential for touch sensation in the mouse. Nature 445:206–209
+39. Brand J et al (2012) A stomatin dimer modulates the activity of acid-sensing ion channels. EMBO J 31:3635–3646
+40. Takekawa N et al (2019) Structure of *Vibrio* FliL, a new stomatin-like protein that assists the bacterial flagellar motor function. *mBio* 10, 10.1128/mbio. 00292–00219
+41. Green JB, Young JPW (2008) Slipins: ancient origin, duplication and diversification of the stomatin protein family. BMC Evol Biol 8:1–12
+42. Tavernarakis N, Driscoll M, Kyrpides NC (1999) The SPFH domain: implicated in regulating targeted protein turnover in stomatins and other membrane-associated proteins. Trends Biochem Sci 24:425–427
+43. Lucena T et al (2011) *Photobacterium aphoticum* sp. nov., isolated from coastal water. Int J Syst Evol Microbiol 61:1579–1584
+44. Yokoyama H, Fujii S, Matsui I (2008) Crystal structure of a core domain of stomatin from *Pyrococcus horikoshii* illustrates a novel trimeric and coiled-coil fold. J Mol Biol 376:868–878
+45. Dannenberg N et al (2022) Mycobacteria form viable cell wall-deficient cells that are undetectable by conventional diagnostics. *bioRxiv*, 2022.2011.2016.516772. https://doi.org/10.1101/2022.11.16.516772
+46. Kieser T, Bibb MJ, Buttner MJ, Chater KF, Hopwood DA (2000) Practical Streptomyces genetics. The John Innes Foundation
+47. Stuttard C (1982) Temperate phages of *Streptomyces venezuelae*: lysogeny and host specificity shown by phages SV1 and SV2. J Gen Microbiol 128:115–121
+48. Taylor RG, Walker DC, McInnes RR (1993) E. coli host strains significantly affect the quality of small scale plasmid DNA preparations used for sequencing. Nucleic Acids Res 21:1677
+49. Kim S et al (2017) Genomic and transcriptomic landscape of *Escherichia coli* BL21 (DE3). Nucleic Acids Res 45:5285–5293
+50. Karimova G, Pidoux J, Ullmann A, Ladant D (1998) A bacterial two-hybrid system based on a reconstituted signal transduction pathway. Proc Natl Acad Sci U S A 95:5752–5756
+51. Hanahan D (1983) Studies on transformation of *Escherichia coli* with plasmids. J Mol Biol 166:557–580
+52. Fernández-Martínez LT et al (2011) A transposon insertion single-gene knockout library and new ordered cosmid library for the model organism *Streptomyces coelicolor* A3 (2). *Antonie Van Leeuwenhoek* 99, 515–522
+53. Koepff J et al (2017) Fast and reliable strain characterization of *Streptomyces lividans* through micro-scale cultivation. Biotechnol Bioeng 114:2011–2022
+54. Krasteva PV et al (2017) Insights into the structure and assembly of a bacterial cellulose secretion system. Nat Commun 8:2065. https://doi.org/10.1038/s41467-017-01523-2
+55. Larson JL, Hershberger CL (1986) The minimal replicon of a streptomycete plasmid produces an ultrahigh level of plasmid DNA. Plasmid 15:199–209
+56. Paul F Jr (2012) Tissue culture: methods and applications. Elsevier
+57. van der Aart LT, Lemmens N, van Wamel WJ, van Wezel GP (2016) Substrate inhibition of VanA by d-alanine reduces vancomycin resistance in a VanX-dependent manner. Antimicrob Agents Chemother 60:4930–4939. https://doi.org/10.1128/AAC.00276-16
+58. Kapteijn R et al (2022) Endocytosis-like DNA uptake by cell wall-deficient bacteria. Nat Commun 13:5524. https://doi.org/10.1038/s41467-022-33054-w
+59. Passot FM, Cantlay S, Flärdh K (2022) Protein phosphatase SppA regulates apical growth and dephosphorylates cell polarity determinant DivIVA in *Streptomyces coelicolor*. Mol Microbiol 117:411–428
+60. Ultee E et al (2020) Teichoic acids anchor distinct cell wall lamellae in an apically growing bacterium. Commun Biol 3:314. https://doi.org/10.1101/714758
+61. Mirdita M et al (2022) ColabFold: making protein folding accessible to all. Nat Methods 19:679–682. https://doi.org/10.1038/s41592-022-01488-1
+
+# Supplementary Files
+
+- [SupplementaryInformation.pdf](https://assets-eu.researchsquare.com/files/rs-3811693/v1/7f977a960d6056b38dcfc7b8.pdf)
+
+- [SupplementaryMovie1.avi](https://assets-eu.researchsquare.com/files/rs-3811693/v1/c14cef1e692d9efedd2ebe01.avi)  
+  Supplementary Movie S1
+
+- [SupplementaryMovie2.avi](https://assets-eu.researchsquare.com/files/rs-3811693/v1/60e759d60a76c82fc041f0ec.avi)  
+  Supplementary Movie S2
+
+- [SupplementaryMovie3.avi](https://assets-eu.researchsquare.com/files/rs-3811693/v1/6e2d5e8ac1dff148bf05cb3f.avi)  
+  Supplementary Movie S3
+
+- [SupplementaryMovie4.avi](https://assets-eu.researchsquare.com/files/rs-3811693/v1/ca9237b64044b2974b079654.avi)  
+  Supplementary Movie S4

0a341ed850ec9151cf2cfa046f9e77fd255dd63ae5c07e37348211f2cd01cb62/preprint/images_list.json

@@ -0,0 +1,74 @@
+[
+    {
+        "type": "image",
+        "img_path": "images/Figure_1.png",
+        "caption": "Multi-model median return period for precipitation in future for the 1-in-10-year precipitation based 1971-2000. a) 1.5 oC warming under RCP4.5 scenario b) 2 oC warming under RCP4.5 scenario c) 1.5 oC warming under RCP8.5 scenario d) 2 oC warming under RCP8.5 scenario.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_2.jpeg",
+        "caption": "Percentage of transportation assets facing a change of return periods. (black line: no change; orange color: 25% decrease; red color: 50% decrease)",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_3.png",
+        "caption": "Absolute global transport infrastructure exposure exceeding 25% shorten in precipitation design return periods. a) 1.5 oC warming under RCP4.5 scenario b) 2 oC warming under RCP4.5 scenario c) 1.5 oC warming under RCP8.5 scenario d) 2 oC warming under RCP8.5 scenario.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_4.png",
+        "caption": "Relative global transport infrastructure exposure exceeding 25% shorten in precipitation design return periods change. a) 1.5 oC warming under RCP4.5 scenario b) 2 oC warming under RCP4.5 scenario c) 1.5 oC warming under RCP8.5 scenario d) 2 oC warming under RCP8.5 scenario",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_5.png",
+        "caption": "Exposure exceeding 25% shorten in precipitation design return periods for four income group a) Absolute exposure under RCP4.5 scenario b) Absolute exposure under RCP8.5 scenario c) Relative exposure under RCP4.5 scenario d) Relative exposure under RCP8.5 scenario. (sky color represents multi-model median results, deep color represents the part of 2/3 models agree)",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_6.png",
+        "caption": "Ranking of countries exposure exceeding 25% change in the precipitation design return periods a) Ranking of countries in absolute exposure under RCP4.5 scenario b) Ranking of countries in absolute exposure under RCP8.5 scenario c) Ranking of countries in relative exposure under RCP4.5 scenario d) Ranking of countries in relative exposure under RCP8.5 scenario(sky color represents multi-model median results, deep color represents the part of 2/3 models agree)",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_7.png",
+        "caption": "Amplification factor to current design standards. a) 1.5 oC warming under RCP4.5 scenario b) 2 oC warming under RCP4.5 scenario c) 1.5 oC warming under RCP8.5 scenario d) 2 oC warming under RCP8.5 scenario",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/[IMAGE_METHODS_1].png",
+        "caption": "",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/[IMAGE_METHODS_2].png",
+        "caption": "",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    }
+]

0a341ed850ec9151cf2cfa046f9e77fd255dd63ae5c07e37348211f2cd01cb62/preprint/preprint.md

@@ -0,0 +1,178 @@
+# Abstract
+
+Transportation infrastructures are generally designed to have multi-decadal service lives. The design of transport infrastructure, however, is largely based on historical conditions. Yet in the face of global warming, climatic variables are likely going to have more intense and frequent extreme events, and this would put infrastructures at severe risk. In this study, we undertake a comprehensive analysis of the exposure of transportation infrastructure assets to changes in precipitation return periods globally. Under RCP8.5, 44.45%/58.64% (1.5°C/2.0°C) of the transportation assets will experience a reduction in the return period ratio of more than 25%. Eastern North America, Northern Western Europe, Central Europe and East Asia experience the highest absolute exposure to changes in precipitation return periods, mainly attributed to their dense transportation networks. We recommend considering future climate change effect during transportation infrastructure design process by introducing a design amplification factor, which is a more robust variable compared to return period change ratio.
+
+# Introduction
+
+Reliable transport infrastructure is the backbone of international trade and well-functioning economies<sup>1</sup>. Yet, the reliability of transport infrastructure is regularly under threat due to natural hazards. Globally, the multi-hazard risk due to direct damage to road and railway assets is estimated to range between 3.1 to 22 billion US dollars<sup>2</sup>. At the same time, local examples of damaged and destroyed infrastructure assets are numerous. In March 2020, for instance, heavy precipitation caused the destruction of an important bridge that connects Tanzania with Rwanda, causing disruptions in cross-border trade<sup>3</sup>. In October 2020, approximately 280 roads were destroyed as a result of tropical storms Eta and Zeta in Jamaica<sup>4</sup>, while several roads were destroyed in France and Italy due to extratropical storm Alex, causing flash floods and rainfall-induced landslides<sup>5</sup>. In July 2021, heavy rainfall caused the destruction of multiple railway segments, roads and bridges in The Netherlands, Belgium and Germany<sup>6</sup>. Similar in Zhengzhou city, which is the key transport hub in China and is one of the five major consolidation centers for China-Europe trains, was inundated by torrential downpours in July 2021, causing the cancellation of thousands of train trips<sup>7</sup>.
+
+As a result of an increasing global mean temperature rise, it is expected that climate extremes may both increase in intensity and frequency towards the future<sup>8–10</sup>. For example, rainfall reached 201.9 mm in one hour in Zhengzhou City on July 20, breaking the extreme hourly rainfall on mainland China<sup>11</sup>. This increased pressure on transport infrastructure may cause further deterioration of infrastructure assets and increased maintenance and replacement costs<sup>1</sup>. As such, building and maintaining resilient, sustainable and reliable infrastructure is one of the key targets of Sustainable Development Goal 9 (SDG9). To achieve this goal, low- and middle-income countries have to spend between 0.5 percent and 3.3 percent of their GDP annually (US $157 billion to 1 trillion) in new transport infrastructure by 2030 – plus an additional 1 percent to 2 percent of GDP to maintain their network – depending on their ambition and their efficiency in service delivery<sup>12</sup>.
+
+To successfully achieve SDG9, through building climate resilient and sustainable infrastructure, one first need to better understand the impacts of climate extremes on infrastructure assets. While much work has been done on understanding the direct economic damage on infrastructure assets<sup>2,13</sup>, and some work on the indirect economic losses through network disruptions<sup>14,15</sup>, little work has been done on understanding how design standards should change towards the future to mitigate the impacts of climate extremes on these infrastructure assets. As such, this study aims to better understand how the probability of extreme precipitation events may change towards the future, and where, and by how much, design standards should change to achieve and maintain reliable and well-functioning infrastructure.
+
+Here, we analyze the future changes in transportation exposure (roads, and railways) to precipitation around the world at 1.5 and 2°C warmer worlds, by employing multi-model projections from the Coupled Model Intercomparison Project Phase 5 (CMIP5). To do so, we first estimated precipitation return period shifts under different global circulation models (GCMs) and then investigate how this change influences transportation exposure by incorporating the design standards of infrastructure drainage system. We show that roughly 93% of the global land mass is projected to have a decrease in return period (increased frequency). Greenland, Northeastern North-America, Northern South-America, Central Africa, Central Siberian plateau, Central India, Southwest and Northeast China, and Southeast Asia, are regions experience some of the most pronounced precipitation return period changes. Nearly 97% of global transportation assets will face a shorter precipitation return periods under Representative Concentration Pathway 4.5 (RCP4.5) scenario, among them, 36.37%/53.81% (1.5°C/2.0°C) of assets will experience a shorten return period exceeding 25% (e.g. what used to be 1-in-10-yr event for the period 1971–2000 is less than 1-in-7.5-yr in the corresponding simulation). The United States, China, and Russia experience the largest absolute exposure given the large quantity of transportation assets, while Small Island Developing Countries are more sensitive to climate change in terms of relative exposure. Canada, Finland, Norway, South Korea, Kazakhstan and Japan rank highest in both absolute and relative exposure, reflecting the dual risk of assets concentration and climate change.
+
+Our study reveals the magnitude of the threat to transportation infrastructures from climate change at the global scale. The results demonstrate that realizing the 1.5°C low warming target would robustly reduce the transportation exposures to precipitation extremes, compared to a warming of 2°C. The avoided impacts are more remarkable for Northern South-America, Central Africa, Central India, Southwest China, and Southeast Asia. Detailed regional differenced amplification factors are suggested to the designed precipitation for transportation drainage system to cope with a changing climate.
+
+# Results
+
+## 2.1. Change in precipitation return period
+
+Figure 1 shows the global spatial distribution of how the probability of a 1-in-10-year precipitation event (an event that has an average of 10% chance of occurring every year) in the baseline situation (1971-2000) changes at 1.5 °C and 2 °C degrees of global warming under RCP4.5 and RCP8.5 scenarios. Roughly 93% of the global land mass experiences a decreasing return period (increased frequency) as a result of global warming. An increasing return period is primarily observed in Northern Africa, South west of South America, and Saudi Arabia. Under RCP4.5 scenario, ~72.2% (1.5 °C) and 81.8% (2 °C) of the world is projected to have a decrease in return periods of up to 25% (i.e. what used to be 1-in-10-yr event for the period 1971-2000 is less than 1-in-7.5-yr in the corresponding simulation). Under RCP 8.5 scenario, the numbers increase to 70.4% and 83.4%, for respectively 1.5 °C and 2 °C. Greenland, Northeastern North-America, Northern South-America, Central Africa, Central Siberian plateau, Central India, Southwest and Northeast China, and Southeast Asia, are regions most sensitive to global warming and will face the most significantly shortened return period of precipitation when the global mean temperature increase rises from 1.5 °C to 2.0 °C.
+
+## 2.2. Exposure Analysis
+
+Infrastructures are most often designed to be able to resist a certain return period rainfall event. When the probability of the rainfall event exceeds this design threshold, damage is expected to occur. In other words, the infrastructure drainage system is expected to protect the infrastructure against rainfall events with return periods smaller than the design return period. Under global warming scenarios, the designed return period based on the historical precipitation record may become shorter, putting infrastructures at greater risk. For instance, an infrastructure was initially designed for 10-year rainfall event, in the future it may become 5-year rainfall event based on historical probability distribution function (PDF), making the infrastructure less reliable as one might anticipate. In this study, we investigate the change of infrastructure design return period under different warming scenarios. Different design standards (i.e., design return periods) of drainage system are assigned for countries with different income groups as well for different assets (See Method and Table S1). Figure 2 shows the percentage distribution of transportation infrastructure assets facing a change in design return period. The design return period of nearly 97% of global transportation assets will become shorter (between 0% and 50%) relative to the historical period 1971–2000, with an average decrease of 23.8%/28.35% (1.5 °C/2 °C, mean of RCP4.5 and RCP8.5) and a standard deviation of 13.7%/14.75% (1.5 °C/2 °C, mean of RCP4.5 and RCP8.5). The percentage distribution under 2.0 °C warming shifts to the left compared to that of 1.5 °C warming situation, especially the 0-25% interval moves significantly towards the 25%-50% interval, and the proportion above 50% change is increasing as well. Under RCP4.5 scenario, 36.37%/53.81% (1.5 °C/2.0 °C) of assets will face a shorten return period exceeding 25%. While under RCP8.5 scenario, the two proportions are, respectively, 44.45% and 58.64% (1.5 °C/2.0 °C).
+
+The exposure of infrastructure assets to changes in precipitation is affected by two factors: the spatial distribution of assets and the change in the design return period at a given location. Figure 3 shows the spatial distribution of absolute exposure of global transportation infrastructure, which is expressed by the sum of all road and railway assets facing a more than 25% decrease in precipitation design return period within a grid of 25x25km (See Method). The individual exposures for each road and railway asset categories can be found in Supplementary Figure 1. The distributions of absolute exposure are similar under RCP4.5 scenario and RCP8.5 scenario. Under RCP4.5 scenario, 8.4 million km (1.5 °C) and 11.4 million km (2.0 °C) of global road and railway assets will be exposed to more frequent extreme precipitation (i.e., a reduction ratio in the design return period of more than 25%), representing 39% and 53% of the global land transport infrastructure. Among these assets, ~24% of the exposed transportation assets are secondary roads, followed by primary roads and railways, both account for ~16%. Under RCP8.5 scenario, the exposed transportation assets increase to 9.8 million km (1.5 °C) and 12.6 million km (2.0 °C), accounting for 45% (1.5 °C) and 59% (2.0 °C) of global transportation assets. Regions that experience high absolute exposure are profoundly in Eastern North America, Northern Western Europe, Central Europe and East Asia, mainly attributed to their dense transportation networks.
+
+Figure 4 shows the relative global transport infrastructure exposure exceeding a 25% reduction in precipitation design return periods. Under RCP4.5 scenario, the grids with an exposure ratio larger than 80% (see Methods) account for ~25.9% (1.5 °C) and ~39.9% (2.0 °C) of the total exposed grids. We find that risk in relative exposure varies widely across the world. Under RCP8.5 scenario, the percentage of grids with an exposure ratio larger than 80%, increases to ~30.4% (1.5 °C) and ~46.6% (2.0 °C). These highly exposed areas are concentrated in the Eastern and Western United States, Southeastern South-America, Central and Northern Europe, Southwestern China, and Southeast Asia. The highly exposed regions expand to Central Africa, Central India and Southeastern Australia. Interestingly, we find some countries that experience a low absolute exposure, but a high relative exposure. As for example, ~61% of grids in Myanmar (total transportation assets of 54,090 km) will face an exposed ratio over 80% under 2.0 °C under RCP8.5 scenario, making a large proportion of transportation infrastructure in Myanmar highly vulnerable to rising temperatures.
+
+Due to the high uncertainty in climate models, we present the multi-model median results and the part with high agreement (defined as more than ~66% climate models supporting the change of exceeding 25% shorten in precipitation return periods<sup><span citationid="CR16" class="CitationRef">16</span>, <span citationid="CR17" class="CitationRef">17</span></sup>). Figure 5 shows the absolute and relative exposure of the four World Bank income groups<sup><span citationid="CR18" class="CitationRef">18</span></sup>. Between 1.5 °C to 2.0 °C, the exposure of the four income groups increases by 1.2-2.7 times. High income countries and upper middle income countries experiences the largest increases in absolute exposure, which is expected as they have dense transport infrastructure and higher design standards of their assets. According to the shape of the PDF of the return period T, a small change in precipitation can lead to a large change in design return period for high design standards. Under RCP8.5, with a 2 °C warming scenario, high income countries may experience an absolute exposure of approximately 7.1 million km, which is around 73% of all infrastructure assets in these countries. The upper middle income countries come second, with approximately 4.8 million km of asset length. At the global perspective, considerably less exposure is found in lower middle and low middle income countries due to the significantly lower amount of infrastructure assets. However, the differences appear to be smaller in relative exposure compared, with 22% in relative exposure for low and low middle income countries while 58% and 73% in upper middle income and high income countries respectively. The smaller relative exposure changes in low and lower middle income countries are due to the lower design standards. The assets in these countries were already severely exposed to precipitation events with low return periods. As such, they are relatively less affected by the estimated changes in this study. Of course, this does not mean their transportation assets are less threatened by global warming, on contrary, they are affected by all rainfall events in excess of their design return period. Compared to high return periods, lower return periods require larger change in precipitation to realize the same proportion change in return period. When we increase the design standards (Table S1), the relative exposure increases to 39.3% for low and low middle income countries.
+
+Figure 6 shows the top-30 countries with the highest absolute and relative exposure under RCP4.5 scenario and RCP8.5 scenario. The most exposed countries can be found in Asia, Europe and America. The United States (US) ranks first in absolute exposure due to its high density of transport infrastructure assets and decreasing precipitation return periods, especially along the East and West coast. Under RCP4.5 scenario, the amount of US exposed assets will reach 1.82 (1.5 °C) and 2.36 (2.0 °C) million km (with a high agreement of 0.77 (1.5 °C)/1.48 (2.0 °C) million km), which accounts for 57% (1.5 °C) and 74% (2.0 °C) of its total assets (with a high agreement of 24% (1.5 °C) /46% (2.0 °C)). China with 1.28 (1.5 °C) and 1.75 (2.0 °C) million km of exposed roads, ranks second at absolute exposure (with a high agreement of 0.68 (1.5 °C)/1.22 (2.0 °C) million km), which accounts for 50% (1.5 °C) and 68% (2.0 °C) of its assets (with a high agreement of 26% (1.5 °C)/47% (2.0 °C)). A total of fourteen European countries can be found in the top-30, with Russia in third place globally due to its large number of assets, followed by France and Germany. Russia’s exposed assets volume will reach 0.88 (1.5 °C) and 1.18 million km (2.0 °C) (with a higher agreement of 0.51/0.85 million km), which accounts for 52% (1.5 °C) and 70% (2.0 °C) of its assets (with a higher agreement of 30% (1.5 °C)/50% (2.0 °C)). There are nine Asian countries in the top30, including China, Japan and India, etc. A rise of 1.5 °C to 2.0 °C has the greatest impact on Romania, Japan and Brazil, resulting in a 253%, 108% and 101% increase in absolute exposure.
+
+Small countries are more sensitive to climate change in relative exposure. Under RCP4.5 scenario, eight countries with a total asset length of less than 5000km, show a high ratio of exposure and high consistency (up to 100%) on all climate models, such as some of the French Overseas Territories (e.g. Guiana, Mayotte). A total of thirteen European countries made the list, with Latvia showing the highest relative exposure, ranking third globally, up to 92.2% (1.5°C) and 94.5% (2.0°C) (with a higher agreement of 76% (1.5°C) and 88% (2.0°C)). Four Asian countries —South Korea, Kazakhstan, Japan and Qatar—made the list. Canada, Finland, Norway, South Korea, Kazakhstan and Japan are both on the absolute and relative exposure lists, reflecting the dual risk of assets concentration and climate change.
+
+The thirty countries that experience the highest absolute exposure for RCP8.5 and RCP 4.5 are very similar, with 29 countries overlapping. Under RCP8.5 scenario, the exposed assets of these 29 countries increase by 0.034 million km on average for both 1.5°C and 2°C (with a higher agreement of 0.017 (1.5°C) and 0.025 (2°C) million km). We find twelve countries exhibit high risk in both absolute and relative exposure, including the United States, Canada, Great Britain, Germany, Finland, and Thailand.
+
+## 2.3. Design amplification factor
+
+Most of transportation infrastructures are designed and built based on the assumption that precipitation will resemble historical patterns<sup><span citationid="CR19" class="CitationRef">19</span></sup>. As shown in our study, change of precipitation in a warmer world would make infrastructures assets in most regions of the world more exposed than it used to be, resulting in a potential decrease in network resilience and reliability due to increased probabilities of network disruptions through damaged assets. We therefore recommend using a design amplification factor for upgrading existing infrastructures or designing new infrastructures to make them maintain the same level of acceptable risk under future warming scenarios. Figure 7 shows the spatial distribution of the amplification factor (see Methods) to current design standards, which represents the ratio of the future precipitation intensity corresponding to the design return period of various transportation assets to the precipitation intensity in the baseline period. For example, the design amplification factor is 1.5 if the precipitation intensity of the same design return period is 20mm and 30 mm for current state and future scenario, respectively. We find that the design amplification factor is mainly distributed between 1 and 1.2 around the world. Under RCP4.5, 86.0% (1.5 °C) and 82.5% (2.0 °C) of the global mass has an amplification factor of 1-1.2; 1.2% (1.5 °C) to 4.8% (2.0 °C) of the global mass has an amplification factor larger than 1.2, which is mainly concentrated in Tibet, Central India and the Andes (South America). Under RCP8.5, 84.6% (1.5 °C) and 76.9% (2.0 °C) of the global land mass has an amplification factor of 1-1.2; 2.0% (1.5 °C) and 9.2% (2.0 °C) of the global land mass has an amplification factor of over 1.2, mainly occurs in India, Southwest China and Indo-China Peninsula, East Africa, Andes of South America and north of 50°N.
+
+We also found that areas with a high amplification factor do not coincide completely with areas with a high degree of return periods shortening (Section 2.1). This again implies that a small increase of precipitation may not make a large difference in an area where design standards for infrastructure are high, and where we observe a large change in design return periods. We also perform a sensitivity analysis to assess the change of design amplification factor while considering uncertainties in design standards (see Methods). With higher design standards assumption, under RCP4.5, 84.5%/ 82.4% (1.5 °C/2.0 °C) of the global mass has an amplification factor between 1 and 1.2; 1.1% /3.9% (1.5 °C/2.0 °C) of the world land area has an amplification factor larger than 1.2. This indicates that the dynamic amplification factor is more robust in guiding engineering design compared to return period change ratio. Considering the uncertainty in the tail of the PDFs of return period, using an amplification factor to current design precipitation intensity would be more feasible and easier.
+
+The design amplification factor achieves a reasonably estimates for the future. Figure 7 shows that a design amplification factor of 1.2, although approximate, works sufficient for most regions of the world for a quick calculation in the design process (86.0%/76.9(1.5 °C/2.0 °C, RCP8.5). We should encourage and accept infrastructure managers or engineers to consider future climate extremes in the design process. By considering a design amplification factor, the transportation assets are expected to maintain the same risk level in the future as they were designed for. Climate researchers, construction designers, and stakeholders should work together to make infrastructures reliable in a warming world.
+
+# Discussion And Conclusion
+
+In this study, we have analyzed the exposure of global transportation assets to changes in precipitation design return periods under different future warming scenarios. The change of extreme precipitation return period challenges the adaptability of design return periods of infrastructure drainage facilities.
+
+Our results show that about 97% of global transportation assets will be exposed to precipitation with higher frequency return periods, with increased frequencies between 0 and 50% for both RCP4.5 and RCP8.5 scenarios. Under RCP4.5 scenario, a total of 8.4 million km (1.5 °C) and 11.4 million km (2.0 °C) of global transportation assets will be exposed to more frequent (frequency change larger than 25%) extreme precipitation. In relative terms, 25.9% (1.5 °C) and 39.9% (2.0 °C) of grids’ relative exposure ratio is over 80%. Under RCP8.5, exposure reaches up to 9.8 million km (1.5 °C) and 12.6 million km (2.0 °C), with the proportion of highly exposed grids proportion increasing to 30.4% (1.5 °C) and 46.6% (2.0 °C), compared to RCP4.5. Regions that experience the highest absolute exposure regions are distributed among Eastern North America, Northern Western Europe, Central Europe and East Asia. This can primarily be explained by their high density of transportation infrastructure (Meijer et al., 2018). Among these regions, the US experiences the largest exposure due to its high asset distribution overlapping with areas with shortened return periods (on both the East and West coast).
+
+A design amplification factor applied to design precipitation intensity determined by historical stationary records provides a practical and useful tool for approximating climate change effect. A change of 1.2 obtains a good approximation of the climate change effect for most regions of the world, which provides a practical and useful tool in design over a future planning horizon. In particular, India, Southwest China and the Indo-China Peninsula, East Africa and the Andes in South America experience higher amplification factors. The amplification factor is less sensitive to uncertainties in design return periods compared to the variations in precipitation design return period, i.e., the value of amplification factor maintains 1–1.2 for most regions of the world regardless the change of design standards. The results of this study also highlight that low income countries are more susceptible to damage (due to lower design standards), but in general they face a relative smaller absolute change in return periods.
+
+While this study did not assess the feasibility of adaptation measures to reduce the impacts, several measures are possible to reduce the damage to roads and railways, of which updating drainage systems is the most prominent one. Updating or implementing a road or railway drainage system, however, will substantially increase the cost of required infrastructure for a specific location. According to a case study of a 1.7 km² street in China, it is estimated that about 0.7 million USD was need to improve its drainage capacity from 1-in-2-year to 1-in-5-year²⁰. To assess whether it is cost-effectivity to implement, a cost-benefit analysis is required. However, it should be emphasized that benefits will be much larger if we consider not only the direct damages, but also the indirect loss caused by transportation disruption.
+
+To conclude, our results help to identify highest risk assets and facilitate decision making about which and where assets to prioritize upgrading drainage system, based on the potential increase in future exposure. We emphasize that uncertainties are a key element in designing resilient infrastructure in a future climate. In particular in the case of infrastructure, which generally has a long lifespan, it is essential to incorporate potential future changes in exposure to keep infrastructures reliable over multi-decadal service lives with growing climatic uncertainty.
+
+# Methods
+
+Future extreme precipitation changes are assessed through two aspects: frequency and intensity. To quantify the transportation assets exposed to extreme precipitation, we use return periods to describe the change between the 1.5°C and 2.0°C warming and the baseline period (1971–2000). A return period indicates the recurrence interval, which is an average time period between two events to occur. For example, the return period of precipitation of 10 years means its probability of occurring being 1/10, or 10% in any one year. To determine the precipitation intensity of different return periods, we fit a distribution function of the annual maximum daily precipitation (RX1D) for the baseline and future time periods<sup>21,22</sup>. RX1D is defined as the annual maxima of daily precipitation amount per year. We fit the RX1D-return period distribution for both historical condition and future warming scenarios. Using these distributions, we estimate the precipitation intensity corresponding to the return period of contemporary asset drainage design into the future distribution function to calculate the return period under the same precipitation intensity. Based on this, we develop an exposure analysis to calculate the amount of assets facing significant return period changes. To describe how precipitation intensity under the contemporary design return period would change in the future, we introduce a design amplification factor to express the ratio of precipitation intensity in the return period of contemporary design to that in the same return period in the future. Figure S1 gives the methodology.
+
+## Base design return period for different infrastructure
+
+Considering that countries with different development levels have different design drainage capabilities of transport infrastructures, we divide the countries into four categories — High income, Upper middle income, Lower middle income and Low income—based on the four levels of development standards of World Bank<sup><a href="https://datahelpdesk.worldbank.org/knowledgebase/articles/906519">https://datahelpdesk.worldbank.org/knowledgebase/articles/906519</a></sup>. According to Kuznet’s theory<sup>23</sup> and the studies of Koks et al.<sup>2</sup>, low income countries usually can’t invest much in infrastructure construction, nor guarantee the expenditure on disaster recovery capacity; Lower middle income countries can invest relatively more in infrastructure construction but also can’t invest much on disaster recovery capacity; Upper middle income and high income can invest much in both infrastructure construction and disaster recovery capacity, with higher investment in high income countries. Accordingly, we assign different design return periods of drainage system for countries with different income groups (See Table S1).
+
+While there are differences between countries with regards to design standards, the same can be said about different assets: different infrastructure types have different design drainage capabilities. In this study, we have divided the assets into three types: (1) Railway; (2) Motorway/Trunk/Primary/Secondary; and (3) Tertiary. Railways tend to be the most strictly designed infrastructure, followed by Motorway/Trunk/Primary/Secondary, which are usually built using similar engineering design standards<sup>24–28</sup>. The third type, tertiary roads, tends to have uneven quality and may not have professional drainage<sup>29,30</sup>. The assumptions of design return periods of drainage system of different types of transportation infrastructure are provided in Table S1. We assumed two different design return periods levels: a higher design return periods and the lower return periods.
+
+## Fitting function for calculating precipitation under different return periods
+
+To obtain the daily precipitation of each grid in different return periods, the generalized extreme value distribution (GEV) is used. The GEV distribution is parameterized with three parameters including location (μ; describing the center of distribution), scale (σ; describing the deviation around the mean) and shape (ξ; describing the tail behavior of the distribution)<sup>31</sup>. We can fit a sample of extremes to the GEV distribution to obtain the parameters that best explains the probability distribution of the extremes. In this study, the GEV parameters were estimated by fitting annual maximum daily precipitation series for each grid, and the GEV parameters were spatially smoothed by averaging the 8 adjacent grids around it. Kolmogorov-Smirnov (KS) test was used to test the fitting degree of the GEV distribution at the significance level of 5%<sup>32</sup>. L-Moments method was used to estimate GEV distribution parameters (based on the python package “lmoments3”,<sup><a href="https://github.com/OpenHydrology/lmoments3">https://github.com/OpenHydrology/lmoments3</a></sup>). The data of 30 years was used for both contemporary and future, and the 30 years data was taken from 15 years before and after the warming arrival year. Spatial smoothing is used to deal with the noise in the models of extreme precipitation variability from the sample, and similar approaches can be found in references<sup>21,33</sup>.
+
+## Estimation of exposure
+
+We calculated the exposure of the assets at a grid scale. Absolute exposure (AE) describes the total amount of assets exposed to "more frequent extreme precipitation" within a grid, expressed as a 25% decrease in the design return period in the future. Relative exposure (RE) expresses the proportion of assets exposed to a "significant frequency change" within a grid, 25% decrease in the design return period in the future. AE and RE can be adapted as eq. 1 and eq. 2:
+
+[IMAGE_METHODS_1]
+
+where AE<sub>g</sub> is the absolute exposure of grid g; RE<sub>g</sub> is the relative exposure of grid g; i is the type of asset, with a total of n types; EL<sub>g</sub> is the length of the assets exposed to a "significant frequency change" within the grid g; TL<sub>g</sub> is the total length of the assets within the grid g.
+
+## Estimation of amplification factor
+
+We used the amplification factor to describe the ratio of future extreme precipitation to contemporary extreme precipitation, as expressed by eq.3:
+
+[IMAGE_METHODS_2]
+
+where AMF<sub>g</sub> is the amplification factor of the grid g; i is the type of asset, with a total of n types; Rx1d<sub>f,g</sub> is the future annual maximum daily precipitation in grid g; Rx1d<sub>p,g</sub> is the contemporary annual maximum daily precipitation in grid g.
+
+## Data
+
+### Climate data
+
+Downscaling data from NASA were used in this study, which is based on the 21 models of CMIP5. The BCSD (Bias Correction and Spatial Disaggregation) methods were used to improve the resolution to ~25km. More information is available on<sup><a href="https://nex.nasa.gov/nex/projects/1356/">https://nex.nasa.gov/nex/projects/1356/</a></sup>. Compared with the coarse resolution of CMIP5, NEX-GDDP data shows stronger ability to simulate extreme precipitation.
+
+The timings arriving 1.5°C and 2.0°C warmer world were defined as the year in which the global average temperature first rise to the corresponding temperature compared to the period of the Industrial Revolution (1861–1890). Following this definition, we calculated the year of arrival for each model using CMIP5 mean temperature data. The models we used and the temperature arrival years are provided in Table S2.
+
+We use the multi-model median projected changes to express the spatial distribution, which is a relatively robust way to prevent the overall results from being affected by one model estimation being too high or too low, and make sure that at least half of the models support changes in the same notation.
+
+### Infrastructure data
+
+Roads and railways were chosen for our study as the main land transport infrastructure. We extracted these data as of March 2020 from OpenStreetMap (OSM). OSM is a large and open global infrastructure database that is now at a high level of completion, which has been widely used in various research<sup>2,34,35</sup>. Based on OSM's classification of roads, we chose the most important five categories of roads, i.e. motorway, trunk, primary, secondary and tertiary, which is principal connectors of the cities, towns and villages. Other roads, such as footway, were not considered as it is often uncertain whether drainage design has been considered when they were built. Table S3 gives the description and length of different categories of transportation assets and their spatial distribution is given in Fig.S2.
+
+# References
+
+1. Hallegatte, S., Rentschler, J. & Rozenberg, J. *LIFELINES: The resilient infrastructure opportunity*. (World Bank Publications, 2019).
+
+2. Koks, E. E. *et al.* A global multi-hazard risk analysis of road and railway infrastructure assets. Nature communications **10**, 1–11 (2019).
+
+3. The New Times. *Relief as traffic resumes on Tanzania highway*, <https://www.newtimes.co.rw/news/relief-traffic-resumes-tanzania-highway> (2020).
+
+4. FloodList. *Jamaica – Over 280 Roads Damaged by Floods From Tropical Storms Eta and Zeta*, <https://floodlist.com/america/jamaica-flood-damage-eta-zeta-november-2020> (2020).
+
+5. BBC News. *Storm Alex: Buildings, roads and bridges destroyed in southern France*, <https://www.bbc.com/news/av/world-europe-54404754> (2020).
+
+6. The Guardian. *Flash floods cause havoc across Europe – in pictures*, <https://www.theguardian.com/environment/gallery/2021/jul/15/flash-floods-cause-havoc-across-europe-in-pictures> (2021).
+
+7. Global Times. *Eurasian train resumes as flood recedes in Zhengzhou*, <https://www.globaltimes.cn/page/202107/1229670.shtml> (2021).
+
+8. Seneviratne S.I. *et al.* Weather and Climate Extreme Events in a Changing Climate. (2021).
+
+9. Easterling, D. R. *et al.* Climate extremes: observations, modeling, and impacts. *science* **289**, 2068–2074 (2000).
+
+10. Fischer, E. M., Beyerle, U. & Knutti, R. Robust spatially aggregated projections of climate extremes. Nature Climate Change **3**, 1033–1038 (2013).
+
+11. CCTV news. *24 hours 622.7mm! Record rainfall has broken in Zhengzhou, Henan Province*, <https://web.archive.org/web/20210721125656/https://www.thepaper.cn/newsDetail_forward_13673710> (2021).
+
+12. Rozenberg, J. & Fay, M. *Beyond the gap: How countries can afford the infrastructure they need while protecting the planet*. (World Bank Publications, 2019).
+
+13. Van Ginkel, K. C., Dottori, F., Alfieri, L., Feyen, L. & Koks, E. E. Flood risk assessment of the European road network. Natural Hazards and Earth System Sciences **21**, 1011–1027 (2021).
+
+14. Colon, C., Brännström, Å., Rovenskaya, E. & Dieckmann, U. Fragmentation of production amplifies systemic risks from extreme events in supply-chain networks. Plos one **15**, e0244196 (2020).
+
+15. Colon, C., Hallegatte, S. & Rozenberg, J. Criticality analysis of a country’s transport network via an agent-based supply chain model. Nature Sustainability **4**, 209–215 (2021).
+
+16. Power, S. B., Delage, F., Colman, R. & Moise, A. Consensus on twenty-first-century rainfall projections in climate models more widespread than previously thought. Journal of Climate **25**, 3792–3809 (2012).
+
+17. Solomon, S. *et al.* Climate change 2007: The physical science basis. *Intergovernmental Panel on Climate Change (IPCC), Cambridge University Press, Cambridge* (2007).
+
+18. World Bank. *New World Bank country classifications by income level: 2020-2021*, <https://blogs.worldbank.org/opendata/new-world-bank-country-classifications-income-level-2020-2021> (2020).
+
+19. Almansour, H., Cusson, D. & Lounis, Z. Climate resilient bridges-state of practice. in *10th International Conference on Short and Medium Span Bridges*.
+
+20. Yang, Y., Li, M., Xiong, M. & Cao, J. Return period for urban rainwater drainage networks based on the lowest total social investment method: a case study in Tianjin, China. Polish Journal of Environmental Studies **28** (2019).
+
+21. Zhang, W., Zhou, T., Zou, L., Zhang, L. & Chen, X. Reduced exposure to extreme precipitation from 0.5oC less warming in global land monsoon regions. Nature Communications **9**, 1–8 (2018).
+
+22. Hirabayashi, Y. *et al.* Global flood risk under climate change. Nature climate change **3**, 816–821 (2013).
+
+23. Stern, D. I. The rise and fall of the environmental Kuznets curve. World development **32**, 1419–1439 (2004).
+
+24. GYMREIG, Y. S. & IRELAND, N. Introduction to the Design Manual for Roads and Bridges. (1997).
+
+25. Spink, T., Duncan, I., Lawrance, A. & Todd, A. Transport infrastructure drainage: condition appraisal and remedial treatment. *CIRIA*, *London* (2014).
+
+26. Sañudo, R., Miranda, M., García, C. & García-Sanchez, D. Drainage in railways. Construction and Building materials **210**, 391–412 (2019).
+
+27. Fei Yueying & Yang, Y. Study on the determination of rainfall return period of railway subgrade drainage facilities. Subgrade Engineering **3**, 100–101 (2009).
+
+28. Han Liu & Weiping, T. Investigation on design of rainfall reappering period in highway drainage facilities. Technology of Highway and Transport, 22–24 (2009).
+
+29. JICA. Preparatory Survey for Road Network Development Project in Conflict-Affected Areas in Mindanao. (Republic of the Philippines Department of Public Works and Highways, 2018).
+
+30. White, S. J. Maintenance and Control of Erosion and Sediment Along Secondary Roads and Tertiary Trails. (Construction Engineering Research Lab (Army) Champaign Il, 1997).
+
+31. Hosking, J. R. M., Wallis, J. R. & Wood, E. F. Estimation of the generalized extreme-value distribution by the method of probability-weighted moments. Technometrics **27**, 251–261 (1985).
+
+32. Massey Jr, F. J. The Kolmogorov-Smirnov test for goodness of fit. Journal of the American statistical Association **46**, 68–78 (1951).
+
+33. Kharin, V. V. & Zwiers, F. W. Estimating extremes in transient climate change simulations. Journal of Climate **18**, 1156–1173 (2005).
+
+34. Meijer, J. R., Huijbregts, M. A., Schotten, K. C. & Schipper, A. M. Global patterns of current and future road infrastructure. Environmental Research Letters **13**, 064006 (2018).
+
+35. Barrington-Leigh, C. & Millard-Ball, A. The world’s user-generated road map is more than 80% complete. PloS one **12**, e0180698 (2017).
+
+# Supplementary Files
+
+- [Supplementaryfile.docx](https://assets-eu.researchsquare.com/files/rs-1285854/v1/1cd6837b44d56813835e9364.docx)  
+  SUPPLEMENTARY INFORMATION

0ab9485f323a84a87f5e9b35a5af0f9f487ed9784339ca91bceacdf95cc00a39/preprint/images_list.json

@@ -0,0 +1,66 @@
+[
+    {
+        "type": "image",
+        "img_path": "images/Figure_1.png",
+        "caption": "Chronic restraint stress (CRS) induces some anxiety phenotypes and robust depression-like behaviors in mice. (A) Experimental strategy and timeline for behavioral assays. (B)-(H) Effects of CRS on behavioral performance in different tests. Since CRS caused similar behavioral changes in the two mouse strains used, we pooled the data for statistical analysis (see Figures S1 and S2 for data in each mouse line and each sex). (B) Open field test: total distance travelled (left; t(38) = 1.507 and p = 0.140) and time in the center zone (right; t(38) = 1.203 and p = 0.236). (C) Light-dark box transition test: latency to the first entry into the dark area (left; p = 0.654) and time spent in the dark area (right; t(38) = 2.152 and p = 0.038). (D) Elevated zero maze test: latency to the first entry into open sections (left; p = 0.008) and time in the open sections (right; t(38) = 5.129 and p = 8.88\u2009\u00d7\u200910\u22126). (E) Forced swimming test: immobility time (t(38) = 7.599 and p = 3.90\u2009\u00d7\u200910\u22129). (F) Tail suspension test: immobility time (t(38) = 6.693 and p = 6.43\u2009\u00d7\u200910\u22128). (G) Sucrose preference test: preference index (p = 7.82\u2009\u00d7\u200910\u22125). (H) Total grooming time (p = 2.77\u2009\u00d7\u200910\u22123). Different cohorts of mice were used in (B-F, H) and (G). n = 10 D1-tdTomato/D2-EGFP mice (circle) and 10 D3-Cre/tdTomato mice (square) in (B-F), (H) and n = 7 per mouse line in (G). Data are expressed as mean\u00b1SEM. Student\u2019s two-tailed unpaired t test for (B) and time in (C)-(F); Mann-Whitney test for (latency in C, D) and (G-H). *p<0.05, **p<0.01, ****p<0.0001; ns, not significant.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_2.png",
+        "caption": "Chronic restraint stress (CRS) significantly decreases neuronal excitability of OT D3 neurons, but not D1/D2 SPNs. (A) and (E) Representative images from coronal sections containing the olfactory tubercle (OT) showing D1-tdTomato and D2-EGFP SPNs (A) as well as D3-Cre/tdTomato neurons (E). Left, low-magnification images from coronal brain sections. Scale bar: 1 mm. Right, enlarged images from corresponding rectangle areas in the left images. Scale bars: 50 \u03bcm (A) and 20 \u03bcm (E). (B) and (F) Representative whole-cell patch-clamp recordings showing current injection-induced firing of D1/D2 SPNs (B) and D3 neurons (F) from controls (black) and CRS-treated (red) mice. (C) and (G) Input resistance of D1/D2 SPNs (C) and D3 neurons (G). D1/D2 SPNs: t(58) = 0.315 and p = 0.754; n = 40 and 20 for control and CRS, respectively. D3 neurons: t(28) = 0.552 and p = 0.585; n = 15 per group. (D) and (H) Firing frequency of D1/D2 SPNs (D) and D3 neurons (H) upon current injections from controls and CRS-treated mice. D1/D2 SPNs: treatment, F(1, 55) = 0.870 and p = 0.355; current, F(2, 96) = 87.937 and p < 0.1\u00d7\u200910\u22127; treatment x current, F(8, 436) = 0.637 and p =0.747; n = 40 and 17 for control (5 mice) and CRS (4 mice), respectively. D3 neurons: treatment, F(1, 18) = 10.289 and p = 0.005; current, F(3, 48) = 30.674 and p < 0.1\u00d7\u200910\u22127; treatment x current, F(7, 126) = 6.451 and p = 1.68\u00d7\u200910\u22125; n = 10 per group (5 mice each group). Holding membrane potential = -60 mV for all recordings. Data are expressed as mean\u00b1SEM. Student\u2019s two-tailed unpaired t tests for (C) and (G). Two-way ANOVA for (D) and (H). ****p<0.0001; ns, not significant.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_3.png",
+        "caption": "Ablation of OT D3 neurons induces depression-like behaviors. (A) Left, schematic showing the viral injection strategy. The AAV8-TurboRFP (control) or Cre-dependent AAV8-mCherry-FLEX-DTA virus (800 nl) was bilaterally injected into the OT. Compared to the control virus (middle), the DTA virus ablated OT D3 neurons as visualized by the absence of most D3-EYFP signals (right). Confocal images were collected 4 weeks post viral injection. Scale bars = 200 \u03bcm. (B) Experimental strategy and timeline of behavioral assays. (C-H) Effects of ablation of OT D3 neurons on behavioral performance in different tests. (C) Open field test: total distance travelled (left; t(12) = 0.487 and p = 0.635) and time in the center zone (right; t(12) = 2.682 and p = 0.020). (D) Light-dark box transition test: latency to the first entry into the dark area (left; t(12) = 1.767 and p = 0.103) and time in the dark area (right; t(12) = 1.244 and p = 0.237). (E) Elevated zero maze test: latency to the first entry into open sections (left; t(12) = 1.860 and p = 0.088) and time in open sections (right; t(12) = 1.607 and p = 0.134). (F) Forced swimming test: immobility time (t(12) = 2.200 and p = 0.048). (G) Tail suspension test: immobility time (t(12) = 2.854 and p = 0.015). (H) Sucrose preference test: preference index (t(12) = 0.453 and p = 0.658). Two different cohorts of mice were used in (C)-(G) and (H). n = 7 mice per group. Data are expressed as mean\u00b1SEM. Student\u2019s two-tailed unpaired t tests. *p<0.05; ns, not significant.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_4.png",
+        "caption": "Chemogenetic inhibition of OT D3 neurons induces anxiety- and depression-like behaviors in mice. (A) Schematic showing strategy for viral injection into the OT and timeline for behavioral assays under inhibitory DREADD manipulations. (B) Ex vivo electrophysiological recordings on KORD-expressing D3 neurons. Left, representative traces showing current injection-induced firing under bath application of ACSF or SALB (in DMSO; 10 \u03bcM). Right, comparison of the firing frequency between ACSF and SALB condition (t(4) = 12.728 and p = 2.20\u00d7\u200910\u22124). n = 5 neurons from 3 mice. (C-I) Effects of inhibition of D3 neurons on behavioral performance in different tests. (C) Open field test: total distance travelled (left; t(6) = 1.816 and p = 0.119) and time in the center zone (right; t(6) =2.955 and p 0.025). (D) Light-dark box transition test: latency to the first entry into the dark area (left; t(6) = 7.773 and p = 2.39\u00d7\u200910\u22124) and time in the dark area (right; t(6) = 4.641 and p = 0.004). (E) Elevated zero maze test: latency to the first entry into open sections (left; t(6) = 3.098 and p = 0.021) and time in open sections (right; t(6) = 2.511 and p = 0.046). (F) Forced swimming test: immobility time (t(6) = 11.978 and p = 2.05\u00d7\u200910\u22125). (G) Tail suspension test: immobility time (t(6) = 10.563 and p = 4.23\u00d7\u200910\u22125). (H) Sucrose preference test: preference index (t(6) = 0.722 and p = 0.497). (I) Total grooming time (t(6) = 2.579 and p = 0.042). n = 7 mice per group. Two different cohorts of mice were used in (C-G, I) and (H). Data are expressed as mean\u00b1SEM. Student\u2019s two-tailed paired t tests. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001; ns, not significant. ACSF, artificial cerebrospinal fluid solution; DREADD, Designer Receptors Exclusively Activated by Designer Drugs; DMSO, dimethyl sulfoxide; SALB, salvinorin B.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_5.png",
+        "caption": "Optogenetic activation of OT D3 neurons normalizes CRS-induced anxiety- and depression-like behaviors. (A) D3 neurons reliably fire action potentials upon blue light stimulation. Top: individual islands of Calleja (IC) in the OT (left) and enlarged view of the dashed area showing tightly packed D3-Cre/ChR2-EYFP neurons in IC (right). Scale bars: 200 \u03bcm (left) and 20 \u03bcm (right). Bottom: a D3-Cre/ChR2 neuron fires reliably upon blue light stimulation at 20 Hz (473 nm; 10 ms pulse). (B) Experimental strategy and timeline of behavioral assays. (C-H) Effects of blue light activation of D3 neurons on behavioral performance of CRS mice in different tests compared to green light stimulation. (C) Open field test: total distance travelled (left; t(12) = 0.220 and p = 0.830) and time in the center zone (right; Mann-Whitney test; p = 0.259). (D) Light-dark box transition test: latency to the first entry into the dark area (left; t(12) = 2.197 and p =0.048) and time in the dark area (right; t(12) = 0.616 and p = 0.550). (E) Elevated zero maze test: latency to the first entry into open sections (left; Mann-Whitney test; p = 0.016) and time in open sections (right; t(12) = 0.713 and p = 0.490). (F) Forced swimming test: immobility time (t(12) = 12.908 and p = 2.14\u00d7\u200910\u22128). (G) Tail suspension test: immobility time (t(12) = 7.278 and p = 9.78\u00d7\u200910\u22126). (H) Sucrose preference test: preference index (t(12) = 1.904 and p = 0.081). n = 7 mice per group. Two different cohorts of mice were used in (C-G) and (H). Data are expressed as mean\u00b1SEM. Student\u2019s two-tailed unpaired t tests; *p<0.05, ****p<0.0001; ns, not significant.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_6.png",
+        "caption": "Chemogenetic activation of OT D3 neurons normalizes CRS-induced depression-like behaviors. (A) Schematic showing strategy for viral injection into the OT and timeline for behavioral assays under excitatory DREADD manipulations. (B) Ex vivo electrophysiological recordings on hM3D(Gq)-expressing D3 neurons. Left, representative traces showing current injection-induced firing under bath application of ACSF or CNO (10 mM). Right, comparison of the firing frequency between ACSF and CNO condition (t(4) = 11.442 and p = 3.33\u00d710\u22124). n = 5 neurons from 3 mice. (C-I) Effects of chemogenetic activation of D3 neurons on behavioral performance in different tests. (C) Open field test: total distance travelled (left; t(6) = 0.816 and p = 0.446) and time in the center zone (right; p = 0.578). (D) Light-dark box transition test: latency to the first entry into the dark area (left; p = 0.579) and time in the dark area (right; t(6) = 1.815 and p = 0.120). (E) Elevated zero maze test: latency to the first entry into open sections (left; t(6) = 1.943 and p = 0.100) and time in open sections (right; t(6) = 1.113 and p = 0.308). (F) Forced swimming test: immobility time (t(6) = 5.739 and p = 0.0012).\u00a0 (G) Tail suspension test: immobility time (t(6) = 5.623 and p =0.0014). (H) Sucrose preference test: preference index (t(6) = 0.722 and p = 0.497). (I) Total grooming time (p = 0.016). n = 7 mice per group. Two different cohorts of mice were used in (C-G, I) and (H). Student\u2019s two-tailed paired t tests for (B), total distance in (C), time in (D), and (E)-(H). Wilcoxon matched-pairs signed rank test for time in (C), latency in (D) and (I). Data are expressed as mean\u00b1SEM. *p<0.05, **p<0.01, ***p<0.001; ns, not significant. ACSF, artificial cerebrospinal fluid solution; DREADD, Designer Receptors Exclusively Activated by Designer Drugs; CNO, Clozapine N-oxide.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_7.png",
+        "caption": "Optogenetic activation of OT D3 neurons induces conditioned place preference (CPP). (A) Representative traces of behavioral tracks of D3-Cre (control) and D3-Cre/ChR2 mice during the pre-conditioning (day 1) and post-conditioning (day 4) sessions. Blue laser stimulation (day 2 and 3) was paired with the least preferred chamber determined in the pre-conditioning session (see Materials and Methods for details). (B) The CPP difference score from the two mouse lines. D3-Cre control mice showed no significant difference in the CPP score of the laser-paired and non-laser-paired chamber (t(7) = 0.470 and p = 0.653). D3-Cre/ChR2 mice had a significantly higher CPP score in the laser-paired than the non-laser-paired chamber (t(7) = 10.031 and p = 2.10\u00d7\u200910\u22125). (C) Time spent in each compartment from the two mouse lines. Compared to the pre-conditioning session, D3-Cre control mice showed no significant difference in the duration of stay in all three corresponding compartments in the post-conditioning session (laser-paired: t(7) = 0.472 and p = 0.651; connecting: t(7) = 0.527 and p = 0.614; non-laser-paired: t(7) = 0.714 and p = 0.499). By contrast, D3-Cre/ChR2 mice spent more time in the laser-paired chamber and less time in the non-laser-paired chamber side without change in the connecting zone after conditioning (laser-paired: t(7) = 5.533 and p = 8.75\u00d7\u200910\u22124; connecting: t(7) = 0.407 and p = 0.696; non-laser-paired: t(7) = 2.980 and p = 0.021). (D) The CPP difference score in the laser-paired or non-laser-paired chamber from mice wearing collar (grooming blocked). Both mouse lines showed no significant difference (D3-Cre: t(6) = 0.120 and p = 0.908; D3-Cre/ChR2: t(6) = 1.781 and p = 0.125). (E) Time spent in each compartment from mice wearing collar (grooming blocked). Compared to the pre-conditioning session, both mouse lines did not show significant difference in the duration of stay in all three corresponding compartments in the post-conditioning session. D3-Cre mice: laser-paired: t(6) = 0.384 and p = 0.706; connecting: t(6) = 0.413 and p = 0.687; non-laser paired: t(6) = 0.270 and p = 0.792. D3-Cre/ChR2 mice: laser paired: t(6) = 1.422 and p = 0.181; connecting: t(6) = 1.595 and p = 0.137; non-laser paired: t(6) = 1.898 and p = 0.082. n = 8 mice per group for (B) and (C), n = 7 mice per group for (D) and (E). Data are expressed as mean\u00b1SEM. Student\u2019s two-tailed paired t tests. *p<0.05, ***p<0.001; ns, not significant.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_8.png",
+        "caption": "OT SPNs make direct inhibitory synaptic connections with NAc-projecting VTA dopaminergic neurons. (A) A model on how OT D3 neurons bidirectionally mediate depression-like behaviors in mice. Left, ablation or inhibition of OT D3 neurons disinhibits OT SPNs which in turn inhibit VTA dopaminergic neurons, reducing dopamine release to NAc. Right, activation of OT D3 neurons inhibits OT SPNs which in turn disinhibit VTA dopaminergic neurons, enhancing dopamine release to NAc. (B) Schematic showing strategy for viral injection and retrograde tracing. AAV1-FLEX-ChR2-EYFP virus (300-500 nl) and cholera toxin subunit B (300-500 nl) were bilaterally injected into the OT and NAc, respectively, in D1-Cre mice (n = 5). (C) Confocal images showing ChR2-EYFP+ OT D1-SPNs (left) and CTB+ NAc neurons (right). Scale bar = 200 \u03bcm (left) and 50 \u03bcm (right). (D) Confocal image showing CTB+ and tyrosine hydroxylase (TH)+ neurons surrounded with ChR2-EYFP+ D1-SPN axonal fibers in VTA (left). High magnification images (right) showing two CTB and TH double positive neurons from the dashed rectangle in the left panel. Scale bar = 50 \u03bcm (left) and 20 \u03bcm (right). (E) D1-SPNs inhibit VTA neurons. Repeated 10 ms blue laser pulses (0.1 Hz) evoked inhibitory postsynaptic currents (IPSCs) in 71.7% (inset, 43 out of 60 neurons) of VTA neurons recorded. (F) Light evoked IPSCs in VTA neurons were not changed by glutamate receptor antagonists (50 \u00b5M AP5 and 20 \u00b5M CNQX) but blocked by GABAA receptor antagonist bicuculline (10 \u00b5M). Left, representative traces for drug applications. Right, quantification of IPSC amplitudes. One-way ANOVA; F(3,16) = 18.517 and p = 2.65E-05 (ACSF vs. AP5 + CNQX + bicuculline: p = 7.35E-06; ACSF vs. AP5 + CNQX: p = 0.378; ACSF vs. wash out: p = 0.371; AP5 + CNQX vs. AP5 + CNQX + bicuculline: p = 3.69E-05; AP5 + CNQX vs. wash out: p = 0.948; AP5 + CNQX + bicuculline vs. wash out: p = 7.72E-05). n = 5 neurons. ****p<0.0001. (G) The majority of recorded VTA CTB+ neurons were dopamine (DA) neurons. Left, representative traces of DA and GABA neurons. Right, pie chart showing the composition of recorded VTA neurons. Baseline membrane potential was kept at -60 mV for all recordings. NAc, nucleus accumbens; OT, olfactory tubercle; VTA, ventral tegmental area. Data are expressed as mean\u00b1SEM.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    }
+]

0ab9485f323a84a87f5e9b35a5af0f9f487ed9784339ca91bceacdf95cc00a39/preprint/preprint.md

@@ -0,0 +1,299 @@
+# Abstract
+
+The ventral striatum, composed of the nucleus accumbens (NAc) and olfactory tubercle (OT), is a key reward center implicated in the pathophysiology of depression. Although the OT is known to regulate motivational and reward-related behaviors, its involvement in depression remains unexplored. We recently report that islands of Calleja, clusters of dopamine D3 receptor-expressing granule cells that are predominately situated in the OT, regulate self-grooming, a repetitive behavior manifested in mood disorders including depression. Here we show that chronic restraint stress (CRS) induces robust depression-like behaviors in mice and decreases excitability of OT D3 neurons. Ablation or inhibition of these neurons leads to depression-like behaviors, whereas their activation ameliorates CRS-induced depressive phenotypes. Moreover, activation of OT D3 neurons has a rewarding effect, which diminishes when grooming is blocked. Finally, we propose a model to explain how OT D3 neurons may influence dopamine release into the NAc via synaptic connections with OT spiny projection neurons (SPNs) that project to midbrain dopamine neurons. Our study reveals a novel role of OT D3 neurons in bidirectionally mediating depressive phenotypes, suggesting an attractive therapeutic target.
+
+**Biological sciences/Neuroscience/Stress and resilience**  
+**Biological sciences/Neuroscience/Diseases of the nervous system/Depression**  
+**Biological sciences/Neuroscience/Emotion/Striatum**  
+**Ventral striatum**  
+**Nucleus accumbens**  
+**Olfactory tubercle**  
+**Islands of Calleja**  
+**Dopamine D3 receptor**  
+**Anxiety**  
+**Depression**  
+**Grooming**
+
+# Introduction
+
+Depression, one of the most common mood disorders, is associated with dysfunction of brain reward systems [1]. In both rodents [2, 3] and humans [4–7], depressive phenotypes are linked to dysfunction of the ventral striatum, a key reward center that integrates brain-wide inputs including from midbrain dopamine neurons and sends inhibitory outputs to downstream structures [8, 9]. The ventral striatum contains two major divisions, the nucleus accumbens (NAc) and the olfactory tubercle (OT; also called tubular striatum [10]). Similar to the NAc, the OT is also involved in motivational and reward-related behaviors in rodent models [11–18]. In contrast to extensive research on the role of NAc in depression [19–27], whether and how the OT circuitry contributes to depression is largely unknown.
+
+Like the rest of the striatum, the OT contains spiny projection neurons (SPNs) and several types of interneurons [28, 29]. SPNs are GABAergic principal neurons that express either D1- or D2-type dopamine receptor [8, 9]. Functional and anatomical changes in NAc D1 and D2-SPNs or interneurons have been linked to depressive behaviors in rodents [19, 26, 30]. The ventral striatum also contains neurons expressing D3-type dopamine receptors, the majority of which are granule cells concentrated in the islands of Calleja of the OT [30]. The drd3 gene encoding the D3 receptor [30] as well as the islands of Calleja [31, 32] have been linked to the pathophysiology of neuropsychiatric diseases. Moreover, cariprazine, an atypical antipsychotic drug prescribed to treat several neuropsychiatric disorders including bipolar depression, predominantly binds to islands of Calleja neurons in the mouse brain [33], supporting a potential role of these neurons in regulating emotion.
+
+We recently discovered that OT D3 neuron activity bidirectionally regulates self-grooming in mice: optogenetic activation of these neurons robustly initiates orofacial grooming while optogenetic inhibition halts ongoing grooming [34]. Self-grooming, an evolutionally-conserved, repetitive behavior, serves important functions including de-arousal and stress reduction [35, 36]. Notably, altered levels of grooming are considered a behavioral biomarker for a number of neurological and neuropsychiatric disorders including depression [35, 36]. Since self-grooming increases dopamine release in the NAc [37], it is possible that the activity of OT D3 neurons and self-grooming are intimately linked to the reward system and emotion, and that these neurons therefore may be integral to the pathophysiology of depressive states.
+
+In this study, by combining optogenetic and chemogenetic manipulations, genetic ablation, ex vivo electrophysiology and mouse behavioral assays, we asked 1) what is the relationship between OT D3 neuron activity and depressive behaviors, 2) does activation of these neurons have a rewarding effect and whether it depends on the elicited self-grooming, and 3) which underlying neural pathway transmits the activity of OT D3 neurons to influence dopamine release? To answer these questions, we used the chronic restraint stress (CRS) model [38–41] to induce robust depression-like behaviors (increased inactivity and anhedonia) in mice and found that CRS significantly decreased excitability of OT D3 neurons without changing that of neighboring D1/D2-SPNs. Loss-of-function of OT D3 neurons led to depression (particularly increased inactivity but not anhedonia), whereas activation of these neurons ameliorated CRS-induced depression-like behaviors. Furthermore, optogenetic activation of OT D3 neurons produced conditioned place preference, indicating a rewarding effect, which diminished when self-grooming was blocked. Finally, we provide ex vivo electrophysiological evidence to support a model in which OT D3 neurons influence dopamine release into the NAc via synaptic connections with OT SPNs that subsequently project to dopamine neurons in the ventral tegmental area (VTA). Our study reveals a critical role of OT D3 neurons in regulating emotional responses, suggesting a new target for treatment of depression.
+
+# Results
+
+CRS induces anxiety- and depression-like behaviors and decreases excitability of OT D3 neurons
+
+To investigate how chronic exposure to stress may influence function of the OT, we subjected double transgenic D1-tdTomato/D2-EGFP mice and transgenic D3-Cre/tdTomato mice (see Materials and Methods for details) to CRS for 14 consecutive days (2 h per day). Twenty-four hours after the last CRS session, the mice underwent multiple behavioral tests (sequentially with stress level from low to high) to assess anxiety- and depression-like behaviors (Fig. 1A; see Materials and Methods for more details). Since CRS caused similar behavioral changes in the two mouse lines used, we pooled the data for statistical analysis in Fig. 1 (see Supplemental Figs. S1 and S2 for data and statistical analysis separated by each mouse line and sex). For anxiety-like behaviors, we used the open field test (OFT), light-dark box transition test (LDT) and elevated zero maze (EZM) test. In the OFT, CRS did not have significant effects on the total distance travelled nor the time spent in the center zone (Fig. 1B), indicating that CRS did not affect general locomotion. In the LDT, compared to the controls, CRS mice exhibited a similar latency for the first entry into the dark area but spent longer time in the dark area (Fig. 1C). In the EZM test, CRS mice showed an increased latency to the first entry into the open sections while shortened the duration stayed in the open sections compared to the controls (Fig. 1D). For depression-like behaviors, we performed the forced swimming test (FST) and tail suspension test (TST). Compared to the controls, CRS mice spent more time in immobility in both tests (Fig. 1E, 1F), inferred as enhanced “helplessness” and “despair” (see Materials and Methods for details). We then conducted the sucrose preference test to investigate CRS-induced anhedonia, another core behavioral symptom of depression. To avoid potential influence from prior behavioral tests, we used another cohort of mice to conduct this test (same for the following sucrose preference tests). CRS decreased the sucrose preference index to about 73% of the controls (Fig. 1G), indicating anhedonia. Overall, CRS induced some anxiety-like phenotypes and robust depression-like behaviors in mice, in general agreement with previous reports [42, 43]. Since stress affects grooming behavior in rodents [37, 44–47], we also quantified the time spent in orofacial grooming and found that CRS mice groomed more than controls (Fig. 1H; see Discussion).
+
+To examine potential effects of CRS on the electrophysiological properties of OT neurons, we performed whole-cell patch-clamp recordings on D1-tdTomato/D2-EGFP SPNs and D3-Cre/tdTomato neurons (hereafter referred to as D1/D2 SPNs or D3 neurons, respectively) in acute brain slices from control and CRS mice (Fig. 2A, 2E). Representative traces of firing patterns of D1/D2 SPNs and D3 neurons are shown in Fig. 2B and Fig. 2F, respectively. Since the data from D1 and D2 SPNs showed very similar trends, they were pooled together. Without altering the input resistance of these neurons (Fig. 2C, 2G), CRS significantly lowered the firing frequencies of D3 neurons but not D1/D2 SPNs upon current injections (Fig. 2D, 2H), indicating that CRS specifically reduces excitability of D3 neurons in the OT.
+
+## Ablation Or Inhibition Of Ot D3 Neurons Induces Robust Depression-like Behaviors
+
+We next explored the potential involvement of OT D3 neurons in anxiety- and depression-like behaviors in physiological conditions (without CRS treatment) by genetically ablating these neurons. We bilaterally injected the Cre-dependent DTA virus or control virus into the OT of D3-Cre/ChR2 mice (ChR2-EYFP as a marker for D3 neurons) (Fig. 3A). We previously showed that four weeks later, this approach efficiently ablated OT D3 neurons revealed by reduced EYFP signals [34]. We performed the same behavioral tests on DTA and control virus mice four weeks post injection (Fig. 3B). Ablation of D3 neurons had little effect on anxiety-like behaviors (except for the longer time spent in the center zone in the OFT, reflecting a mild anxiolytic effect) (Fig. 3C–E). By contrast, ablation of these neurons induced robust depression-like behaviors, leading to significantly longer immobility time in both the FST and TST (Fig. 3F, 3G), similar to CRS effects. Ablation of D3 neurons did not change the sucrose preference index (Fig. 3H), suggesting that these neurons are dispensable for this hedonic behavior.
+
+In addition, we tested the potential role of OT D3 neurons in mediating anxiety- and depression-like behaviors by chemogenetic manipulations. We bilaterally injected a mixture of Cre-dependent excitatory DREADD hM3D(Gq) and inhibitory DREADD KORD (Gi coupled DREADD based on the kappa-opioid receptor template) viruses into the OT of D3-Cre mice (Fig. 4A; Supplemental Fig. S3) to achieve bidirectional manipulations of D3 neuronal activity in the same mice [48] (the results of excitatory DREADD are described later). Ex vivo patch-clamp recordings confirmed that KORD-expressing D3 neurons were inhibited by its ligand salvinorin B (SALB), reflected by significantly decreased firing frequencies compared to control condition (Fig. 4B). Three weeks after viral injection, we conducted the same behavioral tests with subcutaneous injection of either DMSO (as control) or SALB (Fig. 4A). Unlike the mild anxiolytic phenotypes resulting from ablation of D3 neurons, inhibition of these neurons consistently and robustly produced anxiety-like behaviors, characterized by the decreased time spent in the center zone in the OFT (Fig. 4C), shorter latency to the first entry into the dark box and longer time spent in the dark box in the LDT (Fig. 4D), as well as the increased latency to the first entry into the open sections and less time spent in the open sections in the EZM test (Fig. 4E). Similar to ablation of D3 neurons, chemogenetic inhibition of these neurons also induced robust depression-like behaviors. SALB injected mice spent longer time in immobility in both the FST and TST compared to DMSO control conditions (Fig. 4F, 4G). The altered immobility duration was not due to the impaired motor ability since inhibition of D3 neurons did not affect general locomotion in the OFT (Fig. 4C). In addition, inhibition of D3 neurons did not affect the sucrose preference index (Fig. 4H), consistent with the ablation experiment (c.f. Figure 3H). To exclude the potential non-specific effect of SALB, we bilaterally injected the Cre-dependent AAV8-DIO-mCherry virus into the OT of D3-Cre mice as controls. Application of DMSO or SALB did not change behavior in these control mice (Supplemental Fig. S4), supporting that inhibitory DREADD-induced effects result from the interaction between SALB and KORD. Taken together, these results indicate that loss-of-function of OT D3 neurons via both genetic ablation and chemogenetic inhibition reliably induces depression-like behaviors.
+
+## Activation Of D3 Neurons Normalizes Crs-induced Depression-like Behaviors
+
+In order to evaluate whether activation of OT D3 neurons can mitigate CRS-induced anxiety- and depression-like behaviors, we used both optogenetic and chemogenetic approaches. In D3-Cre/ChR2 mice, the islands of Calleja D3 neurons in the OT were visualized by EYFP signals (Fig. 5A, top). These D3-Cre/ChR2 neurons reliably fired action potentials upon blue light stimulation at 20 Hz (473 nm; 10 ms pulse) [34] (Fig. 5A, bottom), and these parameters were applied in the following behavioral tests. Immediately after each CRS session, blue (for activating ChR2-expressing D3 neurons) or green light (less efficiency in activating ChR2 as comparison) was applied for 15 min in the home cage, and all behavioral tests were conducted during blue or green light stimulation, except for the sucrose preference test in which photostimulation was applied immediately before the test (Fig. 5B). Compared to green light, activation of D3 neurons by blue light alleviated some CRS-induced anxiety-like phenotypes (increased latency to the first entry into the dark area in the LDT and shortened latency to the first entry into the open sections in the EZM test) but not others (Fig. 5C–E). By contrast, optogenetic activation of D3 neurons ameliorated CRS-induced depression-like behaviors in both the FST and TST. Compared to green light, activation of D3 neurons by blue light decreased the immobility time by 57% (FST) and 63% (TST) (Fig. 5F, 5G). Furthermore, optogenetic activation of D3 neurons (15 min/day) was not sufficient to weaken CRS-induced anhedonia in the sucrose preference test (Fig. 5H), consistent with the results from the ablation and inhibition experiments (c.f., Figs. 3H and 4H). Since blue light activation of OT D3 neurons induces orofacial grooming [34, 49], it may lead to underestimation of the total distance travelled in the OFT as well as the immobility time in the FST and TST. We therefore rectified potential grooming-related deviations under light stimulation (see Materials and Methods for details), and similar conclusions could be drawn from these behavioral tests (Supplemental Fig. S5A–E).
+
+We next tested the effects of chemogenetic activation of D3 neurons on CRS-induced anxiety- and depression-like behaviors. We performed same behavioral assays on CRS, excitatory DREADD hM3D(Gq) mice with intraperitoneal injection of either saline (as control) or CNO (activating OT D3 neurons) (Fig. 6A). Ex vivo patch-clamp recordings confirmed that hM3D(Gq)-expressing D3 neurons were activated by its ligand CNO, reflected by significantly increased firing frequencies compared to control condition (Fig. 6B). Chemogenetic activation of D3 neurons did not mitigate any CRS-induced anxiety phenotypes (Fig. 6C–E). By contrast, activation of these neurons normalized CRS-induced depression-like behaviors, characterized by less immobility time in both the FST and TST (Fig. 6F, 6G). Similar to optogenetic manipulations, chemogenetic activation of D3 neurons also did not improve CRS-induced anhedonia in the sucrose preference test (Fig. 6H). To exclude the potential non-specific effect of CNO, we bilaterally injected the Cre-dependent AAV8-DIO-mCherry virus into the OT of D3-Cre mice as controls. Intraperitoneal injection of saline or CNO did not influence behavior of these control mice (Supplemental Fig. S6), supporting that excitatory DREADD-induced effects are ascribed to the interaction between CNO and hM3D(Gq).
+
+## Optogenetic Activation Of Ot D3 Neurons Induces Conditioned Place Preference
+
+Since optogenetic or chemogenetic activation of OT D3 neurons has antidepressant effects, we then asked whether activation of these neurons has any inherently rewarding effects (or associated positive valence). We used the conditioned place preference (CPP) assay, a Pavlovian conditioned paradigm, which contained four sessions in four days (see Materials and Methods for details). In the pre-conditioning session, a mouse was allowed to freely explore the three-compartment arena to determine its most and least preferred side chamber (Fig. 7A). In the following two conditioning sessions, the mouse was stimulated by blue light in the least preferred chamber. In the last session, the post-conditioning session, the mouse was again allowed to freely explore the arena, the time it spent in each compartment was measured and the CPP difference score (the time difference between the post- and pre-conditioning session in a chamber) was calculated (Fig. 7B). Unlike the D3-Cre control mice, which maintained their initial bias for the most preferred chamber, D3-Cre/ChR2 mice spent more time in the laser-paired chamber in the post-conditioning session (Fig. 7C), and thus had a significantly higher CPP score (Fig. 7A, 7B). These results suggest that optogenetic activation of OT D3 neurons has a rewarding effect.
+
+We previously reported that optogenetic activation of OT D3 neurons reliably induced self-grooming while inactivation of these neurons halted ongoing grooming [34]. Similarly, chemogenetic manipulations of these neurons also bidirectionally mediated grooming (Fig. 4I and Fig. 6I). We then asked whether grooming induced by D3 neuron activation is required for the ability of these neurons to drive CPP. We performed the same CPP test using D3-Cre/ChR2 mice with a collar around the neck to block grooming by preventing the forepaws from contacting the face and head (Fig. 7D, inset). Ablating self-directed orofacial grooming eliminated CPP caused by OT D3 neuron stimulation (Fig. 7D, 7E), suggesting that grooming elicited by activation of D3 neurons is necessary for the rewarding effect.
+
+## Ot D3 Neurons Inhibit Ot Spns Which Directly Inhibit Nac-projecting Vta Dopamine Neurons
+
+We propose a neural circuit model that may explain bidirectional regulation of grooming and depression-like behaviors via the activity of OT D3 neurons (Fig. 8A). OT D3 neurons directly inhibit OT SPNs which in turn inhibit the ventral tegmental area (VTA) dopamine neurons that mediate dopamine release into the NAc. This model is based on several lines of evidence in the literature: 1) self-grooming (both spontaneous and stress-elicited) induces transient dopamine release into the NAc [37], 2) the VTA◊NAc pathway is implicated in regulating depression-like behaviors [30, 50, 51], 3) OT D3 neurons are local interneurons and provide direct inhibition onto OT D1/D2 SPNs [34], and 4) OT SPNs project directly to VTA [52].
+
+To provide direct evidence to support this model, we examined whether OT SPNs make monosynaptic connections onto NAc-projecting VTA neurons. Since the VTA receives denser innervation from OT D1-SPNs than D2-SPNs [52], we tested functional connections of the OT SPNs◊VTA◊NAc pathway in D1-Cre mice. We bilaterally injected Cre-dependent AAV1-EF1a-DIO-ChR2-EYFP virus and cholera toxin subunit B-555 (CTB) into the OT and NAc, respectively (Fig. 8B). The ChR2-EYFP+ OT D1-SPNs, CTB+ NAc neurons, and retrogradely labeled CTB+ VTA neurons surrounded with dense D1-SPNs axonal fibers were confirmed post mortem (Fig. 8C, 8D). We performed whole-cell patch-clamp recordings on CTB+ VTA neurons in acute brain slices and recorded blue light evoked inhibitory postsynaptic currents (IPSCs, which were inward currents due to high intrapipette [Cl−]; see Materials and Methods for details). In ~72% (43 out of 60) of CTB+ VTA neurons, repeated light pulses evoked IPSCs, which had short latency (~5 ms) and little jitter (<1 ms) (Fig. 8E). These currents were blocked by GABA<sub>A</sub> receptor antagonist bicuculline but not changed by glutamate receptor antagonists, (2R)-amino-5-phosphonovaleric acid (AP5) and cyanquixaline (6-cyano-7-nitroquinoxaline-2,3-dione) (CNQX) (Fig. 8F), supporting the existence of GABA<sub>A</sub> mediated monosynaptic connections from OT D1-SPNs onto these NAc-projecting VTA neurons. Among synaptically connected CTB+ VTA neurons, we classified them into several types based on electrophysiological properties [53, 54]. The majority (~67% or 29 out of 43) displayed characteristics of dopamine neurons with low firing frequency (<5 Hz) upon current injection, apparent spike frequency adaptation and voltage “sag”. Another 23% were deemed as GABAergic neurons as they had higher firing frequency (≥5 Hz) with little spike frequency adaption or voltage “sag”, while 9% belonged to either class (Fig. 8G). Further post-hoc immunostaining with a tyrosine hydroxylase (TH) antibody confirmed that at least some of CTB+ VTA neurons were also TH+ (Fig. 8E), supporting their identity as dopamine neurons. Taken together, these results support that OT SPNs make direct inhibitory synaptic connections onto NAc-projecting VTA neurons.
+
+# Discussion
+
+By using cell-type-specific manipulations, *ex vivo* electrophysiology and behavioral assays, we reveal that OT D3 neurons (mostly in the islands of Calleja) bidirectionally mediate depression-like behaviors in mice. Loss-of-function of OT D3 neurons leads to depression-like behaviors (specifically increased inactivity), gain-of-function of these neurons alleviates CRS-induced depression-related symptoms. We propose a model which links the activity of OT D3 neurons and self-grooming with the reward system and stress induced responses.
+
+We first examined the effects of CRS treatment using multiple behavioral assays to assess anxiety- and depression-like phenotypes in mice. CRS induced robust depressive phenotypes, (i.e., increased immobility in both forced swimming and tail suspension tests) and anhedonia (decreased preference for sucrose) (Fig. <span class="InternalRef" refid="Fig1">1</span> E-G), supporting that CRS effectively causes depression-like behaviors in rodents as previously reported [38–41]. In addition to depressive phenotypes, CRS mice also exhibited anxiety-like behaviors. Unlike the consistent performance in depression-related assays, CRS mice exhibited divergent states in different anxiety assays. Anxiety-like behaviors were observed in light-dark box and elevated zero maze tests, but not in the open field test (Fig. <span class="InternalRef" refid="Fig1">1</span> B-D), suggesting that these behavioral assays may have different sensitivities or test different aspects of anxiety. Given that there is no single ideal animal model for anxiety and that each existing test has its advantages [55], a combination of different behavioral tests produces a better understanding in anxiety-related processes [56]. Our findings further support the necessity of using multiple behavioral assays for testing anxiety-like behaviors.
+
+CRS specifically decreases excitability of OT D3 neurons but not OT D1/D2 SPNs (Fig. <span class="InternalRef" refid="Fig2">2</span>), which is in sharp contrast to results from NAc circuits. Stress produces distinct changes (e.g., morphology, excitability, synaptic transmission) in NAc D1 and D2 SPNs, which have opposing roles in depression-like behaviors [2, 30]. Specifically, chronic stress induces hyperexcitability of NAc D1 SPNs [20, 21]. Here we report that OT D3 neurons are vulnerable but OT SPNs are resilient to stressful and depressive states, suggesting chronic stress exerts differential influences on the OT and NAc circuitry. The role of dopamine receptor-expressing neurons in the pathophysiology of depression-related symptoms is also implicated in other brain areas. For example, dysfunctions of p11 in dopamine D2 receptor-expressing neurons in the lateral habenula and prelimbic cortex contribute to depression-like behaviors [40, 41]. Our results suggest a causal relationship between activity of OT D3 neurons and depressive phenotypes, which is supported by both loss-of-function and gain-of-function experiments. Ablation or chemogenetic inhibition of OT D3 neurons caused depressive behaviors (Figs. <span class="InternalRef" refid="Fig3">3</span>, <span class="InternalRef" refid="Fig4">4</span>), especially increased inactivity, whereas optogenetic or chemogenetic activation of these neurons alleviates CRS-induced depression-like behaviors or makes the mice more resilient (Figs. <span class="InternalRef" refid="Fig5">5</span>, <span class="InternalRef" refid="Fig6">6</span>). These findings add OT to the existing brain areas that are causally linked to depression-related symptoms.
+
+In contrast to the striking effects on the immobility time in forced swimming and tail suspension tests, optogenetic or chemogenetic activation of OT D3 neurons is insufficient to ameliorate anhedonia (reduced sucrose preference) in CRS mice (Figs. <span class="InternalRef" refid="Fig5">5</span>, <span class="InternalRef" refid="Fig6">6</span>). One possibility is that different aspects of CRS-induced depressive phenotypes are mediated by distinct neuronal types and/or brain regions, which is supported by the finding that distinct ventral pallidal neurons mediate separate depressive symptoms [57]. Activation of OT D3 neurons may specifically improve “the decreased motivation for activity” in CRS mice, while anhedonic phenotypes might be mediated by other neuronal subtypes such as cholinergic neurons [25, 58] and/or other brain regions. There is evidence to support that NAc D1-SPNs are critical for the expression of anhedonia [2, 26, 30]. The other possibility is that OT D3 neurons are involved in mediating CRS-induced anhedonia but our relatively short activation of OT D3 neurons (15 min/day in optogenetic manipulation in Fig. <span class="InternalRef" refid="Fig5">5</span> and approximately 2 h chemogenetic manipulation in Fig. <span class="InternalRef" refid="Fig6">6</span>) is not sufficient to improve the reduced sucrose preference. This hypothesis is, to some extent, supported by our finding that activation of OT D3 neurons had a rewarding effect in a Pavlovian conditioned paradigm (Fig. <span class="InternalRef" refid="Fig7">7</span>). A recent report suggests that acute and chronic optogenetic stimulations on certain neural pathways could compensate each other in their effects on normalizing depressive phenotypes [59]. It would be interesting to separate the effects of chronic and repetitive optogenetic activation of OT D3 neurons during CRS treatment versus acute optogenetic activation during the anxiety- and depression-related assays.
+
+In this study, the two loss-of-function approaches produced different effects on anxiety-like behaviors. Chemogenetic inhibition of OT D3 neurons induced anxiety-like behaviors in all three tests (open field, light-dark box transition and elevated zero maze) (Fig. <span class="InternalRef" refid="Fig4">4</span>). However, ablation of these neurons produced a minor anxiolytic effect highlighted by the increased time in the center zone in the open field test, but no significant changes in the other two tests (Fig. <span class="InternalRef" refid="Fig3">3</span>). Unlike transient inhibition via chemogenetic manipulations, ablation of OT D3 neurons permanently alters the OT circuitry. For instance, it may induce compensatory changes in the neuronal activity of neighboring OT D1/D2 SPNs, which in turn affects both local and long-range circuits (e.g., the VTA◊lateral septum pathway) that are involved in regulating anxiety-related behaviors [60].
+
+Consistent with our previous finding that optogenetic activation or inactivation of OT D3 neurons initiates or halts self-grooming [34], here we demonstrate that chemogenetic manipulations also bidirectionally mediate the total grooming time (Figs. <span class="InternalRef" refid="Fig4">4</span>, <span class="InternalRef" refid="Fig6">6</span>). Interestingly, CRS treatment decreased excitability of OT D3 neurons (Fig. <span class="InternalRef" refid="Fig2">2</span>), which should act to reduce the grooming drive from these neurons. However, CRS mice exhibited more grooming than controls (Fig. <span class="InternalRef" refid="Fig1">1</span> H), suggesting that other grooming centers may be more active after CRS treatment to ensure this adaptive behavior in stressed situation [35, 37, 44, 47]. The potential role of D3 receptor in grooming and depression-like behaviors remains elusive. D3 receptor knockout mice display high basal level of grooming [61], consistent with the inhibitory action of this receptor in OT D3 neurons. Whereas some studies suggest that mice lacking D3 receptor are more resistant to stressful conditions and display normal emotional behaviors [62, 63], others show that D3 receptor deficiency results in depression-like behaviors [64, 65]. Since D3 receptor is expressed in multiple brain regions, specific manipulation of D3 receptor expression in distinct subpopulations of D3 neurons would be required to dissect out its role.
+
+As OT D3 neurons are local GABAergic interneurons [34], we propose a model which links these neurons and self-grooming with the NAc reward system through OT SPNs and VTA dopamine neurons (Fig. <span class="InternalRef" refid="Fig8">8</span>). Decreased OT D3 neuron activity would disinhibit monosynaptically connected SPNs, leading to enhanced inhibition onto NAc-projecting VTA dopamine neurons. This would attenuate dopamine release into the NAc and ultimately induce depressive phenotypes in these mice. On the other hand, activation of OT D3 neurons eventually leads to more dopamine release into the NAc, manifested as increased motivation for activity (characteristic of resilience from depressive phenotypes) in CRS mice. This model is supported by the rewarding effect of activation of OT D3 neurons, and interestingly, this effect diminishes when grooming is physically blocked (Fig. <span class="InternalRef" refid="Fig7">7</span>). Furthermore, we provide *ex vivo* electrophysiological evidence to support that OT D1 SPNs directly inhibit NAc-projecting VTA dopamine neurons (Fig. <span class="InternalRef" refid="Fig8">8</span>). However, we do not rule out other potential pathways that link OT D3 neurons and grooming with NAc dopamine signaling.
+
+Taken together, we discovered a novel role of OT D3 neurons in bidirectionally mediating depression-like behaviors. The findings that activation of OT D3 neurons has a rewarding effect and efficiently alleviates CRS-induced depressive phenotypes by increasing the motivation for activity suggest that ventral striatal OT/islands of Calleja D3 neurons are an attractive target for the intervention and treatment of depression.
+
+# Materials And Methods
+
+## Animals
+
+The D1-tdTomato [66] and D2-EGFP (Tg(Drd2-EGFP)S118Gsat) [67] mice were crossed to obtain double-transgenic mice in which dopamine D1 and D2 receptor-expressing neurons are labeled with red and green fluorescence, respectively. Bacterial artificial chromosome (BAC) transgenic D3-Cre (Tg(Drd3-cre)KI198Gsat) was obtained from Mutant Mouse Resource & Research Centers (MMRRC) and crossed with the Cre-dependent tdTomato reporter line (JAX Stock No: 007909 or Ai9 line: Rosa26-floxed-tdTomato) or Cre-dependent channel rhodopsin 2 (ChR2) line (JAX Stock No: 024109 or Ai32 line: Rosa26-floxed-ChR2) [68] to generate D3-Cre/tdTomato or D3-Cre/ChR2 mice, respectively. Drd1-Cre (D1-Cre for simplicity; MMRRC Strain# 37156-Jax) mice were obtained from MMRRC [69]. Both male and female mice (8–12 weeks old) were used in all experiments and the data were pooled. Mice were housed on a 12 h light/dark cycle with food and water available ad libitum. Mice were group-housed until the surgery of viral injection and/or intra-cranial optical fiber implantation and singly-housed afterwards. All experimental procedures were performed in accordance with the guidelines of the National Institutes of Health and were approved by the Institutional Animal Care and Use Committees at the University of Pennsylvania.
+
+## Chronic Restraint Stress Model
+
+Chronic restraint stress (CRS) treatment was performed as previously documented [39–41]. Briefly, mice were individually placed head first into a well-ventilated 50 ml polypropylene conical tube, which was then plugged with a 4.5-cm-long middle tube, and finally tied with the cap of the 50 ml tube. The restraint stress lasted for 2 hours per day (approximately from 9 to 11 am) for consecutive 14 days. After each session of the restraint stress, mice were returned to their home cage, where they were housed in pairs with food and water available ad libitum. Twenty-four hours after the last restraint session, mice were subjected to behavioral assays or electrophysiological recordings.
+
+## Virus/toxin Injection And Optical Fiber Implantation
+
+Mice were anesthetized with isoflurane (~3% in oxygen) and secured in a stereotaxic system (Model 940, David Kopf Instruments). Isoflurane levels were maintained at 1.5-2% for the remainder of the surgery. Body temperature was maintained at 37 ℃ with a heating pad connected to a temperature control system (TC-1000, CWE Inc.). Local anesthetic (bupivacaine, 2 mg/kg, s.c.) was applied before skin incision and hole drilling on the dorsal skull. To target the islands of Calleja in the OT, we used two sets of coordinates from bregma: anteroposterior (AP) 1.2 (or 1.54) mm; mediolateral (ML), ±1.1 (or ±1.15) mm; dorsoventral (DV), -5.5 (or -5.0) mm. For the NAc, in order to retrogradely label more NAc-projecting VTA neurons, we injected CTB (recombinant, Alexa Fluor™ 555 conjugate, Invitrogen #C34776) into both the core and shell subregions: core, AP, 1.6 mm; ML, ±0.8 mm; DV, -4.1 mm; medial shell, AP, 1.5 mm; ML, ±0.55 mm; DV, -4.7 mm; lateral shell, AP, 0.98 mm; ML, ±1.8 mm; DV, -4.92 mm. For viral injection, as appropriate, AAV8-CMV-TurboRFP-WPRE-rBG (2.94x10¹⁰ vg/ml), AAV8-EF1α-mCherry-FlEX-DTA (3.3x10⁹ viral units/ml) (University of North Carolina Viral Vector Core, Chapel Hill, NC) (800 nl each side), AAV1-DIO-ChR2-EYFP (C-34777, Life Technologies) (300–500 nl each side), AAV8-hSyn-DIO-hM3D(Gq)-mCherry (≥4×10¹² vg/ml; Addgene, cat.# 44361), AAV8-hSyn-dF-HA-KORD-IRES-mCitrine (≥7×10¹² vg/ml; Addgene, cat.# 65417) or AAV8-hSyn-DIO-mCherry (≥1×10¹³ vg/ml; Addgene, cat.# 50459) was bilaterally (300–500 nl each side) injected into the OT via a Hamilton syringe (5 µl) with a flow rate of 50 nl/min controlled by an Ultra Micro Pump III (UMP3) with a SYS-micro4 controller attachment (World Precision, Sarasota, USA). In D1-Cre mice, cholera toxin subunit B (300–500 nl) was also bilaterally injected into NAc. The tip of the syringe was left for 10–15 min after the injection. For optical fiber implantation, a cannula (CFMC14L10-Fiber Optic Cannula, Ø2.5 mm Ceramic Ferrule, Ø400 µm Core, 0.39 NA; Thorlabs, Newton, NJ), customized to 6 mm length, was unilaterally placed in the OT at the coordinates aforementioned and fixed on the skull with dental cement in D3-Cre/ChR2 mice. Mice were returned to home cage for recovery for one week before behavioral tests and mice with viral injection had at least three week waiting period before tests. All optical fiber implantation locations were post-hoc verified and only mice with the intended targeted site were included in data analysis.
+
+## Ex vivo electrophysiological recording
+
+Whole-cell patch-clamp recordings were performed as described previously [30]. Briefly, mice were deeply anesthetized (ketamine-xylazine; 200 and 20 mg/kg body weight, respectively) and quickly decapitated. The dissected brain was immediately placed in ice-cold cutting solution containing (in mM) 92 N-Methyl D-glucamine, 2.5 KCl, 1.2 NaH₂PO₄, 30 NaHCO₃, 20 HEPES, 25 glucose, 5 Sodium L-ascorbate, 2 Thiourea, 3 Sodium Pyruvate, 10 MgSO₄, and 0.5 CaCl₂; osmolality ~300 mOsm and pH ~7.3, bubbled with 95% O₂-5% CO₂. Coronal sections (250 µm thick) containing the OT were cut using a Leica VT 1200S vibratome. Brain slices were incubated in oxygenated artificial cerebrospinal fluid (ACSF in mM: 124 NaCl, 3 KCl, 1.3 MgSO₄, 2 CaCl₂, 26 NaHCO₃, 1.25 NaH₂PO₄, 5.5 glucose, and 4.47 sucrose; osmolality ~305 mOsm and pH ~7.3, bubbled with 95% O₂-5% CO₂) for ~30 min at 31ºC and at least 30 minutes at room temperature before use. For recordings, slices were transferred to a recording chamber and continuously perfused with oxygenated ACSF. Fluorescent cells were visualized through a 40X water-immersion objective on an Olympus BX61WI upright microscope equipped with epifluorescence. Whole-cell patch-clamp recordings were made under both current and voltage clamp mode. Recording pipettes were made from borosilicate glass with a Flaming-Brown puller (P-97, Sutter Instruments; tip resistance 5–10 MΩ). The pipette solution contained (in mM) 120 K-gluconate, 10 NaCl, 1 CaCl₂, 10 EGTA, 10 HEPES, 5 Mg-ATP, 0.5 Na-GTP, and 10 phosphocreatine. Light stimulation was delivered through the same objective via pulses of blue laser (473 nm, FTEC2473-V65YF0, Blue Sky Research, Milpitas, USA) with 10 ms light pulse at 20 Hz. For light-evoked inhibitory postsynaptic currents (IPSCs), a high Cl⁻ intrapipette solution (120 mM KCl instead of K-gluconate) was used so that the reversal potential of [Cl⁻] was at ~0 mV and GABAₐ receptor-mediated currents were inward at a holding potential of -60 mV. Light stimulation was delivered through the same objective via 10 ms pulses of blue laser (473 nm, FTEC2473-V65YF0, Blue Sky Research, Milpitas, USA). Viral infection in the OT was confirmed in brain slices during recordings. Pharmacological drugs CNO and SALB (dissolved in DMSO) were bath perfused.
+
+## Immunohistochemistry
+
+After patch clamp recordings, acute brain slices prepared form D1-Cre mice with AAV1-DIO-ChR2-EYFP virus and cholera toxin subunit B bilaterally injected into the OT and NAc, respectively, were immediately incubated in 4% paraformaldehyde (PFA) for 10–15 min, and then transferred into 1X phosphate buffered solution (PBS) overnight. Slices were washed with PBS three times (20 min each), and then incubated with 0.5% Triton X-100 in PBS for 10 min, followed by incubation with 0.5% Triton X-100 and 0.5% Tween-20 in PBS for 10 min. Slices were incubated with a primary antibody against tyrosine hydroxylase (TH) (1:500; Millipore, cat# AB152, host: rabbit) for 24 h at 4°C. After three PBS washes (20 min each), the slices were incubated with a secondary antibody (1:1000; Life Technology, cat# A31573, host: donkey) and incubated for 24 h at 4°C. Slices were washed three more times in 0.5% Triton X-100 and 0.5% Tween-20 in PBS (20 min each) and treated with glycerol in PBS (volume ratio 1:1) for 30 min followed by glycerol in PBS (volume ratio 7:3) for 30 min before being mounted onto superfrost slides for confocal imaging.
+
+## Behavioral Tests
+
+All behavioral procedures were performed during the light cycle (9:00 am -12:00 pm) except for the sucrose preference test which was conducted during the dark cycle (6:00 pm-6:00 am).
+
+All mice tested were transferred to the testing room 1 h before the test for habituation. The following behavioral tests were performed sequentially (stress level from low to high): open field test (OFT), light-dark box transition test (LDT), elevated-zero maze (EZM) test, forced swimming test (FST) and tail suspension test (TST) with an interval of 24 h between two individual behavioral assays. All apparatuses were cleaned with 70% ethanol before and between trials. Mice in all behavioral tests were videotaped using a webcam at 30 frames/sec, and behaviors were scored using the ANY-maze software (Stoelting Co.) or manually by those who were blinded to the experimental conditions.
+
+## Optogenetic and chemogenetic manipulations
+
+For optogenetic experiments, blue light (473 nm, 10–15 mW/mm², 20 Hz, 10 ms pulse, 10 s stimulation every 30 s) was applied to activate ChR2-expressing neurons after each daily CRS treatment for 15 min and during behavioral tests. Green light (532 nm), less efficient in activating these neurons, was applied with the same parameters as control. D3-Cre mice with AAV8-hSyn-DIO-hM3D(Gq)-mCherry (excitatory DREADD) and AAV8-hSyn-dF-HA-KORD-IRES-mCitrine (inhibitory DREADD) or AAV8-hSyn-DIO-mCherry (control virus) in the OT were intraperitoneally injected with saline or CNO (5 mg/kg) (for excitatory DREADD), or subcutaneously injected with DMSO or SALB (10 mg/kg) (for inhibitory DREADD), 30 min before behavioral tests.
+
+## Open field test (OFT)
+
+The mice were placed in an open field arena (40 cm x 40 cm) in a room with dim light and allowed to freely explore the apparatus for 5 min. The total distance travelled and the total time in the central zone (20 cm x 20 cm) were calculated.
+
+## Light-dark box transition test (LDT)
+
+The light-dark box (46 cm x 28 cm x 30 cm) was composed of two compartments. Two-thirds of the box was the light compartment and the remaining part was the dark compartment. Mice were placed in the center of the light box with the head oppositely facing the dark compartment, and were allowed to freely explore the two compartments for 10 min. The latency to the first entry into the dark compartment and the total time spent in the dark compartment were calculated.
+
+## Elevated zero maze (EZM) test
+
+The zero-maze, composed of two open and two closed sections, was elevated 80 cm above the floor. The test mice were placed at the interface of an open and a closed section with the head facing the closed section. The latency to the first entry into the open section and the total time spent in the open sections were calculated.
+
+## Forced swimming test (FST)
+
+The FST, commonly used to assess antidepressant activity [70, 71], was also used for testing depression-like behaviors (or “learned helplessness”) in rodents [72, 73]. Briefly, the mouse was placed in a cylindrical plexiglass tank (40 cm high × 12 cm in diameter) with water (25±2°C) in a depth of 10 cm. Mice were submitted to the swimming test for 6 min on the test day. The time spent in immobility was calculated post-hoc. Immediately after the FST, each mouse was removed from the water, towel-dried, and returned to its home cage. The water was changed and the cylinder was cleaned for each mouse tested.
+
+## Tail suspension test (TST)
+
+The TST is widely used for testing “behavioral despair”, another depression-related symptom [74]. Briefly, the mouse, held by the tail, was suspended 50 cm from the floor for 6 min. The time spent in immobility was recorded. The mice were considered immobile only when they hung down passively and were completely motionless.
+
+## Sucrose preference test
+
+Prior to the test, mice were habituated to the presence of two drinking bottles (one containing 2% sucrose and the other plain water) for 2 days in their home cage. Following this acclimation, mice had the free choice of either drinking the 2% sucrose solution or plain water for a period of 4 days. Water and sucrose solution intake was measured daily, and the positions of the two bottles were switched daily to reduce any confounding side bias. Sucrose preference was calculated as a percentage of the weight of sucrose intake over the total weight of fluid intake and averaged over the 4 days of testing. Some D3-Cre/ChR2 mice with optical fiber implantation in the OT were photostimulated for 15 min after each daily CRS treatment and during the tests with the same parameters as aforementioned. For chemogenetic manipulations, D3-Cre mice with AAV8-hSyn-DIO-hM3D(Gq)-mCherry and AAV8-hSyn-dF-HA-KORD-IRES-mCitrine in the OT were intraperitoneally (I.P.) injected with saline or CNO (5 mg/kg) (after two-week’s CRS treatment), or subcutaneously injected with DMSO or SALB (10 mg/kg) (without CRS treatment), respectively, one time each day with only one drug in a counterbalanced way during the four days test session.
+
+## Conditioned place preference
+
+A custom-built conditioned place preference (CPP) apparatus consisted of a rectangular cage with three compartments: a left black chamber (35 cm × 20 cm) with a metal wire mesh floor, a connecting zone (35 cm × 10 cm) with a smooth gray floor, and a right white chamber (35 cm × 20 cm) with a soft floor. The CCP test was conducted as previously described [75]. Briefly, the CPP test consisted of 4 days. Day 1 (pre-conditioning, 15 min): a preconditioning test was performed to obtain a baseline preference for the apparatus of each mouse tested. The side chamber a mouse spent the most time was assigned as the most preferred side, and the other side chamber as the least preferred side. On days 2 and 3 (conditioning): the mouse was firstly kept in either the most or least preferred side (counterbalanced across mice) for 15 min, then transferred to the other side for 15 min. Mice in the least-preferred side were paired with blue light stimulation (same parameters as aforementioned), and no photostimulations for mice in the most preferred side. On Day 4 (post-conditioning, 15 min): 24 h after the conditioning session on Day 3, the mice were placed back into the arena with all three compartments accessible, to evaluate preference for the stimulation and non-stimulation paired chambers. The CPP difference score was calculated by the time spent in the post-conditioning session minus that in the pre-conditioning session in the corresponding chamber.
+
+## Rectification of potential grooming-related effects in behavioral tests
+
+Since blue light activation of OT D3 neurons induces orofacial grooming, it may deviate some of the measurements of the behavioral tests; e.g., underestimation of the total distance travelled in the OFT and overestimation of the activity time in the FST and TST. We therefore rectified potential grooming-related deviations under light stimulation as follows. We first quantified the total duration of grooming induced by blue light activation of OT D3 neurons in the 5 min OFT - approximately 60 s grooming out of 100 s light stimulation. In the OFT, the rectified total distance travelled = (the actual distance x 300s/240s) as mice did not travel during the 60 s of blue light-induced grooming. For potentially overestimated parameters including the time and latency in anxiety tests (OFT, LDT and EZM) and the activity time in depression tests (FST and TST), we subtracted the presumptive grooming time proportionally to the blue light duration for each measurement.
+
+## Confocal Imaging
+
+Mice were perfused transcardially with 4% paraformaldehyde (PFA) in fresh phosphate buffered saline (PBS). The brain was post fixed in 4% PFA overnight at 4 ºC and then transferred into PBS. Coronal slices (100 µm thick) were prepared using a Leica VT 1200S vibratome. The slices were treated with glycerol in PBS (volume ratio 1:1) for 30 min followed by glycerol in PBS (volume ratio 7:3) for 30 min before being mounted onto superfrost slides for imaging. Confocal imaging was performed by sequential scanning of slices at 10 x and 40 x in a SP5/Leica confocal microscope.
+
+## Statistical analysis
+
+Shapiro-Wilk tests were used to verify normal distribution of datasets. For normally distributed datasets, parametric statistical tests (student’s t test or two-way ANOVA test) were used; otherwise, non-parametric tests (Mann-Whitney or Wilcoxon matched-pairs signed rank test) were applied. Statistical analysis was performed in GraphPad Prism and figures were assembled in Adobe Photoshop.
+
+# References
+
+1. Fox, M.E. and M.K. Lobo, The molecular and cellular mechanisms of depression: a focus on reward circuitry. Mol Psychiatry, 2019. 24(12): p. 1798-1815.
+
+2. Francis, T.C. and M.K. Lobo, Emerging role for nucleus accumbens medium spiny neuron subtypes in depression. Biological psychiatry, 2017. 81(8): p. 645-653.
+
+3. Heshmati, M. and S.J. Russo, Anhedonia and the brain reward circuitry in depression. Current behavioral neuroscience reports, 2015. 2(3): p. 146-153.
+
+4. Malone Jr, D.A., et al., Deep brain stimulation of the ventral capsule/ventral striatum for treatment-resistant depression. Biological psychiatry, 2009. 65(4): p. 267-275.
+
+5. Hou, Z., et al., Increased temporal variability of striatum region facilitating the early antidepressant response in patients with major depressive disorder. Progress in Neuro-Psychopharmacology and Biological Psychiatry, 2018. 85: p. 39-45.
+
+6. Kumar, P., et al., Impaired reward prediction error encoding and striatal-midbrain connectivity in depression. Neuropsychopharmacology, 2018. 43(7): p. 1581.
+
+7. Morgan, J.K., et al., Maternal response to positive affect moderates the impact of familial risk for depression on ventral striatal response to winning reward in 6-to 8-year-old children. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 2022.
+
+8. Burke, D.A., H.G. Rotstein, and V.A. Alvarez, Striatal local circuitry: a new framework for lateral inhibition. Neuron, 2017. 96(2): p. 267-284.
+
+9. Kreitzer, A.C. and R.C. Malenka, Striatal plasticity and basal ganglia circuit function. Neuron, 2008. 60(4): p. 543-554.
+
+10. Wesson, D.W., The tubular striatum. Journal of Neuroscience, 2020. 40(39): p. 7379-7386.
+
+11. Murata, K., et al., Opposing roles of dopamine receptor D1-and D2-expressing neurons in the anteromedial olfactory tubercle in acquisition of place preference in mice. Frontiers in behavioral neuroscience, 2019. 13: p. 50.
+
+12. Zhang, Z., et al., Activation of the dopaminergic pathway from VTA to the medial olfactory tubercle generates odor-preference and reward. Elife, 2017. 6: p. e25423.
+
+13. Gadziola, M.A. and D.W. Wesson, The neural representation of goal-directed actions and outcomes in the ventral striatum's olfactory tubercle. Journal of Neuroscience, 2016. 36(2): p. 548-560.
+
+14. Murata, K., et al., Mapping of learned odor-induced motivated behaviors in the mouse olfactory tubercle. Journal of Neuroscience, 2015. 35(29): p. 10581-10599.
+
+15. Gadziola, M.A., et al., The olfactory tubercle encodes odor valence in behaving mice. Journal of Neuroscience, 2015. 35(11): p. 4515-4527.
+
+16. DiBenedictis, B.T., et al., DREADD-induced silencing of the medial olfactory tubercle disrupts the preference of female mice for opposite-sex chemosignals. Eneuro, 2015. 2(5).
+
+17. Agustín-Pavón, C., F. Martínez-García, and E. Lanuza, Focal lesions within the ventral striato-pallidum abolish attraction for male chemosignals in female mice. Behavioural brain research, 2014. 259: p. 292-296.
+
+18. Ikemoto, S., Involvement of the olfactory tubercle in cocaine reward: intracranial self-administration studies. Journal of Neuroscience, 2003. 23(28): p. 9305-9311.
+
+19. Cheng, J., et al., HCN2 channels in cholinergic interneurons of nucleus accumbens shell regulate depressive behaviors. Neuron, 2019. 101(4): p. 662-672. e5.
+
+20. Francis, T.C., et al., Nucleus accumbens medium spiny neuron subtypes mediate depression-related outcomes to social defeat stress. Biological psychiatry, 2015. 77(3): p. 212-222.
+
+21. Francis, T.C., et al., The Selective RhoA Inhibitor Rhosin Promotes Stress Resiliency Through Enhancing D1-Medium Spiny Neuron Plasticity and Reducing Hyperexcitability. Biological psychiatry, 2019. 85(12): p. 1001-1010.
+
+22. Lim, B.K., et al., Anhedonia requires MC4R-mediated synaptic adaptations in nucleus accumbens. Nature, 2012. 487(7406): p. 183.
+
+23. Wang, J., et al., Dopamine D3 receptor in the nucleus accumbens alleviates neuroinflammation in a mouse model of depressive-like behavior. Brain, behavior, and immunity, 2022. 101: p. 165-179.
+
+24. Miyamoto, Y., et al., Striatal N-acetylaspartate synthetase Shati/Nat8l regulates depression-like behaviors via mGluR3-mediated serotonergic suppression in mice. International Journal of Neuropsychopharmacology, 2017. 20(12): p. 1027-1035.
+
+25. Warner-Schmidt, J.L., et al., Cholinergic interneurons in the nucleus accumbens regulate depression-like behavior. Proceedings of the National Academy of Sciences, 2012. 109(28): p. 11360-11365.
+
+26. Fox, M.E., et al., Dendritic remodeling of D1 neurons by RhoA/Rho-kinase mediates depression-like behavior. Molecular psychiatry, 2018: p. 1.
+
+27. LeGates, T.A., et al., Reward behaviour is regulated by the strength of hippocampus–nucleus accumbens synapses. Nature, 2018. 564(7735): p. 258-262.
+
+28. Martin-Lopez, E., et al., Embryonic and postnatal development of mouse olfactory tubercle. Molecular and Cellular Neuroscience, 2019. 98: p. 82-96.
+
+29. Millhouse, O. and L. Heimer, Cell configurations in the olfactory tubercle of the rat. Journal of Comparative Neurology, 1984. 228(4): p. 571-597.
+
+30. !!! INVALID CITATION !!!
+
+31. Inta, D., et al., Adult neurogenesis in the human striatum: possible implications for psychiatric disorders. Mol. Psychiatry, 2016. 21: p. 446-447.
+
+32. Inta, D., A. Meyer-Lindenberg, and P. Gass, Alterations in postnatal neurogenesis and dopamine dysregulation in schizophrenia: a hypothesis. Schizophrenia bulletin, 2011. 37(4): p. 674-680.
+
+33. Prokop, S., et al., PharmacoSTORM nanoscale pharmacology reveals cariprazine binding on Islands of Calleja granule cells. Nature communications, 2021. 12(1): p. 1-19.
+
+34. Zhang, Y.-F., et al., Ventral striatal islands of Calleja neurons control grooming in mice. Nature neuroscience, 2021. 24(12): p. 1699-1710.
+
+35. Kalueff, A.V., et al., Neurobiology of rodent self-grooming and its value for translational neuroscience. Nature Reviews Neuroscience, 2016. 17(1): p. 45-59.
+
+36. Smolinsky, A.N., et al., Analysis of grooming behavior and its utility in studying animal stress, anxiety, and depression, in Mood and anxiety related phenotypes in mice. 2009, Springer. p. 21-36.
+
+37. Sun, J., et al., Excitatory SST neurons in the medial paralemniscal nucleus control repetitive self-grooming and encode reward. Neuron, 2022.
+
+38. Willner, P., The chronic mild stress (CMS) model of depression: History, evaluation and usage. Neurobiology of stress, 2017. 6: p. 78-93.
+
+39. Oh, Y.-S., et al., SMARCA3, a chromatin-remodeling factor, is required for p11-dependent antidepressant action. Cell, 2013. 152(4): p. 831-843.
+
+40. Seo, J.-S., et al., Cellular and molecular basis for stress-induced depression. Molecular psychiatry, 2016.
+
+41. Seo, J., et al., Elevation of p11 in lateral habenula mediates depression-like behavior. Molecular Psychiatry, 2017.
+
+42. Chiba, S., et al., Chronic restraint stress causes anxiety-and depression-like behaviors, downregulates glucocorticoid receptor expression, and attenuates glutamate release induced by brain-derived neurotrophic factor in the prefrontal cortex. Progress in Neuro-Psychopharmacology and Biological Psychiatry, 2012. 39(1): p. 112-119.
+
+43. Joo, Y., et al., Chronic immobilization stress induces anxiety-and depression-like behaviors and decreases transthyretin in the mouse cortex. Neuroscience letters, 2009. 461(2): p. 121-125.
+
+44. Mu, M.-D., et al., A limbic circuitry involved in emotional stress-induced grooming. Nature communications, 2020. 11(1): p. 1-16.
+
+45. Song, C., K.C. Berridge, and A.V. Kalueff, 'Stressing' rodent self-grooming for neuroscience research. Nature Reviews Neuroscience, 2016. 17(9): p. 591-591.
+
+46. Fernández-Teruel, A. and C. Estanislau, Meanings of self-grooming depend on an inverted U-shaped function with aversiveness. Nature Reviews Neuroscience, 2016. 17(9): p. 591-591.
+
+47. Xu, Y., et al., Identification of a neurocircuit underlying regulation of feeding by stress-related emotional responses. Nature communications, 2019. 10(1): p. 1-16.
+
+48. Vardy, E., et al., A new DREADD facilitates the multiplexed chemogenetic interrogation of behavior. Neuron, 2015. 86(4): p. 936-946.
+
+49. Zhang, Y.-F., et al., Self-directed orofacial grooming promotes social attraction in mice via chemosensory communication. iScience, 2022: p. 104284.
+
+50. Tye, K.M., et al., Dopamine neurons modulate neural encoding and expression of depression-related behaviour. Nature, 2013. 493(7433): p. 537-541.
+
+51. Yadid, G., D.H. Overstreet, and A. Zangen, Limbic dopaminergic adaptation to a stressful stimulus in a rat model of depression. Brain research, 2001. 896(1-2): p. 43-47.
+
+52. Zhang, Z., et al., Whole-brain mapping of the inputs and outputs of the medial part of the olfactory tubercle. Frontiers in Neural Circuits, 2017. 11: p. 52.
+
+53. Koyama, S., et al., Obesity decreases excitability of putative ventral tegmental area GABA ergic neurons. Physiological Reports, 2013. 1(5).
+
+54. Zhang, T.A., A.N. Placzek, and J.A. Dani, In vitro identification and electrophysiological characterization of dopamine neurons in the ventral tegmental area. Neuropharmacology, 2010. 59(6): p. 431-436.
+
+55. Ramos, A., Animal models of anxiety: do I need multiple tests? Trends in pharmacological sciences, 2008. 29(10): p. 493-498.
+
+56. Bouwknecht, J.A. and R. Paylor, Behavioral and physiological mouse assays for anxiety: a survey in nine mouse strains. Behavioural brain research, 2002. 136(2): p. 489-501.
+
+57. Knowland, D., et al., Distinct ventral pallidal neural populations mediate separate symptoms of depression. Cell, 2017. 170(2): p. 284-297. e18.
+
+58. Virk, M.S., et al., Opposing roles for serotonin in cholinergic neurons of the ventral and dorsal striatum. Proceedings of the National Academy of Sciences, 2016. 113(3): p. 734-739.
+
+59. An, K., et al., A circadian rhythm-gated subcortical pathway for nighttime-light-induced depressive-like behaviors in mice. Nature Neuroscience, 2020: p. 1-12.
+
+60. Tong, Q., et al., D1 receptor-expressing neurons in ventral tegmental area alleviate mouse anxiety-like behaviors via glutamatergic projection to lateral septum. Molecular Psychiatry, 2022: p. 1-14.
+
+61. Le Foll, B., J. Diaz, and P. Sokoloff, Neuroadaptations to hyperdopaminergia in dopamine D3 receptor-deficient mice. Life sciences, 2005. 76(11): p. 1281-1296.
+
+62. Chourbaji, S., et al., Dopamine receptor 3 (D3) knockout mice show regular emotional behaviour. Pharmacological research, 2008. 58(5-6): p. 302-307.
+
+63. Xing, B., et al., Effects of immobilization stress on emotional behaviors in dopamine D3 receptor knockout mice. Behavioural brain research, 2013. 243: p. 261-266.
+
+64. Leggio, G.M., et al., Dopamine D3 receptor as a new pharmacological target for the treatment of depression. European journal of pharmacology, 2013. 719(1-3): p. 25-33.
+
+65. Moraga-Amaro, R., et al., Dopamine receptor D3 deficiency results in chronic depression and anxiety. Behavioural brain research, 2014. 274: p. 186-193.
+
+66. Shuen, J.A., et al., Drd1a-tdTomato BAC transgenic mice for simultaneous visualization of medium spiny neurons in the direct and indirect pathways of the basal ganglia. Journal of Neuroscience, 2008. 28(11): p. 2681-2685.
+
+67. Gong, S., et al., A gene expression atlas of the central nervous system based on bacterial artificial chromosomes. Nature, 2003. 425(6961): p. 917.
+
+68. Madisen, L., et al., A robust and high-throughput Cre reporting and characterization system for the whole mouse brain. Nature neuroscience, 2010. 13(1): p. 133.
+
+69. Zhang, J., et al., c-Fos facilitates the acquisition and extinction of cocaine-induced persistent changes. Journal of Neuroscience, 2006. 26(51): p. 13287-13296.
+
+70. Petit-Demouliere, B., F. Chenu, and M. Bourin, Forced swimming test in mice: a review of antidepressant activity. Psychopharmacology (Berl), 2005. 177(3): p. 245-55.
+
+71. Slattery, D.A. and J.F. Cryan, Using the rat forced swim test to assess antidepressant-like activity in rodents. Nature Protocols, 2012. 7(6): p. 1009-1014.
+
+72. Gonzalez, M.M. and G. Aston-Jones, Light deprivation damages monoamine neurons and produces a depressive behavioral phenotype in rats. Proc Natl Acad Sci U S A, 2008. 105(12): p. 4898-903.
+
+73. Zhang, Y.-F., et al., Alterations of motor cortical microcircuit in a depressive-like mouse model produced by light deprivation. Neuroscience, 2016.
+
+74. Steru, L., et al., The tail suspension test: a new method for screening antidepressants in mice. Psychopharmacology (Berl), 1985. 85(3): p. 367-70.
+
+75. Stamatakis, A.M. and G.D. Stuber, Activation of lateral habenula inputs to the ventral midbrain promotes behavioral avoidance. Nature neuroscience, 2012. 15(8): p. 1105.

0d1610230074b14395861cf839ddcbf0c93b942791e2b1291724761b2581bd86/preprint/images_list.json

@@ -0,0 +1,50 @@
+[
+    {
+        "type": "image",
+        "img_path": "images/Figure_1.png",
+        "caption": "Genetic validation of YFP-TM-ERD2 by stable transformation in Nicotiana benthamiana and Physcomitrium patens (A) Schematic of N. benthamiana ERD2ab antisense (AS) construct driven by the strong constitutive CaMV35S promoter (35S), combined with fluorescent markers expressed under the weak TR2 promoter on the same T-DNA. (B) Confocal laser scanning microscopy of stably transformed N. benthamiana in leaf epidermis cells of regenerating shoots in vitro (B1, B2) and root cortex cells of seedlings from the next generation of YFP-TM-ERD2 plants (B3). Notice that ST-YFP-HDEL labels the ER whilst YFP-TM-ERD2 only labels Golgi bodies. In roots, Golgi-stacks are either viewed from the side (white arrow heads) or from top/bottom (white stars), giving rise to the typical doughnut shapes. (C) Schematic of YFP-TM targeted gene knock-in onto PpERD2B-1 (Pp3c9_13230V3.9), in which the YFP-TM coding portion (Silva-Alvim et al., 2018) was recombined into the first protein-coding exon by marker-free transformation, leading to expression of a YFP-TM-ERD2 derivative under the transcriptional control of the native promoter in P. patens.(D) Schematic of PpERDB2-2 (Pp3c15_12830) knockout by complete deletion of the second ERD2 gene by targeted replacement with a selection cassette flanked with loxP sites enabling the subsequent deletion of the marker by transient expression of Cre recombinase. Lines positively tested for both knock-in and knock-out events (Supplementary figures 1, 2) showed normal growth.(E) YFP-TM-PpERD2 expression under its native promoter in P. patens. E1) Notice stronger expression near growing tips and newly formed cell plates. E2) At high magnification, distinguish punctate structures from weak autofluorescence of chloroplasts (Chl.).",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_2.png",
+        "caption": "ERD2 Function and Golgi Residency is Conserved Amongst Eukaryotes(A) Retention assay using protoplasts showing the secretion index (ratio extra/intracellular Amy-HDEL activity) with cargo alone or with co-expressed A. thaliana ERD2b (At) and 12 further ERD2 orthologs from the eukaryotes Ostreococcus lucimarinus (Oi), Acanthamoeba castellanii (Ac), Phytophthora infestans (Pi), Chondrus crispus (Cc), Galdieria sulphuraria (Gs), Homo sapiens (HsERD2), Hypsibius dujardini (Hd), Thalassiosira pseudonana (Tp), Puccinia graminis (Pg), Kluyveromyces lactis (KlERD2), Trypanosoma brucei (Tb) and Saccharomyces cerevisiae (Sc). Transfection efficiencies were normalised by the internal marker GUS established at 5 standard OD units as described in materials and methods. Percentages in brackets refer to the sequence identity with AtERD2b.Notice that cargo retention function is abolished when less than 50% sequence identity remains.(B) CLSM analysis of two human ERD2 fusions (YFP-TM-HsERD2 and HsERD2-YFP constructed as described before23. Notice that only the C-terminal YFP fusion causes partial ER localisation. (C) Retention assays as in A) but either comparing the two HsERD2 fluorescent variants from B) or a comparison of untagged hsERD2 (WT) with point \u2013mutations in the HsERD2 C-terminus indicated above each lane. KKAA refers to the double mutant combining K206A and K207A. LLGG refers to the double mutant combining L208G and L210G. Notice that the C-terminal YFP fusion has lost biological activity. Notice also that only the LLGG mutant has lost biological activity when untagged HsERD2 is analysed. A full dose response for the KKAA double mutant is provided in Supp. Figure 3.(D) C-terminal amino acid sequences of human (KDELR2) and A. thaliana ERD2B. Conserved residues are highlighted grey and the conserved di-leucine motif is highlighted bold.(E) CLSM analysis of selected mutants from C) but in the YFP-TM-HsERD2 configuration. Silent mutations in panel C) retain the Golgi localisation, whilst the inactive LLGG mutant displays partial ER localisation.(F) Schematic of C-terminal fusions to plant and human ERD2 for functional assays (upper) and the fluorescent derivative for CLSM analysis (lower schematic).(G) Secretion index of Amy-HDEL, co-expressed with either wild type human or plant ERD2 compared to the three different C-terminal modifications in each case. Constant levels of ERD2 encoding plasmids were co-transfected (yielding 5 standard OD units). In both instances, the addition of a FLAG or c-myc tag strongly reduced function, whilst most of the activity was maintained for each ortholog after adding the HA tag. (H) Localisation of human and plant ERD2 fluorescent fusions with C-terminal FLAG, c-myc and HA tags. Notice that FLAG and c-myc additions cause an ER-Golgi localisation, whilst the addition of an HA tag does not affect the Golgi localisation of YFP-TM-ERD2 for both orthologs.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_3.png",
+        "caption": "FRAP and redistribution assays identify a Golgi-retention signal at the ERD2 C-terminus(A) Fluorescence recovery after photobleaching comparing wild type ERD2 and the LLGG mutant. ERD2 recovery to 50% (400 seconds) took almost twice the time of the LLGG mutant (240 seconds). LLGG mutant recovery reached 85%, whereas wild type ERD2 remained at around 50%.(B) Schematic of dual expression T-DNA constructs used to co-express fluorescent ERD2 fusions with cargo (either secreted Amy or the ERD2-ligand AmyHDEL). (C) Golgi localisation of YFP-TM-ERD2 co-expressed with Amy and Amy-HDEL. Distribution remains unchanged for both cargo.(D) Dual ER-Golgi localisation of YFP-TM-ERD2\u0394C5 co-expressed with Amy and a more prominent ER localisation when co-expressed with Amy-HDEL.(E) Dual ER-Golgi localisation of ERD2-YFP co-expressed with Amy. The re-distribution of ERD2-YFP to the ER by co-expressed Amy-HDEL is even more drastic compared to that of the deletion mutant in panel D). (F) Schematic of the T-DNA for expression analysis of the ERD2-fusion alone. Golgi bodies (arrows) are labelled by ERD2-YFP more visibly during high cellular expression, whereas punctae are much fainter relative to the ER fluorescence at low expression (imaged at higher detector gain).",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_4.png",
+        "caption": "ERD2 activity is incompatible with COPI-mediated transport.(A) ARF1-RFP localisation in relation to co-expressed Golgi marker ST-YFP together with either co-expressed cytosolic mock protein (PAT, upper panel) as control, or ERD2 (lower panel). Notice that under control conditions the subcellular localisation of ARF1-RFP is mainly in punctate structures, colocalising with the Golgi marker and additional extra-Golgi structures (white arrow heads). Upon co-expression of ERD2, the Golgi-marker redistributes to the ER network, whilst ARF1-RFP is mainly cytosolic. White arrow heads point at post-Golgi structures that remain. (B) Effect of saturating ERD2 overexpression (given in standard GUS OD units below each lane) on secretion of either Amy or Amy-HDEL. Notice that inhibition of constitutive secretion is not observed for the deletion mutant ERD2-\u0394C5).(C) C-terminal amino acid sequences of ERD2 wild type (WT) and two variants in which the last 9 amino acids of ERD2 is replaced by the corresponding region of P24 (underlined.) The proposed COPII ER export signal of p2429 is in bold, as is the dileucine motif in the WT sequence, the relevant lysines of the canonical COPI transport motif in the p24 variant and finally the mutant serines in the KKSS variant.(D) Dose-response assay measuring the influence of co-transfected C-terminal ERD2 variants (given in standard GUS OD units below each lane) on AmyHDEL secretion. ERD2WT mediates strong cell retention whilst the P24 fusion shows no retention activity. The KKSS mutant of the p24 fusion restores the retention activity at the highest dose.(E) Localisation of p24 and KKSS hybrids incorporated into fluorescent ERD2 fusion proteins. The p24 C-terminus mediates complete ER localisation of the resulting ERD2 fusion whilst the KKSS mutant shows high steady state levels at the Golgi.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_5.png",
+        "caption": "Reactivation of ERD2-YFP with a novel cis-Golgi retention motif(A) Comparative Golgi distribution between ST-RFP and YFP-TM-ERD2. Although both labelling punctae, an rS value of 0.54 highlights a cis-trans Golgi segregation of YFP-TM-ERD2 and ST-RFP, respectively. This is also visible in regions identified with white arrow heads.(B) When compared with MNS3-RFP instead, the shared cis-Golgi localisation of YFP-TM-ERD2 is clearly demonstrated by an rS value of 0.92.(C) Retention of AmyHDEL by ERD2 fusion variants at increasing concentrations. As previously published, TM-ERD2 effectively retains AmyHDEL at low and high concentrations. Meanwhile the C-terminal addition of YFP completely abolishes retention, regardless of increasing concentration. However, the N-terminal addition of the LPYS Golgi retention motif does allow significant activity to return with increasing concentration.(D) N-terminal addition of LPYS causes redistribution of TM-ERD2-YFP exclusively to the Golgi, in agreement with reactivation in secretion assays.\u00a0",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_6.png",
+        "caption": "Receptor-recycling cannot explain the observed receptor-ligand stoichiometry Transient expression in which both receptor and ligand were expressed from dual expression vectors harbouring GUS as reference marker for transfection efficiency and expressed at identical GUS units.(A) Immunoprecipitated ERD2-HA and co-expressed Amy-HDEL separated by SDS-page, followed by blotting on nitrocellulose and visualisation via phosphorimaging. Molecular weight markers are given on the right in kilo-daltons. (B) Table showing the total number of cysteine and methionine residues in cargo and receptor. Relative radioactivity units measured for ERD2-HA by phosphorimaging must be multiplied by 1.5 to permit comparison with units from Amy-HDEL to permit calculation of relative number of molecules. (C) Phosphorimaging quantification (arbitrary relative units) of signals from 3 different nitrocellulose blots as in A) showing the radioactivity from transiently expressed ERD2-HA and co-expressed Amy-HDEL (A) from cells and medium. Notice that Amy-HDEL radioactivity is extremely high compared to that of ERD2-HA (955 units), increasing from 22838 to 29284 units in the cells due to co-expressed ERD2-HA. Correcting for the number of cysteine and methionine residues, the introduced ERD2-HA is the equivalent of 1433 units, approximately 4.5 \u2013 fold lower than the increase in cellular Amy-HDEL molecules (6446 units).(D) Retention of AmyHDEL where the maximum receptor levels from panel A are co-transfected (second lane), followed by consecutive dilution of the receptor plasmid up to 100 fold (last lane). Notice that a strong reduction in Amy-HDEL secretion compared to the control (first lane) is still observed even after 100-fold dilution of the receptor plasmid (last lane).",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    }
+]

11268c6e941fa8862c8347d3f39feb5cfe492b57c21063051d1bf8105cdf75bd/preprint/images_list.json

@@ -0,0 +1,50 @@
+[
+    {
+        "type": "image",
+        "img_path": "images/Figure_1.png",
+        "caption": "RNF214 interacts with the TEAD transcription factors\na, Cell proliferation assay of RNF214 +/+, +/- and -/- MEFs. MEF cells were isolated from 13.5 days\u2019 mouse embryos and seeded into 96-well plate for cell viability analysis. The cell viability of MEF cells was quantified by CCK8 assay. Data were presented as mean \u00b1 SD. Student\u2019s t-test was used; n=3. b, Scheme showing APEX2-catalyzed biotinylation. APEX2 (orange) was fused at the N-terminus or C-terminus of RNF214 (blue). Live cells were incubated with biotin-phenol for 30 minutes and then treated for 1 minute with 0.25 mM H2O2 at room temperature to initiate biotinylation. APEX2 catalyzes one-electron oxidation of biotin-phenol into a biotin-phenoxyl radical, which covalently tags proximal endogenous proteins (green). Biotin-labelled proteins (red B =biotin) were enriched by Streptavidin beads and then subjected to mass spectrometry analysis. c, Silver staining to visualize the total biotinylated proteins in each group. The negative control sample, with APEX2 omitted, was also treated with biotin-phenol and H2O2. As shown in silver staining, there were three major bands corresponding to endogenous biotinylated proteins in the negative group. Venn diagram (below) illustrated the number of proteins identified using mass spectrometry. d, Barplot of the KEGG pathway enrichment analysis. 511 proteins shared by both N-terminal and C-terminal groups but excluded in control group were performed for the KEGG pathway enrichment analysis. Only three pathways were enriched. The Hippo pathway was significantly enriched and all four TEAD transcription factors were on the top list and noted above. The red dotted line means p\uff1c0.05. e-f, RNF214 interacts with TEAD1. HEK293T cells were transfected with Flag-RNF214 and Myc-TEAD1. 24 hours after transfection, reciprocal Co-IP and immunoblotting were performed using anti-Myc or anti-Flag antibodies as indicated in the figures. 0.1% input was on the right lane. g-i, Association of RNF214 with TEAD2, TEAD3, TEAD4 was demonstrated using the same procedure as shown in (e-f). j, TEAD1 directly interacts with RNF214. Pulldown assays to detect direct interaction between purified GST-TEAD1 and Strep-RNF214 proteins. GST was used as a negative control. \u2018*\u2019 indicates RNF214 isoform2, which lacks 52-206 amino acids in the N-terminus. k, RNF214 interacts with YAP weakly. HEK293T cells were transfected with HA-RNF214 and Flag-YAP, and Co-IP and western blotting were employed using antibodies as indicated.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_2.png",
+        "caption": "RNF214 augments Hippo-regulated transcription\na, mRNA analysis of TEADs target genes in Hep3b cells. Hep3b cells were transfected with siRNF214 oligos for 96 hours. Total mRNAs were isolated using Trizol reagents and analyzed using qRT-PCR approach. An siRNA-resistant cDNA of RNF214 resumed the mRNA expression levels of three target genes in RNF214-silenced cells. b,mRNA analysis of TEADs target genes in Huh7 cells. Huh7 cells were transfected with siRNF214 oligos for 96 hours. c,Serum induces ANKRD1, CTGF and CYR61 transcription. HEK293A cells were transfected with siRNF214 for 72 hours, starved in serum-free medium for 12 hours and then stimulated with 10% FBS for the indicated times. Cell lysates were subjected to immunoblotting with the indicated antibodies. d, CTGF-luciferase reporter assay. HEK293T cells were transfected with CTGF-luc reporter plasmid, Renilla luciferase plasmid as control and indicated gene expression plasmids. Increasing amounts of RNF214 (0, 100, 200, 400 ng in each transfection) were co-transfected with the reporter system. All values were normalized for transfection efficiency against Renilla luciferase activities. e, RNF214 promotes TEAD1 and TEAD3 transcriptional activities. Myc-TEAD1 or HA-TEAD3 was co-transfected into HEK293T cells with CTGF-luc reporter, YAP and RNF214. f,Gal4-TEAD4/9xUAS-luciferase reporter assay. The transcriptional activities of YAP-TEAD4 were measured based on YAP\u2019s ability to co-activate the Gal4 DNA binding domain fused to TEAD4 (Gal4-TEAD4) on the 9xUAS-luciferase reporter. HEK293T cells were co-transfected with the reporter system as indicated along with increasing amounts of RNF214 plasmid (0, 100, 200, 400 ng in each transfection). Luciferase activities were normalized to \u03b2-gal activities. g-h, RNF214 enhances TEADs transcriptional activities depending on its ubiquitin ligase activity. g, HEK293T cells were transfected with Flag-RNF214 wide type (WT) or Flag-RNF214 RING finger deletion mutant (RD) along with Gal4-TEAD4/9xUAS-luc as indicated. One tenth of the samples were directly subjected to SDS-PAGE to verify expression consistency between WT and RD RNF214 (below). h, qRT-PCR analysis in Huh7 cells. Huh7 cells were transfected with siRNF214 for 96 hours. For RD rescue experiment, an siRNA-resistant cDNA of RNF214 RD mutant was introduced into Huh7 cells stably using a lentivirus infection approach, then endogenous RNF214 was silenced using siRNA. i-j, The coiled-coil domain of RNF214 is essential for its effect. i, HEK293T cells were transfected with Flag-RNF214 WT or the Flag-RNF214 coiled-coil deletion mutant (CCD) along with Gal4-TEAD4/9xUAS-luc as indicated. j,qRT-PCR analysis in Huh7 cells. Huh7 cells were transfected with siRNF214 for 96 hours. For CCD rescue experiment, an siRNA-resistant cDNA of RNF214 CCD mutant was delivered into Huh7 cells stably using a lentivirus infection approach, then endogenous RNF214 was knocked down using siRNA. k-l, The N-terminus of RNF214 is indispensable for its effect. k,HEK293T cells were transfected with Flag-RNF214 WT or the Flag-RNF214 N-terminal deletion (NTD) along with Gal4-TEAD4/9xUAS-luc as indicated. l, qRT-PCR analysis in Huh7 cells. Huh7 cells were transfected with siRNF214 for 96 hours. For NTD rescue experiment, an siRNA-resistant cDNA of RNF214 NTD mutant was expressed into Huh7 cells stably using a lentivirus infection approach, then endogenous RNF214 was depleted using siRNA.\nData were presented as mean \u00b1 SD. Student\u2019s t-test was used; n=3.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_3.png",
+        "caption": "RNF214 promotes nonproteolytic polyubiquitylation of TEADs\na-b,RNF214 promotes TEADs ubiquitylation. a,HEK293T cells were transfected with HA-Ub, Flag-TEAD2 and Myc-RNF214 or the Myc-RNF214 RD mutant plasmids. 24 hours post-transfection, Flag-TEAD2 proteins were pulled out using anti-Flag beads by denaturing immunoprecipitation and the ubiquitylated TEAD2 proteins were detected using anti-HA antibody. b, The ubiquitylation of TEAD3 in HEK293T was detected using the same procedure as shown in (a). c, RNF214 ubiquitylates TEAD2. A biotinylated Avi-tagged TEAD2 (Bio-TEAD2) was expressed in HLF cells. Flag-RNF214 was then transfected into the Bio-TEAD2 HLF cells and biotin (2 mg/mL) was added to culture medium overnight before cell harvest. 36 hours after transfection, biotinylated-TEAD2 proteins were isolated using Streptavidin beads under a denaturing buffer condition. Ubiquitylated TEAD2 proteins were then detected using an anti-ubiquitin antibody. d,Knockout of RNF214 decreased ubiquitylation of TEADs. Halo-ThUBDs proteins were expressed and purified, and then incubated with HLF cell lysates. Ubiquitylated TEADs were detected using a pan-TEAD antibody. Two independent clones were selected from sgRNF214-3 HLF pools. e-f, RNF214 promotes K63 polyubiquitylation of TEAD2. HEK293T cells were transfected with Flag-TEAD2, Myc-RNF214, HA-Ub wide type (WT) and K48R mutant (e) or K63R mutant (f). 24 hours post-transfection, Flag-TEAD2 proteins were pulled out using anti-Flag beads by immunoprecipitation under a denaturing buffer condition and the ubiquitylated TEAD2 proteins were reviewed using anti-HA antibody.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_4.png",
+        "caption": "RNF214 enhances the interaction between TEADs and YAP\na-b,RNF214 increases the interactions between YAP and TEADs. HEK293T cells were transfected with YAP, TEAD and RNF214 WT or RD mutant. 24 hours after transfection, Co-IP and immunoblotting were performed as indicated in the figure. c, Schematic diagram of 4 lysine residues potentially ubiquitylated in TEAD2 YBD domain. d-e, RNF214 ubiquitylates TEAD2 on K345 residue. HEK293T cells were transfected with HA-Ub, Myc-RNF214, Flag-TEAD2 wild type (WT) and 3KR mutant (d) or the K345R mutant (e) plasmids. 24 hours after transfection, Flag-TEAD2 proteins were immunoprecipitated using anti-Flag beads and the ubiquitylated TEAD2 proteins were detected using anti-HA antibody. TEAD2 3KR mutant contains KR substitutions on K280\u3001K281 and K351 residues. f, RNF214 fails to promote the interaction between YAP and the K345R mutant of TEAD2. HEK293T cells were transfected with Myc-RNF214, Flag-YAP and HA-TEAD2 WT or the K345R mutant. 24 hours after transfection, Co-IP and immunoblotting were performed as indicated in the figure. g, The TEAD2 K345R mutant fails to fully support RNF214-induced CTGF-driven luciferase activities in HEK293T cells. Reporter assay were done as described in Fig.2d. Data were plotted as mean \u00b1 SD. Student\u2019s t-test was used; n=3. h-i,The Gal4-TEAD4 K260R mutant fails to rescue the RNF214-induced luciferase activities in TEAD1/3/4 knockdown cells. The K260 residue of TEAD4 in Gal4-TEAD4 corresponds to the K345 residue of TEAD2. HEK293T cells were transfected with Flag-YAP, Flag-RNF214 and Gal4-TEAD4 WT or the K260R mutant along with 9xUAS-luc as indicated. Luciferase activities were normalized to \u03b2-gal activities. Expression consistency is verified by Western blotting. Data were plotted as mean \u00b1 SD. Student\u2019s t-test was used; n=3. j-k, YAP and TAZ directly bind to K48 and K63 polyubiquitin chains in vitro. Pulldown assays to determine whether YAP or TAZ possess polyubiquitin chain binding abilities. GST-YAP or GST-TAZ were purified from BL21 (DE3) bacteria cells and poly-K48 Ub (3-7) and poly-K63 Ub (3-7) chains were purchased from R&D Systems. GST was used as a negative control.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_5.png",
+        "caption": "Overexpression of RNF214 correlates with poor prognosis in HCC\na, Bioinformatic analysis of RNF214 mRNA levels in HCC. mRNA expression levels from TCGA database were analyzed using Student\u2019s t-test. b, Kaplan-Meier survival curves of overall survival based on RNF214 expression in TCGA database. The image was prepared using the Human Protein Atlas. c, Bioinformatic analysis of RNF214 protein abundance in HCC. Protein expression data were from Fanjia database and analyzed using Student\u2019s t-test. d-f, Immunohistochemical (IHC) staining of RNF214 in a tissue microarray. Representative images were presented in (d). Scale bar, 100 mm. The IHC scores between paired tumor and non-tumor tissues from 176 patients were followed by Student\u2019s t-test. g-i, Spearman\u2019s correlation analysis between RNF214 expression and differentiation grades in tumor tissues from 275 patients with HCC. Representative scores were shown in (g). Scale bar, 100 mm. RNF214 (high), IHC score \u22656; RNF214 (low), IHC score < 6. j,Western blotting of RNF214 protein expression in HCC cell lines. k-m, Expression level of RNF214 is positively correlated with YAP/TAZ-TEAD target genes (e.g., AMOTL2, CTGF, and CYR61) in liver cancer patients, as analyzed through TCGA database.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    },
+    {
+        "type": "image",
+        "img_path": "images/Figure_6.png",
+        "caption": "RNF214 is critical for HCC tumorigenesis\na-b, Proliferation in RNF214 knockout HLF cells. a, The cell viability rates of RNF214 knockout HLF cells were quantified by CCK8 assay. b, Colony formation assays in RNF214 knock-out HLF cells. 103 viable cells were seeded into 6-well plate and incubated for 9 days. The total area intensity of colonies was measured and shown in the bar graph (below). c, Rescue experiments of cell viability in Hep3b cells. Hep3b cells were transfected with siCK (control siRNA) and siRNF214. An siRNA-resistant cDNA of RNF214 was stably expressed in Hep3b cells using a lentivirus infection approach to rescue the growth inhibitory effect. d, Transwell assays of migration and invasion. Hep3b or siRNF214-resistant cells were transfected with siCK or siRNF214 for 48 hours and then plated in transwell chambers (with or without Matrigel) for another 48 hours. Scale bar, 100 mm. e, Migration assay in the Tet/on-YAP-S127A Huh7 cell line. The Tet/on-YAP-S127A Huh7 cells were transfected with siRNF214 and 25 ng/ml Dox was added to culture medium after 12 hours. 48 hours post siRNA transfection, cells were plated in transwell chambers (with indicated Dox) for another 48 hours before analysis. Scale bar, 100 mm. f-i, RNF214 knockdown inhibited HCC tumor growth in subcutaneous xenograft model. 2 x 106 control or RNF214-silenced Huh7 cells with Matrigel were injected subcutaneously into 5-week-old male BALB/c nude mice. 21 days after cell implantation, tumors were dissected, photographed and weighted. n=10 tumors for each group. j-k, mRNA and protein levels of CYR61 in RNF214-silenced tumor. The dissected tumors were subjected to qRT-PCR analysis (n=8) and Western blotting (n=6). l, Working model of RNF214 functions in HCC. YAP and TAZ are transcription co-activators that activate TEAD transcriptional activities to promote HCC tumorigenesis. RNF214 induces TEADs non-proteolytic ubiquitylation at a single conserved lysine site, enhances the interactions between TEADs and YAP, and then promotes transactivation of the downstream genes, thereby leading to enhanced tumor progression. The figure is created with BioRender.com and has been granted a license to use the BioRender content.\nData were presented as mean \u00b1 SD. Student\u2019s t-test was used for statistical analysis; the log-rank test and Cox regression analysis were used for tumor occurrence. n=3, unless noted.",
+        "footnote": [],
+        "bbox": [],
+        "page_idx": -1
+    }
+]

11268c6e941fa8862c8347d3f39feb5cfe492b57c21063051d1bf8105cdf75bd/preprint/preprint.md

@@ -0,0 +1,264 @@
+# Abstract
+
+RNF214 is an understudied ubiquitin ligase without any knowledge of its biological functions or specific protein substrates. Using an APEX2-mediated proximity labeling method coupled with the mass spectrometry technique, we identified the TEAD transcription factors in the Hippo pathway as interactors of RNF214. We showed that RNF214 induces non-proteolytic ubiquitylation at a conserved single lysine residue of TEADs, enhances the interactions between TEADs and the transcription coactivators of the Hippo pathway including YAP and TAZ, and then promotes transactivation of the downstream genes of the Hippo signaling. Moreover, we proved that YAP and TAZ could bind polyubiquitin chains, implying the underlying mechanisms by which RNF214 regulates the Hippo pathway. Furthermore, we found that RNF214 is overexpressed in hepatocellular carcinoma (HCC). Clinical and statistical analysis indicated that high expression levels of RNF214 are associated with low differentiation status and poor prognosis of HCC. Consistently, we showcased that RNF214 promotes proliferation, migration and invasion of HCC cells and HCC tumorigenesis in mouse models via the Hippo pathway. Collectively, our data revealed that RNF214 is a critical component in the Hippo pathway by forming a new signaling axis of RNF214-TEAD-YAP, thereby upregulating the transcriptional activities of the YAP/TAZ-TEAD complex. More importantly, our results suggest that RNF214 serves as an oncogene of HCC and could be a potential drug target of HCC therapy.
+
+**Biological sciences/Cell biology/Post-translational modifications/Ubiquitylation**  
+**Biological sciences/Cancer/Oncogenes**  
+**Health sciences/Oncology/Cancer/Oncogenes**
+
+# Introduction
+
+Ubiquitin is a small signaling protein that can be conjugated to its protein substrates. This process, so called ubiquitylation or ubiquitination, is one of the major protein posttranslational modifications in eukaryotes. The ubiquitylation reaction is sequentially catalyzed by a ubiquitin activating enzyme (E1), a ubiquitin conjugating enzyme (E2) and a ubiquitin ligase (E3)¹. Thus far, at least 43948 ubiquitylation sites from over 14692 ubiquitylated proteins have been detected experimentally in humans², implying that ubiquitylation causes complex biology.
+
+The ubiquitin machinery can conjugate not only a single ubiquitin, but also polyubiquitin chains on its target proteins. Polyubiquitin chains are synthesized by forming an isopeptide bond between the C-terminal glycine residue in the donor ubiquitin (Gly76) and the ε-amino group of a lysine residue or the amino group of the N-terminal methionine residue in the acceptor ubiquitin¹. Ubiquitin contains seven lysine residues, therefore, at least eight kinds of polyubiquitin chains could be synthesized by the ubiquitylation machinery. Moreover, mixed and branched polyubiquitin chains have been reported as well¹. The specific linkages of polyubiquitin chains function as ‘ubiquitin codes’ to determine diverse destinies of ubiquitylated substrates³. Certain polyubiquitin chains, such as lysine-11 (K11), lysine-48 (K48) and branched ones often drive protein degradation via the 26S proteasome¹. Indeed, the ubiquitin-proteasome system is the major cellular machinery selectively degrading short-lived and unwanted proteins in eukaryotic cells. Howbeit, proteolysis is not the only fate of ubiquitylated proteins. The polyubiquitin chains via the lysine-63 (K63) or N-terminal methionine (M1) residue of ubiquitin are usually not signals for protein turnover¹. Instead, these non-proteolytic polyubiquitin chains often cause localization change or functional alternation of protein substrates. Because of the complexity of ubiquitin codes, protein ubiquitylation regulates virtually every aspect of human activities and health. Dysregulation of ubiquitylation is often linked to many human diseases, including cancer, autoimmune, neurodegenerative and viral diseases⁴–⁶.
+
+Ubiquitylation is a very specific process and the substrate specificity is mainly determined by ubiquitin ligases. Human genome encodes over 600 ubiquitin ligases which contains either a RING finger or a HECT domain⁷. They could ubiquitylate many human proteins, including components of various signaling pathways²,⁶,⁸.
+
+The Hippo pathway was first discovered in *Drosophila melanogaster* via genetic screens to look for genetic mutations leading to overgrowth phenotypes⁹,¹⁰. Nowadays, it is clear that the key components of the Hippo pathway are highly conserved from Drosophila to human, including MST1/2 and LATS1/2, two pairs of upstream kinases; YAP and TAZ, two downstream effectors and transcription coactivators; and the TEAD family of transcription factors⁹,¹¹. These core players and additional factors coordinate with each other to regulate transcription of Hippo target genes which controls organ size control, cell proliferation, survival, and pathophysiological events¹²–¹⁷. The Hippo pathway is also mediated by several types of protein modifications, including acetylation, methylation, phosphorylation, and ubiquitylation¹⁸–²⁵. Protein ubiquitylation has been shown to control the Hippo pathway by regulating protein stability or localization of several core proteins, such as YAP/TAZ, LATS1/2, MOB1, and MST1/2²⁶–³¹, but no evidence has been found for ubiquitylation to regulate transcription activities of the Hippo-related transcription factors, and no biological significance of ubiquitylation has been characterized for TEAD proteins, although a few ubiquitylation sites were identified in TEAD1, TEAD2, and TEAD4³²,³³.
+
+Here we report that RNF214, a RING finger-containing ubiquitin ligase whose biological functions have never been seriously explored, ubiquitylates the TEAD transcription factors at their C-terminal YAP binding domains (YBD) without affecting their protein stability or localization. However, RNF214 enhances the interactions between TEADs and YAP/TAZ via the recognition of polyubiquitin chains by YAP/TAZ, as well as expression levels of Hippo target genes mediated by YAP and TEADs. Moreover, we found that RNF214 is overexpressed in hepatocellular carcinoma (HCC) and promotes HCC tumorigenesis via the Hippo pathway as an oncogene. Our work uncovers a new mechanism regulating the downstream transcription network of the Hippo pathway by formation of a unique RNF214-TEAD-YAP signaling axis.
+
+# Results
+
+## 1. RNF214 interacts with the TEAD transcription factors
+
+RNF214 is a ubiquitin ligase of the RING finger family and an understudied protein whose biological roles were unknown. To figure out the functions of RNF214, we created RNF214 knockout mice (unpublished) and generated RNF214 knockout (RNF214<sup>−/−</sup>) mouse embryonic fibroblast cells (MEFs). CCK8 assays showcased that both the RNF214<sup>−/−</sup> and the RNF214<sup>+/−</sup> MEFs proliferated significantly slower than the RNF214<sup>+/+</sup> MEFs, whereas the RNF214<sup>−/−</sup> MEFs grew the slowest (Fig. <span class="InternalRef" refid="Fig1">1</span> a). These results implied RNF214 is associated with important biological processes.
+
+To determine biological functions of RNF214, it is critical to identify its interacting proteins and ubiquitylation substrates at first. Because ubiquitin ligases usually stay with its protein substrates transiently, we established an APEX proximity labeling strategy in coupled with mass spectrometry<sup><span citationid="CR34" class="CitationRef">34</span>, <span citationid="CR35" class="CitationRef">35</span></sup> to identify interacting proteins of RNF214. In this approach, we first fused an engineered ascorbate peroxidase (APEX2) to either N-terminus or C-terminus of RNF214, expressed these two fusion proteins in HLF, an HCC cell line, near the endogenous level (Extended Data Fig. <span class="InternalRef" refid="Fig1">1</span> a-b), and generated short-lived radicals around the APEX2-RNF214 fusion proteins to label biotin on nearby interactive proteins by adding hydrogen peroxide (H<sub>2</sub>O<sub>2</sub>) and biotin-phenol (also called biotin-tyramide) transiently. Biotinylated proteins were then isolated using Streptavidin resin for protein identification by mass spectrometry (Fig. <span class="InternalRef" refid="Fig1">1</span> b). Based on this procedure, we identified 511 potential interactors of RNF214 common to both N-terminal and C-terminal labeling (Fig. <span class="InternalRef" refid="Fig1">1</span> c). The KEGG pathway enrichment analysis revealed the Hippo pathway as the most prominent pathway to interact with RNF214 (Fig. <span class="InternalRef" refid="Fig1">1</span> d). Notably, all four human TEAD proteins, which are the final transcription factors of the Hippo pathway<sup><span citationid="CR36" class="CitationRef">36</span></sup>, were on the top of the list among potential interactors of RNF214 (Fig. <span class="InternalRef" refid="Fig1">1</span> d).
+
+Next, we confirmed the interaction between RNF214 and TEAD1 using a reciprocal co-immunoprecipitation method (Co-IP) after co-expressing Flag-RNF214 and Myc-TEAD1 in HEK293T cells (Fig. <span class="InternalRef" refid="Fig1">1</span> e-f). Similar interactions were observed between a Flag-tagged RNF214 and either Myc-tagged TEAD2 (Fig. <span class="InternalRef" refid="Fig1">1</span> g), or HA-tagged TEAD3 (Fig. <span class="InternalRef" refid="Fig1">1</span> h) and TEAD4 (Fig. <span class="InternalRef" refid="Fig1">1</span> i). To determine direct interactions between RNF214 and TEADs, we first purified GST-tagged TEAD1 (GST-TEAD1) recombinant proteins using a bacteria expression system (Extended Data Fig. <span class="InternalRef" refid="Fig1">1</span> c) and Strep-tagged RNF214 (Strep-RNF214) using the baculovirus-insect cell expression system (Extended Data Fig. <span class="InternalRef" refid="Fig1">1</span> d), and then performed a GST pulldown assay using these purified recombinant proteins (Fig. <span class="InternalRef" refid="Fig1">1</span> j). Indeed, GST-TEAD1 interacts with Strep-RNF214 directly. Furthermore, we found that HA-RNF214 interacts weakly with Flag-YAP, a critical effector of the Hippo pathway and a TEAD-binding protein, when HA-RNF214 and Flag-YAP proteins were co-expressed in HEK293T cells (Fig. <span class="InternalRef" refid="Fig1">1</span> k), establishing potential roles of RNF214 in the Hippo pathway.
+
+## 2. RNF214 enhances Hippo-regulated transcription
+
+Since RNF214 associates with multiple downstream transcription factors of the Hippo pathway, we decided to explore functions of RNF214 in gene expression regulated by Hippo signaling. We first knocked down RNF214 in Hep3b, using small interference RNA (siRNA) specifically targeting RNF214 and noticed that mRNA levels of three target genes, including ANKRD1, CTGF, and CYR61, of the TEAD transcription factors were significantly reduced in RNF214-knockdown cells (Fig. <span class="InternalRef" refid="Fig2">2</span> a and Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> a). Expressing an siRNA-resistant RNF214 cDNA reinstituted mRNA levels of these three genes, especially CTGF and CYR61, excluding any potential off-target issues of siRNA (Fig. <span class="InternalRef" refid="Fig2">2</span> a and Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> a). We reconfirmed these results by knocking down RNF214 in Huh7 HCC cell line (Fig. <span class="InternalRef" refid="Fig2">2</span> b and Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> b).
+
+Serum is a stimulating signal for YAP/TAZ activity and regulation of Hippo target genes<sup><span citationid="CR37" class="CitationRef">37</span></sup>. Indeed, the expression of these three target genes was blocked following serum starvation in HEK293A cells, and addition of serum resulted in their transcriptional enhancement as previously reported<sup><span citationid="CR37" class="CitationRef">37</span>, <span citationid="CR38" class="CitationRef">38</span></sup> (Fig. <span class="InternalRef" refid="Fig2">2</span> c). Silencing RNF214 inhibited enhanced expression of these three genes after serum stimulation (Fig. <span class="InternalRef" refid="Fig2">2</span> c), further suggesting that RNF214 participates in regulating expression of Hippo target genes.
+
+Since RNF214 is prominent for expression of CTGF, a <em>bona fide</em> transcription target of the TEAD transcription factors in the Hippo pathway<sup><span citationid="CR36" class="CitationRef">36</span></sup>, we performed a dual luciferase assay using the CTGF promoter controlling expression of the firefly luciferase in HEK293T cells. While TEAD2 alone only produced a small quantity of luciferase activities, co-expressing YAP made a big lift in luciferase activities. Moreover, adding different quantities of Flag-RNF214 further enhanced luciferase activities proportionally (Fig. <span class="InternalRef" refid="Fig2">2</span> d). Consistently, Flag-RNF214 magnified both TEAD1 and TEAD3-induced CTGF-luciferase activities (Fig. <span class="InternalRef" refid="Fig2">2</span> e), suggesting that RNF214 increases transcription activities of TEADs as a whole. We also employed the Gal4-TEAD4/9×UAS luciferase reporter assay<sup><span citationid="CR36" class="CitationRef">36</span>, <span citationid="CR38" class="CitationRef">38</span></sup>. Co-expressing Gal4-TEAD4 and YAP produced some luciferase activities, whereas, adding Flag-RNF214 further augmented luciferase activities significantly (Fig. <span class="InternalRef" refid="Fig2">2</span> f). Of note, the enhancement of luciferase activities was proportional to the amount of Flag-RNF214 expression plasmids transfected (Fig. <span class="InternalRef" refid="Fig2">2</span> f). All together, these data demonstrated RNF214 works together with the TEAD transcription factors to control expression of downstream target genes of the Hippo pathway.
+
+RNF214 is a family member of the RING finger ubiquitin ligases (Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> c). Therefore, we created an RNF214 mutant with its RING finger domain deleted (RD). Unlike the wild type RNF214, the RD mutant behaved like the transfection control in the Gal4-TEAD4/9xUAS luciferase assay when overexpressed in HEK293T cells (Fig. <span class="InternalRef" refid="Fig2">2</span> g). It also could not reinstate expression of ANKRD1, CTGF, and CYR61 when introduced into RNF214 knockdown Huh7 cells (Fig. <span class="InternalRef" refid="Fig2">2</span> h and Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> d), verifying the critical role of RNF214 as a ubiquitin ligase in the Hippo pathway. Beside the RING finger domain at its C-terminus, RNF214 contains a coiled-coil domain (Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> c). We then constructed an RNF214 mutant with its coiled-coil domain removed (CCD), and showcased that this CCD mutant was incapable of rescuing phenotypes of siRNF214 in both luciferase assay (Fig. <span class="InternalRef" refid="Fig2">2</span> i), and expression analysis (Fig. <span class="InternalRef" refid="Fig2">2</span> j and Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> e). The coiled-coil domain is often involved in protein-protein interactions, especially self-oligomerization of proteins harboring it<sup><span citationid="CR39" class="CitationRef">39</span></sup>. Thus, we speculated that the coiled-coil domain is used for RNF214’s oligomerization which is usually employed as a mechanism to activate certain ubiquitin ligases<sup><span citationid="CR7" class="CitationRef">7</span>, <span citationid="CR40" class="CitationRef">40</span>, <span citationid="CR41" class="CitationRef">41</span></sup> (Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> f). We then made both HA-tagged and Flag-tagged RNF214, and found RNF214 did self-associate with each other (Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> g). More importantly, this self-interaction depends on its coiled-coil domain (Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> h). These data might explain the reason why the CCD mutant could not rescue the siRNF214 phenotypes, implying a potential mechanism by which the coiled-coil domain functions in RNF214 activation.
+
+Finally, we turned to the N-terminal part ahead of the coiled-coil domain (Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> c). Deleting this part (NTD) did not affect self-association of RNF214 (Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> h), but reduced the interactions of RNF214 with TEAD1 and TEAD2 (Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> i-j). The NTD mutant showed some defect in the Gal4-TEAD4/9xUAS luciferase assay when overexpressed in HEK293T cells (Fig. <span class="InternalRef" refid="Fig2">2</span> k), and could not rescue the siRNF214 phenotypes in expression analysis of three Hippo target genes (Fig. <span class="InternalRef" refid="Fig2">2</span> l and Extended data Fig. <span class="InternalRef" refid="Fig2">2</span> k). Therefore, it appeared that RNF214 might employ the N-terminal part to recognize the TEAD transcription factors.
+
+## 3. RNF214 promotes nonproteolytic polyubiquitylation of TEADs
+
+RNF214 is a RING finger-containing ubiquitin ligase. As far as we know, no substrate has been identified for RNF214 yet. Having figured out that RNF214 interacts with the TEAD transcription factors and the RING finger domain of RNF214 is important for TEAD-regulated transcription, next we decided to determine whether RNF214 could ubiquitylate TEADs.
+
+Firstly, we co-expressed HA-tagged ubiquitin (HA-Ub) and Flag-tagged TEAD2 or TEAD3 in HEK293T cells, then employed anti-Flag antibody resins to immunoprecipitate Flag-TEAD2 (Fig. <span class="InternalRef" refid="Fig3">3</span> a) or Flag-TEAD3 (Fig. <span class="InternalRef" refid="Fig3">3</span> b). Anti-HA Western blotting showed both Flag-TEAD2(Fig. <span class="InternalRef" refid="Fig3">3</span> a upper panel)and Flag-TEAD3 (Fig. <span class="InternalRef" refid="Fig3">3</span> b upper panel) were heavily ubiquitylated. Overexpressing Myc-tagged RNF214 (Myc-RNF214) did not alter expression levels of TEADs, but greatly enhanced ubiquitylation of either Flag-TEAD2 (Fig. <span class="InternalRef" refid="Fig3">3</span> a) or Flag-TEAD3 (Fig. <span class="InternalRef" refid="Fig3">3</span> b). More significantly, the Myc-RNF214 RD mutant couldn’t increase ubiquitylation of either Flag-TEAD2 or Flag-TEAD3 proteins (Fig. <span class="InternalRef" refid="Fig3">3</span> a-b upper panel). These data confirmed that RNF214 is a ubiquitin ligase of the TEAD family proteins.
+
+Secondly, we expressed an Avi-tagged TEAD1 or TEAD2 in the HLF where we co-expressed BirA, a bacteria biotin ligase which conjugates biotin to the Avi-tag, a biotin-acceptor peptide. We then pulled out biotinylated Avi-TEAD1 (Bio-TEAD1) or Avi-TEAD2 (Bio-TEAD2) proteins using streptavidin resins under a denaturing buffer condition, detected TEAD1 and TEAD2 ubiquitylation using an anti-ubiquitin antibody. Clearly, Flag-RNF214 increased ubiquitylation of both TEAD1 (Extended data Fig. <span class="InternalRef" refid="Fig3">3</span> a) and TEAD2 (Fig. <span class="InternalRef" refid="Fig3">3</span> c). Using the same approach, we noticed that the wild type RNF214 could augment TEAD4 ubiquitylation, while the RD mutant couldn’t (Extended data Fig. <span class="InternalRef" refid="Fig3">3</span> b).
+
+Thirdly, we utilized the Halo-ThUBDs resin, a ubiquitin chain-binding matrix<sup><span citationid="CR42" class="CitationRef">42</span></sup>, to pull ubiquitylated proteins out of HLF cells. Using a pan-TEAD antibody, we confirmed ubiquitylation of TEAD proteins (Fig. <span class="InternalRef" refid="Fig3">3</span> d). When we knocked out RNF214 in HLF cells, TEAD ubiquitylation was largely reduced (Fig. <span class="InternalRef" refid="Fig3">3</span> d). Two independent RNF214-knockout clones produced similar results, indicating the observed phenotypes were not due to off-target effects from sgRNA.
+
+One main outcome of protein ubiquitylation is proteolysis in the proteasome. When RNF214 was either overexpressed (Fig. <span class="InternalRef" refid="Fig1">1</span>) or silenced (Fig. <span class="InternalRef" refid="Fig2">2</span> c and Extended data Fig. <span class="InternalRef" refid="Fig3">3</span> c) or knocked out (Fig. <span class="InternalRef" refid="Fig3">3</span> d), protein levels of TEADs were not changed. Besides, results from cycloheximide chase experiments showed that TEADs were stable proteins in Huh7 and HLF cells (Extended data Fig. <span class="InternalRef" refid="Fig3">3</span> d), implying that RNF214 promotes nonproteolytic ubiquitylation of TEADs. To determine whether RNF214 conjugates nonproteolytic polyubiquitin chains on TEADs, we co-expressed Flag-TEAD2, Myc-RNF214 and wide type or KR mutants of HA-Ub in HEK293T cells, observed that the K63R ubiquitin mutant reduced basal ubiquitylation of TEAD2 and couldn’t support RNF214-induced TEAD2 ubiquitylation (Fig. <span class="InternalRef" refid="Fig3">3</span> e-f and Extended data Fig. <span class="InternalRef" refid="Fig3">3</span> e-f), further indicating that RNF214 conjugates non-proteolytic K63 polyubiquitin chains on TEADs.
+
+## 4. RNF214 enhances the interactions between TEADs and YAP
+
+The TEAD family transcription factors have little transcriptional activity by themselves and require the presence of transcription coactivators YAP or TAZ to induce target gene transcription<sup>36,43−45</sup>. Thus, YAP/TAZ nuclear localization and interactions with TEADs are two critical steps for the TEADs-controlled transcription. Besides, as transcription factors, TEADs’ activities are also regulated by nuclear-cytoplasmic localization upon cellular stress, like many other transcription factors, such as NF-κB and SMAD<sup><span citationid="CR44" class="CitationRef">44</span>, <span citationid="CR46" class="CitationRef">46</span></sup>. Plus, it’s also well documented that non-proteolytic polyubiquitin chains could be signals for changes of protein subcellular localizations<sup><span citationid="CR47" class="CitationRef">47</span>, <span citationid="CR48" class="CitationRef">48</span></sup>. Since RNF214 conjugates nondegradable polyubiquitin chains on TEADs, we first wondered whether RNF214 could influence subcellular localizations of TEADs and YAP. In many cancer cells, YAP is always highly activated and accumulates in the nucleus<sup><span citationid="CR49" class="CitationRef">49</span>, <span citationid="CR50" class="CitationRef">50</span></sup>. As we observed in Huh7 cells, both TEAD1 and YAP mainly localized in the nucleus; silencing RNF214 had no effect on subcellular localization of TEAD1 and YAP (Extended data Fig. <span class="InternalRef" refid="Fig4">4</span> a-b). In the HEK293A cells, YAP mainly localizes in the cytoplasm and translocated into the nucleus under Nocodazole stimulation<sup><span citationid="CR51" class="CitationRef">51</span></sup>. Nocodazole disrupts microtubule polymerization and induces YAP dephosphorylation and nuclear translocation<sup><span citationid="CR51" class="CitationRef">51</span></sup>. The cytoplasmic-nuclear shuttling of YAP wasn’t blocked when RNF214 was silenced (Extended data Fig. <span class="InternalRef" refid="Fig4">4</span> c). In addition, RNF214 knockdown or overexpression had little impact on the protein levels of YAP/TAZ and nonphosphorylated YAP (active YAP) in Hep3b or HEK293A cells (Extended data Fig. <span class="InternalRef" refid="Fig3">3</span> c and Extended data Fig. <span class="InternalRef" refid="Fig4">4</span> d).
+
+Next, we asked whether RNF214 could influence the interactions between YAP and TEADs. As shown in Fig. <span class="InternalRef" refid="Fig4">4</span> a, HA-RNF214 boosted the interaction between Flag-YAP and Myc-TEAD2. Similar results were achieved for the interaction between HA-TEAD4 and Flag-YAP (Fig. <span class="InternalRef" refid="Fig4">4</span> b). Interestingly, the enhancement of interaction between HA-TEAD4 and Flag-YAP largely disappeared when the RD mutant of RNF214 was employed (Fig. <span class="InternalRef" refid="Fig4">4</span> b). Alike results were observed for the interaction between HA-TEAD1 and Flag-YAP (Extended data Fig. <span class="InternalRef" refid="Fig4">4</span> e), further echoing the importance of ubiquitylation activity of RNF214 in regulating the Hippo pathway.
+
+To further demonstrate the ubiquitylation of TEADs by RNF214 is important for their interactions with YAP and their transcriptional activities, we intended to identify the ubiquitylation sites of TEADs in an RNF214-dependent manner. Since RNF214-enhanced interactions between YAP and TEADs depend on its ubiquitin ligase activity (Fig. <span class="InternalRef" refid="Fig4">4</span> b and extended Fig. <span class="InternalRef" refid="Fig4">4</span> e), we turned to the YAP binding domains (YBD) of TEADs. There are eight lysine residues which are conserved among the YBD domains of TEADs. Excluding those lysine residues on the YAP binding surface or those potentially affecting structural stability of TEADs<sup><span additionalcitationids="CR53" citationid="CR52" class="CitationRef">52</span>–<span citationid="CR54" class="CitationRef">54</span></sup>, we focused on four lysine residues of TEAD2 (Fig. <span class="InternalRef" refid="Fig4">4</span> c). We made two TEAD2 mutants containing lysine-to-arginine (KR) substitutions on these lysine residues (K345R and 3KR containing K280R, K281R, and K351R), and analyzed ubiquitylation of TEAD2 mutants in HEK293T cells. In comparison with the wild type, the K345R mutant, rather than the 3KR mutant, completely lost RNF214-mediated ubiquitylation of TEAD2 (Fig. <span class="InternalRef" refid="Fig4">4</span> d-e). Moreover, RNF214 failed to enhance the interaction between YAP and the K345R mutant of TEAD2 (Fig. <span class="InternalRef" refid="Fig4">4</span> f). In comparison to the wild type TEAD2, the K345R mutant fails to fully support RNF214-induced CTGF-driven luciferase activities in HEK293T cells (Fig. <span class="InternalRef" refid="Fig4">4</span> g). Similarly, adding Flag-RNF214 couldn’t rescue the luciferase activities by the K260R mutant of Gal4-TEAD4 as high as the wide type Gal4-TEAD4, in TEAD1/3/4 knockdown HEK293T cells (Fig. <span class="InternalRef" refid="Fig4">4</span> h-i). Together, these data confirmed the importance of RNF214 in the Hippo-mediated transcription via ubiquitylating TEADs at a single lysine site.
+
+Finally, we wondered how TEAD ubiquitylation by RNF214 affects their interactions with YAP. One possibility is that YAP might possess polyubiquitin chain binding feature to enhance YAP recruiting to the TEAD transcriptional complex. To verify this hypothesis, we did a GST pulldown assay using purified recombinant proteins, including GST-YAP or GST-TAZ and synthetic polyubiquitin chains. We observed both YAP and TAZ directly bound to K48 and K63 polyubiquitin chains (Fig. <span class="InternalRef" refid="Fig4">4</span> j-k). These data might explain why TEADs ubiquitylation promotes their interactions with YAP.
+
+All together, these data mechanistically demonstrated the importance of TEADs ubiquitylation by RNF214 in interactions with YAP/TAZ.
+
+## 5. Overexpression of RNF214 correlates with poor prognosis in HCC
+
+YAP and TEAD proteins are the key downstream effectors in the Hippo pathway and oncogenic proteins in common cancer types<sup>13,49,55−57</sup>. Because RNF214 ubiquitylates TEAD proteins and promotes interactions between TEADs and YAP, we wondered whether RNF214 is also an oncogene implicating in tumorigenesis. We first analyzed expression profiles of RNF214 in the cancer-based TCGA database. Interestingly, we found that mRNA levels of RNF214 is upregulated in HCC (Fig. <span class="InternalRef" refid="Fig5">5</span> a). Similar results were obtained from Tiger, another cancer-related database<sup><span citationid="CR58" class="CitationRef">58</span></sup> (Extended data Fig. <span class="InternalRef" refid="Fig5">5</span> a). Compared with non-tumor tissues, HCC tumor samples contain much higher mRNA levels of RNF214 (Fig. <span class="InternalRef" refid="Fig5">5</span> a and Extended data Fig. <span class="InternalRef" refid="Fig5">5</span> a). Further Kaplan-Meier analysis of overall survival and progression-free survival in the TCGA database showed a reverse correlation between RNF214 expression level and the survival probability (Fig. <span class="InternalRef" refid="Fig5">5</span> b and Extended data Fig. <span class="InternalRef" refid="Fig5">5</span> b). We then examined protein expression levels of RNF214 in a published dataset containing protein quantifications of 6478 genes between 159 pairs of tumor and non-tumor samples<sup><span citationid="CR59" class="CitationRef">59</span></sup>, and found that protein expression levels of RNF214 are higher in tumor samples than in paracancerous ones (Fig. <span class="InternalRef" refid="Fig5">5</span> c).
+
+To further validate the results of these statistical analysis, we compared RNF214 expression levels between 176 pairs of HCC tumor samples and paracancerous tissues from Zhejiang Provincial People’s Hospital using an immunohistochemistry (IHC) approach, and noticed that RNF214 was overexpressed in tumor samples among 92 pairs, accounting for 52.3% (Fig. <span class="InternalRef" refid="Fig5">5</span> d-f), indicating that HCC tumor samples in over half of HCC patients possessed upregulated protein levels of RNF214. Meanwhile, we analyzed the correlation between RNF214 expression levels and differentiation grades of 275 cases of HCC patients based on our IHC results, uncovered that more than half of cases with either medium or low differentiation grades displayed high expression level (IHC score≥6) of RNF214 with a R value at 0.26 (Fig. <span class="InternalRef" refid="Fig5">5</span> g-i), implicating RNF214 might contribute to malignancy of HCC. Comparable results were acquired based on the correlation of RNF214 expression and Edmonson-Steiner grade, another crucial prognosticator in HCC (Extended data Fig. <span class="InternalRef" refid="Fig5">5</span> c-d). In this case, 79.6% and 43.3% patients of grade III and IV had high expression levels of RNF214 (IHC score≥6), respectively. Moreover, we observed that RNF214 protein levels were closely associated with serum expression levels of alpha-fetoprotein (AFP), a <em>bona fide</em> liver cancer biomarker (Extended data Fig. <span class="InternalRef" refid="Fig5">5</span> e). To further consolidate our clinical analysis, we measured RNF214 expression in HCC cell lines using a Western blotting approach and noted that RNF214 protein levels were elevated in all seven HCC cell lines examined compared with HL7702, a normal liver cell line (Fig. <span class="InternalRef" refid="Fig5">5</span> j).
+
+We then examined the relationships between RNF214 and Hippo-regulated gene expression in liver cancer samples. Through Spearman’s rank correlation analysis of the TCGA cohort, we observed a positive correlation between RNF214 and YAP/TAZ-TEAD target genes (e.g., <em>AMOTL2</em>, <em>CTGF</em>, <em>CYR61</em>, <em>ANKRD1</em>, <em>AXL</em>, <em>BCL2</em>, <em>CCND1</em>, and <em>CDH2</em>) (Fig. <span class="InternalRef" refid="Fig5">5</span> k-m and Extended data Fig. <span class="InternalRef" refid="Fig5">5</span> f-j). Besides, RNF214 expression positively correlated with YAP and TAZ expression, as well as TEAD1-4 expression (Extended data Fig. <span class="InternalRef" refid="Fig5">5</span> k-l).
+
+Together, these data suggested that RNF214 could be a critical oncogene and tightly associated with enhanced YAP/TAZ-TEAD transcription activities in promoting HCC tumorigenesis.
+
+## 6. RNF214 is critical for HCC tumorigenesis
+
+To investigate the functions of RNF214 in HCC, we first knocked out RNF214 in HLF cells, using the CRISPR/Cas9 method and saw that all of the small guide RNAs (sgRNAs) slowed the growth of HLF cells (Fig. <span class="InternalRef" refid="Fig6">6</span> a). These RNF214 knockout HLF cells produced a smaller number of colonies than control cells in a colony formation assay (Fig. <span class="InternalRef" refid="Fig6">6</span> b upper panel). Quantitative analysis demonstrated that the differences between the RNF214 knockout and control cells were significant statistically (Fig. <span class="InternalRef" refid="Fig6">6</span> b bottom panel). Similar phenotype was detected in Huh7 cells as well when RNF214 was knocked out (Extended data Fig. <span class="InternalRef" refid="Fig6">6</span> a). Meanwhile, we silenced RNF214 in Hep3b cells, using a small hairpin RNA (shRNA) method and demonstrated that RNF214-silenced Hep3b cells produced fewer colonies than control shRNA cells (Extended data Fig. <span class="InternalRef" refid="Fig6">6</span> b). Moreover, we knocked down RNF214 in Hep3b cells using siRNA oligoes and noted that RNF214-silenced cells propagated slower than control cells (Fig. <span class="InternalRef" refid="Fig6">6</span> c). Importantly, an siRNA-resistant cDNA of RNF214 could resume the proliferation rate of Hep3b cells to a large extent (Fig. <span class="InternalRef" refid="Fig6">6</span> c), indicating the authenticity of these phenotypes. Conversely, we overexpressed RNF214 in Huh1, an HCC cell line with relatively low expression of RNF214 (Fig. <span class="InternalRef" refid="Fig5">5</span> j) and observed that the number of colonies was at least doubled when RNF214 was overexpressed (Extended data Fig. <span class="InternalRef" refid="Fig6">6</span> c). Furthermore, we found that RNF214 could foster proliferation of Hep3b cells upon overexpressed (Extended data Fig. <span class="InternalRef" refid="Fig6">6</span> d). Together, these results evidenced that RNF214 is a positive regulator of HCC cell proliferation.
+
+To study the roles of RNF214 in migration and invasion of HCC cells, we first examined the migration ability of Hep3b cells in a wound-healing assay and found that RNF214 knockdown cells migrated slower than control Hep3b cells (Extended data Fig. <span class="InternalRef" refid="Fig6">6</span> e). We also performed the transwell migration assay (without Matrigel) and the transwell invasion assay (with Matrigel), respectively. While control Hep3b cells possess excellent abilities of migration and invasion, knocking down RNF214 using an siRNA oligo reduced abilities of migration, especially invasion remarkably (Fig. <span class="InternalRef" refid="Fig6">6</span> d). An siRNA-resistant cDNA of RNF214 rescued these phenotypes, implicating that these phenotypes were authentic (Fig. <span class="InternalRef" refid="Fig6">6</span> d). Altogether, these data further indicated that RNF214 is an oncogene in HCC.
+
+Both YAP and TEADs play eminent roles in cancer development, progression and metastasis, including HCC tumorigenesis<sup>43,55,60−67</sup>. Having found that RNF214 functions as a positive regulator of YAP/TAZ-TEAD transcriptional complex and promotes tumor cell properties, we decided to determine whether RNF214 is critical for the oncogenic activities of YAP and TEADs in HCC. Phosphorylation of YAP at Serine-127 results in its cytoplasmic retention, whereas the non-phosphorylatable S127A mutant becomes constitutively active in the nucleus<sup><span citationid="CR26" class="CitationRef">26</span></sup>. We first created both HLF and Huh7 cell lines expressing the S127A YAP mutant using the tetracycline-inducible (Tet/on) gene expression system, then knocked down RNF214 using siRNA. Indeed, the S127A mutant induced higher expression levels of these three Hippo target genes in both HLF and Huh7 cells (Extended data Fig. <span class="InternalRef" refid="Fig6">6</span> f-g). More consistently, RNF214 knockdown could decrease mRNA expression levels of three Hippo target genes at both basal and YAP S127A-induced levels (Extended data Fig. <span class="InternalRef" refid="Fig6">6</span> f-g). Overexpressing the S127A mutant of YAP strengthened migration of Huh7 cells profoundly (Fig. <span class="InternalRef" refid="Fig6">6</span> e). However, silencing RNF214 significantly reduced cell migration under both basal and overexpressed conditions of YAP (Fig. <span class="InternalRef" refid="Fig6">6</span> e).
+
+To further evaluate the roles of RNF214 in HCC tumorigenesis, we employed a subcutaneous xenograft mouse model. We subcutaneously injected 2 x 10<sup>6</sup> Huh7 cells with Matrigel into 5-week-old male BALB/c nude mice. Overall, RNF214 knockdown Huh7 cells grew into tumors much slower than the shRNA control cells in nude mice (Fig. <span class="InternalRef" refid="Fig6">6</span> f). Tumors grew from RNF214-silenced Huh7 cells were much smaller than those from shRNA control cells (Fig. <span class="InternalRef" refid="Fig6">6</span> g-i). More relevantly, RNF214-silenced tumors expressed tremendously lower amount of CYR61’s mRNAs and proteins than control tumors (Fig. <span class="InternalRef" refid="Fig6">6</span> j-k). Furthermore, RNF214 knockout Huh7 cells produced smaller tumors than control cells when subcutaneously injected into nude mice (Extended data Fig. <span class="InternalRef" refid="Fig6">6</span> h-i). Together, our data concluded that RNF214 promotes HCC development and progression via governing the downstream effect of the Hippo pathway and is a <em>bona fide</em> oncogene in HCC.
+
+# Discussion
+
+Protein ubiquitylation is pivotal for many essential cellular activities. Components of the ubiquitin signaling pathway have been implicated in tumor initiation, progression and metastasis in both positive and negative ways. Protein ubiquitylation is also a specific process and the specificity is mainly maintained by ubiquitin ligases which consist of the RING finger family and the HECT domain family<sup>7</sup>. RNF214 belongs to the family of the RING finger ubiquitin ligases, but its function is unknown except as a candidate gene potentially in milk lactose regulation based on a GWAS study<sup>68</sup>. By combining an APEX2 proximity labeling method and mass spectrometry, we identified the TEAD family proteins, major transcription factors of the Hippo pathway, as main interactors of RNF214. Human genome encodes four TEAD proteins and all of them emerged in our mass spectrometry analysis. Using a series of biochemical approaches, we validated the interactions of RNF214 with the TEAD family proteins and provided strong evidences supporting RNF214 as the first ubiquitin ligase of the TEAD proteins.
+
+As transcription factors, TEADs orchestrate transcription of genes related to development, cell growth and organ size, and oncogenesis etc. with YAP/TAZ, transcriptional coactivators and major downstream effectors of the Hippo pathway<sup>36,43,44</sup>. Post-translational modifications have been shown to regulate functions of TEAD proteins. For example, phosphorylation of TEAD1 by either protein kinase C or protein kinase A can significantly reduce DNA binding activity of TEAD1<sup>24,25</sup>, whereas, palmitoylation of TEADs is crucial for their proper folding and protein stability maintenance<sup>69–71</sup>. Our data indicated that ubiquitylation of TEADs by RNF214 is important to their functions as downstream transcription factors of the Hippo pathway. We found that the interactions between TEADs and YAP are profoundly increased by the RNF214 ubiquitin ligase. More significantly, the ubiquitylation activity of RNF214 is important for their interactions, and transcription of YAP-TEAD-regulated genes. Although the ubiquitin signaling pathway has been linked to the Hippo pathway by regulating protein stabilities and localizations of several key components in Hippo signaling, such as YAP/TAZ, LATS1/2, MOB1, and MST1/2<sup>8,18</sup>, there were no previous studies to make any connections between ubiquitylation and the biological activities of TEADs. Our data also demonstrated that RNF214 conjugates non-proteolytic polyubiquitin chain on a single lysine site of TEADs. Since ubiquitylation of TEADs is important for their interactions with YAP, we speculated that YAP or additional YAP-associated proteins might possess polyubiquitin chain binding features to enhance YAP recruiting to the TEAD transcriptional complex. Indeed, our results of *in vitro* GST pulldown assays indicated that YAP and TAZ possess ubiquitin-binding abilities and suggested that a conserved domain between YAP and TAZ may act as a polyubiquitin-binding domain. Further studies are needed to answer this question.
+
+The Hippo pathway has been implicated in tumorigenesis, with MST1/2 and LATS1/2 kinases as tumor suppressors, but YAP/TAZ and TEADs as oncogenes<sup>9,17,72</sup>. YAP and TEADs have been proposed to be new therapeutic targets in cancer therapy<sup>73–76</sup>. Indeed, small molecule inhibitors disrupting the interactions between TEADs and YAP are under development as cancer drugs<sup>38,77−80</sup>. By clinic data and biological analysis, we proved that RNF214 is an oncogene of HCC and an important regulator of the YAP-TEAD transcription complex in general. Therefore, adding RNF214 to the axis of YAP-TEAD could offer a new angle to invent unique therapeutic tools to kill cancer cells, especially HCC ones by managing transcriptional activities of the TEAD and YAP/TAZ complex.
+
+# Materials and Methods
+
+## Cell culture and transfection
+HEK293T, HEK293A, HLF, HLE, HepG2, Huh7, Hep3b, Huh1, and MEF cells were maintained in DMEM supplemented with 10% fetal bovine serum. HL7702 and Snu449 were cultured in RPMI1640 supplemented with 10% fetal bovine serum. MEF cells were isolated from 13.5 days’ mouse embryos. All cells were incubated at 37 ℃, with 5% CO₂. Plasmids were transfected into cells using Lipofectamine 3000 (Invitrogen) according to the manufacturer’s protocol.
+
+## Antibodies and reagents
+Antibodies used in this study were as follows: RNF214 (202826-T38, Sino Biological), YAP (sc-101199, Santa Cruz/SC), YAP/TAZ (8418, Cell signaling technology/CST), pan-TEAD (13295, CST), TEAD1 (610922, BD Biosciences), TEAD2 (21159-1-AP, Proteintech), TEAD4 (ab58310, Abcam), ANKRD1 (11427-1-AP, Proteintech), CTGF(sc-365970, SC), CYR61 (sc-374129, SC), Ubiquitin (3936, CST), GAPDH (AC002, ABclonal), actin (AC026, ABclonal), Flag (F3165, Sigma), HA (51064-2-AP, Proteintech), Myc (3800-1, Clontech), Strep (A00626, Genescript). The home-made Anti-RNF214 antibody (J044) was raised in rabbits using peptide corresponding to amino acids 259–273 of human RNF214. Nocodazole was from Selleck (S2775) and stored in DMSO.
+
+## Plasmids
+Human RNF214 coding sequence was amplified using Polymerase Chain Reaction (PCR) from a human cDNA library made in Jianping Jin’s laboratory, then subcloned into a Gateway entry plasmid pENTR-1W and validated by sequencing. Truncated mutants of RNF214 were made by PCR-based mutagenesis and confirmed by sequencing. Expression plasmids for CTGF-Luciferase, Gal4-TEAD4, 9xUAS-Luciferase, CMV-β-galactosidase, Flag-YAP, Flag-YAP-S127A, Myc-TEAD1, and Myc-TEAD2 were generously provided by Bin Zhao’s laboratory, some of them were subcloned into pENTR-1W. pENTR-TEAD3 plasmid was from the Invitrogen ORF Clones library at the core facility of Life Sciences Institute, Zhejiang University. Entry clones were shuttled into different destination vectors through LR reaction (Gateway LR Clonase Ⅱ Enzyme Mix, Invitrogen). pLenti-CRISPR-puro vector was used to construct sgRNA plasmids. The gRNA sequences against RNF214 were as follows:
+sgRNF214-1: GTCTGAGGTTGCTGGTGTTG;
+sgRNF214-2: GAATAACAGAAATGTCCATT;
+sgRNF214-3: GAGCACTCAGAGCAGAATCC;
+sgRNF214-4: GCCTTAAGAGACCGAGTGAC.
+pLKO-ccdB-puro vector was employed to make shRNA plasmids. The shRNA sequence against RNA214 was CCATCCTAAAGGAAGGTAACA.
+
+## Analysis of RNF214 in HCC Microarray
+Human HCC tissues (n = 275) and adjacent non-tumor tissues (n = 256) microarray chips were created in Department of Pathology, Zhejiang Provincial People’s Hospital. 176 cases of the tissue microarray were paired samples. All cases of HCC tissues and non-tumor tissues were diagnosed clinically and pathologically. All samples were received from the patients who underwent surgical resection and signed informed consent before their operations. The protocol was approved by the Medical Ethic Committee of Zhejiang Provincial People's Hospital.
+
+Immunohistochemistry (IHC) was performed to detect protein levels of RNF214 on HCC microarray chips. The degree of immunostaining was reviewed and scored independently by two pathologists based on staining intensity and extent. Staining intensity was classified as 0 (negative), 1 (weak), 2 (moderate) and 3 (strong). Staining extent was divided into 0 (＜5%), 1 (5%-25%), 2 (26%-50%), 3 (51%-75%) and 4 (＞75%) depending on the percentage of positive cells. IHC Score = staining intensity x staining extent.
+
+## Small interference RNA
+siRNAs were transfected into HCC cells using Lipofectamine RNAiMAX Reagent (Invitrogen) according to the manufacturer’s protocol. After 96 hours, cells were harvested. The siRNA sequences were as follows:
+siCK: GAUCCGCAGCGACAUCAACCU;
+siRNF214: CAAAUCCCUACUCCCACUUUA;
+
+## Lentivirus production and stable cell line generation
+Lentiviruses were produced by transfecting lentiviral vectors carrying target gene sequences together with the packing plasmids of psPAX2 and pMD2G into HEK293T cells using PEI. After 48 hours, supernatants containing lentivirus particles were collected to infect host cells using a spin infection method. Stable cells were selected in the presence of puromycin (Sangon Biotech).
+
+## Proliferation and colony formation assays
+For proliferation assay, the viability of HCC cells was quantified by Cell Counting Kit-8 (CCK8, K1018, APExBIO). Cells with indicated treatments were seeded into 96-well plates, incubated for the corresponding days and after 2 hours of incubation with CCK8 reagents at 37 ℃, absorbances at 450 nm were recorded using a microplate reader (TECAN). For colony formation assay, 6-well plates were seeded with 10³ viable cells and incubated for the days as indicated. At the end of the experiments, the colonies were fixed in methanol and then stained with 0.1% crystal violet. The colonies with＞50 cells were counted under the microscope.
+
+## Cell migration and invasion assays
+For wound-healing assays, cells were seeded in 6-well plates, grown to 100% confluence in a monolayer and then starved in serum-free DMEM overnight. After a scratch was made with a sterile pipette tip, the cells were washed with PBS and sequentially fed with serum-free DMEM. Images were acquired immediately following the “wounds” were made, and every 12 hours via a microscope at 4x magnifications.
+
+Transwell chambers (Corning) with and without precoated Matrigel were used to determine cell migration and invasion, respectively. Briefly, 6 x 10⁴ cells in 300 µl serum-free DMEM were plated in transwell inserts and then 500 µl culture medium containing 10% FBS was added to the lower chamber. After 48 hours, the cells in the upper chamber of the transwell were removed with a cotton swab, the migrated cells were fixed in methanol and stained with 0.1% crystal violet. Cells in three randomly selected fields were photographed and statistically analyzed.
+
+## Luciferase reporter assay
+For the CTGF luciferase assay, HEK293T cells were transfected with CTGF-Luc plasmid containing a firefly luciferase under the control of CTGF promoter, a Renilla luciferase plasmid as a transfection control and indicated gene expression plasmids. All values were normalized for transfection efficiency against Renilla luciferase activities. The other reporter assay was carried by transfection HEK293T cells with Gal4-TEAD4, 9xUAS-Luc, CMV-β-gal, and indicated plasmids. Luciferase activities were normalized to β-gal activities. 24 hours after transfection, cells were lysed and luciferase activities were measured using the Dual Luciferase Reporter Assay System (Vazyme).
+
+## RNA extraction and qRT-PCR
+Total RNAs were isolated using Trizol reagent (Sangon Biotech). cDNAs were prepared using HiScript Ⅲ 1st Strand cDNA Synthesis Kit (+ gDNA wiper) (Vazyme) according to the manufacturer’s protocol. The qRT-PCR analysis was performed by the SYBR green method (YEASEN). The sequences of the PCR primers for corresponding human gene were listed as follows (5’-3’):
+ANKRD1: CACTTCTAGCCCACCCTGTGA (Forward),
+CCACAGGTTCCGTAATGATTT (Reverse);
+CTGF: CCAATGACAACGCCTCCTG (Forward),
+TGGTGCAGCCAGAAAGCTC (Reverse);
+CYR61: GGGCTGGAATGCAACTTCG (Forward),
+GGCGCCATCAATACATGTGC (Reverse);
+GAPDH: AGGGCTGCTTTTAACTCTGGT (Forward),
+CCCCTACTTGATTTTGGAGGGA (Reverse).
+
+## Mass spectrometry analysis through APEX2-catalyzed biotinylation
+APEX2 was fused at either the N- or C-terminus of RNF214. Fusion proteins were expressed in HLF cells using a lentivirus infection method and expressed at levels comparable to the endogenous RNF214 proteins. Cells were then incubated with 2 mM biotin-phenol (APExBIO) in the DMEM supplemented with 10% fetal bovine serum for 30 mins at 37 ℃. Consequently, one-minute pulse with 0.25 mM H₂O₂ at room temperature was stopped with ice-cold quenching buffer (5 mM Trolox [Sigma], 10 mM sodium ascorbate [Sigma], and 10 mM sodium azide in PBS). All samples were washed three times with quenching buffer and then harvested.
+
+Cell pellets were lysed in 6M urea buffer (6 M urea, 100 mM Tris-HCl [pH 7.5], 200 mM NaCl and 1% SDS). After a short sonication, lysates were clarified by centrifugation at 15,000 rpm and quantified using the BCA kit. Streptavidin beads (Smart-lifesciences) were washed with lysis buffer. 3 mg of each sample was mixed with 10 µl Streptavidin beads. The suspensions were gently rotated at 25 ℃ overnight. The beads were then washed with 6M urea buffer five times and bound biotinylated proteins were subjected to mass spectrometry analysis. Briefly, after the reduction/alkylation reactions on beads, samples were precipitated by methanol/chloroform. After centrifugation, keep the white middle layer (protein precipitation) and add trypsin into sample solution at 37 ℃ for 12–16 h. After trypsin digestion, the samples were desalted by Ziptip (Millipore) and lyophilized for 2 h. The peptides were loaded into timsTOF Pro (Bruker), and the data were analyzed by PEAKS online.
+
+To reveal the biological pathways of 511 proteins unique to samples from both fusion proteins, KEGG pathway enrichment analysis was performed using 'clusterProfiler' R package.
+
+## Western blotting and co-immunoprecipitation (Co-IP)
+Cells were lysed in lysis buffer (1%SDS and 30 µM Tris-HCl [pH6.8]). Total proteins (10 µg) were separated on SDS-PAGE and then transferred onto PVDF membranes (Millipore). After blocking using 5% nonfat milk, membranes were incubated with the gene-specific primary antibodies, then HRP-conjugated secondary antibody (Jackson ImmunoResearch), and visualized using ECL reagents (YEASEN).
+
+For Co-IP, 24 hours after transfection, cell lysates were lysed in 1% Triton lysis buffer (50 mM Tris-HCl [pH 7.5], 1 mM EDTA, 150 mM NaCl and 1% Triton X-100) containing protease and phosphatase inhibitors. The lysates were subjected to Co-IP using specific antibody-conjugated agarose (Sigma) for 2 hours. After extensive washes, immunoprecipitated proteins were separated on SDS-PAGE, transferred to PVDF membranes and detected by Western blotting with appropriate antibodies.
+
+## Immunofluorescence
+Cells were cultured on glass coverslips for 24 hours. After washing with PBS, cells were incubated with 4% paraformaldehyde for 10 minutes and then permeabilized with 0.2% Triton X-100 for 10 mins at room temperature. The cells were then blocked in 5% BSA and incubated with primary antibody at room temperature for 1 hour, washed three times with PBST (0.1% Tween 20 in PBS) and incubated with Alexa Fluor 488 or 546 antibody (1:1000, Thermo Fisher Scientific) for 1 hour at room temperature. After 3 washes, all coverslips were mounted with ProLong Gold antifade with DAPI reagent (Thermo Fisher Scientific). Fluorescence images were captured by Zeiss Axiovert 200M microscope.
+
+*In vivo* ubiquitylation assays
+To detect ubiquitylation of TEADs in HLF cells, a biotinylated Avi-tagged TEAD (Bio-TEAD) was introduced into HLF cells co-expressing BirA, a bacteria biotin ligase which conjugates biotin to the Avi-tag, a biotin-acceptor peptide using lentivirus expression system. Biotin at 2 µg/mL was added to culture media overnight before cell harvest. Cells were then lysed in 6M urea buffer. After sonication, lysates were cleared using centrifugation and incubated with Streptavidin-agarose resins overnight at room temperature. Subsequently, the pulldown products were washed five times using 6M urea buffer. Ubiquitylated TEADs were detected by Western blotting using an anti-ubiquitin antibody. Alternatively, HEK293T cells were co-transfected with Flag-TEADs, Myc-RNF214 and HA-Ub. Cells were lysed in SDS-denaturing buffer (62.5 mM Tris-HCl [pH 6.8], 2% SDS, 10% glycerol) and sonicated. Cleared cell lysates were then diluted 10 to 15-fold in native lysis buffer (50 mM Tris-HCl [pH 7.5], 0.5% Triton X-100, 200 mM NaCl, 10% glycerol). The supernatants were incubated with anti-Flag beads at 4 ℃ for 2 hours. The immunocomplexes were washed five times using native lysis buffer, resolved on SDS-PAGE, and immunoblotted using anti-HA antibody.
+
+For the Halo-ThUBDs assay, we expressed and purified ThUBDs, the ubiquitin affinity matrix<sup><span citationid="CR42" class="CitationRef">42</span></sup>, which binds selectively to polyubiquitin chains, as Halo-tagged recombinant proteins (Halo-ThUBDs) in BL21(DE3) bacteria cells. Proteins were extracted from HCC cells with 1% Triton lysis buffer containing protease, phosphatase inhibitors and 10 mM N-Ethylmaleimide. A total of 8 µg Halo-ThUBDs recombinant proteins were incubated with 2 mg total lysates from each sample for 3 hours at 4 ℃. The Halo beads were then washed three times and eluted with SDS sample loading buffer, separated on SDS-PAGE, and detected using Western blotting.
+
+## Pulldown Assay
+GST-TEAD1 was expressed and purified from BL21 (DE3) bacteria cells. Strep-RNF214 was purified from SF9 insect cell infected by recombinant baculovirus constructed using Bac-to-Bac™ Baculovirus Expression System (Invitrogen). HA-RNF214 and truncations were expressed using the TNT Coupled Reticulocyte Lysate System (Promega). Proteins bound on beads were mixed with different prey proteins at 4 ℃ for 2 hours in 1% Triton lysis buffer, and then washed five times using the same buffer. The input and pulldown samples were loaded to SDS-PAGE and detected by Ponceau S staining or Western blotting.
+
+For *in vitro* polyubiquitin chain binding assay, GST-YAP and GST-TAZ were expressed and purified from BL21 (DE3) bacteria cells. Poly-K48 Ubiquitin (3–7) and Poly-K63 Ubiquitin (3–7) were purchased from R&D Systems.
+
+## Subcutaneous xenograft model
+A total of 2 x 10⁶ Huh7 cells with indicated treatments were suspended in 100 µl PBS with Matrigel (1:1) and then injected into 5-week-old nude mice. 9 days after injection, the subcutaneous tumors were counted and tumor sizes were measured every 2 days using the Vernier caliper as follows: tumor volume = (L x W²)/2, where L is the long axis and W is the short. After 21 days of injection, mice were sacrificed and tumors were harvested, weighed and photographed. We used a humane protocol in our xenograft tumor growth assay with the endpoints of tumor volume＜1500 mm³.
+
+All animal experiments were approved by the Animal Ethics Committee of Zhejiang University. All mice used were male BALB/c nude mice obtained from GemPharmatech Co., Ltd (Nanjing, China).
+
+## Statistical analysis
+Data are presented as the mean ± SD and three levels of significance (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001) were presented. Statistical analyses used Student’s t-test, Spearman’s correlation analysis, log-rank test and Cox regression analysis with GraphPad Prism software v 7.0 (San Diego, CA. USA). The website links used for statistical analyses of expression and prognosis were provided in Extended data.
+
+# References
+
+1. Swatek, K. N. & Komander, D. Ubiquitin modifications. *Cell research* **26**, 399-422, doi:10.1038/cr.2016.39 (2016).
+2. Nguyen, V. N., Huang, K. Y., Weng, J. T., Lai, K. R. & Lee, T. Y. UbiNet: an online resource for exploring the functional associations and regulatory networks of protein ubiquitylation. *Database (Oxford)* **2016**, doi:10.1093/database/baw054 (2016).
+3. Yau, R. & Rape, M. The increasing complexity of the ubiquitin code. *Nature cell biology* **18**, 579-586, doi:10.1038/ncb3358 (2016).
+4. Popovic, D., Vucic, D. & Dikic, I. Ubiquitination in disease pathogenesis and treatment. *Nature medicine* **20**, 1242-1253, doi:10.1038/nm.3739 (2014).
+5. Rape, M. Ubiquitylation at the crossroads of development and disease. *Nature reviews. Molecular cell biology* **19**, 59-70, doi:10.1038/nrm.2017.83 (2018).
+6. Deng, L., Meng, T., Chen, L., Wei, W. & Wang, P. The role of ubiquitination in tumorigenesis and targeted drug discovery. *Signal Transduct Target Ther* **5**, 11, doi:10.1038/s41392-020-0107-0 (2020).
+7. Zheng, N. & Shabek, N. Ubiquitin Ligases: Structure, Function, and Regulation. *Annu Rev Biochem* **86**, 129-157, doi:10.1146/annurev-biochem-060815-014922 (2017).
+8. Nguyen, T. H. & Kugler, J. M. Ubiquitin-Dependent Regulation of the Mammalian Hippo Pathway: Therapeutic Implications for Cancer. *Cancers* **10**, doi:10.3390/cancers10040121 (2018).
+9. Ma, S., Meng, Z., Chen, R. & Guan, K. L. The Hippo Pathway: Biology and Pathophysiology. *Annu Rev Biochem* **88**, 577-604, doi:10.1146/annurev-biochem-013118-111829 (2019).
+10. Li, F. L. & Guan, K. L. The two sides of Hippo pathway in cancer. *Semin Cancer Biol* **85**, 33-42, doi:10.1016/j.semcancer.2021.07.006 (2022).
+11. Misra, J. R. & Irvine, K. D. The Hippo Signaling Network and Its Biological Functions. *Annu Rev Genet* **52**, 65-87, doi:10.1146/annurev-genet-120417-031621 (2018).
+12. Koo, J. H. & Guan, K. L. Interplay between YAP/TAZ and Metabolism. *Cell Metab* **28**, 196-206, doi:10.1016/j.cmet.2018.07.010 (2018).
+13. Taha, Z., Janse van Rensburg, H. J. & Yang, X. The Hippo Pathway: Immunity and Cancer. *Cancers* **10**, doi:10.3390/cancers10040094 (2018).
+14. White, S. M., Murakami, S. & Yi, C. The complex entanglement of Hippo-Yap/Taz signaling in tumor immunity. *Oncogene* **38**, 2899-2909, doi:10.1038/s41388-018-0649-6 (2019).
+15. Zheng, Y. & Pan, D. The Hippo Signaling Pathway in Development and Disease. *Developmental cell* **50**, 264-282, doi:10.1016/j.devcel.2019.06.003 (2019).
+16. Yimlamai, D. et al. Hippo pathway activity influences liver cell fate. *Cell* **157**, 1324-1338, doi:10.1016/j.cell.2014.03.060 (2014).
+17. Driskill, J. H. & Pan, D. The Hippo Pathway in Liver Homeostasis and Pathophysiology. *Annu Rev Pathol* **16**, 299-322, doi:10.1146/annurev-pathol-030420-105050 (2021).
+18. Yan, F., Qian, M., He, Q., Zhu, H. & Yang, B. The posttranslational modifications of Hippo-YAP pathway in cancer. *Biochim Biophys Acta Gen Subj* **1864**, 129397, doi:10.1016/j.bbagen.2019.07.006 (2020).
+19. Moroishi, T. et al. A YAP/TAZ-induced feedback mechanism regulates Hippo pathway homeostasis. *Genes & development* **29**, 1271-1284, doi:10.1101/gad.262816.115 (2015).
+20. Oudhoff, M. J. et al. Control of the hippo pathway by Set7-dependent methylation of Yap. *Developmental cell* **26**, 188-194, doi:10.1016/j.devcel.2013.05.025 (2013).
+21. Fang, L. et al. SET1A-Mediated Mono-Methylation at K342 Regulates YAP Activation by Blocking Its Nuclear Export and Promotes Tumorigenesis. *Cancer Cell* **34**, 103-118.e109, doi:10.1016/j.ccell.2018.06.002 (2018).
+22. Mao, B. et al. SIRT1 regulates YAP2-mediated cell proliferation and chemoresistance in hepatocellular carcinoma. *Oncogene* **33**, 1468-1474, doi:10.1038/onc.2013.88 (2014).
+23. Peng, C. et al. Regulation of the Hippo-YAP Pathway by Glucose Sensor O-GlcNAcylation. *Molecular cell* **68**, 591-604.e595, doi:10.1016/j.molcel.2017.10.010 (2017).
+24. Gupta, M. P., Kogut, P. & Gupta, M. Protein kinase-A dependent phosphorylation of transcription enhancer factor-1 represses its DNA-binding activity but enhances its gene activation ability. *Nucleic acids research* **28**, 3168-3177, doi:10.1093/nar/28.16.3168 (2000).
+25. Jiang, S. W., Dong, M., Trujillo, M. A., Miller, L. J. & Eberhardt, N. L. DNA binding of TEA/ATTS domain factors is regulated by protein kinase C phosphorylation in human choriocarcinoma cells. *The Journal of biological chemistry* **276**, 23464-23470, doi:10.1074/jbc.M010934200 (2001).
+26. Zhao, B., Li, L., Tumaneng, K., Wang, C.-Y. & Guan, K.-L. A coordinated phosphorylation by Lats and CK1 regulates YAP stability through SCFβ-TRCP. *Genes & development* **24**, 72-85, doi:10.1101/gad.1843810 (2010).
+27. Liu, C. Y. et al. The hippo tumor pathway promotes TAZ degradation by phosphorylating a phosphodegron and recruiting the SCF{beta}-TrCP E3 ligase. *The Journal of biological chemistry* **285**, 37159-37169, doi:10.1074/jbc.M110.152942 (2010).
+28. Li, W. et al. Merlin/NF2 loss-driven tumorigenesis linked to CRL4(DCAF1)-mediated inhibition of the hippo pathway kinases Lats1 and 2 in the nucleus. *Cancer Cell* **26**, 48-60, doi:10.1016/j.ccr.2014.05.001 (2014).
+29. Ma, B. et al. Hypoxia regulates Hippo signalling through the SIAH2 ubiquitin E3 ligase. *Nature cell biology* **17**, 95-103, doi:10.1038/ncb3073 (2015).
+30. Lignitto, L. et al. Proteolysis of MOB1 by the ubiquitin ligase praja2 attenuates Hippo signalling and supports glioblastoma growth. *Nature communications* **4**, 1822, doi:10.1038/ncomms2791 (2013).
+31. Ma, X., Guo, X., Richardson, H. E., Xu, T. & Xue, L. POSH regulates Hippo signaling through ubiquitin-mediated expanded degradation. *Proceedings of the National Academy of Sciences of the United States of America* **115**, 2150-2155, doi:10.1073/pnas.1715165115 (2018).
+32. Akimov, V. et al. UbiSite approach for comprehensive mapping of lysine and N-terminal ubiquitination sites. *Nat Struct Mol Biol* **25**, 631-640, doi:10.1038/s41594-018-0084-y (2018).
+33. Kim, W. et al. Systematic and quantitative assessment of the ubiquitin-modified proteome. *Molecular cell* **44**, 325-340, doi:10.1016/j.molcel.2011.08.025 (2011).
+34. Tan, B. et al. An Optimized Protocol for Proximity Biotinylation in Confluent Epithelial Cell Cultures Using the Peroxidase APEX2. *STAR Protoc* **1**, 100074, doi:10.1016/j.xpro.2020.100074 (2020).
+35. Hung, V. et al. Spatially resolved proteomic mapping in living cells with the engineered peroxidase APEX2. *Nat Protoc* **11**, 456-475, doi:10.1038/nprot.2016.018 (2016).
+36. Zhao, B. et al. TEAD mediates YAP-dependent gene induction and growth control. *Genes & development* **22**, 1962-1971, doi:10.1101/gad.1664408 (2008).
+37. Yu, F. X. et al. Regulation of the Hippo-YAP pathway by G-protein-coupled receptor signaling. *Cell* **150**, 780-791, doi:10.1016/j.cell.2012.06.037 (2012).
+38. Ji, X. et al. Cyclopeptide RA-V Inhibits Organ Enlargement and Tumorigenesis Induced by YAP Activation. *Cancers* **10**, doi:10.3390/cancers10110449 (2018).
+39. Lupas, A. Coiled coils: new structures and new functions. *Trends Biochem Sci* **21**, 375-382 (1996).
+40. Li, Y. et al. Structural insights into the TRIM family of ubiquitin E3 ligases. *Cell research* **24**, 762-765, doi:10.1038/cr.2014.46 (2014).
+41. Welcker, M. et al. Fbw7 dimerization determines the specificity and robustness of substrate degradation. *Genes & development* **27**, 2531-2536, doi:10.1101/gad.229195.113 (2013).
+42. Gao, Y. et al. Enhanced Purification of Ubiquitinated Proteins by Engineered Tandem Hybrid Ubiquitin-binding Domains (ThUBDs). *Mol Cell Proteomics* **15**, 1381-1396, doi:10.1074/mcp.o115.051839 (2016).
+43. Fu, M. et al. The Hippo signalling pathway and its implications in human health and diseases. *Signal Transduction and Targeted Therapy* **7**, doi:10.1038/s41392-022-01191-9 (2022).
+44. Lin, K. C., Park, H. W. & Guan, K. L. Regulation of the Hippo Pathway Transcription Factor TEAD. *Trends Biochem Sci* **42**, 862-872, doi:10.1016/j.tibs.2017.09.003 (2017).
+45. Yagi, R., Chen, L. F., Shigesada, K., Murakami, Y. & Ito, Y. A WW domain-containing yes-associated protein (YAP) is a novel transcriptional co-activator. *The EMBO journal* **18**, 2551-2562, doi:10.1093/emboj/18.9.2551 (1999).
+46. Lin, K. C. et al. Regulation of Hippo pathway transcription factor TEAD by p38 MAPK-induced cytoplasmic translocation. *Nature cell biology* **19**, 996-1002, doi:10.1038/ncb3581 (2017).
+47. Yang, W. L. et al. The E3 ligase TRAF6 regulates Akt ubiquitination and activation. *Science* **325**, 1134-1138, doi:10.1126/science.1175065 (2009).
+48. Song, M. S. et al. The deubiquitinylation and localization of PTEN are regulated by a HAUSP-PML network. *Nature* **455**, 813-817, doi:10.1038/nature07290 (2008).
+49. Dey, A., Varelas, X. & Guan, K. L. Targeting the Hippo pathway in cancer, fibrosis, wound healing and regenerative medicine. *Nature reviews. Drug discovery* **19**, 480-494, doi:10.1038/s41573-020-0070-z (2020).
+50. Maugeri-Saccà, M. & De Maria, R. The Hippo pathway in normal development and cancer. *Pharmacology & therapeutics* **186**, 60-72, doi:10.1016/j.pharmthera.2017.12.011 (2018).
+51. Zhao, B. et al. Cell detachment activates the Hippo pathway via cytoskeleton reorganization to induce anoikis. *Genes & development* **26**, 54-68, doi:10.1101/gad.173435.111 (2012).
+52. Tian, W., Yu, J., Tomchick, D. R., Pan, D. & Luo, X. Structural and functional analysis of the YAP-binding domain of human TEAD2. *Proceedings of the National Academy of Sciences of the United States of America* **107**, 7293-7298, doi:10.1073/pnas.1000293107 (2010).
+53. Chen, L. et al. Structural basis of YAP recognition by TEAD4 in the hippo pathway. *Genes & development* **24**, 290-300, doi:10.1101/gad.1865310 (2010).
+54. Li, Z. et al. Structural insights into the YAP and TEAD complex. *Genes & development* **24**, 235-240, doi:10.1101/gad.1865810 (2010).
+55. Sohn, B. H. et al. Inactivation of Hippo Pathway Is Significantly Associated with Poor Prognosis in Hepatocellular Carcinoma. *Clinical cancer research : an official journal of the American Association for Cancer Research* **22**, 1256-1264, doi:10.1158/1078-0432.Ccr-15-1447 (2016).
+56. Sanchez-Vega, F. et al. Oncogenic Signaling Pathways in The Cancer Genome Atlas. *Cell* **173**, 321-337.e310, doi:10.1016/j.cell.2018.03.035 (2018).
+57. Huh, H. D., Kim, D. H., Jeong, H. S. & Park, H. W. Regulation of TEAD Transcription Factors in Cancer Biology. *Cells* **8**, doi:10.3390/cells8060600 (2019).
+58. Chaisaingmongkol, J. et al. Common Molecular Subtypes Among Asian Hepatocellular Carcinoma and Cholangiocarcinoma. *Cancer Cell* **32**, 57-70.e53, doi:10.1016/j.ccell.2017.05.009 (2017).
+59. Gao, Q. et al. Integrated Proteogenomic Characterization of HBV-Related Hepatocellular Carcinoma. *Cell* **179**, 561-577.e522, doi:10.1016/j.cell.2019.08.052 (2019).
+60. Russell, J. O. & Camargo, F. D. Hippo signalling in the liver: role in development, regeneration and disease. *Nat Rev Gastroenterol Hepatol* **19**, 297-312, doi:10.1038/s41575-021-00571-w (2022).
+61. Wang, H. et al. Distinct and Overlapping Roles of Hippo Effectors YAP and TAZ During Human and Mouse Hepatocarcinogenesis. *Cellular and molecular gastroenterology and hepatology* **11**, 1095-1117, doi:10.1016/j.jcmgh.2020.11.008 (2021).
+62. Zhao, B. et al. Inactivation of YAP oncoprotein by the Hippo pathway is involved in cell contact inhibition and tissue growth control. *Genes & development* **21**, 2747-2761, doi:10.1101/gad.1602907 (2007).
+63. Dimri, M. & Satyanarayana, A. Molecular Signaling Pathways and Therapeutic Targets in Hepatocellular Carcinoma. *Cancers* **12**, doi:10.3390/cancers12020491 (2020).
+64. Park, S. et al. Transcription factors TEAD2 and E2A globally repress acetyl-CoA synthesis to promote tumorigenesis. *Molecular cell* **82**, 4246-4261.e4211, doi:10.1016/j.molcel.2022.10.027 (2022).
+65. Han, Y. Analysis of the role of the Hippo pathway in cancer. *Journal of translational medicine* **17**, 116, doi:10.1186/s12967-019-1869-4 (2019).
+66. Kuenzi, B. M. & Ideker, T. A census of pathway maps in cancer systems biology. *Nat Rev Cancer* **20**, 233-246, doi:10.1038/s41568-020-0240-7 (2020).
+67. Liu, Y., Wang, X. & Yang, Y. Hepatic Hippo signaling inhibits development of hepatocellular carcinoma. *Clin Mol Hepatol* **26**, 742-750, doi:10.3350/cmh.2020.0178 (2020).
+68. Lopdell, T. J. et al. DNA and RNA-sequence based GWAS highlights membrane-transport genes as key modulators of milk lactose content. *BMC genomics* **18**, 968, doi:10.1186/s12864-017-4320-3 (2017).
+69. Noland, C. L. et al. Palmitoylation of TEAD Transcription Factors Is Required for Their Stability and Function in Hippo Pathway Signaling. *Structure (London, England : 1993)* **24**, 179-186, doi:10.1016/j.str.2015.11.005 (2016).
+70. Chan, P. et al. Autopalmitoylation of TEAD proteins regulates transcriptional output of the Hippo pathway. *Nat Chem Biol* **12**, 282-289, doi:10.1038/nchembio.2036 (2016).
+71. Kim, N. G. & Gumbiner, B. M. Cell contact and Nf2/Merlin-dependent regulation of TEAD palmitoylation and activity. *Proceedings of the National Academy of Sciences of the United States of America* **116**, 9877-9882, doi:10.1073/pnas.1819400116 (2019).
+72. Johnson, R. & Halder, G. The two faces of Hippo: targeting the Hippo pathway for regenerative medicine and cancer treatment. *Nature reviews. Drug discovery* **13**, 63-79, doi:10.1038/nrd4161 (2014).
+73. Cunningham, R. & Hansen, C. G. The Hippo pathway in cancer: YAP/TAZ and TEAD as therapeutic targets in cancer. *Clinical science (London, England : 1979)* **136**, 197-222, doi:10.1042/cs20201474 (2022).
+74. Xiao, Y. & Dong, J. The Hippo Signaling Pathway in Cancer: A Cell Cycle Perspective. *Cancers* **13**, doi:10.3390/cancers13246214 (2021).
+75. Thompson, B. J. YAP/TAZ: Drivers of Tumor Growth, Metastasis, and Resistance to Therapy. *Bioessays* **42**, e1900162, doi:10.1002/bies.201900162 (2020).
+76. Pobbati, A. V. & Hong, W. A combat with the YAP/TAZ-TEAD oncoproteins for cancer therapy. *Theranostics* **10**, 3622-3635, doi:10.7150/thno.40889 (2020).
+77. Barrette, A. M. et al. Anti-invasive efficacy and survival benefit of the YAP-TEAD inhibitor verteporfin in preclinical glioblastoma models. *Neuro Oncol* **24**, 694-707, doi:10.1093/neuonc/noab244 (2022).
+78. Jiao, S. et al. A peptide mimicking VGLL4 function acts as a YAP antagonist therapy against gastric cancer. *Cancer Cell* **25**, 166-180, doi:10.1016/j.ccr.2014.01.010 (2014).
+79. Bum-Erdene, K. et al. Small-Molecule Covalent Modification of Conserved Cysteine Leads to Allosteric Inhibition of the TEAD⋅Yap Protein-Protein Interaction. *Cell Chem Biol* **26**, 378-389.e313, doi:10.1016/j.chembiol.2018.11.010 (2019).
+80. Holden, J. K. & Cunningham, C. N. Targeting the Hippo Pathway and Cancer through the TEAD Family of Transcription Factors. *Cancers* **10**, doi:10.3390/cancers10030081 (2018).
+
+# Supplementary Files
+
+- [LinetalRNF214manuscriptsupSubmit.docx](https://assets-eu.researchsquare.com/files/rs-2832184/v1/2a3d3d78fbdda796ff66a708.docx)